diff --git a/vendor/github.com/DataDog/zstd/.travis.yml b/vendor/github.com/DataDog/zstd/.travis.yml
new file mode 100644
index 0000000..88c152e
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/.travis.yml
@@ -0,0 +1,18 @@
+language: go
+
+go:
+  - 1.8.x
+  - 1.9.x
+  - 1.10.x
+
+os:
+  - linux
+  - osx
+
+install:
+  - "wget http://sun.aei.polsl.pl/~sdeor/corpus/mr.bz2"
+  - "bzip2 -d mr.bz2"
+script:
+  - "go build"
+  - "PAYLOAD=`pwd`/mr go test -v"
+  - "PAYLOAD=`pwd`/mr go test -bench ."
diff --git a/vendor/github.com/DataDog/zstd/LICENSE b/vendor/github.com/DataDog/zstd/LICENSE
new file mode 100644
index 0000000..345c1eb
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/LICENSE
@@ -0,0 +1,27 @@
+Simplified BSD License
+
+Copyright (c) 2016, Datadog <info@datadoghq.com>
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+      this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright notice,
+      this list of conditions and the following disclaimer in the documentation
+      and/or other materials provided with the distribution.
+    * Neither the name of the copyright holder nor the names of its contributors
+      may be used to endorse or promote products derived from this software
+      without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
+FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
+OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/vendor/github.com/DataDog/zstd/README.md b/vendor/github.com/DataDog/zstd/README.md
new file mode 100644
index 0000000..6c02e16
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/README.md
@@ -0,0 +1,120 @@
+# Zstd Go Wrapper
+
+[C Zstd Homepage](https://github.com/Cyan4973/zstd)
+
+The current headers and C files are from *v1.3.4* (Commit
+[2555975](https://github.com/facebook/zstd/releases/tag/v1.3.4)).
+
+## Usage
+
+There are two main APIs:
+
+* simple Compress/Decompress
+* streaming API (io.Reader/io.Writer)
+
+The compress/decompress APIs mirror that of lz4, while the streaming API was
+designed to be a drop-in replacement for zlib.
+
+### Simple `Compress/Decompress`
+
+
+```go
+// Compress compresses the byte array given in src and writes it to dst.
+// If you already have a buffer allocated, you can pass it to prevent allocation
+// If not, you can pass nil as dst.
+// If the buffer is too small, it will be reallocated, resized, and returned bu the function
+// If dst is nil, this will allocate the worst case size (CompressBound(src))
+Compress(dst, src []byte) ([]byte, error)
+```
+
+```go
+// CompressLevel is the same as Compress but you can pass another compression level
+CompressLevel(dst, src []byte, level int) ([]byte, error)
+```
+
+```go
+// Decompress will decompress your payload into dst.
+// If you already have a buffer allocated, you can pass it to prevent allocation
+// If not, you can pass nil as dst (allocates a 4*src size as default).
+// If the buffer is too small, it will retry 3 times by doubling the dst size
+// After max retries, it will switch to the slower stream API to be sure to be able
+// to decompress. Currently switches if compression ratio > 4*2**3=32.
+Decompress(dst, src []byte) ([]byte, error)
+```
+
+### Stream API
+
+```go
+// NewWriter creates a new object that can optionally be initialized with
+// a precomputed dictionary. If dict is nil, compress without a dictionary.
+// The dictionary array should not be changed during the use of this object.
+// You MUST CALL Close() to write the last bytes of a zstd stream and free C objects.
+NewWriter(w io.Writer) *Writer
+NewWriterLevel(w io.Writer, level int) *Writer
+NewWriterLevelDict(w io.Writer, level int, dict []byte) *Writer
+
+// Write compresses the input data and write it to the underlying writer
+(w *Writer) Write(p []byte) (int, error)
+
+// Close flushes the buffer and frees C zstd objects
+(w *Writer) Close() error
+```
+
+```go
+// NewReader returns a new io.ReadCloser that will decompress data from the
+// underlying reader.  If a dictionary is provided to NewReaderDict, it must
+// not be modified until Close is called.  It is the caller's responsibility
+// to call Close, which frees up C objects.
+NewReader(r io.Reader) io.ReadCloser
+NewReaderDict(r io.Reader, dict []byte) io.ReadCloser
+```
+
+### Benchmarks (benchmarked with v0.5.0)
+
+The author of Zstd also wrote lz4. Zstd is intended to occupy a speed/ratio
+level similar to what zlib currently provides.  In our tests, the can always
+be made to be better than zlib by chosing an appropriate level while still
+keeping compression and decompression time faster than zlib.
+
+You can run the benchmarks against your own payloads by using the Go benchmarks tool.
+Just export your payload filepath as the `PAYLOAD` environment variable and run the benchmarks:
+
+```go
+go test -bench .
+```
+
+Compression of a 7Mb pdf zstd (this wrapper) vs [czlib](https://github.com/DataDog/czlib):
+```
+BenchmarkCompression               5     221056624 ns/op      67.34 MB/s
+BenchmarkDecompression           100      18370416 ns/op     810.32 MB/s
+
+BenchmarkFzlibCompress             2     610156603 ns/op      24.40 MB/s
+BenchmarkFzlibDecompress          20      81195246 ns/op     183.33 MB/s
+```
+
+Ratio is also better by a margin of ~20%.
+Compression speed is always better than zlib on all the payloads we tested;
+However, [czlib](https://github.com/DataDog/czlib) has optimisations that make it
+faster at decompressiong small payloads:
+
+```
+Testing with size: 11... czlib: 8.97 MB/s, zstd: 3.26 MB/s
+Testing with size: 27... czlib: 23.3 MB/s, zstd: 8.22 MB/s
+Testing with size: 62... czlib: 31.6 MB/s, zstd: 19.49 MB/s
+Testing with size: 141... czlib: 74.54 MB/s, zstd: 42.55 MB/s
+Testing with size: 323... czlib: 155.14 MB/s, zstd: 99.39 MB/s
+Testing with size: 739... czlib: 235.9 MB/s, zstd: 216.45 MB/s
+Testing with size: 1689... czlib: 116.45 MB/s, zstd: 345.64 MB/s
+Testing with size: 3858... czlib: 176.39 MB/s, zstd: 617.56 MB/s
+Testing with size: 8811... czlib: 254.11 MB/s, zstd: 824.34 MB/s
+Testing with size: 20121... czlib: 197.43 MB/s, zstd: 1339.11 MB/s
+Testing with size: 45951... czlib: 201.62 MB/s, zstd: 1951.57 MB/s
+```
+
+zstd starts to shine with payloads > 1KB
+
+### Stability - Current state: STABLE
+
+The C library seems to be pretty stable and according to the author has been tested and fuzzed.
+
+For the Go wrapper, the test cover most usual cases and we have succesfully tested it on all staging and prod data.
diff --git a/vendor/github.com/DataDog/zstd/ZSTD_LICENSE b/vendor/github.com/DataDog/zstd/ZSTD_LICENSE
new file mode 100644
index 0000000..a793a80
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/ZSTD_LICENSE
@@ -0,0 +1,30 @@
+BSD License
+
+For Zstandard software
+
+Copyright (c) 2016-present, Facebook, Inc. All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+
+ * Neither the name Facebook nor the names of its contributors may be used to
+   endorse or promote products derived from this software without specific
+   prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/vendor/github.com/DataDog/zstd/bitstream.h b/vendor/github.com/DataDog/zstd/bitstream.h
new file mode 100644
index 0000000..f7f389f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/bitstream.h
@@ -0,0 +1,471 @@
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2017, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include "mem.h"            /* unaligned access routines */
+#include "error_private.h"  /* error codes and messages */
+
+
+/*-*************************************
+*  Debug
+***************************************/
+#if defined(BIT_DEBUG) && (BIT_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+#define STREAM_ACCUMULATOR_MIN_32  25
+#define STREAM_ACCUMULATOR_MIN_64  57
+#define STREAM_ACCUMULATOR_MIN    ((U32)(MEM_32bits() ? STREAM_ACCUMULATOR_MIN_32 : STREAM_ACCUMULATOR_MIN_64))
+
+
+/*-******************************************
+*  bitStream encoding API (write forward)
+********************************************/
+/* bitStream can mix input from multiple sources.
+ * A critical property of these streams is that they encode and decode in **reverse** direction.
+ * So the first bit sequence you add will be the last to be read, like a LIFO stack.
+ */
+typedef struct
+{
+    size_t bitContainer;
+    unsigned bitPos;
+    char*  startPtr;
+    char*  ptr;
+    char*  endPtr;
+} BIT_CStream_t;
+
+MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC, void* dstBuffer, size_t dstCapacity);
+MEM_STATIC void   BIT_addBits(BIT_CStream_t* bitC, size_t value, unsigned nbBits);
+MEM_STATIC void   BIT_flushBits(BIT_CStream_t* bitC);
+MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC);
+
+/* Start with initCStream, providing the size of buffer to write into.
+*  bitStream will never write outside of this buffer.
+*  `dstCapacity` must be >= sizeof(bitD->bitContainer), otherwise @return will be an error code.
+*
+*  bits are first added to a local register.
+*  Local register is size_t, hence 64-bits on 64-bits systems, or 32-bits on 32-bits systems.
+*  Writing data into memory is an explicit operation, performed by the flushBits function.
+*  Hence keep track how many bits are potentially stored into local register to avoid register overflow.
+*  After a flushBits, a maximum of 7 bits might still be stored into local register.
+*
+*  Avoid storing elements of more than 24 bits if you want compatibility with 32-bits bitstream readers.
+*
+*  Last operation is to close the bitStream.
+*  The function returns the final size of CStream in bytes.
+*  If data couldn't fit into `dstBuffer`, it will return a 0 ( == not storable)
+*/
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+    const char* limitPtr;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+/* Start by invoking BIT_initDStream().
+*  A chunk of the bitStream is then stored into a local register.
+*  Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
+*  You can then retrieve bitFields stored into the local register, **in reverse order**.
+*  Local register is explicitly reloaded from memory by the BIT_reloadDStream() method.
+*  A reload guarantee a minimum of ((8*sizeof(bitD->bitContainer))-7) bits when its result is BIT_DStream_unfinished.
+*  Otherwise, it can be less than that, so proceed accordingly.
+*  Checking if DStream has reached its end can be performed with BIT_endOfDStream().
+*/
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC, size_t value, unsigned nbBits);
+/* faster, but works only if value is "clean", meaning all high bits above nbBits are 0 */
+
+MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC);
+/* unsafe version; does not check buffer overflow */
+
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+    assert(val != 0);
+    {
+#   if defined(_MSC_VER)   /* Visual */
+        unsigned long r=0;
+        _BitScanReverse ( &r, val );
+        return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+        return 31 - __builtin_clz (val);
+#   else   /* Software version */
+        static const unsigned DeBruijnClz[32] = { 0,  9,  1, 10, 13, 21,  2, 29,
+                                                 11, 14, 16, 18, 22, 25,  3, 30,
+                                                  8, 12, 20, 28, 15, 17, 24,  7,
+                                                 19, 27, 23,  6, 26,  5,  4, 31 };
+        U32 v = val;
+        v |= v >> 1;
+        v |= v >> 2;
+        v |= v >> 4;
+        v |= v >> 8;
+        v |= v >> 16;
+        return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+#   endif
+    }
+}
+
+/*=====    Local Constants   =====*/
+static const unsigned BIT_mask[] = {
+    0,          1,         3,         7,         0xF,       0x1F,
+    0x3F,       0x7F,      0xFF,      0x1FF,     0x3FF,     0x7FF,
+    0xFFF,      0x1FFF,    0x3FFF,    0x7FFF,    0xFFFF,    0x1FFFF,
+    0x3FFFF,    0x7FFFF,   0xFFFFF,   0x1FFFFF,  0x3FFFFF,  0x7FFFFF,
+    0xFFFFFF,   0x1FFFFFF, 0x3FFFFFF, 0x7FFFFFF, 0xFFFFFFF, 0x1FFFFFFF,
+    0x3FFFFFFF, 0x7FFFFFFF}; /* up to 31 bits */
+#define BIT_MASK_SIZE (sizeof(BIT_mask) / sizeof(BIT_mask[0]))
+
+/*-**************************************************************
+*  bitStream encoding
+****************************************************************/
+/*! BIT_initCStream() :
+ *  `dstCapacity` must be > sizeof(size_t)
+ *  @return : 0 if success,
+ *            otherwise an error code (can be tested using ERR_isError()) */
+MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
+{
+    bitC->bitContainer = 0;
+    bitC->bitPos = 0;
+    bitC->startPtr = (char*)startPtr;
+    bitC->ptr = bitC->startPtr;
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer);
+    if (dstCapacity <= sizeof(bitC->bitContainer)) return ERROR(dstSize_tooSmall);
+    return 0;
+}
+
+/*! BIT_addBits() :
+ *  can add up to 31 bits into `bitC`.
+ *  Note : does not check for register overflow ! */
+MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
+                            size_t value, unsigned nbBits)
+{
+    MEM_STATIC_ASSERT(BIT_MASK_SIZE == 32);
+    assert(nbBits < BIT_MASK_SIZE);
+    assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
+    bitC->bitPos += nbBits;
+}
+
+/*! BIT_addBitsFast() :
+ *  works only if `value` is _clean_, meaning all high bits above nbBits are 0 */
+MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
+                                size_t value, unsigned nbBits)
+{
+    assert((value>>nbBits) == 0);
+    assert(nbBits + bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    bitC->bitContainer |= value << bitC->bitPos;
+    bitC->bitPos += nbBits;
+}
+
+/*! BIT_flushBitsFast() :
+ *  assumption : bitContainer has not overflowed
+ *  unsafe version; does not check buffer overflow */
+MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC)
+{
+    size_t const nbBytes = bitC->bitPos >> 3;
+    assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    MEM_writeLEST(bitC->ptr, bitC->bitContainer);
+    bitC->ptr += nbBytes;
+    assert(bitC->ptr <= bitC->endPtr);
+    bitC->bitPos &= 7;
+    bitC->bitContainer >>= nbBytes*8;
+}
+
+/*! BIT_flushBits() :
+ *  assumption : bitContainer has not overflowed
+ *  safe version; check for buffer overflow, and prevents it.
+ *  note : does not signal buffer overflow.
+ *  overflow will be revealed later on using BIT_closeCStream() */
+MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
+{
+    size_t const nbBytes = bitC->bitPos >> 3;
+    assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    MEM_writeLEST(bitC->ptr, bitC->bitContainer);
+    bitC->ptr += nbBytes;
+    if (bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
+    bitC->bitPos &= 7;
+    bitC->bitContainer >>= nbBytes*8;
+}
+
+/*! BIT_closeCStream() :
+ *  @return : size of CStream, in bytes,
+ *            or 0 if it could not fit into dstBuffer */
+MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
+{
+    BIT_addBitsFast(bitC, 1, 1);   /* endMark */
+    BIT_flushBits(bitC);
+    if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
+    return (bitC->ptr - bitC->startPtr) + (bitC->bitPos > 0);
+}
+
+
+/*-********************************************************
+*  bitStream decoding
+**********************************************************/
+/*! BIT_initDStream() :
+ *  Initialize a BIT_DStream_t.
+ * `bitD` : a pointer to an already allocated BIT_DStream_t structure.
+ * `srcSize` must be the *exact* size of the bitStream, in bytes.
+ * @return : size of stream (== srcSize), or an errorCode if a problem is detected
+ */
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    bitD->start = (const char*)srcBuffer;
+    bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+    } else {
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+        case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
+                /* fall-through */
+
+        case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
+                /* fall-through */
+
+        case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
+                /* fall-through */
+
+        case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
+                /* fall-through */
+
+        case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
+                /* fall-through */
+
+        case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
+                /* fall-through */
+
+        default: break;
+        }
+        {   BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+            bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
+            if (lastByte == 0) return ERROR(corruption_detected);  /* endMark not present */
+        }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
+{
+    return bitContainer >> start;
+}
+
+MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
+{
+#if defined(__BMI__) && defined(__GNUC__) && __GNUC__*1000+__GNUC_MINOR__ >= 4008  /* experimental */
+#  if defined(__x86_64__)
+    if (sizeof(bitContainer)==8)
+        return _bextr_u64(bitContainer, start, nbBits);
+    else
+#  endif
+        return _bextr_u32(bitContainer, start, nbBits);
+#else
+    assert(nbBits < BIT_MASK_SIZE);
+    return (bitContainer >> start) & BIT_mask[nbBits];
+#endif
+}
+
+MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
+{
+    assert(nbBits < BIT_MASK_SIZE);
+    return bitContainer & BIT_mask[nbBits];
+}
+
+/*! BIT_lookBits() :
+ *  Provides next n bits from local register.
+ *  local register is not modified.
+ *  On 32-bits, maxNbBits==24.
+ *  On 64-bits, maxNbBits==56.
+ * @return : value extracted */
+MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
+{
+#if defined(__BMI__) && defined(__GNUC__)   /* experimental; fails if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8 */
+    return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
+#else
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
+#endif
+}
+
+/*! BIT_lookBitsFast() :
+ *  unsafe version; only works if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
+{
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    assert(nbBits >= 1);
+    return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+/*! BIT_readBits() :
+ *  Read (consume) next n bits from local register and update.
+ *  Pay attention to not read more than nbBits contained into local register.
+ * @return : extracted value. */
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BIT_readBitsFast() :
+ *  unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BIT_lookBitsFast(bitD, nbBits);
+    assert(nbBits >= 1);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BIT_reloadDStream() :
+ *  Refill `bitD` from buffer previously set in BIT_initDStream() .
+ *  This function is safe, it guarantees it will not read beyond src buffer.
+ * @return : status of `BIT_DStream_t` internal register.
+ *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->limitPtr) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    /* start < ptr < limitPtr */
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD->bitContainer), otherwise bitD->ptr == bitD->start */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream() :
+ * @return : 1 if DStream has _exactly_ reached its end (all bits consumed).
+ */
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
diff --git a/vendor/github.com/DataDog/zstd/compiler.h b/vendor/github.com/DataDog/zstd/compiler.h
new file mode 100644
index 0000000..e90a3bc
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/compiler.h
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPILER_H
+#define ZSTD_COMPILER_H
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+/* force inlining */
+#if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#  define INLINE_KEYWORD inline
+#else
+#  define INLINE_KEYWORD
+#endif
+
+#if defined(__GNUC__)
+#  define FORCE_INLINE_ATTR __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#  define FORCE_INLINE_ATTR __forceinline
+#else
+#  define FORCE_INLINE_ATTR
+#endif
+
+/**
+ * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
+ * parameters. They must be inlined for the compiler to elimininate the constant
+ * branches.
+ */
+#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
+/**
+ * HINT_INLINE is used to help the compiler generate better code. It is *not*
+ * used for "templates", so it can be tweaked based on the compilers
+ * performance.
+ *
+ * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the
+ * always_inline attribute.
+ *
+ * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline
+ * attribute.
+ */
+#if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
+#  define HINT_INLINE static INLINE_KEYWORD
+#else
+#  define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR
+#endif
+
+/* force no inlining */
+#ifdef _MSC_VER
+#  define FORCE_NOINLINE static __declspec(noinline)
+#else
+#  ifdef __GNUC__
+#    define FORCE_NOINLINE static __attribute__((__noinline__))
+#  else
+#    define FORCE_NOINLINE static
+#  endif
+#endif
+
+/* target attribute */
+#ifndef __has_attribute
+  #define __has_attribute(x) 0  /* Compatibility with non-clang compilers. */
+#endif
+#if defined(__GNUC__)
+#  define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
+#else
+#  define TARGET_ATTRIBUTE(target)
+#endif
+
+/* Enable runtime BMI2 dispatch based on the CPU.
+ * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
+ */
+#ifndef DYNAMIC_BMI2
+  #if (defined(__clang__) && __has_attribute(__target__)) \
+      || (defined(__GNUC__) \
+          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8))) \
+      && (defined(__x86_64__) || defined(_M_X86)) \
+      && !defined(__BMI2__)
+  #  define DYNAMIC_BMI2 1
+  #else
+  #  define DYNAMIC_BMI2 0
+  #endif
+#endif
+
+/* prefetch */
+#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
+#  include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#  define PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
+#elif defined(__GNUC__)
+#  define PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
+#else
+#  define PREFETCH(ptr)   /* disabled */
+#endif
+
+/* disable warnings */
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+#endif /* ZSTD_COMPILER_H */
diff --git a/vendor/github.com/DataDog/zstd/cover.c b/vendor/github.com/DataDog/zstd/cover.c
new file mode 100644
index 0000000..b5a3957
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/cover.c
@@ -0,0 +1,1048 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* *****************************************************************************
+ * Constructs a dictionary using a heuristic based on the following paper:
+ *
+ * Liao, Petri, Moffat, Wirth
+ * Effective Construction of Relative Lempel-Ziv Dictionaries
+ * Published in WWW 2016.
+ *
+ * Adapted from code originally written by @ot (Giuseppe Ottaviano).
+ ******************************************************************************/
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+
+#include "mem.h" /* read */
+#include "pool.h"
+#include "threading.h"
+#include "zstd_internal.h" /* includes zstd.h */
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+/*-*************************************
+*  Constants
+***************************************/
+#define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((U32)-1) : ((U32)1 GB))
+
+/*-*************************************
+*  Console display
+***************************************/
+static int g_displayLevel = 2;
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
+
+#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
+  if (displayLevel >= l) {                                                     \
+    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
+      g_time = clock();                                                        \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
+static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+
+/*-*************************************
+* Hash table
+***************************************
+* A small specialized hash map for storing activeDmers.
+* The map does not resize, so if it becomes full it will loop forever.
+* Thus, the map must be large enough to store every value.
+* The map implements linear probing and keeps its load less than 0.5.
+*/
+
+#define MAP_EMPTY_VALUE ((U32)-1)
+typedef struct COVER_map_pair_t_s {
+  U32 key;
+  U32 value;
+} COVER_map_pair_t;
+
+typedef struct COVER_map_s {
+  COVER_map_pair_t *data;
+  U32 sizeLog;
+  U32 size;
+  U32 sizeMask;
+} COVER_map_t;
+
+/**
+ * Clear the map.
+ */
+static void COVER_map_clear(COVER_map_t *map) {
+  memset(map->data, MAP_EMPTY_VALUE, map->size * sizeof(COVER_map_pair_t));
+}
+
+/**
+ * Initializes a map of the given size.
+ * Returns 1 on success and 0 on failure.
+ * The map must be destroyed with COVER_map_destroy().
+ * The map is only guaranteed to be large enough to hold size elements.
+ */
+static int COVER_map_init(COVER_map_t *map, U32 size) {
+  map->sizeLog = ZSTD_highbit32(size) + 2;
+  map->size = (U32)1 << map->sizeLog;
+  map->sizeMask = map->size - 1;
+  map->data = (COVER_map_pair_t *)malloc(map->size * sizeof(COVER_map_pair_t));
+  if (!map->data) {
+    map->sizeLog = 0;
+    map->size = 0;
+    return 0;
+  }
+  COVER_map_clear(map);
+  return 1;
+}
+
+/**
+ * Internal hash function
+ */
+static const U32 prime4bytes = 2654435761U;
+static U32 COVER_map_hash(COVER_map_t *map, U32 key) {
+  return (key * prime4bytes) >> (32 - map->sizeLog);
+}
+
+/**
+ * Helper function that returns the index that a key should be placed into.
+ */
+static U32 COVER_map_index(COVER_map_t *map, U32 key) {
+  const U32 hash = COVER_map_hash(map, key);
+  U32 i;
+  for (i = hash;; i = (i + 1) & map->sizeMask) {
+    COVER_map_pair_t *pos = &map->data[i];
+    if (pos->value == MAP_EMPTY_VALUE) {
+      return i;
+    }
+    if (pos->key == key) {
+      return i;
+    }
+  }
+}
+
+/**
+ * Returns the pointer to the value for key.
+ * If key is not in the map, it is inserted and the value is set to 0.
+ * The map must not be full.
+ */
+static U32 *COVER_map_at(COVER_map_t *map, U32 key) {
+  COVER_map_pair_t *pos = &map->data[COVER_map_index(map, key)];
+  if (pos->value == MAP_EMPTY_VALUE) {
+    pos->key = key;
+    pos->value = 0;
+  }
+  return &pos->value;
+}
+
+/**
+ * Deletes key from the map if present.
+ */
+static void COVER_map_remove(COVER_map_t *map, U32 key) {
+  U32 i = COVER_map_index(map, key);
+  COVER_map_pair_t *del = &map->data[i];
+  U32 shift = 1;
+  if (del->value == MAP_EMPTY_VALUE) {
+    return;
+  }
+  for (i = (i + 1) & map->sizeMask;; i = (i + 1) & map->sizeMask) {
+    COVER_map_pair_t *const pos = &map->data[i];
+    /* If the position is empty we are done */
+    if (pos->value == MAP_EMPTY_VALUE) {
+      del->value = MAP_EMPTY_VALUE;
+      return;
+    }
+    /* If pos can be moved to del do so */
+    if (((i - COVER_map_hash(map, pos->key)) & map->sizeMask) >= shift) {
+      del->key = pos->key;
+      del->value = pos->value;
+      del = pos;
+      shift = 1;
+    } else {
+      ++shift;
+    }
+  }
+}
+
+/**
+ * Destroyes a map that is inited with COVER_map_init().
+ */
+static void COVER_map_destroy(COVER_map_t *map) {
+  if (map->data) {
+    free(map->data);
+  }
+  map->data = NULL;
+  map->size = 0;
+}
+
+/*-*************************************
+* Context
+***************************************/
+
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  U32 *suffix;
+  size_t suffixSize;
+  U32 *freqs;
+  U32 *dmerAt;
+  unsigned d;
+} COVER_ctx_t;
+
+/* We need a global context for qsort... */
+static COVER_ctx_t *g_ctx = NULL;
+
+/*-*************************************
+*  Helper functions
+***************************************/
+
+/**
+ * Returns the sum of the sample sizes.
+ */
+static size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
+  size_t sum = 0;
+  size_t i;
+  for (i = 0; i < nbSamples; ++i) {
+    sum += samplesSizes[i];
+  }
+  return sum;
+}
+
+/**
+ * Returns -1 if the dmer at lp is less than the dmer at rp.
+ * Return 0 if the dmers at lp and rp are equal.
+ * Returns 1 if the dmer at lp is greater than the dmer at rp.
+ */
+static int COVER_cmp(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U32 const lhs = *(U32 const *)lp;
+  U32 const rhs = *(U32 const *)rp;
+  return memcmp(ctx->samples + lhs, ctx->samples + rhs, ctx->d);
+}
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U64 const mask = (ctx->d == 8) ? (U64)-1 : (((U64)1 << (8 * ctx->d)) - 1);
+  U64 const lhs = MEM_readLE64(ctx->samples + *(U32 const *)lp) & mask;
+  U64 const rhs = MEM_readLE64(ctx->samples + *(U32 const *)rp) & mask;
+  if (lhs < rhs) {
+    return -1;
+  }
+  return (lhs > rhs);
+}
+
+/**
+ * Same as COVER_cmp() except ties are broken by pointer value
+ * NOTE: g_ctx must be set to call this function.  A global is required because
+ * qsort doesn't take an opaque pointer.
+ */
+static int COVER_strict_cmp(const void *lp, const void *rp) {
+  int result = COVER_cmp(g_ctx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_strict_cmp8(const void *lp, const void *rp) {
+  int result = COVER_cmp8(g_ctx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
+
+/**
+ * Returns the first pointer in [first, last) whose element does not compare
+ * less than value.  If no such element exists it returns last.
+ */
+static const size_t *COVER_lower_bound(const size_t *first, const size_t *last,
+                                       size_t value) {
+  size_t count = last - first;
+  while (count != 0) {
+    size_t step = count / 2;
+    const size_t *ptr = first;
+    ptr += step;
+    if (*ptr < value) {
+      first = ++ptr;
+      count -= step + 1;
+    } else {
+      count = step;
+    }
+  }
+  return first;
+}
+
+/**
+ * Generic groupBy function.
+ * Groups an array sorted by cmp into groups with equivalent values.
+ * Calls grp for each group.
+ */
+static void
+COVER_groupBy(const void *data, size_t count, size_t size, COVER_ctx_t *ctx,
+              int (*cmp)(COVER_ctx_t *, const void *, const void *),
+              void (*grp)(COVER_ctx_t *, const void *, const void *)) {
+  const BYTE *ptr = (const BYTE *)data;
+  size_t num = 0;
+  while (num < count) {
+    const BYTE *grpEnd = ptr + size;
+    ++num;
+    while (num < count && cmp(ctx, ptr, grpEnd) == 0) {
+      grpEnd += size;
+      ++num;
+    }
+    grp(ctx, ptr, grpEnd);
+    ptr = grpEnd;
+  }
+}
+
+/*-*************************************
+*  Cover functions
+***************************************/
+
+/**
+ * Called on each group of positions with the same dmer.
+ * Counts the frequency of each dmer and saves it in the suffix array.
+ * Fills `ctx->dmerAt`.
+ */
+static void COVER_group(COVER_ctx_t *ctx, const void *group,
+                        const void *groupEnd) {
+  /* The group consists of all the positions with the same first d bytes. */
+  const U32 *grpPtr = (const U32 *)group;
+  const U32 *grpEnd = (const U32 *)groupEnd;
+  /* The dmerId is how we will reference this dmer.
+   * This allows us to map the whole dmer space to a much smaller space, the
+   * size of the suffix array.
+   */
+  const U32 dmerId = (U32)(grpPtr - ctx->suffix);
+  /* Count the number of samples this dmer shows up in */
+  U32 freq = 0;
+  /* Details */
+  const size_t *curOffsetPtr = ctx->offsets;
+  const size_t *offsetsEnd = ctx->offsets + ctx->nbSamples;
+  /* Once *grpPtr >= curSampleEnd this occurrence of the dmer is in a
+   * different sample than the last.
+   */
+  size_t curSampleEnd = ctx->offsets[0];
+  for (; grpPtr != grpEnd; ++grpPtr) {
+    /* Save the dmerId for this position so we can get back to it. */
+    ctx->dmerAt[*grpPtr] = dmerId;
+    /* Dictionaries only help for the first reference to the dmer.
+     * After that zstd can reference the match from the previous reference.
+     * So only count each dmer once for each sample it is in.
+     */
+    if (*grpPtr < curSampleEnd) {
+      continue;
+    }
+    freq += 1;
+    /* Binary search to find the end of the sample *grpPtr is in.
+     * In the common case that grpPtr + 1 == grpEnd we can skip the binary
+     * search because the loop is over.
+     */
+    if (grpPtr + 1 != grpEnd) {
+      const size_t *sampleEndPtr =
+          COVER_lower_bound(curOffsetPtr, offsetsEnd, *grpPtr);
+      curSampleEnd = *sampleEndPtr;
+      curOffsetPtr = sampleEndPtr + 1;
+    }
+  }
+  /* At this point we are never going to look at this segment of the suffix
+   * array again.  We take advantage of this fact to save memory.
+   * We store the frequency of the dmer in the first position of the group,
+   * which is dmerId.
+   */
+  ctx->suffix[dmerId] = freq;
+}
+
+/**
+ * A segment is a range in the source as well as the score of the segment.
+ */
+typedef struct {
+  U32 begin;
+  U32 end;
+  U32 score;
+} COVER_segment_t;
+
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of dmer d.
+ * Let S_i be the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer d is in the dictionay we set F(d) = 0.
+ */
+static COVER_segment_t COVER_selectSegment(const COVER_ctx_t *ctx, U32 *freqs,
+                                           COVER_map_t *activeDmers, U32 begin,
+                                           U32 end,
+                                           ZDICT_cover_params_t parameters) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 dmersInK = k - d + 1;
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+  /* Reset the activeDmers in the segment */
+  COVER_map_clear(activeDmers);
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* The dmerId for the dmer at the next position */
+    U32 newDmer = ctx->dmerAt[activeSegment.end];
+    /* The entry in activeDmers for this dmerId */
+    U32 *newDmerOcc = COVER_map_at(activeDmers, newDmer);
+    /* If the dmer isn't already present in the segment add its score. */
+    if (*newDmerOcc == 0) {
+      /* The paper suggest using the L-0.5 norm, but experiments show that it
+       * doesn't help.
+       */
+      activeSegment.score += freqs[newDmer];
+    }
+    /* Add the dmer to the segment */
+    activeSegment.end += 1;
+    *newDmerOcc += 1;
+
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      U32 delDmer = ctx->dmerAt[activeSegment.begin];
+      U32 *delDmerOcc = COVER_map_at(activeDmers, delDmer);
+      activeSegment.begin += 1;
+      *delDmerOcc -= 1;
+      /* If this is the last occurence of the dmer, subtract its score */
+      if (*delDmerOcc == 0) {
+        COVER_map_remove(activeDmers, delDmer);
+        activeSegment.score -= freqs[delDmer];
+      }
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+  {
+    /* Trim off the zero frequency head and tail from the segment. */
+    U32 newBegin = bestSegment.end;
+    U32 newEnd = bestSegment.begin;
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      U32 freq = freqs[ctx->dmerAt[pos]];
+      if (freq != 0) {
+        newBegin = MIN(newBegin, pos);
+        newEnd = pos + 1;
+      }
+    }
+    bestSegment.begin = newBegin;
+    bestSegment.end = newEnd;
+  }
+  {
+    /* Zero out the frequency of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      freqs[ctx->dmerAt[pos]] = 0;
+    }
+  }
+  return bestSegment;
+}
+
+/**
+ * Check the validity of the parameters.
+ * Returns non-zero if the parameters are valid and 0 otherwise.
+ */
+static int COVER_checkParameters(ZDICT_cover_params_t parameters,
+                                 size_t maxDictSize) {
+  /* k and d are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* k <= maxDictSize */
+  if (parameters.k > maxDictSize) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  return 1;
+}
+
+/**
+ * Clean up a context initialized with `COVER_ctx_init()`.
+ */
+static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
+  if (!ctx) {
+    return;
+  }
+  if (ctx->suffix) {
+    free(ctx->suffix);
+    ctx->suffix = NULL;
+  }
+  if (ctx->freqs) {
+    free(ctx->freqs);
+    ctx->freqs = NULL;
+  }
+  if (ctx->dmerAt) {
+    free(ctx->dmerAt);
+    ctx->dmerAt = NULL;
+  }
+  if (ctx->offsets) {
+    free(ctx->offsets);
+    ctx->offsets = NULL;
+  }
+}
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can used multiple
+ * times.
+ * Returns 1 on success or zero on error.
+ * The context must be destroyed with `COVER_ctx_destroy()`.
+ */
+static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
+                          const size_t *samplesSizes, unsigned nbSamples,
+                          unsigned d) {
+  const BYTE *const samples = (const BYTE *)samplesBuffer;
+  const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+  /* Checks */
+  if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+      totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
+    DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
+                 (U32)(totalSamplesSize>>20), (COVER_MAX_SAMPLES_SIZE >> 20));
+    return 0;
+  }
+  /* Zero the context */
+  memset(ctx, 0, sizeof(*ctx));
+  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbSamples,
+               (U32)totalSamplesSize);
+  ctx->samples = samples;
+  ctx->samplesSizes = samplesSizes;
+  ctx->nbSamples = nbSamples;
+  /* Partial suffix array */
+  ctx->suffixSize = totalSamplesSize - MAX(d, sizeof(U64)) + 1;
+  ctx->suffix = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  /* Maps index to the dmerID */
+  ctx->dmerAt = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  /* The offsets of each file */
+  ctx->offsets = (size_t *)malloc((nbSamples + 1) * sizeof(size_t));
+  if (!ctx->suffix || !ctx->dmerAt || !ctx->offsets) {
+    DISPLAYLEVEL(1, "Failed to allocate scratch buffers\n");
+    COVER_ctx_destroy(ctx);
+    return 0;
+  }
+  ctx->freqs = NULL;
+  ctx->d = d;
+
+  /* Fill offsets from the samlesSizes */
+  {
+    U32 i;
+    ctx->offsets[0] = 0;
+    for (i = 1; i <= nbSamples; ++i) {
+      ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+    }
+  }
+  DISPLAYLEVEL(2, "Constructing partial suffix array\n");
+  {
+    /* suffix is a partial suffix array.
+     * It only sorts suffixes by their first parameters.d bytes.
+     * The sort is stable, so each dmer group is sorted by position in input.
+     */
+    U32 i;
+    for (i = 0; i < ctx->suffixSize; ++i) {
+      ctx->suffix[i] = i;
+    }
+    /* qsort doesn't take an opaque pointer, so pass as a global */
+    g_ctx = ctx;
+    qsort(ctx->suffix, ctx->suffixSize, sizeof(U32),
+          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
+  }
+  DISPLAYLEVEL(2, "Computing frequencies\n");
+  /* For each dmer group (group of positions with the same first d bytes):
+   * 1. For each position we set dmerAt[position] = dmerID.  The dmerID is
+   *    (groupBeginPtr - suffix).  This allows us to go from position to
+   *    dmerID so we can look up values in freq.
+   * 2. We calculate how many samples the dmer occurs in and save it in
+   *    freqs[dmerId].
+   */
+  COVER_groupBy(ctx->suffix, ctx->suffixSize, sizeof(U32), ctx,
+                (ctx->d <= 8 ? &COVER_cmp8 : &COVER_cmp), &COVER_group);
+  ctx->freqs = ctx->suffix;
+  ctx->suffix = NULL;
+  return 1;
+}
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
+                                    COVER_map_t *activeDmers, void *dictBuffer,
+                                    size_t dictBufferCapacity,
+                                    ZDICT_cover_params_t parameters) {
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data up into epochs of equal size.
+   * We will select at least one segment from each epoch.
+   */
+  const U32 epochs = (U32)(dictBufferCapacity / parameters.k);
+  const U32 epochSize = (U32)(ctx->suffixSize / epochs);
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n", epochs,
+               epochSize);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs) {
+    const U32 epochBegin = (U32)(epoch * epochSize);
+    const U32 epochEnd = epochBegin + epochSize;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = COVER_selectSegment(
+        ctx, freqs, activeDmers, epochBegin, epochEnd, parameters);
+    /* If the segment covers no dmers, then we are out of content */
+    if (segment.score == 0) {
+      break;
+    }
+    /* Trim the segment if necessary and if it is too small then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize < parameters.d) {
+      break;
+    }
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        2, "\r%u%%       ",
+        (U32)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
+    void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_cover_params_t parameters)
+{
+  BYTE* const dict = (BYTE*)dictBuffer;
+  COVER_ctx_t ctx;
+  COVER_map_t activeDmers;
+
+  /* Initialize global data */
+  g_displayLevel = parameters.zParams.notificationLevel;
+  /* Checks */
+  if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
+    DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+    return ERROR(GENERIC);
+  }
+  if (nbSamples == 0) {
+    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
+    return ERROR(GENERIC);
+  }
+  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                 ZDICT_DICTSIZE_MIN);
+    return ERROR(dstSize_tooSmall);
+  }
+  /* Initialize context and activeDmers */
+  if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                      parameters.d)) {
+    return ERROR(GENERIC);
+  }
+  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
+    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
+    COVER_ctx_destroy(&ctx);
+    return ERROR(GENERIC);
+  }
+
+  DISPLAYLEVEL(2, "Building dictionary\n");
+  {
+    const size_t tail =
+        COVER_buildDictionary(&ctx, ctx.freqs, &activeDmers, dictBuffer,
+                              dictBufferCapacity, parameters);
+    const size_t dictionarySize = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        samplesBuffer, samplesSizes, nbSamples, parameters.zParams);
+    if (!ZSTD_isError(dictionarySize)) {
+      DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                   (U32)dictionarySize);
+    }
+    COVER_ctx_destroy(&ctx);
+    COVER_map_destroy(&activeDmers);
+    return dictionarySize;
+  }
+}
+
+/**
+ * COVER_best_t is used for two purposes:
+ * 1. Synchronizing threads.
+ * 2. Saving the best parameters and dictionary.
+ *
+ * All of the methods except COVER_best_init() are thread safe if zstd is
+ * compiled with multithreaded support.
+ */
+typedef struct COVER_best_s {
+  ZSTD_pthread_mutex_t mutex;
+  ZSTD_pthread_cond_t cond;
+  size_t liveJobs;
+  void *dict;
+  size_t dictSize;
+  ZDICT_cover_params_t parameters;
+  size_t compressedSize;
+} COVER_best_t;
+
+/**
+ * Initialize the `COVER_best_t`.
+ */
+static void COVER_best_init(COVER_best_t *best) {
+  if (best==NULL) return; /* compatible with init on NULL */
+  (void)ZSTD_pthread_mutex_init(&best->mutex, NULL);
+  (void)ZSTD_pthread_cond_init(&best->cond, NULL);
+  best->liveJobs = 0;
+  best->dict = NULL;
+  best->dictSize = 0;
+  best->compressedSize = (size_t)-1;
+  memset(&best->parameters, 0, sizeof(best->parameters));
+}
+
+/**
+ * Wait until liveJobs == 0.
+ */
+static void COVER_best_wait(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  ZSTD_pthread_mutex_lock(&best->mutex);
+  while (best->liveJobs != 0) {
+    ZSTD_pthread_cond_wait(&best->cond, &best->mutex);
+  }
+  ZSTD_pthread_mutex_unlock(&best->mutex);
+}
+
+/**
+ * Call COVER_best_wait() and then destroy the COVER_best_t.
+ */
+static void COVER_best_destroy(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  COVER_best_wait(best);
+  if (best->dict) {
+    free(best->dict);
+  }
+  ZSTD_pthread_mutex_destroy(&best->mutex);
+  ZSTD_pthread_cond_destroy(&best->cond);
+}
+
+/**
+ * Called when a thread is about to be launched.
+ * Increments liveJobs.
+ */
+static void COVER_best_start(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  ZSTD_pthread_mutex_lock(&best->mutex);
+  ++best->liveJobs;
+  ZSTD_pthread_mutex_unlock(&best->mutex);
+}
+
+/**
+ * Called when a thread finishes executing, both on error or success.
+ * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
+ * If this dictionary is the best so far save it and its parameters.
+ */
+static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
+                              ZDICT_cover_params_t parameters, void *dict,
+                              size_t dictSize) {
+  if (!best) {
+    return;
+  }
+  {
+    size_t liveJobs;
+    ZSTD_pthread_mutex_lock(&best->mutex);
+    --best->liveJobs;
+    liveJobs = best->liveJobs;
+    /* If the new dictionary is better */
+    if (compressedSize < best->compressedSize) {
+      /* Allocate space if necessary */
+      if (!best->dict || best->dictSize < dictSize) {
+        if (best->dict) {
+          free(best->dict);
+        }
+        best->dict = malloc(dictSize);
+        if (!best->dict) {
+          best->compressedSize = ERROR(GENERIC);
+          best->dictSize = 0;
+          return;
+        }
+      }
+      /* Save the dictionary, parameters, and size */
+      memcpy(best->dict, dict, dictSize);
+      best->dictSize = dictSize;
+      best->parameters = parameters;
+      best->compressedSize = compressedSize;
+    }
+    ZSTD_pthread_mutex_unlock(&best->mutex);
+    if (liveJobs == 0) {
+      ZSTD_pthread_cond_broadcast(&best->cond);
+    }
+  }
+}
+
+/**
+ * Parameters for COVER_tryParameters().
+ */
+typedef struct COVER_tryParameters_data_s {
+  const COVER_ctx_t *ctx;
+  COVER_best_t *best;
+  size_t dictBufferCapacity;
+  ZDICT_cover_params_t parameters;
+} COVER_tryParameters_data_t;
+
+/**
+ * Tries a set of parameters and upates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void COVER_tryParameters(void *opaque) {
+  /* Save parameters as local variables */
+  COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t *)opaque;
+  const COVER_ctx_t *const ctx = data->ctx;
+  const ZDICT_cover_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Allocate space for hash table, dict, and freqs */
+  COVER_map_t activeDmers;
+  BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
+  U32 *freqs = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
+    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
+    goto _cleanup;
+  }
+  if (!dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ctx->suffixSize * sizeof(U32));
+  /* Build the dictionary */
+  {
+    const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
+                                              dictBufferCapacity, parameters);
+    dictBufferCapacity = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbSamples,
+        parameters.zParams);
+    if (ZDICT_isError(dictBufferCapacity)) {
+      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
+      goto _cleanup;
+    }
+  }
+  /* Check total compressed size */
+  {
+    /* Pointers */
+    ZSTD_CCtx *cctx;
+    ZSTD_CDict *cdict;
+    void *dst;
+    /* Local variables */
+    size_t dstCapacity;
+    size_t i;
+    /* Allocate dst with enough space to compress the maximum sized sample */
+    {
+      size_t maxSampleSize = 0;
+      for (i = 0; i < ctx->nbSamples; ++i) {
+        maxSampleSize = MAX(ctx->samplesSizes[i], maxSampleSize);
+      }
+      dstCapacity = ZSTD_compressBound(maxSampleSize);
+      dst = malloc(dstCapacity);
+    }
+    /* Create the cctx and cdict */
+    cctx = ZSTD_createCCtx();
+    cdict = ZSTD_createCDict(dict, dictBufferCapacity,
+                             parameters.zParams.compressionLevel);
+    if (!dst || !cctx || !cdict) {
+      goto _compressCleanup;
+    }
+    /* Compress each sample and sum their sizes (or error) */
+    totalCompressedSize = dictBufferCapacity;
+    for (i = 0; i < ctx->nbSamples; ++i) {
+      const size_t size = ZSTD_compress_usingCDict(
+          cctx, dst, dstCapacity, ctx->samples + ctx->offsets[i],
+          ctx->samplesSizes[i], cdict);
+      if (ZSTD_isError(size)) {
+        totalCompressedSize = ERROR(GENERIC);
+        goto _compressCleanup;
+      }
+      totalCompressedSize += size;
+    }
+  _compressCleanup:
+    ZSTD_freeCCtx(cctx);
+    ZSTD_freeCDict(cdict);
+    if (dst) {
+      free(dst);
+    }
+  }
+
+_cleanup:
+  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
+                    dictBufferCapacity);
+  free(data);
+  COVER_map_destroy(&activeDmers);
+  if (dict) {
+    free(dict);
+  }
+  if (freqs) {
+    free(freqs);
+  }
+}
+
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
+    const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_cover_params_t *parameters) {
+  /* constants */
+  const unsigned nbThreads = parameters->nbThreads;
+  const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+  const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+  const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+  const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+  const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+  const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+  const unsigned kIterations =
+      (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+  /* Local variables */
+  const int displayLevel = parameters->zParams.notificationLevel;
+  unsigned iteration = 1;
+  unsigned d;
+  unsigned k;
+  COVER_best_t best;
+  POOL_ctx *pool = NULL;
+
+  /* Checks */
+  if (kMinK < kMaxD || kMaxK < kMinK) {
+    LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
+    return ERROR(GENERIC);
+  }
+  if (nbSamples == 0) {
+    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
+    return ERROR(GENERIC);
+  }
+  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                 ZDICT_DICTSIZE_MIN);
+    return ERROR(dstSize_tooSmall);
+  }
+  if (nbThreads > 1) {
+    pool = POOL_create(nbThreads, 1);
+    if (!pool) {
+      return ERROR(memory_allocation);
+    }
+  }
+  /* Initialization */
+  COVER_best_init(&best);
+  /* Turn down global display level to clean up display at level 2 and below */
+  g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
+  /* Loop through d first because each new value needs a new context */
+  LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
+                    kIterations);
+  for (d = kMinD; d <= kMaxD; d += 2) {
+    /* Initialize the context for this value of d */
+    COVER_ctx_t ctx;
+    LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
+    if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d)) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return ERROR(GENERIC);
+    }
+    /* Loop through k reusing the same context */
+    for (k = kMinK; k <= kMaxK; k += kStepSize) {
+      /* Prepare the arguments */
+      COVER_tryParameters_data_t *data = (COVER_tryParameters_data_t *)malloc(
+          sizeof(COVER_tryParameters_data_t));
+      LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
+      if (!data) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
+        COVER_best_destroy(&best);
+        COVER_ctx_destroy(&ctx);
+        POOL_free(pool);
+        return ERROR(GENERIC);
+      }
+      data->ctx = &ctx;
+      data->best = &best;
+      data->dictBufferCapacity = dictBufferCapacity;
+      data->parameters = *parameters;
+      data->parameters.k = k;
+      data->parameters.d = d;
+      data->parameters.steps = kSteps;
+      data->parameters.zParams.notificationLevel = g_displayLevel;
+      /* Check the parameters */
+      if (!COVER_checkParameters(data->parameters, dictBufferCapacity)) {
+        DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+        free(data);
+        continue;
+      }
+      /* Call the function and pass ownership of data to it */
+      COVER_best_start(&best);
+      if (pool) {
+        POOL_add(pool, &COVER_tryParameters, data);
+      } else {
+        COVER_tryParameters(data);
+      }
+      /* Print status */
+      LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
+                         (U32)((iteration * 100) / kIterations));
+      ++iteration;
+    }
+    COVER_best_wait(&best);
+    COVER_ctx_destroy(&ctx);
+  }
+  LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
+  /* Fill the output buffer and parameters with output of the best parameters */
+  {
+    const size_t dictSize = best.dictSize;
+    if (ZSTD_isError(best.compressedSize)) {
+      const size_t compressedSize = best.compressedSize;
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return compressedSize;
+    }
+    *parameters = best.parameters;
+    memcpy(dictBuffer, best.dict, dictSize);
+    COVER_best_destroy(&best);
+    POOL_free(pool);
+    return dictSize;
+  }
+}
diff --git a/vendor/github.com/DataDog/zstd/cpu.h b/vendor/github.com/DataDog/zstd/cpu.h
new file mode 100755
index 0000000..4eb48e3
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/cpu.h
@@ -0,0 +1,216 @@
+/*
+ * Copyright (c) 2018-present, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMMON_CPU_H
+#define ZSTD_COMMON_CPU_H
+
+/**
+ * Implementation taken from folly/CpuId.h
+ * https://github.com/facebook/folly/blob/master/folly/CpuId.h
+ */
+
+#include <string.h>
+
+#include "mem.h"
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
+typedef struct {
+    U32 f1c;
+    U32 f1d;
+    U32 f7b;
+    U32 f7c;
+} ZSTD_cpuid_t;
+
+MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
+    U32 f1c = 0;
+    U32 f1d = 0;
+    U32 f7b = 0;
+    U32 f7c = 0;
+#ifdef _MSC_VER
+    int reg[4];
+    __cpuid((int*)reg, 0);
+    {
+        int const n = reg[0];
+        if (n >= 1) {
+            __cpuid((int*)reg, 1);
+            f1c = (U32)reg[2];
+            f1d = (U32)reg[3];
+        }
+        if (n >= 7) {
+            __cpuidex((int*)reg, 7, 0);
+            f7b = (U32)reg[1];
+            f7c = (U32)reg[2];
+        }
+    }
+#elif defined(__i386__) && defined(__PIC__) && !defined(__clang__) && defined(__GNUC__)
+    /* The following block like the normal cpuid branch below, but gcc
+     * reserves ebx for use of its pic register so we must specially
+     * handle the save and restore to avoid clobbering the register
+     */
+    U32 n;
+    __asm__(
+        "pushl %%ebx\n\t"
+        "cpuid\n\t"
+        "popl %%ebx\n\t"
+        : "=a"(n)
+        : "a"(0)
+        : "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "popl %%ebx\n\t"
+          : "=a"(f1a), "=c"(f1c), "=d"(f1d)
+          : "a"(1)
+          :);
+    }
+    if (n >= 7) {
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "movl %%ebx, %%eax\n\r"
+          "popl %%ebx"
+          : "=a"(f7b), "=c"(f7c)
+          : "a"(7), "c"(0)
+          : "edx");
+    }
+#elif defined(__x86_64__) || defined(_M_X64) || defined(__i386__)
+    U32 n;
+    __asm__("cpuid" : "=a"(n) : "a"(0) : "ebx", "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__("cpuid" : "=a"(f1a), "=c"(f1c), "=d"(f1d) : "a"(1) : "ebx");
+    }
+    if (n >= 7) {
+      U32 f7a;
+      __asm__("cpuid"
+              : "=a"(f7a), "=b"(f7b), "=c"(f7c)
+              : "a"(7), "c"(0)
+              : "edx");
+    }
+#endif
+    {
+        ZSTD_cpuid_t cpuid;
+        cpuid.f1c = f1c;
+        cpuid.f1d = f1d;
+        cpuid.f7b = f7b;
+        cpuid.f7c = f7c;
+        return cpuid;
+    }
+}
+
+#define X(name, r, bit)                                                        \
+  MEM_STATIC int ZSTD_cpuid_##name(ZSTD_cpuid_t const cpuid) {                 \
+    return ((cpuid.r) & (1U << bit)) != 0;                                     \
+  }
+
+/* cpuid(1): Processor Info and Feature Bits. */
+#define C(name, bit) X(name, f1c, bit)
+  C(sse3, 0)
+  C(pclmuldq, 1)
+  C(dtes64, 2)
+  C(monitor, 3)
+  C(dscpl, 4)
+  C(vmx, 5)
+  C(smx, 6)
+  C(eist, 7)
+  C(tm2, 8)
+  C(ssse3, 9)
+  C(cnxtid, 10)
+  C(fma, 12)
+  C(cx16, 13)
+  C(xtpr, 14)
+  C(pdcm, 15)
+  C(pcid, 17)
+  C(dca, 18)
+  C(sse41, 19)
+  C(sse42, 20)
+  C(x2apic, 21)
+  C(movbe, 22)
+  C(popcnt, 23)
+  C(tscdeadline, 24)
+  C(aes, 25)
+  C(xsave, 26)
+  C(osxsave, 27)
+  C(avx, 28)
+  C(f16c, 29)
+  C(rdrand, 30)
+#undef C
+#define D(name, bit) X(name, f1d, bit)
+  D(fpu, 0)
+  D(vme, 1)
+  D(de, 2)
+  D(pse, 3)
+  D(tsc, 4)
+  D(msr, 5)
+  D(pae, 6)
+  D(mce, 7)
+  D(cx8, 8)
+  D(apic, 9)
+  D(sep, 11)
+  D(mtrr, 12)
+  D(pge, 13)
+  D(mca, 14)
+  D(cmov, 15)
+  D(pat, 16)
+  D(pse36, 17)
+  D(psn, 18)
+  D(clfsh, 19)
+  D(ds, 21)
+  D(acpi, 22)
+  D(mmx, 23)
+  D(fxsr, 24)
+  D(sse, 25)
+  D(sse2, 26)
+  D(ss, 27)
+  D(htt, 28)
+  D(tm, 29)
+  D(pbe, 31)
+#undef D
+
+/* cpuid(7): Extended Features. */
+#define B(name, bit) X(name, f7b, bit)
+  B(bmi1, 3)
+  B(hle, 4)
+  B(avx2, 5)
+  B(smep, 7)
+  B(bmi2, 8)
+  B(erms, 9)
+  B(invpcid, 10)
+  B(rtm, 11)
+  B(mpx, 14)
+  B(avx512f, 16)
+  B(avx512dq, 17)
+  B(rdseed, 18)
+  B(adx, 19)
+  B(smap, 20)
+  B(avx512ifma, 21)
+  B(pcommit, 22)
+  B(clflushopt, 23)
+  B(clwb, 24)
+  B(avx512pf, 26)
+  B(avx512er, 27)
+  B(avx512cd, 28)
+  B(sha, 29)
+  B(avx512bw, 30)
+  B(avx512vl, 31)
+#undef B
+#define C(name, bit) X(name, f7c, bit)
+  C(prefetchwt1, 0)
+  C(avx512vbmi, 1)
+#undef C
+
+#undef X
+
+#endif /* ZSTD_COMMON_CPU_H */
diff --git a/vendor/github.com/DataDog/zstd/divsufsort.c b/vendor/github.com/DataDog/zstd/divsufsort.c
new file mode 100644
index 0000000..60cceb0
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/divsufsort.c
@@ -0,0 +1,1913 @@
+/*
+ * divsufsort.c for libdivsufsort-lite
+ * Copyright (c) 2003-2008 Yuta Mori All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/*- Compiler specifics -*/
+#ifdef __clang__
+#pragma clang diagnostic ignored "-Wshorten-64-to-32"
+#endif
+
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4244)
+#  pragma warning(disable : 4127)    /* C4127 : Condition expression is constant */
+#endif
+
+
+/*- Dependencies -*/
+#include <assert.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include "divsufsort.h"
+
+/*- Constants -*/
+#if defined(INLINE)
+# undef INLINE
+#endif
+#if !defined(INLINE)
+# define INLINE __inline
+#endif
+#if defined(ALPHABET_SIZE) && (ALPHABET_SIZE < 1)
+# undef ALPHABET_SIZE
+#endif
+#if !defined(ALPHABET_SIZE)
+# define ALPHABET_SIZE (256)
+#endif
+#define BUCKET_A_SIZE (ALPHABET_SIZE)
+#define BUCKET_B_SIZE (ALPHABET_SIZE * ALPHABET_SIZE)
+#if defined(SS_INSERTIONSORT_THRESHOLD)
+# if SS_INSERTIONSORT_THRESHOLD < 1
+#  undef SS_INSERTIONSORT_THRESHOLD
+#  define SS_INSERTIONSORT_THRESHOLD (1)
+# endif
+#else
+# define SS_INSERTIONSORT_THRESHOLD (8)
+#endif
+#if defined(SS_BLOCKSIZE)
+# if SS_BLOCKSIZE < 0
+#  undef SS_BLOCKSIZE
+#  define SS_BLOCKSIZE (0)
+# elif 32768 <= SS_BLOCKSIZE
+#  undef SS_BLOCKSIZE
+#  define SS_BLOCKSIZE (32767)
+# endif
+#else
+# define SS_BLOCKSIZE (1024)
+#endif
+/* minstacksize = log(SS_BLOCKSIZE) / log(3) * 2 */
+#if SS_BLOCKSIZE == 0
+# define SS_MISORT_STACKSIZE (96)
+#elif SS_BLOCKSIZE <= 4096
+# define SS_MISORT_STACKSIZE (16)
+#else
+# define SS_MISORT_STACKSIZE (24)
+#endif
+#define SS_SMERGE_STACKSIZE (32)
+#define TR_INSERTIONSORT_THRESHOLD (8)
+#define TR_STACKSIZE (64)
+
+
+/*- Macros -*/
+#ifndef SWAP
+# define SWAP(_a, _b) do { t = (_a); (_a) = (_b); (_b) = t; } while(0)
+#endif /* SWAP */
+#ifndef MIN
+# define MIN(_a, _b) (((_a) < (_b)) ? (_a) : (_b))
+#endif /* MIN */
+#ifndef MAX
+# define MAX(_a, _b) (((_a) > (_b)) ? (_a) : (_b))
+#endif /* MAX */
+#define STACK_PUSH(_a, _b, _c, _d)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize++].d = (_d);\
+  } while(0)
+#define STACK_PUSH5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(ssize < STACK_SIZE);\
+    stack[ssize].a = (_a), stack[ssize].b = (_b),\
+    stack[ssize].c = (_c), stack[ssize].d = (_d), stack[ssize++].e = (_e);\
+  } while(0)
+#define STACK_POP(_a, _b, _c, _d)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d;\
+  } while(0)
+#define STACK_POP5(_a, _b, _c, _d, _e)\
+  do {\
+    assert(0 <= ssize);\
+    if(ssize == 0) { return; }\
+    (_a) = stack[--ssize].a, (_b) = stack[ssize].b,\
+    (_c) = stack[ssize].c, (_d) = stack[ssize].d, (_e) = stack[ssize].e;\
+  } while(0)
+#define BUCKET_A(_c0) bucket_A[(_c0)]
+#if ALPHABET_SIZE == 256
+#define BUCKET_B(_c0, _c1) (bucket_B[((_c1) << 8) | (_c0)])
+#define BUCKET_BSTAR(_c0, _c1) (bucket_B[((_c0) << 8) | (_c1)])
+#else
+#define BUCKET_B(_c0, _c1) (bucket_B[(_c1) * ALPHABET_SIZE + (_c0)])
+#define BUCKET_BSTAR(_c0, _c1) (bucket_B[(_c0) * ALPHABET_SIZE + (_c1)])
+#endif
+
+
+/*- Private Functions -*/
+
+static const int lg_table[256]= {
+ -1,0,1,1,2,2,2,2,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,
+  5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7
+};
+
+#if (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE)
+
+static INLINE
+int
+ss_ilg(int n) {
+#if SS_BLOCKSIZE == 0
+  return (n & 0xffff0000) ?
+          ((n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+#elif SS_BLOCKSIZE < 256
+  return lg_table[n];
+#else
+  return (n & 0xff00) ?
+          8 + lg_table[(n >> 8) & 0xff] :
+          0 + lg_table[(n >> 0) & 0xff];
+#endif
+}
+
+#endif /* (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE) */
+
+#if SS_BLOCKSIZE != 0
+
+static const int sqq_table[256] = {
+  0,  16,  22,  27,  32,  35,  39,  42,  45,  48,  50,  53,  55,  57,  59,  61,
+ 64,  65,  67,  69,  71,  73,  75,  76,  78,  80,  81,  83,  84,  86,  87,  89,
+ 90,  91,  93,  94,  96,  97,  98,  99, 101, 102, 103, 104, 106, 107, 108, 109,
+110, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125, 126,
+128, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139, 140, 141, 142,
+143, 144, 144, 145, 146, 147, 148, 149, 150, 150, 151, 152, 153, 154, 155, 155,
+156, 157, 158, 159, 160, 160, 161, 162, 163, 163, 164, 165, 166, 167, 167, 168,
+169, 170, 170, 171, 172, 173, 173, 174, 175, 176, 176, 177, 178, 178, 179, 180,
+181, 181, 182, 183, 183, 184, 185, 185, 186, 187, 187, 188, 189, 189, 190, 191,
+192, 192, 193, 193, 194, 195, 195, 196, 197, 197, 198, 199, 199, 200, 201, 201,
+202, 203, 203, 204, 204, 205, 206, 206, 207, 208, 208, 209, 209, 210, 211, 211,
+212, 212, 213, 214, 214, 215, 215, 216, 217, 217, 218, 218, 219, 219, 220, 221,
+221, 222, 222, 223, 224, 224, 225, 225, 226, 226, 227, 227, 228, 229, 229, 230,
+230, 231, 231, 232, 232, 233, 234, 234, 235, 235, 236, 236, 237, 237, 238, 238,
+239, 240, 240, 241, 241, 242, 242, 243, 243, 244, 244, 245, 245, 246, 246, 247,
+247, 248, 248, 249, 249, 250, 250, 251, 251, 252, 252, 253, 253, 254, 254, 255
+};
+
+static INLINE
+int
+ss_isqrt(int x) {
+  int y, e;
+
+  if(x >= (SS_BLOCKSIZE * SS_BLOCKSIZE)) { return SS_BLOCKSIZE; }
+  e = (x & 0xffff0000) ?
+        ((x & 0xff000000) ?
+          24 + lg_table[(x >> 24) & 0xff] :
+          16 + lg_table[(x >> 16) & 0xff]) :
+        ((x & 0x0000ff00) ?
+           8 + lg_table[(x >>  8) & 0xff] :
+           0 + lg_table[(x >>  0) & 0xff]);
+
+  if(e >= 16) {
+    y = sqq_table[x >> ((e - 6) - (e & 1))] << ((e >> 1) - 7);
+    if(e >= 24) { y = (y + 1 + x / y) >> 1; }
+    y = (y + 1 + x / y) >> 1;
+  } else if(e >= 8) {
+    y = (sqq_table[x >> ((e - 6) - (e & 1))] >> (7 - (e >> 1))) + 1;
+  } else {
+    return sqq_table[x] >> 4;
+  }
+
+  return (x < (y * y)) ? y - 1 : y;
+}
+
+#endif /* SS_BLOCKSIZE != 0 */
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Compares two suffixes. */
+static INLINE
+int
+ss_compare(const unsigned char *T,
+           const int *p1, const int *p2,
+           int depth) {
+  const unsigned char *U1, *U2, *U1n, *U2n;
+
+  for(U1 = T + depth + *p1,
+      U2 = T + depth + *p2,
+      U1n = T + *(p1 + 1) + 2,
+      U2n = T + *(p2 + 1) + 2;
+      (U1 < U1n) && (U2 < U2n) && (*U1 == *U2);
+      ++U1, ++U2) {
+  }
+
+  return U1 < U1n ?
+        (U2 < U2n ? *U1 - *U2 : 1) :
+        (U2 < U2n ? -1 : 0);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+#if (SS_BLOCKSIZE != 1) && (SS_INSERTIONSORT_THRESHOLD != 1)
+
+/* Insertionsort for small size groups */
+static
+void
+ss_insertionsort(const unsigned char *T, const int *PA,
+                 int *first, int *last, int depth) {
+  int *i, *j;
+  int t;
+  int r;
+
+  for(i = last - 2; first <= i; --i) {
+    for(t = *i, j = i + 1; 0 < (r = ss_compare(T, PA + t, PA + *j, depth));) {
+      do { *(j - 1) = *j; } while((++j < last) && (*j < 0));
+      if(last <= j) { break; }
+    }
+    if(r == 0) { *j = ~*j; }
+    *(j - 1) = t;
+  }
+}
+
+#endif /* (SS_BLOCKSIZE != 1) && (SS_INSERTIONSORT_THRESHOLD != 1) */
+
+
+/*---------------------------------------------------------------------------*/
+
+#if (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE)
+
+static INLINE
+void
+ss_fixdown(const unsigned char *Td, const int *PA,
+           int *SA, int i, int size) {
+  int j, k;
+  int v;
+  int c, d, e;
+
+  for(v = SA[i], c = Td[PA[v]]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = Td[PA[SA[k = j++]]];
+    if(d < (e = Td[PA[SA[j]]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+static
+void
+ss_heapsort(const unsigned char *Td, const int *PA, int *SA, int size) {
+  int i, m;
+  int t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(Td[PA[SA[m / 2]]] < Td[PA[SA[m]]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { ss_fixdown(Td, PA, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); ss_fixdown(Td, PA, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    ss_fixdown(Td, PA, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+static INLINE
+int *
+ss_median3(const unsigned char *Td, const int *PA,
+           int *v1, int *v2, int *v3) {
+  int *t;
+  if(Td[PA[*v1]] > Td[PA[*v2]]) { SWAP(v1, v2); }
+  if(Td[PA[*v2]] > Td[PA[*v3]]) {
+    if(Td[PA[*v1]] > Td[PA[*v3]]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+static INLINE
+int *
+ss_median5(const unsigned char *Td, const int *PA,
+           int *v1, int *v2, int *v3, int *v4, int *v5) {
+  int *t;
+  if(Td[PA[*v2]] > Td[PA[*v3]]) { SWAP(v2, v3); }
+  if(Td[PA[*v4]] > Td[PA[*v5]]) { SWAP(v4, v5); }
+  if(Td[PA[*v2]] > Td[PA[*v4]]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(Td[PA[*v1]] > Td[PA[*v3]]) { SWAP(v1, v3); }
+  if(Td[PA[*v1]] > Td[PA[*v4]]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(Td[PA[*v3]] > Td[PA[*v4]]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+static INLINE
+int *
+ss_pivot(const unsigned char *Td, const int *PA, int *first, int *last) {
+  int *middle;
+  int t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return ss_median3(Td, PA, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return ss_median5(Td, PA, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = ss_median3(Td, PA, first, first + t, first + (t << 1));
+  middle = ss_median3(Td, PA, middle - t, middle, middle + t);
+  last   = ss_median3(Td, PA, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return ss_median3(Td, PA, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Binary partition for substrings. */
+static INLINE
+int *
+ss_partition(const int *PA,
+                    int *first, int *last, int depth) {
+  int *a, *b;
+  int t;
+  for(a = first - 1, b = last;;) {
+    for(; (++a < b) && ((PA[*a] + depth) >= (PA[*a + 1] + 1));) { *a = ~*a; }
+    for(; (a < --b) && ((PA[*b] + depth) <  (PA[*b + 1] + 1));) { }
+    if(b <= a) { break; }
+    t = ~*b;
+    *b = *a;
+    *a = t;
+  }
+  if(first < a) { *first = ~*first; }
+  return a;
+}
+
+/* Multikey introsort for medium size groups. */
+static
+void
+ss_mintrosort(const unsigned char *T, const int *PA,
+              int *first, int *last,
+              int depth) {
+#define STACK_SIZE SS_MISORT_STACKSIZE
+  struct { int *a, *b, c; int d; } stack[STACK_SIZE];
+  const unsigned char *Td;
+  int *a, *b, *c, *d, *e, *f;
+  int s, t;
+  int ssize;
+  int limit;
+  int v, x = 0;
+
+  for(ssize = 0, limit = ss_ilg(last - first);;) {
+
+    if((last - first) <= SS_INSERTIONSORT_THRESHOLD) {
+#if 1 < SS_INSERTIONSORT_THRESHOLD
+      if(1 < (last - first)) { ss_insertionsort(T, PA, first, last, depth); }
+#endif
+      STACK_POP(first, last, depth, limit);
+      continue;
+    }
+
+    Td = T + depth;
+    if(limit-- == 0) { ss_heapsort(Td, PA, first, last - first); }
+    if(limit < 0) {
+      for(a = first + 1, v = Td[PA[*first]]; a < last; ++a) {
+        if((x = Td[PA[*a]]) != v) {
+          if(1 < (a - first)) { break; }
+          v = x;
+          first = a;
+        }
+      }
+      if(Td[PA[*first] - 1] < v) {
+        first = ss_partition(PA, first, a, depth);
+      }
+      if((a - first) <= (last - a)) {
+        if(1 < (a - first)) {
+          STACK_PUSH(a, last, depth, -1);
+          last = a, depth += 1, limit = ss_ilg(a - first);
+        } else {
+          first = a, limit = -1;
+        }
+      } else {
+        if(1 < (last - a)) {
+          STACK_PUSH(first, a, depth + 1, ss_ilg(a - first));
+          first = a, limit = -1;
+        } else {
+          last = a, depth += 1, limit = ss_ilg(a - first);
+        }
+      }
+      continue;
+    }
+
+    /* choose pivot */
+    a = ss_pivot(Td, PA, first, last);
+    v = Td[PA[*a]];
+    SWAP(*first, *a);
+
+    /* partition */
+    for(b = first; (++b < last) && ((x = Td[PA[*b]]) == v);) { }
+    if(((a = b) < last) && (x < v)) {
+      for(; (++b < last) && ((x = Td[PA[*b]]) <= v);) {
+        if(x == v) { SWAP(*b, *a); ++a; }
+      }
+    }
+    for(c = last; (b < --c) && ((x = Td[PA[*c]]) == v);) { }
+    if((b < (d = c)) && (x > v)) {
+      for(; (b < --c) && ((x = Td[PA[*c]]) >= v);) {
+        if(x == v) { SWAP(*c, *d); --d; }
+      }
+    }
+    for(; b < c;) {
+      SWAP(*b, *c);
+      for(; (++b < c) && ((x = Td[PA[*b]]) <= v);) {
+        if(x == v) { SWAP(*b, *a); ++a; }
+      }
+      for(; (b < --c) && ((x = Td[PA[*c]]) >= v);) {
+        if(x == v) { SWAP(*c, *d); --d; }
+      }
+    }
+
+    if(a <= d) {
+      c = b - 1;
+
+      if((s = a - first) > (t = b - a)) { s = t; }
+      for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+      if((s = d - c) > (t = last - d - 1)) { s = t; }
+      for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+
+      a = first + (b - a), c = last - (d - c);
+      b = (v <= Td[PA[*a] - 1]) ? a : ss_partition(PA, a, c, depth);
+
+      if((a - first) <= (last - c)) {
+        if((last - c) <= (c - b)) {
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          STACK_PUSH(c, last, depth, limit);
+          last = a;
+        } else if((a - first) <= (c - b)) {
+          STACK_PUSH(c, last, depth, limit);
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          last = a;
+        } else {
+          STACK_PUSH(c, last, depth, limit);
+          STACK_PUSH(first, a, depth, limit);
+          first = b, last = c, depth += 1, limit = ss_ilg(c - b);
+        }
+      } else {
+        if((a - first) <= (c - b)) {
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          STACK_PUSH(first, a, depth, limit);
+          first = c;
+        } else if((last - c) <= (c - b)) {
+          STACK_PUSH(first, a, depth, limit);
+          STACK_PUSH(b, c, depth + 1, ss_ilg(c - b));
+          first = c;
+        } else {
+          STACK_PUSH(first, a, depth, limit);
+          STACK_PUSH(c, last, depth, limit);
+          first = b, last = c, depth += 1, limit = ss_ilg(c - b);
+        }
+      }
+    } else {
+      limit += 1;
+      if(Td[PA[*first] - 1] < v) {
+        first = ss_partition(PA, first, last, depth);
+        limit = ss_ilg(last - first);
+      }
+      depth += 1;
+    }
+  }
+#undef STACK_SIZE
+}
+
+#endif /* (SS_BLOCKSIZE == 0) || (SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE) */
+
+
+/*---------------------------------------------------------------------------*/
+
+#if SS_BLOCKSIZE != 0
+
+static INLINE
+void
+ss_blockswap(int *a, int *b, int n) {
+  int t;
+  for(; 0 < n; --n, ++a, ++b) {
+    t = *a, *a = *b, *b = t;
+  }
+}
+
+static INLINE
+void
+ss_rotate(int *first, int *middle, int *last) {
+  int *a, *b, t;
+  int l, r;
+  l = middle - first, r = last - middle;
+  for(; (0 < l) && (0 < r);) {
+    if(l == r) { ss_blockswap(first, middle, l); break; }
+    if(l < r) {
+      a = last - 1, b = middle - 1;
+      t = *a;
+      do {
+        *a-- = *b, *b-- = *a;
+        if(b < first) {
+          *a = t;
+          last = a;
+          if((r -= l + 1) <= l) { break; }
+          a -= 1, b = middle - 1;
+          t = *a;
+        }
+      } while(1);
+    } else {
+      a = first, b = middle;
+      t = *a;
+      do {
+        *a++ = *b, *b++ = *a;
+        if(last <= b) {
+          *a = t;
+          first = a + 1;
+          if((l -= r + 1) <= r) { break; }
+          a += 1, b = middle;
+          t = *a;
+        }
+      } while(1);
+    }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static
+void
+ss_inplacemerge(const unsigned char *T, const int *PA,
+                int *first, int *middle, int *last,
+                int depth) {
+  const int *p;
+  int *a, *b;
+  int len, half;
+  int q, r;
+  int x;
+
+  for(;;) {
+    if(*(last - 1) < 0) { x = 1; p = PA + ~*(last - 1); }
+    else                { x = 0; p = PA +  *(last - 1); }
+    for(a = first, len = middle - first, half = len >> 1, r = -1;
+        0 < len;
+        len = half, half >>= 1) {
+      b = a + half;
+      q = ss_compare(T, PA + ((0 <= *b) ? *b : ~*b), p, depth);
+      if(q < 0) {
+        a = b + 1;
+        half -= (len & 1) ^ 1;
+      } else {
+        r = q;
+      }
+    }
+    if(a < middle) {
+      if(r == 0) { *a = ~*a; }
+      ss_rotate(a, middle, last);
+      last -= middle - a;
+      middle = a;
+      if(first == middle) { break; }
+    }
+    --last;
+    if(x != 0) { while(*--last < 0) { } }
+    if(middle == last) { break; }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Merge-forward with internal buffer. */
+static
+void
+ss_mergeforward(const unsigned char *T, const int *PA,
+                int *first, int *middle, int *last,
+                int *buf, int depth) {
+  int *a, *b, *c, *bufend;
+  int t;
+  int r;
+
+  bufend = buf + (middle - first) - 1;
+  ss_blockswap(buf, first, middle - first);
+
+  for(t = *(a = first), b = buf, c = middle;;) {
+    r = ss_compare(T, PA + *b, PA + *c, depth);
+    if(r < 0) {
+      do {
+        *a++ = *b;
+        if(bufend <= b) { *bufend = t; return; }
+        *b++ = *a;
+      } while(*b < 0);
+    } else if(r > 0) {
+      do {
+        *a++ = *c, *c++ = *a;
+        if(last <= c) {
+          while(b < bufend) { *a++ = *b, *b++ = *a; }
+          *a = *b, *b = t;
+          return;
+        }
+      } while(*c < 0);
+    } else {
+      *c = ~*c;
+      do {
+        *a++ = *b;
+        if(bufend <= b) { *bufend = t; return; }
+        *b++ = *a;
+      } while(*b < 0);
+
+      do {
+        *a++ = *c, *c++ = *a;
+        if(last <= c) {
+          while(b < bufend) { *a++ = *b, *b++ = *a; }
+          *a = *b, *b = t;
+          return;
+        }
+      } while(*c < 0);
+    }
+  }
+}
+
+/* Merge-backward with internal buffer. */
+static
+void
+ss_mergebackward(const unsigned char *T, const int *PA,
+                 int *first, int *middle, int *last,
+                 int *buf, int depth) {
+  const int *p1, *p2;
+  int *a, *b, *c, *bufend;
+  int t;
+  int r;
+  int x;
+
+  bufend = buf + (last - middle) - 1;
+  ss_blockswap(buf, middle, last - middle);
+
+  x = 0;
+  if(*bufend < 0)       { p1 = PA + ~*bufend; x |= 1; }
+  else                  { p1 = PA +  *bufend; }
+  if(*(middle - 1) < 0) { p2 = PA + ~*(middle - 1); x |= 2; }
+  else                  { p2 = PA +  *(middle - 1); }
+  for(t = *(a = last - 1), b = bufend, c = middle - 1;;) {
+    r = ss_compare(T, p1, p2, depth);
+    if(0 < r) {
+      if(x & 1) { do { *a-- = *b, *b-- = *a; } while(*b < 0); x ^= 1; }
+      *a-- = *b;
+      if(b <= buf) { *buf = t; break; }
+      *b-- = *a;
+      if(*b < 0) { p1 = PA + ~*b; x |= 1; }
+      else       { p1 = PA +  *b; }
+    } else if(r < 0) {
+      if(x & 2) { do { *a-- = *c, *c-- = *a; } while(*c < 0); x ^= 2; }
+      *a-- = *c, *c-- = *a;
+      if(c < first) {
+        while(buf < b) { *a-- = *b, *b-- = *a; }
+        *a = *b, *b = t;
+        break;
+      }
+      if(*c < 0) { p2 = PA + ~*c; x |= 2; }
+      else       { p2 = PA +  *c; }
+    } else {
+      if(x & 1) { do { *a-- = *b, *b-- = *a; } while(*b < 0); x ^= 1; }
+      *a-- = ~*b;
+      if(b <= buf) { *buf = t; break; }
+      *b-- = *a;
+      if(x & 2) { do { *a-- = *c, *c-- = *a; } while(*c < 0); x ^= 2; }
+      *a-- = *c, *c-- = *a;
+      if(c < first) {
+        while(buf < b) { *a-- = *b, *b-- = *a; }
+        *a = *b, *b = t;
+        break;
+      }
+      if(*b < 0) { p1 = PA + ~*b; x |= 1; }
+      else       { p1 = PA +  *b; }
+      if(*c < 0) { p2 = PA + ~*c; x |= 2; }
+      else       { p2 = PA +  *c; }
+    }
+  }
+}
+
+/* D&C based merge. */
+static
+void
+ss_swapmerge(const unsigned char *T, const int *PA,
+             int *first, int *middle, int *last,
+             int *buf, int bufsize, int depth) {
+#define STACK_SIZE SS_SMERGE_STACKSIZE
+#define GETIDX(a) ((0 <= (a)) ? (a) : (~(a)))
+#define MERGE_CHECK(a, b, c)\
+  do {\
+    if(((c) & 1) ||\
+       (((c) & 2) && (ss_compare(T, PA + GETIDX(*((a) - 1)), PA + *(a), depth) == 0))) {\
+      *(a) = ~*(a);\
+    }\
+    if(((c) & 4) && ((ss_compare(T, PA + GETIDX(*((b) - 1)), PA + *(b), depth) == 0))) {\
+      *(b) = ~*(b);\
+    }\
+  } while(0)
+  struct { int *a, *b, *c; int d; } stack[STACK_SIZE];
+  int *l, *r, *lm, *rm;
+  int m, len, half;
+  int ssize;
+  int check, next;
+
+  for(check = 0, ssize = 0;;) {
+    if((last - middle) <= bufsize) {
+      if((first < middle) && (middle < last)) {
+        ss_mergebackward(T, PA, first, middle, last, buf, depth);
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+      continue;
+    }
+
+    if((middle - first) <= bufsize) {
+      if(first < middle) {
+        ss_mergeforward(T, PA, first, middle, last, buf, depth);
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+      continue;
+    }
+
+    for(m = 0, len = MIN(middle - first, last - middle), half = len >> 1;
+        0 < len;
+        len = half, half >>= 1) {
+      if(ss_compare(T, PA + GETIDX(*(middle + m + half)),
+                       PA + GETIDX(*(middle - m - half - 1)), depth) < 0) {
+        m += half + 1;
+        half -= (len & 1) ^ 1;
+      }
+    }
+
+    if(0 < m) {
+      lm = middle - m, rm = middle + m;
+      ss_blockswap(lm, middle, m);
+      l = r = middle, next = 0;
+      if(rm < last) {
+        if(*rm < 0) {
+          *rm = ~*rm;
+          if(first < lm) { for(; *--l < 0;) { } next |= 4; }
+          next |= 1;
+        } else if(first < lm) {
+          for(; *r < 0; ++r) { }
+          next |= 2;
+        }
+      }
+
+      if((l - first) <= (last - r)) {
+        STACK_PUSH(r, rm, last, (next & 3) | (check & 4));
+        middle = lm, last = l, check = (check & 3) | (next & 4);
+      } else {
+        if((next & 2) && (r == middle)) { next ^= 6; }
+        STACK_PUSH(first, lm, l, (check & 3) | (next & 4));
+        first = r, middle = rm, check = (next & 3) | (check & 4);
+      }
+    } else {
+      if(ss_compare(T, PA + GETIDX(*(middle - 1)), PA + *middle, depth) == 0) {
+        *middle = ~*middle;
+      }
+      MERGE_CHECK(first, last, check);
+      STACK_POP(first, middle, last, check);
+    }
+  }
+#undef STACK_SIZE
+}
+
+#endif /* SS_BLOCKSIZE != 0 */
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Substring sort */
+static
+void
+sssort(const unsigned char *T, const int *PA,
+       int *first, int *last,
+       int *buf, int bufsize,
+       int depth, int n, int lastsuffix) {
+  int *a;
+#if SS_BLOCKSIZE != 0
+  int *b, *middle, *curbuf;
+  int j, k, curbufsize, limit;
+#endif
+  int i;
+
+  if(lastsuffix != 0) { ++first; }
+
+#if SS_BLOCKSIZE == 0
+  ss_mintrosort(T, PA, first, last, depth);
+#else
+  if((bufsize < SS_BLOCKSIZE) &&
+      (bufsize < (last - first)) &&
+      (bufsize < (limit = ss_isqrt(last - first)))) {
+    if(SS_BLOCKSIZE < limit) { limit = SS_BLOCKSIZE; }
+    buf = middle = last - limit, bufsize = limit;
+  } else {
+    middle = last, limit = 0;
+  }
+  for(a = first, i = 0; SS_BLOCKSIZE < (middle - a); a += SS_BLOCKSIZE, ++i) {
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+    ss_mintrosort(T, PA, a, a + SS_BLOCKSIZE, depth);
+#elif 1 < SS_BLOCKSIZE
+    ss_insertionsort(T, PA, a, a + SS_BLOCKSIZE, depth);
+#endif
+    curbufsize = last - (a + SS_BLOCKSIZE);
+    curbuf = a + SS_BLOCKSIZE;
+    if(curbufsize <= bufsize) { curbufsize = bufsize, curbuf = buf; }
+    for(b = a, k = SS_BLOCKSIZE, j = i; j & 1; b -= k, k <<= 1, j >>= 1) {
+      ss_swapmerge(T, PA, b - k, b, b + k, curbuf, curbufsize, depth);
+    }
+  }
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+  ss_mintrosort(T, PA, a, middle, depth);
+#elif 1 < SS_BLOCKSIZE
+  ss_insertionsort(T, PA, a, middle, depth);
+#endif
+  for(k = SS_BLOCKSIZE; i != 0; k <<= 1, i >>= 1) {
+    if(i & 1) {
+      ss_swapmerge(T, PA, a - k, a, middle, buf, bufsize, depth);
+      a -= k;
+    }
+  }
+  if(limit != 0) {
+#if SS_INSERTIONSORT_THRESHOLD < SS_BLOCKSIZE
+    ss_mintrosort(T, PA, middle, last, depth);
+#elif 1 < SS_BLOCKSIZE
+    ss_insertionsort(T, PA, middle, last, depth);
+#endif
+    ss_inplacemerge(T, PA, first, middle, last, depth);
+  }
+#endif
+
+  if(lastsuffix != 0) {
+    /* Insert last type B* suffix. */
+    int PAi[2]; PAi[0] = PA[*(first - 1)], PAi[1] = n - 2;
+    for(a = first, i = *(first - 1);
+        (a < last) && ((*a < 0) || (0 < ss_compare(T, &(PAi[0]), PA + *a, depth)));
+        ++a) {
+      *(a - 1) = *a;
+    }
+    *(a - 1) = i;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+int
+tr_ilg(int n) {
+  return (n & 0xffff0000) ?
+          ((n & 0xff000000) ?
+            24 + lg_table[(n >> 24) & 0xff] :
+            16 + lg_table[(n >> 16) & 0xff]) :
+          ((n & 0x0000ff00) ?
+             8 + lg_table[(n >>  8) & 0xff] :
+             0 + lg_table[(n >>  0) & 0xff]);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Simple insertionsort for small size groups. */
+static
+void
+tr_insertionsort(const int *ISAd, int *first, int *last) {
+  int *a, *b;
+  int t, r;
+
+  for(a = first + 1; a < last; ++a) {
+    for(t = *a, b = a - 1; 0 > (r = ISAd[t] - ISAd[*b]);) {
+      do { *(b + 1) = *b; } while((first <= --b) && (*b < 0));
+      if(b < first) { break; }
+    }
+    if(r == 0) { *b = ~*b; }
+    *(b + 1) = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+void
+tr_fixdown(const int *ISAd, int *SA, int i, int size) {
+  int j, k;
+  int v;
+  int c, d, e;
+
+  for(v = SA[i], c = ISAd[v]; (j = 2 * i + 1) < size; SA[i] = SA[k], i = k) {
+    d = ISAd[SA[k = j++]];
+    if(d < (e = ISAd[SA[j]])) { k = j; d = e; }
+    if(d <= c) { break; }
+  }
+  SA[i] = v;
+}
+
+/* Simple top-down heapsort. */
+static
+void
+tr_heapsort(const int *ISAd, int *SA, int size) {
+  int i, m;
+  int t;
+
+  m = size;
+  if((size % 2) == 0) {
+    m--;
+    if(ISAd[SA[m / 2]] < ISAd[SA[m]]) { SWAP(SA[m], SA[m / 2]); }
+  }
+
+  for(i = m / 2 - 1; 0 <= i; --i) { tr_fixdown(ISAd, SA, i, m); }
+  if((size % 2) == 0) { SWAP(SA[0], SA[m]); tr_fixdown(ISAd, SA, 0, m); }
+  for(i = m - 1; 0 < i; --i) {
+    t = SA[0], SA[0] = SA[i];
+    tr_fixdown(ISAd, SA, 0, i);
+    SA[i] = t;
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Returns the median of three elements. */
+static INLINE
+int *
+tr_median3(const int *ISAd, int *v1, int *v2, int *v3) {
+  int *t;
+  if(ISAd[*v1] > ISAd[*v2]) { SWAP(v1, v2); }
+  if(ISAd[*v2] > ISAd[*v3]) {
+    if(ISAd[*v1] > ISAd[*v3]) { return v1; }
+    else { return v3; }
+  }
+  return v2;
+}
+
+/* Returns the median of five elements. */
+static INLINE
+int *
+tr_median5(const int *ISAd,
+           int *v1, int *v2, int *v3, int *v4, int *v5) {
+  int *t;
+  if(ISAd[*v2] > ISAd[*v3]) { SWAP(v2, v3); }
+  if(ISAd[*v4] > ISAd[*v5]) { SWAP(v4, v5); }
+  if(ISAd[*v2] > ISAd[*v4]) { SWAP(v2, v4); SWAP(v3, v5); }
+  if(ISAd[*v1] > ISAd[*v3]) { SWAP(v1, v3); }
+  if(ISAd[*v1] > ISAd[*v4]) { SWAP(v1, v4); SWAP(v3, v5); }
+  if(ISAd[*v3] > ISAd[*v4]) { return v4; }
+  return v3;
+}
+
+/* Returns the pivot element. */
+static INLINE
+int *
+tr_pivot(const int *ISAd, int *first, int *last) {
+  int *middle;
+  int t;
+
+  t = last - first;
+  middle = first + t / 2;
+
+  if(t <= 512) {
+    if(t <= 32) {
+      return tr_median3(ISAd, first, middle, last - 1);
+    } else {
+      t >>= 2;
+      return tr_median5(ISAd, first, first + t, middle, last - 1 - t, last - 1);
+    }
+  }
+  t >>= 3;
+  first  = tr_median3(ISAd, first, first + t, first + (t << 1));
+  middle = tr_median3(ISAd, middle - t, middle, middle + t);
+  last   = tr_median3(ISAd, last - 1 - (t << 1), last - 1 - t, last - 1);
+  return tr_median3(ISAd, first, middle, last);
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+typedef struct _trbudget_t trbudget_t;
+struct _trbudget_t {
+  int chance;
+  int remain;
+  int incval;
+  int count;
+};
+
+static INLINE
+void
+trbudget_init(trbudget_t *budget, int chance, int incval) {
+  budget->chance = chance;
+  budget->remain = budget->incval = incval;
+}
+
+static INLINE
+int
+trbudget_check(trbudget_t *budget, int size) {
+  if(size <= budget->remain) { budget->remain -= size; return 1; }
+  if(budget->chance == 0) { budget->count += size; return 0; }
+  budget->remain += budget->incval - size;
+  budget->chance -= 1;
+  return 1;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+static INLINE
+void
+tr_partition(const int *ISAd,
+             int *first, int *middle, int *last,
+             int **pa, int **pb, int v) {
+  int *a, *b, *c, *d, *e, *f;
+  int t, s;
+  int x = 0;
+
+  for(b = middle - 1; (++b < last) && ((x = ISAd[*b]) == v);) { }
+  if(((a = b) < last) && (x < v)) {
+    for(; (++b < last) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+  }
+  for(c = last; (b < --c) && ((x = ISAd[*c]) == v);) { }
+  if((b < (d = c)) && (x > v)) {
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+  for(; b < c;) {
+    SWAP(*b, *c);
+    for(; (++b < c) && ((x = ISAd[*b]) <= v);) {
+      if(x == v) { SWAP(*b, *a); ++a; }
+    }
+    for(; (b < --c) && ((x = ISAd[*c]) >= v);) {
+      if(x == v) { SWAP(*c, *d); --d; }
+    }
+  }
+
+  if(a <= d) {
+    c = b - 1;
+    if((s = a - first) > (t = b - a)) { s = t; }
+    for(e = first, f = b - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    if((s = d - c) > (t = last - d - 1)) { s = t; }
+    for(e = b, f = last - s; 0 < s; --s, ++e, ++f) { SWAP(*e, *f); }
+    first += (b - a), last -= (d - c);
+  }
+  *pa = first, *pb = last;
+}
+
+static
+void
+tr_copy(int *ISA, const int *SA,
+        int *first, int *a, int *b, int *last,
+        int depth) {
+  /* sort suffixes of middle partition
+     by using sorted order of suffixes of left and right partition. */
+  int *c, *d, *e;
+  int s, v;
+
+  v = b - SA - 1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      ISA[s] = d - SA;
+    }
+  }
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      ISA[s] = d - SA;
+    }
+  }
+}
+
+static
+void
+tr_partialcopy(int *ISA, const int *SA,
+               int *first, int *a, int *b, int *last,
+               int depth) {
+  int *c, *d, *e;
+  int s, v;
+  int rank, lastrank, newrank = -1;
+
+  v = b - SA - 1;
+  lastrank = -1;
+  for(c = first, d = a - 1; c <= d; ++c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *++d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+
+  lastrank = -1;
+  for(e = d; first <= e; --e) {
+    rank = ISA[*e];
+    if(lastrank != rank) { lastrank = rank; newrank = e - SA; }
+    if(newrank != rank) { ISA[*e] = newrank; }
+  }
+
+  lastrank = -1;
+  for(c = last - 1, e = d + 1, d = b; e < d; --c) {
+    if((0 <= (s = *c - depth)) && (ISA[s] == v)) {
+      *--d = s;
+      rank = ISA[s + depth];
+      if(lastrank != rank) { lastrank = rank; newrank = d - SA; }
+      ISA[s] = newrank;
+    }
+  }
+}
+
+static
+void
+tr_introsort(int *ISA, const int *ISAd,
+             int *SA, int *first, int *last,
+             trbudget_t *budget) {
+#define STACK_SIZE TR_STACKSIZE
+  struct { const int *a; int *b, *c; int d, e; }stack[STACK_SIZE];
+  int *a, *b, *c;
+  int t;
+  int v, x = 0;
+  int incr = ISAd - ISA;
+  int limit, next;
+  int ssize, trlink = -1;
+
+  for(ssize = 0, limit = tr_ilg(last - first);;) {
+
+    if(limit < 0) {
+      if(limit == -1) {
+        /* tandem repeat partition */
+        tr_partition(ISAd - incr, first, first, last, &a, &b, last - SA - 1);
+
+        /* update ranks */
+        if(a < last) {
+          for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+        }
+        if(b < last) {
+          for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; }
+        }
+
+        /* push */
+        if(1 < (b - a)) {
+          STACK_PUSH5(NULL, a, b, 0, 0);
+          STACK_PUSH5(ISAd - incr, first, last, -2, trlink);
+          trlink = ssize - 2;
+        }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, tr_ilg(last - b), trlink);
+            last = a, limit = tr_ilg(a - first);
+          } else if(1 < (last - b)) {
+            first = b, limit = tr_ilg(last - b);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, tr_ilg(a - first), trlink);
+            first = b, limit = tr_ilg(last - b);
+          } else if(1 < (a - first)) {
+            last = a, limit = tr_ilg(a - first);
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      } else if(limit == -2) {
+        /* tandem repeat copy */
+        a = stack[--ssize].b, b = stack[ssize].c;
+        if(stack[ssize].d == 0) {
+          tr_copy(ISA, SA, first, a, b, last, ISAd - ISA);
+        } else {
+          if(0 <= trlink) { stack[trlink].d = -1; }
+          tr_partialcopy(ISA, SA, first, a, b, last, ISAd - ISA);
+        }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      } else {
+        /* sorted partition */
+        if(0 <= *first) {
+          a = first;
+          do { ISA[*a] = a - SA; } while((++a < last) && (0 <= *a));
+          first = a;
+        }
+        if(first < last) {
+          a = first; do { *a = ~*a; } while(*++a < 0);
+          next = (ISA[*a] != ISAd[*a]) ? tr_ilg(a - first + 1) : -1;
+          if(++a < last) { for(b = first, v = a - SA - 1; b < a; ++b) { ISA[*b] = v; } }
+
+          /* push */
+          if(trbudget_check(budget, a - first)) {
+            if((a - first) <= (last - a)) {
+              STACK_PUSH5(ISAd, a, last, -3, trlink);
+              ISAd += incr, last = a, limit = next;
+            } else {
+              if(1 < (last - a)) {
+                STACK_PUSH5(ISAd + incr, first, a, next, trlink);
+                first = a, limit = -3;
+              } else {
+                ISAd += incr, last = a, limit = next;
+              }
+            }
+          } else {
+            if(0 <= trlink) { stack[trlink].d = -1; }
+            if(1 < (last - a)) {
+              first = a, limit = -3;
+            } else {
+              STACK_POP5(ISAd, first, last, limit, trlink);
+            }
+          }
+        } else {
+          STACK_POP5(ISAd, first, last, limit, trlink);
+        }
+      }
+      continue;
+    }
+
+    if((last - first) <= TR_INSERTIONSORT_THRESHOLD) {
+      tr_insertionsort(ISAd, first, last);
+      limit = -3;
+      continue;
+    }
+
+    if(limit-- == 0) {
+      tr_heapsort(ISAd, first, last - first);
+      for(a = last - 1; first < a; a = b) {
+        for(x = ISAd[*a], b = a - 1; (first <= b) && (ISAd[*b] == x); --b) { *b = ~*b; }
+      }
+      limit = -3;
+      continue;
+    }
+
+    /* choose pivot */
+    a = tr_pivot(ISAd, first, last);
+    SWAP(*first, *a);
+    v = ISAd[*first];
+
+    /* partition */
+    tr_partition(ISAd, first, first + 1, last, &a, &b, v);
+    if((last - first) != (b - a)) {
+      next = (ISA[*a] != v) ? tr_ilg(b - a) : -1;
+
+      /* update ranks */
+      for(c = first, v = a - SA - 1; c < a; ++c) { ISA[*c] = v; }
+      if(b < last) { for(c = a, v = b - SA - 1; c < b; ++c) { ISA[*c] = v; } }
+
+      /* push */
+      if((1 < (b - a)) && (trbudget_check(budget, b - a))) {
+        if((a - first) <= (last - b)) {
+          if((last - b) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              last = a;
+            } else if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((a - first) <= (b - a)) {
+            if(1 < (a - first)) {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              STACK_PUSH5(ISAd, b, last, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        } else {
+          if((a - first) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              first = b;
+            } else if(1 < (a - first)) {
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              last = a;
+            } else {
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else if((last - b) <= (b - a)) {
+            if(1 < (last - b)) {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              STACK_PUSH5(ISAd + incr, a, b, next, trlink);
+              first = b;
+            } else {
+              STACK_PUSH5(ISAd, first, a, limit, trlink);
+              ISAd += incr, first = a, last = b, limit = next;
+            }
+          } else {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            ISAd += incr, first = a, last = b, limit = next;
+          }
+        }
+      } else {
+        if((1 < (b - a)) && (0 <= trlink)) { stack[trlink].d = -1; }
+        if((a - first) <= (last - b)) {
+          if(1 < (a - first)) {
+            STACK_PUSH5(ISAd, b, last, limit, trlink);
+            last = a;
+          } else if(1 < (last - b)) {
+            first = b;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        } else {
+          if(1 < (last - b)) {
+            STACK_PUSH5(ISAd, first, a, limit, trlink);
+            first = b;
+          } else if(1 < (a - first)) {
+            last = a;
+          } else {
+            STACK_POP5(ISAd, first, last, limit, trlink);
+          }
+        }
+      }
+    } else {
+      if(trbudget_check(budget, last - first)) {
+        limit = tr_ilg(last - first), ISAd += incr;
+      } else {
+        if(0 <= trlink) { stack[trlink].d = -1; }
+        STACK_POP5(ISAd, first, last, limit, trlink);
+      }
+    }
+  }
+#undef STACK_SIZE
+}
+
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Tandem repeat sort */
+static
+void
+trsort(int *ISA, int *SA, int n, int depth) {
+  int *ISAd;
+  int *first, *last;
+  trbudget_t budget;
+  int t, skip, unsorted;
+
+  trbudget_init(&budget, tr_ilg(n) * 2 / 3, n);
+/*  trbudget_init(&budget, tr_ilg(n) * 3 / 4, n); */
+  for(ISAd = ISA + depth; -n < *SA; ISAd += ISAd - ISA) {
+    first = SA;
+    skip = 0;
+    unsorted = 0;
+    do {
+      if((t = *first) < 0) { first -= t; skip += t; }
+      else {
+        if(skip != 0) { *(first + skip) = skip; skip = 0; }
+        last = SA + ISA[t] + 1;
+        if(1 < (last - first)) {
+          budget.count = 0;
+          tr_introsort(ISA, ISAd, SA, first, last, &budget);
+          if(budget.count != 0) { unsorted += budget.count; }
+          else { skip = first - last; }
+        } else if((last - first) == 1) {
+          skip = -1;
+        }
+        first = last;
+      }
+    } while(first < (SA + n));
+    if(skip != 0) { *(first + skip) = skip; }
+    if(unsorted == 0) { break; }
+  }
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/* Sorts suffixes of type B*. */
+static
+int
+sort_typeBstar(const unsigned char *T, int *SA,
+               int *bucket_A, int *bucket_B,
+               int n, int openMP) {
+  int *PAb, *ISAb, *buf;
+#ifdef LIBBSC_OPENMP
+  int *curbuf;
+  int l;
+#endif
+  int i, j, k, t, m, bufsize;
+  int c0, c1;
+#ifdef LIBBSC_OPENMP
+  int d0, d1;
+#endif
+  (void)openMP;
+
+  /* Initialize bucket arrays. */
+  for(i = 0; i < BUCKET_A_SIZE; ++i) { bucket_A[i] = 0; }
+  for(i = 0; i < BUCKET_B_SIZE; ++i) { bucket_B[i] = 0; }
+
+  /* Count the number of occurrences of the first one or two characters of each
+     type A, B and B* suffix. Moreover, store the beginning position of all
+     type B* suffixes into the array SA. */
+  for(i = n - 1, m = n, c0 = T[n - 1]; 0 <= i;) {
+    /* type A suffix. */
+    do { ++BUCKET_A(c1 = c0); } while((0 <= --i) && ((c0 = T[i]) >= c1));
+    if(0 <= i) {
+      /* type B* suffix. */
+      ++BUCKET_BSTAR(c0, c1);
+      SA[--m] = i;
+      /* type B suffix. */
+      for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) <= c1); --i, c1 = c0) {
+        ++BUCKET_B(c0, c1);
+      }
+    }
+  }
+  m = n - m;
+/*
+note:
+  A type B* suffix is lexicographically smaller than a type B suffix that
+  begins with the same first two characters.
+*/
+
+  /* Calculate the index of start/end point of each bucket. */
+  for(c0 = 0, i = 0, j = 0; c0 < ALPHABET_SIZE; ++c0) {
+    t = i + BUCKET_A(c0);
+    BUCKET_A(c0) = i + j; /* start point */
+    i = t + BUCKET_B(c0, c0);
+    for(c1 = c0 + 1; c1 < ALPHABET_SIZE; ++c1) {
+      j += BUCKET_BSTAR(c0, c1);
+      BUCKET_BSTAR(c0, c1) = j; /* end point */
+      i += BUCKET_B(c0, c1);
+    }
+  }
+
+  if(0 < m) {
+    /* Sort the type B* suffixes by their first two characters. */
+    PAb = SA + n - m; ISAb = SA + m;
+    for(i = m - 2; 0 <= i; --i) {
+      t = PAb[i], c0 = T[t], c1 = T[t + 1];
+      SA[--BUCKET_BSTAR(c0, c1)] = i;
+    }
+    t = PAb[m - 1], c0 = T[t], c1 = T[t + 1];
+    SA[--BUCKET_BSTAR(c0, c1)] = m - 1;
+
+    /* Sort the type B* substrings using sssort. */
+#ifdef LIBBSC_OPENMP
+    if (openMP)
+    {
+        buf = SA + m;
+        c0 = ALPHABET_SIZE - 2, c1 = ALPHABET_SIZE - 1, j = m;
+#pragma omp parallel default(shared) private(bufsize, curbuf, k, l, d0, d1)
+        {
+          bufsize = (n - (2 * m)) / omp_get_num_threads();
+          curbuf = buf + omp_get_thread_num() * bufsize;
+          k = 0;
+          for(;;) {
+            #pragma omp critical(sssort_lock)
+            {
+              if(0 < (l = j)) {
+                d0 = c0, d1 = c1;
+                do {
+                  k = BUCKET_BSTAR(d0, d1);
+                  if(--d1 <= d0) {
+                    d1 = ALPHABET_SIZE - 1;
+                    if(--d0 < 0) { break; }
+                  }
+                } while(((l - k) <= 1) && (0 < (l = k)));
+                c0 = d0, c1 = d1, j = k;
+              }
+            }
+            if(l == 0) { break; }
+            sssort(T, PAb, SA + k, SA + l,
+                   curbuf, bufsize, 2, n, *(SA + k) == (m - 1));
+          }
+        }
+    }
+    else
+    {
+        buf = SA + m, bufsize = n - (2 * m);
+        for(c0 = ALPHABET_SIZE - 2, j = m; 0 < j; --c0) {
+          for(c1 = ALPHABET_SIZE - 1; c0 < c1; j = i, --c1) {
+            i = BUCKET_BSTAR(c0, c1);
+            if(1 < (j - i)) {
+              sssort(T, PAb, SA + i, SA + j,
+                     buf, bufsize, 2, n, *(SA + i) == (m - 1));
+            }
+          }
+        }
+    }
+#else
+    buf = SA + m, bufsize = n - (2 * m);
+    for(c0 = ALPHABET_SIZE - 2, j = m; 0 < j; --c0) {
+      for(c1 = ALPHABET_SIZE - 1; c0 < c1; j = i, --c1) {
+        i = BUCKET_BSTAR(c0, c1);
+        if(1 < (j - i)) {
+          sssort(T, PAb, SA + i, SA + j,
+                 buf, bufsize, 2, n, *(SA + i) == (m - 1));
+        }
+      }
+    }
+#endif
+
+    /* Compute ranks of type B* substrings. */
+    for(i = m - 1; 0 <= i; --i) {
+      if(0 <= SA[i]) {
+        j = i;
+        do { ISAb[SA[i]] = i; } while((0 <= --i) && (0 <= SA[i]));
+        SA[i + 1] = i - j;
+        if(i <= 0) { break; }
+      }
+      j = i;
+      do { ISAb[SA[i] = ~SA[i]] = j; } while(SA[--i] < 0);
+      ISAb[SA[i]] = j;
+    }
+
+    /* Construct the inverse suffix array of type B* suffixes using trsort. */
+    trsort(ISAb, SA, m, 1);
+
+    /* Set the sorted order of tyoe B* suffixes. */
+    for(i = n - 1, j = m, c0 = T[n - 1]; 0 <= i;) {
+      for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) >= c1); --i, c1 = c0) { }
+      if(0 <= i) {
+        t = i;
+        for(--i, c1 = c0; (0 <= i) && ((c0 = T[i]) <= c1); --i, c1 = c0) { }
+        SA[ISAb[--j]] = ((t == 0) || (1 < (t - i))) ? t : ~t;
+      }
+    }
+
+    /* Calculate the index of start/end point of each bucket. */
+    BUCKET_B(ALPHABET_SIZE - 1, ALPHABET_SIZE - 1) = n; /* end point */
+    for(c0 = ALPHABET_SIZE - 2, k = m - 1; 0 <= c0; --c0) {
+      i = BUCKET_A(c0 + 1) - 1;
+      for(c1 = ALPHABET_SIZE - 1; c0 < c1; --c1) {
+        t = i - BUCKET_B(c0, c1);
+        BUCKET_B(c0, c1) = i; /* end point */
+
+        /* Move all type B* suffixes to the correct position. */
+        for(i = t, j = BUCKET_BSTAR(c0, c1);
+            j <= k;
+            --i, --k) { SA[i] = SA[k]; }
+      }
+      BUCKET_BSTAR(c0, c0 + 1) = i - BUCKET_B(c0, c0) + 1; /* start point */
+      BUCKET_B(c0, c0) = i; /* end point */
+    }
+  }
+
+  return m;
+}
+
+/* Constructs the suffix array by using the sorted order of type B* suffixes. */
+static
+void
+construct_SA(const unsigned char *T, int *SA,
+             int *bucket_A, int *bucket_B,
+             int n, int m) {
+  int *i, *j, *k;
+  int s;
+  int c0, c1, c2;
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+          *j = ~s;
+          c0 = T[--s];
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j);
+          *k-- = s;
+        } else {
+          assert(((s == 0) && (T[s] == c1)) || (s < 0));
+          *j = ~s;
+        }
+      }
+    }
+  }
+
+  /* Construct the suffix array by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  *k++ = (T[n - 2] < c2) ? ~(n - 1) : (n - 1);
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+      c0 = T[--s];
+      if((s == 0) || (T[s - 1] < c0)) { s = ~s; }
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      *k++ = s;
+    } else {
+      assert(s < 0);
+      *i = ~s;
+    }
+  }
+}
+
+/* Constructs the burrows-wheeler transformed string directly
+   by using the sorted order of type B* suffixes. */
+static
+int
+construct_BWT(const unsigned char *T, int *SA,
+              int *bucket_A, int *bucket_B,
+              int n, int m) {
+  int *i, *j, *k, *orig;
+  int s;
+  int c0, c1, c2;
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+          c0 = T[--s];
+          *j = ~((int)c0);
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j);
+          *k-- = s;
+        } else if(s != 0) {
+          *j = ~s;
+#ifndef NDEBUG
+        } else {
+          assert(T[s] == c1);
+#endif
+        }
+      }
+    }
+  }
+
+  /* Construct the BWTed string by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  *k++ = (T[n - 2] < c2) ? ~((int)T[n - 2]) : (n - 1);
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n, orig = SA; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+      c0 = T[--s];
+      *i = c0;
+      if((0 < s) && (T[s - 1] < c0)) { s = ~((int)T[s - 1]); }
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      *k++ = s;
+    } else if(s != 0) {
+      *i = ~s;
+    } else {
+      orig = i;
+    }
+  }
+
+  return orig - SA;
+}
+
+/* Constructs the burrows-wheeler transformed string directly
+   by using the sorted order of type B* suffixes. */
+static
+int
+construct_BWT_indexes(const unsigned char *T, int *SA,
+                      int *bucket_A, int *bucket_B,
+                      int n, int m,
+                      unsigned char * num_indexes, int * indexes) {
+  int *i, *j, *k, *orig;
+  int s;
+  int c0, c1, c2;
+
+  int mod = n / 8;
+  {
+      mod |= mod >> 1;  mod |= mod >> 2;
+      mod |= mod >> 4;  mod |= mod >> 8;
+      mod |= mod >> 16; mod >>= 1;
+
+      *num_indexes = (unsigned char)((n - 1) / (mod + 1));
+  }
+
+  if(0 < m) {
+    /* Construct the sorted order of type B suffixes by using
+       the sorted order of type B* suffixes. */
+    for(c1 = ALPHABET_SIZE - 2; 0 <= c1; --c1) {
+      /* Scan the suffix array from right to left. */
+      for(i = SA + BUCKET_BSTAR(c1, c1 + 1),
+          j = SA + BUCKET_A(c1 + 1) - 1, k = NULL, c2 = -1;
+          i <= j;
+          --j) {
+        if(0 < (s = *j)) {
+          assert(T[s] == c1);
+          assert(((s + 1) < n) && (T[s] <= T[s + 1]));
+          assert(T[s - 1] <= T[s]);
+
+          if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = j - SA;
+
+          c0 = T[--s];
+          *j = ~((int)c0);
+          if((0 < s) && (T[s - 1] > c0)) { s = ~s; }
+          if(c0 != c2) {
+            if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
+            k = SA + BUCKET_B(c2 = c0, c1);
+          }
+          assert(k < j);
+          *k-- = s;
+        } else if(s != 0) {
+          *j = ~s;
+#ifndef NDEBUG
+        } else {
+          assert(T[s] == c1);
+#endif
+        }
+      }
+    }
+  }
+
+  /* Construct the BWTed string by using
+     the sorted order of type B suffixes. */
+  k = SA + BUCKET_A(c2 = T[n - 1]);
+  if (T[n - 2] < c2) {
+    if (((n - 1) & mod) == 0) indexes[(n - 1) / (mod + 1) - 1] = k - SA;
+    *k++ = ~((int)T[n - 2]);
+  }
+  else {
+    *k++ = n - 1;
+  }
+
+  /* Scan the suffix array from left to right. */
+  for(i = SA, j = SA + n, orig = SA; i < j; ++i) {
+    if(0 < (s = *i)) {
+      assert(T[s - 1] >= T[s]);
+
+      if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = i - SA;
+
+      c0 = T[--s];
+      *i = c0;
+      if(c0 != c2) {
+        BUCKET_A(c2) = k - SA;
+        k = SA + BUCKET_A(c2 = c0);
+      }
+      assert(i < k);
+      if((0 < s) && (T[s - 1] < c0)) {
+          if ((s & mod) == 0) indexes[s / (mod + 1) - 1] = k - SA;
+          *k++ = ~((int)T[s - 1]);
+      } else
+        *k++ = s;
+    } else if(s != 0) {
+      *i = ~s;
+    } else {
+      orig = i;
+    }
+  }
+
+  return orig - SA;
+}
+
+
+/*---------------------------------------------------------------------------*/
+
+/*- Function -*/
+
+int
+divsufsort(const unsigned char *T, int *SA, int n, int openMP) {
+  int *bucket_A, *bucket_B;
+  int m;
+  int err = 0;
+
+  /* Check arguments. */
+  if((T == NULL) || (SA == NULL) || (n < 0)) { return -1; }
+  else if(n == 0) { return 0; }
+  else if(n == 1) { SA[0] = 0; return 0; }
+  else if(n == 2) { m = (T[0] < T[1]); SA[m ^ 1] = 0, SA[m] = 1; return 0; }
+
+  bucket_A = (int *)malloc(BUCKET_A_SIZE * sizeof(int));
+  bucket_B = (int *)malloc(BUCKET_B_SIZE * sizeof(int));
+
+  /* Suffixsort. */
+  if((bucket_A != NULL) && (bucket_B != NULL)) {
+    m = sort_typeBstar(T, SA, bucket_A, bucket_B, n, openMP);
+    construct_SA(T, SA, bucket_A, bucket_B, n, m);
+  } else {
+    err = -2;
+  }
+
+  free(bucket_B);
+  free(bucket_A);
+
+  return err;
+}
+
+int
+divbwt(const unsigned char *T, unsigned char *U, int *A, int n, unsigned char * num_indexes, int * indexes, int openMP) {
+  int *B;
+  int *bucket_A, *bucket_B;
+  int m, pidx, i;
+
+  /* Check arguments. */
+  if((T == NULL) || (U == NULL) || (n < 0)) { return -1; }
+  else if(n <= 1) { if(n == 1) { U[0] = T[0]; } return n; }
+
+  if((B = A) == NULL) { B = (int *)malloc((size_t)(n + 1) * sizeof(int)); }
+  bucket_A = (int *)malloc(BUCKET_A_SIZE * sizeof(int));
+  bucket_B = (int *)malloc(BUCKET_B_SIZE * sizeof(int));
+
+  /* Burrows-Wheeler Transform. */
+  if((B != NULL) && (bucket_A != NULL) && (bucket_B != NULL)) {
+    m = sort_typeBstar(T, B, bucket_A, bucket_B, n, openMP);
+
+    if (num_indexes == NULL || indexes == NULL) {
+        pidx = construct_BWT(T, B, bucket_A, bucket_B, n, m);
+    } else {
+        pidx = construct_BWT_indexes(T, B, bucket_A, bucket_B, n, m, num_indexes, indexes);
+    }
+
+    /* Copy to output string. */
+    U[0] = T[n - 1];
+    for(i = 0; i < pidx; ++i) { U[i + 1] = (unsigned char)B[i]; }
+    for(i += 1; i < n; ++i) { U[i] = (unsigned char)B[i]; }
+    pidx += 1;
+  } else {
+    pidx = -2;
+  }
+
+  free(bucket_B);
+  free(bucket_A);
+  if(A == NULL) { free(B); }
+
+  return pidx;
+}
diff --git a/vendor/github.com/DataDog/zstd/divsufsort.h b/vendor/github.com/DataDog/zstd/divsufsort.h
new file mode 100644
index 0000000..5440994
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/divsufsort.h
@@ -0,0 +1,67 @@
+/*
+ * divsufsort.h for libdivsufsort-lite
+ * Copyright (c) 2003-2008 Yuta Mori All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person
+ * obtaining a copy of this software and associated documentation
+ * files (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use,
+ * copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following
+ * conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
+ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
+ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef _DIVSUFSORT_H
+#define _DIVSUFSORT_H 1
+
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+
+
+/*- Prototypes -*/
+
+/**
+ * Constructs the suffix array of a given string.
+ * @param T [0..n-1] The input string.
+ * @param SA [0..n-1] The output array of suffixes.
+ * @param n The length of the given string.
+ * @param openMP enables OpenMP optimization.
+ * @return 0 if no error occurred, -1 or -2 otherwise.
+ */
+int
+divsufsort(const unsigned char *T, int *SA, int n, int openMP);
+
+/**
+ * Constructs the burrows-wheeler transformed string of a given string.
+ * @param T [0..n-1] The input string.
+ * @param U [0..n-1] The output string. (can be T)
+ * @param A [0..n-1] The temporary array. (can be NULL)
+ * @param n The length of the given string.
+ * @param num_indexes The length of secondary indexes array. (can be NULL)
+ * @param indexes The secondary indexes array. (can be NULL)
+ * @param openMP enables OpenMP optimization.
+ * @return The primary index if no error occurred, -1 or -2 otherwise.
+ */
+int
+divbwt(const unsigned char *T, unsigned char *U, int *A, int n, unsigned char * num_indexes, int * indexes, int openMP);
+
+
+#ifdef __cplusplus
+} /* extern "C" */
+#endif /* __cplusplus */
+
+#endif /* _DIVSUFSORT_H */
diff --git a/vendor/github.com/DataDog/zstd/entropy_common.c b/vendor/github.com/DataDog/zstd/entropy_common.c
new file mode 100644
index 0000000..b37a082
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/entropy_common.c
@@ -0,0 +1,221 @@
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+/* *************************************
+*  Dependencies
+***************************************/
+#include "mem.h"
+#include "error_private.h"       /* ERR_*, ERROR */
+#define FSE_STATIC_LINKING_ONLY  /* FSE_MIN_TABLELOG */
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY  /* HUF_TABLELOG_ABSOLUTEMAX */
+#include "huf.h"
+
+
+/*===   Version   ===*/
+unsigned FSE_versionNumber(void) { return FSE_VERSION_NUMBER; }
+
+
+/*===   Error Management   ===*/
+unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) & (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0 += 24;
+                if (ip < iend-5) {
+                    ip += 2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount   += 16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0 += 3;
+                bitStream >>= 2;
+                bitCount += 2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            } else {
+                bitStream >>= 2;
+        }   }
+        {   int const max = (2*threshold-1) - remaining;
+            int count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = bitStream & (threshold-1);
+                bitCount += nbBits-1;
+            } else {
+                count = bitStream & (2*threshold-1);
+                if (count >= threshold) count -= max;
+                bitCount += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= count < 0 ? -count : count;   /* -1 means +1 */
+            normalizedCounter[charnum++] = (short)count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) & (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(corruption_detected);
+    if (bitCount > 32) return ERROR(corruption_detected);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    return ip-istart;
+}
+
+
+/*! HUF_readStats() :
+    Read compact Huffman tree, saved by HUF_writeCTable().
+    `huffWeight` is destination buffer.
+    `rankStats` is assumed to be a table of at least HUF_TABLELOG_MAX U32.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUF_readCTable() and HUF_readDTableX?() .
+*/
+size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    /* memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128) {  /* special header */
+        oSize = iSize - 127;
+        iSize = ((oSize+1)/2);
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        if (oSize >= hwSize) return ERROR(corruption_detected);
+        ip += 1;
+        {   U32 n;
+            for (n=0; n<oSize; n+=2) {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        FSE_DTable fseWorkspace[FSE_DTABLE_SIZE_U32(6)];  /* 6 is max possible tableLog for HUF header (maybe even 5, to be tested) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress_wksp(huffWeight, hwSize-1, ip+1, iSize, fseWorkspace, 6);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BIT_highbit32(weightTotal) + 1;
+        if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BIT_highbit32(rest);
+            U32 const lastWeight = BIT_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
diff --git a/vendor/github.com/DataDog/zstd/error_private.c b/vendor/github.com/DataDog/zstd/error_private.c
new file mode 100644
index 0000000..d004ee6
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/error_private.c
@@ -0,0 +1,48 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* The purpose of this file is to have a single list of error strings embedded in binary */
+
+#include "error_private.h"
+
+const char* ERR_getErrorString(ERR_enum code)
+{
+    static const char* const notErrorCode = "Unspecified error code";
+    switch( code )
+    {
+    case PREFIX(no_error): return "No error detected";
+    case PREFIX(GENERIC):  return "Error (generic)";
+    case PREFIX(prefix_unknown): return "Unknown frame descriptor";
+    case PREFIX(version_unsupported): return "Version not supported";
+    case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
+    case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
+    case PREFIX(corruption_detected): return "Corrupted block detected";
+    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(parameter_unsupported): return "Unsupported parameter";
+    case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
+    case PREFIX(init_missing): return "Context should be init first";
+    case PREFIX(memory_allocation): return "Allocation error : not enough memory";
+    case PREFIX(workSpace_tooSmall): return "workSpace buffer is not large enough";
+    case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
+    case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
+    case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
+    case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
+    case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
+    case PREFIX(dictionary_wrong): return "Dictionary mismatch";
+    case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
+    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
+    case PREFIX(srcSize_wrong): return "Src size is incorrect";
+        /* following error codes are not stable and may be removed or changed in a future version */
+    case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
+    case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
+    case PREFIX(maxCode):
+    default: return notErrorCode;
+    }
+}
diff --git a/vendor/github.com/DataDog/zstd/error_private.h b/vendor/github.com/DataDog/zstd/error_private.h
new file mode 100644
index 0000000..0d2fa7e
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/error_private.h
@@ -0,0 +1,76 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* Note : this module is expected to remain private, do not expose it */
+
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>        /* size_t */
+#include "zstd_errors.h"  /* enum list */
+
+
+/* ****************************************
+*  Compiler-specific
+******************************************/
+#if defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-****************************************
+*  Customization (error_public.h)
+******************************************/
+typedef ZSTD_ErrorCode ERR_enum;
+#define PREFIX(name) ZSTD_error_##name
+
+
+/*-****************************************
+*  Error codes handling
+******************************************/
+#undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
+#define ERROR(name) ZSTD_ERROR(name)
+#define ZSTD_ERROR(name) ((size_t)-PREFIX(name))
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC ERR_enum ERR_getErrorCode(size_t code) { if (!ERR_isError(code)) return (ERR_enum)0; return (ERR_enum) (0-code); }
+
+
+/*-****************************************
+*  Error Strings
+******************************************/
+
+const char* ERR_getErrorString(ERR_enum code);   /* error_private.c */
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    return ERR_getErrorString(ERR_getErrorCode(code));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
diff --git a/vendor/github.com/DataDog/zstd/errors.go b/vendor/github.com/DataDog/zstd/errors.go
new file mode 100644
index 0000000..38db0d5
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/errors.go
@@ -0,0 +1,35 @@
+package zstd
+
+/*
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+*/
+import "C"
+
+// ErrorCode is an error returned by the zstd library.
+type ErrorCode int
+
+// Error returns the error string given by zstd
+func (e ErrorCode) Error() string {
+	return C.GoString(C.ZSTD_getErrorName(C.size_t(e)))
+}
+
+func cIsError(code int) bool {
+	return int(C.ZSTD_isError(C.size_t(code))) != 0
+}
+
+// getError returns an error for the return code, or nil if it's not an error
+func getError(code int) error {
+	if code < 0 && cIsError(code) {
+		return ErrorCode(code)
+	}
+	return nil
+}
+
+// IsDstSizeTooSmallError returns whether the error correspond to zstd standard sDstSizeTooSmall error
+func IsDstSizeTooSmallError(e error) bool {
+	if e != nil && e.Error() == "Destination buffer is too small" {
+		return true
+	}
+	return false
+}
diff --git a/vendor/github.com/DataDog/zstd/fse.h b/vendor/github.com/DataDog/zstd/fse.h
new file mode 100644
index 0000000..6a1d272
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/fse.h
@@ -0,0 +1,704 @@
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef FSE_H
+#define FSE_H
+
+
+/*-*****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/*-*****************************************
+*  FSE_PUBLIC_API : control library symbols visibility
+******************************************/
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define FSE_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define FSE_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define FSE_PUBLIC_API
+#endif
+
+/*------   Version   ------*/
+#define FSE_VERSION_MAJOR    0
+#define FSE_VERSION_MINOR    9
+#define FSE_VERSION_RELEASE  0
+
+#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE
+#define FSE_QUOTE(str) #str
+#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)
+#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)
+
+#define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
+FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+/*! FSE_compress() :
+    Compress content of buffer 'src', of size 'srcSize', into destination buffer 'dst'.
+    'dst' buffer must be already allocated. Compression runs faster is dstCapacity >= FSE_compressBound(srcSize).
+    @return : size of compressed data (<= dstCapacity).
+    Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
+                     if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression instead.
+                     if FSE_isError(return), compression failed (more details using FSE_getErrorName())
+*/
+FSE_PUBLIC_API size_t FSE_compress(void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
+
+/*! FSE_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSE_isError() .
+
+    ** Important ** : FSE_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+FSE_PUBLIC_API size_t FSE_decompress(void* dst,  size_t dstCapacity,
+                               const void* cSrc, size_t cSrcSize);
+
+
+/*-*****************************************
+*  Tool functions
+******************************************/
+FSE_PUBLIC_API size_t FSE_compressBound(size_t size);       /* maximum compressed size */
+
+/* Error Management */
+FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+/*-*****************************************
+*  FSE advanced functions
+******************************************/
+/*! FSE_compress2() :
+    Same as FSE_compress(), but allows the selection of 'maxSymbolValue' and 'tableLog'
+    Both parameters can be defined as '0' to mean : use default value
+    @return : size of compressed data
+    Special values : if return == 0, srcData is not compressible => Nothing is stored within cSrc !!!
+                     if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression.
+                     if FSE_isError(return), it's an error code.
+*/
+FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSE_compress() does the following:
+1. count symbol occurrence from source[] into table count[]
+2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
+3. save normalized counters to memory buffer using writeNCount()
+4. build encoding table 'CTable' from normalized counters
+5. encode the data stream using encoding table 'CTable'
+
+FSE_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+/* *** COMPRESSION *** */
+
+/*! FSE_count():
+    Provides the precise count of each byte within a table 'count'.
+    'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
+    *maxSymbolValuePtr will be updated if detected smaller than initial value.
+    @return : the count of the most frequent symbol (which is not identified).
+              if return == srcSize, there is only one symbol.
+              Can also return an error code, which can be tested with FSE_isError(). */
+FSE_PUBLIC_API size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+
+/*! FSE_optimalTableLog():
+    dynamically downsize 'tableLog' when conditions are met.
+    It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
+    @return : recommended tableLog (necessarily <= 'maxTableLog') */
+FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
+
+/*! FSE_normalizeCount():
+    normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
+    'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
+    @return : tableLog,
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
+
+/*! FSE_NCountWriteBound():
+    Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
+    Typically useful for allocation purpose. */
+FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_writeNCount():
+    Compactly save 'normalizedCounter' into 'buffer'.
+    @return : size of the compressed table,
+              or an errorCode, which can be tested using FSE_isError(). */
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+
+/*! Constructor and Destructor of FSE_CTable.
+    Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
+FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API void        FSE_freeCTable (FSE_CTable* ct);
+
+/*! FSE_buildCTable():
+    Builds `ct`, which must be already allocated, using FSE_createCTable().
+    @return : 0, or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_compress_usingCTable():
+    Compress `src` using `ct` into `dst` which must be already allocated.
+    @return : size of compressed data (<= `dstCapacity`),
+              or 0 if compressed data could not fit into `dst`,
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
+
+/*!
+Tutorial :
+----------
+The first step is to count all symbols. FSE_count() does this job very fast.
+Result will be saved into 'count', a table of unsigned int, which must be already allocated, and have 'maxSymbolValuePtr[0]+1' cells.
+'src' is a table of bytes of size 'srcSize'. All values within 'src' MUST be <= maxSymbolValuePtr[0]
+maxSymbolValuePtr[0] will be updated, with its real value (necessarily <= original value)
+FSE_count() will return the number of occurrence of the most frequent symbol.
+This can be used to know if there is a single symbol within 'src', and to quickly evaluate its compressibility.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
+
+The next step is to normalize the frequencies.
+FSE_normalizeCount() will ensure that sum of frequencies is == 2 ^'tableLog'.
+It also guarantees a minimum of 1 to any Symbol with frequency >= 1.
+You can use 'tableLog'==0 to mean "use default tableLog value".
+If you are unsure of which tableLog value to use, you can ask FSE_optimalTableLog(),
+which will provide the optimal valid tableLog given sourceSize, maxSymbolValue, and a user-defined maximum (0 means "default").
+
+The result of FSE_normalizeCount() will be saved into a table,
+called 'normalizedCounter', which is a table of signed short.
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValue+1' cells.
+The return value is tableLog if everything proceeded as expected.
+It is 0 if there is a single symbol within distribution.
+If there is an error (ex: invalid tableLog value), the function will return an ErrorCode (which can be tested using FSE_isError()).
+
+'normalizedCounter' can be saved in a compact manner to a memory area using FSE_writeNCount().
+'buffer' must be already allocated.
+For guaranteed success, buffer size must be at least FSE_headerBound().
+The result of the function is the number of bytes written into 'buffer'.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError(); ex : buffer size too small).
+
+'normalizedCounter' can then be used to create the compression table 'CTable'.
+The space required by 'CTable' must be already allocated, using FSE_createCTable().
+You can then use FSE_buildCTable() to fill 'CTable'.
+If there is an error, both functions will return an ErrorCode (which can be tested using FSE_isError()).
+
+'CTable' can then be used to compress 'src', with FSE_compress_usingCTable().
+Similar to FSE_count(), the convention is that 'src' is assumed to be a table of char of size 'srcSize'
+The function returns the size of compressed data (without header), necessarily <= `dstCapacity`.
+If it returns '0', compressed data could not fit into 'dst'.
+If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSE_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSE_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSE_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSE_PUBLIC_API FSE_DTable* FSE_createDTable(unsigned tableLog);
+FSE_PUBLIC_API void        FSE_freeDTable(FSE_DTable* dt);
+
+/*! FSE_buildDTable():
+    Builds 'dt', which must be already allocated, using FSE_createDTable().
+    return : 0, or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSE_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSE_isError() */
+FSE_PUBLIC_API size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
+This is performed by the function FSE_buildDTable().
+The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+`FSE_DTable` can then be used to decompress `cSrc`, with FSE_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
+*/
+
+#endif  /* FSE_H */
+
+#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)
+#define FSE_H_FSE_STATIC_LINKING_ONLY
+
+/* *** Dependency *** */
+#include "bitstream.h"
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
+#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue)   (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
+#define FSE_DTABLE_SIZE(maxTableLog)                   (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+/* FSE_count_wksp() :
+ * Same as FSE_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= `1024` unsigned
+ */
+size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* source, size_t sourceSize, unsigned* workSpace);
+
+/** FSE_countFast() :
+ *  same as FSE_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr
+ */
+size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+
+/* FSE_countFast_wksp() :
+ * Same as FSE_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` must be a table of minimum `1024` unsigned
+ */
+size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* workSpace);
+
+/*! FSE_count_simple() :
+ * Same as FSE_countFast(), but does not use any additional memory (not even on stack).
+ * This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr` (presuming it's also the size of `count`).
+*/
+size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+
+
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
+/**< same as FSE_optimalTableLog(), which used `minus==2` */
+
+/* FSE_compress_wksp() :
+ * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
+ * FSE_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
+ */
+#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
+size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+
+size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
+/**< build a fake FSE_CTable, designed for a flat distribution, where each symbol uses nbBits */
+
+size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
+/**< build a fake FSE_CTable, designed to compress always the same symbolValue */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * `wkspSize` must be >= `(1<<tableLog)`.
+ */
+size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+
+size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
+
+size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/**< build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */
+
+typedef enum {
+   FSE_repeat_none,  /**< Cannot use the previous table */
+   FSE_repeat_check, /**< Can use the previous table but it must be checked */
+   FSE_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+ } FSE_repeat;
+
+/* *****************************************
+*  FSE symbol compression API
+*******************************************/
+/*!
+   This API consists of small unitary functions, which highly benefit from being inlined.
+   Hence their body are included in next section.
+*/
+typedef struct {
+    ptrdiff_t   value;
+    const void* stateTable;
+    const void* symbolTT;
+    unsigned    stateLog;
+} FSE_CState_t;
+
+static void FSE_initCState(FSE_CState_t* CStatePtr, const FSE_CTable* ct);
+
+static void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* CStatePtr, unsigned symbol);
+
+static void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* CStatePtr);
+
+/**<
+These functions are inner components of FSE_compress_usingCTable().
+They allow the creation of custom streams, mixing multiple tables and bit sources.
+
+A key property to keep in mind is that encoding and decoding are done **in reverse direction**.
+So the first symbol you will encode is the last you will decode, like a LIFO stack.
+
+You will need a few variables to track your CStream. They are :
+
+FSE_CTable    ct;         // Provided by FSE_buildCTable()
+BIT_CStream_t bitStream;  // bitStream tracking structure
+FSE_CState_t  state;      // State tracking structure (can have several)
+
+
+The first thing to do is to init bitStream and state.
+    size_t errorCode = BIT_initCStream(&bitStream, dstBuffer, maxDstSize);
+    FSE_initCState(&state, ct);
+
+Note that BIT_initCStream() can produce an error code, so its result should be tested, using FSE_isError();
+You can then encode your input data, byte after byte.
+FSE_encodeSymbol() outputs a maximum of 'tableLog' bits at a time.
+Remember decoding will be done in reverse direction.
+    FSE_encodeByte(&bitStream, &state, symbol);
+
+At any time, you can also add any bit sequence.
+Note : maximum allowed nbBits is 25, for compatibility with 32-bits decoders
+    BIT_addBits(&bitStream, bitField, nbBits);
+
+The above methods don't commit data to memory, they just store it into local register, for speed.
+Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).
+Writing data to memory is a manual operation, performed by the flushBits function.
+    BIT_flushBits(&bitStream);
+
+Your last FSE encoding operation shall be to flush your last state value(s).
+    FSE_flushState(&bitStream, &state);
+
+Finally, you must close the bitStream.
+The function returns the size of CStream in bytes.
+If data couldn't fit into dstBuffer, it will return a 0 ( == not compressible)
+If there is an error, it returns an errorCode (which can be tested using FSE_isError()).
+    size_t size = BIT_closeCStream(&bitStream);
+*/
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct {
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+/**<
+Let's now decompose FSE_decompress_usingDTable() into its unitary components.
+You will decode FSE-encoded symbols from the bitStream,
+and also any other bitFields you put in, **in reverse order**.
+
+You will need a few variables to track your bitStream. They are :
+
+BIT_DStream_t DStream;    // Stream context
+FSE_DState_t  DState;     // State context. Multiple ones are possible
+FSE_DTable*   DTablePtr;  // Decoding table, provided by FSE_buildDTable()
+
+The first thing to do is to init the bitStream.
+    errorCode = BIT_initDStream(&DStream, srcBuffer, srcSize);
+
+You should then retrieve your initial state(s)
+(in reverse flushing order if you have several ones) :
+    errorCode = FSE_initDState(&DState, &DStream, DTablePtr);
+
+You can then decode your data, symbol after symbol.
+For information the maximum number of bits read by FSE_decodeSymbol() is 'tableLog'.
+Keep in mind that symbols are decoded in reverse order, like a LIFO stack (last in, first out).
+    unsigned char symbol = FSE_decodeSymbol(&DState, &DStream);
+
+You can retrieve any bitfield you eventually stored into the bitStream (in reverse order)
+Note : maximum allowed nbBits is 25, for 32-bits compatibility
+    size_t bitField = BIT_readBits(&DStream, nbBits);
+
+All above operations only read from local register (which size depends on size_t).
+Refueling the register from memory is manually performed by the reload method.
+    endSignal = FSE_reloadDStream(&DStream);
+
+BIT_reloadDStream() result tells if there is still some more data to read from DStream.
+BIT_DStream_unfinished : there is still some data left into the DStream.
+BIT_DStream_endOfBuffer : Dstream reached end of buffer. Its container may no longer be completely filled.
+BIT_DStream_completed : Dstream reached its exact end, corresponding in general to decompression completed.
+BIT_DStream_tooFar : Dstream went too far. Decompression result is corrupted.
+
+When reaching end of buffer (BIT_DStream_endOfBuffer), progress slowly, notably if you decode multiple symbols per loop,
+to properly detect the exact end of stream.
+After each decoded symbol, check if DStream is fully consumed using this simple test :
+    BIT_reloadDStream(&DStream) >= BIT_DStream_completed
+
+When it's done, verify decompression is fully completed, by checking both DStream and the relevant states.
+Checking if DStream has reached its end is performed by :
+    BIT_endOfDStream(&DStream);
+Check also the states. There might be some symbols left there, if some high probability ones (>50%) are possible.
+    FSE_endOfDState(&DState);
+*/
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+typedef struct {
+    int deltaFindState;
+    U32 deltaNbBits;
+} FSE_symbolCompressionTransform; /* total 8 bytes */
+
+MEM_STATIC void FSE_initCState(FSE_CState_t* statePtr, const FSE_CTable* ct)
+{
+    const void* ptr = ct;
+    const U16* u16ptr = (const U16*) ptr;
+    const U32 tableLog = MEM_read16(ptr);
+    statePtr->value = (ptrdiff_t)1<<tableLog;
+    statePtr->stateTable = u16ptr+2;
+    statePtr->symbolTT = ((const U32*)ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1));
+    statePtr->stateLog = tableLog;
+}
+
+
+/*! FSE_initCState2() :
+*   Same as FSE_initCState(), but the first symbol to include (which will be the last to be read)
+*   uses the smallest state value possible, saving the cost of this symbol */
+MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U32 symbol)
+{
+    FSE_initCState(statePtr, ct);
+    {   const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+        const U16* stateTable = (const U16*)(statePtr->stateTable);
+        U32 nbBitsOut  = (U32)((symbolTT.deltaNbBits + (1<<15)) >> 16);
+        statePtr->value = (nbBitsOut << 16) - symbolTT.deltaNbBits;
+        statePtr->value = stateTable[(statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
+    }
+}
+
+MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, U32 symbol)
+{
+    FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+    const U16* const stateTable = (const U16*)(statePtr->stateTable);
+    U32 const nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
+    BIT_addBits(bitC, statePtr->value, nbBitsOut);
+    statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
+}
+
+MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePtr)
+{
+    BIT_addBits(bitC, statePtr->value, statePtr->stateLog);
+    BIT_flushBits(bitC);
+}
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_peekSymbol(const FSE_DState_t* DStatePtr)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSE_updateState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSE_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#ifndef FSE_MAX_MEMORY_USAGE
+#  define FSE_MAX_MEMORY_USAGE 14
+#endif
+#ifndef FSE_DEFAULT_MEMORY_USAGE
+#  define FSE_DEFAULT_MEMORY_USAGE 13
+#endif
+
+/*!FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#ifndef FSE_MAX_SYMBOL_VALUE
+#  define FSE_MAX_SYMBOL_VALUE 255
+#endif
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#  error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSE_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#endif /* FSE_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/vendor/github.com/DataDog/zstd/fse_compress.c b/vendor/github.com/DataDog/zstd/fse_compress.c
new file mode 100644
index 0000000..cb8f1fa
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/fse_compress.c
@@ -0,0 +1,849 @@
+/* ******************************************************************
+   FSE : Finite State Entropy encoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+#include "bitstream.h"
+#include "compiler.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#include "error_private.h"
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_isError ERR_isError
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
+ * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
+ */
+size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    U32 const tableSize = 1 << tableLog;
+    U32 const tableMask = tableSize - 1;
+    void* const ptr = ct;
+    U16* const tableU16 = ( (U16*) ptr) + 2;
+    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
+    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
+    U32 const step = FSE_TABLESTEP(tableSize);
+    U32 cumul[FSE_MAX_SYMBOL_VALUE+2];
+
+    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)workSpace;
+    U32 highThreshold = tableSize-1;
+
+    /* CTable header */
+    if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
+    tableU16[-2] = (U16) tableLog;
+    tableU16[-1] = (U16) maxSymbolValue;
+
+    /* For explanations on how to distribute symbol values over the table :
+    *  http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+    /* symbol start positions */
+    {   U32 u;
+        cumul[0] = 0;
+        for (u=1; u<=maxSymbolValue+1; u++) {
+            if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
+                cumul[u] = cumul[u-1] + 1;
+                tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
+            } else {
+                cumul[u] = cumul[u-1] + normalizedCounter[u-1];
+        }   }
+        cumul[maxSymbolValue+1] = tableSize+1;
+    }
+
+    /* Spread symbols */
+    {   U32 position = 0;
+        U32 symbol;
+        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+            int nbOccurences;
+            for (nbOccurences=0; nbOccurences<normalizedCounter[symbol]; nbOccurences++) {
+                tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* Low proba area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* Must have gone through all positions */
+    }
+
+    /* Build table */
+    {   U32 u; for (u=0; u<tableSize; u++) {
+        FSE_FUNCTION_TYPE s = tableSymbol[u];   /* note : static analyzer may not understand tableSymbol is properly initialized */
+        tableU16[cumul[s]++] = (U16) (tableSize+u);   /* TableU16 : sorted by symbol order; gives next state value */
+    }   }
+
+    /* Build Symbol Transformation Table */
+    {   unsigned total = 0;
+        unsigned s;
+        for (s=0; s<=maxSymbolValue; s++) {
+            switch (normalizedCounter[s])
+            {
+            case  0: break;
+
+            case -1:
+            case  1:
+                symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
+                symbolTT[s].deltaFindState = total - 1;
+                total ++;
+                break;
+            default :
+                {
+                    U32 const maxBitsOut = tableLog - BIT_highbit32 (normalizedCounter[s]-1);
+                    U32 const minStatePlus = normalizedCounter[s] << maxBitsOut;
+                    symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
+                    symbolTT[s].deltaFindState = total - normalizedCounter[s];
+                    total +=  normalizedCounter[s];
+    }   }   }   }
+
+    return 0;
+}
+
+
+size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE];   /* memset() is not necessary, even if static analyzer complain about it */
+    return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
+{
+    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
+    return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
+}
+
+static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                                       const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                                       unsigned writeIsSafe)
+{
+    BYTE* const ostart = (BYTE*) header;
+    BYTE* out = ostart;
+    BYTE* const oend = ostart + headerBufferSize;
+    int nbBits;
+    const int tableSize = 1 << tableLog;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    bitStream = 0;
+    bitCount  = 0;
+    /* Table Size */
+    bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
+    bitCount  += 4;
+
+    /* Init */
+    remaining = tableSize+1;   /* +1 for extra accuracy */
+    threshold = tableSize;
+    nbBits = tableLog+1;
+
+    while (remaining>1) {  /* stops at 1 */
+        if (previous0) {
+            unsigned start = charnum;
+            while (!normalizedCounter[charnum]) charnum++;
+            while (charnum >= start+24) {
+                start+=24;
+                bitStream += 0xFFFFU << bitCount;
+                if ((!writeIsSafe) && (out > oend-2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE) bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out+=2;
+                bitStream>>=16;
+            }
+            while (charnum >= start+3) {
+                start+=3;
+                bitStream += 3 << bitCount;
+                bitCount += 2;
+            }
+            bitStream += (charnum-start) << bitCount;
+            bitCount += 2;
+            if (bitCount>16) {
+                if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE)bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out += 2;
+                bitStream >>= 16;
+                bitCount -= 16;
+        }   }
+        {   int count = normalizedCounter[charnum++];
+            int const max = (2*threshold-1)-remaining;
+            remaining -= count < 0 ? -count : count;
+            count++;   /* +1 for extra accuracy */
+            if (count>=threshold) count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            bitStream += count << bitCount;
+            bitCount  += nbBits;
+            bitCount  -= (count<max);
+            previous0  = (count==1);
+            if (remaining<1) return ERROR(GENERIC);
+            while (remaining<threshold) { nbBits--; threshold>>=1; }
+        }
+        if (bitCount>16) {
+            if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            out[0] = (BYTE)bitStream;
+            out[1] = (BYTE)(bitStream>>8);
+            out += 2;
+            bitStream >>= 16;
+            bitCount -= 16;
+    }   }
+
+    /* flush remaining bitStream */
+    if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    out[0] = (BYTE)bitStream;
+    out[1] = (BYTE)(bitStream>>8);
+    out+= (bitCount+7) /8;
+
+    if (charnum > maxSymbolValue + 1) return ERROR(GENERIC);
+
+    return (out-ostart);
+}
+
+
+size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
+
+    if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
+        return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
+
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1);
+}
+
+
+
+/*-**************************************************************
+*  Counting histogram
+****************************************************************/
+/*! FSE_count_simple
+    This function counts byte values within `src`, and store the histogram into table `count`.
+    It doesn't use any additional memory.
+    But this function is unsafe : it doesn't check that all values within `src` can fit into `count`.
+    For this reason, prefer using a table `count` with 256 elements.
+    @return : count of most numerous element.
+*/
+size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                        const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned max=0;
+
+    memset(count, 0, (maxSymbolValue+1)*sizeof(*count));
+    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
+
+    while (ip<end) {
+        assert(*ip <= maxSymbolValue);
+        count[*ip++]++;
+    }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+
+    { U32 s; for (s=0; s<=maxSymbolValue; s++) if (count[s] > max) max = count[s]; }
+
+    return (size_t)max;
+}
+
+
+/* FSE_count_parallel_wksp() :
+ * Same as FSE_count_parallel(), but using an externally provided scratch buffer.
+ * `workSpace` size must be a minimum of `1024 * sizeof(unsigned)`.
+ * @return : largest histogram frequency, or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
+static size_t FSE_count_parallel_wksp(
+                                unsigned* count, unsigned* maxSymbolValuePtr,
+                                const void* source, size_t sourceSize,
+                                unsigned checkMax, unsigned* const workSpace)
+{
+    const BYTE* ip = (const BYTE*)source;
+    const BYTE* const iend = ip+sourceSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned max=0;
+    U32* const Counting1 = workSpace;
+    U32* const Counting2 = Counting1 + 256;
+    U32* const Counting3 = Counting2 + 256;
+    U32* const Counting4 = Counting3 + 256;
+
+    memset(workSpace, 0, 4*256*sizeof(unsigned));
+
+    /* safety checks */
+    if (!sourceSize) {
+        memset(count, 0, maxSymbolValue + 1);
+        *maxSymbolValuePtr = 0;
+        return 0;
+    }
+    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
+
+    /* by stripes of 16 bytes */
+    {   U32 cached = MEM_read32(ip); ip += 4;
+        while (ip < iend-15) {
+            U32 c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+        }
+        ip-=4;
+    }
+
+    /* finish last symbols */
+    while (ip<iend) Counting1[*ip++]++;
+
+    if (checkMax) {   /* verify stats will fit into destination table */
+        U32 s; for (s=255; s>maxSymbolValue; s--) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
+    }   }
+
+    {   U32 s;
+        if (maxSymbolValue > 255) maxSymbolValue = 255;
+        for (s=0; s<=maxSymbolValue; s++) {
+            count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
+            if (count[s] > max) max = count[s];
+    }   }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+    return (size_t)max;
+}
+
+/* FSE_countFast_wksp() :
+ * Same as FSE_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= `1024` unsigned */
+size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                          const void* source, size_t sourceSize,
+                          unsigned* workSpace)
+{
+    if (sourceSize < 1500) /* heuristic threshold */
+        return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 0, workSpace);
+}
+
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[1024];
+    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters);
+}
+
+/* FSE_count_wksp() :
+ * Same as FSE_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= `1024` unsigned */
+size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* source, size_t sourceSize, unsigned* workSpace)
+{
+    if (*maxSymbolValuePtr < 255)
+        return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 1, workSpace);
+    *maxSymbolValuePtr = 255;
+    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace);
+}
+
+size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* src, size_t srcSize)
+{
+    unsigned tmpCounters[1024];
+    return FSE_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters);
+}
+
+
+
+/*-**************************************************************
+*  FSE Compression Code
+****************************************************************/
+/*! FSE_sizeof_CTable() :
+    FSE_CTable is a variable size structure which contains :
+    `U16 tableLog;`
+    `U16 maxSymbolValue;`
+    `U16 nextStateNumber[1 << tableLog];`                         // This size is variable
+    `FSE_symbolCompressionTransform symbolTT[maxSymbolValue+1];`  // This size is variable
+Allocation is manual (C standard does not support variable-size structures).
+*/
+size_t FSE_sizeof_CTable (unsigned maxSymbolValue, unsigned tableLog)
+{
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    return FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
+}
+
+FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
+{
+    size_t size;
+    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
+    size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
+    return (FSE_CTable*)malloc(size);
+}
+
+void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
+
+/* provides the minimum logSize to safely represent a distribution */
+static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
+{
+    U32 minBitsSrc = BIT_highbit32((U32)(srcSize - 1)) + 1;
+    U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
+    U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    return minBits;
+}
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
+{
+    U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
+    U32 tableLog = maxTableLog;
+    U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    assert(srcSize > 1); /* Not supported, RLE should be used instead */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
+    if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
+    if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
+    if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
+    return tableLog;
+}
+
+unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+{
+    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
+}
+
+
+/* Secondary normalization method.
+   To be used when primary method fails. */
+
+static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue)
+{
+    short const NOT_YET_ASSIGNED = -2;
+    U32 s;
+    U32 distributed = 0;
+    U32 ToDistribute;
+
+    /* Init */
+    U32 const lowThreshold = (U32)(total >> tableLog);
+    U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
+
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (count[s] == 0) {
+            norm[s]=0;
+            continue;
+        }
+        if (count[s] <= lowThreshold) {
+            norm[s] = -1;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+        if (count[s] <= lowOne) {
+            norm[s] = 1;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+
+        norm[s]=NOT_YET_ASSIGNED;
+    }
+    ToDistribute = (1 << tableLog) - distributed;
+
+    if ((total / ToDistribute) > lowOne) {
+        /* risk of rounding to zero */
+        lowOne = (U32)((total * 3) / (ToDistribute * 2));
+        for (s=0; s<=maxSymbolValue; s++) {
+            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
+                norm[s] = 1;
+                distributed++;
+                total -= count[s];
+                continue;
+        }   }
+        ToDistribute = (1 << tableLog) - distributed;
+    }
+
+    if (distributed == maxSymbolValue+1) {
+        /* all values are pretty poor;
+           probably incompressible data (should have already been detected);
+           find max, then give all remaining points to max */
+        U32 maxV = 0, maxC = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > maxC) { maxV=s; maxC=count[s]; }
+        norm[maxV] += (short)ToDistribute;
+        return 0;
+    }
+
+    if (total == 0) {
+        /* all of the symbols were low enough for the lowOne or lowThreshold */
+        for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
+            if (norm[s] > 0) { ToDistribute--; norm[s]++; }
+        return 0;
+    }
+
+    {   U64 const vStepLog = 62 - tableLog;
+        U64 const mid = (1ULL << (vStepLog-1)) - 1;
+        U64 const rStep = ((((U64)1<<vStepLog) * ToDistribute) + mid) / total;   /* scale on remaining */
+        U64 tmpTotal = mid;
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (norm[s]==NOT_YET_ASSIGNED) {
+                U64 const end = tmpTotal + (count[s] * rStep);
+                U32 const sStart = (U32)(tmpTotal >> vStepLog);
+                U32 const sEnd = (U32)(end >> vStepLog);
+                U32 const weight = sEnd - sStart;
+                if (weight < 1)
+                    return ERROR(GENERIC);
+                norm[s] = (short)weight;
+                tmpTotal = end;
+    }   }   }
+
+    return 0;
+}
+
+
+size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
+                           const unsigned* count, size_t total,
+                           unsigned maxSymbolValue)
+{
+    /* Sanity checks */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported size */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
+    if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
+
+    {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+        U64 const scale = 62 - tableLog;
+        U64 const step = ((U64)1<<62) / total;   /* <== here, one division ! */
+        U64 const vStep = 1ULL<<(scale-20);
+        int stillToDistribute = 1<<tableLog;
+        unsigned s;
+        unsigned largest=0;
+        short largestP=0;
+        U32 lowThreshold = (U32)(total >> tableLog);
+
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (count[s] == total) return 0;   /* rle special case */
+            if (count[s] == 0) { normalizedCounter[s]=0; continue; }
+            if (count[s] <= lowThreshold) {
+                normalizedCounter[s] = -1;
+                stillToDistribute--;
+            } else {
+                short proba = (short)((count[s]*step) >> scale);
+                if (proba<8) {
+                    U64 restToBeat = vStep * rtbTable[proba];
+                    proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
+                }
+                if (proba > largestP) { largestP=proba; largest=s; }
+                normalizedCounter[s] = proba;
+                stillToDistribute -= proba;
+        }   }
+        if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
+            /* corner case, need another normalization method */
+            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
+            if (FSE_isError(errorCode)) return errorCode;
+        }
+        else normalizedCounter[largest] += (short)stillToDistribute;
+    }
+
+#if 0
+    {   /* Print Table (debug) */
+        U32 s;
+        U32 nTotal = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            printf("%3i: %4i \n", s, normalizedCounter[s]);
+        for (s=0; s<=maxSymbolValue; s++)
+            nTotal += abs(normalizedCounter[s]);
+        if (nTotal != (1U<<tableLog))
+            printf("Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
+        getchar();
+    }
+#endif
+
+    return tableLog;
+}
+
+
+/* fake FSE_CTable, for raw (uncompressed) input */
+size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
+{
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    void* const ptr = ct;
+    U16* const tableU16 = ( (U16*) ptr) + 2;
+    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableSize>>1);   /* assumption : tableLog >= 1 */
+    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);             /* min size */
+
+    /* header */
+    tableU16[-2] = (U16) nbBits;
+    tableU16[-1] = (U16) maxSymbolValue;
+
+    /* Build table */
+    for (s=0; s<tableSize; s++)
+        tableU16[s] = (U16)(tableSize + s);
+
+    /* Build Symbol Transformation Table */
+    {   const U32 deltaNbBits = (nbBits << 16) - (1 << nbBits);
+        for (s=0; s<=maxSymbolValue; s++) {
+            symbolTT[s].deltaNbBits = deltaNbBits;
+            symbolTT[s].deltaFindState = s-1;
+    }   }
+
+    return 0;
+}
+
+/* fake FSE_CTable, for rle input (always same symbol) */
+size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
+{
+    void* ptr = ct;
+    U16* tableU16 = ( (U16*) ptr) + 2;
+    void* FSCTptr = (U32*)ptr + 2;
+    FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
+
+    /* header */
+    tableU16[-2] = (U16) 0;
+    tableU16[-1] = (U16) symbolValue;
+
+    /* Build table */
+    tableU16[0] = 0;
+    tableU16[1] = 0;   /* just in case */
+
+    /* Build Symbol Transformation Table */
+    symbolTT[symbolValue].deltaNbBits = 0;
+    symbolTT[symbolValue].deltaFindState = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct, const unsigned fast)
+{
+    const BYTE* const istart = (const BYTE*) src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip=iend;
+
+    BIT_CStream_t bitC;
+    FSE_CState_t CState1, CState2;
+
+    /* init */
+    if (srcSize <= 2) return 0;
+    { size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
+      if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
+
+#define FSE_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+
+    if (srcSize & 1) {
+        FSE_initCState2(&CState1, ct, *--ip);
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    } else {
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_initCState2(&CState1, ct, *--ip);
+    }
+
+    /* join to mod 4 */
+    srcSize -= 2;
+    if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) {  /* test bit 2 */
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    /* 2 or 4 encoding per loop */
+    while ( ip>istart ) {
+
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 )   /* this test must be static */
+            FSE_FLUSHBITS(&bitC);
+
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) {  /* this test must be static */
+            FSE_encodeSymbol(&bitC, &CState2, *--ip);
+            FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        }
+
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    FSE_flushCState(&bitC, &CState2);
+    FSE_flushCState(&bitC, &CState1);
+    return BIT_closeCStream(&bitC);
+}
+
+size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct)
+{
+    unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
+
+    if (fast)
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
+    else
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
+}
+
+
+size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
+
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
+#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+
+/* FSE_compress_wksp() :
+ * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
+ * `wkspSize` size must be `(1<<tableLog)`.
+ */
+size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstSize;
+
+    U32   count[FSE_MAX_SYMBOL_VALUE+1];
+    S16   norm[FSE_MAX_SYMBOL_VALUE+1];
+    FSE_CTable* CTable = (FSE_CTable*)workSpace;
+    size_t const CTableSize = FSE_CTABLE_SIZE_U32(tableLog, maxSymbolValue);
+    void* scratchBuffer = (void*)(CTable + CTableSize);
+    size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
+
+    /* init conditions */
+    if (wkspSize < FSE_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
+    if (srcSize <= 1) return 0;  /* Not compressible */
+    if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(maxCount, FSE_count_wksp(count, &maxSymbolValue, src, srcSize, (unsigned*)scratchBuffer) );
+        if (maxCount == srcSize) return 1;   /* only a single symbol in src : rle */
+        if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
+        if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
+    }
+
+    tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue) );
+
+    /* Write table description header */
+    {   CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
+        op += nc_err;
+    }
+
+    /* Compress */
+    CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, scratchBufferSize) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, src, srcSize, CTable) );
+        if (cSize == 0) return 0;   /* not enough space for compressed data */
+        op += cSize;
+    }
+
+    /* check compressibility */
+    if ( (size_t)(op-ostart) >= srcSize-1 ) return 0;
+
+    return op-ostart;
+}
+
+typedef struct {
+    FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
+    BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
+} fseWkspMax_t;
+
+size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
+{
+    fseWkspMax_t scratchBuffer;
+    FSE_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
+}
+
+size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
+}
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/vendor/github.com/DataDog/zstd/fse_decompress.c b/vendor/github.com/DataDog/zstd/fse_decompress.c
new file mode 100644
index 0000000..4c66c3b
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/fse_decompress.c
@@ -0,0 +1,309 @@
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include "bitstream.h"
+#include "compiler.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#include "error_private.h"
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_isError ERR_isError
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+/* check and forward error code */
+#define CHECK_F(f) { size_t const e = f; if (FSE_isError(e)) return e; }
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+FSE_DTable* FSE_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
+    return (FSE_DTable*)malloc( FSE_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSE_freeDTable (FSE_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSE_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const cell = (FSE_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE_TEMPLATE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+
+    /* Init */
+    CHECK_F(BIT_initDStream(&bitD, cSrc, cSrcSize));
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) & (op<olimit) ; op+=4) {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+        *op++ = FSE_GETSYMBOL(&state1);
+        if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
+            *op++ = FSE_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+        *op++ = FSE_GETSYMBOL(&state2);
+        if (BIT_reloadDStream(&bitD)==BIT_DStream_overflow) {
+            *op++ = FSE_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSE_DTableHeader* DTableH = (const FSE_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+
+    /* normal FSE decoding mode */
+    size_t const NCountLength = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(NCountLength)) return NCountLength;
+    //if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size; supposed to be already checked in NCountLength, only remaining case : NCountLength==cSrcSize */
+    if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
+    ip += NCountLength;
+    cSrcSize -= NCountLength;
+
+    CHECK_F( FSE_buildDTable (workSpace, counting, maxSymbolValue, tableLog) );
+
+    return FSE_decompress_usingDTable (dst, dstCapacity, ip, cSrcSize, workSpace);   /* always return, even if it is an error code */
+}
+
+
+typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
+{
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, dt, FSE_MAX_TABLELOG);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/vendor/github.com/DataDog/zstd/huf.h b/vendor/github.com/DataDog/zstd/huf.h
new file mode 100644
index 0000000..b4645b4
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/huf.h
@@ -0,0 +1,327 @@
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef HUF_H_298734234
+#define HUF_H_298734234
+
+/* *** Dependencies *** */
+#include <stddef.h>    /* size_t */
+
+
+/* *** library symbols visibility *** */
+/* Note : when linking with -fvisibility=hidden on gcc, or by default on Visual,
+ *        HUF symbols remain "private" (internal symbols for library only).
+ *        Set macro FSE_DLL_EXPORT to 1 if you want HUF symbols visible on DLL interface */
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define HUF_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define HUF_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define HUF_PUBLIC_API __declspec(dllimport)  /* not required, just to generate faster code (saves a function pointer load from IAT and an indirect jump) */
+#else
+#  define HUF_PUBLIC_API
+#endif
+
+
+/* ========================== */
+/* ***  simple functions  *** */
+/* ========================== */
+
+/** HUF_compress() :
+ *  Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
+ * 'dst' buffer must be already allocated.
+ *  Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
+ * `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
+ * @return : size of compressed data (<= `dstCapacity`).
+ *  Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
+ *                   if HUF_isError(return), compression failed (more details using HUF_getErrorName())
+ */
+HUF_PUBLIC_API size_t HUF_compress(void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
+
+/** HUF_decompress() :
+ *  Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+ *  into already allocated buffer 'dst', of minimum size 'dstSize'.
+ * `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
+ *  Note : in contrast with FSE, HUF_decompress can regenerate
+ *         RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+ *         because it knows size to regenerate (originalSize).
+ * @return : size of regenerated data (== originalSize),
+ *           or an error code, which can be tested using HUF_isError()
+ */
+HUF_PUBLIC_API size_t HUF_decompress(void* dst,  size_t originalSize,
+                               const void* cSrc, size_t cSrcSize);
+
+
+/* ***   Tool functions *** */
+#define HUF_BLOCKSIZE_MAX (128 * 1024)                  /**< maximum input size for a single block compressed with HUF_compress */
+HUF_PUBLIC_API size_t HUF_compressBound(size_t size);   /**< maximum compressed size (worst case) */
+
+/* Error Management */
+HUF_PUBLIC_API unsigned    HUF_isError(size_t code);       /**< tells if a return value is an error code */
+HUF_PUBLIC_API const char* HUF_getErrorName(size_t code);  /**< provides error code string (useful for debugging) */
+
+
+/* ***   Advanced function   *** */
+
+/** HUF_compress2() :
+ *  Same as HUF_compress(), but offers control over `maxSymbolValue` and `tableLog`.
+ * `maxSymbolValue` must be <= HUF_SYMBOLVALUE_MAX .
+ * `tableLog` must be `<= HUF_TABLELOG_MAX` . */
+HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               unsigned maxSymbolValue, unsigned tableLog);
+
+/** HUF_compress4X_wksp() :
+ *  Same as HUF_compress2(), but uses externally allocated `workSpace`.
+ * `workspace` must have minimum alignment of 4, and be at least as large as HUF_WORKSPACE_SIZE */
+#define HUF_WORKSPACE_SIZE (6 << 10)
+#define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
+HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     unsigned maxSymbolValue, unsigned tableLog,
+                                     void* workSpace, size_t wkspSize);
+
+#endif   /* HUF_H_298734234 */
+
+/* ******************************************************************
+ *  WARNING !!
+ *  The following section contains advanced and experimental definitions
+ *  which shall never be used in the context of a dynamic library,
+ *  because they are not guaranteed to remain stable in the future.
+ *  Only consider them in association with static linking.
+ * *****************************************************************/
+#if defined(HUF_STATIC_LINKING_ONLY) && !defined(HUF_H_HUF_STATIC_LINKING_ONLY)
+#define HUF_H_HUF_STATIC_LINKING_ONLY
+
+/* *** Dependencies *** */
+#include "mem.h"   /* U32 */
+
+
+/* *** Constants *** */
+#define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_TABLELOG_DEFAULT  11      /* default tableLog value when none specified */
+#define HUF_SYMBOLVALUE_MAX  255
+
+#define HUF_TABLELOG_ABSOLUTEMAX  15  /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
+#  error "HUF_TABLELOG_MAX is too large !"
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true when incompressible is pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of HUF's Compression Table */
+#define HUF_CTABLE_SIZE_U32(maxSymbolValue)   ((maxSymbolValue)+1)   /* Use tables of U32, for proper alignment */
+#define HUF_CTABLE_SIZE(maxSymbolValue)       (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
+#define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
+    U32 name##hb[HUF_CTABLE_SIZE_U32(maxSymbolValue)]; \
+    void* name##hv = &(name##hb); \
+    HUF_CElt* name = (HUF_CElt*)(name##hv)   /* no final ; */
+
+/* static allocation of HUF's DTable */
+typedef U32 HUF_DTable;
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
+size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize); /**< considers RLE and uncompressed as errors */
+size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+
+
+/* ****************************************
+ *  HUF detailed API
+ * ****************************************/
+
+/*! HUF_compress() does the following:
+ *  1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
+ *  2. (optional) refine tableLog using HUF_optimalTableLog()
+ *  3. build Huffman table from count using HUF_buildCTable()
+ *  4. save Huffman table to memory buffer using HUF_writeCTable()
+ *  5. encode the data stream using HUF_compress4X_usingCTable()
+ *
+ *  The following API allows targeting specific sub-functions for advanced tasks.
+ *  For example, it's possible to compress several blocks using the same 'CTable',
+ *  or to save and regenerate 'CTable' using external methods.
+ */
+unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
+typedef struct HUF_CElt_s HUF_CElt;   /* incomplete type */
+size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);   /* @return : maxNbBits; CTable and count can overlap. In which case, CTable will overwrite count content */
+size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+
+typedef enum {
+   HUF_repeat_none,  /**< Cannot use the previous table */
+   HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
+   HUF_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+ } HUF_repeat;
+/** HUF_compress4X_repeat() :
+ *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,    /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
+
+/** HUF_buildCTable_wksp() :
+ *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
+ * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
+ */
+#define HUF_CTABLE_WORKSPACE_SIZE_U32 (2*HUF_SYMBOLVALUE_MAX +1 +1)
+#define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
+size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize);
+
+/*! HUF_readStats() :
+ *  Read compact Huffman tree, saved by HUF_writeCTable().
+ * `huffWeight` is destination buffer.
+ * @return : size read from `src` , or an error Code .
+ *  Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
+size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
+                     U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize);
+
+/** HUF_readCTable() :
+ *  Loading a CTable saved with HUF_writeCTable() */
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+
+
+/*
+ * HUF_decompress() does the following:
+ * 1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
+ * 2. build Huffman table from save, using HUF_readDTableX?()
+ * 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
+ */
+
+/** HUF_selectDecoder() :
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
+ *  Assumption : 0 < dstSize <= 128 KB */
+U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
+
+/**
+ *  The minimum workspace size for the `workSpace` used in
+ *  HUF_readDTableX2_wksp() and HUF_readDTableX4_wksp().
+ *
+ *  The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
+ *  HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
+ *  Buffer overflow errors may potentially occur if code modifications result in
+ *  a required workspace size greater than that specified in the following
+ *  macro.
+ */
+#define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
+#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
+
+size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize);
+size_t HUF_readDTableX2_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
+size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize);
+size_t HUF_readDTableX4_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
+
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+
+
+/* ====================== */
+/* single stream variants */
+/* ====================== */
+
+size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+/** HUF_compress1X_repeat() :
+ *  Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,   /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
+
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+size_t HUF_decompress1X_DCtx_wksp (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);
+size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /**< automatic selection of sing or double symbol decoder, based on DTable */
+size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+
+/* BMI2 variants.
+ * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
+ */
+size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress1X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+
+#endif /* HUF_STATIC_LINKING_ONLY */
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/vendor/github.com/DataDog/zstd/huf_compress.c b/vendor/github.com/DataDog/zstd/huf_compress.c
new file mode 100644
index 0000000..83230b4
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/huf_compress.c
@@ -0,0 +1,788 @@
+/* ******************************************************************
+   Huffman encoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+#include "bitstream.h"
+#include "compiler.h"
+#define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
+#include "fse.h"        /* header compression */
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "error_private.h"
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_isError ERR_isError
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
+#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+
+
+/* **************************************************************
+*  Utils
+****************************************************************/
+unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+{
+    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+}
+
+
+/* *******************************************************
+*  HUF : Huffman block compression
+*********************************************************/
+/* HUF_compressWeights() :
+ * Same as FSE_compress(), but dedicated to huff0's weights compression.
+ * The use case needs much less stack memory.
+ * Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
+ */
+#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
+size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstSize;
+
+    U32 maxSymbolValue = HUF_TABLELOG_MAX;
+    U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
+
+    FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
+    BYTE scratchBuffer[1<<MAX_FSE_TABLELOG_FOR_HUFF_HEADER];
+
+    U32 count[HUF_TABLELOG_MAX+1];
+    S16 norm[HUF_TABLELOG_MAX+1];
+
+    /* init conditions */
+    if (wtSize <= 1) return 0;  /* Not compressible */
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(maxCount, FSE_count_simple(count, &maxSymbolValue, weightTable, wtSize) );
+        if (maxCount == wtSize) return 1;   /* only a single symbol in src : rle */
+        if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
+    }
+
+    tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, wtSize, maxSymbolValue) );
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
+        op += hSize;
+    }
+
+    /* Compress */
+    CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, sizeof(scratchBuffer)) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, weightTable, wtSize, CTable) );
+        if (cSize == 0) return 0;   /* not enough space for compressed data */
+        op += cSize;
+    }
+
+    return op-ostart;
+}
+
+
+struct HUF_CElt_s {
+  U16  val;
+  BYTE nbBits;
+};   /* typedef'd to HUF_CElt within "huf.h" */
+
+/*! HUF_writeCTable() :
+    `CTable` : Huffman tree to save, using huf representation.
+    @return : size of saved CTable */
+size_t HUF_writeCTable (void* dst, size_t maxDstSize,
+                        const HUF_CElt* CTable, U32 maxSymbolValue, U32 huffLog)
+{
+    BYTE bitsToWeight[HUF_TABLELOG_MAX + 1];   /* precomputed conversion table */
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
+    BYTE* op = (BYTE*)dst;
+    U32 n;
+
+     /* check conditions */
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+
+    /* convert to weight */
+    bitsToWeight[0] = 0;
+    for (n=1; n<huffLog+1; n++)
+        bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
+    for (n=0; n<maxSymbolValue; n++)
+        huffWeight[n] = bitsToWeight[CTable[n].nbBits];
+
+    /* attempt weights compression by FSE */
+    {   CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, huffWeight, maxSymbolValue) );
+        if ((hSize>1) & (hSize < maxSymbolValue/2)) {   /* FSE compressed */
+            op[0] = (BYTE)hSize;
+            return hSize+1;
+    }   }
+
+    /* write raw values as 4-bits (max : 15) */
+    if (maxSymbolValue > (256-128)) return ERROR(GENERIC);   /* should not happen : likely means source cannot be compressed */
+    if (((maxSymbolValue+1)/2) + 1 > maxDstSize) return ERROR(dstSize_tooSmall);   /* not enough space within dst buffer */
+    op[0] = (BYTE)(128 /*special case*/ + (maxSymbolValue-1));
+    huffWeight[maxSymbolValue] = 0;   /* to be sure it doesn't cause msan issue in final combination */
+    for (n=0; n<maxSymbolValue; n+=2)
+        op[(n/2)+1] = (BYTE)((huffWeight[n] << 4) + huffWeight[n+1]);
+    return ((maxSymbolValue+1)/2) + 1;
+}
+
+
+size_t HUF_readCTable (HUF_CElt* CTable, U32* maxSymbolValuePtr, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];   /* init not required, even though some static analyzer may complain */
+    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+
+    /* get symbol weights */
+    CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
+
+    /* check result */
+    if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
+
+    /* Prepare base value per rank */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<=tableLog; n++) {
+            U32 current = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = current;
+    }   }
+
+    /* fill nbBits */
+    {   U32 n; for (n=0; n<nbSymbols; n++) {
+            const U32 w = huffWeight[n];
+            CTable[n].nbBits = (BYTE)(tableLog + 1 - w);
+    }   }
+
+    /* fill val */
+    {   U16 nbPerRank[HUF_TABLELOG_MAX+2]  = {0};  /* support w=0=>n=tableLog+1 */
+        U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
+        { U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[CTable[n].nbBits]++; }
+        /* determine stating value per rank */
+        valPerRank[tableLog+1] = 0;   /* for w==0 */
+        {   U16 min = 0;
+            U32 n; for (n=tableLog; n>0; n--) {  /* start at n=tablelog <-> w=1 */
+                valPerRank[n] = min;     /* get starting value within each rank */
+                min += nbPerRank[n];
+                min >>= 1;
+        }   }
+        /* assign value within rank, symbol order */
+        { U32 n; for (n=0; n<nbSymbols; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
+    }
+
+    *maxSymbolValuePtr = nbSymbols - 1;
+    return readSize;
+}
+
+
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
+{
+    const U32 largestBits = huffNode[lastNonNull].nbBits;
+    if (largestBits <= maxNbBits) return largestBits;   /* early exit : no elt > maxNbBits */
+
+    /* there are several too large elements (at least >= 2) */
+    {   int totalCost = 0;
+        const U32 baseCost = 1 << (largestBits - maxNbBits);
+        U32 n = lastNonNull;
+
+        while (huffNode[n].nbBits > maxNbBits) {
+            totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
+            huffNode[n].nbBits = (BYTE)maxNbBits;
+            n --;
+        }  /* n stops at huffNode[n].nbBits <= maxNbBits */
+        while (huffNode[n].nbBits == maxNbBits) n--;   /* n end at index of smallest symbol using < maxNbBits */
+
+        /* renorm totalCost */
+        totalCost >>= (largestBits - maxNbBits);  /* note : totalCost is necessarily a multiple of baseCost */
+
+        /* repay normalized cost */
+        {   U32 const noSymbol = 0xF0F0F0F0;
+            U32 rankLast[HUF_TABLELOG_MAX+2];
+            int pos;
+
+            /* Get pos of last (smallest) symbol per rank */
+            memset(rankLast, 0xF0, sizeof(rankLast));
+            {   U32 currentNbBits = maxNbBits;
+                for (pos=n ; pos >= 0; pos--) {
+                    if (huffNode[pos].nbBits >= currentNbBits) continue;
+                    currentNbBits = huffNode[pos].nbBits;   /* < maxNbBits */
+                    rankLast[maxNbBits-currentNbBits] = pos;
+            }   }
+
+            while (totalCost > 0) {
+                U32 nBitsToDecrease = BIT_highbit32(totalCost) + 1;
+                for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
+                    U32 highPos = rankLast[nBitsToDecrease];
+                    U32 lowPos = rankLast[nBitsToDecrease-1];
+                    if (highPos == noSymbol) continue;
+                    if (lowPos == noSymbol) break;
+                    {   U32 const highTotal = huffNode[highPos].count;
+                        U32 const lowTotal = 2 * huffNode[lowPos].count;
+                        if (highTotal <= lowTotal) break;
+                }   }
+                /* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
+                /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
+                while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))
+                    nBitsToDecrease ++;
+                totalCost -= 1 << (nBitsToDecrease-1);
+                if (rankLast[nBitsToDecrease-1] == noSymbol)
+                    rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];   /* this rank is no longer empty */
+                huffNode[rankLast[nBitsToDecrease]].nbBits ++;
+                if (rankLast[nBitsToDecrease] == 0)    /* special case, reached largest symbol */
+                    rankLast[nBitsToDecrease] = noSymbol;
+                else {
+                    rankLast[nBitsToDecrease]--;
+                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease)
+                        rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
+            }   }   /* while (totalCost > 0) */
+
+            while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
+                if (rankLast[1] == noSymbol) {  /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
+                    while (huffNode[n].nbBits == maxNbBits) n--;
+                    huffNode[n+1].nbBits--;
+                    rankLast[1] = n+1;
+                    totalCost++;
+                    continue;
+                }
+                huffNode[ rankLast[1] + 1 ].nbBits--;
+                rankLast[1]++;
+                totalCost ++;
+    }   }   }   /* there are several too large elements (at least >= 2) */
+
+    return maxNbBits;
+}
+
+
+typedef struct {
+    U32 base;
+    U32 current;
+} rankPos;
+
+static void HUF_sort(nodeElt* huffNode, const U32* count, U32 maxSymbolValue)
+{
+    rankPos rank[32];
+    U32 n;
+
+    memset(rank, 0, sizeof(rank));
+    for (n=0; n<=maxSymbolValue; n++) {
+        U32 r = BIT_highbit32(count[n] + 1);
+        rank[r].base ++;
+    }
+    for (n=30; n>0; n--) rank[n-1].base += rank[n].base;
+    for (n=0; n<32; n++) rank[n].current = rank[n].base;
+    for (n=0; n<=maxSymbolValue; n++) {
+        U32 const c = count[n];
+        U32 const r = BIT_highbit32(c+1) + 1;
+        U32 pos = rank[r].current++;
+        while ((pos > rank[r].base) && (c > huffNode[pos-1].count)) {
+            huffNode[pos] = huffNode[pos-1];
+            pos--;
+        }
+        huffNode[pos].count = c;
+        huffNode[pos].byte  = (BYTE)n;
+    }
+}
+
+
+/** HUF_buildCTable_wksp() :
+ *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
+ *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of HUF_CTABLE_WORKSPACE_SIZE_U32 unsigned.
+ */
+#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
+typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
+size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
+{
+    nodeElt* const huffNode0 = (nodeElt*)workSpace;
+    nodeElt* const huffNode = huffNode0+1;
+    U32 n, nonNullRank;
+    int lowS, lowN;
+    U16 nodeNb = STARTNODE;
+    U32 nodeRoot;
+
+    /* safety checks */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < sizeof(huffNodeTable)) return ERROR(workSpace_tooSmall);
+    if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+    memset(huffNode0, 0, sizeof(huffNodeTable));
+
+    /* sort, decreasing order */
+    HUF_sort(huffNode, count, maxSymbolValue);
+
+    /* init for parents */
+    nonNullRank = maxSymbolValue;
+    while(huffNode[nonNullRank].count == 0) nonNullRank--;
+    lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
+    huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
+    huffNode[lowS].parent = huffNode[lowS-1].parent = nodeNb;
+    nodeNb++; lowS-=2;
+    for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
+    huffNode0[0].count = (U32)(1U<<31);  /* fake entry, strong barrier */
+
+    /* create parents */
+    while (nodeNb <= nodeRoot) {
+        U32 n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        U32 n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
+        huffNode[n1].parent = huffNode[n2].parent = nodeNb;
+        nodeNb++;
+    }
+
+    /* distribute weights (unlimited tree height) */
+    huffNode[nodeRoot].nbBits = 0;
+    for (n=nodeRoot-1; n>=STARTNODE; n--)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+    for (n=0; n<=nonNullRank; n++)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+
+    /* enforce maxTableLog */
+    maxNbBits = HUF_setMaxHeight(huffNode, nonNullRank, maxNbBits);
+
+    /* fill result into tree (val, nbBits) */
+    {   U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
+        U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+        if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
+        for (n=0; n<=nonNullRank; n++)
+            nbPerRank[huffNode[n].nbBits]++;
+        /* determine stating value per rank */
+        {   U16 min = 0;
+            for (n=maxNbBits; n>0; n--) {
+                valPerRank[n] = min;      /* get starting value within each rank */
+                min += nbPerRank[n];
+                min >>= 1;
+        }   }
+        for (n=0; n<=maxSymbolValue; n++)
+            tree[huffNode[n].byte].nbBits = huffNode[n].nbBits;   /* push nbBits per symbol, symbol order */
+        for (n=0; n<=maxSymbolValue; n++)
+            tree[n].val = valPerRank[tree[n].nbBits]++;   /* assign value within rank, symbol order */
+    }
+
+    return maxNbBits;
+}
+
+/** HUF_buildCTable() :
+ * @return : maxNbBits
+ *  Note : count is used before tree is written, so they can safely overlap
+ */
+size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits)
+{
+    huffNodeTable nodeTable;
+    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, nodeTable, sizeof(nodeTable));
+}
+
+static size_t HUF_estimateCompressedSize(HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
+{
+    size_t nbBits = 0;
+    int s;
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        nbBits += CTable[s].nbBits * count[s];
+    }
+    return nbBits >> 3;
+}
+
+static int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+  int bad = 0;
+  int s;
+  for (s = 0; s <= (int)maxSymbolValue; ++s) {
+    bad |= (count[s] != 0) & (CTable[s].nbBits == 0);
+  }
+  return !bad;
+}
+
+size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
+
+FORCE_INLINE_TEMPLATE void
+HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
+{
+    BIT_addBitsFast(bitCPtr, CTable[symbol].val, CTable[symbol].nbBits);
+}
+
+#define HUF_FLUSHBITS(s)  BIT_flushBits(s)
+
+#define HUF_FLUSHBITS_1(stream) \
+    if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*2+7) HUF_FLUSHBITS(stream)
+
+#define HUF_FLUSHBITS_2(stream) \
+    if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*4+7) HUF_FLUSHBITS(stream)
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    const BYTE* ip = (const BYTE*) src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+    size_t n;
+    BIT_CStream_t bitC;
+
+    /* init */
+    if (dstSize < 8) return 0;   /* not enough space to compress */
+    { size_t const initErr = BIT_initCStream(&bitC, op, oend-op);
+      if (HUF_isError(initErr)) return 0; }
+
+    n = srcSize & ~3;  /* join to mod 4 */
+    switch (srcSize & 3)
+    {
+        case 3 : HUF_encodeSymbol(&bitC, ip[n+ 2], CTable);
+                 HUF_FLUSHBITS_2(&bitC);
+		 /* fall-through */
+        case 2 : HUF_encodeSymbol(&bitC, ip[n+ 1], CTable);
+                 HUF_FLUSHBITS_1(&bitC);
+		 /* fall-through */
+        case 1 : HUF_encodeSymbol(&bitC, ip[n+ 0], CTable);
+                 HUF_FLUSHBITS(&bitC);
+		 /* fall-through */
+        case 0 : /* fall-through */
+        default: break;
+    }
+
+    for (; n>0; n-=4) {  /* note : n&3==0 at this stage */
+        HUF_encodeSymbol(&bitC, ip[n- 1], CTable);
+        HUF_FLUSHBITS_1(&bitC);
+        HUF_encodeSymbol(&bitC, ip[n- 2], CTable);
+        HUF_FLUSHBITS_2(&bitC);
+        HUF_encodeSymbol(&bitC, ip[n- 3], CTable);
+        HUF_FLUSHBITS_1(&bitC);
+        HUF_encodeSymbol(&bitC, ip[n- 4], CTable);
+        HUF_FLUSHBITS(&bitC);
+    }
+
+    return BIT_closeCStream(&bitC);
+}
+
+#if DYNAMIC_BMI2
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
+                                      const void* src, size_t srcSize,
+                                      const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int bmi2)
+{
+    if (bmi2) {
+        return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
+    }
+    return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
+}
+
+#else
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int bmi2)
+{
+    (void)bmi2;
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+#endif
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+}
+
+
+static size_t
+HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, int bmi2)
+{
+    size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    if (dstSize < 6 + 1 + 1 + 1 + 8) return 0;   /* minimum space to compress successfully */
+    if (srcSize < 12) return 0;   /* no saving possible : too small input */
+    op += 6;   /* jumpTable */
+
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, segmentSize, CTable, bmi2) );
+        if (cSize==0) return 0;
+        assert(cSize <= 65535);
+        MEM_writeLE16(ostart, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, segmentSize, CTable, bmi2) );
+        if (cSize==0) return 0;
+        assert(cSize <= 65535);
+        MEM_writeLE16(ostart+2, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, segmentSize, CTable, bmi2) );
+        if (cSize==0) return 0;
+        assert(cSize <= 65535);
+        MEM_writeLE16(ostart+4, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, oend-op, ip, iend-ip, CTable, bmi2) );
+        if (cSize==0) return 0;
+        op += cSize;
+    }
+
+    return op-ostart;
+}
+
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+}
+
+
+static size_t HUF_compressCTable_internal(
+                BYTE* const ostart, BYTE* op, BYTE* const oend,
+                const void* src, size_t srcSize,
+                unsigned singleStream, const HUF_CElt* CTable, const int bmi2)
+{
+    size_t const cSize = singleStream ?
+                         HUF_compress1X_usingCTable_internal(op, oend - op, src, srcSize, CTable, bmi2) :
+                         HUF_compress4X_usingCTable_internal(op, oend - op, src, srcSize, CTable, bmi2);
+    if (HUF_isError(cSize)) { return cSize; }
+    if (cSize==0) { return 0; }   /* uncompressible */
+    op += cSize;
+    /* check compressibility */
+    if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
+    return op-ostart;
+}
+
+typedef struct {
+    U32 count[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_CElt CTable[HUF_SYMBOLVALUE_MAX + 1];
+    huffNodeTable nodeTable;
+} HUF_compress_tables_t;
+
+/* HUF_compress_internal() :
+ * `workSpace` must a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
+static size_t HUF_compress_internal (
+                void* dst, size_t dstSize,
+                const void* src, size_t srcSize,
+                unsigned maxSymbolValue, unsigned huffLog,
+                unsigned singleStream,
+                void* workSpace, size_t wkspSize,
+                HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
+                const int bmi2)
+{
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    /* checks & inits */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < sizeof(*table)) return ERROR(workSpace_tooSmall);
+    if (!srcSize) return 0;  /* Uncompressed */
+    if (!dstSize) return 0;  /* cannot fit anything within dst budget */
+    if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
+    if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+    if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
+
+    /* Heuristic : If old table is valid, use it for small inputs */
+    if (preferRepeat && repeat && *repeat == HUF_repeat_valid) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           singleStream, oldHufTable, bmi2);
+    }
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(largest, FSE_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, table->count) );
+        if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
+        if (largest <= (srcSize >> 7)+1) return 0;   /* heuristic : probably not compressible enough */
+    }
+
+    /* Check validity of previous table */
+    if ( repeat
+      && *repeat == HUF_repeat_check
+      && !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
+        *repeat = HUF_repeat_none;
+    }
+    /* Heuristic : use existing table for small inputs */
+    if (preferRepeat && repeat && *repeat != HUF_repeat_none) {
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           singleStream, oldHufTable, bmi2);
+    }
+
+    /* Build Huffman Tree */
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    {   CHECK_V_F(maxBits, HUF_buildCTable_wksp(table->CTable, table->count,
+                                                maxSymbolValue, huffLog,
+                                                table->nodeTable, sizeof(table->nodeTable)) );
+        huffLog = (U32)maxBits;
+        /* Zero unused symbols in CTable, so we can check it for validity */
+        memset(table->CTable + (maxSymbolValue + 1), 0,
+               sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
+    }
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, table->CTable, maxSymbolValue, huffLog) );
+        /* Check if using previous huffman table is beneficial */
+        if (repeat && *repeat != HUF_repeat_none) {
+            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
+            size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
+            if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
+                return HUF_compressCTable_internal(ostart, op, oend,
+                                                   src, srcSize,
+                                                   singleStream, oldHufTable, bmi2);
+        }   }
+
+        /* Use the new huffman table */
+        if (hSize + 12ul >= srcSize) { return 0; }
+        op += hSize;
+        if (repeat) { *repeat = HUF_repeat_none; }
+        if (oldHufTable)
+            memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
+    }
+    return HUF_compressCTable_internal(ostart, op, oend,
+                                       src, srcSize,
+                                       singleStream, table->CTable, bmi2);
+}
+
+
+size_t HUF_compress1X_wksp (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 1 /*single stream*/,
+                                 workSpace, wkspSize,
+                                 NULL, NULL, 0, 0 /*bmi2*/);
+}
+
+size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 1 /*single stream*/,
+                                 workSpace, wkspSize, hufTable,
+                                 repeat, preferRepeat, bmi2);
+}
+
+size_t HUF_compress1X (void* dst, size_t dstSize,
+                 const void* src, size_t srcSize,
+                 unsigned maxSymbolValue, unsigned huffLog)
+{
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
+    return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
+}
+
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * provide workspace to generate compression tables */
+size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 0 /*4 streams*/,
+                                 workSpace, wkspSize,
+                                 NULL, NULL, 0, 0 /*bmi2*/);
+}
+
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * re-use an existing huffman compression table */
+size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, 0 /* 4 streams */,
+                                 workSpace, wkspSize,
+                                 hufTable, repeat, preferRepeat, bmi2);
+}
+
+size_t HUF_compress2 (void* dst, size_t dstSize,
+                const void* src, size_t srcSize,
+                unsigned maxSymbolValue, unsigned huffLog)
+{
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
+    return HUF_compress4X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
+}
+
+size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return HUF_compress2(dst, maxDstSize, src, srcSize, 255, HUF_TABLELOG_DEFAULT);
+}
diff --git a/vendor/github.com/DataDog/zstd/huf_decompress.c b/vendor/github.com/DataDog/zstd/huf_decompress.c
new file mode 100644
index 0000000..73f5c46
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/huf_decompress.c
@@ -0,0 +1,1096 @@
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Dependencies
+****************************************************************/
+#include <string.h>     /* memcpy, memset */
+#include "bitstream.h"  /* BIT_* */
+#include "compiler.h"
+#include "fse.h"        /* header compression */
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "error_private.h"
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_isError ERR_isError
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+#define CHECK_F(f) { size_t const err_ = (f); if (HUF_isError(err_)) return err_; }
+
+
+/* **************************************************************
+*  Byte alignment for workSpace management
+****************************************************************/
+#define HUF_ALIGN(x, a)         HUF_ALIGN_MASK((x), (a) - 1)
+#define HUF_ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask))
+
+
+/*-***************************/
+/*  generic DTableDesc       */
+/*-***************************/
+typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
+
+static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
+{
+    DTableDesc dtd;
+    memcpy(&dtd, table, sizeof(dtd));
+    return dtd;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
+{
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    size_t iSize;
+    void* const dtPtr = DTable + 1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+
+    U32* rankVal;
+    BYTE* huffWeight;
+    size_t spaceUsed32 = 0;
+
+    rankVal = (U32 *)workSpace + spaceUsed32;
+    spaceUsed32 += HUF_TABLELOG_ABSOLUTEMAX + 1;
+    huffWeight = (BYTE *)((U32 *)workSpace + spaceUsed32);
+    spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
+
+    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
+
+    HUF_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
+    /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* Table header */
+    {   DTableDesc dtd = HUF_getDTableDesc(DTable);
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, Huffman tree cannot fit in */
+        dtd.tableType = 0;
+        dtd.tableLog = (BYTE)tableLog;
+        memcpy(DTable, &dtd, sizeof(dtd));
+    }
+
+    /* Calculate starting value for each rank */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<tableLog+1; n++) {
+            U32 const current = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = current;
+    }   }
+
+    /* fill DTable */
+    {   U32 n;
+        for (n=0; n<nbSymbols; n++) {
+            U32 const w = huffWeight[n];
+            U32 const length = (1 << w) >> 1;
+            U32 u;
+            HUF_DEltX2 D;
+            D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+            for (u = rankVal[w]; u < rankVal[w] + length; u++)
+                dt[u] = D;
+            rankVal[w] += length;
+    }   }
+
+    return iSize;
+}
+
+size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_readDTableX2_wksp(DTable, src, srcSize,
+                                 workSpace, sizeof(workSpace));
+}
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+FORCE_INLINE_TEMPLATE BYTE
+HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    BYTE const c = dt[val].byte;
+    BIT_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr)  \
+    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+HINT_INLINE size_t
+HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* [0-3] symbols remaining */
+    if (MEM_32bits())
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd))
+            HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const void* dtPtr = DTable + 1;
+    const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+    BIT_DStream_t bitD;
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    U32 const dtLog = dtd.tableLog;
+
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    HUF_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X2_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable + 1;
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal = BIT_DStream_unfinished;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        while ( (endSignal==BIT_DStream_unfinished) && (op4<(oend-3)) ) {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+            BIT_reloadDStream(&bitD1);
+            BIT_reloadDStream(&bitD2);
+            BIT_reloadDStream(&bitD3);
+            BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        /* note : should not be necessary : op# advance in lock step, and we control op4.
+         *        but curiously, binary generated by gcc 7.2 & 7.3 with -mbmi2 runs faster when >=1 test is present */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);
+    }   }
+    return 1;
+}
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+HINT_INLINE size_t
+HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
+                const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X4_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    BIT_DStream_t bitD;
+
+    /* Init */
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        HUF_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X4_usingDTable_internal_body(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;
+        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUF_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) & (op4<(oend-(sizeof(bitD4.bitContainer)-1))) ; ) {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
+                                               const void *cSrc,
+                                               size_t cSrcSize,
+                                               const HUF_DTable *DTable);
+#if DYNAMIC_BMI2
+
+#define X(fn)                                                               \
+                                                                            \
+    static size_t fn##_default(                                             \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static TARGET_ATTRIBUTE("bmi2") size_t fn##_bmi2(                       \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+    {                                                                       \
+        if (bmi2) {                                                         \
+            return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
+        }                                                                   \
+        return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
+    }
+
+#else
+
+#define X(fn)                                                               \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+    {                                                                       \
+        (void)bmi2;                                                         \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }
+
+#endif
+
+X(HUF_decompress1X2_usingDTable_internal)
+X(HUF_decompress4X2_usingDTable_internal)
+X(HUF_decompress1X4_usingDTable_internal)
+X(HUF_decompress4X4_usingDTable_internal)
+
+#undef X
+
+
+size_t HUF_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp (dctx, cSrc, cSrcSize,
+                                                workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
+}
+
+
+size_t HUF_decompress4X2_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/* HUF_fillDTableX4Level2() :
+ * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_TABLELOG_MAX + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    {   U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+            const U32 symbol = sortedSymbols[s].symbol;
+            const U32 weight = sortedSymbols[s].weight;
+            const U32 nbBits = nbBitsBaseline - weight;
+            const U32 length = 1 << (sizeLog-nbBits);
+            const U32 start = rankVal[weight];
+            U32 i = start;
+            const U32 end = start + length;
+
+            MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+            DElt.nbBits = (BYTE)(nbBits + consumed);
+            DElt.length = 2;
+            do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+            rankVal[weight] += length;
+    }   }
+}
+
+typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
+typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_TABLELOG_MAX + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUF_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 const end = start + length;
+                U32 u;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
+                             size_t srcSize, void* workSpace,
+                             size_t wkspSize)
+{
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    U32 const maxTableLog = dtd.maxTableLog;
+    size_t iSize;
+    void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
+    HUF_DEltX4* const dt = (HUF_DEltX4*)dtPtr;
+    U32 *rankStart;
+
+    rankValCol_t* rankVal;
+    U32* rankStats;
+    U32* rankStart0;
+    sortedSymbol_t* sortedSymbol;
+    BYTE* weightList;
+    size_t spaceUsed32 = 0;
+
+    rankVal = (rankValCol_t *)((U32 *)workSpace + spaceUsed32);
+    spaceUsed32 += (sizeof(rankValCol_t) * HUF_TABLELOG_MAX) >> 2;
+    rankStats = (U32 *)workSpace + spaceUsed32;
+    spaceUsed32 += HUF_TABLELOG_MAX + 1;
+    rankStart0 = (U32 *)workSpace + spaceUsed32;
+    spaceUsed32 += HUF_TABLELOG_MAX + 2;
+    sortedSymbol = (sortedSymbol_t *)workSpace + (spaceUsed32 * sizeof(U32)) / sizeof(sortedSymbol_t);
+    spaceUsed32 += HUF_ALIGN(sizeof(sortedSymbol_t) * (HUF_SYMBOLVALUE_MAX + 1), sizeof(U32)) >> 2;
+    weightList = (BYTE *)((U32 *)workSpace + spaceUsed32);
+    spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
+
+    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
+
+    rankStart = rankStart0 + 1;
+    memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
+    if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    /* memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (maxTableLog-tableLog) - 1;   /* tableLog <= maxTableLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUF_fillDTableX4(dt, maxTableLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    dtd.tableLog = (BYTE)maxTableLog;
+    dtd.tableType = 1;
+    memcpy(DTable, &dtd, sizeof(dtd));
+    return iSize;
+}
+
+size_t HUF_readDTableX4(HUF_DTable* DTable, const void* src, size_t srcSize)
+{
+  U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+  return HUF_readDTableX4_wksp(DTable, src, srcSize,
+                               workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX4_wksp(DCtx, cSrc, cSrcSize,
+                                               workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress1X4_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X4_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+static size_t HUF_decompress4X4_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4_wksp(dctx, cSrc, cSrcSize,
+                                         workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    return HUF_decompress4X4_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUF_DTable* DTable)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUF_DTable* DTable)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+/** HUF_selectDecoder() :
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
+ *  Assumption : 0 < dstSize <= 128 KB */
+U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
+{
+    assert(dstSize > 0);
+    assert(dstSize <= 128 KB);
+    /* decoder timing evaluation */
+    {   U32 const Q = (cSrcSize >= dstSize) ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+        U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+        DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, to reduce cache eviction */
+        return DTime1 < DTime0;
+}   }
+
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[2] = { HUF_decompress4X2, HUF_decompress4X4 };
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+}
+
+size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUF_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress4X_hufOnly_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                         workSpace, sizeof(workSpace));
+}
+
+
+size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
+                                     size_t dstSize, const void* cSrc,
+                                     size_t cSrcSize, void* workSpace,
+                                     size_t wkspSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize == 0) return ERROR(corruption_detected);
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize):
+                        HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
+    }
+}
+
+size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                  const void* cSrc, size_t cSrcSize,
+                                  void* workSpace, size_t wkspSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUF_decompress1X4_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize):
+                        HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize);
+    }
+}
+
+size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+                             const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+                                      workSpace, sizeof(workSpace));
+}
+
+
+size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+}
+
+size_t HUF_decompress1X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+}
+
+size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize == 0) return ERROR(corruption_detected);
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUF_decompress4X4_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
+                        HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+    }
+}
diff --git a/vendor/github.com/DataDog/zstd/mem.h b/vendor/github.com/DataDog/zstd/mem.h
new file mode 100644
index 0000000..47d2300
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/mem.h
@@ -0,0 +1,362 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>     /* size_t, ptrdiff_t */
+#include <string.h>     /* memcpy */
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __inline __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+/* code only tested on 32 and 64 bits systems */
+#define MEM_STATIC_ASSERT(c)   { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
+MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include <stdint.h>
+  typedef   uint8_t BYTE;
+  typedef  uint16_t U16;
+  typedef   int16_t S16;
+  typedef  uint32_t U32;
+  typedef   int32_t S32;
+  typedef  uint64_t U64;
+  typedef   int64_t S64;
+#else
+  typedef unsigned char      BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS :
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets depending on alignment.
+ *            In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif defined(__INTEL_COMPILER) || defined(__GNUC__)
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard, by lying on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+#if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
+    __pragma( pack(push, 1) )
+    typedef struct { U16 v; } unalign16;
+    typedef struct { U32 v; } unalign32;
+    typedef struct { U64 v; } unalign64;
+    typedef struct { size_t v; } unalignArch;
+    __pragma( pack(pop) )
+#else
+    typedef struct { U16 v; } __attribute__((packed)) unalign16;
+    typedef struct { U32 v; } __attribute__((packed)) unalign32;
+    typedef struct { U64 v; } __attribute__((packed)) unalign64;
+    typedef struct { size_t v; } __attribute__((packed)) unalignArch;
+#endif
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; }
+MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC size_t MEM_readST(const void* memPtr)
+{
+    size_t val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write32(void* memPtr, U32 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write64(void* memPtr, U64 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* MEM_FORCE_MEMORY_ACCESS */
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+MEM_STATIC size_t MEM_swapST(size_t in)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_swap32((U32)in);
+    else
+        return (size_t)MEM_swap64((U64)in);
+}
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
+MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val)
+{
+    MEM_writeLE16(memPtr, (U16)val);
+    ((BYTE*)memPtr)[2] = (BYTE)(val>>16);
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32)
+{
+    if (MEM_isLittleEndian())
+        MEM_write32(memPtr, val32);
+    else
+        MEM_write32(memPtr, MEM_swap32(val32));
+}
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64)
+{
+    if (MEM_isLittleEndian())
+        MEM_write64(memPtr, val64);
+    else
+        MEM_write64(memPtr, MEM_swap64(val64));
+}
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val)
+{
+    if (MEM_32bits())
+        MEM_writeLE32(memPtr, (U32)val);
+    else
+        MEM_writeLE64(memPtr, (U64)val);
+}
+
+/*=== Big endian r/w ===*/
+
+MEM_STATIC U32 MEM_readBE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_swap32(MEM_read32(memPtr));
+    else
+        return MEM_read32(memPtr);
+}
+
+MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32)
+{
+    if (MEM_isLittleEndian())
+        MEM_write32(memPtr, MEM_swap32(val32));
+    else
+        MEM_write32(memPtr, val32);
+}
+
+MEM_STATIC U64 MEM_readBE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_swap64(MEM_read64(memPtr));
+    else
+        return MEM_read64(memPtr);
+}
+
+MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64)
+{
+    if (MEM_isLittleEndian())
+        MEM_write64(memPtr, MEM_swap64(val64));
+    else
+        MEM_write64(memPtr, val64);
+}
+
+MEM_STATIC size_t MEM_readBEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readBE32(memPtr);
+    else
+        return (size_t)MEM_readBE64(memPtr);
+}
+
+MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
+{
+    if (MEM_32bits())
+        MEM_writeBE32(memPtr, (U32)val);
+    else
+        MEM_writeBE64(memPtr, (U64)val);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
diff --git a/vendor/github.com/DataDog/zstd/pool.c b/vendor/github.com/DataDog/zstd/pool.c
new file mode 100644
index 0000000..773488b
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/pool.c
@@ -0,0 +1,283 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ======   Dependencies   ======= */
+#include <stddef.h>  /* size_t */
+#include "pool.h"
+#include "zstd_internal.h"  /* ZSTD_malloc, ZSTD_free */
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+#ifdef ZSTD_MULTITHREAD
+
+#include "threading.h"   /* pthread adaptation */
+
+/* A job is a function and an opaque argument */
+typedef struct POOL_job_s {
+    POOL_function function;
+    void *opaque;
+} POOL_job;
+
+struct POOL_ctx_s {
+    ZSTD_customMem customMem;
+    /* Keep track of the threads */
+    ZSTD_pthread_t *threads;
+    size_t numThreads;
+
+    /* The queue is a circular buffer */
+    POOL_job *queue;
+    size_t queueHead;
+    size_t queueTail;
+    size_t queueSize;
+
+    /* The number of threads working on jobs */
+    size_t numThreadsBusy;
+    /* Indicates if the queue is empty */
+    int queueEmpty;
+
+    /* The mutex protects the queue */
+    ZSTD_pthread_mutex_t queueMutex;
+    /* Condition variable for pushers to wait on when the queue is full */
+    ZSTD_pthread_cond_t queuePushCond;
+    /* Condition variables for poppers to wait on when the queue is empty */
+    ZSTD_pthread_cond_t queuePopCond;
+    /* Indicates if the queue is shutting down */
+    int shutdown;
+};
+
+/* POOL_thread() :
+   Work thread for the thread pool.
+   Waits for jobs and executes them.
+   @returns : NULL on failure else non-null.
+*/
+static void* POOL_thread(void* opaque) {
+    POOL_ctx* const ctx = (POOL_ctx*)opaque;
+    if (!ctx) { return NULL; }
+    for (;;) {
+        /* Lock the mutex and wait for a non-empty queue or until shutdown */
+        ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+
+        while (ctx->queueEmpty && !ctx->shutdown) {
+            ZSTD_pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
+        }
+        /* empty => shutting down: so stop */
+        if (ctx->queueEmpty) {
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+            return opaque;
+        }
+        /* Pop a job off the queue */
+        {   POOL_job const job = ctx->queue[ctx->queueHead];
+            ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
+            ctx->numThreadsBusy++;
+            ctx->queueEmpty = ctx->queueHead == ctx->queueTail;
+            /* Unlock the mutex, signal a pusher, and run the job */
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+            ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+
+            job.function(job.opaque);
+
+            /* If the intended queue size was 0, signal after finishing job */
+            if (ctx->queueSize == 1) {
+                ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+                ctx->numThreadsBusy--;
+                ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+                ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+        }   }
+    }  /* for (;;) */
+    /* Unreachable */
+}
+
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
+    return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
+}
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
+    POOL_ctx* ctx;
+    /* Check the parameters */
+    if (!numThreads) { return NULL; }
+    /* Allocate the context and zero initialize */
+    ctx = (POOL_ctx*)ZSTD_calloc(sizeof(POOL_ctx), customMem);
+    if (!ctx) { return NULL; }
+    /* Initialize the job queue.
+     * It needs one extra space since one space is wasted to differentiate empty
+     * and full queues.
+     */
+    ctx->queueSize = queueSize + 1;
+    ctx->queue = (POOL_job*)ZSTD_malloc(ctx->queueSize * sizeof(POOL_job), customMem);
+    ctx->queueHead = 0;
+    ctx->queueTail = 0;
+    ctx->numThreadsBusy = 0;
+    ctx->queueEmpty = 1;
+    (void)ZSTD_pthread_mutex_init(&ctx->queueMutex, NULL);
+    (void)ZSTD_pthread_cond_init(&ctx->queuePushCond, NULL);
+    (void)ZSTD_pthread_cond_init(&ctx->queuePopCond, NULL);
+    ctx->shutdown = 0;
+    /* Allocate space for the thread handles */
+    ctx->threads = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
+    ctx->numThreads = 0;
+    ctx->customMem = customMem;
+    /* Check for errors */
+    if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
+    /* Initialize the threads */
+    {   size_t i;
+        for (i = 0; i < numThreads; ++i) {
+            if (ZSTD_pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
+                ctx->numThreads = i;
+                POOL_free(ctx);
+                return NULL;
+        }   }
+        ctx->numThreads = numThreads;
+    }
+    return ctx;
+}
+
+/*! POOL_join() :
+    Shutdown the queue, wake any sleeping threads, and join all of the threads.
+*/
+static void POOL_join(POOL_ctx* ctx) {
+    /* Shut down the queue */
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    ctx->shutdown = 1;
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    /* Wake up sleeping threads */
+    ZSTD_pthread_cond_broadcast(&ctx->queuePushCond);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
+    /* Join all of the threads */
+    {   size_t i;
+        for (i = 0; i < ctx->numThreads; ++i) {
+            ZSTD_pthread_join(ctx->threads[i], NULL);
+    }   }
+}
+
+void POOL_free(POOL_ctx *ctx) {
+    if (!ctx) { return; }
+    POOL_join(ctx);
+    ZSTD_pthread_mutex_destroy(&ctx->queueMutex);
+    ZSTD_pthread_cond_destroy(&ctx->queuePushCond);
+    ZSTD_pthread_cond_destroy(&ctx->queuePopCond);
+    ZSTD_free(ctx->queue, ctx->customMem);
+    ZSTD_free(ctx->threads, ctx->customMem);
+    ZSTD_free(ctx, ctx->customMem);
+}
+
+size_t POOL_sizeof(POOL_ctx *ctx) {
+    if (ctx==NULL) return 0;  /* supports sizeof NULL */
+    return sizeof(*ctx)
+        + ctx->queueSize * sizeof(POOL_job)
+        + ctx->numThreads * sizeof(ZSTD_pthread_t);
+}
+
+/**
+ * Returns 1 if the queue is full and 0 otherwise.
+ *
+ * If the queueSize is 1 (the pool was created with an intended queueSize of 0),
+ * then a queue is empty if there is a thread free and no job is waiting.
+ */
+static int isQueueFull(POOL_ctx const* ctx) {
+    if (ctx->queueSize > 1) {
+        return ctx->queueHead == ((ctx->queueTail + 1) % ctx->queueSize);
+    } else {
+        return ctx->numThreadsBusy == ctx->numThreads ||
+               !ctx->queueEmpty;
+    }
+}
+
+
+static void POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
+{
+    POOL_job const job = {function, opaque};
+    assert(ctx != NULL);
+    if (ctx->shutdown) return;
+
+    ctx->queueEmpty = 0;
+    ctx->queue[ctx->queueTail] = job;
+    ctx->queueTail = (ctx->queueTail + 1) % ctx->queueSize;
+    ZSTD_pthread_cond_signal(&ctx->queuePopCond);
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    /* Wait until there is space in the queue for the new job */
+    while (isQueueFull(ctx) && (!ctx->shutdown)) {
+        ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
+    }
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+}
+
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    if (isQueueFull(ctx)) {
+        ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        return 0;
+    }
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return 1;
+}
+
+
+#else  /* ZSTD_MULTITHREAD  not defined */
+
+/* ========================== */
+/* No multi-threading support */
+/* ========================== */
+
+
+/* We don't need any data, but if it is empty, malloc() might return NULL. */
+struct POOL_ctx_s {
+    int dummy;
+};
+static POOL_ctx g_ctx;
+
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
+    return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
+}
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
+    (void)numThreads;
+    (void)queueSize;
+    (void)customMem;
+    return &g_ctx;
+}
+
+void POOL_free(POOL_ctx* ctx) {
+    assert(!ctx || ctx == &g_ctx);
+    (void)ctx;
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+}
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+    return 1;
+}
+
+size_t POOL_sizeof(POOL_ctx* ctx) {
+    if (ctx==NULL) return 0;  /* supports sizeof NULL */
+    assert(ctx == &g_ctx);
+    return sizeof(*ctx);
+}
+
+#endif  /* ZSTD_MULTITHREAD */
diff --git a/vendor/github.com/DataDog/zstd/pool.h b/vendor/github.com/DataDog/zstd/pool.h
new file mode 100644
index 0000000..a57e9b4
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/pool.h
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef POOL_H
+#define POOL_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+#include <stddef.h>   /* size_t */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_customMem */
+#include "zstd.h"
+
+typedef struct POOL_ctx_s POOL_ctx;
+
+/*! POOL_create() :
+ *  Create a thread pool with at most `numThreads` threads.
+ * `numThreads` must be at least 1.
+ *  The maximum number of queued jobs before blocking is `queueSize`.
+ * @return : POOL_ctx pointer on success, else NULL.
+*/
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize);
+
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem);
+
+/*! POOL_free() :
+    Free a thread pool returned by POOL_create().
+*/
+void POOL_free(POOL_ctx* ctx);
+
+/*! POOL_sizeof() :
+    return memory usage of pool returned by POOL_create().
+*/
+size_t POOL_sizeof(POOL_ctx* ctx);
+
+/*! POOL_function :
+    The function type that can be added to a thread pool.
+*/
+typedef void (*POOL_function)(void*);
+/*! POOL_add_function :
+    The function type for a generic thread pool add function.
+*/
+typedef void (*POOL_add_function)(void*, POOL_function, void*);
+
+/*! POOL_add() :
+    Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
+    Possibly blocks until there is room in the queue.
+    Note : The function may be executed asynchronously, so `opaque` must live until the function has been completed.
+*/
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
+
+
+/*! POOL_tryAdd() :
+    Add the job `function(opaque)` to the thread pool if a worker is available.
+    return immediately otherwise.
+   @return : 1 if successful, 0 if not.
+*/
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif
diff --git a/vendor/github.com/DataDog/zstd/threading.c b/vendor/github.com/DataDog/zstd/threading.c
new file mode 100644
index 0000000..8be8c8d
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/threading.c
@@ -0,0 +1,75 @@
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ */
+
+/**
+ * This file will hold wrapper for systems, which do not support pthreads
+ */
+
+/* create fake symbol to avoid empty trnaslation unit warning */
+int g_ZSTD_threading_useles_symbol;
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper, based on :
+ * http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ */
+
+
+/* ===  Dependencies  === */
+#include <process.h>
+#include <errno.h>
+#include "threading.h"
+
+
+/* ===  Implementation  === */
+
+static unsigned __stdcall worker(void *arg)
+{
+    ZSTD_pthread_t* const thread = (ZSTD_pthread_t*) arg;
+    thread->arg = thread->start_routine(thread->arg);
+    return 0;
+}
+
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
+            void* (*start_routine) (void*), void* arg)
+{
+    (void)unused;
+    thread->arg = arg;
+    thread->start_routine = start_routine;
+    thread->handle = (HANDLE) _beginthreadex(NULL, 0, worker, thread, 0, NULL);
+
+    if (!thread->handle)
+        return errno;
+    else
+        return 0;
+}
+
+int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
+{
+    DWORD result;
+
+    if (!thread.handle) return 0;
+
+    result = WaitForSingleObject(thread.handle, INFINITE);
+    switch (result) {
+    case WAIT_OBJECT_0:
+        if (value_ptr) *value_ptr = thread.arg;
+        return 0;
+    case WAIT_ABANDONED:
+        return EINVAL;
+    default:
+        return GetLastError();
+    }
+}
+
+#endif   /* ZSTD_MULTITHREAD */
diff --git a/vendor/github.com/DataDog/zstd/threading.h b/vendor/github.com/DataDog/zstd/threading.h
new file mode 100644
index 0000000..d806c89
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/threading.h
@@ -0,0 +1,123 @@
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ */
+
+#ifndef THREADING_H_938743
+#define THREADING_H_938743
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper, based on :
+ * http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ */
+#ifdef WINVER
+#  undef WINVER
+#endif
+#define WINVER       0x0600
+
+#ifdef _WIN32_WINNT
+#  undef _WIN32_WINNT
+#endif
+#define _WIN32_WINNT 0x0600
+
+#ifndef WIN32_LEAN_AND_MEAN
+#  define WIN32_LEAN_AND_MEAN
+#endif
+
+#undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
+#include <windows.h>
+#undef ERROR
+#define ERROR(name) ZSTD_ERROR(name)
+
+
+/* mutex */
+#define ZSTD_pthread_mutex_t           CRITICAL_SECTION
+#define ZSTD_pthread_mutex_init(a, b)  ((void)(b), InitializeCriticalSection((a)), 0)
+#define ZSTD_pthread_mutex_destroy(a)  DeleteCriticalSection((a))
+#define ZSTD_pthread_mutex_lock(a)     EnterCriticalSection((a))
+#define ZSTD_pthread_mutex_unlock(a)   LeaveCriticalSection((a))
+
+/* condition variable */
+#define ZSTD_pthread_cond_t             CONDITION_VARIABLE
+#define ZSTD_pthread_cond_init(a, b)    ((void)(b), InitializeConditionVariable((a)), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
+#define ZSTD_pthread_cond_signal(a)     WakeConditionVariable((a))
+#define ZSTD_pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
+
+/* ZSTD_pthread_create() and ZSTD_pthread_join() */
+typedef struct {
+    HANDLE handle;
+    void* (*start_routine)(void*);
+    void* arg;
+} ZSTD_pthread_t;
+
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
+                   void* (*start_routine) (void*), void* arg);
+
+int ZSTD_pthread_join(ZSTD_pthread_t thread, void** value_ptr);
+
+/**
+ * add here more wrappers as required
+ */
+
+
+#elif defined(ZSTD_MULTITHREAD)   /* posix assumed ; need a better detection method */
+/* ===   POSIX Systems   === */
+#  include <pthread.h>
+
+#define ZSTD_pthread_mutex_t            pthread_mutex_t
+#define ZSTD_pthread_mutex_init(a, b)   pthread_mutex_init((a), (b))
+#define ZSTD_pthread_mutex_destroy(a)   pthread_mutex_destroy((a))
+#define ZSTD_pthread_mutex_lock(a)      pthread_mutex_lock((a))
+#define ZSTD_pthread_mutex_unlock(a)    pthread_mutex_unlock((a))
+
+#define ZSTD_pthread_cond_t             pthread_cond_t
+#define ZSTD_pthread_cond_init(a, b)    pthread_cond_init((a), (b))
+#define ZSTD_pthread_cond_destroy(a)    pthread_cond_destroy((a))
+#define ZSTD_pthread_cond_wait(a, b)    pthread_cond_wait((a), (b))
+#define ZSTD_pthread_cond_signal(a)     pthread_cond_signal((a))
+#define ZSTD_pthread_cond_broadcast(a)  pthread_cond_broadcast((a))
+
+#define ZSTD_pthread_t                  pthread_t
+#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
+#define ZSTD_pthread_join(a, b)         pthread_join((a),(b))
+
+#else  /* ZSTD_MULTITHREAD not defined */
+/* No multithreading support */
+
+typedef int ZSTD_pthread_mutex_t;
+#define ZSTD_pthread_mutex_init(a, b)   ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_mutex_destroy(a)   ((void)(a))
+#define ZSTD_pthread_mutex_lock(a)      ((void)(a))
+#define ZSTD_pthread_mutex_unlock(a)    ((void)(a))
+
+typedef int ZSTD_pthread_cond_t;
+#define ZSTD_pthread_cond_init(a, b)    ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    ((void)(a), (void)(b))
+#define ZSTD_pthread_cond_signal(a)     ((void)(a))
+#define ZSTD_pthread_cond_broadcast(a)  ((void)(a))
+
+/* do not use ZSTD_pthread_t */
+
+#endif /* ZSTD_MULTITHREAD */
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* THREADING_H_938743 */
diff --git a/vendor/github.com/DataDog/zstd/update.txt b/vendor/github.com/DataDog/zstd/update.txt
new file mode 100644
index 0000000..1de939f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/update.txt
@@ -0,0 +1,56 @@
+./lib/common/bitstream.h
+./lib/common/compiler.h
+./lib/compress/zstd_compress_internal.h
+./lib/compress/zstd_fast.h
+./lib/compress/zstd_double_fast.h
+./lib/compress/zstd_lazy.h
+./lib/compress/zstd_ldm.h
+./lib/dictBuilder/cover.c
+./lib/dictBuilder/divsufsort.c
+./lib/dictBuilder/divsufsort.h
+./lib/common/entropy_common.c
+./lib/common/error_private.c
+./lib/common/error_private.h
+./lib/compress/fse_compress.c
+./lib/common/fse_decompress.c
+./lib/common/fse.h
+./lib/compress/huf_compress.c
+./lib/decompress/huf_decompress.c
+./lib/common/huf.h
+./lib/common/mem.h
+./lib/common/pool.c
+./lib/common/pool.h
+./lib/common/threading.c
+./lib/common/threading.h
+./lib/common/xxhash.c
+./lib/common/xxhash.h
+./lib/deprecated/zbuff_common.c
+./lib/deprecated/zbuff_compress.c
+./lib/deprecated/zbuff_decompress.c
+./lib/deprecated/zbuff.h
+./lib/dictBuilder/zdict.c
+./lib/dictBuilder/zdict.h
+./lib/common/zstd_common.c
+./lib/compress/zstd_compress.c
+./lib/decompress/zstd_decompress.c
+./lib/common/zstd_errors.h
+./lib/zstd.h
+./lib/common/zstd_internal.h
+./lib/legacy/zstd_legacy.h
+./lib/compress/zstd_opt.c
+./lib/compress/zstd_opt.h
+./lib/legacy/zstd_v01.c
+./lib/legacy/zstd_v01.h
+./lib/legacy/zstd_v02.c
+./lib/legacy/zstd_v02.h
+./lib/legacy/zstd_v03.c
+./lib/legacy/zstd_v03.h
+./lib/legacy/zstd_v04.c
+./lib/legacy/zstd_v04.h
+./lib/legacy/zstd_v05.c
+./lib/legacy/zstd_v05.h
+./lib/legacy/zstd_v06.c
+./lib/legacy/zstd_v06.h
+./lib/legacy/zstd_v07.c
+./lib/legacy/zstd_v07.h
+
diff --git a/vendor/github.com/DataDog/zstd/xxhash.c b/vendor/github.com/DataDog/zstd/xxhash.c
new file mode 100644
index 0000000..9d9c0e9
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/xxhash.c
@@ -0,0 +1,875 @@
+/*
+*  xxHash - Fast Hash algorithm
+*  Copyright (C) 2012-2016, Yann Collet
+*
+*  BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+*
+*  Redistribution and use in source and binary forms, with or without
+*  modification, are permitted provided that the following conditions are
+*  met:
+*
+*  * Redistributions of source code must retain the above copyright
+*  notice, this list of conditions and the following disclaimer.
+*  * Redistributions in binary form must reproduce the above
+*  copyright notice, this list of conditions and the following disclaimer
+*  in the documentation and/or other materials provided with the
+*  distribution.
+*
+*  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+*  "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+*  LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+*  A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+*  OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+*  SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+*  LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+*  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+*  THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+*  (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+*  OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*
+*  You can contact the author at :
+*  - xxHash homepage: http://www.xxhash.com
+*  - xxHash source repository : https://github.com/Cyan4973/xxHash
+*/
+
+
+/* *************************************
+*  Tuning parameters
+***************************************/
+/*!XXH_FORCE_MEMORY_ACCESS :
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method doesn't depend on compiler but violate C standard.
+ *            It can generate buggy code on targets which do not support unaligned memory accesses.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://stackoverflow.com/a/32095106/646947 for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define XXH_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define XXH_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+/*!XXH_ACCEPT_NULL_INPUT_POINTER :
+ * If the input pointer is a null pointer, xxHash default behavior is to trigger a memory access error, since it is a bad pointer.
+ * When this option is enabled, xxHash output for null input pointers will be the same as a null-length input.
+ * By default, this option is disabled. To enable it, uncomment below define :
+ */
+/* #define XXH_ACCEPT_NULL_INPUT_POINTER 1 */
+
+/*!XXH_FORCE_NATIVE_FORMAT :
+ * By default, xxHash library provides endian-independant Hash values, based on little-endian convention.
+ * Results are therefore identical for little-endian and big-endian CPU.
+ * This comes at a performance cost for big-endian CPU, since some swapping is required to emulate little-endian format.
+ * Should endian-independance be of no importance for your application, you may set the #define below to 1,
+ * to improve speed for Big-endian CPU.
+ * This option has no impact on Little_Endian CPU.
+ */
+#ifndef XXH_FORCE_NATIVE_FORMAT   /* can be defined externally */
+#  define XXH_FORCE_NATIVE_FORMAT 0
+#endif
+
+/*!XXH_FORCE_ALIGN_CHECK :
+ * This is a minor performance trick, only useful with lots of very small keys.
+ * It means : check for aligned/unaligned input.
+ * The check costs one initial branch per hash; set to 0 when the input data
+ * is guaranteed to be aligned.
+ */
+#ifndef XXH_FORCE_ALIGN_CHECK /* can be defined externally */
+#  if defined(__i386) || defined(_M_IX86) || defined(__x86_64__) || defined(_M_X64)
+#    define XXH_FORCE_ALIGN_CHECK 0
+#  else
+#    define XXH_FORCE_ALIGN_CHECK 1
+#  endif
+#endif
+
+
+/* *************************************
+*  Includes & Memory related functions
+***************************************/
+/* Modify the local functions below should you wish to use some other memory routines */
+/* for malloc(), free() */
+#include <stdlib.h>
+static void* XXH_malloc(size_t s) { return malloc(s); }
+static void  XXH_free  (void* p)  { free(p); }
+/* for memcpy() */
+#include <string.h>
+static void* XXH_memcpy(void* dest, const void* src, size_t size) { return memcpy(dest,src,size); }
+
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY
+#endif
+#include "xxhash.h"
+
+
+/* *************************************
+*  Compiler Specific Options
+***************************************/
+#if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#  define INLINE_KEYWORD inline
+#else
+#  define INLINE_KEYWORD
+#endif
+
+#if defined(__GNUC__)
+#  define FORCE_INLINE_ATTR __attribute__((always_inline))
+#elif defined(_MSC_VER)
+#  define FORCE_INLINE_ATTR __forceinline
+#else
+#  define FORCE_INLINE_ATTR
+#endif
+
+#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
+
+
+#ifdef _MSC_VER
+#  pragma warning(disable : 4127)      /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* *************************************
+*  Basic Types
+***************************************/
+#ifndef MEM_MODULE
+# define MEM_MODULE
+# if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#   include <stdint.h>
+    typedef uint8_t  BYTE;
+    typedef uint16_t U16;
+    typedef uint32_t U32;
+    typedef  int32_t S32;
+    typedef uint64_t U64;
+#  else
+    typedef unsigned char      BYTE;
+    typedef unsigned short     U16;
+    typedef unsigned int       U32;
+    typedef   signed int       S32;
+    typedef unsigned long long U64;   /* if your compiler doesn't support unsigned long long, replace by another 64-bit type here. Note that xxhash.h will also need to be updated. */
+#  endif
+#endif
+
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
+static U32 XXH_read32(const void* memPtr) { return *(const U32*) memPtr; }
+static U64 XXH_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U32 u32; U64 u64; } __attribute__((packed)) unalign;
+
+static U32 XXH_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+static U64 XXH_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+#else
+
+/* portable and safe solution. Generally efficient.
+ * see : http://stackoverflow.com/a/32095106/646947
+ */
+
+static U32 XXH_read32(const void* memPtr)
+{
+    U32 val;
+    memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+static U64 XXH_read64(const void* memPtr)
+{
+    U64 val;
+    memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+
+/* ****************************************
+*  Compiler-specific Functions and Macros
+******************************************/
+#define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
+/* Note : although _rotl exists for minGW (GCC under windows), performance seems poor */
+#if defined(_MSC_VER)
+#  define XXH_rotl32(x,r) _rotl(x,r)
+#  define XXH_rotl64(x,r) _rotl64(x,r)
+#else
+#  define XXH_rotl32(x,r) ((x << r) | (x >> (32 - r)))
+#  define XXH_rotl64(x,r) ((x << r) | (x >> (64 - r)))
+#endif
+
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap32 _byteswap_ulong
+#  define XXH_swap64 _byteswap_uint64
+#elif GCC_VERSION >= 403
+#  define XXH_swap32 __builtin_bswap32
+#  define XXH_swap64 __builtin_bswap64
+#else
+static U32 XXH_swap32 (U32 x)
+{
+    return  ((x << 24) & 0xff000000 ) |
+            ((x <<  8) & 0x00ff0000 ) |
+            ((x >>  8) & 0x0000ff00 ) |
+            ((x >> 24) & 0x000000ff );
+}
+static U64 XXH_swap64 (U64 x)
+{
+    return  ((x << 56) & 0xff00000000000000ULL) |
+            ((x << 40) & 0x00ff000000000000ULL) |
+            ((x << 24) & 0x0000ff0000000000ULL) |
+            ((x << 8)  & 0x000000ff00000000ULL) |
+            ((x >> 8)  & 0x00000000ff000000ULL) |
+            ((x >> 24) & 0x0000000000ff0000ULL) |
+            ((x >> 40) & 0x000000000000ff00ULL) |
+            ((x >> 56) & 0x00000000000000ffULL);
+}
+#endif
+
+
+/* *************************************
+*  Architecture Macros
+***************************************/
+typedef enum { XXH_bigEndian=0, XXH_littleEndian=1 } XXH_endianess;
+
+/* XXH_CPU_LITTLE_ENDIAN can be defined externally, for example on the compiler command line */
+#ifndef XXH_CPU_LITTLE_ENDIAN
+    static const int g_one = 1;
+#   define XXH_CPU_LITTLE_ENDIAN   (*(const char*)(&g_one))
+#endif
+
+
+/* ***************************
+*  Memory reads
+*****************************/
+typedef enum { XXH_aligned, XXH_unaligned } XXH_alignment;
+
+FORCE_INLINE_TEMPLATE U32 XXH_readLE32_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return endian==XXH_littleEndian ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
+    else
+        return endian==XXH_littleEndian ? *(const U32*)ptr : XXH_swap32(*(const U32*)ptr);
+}
+
+FORCE_INLINE_TEMPLATE U32 XXH_readLE32(const void* ptr, XXH_endianess endian)
+{
+    return XXH_readLE32_align(ptr, endian, XXH_unaligned);
+}
+
+static U32 XXH_readBE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
+}
+
+FORCE_INLINE_TEMPLATE U64 XXH_readLE64_align(const void* ptr, XXH_endianess endian, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return endian==XXH_littleEndian ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
+    else
+        return endian==XXH_littleEndian ? *(const U64*)ptr : XXH_swap64(*(const U64*)ptr);
+}
+
+FORCE_INLINE_TEMPLATE U64 XXH_readLE64(const void* ptr, XXH_endianess endian)
+{
+    return XXH_readLE64_align(ptr, endian, XXH_unaligned);
+}
+
+static U64 XXH_readBE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+}
+
+
+/* *************************************
+*  Macros
+***************************************/
+#define XXH_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(int)(!!(c)) }; }    /* use only *after* variable declarations */
+
+
+/* *************************************
+*  Constants
+***************************************/
+static const U32 PRIME32_1 = 2654435761U;
+static const U32 PRIME32_2 = 2246822519U;
+static const U32 PRIME32_3 = 3266489917U;
+static const U32 PRIME32_4 =  668265263U;
+static const U32 PRIME32_5 =  374761393U;
+
+static const U64 PRIME64_1 = 11400714785074694791ULL;
+static const U64 PRIME64_2 = 14029467366897019727ULL;
+static const U64 PRIME64_3 =  1609587929392839161ULL;
+static const U64 PRIME64_4 =  9650029242287828579ULL;
+static const U64 PRIME64_5 =  2870177450012600261ULL;
+
+XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
+
+
+/* **************************
+*  Utils
+****************************/
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dstState, const XXH32_state_t* restrict srcState)
+{
+    memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dstState, const XXH64_state_t* restrict srcState)
+{
+    memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+
+/* ***************************
+*  Simple Hash Functions
+*****************************/
+
+static U32 XXH32_round(U32 seed, U32 input)
+{
+    seed += input * PRIME32_2;
+    seed  = XXH_rotl32(seed, 13);
+    seed *= PRIME32_1;
+    return seed;
+}
+
+FORCE_INLINE_TEMPLATE U32 XXH32_endian_align(const void* input, size_t len, U32 seed, XXH_endianess endian, XXH_alignment align)
+{
+    const BYTE* p = (const BYTE*)input;
+    const BYTE* bEnd = p + len;
+    U32 h32;
+#define XXH_get32bits(p) XXH_readLE32_align(p, endian, align)
+
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
+    if (p==NULL) {
+        len=0;
+        bEnd=p=(const BYTE*)(size_t)16;
+    }
+#endif
+
+    if (len>=16) {
+        const BYTE* const limit = bEnd - 16;
+        U32 v1 = seed + PRIME32_1 + PRIME32_2;
+        U32 v2 = seed + PRIME32_2;
+        U32 v3 = seed + 0;
+        U32 v4 = seed - PRIME32_1;
+
+        do {
+            v1 = XXH32_round(v1, XXH_get32bits(p)); p+=4;
+            v2 = XXH32_round(v2, XXH_get32bits(p)); p+=4;
+            v3 = XXH32_round(v3, XXH_get32bits(p)); p+=4;
+            v4 = XXH32_round(v4, XXH_get32bits(p)); p+=4;
+        } while (p<=limit);
+
+        h32 = XXH_rotl32(v1, 1) + XXH_rotl32(v2, 7) + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
+    } else {
+        h32  = seed + PRIME32_5;
+    }
+
+    h32 += (U32) len;
+
+    while (p+4<=bEnd) {
+        h32 += XXH_get32bits(p) * PRIME32_3;
+        h32  = XXH_rotl32(h32, 17) * PRIME32_4 ;
+        p+=4;
+    }
+
+    while (p<bEnd) {
+        h32 += (*p) * PRIME32_5;
+        h32 = XXH_rotl32(h32, 11) * PRIME32_1 ;
+        p++;
+    }
+
+    h32 ^= h32 >> 15;
+    h32 *= PRIME32_2;
+    h32 ^= h32 >> 13;
+    h32 *= PRIME32_3;
+    h32 ^= h32 >> 16;
+
+    return h32;
+}
+
+
+XXH_PUBLIC_API unsigned int XXH32 (const void* input, size_t len, unsigned int seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH32_CREATESTATE_STATIC(state);
+    XXH32_reset(state, seed);
+    XXH32_update(state, input, len);
+    return XXH32_digest(state);
+#else
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
+            if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+                return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
+            else
+                return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
+    }   }
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH32_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
+    else
+        return XXH32_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
+#endif
+}
+
+
+static U64 XXH64_round(U64 acc, U64 input)
+{
+    acc += input * PRIME64_2;
+    acc  = XXH_rotl64(acc, 31);
+    acc *= PRIME64_1;
+    return acc;
+}
+
+static U64 XXH64_mergeRound(U64 acc, U64 val)
+{
+    val  = XXH64_round(0, val);
+    acc ^= val;
+    acc  = acc * PRIME64_1 + PRIME64_4;
+    return acc;
+}
+
+FORCE_INLINE_TEMPLATE U64 XXH64_endian_align(const void* input, size_t len, U64 seed, XXH_endianess endian, XXH_alignment align)
+{
+    const BYTE* p = (const BYTE*)input;
+    const BYTE* const bEnd = p + len;
+    U64 h64;
+#define XXH_get64bits(p) XXH_readLE64_align(p, endian, align)
+
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
+    if (p==NULL) {
+        len=0;
+        bEnd=p=(const BYTE*)(size_t)32;
+    }
+#endif
+
+    if (len>=32) {
+        const BYTE* const limit = bEnd - 32;
+        U64 v1 = seed + PRIME64_1 + PRIME64_2;
+        U64 v2 = seed + PRIME64_2;
+        U64 v3 = seed + 0;
+        U64 v4 = seed - PRIME64_1;
+
+        do {
+            v1 = XXH64_round(v1, XXH_get64bits(p)); p+=8;
+            v2 = XXH64_round(v2, XXH_get64bits(p)); p+=8;
+            v3 = XXH64_round(v3, XXH_get64bits(p)); p+=8;
+            v4 = XXH64_round(v4, XXH_get64bits(p)); p+=8;
+        } while (p<=limit);
+
+        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
+        h64 = XXH64_mergeRound(h64, v1);
+        h64 = XXH64_mergeRound(h64, v2);
+        h64 = XXH64_mergeRound(h64, v3);
+        h64 = XXH64_mergeRound(h64, v4);
+
+    } else {
+        h64  = seed + PRIME64_5;
+    }
+
+    h64 += (U64) len;
+
+    while (p+8<=bEnd) {
+        U64 const k1 = XXH64_round(0, XXH_get64bits(p));
+        h64 ^= k1;
+        h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
+        p+=8;
+    }
+
+    if (p+4<=bEnd) {
+        h64 ^= (U64)(XXH_get32bits(p)) * PRIME64_1;
+        h64 = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
+        p+=4;
+    }
+
+    while (p<bEnd) {
+        h64 ^= (*p) * PRIME64_5;
+        h64 = XXH_rotl64(h64, 11) * PRIME64_1;
+        p++;
+    }
+
+    h64 ^= h64 >> 33;
+    h64 *= PRIME64_2;
+    h64 ^= h64 >> 29;
+    h64 *= PRIME64_3;
+    h64 ^= h64 >> 32;
+
+    return h64;
+}
+
+
+XXH_PUBLIC_API unsigned long long XXH64 (const void* input, size_t len, unsigned long long seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH64_CREATESTATE_STATIC(state);
+    XXH64_reset(state, seed);
+    XXH64_update(state, input, len);
+    return XXH64_digest(state);
+#else
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
+            if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+                return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_aligned);
+            else
+                return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_aligned);
+    }   }
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH64_endian_align(input, len, seed, XXH_littleEndian, XXH_unaligned);
+    else
+        return XXH64_endian_align(input, len, seed, XXH_bigEndian, XXH_unaligned);
+#endif
+}
+
+
+/* **************************************************
+*  Advanced Hash Functions
+****************************************************/
+
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
+{
+    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
+}
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
+{
+    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
+}
+XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+
+/*** Hash feed ***/
+
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, unsigned int seed)
+{
+    XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
+    memset(&state, 0, sizeof(state)-4);   /* do not write into reserved, for future removal */
+    state.v1 = seed + PRIME32_1 + PRIME32_2;
+    state.v2 = seed + PRIME32_2;
+    state.v3 = seed + 0;
+    state.v4 = seed - PRIME32_1;
+    memcpy(statePtr, &state, sizeof(state));
+    return XXH_OK;
+}
+
+
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, unsigned long long seed)
+{
+    XXH64_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
+    memset(&state, 0, sizeof(state)-8);   /* do not write into reserved, for future removal */
+    state.v1 = seed + PRIME64_1 + PRIME64_2;
+    state.v2 = seed + PRIME64_2;
+    state.v3 = seed + 0;
+    state.v4 = seed - PRIME64_1;
+    memcpy(statePtr, &state, sizeof(state));
+    return XXH_OK;
+}
+
+
+FORCE_INLINE_TEMPLATE XXH_errorcode XXH32_update_endian (XXH32_state_t* state, const void* input, size_t len, XXH_endianess endian)
+{
+    const BYTE* p = (const BYTE*)input;
+    const BYTE* const bEnd = p + len;
+
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
+    if (input==NULL) return XXH_ERROR;
+#endif
+
+    state->total_len_32 += (unsigned)len;
+    state->large_len |= (len>=16) | (state->total_len_32>=16);
+
+    if (state->memsize + len < 16)  {   /* fill in tmp buffer */
+        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, len);
+        state->memsize += (unsigned)len;
+        return XXH_OK;
+    }
+
+    if (state->memsize) {   /* some data left from previous update */
+        XXH_memcpy((BYTE*)(state->mem32) + state->memsize, input, 16-state->memsize);
+        {   const U32* p32 = state->mem32;
+            state->v1 = XXH32_round(state->v1, XXH_readLE32(p32, endian)); p32++;
+            state->v2 = XXH32_round(state->v2, XXH_readLE32(p32, endian)); p32++;
+            state->v3 = XXH32_round(state->v3, XXH_readLE32(p32, endian)); p32++;
+            state->v4 = XXH32_round(state->v4, XXH_readLE32(p32, endian)); p32++;
+        }
+        p += 16-state->memsize;
+        state->memsize = 0;
+    }
+
+    if (p <= bEnd-16) {
+        const BYTE* const limit = bEnd - 16;
+        U32 v1 = state->v1;
+        U32 v2 = state->v2;
+        U32 v3 = state->v3;
+        U32 v4 = state->v4;
+
+        do {
+            v1 = XXH32_round(v1, XXH_readLE32(p, endian)); p+=4;
+            v2 = XXH32_round(v2, XXH_readLE32(p, endian)); p+=4;
+            v3 = XXH32_round(v3, XXH_readLE32(p, endian)); p+=4;
+            v4 = XXH32_round(v4, XXH_readLE32(p, endian)); p+=4;
+        } while (p<=limit);
+
+        state->v1 = v1;
+        state->v2 = v2;
+        state->v3 = v3;
+        state->v4 = v4;
+    }
+
+    if (p < bEnd) {
+        XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
+        state->memsize = (unsigned)(bEnd-p);
+    }
+
+    return XXH_OK;
+}
+
+XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* state_in, const void* input, size_t len)
+{
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH32_update_endian(state_in, input, len, XXH_littleEndian);
+    else
+        return XXH32_update_endian(state_in, input, len, XXH_bigEndian);
+}
+
+
+
+FORCE_INLINE_TEMPLATE U32 XXH32_digest_endian (const XXH32_state_t* state, XXH_endianess endian)
+{
+    const BYTE * p = (const BYTE*)state->mem32;
+    const BYTE* const bEnd = (const BYTE*)(state->mem32) + state->memsize;
+    U32 h32;
+
+    if (state->large_len) {
+        h32 = XXH_rotl32(state->v1, 1) + XXH_rotl32(state->v2, 7) + XXH_rotl32(state->v3, 12) + XXH_rotl32(state->v4, 18);
+    } else {
+        h32 = state->v3 /* == seed */ + PRIME32_5;
+    }
+
+    h32 += state->total_len_32;
+
+    while (p+4<=bEnd) {
+        h32 += XXH_readLE32(p, endian) * PRIME32_3;
+        h32  = XXH_rotl32(h32, 17) * PRIME32_4;
+        p+=4;
+    }
+
+    while (p<bEnd) {
+        h32 += (*p) * PRIME32_5;
+        h32  = XXH_rotl32(h32, 11) * PRIME32_1;
+        p++;
+    }
+
+    h32 ^= h32 >> 15;
+    h32 *= PRIME32_2;
+    h32 ^= h32 >> 13;
+    h32 *= PRIME32_3;
+    h32 ^= h32 >> 16;
+
+    return h32;
+}
+
+
+XXH_PUBLIC_API unsigned int XXH32_digest (const XXH32_state_t* state_in)
+{
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH32_digest_endian(state_in, XXH_littleEndian);
+    else
+        return XXH32_digest_endian(state_in, XXH_bigEndian);
+}
+
+
+
+/* **** XXH64 **** */
+
+FORCE_INLINE_TEMPLATE XXH_errorcode XXH64_update_endian (XXH64_state_t* state, const void* input, size_t len, XXH_endianess endian)
+{
+    const BYTE* p = (const BYTE*)input;
+    const BYTE* const bEnd = p + len;
+
+#ifdef XXH_ACCEPT_NULL_INPUT_POINTER
+    if (input==NULL) return XXH_ERROR;
+#endif
+
+    state->total_len += len;
+
+    if (state->memsize + len < 32) {  /* fill in tmp buffer */
+        XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, len);
+        state->memsize += (U32)len;
+        return XXH_OK;
+    }
+
+    if (state->memsize) {   /* tmp buffer is full */
+        XXH_memcpy(((BYTE*)state->mem64) + state->memsize, input, 32-state->memsize);
+        state->v1 = XXH64_round(state->v1, XXH_readLE64(state->mem64+0, endian));
+        state->v2 = XXH64_round(state->v2, XXH_readLE64(state->mem64+1, endian));
+        state->v3 = XXH64_round(state->v3, XXH_readLE64(state->mem64+2, endian));
+        state->v4 = XXH64_round(state->v4, XXH_readLE64(state->mem64+3, endian));
+        p += 32-state->memsize;
+        state->memsize = 0;
+    }
+
+    if (p+32 <= bEnd) {
+        const BYTE* const limit = bEnd - 32;
+        U64 v1 = state->v1;
+        U64 v2 = state->v2;
+        U64 v3 = state->v3;
+        U64 v4 = state->v4;
+
+        do {
+            v1 = XXH64_round(v1, XXH_readLE64(p, endian)); p+=8;
+            v2 = XXH64_round(v2, XXH_readLE64(p, endian)); p+=8;
+            v3 = XXH64_round(v3, XXH_readLE64(p, endian)); p+=8;
+            v4 = XXH64_round(v4, XXH_readLE64(p, endian)); p+=8;
+        } while (p<=limit);
+
+        state->v1 = v1;
+        state->v2 = v2;
+        state->v3 = v3;
+        state->v4 = v4;
+    }
+
+    if (p < bEnd) {
+        XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
+        state->memsize = (unsigned)(bEnd-p);
+    }
+
+    return XXH_OK;
+}
+
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* state_in, const void* input, size_t len)
+{
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH64_update_endian(state_in, input, len, XXH_littleEndian);
+    else
+        return XXH64_update_endian(state_in, input, len, XXH_bigEndian);
+}
+
+
+
+FORCE_INLINE_TEMPLATE U64 XXH64_digest_endian (const XXH64_state_t* state, XXH_endianess endian)
+{
+    const BYTE * p = (const BYTE*)state->mem64;
+    const BYTE* const bEnd = (const BYTE*)state->mem64 + state->memsize;
+    U64 h64;
+
+    if (state->total_len >= 32) {
+        U64 const v1 = state->v1;
+        U64 const v2 = state->v2;
+        U64 const v3 = state->v3;
+        U64 const v4 = state->v4;
+
+        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
+        h64 = XXH64_mergeRound(h64, v1);
+        h64 = XXH64_mergeRound(h64, v2);
+        h64 = XXH64_mergeRound(h64, v3);
+        h64 = XXH64_mergeRound(h64, v4);
+    } else {
+        h64  = state->v3 + PRIME64_5;
+    }
+
+    h64 += (U64) state->total_len;
+
+    while (p+8<=bEnd) {
+        U64 const k1 = XXH64_round(0, XXH_readLE64(p, endian));
+        h64 ^= k1;
+        h64  = XXH_rotl64(h64,27) * PRIME64_1 + PRIME64_4;
+        p+=8;
+    }
+
+    if (p+4<=bEnd) {
+        h64 ^= (U64)(XXH_readLE32(p, endian)) * PRIME64_1;
+        h64  = XXH_rotl64(h64, 23) * PRIME64_2 + PRIME64_3;
+        p+=4;
+    }
+
+    while (p<bEnd) {
+        h64 ^= (*p) * PRIME64_5;
+        h64  = XXH_rotl64(h64, 11) * PRIME64_1;
+        p++;
+    }
+
+    h64 ^= h64 >> 33;
+    h64 *= PRIME64_2;
+    h64 ^= h64 >> 29;
+    h64 *= PRIME64_3;
+    h64 ^= h64 >> 32;
+
+    return h64;
+}
+
+
+XXH_PUBLIC_API unsigned long long XXH64_digest (const XXH64_state_t* state_in)
+{
+    XXH_endianess endian_detected = (XXH_endianess)XXH_CPU_LITTLE_ENDIAN;
+
+    if ((endian_detected==XXH_littleEndian) || XXH_FORCE_NATIVE_FORMAT)
+        return XXH64_digest_endian(state_in, XXH_littleEndian);
+    else
+        return XXH64_digest_endian(state_in, XXH_bigEndian);
+}
+
+
+/* **************************
+*  Canonical representation
+****************************/
+
+/*! Default XXH result types are basic unsigned 32 and 64 bits.
+*   The canonical representation follows human-readable write convention, aka big-endian (large digits first).
+*   These functions allow transformation of hash result into and from its canonical format.
+*   This way, hash values can be written into a file or buffer, and remain comparable across different systems and programs.
+*/
+
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+    memcpy(dst, &hash, sizeof(*dst));
+}
+
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
+    memcpy(dst, &hash, sizeof(*dst));
+}
+
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
+{
+    return XXH_readBE32(src);
+}
+
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
+{
+    return XXH_readBE64(src);
+}
diff --git a/vendor/github.com/DataDog/zstd/xxhash.h b/vendor/github.com/DataDog/zstd/xxhash.h
new file mode 100644
index 0000000..9bad1f5
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/xxhash.h
@@ -0,0 +1,305 @@
+/*
+   xxHash - Extremely Fast Hash algorithm
+   Header File
+   Copyright (C) 2012-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - xxHash source repository : https://github.com/Cyan4973/xxHash
+*/
+
+/* Notice extracted from xxHash homepage :
+
+xxHash is an extremely fast Hash algorithm, running at RAM speed limits.
+It also successfully passes all tests from the SMHasher suite.
+
+Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)
+
+Name            Speed       Q.Score   Author
+xxHash          5.4 GB/s     10
+CrapWow         3.2 GB/s      2       Andrew
+MumurHash 3a    2.7 GB/s     10       Austin Appleby
+SpookyHash      2.0 GB/s     10       Bob Jenkins
+SBox            1.4 GB/s      9       Bret Mulvey
+Lookup3         1.2 GB/s      9       Bob Jenkins
+SuperFastHash   1.2 GB/s      1       Paul Hsieh
+CityHash64      1.05 GB/s    10       Pike & Alakuijala
+FNV             0.55 GB/s     5       Fowler, Noll, Vo
+CRC32           0.43 GB/s     9
+MD5-32          0.33 GB/s    10       Ronald L. Rivest
+SHA1-32         0.28 GB/s    10
+
+Q.Score is a measure of quality of the hash function.
+It depends on successfully passing SMHasher test set.
+10 is a perfect score.
+
+A 64-bits version, named XXH64, is available since r35.
+It offers much better speed, but for 64-bits applications only.
+Name     Speed on 64 bits    Speed on 32 bits
+XXH64       13.8 GB/s            1.9 GB/s
+XXH32        6.8 GB/s            6.0 GB/s
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef XXHASH_H_5627135585666179
+#define XXHASH_H_5627135585666179 1
+
+
+/* ****************************
+*  Definitions
+******************************/
+#include <stddef.h>   /* size_t */
+typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
+
+
+/* ****************************
+*  API modifier
+******************************/
+/** XXH_PRIVATE_API
+*   This is useful if you want to include xxhash functions in `static` mode
+*   in order to inline them, and remove their symbol from the public list.
+*   Methodology :
+*     #define XXH_PRIVATE_API
+*     #include "xxhash.h"
+*   `xxhash.c` is automatically included.
+*   It's not useful to compile and link it as a separate module anymore.
+*/
+#ifdef XXH_PRIVATE_API
+#  ifndef XXH_STATIC_LINKING_ONLY
+#    define XXH_STATIC_LINKING_ONLY
+#  endif
+#  if defined(__GNUC__)
+#    define XXH_PUBLIC_API static __inline __attribute__((unused))
+#  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#    define XXH_PUBLIC_API static inline
+#  elif defined(_MSC_VER)
+#    define XXH_PUBLIC_API static __inline
+#  else
+#    define XXH_PUBLIC_API static   /* this version may generate warnings for unused static functions; disable the relevant warning */
+#  endif
+#else
+#  define XXH_PUBLIC_API   /* do nothing */
+#endif /* XXH_PRIVATE_API */
+
+/*!XXH_NAMESPACE, aka Namespace Emulation :
+
+If you want to include _and expose_ xxHash functions from within your own library,
+but also want to avoid symbol collisions with another library which also includes xxHash,
+
+you can use XXH_NAMESPACE, to automatically prefix any public symbol from xxhash library
+with the value of XXH_NAMESPACE (so avoid to keep it NULL and avoid numeric values).
+
+Note that no change is required within the calling program as long as it includes `xxhash.h` :
+regular symbol name will be automatically translated by this header.
+*/
+#ifdef XXH_NAMESPACE
+#  define XXH_CAT(A,B) A##B
+#  define XXH_NAME2(A,B) XXH_CAT(A,B)
+#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
+#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+#  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+#  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
+#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+#  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
+#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+#  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
+#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+#  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
+#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+#  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
+#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+#  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
+#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+#  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
+#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
+#  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+#  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+#endif
+
+
+/* *************************************
+*  Version
+***************************************/
+#define XXH_VERSION_MAJOR    0
+#define XXH_VERSION_MINOR    6
+#define XXH_VERSION_RELEASE  2
+#define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
+XXH_PUBLIC_API unsigned XXH_versionNumber (void);
+
+
+/* ****************************
+*  Simple Hash Functions
+******************************/
+typedef unsigned int       XXH32_hash_t;
+typedef unsigned long long XXH64_hash_t;
+
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, unsigned int seed);
+XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t length, unsigned long long seed);
+
+/*!
+XXH32() :
+    Calculate the 32-bits hash of sequence "length" bytes stored at memory address "input".
+    The memory between input & input+length must be valid (allocated and read-accessible).
+    "seed" can be used to alter the result predictably.
+    Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark) : 5.4 GB/s
+XXH64() :
+    Calculate the 64-bits hash of sequence of length "len" stored at memory address "input".
+    "seed" can be used to alter the result predictably.
+    This function runs 2x faster on 64-bits systems, but slower on 32-bits systems (see benchmark).
+*/
+
+
+/* ****************************
+*  Streaming Hash Functions
+******************************/
+typedef struct XXH32_state_s XXH32_state_t;   /* incomplete type */
+typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+
+/*! State allocation, compatible with dynamic libraries */
+
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
+XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
+
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
+XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+
+
+/* hash streaming */
+
+XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, unsigned int seed);
+XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH32_hash_t  XXH32_digest (const XXH32_state_t* statePtr);
+
+XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, unsigned long long seed);
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
+
+/*
+These functions generate the xxHash of an input provided in multiple segments.
+Note that, for small input, they are slower than single-call functions, due to state management.
+For small input, prefer `XXH32()` and `XXH64()` .
+
+XXH state must first be allocated, using XXH*_createState() .
+
+Start a new hash by initializing state with a seed, using XXH*_reset().
+
+Then, feed the hash state by calling XXH*_update() as many times as necessary.
+Obviously, input must be allocated and read accessible.
+The function returns an error code, with 0 meaning OK, and any other value meaning there is an error.
+
+Finally, a hash value can be produced anytime, by using XXH*_digest().
+This function returns the nn-bits hash as an int or long long.
+
+It's still possible to continue inserting input into the hash state after a digest,
+and generate some new hashes later on, by calling again XXH*_digest().
+
+When done, free XXH state space if it was allocated dynamically.
+*/
+
+
+/* **************************
+*  Utils
+****************************/
+#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* ! C99 */
+#  define restrict   /* disable restrict */
+#endif
+
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* restrict dst_state, const XXH32_state_t* restrict src_state);
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* restrict dst_state, const XXH64_state_t* restrict src_state);
+
+
+/* **************************
+*  Canonical representation
+****************************/
+/* Default result type for XXH functions are primitive unsigned 32 and 64 bits.
+*  The canonical representation uses human-readable write convention, aka big-endian (large digits first).
+*  These functions allow transformation of hash result into and from its canonical format.
+*  This way, hash values can be written into a file / memory, and remain comparable on different systems and programs.
+*/
+typedef struct { unsigned char digest[4]; } XXH32_canonical_t;
+typedef struct { unsigned char digest[8]; } XXH64_canonical_t;
+
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
+
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
+
+#endif /* XXHASH_H_5627135585666179 */
+
+
+
+/* ================================================================================================
+   This section contains definitions which are not guaranteed to remain stable.
+   They may change in future versions, becoming incompatible with a different version of the library.
+   They shall only be used with static linking.
+   Never use these definitions in association with dynamic linking !
+=================================================================================================== */
+#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXH_STATIC_H_3543687687345)
+#define XXH_STATIC_H_3543687687345
+
+/* These definitions are only meant to allow allocation of XXH state
+   statically, on stack, or in a struct for example.
+   Do not use members directly. */
+
+   struct XXH32_state_s {
+       unsigned total_len_32;
+       unsigned large_len;
+       unsigned v1;
+       unsigned v2;
+       unsigned v3;
+       unsigned v4;
+       unsigned mem32[4];   /* buffer defined as U32 for alignment */
+       unsigned memsize;
+       unsigned reserved;   /* never read nor write, will be removed in a future version */
+   };   /* typedef'd to XXH32_state_t */
+
+   struct XXH64_state_s {
+       unsigned long long total_len;
+       unsigned long long v1;
+       unsigned long long v2;
+       unsigned long long v3;
+       unsigned long long v4;
+       unsigned long long mem64[4];   /* buffer defined as U64 for alignment */
+       unsigned memsize;
+       unsigned reserved[2];          /* never read nor write, will be removed in a future version */
+   };   /* typedef'd to XXH64_state_t */
+
+
+#  ifdef XXH_PRIVATE_API
+#    include "xxhash.c"   /* include xxhash functions as `static`, for inlining */
+#  endif
+
+#endif /* XXH_STATIC_LINKING_ONLY && XXH_STATIC_H_3543687687345 */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/vendor/github.com/DataDog/zstd/zbuff.h b/vendor/github.com/DataDog/zstd/zbuff.h
new file mode 100644
index 0000000..a93115d
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zbuff.h
@@ -0,0 +1,213 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* ***************************************************************
+*  NOTES/WARNINGS
+******************************************************************/
+/* The streaming API defined here is deprecated.
+ * Consider migrating towards ZSTD_compressStream() API in `zstd.h`
+ * See 'lib/README.md'.
+ *****************************************************************/
+
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef ZSTD_BUFFERED_H_23987
+#define ZSTD_BUFFERED_H_23987
+
+/* *************************************
+*  Dependencies
+***************************************/
+#include <stddef.h>      /* size_t */
+#include "zstd.h"        /* ZSTD_CStream, ZSTD_DStream, ZSTDLIB_API */
+
+
+/* ***************************************************************
+*  Compiler specifics
+*****************************************************************/
+/* Deprecation warnings */
+/* Should these warnings be a problem,
+   it is generally possible to disable them,
+   typically with -Wno-deprecated-declarations for gcc
+   or _CRT_SECURE_NO_WARNINGS in Visual.
+   Otherwise, it's also possible to define ZBUFF_DISABLE_DEPRECATE_WARNINGS */
+#ifdef ZBUFF_DISABLE_DEPRECATE_WARNINGS
+#  define ZBUFF_DEPRECATED(message) ZSTDLIB_API  /* disable deprecation warnings */
+#else
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZBUFF_DEPRECATED(message) [[deprecated(message)]] ZSTDLIB_API
+#  elif (defined(__GNUC__) && (__GNUC__ >= 5)) || defined(__clang__)
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API __attribute__((deprecated(message)))
+#  elif defined(__GNUC__) && (__GNUC__ >= 3)
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZBUFF_DEPRECATED for this compiler")
+#    define ZBUFF_DEPRECATED(message) ZSTDLIB_API
+#  endif
+#endif /* ZBUFF_DISABLE_DEPRECATE_WARNINGS */
+
+
+/* *************************************
+*  Streaming functions
+***************************************/
+/* This is the easier "buffered" streaming API,
+*  using an internal buffer to lift all restrictions on user-provided buffers
+*  which can be any size, any place, for both input and output.
+*  ZBUFF and ZSTD are 100% interoperable,
+*  frames created by one can be decoded by the other one */
+
+typedef ZSTD_CStream ZBUFF_CCtx;
+ZBUFF_DEPRECATED("use ZSTD_createCStream") ZBUFF_CCtx* ZBUFF_createCCtx(void);
+ZBUFF_DEPRECATED("use ZSTD_freeCStream")   size_t      ZBUFF_freeCCtx(ZBUFF_CCtx* cctx);
+
+ZBUFF_DEPRECATED("use ZSTD_initCStream")           size_t ZBUFF_compressInit(ZBUFF_CCtx* cctx, int compressionLevel);
+ZBUFF_DEPRECATED("use ZSTD_initCStream_usingDict") size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+
+ZBUFF_DEPRECATED("use ZSTD_compressStream") size_t ZBUFF_compressContinue(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr, const void* src, size_t* srcSizePtr);
+ZBUFF_DEPRECATED("use ZSTD_flushStream")    size_t ZBUFF_compressFlush(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr);
+ZBUFF_DEPRECATED("use ZSTD_endStream")      size_t ZBUFF_compressEnd(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr);
+
+/*-*************************************************
+*  Streaming compression - howto
+*
+*  A ZBUFF_CCtx object is required to track streaming operation.
+*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
+*  ZBUFF_CCtx objects can be reused multiple times.
+*
+*  Start by initializing ZBUF_CCtx.
+*  Use ZBUFF_compressInit() to start a new compression operation.
+*  Use ZBUFF_compressInitDictionary() for a compression which requires a dictionary.
+*
+*  Use ZBUFF_compressContinue() repetitively to consume input stream.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written within *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present again remaining data.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each call, so save its content if it matters or change @dst .
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's just a hint, to improve latency)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  At any moment, it's possible to flush whatever data remains within buffer, using ZBUFF_compressFlush().
+*  The nb of bytes written into `dst` will be reported into *dstCapacityPtr.
+*  Note that the function cannot output more than *dstCapacityPtr,
+*  therefore, some content might still be left into internal buffer if *dstCapacityPtr is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressEnd() instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  The epilogue is required for decoders to consider a frame completed.
+*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
+*  In which case, call again ZBUFF_compressFlush() to complete the flush.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : _recommended buffer_ sizes (not compulsory) : ZBUFF_recommendedCInSize() / ZBUFF_recommendedCOutSize()
+*  input : ZBUFF_recommendedCInSize==128 KB block size is the internal unit, use this value to reduce intermediate stages (better latency)
+*  output : ZBUFF_recommendedCOutSize==ZSTD_compressBound(128 KB) + 3 + 3 : ensures it's always possible to write/flush/end a full block. Skip some buffering.
+*  By using both, it ensures that input will be entirely consumed, and output will always contain the result, reducing intermediate buffering.
+* **************************************************/
+
+
+typedef ZSTD_DStream ZBUFF_DCtx;
+ZBUFF_DEPRECATED("use ZSTD_createDStream") ZBUFF_DCtx* ZBUFF_createDCtx(void);
+ZBUFF_DEPRECATED("use ZSTD_freeDStream")   size_t      ZBUFF_freeDCtx(ZBUFF_DCtx* dctx);
+
+ZBUFF_DEPRECATED("use ZSTD_initDStream")           size_t ZBUFF_decompressInit(ZBUFF_DCtx* dctx);
+ZBUFF_DEPRECATED("use ZSTD_initDStream_usingDict") size_t ZBUFF_decompressInitDictionary(ZBUFF_DCtx* dctx, const void* dict, size_t dictSize);
+
+ZBUFF_DEPRECATED("use ZSTD_decompressStream") size_t ZBUFF_decompressContinue(ZBUFF_DCtx* dctx,
+                                            void* dst, size_t* dstCapacityPtr,
+                                      const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFF_DCtx object is required to track streaming operations.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation,
+*   or ZBUFF_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFF_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
+*  @return : 0 when a frame is completely decoded and fully flushed,
+*            1 when there is still some data left within internal buffer to flush,
+*            >1 when more data is expected, with value being a suggested next input size (it's just a hint, which helps latency),
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFF_recommendedDInSize() and ZBUFF_recommendedDOutSize()
+*  output : ZBUFF_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFF_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+ZBUFF_DEPRECATED("use ZSTD_isError")      unsigned ZBUFF_isError(size_t errorCode);
+ZBUFF_DEPRECATED("use ZSTD_getErrorName") const char* ZBUFF_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, they tend to offer better latency */
+ZBUFF_DEPRECATED("use ZSTD_CStreamInSize")  size_t ZBUFF_recommendedCInSize(void);
+ZBUFF_DEPRECATED("use ZSTD_CStreamOutSize") size_t ZBUFF_recommendedCOutSize(void);
+ZBUFF_DEPRECATED("use ZSTD_DStreamInSize")  size_t ZBUFF_recommendedDInSize(void);
+ZBUFF_DEPRECATED("use ZSTD_DStreamOutSize") size_t ZBUFF_recommendedDOutSize(void);
+
+#endif  /* ZSTD_BUFFERED_H_23987 */
+
+
+#ifdef ZBUFF_STATIC_LINKING_ONLY
+#ifndef ZBUFF_STATIC_H_30298098432
+#define ZBUFF_STATIC_H_30298098432
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used in association with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+/*--- Dependency ---*/
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters, ZSTD_customMem */
+#include "zstd.h"
+
+
+/*--- Custom memory allocator ---*/
+/*! ZBUFF_createCCtx_advanced() :
+ *  Create a ZBUFF compression context using external alloc and free functions */
+ZBUFF_DEPRECATED("use ZSTD_createCStream_advanced") ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem);
+
+/*! ZBUFF_createDCtx_advanced() :
+ *  Create a ZBUFF decompression context using external alloc and free functions */
+ZBUFF_DEPRECATED("use ZSTD_createDStream_advanced") ZBUFF_DCtx* ZBUFF_createDCtx_advanced(ZSTD_customMem customMem);
+
+
+/*--- Advanced Streaming Initialization ---*/
+ZBUFF_DEPRECATED("use ZSTD_initDStream_usingDict") size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
+                                               const void* dict, size_t dictSize,
+                                               ZSTD_parameters params, unsigned long long pledgedSrcSize);
+
+
+#endif    /* ZBUFF_STATIC_H_30298098432 */
+#endif    /* ZBUFF_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/vendor/github.com/DataDog/zstd/zbuff_common.c b/vendor/github.com/DataDog/zstd/zbuff_common.c
new file mode 100644
index 0000000..661b9b0
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zbuff_common.c
@@ -0,0 +1,26 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "error_private.h"
+#include "zbuff.h"
+
+/*-****************************************
+*  ZBUFF Error Management  (deprecated)
+******************************************/
+
+/*! ZBUFF_isError() :
+*   tells if a return value is an error code */
+unsigned ZBUFF_isError(size_t errorCode) { return ERR_isError(errorCode); }
+/*! ZBUFF_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZBUFF_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
diff --git a/vendor/github.com/DataDog/zstd/zbuff_compress.c b/vendor/github.com/DataDog/zstd/zbuff_compress.c
new file mode 100644
index 0000000..f39c60d
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zbuff_compress.c
@@ -0,0 +1,147 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/* *************************************
+*  Dependencies
+***************************************/
+#define ZBUFF_STATIC_LINKING_ONLY
+#include "zbuff.h"
+
+
+/*-***********************************************************
+*  Streaming compression
+*
+*  A ZBUFF_CCtx object is required to track streaming operation.
+*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
+*  Use ZBUFF_compressInit() to start a new compression operation.
+*  ZBUFF_CCtx objects can be reused multiple times.
+*
+*  Use ZBUFF_compressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters or change dst .
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressFlush() can be used to instruct ZBUFF to compress and output whatever remains within its buffer.
+*  Note that it will not output more than *dstCapacityPtr.
+*  Therefore, some content might still be left into its internal buffer if dst buffer is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  ZBUFF_compressEnd() instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
+*  @return : nb of bytes still present into internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  input : ZSTD_BLOCKSIZE_MAX (128 KB), internal unit size, it improves latency to use this value.
+*  output : ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + ZBUFF_endFrameSize : ensures it's always possible to write/flush/end a full block at best speed.
+* ***********************************************************/
+
+ZBUFF_CCtx* ZBUFF_createCCtx(void)
+{
+    return ZSTD_createCStream();
+}
+
+ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createCStream_advanced(customMem);
+}
+
+size_t ZBUFF_freeCCtx(ZBUFF_CCtx* zbc)
+{
+    return ZSTD_freeCStream(zbc);
+}
+
+
+/* ======   Initialization   ====== */
+
+size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
+                                   const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* preserve "0 == unknown" behavior */
+    return ZSTD_initCStream_advanced(zbc, dict, dictSize, params, pledgedSrcSize);
+}
+
+
+size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* zbc, const void* dict, size_t dictSize, int compressionLevel)
+{
+    return ZSTD_initCStream_usingDict(zbc, dict, dictSize, compressionLevel);
+}
+
+size_t ZBUFF_compressInit(ZBUFF_CCtx* zbc, int compressionLevel)
+{
+    return ZSTD_initCStream(zbc, compressionLevel);
+}
+
+/* ======   Compression   ====== */
+
+
+size_t ZBUFF_compressContinue(ZBUFF_CCtx* zbc,
+                              void* dst, size_t* dstCapacityPtr,
+                        const void* src, size_t* srcSizePtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    ZSTD_inBuffer inBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    inBuff.src = src;
+    inBuff.pos = 0;
+    inBuff.size = *srcSizePtr;
+    result = ZSTD_compressStream(zbc, &outBuff, &inBuff);
+    *dstCapacityPtr = outBuff.pos;
+    *srcSizePtr = inBuff.pos;
+    return result;
+}
+
+
+
+/* ======   Finalize   ====== */
+
+size_t ZBUFF_compressFlush(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    result = ZSTD_flushStream(zbc, &outBuff);
+    *dstCapacityPtr = outBuff.pos;
+    return result;
+}
+
+
+size_t ZBUFF_compressEnd(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
+{
+    size_t result;
+    ZSTD_outBuffer outBuff;
+    outBuff.dst = dst;
+    outBuff.pos = 0;
+    outBuff.size = *dstCapacityPtr;
+    result = ZSTD_endStream(zbc, &outBuff);
+    *dstCapacityPtr = outBuff.pos;
+    return result;
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFF_recommendedCInSize(void)  { return ZSTD_CStreamInSize(); }
+size_t ZBUFF_recommendedCOutSize(void) { return ZSTD_CStreamOutSize(); }
diff --git a/vendor/github.com/DataDog/zstd/zbuff_decompress.c b/vendor/github.com/DataDog/zstd/zbuff_decompress.c
new file mode 100644
index 0000000..923c22b
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zbuff_decompress.c
@@ -0,0 +1,75 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/* *************************************
+*  Dependencies
+***************************************/
+#define ZBUFF_STATIC_LINKING_ONLY
+#include "zbuff.h"
+
+
+ZBUFF_DCtx* ZBUFF_createDCtx(void)
+{
+    return ZSTD_createDStream();
+}
+
+ZBUFF_DCtx* ZBUFF_createDCtx_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDStream_advanced(customMem);
+}
+
+size_t ZBUFF_freeDCtx(ZBUFF_DCtx* zbd)
+{
+    return ZSTD_freeDStream(zbd);
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFF_decompressInitDictionary(ZBUFF_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    return ZSTD_initDStream_usingDict(zbd, dict, dictSize);
+}
+
+size_t ZBUFF_decompressInit(ZBUFF_DCtx* zbd)
+{
+    return ZSTD_initDStream(zbd);
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFF_decompressContinue(ZBUFF_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    ZSTD_outBuffer outBuff;
+    ZSTD_inBuffer inBuff;
+    size_t result;
+    outBuff.dst  = dst;
+    outBuff.pos  = 0;
+    outBuff.size = *dstCapacityPtr;
+    inBuff.src  = src;
+    inBuff.pos  = 0;
+    inBuff.size = *srcSizePtr;
+    result = ZSTD_decompressStream(zbd, &outBuff, &inBuff);
+    *dstCapacityPtr = outBuff.pos;
+    *srcSizePtr = inBuff.pos;
+    return result;
+}
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFF_recommendedDInSize(void)  { return ZSTD_DStreamInSize(); }
+size_t ZBUFF_recommendedDOutSize(void) { return ZSTD_DStreamOutSize(); }
diff --git a/vendor/github.com/DataDog/zstd/zdict.c b/vendor/github.com/DataDog/zstd/zdict.c
new file mode 100644
index 0000000..7d24e49
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zdict.c
@@ -0,0 +1,1108 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*-**************************************
+*  Tuning parameters
+****************************************/
+#define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
+#define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
+#define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
+
+
+/*-**************************************
+*  Compiler Options
+****************************************/
+/* Unix Large Files support (>4GB) */
+#define _FILE_OFFSET_BITS 64
+#if (defined(__sun__) && (!defined(__LP64__)))   /* Sun Solaris 32-bits requires specific definitions */
+#  define _LARGEFILE_SOURCE
+#elif ! defined(__LP64__)                        /* No point defining Large file for 64 bit */
+#  define _LARGEFILE64_SOURCE
+#endif
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdlib.h>        /* malloc, free */
+#include <string.h>        /* memset */
+#include <stdio.h>         /* fprintf, fopen, ftello64 */
+#include <time.h>          /* clock */
+
+#include "mem.h"           /* read */
+#include "fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"           /* HUF_buildCTable, HUF_writeCTable */
+#include "zstd_internal.h" /* includes zstd.h */
+#include "xxhash.h"        /* XXH64 */
+#include "divsufsort.h"
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#  define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define DICTLISTSIZE_DEFAULT 10000
+
+#define NOISELENGTH 32
+
+static const int g_compressionLevel_default = 3;
+static const U32 g_selectivity_default = 9;
+
+
+/*-*************************************
+*  Console display
+***************************************/
+#define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
+#define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); }    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+
+static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
+
+static void ZDICT_printHex(const void* ptr, size_t length)
+{
+    const BYTE* const b = (const BYTE*)ptr;
+    size_t u;
+    for (u=0; u<length; u++) {
+        BYTE c = b[u];
+        if (c<32 || c>126) c = '.';   /* non-printable char */
+        DISPLAY("%c", c);
+    }
+}
+
+
+/*-********************************************************
+*  Helper functions
+**********************************************************/
+unsigned ZDICT_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZDICT_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dictBuffer + 4);
+}
+
+
+/*-********************************************************
+*  Dictionary training functions
+**********************************************************/
+static unsigned ZDICT_NbCommonBytes (size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanForward64( &r, (U64)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctzll((U64)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
+            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r=0;
+            _BitScanForward( &r, (U32)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctz((U32)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
+            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#       endif
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanReverse64( &r, val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clzll(val) >> 3);
+#       else
+            unsigned r;
+            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
+            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r = 0;
+            _BitScanReverse( &r, (unsigned long)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clz((U32)val) >> 3);
+#       else
+            unsigned r;
+            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+    }   }
+}
+
+
+/*! ZDICT_count() :
+    Count the nb of common bytes between 2 pointers.
+    Note : this function presumes end of buffer followed by noisy guard band.
+*/
+static size_t ZDICT_count(const void* pIn, const void* pMatch)
+{
+    const char* const pStart = (const char*)pIn;
+    for (;;) {
+        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+        if (!diff) {
+            pIn = (const char*)pIn+sizeof(size_t);
+            pMatch = (const char*)pMatch+sizeof(size_t);
+            continue;
+        }
+        pIn = (const char*)pIn+ZDICT_NbCommonBytes(diff);
+        return (size_t)((const char*)pIn - pStart);
+    }
+}
+
+
+typedef struct {
+    U32 pos;
+    U32 length;
+    U32 savings;
+} dictItem;
+
+static void ZDICT_initDictItem(dictItem* d)
+{
+    d->pos = 1;
+    d->length = 0;
+    d->savings = (U32)(-1);
+}
+
+
+#define LLIMIT 64          /* heuristic determined experimentally */
+#define MINMATCHLENGTH 7   /* heuristic determined experimentally */
+static dictItem ZDICT_analyzePos(
+                       BYTE* doneMarks,
+                       const int* suffix, U32 start,
+                       const void* buffer, U32 minRatio, U32 notificationLevel)
+{
+    U32 lengthList[LLIMIT] = {0};
+    U32 cumulLength[LLIMIT] = {0};
+    U32 savings[LLIMIT] = {0};
+    const BYTE* b = (const BYTE*)buffer;
+    size_t maxLength = LLIMIT;
+    size_t pos = suffix[start];
+    U32 end = start;
+    dictItem solution;
+
+    /* init */
+    memset(&solution, 0, sizeof(solution));
+    doneMarks[pos] = 1;
+
+    /* trivial repetition cases */
+    if ( (MEM_read16(b+pos+0) == MEM_read16(b+pos+2))
+       ||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
+       ||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
+        /* skip and mark segment */
+        U16 const pattern16 = MEM_read16(b+pos+4);
+        U32 u, patternEnd = 6;
+        while (MEM_read16(b+pos+patternEnd) == pattern16) patternEnd+=2 ;
+        if (b[pos+patternEnd] == b[pos+patternEnd-1]) patternEnd++;
+        for (u=1; u<patternEnd; u++)
+            doneMarks[pos+u] = 1;
+        return solution;
+    }
+
+    /* look forward */
+    {   size_t length;
+        do {
+            end++;
+            length = ZDICT_count(b + pos, b + suffix[end]);
+        } while (length >= MINMATCHLENGTH);
+    }
+
+    /* look backward */
+    {   size_t length;
+        do {
+            length = ZDICT_count(b + pos, b + *(suffix+start-1));
+            if (length >=MINMATCHLENGTH) start--;
+        } while(length >= MINMATCHLENGTH);
+    }
+
+    /* exit if not found a minimum nb of repetitions */
+    if (end-start < minRatio) {
+        U32 idx;
+        for(idx=start; idx<end; idx++)
+            doneMarks[suffix[idx]] = 1;
+        return solution;
+    }
+
+    {   int i;
+        U32 searchLength;
+        U32 refinedStart = start;
+        U32 refinedEnd = end;
+
+        DISPLAYLEVEL(4, "\n");
+        DISPLAYLEVEL(4, "found %3u matches of length >= %i at pos %7u  ", (U32)(end-start), MINMATCHLENGTH, (U32)pos);
+        DISPLAYLEVEL(4, "\n");
+
+        for (searchLength = MINMATCHLENGTH ; ; searchLength++) {
+            BYTE currentChar = 0;
+            U32 currentCount = 0;
+            U32 currentID = refinedStart;
+            U32 id;
+            U32 selectedCount = 0;
+            U32 selectedID = currentID;
+            for (id =refinedStart; id < refinedEnd; id++) {
+                if (b[suffix[id] + searchLength] != currentChar) {
+                    if (currentCount > selectedCount) {
+                        selectedCount = currentCount;
+                        selectedID = currentID;
+                    }
+                    currentID = id;
+                    currentChar = b[ suffix[id] + searchLength];
+                    currentCount = 0;
+                }
+                currentCount ++;
+            }
+            if (currentCount > selectedCount) {  /* for last */
+                selectedCount = currentCount;
+                selectedID = currentID;
+            }
+
+            if (selectedCount < minRatio)
+                break;
+            refinedStart = selectedID;
+            refinedEnd = refinedStart + selectedCount;
+        }
+
+        /* evaluate gain based on new ref */
+        start = refinedStart;
+        pos = suffix[refinedStart];
+        end = start;
+        memset(lengthList, 0, sizeof(lengthList));
+
+        /* look forward */
+        {   size_t length;
+            do {
+                end++;
+                length = ZDICT_count(b + pos, b + suffix[end]);
+                if (length >= LLIMIT) length = LLIMIT-1;
+                lengthList[length]++;
+            } while (length >=MINMATCHLENGTH);
+        }
+
+        /* look backward */
+        {   size_t length = MINMATCHLENGTH;
+            while ((length >= MINMATCHLENGTH) & (start > 0)) {
+                length = ZDICT_count(b + pos, b + suffix[start - 1]);
+                if (length >= LLIMIT) length = LLIMIT - 1;
+                lengthList[length]++;
+                if (length >= MINMATCHLENGTH) start--;
+            }
+        }
+
+        /* largest useful length */
+        memset(cumulLength, 0, sizeof(cumulLength));
+        cumulLength[maxLength-1] = lengthList[maxLength-1];
+        for (i=(int)(maxLength-2); i>=0; i--)
+            cumulLength[i] = cumulLength[i+1] + lengthList[i];
+
+        for (i=LLIMIT-1; i>=MINMATCHLENGTH; i--) if (cumulLength[i]>=minRatio) break;
+        maxLength = i;
+
+        /* reduce maxLength in case of final into repetitive data */
+        {   U32 l = (U32)maxLength;
+            BYTE const c = b[pos + maxLength-1];
+            while (b[pos+l-2]==c) l--;
+            maxLength = l;
+        }
+        if (maxLength < MINMATCHLENGTH) return solution;   /* skip : no long-enough solution */
+
+        /* calculate savings */
+        savings[5] = 0;
+        for (i=MINMATCHLENGTH; i<=(int)maxLength; i++)
+            savings[i] = savings[i-1] + (lengthList[i] * (i-3));
+
+        DISPLAYLEVEL(4, "Selected ref at position %u, of length %u : saves %u (ratio: %.2f)  \n",
+                     (U32)pos, (U32)maxLength, savings[maxLength], (double)savings[maxLength] / maxLength);
+
+        solution.pos = (U32)pos;
+        solution.length = (U32)maxLength;
+        solution.savings = savings[maxLength];
+
+        /* mark positions done */
+        {   U32 id;
+            for (id=start; id<end; id++) {
+                U32 p, pEnd, length;
+                U32 const testedPos = suffix[id];
+                if (testedPos == pos)
+                    length = solution.length;
+                else {
+                    length = (U32)ZDICT_count(b+pos, b+testedPos);
+                    if (length > solution.length) length = solution.length;
+                }
+                pEnd = (U32)(testedPos + length);
+                for (p=testedPos; p<pEnd; p++)
+                    doneMarks[p] = 1;
+    }   }   }
+
+    return solution;
+}
+
+
+static int isIncluded(const void* in, const void* container, size_t length)
+{
+    const char* const ip = (const char*) in;
+    const char* const into = (const char*) container;
+    size_t u;
+
+    for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
+        if (ip[u] != into[u]) break;
+    }
+
+    return u==length;
+}
+
+/*! ZDICT_tryMerge() :
+    check if dictItem can be merged, do it if possible
+    @return : id of destination elt, 0 if not merged
+*/
+static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
+{
+    const U32 tableSize = table->pos;
+    const U32 eltEnd = elt.pos + elt.length;
+    const char* const buf = (const char*) buffer;
+
+    /* tail overlap */
+    U32 u; for (u=1; u<tableSize; u++) {
+        if (u==eltNbToSkip) continue;
+        if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
+            /* append */
+            U32 const addedLength = table[u].pos - elt.pos;
+            table[u].length += addedLength;
+            table[u].pos = elt.pos;
+            table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            elt = table[u];
+            /* sort : improve rank */
+            while ((u>1) && (table[u-1].savings < elt.savings))
+            table[u] = table[u-1], u--;
+            table[u] = elt;
+            return u;
+    }   }
+
+    /* front overlap */
+    for (u=1; u<tableSize; u++) {
+        if (u==eltNbToSkip) continue;
+
+        if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
+            /* append */
+            int const addedLength = (int)eltEnd - (table[u].pos + table[u].length);
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            if (addedLength > 0) {   /* otherwise, elt fully included into existing */
+                table[u].length += addedLength;
+                table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
+            }
+            /* sort : improve rank */
+            elt = table[u];
+            while ((u>1) && (table[u-1].savings < elt.savings))
+                table[u] = table[u-1], u--;
+            table[u] = elt;
+            return u;
+        }
+
+        if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
+            if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
+                size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 );
+                table[u].pos = elt.pos;
+                table[u].savings += (U32)(elt.savings * addedLength / elt.length);
+                table[u].length = MIN(elt.length, table[u].length + 1);
+                return u;
+            }
+        }
+    }
+
+    return 0;
+}
+
+
+static void ZDICT_removeDictItem(dictItem* table, U32 id)
+{
+    /* convention : table[0].pos stores nb of elts */
+    U32 const max = table[0].pos;
+    U32 u;
+    if (!id) return;   /* protection, should never happen */
+    for (u=id; u<max-1; u++)
+        table[u] = table[u+1];
+    table->pos--;
+}
+
+
+static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
+{
+    /* merge if possible */
+    U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
+    if (mergeId) {
+        U32 newMerge = 1;
+        while (newMerge) {
+            newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
+            if (newMerge) ZDICT_removeDictItem(table, mergeId);
+            mergeId = newMerge;
+        }
+        return;
+    }
+
+    /* insert */
+    {   U32 current;
+        U32 nextElt = table->pos;
+        if (nextElt >= maxSize) nextElt = maxSize-1;
+        current = nextElt-1;
+        while (table[current].savings < elt.savings) {
+            table[current+1] = table[current];
+            current--;
+        }
+        table[current+1] = elt;
+        table->pos = nextElt+1;
+    }
+}
+
+
+static U32 ZDICT_dictSize(const dictItem* dictList)
+{
+    U32 u, dictSize = 0;
+    for (u=1; u<dictList[0].pos; u++)
+        dictSize += dictList[u].length;
+    return dictSize;
+}
+
+
+static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
+                            const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
+                            const size_t* fileSizes, unsigned nbFiles,
+                            U32 minRatio, U32 notificationLevel)
+{
+    int* const suffix0 = (int*)malloc((bufferSize+2)*sizeof(*suffix0));
+    int* const suffix = suffix0+1;
+    U32* reverseSuffix = (U32*)malloc((bufferSize)*sizeof(*reverseSuffix));
+    BYTE* doneMarks = (BYTE*)malloc((bufferSize+16)*sizeof(*doneMarks));   /* +16 for overflow security */
+    U32* filePos = (U32*)malloc(nbFiles * sizeof(*filePos));
+    size_t result = 0;
+    clock_t displayClock = 0;
+    clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
+
+#   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
+            if (ZDICT_clockSpan(displayClock) > refreshRate)  \
+            { displayClock = clock(); DISPLAY(__VA_ARGS__); \
+            if (notificationLevel>=4) fflush(stderr); } }
+
+    /* init */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    if (!suffix0 || !reverseSuffix || !doneMarks || !filePos) {
+        result = ERROR(memory_allocation);
+        goto _cleanup;
+    }
+    if (minRatio < MINRATIO) minRatio = MINRATIO;
+    memset(doneMarks, 0, bufferSize+16);
+
+    /* limit sample set size (divsufsort limitation)*/
+    if (bufferSize > ZDICT_MAX_SAMPLES_SIZE) DISPLAYLEVEL(3, "sample set too large : reduced to %u MB ...\n", (U32)(ZDICT_MAX_SAMPLES_SIZE>>20));
+    while (bufferSize > ZDICT_MAX_SAMPLES_SIZE) bufferSize -= fileSizes[--nbFiles];
+
+    /* sort */
+    DISPLAYLEVEL(2, "sorting %u files of total size %u MB ...\n", nbFiles, (U32)(bufferSize>>20));
+    {   int const divSuftSortResult = divsufsort((const unsigned char*)buffer, suffix, (int)bufferSize, 0);
+        if (divSuftSortResult != 0) { result = ERROR(GENERIC); goto _cleanup; }
+    }
+    suffix[bufferSize] = (int)bufferSize;   /* leads into noise */
+    suffix0[0] = (int)bufferSize;           /* leads into noise */
+    /* build reverse suffix sort */
+    {   size_t pos;
+        for (pos=0; pos < bufferSize; pos++)
+            reverseSuffix[suffix[pos]] = (U32)pos;
+        /* note filePos tracks borders between samples.
+           It's not used at this stage, but planned to become useful in a later update */
+        filePos[0] = 0;
+        for (pos=1; pos<nbFiles; pos++)
+            filePos[pos] = (U32)(filePos[pos-1] + fileSizes[pos-1]);
+    }
+
+    DISPLAYLEVEL(2, "finding patterns ... \n");
+    DISPLAYLEVEL(3, "minimum ratio : %u \n", minRatio);
+
+    {   U32 cursor; for (cursor=0; cursor < bufferSize; ) {
+            dictItem solution;
+            if (doneMarks[cursor]) { cursor++; continue; }
+            solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
+            if (solution.length==0) { cursor++; continue; }
+            ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
+            cursor += solution.length;
+            DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / bufferSize * 100);
+    }   }
+
+_cleanup:
+    free(suffix0);
+    free(reverseSuffix);
+    free(doneMarks);
+    free(filePos);
+    return result;
+}
+
+
+static void ZDICT_fillNoise(void* buffer, size_t length)
+{
+    unsigned const prime1 = 2654435761U;
+    unsigned const prime2 = 2246822519U;
+    unsigned acc = prime1;
+    size_t p=0;;
+    for (p=0; p<length; p++) {
+        acc *= prime2;
+        ((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
+    }
+}
+
+
+typedef struct
+{
+    ZSTD_CCtx* ref;    /* contains reference to dictionary */
+    ZSTD_CCtx* zc;     /* working context */
+    void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
+} EStats_ress_t;
+
+#define MAXREPOFFSET 1024
+
+static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
+                              U32* countLit, U32* offsetcodeCount, U32* matchlengthCount, U32* litlengthCount, U32* repOffsets,
+                              const void* src, size_t srcSize,
+                              U32 notificationLevel)
+{
+    size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params.cParams.windowLog);
+    size_t cSize;
+
+    if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
+    {   size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref, 0);
+        if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
+    }
+    cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
+    if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
+
+    if (cSize) {  /* if == 0; block is not compressible */
+        const seqStore_t* const seqStorePtr = ZSTD_getSeqStore(esr.zc);
+
+        /* literals stats */
+        {   const BYTE* bytePtr;
+            for(bytePtr = seqStorePtr->litStart; bytePtr < seqStorePtr->lit; bytePtr++)
+                countLit[*bytePtr]++;
+        }
+
+        /* seqStats */
+        {   U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+            ZSTD_seqToCodes(seqStorePtr);
+
+            {   const BYTE* codePtr = seqStorePtr->ofCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) offsetcodeCount[codePtr[u]]++;
+            }
+
+            {   const BYTE* codePtr = seqStorePtr->mlCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) matchlengthCount[codePtr[u]]++;
+            }
+
+            {   const BYTE* codePtr = seqStorePtr->llCode;
+                U32 u;
+                for (u=0; u<nbSeq; u++) litlengthCount[codePtr[u]]++;
+            }
+
+            if (nbSeq >= 2) { /* rep offsets */
+                const seqDef* const seq = seqStorePtr->sequencesStart;
+                U32 offset1 = seq[0].offset - 3;
+                U32 offset2 = seq[1].offset - 3;
+                if (offset1 >= MAXREPOFFSET) offset1 = 0;
+                if (offset2 >= MAXREPOFFSET) offset2 = 0;
+                repOffsets[offset1] += 3;
+                repOffsets[offset2] += 1;
+    }   }   }
+}
+
+static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles)
+{
+    size_t total=0;
+    unsigned u;
+    for (u=0; u<nbFiles; u++) total += fileSizes[u];
+    return total;
+}
+
+typedef struct { U32 offset; U32 count; } offsetCount_t;
+
+static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val, U32 count)
+{
+    U32 u;
+    table[ZSTD_REP_NUM].offset = val;
+    table[ZSTD_REP_NUM].count = count;
+    for (u=ZSTD_REP_NUM; u>0; u--) {
+        offsetCount_t tmp;
+        if (table[u-1].count >= table[u].count) break;
+        tmp = table[u-1];
+        table[u-1] = table[u];
+        table[u] = tmp;
+    }
+}
+
+/* ZDICT_flatLit() :
+ * rewrite `countLit` to contain a mostly flat but still compressible distribution of literals.
+ * necessary to avoid generating a non-compressible distribution that HUF_writeCTable() cannot encode.
+ */
+static void ZDICT_flatLit(U32* countLit)
+{
+    int u;
+    for (u=1; u<256; u++) countLit[u] = 2;
+    countLit[0]   = 4;
+    countLit[253] = 1;
+    countLit[254] = 1;
+}
+
+#define OFFCODE_MAX 30  /* only applicable to first block */
+static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
+                                   unsigned compressionLevel,
+                             const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
+                             const void* dictBuffer, size_t  dictBufferSize,
+                                   unsigned notificationLevel)
+{
+    U32 countLit[256];
+    HUF_CREATE_STATIC_CTABLE(hufTable, 255);
+    U32 offcodeCount[OFFCODE_MAX+1];
+    short offcodeNCount[OFFCODE_MAX+1];
+    U32 offcodeMax = ZSTD_highbit32((U32)(dictBufferSize + 128 KB));
+    U32 matchLengthCount[MaxML+1];
+    short matchLengthNCount[MaxML+1];
+    U32 litLengthCount[MaxLL+1];
+    short litLengthNCount[MaxLL+1];
+    U32 repOffset[MAXREPOFFSET];
+    offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
+    EStats_ress_t esr;
+    ZSTD_parameters params;
+    U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
+    size_t pos = 0, errorCode;
+    size_t eSize = 0;
+    size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles);
+    size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles);
+    BYTE* dstPtr = (BYTE*)dstBuffer;
+
+    /* init */
+    DEBUGLOG(4, "ZDICT_analyzeEntropy");
+    esr.ref = ZSTD_createCCtx();
+    esr.zc = ZSTD_createCCtx();
+    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
+    if (!esr.ref || !esr.zc || !esr.workPlace) {
+        eSize = ERROR(memory_allocation);
+        DISPLAYLEVEL(1, "Not enough memory \n");
+        goto _cleanup;
+    }
+    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
+    for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
+    for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
+    for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
+    for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
+    memset(repOffset, 0, sizeof(repOffset));
+    repOffset[1] = repOffset[4] = repOffset[8] = 1;
+    memset(bestRepOffset, 0, sizeof(bestRepOffset));
+    if (compressionLevel<=0) compressionLevel = g_compressionLevel_default;
+    params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
+    {   size_t const beginResult = ZSTD_compressBegin_advanced(esr.ref, dictBuffer, dictBufferSize, params, 0);
+        if (ZSTD_isError(beginResult)) {
+            DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced() failed : %s \n", ZSTD_getErrorName(beginResult));
+            eSize = ERROR(GENERIC);
+            goto _cleanup;
+    }   }
+
+    /* collect stats on all samples */
+    for (u=0; u<nbFiles; u++) {
+        ZDICT_countEStats(esr, params,
+                          countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
+                         (const char*)srcBuffer + pos, fileSizes[u],
+                          notificationLevel);
+        pos += fileSizes[u];
+    }
+
+    /* analyze, build stats, starting with literals */
+    {   size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
+        if (HUF_isError(maxNbBits)) {
+            eSize = ERROR(GENERIC);
+            DISPLAYLEVEL(1, " HUF_buildCTable error \n");
+            goto _cleanup;
+        }
+        if (maxNbBits==8) {  /* not compressible : will fail on HUF_writeCTable() */
+            DISPLAYLEVEL(2, "warning : pathological dataset : literals are not compressible : samples are noisy or too regular \n");
+            ZDICT_flatLit(countLit);  /* replace distribution by a fake "mostly flat but still compressible" distribution, that HUF_writeCTable() can encode */
+            maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
+            assert(maxNbBits==9);
+        }
+        huffLog = (U32)maxNbBits;
+    }
+
+    /* looking for most common first offsets */
+    {   U32 offset;
+        for (offset=1; offset<MAXREPOFFSET; offset++)
+            ZDICT_insertSortCount(bestRepOffset, offset, repOffset[offset]);
+    }
+    /* note : the result of this phase should be used to better appreciate the impact on statistics */
+
+    total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
+    errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax);
+    if (FSE_isError(errorCode)) {
+        eSize = ERROR(GENERIC);
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
+        goto _cleanup;
+    }
+    Offlog = (U32)errorCode;
+
+    total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
+    errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML);
+    if (FSE_isError(errorCode)) {
+        eSize = ERROR(GENERIC);
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
+        goto _cleanup;
+    }
+    mlLog = (U32)errorCode;
+
+    total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
+    errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL);
+    if (FSE_isError(errorCode)) {
+        eSize = ERROR(GENERIC);
+        DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
+        goto _cleanup;
+    }
+    llLog = (U32)errorCode;
+
+    /* write result to buffer */
+    {   size_t const hhSize = HUF_writeCTable(dstPtr, maxDstSize, hufTable, 255, huffLog);
+        if (HUF_isError(hhSize)) {
+            eSize = ERROR(GENERIC);
+            DISPLAYLEVEL(1, "HUF_writeCTable error \n");
+            goto _cleanup;
+        }
+        dstPtr += hhSize;
+        maxDstSize -= hhSize;
+        eSize += hhSize;
+    }
+
+    {   size_t const ohSize = FSE_writeNCount(dstPtr, maxDstSize, offcodeNCount, OFFCODE_MAX, Offlog);
+        if (FSE_isError(ohSize)) {
+            eSize = ERROR(GENERIC);
+            DISPLAYLEVEL(1, "FSE_writeNCount error with offcodeNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += ohSize;
+        maxDstSize -= ohSize;
+        eSize += ohSize;
+    }
+
+    {   size_t const mhSize = FSE_writeNCount(dstPtr, maxDstSize, matchLengthNCount, MaxML, mlLog);
+        if (FSE_isError(mhSize)) {
+            eSize = ERROR(GENERIC);
+            DISPLAYLEVEL(1, "FSE_writeNCount error with matchLengthNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += mhSize;
+        maxDstSize -= mhSize;
+        eSize += mhSize;
+    }
+
+    {   size_t const lhSize = FSE_writeNCount(dstPtr, maxDstSize, litLengthNCount, MaxLL, llLog);
+        if (FSE_isError(lhSize)) {
+            eSize = ERROR(GENERIC);
+            DISPLAYLEVEL(1, "FSE_writeNCount error with litlengthNCount \n");
+            goto _cleanup;
+        }
+        dstPtr += lhSize;
+        maxDstSize -= lhSize;
+        eSize += lhSize;
+    }
+
+    if (maxDstSize<12) {
+        eSize = ERROR(GENERIC);
+        DISPLAYLEVEL(1, "not enough space to write RepOffsets \n");
+        goto _cleanup;
+    }
+# if 0
+    MEM_writeLE32(dstPtr+0, bestRepOffset[0].offset);
+    MEM_writeLE32(dstPtr+4, bestRepOffset[1].offset);
+    MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
+#else
+    /* at this stage, we don't use the result of "most common first offset",
+       as the impact of statistics is not properly evaluated */
+    MEM_writeLE32(dstPtr+0, repStartValue[0]);
+    MEM_writeLE32(dstPtr+4, repStartValue[1]);
+    MEM_writeLE32(dstPtr+8, repStartValue[2]);
+#endif
+    eSize += 12;
+
+_cleanup:
+    ZSTD_freeCCtx(esr.ref);
+    ZSTD_freeCCtx(esr.zc);
+    free(esr.workPlace);
+
+    return eSize;
+}
+
+
+
+size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
+                          const void* customDictContent, size_t dictContentSize,
+                          const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                          ZDICT_params_t params)
+{
+    size_t hSize;
+#define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
+    BYTE header[HBUFFSIZE];
+    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+
+    /* check conditions */
+    DEBUGLOG(4, "ZDICT_finalizeDictionary");
+    if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
+    if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
+
+    /* dictionary header */
+    MEM_writeLE32(header, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32(header+4, dictID);
+    }
+    hSize = 8;
+
+    /* entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy(header+hSize, HBUFFSIZE-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  customDictContent, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* copy elements in final buffer ; note : src and dst buffer can overlap */
+    if (hSize + dictContentSize > dictBufferCapacity) dictContentSize = dictBufferCapacity - hSize;
+    {   size_t const dictSize = hSize + dictContentSize;
+        char* dictEnd = (char*)dictBuffer + dictSize;
+        memmove(dictEnd - dictContentSize, customDictContent, dictContentSize);
+        memcpy(dictBuffer, header, hSize);
+        return dictSize;
+    }
+}
+
+
+size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                                                 ZDICT_params_t params)
+{
+    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+    size_t hSize = 8;
+
+    /* calculate entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* add dictionary header (after entropy tables) */
+    MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY);
+    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32((char*)dictBuffer+4, dictID);
+    }
+
+    if (hSize + dictContentSize < dictBufferCapacity)
+        memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
+    return MIN(dictBufferCapacity, hSize+dictContentSize);
+}
+
+
+/*! ZDICT_trainFromBuffer_unsafe_legacy() :
+*   Warning : `samplesBuffer` must be followed by noisy guard band.
+*   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
+*/
+size_t ZDICT_trainFromBuffer_unsafe_legacy(
+                            void* dictBuffer, size_t maxDictSize,
+                            const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                            ZDICT_legacy_params_t params)
+{
+    U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16));
+    dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
+    unsigned const selectivity = params.selectivityLevel == 0 ? g_selectivity_default : params.selectivityLevel;
+    unsigned const minRep = (selectivity > 30) ? MINRATIO : nbSamples >> selectivity;
+    size_t const targetDictSize = maxDictSize;
+    size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
+    size_t dictSize = 0;
+    U32 const notificationLevel = params.zParams.notificationLevel;
+
+    /* checks */
+    if (!dictList) return ERROR(memory_allocation);
+    if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
+    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
+
+    /* init */
+    ZDICT_initDictItem(dictList);
+
+    /* build dictionary */
+    ZDICT_trainBuffer_legacy(dictList, dictListSize,
+                       samplesBuffer, samplesBuffSize,
+                       samplesSizes, nbSamples,
+                       minRep, notificationLevel);
+
+    /* display best matches */
+    if (params.zParams.notificationLevel>= 3) {
+        U32 const nb = MIN(25, dictList[0].pos);
+        U32 const dictContentSize = ZDICT_dictSize(dictList);
+        U32 u;
+        DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", dictList[0].pos-1, dictContentSize);
+        DISPLAYLEVEL(3, "list %u best segments \n", nb-1);
+        for (u=1; u<nb; u++) {
+            U32 const pos = dictList[u].pos;
+            U32 const length = dictList[u].length;
+            U32 const printedLength = MIN(40, length);
+            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize))
+                return ERROR(GENERIC);   /* should never happen */
+            DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
+                         u, length, pos, dictList[u].savings);
+            ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
+            DISPLAYLEVEL(3, "| \n");
+    }   }
+
+
+    /* create dictionary */
+    {   U32 dictContentSize = ZDICT_dictSize(dictList);
+        if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
+        if (dictContentSize < targetDictSize/4) {
+            DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (U32)maxDictSize);
+            if (samplesBuffSize < 10 * targetDictSize)
+                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (U32)(samplesBuffSize>>20));
+            if (minRep > MINRATIO) {
+                DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
+                DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
+            }
+        }
+
+        if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
+            U32 proposedSelectivity = selectivity-1;
+            while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
+            DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (U32)maxDictSize);
+            DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
+            DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
+        }
+
+        /* limit dictionary size */
+        {   U32 const max = dictList->pos;   /* convention : nb of useful elts within dictList */
+            U32 currentSize = 0;
+            U32 n; for (n=1; n<max; n++) {
+                currentSize += dictList[n].length;
+                if (currentSize > targetDictSize) { currentSize -= dictList[n].length; break; }
+            }
+            dictList->pos = n;
+            dictContentSize = currentSize;
+        }
+
+        /* build dict content */
+        {   U32 u;
+            BYTE* ptr = (BYTE*)dictBuffer + maxDictSize;
+            for (u=1; u<dictList->pos; u++) {
+                U32 l = dictList[u].length;
+                ptr -= l;
+                if (ptr<(BYTE*)dictBuffer) { free(dictList); return ERROR(GENERIC); }   /* should not happen */
+                memcpy(ptr, (const char*)samplesBuffer+dictList[u].pos, l);
+        }   }
+
+        dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
+                                                             samplesBuffer, samplesSizes, nbSamples,
+                                                             params.zParams);
+    }
+
+    /* clean up */
+    free(dictList);
+    return dictSize;
+}
+
+
+/* ZDICT_trainFromBuffer_legacy() :
+ * issue : samplesBuffer need to be followed by a noisy guard band.
+ * work around : duplicate the buffer, and add the noise */
+size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
+                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                              ZDICT_legacy_params_t params)
+{
+    size_t result;
+    void* newBuff;
+    size_t const sBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
+    if (sBuffSize < ZDICT_MIN_SAMPLES_SIZE) return 0;   /* not enough content => no dictionary */
+
+    newBuff = malloc(sBuffSize + NOISELENGTH);
+    if (!newBuff) return ERROR(memory_allocation);
+
+    memcpy(newBuff, samplesBuffer, sBuffSize);
+    ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH);   /* guard band, for end of buffer condition */
+
+    result =
+        ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff,
+                                            samplesSizes, nbSamples, params);
+    free(newBuff);
+    return result;
+}
+
+
+size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                             const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+{
+    ZDICT_cover_params_t params;
+    DEBUGLOG(3, "ZDICT_trainFromBuffer");
+    memset(&params, 0, sizeof(params));
+    params.d = 8;
+    params.steps = 4;
+    /* Default to level 6 since no compression level information is available */
+    params.zParams.compressionLevel = 6;
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
+    params.zParams.notificationLevel = ZSTD_DEBUG;
+#endif
+    return ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, dictBufferCapacity,
+                                               samplesBuffer, samplesSizes, nbSamples,
+                                               &params);
+}
+
+size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
+{
+    ZDICT_params_t params;
+    memset(&params, 0, sizeof(params));
+    return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity,
+                                                     samplesBuffer, samplesSizes, nbSamples,
+                                                     params);
+}
diff --git a/vendor/github.com/DataDog/zstd/zdict.h b/vendor/github.com/DataDog/zstd/zdict.h
new file mode 100644
index 0000000..ad459c2
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zdict.h
@@ -0,0 +1,212 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef DICTBUILDER_H_001
+#define DICTBUILDER_H_001
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*======  Dependencies  ======*/
+#include <stddef.h>  /* size_t */
+
+
+/* =====   ZDICTLIB_API : control library symbols visibility   ===== */
+#ifndef ZDICTLIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define ZDICTLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define ZDICTLIB_VISIBILITY
+#  endif
+#endif
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBILITY
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZDICTLIB_API ZDICTLIB_VISIBILITY
+#endif
+
+
+/*! ZDICT_trainFromBuffer():
+ *  Train a dictionary from an array of samples.
+ *  Redirect towards ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note: ZDICT_trainFromBuffer() requires about 9 bytes of memory for each input byte.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                                    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+
+
+/*======   Helper functions   ======*/
+ZDICTLIB_API unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize);  /**< extracts dictID; @return zero if error (not a valid dictionary) */
+ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode);
+ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
+
+
+
+#ifdef ZDICT_STATIC_LINKING_ONLY
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+typedef struct {
+    int      compressionLevel;   /* optimize for a specific zstd compression level; 0 means default */
+    unsigned notificationLevel;  /* Write log to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
+    unsigned dictID;             /* force dictID value; 0 means auto mode (32-bits random value) */
+} ZDICT_params_t;
+
+/*! ZDICT_cover_params_t:
+ *  k and d are the only required parameters.
+ *  For others, value 0 means default.
+ */
+typedef struct {
+    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
+    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (32) : Higher means more parameters checked */
+    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    ZDICT_params_t zParams;
+} ZDICT_cover_params_t;
+
+
+/*! ZDICT_trainFromBuffer_cover():
+ *  Train a dictionary from an array of samples using the COVER algorithm.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
+          void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+          ZDICT_cover_params_t parameters);
+
+/*! ZDICT_optimizeTrainFromBuffer_cover():
+ * The same requirements as above hold for all the parameters except `parameters`.
+ * This function tries many parameter combinations and picks the best parameters.
+ * `*parameters` is filled with the best parameters found,
+ * dictionary constructed with those parameters is stored in `dictBuffer`.
+ *
+ * All of the parameters d, k, steps are optional.
+ * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
+ * if steps is zero it defaults to its default value.
+ * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *           or an error code, which can be tested with ZDICT_isError().
+ *           On success `*parameters` contains the parameters selected.
+ * Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
+ */
+ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
+          void* dictBuffer, size_t dictBufferCapacity,
+    const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+          ZDICT_cover_params_t* parameters);
+
+/*! ZDICT_finalizeDictionary():
+ * Given a custom content as a basis for dictionary, and a set of samples,
+ * finalize dictionary by adding headers and statistics.
+ *
+ * Samples must be stored concatenated in a flat buffer `samplesBuffer`,
+ * supplied with an array of sizes `samplesSizes`, providing the size of each sample in order.
+ *
+ * dictContentSize must be >= ZDICT_CONTENTSIZE_MIN bytes.
+ * maxDictSize must be >= dictContentSize, and must be >= ZDICT_DICTSIZE_MIN bytes.
+ *
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`),
+ *           or an error code, which can be tested by ZDICT_isError().
+ * Note: ZDICT_finalizeDictionary() will push notifications into stderr if instructed to, using notificationLevel>0.
+ * Note 2: dictBuffer and dictContent can overlap
+ */
+#define ZDICT_CONTENTSIZE_MIN 128
+#define ZDICT_DICTSIZE_MIN    256
+ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
+                                const void* dictContent, size_t dictContentSize,
+                                const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                                ZDICT_params_t parameters);
+
+typedef struct {
+    unsigned selectivityLevel;   /* 0 means default; larger => select more => larger dictionary */
+    ZDICT_params_t zParams;
+} ZDICT_legacy_params_t;
+
+/*! ZDICT_trainFromBuffer_legacy():
+ *  Train a dictionary from an array of samples.
+ *  Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+ *  supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+ *  The resulting dictionary will be saved into `dictBuffer`.
+ * `parameters` is optional and can be provided with values set to 0 to mean "default".
+ * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+ *          or an error code, which can be tested with ZDICT_isError().
+ *  Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
+ *        It's possible to select smaller or larger size, just by specifying `dictBufferCapacity`.
+ *        In general, it's recommended to provide a few thousands samples, though this can vary a lot.
+ *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+ *  Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
+ */
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
+    void *dictBuffer, size_t dictBufferCapacity,
+    const void *samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+    ZDICT_legacy_params_t parameters);
+
+/* Deprecation warnings */
+/* It is generally possible to disable deprecation warnings from compiler,
+   for example with -Wno-deprecated-declarations for gcc
+   or _CRT_SECURE_NO_WARNINGS in Visual.
+   Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
+#ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
+#  define ZDICT_DEPRECATED(message) ZDICTLIB_API   /* disable deprecation warnings */
+#else
+#  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
+#  elif (ZDICT_GCC_VERSION >= 405) || defined(__clang__)
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
+#  elif (ZDICT_GCC_VERSION >= 301)
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API
+#  endif
+#endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
+
+ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
+size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+
+
+#endif   /* ZDICT_STATIC_LINKING_ONLY */
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* DICTBUILDER_H_001 */
diff --git a/vendor/github.com/DataDog/zstd/zstd.go b/vendor/github.com/DataDog/zstd/zstd.go
new file mode 100644
index 0000000..ac3839f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd.go
@@ -0,0 +1,139 @@
+package zstd
+
+/*
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+#include "stdint.h"  // for uintptr_t
+
+// The following *_wrapper function are used for removing superflouos
+// memory allocations when calling the wrapped functions from Go code.
+// See https://github.com/golang/go/issues/24450 for details.
+
+static size_t ZSTD_compress_wrapper(uintptr_t dst, size_t maxDstSize, const uintptr_t src, size_t srcSize, int compressionLevel) {
+	return ZSTD_compress((void*)dst, maxDstSize, (const void*)src, srcSize, compressionLevel);
+}
+
+static size_t ZSTD_decompress_wrapper(uintptr_t dst, size_t maxDstSize, uintptr_t src, size_t srcSize) {
+	return ZSTD_decompress((void*)dst, maxDstSize, (const void *)src, srcSize);
+}
+
+*/
+import "C"
+import (
+	"bytes"
+	"errors"
+	"io/ioutil"
+	"unsafe"
+)
+
+// Defines best and standard values for zstd cli
+const (
+	BestSpeed          = 1
+	BestCompression    = 20
+	DefaultCompression = 5
+)
+
+var (
+	// ErrEmptySlice is returned when there is nothing to compress
+	ErrEmptySlice = errors.New("Bytes slice is empty")
+)
+
+// CompressBound returns the worst case size needed for a destination buffer,
+// which can be used to preallocate a destination buffer or select a previously
+// allocated buffer from a pool.
+// See zstd.h to mirror implementation of ZSTD_COMPRESSBOUND
+func CompressBound(srcSize int) int {
+	lowLimit := 128 << 10 // 128 kB
+	var margin int
+	if srcSize < lowLimit {
+		margin = (lowLimit - srcSize) >> 11
+	}
+	return srcSize + (srcSize >> 8) + margin
+}
+
+// cCompressBound is a cgo call to check the go implementation above against the c code.
+func cCompressBound(srcSize int) int {
+	return int(C.ZSTD_compressBound(C.size_t(srcSize)))
+}
+
+// Compress src into dst.  If you have a buffer to use, you can pass it to
+// prevent allocation.  If it is too small, or if nil is passed, a new buffer
+// will be allocated and returned.
+func Compress(dst, src []byte) ([]byte, error) {
+	return CompressLevel(dst, src, DefaultCompression)
+}
+
+// CompressLevel is the same as Compress but you can pass a compression level
+func CompressLevel(dst, src []byte, level int) ([]byte, error) {
+	if len(src) == 0 {
+		return []byte{}, ErrEmptySlice
+	}
+	bound := CompressBound(len(src))
+	if cap(dst) >= bound {
+		dst = dst[0:bound] // Reuse dst buffer
+	} else {
+		dst = make([]byte, bound)
+	}
+
+	cWritten := C.ZSTD_compress_wrapper(
+		C.uintptr_t(uintptr(unsafe.Pointer(&dst[0]))),
+		C.size_t(len(dst)),
+		C.uintptr_t(uintptr(unsafe.Pointer(&src[0]))),
+		C.size_t(len(src)),
+		C.int(level))
+
+	written := int(cWritten)
+	// Check if the return is an Error code
+	if err := getError(written); err != nil {
+		return nil, err
+	}
+	return dst[:written], nil
+}
+
+// Decompress src into dst.  If you have a buffer to use, you can pass it to
+// prevent allocation.  If it is too small, or if nil is passed, a new buffer
+// will be allocated and returned.
+func Decompress(dst, src []byte) ([]byte, error) {
+	decompress := func(dst, src []byte) ([]byte, error) {
+
+		cWritten := C.ZSTD_decompress_wrapper(
+			C.uintptr_t(uintptr(unsafe.Pointer(&dst[0]))),
+			C.size_t(len(dst)),
+			C.uintptr_t(uintptr(unsafe.Pointer(&src[0]))),
+			C.size_t(len(src)))
+
+		written := int(cWritten)
+		// Check error
+		if err := getError(written); err != nil {
+			return nil, err
+		}
+		return dst[:written], nil
+	}
+
+	if dst == nil {
+		// Attempt to use zStd to determine decompressed size (may result in error or 0)
+		size := int(C.size_t(C.ZSTD_getDecompressedSize(unsafe.Pointer(&src[0]), C.size_t(len(src)))))
+
+		if err := getError(size); err != nil {
+			return nil, err
+		}
+
+		if size > 0 {
+			dst = make([]byte, size)
+		} else {
+			dst = make([]byte, len(src)*3) // starting guess
+		}
+	}
+	for i := 0; i < 3; i++ { // 3 tries to allocate a bigger buffer
+		result, err := decompress(dst, src)
+		if !IsDstSizeTooSmallError(err) {
+			return result, err
+		}
+		dst = make([]byte, len(dst)*2) // Grow buffer by 2
+	}
+
+	// We failed getting a dst buffer of correct size, use stream API
+	r := NewReader(bytes.NewReader(src))
+	defer r.Close()
+	return ioutil.ReadAll(r)
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd.h b/vendor/github.com/DataDog/zstd/zstd.h
new file mode 100644
index 0000000..6405da6
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd.h
@@ -0,0 +1,1399 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#ifndef ZSTD_H_235446
+#define ZSTD_H_235446
+
+/* ======   Dependency   ======*/
+#include <stddef.h>   /* size_t */
+
+
+/* =====   ZSTDLIB_API : control library symbols visibility   ===== */
+#ifndef ZSTDLIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define ZSTDLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDLIB_VISIBILITY
+#  endif
+#endif
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBILITY
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZSTDLIB_API __declspec(dllimport) ZSTDLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZSTDLIB_API ZSTDLIB_VISIBILITY
+#endif
+
+
+/*******************************************************************************************************
+  Introduction
+
+  zstd, short for Zstandard, is a fast lossless compression algorithm,
+  targeting real-time compression scenarios at zlib-level and better compression ratios.
+  The zstd compression library provides in-memory compression and decompression functions.
+  The library supports compression levels from 1 up to ZSTD_maxCLevel() which is currently 22.
+  Levels >= 20, labeled `--ultra`, should be used with caution, as they require more memory.
+  Compression can be done in:
+    - a single step (described as Simple API)
+    - a single step, reusing a context (described as Explicit context)
+    - unbounded multiple steps (described as Streaming compression)
+  The compression ratio achievable on small data can be highly improved using a dictionary in:
+    - a single step (described as Simple dictionary API)
+    - a single step, reusing a dictionary (described as Bulk-processing dictionary API)
+
+  Advanced experimental functions can be accessed using #define ZSTD_STATIC_LINKING_ONLY before including zstd.h.
+  Advanced experimental APIs shall never be used with a dynamic library.
+  They are not "stable", their definition may change in the future. Only static linking is allowed.
+*********************************************************************************************************/
+
+/*------   Version   ------*/
+#define ZSTD_VERSION_MAJOR    1
+#define ZSTD_VERSION_MINOR    3
+#define ZSTD_VERSION_RELEASE  4
+
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< useful to check dll version */
+
+#define ZSTD_LIB_VERSION ZSTD_VERSION_MAJOR.ZSTD_VERSION_MINOR.ZSTD_VERSION_RELEASE
+#define ZSTD_QUOTE(str) #str
+#define ZSTD_EXPAND_AND_QUOTE(str) ZSTD_QUOTE(str)
+#define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
+ZSTDLIB_API const char* ZSTD_versionString(void);   /* added in v1.3.0 */
+
+
+/***************************************
+*  Simple API
+***************************************/
+/*! ZSTD_compress() :
+ *  Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
+ *  Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
+ *  @return : compressed size written into `dst` (<= `dstCapacity),
+ *            or an error code if it fails (which can be tested using ZSTD_isError()). */
+ZSTDLIB_API size_t ZSTD_compress( void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                                  int compressionLevel);
+
+/*! ZSTD_decompress() :
+ *  `compressedSize` : must be the _exact_ size of some number of compressed and/or skippable frames.
+ *  `dstCapacity` is an upper bound of originalSize to regenerate.
+ *  If user cannot imply a maximum upper bound, it's better to use streaming mode to decompress data.
+ *  @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+ *            or an errorCode if it fails (which can be tested using ZSTD_isError()). */
+ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
+                              const void* src, size_t compressedSize);
+
+/*! ZSTD_getFrameContentSize() : added in v1.3.0
+ *  `src` should point to the start of a ZSTD encoded frame.
+ *  `srcSize` must be at least as large as the frame header.
+ *            hint : any size >= `ZSTD_frameHeaderSize_max` is large enough.
+ *  @return : - decompressed size of the frame in `src`, if known
+ *            - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *            - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small)
+ *   note 1 : a 0 return value means the frame is valid but "empty".
+ *   note 2 : decompressed size is an optional field, it may not be present, typically in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *            Optionally, application can rely on some implicit limit,
+ *            as ZSTD_decompress() only needs an upper bound of decompressed size.
+ *            (For example, data could be necessarily cut into blocks <= 16 KB).
+ *   note 3 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 4 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 5 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure return value fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 6 : This function replaces ZSTD_getDecompressedSize() */
+#define ZSTD_CONTENTSIZE_UNKNOWN (0ULL - 1)
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize);
+
+/*! ZSTD_getDecompressedSize() :
+ *  NOTE: This function is now obsolete, in favor of ZSTD_getFrameContentSize().
+ *  Both functions work the same way, but ZSTD_getDecompressedSize() blends
+ *  "empty", "unknown" and "error" results to the same return value (0),
+ *  while ZSTD_getFrameContentSize() gives them separate return values.
+ * `src` is the start of a zstd compressed frame.
+ * @return : content size to be decompressed, as a 64-bits value _if known and not empty_, 0 otherwise. */
+ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
+
+
+/*======  Helper functions  ======*/
+#define ZSTD_COMPRESSBOUND(srcSize)   ((srcSize) + ((srcSize)>>8) + (((srcSize) < (128<<10)) ? (((128<<10) - (srcSize)) >> 11) /* margin, from 64 to 0 */ : 0))  /* this formula ensures that bound(A) + bound(B) <= bound(A+B) as long as A and B >= 128 KB */
+ZSTDLIB_API size_t      ZSTD_compressBound(size_t srcSize); /*!< maximum compressed size in worst case single-pass scenario */
+ZSTDLIB_API unsigned    ZSTD_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+ZSTDLIB_API const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
+ZSTDLIB_API int         ZSTD_maxCLevel(void);               /*!< maximum compression level available */
+
+
+/***************************************
+*  Explicit context
+***************************************/
+/*= Compression context
+ *  When compressing many times,
+ *  it is recommended to allocate a context just once, and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution in multi-threaded environments. */
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;
+ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx(void);
+ZSTDLIB_API size_t     ZSTD_freeCCtx(ZSTD_CCtx* cctx);
+
+/*! ZSTD_compressCCtx() :
+ *  Same as ZSTD_compress(), requires an allocated ZSTD_CCtx (see ZSTD_createCCtx()). */
+ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* ctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     int compressionLevel);
+
+/*= Decompression context
+ *  When decompressing many times,
+ *  it is recommended to allocate a context only once,
+ *  and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution. */
+typedef struct ZSTD_DCtx_s ZSTD_DCtx;
+ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx(void);
+ZSTDLIB_API size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
+
+/*! ZSTD_decompressDCtx() :
+ *  Same as ZSTD_decompress(), requires an allocated ZSTD_DCtx (see ZSTD_createDCtx()) */
+ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* ctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize);
+
+
+/**************************
+*  Simple dictionary API
+***************************/
+/*! ZSTD_compress_usingDict() :
+ *  Compression using a predefined Dictionary (see dictBuilder/zdict.h).
+ *  Note : This function loads the dictionary, resulting in significant startup delay.
+ *  Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
+ZSTDLIB_API size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const void* dict,size_t dictSize,
+                                           int compressionLevel);
+
+/*! ZSTD_decompress_usingDict() :
+ *  Decompression using a predefined Dictionary (see dictBuilder/zdict.h).
+ *  Dictionary must be identical to the one used during compression.
+ *  Note : This function loads the dictionary, resulting in significant startup delay.
+ *  Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
+ZSTDLIB_API size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
+                                             void* dst, size_t dstCapacity,
+                                       const void* src, size_t srcSize,
+                                       const void* dict,size_t dictSize);
+
+
+/**********************************
+ *  Bulk processing dictionary API
+ *********************************/
+typedef struct ZSTD_CDict_s ZSTD_CDict;
+
+/*! ZSTD_createCDict() :
+ *  When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
+ *  ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
+ *  ZSTD_CDict can be created once and shared by multiple threads concurrently, since its usage is read-only.
+ *  `dictBuffer` can be released after ZSTD_CDict creation, since its content is copied within CDict */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize,
+                                         int compressionLevel);
+
+/*! ZSTD_freeCDict() :
+ *  Function frees memory allocated by ZSTD_createCDict(). */
+ZSTDLIB_API size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
+
+/*! ZSTD_compress_usingCDict() :
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression level is decided during dictionary creation.
+ *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
+ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
+                                            void* dst, size_t dstCapacity,
+                                      const void* src, size_t srcSize,
+                                      const ZSTD_CDict* cdict);
+
+
+typedef struct ZSTD_DDict_s ZSTD_DDict;
+
+/*! ZSTD_createDDict() :
+ *  Create a digested dictionary, ready to start decompression operation without startup delay.
+ *  dictBuffer can be released after DDict creation, as its content is copied inside DDict */
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict(const void* dictBuffer, size_t dictSize);
+
+/*! ZSTD_freeDDict() :
+ *  Function frees memory allocated with ZSTD_createDDict() */
+ZSTDLIB_API size_t      ZSTD_freeDDict(ZSTD_DDict* ddict);
+
+/*! ZSTD_decompress_usingDDict() :
+ *  Decompression using a digested Dictionary.
+ *  Faster startup than ZSTD_decompress_usingDict(), recommended when same dictionary is used multiple times. */
+ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
+                                              void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize,
+                                        const ZSTD_DDict* ddict);
+
+
+/****************************
+*  Streaming
+****************************/
+
+typedef struct ZSTD_inBuffer_s {
+  const void* src;    /**< start of input buffer */
+  size_t size;        /**< size of input buffer */
+  size_t pos;         /**< position where reading stopped. Will be updated. Necessarily 0 <= pos <= size */
+} ZSTD_inBuffer;
+
+typedef struct ZSTD_outBuffer_s {
+  void*  dst;         /**< start of output buffer */
+  size_t size;        /**< size of output buffer */
+  size_t pos;         /**< position where writing stopped. Will be updated. Necessarily 0 <= pos <= size */
+} ZSTD_outBuffer;
+
+
+
+/*-***********************************************************************
+*  Streaming compression - HowTo
+*
+*  A ZSTD_CStream object is required to track streaming operation.
+*  Use ZSTD_createCStream() and ZSTD_freeCStream() to create/release resources.
+*  ZSTD_CStream objects can be reused multiple times on consecutive compression operations.
+*  It is recommended to re-use ZSTD_CStream in situations where many streaming operations will be achieved consecutively,
+*  since it will play nicer with system's memory, by re-using already allocated memory.
+*  Use one separate ZSTD_CStream per thread for parallel execution.
+*
+*  Start a new compression by initializing ZSTD_CStream.
+*  Use ZSTD_initCStream() to start a new compression operation.
+*  Use ZSTD_initCStream_usingDict() or ZSTD_initCStream_usingCDict() for a compression which requires a dictionary (experimental section)
+*
+*  Use ZSTD_compressStream() repetitively to consume input stream.
+*  The function will automatically update both `pos` fields.
+*  Note that it may not consume the entire input, in which case `pos < size`,
+*  and it's up to the caller to present again remaining data.
+*  @return : a size hint, preferred nb of bytes to use as input for next function call
+*            or an error code, which can be tested using ZSTD_isError().
+*            Note 1 : it's just a hint, to help latency a little, any other value will work fine.
+*            Note 2 : size hint is guaranteed to be <= ZSTD_CStreamInSize()
+*
+*  At any moment, it's possible to flush whatever data remains within internal buffer, using ZSTD_flushStream().
+*  `output->pos` will be updated.
+*  Note that some content might still be left within internal buffer if `output->size` is too small.
+*  @return : nb of bytes still present within internal buffer (0 if it's empty)
+*            or an error code, which can be tested using ZSTD_isError().
+*
+*  ZSTD_endStream() instructs to finish a frame.
+*  It will perform a flush and write frame epilogue.
+*  The epilogue is required for decoders to consider a frame completed.
+*  ZSTD_endStream() may not be able to flush full data if `output->size` is too small.
+*  In which case, call again ZSTD_endStream() to complete the flush.
+*  @return : 0 if frame fully completed and fully flushed,
+             or >0 if some data is still present within internal buffer
+                  (value is minimum size estimation for remaining data to flush, but it could be more)
+*            or an error code, which can be tested using ZSTD_isError().
+*
+* *******************************************************************/
+
+typedef ZSTD_CCtx ZSTD_CStream;  /**< CCtx and CStream are now effectively same object (>= v1.3.0) */
+                                 /* Continue to distinguish them for compatibility with versions <= v1.2.0 */
+/*===== ZSTD_CStream management functions =====*/
+ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream(void);
+ZSTDLIB_API size_t ZSTD_freeCStream(ZSTD_CStream* zcs);
+
+/*===== Streaming compression functions =====*/
+ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+ZSTDLIB_API size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
+ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
+
+ZSTDLIB_API size_t ZSTD_CStreamInSize(void);    /**< recommended size for input buffer */
+ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output buffer. Guarantee to successfully flush at least one complete compressed block in all circumstances. */
+
+
+
+/*-***************************************************************************
+*  Streaming decompression - HowTo
+*
+*  A ZSTD_DStream object is required to track streaming operations.
+*  Use ZSTD_createDStream() and ZSTD_freeDStream() to create/release resources.
+*  ZSTD_DStream objects can be re-used multiple times.
+*
+*  Use ZSTD_initDStream() to start a new decompression operation,
+*   or ZSTD_initDStream_usingDict() if decompression requires a dictionary.
+*   @return : recommended first input size
+*
+*  Use ZSTD_decompressStream() repetitively to consume your input.
+*  The function will update both `pos` fields.
+*  If `input.pos < input.size`, some input has not been consumed.
+*  It's up to the caller to present again remaining data.
+*  If `output.pos < output.size`, decoder has flushed everything it could.
+*  @return : 0 when a frame is completely decoded and fully flushed,
+*            an error code, which can be tested using ZSTD_isError(),
+*            any other value > 0, which means there is still some decoding to do to complete current frame.
+*            The return value is a suggested next input size (a hint to improve latency) that will never load more than the current frame.
+* *******************************************************************************/
+
+typedef ZSTD_DCtx ZSTD_DStream;  /**< DCtx and DStream are now effectively same object (>= v1.3.0) */
+                                 /* For compatibility with versions <= v1.2.0, continue to consider them separated. */
+/*===== ZSTD_DStream management functions =====*/
+ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream(void);
+ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);
+
+/*===== Streaming decompression functions =====*/
+ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
+ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+
+ZSTDLIB_API size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
+ZSTDLIB_API size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output buffer. Guarantee to successfully flush at least one complete block in all circumstances. */
+
+#endif  /* ZSTD_H_235446 */
+
+
+
+/****************************************************************************************
+ * START OF ADVANCED AND EXPERIMENTAL FUNCTIONS
+ * The definitions in this section are considered experimental.
+ * They should never be used with a dynamic library, as prototypes may change in the future.
+ * They are provided for advanced scenarios.
+ * Use them only in association with static linking.
+ * ***************************************************************************************/
+
+#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
+#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
+
+/* --- Constants ---*/
+#define ZSTD_MAGICNUMBER            0xFD2FB528   /* >= v0.8.0 */
+#define ZSTD_MAGIC_SKIPPABLE_START  0x184D2A50U
+#define ZSTD_MAGIC_DICTIONARY       0xEC30A437   /* >= v0.7.0 */
+
+#define ZSTD_WINDOWLOG_MAX_32   30
+#define ZSTD_WINDOWLOG_MAX_64   31
+#define ZSTD_WINDOWLOG_MAX    ((unsigned)(sizeof(size_t) == 4 ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
+#define ZSTD_WINDOWLOG_MIN      10
+#define ZSTD_HASHLOG_MAX      ((ZSTD_WINDOWLOG_MAX < 30) ? ZSTD_WINDOWLOG_MAX : 30)
+#define ZSTD_HASHLOG_MIN         6
+#define ZSTD_CHAINLOG_MAX_32    29
+#define ZSTD_CHAINLOG_MAX_64    30
+#define ZSTD_CHAINLOG_MAX     ((unsigned)(sizeof(size_t) == 4 ? ZSTD_CHAINLOG_MAX_32 : ZSTD_CHAINLOG_MAX_64))
+#define ZSTD_CHAINLOG_MIN       ZSTD_HASHLOG_MIN
+#define ZSTD_HASHLOG3_MAX       17
+#define ZSTD_SEARCHLOG_MAX     (ZSTD_WINDOWLOG_MAX-1)
+#define ZSTD_SEARCHLOG_MIN       1
+#define ZSTD_SEARCHLENGTH_MAX    7   /* only for ZSTD_fast, other strategies are limited to 6 */
+#define ZSTD_SEARCHLENGTH_MIN    3   /* only for ZSTD_btopt, other strategies are limited to 4 */
+#define ZSTD_TARGETLENGTH_MIN    1   /* only used by btopt, btultra and btfast */
+#define ZSTD_LDM_MINMATCH_MIN    4
+#define ZSTD_LDM_MINMATCH_MAX 4096
+#define ZSTD_LDM_BUCKETSIZELOG_MAX 8
+
+#define ZSTD_FRAMEHEADERSIZE_PREFIX 5   /* minimum input size to know frame header size */
+#define ZSTD_FRAMEHEADERSIZE_MIN    6
+#define ZSTD_FRAMEHEADERSIZE_MAX   18   /* for static allocation */
+static const size_t ZSTD_frameHeaderSize_prefix = ZSTD_FRAMEHEADERSIZE_PREFIX;
+static const size_t ZSTD_frameHeaderSize_min = ZSTD_FRAMEHEADERSIZE_MIN;
+static const size_t ZSTD_frameHeaderSize_max = ZSTD_FRAMEHEADERSIZE_MAX;
+static const size_t ZSTD_skippableHeaderSize = 8;  /* magic number + skippable frame length */
+
+
+/*--- Advanced types ---*/
+typedef enum { ZSTD_fast=1, ZSTD_dfast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2,
+               ZSTD_btlazy2, ZSTD_btopt, ZSTD_btultra } ZSTD_strategy;   /* from faster to stronger */
+
+typedef struct {
+    unsigned windowLog;      /**< largest match distance : larger == more compression, more memory needed during decompression */
+    unsigned chainLog;       /**< fully searched segment : larger == more compression, slower, more memory (useless for fast) */
+    unsigned hashLog;        /**< dispatch table : larger == faster, more memory */
+    unsigned searchLog;      /**< nb of searches : larger == more compression, slower */
+    unsigned searchLength;   /**< match length searched : larger == faster decompression, sometimes less compression */
+    unsigned targetLength;   /**< acceptable match size for optimal parser (only) : larger == more compression, slower */
+    ZSTD_strategy strategy;
+} ZSTD_compressionParameters;
+
+typedef struct {
+    unsigned contentSizeFlag; /**< 1: content size will be in frame header (when known) */
+    unsigned checksumFlag;    /**< 1: generate a 32-bits checksum at end of frame, for error detection */
+    unsigned noDictIDFlag;    /**< 1: no dictID will be saved into frame header (if dictionary compression) */
+} ZSTD_frameParameters;
+
+typedef struct {
+    ZSTD_compressionParameters cParams;
+    ZSTD_frameParameters fParams;
+} ZSTD_parameters;
+
+typedef struct ZSTD_CCtx_params_s ZSTD_CCtx_params;
+
+typedef enum {
+    ZSTD_dct_auto=0,      /* dictionary is "full" when starting with ZSTD_MAGIC_DICTIONARY, otherwise it is "rawContent" */
+    ZSTD_dct_rawContent,  /* ensures dictionary is always loaded as rawContent, even if it starts with ZSTD_MAGIC_DICTIONARY */
+    ZSTD_dct_fullDict     /* refuses to load a dictionary if it does not respect Zstandard's specification */
+} ZSTD_dictContentType_e;
+
+typedef enum {
+    ZSTD_dlm_byCopy = 0, /**< Copy dictionary content internally */
+    ZSTD_dlm_byRef,      /**< Reference dictionary content -- the dictionary buffer must outlive its users. */
+} ZSTD_dictLoadMethod_e;
+
+
+
+/***************************************
+*  Frame size functions
+***************************************/
+
+/*! ZSTD_findFrameCompressedSize() :
+ *  `src` should point to the start of a ZSTD encoded frame or skippable frame
+ *  `srcSize` must be >= first frame size
+ *  @return : the compressed size of the first frame starting at `src`,
+ *            suitable to pass to `ZSTD_decompress` or similar,
+ *            or an error code if input is invalid */
+ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_findDecompressedSize() :
+ *  `src` should point the start of a series of ZSTD encoded and/or skippable frames
+ *  `srcSize` must be the _exact_ size of this series
+ *       (i.e. there should be a frame boundary exactly at `srcSize` bytes after `src`)
+ *  @return : - decompressed size of all data in all successive frames
+ *            - if the decompressed size cannot be determined: ZSTD_CONTENTSIZE_UNKNOWN
+ *            - if an error occurred: ZSTD_CONTENTSIZE_ERROR
+ *
+ *   note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 2 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 3 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure result fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 5 : ZSTD_findDecompressedSize handles multiple frames, and so it must traverse the input to
+ *            read each contained frame header.  This is fast as most of the data is skipped,
+ *            however it does mean that all frame data must be present and valid. */
+ZSTDLIB_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_frameHeaderSize() :
+*   `src` should point to the start of a ZSTD frame
+*   `srcSize` must be >= ZSTD_frameHeaderSize_prefix.
+*   @return : size of the Frame Header */
+ZSTDLIB_API size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize);
+
+
+/***************************************
+*  Memory management
+***************************************/
+
+/*! ZSTD_sizeof_*() :
+ *  These functions give the current memory usage of selected object.
+ *  Object memory usage can evolve when re-used. */
+ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
+ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
+ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
+ZSTDLIB_API size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict);
+ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
+
+/*! ZSTD_estimate*() :
+ *  These functions make it possible to estimate memory usage
+ *  of a future {D,C}Ctx, before its creation.
+ *  ZSTD_estimateCCtxSize() will provide a budget large enough for any compression level up to selected one.
+ *  It will also consider src size to be arbitrarily "large", which is worst case.
+ *  If srcSize is known to always be small, ZSTD_estimateCCtxSize_usingCParams() can provide a tighter estimation.
+ *  ZSTD_estimateCCtxSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
+ *  ZSTD_estimateCCtxSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParam_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_p_nbWorkers is >= 1.
+ *  Note : CCtx size estimation is only correct for single-threaded compression. */
+ZSTDLIB_API size_t ZSTD_estimateCCtxSize(int compressionLevel);
+ZSTDLIB_API size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_API size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
+
+/*! ZSTD_estimateCStreamSize() :
+ *  ZSTD_estimateCStreamSize() will provide a budget large enough for any compression level up to selected one.
+ *  It will also consider src size to be arbitrarily "large", which is worst case.
+ *  If srcSize is known to always be small, ZSTD_estimateCStreamSize_usingCParams() can provide a tighter estimation.
+ *  ZSTD_estimateCStreamSize_usingCParams() can be used in tandem with ZSTD_getCParams() to create cParams from compressionLevel.
+ *  ZSTD_estimateCStreamSize_usingCCtxParams() can be used in tandem with ZSTD_CCtxParam_setParameter(). Only single-threaded compression is supported. This function will return an error code if ZSTD_p_nbWorkers is >= 1.
+ *  Note : CStream size estimation is only correct for single-threaded compression.
+ *  ZSTD_DStream memory budget depends on window Size.
+ *  This information can be passed manually, using ZSTD_estimateDStreamSize,
+ *  or deducted from a valid frame Header, using ZSTD_estimateDStreamSize_fromFrame();
+ *  Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
+ *         an internal ?Dict will be created, which additional size is not estimated here.
+ *         In this case, get total size by adding ZSTD_estimate?DictSize */
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize(int compressionLevel);
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams);
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params);
+ZSTDLIB_API size_t ZSTD_estimateDStreamSize(size_t windowSize);
+ZSTDLIB_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize);
+
+/*! ZSTD_estimate?DictSize() :
+ *  ZSTD_estimateCDictSize() will bet that src size is relatively "small", and content is copied, like ZSTD_createCDict().
+ *  ZSTD_estimateCDictSize_advanced() makes it possible to control compression parameters precisely, like ZSTD_createCDict_advanced().
+ *  Note : dictionaries created by reference (`ZSTD_dlm_byRef`) are logically smaller.
+ */
+ZSTDLIB_API size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_estimateCDictSize_advanced(size_t dictSize, ZSTD_compressionParameters cParams, ZSTD_dictLoadMethod_e dictLoadMethod);
+ZSTDLIB_API size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod);
+
+/*! ZSTD_initStatic*() :
+ *  Initialize an object using a pre-allocated fixed-size buffer.
+ *  workspace: The memory area to emplace the object into.
+ *             Provided pointer *must be 8-bytes aligned*.
+ *             Buffer must outlive object.
+ *  workspaceSize: Use ZSTD_estimate*Size() to determine
+ *                 how large workspace must be to support target scenario.
+ * @return : pointer to object (same address as workspace, just different type),
+ *           or NULL if error (size too small, incorrect alignment, etc.)
+ *  Note : zstd will never resize nor malloc() when using a static buffer.
+ *         If the object requires more memory than available,
+ *         zstd will just error out (typically ZSTD_error_memory_allocation).
+ *  Note 2 : there is no corresponding "free" function.
+ *           Since workspace is allocated externally, it must be freed externally too.
+ *  Note 3 : cParams : use ZSTD_getCParams() to convert a compression level
+ *           into its associated cParams.
+ *  Limitation 1 : currently not compatible with internal dictionary creation, triggered by
+ *                 ZSTD_CCtx_loadDictionary(), ZSTD_initCStream_usingDict() or ZSTD_initDStream_usingDict().
+ *  Limitation 2 : static cctx currently not compatible with multi-threading.
+ *  Limitation 3 : static dctx is incompatible with legacy support.
+ */
+ZSTDLIB_API ZSTD_CCtx*    ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_API ZSTD_CStream* ZSTD_initStaticCStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticCCtx() */
+
+ZSTDLIB_API ZSTD_DCtx*    ZSTD_initStaticDCtx(void* workspace, size_t workspaceSize);
+ZSTDLIB_API ZSTD_DStream* ZSTD_initStaticDStream(void* workspace, size_t workspaceSize);    /**< same as ZSTD_initStaticDCtx() */
+
+ZSTDLIB_API const ZSTD_CDict* ZSTD_initStaticCDict(
+                                        void* workspace, size_t workspaceSize,
+                                        const void* dict, size_t dictSize,
+                                        ZSTD_dictLoadMethod_e dictLoadMethod,
+                                        ZSTD_dictContentType_e dictContentType,
+                                        ZSTD_compressionParameters cParams);
+
+ZSTDLIB_API const ZSTD_DDict* ZSTD_initStaticDDict(
+                                        void* workspace, size_t workspaceSize,
+                                        const void* dict, size_t dictSize,
+                                        ZSTD_dictLoadMethod_e dictLoadMethod,
+                                        ZSTD_dictContentType_e dictContentType);
+
+/*! Custom memory allocation :
+ *  These prototypes make it possible to pass your own allocation/free functions.
+ *  ZSTD_customMem is provided at creation time, using ZSTD_create*_advanced() variants listed below.
+ *  All allocation/free operations will be completed using these custom variants instead of regular <stdlib.h> ones.
+ */
+typedef void* (*ZSTD_allocFunction) (void* opaque, size_t size);
+typedef void  (*ZSTD_freeFunction) (void* opaque, void* address);
+typedef struct { ZSTD_allocFunction customAlloc; ZSTD_freeFunction customFree; void* opaque; } ZSTD_customMem;
+static ZSTD_customMem const ZSTD_defaultCMem = { NULL, NULL, NULL };  /**< this constant defers to stdlib's functions */
+
+ZSTDLIB_API ZSTD_CCtx*    ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_DCtx*    ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
+
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
+                                                  ZSTD_dictLoadMethod_e dictLoadMethod,
+                                                  ZSTD_dictContentType_e dictContentType,
+                                                  ZSTD_compressionParameters cParams,
+                                                  ZSTD_customMem customMem);
+
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                                  ZSTD_dictLoadMethod_e dictLoadMethod,
+                                                  ZSTD_dictContentType_e dictContentType,
+                                                  ZSTD_customMem customMem);
+
+
+
+/***************************************
+*  Advanced compression functions
+***************************************/
+
+/*! ZSTD_createCDict_byReference() :
+ *  Create a digested dictionary for compression
+ *  Dictionary content is simply referenced, and therefore stays in dictBuffer.
+ *  It is important that dictBuffer outlives CDict, it must remain read accessible throughout the lifetime of CDict */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
+
+/*! ZSTD_getCParams() :
+*   @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
+*   `estimatedSrcSize` value is optional, select 0 if not known */
+ZSTDLIB_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+
+/*! ZSTD_getParams() :
+*   same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
+*   All fields of `ZSTD_frameParameters` are set to default : contentSize=1, checksum=0, noDictID=0 */
+ZSTDLIB_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
+
+/*! ZSTD_checkCParams() :
+*   Ensure param values remain within authorized range */
+ZSTDLIB_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
+
+/*! ZSTD_adjustCParams() :
+ *  optimize params for a given `srcSize` and `dictSize`.
+ *  both values are optional, select `0` if unknown. */
+ZSTDLIB_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
+
+/*! ZSTD_compress_advanced() :
+*   Same as ZSTD_compress_usingDict(), with fine-tune control over each compression parameter */
+ZSTDLIB_API size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const void* dict,size_t dictSize,
+                                  ZSTD_parameters params);
+
+/*! ZSTD_compress_usingCDict_advanced() :
+*   Same as ZSTD_compress_usingCDict(), with fine-tune control over frame parameters */
+ZSTDLIB_API size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_CDict* cdict, ZSTD_frameParameters fParams);
+
+
+/*--- Advanced decompression functions ---*/
+
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+ZSTDLIB_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
+
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, ready to start decompression operation without startup delay.
+ *  Dictionary content is referenced, and therefore stays in dictBuffer.
+ *  It is important that dictBuffer outlives DDict,
+ *  it must remain read accessible throughout the lifetime of DDict */
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
+
+
+/*! ZSTD_getDictID_fromDict() :
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize);
+
+/*! ZSTD_getDictID_fromDDict() :
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict);
+
+/*! ZSTD_getDictID_fromFrame() :
+ *  Provides the dictID required to decompressed the frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary to be decoded (most common case).
+ *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use ZSTD_getFrameHeader(), which will provide a more precise error code. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
+
+
+/********************************************************************
+*  Advanced streaming functions
+********************************************************************/
+
+/*=====   Advanced Streaming compression functions  =====*/
+ZSTDLIB_API size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize);   /**< pledgedSrcSize must be correct. If it is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN. Note that, for compatibility with older programs, "0" also disables frame content size field. It may be enabled in the future. */
+ZSTDLIB_API size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel); /**< creates of an internal CDict (incompatible with static CCtx), except if dict == NULL or dictSize < 8, in which case no dict is used. Note: dict is loaded with ZSTD_dm_auto (treated as a full zstd dictionary if it begins with ZSTD_MAGIC_DICTIONARY, else as raw content) and ZSTD_dlm_byCopy.*/
+ZSTDLIB_API size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs, const void* dict, size_t dictSize,
+                                             ZSTD_parameters params, unsigned long long pledgedSrcSize);  /**< pledgedSrcSize must be correct. If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN. dict is loaded with ZSTD_dm_auto and ZSTD_dlm_byCopy. */
+ZSTDLIB_API size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);  /**< note : cdict will just be referenced, and must outlive compression session */
+ZSTDLIB_API size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs, const ZSTD_CDict* cdict, ZSTD_frameParameters fParams, unsigned long long pledgedSrcSize);  /**< same as ZSTD_initCStream_usingCDict(), with control over frame parameters. pledgedSrcSize must be correct. If srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN. */
+
+/*! ZSTD_resetCStream() :
+ *  start a new compression job, using same parameters from previous job.
+ *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place..
+ *  Note that zcs must be init at least once before using ZSTD_resetCStream().
+ *  If pledgedSrcSize is not known at reset time, use macro ZSTD_CONTENTSIZE_UNKNOWN.
+ *  If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
+ *  For the time being, pledgedSrcSize==0 is interpreted as "srcSize unknown" for compatibility with older programs,
+ *  but it will change to mean "empty" in future version, so use macro ZSTD_CONTENTSIZE_UNKNOWN instead.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
+
+
+typedef struct {
+    unsigned long long ingested;
+    unsigned long long consumed;
+    unsigned long long produced;
+} ZSTD_frameProgression;
+
+/* ZSTD_getFrameProgression():
+ * tells how much data has been ingested (read from input)
+ * consumed (input actually compressed) and produced (output) for current frame.
+ * Therefore, (ingested - consumed) is amount of input data buffered internally, not yet compressed.
+ * Can report progression inside worker threads (multi-threading and non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx);
+
+
+
+/*=====   Advanced Streaming decompression functions  =====*/
+typedef enum { DStream_p_maxWindowSize } ZSTD_DStreamParameter_e;
+ZSTDLIB_API size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds, ZSTD_DStreamParameter_e paramType, unsigned paramValue);   /* obsolete : this API will be removed in a future version */
+ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize); /**< note: no dictionary will be used if dict == NULL or dictSize < 8 */
+ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);  /**< note : ddict is referenced, it must outlive decompression session */
+ZSTDLIB_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);  /**< re-use decompression parameters from previous init; saves dictionary loading */
+
+
+/*********************************************************************
+*  Buffer-less and synchronous inner streaming functions
+*
+*  This is an advanced API, giving full control over buffer management, for users which need direct control over memory.
+*  But it's also a complex one, with several restrictions, documented below.
+*  Prefer normal streaming API for an easier experience.
+********************************************************************* */
+
+/**
+  Buffer-less streaming compression (synchronous mode)
+
+  A ZSTD_CCtx object is required to track streaming operations.
+  Use ZSTD_createCCtx() / ZSTD_freeCCtx() to manage resource.
+  ZSTD_CCtx object can be re-used multiple times within successive compression operations.
+
+  Start by initializing a context.
+  Use ZSTD_compressBegin(), or ZSTD_compressBegin_usingDict() for dictionary compression,
+  or ZSTD_compressBegin_advanced(), for finer parameter control.
+  It's also possible to duplicate a reference context which has already been initialized, using ZSTD_copyCCtx()
+
+  Then, consume your input using ZSTD_compressContinue().
+  There are some important considerations to keep in mind when using this advanced function :
+  - ZSTD_compressContinue() has no internal buffer. It uses externally provided buffers only.
+  - Interface is synchronous : input is consumed entirely and produces 1+ compressed blocks.
+  - Caller must ensure there is enough space in `dst` to store compressed data under worst case scenario.
+    Worst case evaluation is provided by ZSTD_compressBound().
+    ZSTD_compressContinue() doesn't guarantee recover after a failed compression.
+  - ZSTD_compressContinue() presumes prior input ***is still accessible and unmodified*** (up to maximum distance size, see WindowLog).
+    It remembers all previous contiguous blocks, plus one separated memory segment (which can itself consists of multiple contiguous blocks)
+  - ZSTD_compressContinue() detects that prior input has been overwritten when `src` buffer overlaps.
+    In which case, it will "discard" the relevant memory section from its history.
+
+  Finish a frame with ZSTD_compressEnd(), which will write the last block(s) and optional checksum.
+  It's possible to use srcSize==0, in which case, it will write a final empty block to end the frame.
+  Without last block mark, frames are considered unfinished (hence corrupted) by compliant decoders.
+
+  `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress again.
+*/
+
+/*=====   Buffer-less streaming compression functions  =====*/
+ZSTDLIB_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize : If srcSize is not known at init time, use ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
+ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   /* compression parameters are already set within cdict. pledgedSrcSize must be correct. If srcSize is not known, use macro ZSTD_CONTENTSIZE_UNKNOWN */
+ZSTDLIB_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize is not known, use ZSTD_CONTENTSIZE_UNKNOWN */
+
+ZSTDLIB_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-
+  Buffer-less streaming decompression (synchronous mode)
+
+  A ZSTD_DCtx object is required to track streaming operations.
+  Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
+  A ZSTD_DCtx object can be re-used multiple times.
+
+  First typical operation is to retrieve frame parameters, using ZSTD_getFrameHeader().
+  Frame header is extracted from the beginning of compressed frame, so providing only the frame's beginning is enough.
+  Data fragment must be large enough to ensure successful decoding.
+ `ZSTD_frameHeaderSize_max` bytes is guaranteed to always be large enough.
+  @result : 0 : successful decoding, the `ZSTD_frameHeader` structure is correctly filled.
+           >0 : `srcSize` is too small, please provide at least @result bytes on next attempt.
+           errorCode, which can be tested using ZSTD_isError().
+
+  It fills a ZSTD_frameHeader structure with important information to correctly decode the frame,
+  such as the dictionary ID, content size, or maximum back-reference distance (`windowSize`).
+  Note that these values could be wrong, either because of data corruption, or because a 3rd party deliberately spoofs false information.
+  As a consequence, check that values remain within valid application range.
+  For example, do not allocate memory blindly, check that `windowSize` is within expectation.
+  Each application can set its own limits, depending on local restrictions.
+  For extended interoperability, it is recommended to support `windowSize` of at least 8 MB.
+
+  ZSTD_decompressContinue() needs previous data blocks during decompression, up to `windowSize` bytes.
+  ZSTD_decompressContinue() is very sensitive to contiguity,
+  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
+  or that previous contiguous segment is large enough to properly handle maximum back-reference distance.
+  There are multiple ways to guarantee this condition.
+
+  The most memory efficient way is to use a round buffer of sufficient size.
+  Sufficient size is determined by invoking ZSTD_decodingBufferSize_min(),
+  which can @return an error code if required value is too large for current system (in 32-bits mode).
+  In a round buffer methodology, ZSTD_decompressContinue() decompresses each block next to previous one,
+  up to the moment there is not enough room left in the buffer to guarantee decoding another full block,
+  which maximum size is provided in `ZSTD_frameHeader` structure, field `blockSizeMax`.
+  At which point, decoding can resume from the beginning of the buffer.
+  Note that already decoded data stored in the buffer should be flushed before being overwritten.
+
+  There are alternatives possible, for example using two or more buffers of size `windowSize` each, though they consume more memory.
+
+  Finally, if you control the compression process, you can also ignore all buffer size rules,
+  as long as the encoder and decoder progress in "lock-step",
+  aka use exactly the same buffer sizes, break contiguity at the same place, etc.
+
+  Once buffers are setup, start decompression, with ZSTD_decompressBegin().
+  If decompression requires a dictionary, use ZSTD_decompressBegin_usingDict() or ZSTD_decompressBegin_usingDDict().
+
+  Then use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how many bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() requires this _exact_ amount of bytes, or it will fail.
+
+ @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  It can be zero : it just means ZSTD_decompressContinue() has decoded some metadata item.
+  It can also be an error code, which can be tested with ZSTD_isError().
+
+  A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+
+  Note : it's possible to know if next input to present is a header or a block, using ZSTD_nextInputType().
+  This information is not required to properly decode a frame.
+
+  == Special case : skippable frames ==
+
+  Skippable frames allow integration of user-defined data into a flow of concatenated frames.
+  Skippable frames will be ignored (skipped) by decompressor.
+  The format of skippable frames is as follows :
+  a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
+  b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
+  c) Frame Content - any content (User Data) of length equal to Frame Size
+  For skippable frames ZSTD_getFrameHeader() returns zfhPtr->frameType==ZSTD_skippableFrame.
+  For skippable frames ZSTD_decompressContinue() always returns 0 : it only skips the content.
+*/
+
+/*=====   Buffer-less streaming decompression functions  =====*/
+typedef enum { ZSTD_frame, ZSTD_skippableFrame } ZSTD_frameType_e;
+typedef struct {
+    unsigned long long frameContentSize; /* if == ZSTD_CONTENTSIZE_UNKNOWN, it means this field is not available. 0 means "empty" */
+    unsigned long long windowSize;       /* can be very large, up to <= frameContentSize */
+    unsigned blockSizeMax;
+    ZSTD_frameType_e frameType;          /* if == ZSTD_skippableFrame, frameContentSize is the size of skippable content */
+    unsigned headerSize;
+    unsigned dictID;
+    unsigned checksumFlag;
+} ZSTD_frameHeader;
+ZSTDLIB_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+ZSTDLIB_API size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize);  /**< when frame content size is not known, pass in frameContentSize == ZSTD_CONTENTSIZE_UNKNOWN */
+
+ZSTDLIB_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+ZSTDLIB_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/* misc */
+ZSTDLIB_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
+typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
+ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
+
+
+
+/* ============================================ */
+/**       New advanced API (experimental)       */
+/* ============================================ */
+
+/* notes on API design :
+ *   In this proposal, parameters are pushed one by one into an existing context,
+ *   and then applied on all subsequent compression jobs.
+ *   When no parameter is ever provided, CCtx is created with compression level ZSTD_CLEVEL_DEFAULT.
+ *
+ *   This API is intended to replace all others advanced / experimental API entry points.
+ *   But it stands a reasonable chance to become "stable", after a reasonable testing period.
+ */
+
+/* note on naming convention :
+ *   Initially, the API favored names like ZSTD_setCCtxParameter() .
+ *   In this proposal, convention is changed towards ZSTD_CCtx_setParameter() .
+ *   The main driver is that it identifies more clearly the target object type.
+ *   It feels clearer when considering multiple targets :
+ *   ZSTD_CDict_setParameter() (rather than ZSTD_setCDictParameter())
+ *   ZSTD_CCtxParams_setParameter()  (rather than ZSTD_setCCtxParamsParameter() )
+ *   etc...
+ */
+
+/* note on enum design :
+ * All enum will be pinned to explicit values before reaching "stable API" status */
+
+typedef enum {
+    /* Opened question : should we have a format ZSTD_f_auto ?
+     * Today, it would mean exactly the same as ZSTD_f_zstd1.
+     * But, in the future, should several formats become supported,
+     * on the compression side, it would mean "default format".
+     * On the decompression side, it would mean "automatic format detection",
+     * so that ZSTD_f_zstd1 would mean "accept *only* zstd frames".
+     * Since meaning is a little different, another option could be to define different enums for compression and decompression.
+     * This question could be kept for later, when there are actually multiple formats to support,
+     * but there is also the question of pinning enum values, and pinning value `0` is especially important */
+    ZSTD_f_zstd1 = 0,        /* zstd frame format, specified in zstd_compression_format.md (default) */
+    ZSTD_f_zstd1_magicless,  /* Variant of zstd frame format, without initial 4-bytes magic number.
+                              * Useful to save 4 bytes per generated frame.
+                              * Decoder cannot recognise automatically this format, requiring instructions. */
+} ZSTD_format_e;
+
+typedef enum {
+    /* compression format */
+    ZSTD_p_format = 10,      /* See ZSTD_format_e enum definition.
+                              * Cast selected format as unsigned for ZSTD_CCtx_setParameter() compatibility. */
+
+    /* compression parameters */
+    ZSTD_p_compressionLevel=100, /* Update all compression parameters according to pre-defined cLevel table
+                              * Default level is ZSTD_CLEVEL_DEFAULT==3.
+                              * Special: value 0 means "do not change cLevel".
+                              * Note 1 : it's possible to pass a negative compression level by casting it to unsigned type.
+                              * Note 2 : setting a level sets all default values of other compression parameters.
+                              * Note 3 : setting compressionLevel automatically updates ZSTD_p_compressLiterals. */
+    ZSTD_p_windowLog,        /* Maximum allowed back-reference distance, expressed as power of 2.
+                              * Must be clamped between ZSTD_WINDOWLOG_MIN and ZSTD_WINDOWLOG_MAX.
+                              * Special: value 0 means "use default windowLog".
+                              * Note: Using a window size greater than ZSTD_MAXWINDOWSIZE_DEFAULT (default: 2^27)
+                              *       requires explicitly allowing such window size during decompression stage. */
+    ZSTD_p_hashLog,          /* Size of the probe table, as a power of 2.
+                              * Resulting table size is (1 << (hashLog+2)).
+                              * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX.
+                              * Larger tables improve compression ratio of strategies <= dFast,
+                              * and improve speed of strategies > dFast.
+                              * Special: value 0 means "use default hashLog". */
+    ZSTD_p_chainLog,         /* Size of the full-search table, as a power of 2.
+                              * Resulting table size is (1 << (chainLog+2)).
+                              * Larger tables result in better and slower compression.
+                              * This parameter is useless when using "fast" strategy.
+                              * Special: value 0 means "use default chainLog". */
+    ZSTD_p_searchLog,        /* Number of search attempts, as a power of 2.
+                              * More attempts result in better and slower compression.
+                              * This parameter is useless when using "fast" and "dFast" strategies.
+                              * Special: value 0 means "use default searchLog". */
+    ZSTD_p_minMatch,         /* Minimum size of searched matches (note : repCode matches can be smaller).
+                              * Larger values make faster compression and decompression, but decrease ratio.
+                              * Must be clamped between ZSTD_SEARCHLENGTH_MIN and ZSTD_SEARCHLENGTH_MAX.
+                              * Note that currently, for all strategies < btopt, effective minimum is 4.
+                              *                    , for all strategies > fast, effective maximum is 6.
+                              * Special: value 0 means "use default minMatchLength". */
+    ZSTD_p_targetLength,     /* Impact of this field depends on strategy.
+                              * For strategies btopt & btultra:
+                              *     Length of Match considered "good enough" to stop search.
+                              *     Larger values make compression stronger, and slower.
+                              * For strategy fast:
+                              *     Distance between match sampling.
+                              *     Larger values make compression faster, and weaker.
+                              * Special: value 0 means "use default targetLength". */
+    ZSTD_p_compressionStrategy, /* See ZSTD_strategy enum definition.
+                              * Cast selected strategy as unsigned for ZSTD_CCtx_setParameter() compatibility.
+                              * The higher the value of selected strategy, the more complex it is,
+                              * resulting in stronger and slower compression.
+                              * Special: value 0 means "use default strategy". */
+
+    ZSTD_p_enableLongDistanceMatching=160, /* Enable long distance matching.
+                                         * This parameter is designed to improve compression ratio
+                                         * for large inputs, by finding large matches at long distance.
+                                         * It increases memory usage and window size.
+                                         * Note: enabling this parameter increases ZSTD_p_windowLog to 128 MB
+                                         * except when expressly set to a different value. */
+    ZSTD_p_ldmHashLog,       /* Size of the table for long distance matching, as a power of 2.
+                              * Larger values increase memory usage and compression ratio,
+                              * but decrease compression speed.
+                              * Must be clamped between ZSTD_HASHLOG_MIN and ZSTD_HASHLOG_MAX
+                              * default: windowlog - 7.
+                              * Special: value 0 means "automatically determine hashlog". */
+    ZSTD_p_ldmMinMatch,      /* Minimum match size for long distance matcher.
+                              * Larger/too small values usually decrease compression ratio.
+                              * Must be clamped between ZSTD_LDM_MINMATCH_MIN and ZSTD_LDM_MINMATCH_MAX.
+                              * Special: value 0 means "use default value" (default: 64). */
+    ZSTD_p_ldmBucketSizeLog, /* Log size of each bucket in the LDM hash table for collision resolution.
+                              * Larger values improve collision resolution but decrease compression speed.
+                              * The maximum value is ZSTD_LDM_BUCKETSIZELOG_MAX .
+                              * Special: value 0 means "use default value" (default: 3). */
+    ZSTD_p_ldmHashEveryLog,  /* Frequency of inserting/looking up entries in the LDM hash table.
+                              * Must be clamped between 0 and (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN).
+                              * Default is MAX(0, (windowLog - ldmHashLog)), optimizing hash table usage.
+                              * Larger values improve compression speed.
+                              * Deviating far from default value will likely result in a compression ratio decrease.
+                              * Special: value 0 means "automatically determine hashEveryLog". */
+
+    /* frame parameters */
+    ZSTD_p_contentSizeFlag=200, /* Content size will be written into frame header _whenever known_ (default:1)
+                              * Content size must be known at the beginning of compression,
+                              * it is provided using ZSTD_CCtx_setPledgedSrcSize() */
+    ZSTD_p_checksumFlag,     /* A 32-bits checksum of content is written at end of frame (default:0) */
+    ZSTD_p_dictIDFlag,       /* When applicable, dictionary's ID is written into frame header (default:1) */
+
+    /* multi-threading parameters */
+    /* These parameters are only useful if multi-threading is enabled (ZSTD_MULTITHREAD).
+     * They return an error otherwise. */
+    ZSTD_p_nbWorkers=400,    /* Select how many threads will be spawned to compress in parallel.
+                              * When nbWorkers >= 1, triggers asynchronous mode :
+                              * ZSTD_compress_generic() consumes some input, flush some output if possible, and immediately gives back control to caller,
+                              * while compression work is performed in parallel, within worker threads.
+                              * (note : a strong exception to this rule is when first invocation sets ZSTD_e_end : it becomes a blocking call).
+                              * More workers improve speed, but also increase memory usage.
+                              * Default value is `0`, aka "single-threaded mode" : no worker is spawned, compression is performed inside Caller's thread, all invocations are blocking */
+    ZSTD_p_jobSize,          /* Size of a compression job. This value is enforced only in non-blocking mode.
+                              * Each compression job is completed in parallel, so this value indirectly controls the nb of active threads.
+                              * 0 means default, which is dynamically determined based on compression parameters.
+                              * Job size must be a minimum of overlapSize, or 1 MB, whichever is largest.
+                              * The minimum size is automatically and transparently enforced */
+    ZSTD_p_overlapSizeLog,   /* Size of previous input reloaded at the beginning of each job.
+                              * 0 => no overlap, 6(default) => use 1/8th of windowSize, >=9 => use full windowSize */
+
+    /* =================================================================== */
+    /* experimental parameters - no stability guaranteed                   */
+    /* =================================================================== */
+
+    ZSTD_p_compressLiterals=1000, /* control huffman compression of literals (enabled) by default.
+                              * disabling it improves speed and decreases compression ratio by a large amount.
+                              * note : this setting is automatically updated when changing compression level.
+                              *        positive compression levels set ZSTD_p_compressLiterals to 1.
+                              *        negative compression levels set ZSTD_p_compressLiterals to 0. */
+
+    ZSTD_p_forceMaxWindow=1100, /* Force back-reference distances to remain < windowSize,
+                              * even when referencing into Dictionary content (default:0) */
+
+} ZSTD_cParameter;
+
+
+/*! ZSTD_CCtx_setParameter() :
+ *  Set one compression parameter, selected by enum ZSTD_cParameter.
+ *  Setting a parameter is generally only possible during frame initialization (before starting compression),
+ *  except for a few exceptions which can be updated during compression: compressionLevel, hashLog, chainLog, searchLog, minMatch, targetLength and strategy.
+ *  Note : when `value` is an enum, cast it to unsigned for proper type checking.
+ *  @result : informational value (typically, value being set clamped correctly),
+ *            or an error code (which can be tested with ZSTD_isError()). */
+ZSTDLIB_API size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value);
+
+/*! ZSTD_CCtx_setPledgedSrcSize() :
+ *  Total input data size to be compressed as a single frame.
+ *  This value will be controlled at the end, and result in error if not respected.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : 0 means zero, empty.
+ *           In order to mean "unknown content size", pass constant ZSTD_CONTENTSIZE_UNKNOWN.
+ *           ZSTD_CONTENTSIZE_UNKNOWN is default value for any new compression job.
+ *  Note 2 : If all data is provided and consumed in a single round,
+ *           this value is overriden by srcSize instead. */
+ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize);
+
+/*! ZSTD_CCtx_loadDictionary() :
+ *  Create an internal CDict from `dict` buffer.
+ *  Decompression will have to use same dictionary.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: Adding a NULL (or 0-size) dictionary invalidates previous dictionary,
+ *           meaning "return to no-dictionary mode".
+ *  Note 1 : Dictionary will be used for all future compression jobs.
+ *           To return to "no-dictionary" situation, load a NULL dictionary
+ *  Note 2 : Loading a dictionary involves building tables, which are dependent on compression parameters.
+ *           For this reason, compression parameters cannot be changed anymore after loading a dictionary.
+ *           It's also a CPU consuming operation, with non-negligible impact on latency.
+ *  Note 3 :`dict` content will be copied internally.
+ *           Use ZSTD_CCtx_loadDictionary_byReference() to reference dictionary content instead.
+ *           In such a case, dictionary buffer must outlive its users.
+ *  Note 4 : Use ZSTD_CCtx_loadDictionary_advanced()
+ *           to precisely select how dictionary content must be interpreted. */
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(ZSTD_CCtx* cctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
+
+/*! ZSTD_CCtx_refCDict() :
+ *  Reference a prepared dictionary, to be used for all next compression jobs.
+ *  Note that compression parameters are enforced from within CDict,
+ *  and supercede any compression parameter previously set within CCtx.
+ *  The dictionary will remain valid for future compression jobs using same CCtx.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special : adding a NULL CDict means "return to no-dictionary mode".
+ *  Note 1 : Currently, only one dictionary can be managed.
+ *           Adding a new dictionary effectively "discards" any previous one.
+ *  Note 2 : CDict is just referenced, its lifetime must outlive CCtx. */
+ZSTDLIB_API size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
+
+/*! ZSTD_CCtx_refPrefix() :
+ *  Reference a prefix (single-usage dictionary) for next compression job.
+ *  Decompression need same prefix to properly regenerate data.
+ *  Prefix is **only used once**. Tables are discarded at end of compression job.
+ *  Subsequent compression jobs will be done without prefix (if none is explicitly referenced).
+ *  If there is a need to use same prefix multiple times, consider embedding it into a ZSTD_CDict instead.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special: Adding any prefix (including NULL) invalidates any previous prefix or dictionary
+ *  Note 1 : Prefix buffer is referenced. It must outlive compression job.
+ *  Note 2 : Referencing a prefix involves building tables, which are dependent on compression parameters.
+ *           It's a CPU consuming operation, with non-negligible impact on latency.
+ *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+ *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode. */
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize);
+ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
+/*! ZSTD_CCtx_reset() :
+ *  Return a CCtx to clean state.
+ *  Useful after an error, or to interrupt an ongoing compression job and start a new one.
+ *  Any internal data not yet flushed is cancelled.
+ *  Dictionary (if any) is dropped.
+ *  All parameters are back to default values.
+ *  It's possible to modify compression parameters after a reset.
+ */
+ZSTDLIB_API void ZSTD_CCtx_reset(ZSTD_CCtx* cctx);
+
+
+
+typedef enum {
+    ZSTD_e_continue=0, /* collect more data, encoder decides when to output compressed result, for optimal conditions */
+    ZSTD_e_flush,      /* flush any data provided so far - frame will continue, future data can still reference previous data for better compression */
+    ZSTD_e_end         /* flush any remaining data and close current frame. Any additional data starts a new frame. */
+} ZSTD_EndDirective;
+
+/*! ZSTD_compress_generic() :
+ *  Behave about the same as ZSTD_compressStream. To note :
+ *  - Compression parameters are pushed into CCtx before starting compression, using ZSTD_CCtx_setParameter()
+ *  - Compression parameters cannot be changed once compression is started.
+ *  - outpot->pos must be <= dstCapacity, input->pos must be <= srcSize
+ *  - outpot->pos and input->pos will be updated. They are guaranteed to remain below their respective limit.
+ *  - In single-thread mode (default), function is blocking : it completed its job before returning to caller.
+ *  - In multi-thread mode, function is non-blocking : it just acquires a copy of input, and distribute job to internal worker threads,
+ *                                                     and then immediately returns, just indicating that there is some data remaining to be flushed.
+ *                                                     The function nonetheless guarantees forward progress : it will return only after it reads or write at least 1+ byte.
+ *  - Exception : in multi-threading mode, if the first call requests a ZSTD_e_end directive, it is blocking : it will complete compression before giving back control to caller.
+ *  - @return provides a minimum amount of data remaining to be flushed from internal buffers
+ *            or an error code, which can be tested using ZSTD_isError().
+ *            if @return != 0, flush is not fully completed, there is still some data left within internal buffers.
+ *            This is useful for ZSTD_e_flush, since in this case more flushes are necessary to empty all buffers.
+ *            For ZSTD_e_end, @return == 0 when internal buffers are fully flushed and frame is completed.
+ *  - after a ZSTD_e_end directive, if internal buffer is not fully flushed (@return != 0),
+ *            only ZSTD_e_end or ZSTD_e_flush operations are allowed.
+ *            Before starting a new compression job, or changing compression parameters,
+ *            it is required to fully flush internal buffers.
+ */
+ZSTDLIB_API size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
+                                          ZSTD_outBuffer* output,
+                                          ZSTD_inBuffer* input,
+                                          ZSTD_EndDirective endOp);
+
+
+/*! ZSTD_compress_generic_simpleArgs() :
+ *  Same as ZSTD_compress_generic(),
+ *  but using only integral types as arguments.
+ *  Argument list is larger than ZSTD_{in,out}Buffer,
+ *  but can be helpful for binders from dynamic languages
+ *  which have troubles handling structures containing memory pointers.
+ */
+ZSTDLIB_API size_t ZSTD_compress_generic_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp);
+
+
+/*! ZSTD_CCtx_params :
+ *  Quick howto :
+ *  - ZSTD_createCCtxParams() : Create a ZSTD_CCtx_params structure
+ *  - ZSTD_CCtxParam_setParameter() : Push parameters one by one into
+ *                                    an existing ZSTD_CCtx_params structure.
+ *                                    This is similar to
+ *                                    ZSTD_CCtx_setParameter().
+ *  - ZSTD_CCtx_setParametersUsingCCtxParams() : Apply parameters to
+ *                                    an existing CCtx.
+ *                                    These parameters will be applied to
+ *                                    all subsequent compression jobs.
+ *  - ZSTD_compress_generic() : Do compression using the CCtx.
+ *  - ZSTD_freeCCtxParams() : Free the memory.
+ *
+ *  This can be used with ZSTD_estimateCCtxSize_advanced_usingCCtxParams()
+ *  for static allocation for single-threaded compression.
+ */
+ZSTDLIB_API ZSTD_CCtx_params* ZSTD_createCCtxParams(void);
+ZSTDLIB_API size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params);
+
+
+/*! ZSTD_CCtxParams_reset() :
+ *  Reset params to default values.
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params);
+
+/*! ZSTD_CCtxParams_init() :
+ *  Initializes the compression parameters of cctxParams according to
+ *  compression level. All other parameters are reset to their default values.
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel);
+
+/*! ZSTD_CCtxParams_init_advanced() :
+ *  Initializes the compression and frame parameters of cctxParams according to
+ *  params. All other parameters are reset to their default values.
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params);
+
+
+/*! ZSTD_CCtxParam_setParameter() :
+ *  Similar to ZSTD_CCtx_setParameter.
+ *  Set one compression parameter, selected by enum ZSTD_cParameter.
+ *  Parameters must be applied to a ZSTD_CCtx using ZSTD_CCtx_setParametersUsingCCtxParams().
+ *  Note : when `value` is an enum, cast it to unsigned for proper type checking.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_CCtxParam_setParameter(ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned value);
+
+/*! ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  Apply a set of ZSTD_CCtx_params to the compression context.
+ *  This can be done even after compression is started,
+ *    if nbWorkers==0, this will have no impact until a new compression is started.
+ *    if nbWorkers>=1, new parameters will be picked up at next job,
+ *       with a few restrictions (windowLog, pledgedSrcSize, nbWorkers, jobSize, and overlapLog are not updated).
+ */
+ZSTDLIB_API size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params);
+
+
+/*===   Advanced parameters for decompression API  ===*/
+
+/* The following parameters must be set after creating a ZSTD_DCtx* (or ZSTD_DStream*) object,
+ * but before starting decompression of a frame.
+ */
+
+/*! ZSTD_DCtx_loadDictionary() :
+ *  Create an internal DDict from dict buffer,
+ *  to be used to decompress next frames.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Special : Adding a NULL (or 0-size) dictionary invalidates any previous dictionary,
+ *            meaning "return to no-dictionary mode".
+ *  Note 1 : `dict` content will be copied internally.
+ *            Use ZSTD_DCtx_loadDictionary_byReference()
+ *            to reference dictionary content instead.
+ *            In which case, the dictionary buffer must outlive its users.
+ *  Note 2 : Loading a dictionary involves building tables,
+ *           which has a non-negligible impact on CPU usage and latency.
+ *  Note 3 : Use ZSTD_DCtx_loadDictionary_advanced() to select
+ *           how dictionary content will be interpreted and loaded.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType);
+
+
+/*! ZSTD_DCtx_refDDict() :
+ *  Reference a prepared dictionary, to be used to decompress next frames.
+ *  The dictionary remains active for decompression of future frames using same DCtx.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : Currently, only one dictionary can be managed.
+ *           Referencing a new dictionary effectively "discards" any previous one.
+ *  Special : adding a NULL DDict means "return to no-dictionary mode".
+ *  Note 2 : DDict is just referenced, its lifetime must outlive its usage from DCtx.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+
+/*! ZSTD_DCtx_refPrefix() :
+ *  Reference a prefix (single-usage dictionary) for next compression job.
+ *  Prefix is **only used once**. It must be explicitly referenced before each frame.
+ *  If there is a need to use same prefix multiple times, consider embedding it into a ZSTD_DDict instead.
+ * @result : 0, or an error code (which can be tested with ZSTD_isError()).
+ *  Note 1 : Adding any prefix (including NULL) invalidates any previously set prefix or dictionary
+ *  Note 2 : Prefix buffer is referenced. It must outlive compression job.
+ *  Note 3 : By default, the prefix is treated as raw content (ZSTD_dm_rawContent).
+ *           Use ZSTD_CCtx_refPrefix_advanced() to alter dictMode.
+ *  Note 4 : Referencing a raw content prefix has almost no cpu nor memory cost.
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize);
+ZSTDLIB_API size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType);
+
+
+/*! ZSTD_DCtx_setMaxWindowSize() :
+ *  Refuses allocating internal buffers for frames requiring a window size larger than provided limit.
+ *  This is useful to prevent a decoder context from reserving too much memory for itself (potential attack scenario).
+ *  This parameter is only useful in streaming mode, since no internal buffer is allocated in direct mode.
+ *  By default, a decompression context accepts all window sizes <= (1 << ZSTD_WINDOWLOG_MAX)
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize);
+
+
+/*! ZSTD_DCtx_setFormat() :
+ *  Instruct the decoder context about what kind of data to decode next.
+ *  This instruction is mandatory to decode data without a fully-formed header,
+ *  such ZSTD_f_zstd1_magicless for example.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()).
+ */
+ZSTDLIB_API size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format);
+
+
+/*! ZSTD_decompress_generic() :
+ *  Behave the same as ZSTD_decompressStream.
+ *  Decompression parameters cannot be changed once decompression is started.
+ * @return : an error code, which can be tested using ZSTD_isError()
+ *           if >0, a hint, nb of expected input bytes for next invocation.
+ *           `0` means : a frame has just been fully decoded and flushed.
+ */
+ZSTDLIB_API size_t ZSTD_decompress_generic(ZSTD_DCtx* dctx,
+                                           ZSTD_outBuffer* output,
+                                           ZSTD_inBuffer* input);
+
+
+/*! ZSTD_decompress_generic_simpleArgs() :
+ *  Same as ZSTD_decompress_generic(),
+ *  but using only integral types as arguments.
+ *  Argument list is larger than ZSTD_{in,out}Buffer,
+ *  but can be helpful for binders from dynamic languages
+ *  which have troubles handling structures containing memory pointers.
+ */
+ZSTDLIB_API size_t ZSTD_decompress_generic_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos);
+
+
+/*! ZSTD_DCtx_reset() :
+ *  Return a DCtx to clean state.
+ *  If a decompression was ongoing, any internal data not yet flushed is cancelled.
+ *  All parameters are back to default values, including sticky ones.
+ *  Dictionary (if any) is dropped.
+ *  Parameters can be modified again after a reset.
+ */
+ZSTDLIB_API void ZSTD_DCtx_reset(ZSTD_DCtx* dctx);
+
+
+
+/* ============================ */
+/**       Block level API       */
+/* ============================ */
+
+/*!
+    Block functions produce and decode raw zstd blocks, without frame metadata.
+    Frame metadata cost is typically ~18 bytes, which can be non-negligible for very small blocks (< 100 bytes).
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Compressing and decompressing require a context structure
+      + Use ZSTD_createCCtx() and ZSTD_createDCtx()
+    - It is necessary to init context before starting
+      + compression : any ZSTD_compressBegin*() variant, including with dictionary
+      + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
+      + copyCCtx() and copyDCtx() can be used too
+    - Block size is limited, it must be <= ZSTD_getBlockSize() <= ZSTD_BLOCKSIZE_MAX == 128 KB
+      + If input is larger than a block size, it's necessary to split input data into multiple blocks
+      + For inputs larger than a single block size, consider using the regular ZSTD_compress() instead.
+        Frame metadata is not that costly, and quickly becomes negligible as source size grows larger.
+    - When a block is considered not compressible enough, ZSTD_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTD_decompressBlock() doesn't accept uncompressed data as input !!!
+      + In case of multiple successive blocks, should some of them be uncompressed,
+        decoder must be informed of their existence in order to follow proper history.
+        Use ZSTD_insertBlock() for such a case.
+*/
+
+#define ZSTD_BLOCKSIZELOG_MAX 17
+#define ZSTD_BLOCKSIZE_MAX   (1<<ZSTD_BLOCKSIZELOG_MAX)   /* define, for static allocation */
+/*=====   Raw zstd block functions  =====*/
+ZSTDLIB_API size_t ZSTD_getBlockSize   (const ZSTD_CCtx* cctx);
+ZSTDLIB_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_API size_t ZSTD_insertBlock    (ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert uncompressed block into `dctx` history. Useful for multi-blocks decompression. */
+
+
+#endif   /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/vendor/github.com/DataDog/zstd/zstd_common.c b/vendor/github.com/DataDog/zstd/zstd_common.c
new file mode 100644
index 0000000..bccc948
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_common.c
@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdlib.h>      /* malloc, calloc, free */
+#include <string.h>      /* memset */
+#include "error_private.h"
+#include "zstd_internal.h"
+
+
+/*-****************************************
+*  Version
+******************************************/
+unsigned ZSTD_versionNumber(void) { return ZSTD_VERSION_NUMBER; }
+
+const char* ZSTD_versionString(void) { return ZSTD_VERSION_STRING; }
+
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTD_isError() :
+ *  tells if a return value is an error code */
+unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTD_getErrorName() :
+ *  provides error code string from function result (useful for debugging) */
+const char* ZSTD_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+/*! ZSTD_getError() :
+ *  convert a `size_t` function result into a proper ZSTD_errorCode enum */
+ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
+
+/*! ZSTD_getErrorString() :
+ *  provides error code string from enum */
+const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
+
+/*! g_debuglog_enable :
+ *  turn on/off debug traces (global switch) */
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG >= 2)
+int g_debuglog_enable = 1;
+#endif
+
+
+/*=**************************************************************
+*  Custom allocator
+****************************************************************/
+void* ZSTD_malloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc)
+        return customMem.customAlloc(customMem.opaque, size);
+    return malloc(size);
+}
+
+void* ZSTD_calloc(size_t size, ZSTD_customMem customMem)
+{
+    if (customMem.customAlloc) {
+        /* calloc implemented as malloc+memset;
+         * not as efficient as calloc, but next best guess for custom malloc */
+        void* const ptr = customMem.customAlloc(customMem.opaque, size);
+        memset(ptr, 0, size);
+        return ptr;
+    }
+    return calloc(1, size);
+}
+
+void ZSTD_free(void* ptr, ZSTD_customMem customMem)
+{
+    if (ptr!=NULL) {
+        if (customMem.customFree)
+            customMem.customFree(customMem.opaque, ptr);
+        else
+            free(ptr);
+    }
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_compress.c b/vendor/github.com/DataDog/zstd/zstd_compress.c
new file mode 100644
index 0000000..2aa26da
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_compress.c
@@ -0,0 +1,3449 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*-*************************************
+*  Tuning parameters
+***************************************/
+#ifndef ZSTD_CLEVEL_DEFAULT
+#  define ZSTD_CLEVEL_DEFAULT 3
+#endif
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <string.h>         /* memset */
+#include "cpu.h"
+#include "mem.h"
+#define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "zstd_compress_internal.h"
+#include "zstd_fast.h"
+#include "zstd_double_fast.h"
+#include "zstd_lazy.h"
+#include "zstd_opt.h"
+#include "zstd_ldm.h"
+
+
+/*-*************************************
+*  Helper functions
+***************************************/
+size_t ZSTD_compressBound(size_t srcSize) {
+    return ZSTD_COMPRESSBOUND(srcSize);
+}
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+struct ZSTD_CDict_s {
+    void* dictBuffer;
+    const void* dictContent;
+    size_t dictContentSize;
+    void* workspace;
+    size_t workspaceSize;
+    ZSTD_matchState_t matchState;
+    ZSTD_compressedBlockState_t cBlockState;
+    ZSTD_compressionParameters cParams;
+    ZSTD_customMem customMem;
+    U32 dictID;
+};  /* typedef'd to ZSTD_CDict within "zstd.h" */
+
+ZSTD_CCtx* ZSTD_createCCtx(void)
+{
+    return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    ZSTD_STATIC_ASSERT(zcss_init==0);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_calloc(sizeof(ZSTD_CCtx), customMem);
+        if (!cctx) return NULL;
+        cctx->customMem = customMem;
+        cctx->requestedParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        cctx->requestedParams.fParams.contentSizeFlag = 1;
+        cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+        return cctx;
+    }
+}
+
+ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_CCtx* const cctx = (ZSTD_CCtx*) workspace;
+    if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL;  /* minimum size */
+    if ((size_t)workspace & 7) return NULL;  /* must be 8-aligned */
+    memset(workspace, 0, workspaceSize);   /* may be a bit generous, could memset be smaller ? */
+    cctx->staticSize = workspaceSize;
+    cctx->workSpace = (void*)(cctx+1);
+    cctx->workSpaceSize = workspaceSize - sizeof(ZSTD_CCtx);
+
+    /* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+    if (cctx->workSpaceSize < HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t)) return NULL;
+    assert(((size_t)cctx->workSpace & (sizeof(void*)-1)) == 0);   /* ensure correct alignment */
+    cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)cctx->workSpace;
+    cctx->blockState.nextCBlock = cctx->blockState.prevCBlock + 1;
+    {
+        void* const ptr = cctx->blockState.nextCBlock + 1;
+        cctx->entropyWorkspace = (U32*)ptr;
+    }
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    return cctx;
+}
+
+size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support free on NULL */
+    if (cctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static CCtx */
+    ZSTD_free(cctx->workSpace, cctx->customMem); cctx->workSpace = NULL;
+    ZSTD_freeCDict(cctx->cdictLocal); cctx->cdictLocal = NULL;
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
+#endif
+    ZSTD_free(cctx, cctx->customMem);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+
+static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    return ZSTDMT_sizeof_CCtx(cctx->mtctx);
+#else
+    (void) cctx;
+    return 0;
+#endif
+}
+
+
+size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*cctx) + cctx->workSpaceSize
+           + ZSTD_sizeof_CDict(cctx->cdictLocal)
+           + ZSTD_sizeof_mtctx(cctx);
+}
+
+size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
+{
+    return ZSTD_sizeof_CCtx(zcs);  /* same object */
+}
+
+/* private API call, for dictBuilder only */
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(CCtxParams->compressionLevel, srcSizeHint, dictSize);
+    if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    if (CCtxParams->cParams.windowLog) cParams.windowLog = CCtxParams->cParams.windowLog;
+    if (CCtxParams->cParams.hashLog) cParams.hashLog = CCtxParams->cParams.hashLog;
+    if (CCtxParams->cParams.chainLog) cParams.chainLog = CCtxParams->cParams.chainLog;
+    if (CCtxParams->cParams.searchLog) cParams.searchLog = CCtxParams->cParams.searchLog;
+    if (CCtxParams->cParams.searchLength) cParams.searchLength = CCtxParams->cParams.searchLength;
+    if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
+    if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
+    return cParams;
+}
+
+static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
+        ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params cctxParams;
+    memset(&cctxParams, 0, sizeof(cctxParams));
+    cctxParams.cParams = cParams;
+    cctxParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;  /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(cParams));
+    cctxParams.fParams.contentSizeFlag = 1;
+    return cctxParams;
+}
+
+static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
+        ZSTD_customMem customMem)
+{
+    ZSTD_CCtx_params* params;
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    params = (ZSTD_CCtx_params*)ZSTD_calloc(
+            sizeof(ZSTD_CCtx_params), customMem);
+    if (!params) { return NULL; }
+    params->customMem = customMem;
+    params->compressionLevel = ZSTD_CLEVEL_DEFAULT;
+    params->fParams.contentSizeFlag = 1;
+    return params;
+}
+
+ZSTD_CCtx_params* ZSTD_createCCtxParams(void)
+{
+    return ZSTD_createCCtxParams_advanced(ZSTD_defaultCMem);
+}
+
+size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
+{
+    if (params == NULL) { return 0; }
+    ZSTD_free(params, params->customMem);
+    return 0;
+}
+
+size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
+{
+    return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
+}
+
+size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
+    if (!cctxParams) { return ERROR(GENERIC); }
+    memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->compressionLevel = compressionLevel;
+    cctxParams->fParams.contentSizeFlag = 1;
+    return 0;
+}
+
+size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
+{
+    if (!cctxParams) { return ERROR(GENERIC); }
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    memset(cctxParams, 0, sizeof(*cctxParams));
+    cctxParams->cParams = params.cParams;
+    cctxParams->fParams = params.fParams;
+    cctxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(params.cParams));
+    return 0;
+}
+
+/* ZSTD_assignParamsToCCtxParams() :
+ * params is presumed valid at this stage */
+static ZSTD_CCtx_params ZSTD_assignParamsToCCtxParams(
+        ZSTD_CCtx_params cctxParams, ZSTD_parameters params)
+{
+    ZSTD_CCtx_params ret = cctxParams;
+    ret.cParams = params.cParams;
+    ret.fParams = params.fParams;
+    ret.compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(params.cParams));
+    return ret;
+}
+
+#define CLAMPCHECK(val,min,max) {            \
+    if (((val)<(min)) | ((val)>(max))) {     \
+        return ERROR(parameter_outOfBound);  \
+}   }
+
+
+static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
+{
+    switch(param)
+    {
+    case ZSTD_p_compressionLevel:
+    case ZSTD_p_hashLog:
+    case ZSTD_p_chainLog:
+    case ZSTD_p_searchLog:
+    case ZSTD_p_minMatch:
+    case ZSTD_p_targetLength:
+    case ZSTD_p_compressionStrategy:
+    case ZSTD_p_compressLiterals:
+        return 1;
+
+    case ZSTD_p_format:
+    case ZSTD_p_windowLog:
+    case ZSTD_p_contentSizeFlag:
+    case ZSTD_p_checksumFlag:
+    case ZSTD_p_dictIDFlag:
+    case ZSTD_p_forceMaxWindow :
+    case ZSTD_p_nbWorkers:
+    case ZSTD_p_jobSize:
+    case ZSTD_p_overlapSizeLog:
+    case ZSTD_p_enableLongDistanceMatching:
+    case ZSTD_p_ldmHashLog:
+    case ZSTD_p_ldmMinMatch:
+    case ZSTD_p_ldmBucketSizeLog:
+    case ZSTD_p_ldmHashEveryLog:
+    default:
+        return 0;
+    }
+}
+
+size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParameter (%u, %u)", (U32)param, value);
+    if (cctx->streamStage != zcss_init) {
+        if (ZSTD_isUpdateAuthorized(param)) {
+            cctx->cParamsChanged = 1;
+        } else {
+            return ERROR(stage_wrong);
+    }   }
+
+    switch(param)
+    {
+    case ZSTD_p_format :
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_compressionLevel:
+        if (cctx->cdict) return ERROR(stage_wrong);
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_windowLog:
+    case ZSTD_p_hashLog:
+    case ZSTD_p_chainLog:
+    case ZSTD_p_searchLog:
+    case ZSTD_p_minMatch:
+    case ZSTD_p_targetLength:
+    case ZSTD_p_compressionStrategy:
+        if (cctx->cdict) return ERROR(stage_wrong);
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_compressLiterals:
+    case ZSTD_p_contentSizeFlag:
+    case ZSTD_p_checksumFlag:
+    case ZSTD_p_dictIDFlag:
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_forceMaxWindow :  /* Force back-references to remain < windowSize,
+                                   * even when referencing into Dictionary content.
+                                   * default : 0 when using a CDict, 1 when using a Prefix */
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_nbWorkers:
+        if ((value>0) && cctx->staticSize) {
+            return ERROR(parameter_unsupported);  /* MT not compatible with static alloc */
+        }
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_jobSize:
+    case ZSTD_p_overlapSizeLog:
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    case ZSTD_p_enableLongDistanceMatching:
+    case ZSTD_p_ldmHashLog:
+    case ZSTD_p_ldmMinMatch:
+    case ZSTD_p_ldmBucketSizeLog:
+    case ZSTD_p_ldmHashEveryLog:
+        if (cctx->cdict) return ERROR(stage_wrong);
+        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+
+    default: return ERROR(parameter_unsupported);
+    }
+}
+
+size_t ZSTD_CCtxParam_setParameter(
+        ZSTD_CCtx_params* CCtxParams, ZSTD_cParameter param, unsigned value)
+{
+    DEBUGLOG(4, "ZSTD_CCtxParam_setParameter (%u, %u)", (U32)param, value);
+    switch(param)
+    {
+    case ZSTD_p_format :
+        if (value > (unsigned)ZSTD_f_zstd1_magicless)
+            return ERROR(parameter_unsupported);
+        CCtxParams->format = (ZSTD_format_e)value;
+        return (size_t)CCtxParams->format;
+
+    case ZSTD_p_compressionLevel : {
+        int cLevel = (int)value;  /* cast expected to restore negative sign */
+        if (cLevel > ZSTD_maxCLevel()) cLevel = ZSTD_maxCLevel();
+        if (cLevel) {  /* 0 : does not change current level */
+            CCtxParams->disableLiteralCompression = (cLevel<0);  /* negative levels disable huffman */
+            CCtxParams->compressionLevel = cLevel;
+        }
+        if (CCtxParams->compressionLevel >= 0) return CCtxParams->compressionLevel;
+        return 0;  /* return type (size_t) cannot represent negative values */
+    }
+
+    case ZSTD_p_windowLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+        CCtxParams->cParams.windowLog = value;
+        return CCtxParams->cParams.windowLog;
+
+    case ZSTD_p_hashLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+        CCtxParams->cParams.hashLog = value;
+        return CCtxParams->cParams.hashLog;
+
+    case ZSTD_p_chainLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
+        CCtxParams->cParams.chainLog = value;
+        return CCtxParams->cParams.chainLog;
+
+    case ZSTD_p_searchLog :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
+        CCtxParams->cParams.searchLog = value;
+        return value;
+
+    case ZSTD_p_minMatch :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+        CCtxParams->cParams.searchLength = value;
+        return CCtxParams->cParams.searchLength;
+
+    case ZSTD_p_targetLength :
+        /* all values are valid. 0 => use default */
+        CCtxParams->cParams.targetLength = value;
+        return CCtxParams->cParams.targetLength;
+
+    case ZSTD_p_compressionStrategy :
+        if (value>0)   /* 0 => use default */
+            CLAMPCHECK(value, (unsigned)ZSTD_fast, (unsigned)ZSTD_btultra);
+        CCtxParams->cParams.strategy = (ZSTD_strategy)value;
+        return (size_t)CCtxParams->cParams.strategy;
+
+    case ZSTD_p_compressLiterals:
+        CCtxParams->disableLiteralCompression = !value;
+        return !CCtxParams->disableLiteralCompression;
+
+    case ZSTD_p_contentSizeFlag :
+        /* Content size written in frame header _when known_ (default:1) */
+        DEBUGLOG(4, "set content size flag = %u", (value>0));
+        CCtxParams->fParams.contentSizeFlag = value > 0;
+        return CCtxParams->fParams.contentSizeFlag;
+
+    case ZSTD_p_checksumFlag :
+        /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
+        CCtxParams->fParams.checksumFlag = value > 0;
+        return CCtxParams->fParams.checksumFlag;
+
+    case ZSTD_p_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
+        DEBUGLOG(4, "set dictIDFlag = %u", (value>0));
+        CCtxParams->fParams.noDictIDFlag = !value;
+        return !CCtxParams->fParams.noDictIDFlag;
+
+    case ZSTD_p_forceMaxWindow :
+        CCtxParams->forceWindow = (value > 0);
+        return CCtxParams->forceWindow;
+
+    case ZSTD_p_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        if (value>0) return ERROR(parameter_unsupported);
+        return 0;
+#else
+        return ZSTDMT_CCtxParam_setNbWorkers(CCtxParams, value);
+#endif
+
+    case ZSTD_p_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        return ERROR(parameter_unsupported);
+#else
+        return ZSTDMT_CCtxParam_setMTCtxParameter(CCtxParams, ZSTDMT_p_jobSize, value);
+#endif
+
+    case ZSTD_p_overlapSizeLog :
+#ifndef ZSTD_MULTITHREAD
+        return ERROR(parameter_unsupported);
+#else
+        return ZSTDMT_CCtxParam_setMTCtxParameter(CCtxParams, ZSTDMT_p_overlapSectionLog, value);
+#endif
+
+    case ZSTD_p_enableLongDistanceMatching :
+        CCtxParams->ldmParams.enableLdm = (value>0);
+        return CCtxParams->ldmParams.enableLdm;
+
+    case ZSTD_p_ldmHashLog :
+        if (value>0)   /* 0 ==> auto */
+            CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+        CCtxParams->ldmParams.hashLog = value;
+        return CCtxParams->ldmParams.hashLog;
+
+    case ZSTD_p_ldmMinMatch :
+        if (value>0)   /* 0 ==> default */
+            CLAMPCHECK(value, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX);
+        CCtxParams->ldmParams.minMatchLength = value;
+        return CCtxParams->ldmParams.minMatchLength;
+
+    case ZSTD_p_ldmBucketSizeLog :
+        if (value > ZSTD_LDM_BUCKETSIZELOG_MAX)
+            return ERROR(parameter_outOfBound);
+        CCtxParams->ldmParams.bucketSizeLog = value;
+        return CCtxParams->ldmParams.bucketSizeLog;
+
+    case ZSTD_p_ldmHashEveryLog :
+        if (value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN)
+            return ERROR(parameter_outOfBound);
+        CCtxParams->ldmParams.hashEveryLog = value;
+        return CCtxParams->ldmParams.hashEveryLog;
+
+    default: return ERROR(parameter_unsupported);
+    }
+}
+
+/** ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  just applies `params` into `cctx`
+ *  no action is performed, parameters are merely stored.
+ *  If ZSTDMT is enabled, parameters are pushed to cctx->mtctx.
+ *    This is possible even if a compression is ongoing.
+ *    In which case, new parameters will be applied on the fly, starting with next compression job.
+ */
+size_t ZSTD_CCtx_setParametersUsingCCtxParams(
+        ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    if (cctx->cdict) return ERROR(stage_wrong);
+
+    cctx->requestedParams = *params;
+    return 0;
+}
+
+ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %u bytes", (U32)pledgedSrcSize);
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+    return 0;
+}
+
+size_t ZSTD_CCtx_loadDictionary_advanced(
+        ZSTD_CCtx* cctx, const void* dict, size_t dictSize,
+        ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    if (cctx->staticSize) return ERROR(memory_allocation);  /* no malloc for static CCtx */
+    DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
+    ZSTD_freeCDict(cctx->cdictLocal);  /* in case one already exists */
+    if (dict==NULL || dictSize==0) {   /* no dictionary mode */
+        cctx->cdictLocal = NULL;
+        cctx->cdict = NULL;
+    } else {
+        ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(&cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, dictSize);
+        cctx->cdictLocal = ZSTD_createCDict_advanced(
+                                dict, dictSize,
+                                dictLoadMethod, dictContentType,
+                                cParams, cctx->customMem);
+        cctx->cdict = cctx->cdictLocal;
+        if (cctx->cdictLocal == NULL)
+            return ERROR(memory_allocation);
+    }
+    return 0;
+}
+
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(
+      ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_CCtx_loadDictionary_advanced(
+            cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+
+size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    cctx->cdict = cdict;
+    memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));  /* exclusive */
+    return 0;
+}
+
+size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+size_t ZSTD_CCtx_refPrefix_advanced(
+        ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    cctx->cdict = NULL;   /* prefix discards any prior cdict */
+    cctx->prefixDict.dict = prefix;
+    cctx->prefixDict.dictSize = prefixSize;
+    cctx->prefixDict.dictContentType = dictContentType;
+    return 0;
+}
+
+static void ZSTD_startNewCompression(ZSTD_CCtx* cctx)
+{
+    cctx->streamStage = zcss_init;
+    cctx->pledgedSrcSizePlusOne = 0;
+}
+
+/*! ZSTD_CCtx_reset() :
+ *  Also dumps dictionary */
+void ZSTD_CCtx_reset(ZSTD_CCtx* cctx)
+{
+    ZSTD_startNewCompression(cctx);
+    cctx->cdict = NULL;
+}
+
+/** ZSTD_checkCParams() :
+    control CParam values remain within authorized range.
+    @return : 0, or an error code if one value is beyond authorized range */
+size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
+{
+    CLAMPCHECK(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+    CLAMPCHECK(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
+    CLAMPCHECK(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+    CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
+    CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+    if ((U32)(cParams.targetLength) < ZSTD_TARGETLENGTH_MIN)
+        return ERROR(parameter_unsupported);
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra)
+        return ERROR(parameter_unsupported);
+    return 0;
+}
+
+/** ZSTD_clampCParams() :
+ *  make CParam values within valid range.
+ *  @return : valid CParams */
+static ZSTD_compressionParameters ZSTD_clampCParams(ZSTD_compressionParameters cParams)
+{
+#   define CLAMP(val,min,max) {      \
+        if (val<min) val=min;        \
+        else if (val>max) val=max;   \
+    }
+    CLAMP(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+    CLAMP(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
+    CLAMP(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+    CLAMP(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
+    CLAMP(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+    if ((U32)(cParams.targetLength) < ZSTD_TARGETLENGTH_MIN) cParams.targetLength = ZSTD_TARGETLENGTH_MIN;
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) cParams.strategy = ZSTD_btultra;
+    return cParams;
+}
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
+{
+    U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
+    return hashLog - btScale;
+}
+
+/** ZSTD_adjustCParams_internal() :
+    optimize `cPar` for a given input (`srcSize` and `dictSize`).
+    mostly downsizing to reduce memory consumption and initialization latency.
+    Both `srcSize` and `dictSize` are optional (use 0 if unknown).
+    Note : cPar is considered validated at this stage. Use ZSTD_checkCParams() to ensure that condition. */
+ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+{
+    static const U64 minSrcSize = 513; /* (1<<9) + 1 */
+    static const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
+    assert(ZSTD_checkCParams(cPar)==0);
+
+    if (dictSize && (srcSize+1<2) /* srcSize unknown */ )
+        srcSize = minSrcSize;  /* presumed small when there is a dictionary */
+    else if (srcSize == 0)
+        srcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* 0 == unknown : presumed large */
+
+    /* resize windowLog if input is small enough, to use less memory */
+    if ( (srcSize < maxWindowResize)
+      && (dictSize < maxWindowResize) )  {
+        U32 const tSize = (U32)(srcSize + dictSize);
+        static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
+        U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
+                            ZSTD_highbit32(tSize-1) + 1;
+        if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
+    }
+    if (cPar.hashLog > cPar.windowLog) cPar.hashLog = cPar.windowLog;
+    {   U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
+        if (cycleLog > cPar.windowLog)
+            cPar.chainLog -= (cycleLog - cPar.windowLog);
+    }
+
+    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
+        cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* required for frame header */
+
+    return cPar;
+}
+
+ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+{
+    cPar = ZSTD_clampCParams(cPar);
+    return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
+}
+
+static size_t ZSTD_sizeof_matchState(ZSTD_compressionParameters const* cParams, const U32 forCCtx)
+{
+    size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->searchLength==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = ((size_t)1) << hashLog3;
+    size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+    size_t const optPotentialSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits)) * sizeof(U32)
+                          + (ZSTD_OPT_NUM+1) * (sizeof(ZSTD_match_t)+sizeof(ZSTD_optimal_t));
+    size_t const optSpace = (forCCtx && ((cParams->strategy == ZSTD_btopt) ||
+                                         (cParams->strategy == ZSTD_btultra)))
+                                ? optPotentialSpace
+                                : 0;
+    DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
+                (U32)chainSize, (U32)hSize, (U32)h3Size);
+    return tableSpace + optSpace;
+}
+
+size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    /* Estimate CCtx size is supported for single-threaded compression only. */
+    if (params->nbWorkers > 0) { return ERROR(GENERIC); }
+    {   ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, 0, 0);
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
+        U32    const divider = (cParams.searchLength==3) ? 3 : 4;
+        size_t const maxNbSeq = blockSize / divider;
+        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const entropySpace = HUF_WORKSPACE_SIZE;
+        size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
+        size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
+
+        size_t const ldmSpace = ZSTD_ldm_getTableSize(params->ldmParams);
+        size_t const ldmSeqSpace = ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize) * sizeof(rawSeq);
+
+        size_t const neededSpace = entropySpace + blockStateSpace + tokenSpace +
+                                   matchStateSize + ldmSpace + ldmSeqSpace;
+
+        DEBUGLOG(5, "sizeof(ZSTD_CCtx) : %u", (U32)sizeof(ZSTD_CCtx));
+        DEBUGLOG(5, "estimate workSpace : %u", (U32)neededSpace);
+        return sizeof(ZSTD_CCtx) + neededSpace;
+    }
+}
+
+size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
+    return ZSTD_estimateCCtxSize_usingCCtxParams(&params);
+}
+
+static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
+    return ZSTD_estimateCCtxSize_usingCParams(cParams);
+}
+
+size_t ZSTD_estimateCCtxSize(int compressionLevel)
+{
+    int level;
+    size_t memBudget = 0;
+    for (level=1; level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCCtxSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    if (params->nbWorkers > 0) { return ERROR(GENERIC); }
+    {   size_t const CCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(params);
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << params->cParams.windowLog);
+        size_t const inBuffSize = ((size_t)1 << params->cParams.windowLog) + blockSize;
+        size_t const outBuffSize = ZSTD_compressBound(blockSize) + 1;
+        size_t const streamingSize = inBuffSize + outBuffSize;
+
+        return CCtxSize + streamingSize;
+    }
+}
+
+size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
+{
+    ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
+    return ZSTD_estimateCStreamSize_usingCCtxParams(&params);
+}
+
+static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel) {
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
+    return ZSTD_estimateCStreamSize_usingCParams(cParams);
+}
+
+size_t ZSTD_estimateCStreamSize(int compressionLevel) {
+    int level;
+    size_t memBudget = 0;
+    for (level=1; level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCStreamSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
+}
+
+/* ZSTD_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads (non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_getFrameProgression(cctx->mtctx);
+    }
+#endif
+    {   ZSTD_frameProgression fp;
+        size_t const buffered = (cctx->inBuff == NULL) ? 0 :
+                                cctx->inBuffPos - cctx->inToCompress;
+        if (buffered) assert(cctx->inBuffPos >= cctx->inToCompress);
+        assert(buffered <= ZSTD_BLOCKSIZE_MAX);
+        fp.ingested = cctx->consumedSrcSize + buffered;
+        fp.consumed = cctx->consumedSrcSize;
+        fp.produced = cctx->producedCSize;
+        return fp;
+}   }
+
+
+static U32 ZSTD_equivalentCParams(ZSTD_compressionParameters cParams1,
+                                  ZSTD_compressionParameters cParams2)
+{
+    return (cParams1.hashLog  == cParams2.hashLog)
+         & (cParams1.chainLog == cParams2.chainLog)
+         & (cParams1.strategy == cParams2.strategy)   /* opt parser space */
+         & ((cParams1.searchLength==3) == (cParams2.searchLength==3));  /* hashlog3 space */
+}
+
+/** The parameters are equivalent if ldm is not enabled in both sets or
+ *  all the parameters are equivalent. */
+static U32 ZSTD_equivalentLdmParams(ldmParams_t ldmParams1,
+                                    ldmParams_t ldmParams2)
+{
+    return (!ldmParams1.enableLdm && !ldmParams2.enableLdm) ||
+           (ldmParams1.enableLdm == ldmParams2.enableLdm &&
+            ldmParams1.hashLog == ldmParams2.hashLog &&
+            ldmParams1.bucketSizeLog == ldmParams2.bucketSizeLog &&
+            ldmParams1.minMatchLength == ldmParams2.minMatchLength &&
+            ldmParams1.hashEveryLog == ldmParams2.hashEveryLog);
+}
+
+typedef enum { ZSTDb_not_buffered, ZSTDb_buffered } ZSTD_buffered_policy_e;
+
+/* ZSTD_sufficientBuff() :
+ * check internal buffers exist for streaming if buffPol == ZSTDb_buffered .
+ * Note : they are assumed to be correctly sized if ZSTD_equivalentCParams()==1 */
+static U32 ZSTD_sufficientBuff(size_t bufferSize1, size_t blockSize1,
+                            ZSTD_buffered_policy_e buffPol2,
+                            ZSTD_compressionParameters cParams2,
+                            U64 pledgedSrcSize)
+{
+    size_t const windowSize2 = MAX(1, (size_t)MIN(((U64)1 << cParams2.windowLog), pledgedSrcSize));
+    size_t const blockSize2 = MIN(ZSTD_BLOCKSIZE_MAX, windowSize2);
+    size_t const neededBufferSize2 = (buffPol2==ZSTDb_buffered) ? windowSize2 + blockSize2 : 0;
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is windowSize2=%u <= wlog1=%u",
+                (U32)windowSize2, cParams2.windowLog);
+    DEBUGLOG(4, "ZSTD_sufficientBuff: is blockSize2=%u <= blockSize1=%u",
+                (U32)blockSize2, (U32)blockSize1);
+    return (blockSize2 <= blockSize1) /* seqStore space depends on blockSize */
+         & (neededBufferSize2 <= bufferSize1);
+}
+
+/** Equivalence for resetCCtx purposes */
+static U32 ZSTD_equivalentParams(ZSTD_CCtx_params params1,
+                                 ZSTD_CCtx_params params2,
+                                 size_t buffSize1, size_t blockSize1,
+                                 ZSTD_buffered_policy_e buffPol2,
+                                 U64 pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_equivalentParams: pledgedSrcSize=%u", (U32)pledgedSrcSize);
+    return ZSTD_equivalentCParams(params1.cParams, params2.cParams) &&
+           ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams) &&
+           ZSTD_sufficientBuff(buffSize1, blockSize1, buffPol2, params2.cParams, pledgedSrcSize);
+}
+
+static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
+{
+    int i;
+    for (i = 0; i < ZSTD_REP_NUM; ++i)
+        bs->rep[i] = repStartValue[i];
+    bs->entropy.hufCTable_repeatMode = HUF_repeat_none;
+    bs->entropy.offcode_repeatMode = FSE_repeat_none;
+    bs->entropy.matchlength_repeatMode = FSE_repeat_none;
+    bs->entropy.litlength_repeatMode = FSE_repeat_none;
+}
+
+/*! ZSTD_invalidateMatchState()
+ * Invalidate all the matches in the match finder tables.
+ * Requires nextSrc and base to be set (can be NULL).
+ */
+static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
+{
+    ZSTD_window_clear(&ms->window);
+
+    ms->nextToUpdate = ms->window.dictLimit + 1;
+    ms->loadedDictEnd = 0;
+    ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
+}
+
+/*! ZSTD_continueCCtx() :
+ *  reuse CCtx without reset (note : requires no dictionary) */
+static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pledgedSrcSize)
+{
+    size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
+    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
+    DEBUGLOG(4, "ZSTD_continueCCtx: re-use context in place");
+
+    cctx->blockSize = blockSize;   /* previous block size could be different even for same windowLog, due to pledgedSrcSize */
+    cctx->appliedParams = params;
+    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+    cctx->consumedSrcSize = 0;
+    cctx->producedCSize = 0;
+    if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+        cctx->appliedParams.fParams.contentSizeFlag = 0;
+    DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+        (U32)pledgedSrcSize, cctx->appliedParams.fParams.contentSizeFlag);
+    cctx->stage = ZSTDcs_init;
+    cctx->dictID = 0;
+    if (params.ldmParams.enableLdm)
+        ZSTD_window_clear(&cctx->ldmState.window);
+    ZSTD_referenceExternalSequences(cctx, NULL, 0);
+    ZSTD_invalidateMatchState(&cctx->blockState.matchState);
+    ZSTD_reset_compressedBlockState(cctx->blockState.prevCBlock);
+    XXH64_reset(&cctx->xxhState, 0);
+    return 0;
+}
+
+typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e;
+
+static void* ZSTD_reset_matchState(ZSTD_matchState_t* ms, void* ptr, ZSTD_compressionParameters const* cParams, ZSTD_compResetPolicy_e const crp, U32 const forCCtx)
+{
+    size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->searchLength==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = ((size_t)1) << hashLog3;
+    size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+
+    assert(((size_t)ptr & 3) == 0);
+
+    ms->hashLog3 = hashLog3;
+    memset(&ms->window, 0, sizeof(ms->window));
+    ZSTD_invalidateMatchState(ms);
+
+    /* opt parser space */
+    if (forCCtx && ((cParams->strategy == ZSTD_btopt) | (cParams->strategy == ZSTD_btultra))) {
+        DEBUGLOG(4, "reserving optimal parser space");
+        ms->opt.litFreq = (U32*)ptr;
+        ms->opt.litLengthFreq = ms->opt.litFreq + (1<<Litbits);
+        ms->opt.matchLengthFreq = ms->opt.litLengthFreq + (MaxLL+1);
+        ms->opt.offCodeFreq = ms->opt.matchLengthFreq + (MaxML+1);
+        ptr = ms->opt.offCodeFreq + (MaxOff+1);
+        ms->opt.matchTable = (ZSTD_match_t*)ptr;
+        ptr = ms->opt.matchTable + ZSTD_OPT_NUM+1;
+        ms->opt.priceTable = (ZSTD_optimal_t*)ptr;
+        ptr = ms->opt.priceTable + ZSTD_OPT_NUM+1;
+    }
+
+    /* table Space */
+    DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_noMemset);
+    assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
+    if (crp!=ZSTDcrp_noMemset) memset(ptr, 0, tableSpace);   /* reset tables only */
+    ms->hashTable = (U32*)(ptr);
+    ms->chainTable = ms->hashTable + hSize;
+    ms->hashTable3 = ms->chainTable + chainSize;
+    ptr = ms->hashTable3 + h3Size;
+
+    assert(((size_t)ptr & 3) == 0);
+    return ptr;
+}
+
+/*! ZSTD_resetCCtx_internal() :
+    note : `params` are assumed fully validated at this stage */
+static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
+                                      ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                                      ZSTD_compResetPolicy_e const crp,
+                                      ZSTD_buffered_policy_e const zbuff)
+{
+    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u",
+                (U32)pledgedSrcSize, params.cParams.windowLog);
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+
+    if (crp == ZSTDcrp_continue) {
+        if (ZSTD_equivalentParams(zc->appliedParams, params,
+                                zc->inBuffSize, zc->blockSize,
+                                zbuff, pledgedSrcSize)) {
+            DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%u)",
+                        zc->appliedParams.cParams.windowLog, (U32)zc->blockSize);
+            return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
+    }   }
+    DEBUGLOG(4, "ZSTD_equivalentParams()==0 -> reset CCtx");
+
+    if (params.ldmParams.enableLdm) {
+        /* Adjust long distance matching parameters */
+        params.ldmParams.windowLog = params.cParams.windowLog;
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashEveryLog < 32);
+        zc->ldmState.hashPower =
+                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+    }
+
+    {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
+        U32    const divider = (params.cParams.searchLength==3) ? 3 : 4;
+        size_t const maxNbSeq = blockSize / divider;
+        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const buffOutSize = (zbuff==ZSTDb_buffered) ? ZSTD_compressBound(blockSize)+1 : 0;
+        size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0;
+        size_t const matchStateSize = ZSTD_sizeof_matchState(&params.cParams, /* forCCtx */ 1);
+        size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize);
+        void* ptr;
+
+        /* Check if workSpace is large enough, alloc a new one if needed */
+        {   size_t const entropySpace = HUF_WORKSPACE_SIZE;
+            size_t const blockStateSpace = 2 * sizeof(ZSTD_compressedBlockState_t);
+            size_t const bufferSpace = buffInSize + buffOutSize;
+            size_t const ldmSpace = ZSTD_ldm_getTableSize(params.ldmParams);
+            size_t const ldmSeqSpace = maxNbLdmSeq * sizeof(rawSeq);
+
+            size_t const neededSpace = entropySpace + blockStateSpace + ldmSpace +
+                                       ldmSeqSpace + matchStateSize + tokenSpace +
+                                       bufferSpace;
+            DEBUGLOG(4, "Need %uKB workspace, including %uKB for match state, and %uKB for buffers",
+                        (U32)(neededSpace>>10), (U32)(matchStateSize>>10), (U32)(bufferSpace>>10));
+            DEBUGLOG(4, "windowSize: %u - blockSize: %u", (U32)windowSize, (U32)blockSize);
+
+            if (zc->workSpaceSize < neededSpace) {  /* too small : resize */
+                DEBUGLOG(4, "Need to update workSpaceSize from %uK to %uK",
+                            (unsigned)(zc->workSpaceSize>>10),
+                            (unsigned)(neededSpace>>10));
+                /* static cctx : no resize, error out */
+                if (zc->staticSize) return ERROR(memory_allocation);
+
+                zc->workSpaceSize = 0;
+                ZSTD_free(zc->workSpace, zc->customMem);
+                zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
+                if (zc->workSpace == NULL) return ERROR(memory_allocation);
+                zc->workSpaceSize = neededSpace;
+                ptr = zc->workSpace;
+
+                /* Statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+                assert(((size_t)zc->workSpace & 3) == 0);   /* ensure correct alignment */
+                assert(zc->workSpaceSize >= 2 * sizeof(ZSTD_compressedBlockState_t));
+                zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)zc->workSpace;
+                zc->blockState.nextCBlock = zc->blockState.prevCBlock + 1;
+                ptr = zc->blockState.nextCBlock + 1;
+                zc->entropyWorkspace = (U32*)ptr;
+        }   }
+
+        /* init params */
+        zc->appliedParams = params;
+        zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
+        zc->consumedSrcSize = 0;
+        zc->producedCSize = 0;
+        if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+            zc->appliedParams.fParams.contentSizeFlag = 0;
+        DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+            (U32)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
+        zc->blockSize = blockSize;
+
+        XXH64_reset(&zc->xxhState, 0);
+        zc->stage = ZSTDcs_init;
+        zc->dictID = 0;
+
+        ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
+
+        ptr = zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32;
+
+        /* ldm hash table */
+        /* initialize bucketOffsets table later for pointer alignment */
+        if (params.ldmParams.enableLdm) {
+            size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
+            memset(ptr, 0, ldmHSize * sizeof(ldmEntry_t));
+            assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
+            zc->ldmState.hashTable = (ldmEntry_t*)ptr;
+            ptr = zc->ldmState.hashTable + ldmHSize;
+            zc->ldmSequences = (rawSeq*)ptr;
+            ptr = zc->ldmSequences + maxNbLdmSeq;
+            zc->maxNbLdmSequences = maxNbLdmSeq;
+
+            memset(&zc->ldmState.window, 0, sizeof(zc->ldmState.window));
+        }
+        assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
+
+        ptr = ZSTD_reset_matchState(&zc->blockState.matchState, ptr, &params.cParams, crp, /* forCCtx */ 1);
+
+        /* sequences storage */
+        zc->seqStore.sequencesStart = (seqDef*)ptr;
+        ptr = zc->seqStore.sequencesStart + maxNbSeq;
+        zc->seqStore.llCode = (BYTE*) ptr;
+        zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
+        zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
+        zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
+        ptr = zc->seqStore.litStart + blockSize;
+
+        /* ldm bucketOffsets table */
+        if (params.ldmParams.enableLdm) {
+            size_t const ldmBucketSize =
+                  ((size_t)1) << (params.ldmParams.hashLog -
+                                  params.ldmParams.bucketSizeLog);
+            memset(ptr, 0, ldmBucketSize);
+            zc->ldmState.bucketOffsets = (BYTE*)ptr;
+            ptr = zc->ldmState.bucketOffsets + ldmBucketSize;
+            ZSTD_window_clear(&zc->ldmState.window);
+        }
+        ZSTD_referenceExternalSequences(zc, NULL, 0);
+
+        /* buffers */
+        zc->inBuffSize = buffInSize;
+        zc->inBuff = (char*)ptr;
+        zc->outBuffSize = buffOutSize;
+        zc->outBuff = zc->inBuff + buffInSize;
+
+        return 0;
+    }
+}
+
+/* ZSTD_invalidateRepCodes() :
+ * ensures next compression will not use repcodes from previous block.
+ * Note : only works with regular variant;
+ *        do not use with extDict variant ! */
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
+    int i;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->blockState.prevCBlock->rep[i] = 0;
+    assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
+}
+
+static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            unsigned windowLog,
+                            ZSTD_frameParameters fParams,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    {   ZSTD_CCtx_params params = cctx->requestedParams;
+        /* Copy only compression parameters related to tables. */
+        params.cParams = cdict->cParams;
+        if (windowLog) params.cParams.windowLog = windowLog;
+        params.fParams = fParams;
+        ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                ZSTDcrp_noMemset, zbuff);
+        assert(cctx->appliedParams.cParams.strategy == cdict->cParams.strategy);
+        assert(cctx->appliedParams.cParams.hashLog == cdict->cParams.hashLog);
+        assert(cctx->appliedParams.cParams.chainLog == cdict->cParams.chainLog);
+    }
+
+    /* copy tables */
+    {   size_t const chainSize = (cdict->cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << cdict->cParams.chainLog);
+        size_t const hSize =  (size_t)1 << cdict->cParams.hashLog;
+        size_t const tableSpace = (chainSize + hSize) * sizeof(U32);
+        assert((U32*)cctx->blockState.matchState.chainTable == (U32*)cctx->blockState.matchState.hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)cctx->blockState.matchState.hashTable3 == (U32*)cctx->blockState.matchState.chainTable + chainSize);
+        assert((U32*)cdict->matchState.chainTable == (U32*)cdict->matchState.hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)cdict->matchState.hashTable3 == (U32*)cdict->matchState.chainTable + chainSize);
+        memcpy(cctx->blockState.matchState.hashTable, cdict->matchState.hashTable, tableSpace);   /* presumes all tables follow each other */
+    }
+    /* Zero the hashTable3, since the cdict never fills it */
+    {   size_t const h3Size = (size_t)1 << cctx->blockState.matchState.hashLog3;
+        assert(cdict->matchState.hashLog3 == 0);
+        memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
+    }
+
+    /* copy dictionary offsets */
+    {
+        ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
+        ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+    cctx->dictID = cdict->dictID;
+
+    /* copy block state */
+    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+/*! ZSTD_copyCCtx_internal() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  The "context", in this case, refers to the hash and chain tables,
+ *  entropy tables, and dictionary references.
+ * `windowLog` value is enforced if != 0, otherwise value is copied from srcCCtx.
+ * @return : 0, or an error code */
+static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
+                            const ZSTD_CCtx* srcCCtx,
+                            ZSTD_frameParameters fParams,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    DEBUGLOG(5, "ZSTD_copyCCtx_internal");
+    if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
+
+    memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
+    {   ZSTD_CCtx_params params = dstCCtx->requestedParams;
+        /* Copy only compression parameters related to tables. */
+        params.cParams = srcCCtx->appliedParams.cParams;
+        params.fParams = fParams;
+        ZSTD_resetCCtx_internal(dstCCtx, params, pledgedSrcSize,
+                                ZSTDcrp_noMemset, zbuff);
+        assert(dstCCtx->appliedParams.cParams.windowLog == srcCCtx->appliedParams.cParams.windowLog);
+        assert(dstCCtx->appliedParams.cParams.strategy == srcCCtx->appliedParams.cParams.strategy);
+        assert(dstCCtx->appliedParams.cParams.hashLog == srcCCtx->appliedParams.cParams.hashLog);
+        assert(dstCCtx->appliedParams.cParams.chainLog == srcCCtx->appliedParams.cParams.chainLog);
+        assert(dstCCtx->blockState.matchState.hashLog3 == srcCCtx->blockState.matchState.hashLog3);
+    }
+
+    /* copy tables */
+    {   size_t const chainSize = (srcCCtx->appliedParams.cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << srcCCtx->appliedParams.cParams.chainLog);
+        size_t const hSize =  (size_t)1 << srcCCtx->appliedParams.cParams.hashLog;
+        size_t const h3Size = (size_t)1 << srcCCtx->blockState.matchState.hashLog3;
+        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+        assert((U32*)dstCCtx->blockState.matchState.chainTable == (U32*)dstCCtx->blockState.matchState.hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)dstCCtx->blockState.matchState.hashTable3 == (U32*)dstCCtx->blockState.matchState.chainTable + chainSize);
+        memcpy(dstCCtx->blockState.matchState.hashTable, srcCCtx->blockState.matchState.hashTable, tableSpace);   /* presumes all tables follow each other */
+    }
+
+    /* copy dictionary offsets */
+    {
+        ZSTD_matchState_t const* srcMatchState = &srcCCtx->blockState.matchState;
+        ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+    dstCCtx->dictID = srcCCtx->dictID;
+
+    /* copy block state */
+    memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
+
+    return 0;
+}
+
+/*! ZSTD_copyCCtx() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  pledgedSrcSize==0 means "unknown".
+*   @return : 0, or an error code */
+size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
+{
+    ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    ZSTD_buffered_policy_e const zbuff = (ZSTD_buffered_policy_e)(srcCCtx->inBuffSize>0);
+    ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
+
+    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx,
+                                fParams, pledgedSrcSize,
+                                zbuff);
+}
+
+
+#define ZSTD_ROWSIZE 16
+/*! ZSTD_reduceTable() :
+ *  reduce table indexes by `reducerValue`, or squash to zero.
+ *  PreserveMark preserves "unsorted mark" for btlazy2 strategy.
+ *  It must be set to a clear 0/1 value, to remove branch during inlining.
+ *  Presume table size is a multiple of ZSTD_ROWSIZE
+ *  to help auto-vectorization */
+FORCE_INLINE_TEMPLATE void
+ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerValue, int const preserveMark)
+{
+    int const nbRows = (int)size / ZSTD_ROWSIZE;
+    int cellNb = 0;
+    int rowNb;
+    assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
+    assert(size < (1U<<31));   /* can be casted to int */
+    for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
+        int column;
+        for (column=0; column<ZSTD_ROWSIZE; column++) {
+            if (preserveMark) {
+                U32 const adder = (table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) ? reducerValue : 0;
+                table[cellNb] += adder;
+            }
+            if (table[cellNb] < reducerValue) table[cellNb] = 0;
+            else table[cellNb] -= reducerValue;
+            cellNb++;
+    }   }
+}
+
+static void ZSTD_reduceTable(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 0);
+}
+
+static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 1);
+}
+
+/*! ZSTD_reduceIndex() :
+*   rescale all indexes to avoid future overflow (indexes are U32) */
+static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
+{
+    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
+    {   U32 const hSize = (U32)1 << zc->appliedParams.cParams.hashLog;
+        ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
+    }
+
+    if (zc->appliedParams.cParams.strategy != ZSTD_fast) {
+        U32 const chainSize = (U32)1 << zc->appliedParams.cParams.chainLog;
+        if (zc->appliedParams.cParams.strategy == ZSTD_btlazy2)
+            ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
+        else
+            ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
+    }
+
+    if (ms->hashLog3) {
+        U32 const h3Size = (U32)1 << ms->hashLog3;
+        ZSTD_reduceTable(ms->hashTable3, h3Size, reducerValue);
+    }
+}
+
+
+/*-*******************************************************
+*  Block entropic compression
+*********************************************************/
+
+/* See doc/zstd_compression_format.md for detailed format description */
+
+size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (srcSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
+    MEM_writeLE24(dst, (U32)(srcSize << 2) + (U32)bt_raw);
+    return ZSTD_blockHeaderSize+srcSize;
+}
+
+
+static size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE* const)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    if (srcSize + flSize > dstCapacity) return ERROR(dstSize_tooSmall);
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
+            break;
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
+            break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
+    }
+
+    memcpy(ostart + flSize, src, srcSize);
+    return srcSize + flSize;
+}
+
+static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE* const)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    (void)dstCapacity;  /* dstCapacity already guaranteed to be >=4, hence large enough */
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
+            break;
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
+            break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
+    }
+
+    ostart[flSize] = *(const BYTE*)src;
+    return flSize+1;
+}
+
+
+static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
+
+static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t const* prevEntropy,
+                                     ZSTD_entropyCTables_t* nextEntropy,
+                                     ZSTD_strategy strategy, int disableLiteralCompression,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     U32* workspace, const int bmi2)
+{
+    size_t const minGain = ZSTD_minGain(srcSize);
+    size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
+    BYTE*  const ostart = (BYTE*)dst;
+    U32 singleStream = srcSize < 256;
+    symbolEncodingType_e hType = set_compressed;
+    size_t cLitSize;
+
+    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i)",
+                disableLiteralCompression);
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    nextEntropy->hufCTable_repeatMode = prevEntropy->hufCTable_repeatMode;
+    memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable,
+           sizeof(prevEntropy->hufCTable));
+
+    if (disableLiteralCompression)
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+
+    /* small ? don't even attempt compression (speed opt) */
+#   define COMPRESS_LITERALS_SIZE_MIN 63
+    {   size_t const minLitSize = (prevEntropy->hufCTable_repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
+        if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }
+
+    if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall);   /* not enough space for compression */
+    {   HUF_repeat repeat = prevEntropy->hufCTable_repeatMode;
+        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
+        cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
+                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2)
+                                : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
+                                      workspace, HUF_WORKSPACE_SIZE, (HUF_CElt*)nextEntropy->hufCTable, &repeat, preferRepeat, bmi2);
+        if (repeat != HUF_repeat_none) {
+            /* reused the existing table */
+            hType = set_repeat;
+        }
+    }
+
+    if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
+        memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }
+    if (cLitSize==1) {
+        memcpy(nextEntropy->hufCTable, prevEntropy->hufCTable, sizeof(prevEntropy->hufCTable));
+        return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }
+
+    if (hType == set_compressed) {
+        /* using a newly constructed table */
+        nextEntropy->hufCTable_repeatMode = HUF_repeat_check;
+    }
+
+    /* Build header */
+    switch(lhSize)
+    {
+    case 3: /* 2 - 2 - 10 - 10 */
+        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
+            MEM_writeLE24(ostart, lhc);
+            break;
+        }
+    case 4: /* 2 - 2 - 14 - 14 */
+        {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
+            MEM_writeLE32(ostart, lhc);
+            break;
+        }
+    case 5: /* 2 - 2 - 18 - 18 */
+        {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
+            MEM_writeLE32(ostart, lhc);
+            ostart[4] = (BYTE)(cLitSize >> 10);
+            break;
+        }
+    default:  /* not possible : lhSize is {3,4,5} */
+        assert(0);
+    }
+    return lhSize+cLitSize;
+}
+
+
+void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
+{
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    BYTE* const llCodeTable = seqStorePtr->llCode;
+    BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    U32 u;
+    for (u=0; u<nbSeq; u++) {
+        U32 const llv = sequences[u].litLength;
+        U32 const mlv = sequences[u].matchLength;
+        llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
+        ofCodeTable[u] = (BYTE)ZSTD_highbit32(sequences[u].offset);
+        mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
+    }
+    if (seqStorePtr->longLengthID==1)
+        llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
+    if (seqStorePtr->longLengthID==2)
+        mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
+}
+
+typedef enum {
+    ZSTD_defaultDisallowed = 0,
+    ZSTD_defaultAllowed = 1
+} ZSTD_defaultPolicy_e;
+
+MEM_STATIC
+symbolEncodingType_e ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, size_t const mostFrequent, size_t nbSeq,
+        U32 defaultNormLog, ZSTD_defaultPolicy_e const isDefaultAllowed)
+{
+#define MIN_SEQ_FOR_DYNAMIC_FSE   64
+#define MAX_SEQ_FOR_STATIC_FSE  1000
+    ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
+    if ((mostFrequent == nbSeq) && (!isDefaultAllowed || nbSeq > 2)) {
+        DEBUGLOG(5, "Selected set_rle");
+        /* Prefer set_basic over set_rle when there are 2 or less symbols,
+         * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
+         * If basic encoding isn't possible, always choose RLE.
+         */
+        *repeatMode = FSE_repeat_check;
+        return set_rle;
+    }
+    if ( isDefaultAllowed
+      && (*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+        DEBUGLOG(5, "Selected set_repeat");
+        return set_repeat;
+    }
+    if ( isDefaultAllowed
+      && ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) ) {
+        DEBUGLOG(5, "Selected set_basic");
+        /* The format allows default tables to be repeated, but it isn't useful.
+         * When using simple heuristics to select encoding type, we don't want
+         * to confuse these tables with dictionaries. When running more careful
+         * analysis, we don't need to waste time checking both repeating tables
+         * and default tables.
+         */
+        *repeatMode = FSE_repeat_none;
+        return set_basic;
+    }
+    DEBUGLOG(5, "Selected set_compressed");
+    *repeatMode = FSE_repeat_check;
+    return set_compressed;
+}
+
+MEM_STATIC
+size_t ZSTD_buildCTable(void* dst, size_t dstCapacity,
+        FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+        U32* count, U32 max,
+        BYTE const* codeTable, size_t nbSeq,
+        S16 const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+        FSE_CTable const* prevCTable, size_t prevCTableSize,
+        void* workspace, size_t workspaceSize)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE const* const oend = op + dstCapacity;
+
+    switch (type) {
+    case set_rle:
+        *op = codeTable[0];
+        CHECK_F(FSE_buildCTable_rle(nextCTable, (BYTE)max));
+        return 1;
+    case set_repeat:
+        memcpy(nextCTable, prevCTable, prevCTableSize);
+        return 0;
+    case set_basic:
+        CHECK_F(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize));  /* note : could be pre-calculated */
+        return 0;
+    case set_compressed: {
+        S16 norm[MaxSeq + 1];
+        size_t nbSeq_1 = nbSeq;
+        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+        if (count[codeTable[nbSeq-1]] > 1) {
+            count[codeTable[nbSeq-1]]--;
+            nbSeq_1--;
+        }
+        assert(nbSeq_1 > 1);
+        CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
+        {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
+            if (FSE_isError(NCountSize)) return NCountSize;
+            CHECK_F(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, workspace, workspaceSize));
+            return NCountSize;
+        }
+    }
+    default: return assert(0), ERROR(GENERIC);
+    }
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_encodeSequences_body(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    BIT_CStream_t blockStream;
+    FSE_CState_t  stateMatchLength;
+    FSE_CState_t  stateOffsetBits;
+    FSE_CState_t  stateLitLength;
+
+    CHECK_E(BIT_initCStream(&blockStream, dst, dstCapacity), dstSize_tooSmall); /* not enough space remaining */
+
+    /* first symbols */
+    FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    if (longOffsets) {
+        U32 const ofBits = ofCodeTable[nbSeq-1];
+        int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+        if (extraBits) {
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
+            BIT_flushBits(&blockStream);
+        }
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
+                    ofBits - extraBits);
+    } else {
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+    }
+    BIT_flushBits(&blockStream);
+
+    {   size_t n;
+        for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
+            BYTE const llCode = llCodeTable[n];
+            BYTE const ofCode = ofCodeTable[n];
+            BYTE const mlCode = mlCodeTable[n];
+            U32  const llBits = LL_bits[llCode];
+            U32  const ofBits = ofCode;
+            U32  const mlBits = ML_bits[mlCode];
+            DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
+                        sequences[n].litLength,
+                        sequences[n].matchLength + MINMATCH,
+                        sequences[n].offset);
+                                                                            /* 32b*/  /* 64b*/
+                                                                            /* (7)*/  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
+            FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
+            if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
+            if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
+                BIT_flushBits(&blockStream);                                /* (7)*/
+            BIT_addBits(&blockStream, sequences[n].litLength, llBits);
+            if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
+            BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
+            if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
+            if (longOffsets) {
+                int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+                if (extraBits) {
+                    BIT_addBits(&blockStream, sequences[n].offset, extraBits);
+                    BIT_flushBits(&blockStream);                            /* (7)*/
+                }
+                BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
+                            ofBits - extraBits);                            /* 31 */
+            } else {
+                BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
+            }
+            BIT_flushBits(&blockStream);                                    /* (7)*/
+    }   }
+
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
+    FSE_flushCState(&blockStream, &stateMatchLength);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
+    FSE_flushCState(&blockStream, &stateOffsetBits);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
+    FSE_flushCState(&blockStream, &stateLitLength);
+
+    {   size_t const streamSize = BIT_closeCStream(&blockStream);
+        if (streamSize==0) return ERROR(dstSize_tooSmall);   /* not enough space */
+        return streamSize;
+    }
+}
+
+static size_t
+ZSTD_encodeSequences_default(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+
+#if DYNAMIC_BMI2
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_encodeSequences_bmi2(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+#endif
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
+{
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
+                                         CTable_MatchLength, mlCodeTable,
+                                         CTable_OffsetBits, ofCodeTable,
+                                         CTable_LitLength, llCodeTable,
+                                         sequences, nbSeq, longOffsets);
+    }
+#endif
+    (void)bmi2;
+    return ZSTD_encodeSequences_default(dst, dstCapacity,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq, longOffsets);
+}
+
+MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
+                              ZSTD_entropyCTables_t const* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                              ZSTD_CCtx_params const* cctxParams,
+                              void* dst, size_t dstCapacity, U32* workspace,
+                              const int bmi2)
+{
+    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
+    U32 count[MaxSeq+1];
+    FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
+    U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    BYTE* seqHead;
+
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+
+    /* Compress literals */
+    {   const BYTE* const literals = seqStorePtr->litStart;
+        size_t const litSize = seqStorePtr->lit - literals;
+        size_t const cSize = ZSTD_compressLiterals(
+                                    prevEntropy, nextEntropy,
+                                    cctxParams->cParams.strategy, cctxParams->disableLiteralCompression,
+                                    op, dstCapacity,
+                                    literals, litSize,
+                                    workspace, bmi2);
+        if (ZSTD_isError(cSize))
+          return cSize;
+        assert(cSize <= dstCapacity);
+        op += cSize;
+    }
+
+    /* Sequences Header */
+    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/) return ERROR(dstSize_tooSmall);
+    if (nbSeq < 0x7F)
+        *op++ = (BYTE)nbSeq;
+    else if (nbSeq < LONGNBSEQ)
+        op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
+    else
+        op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    if (nbSeq==0) {
+      memcpy(nextEntropy->litlengthCTable, prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable));
+      nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+      memcpy(nextEntropy->offcodeCTable, prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable));
+      nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+      memcpy(nextEntropy->matchlengthCTable, prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable));
+      nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+      return op - ostart;
+    }
+
+    /* seqHead : flags for FSE encoding type */
+    seqHead = op++;
+
+    /* convert length/distances into codes */
+    ZSTD_seqToCodes(seqStorePtr);
+    /* build CTable for Literal Lengths */
+    {   U32 max = MaxLL;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, workspace);
+        DEBUGLOG(5, "Building LL table");
+        nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+        LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog, ZSTD_defaultAllowed);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
+                    count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                    prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
+                    workspace, HUF_WORKSPACE_SIZE);
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
+    }   }
+    /* build CTable for Offsets */
+    {   U32 max = MaxOff;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, workspace);
+        /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
+        ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
+        DEBUGLOG(5, "Building OF table");
+        nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+        Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode, mostFrequent, nbSeq, OF_defaultNormLog, defaultPolicy);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
+                    count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                    prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
+                    workspace, HUF_WORKSPACE_SIZE);
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
+    }   }
+    /* build CTable for MatchLengths */
+    {   U32 max = MaxML;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, workspace);
+        DEBUGLOG(5, "Building ML table");
+        nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+        MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog, ZSTD_defaultAllowed);
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
+                    count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
+                    prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
+                    workspace, HUF_WORKSPACE_SIZE);
+            if (ZSTD_isError(countSize)) return countSize;
+            op += countSize;
+    }   }
+
+    *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
+
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, oend - op,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        if (ZSTD_isError(bitstreamSize)) return bitstreamSize;
+        op += bitstreamSize;
+    }
+
+    return op - ostart;
+}
+
+MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
+                              ZSTD_entropyCTables_t const* prevEntropy,
+                              ZSTD_entropyCTables_t* nextEntropy,
+                              ZSTD_CCtx_params const* cctxParams,
+                              void* dst, size_t dstCapacity,
+                              size_t srcSize, U32* workspace, int bmi2)
+{
+    size_t const cSize = ZSTD_compressSequences_internal(
+            seqStorePtr, prevEntropy, nextEntropy, cctxParams, dst, dstCapacity,
+            workspace, bmi2);
+    /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
+     * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
+     */
+    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
+        return 0;  /* block not compressed */
+    if (ZSTD_isError(cSize)) return cSize;
+
+    /* Check compressibility */
+    {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize);  /* note : fixed formula, maybe should depend on compression level, or strategy */
+        if (cSize >= maxCSize) return 0;  /* block not compressed */
+    }
+
+    /* We check that dictionaries have offset codes available for the first
+     * block. After the first block, the offcode table might not have large
+     * enough codes to represent the offsets in the data.
+     */
+    if (nextEntropy->offcode_repeatMode == FSE_repeat_valid)
+        nextEntropy->offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+/* ZSTD_selectBlockCompressor() :
+ * Not static, but internal use only (used by long distance matcher)
+ * assumption : strat is a valid strategy */
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
+{
+    static const ZSTD_blockCompressor blockCompressor[2][(unsigned)ZSTD_btultra+1] = {
+        { ZSTD_compressBlock_fast  /* default for 0 */,
+          ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy,
+          ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2,
+          ZSTD_compressBlock_btopt, ZSTD_compressBlock_btultra },
+        { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
+          ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict,
+          ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict,
+          ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btultra_extDict }
+    };
+    ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
+
+    assert((U32)strat >= (U32)ZSTD_fast);
+    assert((U32)strat <= (U32)ZSTD_btultra);
+    return blockCompressor[extDict!=0][(U32)strat];
+}
+
+static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
+                                   const BYTE* anchor, size_t lastLLSize)
+{
+    memcpy(seqStorePtr->lit, anchor, lastLLSize);
+    seqStorePtr->lit += lastLLSize;
+}
+
+static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
+{
+    ssPtr->lit = ssPtr->litStart;
+    ssPtr->sequences = ssPtr->sequencesStart;
+    ssPtr->longLengthID = 0;
+}
+
+static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                                        void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize)
+{
+    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (U32)dstCapacity, ms->window.dictLimit, ms->nextToUpdate);
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+        ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.searchLength);
+        return 0;   /* don't even attempt compression below a certain srcSize */
+    }
+    ZSTD_resetSeqStore(&(zc->seqStore));
+
+    /* limited update after a very long match */
+    {   const BYTE* const base = ms->window.base;
+        const BYTE* const istart = (const BYTE*)src;
+        const U32 current = (U32)(istart-base);
+        if (current > ms->nextToUpdate + 384)
+            ms->nextToUpdate = current - MIN(192, (U32)(current - ms->nextToUpdate - 384));
+    }
+
+    /* select and store sequences */
+    {   U32 const extDict = ZSTD_window_hasExtDict(ms->window);
+        size_t lastLLSize;
+        {   int i;
+            for (i = 0; i < ZSTD_REP_NUM; ++i)
+                zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
+        }
+        if (zc->externSeqStore.pos < zc->externSeqStore.size) {
+            assert(!zc->appliedParams.ldmParams.enableLdm);
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&zc->externSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       &zc->appliedParams.cParams,
+                                       src, srcSize, extDict);
+            assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
+        } else if (zc->appliedParams.ldmParams.enableLdm) {
+            rawSeqStore_t ldmSeqStore = {NULL, 0, 0, 0};
+
+            ldmSeqStore.seq = zc->ldmSequences;
+            ldmSeqStore.capacity = zc->maxNbLdmSequences;
+            /* Updates ldmSeqStore.size */
+            CHECK_F(ZSTD_ldm_generateSequences(&zc->ldmState, &ldmSeqStore,
+                                               &zc->appliedParams.ldmParams,
+                                               src, srcSize));
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&ldmSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       &zc->appliedParams.cParams,
+                                       src, srcSize, extDict);
+            assert(ldmSeqStore.pos == ldmSeqStore.size);
+        } else {   /* not long range mode */
+            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, extDict);
+            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, &zc->appliedParams.cParams, src, srcSize);
+        }
+        {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
+            ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
+    }   }
+
+    /* encode sequences and literals */
+    {   size_t const cSize = ZSTD_compressSequences(&zc->seqStore,
+                                &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+                                &zc->appliedParams,
+                                dst, dstCapacity,
+                                srcSize, zc->entropyWorkspace, zc->bmi2);
+        if (ZSTD_isError(cSize) || cSize == 0) return cSize;
+        /* confirm repcodes and entropy tables */
+        {   ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
+            zc->blockState.prevCBlock = zc->blockState.nextCBlock;
+            zc->blockState.nextCBlock = tmp;
+        }
+        return cSize;
+    }
+}
+
+
+/*! ZSTD_compress_frameChunk() :
+*   Compress a chunk of data into one or multiple blocks.
+*   All blocks will be terminated, all input will be consumed.
+*   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
+*   Frame is supposed already started (header already produced)
+*   @return : compressed size, or an error code
+*/
+static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     U32 lastFrameChunk)
+{
+    size_t blockSize = cctx->blockSize;
+    size_t remaining = srcSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
+    assert(cctx->appliedParams.cParams.windowLog <= 31);
+
+    DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (U32)blockSize);
+    if (cctx->appliedParams.fParams.checksumFlag && srcSize)
+        XXH64_update(&cctx->xxhState, src, srcSize);
+
+    while (remaining) {
+        ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
+        U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
+
+        if (dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE)
+            return ERROR(dstSize_tooSmall);   /* not enough space to store compressed block */
+        if (remaining < blockSize) blockSize = remaining;
+
+        if (ZSTD_window_needOverflowCorrection(ms->window, ip + blockSize)) {
+            U32 const cycleLog = ZSTD_cycleLog(cctx->appliedParams.cParams.chainLog, cctx->appliedParams.cParams.strategy);
+            U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
+            ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+
+            ZSTD_reduceIndex(cctx, correction);
+            if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+            else ms->nextToUpdate -= correction;
+            ms->loadedDictEnd = 0;
+        }
+        ZSTD_window_enforceMaxDist(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd);
+        if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
+
+        {   size_t cSize = ZSTD_compressBlock_internal(cctx,
+                                op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
+                                ip, blockSize);
+            if (ZSTD_isError(cSize)) return cSize;
+
+            if (cSize == 0) {  /* block is not compressible */
+                U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
+                if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
+                MEM_writeLE32(op, cBlockHeader24);   /* 4th byte will be overwritten */
+                memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
+                cSize = ZSTD_blockHeaderSize + blockSize;
+            } else {
+                U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+                MEM_writeLE24(op, cBlockHeader24);
+                cSize += ZSTD_blockHeaderSize;
+            }
+
+            ip += blockSize;
+            assert(remaining >= blockSize);
+            remaining -= blockSize;
+            op += cSize;
+            assert(dstCapacity >= cSize);
+            dstCapacity -= cSize;
+            DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
+                        (U32)cSize);
+    }   }
+
+    if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
+    return op-ostart;
+}
+
+
+static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
+                                    ZSTD_CCtx_params params, U64 pledgedSrcSize, U32 dictID)
+{   BYTE* const op = (BYTE*)dst;
+    U32   const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    U32   const dictIDSizeCode = params.fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
+    U32   const checksumFlag = params.fParams.checksumFlag>0;
+    U32   const windowSize = (U32)1 << params.cParams.windowLog;
+    U32   const singleSegment = params.fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
+    BYTE  const windowLogByte = (BYTE)((params.cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
+    U32   const fcsCode = params.fParams.contentSizeFlag ?
+                     (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) : 0;  /* 0-3 */
+    BYTE  const frameHeaderDecriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
+    size_t pos=0;
+
+    if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
+    DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
+                !params.fParams.noDictIDFlag, dictID,  dictIDSizeCode);
+
+    if (params.format == ZSTD_f_zstd1) {
+        MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
+        pos = 4;
+    }
+    op[pos++] = frameHeaderDecriptionByte;
+    if (!singleSegment) op[pos++] = windowLogByte;
+    switch(dictIDSizeCode)
+    {
+        default:  assert(0); /* impossible */
+        case 0 : break;
+        case 1 : op[pos] = (BYTE)(dictID); pos++; break;
+        case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
+        case 3 : MEM_writeLE32(op+pos, dictID); pos+=4; break;
+    }
+    switch(fcsCode)
+    {
+        default:  assert(0); /* impossible */
+        case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
+        case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
+        case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
+        case 3 : MEM_writeLE64(op+pos, (U64)(pledgedSrcSize)); pos+=8; break;
+    }
+    return pos;
+}
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapcity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
+{
+    if (dstCapacity < ZSTD_blockHeaderSize) return ERROR(dstSize_tooSmall);
+    {   U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1);  /* 0 size */
+        MEM_writeLE24(dst, cBlockHeader24);
+        return ZSTD_blockHeaderSize;
+    }
+}
+
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
+{
+    if (cctx->stage != ZSTDcs_init)
+        return ERROR(stage_wrong);
+    if (cctx->appliedParams.ldmParams.enableLdm)
+        return ERROR(parameter_unsupported);
+    cctx->externSeqStore.seq = seq;
+    cctx->externSeqStore.size = nbSeq;
+    cctx->externSeqStore.capacity = nbSeq;
+    cctx->externSeqStore.pos = 0;
+    return 0;
+}
+
+
+static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                               U32 frame, U32 lastFrameChunk)
+{
+    ZSTD_matchState_t* ms = &cctx->blockState.matchState;
+    size_t fhSize = 0;
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
+                cctx->stage, (U32)srcSize);
+    if (cctx->stage==ZSTDcs_created) return ERROR(stage_wrong);   /* missing init (ZSTD_compressBegin) */
+
+    if (frame && (cctx->stage==ZSTDcs_init)) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams,
+                                       cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
+        if (ZSTD_isError(fhSize)) return fhSize;
+        dstCapacity -= fhSize;
+        dst = (char*)dst + fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (!srcSize) return fhSize;  /* do not generate an empty block if no input */
+
+    if (!ZSTD_window_update(&ms->window, src, srcSize)) {
+        ms->nextToUpdate = ms->window.dictLimit;
+    }
+    if (cctx->appliedParams.ldmParams.enableLdm)
+        ZSTD_window_update(&cctx->ldmState.window, src, srcSize);
+
+    DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (U32)cctx->blockSize);
+    {   size_t const cSize = frame ?
+                             ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
+                             ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize);
+        if (ZSTD_isError(cSize)) return cSize;
+        cctx->consumedSrcSize += srcSize;
+        cctx->producedCSize += (cSize + fhSize);
+        if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
+            if (cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne) {
+                DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize >= %u",
+                    (U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
+                return ERROR(srcSize_wrong);
+            }
+        }
+        return cSize + fhSize;
+    }
+}
+
+size_t ZSTD_compressContinue (ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (U32)srcSize);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
+}
+
+
+size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
+{
+    ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
+    assert(!ZSTD_checkCParams(cParams));
+    return MIN (ZSTD_BLOCKSIZE_MAX, (U32)1 << cParams.windowLog);
+}
+
+size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
+    if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
+}
+
+/*! ZSTD_loadDictionaryContent() :
+ *  @return : 0, or an error code
+ */
+static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* src, size_t srcSize)
+{
+    const BYTE* const ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    ZSTD_compressionParameters const* cParams = &params->cParams;
+
+    ZSTD_window_update(&ms->window, src, srcSize);
+    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
+
+    if (srcSize <= HASH_READ_SIZE) return 0;
+
+    switch(params->cParams.strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, cParams, iend);
+        break;
+    case ZSTD_dfast:
+        ZSTD_fillDoubleHashTable(ms, cParams, iend);
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_insertAndFindFirstIndex(ms, cParams, iend-HASH_READ_SIZE);
+        break;
+
+    case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_updateTree(ms, cParams, iend-HASH_READ_SIZE, iend);
+        break;
+
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    ms->nextToUpdate = (U32)(iend - ms->window.base);
+    return 0;
+}
+
+
+/* Dictionaries that assign zero probability to symbols that show up causes problems
+   when FSE encoding.  Refuse dictionaries that assign zero probability to symbols
+   that we may encounter during compression.
+   NOTE: This behavior is not standard and could be improved in the future. */
+static size_t ZSTD_checkDictNCount(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue) {
+    U32 s;
+    if (dictMaxSymbolValue < maxSymbolValue) return ERROR(dictionary_corrupted);
+    for (s = 0; s <= maxSymbolValue; ++s) {
+        if (normalizedCounter[s] == 0) return ERROR(dictionary_corrupted);
+    }
+    return 0;
+}
+
+
+/* Dictionary format :
+ * See :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
+ */
+/*! ZSTD_loadZstdDictionary() :
+ * @return : dictID, or an error code
+ *  assumptions : magic number supposed already checked
+ *                dictSize supposed > 8
+ */
+static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const void* dict, size_t dictSize, void* workspace)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue = MaxOff;
+    size_t dictID;
+
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+
+    dictPtr += 4;   /* skip magic number */
+    dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr);
+    dictPtr += 4;
+
+    {   unsigned maxSymbolValue = 255;
+        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.hufCTable, &maxSymbolValue, dictPtr, dictEnd-dictPtr);
+        if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
+        if (maxSymbolValue < 255) return ERROR(dictionary_corrupted);
+        dictPtr += hufHeaderSize;
+    }
+
+    {   unsigned offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, workspace, HUF_WORKSPACE_SIZE),
+                 dictionary_corrupted);
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        /* Every match length code must have non-zero probability */
+        CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, workspace, HUF_WORKSPACE_SIZE),
+                 dictionary_corrupted);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        /* Every literal length code must have non-zero probability */
+        CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
+        CHECK_E( FSE_buildCTable_wksp(bs->entropy.litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, workspace, HUF_WORKSPACE_SIZE),
+                 dictionary_corrupted);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    bs->rep[0] = MEM_readLE32(dictPtr+0);
+    bs->rep[1] = MEM_readLE32(dictPtr+4);
+    bs->rep[2] = MEM_readLE32(dictPtr+8);
+    dictPtr += 12;
+
+    {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        U32 offcodeMax = MaxOff;
+        if (dictContentSize <= ((U32)-1) - 128 KB) {
+            U32 const maxOffset = (U32)dictContentSize + 128 KB; /* The maximum offset that must be supported */
+            offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
+        }
+        /* All offset values <= dictContentSize + 128 KB must be representable */
+        CHECK_F (ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)));
+        /* All repCodes must be <= dictContentSize and != 0*/
+        {   U32 u;
+            for (u=0; u<3; u++) {
+                if (bs->rep[u] == 0) return ERROR(dictionary_corrupted);
+                if (bs->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
+        }   }
+
+        bs->entropy.hufCTable_repeatMode = HUF_repeat_valid;
+        bs->entropy.offcode_repeatMode = FSE_repeat_valid;
+        bs->entropy.matchlength_repeatMode = FSE_repeat_valid;
+        bs->entropy.litlength_repeatMode = FSE_repeat_valid;
+        CHECK_F(ZSTD_loadDictionaryContent(ms, params, dictPtr, dictContentSize));
+        return dictID;
+    }
+}
+
+/** ZSTD_compress_insertDictionary() :
+*   @return : dictID, or an error code */
+static size_t ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs, ZSTD_matchState_t* ms,
+                                             ZSTD_CCtx_params const* params,
+                                       const void* dict, size_t dictSize,
+                                             ZSTD_dictContentType_e dictContentType,
+                                             void* workspace)
+{
+    DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
+    if ((dict==NULL) || (dictSize<=8)) return 0;
+
+    ZSTD_reset_compressedBlockState(bs);
+
+    /* dict restricted modes */
+    if (dictContentType == ZSTD_dct_rawContent)
+        return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
+
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
+        if (dictContentType == ZSTD_dct_auto) {
+            DEBUGLOG(4, "raw content dictionary detected");
+            return ZSTD_loadDictionaryContent(ms, params, dict, dictSize);
+        }
+        if (dictContentType == ZSTD_dct_fullDict)
+            return ERROR(dictionary_wrong);
+        assert(0);   /* impossible */
+    }
+
+    /* dict as full zstd dictionary */
+    return ZSTD_loadZstdDictionary(bs, ms, params, dict, dictSize, workspace);
+}
+
+/*! ZSTD_compressBegin_internal() :
+ * @return : 0, or an error code */
+size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
+                             const void* dict, size_t dictSize,
+                             ZSTD_dictContentType_e dictContentType,
+                             const ZSTD_CDict* cdict,
+                             ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                             ZSTD_buffered_policy_e zbuff)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params.cParams.windowLog);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    if (cdict && cdict->dictContentSize>0) {
+        cctx->requestedParams = params;
+        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params.cParams.windowLog,
+                                         params.fParams, pledgedSrcSize, zbuff);
+    }
+
+    CHECK_F( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                     ZSTDcrp_continue, zbuff) );
+    {
+        size_t const dictID = ZSTD_compress_insertDictionary(
+                cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                &params, dict, dictSize, dictContentType, cctx->entropyWorkspace);
+        if (ZSTD_isError(dictID)) return dictID;
+        assert(dictID <= (size_t)(U32)-1);
+        cctx->dictID = (U32)dictID;
+    }
+    return 0;
+}
+
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    const ZSTD_CDict* cdict,
+                                    ZSTD_CCtx_params params,
+                                    unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params.cParams.windowLog);
+    /* compression parameters verification and optimization */
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    return ZSTD_compressBegin_internal(cctx,
+                                       dict, dictSize, dictContentType,
+                                       cdict,
+                                       params, pledgedSrcSize,
+                                       ZSTDb_not_buffered);
+}
+
+/*! ZSTD_compressBegin_advanced() :
+*   @return : 0, or an error code */
+size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    return ZSTD_compressBegin_advanced_internal(cctx,
+                                            dict, dictSize, ZSTD_dct_auto,
+                                            NULL /*cdict*/,
+                                            cctxParams, pledgedSrcSize);
+}
+
+size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (U32)dictSize);
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, NULL,
+                                       cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
+}
+
+size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict(cctx, NULL, 0, compressionLevel);
+}
+
+
+/*! ZSTD_writeEpilogue() :
+*   Ends a frame.
+*   @return : nb of bytes written into dst (or an error code) */
+static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    size_t fhSize = 0;
+
+    DEBUGLOG(4, "ZSTD_writeEpilogue");
+    if (cctx->stage == ZSTDcs_created) return ERROR(stage_wrong);  /* init missing */
+
+    /* special case : empty frame */
+    if (cctx->stage == ZSTDcs_init) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams, 0, 0);
+        if (ZSTD_isError(fhSize)) return fhSize;
+        dstCapacity -= fhSize;
+        op += fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (cctx->stage != ZSTDcs_ending) {
+        /* write one last empty block, make it the "last" block */
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
+        if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        MEM_writeLE32(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+    }
+
+    if (cctx->appliedParams.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", checksum);
+        MEM_writeLE32(op, checksum);
+        op += 4;
+    }
+
+    cctx->stage = ZSTDcs_created;  /* return to "created but no init" status */
+    return op-ostart;
+}
+
+size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
+                         void* dst, size_t dstCapacity,
+                   const void* src, size_t srcSize)
+{
+    size_t endResult;
+    size_t const cSize = ZSTD_compressContinue_internal(cctx,
+                                dst, dstCapacity, src, srcSize,
+                                1 /* frame mode */, 1 /* last chunk */);
+    if (ZSTD_isError(cSize)) return cSize;
+    endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
+    if (ZSTD_isError(endResult)) return endResult;
+    if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
+        DEBUGLOG(4, "end of frame : controlling src size");
+        if (cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1) {
+            DEBUGLOG(4, "error : pledgedSrcSize = %u, while realSrcSize = %u",
+                (U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
+            return ERROR(srcSize_wrong);
+    }   }
+    return cSize + endResult;
+}
+
+
+static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict,size_t dictSize,
+                               ZSTD_parameters params)
+{
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    DEBUGLOG(4, "ZSTD_compress_internal");
+    return ZSTD_compress_advanced_internal(cctx,
+                                          dst, dstCapacity,
+                                          src, srcSize,
+                                          dict, dictSize,
+                                          cctxParams);
+}
+
+size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict,size_t dictSize,
+                               ZSTD_parameters params)
+{
+    DEBUGLOG(4, "ZSTD_compress_advanced");
+    CHECK_F(ZSTD_checkCParams(params.cParams));
+    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+}
+
+/* Internal */
+size_t ZSTD_compress_advanced_internal(
+        ZSTD_CCtx* cctx,
+        void* dst, size_t dstCapacity,
+        const void* src, size_t srcSize,
+        const void* dict,size_t dictSize,
+        ZSTD_CCtx_params params)
+{
+    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)",
+                (U32)srcSize);
+    CHECK_F( ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, NULL,
+                                         params, srcSize, ZSTDb_not_buffered) );
+    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
+}
+
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize,
+                               const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, srcSize ? srcSize : 1, dict ? dictSize : 0);
+    ZSTD_CCtx_params cctxParams = ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+    assert(params.fParams.contentSizeFlag == 1);
+    ZSTD_CCtxParam_setParameter(&cctxParams, ZSTD_p_compressLiterals, compressionLevel>=0);
+    return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, cctxParams);
+}
+
+size_t ZSTD_compressCCtx (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (U32)srcSize);
+    return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
+}
+
+size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+{
+    size_t result;
+    ZSTD_CCtx ctxBody;
+    memset(&ctxBody, 0, sizeof(ctxBody));
+    ctxBody.customMem = ZSTD_defaultCMem;
+    result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_free(ctxBody.workSpace, ZSTD_defaultCMem);  /* can't free ctxBody itself, as it's on stack; free only heap content */
+    return result;
+}
+
+
+/* =====  Dictionary API  ===== */
+
+/*! ZSTD_estimateCDictSize_advanced() :
+ *  Estimate amount of memory that will be needed to create a dictionary with following arguments */
+size_t ZSTD_estimateCDictSize_advanced(
+        size_t dictSize, ZSTD_compressionParameters cParams,
+        ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (U32)sizeof(ZSTD_CDict));
+    return sizeof(ZSTD_CDict) + HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0)
+           + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+}
+
+size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
+}
+
+size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support sizeof on NULL */
+    DEBUGLOG(5, "sizeof(*cdict) : %u", (U32)sizeof(*cdict));
+    return cdict->workspaceSize + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict);
+}
+
+static size_t ZSTD_initCDict_internal(
+                    ZSTD_CDict* cdict,
+              const void* dictBuffer, size_t dictSize,
+                    ZSTD_dictLoadMethod_e dictLoadMethod,
+                    ZSTD_dictContentType_e dictContentType,
+                    ZSTD_compressionParameters cParams)
+{
+    DEBUGLOG(3, "ZSTD_initCDict_internal, dictContentType %u", (U32)dictContentType);
+    assert(!ZSTD_checkCParams(cParams));
+    cdict->cParams = cParams;
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
+        cdict->dictBuffer = NULL;
+        cdict->dictContent = dictBuffer;
+    } else {
+        void* const internalBuffer = ZSTD_malloc(dictSize, cdict->customMem);
+        cdict->dictBuffer = internalBuffer;
+        cdict->dictContent = internalBuffer;
+        if (!internalBuffer) return ERROR(memory_allocation);
+        memcpy(internalBuffer, dictBuffer, dictSize);
+    }
+    cdict->dictContentSize = dictSize;
+
+    /* Reset the state to no dictionary */
+    ZSTD_reset_compressedBlockState(&cdict->cBlockState);
+    {   void* const end = ZSTD_reset_matchState(
+                &cdict->matchState,
+                (U32*)cdict->workspace + HUF_WORKSPACE_SIZE_U32,
+                &cParams, ZSTDcrp_continue, /* forCCtx */ 0);
+        assert(end == (char*)cdict->workspace + cdict->workspaceSize);
+        (void)end;
+    }
+    /* (Maybe) load the dictionary
+     * Skips loading the dictionary if it is <= 8 bytes.
+     */
+    {   ZSTD_CCtx_params params;
+        memset(&params, 0, sizeof(params));
+        params.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        params.fParams.contentSizeFlag = 1;
+        params.cParams = cParams;
+        {   size_t const dictID = ZSTD_compress_insertDictionary(
+                    &cdict->cBlockState, &cdict->matchState, &params,
+                    cdict->dictContent, cdict->dictContentSize,
+                    dictContentType, cdict->workspace);
+            if (ZSTD_isError(dictID)) return dictID;
+            assert(dictID <= (size_t)(U32)-1);
+            cdict->dictID = (U32)dictID;
+        }
+    }
+
+    return 0;
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
+{
+    DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (U32)dictContentType);
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    {   ZSTD_CDict* const cdict = (ZSTD_CDict*)ZSTD_malloc(sizeof(ZSTD_CDict), customMem);
+        size_t const workspaceSize = HUF_WORKSPACE_SIZE + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
+        void* const workspace = ZSTD_malloc(workspaceSize, customMem);
+
+        if (!cdict || !workspace) {
+            ZSTD_free(cdict, customMem);
+            ZSTD_free(workspace, customMem);
+            return NULL;
+        }
+        cdict->customMem = customMem;
+        cdict->workspace = workspace;
+        cdict->workspaceSize = workspaceSize;
+        if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                        dictBuffer, dictSize,
+                                        dictLoadMethod, dictContentType,
+                                        cParams) )) {
+            ZSTD_freeCDict(cdict);
+            return NULL;
+        }
+
+        return cdict;
+    }
+}
+
+ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_createCDict_advanced(dict, dictSize,
+                                     ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                     cParams, ZSTD_defaultCMem);
+}
+
+ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_createCDict_advanced(dict, dictSize,
+                                     ZSTD_dlm_byRef, ZSTD_dct_auto,
+                                     cParams, ZSTD_defaultCMem);
+}
+
+size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = cdict->customMem;
+        ZSTD_free(cdict->workspace, cMem);
+        ZSTD_free(cdict->dictBuffer, cMem);
+        ZSTD_free(cdict, cMem);
+        return 0;
+    }
+}
+
+/*! ZSTD_initStaticCDict_advanced() :
+ *  Generate a digested dictionary in provided memory area.
+ *  workspace: The memory area to emplace the dictionary into.
+ *             Provided pointer must 8-bytes aligned.
+ *             It must outlive dictionary usage.
+ *  workspaceSize: Use ZSTD_estimateCDictSize()
+ *                 to determine how large workspace must be.
+ *  cParams : use ZSTD_getCParams() to transform a compression level
+ *            into its relevants cParams.
+ * @return : pointer to ZSTD_CDict*, or NULL if error (size too small)
+ *  Note : there is no corresponding "free" function.
+ *         Since workspace was allocated externally, it must be freed externally.
+ */
+const ZSTD_CDict* ZSTD_initStaticCDict(
+                                 void* workspace, size_t workspaceSize,
+                           const void* dict, size_t dictSize,
+                                 ZSTD_dictLoadMethod_e dictLoadMethod,
+                                 ZSTD_dictContentType_e dictContentType,
+                                 ZSTD_compressionParameters cParams)
+{
+    size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
+    size_t const neededSize = sizeof(ZSTD_CDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize)
+                            + HUF_WORKSPACE_SIZE + matchStateSize;
+    ZSTD_CDict* const cdict = (ZSTD_CDict*) workspace;
+    void* ptr;
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u",
+        (U32)workspaceSize, (U32)neededSize, (U32)(workspaceSize < neededSize));
+    if (workspaceSize < neededSize) return NULL;
+
+    if (dictLoadMethod == ZSTD_dlm_byCopy) {
+        memcpy(cdict+1, dict, dictSize);
+        dict = cdict+1;
+        ptr = (char*)workspace + sizeof(ZSTD_CDict) + dictSize;
+    } else {
+        ptr = cdict+1;
+    }
+    cdict->workspace = ptr;
+    cdict->workspaceSize = HUF_WORKSPACE_SIZE + matchStateSize;
+
+    if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
+                                              dict, dictSize,
+                                              ZSTD_dlm_byRef, dictContentType,
+                                              cParams) ))
+        return NULL;
+
+    return cdict;
+}
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
+{
+    assert(cdict != NULL);
+    return cdict->cParams;
+}
+
+/* ZSTD_compressBegin_usingCDict_advanced() :
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_advanced(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_advanced");
+    if (cdict==NULL) return ERROR(dictionary_wrong);
+    {   ZSTD_CCtx_params params = cctx->requestedParams;
+        params.cParams = ZSTD_getCParamsFromCDict(cdict);
+        /* Increase window log to fit the entire dictionary and source if the
+         * source size is known. Limit the increase to 19, which is the
+         * window log for compression level 1 with the largest source size.
+         */
+        if (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+            U32 const limitedSrcSize = (U32)MIN(pledgedSrcSize, 1U << 19);
+            U32 const limitedSrcLog = limitedSrcSize > 1 ? ZSTD_highbit32(limitedSrcSize - 1) + 1 : 1;
+            params.cParams.windowLog = MAX(params.cParams.windowLog, limitedSrcLog);
+        }
+        params.fParams = fParams;
+        return ZSTD_compressBegin_internal(cctx,
+                                           NULL, 0, ZSTD_dct_auto,
+                                           cdict,
+                                           params, pledgedSrcSize,
+                                           ZSTDb_not_buffered);
+    }
+}
+
+/* ZSTD_compressBegin_usingCDict() :
+ * pledgedSrcSize=0 means "unknown"
+ * if pledgedSrcSize>0, it will enable contentSizeFlag */
+size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u", !fParams.noDictIDFlag);
+    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, 0);
+}
+
+size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    CHECK_F (ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, srcSize));   /* will check if cdict != NULL */
+    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
+}
+
+/*! ZSTD_compress_usingCDict() :
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression parameters are decided at CDict creation time
+ *  while frame parameters are hardcoded */
+size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
+}
+
+
+
+/* ******************************************************************
+*  Streaming
+********************************************************************/
+
+ZSTD_CStream* ZSTD_createCStream(void)
+{
+    DEBUGLOG(3, "ZSTD_createCStream");
+    return ZSTD_createCStream_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_CStream* ZSTD_initStaticCStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticCCtx(workspace, workspaceSize);
+}
+
+ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem)
+{   /* CStream and CCtx are now same object */
+    return ZSTD_createCCtx_advanced(customMem);
+}
+
+size_t ZSTD_freeCStream(ZSTD_CStream* zcs)
+{
+    return ZSTD_freeCCtx(zcs);   /* same object */
+}
+
+
+
+/*======   Initialization   ======*/
+
+size_t ZSTD_CStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_CStreamOutSize(void)
+{
+    return ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
+}
+
+static size_t ZSTD_resetCStream_internal(ZSTD_CStream* cctx,
+                    const void* const dict, size_t const dictSize, ZSTD_dictContentType_e const dictContentType,
+                    const ZSTD_CDict* const cdict,
+                    ZSTD_CCtx_params const params, unsigned long long const pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_resetCStream_internal (disableLiteralCompression=%i)",
+                params.disableLiteralCompression);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    CHECK_F( ZSTD_compressBegin_internal(cctx,
+                                         dict, dictSize, dictContentType,
+                                         cdict,
+                                         params, pledgedSrcSize,
+                                         ZSTDb_buffered) );
+
+    cctx->inToCompress = 0;
+    cctx->inBuffPos = 0;
+    cctx->inBuffTarget = cctx->blockSize
+                      + (cctx->blockSize == pledgedSrcSize);   /* for small input: avoid automatic flush on reaching end of block, since it would require to add a 3-bytes null block to end frame */
+    cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
+    cctx->streamStage = zcss_load;
+    cctx->frameEnded = 0;
+    return 0;   /* ready to go */
+}
+
+/* ZSTD_resetCStream():
+ * pledgedSrcSize == 0 means "unknown" */
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params params = zcs->requestedParams;
+    DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (U32)pledgedSrcSize);
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    params.fParams.contentSizeFlag = 1;
+    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, 0);
+    return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dct_auto, zcs->cdict, params, pledgedSrcSize);
+}
+
+/*! ZSTD_initCStream_internal() :
+ *  Note : for lib/compress only. Used by zstdmt_compress.c.
+ *  Assumption 1 : params are valid
+ *  Assumption 2 : either dict, or cdict, is defined, not both */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                    const void* dict, size_t dictSize, const ZSTD_CDict* cdict,
+                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_internal");
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+
+    if (dict && dictSize >= 8) {
+        DEBUGLOG(4, "loading dictionary of size %u", (U32)dictSize);
+        if (zcs->staticSize) {   /* static CCtx : never uses malloc */
+            /* incompatible with internal cdict creation */
+            return ERROR(memory_allocation);
+        }
+        ZSTD_freeCDict(zcs->cdictLocal);
+        zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                            ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                            params.cParams, zcs->customMem);
+        zcs->cdict = zcs->cdictLocal;
+        if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        if (cdict) {
+            params.cParams = ZSTD_getCParamsFromCDict(cdict);  /* cParams are enforced from cdict; it includes windowLog */
+        }
+        ZSTD_freeCDict(zcs->cdictLocal);
+        zcs->cdictLocal = NULL;
+        zcs->cdict = cdict;
+    }
+
+    return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dct_auto, zcs->cdict, params, pledgedSrcSize);
+}
+
+/* ZSTD_initCStream_usingCDict_advanced() :
+ * same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
+size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
+                                            const ZSTD_CDict* cdict,
+                                            ZSTD_frameParameters fParams,
+                                            unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict_advanced");
+    if (!cdict) return ERROR(dictionary_wrong); /* cannot handle NULL cdict (does not know what to do) */
+    {   ZSTD_CCtx_params params = zcs->requestedParams;
+        params.cParams = ZSTD_getCParamsFromCDict(cdict);
+        params.fParams = fParams;
+        return ZSTD_initCStream_internal(zcs,
+                                NULL, 0, cdict,
+                                params, pledgedSrcSize);
+    }
+}
+
+/* note : cdict must outlive compression session */
+size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /* contentSizeFlag */, 0 /* checksum */, 0 /* hideDictID */ };
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
+    return ZSTD_initCStream_usingCDict_advanced(zcs, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);  /* note : will check that cdict != NULL */
+}
+
+
+/* ZSTD_initCStream_advanced() :
+ * pledgedSrcSize must be exact.
+ * if srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
+ * dict is loaded with default parameters ZSTD_dm_auto and ZSTD_dlm_byCopy. */
+size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+                                 const void* dict, size_t dictSize,
+                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTD_initCStream_advanced: pledgedSrcSize=%u, flag=%u",
+                (U32)pledgedSrcSize, params.fParams.contentSizeFlag);
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    if ((pledgedSrcSize==0) && (params.fParams.contentSizeFlag==0)) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;  /* for compatibility with older programs relying on this behavior. Users should now specify ZSTD_CONTENTSIZE_UNKNOWN. This line will be removed in the future. */
+    {   ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+        return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL /*cdict*/, cctxParams, pledgedSrcSize);
+    }
+}
+
+size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
+    ZSTD_CCtx_params const cctxParams =
+            ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
+}
+
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
+{
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;  /* temporary : 0 interpreted as "unknown" during transition period. Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN. `0` will be interpreted as "empty" in the future */
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
+    ZSTD_CCtx_params const cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
+    return ZSTD_initCStream_internal(zcs, NULL, 0, NULL, cctxParams, pledgedSrcSize);
+}
+
+size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
+{
+    DEBUGLOG(4, "ZSTD_initCStream");
+    return ZSTD_initCStream_srcSize(zcs, compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN);
+}
+
+/*======   Compression   ======*/
+
+MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity,
+                           const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    if (length) memcpy(dst, src, length);
+    return length;
+}
+
+/** ZSTD_compressStream_generic():
+ *  internal function for all *compressStream*() variants and *compress_generic()
+ *  non-static, because can be called from zstdmt_compress.c
+ * @return : hint size for next input */
+size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                   ZSTD_outBuffer* output,
+                                   ZSTD_inBuffer* input,
+                                   ZSTD_EndDirective const flushMode)
+{
+    const char* const istart = (const char*)input->src;
+    const char* const iend = istart + input->size;
+    const char* ip = istart + input->pos;
+    char* const ostart = (char*)output->dst;
+    char* const oend = ostart + output->size;
+    char* op = ostart + output->pos;
+    U32 someMoreWork = 1;
+
+    /* check expectations */
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (U32)flushMode);
+    assert(zcs->inBuff != NULL);
+    assert(zcs->inBuffSize > 0);
+    assert(zcs->outBuff !=  NULL);
+    assert(zcs->outBuffSize > 0);
+    assert(output->pos <= output->size);
+    assert(input->pos <= input->size);
+
+    while (someMoreWork) {
+        switch(zcs->streamStage)
+        {
+        case zcss_init:
+            /* call ZSTD_initCStream() first ! */
+            return ERROR(init_missing);
+
+        case zcss_load:
+            if ( (flushMode == ZSTD_e_end)
+              && ((size_t)(oend-op) >= ZSTD_compressBound(iend-ip))  /* enough dstCapacity */
+              && (zcs->inBuffPos == 0) ) {
+                /* shortcut to compression pass directly into output buffer */
+                size_t const cSize = ZSTD_compressEnd(zcs,
+                                                op, oend-op, ip, iend-ip);
+                DEBUGLOG(4, "ZSTD_compressEnd : %u", (U32)cSize);
+                if (ZSTD_isError(cSize)) return cSize;
+                ip = iend;
+                op += cSize;
+                zcs->frameEnded = 1;
+                ZSTD_startNewCompression(zcs);
+                someMoreWork = 0; break;
+            }
+            /* complete loading into inBuffer */
+            {   size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
+                size_t const loaded = ZSTD_limitCopy(
+                                        zcs->inBuff + zcs->inBuffPos, toLoad,
+                                        ip, iend-ip);
+                zcs->inBuffPos += loaded;
+                ip += loaded;
+                if ( (flushMode == ZSTD_e_continue)
+                  && (zcs->inBuffPos < zcs->inBuffTarget) ) {
+                    /* not enough input to fill full block : stop here */
+                    someMoreWork = 0; break;
+                }
+                if ( (flushMode == ZSTD_e_flush)
+                  && (zcs->inBuffPos == zcs->inToCompress) ) {
+                    /* empty */
+                    someMoreWork = 0; break;
+                }
+            }
+            /* compress current block (note : this stage cannot be stopped in the middle) */
+            DEBUGLOG(5, "stream compression stage (flushMode==%u)", flushMode);
+            {   void* cDst;
+                size_t cSize;
+                size_t const iSize = zcs->inBuffPos - zcs->inToCompress;
+                size_t oSize = oend-op;
+                unsigned const lastBlock = (flushMode == ZSTD_e_end) && (ip==iend);
+                if (oSize >= ZSTD_compressBound(iSize))
+                    cDst = op;   /* compress into output buffer, to skip flush stage */
+                else
+                    cDst = zcs->outBuff, oSize = zcs->outBuffSize;
+                cSize = lastBlock ?
+                        ZSTD_compressEnd(zcs, cDst, oSize,
+                                    zcs->inBuff + zcs->inToCompress, iSize) :
+                        ZSTD_compressContinue(zcs, cDst, oSize,
+                                    zcs->inBuff + zcs->inToCompress, iSize);
+                if (ZSTD_isError(cSize)) return cSize;
+                zcs->frameEnded = lastBlock;
+                /* prepare next block */
+                zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
+                if (zcs->inBuffTarget > zcs->inBuffSize)
+                    zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
+                DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
+                         (U32)zcs->inBuffTarget, (U32)zcs->inBuffSize);
+                if (!lastBlock)
+                    assert(zcs->inBuffTarget <= zcs->inBuffSize);
+                zcs->inToCompress = zcs->inBuffPos;
+                if (cDst == op) {  /* no need to flush */
+                    op += cSize;
+                    if (zcs->frameEnded) {
+                        DEBUGLOG(5, "Frame completed directly in outBuffer");
+                        someMoreWork = 0;
+                        ZSTD_startNewCompression(zcs);
+                    }
+                    break;
+                }
+                zcs->outBuffContentSize = cSize;
+                zcs->outBuffFlushedSize = 0;
+                zcs->streamStage = zcss_flush; /* pass-through to flush stage */
+            }
+	    /* fall-through */
+        case zcss_flush:
+            DEBUGLOG(5, "flush stage");
+            {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
+                size_t const flushed = ZSTD_limitCopy(op, oend-op,
+                            zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
+                            (U32)toFlush, (U32)(oend-op), (U32)flushed);
+                op += flushed;
+                zcs->outBuffFlushedSize += flushed;
+                if (toFlush!=flushed) {
+                    /* flush not fully completed, presumably because dst is too small */
+                    assert(op==oend);
+                    someMoreWork = 0;
+                    break;
+                }
+                zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
+                if (zcs->frameEnded) {
+                    DEBUGLOG(5, "Frame completed on flush");
+                    someMoreWork = 0;
+                    ZSTD_startNewCompression(zcs);
+                    break;
+                }
+                zcs->streamStage = zcss_load;
+                break;
+            }
+
+        default: /* impossible */
+            assert(0);
+        }
+    }
+
+    input->pos = ip - istart;
+    output->pos = op - ostart;
+    if (zcs->frameEnded) return 0;
+    {   size_t hintInSize = zcs->inBuffTarget - zcs->inBuffPos;
+        if (hintInSize==0) hintInSize = zcs->blockSize;
+        return hintInSize;
+    }
+}
+
+size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    /* check conditions */
+    if (output->pos > output->size) return ERROR(GENERIC);
+    if (input->pos  > input->size)  return ERROR(GENERIC);
+
+    return ZSTD_compressStream_generic(zcs, output, input, ZSTD_e_continue);
+}
+
+
+size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
+                              ZSTD_outBuffer* output,
+                              ZSTD_inBuffer* input,
+                              ZSTD_EndDirective endOp)
+{
+    DEBUGLOG(5, "ZSTD_compress_generic, endOp=%u ", (U32)endOp);
+    /* check conditions */
+    if (output->pos > output->size) return ERROR(GENERIC);
+    if (input->pos  > input->size)  return ERROR(GENERIC);
+    assert(cctx!=NULL);
+
+    /* transparent initialization stage */
+    if (cctx->streamStage == zcss_init) {
+        ZSTD_CCtx_params params = cctx->requestedParams;
+        ZSTD_prefixDict const prefixDict = cctx->prefixDict;
+        memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));  /* single usage */
+        assert(prefixDict.dict==NULL || cctx->cdict==NULL);   /* only one can be set */
+        DEBUGLOG(4, "ZSTD_compress_generic : transparent init stage");
+        if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = input->size + 1;  /* auto-fix pledgedSrcSize */
+        params.cParams = ZSTD_getCParamsFromCCtxParams(
+                &cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, 0 /*dictSize*/);
+
+#ifdef ZSTD_MULTITHREAD
+        if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
+            params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
+        }
+        if (params.nbWorkers > 0) {
+            /* mt context creation */
+            if (cctx->mtctx == NULL || (params.nbWorkers != ZSTDMT_getNbWorkers(cctx->mtctx))) {
+                DEBUGLOG(4, "ZSTD_compress_generic: creating new mtctx for nbWorkers=%u",
+                            params.nbWorkers);
+                if (cctx->mtctx != NULL)
+                    DEBUGLOG(4, "ZSTD_compress_generic: previous nbWorkers was %u",
+                                ZSTDMT_getNbWorkers(cctx->mtctx));
+                ZSTDMT_freeCCtx(cctx->mtctx);
+                cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbWorkers, cctx->customMem);
+                if (cctx->mtctx == NULL) return ERROR(memory_allocation);
+            }
+            /* mt compression */
+            DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
+            CHECK_F( ZSTDMT_initCStream_internal(
+                        cctx->mtctx,
+                        prefixDict.dict, prefixDict.dictSize, ZSTD_dct_rawContent,
+                        cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) );
+            cctx->streamStage = zcss_load;
+            cctx->appliedParams.nbWorkers = params.nbWorkers;
+        } else
+#endif
+        {   CHECK_F( ZSTD_resetCStream_internal(cctx,
+                            prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                            cctx->cdict,
+                            params, cctx->pledgedSrcSizePlusOne-1) );
+            assert(cctx->streamStage == zcss_load);
+            assert(cctx->appliedParams.nbWorkers == 0);
+    }   }
+
+    /* compression stage */
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        if (cctx->cParamsChanged) {
+            ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
+            cctx->cParamsChanged = 0;
+        }
+        {   size_t const flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
+            if ( ZSTD_isError(flushMin)
+              || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
+                ZSTD_startNewCompression(cctx);
+            }
+            return flushMin;
+    }   }
+#endif
+    CHECK_F( ZSTD_compressStream_generic(cctx, output, input, endOp) );
+    DEBUGLOG(5, "completed ZSTD_compress_generic");
+    return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
+}
+
+size_t ZSTD_compress_generic_simpleArgs (
+                            ZSTD_CCtx* cctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos,
+                            ZSTD_EndDirective endOp)
+{
+    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
+    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
+    size_t const cErr = ZSTD_compress_generic(cctx, &output, &input, endOp);
+    *dstPos = output.pos;
+    *srcPos = input.pos;
+    return cErr;
+}
+
+
+/*======   Finalize   ======*/
+
+/*! ZSTD_flushStream() :
+ * @return : amount of data remaining to flush */
+size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    ZSTD_inBuffer input = { NULL, 0, 0 };
+    if (output->pos > output->size) return ERROR(GENERIC);
+    CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_flush) );
+    return zcs->outBuffContentSize - zcs->outBuffFlushedSize;  /* remaining to flush */
+}
+
+
+size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    ZSTD_inBuffer input = { NULL, 0, 0 };
+    if (output->pos > output->size) return ERROR(GENERIC);
+    CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_end) );
+    {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
+        size_t const checksumSize = zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4;
+        size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize + lastBlockSize + checksumSize;
+        DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (U32)toFlush);
+        return toFlush;
+    }
+}
+
+
+/*-=====  Pre-defined compression levels  =====-*/
+
+#define ZSTD_MAX_CLEVEL     22
+int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
+
+static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
+{   /* "default" - guarantees a monotonically increasing memory budget */
+    /* W,  C,  H,  S,  L, TL, strat */
+    { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
+    { 19, 13, 14,  1,  7,  1, ZSTD_fast    },  /* level  1 */
+    { 19, 15, 16,  1,  6,  1, ZSTD_fast    },  /* level  2 */
+    { 20, 16, 17,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
+    { 20, 17, 18,  1,  5,  8, ZSTD_dfast   },  /* level  4 */
+    { 20, 17, 18,  2,  5, 16, ZSTD_greedy  },  /* level  5 */
+    { 21, 17, 19,  2,  5, 16, ZSTD_lazy    },  /* level  6 */
+    { 21, 18, 19,  3,  5, 16, ZSTD_lazy    },  /* level  7 */
+    { 21, 18, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
+    { 21, 19, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 21, 19, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
+    { 22, 20, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
+    { 22, 20, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
+    { 22, 21, 22,  4,  5, 32, ZSTD_btlazy2 },  /* level 13 */
+    { 22, 21, 22,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
+    { 22, 22, 22,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
+    { 22, 21, 22,  4,  5, 48, ZSTD_btopt   },  /* level 16 */
+    { 23, 22, 22,  4,  4, 48, ZSTD_btopt   },  /* level 17 */
+    { 23, 22, 22,  5,  3, 64, ZSTD_btopt   },  /* level 18 */
+    { 23, 23, 22,  7,  3,128, ZSTD_btopt   },  /* level 19 */
+    { 25, 25, 23,  7,  3,128, ZSTD_btultra },  /* level 20 */
+    { 26, 26, 24,  7,  3,256, ZSTD_btultra },  /* level 21 */
+    { 27, 27, 25,  9,  3,512, ZSTD_btultra },  /* level 22 */
+},
+{   /* for srcSize <= 256 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 18, 13, 14,  1,  6,  1, ZSTD_fast    },  /* level  1 */
+    { 18, 14, 13,  1,  5,  8, ZSTD_dfast   },  /* level  2 */
+    { 18, 16, 15,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
+    { 18, 15, 17,  1,  5,  8, ZSTD_greedy  },  /* level  4.*/
+    { 18, 16, 17,  4,  5,  8, ZSTD_greedy  },  /* level  5.*/
+    { 18, 16, 17,  3,  5,  8, ZSTD_lazy    },  /* level  6.*/
+    { 18, 17, 17,  4,  4,  8, ZSTD_lazy    },  /* level  7 */
+    { 18, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 18, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 18, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 18, 18, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 11.*/
+    { 18, 18, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 12.*/
+    { 18, 19, 17,  7,  4,  8, ZSTD_btlazy2 },  /* level 13 */
+    { 18, 18, 18,  4,  4, 16, ZSTD_btopt   },  /* level 14.*/
+    { 18, 18, 18,  4,  3, 16, ZSTD_btopt   },  /* level 15.*/
+    { 18, 19, 18,  6,  3, 32, ZSTD_btopt   },  /* level 16.*/
+    { 18, 19, 18,  8,  3, 64, ZSTD_btopt   },  /* level 17.*/
+    { 18, 19, 18,  9,  3,128, ZSTD_btopt   },  /* level 18.*/
+    { 18, 19, 18, 10,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 18, 19, 18, 11,  3,512, ZSTD_btultra },  /* level 20.*/
+    { 18, 19, 18, 12,  3,512, ZSTD_btultra },  /* level 21.*/
+    { 18, 19, 18, 13,  3,512, ZSTD_btultra },  /* level 22.*/
+},
+{   /* for srcSize <= 128 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* level  0 - not used */
+    { 17, 12, 13,  1,  6,  1, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 16,  1,  5,  1, ZSTD_fast    },  /* level  2 */
+    { 17, 16, 16,  2,  5,  8, ZSTD_dfast   },  /* level  3 */
+    { 17, 13, 15,  3,  4,  8, ZSTD_greedy  },  /* level  4 */
+    { 17, 15, 17,  4,  4,  8, ZSTD_greedy  },  /* level  5 */
+    { 17, 16, 17,  3,  4,  8, ZSTD_lazy    },  /* level  6 */
+    { 17, 15, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 17, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 17, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 17, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 17, 17, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 11 */
+    { 17, 17, 17,  8,  4,  8, ZSTD_lazy2   },  /* level 12 */
+    { 17, 18, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13.*/
+    { 17, 17, 17,  7,  3,  8, ZSTD_btopt   },  /* level 14.*/
+    { 17, 17, 17,  7,  3, 16, ZSTD_btopt   },  /* level 15.*/
+    { 17, 18, 17,  7,  3, 32, ZSTD_btopt   },  /* level 16.*/
+    { 17, 18, 17,  7,  3, 64, ZSTD_btopt   },  /* level 17.*/
+    { 17, 18, 17,  7,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 17, 18, 17,  9,  3,256, ZSTD_btultra },  /* level 20.*/
+    { 17, 18, 17, 10,  3,256, ZSTD_btultra },  /* level 21.*/
+    { 17, 18, 17, 11,  3,512, ZSTD_btultra },  /* level 22.*/
+},
+{   /* for srcSize <= 16 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 14, 14, 14,  1,  6,  1, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 14,  1,  4,  1, ZSTD_fast    },  /* level  2 */
+    { 14, 14, 14,  1,  4,  6, ZSTD_dfast   },  /* level  3.*/
+    { 14, 14, 14,  4,  4,  6, ZSTD_greedy  },  /* level  4.*/
+    { 14, 14, 14,  3,  4,  6, ZSTD_lazy    },  /* level  5.*/
+    { 14, 14, 14,  4,  4,  6, ZSTD_lazy2   },  /* level  6 */
+    { 14, 14, 14,  5,  4,  6, ZSTD_lazy2   },  /* level  7 */
+    { 14, 14, 14,  6,  4,  6, ZSTD_lazy2   },  /* level  8.*/
+    { 14, 15, 14,  6,  4,  6, ZSTD_btlazy2 },  /* level  9.*/
+    { 14, 15, 14,  3,  3,  6, ZSTD_btopt   },  /* level 10.*/
+    { 14, 15, 14,  6,  3,  8, ZSTD_btopt   },  /* level 11.*/
+    { 14, 15, 14,  6,  3, 16, ZSTD_btopt   },  /* level 12.*/
+    { 14, 15, 14,  6,  3, 24, ZSTD_btopt   },  /* level 13.*/
+    { 14, 15, 15,  6,  3, 48, ZSTD_btopt   },  /* level 14.*/
+    { 14, 15, 15,  6,  3, 64, ZSTD_btopt   },  /* level 15.*/
+    { 14, 15, 15,  6,  3, 96, ZSTD_btopt   },  /* level 16.*/
+    { 14, 15, 15,  6,  3,128, ZSTD_btopt   },  /* level 17.*/
+    { 14, 15, 15,  6,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btultra },  /* level 20.*/
+    { 14, 15, 15,  9,  3,256, ZSTD_btultra },  /* level 21.*/
+    { 14, 15, 15, 10,  3,256, ZSTD_btultra },  /* level 22.*/
+},
+};
+
+/*! ZSTD_getCParams() :
+*  @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+*   Size values are optional, provide 0 if not known or unused */
+ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
+{
+    size_t const addedSize = srcSizeHint ? 0 : 500;
+    U64 const rSize = srcSizeHint+dictSize ? srcSizeHint+dictSize+addedSize : (U64)-1;
+    U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);   /* intentional underflow for srcSizeHint == 0 */
+    int row = compressionLevel;
+    DEBUGLOG(5, "ZSTD_getCParams (cLevel=%i)", compressionLevel);
+    if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT;   /* 0 == default */
+    if (compressionLevel < 0) row = 0;   /* entry 0 is baseline for fast mode */
+    if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
+    {   ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
+        if (compressionLevel < 0) cp.targetLength = (unsigned)(-compressionLevel);   /* acceleration factor */
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize); }
+
+}
+
+/*! ZSTD_getParams() :
+*   same as ZSTD_getCParams(), but @return a `ZSTD_parameters` object (instead of `ZSTD_compressionParameters`).
+*   All fields of `ZSTD_frameParameters` are set to default (0) */
+ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
+    ZSTD_parameters params;
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, srcSizeHint, dictSize);
+    DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
+    memset(&params, 0, sizeof(params));
+    params.cParams = cParams;
+    params.fParams.contentSizeFlag = 1;
+    return params;
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_compress_internal.h b/vendor/github.com/DataDog/zstd/zstd_compress_internal.h
new file mode 100644
index 0000000..81f12ca
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_compress_internal.h
@@ -0,0 +1,709 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This header contains definitions
+ * that shall **only** be used by modules within lib/compress.
+ */
+
+#ifndef ZSTD_COMPRESS_H
+#define ZSTD_COMPRESS_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "zstd_internal.h"
+#ifdef ZSTD_MULTITHREAD
+#  include "zstdmt_compress.h"
+#endif
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Constants
+***************************************/
+#define kSearchStrength      8
+#define HASH_READ_SIZE       8
+#define ZSTD_DUBT_UNSORTED_MARK 1   /* For btlazy2 strategy, index 1 now means "unsorted".
+                                       It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
+                                       It's not a big deal though : candidate will just be sorted again.
+                                       Additionnally, candidate position 1 will be lost.
+                                       But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be misdhandled after table re-use with a different strategy */
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
+typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
+
+typedef struct ZSTD_prefixDict_s {
+    const void* dict;
+    size_t dictSize;
+    ZSTD_dictContentType_e dictContentType;
+} ZSTD_prefixDict;
+
+typedef struct {
+    U32 hufCTable[HUF_CTABLE_SIZE_U32(255)];
+    FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
+    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
+    FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
+    HUF_repeat hufCTable_repeatMode;
+    FSE_repeat offcode_repeatMode;
+    FSE_repeat matchlength_repeatMode;
+    FSE_repeat litlength_repeatMode;
+} ZSTD_entropyCTables_t;
+
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTD_match_t;
+
+typedef struct {
+    int price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_optimal_t;
+
+typedef struct {
+    /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
+    U32* litFreq;               /* table of literals statistics, of size 256 */
+    U32* litLengthFreq;         /* table of litLength statistics, of size (MaxLL+1) */
+    U32* matchLengthFreq;       /* table of matchLength statistics, of size (MaxML+1) */
+    U32* offCodeFreq;           /* table of offCode statistics, of size (MaxOff+1) */
+    ZSTD_match_t* matchTable;   /* list of found matches, of size ZSTD_OPT_NUM+1 */
+    ZSTD_optimal_t* priceTable; /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
+
+    U32  litSum;                 /* nb of literals */
+    U32  litLengthSum;           /* nb of litLength codes */
+    U32  matchLengthSum;         /* nb of matchLength codes */
+    U32  offCodeSum;             /* nb of offset codes */
+    /* begin updated by ZSTD_setLog2Prices */
+    U32  log2litSum;             /* pow2 to compare log2(litfreq) to */
+    U32  log2litLengthSum;       /* pow2 to compare log2(llfreq) to */
+    U32  log2matchLengthSum;     /* pow2 to compare log2(mlfreq) to */
+    U32  log2offCodeSum;         /* pow2 to compare log2(offreq) to */
+    /* end : updated by ZSTD_setLog2Prices */
+    U32  staticPrices;           /* prices follow a pre-defined cost structure, statistics are irrelevant */
+} optState_t;
+
+typedef struct {
+  ZSTD_entropyCTables_t entropy;
+  U32 rep[ZSTD_REP_NUM];
+} ZSTD_compressedBlockState_t;
+
+typedef struct {
+    BYTE const* nextSrc;    /* next block here to continue on current prefix */
+    BYTE const* base;       /* All regular indexes relative to this position */
+    BYTE const* dictBase;   /* extDict indexes relative to this position */
+    U32 dictLimit;          /* below that point, need extDict */
+    U32 lowLimit;           /* below that point, no more data */
+} ZSTD_window_t;
+
+typedef struct {
+    ZSTD_window_t window;      /* State for window round buffer management */
+    U32 loadedDictEnd;         /* index of end of dictionary */
+    U32 nextToUpdate;          /* index from which to continue table update */
+    U32 nextToUpdate3;         /* index from which to continue table update */
+    U32 hashLog3;              /* dispatch table : larger == faster, more memory */
+    U32* hashTable;
+    U32* hashTable3;
+    U32* chainTable;
+    optState_t opt;         /* optimal parser state */
+} ZSTD_matchState_t;
+
+typedef struct {
+    ZSTD_compressedBlockState_t* prevCBlock;
+    ZSTD_compressedBlockState_t* nextCBlock;
+    ZSTD_matchState_t matchState;
+} ZSTD_blockState_t;
+
+typedef struct {
+    U32 offset;
+    U32 checksum;
+} ldmEntry_t;
+
+typedef struct {
+    ZSTD_window_t window;   /* State for the window round buffer management */
+    ldmEntry_t* hashTable;
+    BYTE* bucketOffsets;    /* Next position in bucket to insert entry */
+    U64 hashPower;          /* Used to compute the rolling hash.
+                             * Depends on ldmParams.minMatchLength */
+} ldmState_t;
+
+typedef struct {
+    U32 enableLdm;          /* 1 if enable long distance matching */
+    U32 hashLog;            /* Log size of hashTable */
+    U32 bucketSizeLog;      /* Log bucket size for collision resolution, at most 8 */
+    U32 minMatchLength;     /* Minimum match length */
+    U32 hashEveryLog;       /* Log number of entries to skip */
+    U32 windowLog;          /* Window log for the LDM */
+} ldmParams_t;
+
+typedef struct {
+    U32 offset;
+    U32 litLength;
+    U32 matchLength;
+} rawSeq;
+
+typedef struct {
+  rawSeq* seq;     /* The start of the sequences */
+  size_t pos;      /* The position where reading stopped. <= size. */
+  size_t size;     /* The number of sequences. <= capacity. */
+  size_t capacity; /* The capacity of the `seq` pointer */
+} rawSeqStore_t;
+
+struct ZSTD_CCtx_params_s {
+    ZSTD_format_e format;
+    ZSTD_compressionParameters cParams;
+    ZSTD_frameParameters fParams;
+
+    int compressionLevel;
+    int disableLiteralCompression;
+    int forceWindow;           /* force back-references to respect limit of
+                                * 1<<wLog, even for dictionary */
+
+    /* Multithreading: used to pass parameters to mtctx */
+    unsigned nbWorkers;
+    unsigned jobSize;
+    unsigned overlapSizeLog;
+
+    /* Long distance matching parameters */
+    ldmParams_t ldmParams;
+
+    /* Internal use, for createCCtxParams() and freeCCtxParams() only */
+    ZSTD_customMem customMem;
+};  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
+
+struct ZSTD_CCtx_s {
+    ZSTD_compressionStage_e stage;
+    int cParamsChanged;                  /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
+    int bmi2;                            /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+    ZSTD_CCtx_params requestedParams;
+    ZSTD_CCtx_params appliedParams;
+    U32   dictID;
+    void* workSpace;
+    size_t workSpaceSize;
+    size_t blockSize;
+    unsigned long long pledgedSrcSizePlusOne;  /* this way, 0 (default) == unknown */
+    unsigned long long consumedSrcSize;
+    unsigned long long producedCSize;
+    XXH64_state_t xxhState;
+    ZSTD_customMem customMem;
+    size_t staticSize;
+
+    seqStore_t seqStore;      /* sequences storage ptrs */
+    ldmState_t ldmState;      /* long distance matching state */
+    rawSeq* ldmSequences;     /* Storage for the ldm output sequences */
+    size_t maxNbLdmSequences;
+    rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
+    ZSTD_blockState_t blockState;
+    U32* entropyWorkspace;  /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+
+    /* streaming */
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inToCompress;
+    size_t inBuffPos;
+    size_t inBuffTarget;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outBuffContentSize;
+    size_t outBuffFlushedSize;
+    ZSTD_cStreamStage streamStage;
+    U32    frameEnded;
+
+    /* Dictionary */
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+    ZSTD_prefixDict prefixDict;   /* single-usage dictionary */
+
+    /* Multi-threading */
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_CCtx* mtctx;
+#endif
+};
+
+
+typedef size_t (*ZSTD_blockCompressor) (
+        ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict);
+
+
+MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
+{
+    static const BYTE LL_Code[64] = {  0,  1,  2,  3,  4,  5,  6,  7,
+                                       8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 16, 17, 17, 18, 18, 19, 19,
+                                      20, 20, 20, 20, 21, 21, 21, 21,
+                                      22, 22, 22, 22, 22, 22, 22, 22,
+                                      23, 23, 23, 23, 23, 23, 23, 23,
+                                      24, 24, 24, 24, 24, 24, 24, 24,
+                                      24, 24, 24, 24, 24, 24, 24, 24 };
+    static const U32 LL_deltaCode = 19;
+    return (litLength > 63) ? ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+}
+
+/* ZSTD_MLcode() :
+ * note : mlBase = matchLength - MINMATCH;
+ *        because it's the format it's stored in seqStore->sequences */
+MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
+{
+    static const BYTE ML_Code[128] = { 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
+                                      32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
+                                      38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
+                                      40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
+                                      41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
+    static const U32 ML_deltaCode = 36;
+    return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
+}
+
+/*! ZSTD_storeSeq() :
+ *  Store a sequence (literal length, literals, offset code and match length code) into seqStore_t.
+ *  `offsetCode` : distance to match + 3 (values 1-3 are repCodes).
+ *  `mlBase` : matchLength - MINMATCH
+*/
+MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const void* literals, U32 offsetCode, size_t mlBase)
+{
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG >= 6)
+    static const BYTE* g_start = NULL;
+    if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
+    {   U32 const pos = (U32)((const BYTE*)literals - g_start);
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%3u bytes at dist.code%7u",
+               pos, (U32)litLength, (U32)mlBase+MINMATCH, (U32)offsetCode);
+    }
+#endif
+    /* copy Literals */
+    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + 128 KB);
+    ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
+    seqStorePtr->lit += litLength;
+
+    /* literal Length */
+    if (litLength>0xFFFF) {
+        assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
+        seqStorePtr->longLengthID = 1;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].litLength = (U16)litLength;
+
+    /* match offset */
+    seqStorePtr->sequences[0].offset = offsetCode + 1;
+
+    /* match Length */
+    if (mlBase>0xFFFF) {
+        assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
+        seqStorePtr->longLengthID = 2;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].matchLength = (U16)mlBase;
+
+    seqStorePtr->sequences++;
+}
+
+
+/*-*************************************
+*  Match length counter
+***************************************/
+static unsigned ZSTD_NbCommonBytes (size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanForward64( &r, (U64)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 4)
+            return (__builtin_ctzll((U64)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
+                                                     0, 3, 1, 3, 1, 4, 2, 7,
+                                                     0, 2, 3, 6, 1, 5, 3, 5,
+                                                     1, 3, 4, 4, 2, 5, 6, 7,
+                                                     7, 0, 1, 2, 3, 3, 4, 6,
+                                                     2, 6, 5, 5, 3, 4, 5, 6,
+                                                     7, 1, 2, 4, 6, 4, 4, 5,
+                                                     7, 2, 6, 5, 7, 6, 7, 7 };
+            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r=0;
+            _BitScanForward( &r, (U32)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctz((U32)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
+                                                     3, 2, 2, 1, 3, 2, 0, 1,
+                                                     3, 3, 1, 2, 2, 2, 2, 0,
+                                                     3, 1, 2, 0, 1, 0, 1, 1 };
+            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#       endif
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanReverse64( &r, val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 4)
+            return (__builtin_clzll(val) >> 3);
+#       else
+            unsigned r;
+            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
+            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r = 0;
+            _BitScanReverse( &r, (unsigned long)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clz((U32)val) >> 3);
+#       else
+            unsigned r;
+            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+    }   }
+}
+
+
+MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
+{
+    const BYTE* const pStart = pIn;
+    const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
+
+    if (pIn < pInLoopLimit) {
+        { size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+          if (diff) return ZSTD_NbCommonBytes(diff); }
+        pIn+=sizeof(size_t); pMatch+=sizeof(size_t);
+        while (pIn < pInLoopLimit) {
+            size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+            if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
+            pIn += ZSTD_NbCommonBytes(diff);
+            return (size_t)(pIn - pStart);
+    }   }
+    if (MEM_64bits() && (pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (size_t)(pIn - pStart);
+}
+
+/** ZSTD_count_2segments() :
+ *  can count match length with `ip` & `match` in 2 different segments.
+ *  convention : on reaching mEnd, match count continue starting from iStart
+ */
+MEM_STATIC size_t
+ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
+                     const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
+{
+    const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
+    size_t const matchLength = ZSTD_count(ip, match, vEnd);
+    if (match + matchLength != mEnd) return matchLength;
+    return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
+}
+
+
+/*-*************************************
+ *  Hashes
+ ***************************************/
+static const U32 prime3bytes = 506832829U;
+static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
+MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); } /* only in zstd_opt.h */
+
+static const U32 prime4bytes = 2654435761U;
+static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
+
+static const U64 prime5bytes = 889523592379ULL;
+static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
+static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
+
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+
+static const U64 prime7bytes = 58295818150454627ULL;
+static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
+static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+
+MEM_STATIC size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
+{
+    switch(mls)
+    {
+    default:
+    case 4: return ZSTD_hash4Ptr(p, hBits);
+    case 5: return ZSTD_hash5Ptr(p, hBits);
+    case 6: return ZSTD_hash6Ptr(p, hBits);
+    case 7: return ZSTD_hash7Ptr(p, hBits);
+    case 8: return ZSTD_hash8Ptr(p, hBits);
+    }
+}
+
+/*-*************************************
+*  Round buffer management
+***************************************/
+/* Max current allowed */
+#define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
+/* Maximum chunk size before overflow correction needs to be called again */
+#define ZSTD_CHUNKSIZE_MAX                                                     \
+    ( ((U32)-1)                  /* Maximum ending current index */            \
+    - ZSTD_CURRENT_MAX)          /* Maximum beginning lowLimit */
+
+/**
+ * ZSTD_window_clear():
+ * Clears the window containing the history by simply setting it to empty.
+ */
+MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
+{
+    size_t const endT = (size_t)(window->nextSrc - window->base);
+    U32 const end = (U32)endT;
+
+    window->lowLimit = end;
+    window->dictLimit = end;
+}
+
+/**
+ * ZSTD_window_hasExtDict():
+ * Returns non-zero if the window has a non-empty extDict.
+ */
+MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
+{
+    return window.lowLimit < window.dictLimit;
+}
+
+/**
+ * ZSTD_window_needOverflowCorrection():
+ * Returns non-zero if the indices are getting too large and need overflow
+ * protection.
+ */
+MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
+                                                  void const* srcEnd)
+{
+    U32 const current = (U32)((BYTE const*)srcEnd - window.base);
+    return current > ZSTD_CURRENT_MAX;
+}
+
+/**
+ * ZSTD_window_correctOverflow():
+ * Reduces the indices to protect from index overflow.
+ * Returns the correction made to the indices, which must be applied to every
+ * stored index.
+ *
+ * The least significant cycleLog bits of the indices must remain the same,
+ * which may be 0. Every index up to maxDist in the past must be valid.
+ * NOTE: (maxDist & cycleMask) must be zero.
+ */
+MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
+                                           U32 maxDist, void const* src)
+{
+    /* preemptive overflow correction:
+     * 1. correction is large enough:
+     *    lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog
+     *    1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
+     *
+     *    current - newCurrent
+     *    > (3<<29 + 1<<windowLog) - (1<<windowLog + 1<<chainLog)
+     *    > (3<<29) - (1<<chainLog)
+     *    > (3<<29) - (1<<30)             (NOTE: chainLog <= 30)
+     *    > 1<<29
+     *
+     * 2. (ip+ZSTD_CHUNKSIZE_MAX - cctx->base) doesn't overflow:
+     *    After correction, current is less than (1<<chainLog + 1<<windowLog).
+     *    In 64-bit mode we are safe, because we have 64-bit ptrdiff_t.
+     *    In 32-bit mode we are safe, because (chainLog <= 29), so
+     *    ip+ZSTD_CHUNKSIZE_MAX - cctx->base < 1<<32.
+     * 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
+     *    windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
+     */
+    U32 const cycleMask = (1U << cycleLog) - 1;
+    U32 const current = (U32)((BYTE const*)src - window->base);
+    U32 const newCurrent = (current & cycleMask) + maxDist;
+    U32 const correction = current - newCurrent;
+    assert((maxDist & cycleMask) == 0);
+    assert(current > newCurrent);
+    /* Loose bound, should be around 1<<29 (see above) */
+    assert(correction > 1<<28);
+
+    window->base += correction;
+    window->dictBase += correction;
+    window->lowLimit -= correction;
+    window->dictLimit -= correction;
+
+    DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction,
+             window->lowLimit);
+    return correction;
+}
+
+/**
+ * ZSTD_window_enforceMaxDist():
+ * Updates lowLimit so that:
+ *    (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
+ * This allows a simple check that index >= lowLimit to see if index is valid.
+ * This must be called before a block compression call, with srcEnd as the block
+ * source end.
+ * If loadedDictEndPtr is not NULL, we set it to zero once we update lowLimit.
+ * This is because dictionaries are allowed to be referenced as long as the last
+ * byte of the dictionary is in the window, but once they are out of range,
+ * they cannot be referenced. If loadedDictEndPtr is NULL, we use
+ * loadedDictEnd == 0.
+ */
+MEM_STATIC void ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
+                                           void const* srcEnd, U32 maxDist,
+                                           U32* loadedDictEndPtr)
+{
+    U32 const current = (U32)((BYTE const*)srcEnd - window->base);
+    U32 loadedDictEnd = loadedDictEndPtr != NULL ? *loadedDictEndPtr : 0;
+    if (current > maxDist + loadedDictEnd) {
+        U32 const newLowLimit = current - maxDist;
+        if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
+        if (window->dictLimit < window->lowLimit) {
+            DEBUGLOG(5, "Update dictLimit from %u to %u", window->dictLimit,
+                     window->lowLimit);
+            window->dictLimit = window->lowLimit;
+        }
+        if (loadedDictEndPtr)
+            *loadedDictEndPtr = 0;
+    }
+}
+
+/**
+ * ZSTD_window_update():
+ * Updates the window by appending [src, src + srcSize) to the window.
+ * If it is not contiguous, the current prefix becomes the extDict, and we
+ * forget about the extDict. Handles overlap of the prefix and extDict.
+ * Returns non-zero if the segment is contiguous.
+ */
+MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
+                                  void const* src, size_t srcSize)
+{
+    BYTE const* const ip = (BYTE const*)src;
+    U32 contiguous = 1;
+    /* Check if blocks follow each other */
+    if (src != window->nextSrc) {
+        /* not contiguous */
+        size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
+        DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u",
+                 window->dictLimit);
+        window->lowLimit = window->dictLimit;
+        assert(distanceFromBase == (size_t)(U32)distanceFromBase);  /* should never overflow */
+        window->dictLimit = (U32)distanceFromBase;
+        window->dictBase = window->base;
+        window->base = ip - distanceFromBase;
+        // ms->nextToUpdate = window->dictLimit;
+        if (window->dictLimit - window->lowLimit < HASH_READ_SIZE) window->lowLimit = window->dictLimit;   /* too small extDict */
+        contiguous = 0;
+    }
+    window->nextSrc = ip + srcSize;
+    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
+    if ( (ip+srcSize > window->dictBase + window->lowLimit)
+       & (ip < window->dictBase + window->dictLimit)) {
+        ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
+        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
+        window->lowLimit = lowLimitMax;
+    }
+    return contiguous;
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+
+/* ==============================================================
+ * Private declarations
+ * These prototypes shall only be called from within lib/compress
+ * ============================================================== */
+
+/* ZSTD_getCParamsFromCCtxParams() :
+ * cParams are built depending on compressionLevel, src size hints, 
+ * LDM and manually set compression parameters.
+ */
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize);
+
+/*! ZSTD_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  @return : 0, or an error code */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                     const void* dict, size_t dictSize,
+                     const ZSTD_CDict* cdict,
+                     ZSTD_CCtx_params  params, unsigned long long pledgedSrcSize);
+
+/*! ZSTD_compressStream_generic() :
+ *  Private use only. To be called from zstdmt_compress.c in single-thread mode. */
+size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                   ZSTD_outBuffer* output,
+                                   ZSTD_inBuffer* input,
+                                   ZSTD_EndDirective const flushMode);
+
+/*! ZSTD_getCParamsFromCDict() :
+ *  as the name implies */
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
+
+/* ZSTD_compressBegin_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    const ZSTD_CDict* cdict,
+                                    ZSTD_CCtx_params params,
+                                    unsigned long long pledgedSrcSize);
+
+/* ZSTD_compress_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict,size_t dictSize,
+                                 ZSTD_CCtx_params params);
+
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapcity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
+
+
+/* ZSTD_referenceExternalSequences() :
+ * Must be called before starting a compression operation.
+ * seqs must parse a prefix of the source.
+ * This cannot be used when long range matching is enabled.
+ * Zstd will use these sequences, and pass the literals to a secondary block
+ * compressor.
+ * @return : An error code on failure.
+ * NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
+ * access and data corruption.
+ */
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
+
+
+#endif /* ZSTD_COMPRESS_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_decompress.c b/vendor/github.com/DataDog/zstd/zstd_decompress.c
new file mode 100644
index 0000000..3ec6a1c
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_decompress.c
@@ -0,0 +1,3003 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTD_decompress() allocates its context,
+ * on stack (0), or into heap (1, default; requires malloc()).
+ * Note that functions with explicit context such as ZSTD_decompressDCtx() are unaffected.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif
+
+/*!
+*  LEGACY_SUPPORT :
+*  if set to 1+, ZSTD_decompress() can decode older formats (v0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 0
+#endif
+
+/*!
+ *  MAXWINDOWSIZE_DEFAULT :
+ *  maximum window size accepted by DStream __by default__.
+ *  Frames requiring more memory will be rejected.
+ *  It's possible to set a different limit using ZSTD_DCtx_setMaxWindowSize().
+ */
+#ifndef ZSTD_MAXWINDOWSIZE_DEFAULT
+#  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_DEFAULTMAX) + 1)
+#endif
+
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include <string.h>      /* memcpy, memmove, memset */
+#include "cpu.h"
+#include "mem.h"         /* low level memory routines */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "zstd_internal.h"
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+#  include "zstd_legacy.h"
+#endif
+
+
+/*-*************************************
+*  Errors
+***************************************/
+#define ZSTD_isError ERR_isError   /* for inlining */
+#define FSE_isError  ERR_isError
+#define HUF_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
+               ZSTDds_decompressLastBlock, ZSTDds_checkChecksum,
+               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTD_dStage;
+
+typedef enum { zdss_init=0, zdss_loadHeader,
+               zdss_read, zdss_load, zdss_flush } ZSTD_dStreamStage;
+
+
+typedef struct {
+    U32 fastMode;
+    U32 tableLog;
+} ZSTD_seqSymbol_header;
+
+typedef struct {
+    U16  nextState;
+    BYTE nbAdditionalBits;
+    BYTE nbBits;
+    U32  baseValue;
+} ZSTD_seqSymbol;
+
+#define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
+
+typedef struct {
+    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];
+    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];
+    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];
+    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
+    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_entropyDTables_t;
+
+struct ZSTD_DCtx_s
+{
+    const ZSTD_seqSymbol* LLTptr;
+    const ZSTD_seqSymbol* MLTptr;
+    const ZSTD_seqSymbol* OFTptr;
+    const HUF_DTable* HUFptr;
+    ZSTD_entropyDTables_t entropy;
+    const void* previousDstEnd;   /* detect continuity */
+    const void* base;             /* start of current segment */
+    const void* vBase;            /* virtual start of previous segment if it was just before current one */
+    const void* dictEnd;          /* end of previous segment */
+    size_t expected;
+    ZSTD_frameHeader fParams;
+    U64 decodedSize;
+    blockType_e bType;            /* used in ZSTD_decompressContinue(), store blockType between block header decoding and block decompression stages */
+    ZSTD_dStage stage;
+    U32 litEntropy;
+    U32 fseEntropy;
+    XXH64_state_t xxhState;
+    size_t headerSize;
+    U32 dictID;
+    ZSTD_format_e format;
+    const BYTE* litPtr;
+    ZSTD_customMem customMem;
+    size_t litSize;
+    size_t rleSize;
+    size_t staticSize;
+    int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+
+    /* streaming */
+    ZSTD_DDict* ddictLocal;
+    const ZSTD_DDict* ddict;
+    ZSTD_dStreamStage streamStage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    size_t maxWindowSize;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t lhSize;
+    void* legacyContext;
+    U32 previousLegacyVersion;
+    U32 legacyVersion;
+    U32 hostageByte;
+
+    /* workspace */
+    BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTD_DCtx within "zstd.h" */
+
+size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support sizeof NULL */
+    return sizeof(*dctx)
+           + ZSTD_sizeof_DDict(dctx->ddictLocal)
+           + dctx->inBuffSize + dctx->outBuffSize;
+}
+
+size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
+
+
+static size_t ZSTD_startingInputLength(ZSTD_format_e format)
+{
+    size_t const startingInputLength = (format==ZSTD_f_zstd1_magicless) ?
+                    ZSTD_frameHeaderSize_prefix - ZSTD_frameIdSize :
+                    ZSTD_frameHeaderSize_prefix;
+    ZSTD_STATIC_ASSERT(ZSTD_FRAMEHEADERSIZE_PREFIX >= ZSTD_FRAMEIDSIZE);
+    /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
+    assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
+    return startingInputLength;
+}
+
+static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
+{
+    dctx->format = ZSTD_f_zstd1;  /* ZSTD_decompressBegin() invokes ZSTD_startingInputLength() with argument dctx->format */
+    dctx->staticSize  = 0;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+    dctx->ddict       = NULL;
+    dctx->ddictLocal  = NULL;
+    dctx->inBuff      = NULL;
+    dctx->inBuffSize  = 0;
+    dctx->outBuffSize = 0;
+    dctx->streamStage = zdss_init;
+    dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+}
+
+ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_DCtx* const dctx = (ZSTD_DCtx*) workspace;
+
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < sizeof(ZSTD_DCtx)) return NULL;  /* minimum size */
+
+    ZSTD_initDCtx_internal(dctx);
+    dctx->staticSize = workspaceSize;
+    dctx->inBuff = (char*)(dctx+1);
+    return dctx;
+}
+
+ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
+{
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
+        if (!dctx) return NULL;
+        dctx->customMem = customMem;
+        dctx->legacyContext = NULL;
+        dctx->previousLegacyVersion = 0;
+        ZSTD_initDCtx_internal(dctx);
+        return dctx;
+    }
+}
+
+ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    DEBUGLOG(3, "ZSTD_createDCtx");
+    return ZSTD_createDCtx_advanced(ZSTD_defaultCMem);
+}
+
+size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support free on NULL */
+    if (dctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static DCtx */
+    {   ZSTD_customMem const cMem = dctx->customMem;
+        ZSTD_freeDDict(dctx->ddictLocal);
+        dctx->ddictLocal = NULL;
+        ZSTD_free(dctx->inBuff, cMem);
+        dctx->inBuff = NULL;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (dctx->legacyContext)
+            ZSTD_freeLegacyStreamContext(dctx->legacyContext, dctx->previousLegacyVersion);
+#endif
+        ZSTD_free(dctx, cMem);
+        return 0;
+    }
+}
+
+/* no longer useful */
+void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
+{
+    size_t const toCopy = (size_t)((char*)(&dstDCtx->inBuff) - (char*)dstDCtx);
+    memcpy(dstDCtx, srcDCtx, toCopy);  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+ *   Frame header decoding
+ ***************************************************************/
+
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+unsigned ZSTD_isFrame(const void* buffer, size_t size)
+{
+    if (size < ZSTD_frameIdSize) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if (magic == ZSTD_MAGICNUMBER) return 1;
+        if ((magic & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(buffer, size)) return 1;
+#endif
+    return 0;
+}
+
+/** ZSTD_frameHeaderSize_internal() :
+ *  srcSize must be large enough to reach header size fields.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
+ * @return : size of the Frame Header
+ *           or an error code, which can be tested with ZSTD_isError() */
+static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+    if (srcSize < minInputSize) return ERROR(srcSize_wrong);
+
+    {   BYTE const fhd = ((const BYTE*)src)[minInputSize-1];
+        U32 const dictID= fhd & 3;
+        U32 const singleSegment = (fhd >> 5) & 1;
+        U32 const fcsId = fhd >> 6;
+        return minInputSize + !singleSegment
+             + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
+             + (singleSegment && !fcsId);
+    }
+}
+
+/** ZSTD_frameHeaderSize() :
+ *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
+ * @return : size of the Frame Header */
+size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+{
+    return ZSTD_frameHeaderSize_internal(src, srcSize, ZSTD_f_zstd1);
+}
+
+
+/** ZSTD_getFrameHeader_internal() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_getFrameHeader_internal(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+
+    if (srcSize < minInputSize) return minInputSize;
+
+    if ( (format != ZSTD_f_zstd1_magicless)
+      && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+            /* skippable frame */
+            if (srcSize < ZSTD_skippableHeaderSize)
+                return ZSTD_skippableHeaderSize; /* magic number + frame length */
+            memset(zfhPtr, 0, sizeof(*zfhPtr));
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_frameIdSize);
+            zfhPtr->frameType = ZSTD_skippableFrame;
+            return 0;
+        }
+        return ERROR(prefix_unknown);
+    }
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    {   size_t const fhsize = ZSTD_frameHeaderSize_internal(src, srcSize, format);
+        if (srcSize < fhsize) return fhsize;
+        zfhPtr->headerSize = (U32)fhsize;
+    }
+
+    {   BYTE const fhdByte = ip[minInputSize-1];
+        size_t pos = minInputSize;
+        U32 const dictIDSizeCode = fhdByte&3;
+        U32 const checksumFlag = (fhdByte>>2)&1;
+        U32 const singleSegment = (fhdByte>>5)&1;
+        U32 const fcsID = fhdByte>>6;
+        U64 windowSize = 0;
+        U32 dictID = 0;
+        U64 frameContentSize = ZSTD_CONTENTSIZE_UNKNOWN;
+        if ((fhdByte & 0x08) != 0)
+            return ERROR(frameParameter_unsupported); /* reserved bits, must be zero */
+
+        if (!singleSegment) {
+            BYTE const wlByte = ip[pos++];
+            U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
+            if (windowLog > ZSTD_WINDOWLOG_MAX)
+                return ERROR(frameParameter_windowTooLarge);
+            windowSize = (1ULL << windowLog);
+            windowSize += (windowSize >> 3) * (wlByte&7);
+        }
+        switch(dictIDSizeCode)
+        {
+            default: assert(0);  /* impossible */
+            case 0 : break;
+            case 1 : dictID = ip[pos]; pos++; break;
+            case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
+            case 3 : dictID = MEM_readLE32(ip+pos); pos+=4; break;
+        }
+        switch(fcsID)
+        {
+            default: assert(0);  /* impossible */
+            case 0 : if (singleSegment) frameContentSize = ip[pos]; break;
+            case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
+            case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
+            case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
+        }
+        if (singleSegment) windowSize = frameContentSize;
+
+        zfhPtr->frameType = ZSTD_frame;
+        zfhPtr->frameContentSize = frameContentSize;
+        zfhPtr->windowSize = windowSize;
+        zfhPtr->blockSizeMax = (unsigned) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+        zfhPtr->dictID = dictID;
+        zfhPtr->checksumFlag = checksumFlag;
+    }
+    return 0;
+}
+
+/** ZSTD_getFrameHeader() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : this function does not consume input, it only reads it.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
+{
+    return ZSTD_getFrameHeader_internal(zfhPtr, src, srcSize, ZSTD_f_zstd1);
+}
+
+
+/** ZSTD_getFrameContentSize() :
+ *  compatible with legacy mode
+ * @return : decompressed size of the single frame pointed to be `src` if known, otherwise
+ *         - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *         - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small) */
+unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize)) {
+        unsigned long long const ret = ZSTD_getDecompressedSize_legacy(src, srcSize);
+        return ret == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : ret;
+    }
+#endif
+    {   ZSTD_frameHeader zfh;
+        if (ZSTD_getFrameHeader(&zfh, src, srcSize) != 0)
+            return ZSTD_CONTENTSIZE_ERROR;
+        if (zfh.frameType == ZSTD_skippableFrame) {
+            return 0;
+        } else {
+            return zfh.frameContentSize;
+    }   }
+}
+
+/** ZSTD_findDecompressedSize() :
+ *  compatible with legacy mode
+ *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
+ *      skippable frames
+ *  @return : decompressed size of the frames contained */
+unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long totalDstSize = 0;
+
+    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
+        U32 const magicNumber = MEM_readLE32(src);
+
+        if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+            size_t skippableSize;
+            if (srcSize < ZSTD_skippableHeaderSize)
+                return ERROR(srcSize_wrong);
+            skippableSize = MEM_readLE32((const BYTE *)src + ZSTD_frameIdSize)
+                          + ZSTD_skippableHeaderSize;
+            if (srcSize < skippableSize) {
+                return ZSTD_CONTENTSIZE_ERROR;
+            }
+
+            src = (const BYTE *)src + skippableSize;
+            srcSize -= skippableSize;
+            continue;
+        }
+
+        {   unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+            if (ret >= ZSTD_CONTENTSIZE_ERROR) return ret;
+
+            /* check for overflow */
+            if (totalDstSize + ret < totalDstSize) return ZSTD_CONTENTSIZE_ERROR;
+            totalDstSize += ret;
+        }
+        {   size_t const frameSrcSize = ZSTD_findFrameCompressedSize(src, srcSize);
+            if (ZSTD_isError(frameSrcSize)) {
+                return ZSTD_CONTENTSIZE_ERROR;
+            }
+
+            src = (const BYTE *)src + frameSrcSize;
+            srcSize -= frameSrcSize;
+        }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    if (srcSize) return ZSTD_CONTENTSIZE_ERROR;
+
+    return totalDstSize;
+}
+
+/** ZSTD_getDecompressedSize() :
+*   compatible with legacy mode
+*   @return : decompressed size if known, 0 otherwise
+              note : 0 can mean any of the following :
+                   - frame content is empty
+                   - decompressed size field is not present in frame header
+                   - frame header unknown / not supported
+                   - frame header not complete (`srcSize` too small) */
+unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
+{
+    unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_ERROR < ZSTD_CONTENTSIZE_UNKNOWN);
+    return (ret >= ZSTD_CONTENTSIZE_ERROR) ? 0 : ret;
+}
+
+
+/** ZSTD_decodeFrameHeader() :
+*   `headerSize` must be the size provided by ZSTD_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
+{
+    size_t const result = ZSTD_getFrameHeader_internal(&(dctx->fParams), src, headerSize, dctx->format);
+    if (ZSTD_isError(result)) return result;    /* invalid header */
+    if (result>0) return ERROR(srcSize_wrong);  /* headerSize too small */
+    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID))
+        return ERROR(dictionary_wrong);
+    if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
+    return 0;
+}
+
+
+/*-*************************************************************
+ *   Block decoding
+ ***************************************************************/
+
+/*! ZSTD_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr)
+{
+    if (srcSize < ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    {   U32 const cBlockHeader = MEM_readLE24(src);
+        U32 const cSize = cBlockHeader >> 3;
+        bpPtr->lastBlock = cBlockHeader & 1;
+        bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3);
+        bpPtr->origSize = cSize;   /* only useful for RLE */
+        if (bpPtr->blockType == bt_rle) return 1;
+        if (bpPtr->blockType == bt_reserved) return ERROR(corruption_detected);
+        return cSize;
+    }
+}
+
+
+static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
+                          const void* src, size_t srcSize)
+{
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                               size_t regenSize)
+{
+    if (srcSize != 1) return ERROR(srcSize_wrong);
+    if (regenSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memset(dst, *(const BYTE*)src, regenSize);
+    return regenSize;
+}
+
+/*! ZSTD_decodeLiteralsBlock() :
+ * @return : nb of bytes read from src (< srcSize )
+ *  note : symbol not declared but exposed for fullbench */
+size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    {   const BYTE* const istart = (const BYTE*) src;
+        symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3);
+
+        switch(litEncType)
+        {
+        case set_repeat:
+            if (dctx->litEntropy==0) return ERROR(dictionary_corrupted);
+            /* fall-through */
+        case set_compressed:
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
+            {   size_t lhSize, litSize, litCSize;
+                U32 singleStream=0;
+                U32 const lhlCode = (istart[0] >> 2) & 3;
+                U32 const lhc = MEM_readLE32(istart);
+                switch(lhlCode)
+                {
+                case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    /* 2 - 2 - 10 - 10 */
+                    singleStream = !lhlCode;
+                    lhSize = 3;
+                    litSize  = (lhc >> 4) & 0x3FF;
+                    litCSize = (lhc >> 14) & 0x3FF;
+                    break;
+                case 2:
+                    /* 2 - 2 - 14 - 14 */
+                    lhSize = 4;
+                    litSize  = (lhc >> 4) & 0x3FFF;
+                    litCSize = lhc >> 18;
+                    break;
+                case 3:
+                    /* 2 - 2 - 18 - 18 */
+                    lhSize = 5;
+                    litSize  = (lhc >> 4) & 0x3FFFF;
+                    litCSize = (lhc >> 22) + (istart[4] << 10);
+                    break;
+                }
+                if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+                if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+                if (HUF_isError((litEncType==set_repeat) ?
+                                    ( singleStream ?
+                                        HUF_decompress1X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) :
+                                        HUF_decompress4X_usingDTable_bmi2(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr, dctx->bmi2) ) :
+                                    ( singleStream ?
+                                        HUF_decompress1X2_DCtx_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                                                                         dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2) :
+                                        HUF_decompress4X_hufOnly_wksp_bmi2(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                                                                           dctx->entropy.workspace, sizeof(dctx->entropy.workspace), dctx->bmi2))))
+                    return ERROR(corruption_detected);
+
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                dctx->litEntropy = 1;
+                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return litCSize + lhSize;
+            }
+
+        case set_basic:
+            {   size_t litSize, lhSize;
+                U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    litSize = MEM_readLE24(istart) >> 4;
+                    break;
+                }
+
+                if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                    if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                    memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                    dctx->litPtr = dctx->litBuffer;
+                    dctx->litSize = litSize;
+                    memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                    return lhSize+litSize;
+                }
+                /* direct reference into compressed stream */
+                dctx->litPtr = istart+lhSize;
+                dctx->litSize = litSize;
+                return lhSize+litSize;
+            }
+
+        case set_rle:
+            {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t litSize, lhSize;
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    litSize = MEM_readLE24(istart) >> 4;
+                    if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                    break;
+                }
+                if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+                memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                return lhSize+1;
+            }
+        default:
+            return ERROR(corruption_detected);   /* impossible */
+        }
+    }
+}
+
+/* Default FSE distribution tables.
+ * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#default-distributions
+ * They were generated programmatically with following method :
+ * - start from default distributions, present in /lib/common/zstd_internal.h
+ * - generate tables normally, using ZSTD_buildFSETable()
+ * - printout the content of tables
+ * - pretify output, report below, test with fuzzer to ensure it's correct */
+
+/* Default FSE distribution table for Literal Lengths */
+static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
+     {  1,  1,  1, LL_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+     /* nextState, nbAddBits, nbBits, baseVal */
+     {  0,  0,  4,    0},  { 16,  0,  4,    0},
+     { 32,  0,  5,    1},  {  0,  0,  5,    3},
+     {  0,  0,  5,    4},  {  0,  0,  5,    6},
+     {  0,  0,  5,    7},  {  0,  0,  5,    9},
+     {  0,  0,  5,   10},  {  0,  0,  5,   12},
+     {  0,  0,  6,   14},  {  0,  1,  5,   16},
+     {  0,  1,  5,   20},  {  0,  1,  5,   22},
+     {  0,  2,  5,   28},  {  0,  3,  5,   32},
+     {  0,  4,  5,   48},  { 32,  6,  5,   64},
+     {  0,  7,  5,  128},  {  0,  8,  6,  256},
+     {  0, 10,  6, 1024},  {  0, 12,  6, 4096},
+     { 32,  0,  4,    0},  {  0,  0,  4,    1},
+     {  0,  0,  5,    2},  { 32,  0,  5,    4},
+     {  0,  0,  5,    5},  { 32,  0,  5,    7},
+     {  0,  0,  5,    8},  { 32,  0,  5,   10},
+     {  0,  0,  5,   11},  {  0,  0,  6,   13},
+     { 32,  1,  5,   16},  {  0,  1,  5,   18},
+     { 32,  1,  5,   22},  {  0,  2,  5,   24},
+     { 32,  3,  5,   32},  {  0,  3,  5,   40},
+     {  0,  6,  4,   64},  { 16,  6,  4,   64},
+     { 32,  7,  5,  128},  {  0,  9,  6,  512},
+     {  0, 11,  6, 2048},  { 48,  0,  4,    0},
+     { 16,  0,  4,    1},  { 32,  0,  5,    2},
+     { 32,  0,  5,    3},  { 32,  0,  5,    5},
+     { 32,  0,  5,    6},  { 32,  0,  5,    8},
+     { 32,  0,  5,    9},  { 32,  0,  5,   11},
+     { 32,  0,  5,   12},  {  0,  0,  6,   15},
+     { 32,  1,  5,   18},  { 32,  1,  5,   20},
+     { 32,  2,  5,   24},  { 32,  2,  5,   28},
+     { 32,  3,  5,   40},  { 32,  4,  5,   48},
+     {  0, 16,  6,65536},  {  0, 15,  6,32768},
+     {  0, 14,  6,16384},  {  0, 13,  6, 8192},
+};   /* LL_defaultDTable */
+
+/* Default FSE distribution table for Offset Codes */
+static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, OF_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  5,    0},     {  0,  6,  4,   61},
+    {  0,  9,  5,  509},     {  0, 15,  5,32765},
+    {  0, 21,  5,2097149},   {  0,  3,  5,    5},
+    {  0,  7,  4,  125},     {  0, 12,  5, 4093},
+    {  0, 18,  5,262141},    {  0, 23,  5,8388605},
+    {  0,  5,  5,   29},     {  0,  8,  4,  253},
+    {  0, 14,  5,16381},     {  0, 20,  5,1048573},
+    {  0,  2,  5,    1},     { 16,  7,  4,  125},
+    {  0, 11,  5, 2045},     {  0, 17,  5,131069},
+    {  0, 22,  5,4194301},   {  0,  4,  5,   13},
+    { 16,  8,  4,  253},     {  0, 13,  5, 8189},
+    {  0, 19,  5,524285},    {  0,  1,  5,    1},
+    { 16,  6,  4,   61},     {  0, 10,  5, 1021},
+    {  0, 16,  5,65533},     {  0, 28,  5,268435453},
+    {  0, 27,  5,134217725}, {  0, 26,  5,67108861},
+    {  0, 25,  5,33554429},  {  0, 24,  5,16777213},
+};   /* OF_defaultDTable */
+
+
+/* Default FSE distribution table for Match Lengths */
+static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, ML_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  6,    3},  {  0,  0,  4,    4},
+    { 32,  0,  5,    5},  {  0,  0,  5,    6},
+    {  0,  0,  5,    8},  {  0,  0,  5,    9},
+    {  0,  0,  5,   11},  {  0,  0,  6,   13},
+    {  0,  0,  6,   16},  {  0,  0,  6,   19},
+    {  0,  0,  6,   22},  {  0,  0,  6,   25},
+    {  0,  0,  6,   28},  {  0,  0,  6,   31},
+    {  0,  0,  6,   34},  {  0,  1,  6,   37},
+    {  0,  1,  6,   41},  {  0,  2,  6,   47},
+    {  0,  3,  6,   59},  {  0,  4,  6,   83},
+    {  0,  7,  6,  131},  {  0,  9,  6,  515},
+    { 16,  0,  4,    4},  {  0,  0,  4,    5},
+    { 32,  0,  5,    6},  {  0,  0,  5,    7},
+    { 32,  0,  5,    9},  {  0,  0,  5,   10},
+    {  0,  0,  6,   12},  {  0,  0,  6,   15},
+    {  0,  0,  6,   18},  {  0,  0,  6,   21},
+    {  0,  0,  6,   24},  {  0,  0,  6,   27},
+    {  0,  0,  6,   30},  {  0,  0,  6,   33},
+    {  0,  1,  6,   35},  {  0,  1,  6,   39},
+    {  0,  2,  6,   43},  {  0,  3,  6,   51},
+    {  0,  4,  6,   67},  {  0,  5,  6,   99},
+    {  0,  8,  6,  259},  { 32,  0,  4,    4},
+    { 48,  0,  4,    4},  { 16,  0,  4,    5},
+    { 32,  0,  5,    7},  { 32,  0,  5,    8},
+    { 32,  0,  5,   10},  { 32,  0,  5,   11},
+    {  0,  0,  6,   14},  {  0,  0,  6,   17},
+    {  0,  0,  6,   20},  {  0,  0,  6,   23},
+    {  0,  0,  6,   26},  {  0,  0,  6,   29},
+    {  0,  0,  6,   32},  {  0, 16,  6,65539},
+    {  0, 15,  6,32771},  {  0, 14,  6,16387},
+    {  0, 13,  6, 8195},  {  0, 12,  6, 4099},
+    {  0, 11,  6, 2051},  {  0, 10,  6, 1027},
+};   /* ML_defaultDTable */
+
+
+static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddBits)
+{
+    void* ptr = dt;
+    ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr;
+    ZSTD_seqSymbol* const cell = dt + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->nbBits = 0;
+    cell->nextState = 0;
+    assert(nbAddBits < 255);
+    cell->nbAdditionalBits = (BYTE)nbAddBits;
+    cell->baseValue = baseValue;
+}
+
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off) */
+static void
+ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+    const short* normalizedCounter, unsigned maxSymbolValue,
+    const U32* baseValue, const U32* nbAdditionalBits,
+    unsigned tableLog)
+{
+    ZSTD_seqSymbol* const tableDecode = dt+1;
+    U16 symbolNext[MaxSeq+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    assert(maxSymbolValue <= MaxSeq);
+    assert(tableLog <= MaxFSELog);
+
+    /* Init, lay down lowprob symbols */
+    {   ZSTD_seqSymbol_header DTableH;
+        DTableH.tableLog = tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].baseValue = s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].baseValue = s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            U32 const symbol = tableDecode[u].baseValue;
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
+            tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+            assert(nbAdditionalBits[symbol] < 255);
+            tableDecode[u].nbAdditionalBits = (BYTE)nbAdditionalBits[symbol];
+            tableDecode[u].baseValue = baseValue[symbol];
+    }   }
+}
+
+
+/*! ZSTD_buildSeqTable() :
+ * @return : nb bytes read from src,
+ *           or an error code if it fails */
+static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr,
+                                 symbolEncodingType_e type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const U32* baseValue, const U32* nbAdditionalBits,
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case set_rle :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        {   U32 const symbol = *(const BYTE*)src;
+            U32 const baseline = baseValue[symbol];
+            U32 const nbBits = nbAdditionalBits[symbol];
+            ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
+        }
+        *DTablePtr = DTableSpace;
+        return 1;
+    case set_basic :
+        *DTablePtr = defaultTable;
+        return 0;
+    case set_repeat:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    case set_compressed :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSE_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog);
+            *DTablePtr = DTableSpace;
+            return headerSize;
+        }
+    default :   /* impossible */
+        assert(0);
+        return ERROR(GENERIC);
+    }
+}
+
+static const U32 LL_base[MaxLL+1] = {
+                 0,    1,    2,     3,     4,     5,     6,      7,
+                 8,    9,   10,    11,    12,    13,    14,     15,
+                16,   18,   20,    22,    24,    28,    32,     40,
+                48,   64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                0x2000, 0x4000, 0x8000, 0x10000 };
+
+static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD, 0x1FFFFFFD, 0x3FFFFFFD, 0x7FFFFFFD };
+
+static const U32 OF_bits[MaxOff+1] = {
+                     0,  1,  2,  3,  4,  5,  6,  7,
+                     8,  9, 10, 11, 12, 13, 14, 15,
+                    16, 17, 18, 19, 20, 21, 22, 23,
+                    24, 25, 26, 27, 28, 29, 30, 31 };
+
+static const U32 ML_base[MaxML+1] = {
+                     3,  4,  5,    6,     7,     8,     9,    10,
+                    11, 12, 13,   14,    15,    16,    17,    18,
+                    19, 20, 21,   22,    23,    24,    25,    26,
+                    27, 28, 29,   30,    31,    32,    33,    34,
+                    35, 37, 39,   41,    43,    47,    51,    59,
+                    67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                    0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+    DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    if (ip+4 > iend) return ERROR(srcSize_wrong); /* minimum possible size */
+    {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
+        symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
+        symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
+        ip++;
+
+        /* Build DTables */
+        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
+                                                      LLtype, MaxLL, LLFSELog,
+                                                      ip, iend-ip,
+                                                      LL_base, LL_bits,
+                                                      LL_defaultDTable, dctx->fseEntropy);
+            if (ZSTD_isError(llhSize)) return ERROR(corruption_detected);
+            ip += llhSize;
+        }
+
+        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
+                                                      OFtype, MaxOff, OffFSELog,
+                                                      ip, iend-ip,
+                                                      OF_base, OF_bits,
+                                                      OF_defaultDTable, dctx->fseEntropy);
+            if (ZSTD_isError(ofhSize)) return ERROR(corruption_detected);
+            ip += ofhSize;
+        }
+
+        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
+                                                      MLtype, MaxML, MLFSELog,
+                                                      ip, iend-ip,
+                                                      ML_base, ML_bits,
+                                                      ML_defaultDTable, dctx->fseEntropy);
+            if (ZSTD_isError(mlhSize)) return ERROR(corruption_detected);
+            ip += mlhSize;
+        }
+    }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+    const BYTE* match;
+} seq_t;
+
+typedef struct {
+    size_t state;
+    const ZSTD_seqSymbol* table;
+} ZSTD_fseState;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    ZSTD_fseState stateLL;
+    ZSTD_fseState stateOffb;
+    ZSTD_fseState stateML;
+    size_t prevOffset[ZSTD_REP_NUM];
+    const BYTE* prefixStart;
+    const BYTE* dictEnd;
+    size_t pos;
+} seqState_t;
+
+
+FORCE_NOINLINE
+size_t ZSTD_execSequenceLast7(BYTE* op,
+                              BYTE* const oend, seq_t sequence,
+                              const BYTE** litPtr, const BYTE* const litLimit,
+                              const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd <= oend_w) return ERROR(GENERIC);   /* Precondition */
+
+    /* copy literals */
+    if (op < oend_w) {
+        ZSTD_wildcopy(op, *litPtr, oend_w - op);
+        *litPtr += oend_w - op;
+        op = oend_w;
+    }
+    while (op < oLitEnd) *op++ = *(*litPtr)++;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+    }   }
+    while (op < oMatchEnd) *op++ = *match++;
+    return sequenceLength;
+}
+
+
+HINT_INLINE
+size_t ZSTD_execSequence(BYTE* op,
+                         BYTE* const oend, seq_t sequence,
+                         const BYTE** litPtr, const BYTE* const litLimit,
+                         const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
+
+    /* copy Literals */
+    ZSTD_copy8(op, *litPtr);
+    if (sequence.litLength > 8)
+        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix -> go into extDict */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd + (match - base);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              U32 i;
+              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTD_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+HINT_INLINE
+size_t ZSTD_execSequenceLong(BYTE* op,
+                             BYTE* const oend, seq_t sequence,
+                             const BYTE** litPtr, const BYTE* const litLimit,
+                             const BYTE* const prefixStart, const BYTE* const dictStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = sequence.match;
+
+    /* check */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd > oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, dictStart, dictEnd);
+
+    /* copy Literals */
+    ZSTD_copy8(op, *litPtr);  /* note : op <= oLitEnd <= oend_w == oend - 8 */
+    if (sequence.litLength > 8)
+        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - dictStart)) return ERROR(corruption_detected);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              U32 i;
+              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
+              return sequenceLength;
+            }
+    }   }
+    assert(op <= oend_w);
+    assert(sequence.matchLength >= MINMATCH);
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTD_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+static void
+ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
+{
+    const void* ptr = dt;
+    const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits",
+                (U32)DStatePtr->state, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+FORCE_INLINE_TEMPLATE void
+ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD)
+{
+    ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.nextState + lowBits;
+}
+
+/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
+ * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
+ * bits before reloading. This value is the maximum number of bytes we read
+ * after reloading when we are decoding long offets.
+ */
+#define LONG_OFFSETS_MAX_EXTRA_BITS_32                       \
+    (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32       \
+        ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32  \
+        : 0)
+
+typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
+
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
+{
+    seq_t seq;
+    U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits;
+    U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits;
+    U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits;
+    U32 const totalBits = llBits+mlBits+ofBits;
+    U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue;
+    U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue;
+    U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue;
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofBits)
+            offset = 0;
+        else {
+            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+            ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+            assert(ofBits <= MaxOff);
+            if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
+                U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
+                offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                BIT_reloadDStream(&seqState->DStream);
+                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+                assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32);   /* to avoid another reload */
+            } else {
+                offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            }
+        }
+
+        if (ofBits <= 1) {
+            offset += (llBase==0);
+            if (offset) {
+                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {  /* offset == 0 */
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = mlBase
+                    + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/) : 0);  /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+        BIT_reloadDStream(&seqState->DStream);
+    if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+        BIT_reloadDStream(&seqState->DStream);
+    /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
+    ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+    seq.litLength = llBase
+                  + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits/*>0*/) : 0);    /* <=  16 bits */
+    if (MEM_32bits())
+        BIT_reloadDStream(&seqState->DStream);
+
+    DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
+                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+
+    /* ANS state update */
+    ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
+    ZSTD_updateFseState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
+    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+    ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
+
+    return seq;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences");
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            {   seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
+                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences: after decode loop, remaining nbSeq : %i", nbSeq);
+        if (nbSeq) return ERROR(corruption_detected);
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
+        memcpy(op, litPtr, lastLLSize);
+        op += lastLLSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t
+ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+
+
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequenceLong(seqState_t* seqState, ZSTD_longOffset_e const longOffsets)
+{
+    seq_t seq;
+    U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits;
+    U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits;
+    U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits;
+    U32 const totalBits = llBits+mlBits+ofBits;
+    U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue;
+    U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue;
+    U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue;
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofBits)
+            offset = 0;
+        else {
+            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+            ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+            assert(ofBits <= MaxOff);
+            if (MEM_32bits() && longOffsets) {
+                U32 const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN_32-1);
+                offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream);
+                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+            } else {
+                offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            }
+        }
+
+        if (ofBits <= 1) {
+            offset += (llBase==0);
+            if (offset) {
+                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = mlBase + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+        BIT_reloadDStream(&seqState->DStream);
+    if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+        BIT_reloadDStream(&seqState->DStream);
+    /* Verify that there is enough bits to read the rest of the data in 64-bit mode. */
+    ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+    seq.litLength = llBase + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
+    if (MEM_32bits())
+        BIT_reloadDStream(&seqState->DStream);
+
+    {   size_t const pos = seqState->pos + seq.litLength;
+        const BYTE* const matchBase = (seq.offset > pos) ? seqState->dictEnd : seqState->prefixStart;
+        seq.match = matchBase + pos - seq.offset;  /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
+                                                    * No consequence though : no memory access will occur, overly large offset will be detected in ZSTD_execSequenceLong() */
+        seqState->pos = pos + seq.matchLength;
+    }
+
+    /* ANS state update */
+    ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
+    ZSTD_updateFseState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
+    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+    ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
+
+    return seq;
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequencesLong_body(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->base);
+    const BYTE* const dictStart = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Regen sequences */
+    if (nbSeq) {
+#define STORED_SEQS 4
+#define STOSEQ_MASK (STORED_SEQS-1)
+#define ADVANCED_SEQS 4
+        seq_t sequences[STORED_SEQS];
+        int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
+        seqState_t seqState;
+        int seqNb;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        seqState.prefixStart = prefixStart;
+        seqState.pos = (size_t)(op-prefixStart);
+        seqState.dictEnd = dictEnd;
+        CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+
+        /* prepare in advance */
+        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
+            sequences[seqNb] = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
+        }
+        if (seqNb<seqAdvance) return ERROR(corruption_detected);
+
+        /* decode and decompress */
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
+            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STOSEQ_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            PREFETCH(sequence.match);  /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
+            sequences[seqNb&STOSEQ_MASK] = sequence;
+            op += oneSeqSize;
+        }
+        if (seqNb<nbSeq) return ERROR(corruption_detected);
+
+        /* finish queue */
+        seqNb -= seqAdvance;
+        for ( ; seqNb<nbSeq ; seqNb++) {
+            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STOSEQ_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+#undef STORED_SEQS
+#undef STOSEQ_MASK
+#undef ADVANCED_SEQS
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
+        memcpy(op, litPtr, lastLLSize);
+        op += lastLLSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t
+ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+
+
+#if DYNAMIC_BMI2
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+#endif
+
+typedef size_t (*ZSTD_decompressSequences_t)(
+    ZSTD_DCtx *dctx, void *dst, size_t maxDstSize,
+    const void *seqStart, size_t seqSize, int nbSeq,
+    const ZSTD_longOffset_e isLongOffset);
+
+static size_t ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                                const void* seqStart, size_t seqSize, int nbSeq,
+                                const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequences");
+#if DYNAMIC_BMI2
+    if (dctx->bmi2) {
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+  return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+static size_t ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
+                                void* dst, size_t maxDstSize,
+                                const void* seqStart, size_t seqSize, int nbSeq,
+                                const ZSTD_longOffset_e isLongOffset)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesLong");
+#if DYNAMIC_BMI2
+    if (dctx->bmi2) {
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+    }
+#endif
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset);
+}
+
+/* ZSTD_getLongOffsetsShare() :
+ * condition : offTable must be valid
+ * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
+ *           compared to maximum possible of (1<<OffFSELog) */
+static unsigned
+ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
+{
+    const void* ptr = offTable;
+    U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+    const ZSTD_seqSymbol* table = offTable + 1;
+    U32 const max = 1 << tableLog;
+    U32 u, total = 0;
+    DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+    assert(max <= (1 << OffFSELog));  /* max not too large */
+    for (u=0; u<max; u++) {
+        if (table[u].nbAdditionalBits > 22) total += 1;
+    }
+
+    assert(tableLog <= OffFSELog);
+    total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+
+    return total;
+}
+
+
+static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize, const int frame)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    /* isLongOffset must be true if there are long offsets.
+     * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
+     * We don't expect that to be the case in 64-bit mode.
+     * In block mode, window size is not known, so we have to be conservative. (note: but it could be evaluated from current-lowLimit)
+     */
+    ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)));
+    DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
+
+    if (srcSize >= ZSTD_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals section */
+    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
+        if (ZSTD_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+
+    /* Build Decoding Tables */
+    {   int nbSeq;
+        size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        srcSize -= seqHSize;
+
+        if ( (!frame || dctx->fParams.windowSize > (1<<24))
+          && (nbSeq>0) ) {  /* could probably use a larger nbSeq limit */
+            U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
+            U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+            if (shareLongOffsets >= minShare)
+                return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+        }
+
+        return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset);
+    }
+}
+
+
+static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    size_t dSize;
+    ZSTD_checkContinuity(dctx, dst);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
+
+
+/** ZSTD_insertBlock() :
+    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+ZSTDLIB_API size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
+{
+    ZSTD_checkContinuity(dctx, blockStart);
+    dctx->previousDstEnd = (const char*)blockStart + blockSize;
+    return blockSize;
+}
+
+
+static size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+{
+    if (length > dstCapacity) return ERROR(dstSize_tooSmall);
+    memset(dst, byte, length);
+    return length;
+}
+
+/** ZSTD_findFrameCompressedSize() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the compressed size of the frame starting at `src` */
+size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize))
+        return ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+#endif
+    if ( (srcSize >= ZSTD_skippableHeaderSize)
+      && (MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START ) {
+        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize);
+    } else {
+        const BYTE* ip = (const BYTE*)src;
+        const BYTE* const ipstart = ip;
+        size_t remainingSize = srcSize;
+        ZSTD_frameHeader zfh;
+
+        /* Extract Frame Header */
+        {   size_t const ret = ZSTD_getFrameHeader(&zfh, src, srcSize);
+            if (ZSTD_isError(ret)) return ret;
+            if (ret > 0) return ERROR(srcSize_wrong);
+        }
+
+        ip += zfh.headerSize;
+        remainingSize -= zfh.headerSize;
+
+        /* Loop on each block */
+        while (1) {
+            blockProperties_t blockProperties;
+            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+            if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize)
+                return ERROR(srcSize_wrong);
+
+            ip += ZSTD_blockHeaderSize + cBlockSize;
+            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
+
+            if (blockProperties.lastBlock) break;
+        }
+
+        if (zfh.checksumFlag) {   /* Final frame content checksum */
+            if (remainingSize < 4) return ERROR(srcSize_wrong);
+            ip += 4;
+            remainingSize -= 4;
+        }
+
+        return ip - ipstart;
+    }
+}
+
+/*! ZSTD_decompressFrame() :
+*   @dctx must be properly initialized */
+static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
+                                   void* dst, size_t dstCapacity,
+                             const void** srcPtr, size_t *srcSizePtr)
+{
+    const BYTE* ip = (const BYTE*)(*srcPtr);
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t remainingSize = *srcSizePtr;
+
+    /* check */
+    if (remainingSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize)
+        return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(ip, ZSTD_frameHeaderSize_prefix);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+        if (remainingSize < frameHeaderSize+ZSTD_blockHeaderSize)
+            return ERROR(srcSize_wrong);
+        CHECK_F( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) );
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize;
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize, /* frame */ 1);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            decodedSize = ZSTD_generateNxBytes(op, oend-op, *ip, blockProperties.origSize);
+            break;
+        case bt_reserved :
+        default:
+            return ERROR(corruption_detected);
+        }
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        if (dctx->fParams.checksumFlag)
+            XXH64_update(&dctx->xxhState, op, decodedSize);
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+        if (blockProperties.lastBlock) break;
+    }
+
+    if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        if ((U64)(op-ostart) != dctx->fParams.frameContentSize) {
+            return ERROR(corruption_detected);
+    }   }
+    if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
+        U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
+        U32 checkRead;
+        if (remainingSize<4) return ERROR(checksum_wrong);
+        checkRead = MEM_readLE32(ip);
+        if (checkRead != checkCalc) return ERROR(checksum_wrong);
+        ip += 4;
+        remainingSize -= 4;
+    }
+
+    /* Allow caller to get size read */
+    *srcPtr = ip;
+    *srcSizePtr = remainingSize;
+    return op-ostart;
+}
+
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
+
+static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize,
+                                  const void* dict, size_t dictSize,
+                                  const ZSTD_DDict* ddict)
+{
+    void* const dststart = dst;
+    assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
+
+    if (ddict) {
+        dict = ZSTD_DDictDictContent(ddict);
+        dictSize = ZSTD_DDictDictSize(ddict);
+    }
+
+    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
+        U32 magicNumber;
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (ZSTD_isLegacy(src, srcSize)) {
+            size_t decodedSize;
+            size_t const frameSize = ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+            if (ZSTD_isError(frameSize)) return frameSize;
+            /* legacy support is not compatible with static dctx */
+            if (dctx->staticSize) return ERROR(memory_allocation);
+
+            decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
+
+            dst = (BYTE*)dst + decodedSize;
+            dstCapacity -= decodedSize;
+
+            src = (const BYTE*)src + frameSize;
+            srcSize -= frameSize;
+
+            continue;
+        }
+#endif
+
+        magicNumber = MEM_readLE32(src);
+        DEBUGLOG(4, "reading magic number %08X (expecting %08X)",
+                    (U32)magicNumber, (U32)ZSTD_MAGICNUMBER);
+        if (magicNumber != ZSTD_MAGICNUMBER) {
+            if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+                size_t skippableSize;
+                if (srcSize < ZSTD_skippableHeaderSize)
+                    return ERROR(srcSize_wrong);
+                skippableSize = MEM_readLE32((const BYTE*)src + ZSTD_frameIdSize)
+                              + ZSTD_skippableHeaderSize;
+                if (srcSize < skippableSize) return ERROR(srcSize_wrong);
+
+                src = (const BYTE *)src + skippableSize;
+                srcSize -= skippableSize;
+                continue;
+            }
+            return ERROR(prefix_unknown);
+        }
+
+        if (ddict) {
+            /* we were called from ZSTD_decompress_usingDDict */
+            CHECK_F(ZSTD_decompressBegin_usingDDict(dctx, ddict));
+        } else {
+            /* this will initialize correctly with no dict if dict == NULL, so
+             * use this in all cases but ddict */
+            CHECK_F(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize));
+        }
+        ZSTD_checkContinuity(dctx, dst);
+
+        {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
+                                                    &src, &srcSize);
+            if (ZSTD_isError(res)) return res;
+            /* no need to bound check, ZSTD_decompressFrame already has */
+            dst = (BYTE*)dst + res;
+            dstCapacity -= res;
+        }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
+
+    if (srcSize) return ERROR(srcSize_wrong); /* input not entirely consumed */
+
+    return (BYTE*)dst - (BYTE*)dststart;
+}
+
+size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                           const void* src, size_t srcSize,
+                           const void* dict, size_t dictSize)
+{
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize, dict, dictSize, NULL);
+}
+
+
+size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTD_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
+    size_t regenSize;
+    ZSTD_DCtx* const dctx = ZSTD_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTD_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTD_DCtx dctx;
+    return ZSTD_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+
+/*-**************************************
+*   Advanced Streaming Decompression API
+*   Bufferless and synchronous
+****************************************/
+size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
+
+ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
+    switch(dctx->stage)
+    {
+    default:   /* should not happen */
+        assert(0);
+    case ZSTDds_getFrameHeaderSize:
+    case ZSTDds_decodeFrameHeader:
+        return ZSTDnit_frameHeader;
+    case ZSTDds_decodeBlockHeader:
+        return ZSTDnit_blockHeader;
+    case ZSTDds_decompressBlock:
+        return ZSTDnit_block;
+    case ZSTDds_decompressLastBlock:
+        return ZSTDnit_lastBlock;
+    case ZSTDds_checkChecksum:
+        return ZSTDnit_checksum;
+    case ZSTDds_decodeSkippableHeader:
+    case ZSTDds_skipFrame:
+        return ZSTDnit_skippableFrame;
+    }
+}
+
+static int ZSTD_isSkipFrame(ZSTD_DCtx* dctx) { return dctx->stage == ZSTDds_skipFrame; }
+
+/** ZSTD_decompressContinue() :
+ *  srcSize : must be the exact nb of bytes expected (see ZSTD_nextSrcSizeToDecompress())
+ *  @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+ *            or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (U32)srcSize);
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);  /* not allowed */
+    if (dstCapacity) ZSTD_checkContinuity(dctx, dst);
+
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        assert(src != NULL);
+        if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
+            assert(srcSize >= ZSTD_frameIdSize);  /* to read skippable magic number */
+            if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
+                memcpy(dctx->headerBuffer, src, srcSize);
+                dctx->expected = ZSTD_skippableHeaderSize - srcSize;  /* remaining to load to get full skippable frame header */
+                dctx->stage = ZSTDds_decodeSkippableHeader;
+                return 0;
+        }   }
+        dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
+        if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, srcSize);
+        dctx->expected = dctx->headerSize - srcSize;
+        dctx->stage = ZSTDds_decodeFrameHeader;
+        return 0;
+
+    case ZSTDds_decodeFrameHeader:
+        assert(src != NULL);
+        memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
+        CHECK_F(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize));
+        dctx->expected = ZSTD_blockHeaderSize;
+        dctx->stage = ZSTDds_decodeBlockHeader;
+        return 0;
+
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+            if (ZSTD_isError(cBlockSize)) return cBlockSize;
+            dctx->expected = cBlockSize;
+            dctx->bType = bp.blockType;
+            dctx->rleSize = bp.origSize;
+            if (cBlockSize) {
+                dctx->stage = bp.lastBlock ? ZSTDds_decompressLastBlock : ZSTDds_decompressBlock;
+                return 0;
+            }
+            /* empty block */
+            if (bp.lastBlock) {
+                if (dctx->fParams.checksumFlag) {
+                    dctx->expected = 4;
+                    dctx->stage = ZSTDds_checkChecksum;
+                } else {
+                    dctx->expected = 0; /* end of frame */
+                    dctx->stage = ZSTDds_getFrameHeaderSize;
+                }
+            } else {
+                dctx->expected = ZSTD_blockHeaderSize;  /* jump to next header */
+                dctx->stage = ZSTDds_decodeBlockHeader;
+            }
+            return 0;
+        }
+
+    case ZSTDds_decompressLastBlock:
+    case ZSTDds_decompressBlock:
+        DEBUGLOG(5, "ZSTD_decompressContinue: case ZSTDds_decompressBlock");
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
+                rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
+                break;
+            case bt_raw :
+                rSize = ZSTD_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                rSize = ZSTD_setRleBlock(dst, dstCapacity, src, srcSize, dctx->rleSize);
+                break;
+            case bt_reserved :   /* should never happen */
+            default:
+                return ERROR(corruption_detected);
+            }
+            if (ZSTD_isError(rSize)) return rSize;
+            DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (U32)rSize);
+            dctx->decodedSize += rSize;
+            if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+
+            if (dctx->stage == ZSTDds_decompressLastBlock) {   /* end of frame */
+                DEBUGLOG(4, "ZSTD_decompressContinue: decoded size from frame : %u", (U32)dctx->decodedSize);
+                if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+                    if (dctx->decodedSize != dctx->fParams.frameContentSize) {
+                        return ERROR(corruption_detected);
+                }   }
+                if (dctx->fParams.checksumFlag) {  /* another round for frame checksum */
+                    dctx->expected = 4;
+                    dctx->stage = ZSTDds_checkChecksum;
+                } else {
+                    dctx->expected = 0;   /* ends here */
+                    dctx->stage = ZSTDds_getFrameHeaderSize;
+                }
+            } else {
+                dctx->stage = ZSTDds_decodeBlockHeader;
+                dctx->expected = ZSTD_blockHeaderSize;
+                dctx->previousDstEnd = (char*)dst + rSize;
+            }
+            return rSize;
+        }
+
+    case ZSTDds_checkChecksum:
+        assert(srcSize == 4);  /* guaranteed by dctx->expected */
+        {   U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
+            U32 const check32 = MEM_readLE32(src);
+            DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", h32, check32);
+            if (check32 != h32) return ERROR(checksum_wrong);
+            dctx->expected = 0;
+            dctx->stage = ZSTDds_getFrameHeaderSize;
+            return 0;
+        }
+
+    case ZSTDds_decodeSkippableHeader:
+        assert(src != NULL);
+        assert(srcSize <= ZSTD_skippableHeaderSize);
+        memcpy(dctx->headerBuffer + (ZSTD_skippableHeaderSize - srcSize), src, srcSize);   /* complete skippable header */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_frameIdSize);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->stage = ZSTDds_skipFrame;
+        return 0;
+
+    case ZSTDds_skipFrame:
+        dctx->expected = 0;
+        dctx->stage = ZSTDds_getFrameHeaderSize;
+        return 0;
+
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+    return 0;
+}
+
+/* ZSTD_loadEntropy() :
+ * dict : must point at beginning of a valid zstd dictionary
+ * @return : size of entropy tables read */
+static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const dict, size_t const dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+
+    if (dictSize <= 8) return ERROR(dictionary_corrupted);
+    dictPtr += 8;   /* skip header = magic + dictID */
+
+
+    {   size_t const hSize = HUF_readDTableX4_wksp(
+            entropy->hufTable, dictPtr, dictEnd - dictPtr,
+            entropy->workspace, sizeof(entropy->workspace));
+        if (HUF_isError(hSize)) return ERROR(dictionary_corrupted);
+        dictPtr += hSize;
+    }
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue = MaxOff, offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeMaxValue > MaxOff) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        ZSTD_buildFSETable(entropy->OFTable,
+                            offcodeNCount, offcodeMaxValue,
+                            OF_base, OF_bits,
+                            offcodeLog);
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthMaxValue > MaxML) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        ZSTD_buildFSETable(entropy->MLTable,
+                            matchlengthNCount, matchlengthMaxValue,
+                            ML_base, ML_bits,
+                            matchlengthLog);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthMaxValue > MaxLL) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        ZSTD_buildFSETable(entropy->LLTable,
+                            litlengthNCount, litlengthMaxValue,
+                            LL_base, LL_bits,
+                            litlengthLog);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    {   int i;
+        size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
+        for (i=0; i<3; i++) {
+            U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
+            if (rep==0 || rep >= dictContentSize) return ERROR(dictionary_corrupted);
+            entropy->rep[i] = rep;
+    }   }
+
+    return dictPtr - (const BYTE*)dict;
+}
+
+static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return ZSTD_refDictContent(dctx, dict, dictSize);
+    {   U32 const magic = MEM_readLE32(dict);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
+    }   }
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
+
+    /* load entropy tables */
+    {   size_t const eSize = ZSTD_loadEntropy(&dctx->entropy, dict, dictSize);
+        if (ZSTD_isError(eSize)) return ERROR(dictionary_corrupted);
+        dict = (const char*)dict + eSize;
+        dictSize -= eSize;
+    }
+    dctx->litEntropy = dctx->fseEntropy = 1;
+
+    /* reference dictionary content */
+    return ZSTD_refDictContent(dctx, dict, dictSize);
+}
+
+/* Note : this function cannot fail */
+size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
+{
+    assert(dctx != NULL);
+    dctx->expected = ZSTD_startingInputLength(dctx->format);  /* dctx->format must be properly set */
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->decodedSize = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
+    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    dctx->LLTptr = dctx->entropy.LLTable;
+    dctx->MLTptr = dctx->entropy.MLTable;
+    dctx->OFTptr = dctx->entropy.OFTable;
+    dctx->HUFptr = dctx->entropy.hufTable;
+    return 0;
+}
+
+size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    CHECK_F( ZSTD_decompressBegin(dctx) );
+    if (dict && dictSize)
+        CHECK_E(ZSTD_decompress_insertDictionary(dctx, dict, dictSize), dictionary_corrupted);
+    return 0;
+}
+
+
+/* ======   ZSTD_DDict   ====== */
+
+struct ZSTD_DDict_s {
+    void* dictBuffer;
+    const void* dictContent;
+    size_t dictSize;
+    ZSTD_entropyDTables_t entropy;
+    U32 dictID;
+    U32 entropyPresent;
+    ZSTD_customMem cMem;
+};  /* typedef'd to ZSTD_DDict within "zstd.h" */
+
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict)
+{
+    return ddict->dictContent;
+}
+
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict)
+{
+    return ddict->dictSize;
+}
+
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict)
+{
+    CHECK_F( ZSTD_decompressBegin(dstDCtx) );
+    if (ddict) {   /* support begin on NULL */
+        dstDCtx->dictID = ddict->dictID;
+        dstDCtx->base = ddict->dictContent;
+        dstDCtx->vBase = ddict->dictContent;
+        dstDCtx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+        dstDCtx->previousDstEnd = dstDCtx->dictEnd;
+        if (ddict->entropyPresent) {
+            dstDCtx->litEntropy = 1;
+            dstDCtx->fseEntropy = 1;
+            dstDCtx->LLTptr = ddict->entropy.LLTable;
+            dstDCtx->MLTptr = ddict->entropy.MLTable;
+            dstDCtx->OFTptr = ddict->entropy.OFTable;
+            dstDCtx->HUFptr = ddict->entropy.hufTable;
+            dstDCtx->entropy.rep[0] = ddict->entropy.rep[0];
+            dstDCtx->entropy.rep[1] = ddict->entropy.rep[1];
+            dstDCtx->entropy.rep[2] = ddict->entropy.rep[2];
+        } else {
+            dstDCtx->litEntropy = 0;
+            dstDCtx->fseEntropy = 0;
+        }
+    }
+    return 0;
+}
+
+static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict, ZSTD_dictContentType_e dictContentType)
+{
+    ddict->dictID = 0;
+    ddict->entropyPresent = 0;
+    if (dictContentType == ZSTD_dct_rawContent) return 0;
+
+    if (ddict->dictSize < 8) {
+        if (dictContentType == ZSTD_dct_fullDict)
+            return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+        return 0;   /* pure content mode */
+    }
+    {   U32 const magic = MEM_readLE32(ddict->dictContent);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            if (dictContentType == ZSTD_dct_fullDict)
+                return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+            return 0;   /* pure content mode */
+        }
+    }
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_frameIdSize);
+
+    /* load entropy tables */
+    CHECK_E( ZSTD_loadEntropy(&ddict->entropy, ddict->dictContent, ddict->dictSize), dictionary_corrupted );
+    ddict->entropyPresent = 1;
+    return 0;
+}
+
+
+static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType)
+{
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
+        ddict->dictBuffer = NULL;
+        ddict->dictContent = dict;
+    } else {
+        void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
+        ddict->dictBuffer = internalBuffer;
+        ddict->dictContent = internalBuffer;
+        if (!internalBuffer) return ERROR(memory_allocation);
+        memcpy(internalBuffer, dict, dictSize);
+    }
+    ddict->dictSize = dictSize;
+    ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+
+    /* parse dictionary content */
+    CHECK_F( ZSTD_loadEntropy_inDDict(ddict, dictContentType) );
+
+    return 0;
+}
+
+ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_customMem customMem)
+{
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
+        if (!ddict) return NULL;
+        ddict->cMem = customMem;
+
+        if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, dictLoadMethod, dictContentType) )) {
+            ZSTD_freeDDict(ddict);
+            return NULL;
+        }
+
+        return ddict;
+    }
+}
+
+/*! ZSTD_createDDict() :
+*   Create a digested dictionary, to start decompression without startup delay.
+*   `dict` content is copied inside DDict.
+*   Consequently, `dict` can be released after `ZSTD_DDict` creation */
+ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto, allocator);
+}
+
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, to start decompression without startup delay.
+ *  Dictionary content is simply referenced, it will be accessed during decompression.
+ *  Warning : dictBuffer must outlive DDict (DDict must be freed before dictBuffer) */
+ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, allocator);
+}
+
+
+const ZSTD_DDict* ZSTD_initStaticDDict(
+                                void* workspace, size_t workspaceSize,
+                                const void* dict, size_t dictSize,
+                                ZSTD_dictLoadMethod_e dictLoadMethod,
+                                ZSTD_dictContentType_e dictContentType)
+{
+    size_t const neededSpace =
+            sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    ZSTD_DDict* const ddict = (ZSTD_DDict*)workspace;
+    assert(workspace != NULL);
+    assert(dict != NULL);
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < neededSpace) return NULL;
+    if (dictLoadMethod == ZSTD_dlm_byCopy) {
+        memcpy(ddict+1, dict, dictSize);  /* local copy */
+        dict = ddict+1;
+    }
+    if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, ZSTD_dlm_byRef, dictContentType) ))
+        return NULL;
+    return ddict;
+}
+
+
+size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = ddict->cMem;
+        ZSTD_free(ddict->dictBuffer, cMem);
+        ZSTD_free(ddict, cMem);
+        return 0;
+    }
+}
+
+/*! ZSTD_estimateDDictSize() :
+ *  Estimate amount of memory that will be needed to create a dictionary for decompression.
+ *  Note : dictionary created by reference using ZSTD_dlm_byRef are smaller */
+size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    return sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+}
+
+size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*ddict) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
+}
+
+/*! ZSTD_getDictID_fromDict() :
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
+    return MEM_readLE32((const char*)dict + ZSTD_frameIdSize);
+}
+
+/*! ZSTD_getDictID_fromDDict() :
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;
+    return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
+}
+
+/*! ZSTD_getDictID_fromFrame() :
+ *  Provides the dictID required to decompresse frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary (most common case).
+ *  - The frame was built with dictID intentionally removed.
+ *    Needed dictionary is a hidden information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, frame header could not be decoded.
+ *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use
+ *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
+unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
+{
+    ZSTD_frameHeader zfp = { 0, 0, 0, ZSTD_frame, 0, 0, 0 };
+    size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
+    if (ZSTD_isError(hError)) return 0;
+    return zfp.dictID;
+}
+
+
+/*! ZSTD_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Use dictionary without significant overhead. */
+size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_DDict* ddict)
+{
+    /* pass content and size in case legacy frames are encountered */
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize,
+                                     NULL, 0,
+                                     ddict);
+}
+
+
+/*=====================================
+*   Streaming decompression
+*====================================*/
+
+ZSTD_DStream* ZSTD_createDStream(void)
+{
+    DEBUGLOG(3, "ZSTD_createDStream");
+    return ZSTD_createDStream_advanced(ZSTD_defaultCMem);
+}
+
+ZSTD_DStream* ZSTD_initStaticDStream(void *workspace, size_t workspaceSize)
+{
+    return ZSTD_initStaticDCtx(workspace, workspaceSize);
+}
+
+ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
+{
+    return ZSTD_createDCtx_advanced(customMem);
+}
+
+size_t ZSTD_freeDStream(ZSTD_DStream* zds)
+{
+    return ZSTD_freeDCtx(zds);
+}
+
+
+/* *** Initialization *** */
+
+size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
+size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
+
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    ZSTD_freeDDict(dctx->ddictLocal);
+    if (dict && dictSize >= 8) {
+        dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
+        if (dctx->ddictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        dctx->ddictLocal = NULL;
+    }
+    dctx->ddict = dctx->ddictLocal;
+    return 0;
+}
+
+size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType);
+}
+
+size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_DCtx_refPrefix_advanced(dctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+
+/* ZSTD_initDStream_usingDict() :
+ * return : expected size, aka ZSTD_frameHeaderSize_prefix.
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
+{
+    DEBUGLOG(4, "ZSTD_initDStream_usingDict");
+    zds->streamStage = zdss_init;
+    CHECK_F( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) );
+    return ZSTD_frameHeaderSize_prefix;
+}
+
+/* note : this variant can't fail */
+size_t ZSTD_initDStream(ZSTD_DStream* zds)
+{
+    DEBUGLOG(4, "ZSTD_initDStream");
+    return ZSTD_initDStream_usingDict(zds, NULL, 0);
+}
+
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->ddict = ddict;
+    return 0;
+}
+
+/* ZSTD_initDStream_usingDDict() :
+ * ddict will just be referenced, and must outlive decompression session
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
+{
+    size_t const initResult = ZSTD_initDStream(dctx);
+    dctx->ddict = ddict;
+    return initResult;
+}
+
+/* ZSTD_resetDStream() :
+ * return : expected size, aka ZSTD_frameHeaderSize_prefix.
+ * this function cannot fail */
+size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
+{
+    DEBUGLOG(4, "ZSTD_resetDStream");
+    dctx->streamStage = zdss_loadHeader;
+    dctx->lhSize = dctx->inPos = dctx->outStart = dctx->outEnd = 0;
+    dctx->legacyVersion = 0;
+    dctx->hostageByte = 0;
+    return ZSTD_frameHeaderSize_prefix;
+}
+
+size_t ZSTD_setDStreamParameter(ZSTD_DStream* dctx,
+                                ZSTD_DStreamParameter_e paramType, unsigned paramValue)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    switch(paramType)
+    {
+        default : return ERROR(parameter_unsupported);
+        case DStream_p_maxWindowSize :
+            DEBUGLOG(4, "setting maxWindowSize = %u KB", paramValue >> 10);
+            dctx->maxWindowSize = paramValue ? paramValue : (U32)(-1);
+            break;
+    }
+    return 0;
+}
+
+size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
+{
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->maxWindowSize = maxWindowSize;
+    return 0;
+}
+
+size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
+{
+    DEBUGLOG(4, "ZSTD_DCtx_setFormat : %u", (unsigned)format);
+    if (dctx->streamStage != zdss_init) return ERROR(stage_wrong);
+    dctx->format = format;
+    return 0;
+}
+
+
+size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
+{
+    return ZSTD_sizeof_DCtx(dctx);
+}
+
+size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
+{
+    size_t const blockSize = (size_t) MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    unsigned long long const neededRBSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
+    unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
+    size_t const minRBSize = (size_t) neededSize;
+    if ((unsigned long long)minRBSize != neededSize) return ERROR(frameParameter_windowTooLarge);
+    return minRBSize;
+}
+
+size_t ZSTD_estimateDStreamSize(size_t windowSize)
+{
+    size_t const blockSize = MIN(windowSize, ZSTD_BLOCKSIZE_MAX);
+    size_t const inBuffSize = blockSize;  /* no block can be larger */
+    size_t const outBuffSize = ZSTD_decodingBufferSize_min(windowSize, ZSTD_CONTENTSIZE_UNKNOWN);
+    return ZSTD_estimateDCtxSize() + inBuffSize + outBuffSize;
+}
+
+size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
+{
+    U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;   /* note : should be user-selectable */
+    ZSTD_frameHeader zfh;
+    size_t const err = ZSTD_getFrameHeader(&zfh, src, srcSize);
+    if (ZSTD_isError(err)) return err;
+    if (err>0) return ERROR(srcSize_wrong);
+    if (zfh.windowSize > windowSizeMax)
+        return ERROR(frameParameter_windowTooLarge);
+    return ZSTD_estimateDStreamSize((size_t)zfh.windowSize);
+}
+
+
+/* *****   Decompression   ***** */
+
+MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    memcpy(dst, src, length);
+    return length;
+}
+
+
+size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    const char* const istart = (const char*)(input->src) + input->pos;
+    const char* const iend = (const char*)(input->src) + input->size;
+    const char* ip = istart;
+    char* const ostart = (char*)(output->dst) + output->pos;
+    char* const oend = (char*)(output->dst) + output->size;
+    char* op = ostart;
+    U32 someMoreWork = 1;
+
+    DEBUGLOG(5, "ZSTD_decompressStream");
+    if (input->pos > input->size) {  /* forbidden */
+        DEBUGLOG(5, "in: pos: %u   vs size: %u",
+                    (U32)input->pos, (U32)input->size);
+        return ERROR(srcSize_wrong);
+    }
+    if (output->pos > output->size) {  /* forbidden */
+        DEBUGLOG(5, "out: pos: %u   vs size: %u",
+                    (U32)output->pos, (U32)output->size);
+        return ERROR(dstSize_tooSmall);
+    }
+    DEBUGLOG(5, "input size : %u", (U32)(input->size - input->pos));
+
+    while (someMoreWork) {
+        switch(zds->streamStage)
+        {
+        case zdss_init :
+            DEBUGLOG(5, "stage zdss_init => transparent reset ");
+            ZSTD_resetDStream(zds);   /* transparent reset on starting decoding a new frame */
+            /* fall-through */
+
+        case zdss_loadHeader :
+            DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+            if (zds->legacyVersion) {
+                /* legacy support is incompatible with static dctx */
+                if (zds->staticSize) return ERROR(memory_allocation);
+                {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
+                    if (hint==0) zds->streamStage = zdss_init;
+                    return hint;
+            }   }
+#endif
+            {   size_t const hSize = ZSTD_getFrameHeader_internal(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+                DEBUGLOG(5, "header size : %u", (U32)hSize);
+                if (ZSTD_isError(hSize)) {
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+                    U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
+                    if (legacyVersion) {
+                        const void* const dict = zds->ddict ? zds->ddict->dictContent : NULL;
+                        size_t const dictSize = zds->ddict ? zds->ddict->dictSize : 0;
+                        DEBUGLOG(5, "ZSTD_decompressStream: detected legacy version v0.%u", legacyVersion);
+                        /* legacy support is incompatible with static dctx */
+                        if (zds->staticSize) return ERROR(memory_allocation);
+                        CHECK_F(ZSTD_initLegacyStream(&zds->legacyContext,
+                                    zds->previousLegacyVersion, legacyVersion,
+                                    dict, dictSize));
+                        zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
+                        {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
+                            if (hint==0) zds->streamStage = zdss_init;   /* or stay in stage zdss_loadHeader */
+                            return hint;
+                    }   }
+#endif
+                    return hSize;   /* error */
+                }
+                if (hSize != 0) {   /* need more input */
+                    size_t const toLoad = hSize - zds->lhSize;   /* if hSize!=0, hSize > zds->lhSize */
+                    size_t const remainingInput = (size_t)(iend-ip);
+                    assert(iend >= ip);
+                    if (toLoad > remainingInput) {   /* not enough input to load full header */
+                        if (remainingInput > 0) {
+                            memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            zds->lhSize += remainingInput;
+                        }
+                        input->pos = input->size;
+                        return (MAX(ZSTD_frameHeaderSize_min, hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    assert(ip != NULL);
+                    memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* check for single-pass mode opportunity */
+            if (zds->fParams.frameContentSize && zds->fParams.windowSize /* skippable frame if == 0 */
+                && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
+                size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
+                if (cSize <= (size_t)(iend-istart)) {
+                    /* shortcut : using single-pass mode */
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, oend-op, istart, cSize, zds->ddict);
+                    if (ZSTD_isError(decompressedSize)) return decompressedSize;
+                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
+                    ip = istart + cSize;
+                    op += decompressedSize;
+                    zds->expected = 0;
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+            }   }
+
+            /* Consume header (see ZSTDds_decodeFrameHeader) */
+            DEBUGLOG(4, "Consume header");
+            CHECK_F(ZSTD_decompressBegin_usingDDict(zds, zds->ddict));
+
+            if ((MEM_readLE32(zds->headerBuffer) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_frameIdSize);
+                zds->stage = ZSTDds_skipFrame;
+            } else {
+                CHECK_F(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize));
+                zds->expected = ZSTD_blockHeaderSize;
+                zds->stage = ZSTDds_decodeBlockHeader;
+            }
+
+            /* control buffer memory usage */
+            DEBUGLOG(4, "Control max memory usage (%u KB <= max %u KB)",
+                        (U32)(zds->fParams.windowSize >>10),
+                        (U32)(zds->maxWindowSize >> 10) );
+            zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
+            if (zds->fParams.windowSize > zds->maxWindowSize) return ERROR(frameParameter_windowTooLarge);
+
+            /* Adapt buffer sizes to frame header instructions */
+            {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
+                size_t const neededOutBuffSize = ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize);
+                if ((zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize)) {
+                    size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
+                    DEBUGLOG(4, "inBuff  : from %u to %u",
+                                (U32)zds->inBuffSize, (U32)neededInBuffSize);
+                    DEBUGLOG(4, "outBuff : from %u to %u",
+                                (U32)zds->outBuffSize, (U32)neededOutBuffSize);
+                    if (zds->staticSize) {  /* static DCtx */
+                        DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
+                        assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
+                        if (bufferSize > zds->staticSize - sizeof(ZSTD_DCtx))
+                            return ERROR(memory_allocation);
+                    } else {
+                        ZSTD_free(zds->inBuff, zds->customMem);
+                        zds->inBuffSize = 0;
+                        zds->outBuffSize = 0;
+                        zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
+                        if (zds->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    zds->inBuffSize = neededInBuffSize;
+                    zds->outBuff = zds->inBuff + zds->inBuffSize;
+                    zds->outBuffSize = neededOutBuffSize;
+            }   }
+            zds->streamStage = zdss_read;
+            /* fall-through */
+
+        case zdss_read:
+            DEBUGLOG(5, "stage zdss_read");
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+                DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
+                if (neededInSize==0) {  /* end of frame */
+                    zds->streamStage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    int const isSkipFrame = ZSTD_isSkipFrame(zds);
+                    size_t const decodedSize = ZSTD_decompressContinue(zds,
+                        zds->outBuff + zds->outStart, (isSkipFrame ? 0 : zds->outBuffSize - zds->outStart),
+                        ip, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
+                    zds->outEnd = zds->outStart + decodedSize;
+                    zds->streamStage = zdss_flush;
+                    break;
+            }   }
+            if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
+            zds->streamStage = zdss_load;
+            /* fall-through */
+
+        case zdss_load:
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+                size_t const toLoad = neededInSize - zds->inPos;
+                int const isSkipFrame = ZSTD_isSkipFrame(zds);
+                size_t loadedSize;
+                if (isSkipFrame) {
+                    loadedSize = MIN(toLoad, (size_t)(iend-ip));
+                } else {
+                    if (toLoad > zds->inBuffSize - zds->inPos) return ERROR(corruption_detected);   /* should never happen */
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                }
+                ip += loadedSize;
+                zds->inPos += loadedSize;
+                if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {   size_t const decodedSize = ZSTD_decompressContinue(zds,
+                        zds->outBuff + zds->outStart, zds->outBuffSize - zds->outStart,
+                        zds->inBuff, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    zds->inPos = 0;   /* input is consumed */
+                    if (!decodedSize && !isSkipFrame) { zds->streamStage = zdss_read; break; }   /* this was just a header */
+                    zds->outEnd = zds->outStart +  decodedSize;
+            }   }
+            zds->streamStage = zdss_flush;
+            /* fall-through */
+
+        case zdss_flush:
+            {   size_t const toFlushSize = zds->outEnd - zds->outStart;
+                size_t const flushedSize = ZSTD_limitCopy(op, oend-op, zds->outBuff + zds->outStart, toFlushSize);
+                op += flushedSize;
+                zds->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {  /* flush completed */
+                    zds->streamStage = zdss_read;
+                    if ( (zds->outBuffSize < zds->fParams.frameContentSize)
+                      && (zds->outStart + zds->fParams.blockSizeMax > zds->outBuffSize) ) {
+                        DEBUGLOG(5, "restart filling outBuff from beginning (left:%i, needed:%u)",
+                                (int)(zds->outBuffSize - zds->outStart),
+                                (U32)zds->fParams.blockSizeMax);
+                        zds->outStart = zds->outEnd = 0;
+                    }
+                    break;
+            }   }
+            /* cannot complete flush */
+            someMoreWork = 0;
+            break;
+
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    input->pos += (size_t)(ip-istart);
+    output->pos += (size_t)(op-ostart);
+    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds);
+        if (!nextSrcSizeHint) {   /* frame fully decoded */
+            if (zds->outEnd == zds->outStart) {  /* output fully flushed */
+                if (zds->hostageByte) {
+                    if (input->pos >= input->size) {
+                        /* can't release hostage (not present) */
+                        zds->streamStage = zdss_read;
+                        return 1;
+                    }
+                    input->pos++;  /* release hostage */
+                }   /* zds->hostageByte */
+                return 0;
+            }  /* zds->outEnd == zds->outStart */
+            if (!zds->hostageByte) { /* output not fully flushed; keep last byte as hostage; will be released when all output is flushed */
+                input->pos--;   /* note : pos > 0, otherwise, impossible to finish reading last block */
+                zds->hostageByte=1;
+            }
+            return 1;
+        }  /* nextSrcSizeHint==0 */
+        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds) == ZSTDnit_block);   /* preload header of next block */
+        assert(zds->inPos <= nextSrcSizeHint);
+        nextSrcSizeHint -= zds->inPos;   /* part already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+size_t ZSTD_decompress_generic(ZSTD_DCtx* dctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    return ZSTD_decompressStream(dctx, output, input);
+}
+
+size_t ZSTD_decompress_generic_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos)
+{
+    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
+    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
+    size_t const cErr = ZSTD_decompress_generic(dctx, &output, &input);
+    *dstPos = output.pos;
+    *srcPos = input.pos;
+    return cErr;
+}
+
+void ZSTD_DCtx_reset(ZSTD_DCtx* dctx)
+{
+    (void)ZSTD_initDStream(dctx);
+    dctx->format = ZSTD_f_zstd1;
+    dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_double_fast.c b/vendor/github.com/DataDog/zstd/zstd_double_fast.c
new file mode 100644
index 0000000..86e6b39
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_double_fast.c
@@ -0,0 +1,327 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_double_fast.h"
+
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              ZSTD_compressionParameters const* cParams,
+                              void const* end)
+{
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32  const mls = cParams->searchLength;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const current = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0)
+                hashSmall[smHash] = current + i;
+            if (i == 0 || hashLarge[lgHash] == 0)
+                hashLarge[lgHash] = current + i;
+        }
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_doubleFast_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32 lowestIndex = ms->window.dictLimit;
+    const BYTE* const lowest = base + lowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved = 0;
+
+    /* init */
+    ip += (ip==lowest);
+    {   U32 const maxRep = (U32)(ip-lowest);
+        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    }
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
+        size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        U32 const current = (U32)(ip-base);
+        U32 const matchIndexL = hashLong[h2];
+        U32 const matchIndexS = hashSmall[h];
+        const BYTE* matchLong = base + matchIndexL;
+        const BYTE* match = base + matchIndexS;
+        hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
+
+        assert(offset_1 <= current);   /* supposed guaranteed by construction */
+        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+            /* favor repcode */
+            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            U32 offset;
+            if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) {
+                mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
+                offset = (U32)(ip-matchLong);
+                while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+            } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) {
+                size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndexL3 = hashLong[hl3];
+                const BYTE* matchL3 = base + matchIndexL3;
+                hashLong[hl3] = current + 1;
+                if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) {
+                    mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
+                    ip++;
+                    offset = (U32)(ip-matchL3);
+                    while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                } else {
+                    mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+                    offset = (U32)(ip-match);
+                    while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                }
+            } else {
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
+                continue;
+            }
+
+            offset_2 = offset_1;
+            offset_1 = offset;
+
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }
+
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] =
+                hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] =
+                hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
+                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    const U32 mls = cParams->searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, cParams, src, srcSize, 7);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    U32* const hashLong = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ms->window.lowLimit;
+    const BYTE* const dictStart = dictBase + lowestIndex;
+    const U32   dictLimit = ms->window.dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
+        const U32 matchIndex = hashSmall[hSmall];
+        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+
+        const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
+        const U32 matchLongIndex = hashLong[hLong];
+        const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base;
+        const BYTE* matchLong = matchLongBase + matchLongIndex;
+
+        const U32 current = (U32)(ip-base);
+        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
+        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
+        const BYTE* repMatch = repBase + repIndex;
+        size_t mLength;
+        hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
+
+        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8;
+                offset = current - matchLongIndex;
+                while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+            } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndex3 = hashLong[h3];
+                const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base;
+                const BYTE* match3 = match3Base + matchIndex3;
+                U32 offset;
+                hashLong[h3] = current + 1;
+                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend;
+                    const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8;
+                    ip++;
+                    offset = current+1 - matchIndex3;
+                    while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
+                } else {
+                    const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
+                    const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                    offset = current - matchIndex;
+                    while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                }
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+            } else {
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
+                continue;
+        }   }
+
+        /* found a match : store it */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;
+            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2;
+            hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    U32 const mls = cParams->searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 7);
+    }
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_double_fast.h b/vendor/github.com/DataDog/zstd/zstd_double_fast.h
new file mode 100644
index 0000000..6d80b27
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_double_fast.h
@@ -0,0 +1,36 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_DOUBLE_FAST_H
+#define ZSTD_DOUBLE_FAST_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "mem.h"      /* U32 */
+#include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              ZSTD_compressionParameters const* cParams,
+                              void const* end);
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_DOUBLE_FAST_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_errors.h b/vendor/github.com/DataDog/zstd/zstd_errors.h
new file mode 100644
index 0000000..57533f2
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_errors.h
@@ -0,0 +1,92 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_ERRORS_H_398273423
+#define ZSTD_ERRORS_H_398273423
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*===== dependency =====*/
+#include <stddef.h>   /* size_t */
+
+
+/* =====   ZSTDERRORLIB_API : control library symbols visibility   ===== */
+#ifndef ZSTDERRORLIB_VISIBILITY
+#  if defined(__GNUC__) && (__GNUC__ >= 4)
+#    define ZSTDERRORLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#  else
+#    define ZSTDERRORLIB_VISIBILITY
+#  endif
+#endif
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZSTDERRORLIB_API __declspec(dllexport) ZSTDERRORLIB_VISIBILITY
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZSTDERRORLIB_API __declspec(dllimport) ZSTDERRORLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBILITY
+#endif
+
+/*-*********************************************
+ *  Error codes list
+ *-*********************************************
+ *  Error codes _values_ are pinned down since v1.3.1 only.
+ *  Therefore, don't rely on values if you may link to any version < v1.3.1.
+ *
+ *  Only values < 100 are considered stable.
+ *
+ *  note 1 : this API shall be used with static linking only.
+ *           dynamic linking is not yet officially supported.
+ *  note 2 : Prefer relying on the enum than on its value whenever possible
+ *           This is the only supported way to use the error list < v1.3.1
+ *  note 3 : ZSTD_isError() is always correct, whatever the library version.
+ **********************************************/
+typedef enum {
+  ZSTD_error_no_error = 0,
+  ZSTD_error_GENERIC  = 1,
+  ZSTD_error_prefix_unknown                = 10,
+  ZSTD_error_version_unsupported           = 12,
+  ZSTD_error_frameParameter_unsupported    = 14,
+  ZSTD_error_frameParameter_windowTooLarge = 16,
+  ZSTD_error_corruption_detected = 20,
+  ZSTD_error_checksum_wrong      = 22,
+  ZSTD_error_dictionary_corrupted      = 30,
+  ZSTD_error_dictionary_wrong          = 32,
+  ZSTD_error_dictionaryCreation_failed = 34,
+  ZSTD_error_parameter_unsupported   = 40,
+  ZSTD_error_parameter_outOfBound    = 42,
+  ZSTD_error_tableLog_tooLarge       = 44,
+  ZSTD_error_maxSymbolValue_tooLarge = 46,
+  ZSTD_error_maxSymbolValue_tooSmall = 48,
+  ZSTD_error_stage_wrong       = 60,
+  ZSTD_error_init_missing      = 62,
+  ZSTD_error_memory_allocation = 64,
+  ZSTD_error_workSpace_tooSmall= 66,
+  ZSTD_error_dstSize_tooSmall = 70,
+  ZSTD_error_srcSize_wrong    = 72,
+  /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
+  ZSTD_error_frameIndex_tooLarge = 100,
+  ZSTD_error_seekableIO          = 102,
+  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
+} ZSTD_ErrorCode;
+
+/*! ZSTD_getErrorCode() :
+    convert a `size_t` function result into a `ZSTD_ErrorCode` enum type,
+    which can be used to compare with enum list published above */
+ZSTDERRORLIB_API ZSTD_ErrorCode ZSTD_getErrorCode(size_t functionResult);
+ZSTDERRORLIB_API const char* ZSTD_getErrorString(ZSTD_ErrorCode code);   /**< Same as ZSTD_getErrorName, but using a `ZSTD_ErrorCode` enum argument */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_ERRORS_H_398273423 */
diff --git a/vendor/github.com/DataDog/zstd/zstd_fast.c b/vendor/github.com/DataDog/zstd/zstd_fast.c
new file mode 100644
index 0000000..df4d28b
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_fast.c
@@ -0,0 +1,259 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_fast.h"
+
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        ZSTD_compressionParameters const* cParams,
+                        void const* end)
+{
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog;
+    U32  const mls = cParams->searchLength;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const current = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const hash = ZSTD_hashPtr(ip + i, hBits, mls);
+            if (i == 0 || hashTable[hash] == 0)
+                hashTable[hash] = current + i;
+        }
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_fast_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const hlog, U32 const stepSize, U32 const mls)
+{
+    U32* const hashTable = ms->hashTable;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ms->window.dictLimit;
+    const BYTE* const lowest = base + lowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved = 0;
+
+    /* init */
+    ip += (ip==lowest);
+    {   U32 const maxRep = (U32)(ip-lowest);
+        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    }
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        size_t const h = ZSTD_hashPtr(ip, hlog, mls);
+        U32 const current = (U32)(ip-base);
+        U32 const matchIndex = hashTable[h];
+        const BYTE* match = base + matchIndex;
+        hashTable[h] = current;   /* update hash table */
+
+        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ( (matchIndex <= lowestIndex)
+              || (MEM_read32(match) != MEM_read32(ip)) ) {
+                assert(stepSize >= 1);
+                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+                continue;
+            }
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            {   U32 const offset = (U32)(ip-match);
+                while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }   }
+
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; }  /* swap offset_2 <=> offset_1 */
+                hashTable[ZSTD_hashPtr(ip, hlog, mls)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    U32 const hlog = cParams->hashLog;
+    U32 const mls = cParams->searchLength;
+    U32 const stepSize = cParams->targetLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 7);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_fast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const hlog, U32 const stepSize, U32 const mls)
+{
+    U32* hashTable = ms->hashTable;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ms->window.lowLimit;
+    const BYTE* const dictStart = dictBase + lowestIndex;
+    const U32   dictLimit = ms->window.dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t h = ZSTD_hashPtr(ip, hlog, mls);
+        const U32 matchIndex = hashTable[h];
+        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+        const U32 current = (U32)(ip-base);
+        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
+        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
+        const BYTE* repMatch = repBase + repIndex;
+        size_t mLength;
+        hashTable[h] = current;   /* update hash table */
+
+        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ( (matchIndex < lowestIndex) ||
+                 (MEM_read32(match) != MEM_read32(ip)) ) {
+                assert(stepSize >= 1);
+                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+                continue;
+            }
+            {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                offset = current - matchIndex;
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }   }
+
+        /* found a match : store it */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;
+            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH);
+                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    U32 const hlog = cParams->hashLog;
+    U32 const mls = cParams->searchLength;
+    U32 const stepSize = cParams->targetLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, hlog, stepSize, 7);
+    }
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_fast.h b/vendor/github.com/DataDog/zstd/zstd_fast.h
new file mode 100644
index 0000000..f0438ad
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_fast.h
@@ -0,0 +1,35 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_FAST_H
+#define ZSTD_FAST_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "mem.h"      /* U32 */
+#include "zstd_compress_internal.h"
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        ZSTD_compressionParameters const* cParams,
+                        void const* end);
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FAST_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_internal.h b/vendor/github.com/DataDog/zstd/zstd_internal.h
new file mode 100644
index 0000000..65c08a8
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_internal.h
@@ -0,0 +1,290 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+/* this module contains definitions which must be identical
+ * across compression, decompression and dictBuilder.
+ * It also contains a few functions useful to at least 2 of them
+ * and which benefit from being inlined */
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "compiler.h"
+#include "mem.h"
+#include "error_private.h"
+#define ZSTD_STATIC_LINKING_ONLY
+#include "zstd.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY  /* XXH64_state_t */
+#endif
+#include "xxhash.h"                /* XXH_reset, update, digest */
+
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*-*************************************
+*  Debug
+***************************************/
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
+
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
+#  include <stdio.h>
+extern int g_debuglog_enable;
+/* recommended values for ZSTD_DEBUG display levels :
+ * 1 : no display, enables assert() only
+ * 2 : reserved for currently active debug path
+ * 3 : events once per object lifetime (CCtx, CDict, etc.)
+ * 4 : events once per frame
+ * 5 : events once per block
+ * 6 : events once per sequence (*very* verbose) */
+#  define RAWLOG(l, ...) {                                      \
+                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
+                    fprintf(stderr, __VA_ARGS__);               \
+            }   }
+#  define DEBUGLOG(l, ...) {                                    \
+                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
+                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
+                    fprintf(stderr, " \n");                     \
+            }   }
+#else
+#  define RAWLOG(l, ...)      {}    /* disabled */
+#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#endif
+
+
+/*-*************************************
+*  shared macros
+***************************************/
+#undef MIN
+#undef MAX
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+#define CHECK_F(f) { size_t const errcod = f; if (ERR_isError(errcod)) return errcod; }  /* check and Forward error code */
+#define CHECK_E(f, e) { size_t const errcod = f; if (ERR_isError(errcod)) return ERROR(e); }  /* check and send Error code */
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTD_OPT_NUM    (1<<12)
+
+#define ZSTD_REP_NUM      3                 /* number of repcodes */
+#define ZSTD_REP_MOVE     (ZSTD_REP_NUM-1)
+static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTD_WINDOWLOG_ABSOLUTEMIN 10
+#define ZSTD_WINDOWLOG_DEFAULTMAX 27 /* Default maximum allowed window log */
+static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
+static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
+
+#define ZSTD_FRAMEIDSIZE 4
+static const size_t ZSTD_frameIdSize = ZSTD_FRAMEIDSIZE;  /* magic number size */
+
+#define ZSTD_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
+static const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
+typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define HufLog 12
+typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingType_e;
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML   52
+#define MaxLL   35
+#define DefaultMaxOff 28
+#define MaxOff  31
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+#define MaxFSELog  MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3,
+                                      4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2,
+                                             2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2,
+                                             2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+#define LL_DEFAULTNORMLOG 6  /* for static allocation */
+static const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3,
+                                      4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2,
+                                             2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+#define ML_DEFAULTNORMLOG 6  /* for static allocation */
+static const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
+
+static const S16 OF_defaultNorm[DefaultMaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2,
+                                                     2, 1, 1, 1, 1, 1, 1, 1,
+                                                     1, 1, 1, 1, 1, 1, 1, 1,
+                                                    -1,-1,-1,-1,-1 };
+#define OF_DEFAULTNORMLOG 5  /* for static allocation */
+static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy() :
+ *  custom version of memcpy(), can overwrite up to WILDCOPY_OVERLENGTH bytes (if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+MEM_STATIC void ZSTD_wildcopy_e(void* dst, const void* src, void* dstEnd)   /* should be faster for decoding, but strangely, not verified on all platform */
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = (BYTE*)dstEnd;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private declarations
+*********************************************/
+typedef struct seqDef_s {
+    U32 offset;
+    U16 litLength;
+    U16 matchLength;
+} seqDef;
+
+typedef struct {
+    seqDef* sequencesStart;
+    seqDef* sequences;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* llCode;
+    BYTE* mlCode;
+    BYTE* ofCode;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+} seqStore_t;
+
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBuilder */
+void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
+
+/* custom memory allocation functions */
+void* ZSTD_malloc(size_t size, ZSTD_customMem customMem);
+void* ZSTD_calloc(size_t size, ZSTD_customMem customMem);
+void ZSTD_free(void* ptr, ZSTD_customMem customMem);
+
+
+MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
+{
+    assert(val != 0);
+    {
+#   if defined(_MSC_VER)   /* Visual */
+        unsigned long r=0;
+        _BitScanReverse(&r, val);
+        return (unsigned)r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
+        return 31 - __builtin_clz(val);
+#   else   /* Software version */
+        static const U32 DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+        U32 v = val;
+        v |= v >> 1;
+        v |= v >> 2;
+        v |= v >> 4;
+        v |= v >> 8;
+        v |= v >> 16;
+        return DeBruijnClz[(v * 0x07C4ACDDU) >> 27];
+#   endif
+    }
+}
+
+
+/* ZSTD_invalidateRepCodes() :
+ * ensures next compression will not use repcodes from previous block.
+ * Note : only works with regular variant;
+ *        do not use with extDict variant ! */
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx);   /* zstdmt, adaptive_compression (shouldn't get this definition from here) */
+
+
+typedef struct {
+    blockType_e blockType;
+    U32 lastBlock;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTD_getcBlockSize() :
+ *  Provides the size of compressed block from block header `src` */
+/* Used by: decompress, fullbench (does not get its definition from here) */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTD_CCOMMON_H_MODULE */
diff --git a/vendor/github.com/DataDog/zstd/zstd_lazy.c b/vendor/github.com/DataDog/zstd/zstd_lazy.c
new file mode 100644
index 0000000..9f15812
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_lazy.c
@@ -0,0 +1,824 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_lazy.h"
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+
+void ZSTD_updateDUBT(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* ip, const BYTE* iend,
+                U32 mls)
+{
+    U32* const hashTable = ms->hashTable;
+    U32  const hashLog = cParams->hashLog;
+
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    if (idx != target)
+        DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)",
+                    idx, target, ms->window.dictLimit);
+    assert(ip + 8 <= iend);   /* condition for ZSTD_hashPtr */
+    (void)iend;
+
+    assert(idx >= ms->window.dictLimit);   /* condition for valid base+idx */
+    for ( ; idx < target ; idx++) {
+        size_t const h  = ZSTD_hashPtr(base + idx, hashLog, mls);   /* assumption : ip + 8 <= iend */
+        U32    const matchIndex = hashTable[h];
+
+        U32*   const nextCandidatePtr = bt + 2*(idx&btMask);
+        U32*   const sortMarkPtr  = nextCandidatePtr + 1;
+
+        DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx);
+        hashTable[h] = idx;   /* Update Hash Table */
+        *nextCandidatePtr = matchIndex;   /* update BT like a chain */
+        *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK;
+    }
+    ms->nextToUpdate = target;
+}
+
+
+/** ZSTD_insertDUBT1() :
+ *  sort one already inserted but unsorted position
+ *  assumption : current >= btlow == (current - btmask)
+ *  doesn't fail */
+static void ZSTD_insertDUBT1(
+                 ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                 U32 current, const BYTE* inputEnd,
+                 U32 nbCompares, U32 btLow, int extDict)
+{
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const ip = (current>=dictLimit) ? base + current : dictBase + current;
+    const BYTE* const iend = (current>=dictLimit) ? inputEnd : dictBase + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 matchIndex = *smallerPtr;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowLow = ms->window.lowLimit;
+
+    DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
+                current, dictLimit, windowLow);
+    assert(current >= btLow);
+    assert(ip < iend);   /* condition for ZSTD_count */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < current);
+
+        if ( (!extDict)
+          || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
+          || (current < dictLimit) /* both in extDict */) {
+            const BYTE* const mBase = !extDict || ((matchIndex+matchLength) >= dictLimit) ? base : dictBase;
+            assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
+                 || (current < dictLimit) );
+            match = mBase + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
+                    current, matchIndex, (U32)matchLength);
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u",
+                        matchIndex, btLow, nextPtr[1]);
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u",
+                        matchIndex, btLow, nextPtr[0]);
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+}
+
+
+static size_t ZSTD_DUBT_findBestMatch (
+                            ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                            const BYTE* const ip, const BYTE* const iend,
+                            size_t* offsetPtr,
+                            U32 const mls,
+                            U32 const extDict)
+{
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32          matchIndex  = hashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    U32    const current = (U32)(ip-base);
+    U32    const windowLow = ms->window.lowLimit;
+
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32    const btLow = (btMask >= current) ? 0 : current - btMask;
+    U32    const unsortLimit = MAX(btLow, windowLow);
+
+    U32*         nextCandidate = bt + 2*(matchIndex&btMask);
+    U32*         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+    U32          nbCompares = 1U << cParams->searchLog;
+    U32          nbCandidates = nbCompares;
+    U32          previousCandidate = 0;
+
+    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", current);
+    assert(ip <= iend-8);   /* required for h calculation */
+
+    /* reach end of unsorted candidates list */
+    while ( (matchIndex > unsortLimit)
+         && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK)
+         && (nbCandidates > 1) ) {
+        DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted",
+                    matchIndex);
+        *unsortedMark = previousCandidate;
+        previousCandidate = matchIndex;
+        matchIndex = *nextCandidate;
+        nextCandidate = bt + 2*(matchIndex&btMask);
+        unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+        nbCandidates --;
+    }
+
+    if ( (matchIndex > unsortLimit)
+      && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) {
+        DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u",
+                    matchIndex);
+        *nextCandidate = *unsortedMark = 0;   /* nullify next candidate if it's still unsorted (note : simplification, detrimental to compression ratio, beneficial for speed) */
+    }
+
+    /* batch sort stacked candidates */
+    matchIndex = previousCandidate;
+    while (matchIndex) {  /* will end on matchIndex == 0 */
+        U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
+        U32 const nextCandidateIdx = *nextCandidateIdxPtr;
+        ZSTD_insertDUBT1(ms, cParams, matchIndex, iend,
+                         nbCandidates, unsortLimit, extDict);
+        matchIndex = nextCandidateIdx;
+        nbCandidates++;
+    }
+
+    /* find longest match */
+    {   size_t commonLengthSmaller=0, commonLengthLarger=0;
+        const BYTE* const dictBase = ms->window.dictBase;
+        const U32 dictLimit = ms->window.dictLimit;
+        const BYTE* const dictEnd = dictBase + dictLimit;
+        const BYTE* const prefixStart = base + dictLimit;
+        U32* smallerPtr = bt + 2*(current&btMask);
+        U32* largerPtr  = bt + 2*(current&btMask) + 1;
+        U32 matchEndIdx = current+8+1;
+        U32 dummy32;   /* to be nullified at the end */
+        size_t bestLength = 0;
+
+        matchIndex  = hashTable[h];
+        hashTable[h] = current;   /* Update Hash Table */
+
+        while (nbCompares-- && (matchIndex > windowLow)) {
+            U32* const nextPtr = bt + 2*(matchIndex & btMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match;
+
+            if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+                match = base + matchIndex;
+                matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+            } else {
+                match = dictBase + matchIndex;
+                matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+                if (matchIndex+matchLength >= dictLimit)
+                    match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+            }
+
+            if (matchLength > bestLength) {
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
+                    bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
+                if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
+
+            if (match[matchLength] < ip[matchLength]) {
+                /* match is smaller than current */
+                *smallerPtr = matchIndex;             /* update smaller idx */
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+                matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                *largerPtr = matchIndex;
+                commonLengthLarger = matchLength;
+                if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                largerPtr = nextPtr;
+                matchIndex = nextPtr[0];
+        }   }
+
+        *smallerPtr = *largerPtr = 0;
+
+        assert(matchEndIdx > current+8); /* ensure nextToUpdate is increased */
+        ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
+        if (bestLength >= MINMATCH) {
+            U32 const mIndex = current - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+            DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                        current, (U32)bestLength, (U32)*offsetPtr, mIndex);
+        }
+        return bestLength;
+    }
+}
+
+
+/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
+static size_t ZSTD_BtFindBestMatch (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls /* template */)
+{
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, cParams, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, cParams, ip, iLimit, offsetPtr, mls, 0);
+}
+
+
+static size_t ZSTD_BtFindBestMatch_selectMLS (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 4);
+    case 5 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 5);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch(ms, cParams, ip, iLimit, offsetPtr, 6);
+    }
+}
+
+
+/** Tree updater, providing best match */
+static size_t ZSTD_BtFindBestMatch_extDict (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls)
+{
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch_extDict");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, cParams, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, cParams, ip, iLimit, offsetPtr, mls, 1);
+}
+
+
+static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 4);
+    case 5 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 5);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch_extDict(ms, cParams, ip, iLimit, offsetPtr, 6);
+    }
+}
+
+
+
+/* *********************************
+*  Hash Chain
+***********************************/
+#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & mask]
+
+/* Update chains up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+static U32 ZSTD_insertAndFindFirstIndex_internal(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, U32 const mls)
+{
+    U32* const hashTable  = ms->hashTable;
+    const U32 hashLog = cParams->hashLog;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainMask = (1 << cParams->chainLog) - 1;
+    const BYTE* const base = ms->window.base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    while(idx < target) { /* catch up */
+        size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
+        NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
+        hashTable[h] = idx;
+        idx++;
+    }
+
+    ms->nextToUpdate = target;
+    return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
+}
+
+U32 ZSTD_insertAndFindFirstIndex(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip)
+{
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, cParams->searchLength);
+}
+
+
+/* inlining is important to hardwire a hot branch (template emulation) */
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_HcFindBestMatch_generic (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 mls, const U32 extDict)
+{
+    U32* const chainTable = ms->chainTable;
+    const U32 chainSize = (1 << cParams->chainLog);
+    const U32 chainMask = chainSize-1;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 lowLimit = ms->window.lowLimit;
+    const U32 current = (U32)(ip-base);
+    const U32 minChain = current > chainSize ? current - chainSize : 0;
+    U32 nbAttempts = 1U << cParams->searchLog;
+    size_t ml=4-1;
+
+    /* HC4 match finder */
+    U32 matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls);
+
+    for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
+        size_t currentMl=0;
+        if ((!extDict) || matchIndex >= dictLimit) {
+            const BYTE* const match = base + matchIndex;
+            if (match[ml] == ip[ml])   /* potentially better */
+                currentMl = ZSTD_count(ip, match, iLimit);
+        } else {
+            const BYTE* const match = dictBase + matchIndex;
+            assert(match+4 <= dictEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = current - matchIndex + ZSTD_REP_MOVE;
+            if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+        }
+
+        if (matchIndex <= minChain) break;
+        matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
+    }
+
+    return ml;
+}
+
+
+FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_selectMLS (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 4, 0);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 5, 0);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 6, 0);
+    }
+}
+
+
+FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* const offsetPtr)
+{
+    switch(cParams->searchLength)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 4, 1);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 5, 1);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, cParams, ip, iLimit, offsetPtr, 6, 1);
+    }
+}
+
+
+/* *******************************
+*  Common parser - lazy strategy
+*********************************/
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_lazy_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        ZSTD_compressionParameters const* cParams,
+                        const void* src, size_t srcSize,
+                        const U32 searchMethod, const U32 depth)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base + ms->window.dictLimit;
+
+    typedef size_t (*searchMax_f)(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
+    U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
+
+    /* init */
+    ip += (ip==base);
+    ms->nextToUpdate3 = ms->nextToUpdate;
+    {   U32 const maxRep = (U32)(ip-base);
+        if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
+    }
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offset=0;
+        const BYTE* start=ip+1;
+
+        /* check repCode */
+        if ((offset_1>0) & (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1))) {
+            /* repcode : we take it */
+            matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            if (depth==0) goto _storeSequence;
+        }
+
+        /* first search (depth 0) */
+        {   size_t offsetFound = 99999999;
+            size_t const ml2 = searchMax(ms, cParams, ip, iend, &offsetFound);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offset=offsetFound;
+        }
+
+        if (matchLength < 4) {
+            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                int const gain2 = (int)(mlRep * 3);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                if ((mlRep >= 4) && (gain2 > gain1))
+                    matchLength = mlRep, offset = 0, start = ip;
+            }
+            {   size_t offset2=99999999;
+                size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offset = offset2, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const ml2 = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(ml2 * 4);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((ml2 >= 4) && (gain2 > gain1))
+                        matchLength = ml2, offset = 0, start = ip;
+                }
+                {   size_t offset2=99999999;
+                    size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offset = offset2, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* NOTE:
+         * start[-offset+ZSTD_REP_MOVE-1] is undefined behavior.
+         * (-offset+ZSTD_REP_MOVE-1) is unsigned, and is added to start, which
+         * overflows the pointer, which is undefined behavior.
+         */
+        /* catch up */
+        if (offset) {
+            while ( ((start > anchor) & (start - (offset-ZSTD_REP_MOVE) > base))
+                 && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
+                { start--; matchLength++; }
+            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+        }
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = start - anchor;
+            ZSTD_storeSeq(seqStore, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            anchor = ip = start + matchLength;
+        }
+
+        /* check immediate repcode */
+        while ( ((ip <= ilimit) & (offset_2>0))
+             && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
+            /* store sequence */
+            matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+            offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
+            ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
+            ip += matchLength;
+            anchor = ip;
+            continue;   /* faster when present ... (?) */
+    }   }
+
+    /* Save reps for next block */
+    rep[0] = offset_1 ? offset_1 : savedOffset;
+    rep[1] = offset_2 ? offset_2 : savedOffset;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 1, 2);
+}
+
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 2);
+}
+
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 1);
+}
+
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 0);
+}
+
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_lazy_extDict_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        ZSTD_compressionParameters const* cParams,
+                        const void* src, size_t srcSize,
+                        const U32 searchMethod, const U32 depth)
+{
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const U32 dictLimit = ms->window.dictLimit;
+    const U32 lowestIndex = ms->window.lowLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictEnd  = dictBase + dictLimit;
+    const BYTE* const dictStart  = dictBase + lowestIndex;
+
+    typedef size_t (*searchMax_f)(
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
+
+    U32 offset_1 = rep[0], offset_2 = rep[1];
+
+    /* init */
+    ms->nextToUpdate3 = ms->nextToUpdate;
+    ip += (ip == prefixStart);
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offset=0;
+        const BYTE* start=ip+1;
+        U32 current = (U32)(ip-base);
+
+        /* check repCode */
+        {   const U32 repIndex = (U32)(current+1 - offset_1);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))   /* intentional overflow */
+            if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                if (depth==0) goto _storeSequence;
+        }   }
+
+        /* first search (depth 0) */
+        {   size_t offsetFound = 99999999;
+            size_t const ml2 = searchMax(ms, cParams, ip, iend, &offsetFound);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offset=offsetFound;
+        }
+
+         if (matchLength < 4) {
+            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            current++;
+            /* check repCode */
+            if (offset) {
+                const U32 repIndex = (U32)(current - offset_1);
+                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                const BYTE* const repMatch = repBase + repIndex;
+                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                    /* repcode detected */
+                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                    size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                    int const gain2 = (int)(repLength * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((repLength >= 4) && (gain2 > gain1))
+                        matchLength = repLength, offset = 0, start = ip;
+            }   }
+
+            /* search match, depth 1 */
+            {   size_t offset2=99999999;
+                size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offset = offset2, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                current++;
+                /* check repCode */
+                if (offset) {
+                    const U32 repIndex = (U32)(current - offset_1);
+                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                    const BYTE* const repMatch = repBase + repIndex;
+                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                    if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                        /* repcode detected */
+                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                        size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                        int const gain2 = (int)(repLength * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        if ((repLength >= 4) && (gain2 > gain1))
+                            matchLength = repLength, offset = 0, start = ip;
+                }   }
+
+                /* search match, depth 2 */
+                {   size_t offset2=99999999;
+                    size_t const ml2 = searchMax(ms, cParams, ip, iend, &offset2);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offset = offset2, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* catch up */
+        if (offset) {
+            U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+            const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
+            const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
+            while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+        }
+
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = start - anchor;
+            ZSTD_storeSeq(seqStore, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            anchor = ip = start + matchLength;
+        }
+
+        /* check immediate repcode */
+        while (ip <= ilimit) {
+            const U32 repIndex = (U32)((ip-base) - offset_2);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+            if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset history */
+                ZSTD_storeSeq(seqStore, 0, anchor, 0, matchLength-MINMATCH);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+            break;
+    }   }
+
+    /* Save reps for next block */
+    rep[0] = offset_1;
+    rep[1] = offset_2;
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 0);
+}
+
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 1);
+}
+
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 0, 2);
+}
+
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+
+{
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, cParams, src, srcSize, 1, 2);
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_lazy.h b/vendor/github.com/DataDog/zstd/zstd_lazy.h
new file mode 100644
index 0000000..bda064f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_lazy.h
@@ -0,0 +1,56 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LAZY_H
+#define ZSTD_LAZY_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"
+
+U32 ZSTD_insertAndFindFirstIndex(
+        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+        const BYTE* ip);
+
+void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). pre-emptively increase value of ZSTD_DUBT_UNSORTED_MARK */
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_LAZY_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_ldm.c b/vendor/github.com/DataDog/zstd/zstd_ldm.c
new file mode 100644
index 0000000..bffd8a3
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_ldm.c
@@ -0,0 +1,653 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ */
+
+#include "zstd_ldm.h"
+
+#include "zstd_fast.h"          /* ZSTD_fillHashTable() */
+#include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
+
+#define LDM_BUCKET_SIZE_LOG 3
+#define LDM_MIN_MATCH_LENGTH 64
+#define LDM_HASH_RLOG 7
+#define LDM_HASH_CHAR_OFFSET 10
+
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams)
+{
+    U32 const windowLog = cParams->windowLog;
+    ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
+    DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
+    if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
+    if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
+    if (cParams->strategy >= ZSTD_btopt) {
+      /* Get out of the way of the optimal parser */
+      U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
+      assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
+      assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
+      params->minMatchLength = minMatch;
+    }
+    if (params->hashLog == 0) {
+        params->hashLog = MAX(ZSTD_HASHLOG_MIN, windowLog - LDM_HASH_RLOG);
+        assert(params->hashLog <= ZSTD_HASHLOG_MAX);
+    }
+    if (params->hashEveryLog == 0) {
+        params->hashEveryLog =
+                windowLog < params->hashLog ? 0 : windowLog - params->hashLog;
+    }
+    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
+}
+
+size_t ZSTD_ldm_getTableSize(ldmParams_t params)
+{
+    size_t const ldmHSize = ((size_t)1) << params.hashLog;
+    size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
+    size_t const ldmBucketSize =
+        ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
+    size_t const totalSize = ldmBucketSize + ldmHSize * sizeof(ldmEntry_t);
+    return params.enableLdm ? totalSize : 0;
+}
+
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
+{
+    return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
+}
+
+/** ZSTD_ldm_getSmallHash() :
+ *  numBits should be <= 32
+ *  If numBits==0, returns 0.
+ *  @return : the most significant numBits of value. */
+static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
+{
+    assert(numBits <= 32);
+    return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
+}
+
+/** ZSTD_ldm_getChecksum() :
+ *  numBitsToDiscard should be <= 32
+ *  @return : the next most significant 32 bits after numBitsToDiscard */
+static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
+{
+    assert(numBitsToDiscard <= 32);
+    return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
+}
+
+/** ZSTD_ldm_getTag() ;
+ *  Given the hash, returns the most significant numTagBits bits
+ *  after (32 + hbits) bits.
+ *
+ *  If there are not enough bits remaining, return the last
+ *  numTagBits bits. */
+static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
+{
+    assert(numTagBits < 32 && hbits <= 32);
+    if (32 - hbits < numTagBits) {
+        return hash & (((U32)1 << numTagBits) - 1);
+    } else {
+        return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
+    }
+}
+
+/** ZSTD_ldm_getBucket() :
+ *  Returns a pointer to the start of the bucket associated with hash. */
+static ldmEntry_t* ZSTD_ldm_getBucket(
+        ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
+{
+    return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
+}
+
+/** ZSTD_ldm_insertEntry() :
+ *  Insert the entry with corresponding hash into the hash table */
+static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
+                                 size_t const hash, const ldmEntry_t entry,
+                                 ldmParams_t const ldmParams)
+{
+    BYTE* const bucketOffsets = ldmState->bucketOffsets;
+    *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
+    bucketOffsets[hash]++;
+    bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
+}
+
+/** ZSTD_ldm_makeEntryAndInsertByTag() :
+ *
+ *  Gets the small hash, checksum, and tag from the rollingHash.
+ *
+ *  If the tag matches (1 << ldmParams.hashEveryLog)-1, then
+ *  creates an ldmEntry from the offset, and inserts it into the hash table.
+ *
+ *  hBits is the length of the small hash, which is the most significant hBits
+ *  of rollingHash. The checksum is the next 32 most significant bits, followed
+ *  by ldmParams.hashEveryLog bits that make up the tag. */
+static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
+                                             U64 const rollingHash,
+                                             U32 const hBits,
+                                             U32 const offset,
+                                             ldmParams_t const ldmParams)
+{
+    U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog);
+    U32 const tagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
+    if (tag == tagMask) {
+        U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
+        U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
+        ldmEntry_t entry;
+        entry.offset = offset;
+        entry.checksum = checksum;
+        ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
+    }
+}
+
+/** ZSTD_ldm_getRollingHash() :
+ *  Get a 64-bit hash using the first len bytes from buf.
+ *
+ *  Giving bytes s = s_1, s_2, ... s_k, the hash is defined to be
+ *  H(s) = s_1*(a^(k-1)) + s_2*(a^(k-2)) + ... + s_k*(a^0)
+ *
+ *  where the constant a is defined to be prime8bytes.
+ *
+ *  The implementation adds an offset to each byte, so
+ *  H(s) = (s_1 + HASH_CHAR_OFFSET)*(a^(k-1)) + ... */
+static U64 ZSTD_ldm_getRollingHash(const BYTE* buf, U32 len)
+{
+    U64 ret = 0;
+    U32 i;
+    for (i = 0; i < len; i++) {
+        ret *= prime8bytes;
+        ret += buf[i] + LDM_HASH_CHAR_OFFSET;
+    }
+    return ret;
+}
+
+/** ZSTD_ldm_ipow() :
+ *  Return base^exp. */
+static U64 ZSTD_ldm_ipow(U64 base, U64 exp)
+{
+    U64 ret = 1;
+    while (exp) {
+        if (exp & 1) { ret *= base; }
+        exp >>= 1;
+        base *= base;
+    }
+    return ret;
+}
+
+U64 ZSTD_ldm_getHashPower(U32 minMatchLength) {
+    DEBUGLOG(4, "ZSTD_ldm_getHashPower: mml=%u", minMatchLength);
+    assert(minMatchLength >= ZSTD_LDM_MINMATCH_MIN);
+    return ZSTD_ldm_ipow(prime8bytes, minMatchLength - 1);
+}
+
+/** ZSTD_ldm_updateHash() :
+ *  Updates hash by removing toRemove and adding toAdd. */
+static U64 ZSTD_ldm_updateHash(U64 hash, BYTE toRemove, BYTE toAdd, U64 hashPower)
+{
+    hash -= ((toRemove + LDM_HASH_CHAR_OFFSET) * hashPower);
+    hash *= prime8bytes;
+    hash += toAdd + LDM_HASH_CHAR_OFFSET;
+    return hash;
+}
+
+/** ZSTD_ldm_countBackwardsMatch() :
+ *  Returns the number of bytes that match backwards before pIn and pMatch.
+ *
+ *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
+static size_t ZSTD_ldm_countBackwardsMatch(
+            const BYTE* pIn, const BYTE* pAnchor,
+            const BYTE* pMatch, const BYTE* pBase)
+{
+    size_t matchLength = 0;
+    while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
+        pIn--;
+        pMatch--;
+        matchLength++;
+    }
+    return matchLength;
+}
+
+/** ZSTD_ldm_fillFastTables() :
+ *
+ *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
+ *  This is similar to ZSTD_loadDictionaryContent.
+ *
+ *  The tables for the other strategies are filled within their
+ *  block compressors. */
+static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
+                                      ZSTD_compressionParameters const* cParams,
+                                      void const* end)
+{
+    const BYTE* const iend = (const BYTE*)end;
+
+    switch(cParams->strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable(ms, cParams, iend);
+        ms->nextToUpdate = (U32)(iend - ms->window.base);
+        break;
+
+    case ZSTD_dfast:
+        ZSTD_fillDoubleHashTable(ms, cParams, iend);
+        ms->nextToUpdate = (U32)(iend - ms->window.base);
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+    case ZSTD_btlazy2:
+    case ZSTD_btopt:
+    case ZSTD_btultra:
+        break;
+    default:
+        assert(0);  /* not possible : not a valid strategy id */
+    }
+
+    return 0;
+}
+
+/** ZSTD_ldm_fillLdmHashTable() :
+ *
+ *  Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
+ *  lastHash is the rolling hash that corresponds to lastHashed.
+ *
+ *  Returns the rolling hash corresponding to position iend-1. */
+static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
+                                     U64 lastHash, const BYTE* lastHashed,
+                                     const BYTE* iend, const BYTE* base,
+                                     U32 hBits, ldmParams_t const ldmParams)
+{
+    U64 rollingHash = lastHash;
+    const BYTE* cur = lastHashed + 1;
+
+    while (cur < iend) {
+        rollingHash = ZSTD_ldm_updateHash(rollingHash, cur[-1],
+                                          cur[ldmParams.minMatchLength-1],
+                                          state->hashPower);
+        ZSTD_ldm_makeEntryAndInsertByTag(state,
+                                         rollingHash, hBits,
+                                         (U32)(cur - base), ldmParams);
+        ++cur;
+    }
+    return rollingHash;
+}
+
+
+/** ZSTD_ldm_limitTableUpdate() :
+ *
+ *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
+ *  if it is far way
+ *  (after a long match, only update tables a limited amount). */
+static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
+{
+    U32 const current = (U32)(anchor - ms->window.base);
+    if (current > ms->nextToUpdate + 1024) {
+        ms->nextToUpdate =
+            current - MIN(512, current - ms->nextToUpdate - 1024);
+    }
+}
+
+static size_t ZSTD_ldm_generateSequences_internal(
+        ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    /* LDM parameters */
+    int const extDict = ZSTD_window_hasExtDict(ldmState->window);
+    U32 const minMatchLength = params->minMatchLength;
+    U64 const hashPower = ldmState->hashPower;
+    U32 const hBits = params->hashLog - params->bucketSizeLog;
+    U32 const ldmBucketSize = 1U << params->bucketSizeLog;
+    U32 const hashEveryLog = params->hashEveryLog;
+    U32 const ldmTagMask = (1U << params->hashEveryLog) - 1;
+    /* Prefix and extDict parameters */
+    U32 const dictLimit = ldmState->window.dictLimit;
+    U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
+    BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
+    BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
+    BYTE const* const lowPrefixPtr = base + dictLimit;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
+    /* Input positions */
+    BYTE const* anchor = istart;
+    BYTE const* ip = istart;
+    /* Rolling hash */
+    BYTE const* lastHashed = NULL;
+    U64 rollingHash = 0;
+
+    while (ip <= ilimit) {
+        size_t mLength;
+        U32 const current = (U32)(ip - base);
+        size_t forwardMatchLength = 0, backwardMatchLength = 0;
+        ldmEntry_t* bestEntry = NULL;
+        if (ip != istart) {
+            rollingHash = ZSTD_ldm_updateHash(rollingHash, lastHashed[0],
+                                              lastHashed[minMatchLength],
+                                              hashPower);
+        } else {
+            rollingHash = ZSTD_ldm_getRollingHash(ip, minMatchLength);
+        }
+        lastHashed = ip;
+
+        /* Do not insert and do not look for a match */
+        if (ZSTD_ldm_getTag(rollingHash, hBits, hashEveryLog) != ldmTagMask) {
+           ip++;
+           continue;
+        }
+
+        /* Get the best entry and compute the match lengths */
+        {
+            ldmEntry_t* const bucket =
+                ZSTD_ldm_getBucket(ldmState,
+                                   ZSTD_ldm_getSmallHash(rollingHash, hBits),
+                                   *params);
+            ldmEntry_t* cur;
+            size_t bestMatchLength = 0;
+            U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
+
+            for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
+                size_t curForwardMatchLength, curBackwardMatchLength,
+                       curTotalMatchLength;
+                if (cur->checksum != checksum || cur->offset <= lowestIndex) {
+                    continue;
+                }
+                if (extDict) {
+                    BYTE const* const curMatchBase =
+                        cur->offset < dictLimit ? dictBase : base;
+                    BYTE const* const pMatch = curMatchBase + cur->offset;
+                    BYTE const* const matchEnd =
+                        cur->offset < dictLimit ? dictEnd : iend;
+                    BYTE const* const lowMatchPtr =
+                        cur->offset < dictLimit ? dictStart : lowPrefixPtr;
+
+                    curForwardMatchLength = ZSTD_count_2segments(
+                                                ip, pMatch, iend,
+                                                matchEnd, lowPrefixPtr);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
+                                                     lowMatchPtr);
+                    curTotalMatchLength = curForwardMatchLength +
+                                          curBackwardMatchLength;
+                } else { /* !extDict */
+                    BYTE const* const pMatch = base + cur->offset;
+                    curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
+                                                     lowPrefixPtr);
+                    curTotalMatchLength = curForwardMatchLength +
+                                          curBackwardMatchLength;
+                }
+
+                if (curTotalMatchLength > bestMatchLength) {
+                    bestMatchLength = curTotalMatchLength;
+                    forwardMatchLength = curForwardMatchLength;
+                    backwardMatchLength = curBackwardMatchLength;
+                    bestEntry = cur;
+                }
+            }
+        }
+
+        /* No match found -- continue searching */
+        if (bestEntry == NULL) {
+            ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
+                                             hBits, current,
+                                             *params);
+            ip++;
+            continue;
+        }
+
+        /* Match found */
+        mLength = forwardMatchLength + backwardMatchLength;
+        ip -= backwardMatchLength;
+
+        {
+            /* Store the sequence:
+             * ip = current - backwardMatchLength
+             * The match is at (bestEntry->offset - backwardMatchLength)
+             */
+            U32 const matchIndex = bestEntry->offset;
+            U32 const offset = current - matchIndex;
+            rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
+
+            /* Out of sequence storage */
+            if (rawSeqStore->size == rawSeqStore->capacity)
+                return ERROR(dstSize_tooSmall);
+            seq->litLength = (U32)(ip - anchor);
+            seq->matchLength = (U32)mLength;
+            seq->offset = offset;
+            rawSeqStore->size++;
+        }
+
+        /* Insert the current entry into the hash table */
+        ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
+                                         (U32)(lastHashed - base),
+                                         *params);
+
+        assert(ip + backwardMatchLength == lastHashed);
+
+        /* Fill the hash table from lastHashed+1 to ip+mLength*/
+        /* Heuristic: don't need to fill the entire table at end of block */
+        if (ip + mLength <= ilimit) {
+            rollingHash = ZSTD_ldm_fillLdmHashTable(
+                              ldmState, rollingHash, lastHashed,
+                              ip + mLength, base, hBits, *params);
+            lastHashed = ip + mLength - 1;
+        }
+        ip += mLength;
+        anchor = ip;
+    }
+    return iend - anchor;
+}
+
+/*! ZSTD_ldm_reduceTable() :
+ *  reduce table indexes by `reducerValue` */
+static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
+                                 U32 const reducerValue)
+{
+    U32 u;
+    for (u = 0; u < size; u++) {
+        if (table[u].offset < reducerValue) table[u].offset = 0;
+        else table[u].offset -= reducerValue;
+    }
+}
+
+size_t ZSTD_ldm_generateSequences(
+        ldmState_t* ldmState, rawSeqStore_t* sequences,
+        ldmParams_t const* params, void const* src, size_t srcSize)
+{
+    U32 const maxDist = 1U << params->windowLog;
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    size_t const kMaxChunkSize = 1 << 20;
+    size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
+    size_t chunk;
+    size_t leftoverSize = 0;
+
+    assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
+    /* Check that ZSTD_window_update() has been called for this chunk prior
+     * to passing it to this function.
+     */
+    assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
+    /* The input could be very large (in zstdmt), so it must be broken up into
+     * chunks to enforce the maximmum distance and handle overflow correction.
+     */
+    assert(sequences->pos <= sequences->size);
+    assert(sequences->size <= sequences->capacity);
+    for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
+        BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
+        size_t const remaining = (size_t)(iend - chunkStart);
+        BYTE const *const chunkEnd =
+            (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
+        size_t const chunkSize = chunkEnd - chunkStart;
+        size_t newLeftoverSize;
+        size_t const prevSize = sequences->size;
+
+        assert(chunkStart < iend);
+        /* 1. Perform overflow correction if necessary. */
+        if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
+            U32 const ldmHSize = 1U << params->hashLog;
+            U32 const correction = ZSTD_window_correctOverflow(
+                &ldmState->window, /* cycleLog */ 0, maxDist, src);
+            ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
+        }
+        /* 2. We enforce the maximum offset allowed.
+         *
+         * kMaxChunkSize should be small enough that we don't lose too much of
+         * the window through early invalidation.
+         * TODO: * Test the chunk size.
+         *       * Try invalidation after the sequence generation and test the
+         *         the offset against maxDist directly.
+         */
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL);
+        /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
+        newLeftoverSize = ZSTD_ldm_generateSequences_internal(
+            ldmState, sequences, params, chunkStart, chunkSize);
+        if (ZSTD_isError(newLeftoverSize))
+            return newLeftoverSize;
+        /* 4. We add the leftover literals from previous iterations to the first
+         *    newly generated sequence, or add the `newLeftoverSize` if none are
+         *    generated.
+         */
+        /* Prepend the leftover literals from the last call */
+        if (prevSize < sequences->size) {
+            sequences->seq[prevSize].litLength += (U32)leftoverSize;
+            leftoverSize = newLeftoverSize;
+        } else {
+            assert(newLeftoverSize == chunkSize);
+            leftoverSize += chunkSize;
+        }
+    }
+    return 0;
+}
+
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
+    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
+        if (srcSize <= seq->litLength) {
+            /* Skip past srcSize literals */
+            seq->litLength -= (U32)srcSize;
+            return;
+        }
+        srcSize -= seq->litLength;
+        seq->litLength = 0;
+        if (srcSize < seq->matchLength) {
+            /* Skip past the first srcSize of the match */
+            seq->matchLength -= (U32)srcSize;
+            if (seq->matchLength < minMatch) {
+                /* The match is too short, omit it */
+                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
+                    seq[1].litLength += seq[0].matchLength;
+                }
+                rawSeqStore->pos++;
+            }
+            return;
+        }
+        srcSize -= seq->matchLength;
+        seq->matchLength = 0;
+        rawSeqStore->pos++;
+    }
+}
+
+/**
+ * If the sequence length is longer than remaining then the sequence is split
+ * between this block and the next.
+ *
+ * Returns the current sequence to handle, or if the rest of the block should
+ * be literals, it returns a sequence with offset == 0.
+ */
+static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
+                                 U32 const remaining, U32 const minMatch)
+{
+    rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
+    assert(sequence.offset > 0);
+    /* Likely: No partial sequence */
+    if (remaining >= sequence.litLength + sequence.matchLength) {
+        rawSeqStore->pos++;
+        return sequence;
+    }
+    /* Cut the sequence short (offset == 0 ==> rest is literals). */
+    if (remaining <= sequence.litLength) {
+        sequence.offset = 0;
+    } else if (remaining < sequence.litLength + sequence.matchLength) {
+        sequence.matchLength = remaining - sequence.litLength;
+        if (sequence.matchLength < minMatch) {
+            sequence.offset = 0;
+        }
+    }
+    /* Skip past `remaining` bytes for the future sequences. */
+    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
+    return sequence;
+}
+
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+    ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+    ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
+    int const extDict)
+{
+    unsigned const minMatch = cParams->searchLength;
+    ZSTD_blockCompressor const blockCompressor =
+        ZSTD_selectBlockCompressor(cParams->strategy, extDict);
+    BYTE const* const base = ms->window.base;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    /* Input positions */
+    BYTE const* ip = istart;
+
+    assert(rawSeqStore->pos <= rawSeqStore->size);
+    assert(rawSeqStore->size <= rawSeqStore->capacity);
+    /* Loop through each sequence and apply the block compressor to the lits */
+    while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
+        /* maybeSplitSequence updates rawSeqStore->pos */
+        rawSeq const sequence = maybeSplitSequence(rawSeqStore,
+                                                   (U32)(iend - ip), minMatch);
+        int i;
+        /* End signal */
+        if (sequence.offset == 0)
+            break;
+
+        assert(sequence.offset <= (1U << cParams->windowLog));
+        assert(ip + sequence.litLength + sequence.matchLength <= iend);
+
+        /* Fill tables for block compressor */
+        ZSTD_ldm_limitTableUpdate(ms, ip);
+        ZSTD_ldm_fillFastTables(ms, cParams, ip);
+        /* Run the block compressor */
+        {
+            size_t const newLitLength =
+                blockCompressor(ms, seqStore, rep, cParams, ip,
+                                sequence.litLength);
+            ip += sequence.litLength;
+            ms->nextToUpdate = (U32)(ip - base);
+            /* Update the repcodes */
+            for (i = ZSTD_REP_NUM - 1; i > 0; i--)
+                rep[i] = rep[i-1];
+            rep[0] = sequence.offset;
+            /* Store the sequence */
+            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength,
+                          sequence.offset + ZSTD_REP_MOVE,
+                          sequence.matchLength - MINMATCH);
+            ip += sequence.matchLength;
+        }
+    }
+    /* Fill the tables for the block compressor */
+    ZSTD_ldm_limitTableUpdate(ms, ip);
+    ZSTD_ldm_fillFastTables(ms, cParams, ip);
+    /* Compress the last literals */
+    {
+        size_t const lastLiterals = blockCompressor(ms, seqStore, rep, cParams,
+                                                    ip, iend - ip);
+        ms->nextToUpdate = (U32)(iend - base);
+        return lastLiterals;
+    }
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_ldm.h b/vendor/github.com/DataDog/zstd/zstd_ldm.h
new file mode 100644
index 0000000..0c3789f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_ldm.h
@@ -0,0 +1,111 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ */
+
+#ifndef ZSTD_LDM_H
+#define ZSTD_LDM_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"   /* ldmParams_t, U32 */
+#include "zstd.h"   /* ZSTD_CCtx, size_t */
+
+/*-*************************************
+*  Long distance matching
+***************************************/
+
+#define ZSTD_LDM_DEFAULT_WINDOW_LOG ZSTD_WINDOWLOG_DEFAULTMAX
+
+/**
+ * ZSTD_ldm_generateSequences():
+ *
+ * Generates the sequences using the long distance match finder.
+ * Generates long range matching sequences in `sequences`, which parse a prefix
+ * of the source. `sequences` must be large enough to store every sequence,
+ * which can be checked with `ZSTD_ldm_getMaxNbSeq()`.
+ * @returns 0 or an error code.
+ *
+ * NOTE: The user must have called ZSTD_window_update() for all of the input
+ * they have, even if they pass it to ZSTD_ldm_generateSequences() in chunks.
+ * NOTE: This function returns an error if it runs out of space to store
+ *       sequences.
+ */
+size_t ZSTD_ldm_generateSequences(
+            ldmState_t* ldms, rawSeqStore_t* sequences,
+            ldmParams_t const* params, void const* src, size_t srcSize);
+
+/**
+ * ZSTD_ldm_blockCompress():
+ *
+ * Compresses a block using the predefined sequences, along with a secondary
+ * block compressor. The literals section of every sequence is passed to the
+ * secondary block compressor, and those sequences are interspersed with the
+ * predefined sequences. Returns the length of the last literals.
+ * Updates `rawSeqStore.pos` to indicate how many sequences have been consumed.
+ * `rawSeqStore.seq` may also be updated to split the last sequence between two
+ * blocks.
+ * @return The length of the last literals.
+ *
+ * NOTE: The source must be at most the maximum block size, but the predefined
+ * sequences can be any size, and may be longer than the block. In the case that
+ * they are longer than the block, the last sequences may need to be split into
+ * two. We handle that case correctly, and update `rawSeqStore` appropriately.
+ * NOTE: This function does not return any errors.
+ */
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+            ZSTD_compressionParameters const* cParams,
+            void const* src, size_t srcSize,
+            int const extDict);
+
+/**
+ * ZSTD_ldm_skipSequences():
+ *
+ * Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
+ * Avoids emitting matches less than `minMatch` bytes.
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
+    U32 const minMatch);
+
+
+/** ZSTD_ldm_getTableSize() :
+ *  Estimate the space needed for long distance matching tables or 0 if LDM is
+ *  disabled.
+ */
+size_t ZSTD_ldm_getTableSize(ldmParams_t params);
+
+/** ZSTD_ldm_getSeqSpace() :
+ *  Return an upper bound on the number of sequences that can be produced by
+ *  the long distance matcher, or 0 if LDM is disabled.
+ */
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize);
+
+/** ZSTD_ldm_getTableSize() :
+ *  Return prime8bytes^(minMatchLength-1) */
+U64 ZSTD_ldm_getHashPower(U32 minMatchLength);
+
+/** ZSTD_ldm_adjustParameters() :
+ *  If the params->hashEveryLog is not set, set it to its default value based on
+ *  windowLog and params->hashLog.
+ *
+ *  Ensures that params->bucketSizeLog is <= params->hashLog (setting it to
+ *  params->hashLog if it is not).
+ *
+ *  Ensures that the minMatchLength >= targetLength during optimal parsing.
+ */
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_FAST_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_legacy.h b/vendor/github.com/DataDog/zstd/zstd_legacy.h
new file mode 100644
index 0000000..5893cb9
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_legacy.h
@@ -0,0 +1,381 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LEGACY_H
+#define ZSTD_LEGACY_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include "mem.h"            /* MEM_STATIC */
+#include "error_private.h"  /* ERROR */
+#include "zstd.h"           /* ZSTD_inBuffer, ZSTD_outBuffer */
+
+#if !defined (ZSTD_LEGACY_SUPPORT) || (ZSTD_LEGACY_SUPPORT == 0)
+#  undef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 8
+#endif
+
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+#  include "zstd_v01.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+#  include "zstd_v02.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+#  include "zstd_v03.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+#  include "zstd_v04.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+#  include "zstd_v05.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+#  include "zstd_v06.h"
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+#  include "zstd_v07.h"
+#endif
+
+/** ZSTD_isLegacy() :
+    @return : > 0 if supported by legacy decoder. 0 otherwise.
+              return value is the version.
+*/
+MEM_STATIC unsigned ZSTD_isLegacy(const void* src, size_t srcSize)
+{
+    U32 magicNumberLE;
+    if (srcSize<4) return 0;
+    magicNumberLE = MEM_readLE32(src);
+    switch(magicNumberLE)
+    {
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+        case ZSTDv01_magicNumberLE:return 1;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+        case ZSTDv02_magicNumber : return 2;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+        case ZSTDv03_magicNumber : return 3;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case ZSTDv04_magicNumber : return 4;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case ZSTDv05_MAGICNUMBER : return 5;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case ZSTDv06_MAGICNUMBER : return 6;
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case ZSTDv07_MAGICNUMBER : return 7;
+#endif
+        default : return 0;
+    }
+}
+
+
+MEM_STATIC unsigned long long ZSTD_getDecompressedSize_legacy(const void* src, size_t srcSize)
+{
+    U32 const version = ZSTD_isLegacy(src, srcSize);
+    if (version < 5) return 0;  /* no decompressed size in frame header, or not a legacy format */
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+    if (version==5) {
+        ZSTDv05_parameters fParams;
+        size_t const frResult = ZSTDv05_getFrameParams(&fParams, src, srcSize);
+        if (frResult != 0) return 0;
+        return fParams.srcSize;
+    }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+    if (version==6) {
+        ZSTDv06_frameParams fParams;
+        size_t const frResult = ZSTDv06_getFrameParams(&fParams, src, srcSize);
+        if (frResult != 0) return 0;
+        return fParams.frameContentSize;
+    }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+    if (version==7) {
+        ZSTDv07_frameParams fParams;
+        size_t const frResult = ZSTDv07_getFrameParams(&fParams, src, srcSize);
+        if (frResult != 0) return 0;
+        return fParams.frameContentSize;
+    }
+#endif
+    return 0;   /* should not be possible */
+}
+
+
+MEM_STATIC size_t ZSTD_decompressLegacy(
+                     void* dst, size_t dstCapacity,
+               const void* src, size_t compressedSize,
+               const void* dict,size_t dictSize)
+{
+    U32 const version = ZSTD_isLegacy(src, compressedSize);
+    (void)dst; (void)dstCapacity; (void)dict; (void)dictSize;  /* unused when ZSTD_LEGACY_SUPPORT >= 8 */
+    switch(version)
+    {
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+        case 1 :
+            return ZSTDv01_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+        case 2 :
+            return ZSTDv02_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+        case 3 :
+            return ZSTDv03_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+            return ZSTDv04_decompress(dst, dstCapacity, src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+            {   size_t result;
+                ZSTDv05_DCtx* const zd = ZSTDv05_createDCtx();
+                if (zd==NULL) return ERROR(memory_allocation);
+                result = ZSTDv05_decompress_usingDict(zd, dst, dstCapacity, src, compressedSize, dict, dictSize);
+                ZSTDv05_freeDCtx(zd);
+                return result;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+            {   size_t result;
+                ZSTDv06_DCtx* const zd = ZSTDv06_createDCtx();
+                if (zd==NULL) return ERROR(memory_allocation);
+                result = ZSTDv06_decompress_usingDict(zd, dst, dstCapacity, src, compressedSize, dict, dictSize);
+                ZSTDv06_freeDCtx(zd);
+                return result;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+            {   size_t result;
+                ZSTDv07_DCtx* const zd = ZSTDv07_createDCtx();
+                if (zd==NULL) return ERROR(memory_allocation);
+                result = ZSTDv07_decompress_usingDict(zd, dst, dstCapacity, src, compressedSize, dict, dictSize);
+                ZSTDv07_freeDCtx(zd);
+                return result;
+            }
+#endif
+        default :
+            return ERROR(prefix_unknown);
+    }
+}
+
+MEM_STATIC size_t ZSTD_findFrameCompressedSizeLegacy(const void *src,
+                                             size_t compressedSize)
+{
+    U32 const version = ZSTD_isLegacy(src, compressedSize);
+    switch(version)
+    {
+#if (ZSTD_LEGACY_SUPPORT <= 1)
+        case 1 :
+            return ZSTDv01_findFrameCompressedSize(src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 2)
+        case 2 :
+            return ZSTDv02_findFrameCompressedSize(src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 3)
+        case 3 :
+            return ZSTDv03_findFrameCompressedSize(src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+            return ZSTDv04_findFrameCompressedSize(src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+            return ZSTDv05_findFrameCompressedSize(src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+            return ZSTDv06_findFrameCompressedSize(src, compressedSize);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+            return ZSTDv07_findFrameCompressedSize(src, compressedSize);
+#endif
+        default :
+            return ERROR(prefix_unknown);
+    }
+}
+
+MEM_STATIC size_t ZSTD_freeLegacyStreamContext(void* legacyContext, U32 version)
+{
+    switch(version)
+    {
+        default :
+        case 1 :
+        case 2 :
+        case 3 :
+            (void)legacyContext;
+            return ERROR(version_unsupported);
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 : return ZBUFFv04_freeDCtx((ZBUFFv04_DCtx*)legacyContext);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 : return ZBUFFv05_freeDCtx((ZBUFFv05_DCtx*)legacyContext);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 : return ZBUFFv06_freeDCtx((ZBUFFv06_DCtx*)legacyContext);
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 : return ZBUFFv07_freeDCtx((ZBUFFv07_DCtx*)legacyContext);
+#endif
+    }
+}
+
+
+MEM_STATIC size_t ZSTD_initLegacyStream(void** legacyContext, U32 prevVersion, U32 newVersion,
+                                        const void* dict, size_t dictSize)
+{
+    DEBUGLOG(5, "ZSTD_initLegacyStream for v0.%u", newVersion);
+    if (prevVersion != newVersion) ZSTD_freeLegacyStreamContext(*legacyContext, prevVersion);
+    switch(newVersion)
+    {
+        default :
+        case 1 :
+        case 2 :
+        case 3 :
+            (void)dict; (void)dictSize;
+            return 0;
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+        {
+            ZBUFFv04_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv04_createDCtx() : (ZBUFFv04_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv04_decompressInit(dctx);
+            ZBUFFv04_decompressWithDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+        {
+            ZBUFFv05_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv05_createDCtx() : (ZBUFFv05_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv05_decompressInitDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+        {
+            ZBUFFv06_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv06_createDCtx() : (ZBUFFv06_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv06_decompressInitDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+        {
+            ZBUFFv07_DCtx* dctx = (prevVersion != newVersion) ? ZBUFFv07_createDCtx() : (ZBUFFv07_DCtx*)*legacyContext;
+            if (dctx==NULL) return ERROR(memory_allocation);
+            ZBUFFv07_decompressInitDictionary(dctx, dict, dictSize);
+            *legacyContext = dctx;
+            return 0;
+        }
+#endif
+    }
+}
+
+
+
+MEM_STATIC size_t ZSTD_decompressLegacyStream(void* legacyContext, U32 version,
+                                              ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    DEBUGLOG(5, "ZSTD_decompressLegacyStream for v0.%u", version);
+    switch(version)
+    {
+        default :
+        case 1 :
+        case 2 :
+        case 3 :
+            (void)legacyContext; (void)output; (void)input;
+            return ERROR(version_unsupported);
+#if (ZSTD_LEGACY_SUPPORT <= 4)
+        case 4 :
+            {
+                ZBUFFv04_DCtx* dctx = (ZBUFFv04_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv04_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 5)
+        case 5 :
+            {
+                ZBUFFv05_DCtx* dctx = (ZBUFFv05_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv05_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 6)
+        case 6 :
+            {
+                ZBUFFv06_DCtx* dctx = (ZBUFFv06_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv06_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+#if (ZSTD_LEGACY_SUPPORT <= 7)
+        case 7 :
+            {
+                ZBUFFv07_DCtx* dctx = (ZBUFFv07_DCtx*) legacyContext;
+                const void* src = (const char*)input->src + input->pos;
+                size_t readSize = input->size - input->pos;
+                void* dst = (char*)output->dst + output->pos;
+                size_t decodedSize = output->size - output->pos;
+                size_t const hintSize = ZBUFFv07_decompressContinue(dctx, dst, &decodedSize, src, &readSize);
+                output->pos += decodedSize;
+                input->pos += readSize;
+                return hintSize;
+            }
+#endif
+    }
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTD_LEGACY_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_opt.c b/vendor/github.com/DataDog/zstd/zstd_opt.c
new file mode 100644
index 0000000..f63f0c5
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_opt.c
@@ -0,0 +1,923 @@
+/*
+ * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#include "zstd_compress_internal.h"
+#include "zstd_opt.h"
+
+
+#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats. Also used for matchSum (?) */
+#define ZSTD_FREQ_DIV       4   /* log factor when using previous stats to init next stats */
+#define ZSTD_MAX_PRICE     (1<<30)
+
+
+/*-*************************************
+*  Price functions for optimal parser
+***************************************/
+static void ZSTD_setLog2Prices(optState_t* optPtr)
+{
+    optPtr->log2litSum = ZSTD_highbit32(optPtr->litSum+1);
+    optPtr->log2litLengthSum = ZSTD_highbit32(optPtr->litLengthSum+1);
+    optPtr->log2matchLengthSum = ZSTD_highbit32(optPtr->matchLengthSum+1);
+    optPtr->log2offCodeSum = ZSTD_highbit32(optPtr->offCodeSum+1);
+}
+
+
+static void ZSTD_rescaleFreqs(optState_t* const optPtr,
+                              const BYTE* const src, size_t const srcSize)
+{
+    optPtr->staticPrices = 0;
+
+    if (optPtr->litLengthSum == 0) {  /* first init */
+        unsigned u;
+        if (srcSize <= 1024) optPtr->staticPrices = 1;
+
+        assert(optPtr->litFreq!=NULL);
+        for (u=0; u<=MaxLit; u++)
+            optPtr->litFreq[u] = 0;
+        for (u=0; u<srcSize; u++)
+            optPtr->litFreq[src[u]]++;
+        optPtr->litSum = 0;
+        for (u=0; u<=MaxLit; u++) {
+            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u] >> ZSTD_FREQ_DIV);
+            optPtr->litSum += optPtr->litFreq[u];
+        }
+
+        for (u=0; u<=MaxLL; u++)
+            optPtr->litLengthFreq[u] = 1;
+        optPtr->litLengthSum = MaxLL+1;
+        for (u=0; u<=MaxML; u++)
+            optPtr->matchLengthFreq[u] = 1;
+        optPtr->matchLengthSum = MaxML+1;
+        for (u=0; u<=MaxOff; u++)
+            optPtr->offCodeFreq[u] = 1;
+        optPtr->offCodeSum = (MaxOff+1);
+
+    } else {
+        unsigned u;
+
+        optPtr->litSum = 0;
+        for (u=0; u<=MaxLit; u++) {
+            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u] >> (ZSTD_FREQ_DIV+1));
+            optPtr->litSum += optPtr->litFreq[u];
+        }
+        optPtr->litLengthSum = 0;
+        for (u=0; u<=MaxLL; u++) {
+            optPtr->litLengthFreq[u] = 1 + (optPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
+            optPtr->litLengthSum += optPtr->litLengthFreq[u];
+        }
+        optPtr->matchLengthSum = 0;
+        for (u=0; u<=MaxML; u++) {
+            optPtr->matchLengthFreq[u] = 1 + (optPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->matchLengthSum += optPtr->matchLengthFreq[u];
+        }
+        optPtr->offCodeSum = 0;
+        for (u=0; u<=MaxOff; u++) {
+            optPtr->offCodeFreq[u] = 1 + (optPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
+            optPtr->offCodeSum += optPtr->offCodeFreq[u];
+        }
+    }
+
+    ZSTD_setLog2Prices(optPtr);
+}
+
+
+/* ZSTD_rawLiteralsCost() :
+ * cost of literals (only) in given segment (which length can be null)
+ * does not include cost of literalLength symbol */
+static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                const optState_t* const optPtr)
+{
+    if (optPtr->staticPrices) return (litLength*6);  /* 6 bit per literal - no statistic used */
+    if (litLength == 0) return 0;
+
+    /* literals */
+    {   U32 u;
+        U32 cost = litLength * optPtr->log2litSum;
+        for (u=0; u < litLength; u++)
+            cost -= ZSTD_highbit32(optPtr->litFreq[literals[u]]+1);
+        return cost;
+    }
+}
+
+/* ZSTD_litLengthPrice() :
+ * cost of literalLength symbol */
+static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr)
+{
+    if (optPtr->staticPrices) return ZSTD_highbit32((U32)litLength+1);
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        U32 const price = LL_bits[llCode] + optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
+        return price;
+    }
+}
+
+/* ZSTD_litLengthPrice() :
+ * cost of the literal part of a sequence,
+ * including literals themselves, and literalLength symbol */
+static U32 ZSTD_fullLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                 const optState_t* const optPtr)
+{
+    return ZSTD_rawLiteralsCost(literals, litLength, optPtr)
+         + ZSTD_litLengthPrice(litLength, optPtr);
+}
+
+/* ZSTD_litLengthContribution() :
+ * @return ( cost(litlength) - cost(0) )
+ * this value can then be added to rawLiteralsCost()
+ * to provide a cost which is directly comparable to a match ending at same position */
+static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr)
+{
+    if (optPtr->staticPrices) return ZSTD_highbit32(litLength+1);
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        int const contribution = LL_bits[llCode]
+                        + ZSTD_highbit32(optPtr->litLengthFreq[0]+1)
+                        - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
+#if 1
+        return contribution;
+#else
+        return MAX(0, contribution); /* sometimes better, sometimes not ... */
+#endif
+    }
+}
+
+/* ZSTD_literalsContribution() :
+ * creates a fake cost for the literals part of a sequence
+ * which can be compared to the ending cost of a match
+ * should a new match start at this position */
+static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLength,
+                                     const optState_t* const optPtr)
+{
+    int const contribution = ZSTD_rawLiteralsCost(literals, litLength, optPtr)
+                           + ZSTD_litLengthContribution(litLength, optPtr);
+    return contribution;
+}
+
+/* ZSTD_getMatchPrice() :
+ * Provides the cost of the match part (offset + matchLength) of a sequence
+ * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
+ * optLevel: when <2, favors small offset for decompression speed (improved cache efficiency) */
+FORCE_INLINE_TEMPLATE U32 ZSTD_getMatchPrice(
+                                    U32 const offset, U32 const matchLength,
+                                    const optState_t* const optPtr,
+                                    int const optLevel)
+{
+    U32 price;
+    U32 const offCode = ZSTD_highbit32(offset+1);
+    U32 const mlBase = matchLength - MINMATCH;
+    assert(matchLength >= MINMATCH);
+
+    if (optPtr->staticPrices)  /* fixed scheme, do not use statistics */
+        return ZSTD_highbit32((U32)mlBase+1) + 16 + offCode;
+
+    price = offCode + optPtr->log2offCodeSum - ZSTD_highbit32(optPtr->offCodeFreq[offCode]+1);
+    if ((optLevel<2) /*static*/ && offCode >= 20) price += (offCode-19)*2; /* handicap for long distance offsets, favor decompression speed */
+
+    /* match Length */
+    {   U32 const mlCode = ZSTD_MLcode(mlBase);
+        price += ML_bits[mlCode] + optPtr->log2matchLengthSum - ZSTD_highbit32(optPtr->matchLengthFreq[mlCode]+1);
+    }
+
+    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
+    return price;
+}
+
+static void ZSTD_updateStats(optState_t* const optPtr,
+                             U32 litLength, const BYTE* literals,
+                             U32 offsetCode, U32 matchLength)
+{
+    /* literals */
+    {   U32 u;
+        for (u=0; u < litLength; u++)
+            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    }
+
+    /* literal Length */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        optPtr->litLengthFreq[llCode]++;
+        optPtr->litLengthSum++;
+    }
+
+    /* match offset code (0-2=>repCode; 3+=>offset+2) */
+    {   U32 const offCode = ZSTD_highbit32(offsetCode+1);
+        assert(offCode <= MaxOff);
+        optPtr->offCodeFreq[offCode]++;
+        optPtr->offCodeSum++;
+    }
+
+    /* match Length */
+    {   U32 const mlBase = matchLength - MINMATCH;
+        U32 const mlCode = ZSTD_MLcode(mlBase);
+        optPtr->matchLengthFreq[mlCode]++;
+        optPtr->matchLengthSum++;
+    }
+}
+
+
+/* ZSTD_readMINMATCH() :
+ * function safe only for comparisons
+ * assumption : memPtr must be at least 4 bytes before end of buffer */
+MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
+{
+    switch (length)
+    {
+    default :
+    case 4 : return MEM_read32(memPtr);
+    case 3 : if (MEM_isLittleEndian())
+                return MEM_read32(memPtr)<<8;
+             else
+                return MEM_read32(memPtr)>>8;
+    }
+}
+
+
+/* Update hashTable3 up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE* const ip)
+{
+    U32* const hashTable3 = ms->hashTable3;
+    U32 const hashLog3 = ms->hashLog3;
+    const BYTE* const base = ms->window.base;
+    U32 idx = ms->nextToUpdate3;
+    U32 const target = ms->nextToUpdate3 = (U32)(ip - base);
+    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+    assert(hashLog3 > 0);
+
+    while(idx < target) {
+        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
+        idx++;
+    }
+
+    return hashTable3[hash3];
+}
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+/** ZSTD_insertBt1() : add one or multiple positions to tree.
+ *  ip : assumed <= iend-8 .
+ * @return : nb of positions added */
+static U32 ZSTD_insertBt1(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* const ip, const BYTE* const iend,
+                U32 const mls, U32 const extDict)
+{
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
+    const U32 current = (U32)(ip-base);
+    const U32 btLow = btMask >= current ? 0 : current - btMask;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowLow = ms->window.lowLimit;
+    U32 matchEndIdx = current+8+1;
+    size_t bestLength = 8;
+    U32 nbCompares = 1U << cParams->searchLog;
+#ifdef ZSTD_C_PREDICT
+    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
+    predictedSmall += (predictedSmall>0);
+    predictedLarge += (predictedLarge>0);
+#endif /* ZSTD_C_PREDICT */
+
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);
+
+    assert(ip <= iend-8);   /* required for h calculation */
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < current);
+
+#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
+        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
+        if (matchIndex == predictedSmall) {
+            /* no need to check length, result known */
+            *smallerPtr = matchIndex;
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
+            continue;
+        }
+        if (matchIndex == predictedLarge) {
+            *largerPtr = matchIndex;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
+            continue;
+        }
+#endif
+
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if extDict is incorrectly set to 0 */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            bestLength = matchLength;
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+        }
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+    assert(matchEndIdx > current + 8);
+    return matchEndIdx - (current + 8);
+}
+
+FORCE_INLINE_TEMPLATE
+void ZSTD_updateTree_internal(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* const ip, const BYTE* const iend,
+                const U32 mls, const U32 extDict)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+    DEBUGLOG(7, "ZSTD_updateTree_internal, from %u to %u  (extDict:%u)",
+                idx, target, extDict);
+
+    while(idx < target)
+        idx += ZSTD_insertBt1(ms, cParams, base+idx, iend, mls, extDict);
+    ms->nextToUpdate = target;
+}
+
+void ZSTD_updateTree(
+                ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                const BYTE* ip, const BYTE* iend)
+{
+    ZSTD_updateTree_internal(ms, cParams, ip, iend, cParams->searchLength, 0 /*extDict*/);
+}
+
+FORCE_INLINE_TEMPLATE
+U32 ZSTD_insertBtAndGetAllMatches (
+                    ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                    const BYTE* const ip, const BYTE* const iLimit, int const extDict,
+                    U32 rep[ZSTD_REP_NUM], U32 const ll0,
+                    ZSTD_match_t* matches, const U32 lengthToBeat, U32 const mls /* template */)
+{
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    const BYTE* const base = ms->window.base;
+    U32 const current = (U32)(ip-base);
+    U32 const hashLog = cParams->hashLog;
+    U32 const minMatch = (mls==3) ? 3 : 4;
+    U32* const hashTable = ms->hashTable;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32 matchIndex  = hashTable[h];
+    U32* const bt   = ms->chainTable;
+    U32 const btLog = cParams->chainLog - 1;
+    U32 const btMask= (1U << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const dictBase = ms->window.dictBase;
+    U32 const dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    U32 const btLow = btMask >= current ? 0 : current - btMask;
+    U32 const windowLow = ms->window.lowLimit;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = bt + 2*(current&btMask) + 1;
+    U32 matchEndIdx = current+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
+    U32 dummy32;   /* to be nullified at the end */
+    U32 mnum = 0;
+    U32 nbCompares = 1U << cParams->searchLog;
+
+    size_t bestLength = lengthToBeat-1;
+    DEBUGLOG(7, "ZSTD_insertBtAndGetAllMatches");
+
+    /* check repCode */
+    {   U32 const lastR = ZSTD_REP_NUM + ll0;
+        U32 repCode;
+        for (repCode = ll0; repCode < lastR; repCode++) {
+            U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            U32 const repIndex = current - repOffset;
+            U32 repLen = 0;
+            assert(current >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < current-dictLimit) {  /* equivalent to `current > repIndex >= dictLimit` */
+                if (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch)) {
+                    repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
+                }
+            } else {  /* repIndex < dictLimit || repIndex >= current */
+                const BYTE* const repMatch = dictBase + repIndex;
+                assert(current >= windowLow);
+                if ( extDict /* this case only valid in extDict mode */
+                  && ( ((repOffset-1) /*intentional overflow*/ < current - windowLow)  /* equivalent to `current > repIndex >= windowLow` */
+                     & (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
+            }   }
+            /* save longer solution */
+            if (repLen > bestLength) {
+                DEBUGLOG(8, "found rep-match %u of length %u",
+                            repCode - ll0, (U32)repLen);
+                bestLength = repLen;
+                matches[mnum].off = repCode - ll0;
+                matches[mnum].len = (U32)repLen;
+                mnum++;
+                if ( (repLen > sufficient_len)
+                   | (ip+repLen == iLimit) ) {  /* best possible */
+                    return mnum;
+    }   }   }   }
+
+    /* HC3 match finder */
+    if ((mls == 3) /*static*/ && (bestLength < mls)) {
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, ip);
+        if ((matchIndex3 > windowLow)
+          & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+            size_t mlen;
+            if ((!extDict) /*static*/ || (matchIndex3 >= dictLimit)) {
+                const BYTE* const match = base + matchIndex3;
+                mlen = ZSTD_count(ip, match, iLimit);
+            } else {
+                const BYTE* const match = dictBase + matchIndex3;
+                mlen = ZSTD_count_2segments(ip, match, iLimit, dictEnd, prefixStart);
+            }
+
+            /* save best solution */
+            if (mlen >= mls /* == 3 > bestLength */) {
+                DEBUGLOG(8, "found small match with hlog3, of length %u",
+                            (U32)mlen);
+                bestLength = mlen;
+                assert(current > matchIndex3);
+                assert(mnum==0);  /* no prior solution */
+                matches[0].off = (current - matchIndex3) + ZSTD_REP_MOVE;
+                matches[0].len = (U32)mlen;
+                mnum = 1;
+                if ( (mlen > sufficient_len) |
+                     (ip+mlen == iLimit) ) {  /* best possible length */
+                    ms->nextToUpdate = current+1;  /* skip insertion */
+                    return 1;
+    }   }   }   }
+
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match;
+        assert(current > matchIndex);
+
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* prepare for match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            DEBUGLOG(8, "found match of length %u at distance %u",
+                    (U32)matchLength, current - matchIndex);
+            assert(matchEndIdx > matchIndex);
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+            bestLength = matchLength;
+            matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+            matches[mnum].len = (U32)matchLength;
+            mnum++;
+            if (matchLength > ZSTD_OPT_NUM) break;
+            if (ip+matchLength == iLimit) {  /* equal : no way to know if inf or sup */
+                break;   /* drop, to preserve bt consistency (miss a little bit of compression) */
+            }
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            /* match smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new candidate => larger than match, which was smaller than current */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous, closer to current */
+        } else {
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+
+    assert(matchEndIdx > current+8);
+    ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
+    return mnum;
+}
+
+
+FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
+                        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+                        const BYTE* ip, const BYTE* const iHighLimit, int const extDict,
+                        U32 rep[ZSTD_REP_NUM], U32 const ll0,
+                        ZSTD_match_t* matches, U32 const lengthToBeat)
+{
+    U32 const matchLengthSearch = cParams->searchLength;
+    DEBUGLOG(7, "ZSTD_BtGetAllMatches");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree_internal(ms, cParams, ip, iHighLimit, matchLengthSearch, extDict);
+    switch(matchLengthSearch)
+    {
+    case 3 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 3);
+    default :
+    case 4 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 4);
+    case 5 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 5);
+    case 7 :
+    case 6 : return ZSTD_insertBtAndGetAllMatches(ms, cParams, ip, iHighLimit, extDict, rep, ll0, matches, lengthToBeat, 6);
+    }
+}
+
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+typedef struct repcodes_s {
+    U32 rep[3];
+} repcodes_t;
+
+repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
+{
+    repcodes_t newReps;
+    if (offset >= ZSTD_REP_NUM) {  /* full offset */
+        newReps.rep[2] = rep[1];
+        newReps.rep[1] = rep[0];
+        newReps.rep[0] = offset - ZSTD_REP_MOVE;
+    } else {   /* repcode */
+        U32 const repCode = offset + ll0;
+        if (repCode > 0) {  /* note : if repCode==0, no change */
+            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            newReps.rep[2] = (repCode >= 2) ? rep[1] : rep[2];
+            newReps.rep[1] = rep[0];
+            newReps.rep[0] = currentOffset;
+        } else {   /* repCode == 0 */
+            memcpy(&newReps, rep, sizeof(newReps));
+        }
+    }
+    return newReps;
+}
+
+
+typedef struct {
+    const BYTE* anchor;
+    U32 litlen;
+    U32 rawLitCost;
+} cachedLiteralPrice_t;
+
+static U32 ZSTD_rawLiteralsCost_cached(
+                            cachedLiteralPrice_t* const cachedLitPrice,
+                            const BYTE* const anchor, U32 const litlen,
+                            const optState_t* const optStatePtr)
+{
+    U32 startCost;
+    U32 remainingLength;
+    const BYTE* startPosition;
+
+    if (anchor == cachedLitPrice->anchor) {
+        startCost = cachedLitPrice->rawLitCost;
+        startPosition = anchor + cachedLitPrice->litlen;
+        assert(litlen >= cachedLitPrice->litlen);
+        remainingLength = litlen - cachedLitPrice->litlen;
+    } else {
+        startCost = 0;
+        startPosition = anchor;
+        remainingLength = litlen;
+    }
+
+    {   U32 const rawLitCost = startCost + ZSTD_rawLiteralsCost(startPosition, remainingLength, optStatePtr);
+        cachedLitPrice->anchor = anchor;
+        cachedLitPrice->litlen = litlen;
+        cachedLitPrice->rawLitCost = rawLitCost;
+        return rawLitCost;
+    }
+}
+
+static U32 ZSTD_fullLiteralsCost_cached(
+                            cachedLiteralPrice_t* const cachedLitPrice,
+                            const BYTE* const anchor, U32 const litlen,
+                            const optState_t* const optStatePtr)
+{
+    return ZSTD_rawLiteralsCost_cached(cachedLitPrice, anchor, litlen, optStatePtr)
+         + ZSTD_litLengthPrice(litlen, optStatePtr);
+}
+
+static int ZSTD_literalsContribution_cached(
+                            cachedLiteralPrice_t* const cachedLitPrice,
+                            const BYTE* const anchor, U32 const litlen,
+                            const optState_t* const optStatePtr)
+{
+    int const contribution = ZSTD_rawLiteralsCost_cached(cachedLitPrice, anchor, litlen, optStatePtr)
+                           + ZSTD_litLengthContribution(litlen, optStatePtr);
+    return contribution;
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,seqStore_t* seqStore,
+                                      U32 rep[ZSTD_REP_NUM],
+                                      ZSTD_compressionParameters const* cParams,
+                                      const void* src, size_t srcSize,
+                                      const int optLevel, const int extDict)
+{
+    optState_t* const optStatePtr = &ms->opt;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    U32 const minMatch = (cParams->searchLength == 3) ? 3 : 4;
+
+    ZSTD_optimal_t* const opt = optStatePtr->priceTable;
+    ZSTD_match_t* const matches = optStatePtr->matchTable;
+    cachedLiteralPrice_t cachedLitPrice;
+
+    /* init */
+    DEBUGLOG(5, "ZSTD_compressBlock_opt_generic");
+    ms->nextToUpdate3 = ms->nextToUpdate;
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
+    ip += (ip==prefixStart);
+    memset(&cachedLitPrice, 0, sizeof(cachedLitPrice));
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        U32 cur, last_pos = 0;
+        U32 best_mlen, best_off;
+
+        /* find first match */
+        {   U32 const litlen = (U32)(ip - anchor);
+            U32 const ll0 = !litlen;
+            U32 const nbMatches = ZSTD_BtGetAllMatches(ms, cParams, ip, iend, extDict, rep, ll0, matches, minMatch);
+            if (!nbMatches) { ip++; continue; }
+
+            /* initialize opt[0] */
+            { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
+            opt[0].mlen = 1;
+            opt[0].litlen = litlen;
+
+            /* large match -> immediate encoding */
+            {   U32 const maxML = matches[nbMatches-1].len;
+                DEBUGLOG(7, "found %u matches of maxLength=%u and offset=%u at cPos=%u => start new serie",
+                            nbMatches, maxML, matches[nbMatches-1].off, (U32)(ip-prefixStart));
+
+                if (maxML > sufficient_len) {
+                    best_mlen = maxML;
+                    best_off = matches[nbMatches-1].off;
+                    DEBUGLOG(7, "large match (%u>%u), immediate encoding",
+                                best_mlen, sufficient_len);
+                    cur = 0;
+                    last_pos = 1;
+                    goto _shortestPath;
+            }   }
+
+            /* set prices for first matches starting position == 0 */
+            {   U32 const literalsPrice = ZSTD_fullLiteralsCost_cached(&cachedLitPrice, anchor, litlen, optStatePtr);
+                U32 pos;
+                U32 matchNb;
+                for (pos = 0; pos < minMatch; pos++) {
+                    opt[pos].mlen = 1;
+                    opt[pos].price = ZSTD_MAX_PRICE;
+                }
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    U32 const end = matches[matchNb].len;
+                    repcodes_t const repHistory = ZSTD_updateRep(rep, offset, ll0);
+                    for ( ; pos <= end ; pos++ ) {
+                        U32 const matchPrice = literalsPrice + ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
+                        DEBUGLOG(7, "rPos:%u => set initial price : %u",
+                                    pos, matchPrice);
+                        opt[pos].mlen = pos;
+                        opt[pos].off = offset;
+                        opt[pos].litlen = litlen;
+                        opt[pos].price = matchPrice;
+                        memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
+                }   }
+                last_pos = pos-1;
+            }
+        }
+
+        /* check further positions */
+        for (cur = 1; cur <= last_pos; cur++) {
+            const BYTE* const inr = ip + cur;
+            assert(cur < ZSTD_OPT_NUM);
+
+            /* Fix current position with one literal if cheaper */
+            {   U32 const litlen = (opt[cur-1].mlen == 1) ? opt[cur-1].litlen + 1 : 1;
+                int price;  /* note : contribution can be negative */
+                if (cur > litlen) {
+                    price = opt[cur - litlen].price + ZSTD_literalsContribution(inr-litlen, litlen, optStatePtr);
+                } else {
+                    price = ZSTD_literalsContribution_cached(&cachedLitPrice, anchor, litlen, optStatePtr);
+                }
+                assert(price < 1000000000); /* overflow check */
+                if (price <= opt[cur].price) {
+                    DEBUGLOG(7, "rPos:%u : better price (%u<%u) using literal",
+                                cur, price, opt[cur].price);
+                    opt[cur].mlen = 1;
+                    opt[cur].off = 0;
+                    opt[cur].litlen = litlen;
+                    opt[cur].price = price;
+                    memcpy(opt[cur].rep, opt[cur-1].rep, sizeof(opt[cur].rep));
+            }   }
+
+            /* last match must start at a minimum distance of 8 from oend */
+            if (inr > ilimit) continue;
+
+            if (cur == last_pos) break;
+
+             if ( (optLevel==0) /*static*/
+               && (opt[cur+1].price <= opt[cur].price) )
+                continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
+
+            {   U32 const ll0 = (opt[cur].mlen != 1);
+                U32 const litlen = (opt[cur].mlen == 1) ? opt[cur].litlen : 0;
+                U32 const previousPrice = (cur > litlen) ? opt[cur-litlen].price : 0;
+                U32 const basePrice = previousPrice + ZSTD_fullLiteralsCost(inr-litlen, litlen, optStatePtr);
+                U32 const nbMatches = ZSTD_BtGetAllMatches(ms, cParams, inr, iend, extDict, opt[cur].rep, ll0, matches, minMatch);
+                U32 matchNb;
+                if (!nbMatches) continue;
+
+                {   U32 const maxML = matches[nbMatches-1].len;
+                    DEBUGLOG(7, "rPos:%u, found %u matches, of maxLength=%u",
+                                cur, nbMatches, maxML);
+
+                    if ( (maxML > sufficient_len)
+                       | (cur + maxML >= ZSTD_OPT_NUM) ) {
+                        best_mlen = maxML;
+                        best_off = matches[nbMatches-1].off;
+                        last_pos = cur + 1;
+                        goto _shortestPath;
+                    }
+                }
+
+                /* set prices using matches found at position == cur */
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    repcodes_t const repHistory = ZSTD_updateRep(opt[cur].rep, offset, ll0);
+                    U32 const lastML = matches[matchNb].len;
+                    U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
+                    U32 mlen;
+
+                    DEBUGLOG(7, "testing match %u => offCode=%u, mlen=%u, llen=%u",
+                                matchNb, matches[matchNb].off, lastML, litlen);
+
+                    for (mlen = lastML; mlen >= startML; mlen--) {
+                        U32 const pos = cur + mlen;
+                        int const price = basePrice + ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+
+                        if ((pos > last_pos) || (price < opt[pos].price)) {
+                            DEBUGLOG(7, "rPos:%u => new better price (%u<%u)",
+                                        pos, price, opt[pos].price);
+                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }
+                            opt[pos].mlen = mlen;
+                            opt[pos].off = offset;
+                            opt[pos].litlen = litlen;
+                            opt[pos].price = price;
+                            memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
+                        } else {
+                            if (optLevel==0) break;  /* gets ~+10% speed for about -0.01 ratio loss */
+                        }
+            }   }   }
+        }  /* for (cur = 1; cur <= last_pos; cur++) */
+
+        best_mlen = opt[last_pos].mlen;
+        best_off = opt[last_pos].off;
+        cur = last_pos - best_mlen;
+
+_shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
+        assert(opt[0].mlen == 1);
+
+        /* reverse traversal */
+        DEBUGLOG(7, "start reverse traversal (last_pos:%u, cur:%u)",
+                    last_pos, cur);
+        {   U32 selectedMatchLength = best_mlen;
+            U32 selectedOffset = best_off;
+            U32 pos = cur;
+            while (1) {
+                U32 const mlen = opt[pos].mlen;
+                U32 const off = opt[pos].off;
+                opt[pos].mlen = selectedMatchLength;
+                opt[pos].off = selectedOffset;
+                selectedMatchLength = mlen;
+                selectedOffset = off;
+                if (mlen > pos) break;
+                pos -= mlen;
+        }   }
+
+        /* save sequences */
+        {   U32 pos;
+            for (pos=0; pos < last_pos; ) {
+                U32 const llen = (U32)(ip - anchor);
+                U32 const mlen = opt[pos].mlen;
+                U32 const offset = opt[pos].off;
+                if (mlen == 1) { ip++; pos++; continue; }  /* literal position => move on */
+                pos += mlen; ip += mlen;
+
+                /* repcodes update : like ZSTD_updateRep(), but update in place */
+                if (offset >= ZSTD_REP_NUM) {  /* full offset */
+                    rep[2] = rep[1];
+                    rep[1] = rep[0];
+                    rep[0] = offset - ZSTD_REP_MOVE;
+                } else {   /* repcode */
+                    U32 const repCode = offset + (llen==0);
+                    if (repCode) {  /* note : if repCode==0, no change */
+                        U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+                        if (repCode >= 2) rep[2] = rep[1];
+                        rep[1] = rep[0];
+                        rep[0] = currentOffset;
+                    }
+                }
+
+                ZSTD_updateStats(optStatePtr, llen, anchor, offset, mlen);
+                ZSTD_storeSeq(seqStore, llen, anchor, offset, mlen-MINMATCH);
+                anchor = ip;
+        }   }
+        ZSTD_setLog2Prices(optStatePtr);
+    }   /* while (ip < ilimit) */
+
+    /* Return the last literals size */
+    return iend - anchor;
+}
+
+
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btopt");
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 0 /*optLevel*/, 0 /*extDict*/);
+}
+
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 2 /*optLevel*/, 0 /*extDict*/);
+}
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 0 /*optLevel*/, 1 /*extDict*/);
+}
+
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, cParams, src, srcSize, 2 /*optLevel*/, 1 /*extDict*/);
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_opt.h b/vendor/github.com/DataDog/zstd/zstd_opt.h
new file mode 100644
index 0000000..b8dc389
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_opt.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_OPT_H
+#define ZSTD_OPT_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#include "zstd_compress_internal.h"
+
+void ZSTD_updateTree(
+        ZSTD_matchState_t* ms, ZSTD_compressionParameters const* cParams,
+        const BYTE* ip, const BYTE* iend);  /* used in ZSTD_loadDictionaryContent() */
+
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize);
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_OPT_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_stream.go b/vendor/github.com/DataDog/zstd/zstd_stream.go
new file mode 100644
index 0000000..28dfd3e
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_stream.go
@@ -0,0 +1,269 @@
+package zstd
+
+/*
+#define ZSTD_STATIC_LINKING_ONLY
+#define ZBUFF_DISABLE_DEPRECATE_WARNINGS
+#include "zstd.h"
+#include "zbuff.h"
+*/
+import "C"
+import (
+	"fmt"
+	"io"
+	"unsafe"
+)
+
+// Writer is an io.WriteCloser that zstd-compresses its input.
+type Writer struct {
+	CompressionLevel int
+
+	ctx              *C.ZSTD_CCtx
+	dict             []byte
+	dstBuffer        []byte
+	firstError       error
+	underlyingWriter io.Writer
+}
+
+func resize(in []byte, newSize int) []byte {
+	if in == nil {
+		return make([]byte, newSize)
+	}
+	if newSize <= cap(in) {
+		return in[:newSize]
+	}
+	toAdd := newSize - len(in)
+	return append(in, make([]byte, toAdd)...)
+}
+
+// NewWriter creates a new Writer with default compression options.  Writes to
+// the writer will be written in compressed form to w.
+func NewWriter(w io.Writer) *Writer {
+	return NewWriterLevelDict(w, DefaultCompression, nil)
+}
+
+// NewWriterLevel is like NewWriter but specifies the compression level instead
+// of assuming default compression.
+//
+// The level can be DefaultCompression or any integer value between BestSpeed
+// and BestCompression inclusive.
+func NewWriterLevel(w io.Writer, level int) *Writer {
+	return NewWriterLevelDict(w, level, nil)
+
+}
+
+// NewWriterLevelDict is like NewWriterLevel but specifies a dictionary to
+// compress with.  If the dictionary is empty or nil it is ignored. The dictionary
+// should not be modified until the writer is closed.
+func NewWriterLevelDict(w io.Writer, level int, dict []byte) *Writer {
+	var err error
+	ctx := C.ZSTD_createCCtx()
+
+	if dict == nil {
+		err = getError(int(C.ZSTD_compressBegin(ctx,
+			C.int(level))))
+	} else {
+		err = getError(int(C.ZSTD_compressBegin_usingDict(
+			ctx,
+			unsafe.Pointer(&dict[0]),
+			C.size_t(len(dict)),
+			C.int(level))))
+	}
+
+	return &Writer{
+		CompressionLevel: level,
+		ctx:              ctx,
+		dict:             dict,
+		dstBuffer:        make([]byte, CompressBound(1024)),
+		firstError:       err,
+		underlyingWriter: w,
+	}
+}
+
+// Write writes a compressed form of p to the underlying io.Writer.
+func (w *Writer) Write(p []byte) (int, error) {
+	if w.firstError != nil {
+		return 0, w.firstError
+	}
+	if len(p) == 0 {
+		return 0, nil
+	}
+	// Check if dstBuffer is enough
+	if len(w.dstBuffer) < CompressBound(len(p)) {
+		w.dstBuffer = make([]byte, CompressBound(len(p)))
+	}
+
+	retCode := C.ZSTD_compressContinue(
+		w.ctx,
+		unsafe.Pointer(&w.dstBuffer[0]),
+		C.size_t(len(w.dstBuffer)),
+		unsafe.Pointer(&p[0]),
+		C.size_t(len(p)))
+
+	if err := getError(int(retCode)); err != nil {
+		return 0, err
+	}
+	written := int(retCode)
+
+	// Write to underlying buffer
+	_, err := w.underlyingWriter.Write(w.dstBuffer[:written])
+
+	// Same behaviour as zlib, we can't know how much data we wrote, only
+	// if there was an error
+	if err != nil {
+		return 0, err
+	}
+	return len(p), err
+}
+
+// Close closes the Writer, flushing any unwritten data to the underlying
+// io.Writer and freeing objects, but does not close the underlying io.Writer.
+func (w *Writer) Close() error {
+	retCode := C.ZSTD_compressEnd(
+		w.ctx,
+		unsafe.Pointer(&w.dstBuffer[0]),
+		C.size_t(len(w.dstBuffer)),
+		unsafe.Pointer(nil),
+		C.size_t(0))
+
+	if err := getError(int(retCode)); err != nil {
+		return err
+	}
+	written := int(retCode)
+	retCode = C.ZSTD_freeCCtx(w.ctx) // Safely close buffer before writing the end
+
+	if err := getError(int(retCode)); err != nil {
+		return err
+	}
+
+	_, err := w.underlyingWriter.Write(w.dstBuffer[:written])
+	if err != nil {
+		return err
+	}
+	return nil
+}
+
+// reader is an io.ReadCloser that decompresses when read from.
+type reader struct {
+	ctx                 *C.ZBUFF_DCtx
+	compressionBuffer   []byte
+	compressionLeft     int
+	decompressionBuffer []byte
+	decompOff           int
+	decompSize          int
+	dict                []byte
+	firstError          error
+	recommendedSrcSize  int
+	underlyingReader    io.Reader
+}
+
+// NewReader creates a new io.ReadCloser.  Reads from the returned ReadCloser
+// read and decompress data from r.  It is the caller's responsibility to call
+// Close on the ReadCloser when done.  If this is not done, underlying objects
+// in the zstd library will not be freed.
+func NewReader(r io.Reader) io.ReadCloser {
+	return NewReaderDict(r, nil)
+}
+
+// NewReaderDict is like NewReader but uses a preset dictionary.  NewReaderDict
+// ignores the dictionary if it is nil.
+func NewReaderDict(r io.Reader, dict []byte) io.ReadCloser {
+	var err error
+	ctx := C.ZBUFF_createDCtx()
+	if len(dict) == 0 {
+		err = getError(int(C.ZBUFF_decompressInit(ctx)))
+	} else {
+		err = getError(int(C.ZBUFF_decompressInitDictionary(
+			ctx,
+			unsafe.Pointer(&dict[0]),
+			C.size_t(len(dict)))))
+	}
+	cSize := int(C.ZBUFF_recommendedDInSize())
+	dSize := int(C.ZBUFF_recommendedDOutSize())
+	if cSize <= 0 {
+		panic(fmt.Errorf("ZBUFF_recommendedDInSize() returned invalid size: %v", cSize))
+	}
+	if dSize <= 0 {
+		panic(fmt.Errorf("ZBUFF_recommendedDOutSize() returned invalid size: %v", dSize))
+	}
+
+	compressionBuffer := make([]byte, cSize)
+	decompressionBuffer := make([]byte, dSize)
+	return &reader{
+		ctx:                 ctx,
+		dict:                dict,
+		compressionBuffer:   compressionBuffer,
+		decompressionBuffer: decompressionBuffer,
+		firstError:          err,
+		recommendedSrcSize:  cSize,
+		underlyingReader:    r,
+	}
+}
+
+// Close frees the allocated C objects
+func (r *reader) Close() error {
+	return getError(int(C.ZBUFF_freeDCtx(r.ctx)))
+}
+
+func (r *reader) Read(p []byte) (int, error) {
+
+	// If we already have enough bytes, return
+	if r.decompSize-r.decompOff >= len(p) {
+		copy(p, r.decompressionBuffer[r.decompOff:])
+		r.decompOff += len(p)
+		return len(p), nil
+	}
+
+	copy(p, r.decompressionBuffer[r.decompOff:r.decompSize])
+	got := r.decompSize - r.decompOff
+	r.decompSize = 0
+	r.decompOff = 0
+
+	for got < len(p) {
+		// Populate src
+		src := r.compressionBuffer
+		reader := r.underlyingReader
+		n, err := io.ReadFull(reader, src[r.compressionLeft:])
+		if err == io.EOF && r.compressionLeft == 0 {
+			return got, io.EOF
+		} else if err != nil && err != io.EOF && err != io.ErrUnexpectedEOF {
+			return 0, fmt.Errorf("failed to read from underlying reader: %s", err)
+		}
+		src = src[:r.compressionLeft+n]
+
+		// C code
+		cSrcSize := C.size_t(len(src))
+		cDstSize := C.size_t(len(r.decompressionBuffer))
+		retCode := int(C.ZBUFF_decompressContinue(
+			r.ctx,
+			unsafe.Pointer(&r.decompressionBuffer[0]),
+			&cDstSize,
+			unsafe.Pointer(&src[0]),
+			&cSrcSize))
+
+		if err = getError(retCode); err != nil {
+			return 0, fmt.Errorf("failed to decompress: %s", err)
+		}
+
+		// Put everything in buffer
+		if int(cSrcSize) < len(src) {
+			left := src[int(cSrcSize):]
+			copy(r.compressionBuffer, left)
+		}
+		r.compressionLeft = len(src) - int(cSrcSize)
+		r.decompSize = int(cDstSize)
+		r.decompOff = copy(p[got:], r.decompressionBuffer[:r.decompSize])
+		got += r.decompOff
+
+		// Resize buffers
+		nsize := retCode // Hint for next src buffer size
+		if nsize <= 0 {
+			// Reset to recommended size
+			nsize = r.recommendedSrcSize
+		}
+		if nsize < r.compressionLeft {
+			nsize = r.compressionLeft
+		}
+		r.compressionBuffer = resize(r.compressionBuffer, nsize)
+	}
+	return got, nil
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_v01.c b/vendor/github.com/DataDog/zstd/zstd_v01.c
new file mode 100644
index 0000000..ae1cb2c
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v01.c
@@ -0,0 +1,2127 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v01.h"
+#include "error_private.h"
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+/* You can statically allocate Huff0 DTable as a table of unsigned short using below macro */
+#define HUF_DTABLE_SIZE_U16(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUF_CREATE_STATIC_DTABLE(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE_U16(maxTableLog)] = { maxTableLog }
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define FSE_LIST_ERRORS(ITEM) \
+        ITEM(FSE_OK_NoError) ITEM(FSE_ERROR_GENERIC) \
+        ITEM(FSE_ERROR_tableLog_tooLarge) ITEM(FSE_ERROR_maxSymbolValue_tooLarge) ITEM(FSE_ERROR_maxSymbolValue_tooSmall) \
+        ITEM(FSE_ERROR_dstSize_tooSmall) ITEM(FSE_ERROR_srcSize_wrong)\
+        ITEM(FSE_ERROR_corruptionDetected) \
+        ITEM(FSE_ERROR_maxCode)
+
+#define FSE_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { FSE_LIST_ERRORS(FSE_GENERATE_ENUM) } FSE_errorCodes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+
+/******************************************
+*  FSE symbol compression API
+******************************************/
+/*
+   This API consists of small unitary functions, which highly benefit from being inlined.
+   You will want to enable link-time-optimization to ensure these functions are properly inlined in your binary.
+   Visual seems to do it automatically.
+   For gcc or clang, you'll need to add -flto flag at compilation and linking stages.
+   If none of these solutions is applicable, include "fse.c" directly.
+*/
+
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+typedef struct
+{
+    size_t bitContainer;
+    int    bitPos;
+    char*  startPtr;
+    char*  ptr;
+    char*  endPtr;
+} FSE_CStream_t;
+
+typedef struct
+{
+    ptrdiff_t   value;
+    const void* stateTable;
+    const void* symbolTT;
+    unsigned    stateLog;
+} FSE_CState_t;
+
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} FSE_DStream_t;
+
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+typedef enum { FSE_DStream_unfinished = 0,
+               FSE_DStream_endOfBuffer = 1,
+               FSE_DStream_completed = 2,
+               FSE_DStream_tooFar = 3 } FSE_DStream_status;  /* result of FSE_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... ?! */
+
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+#ifndef MEM_ACCESS_MODULE
+#define MEM_ACCESS_MODULE
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+# include <stdint.h>
+typedef  uint8_t BYTE;
+typedef uint16_t U16;
+typedef  int16_t S16;
+typedef uint32_t U32;
+typedef  int32_t S32;
+typedef uint64_t U64;
+typedef  int64_t S64;
+#else
+typedef unsigned char       BYTE;
+typedef unsigned short      U16;
+typedef   signed short      S16;
+typedef unsigned int        U32;
+typedef   signed int        S32;
+typedef unsigned long long  U64;
+typedef   signed long long  S64;
+#endif
+
+#endif   /* MEM_ACCESS_MODULE */
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+/* FSE_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets generating assembly depending on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef FSE_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define FSE_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define FSE_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+
+static unsigned FSE_32bits(void)
+{
+    return sizeof(void*)==4;
+}
+
+static unsigned FSE_isLittleEndian(void)
+{
+    const union { U32 i; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(FSE_FORCE_MEMORY_ACCESS) && (FSE_FORCE_MEMORY_ACCESS==2)
+
+static U16 FSE_read16(const void* memPtr) { return *(const U16*) memPtr; }
+static U32 FSE_read32(const void* memPtr) { return *(const U32*) memPtr; }
+static U64 FSE_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+#elif defined(FSE_FORCE_MEMORY_ACCESS) && (FSE_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
+
+static U16 FSE_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+static U32 FSE_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+static U64 FSE_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+#else
+
+static U16 FSE_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U32 FSE_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+static U64 FSE_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+#endif // FSE_FORCE_MEMORY_ACCESS
+
+static U16 FSE_readLE16(const void* memPtr)
+{
+    if (FSE_isLittleEndian())
+        return FSE_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+static U32 FSE_readLE32(const void* memPtr)
+{
+    if (FSE_isLittleEndian())
+        return FSE_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+static U64 FSE_readLE64(const void* memPtr)
+{
+    if (FSE_isLittleEndian())
+        return FSE_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+static size_t FSE_readLEST(const void* memPtr)
+{
+    if (FSE_32bits())
+        return (size_t)FSE_readLE32(memPtr);
+    else
+        return (size_t)FSE_readLE64(memPtr);
+}
+
+
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef struct
+{
+    int deltaFindState;
+    U32 deltaNbBits;
+} FSE_symbolCompressionTransform; /* total 8 bytes */
+
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+/****************************************************************
+*  Internal functions
+****************************************************************/
+FORCE_INLINE unsigned FSE_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (GCC_VERSION >= 304)   /* GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)(ptr) + 1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return (size_t)-FSE_ERROR_maxSymbolValue_tooLarge;
+    if (tableLog > FSE_MAX_TABLELOG) return (size_t)-FSE_ERROR_tableLog_tooLarge;
+
+    /* Init, lay down lowprob symbols */
+    DTableH[0].tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return (size_t)-FSE_ERROR_GENERIC;   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - FSE_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH->fastMode = (U16)noLarge;
+    return 0;
+}
+
+
+/******************************************
+*  FSE byte symbol
+******************************************/
+#ifndef FSE_COMMONDEFS_ONLY
+
+static unsigned FSE_isError(size_t code) { return (code > (size_t)(-FSE_ERROR_maxCode)); }
+
+static short FSE_abs(short a)
+{
+    return a<0? -a : a;
+}
+
+
+/****************************************************************
+*  Header bitstream management
+****************************************************************/
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return (size_t)-FSE_ERROR_srcSize_wrong;
+    bitStream = FSE_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return (size_t)-FSE_ERROR_tableLog_tooLarge;
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = FSE_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return (size_t)-FSE_ERROR_maxSymbolValue_tooSmall;
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = FSE_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = FSE_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return (size_t)-FSE_ERROR_GENERIC;
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return (size_t)-FSE_ERROR_GENERIC;             /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+
+/* FSE_initDStream
+ * Initialize a FSE_DStream_t.
+ * srcBuffer must point at the beginning of an FSE block.
+ * The function result is the size of the FSE_block (== srcSize).
+ * If srcSize is too small, the function will return an errorCode;
+ */
+static size_t FSE_initDStream(FSE_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) return (size_t)-FSE_ERROR_srcSize_wrong;
+
+    if (srcSize >=  sizeof(size_t))
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = FSE_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return (size_t)-FSE_ERROR_GENERIC;   /* stop bit not present */
+        bitD->bitsConsumed = 8 - FSE_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return (size_t)-FSE_ERROR_GENERIC;   /* stop bit not present */
+        bitD->bitsConsumed = 8 - FSE_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+/*!FSE_lookBits
+ * Provides next n bits from the bitContainer.
+ * bitContainer is not modified (bits are still present for next read/look)
+ * On 32-bits, maxNbBits==25
+ * On 64-bits, maxNbBits==57
+ * return : value extracted.
+ */
+static size_t FSE_lookBits(FSE_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+static size_t FSE_lookBitsFast(FSE_DStream_t* bitD, U32 nbBits)   /* only if nbBits >= 1 !! */
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+static void FSE_skipBits(FSE_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+
+/*!FSE_readBits
+ * Read next n bits from the bitContainer.
+ * On 32-bits, don't read more than maxNbBits==25
+ * On 64-bits, don't read more than maxNbBits==57
+ * Use the fast variant *only* if n >= 1.
+ * return : value extracted.
+ */
+static size_t FSE_readBits(FSE_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = FSE_lookBits(bitD, nbBits);
+    FSE_skipBits(bitD, nbBits);
+    return value;
+}
+
+static size_t FSE_readBitsFast(FSE_DStream_t* bitD, U32 nbBits)   /* only if nbBits >= 1 !! */
+{
+    size_t value = FSE_lookBitsFast(bitD, nbBits);
+    FSE_skipBits(bitD, nbBits);
+    return value;
+}
+
+static unsigned FSE_reloadDStream(FSE_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return FSE_DStream_tooFar;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = FSE_readLEST(bitD->ptr);
+        return FSE_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return FSE_DStream_endOfBuffer;
+        return FSE_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        U32 result = FSE_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = FSE_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = FSE_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+
+static void FSE_initDState(FSE_DState_t* DStatePtr, FSE_DStream_t* bitD, const FSE_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;
+    DStatePtr->state = FSE_readBits(bitD, DTableH->tableLog);
+    FSE_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+static BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, FSE_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = FSE_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+static BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, FSE_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = FSE_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/* FSE_endOfDStream
+   Tells if bitD has reached end of bitStream or not */
+
+static unsigned FSE_endOfDStream(const FSE_DStream_t* bitD)
+{
+    return ((bitD->ptr == bitD->start) && (bitD->bitsConsumed == sizeof(bitD->bitContainer)*8));
+}
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    FSE_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = FSE_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (FSE_reloadDStream(&bitD)==FSE_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            FSE_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (FSE_reloadDStream(&bitD) > FSE_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            FSE_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : FSE_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly FSE_DStream_completed */
+    while (1)
+    {
+        if ( (FSE_reloadDStream(&bitD)>FSE_DStream_completed) || (op==omax) || (FSE_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (FSE_reloadDStream(&bitD)>FSE_DStream_completed) || (op==omax) || (FSE_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (FSE_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall;   /* dst buffer is full, but cSrc unfinished */
+
+    return (size_t)-FSE_ERROR_corruptionDetected;
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));   /* memcpy() into local variable, to avoid strict aliasing warning */
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return (size_t)-FSE_ERROR_srcSize_wrong;   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+/* *******************************************************
+*  Huff0 : Huffman block compression
+*********************************************************/
+#define HUF_MAX_SYMBOL_VALUE 255
+#define HUF_DEFAULT_TABLELOG  12       /* used by default, when not specified */
+#define HUF_MAX_TABLELOG  12           /* max possible tableLog; for allocation purpose; can be modified */
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+typedef struct HUF_CElt_s {
+  U16  val;
+  BYTE nbBits;
+} HUF_CElt ;
+
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+
+/* *******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct {
+    BYTE byte;
+    BYTE nbBits;
+} HUF_DElt;
+
+static size_t HUF_readDTable (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];  /* large enough for values from 0 to 16 */
+    U32 weightTotal;
+    U32 maxBits;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DElt* const dt = (HUF_DElt*)ptr;
+
+    if (!srcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+    iSize = ip[0];
+
+    FSE_STATIC_ASSERT(sizeof(HUF_DElt) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* should not be necessary, but some analyzer complain ... */
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, sizeof(huffWeight));
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+        oSize = FSE_decompress(huffWeight, HUF_MAX_SYMBOL_VALUE, ip+1, iSize);   /* max 255 values decoded, last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankVal, 0, sizeof(rankVal));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return (size_t)-FSE_ERROR_corruptionDetected;
+        rankVal[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return (size_t)-FSE_ERROR_corruptionDetected;
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    maxBits = FSE_highbit32(weightTotal) + 1;
+    if (maxBits > DTable[0]) return (size_t)-FSE_ERROR_tableLog_tooLarge;   /* DTable is too small */
+    DTable[0] = (U16)maxBits;
+    {
+        U32 total = 1 << maxBits;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << FSE_highbit32(rest);
+        U32 lastWeight = FSE_highbit32(rest) + 1;
+        if (verif != rest) return (size_t)-FSE_ERROR_corruptionDetected;    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankVal[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankVal[1] < 2) || (rankVal[1] & 1)) return (size_t)-FSE_ERROR_corruptionDetected;   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=maxBits; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<=oSize; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DElt D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(maxBits + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize+1;
+}
+
+
+static BYTE HUF_decodeSymbol(FSE_DStream_t* Dstream, const HUF_DElt* dt, const U32 dtLog)
+{
+        const size_t val = FSE_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        FSE_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+static size_t HUF_decompress_usingDTable(   /* -3% slower when non static */
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-15;
+
+    const void* ptr = DTable;
+    const HUF_DElt* const dt = (const HUF_DElt*)(ptr)+1;
+    const U32 dtLog = DTable[0];
+    size_t errorCode;
+    U32 reloadStatus;
+
+    /* Init */
+
+    const U16* jumpTable = (const U16*)cSrc;
+    const size_t length1 = FSE_readLE16(jumpTable);
+    const size_t length2 = FSE_readLE16(jumpTable+1);
+    const size_t length3 = FSE_readLE16(jumpTable+2);
+    const size_t length4 = cSrcSize - 6 - length1 - length2 - length3;   // check coherency !!
+    const char* const start1 = (const char*)(cSrc) + 6;
+    const char* const start2 = start1 + length1;
+    const char* const start3 = start2 + length2;
+    const char* const start4 = start3 + length3;
+    FSE_DStream_t bitD1, bitD2, bitD3, bitD4;
+
+    if (length1+length2+length3+6 >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+
+    errorCode = FSE_initDStream(&bitD1, start1, length1);
+    if (FSE_isError(errorCode)) return errorCode;
+    errorCode = FSE_initDStream(&bitD2, start2, length2);
+    if (FSE_isError(errorCode)) return errorCode;
+    errorCode = FSE_initDStream(&bitD3, start3, length3);
+    if (FSE_isError(errorCode)) return errorCode;
+    errorCode = FSE_initDStream(&bitD4, start4, length4);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    reloadStatus=FSE_reloadDStream(&bitD2);
+
+    /* 16 symbols per loop */
+    for ( ; (reloadStatus<FSE_DStream_completed) && (op<olimit);  /* D2-3-4 are supposed to be synchronized and finish together */
+        op+=16, reloadStatus = FSE_reloadDStream(&bitD2) | FSE_reloadDStream(&bitD3) | FSE_reloadDStream(&bitD4), FSE_reloadDStream(&bitD1))
+    {
+#define HUF_DECODE_SYMBOL_0(n, Dstream) \
+        op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog);
+
+#define HUF_DECODE_SYMBOL_1(n, Dstream) \
+        op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
+        if (FSE_32bits() && (HUF_MAX_TABLELOG>12)) FSE_reloadDStream(&Dstream)
+
+#define HUF_DECODE_SYMBOL_2(n, Dstream) \
+        op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
+        if (FSE_32bits()) FSE_reloadDStream(&Dstream)
+
+        HUF_DECODE_SYMBOL_1( 0, bitD1);
+        HUF_DECODE_SYMBOL_1( 1, bitD2);
+        HUF_DECODE_SYMBOL_1( 2, bitD3);
+        HUF_DECODE_SYMBOL_1( 3, bitD4);
+        HUF_DECODE_SYMBOL_2( 4, bitD1);
+        HUF_DECODE_SYMBOL_2( 5, bitD2);
+        HUF_DECODE_SYMBOL_2( 6, bitD3);
+        HUF_DECODE_SYMBOL_2( 7, bitD4);
+        HUF_DECODE_SYMBOL_1( 8, bitD1);
+        HUF_DECODE_SYMBOL_1( 9, bitD2);
+        HUF_DECODE_SYMBOL_1(10, bitD3);
+        HUF_DECODE_SYMBOL_1(11, bitD4);
+        HUF_DECODE_SYMBOL_0(12, bitD1);
+        HUF_DECODE_SYMBOL_0(13, bitD2);
+        HUF_DECODE_SYMBOL_0(14, bitD3);
+        HUF_DECODE_SYMBOL_0(15, bitD4);
+    }
+
+    if (reloadStatus!=FSE_DStream_completed)   /* not complete : some bitStream might be FSE_DStream_unfinished */
+        return (size_t)-FSE_ERROR_corruptionDetected;
+
+    /* tail */
+    {
+        // bitTail = bitD1;   // *much* slower : -20% !??!
+        FSE_DStream_t bitTail;
+        bitTail.ptr = bitD1.ptr;
+        bitTail.bitsConsumed = bitD1.bitsConsumed;
+        bitTail.bitContainer = bitD1.bitContainer;   // required in case of FSE_DStream_endOfBuffer
+        bitTail.start = start1;
+        for ( ; (FSE_reloadDStream(&bitTail) < FSE_DStream_completed) && (op<omax) ; op++)
+        {
+            HUF_DECODE_SYMBOL_0(0, bitTail);
+        }
+
+        if (FSE_endOfDStream(&bitTail))
+            return op-ostart;
+    }
+
+    if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall;   /* dst buffer is full, but cSrc unfinished */
+
+    return (size_t)-FSE_ERROR_corruptionDetected;
+}
+
+
+static size_t HUF_decompress (void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLE(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTable (DTable, cSrc, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, DTable);
+}
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/****************************************************************
+*  Tuning parameters
+*****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect */
+#define ZSTD_MEMORY_USAGE 17
+
+
+/**************************************
+   CPU Feature Detection
+**************************************/
+/*
+ * Automated efficient unaligned memory access detection
+ * Based on known hardware architectures
+ * This list will be updated thanks to feedbacks
+ */
+#if defined(CPU_HAS_EFFICIENT_UNALIGNED_MEMORY_ACCESS) \
+    || defined(__ARM_FEATURE_UNALIGNED) \
+    || defined(__i386__) || defined(__x86_64__) \
+    || defined(_M_IX86) || defined(_M_X64) \
+    || defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_8__) \
+    || (defined(_M_ARM) && (_M_ARM >= 7))
+#  define ZSTD_UNALIGNED_ACCESS 1
+#else
+#  define ZSTD_UNALIGNED_ACCESS 0
+#endif
+
+
+/********************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/********************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+#ifndef MEM_ACCESS_MODULE
+#define MEM_ACCESS_MODULE
+/********************************************************
+*  Basic Types
+*********************************************************/
+#if defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+# include <stdint.h>
+typedef  uint8_t BYTE;
+typedef uint16_t U16;
+typedef  int16_t S16;
+typedef uint32_t U32;
+typedef  int32_t S32;
+typedef uint64_t U64;
+#else
+typedef unsigned char       BYTE;
+typedef unsigned short      U16;
+typedef   signed short      S16;
+typedef unsigned int        U32;
+typedef   signed int        S32;
+typedef unsigned long long  U64;
+#endif
+
+#endif   /* MEM_ACCESS_MODULE */
+
+
+/********************************************************
+*  Constants
+*********************************************************/
+static const U32 ZSTD_magicNumber = 0xFD2FB51E;   /* 3rd version : seqNb header */
+
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff ((1<<Offbits)-1)
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/********************************************************
+*  Memory operations
+*********************************************************/
+static unsigned ZSTD_32bits(void) { return sizeof(void*)==4; }
+
+static unsigned ZSTD_isLittleEndian(void)
+{
+    const union { U32 i; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+static U16    ZSTD_read16(const void* p) { U16 r; memcpy(&r, p, sizeof(r)); return r; }
+
+static U32    ZSTD_read32(const void* p) { U32 r; memcpy(&r, p, sizeof(r)); return r; }
+
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s)    { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    while (op < oend) COPY8(op, ip);
+}
+
+static U16 ZSTD_readLE16(const void* memPtr)
+{
+    if (ZSTD_isLittleEndian()) return ZSTD_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)((U16)p[0] + ((U16)p[1]<<8));
+    }
+}
+
+
+static U32 ZSTD_readLE32(const void* memPtr)
+{
+    if (ZSTD_isLittleEndian())
+        return ZSTD_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+static U32 ZSTD_readBE32(const void* memPtr)
+{
+    const BYTE* p = (const BYTE*)memPtr;
+    return (U32)(((U32)p[0]<<24) + ((U32)p[1]<<16) + ((U32)p[2]<<8) + ((U32)p[3]<<0));
+}
+
+
+/**************************************
+*  Local structures
+***************************************/
+typedef struct ZSTD_Cctx_s ZSTD_Cctx;
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} seqStore_t;
+
+
+typedef struct ZSTD_Cctx_s
+{
+    const BYTE* base;
+    U32 current;
+    U32 nextUpdate;
+    seqStore_t seqStore;
+#ifdef __AVX2__
+    __m256i hashTable[HASH_TABLESIZE>>3];
+#else
+    U32 hashTable[HASH_TABLESIZE];
+#endif
+    BYTE buffer[WORKPLACESIZE];
+} cctxi_t;
+
+
+
+
+/**************************************
+*  Error Management
+**************************************/
+/* published entry point */
+unsigned ZSTDv01_isError(size_t code) { return ERR_isError(code); }
+
+
+/**************************************
+*  Tool functions
+**************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    1    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  3    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+/**************************************************************
+*   Decompression code
+**************************************************************/
+
+size_t ZSTDv01_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+static size_t ZSTD_decompressLiterals(void* ctx,
+                                      void* dst, size_t maxDstSize,
+                                const void* src, size_t srcSize)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + maxDstSize;
+    const BYTE* ip = (const BYTE*)src;
+    size_t errorCode;
+    size_t litSize;
+
+    /* check : minimum 2, for litSize, +1, for content */
+    if (srcSize <= 3) return ERROR(corruption_detected);
+
+    litSize = ip[1] + (ip[0]<<8);
+    litSize += ((ip[-3] >> 3) & 7) << 16;   // mmmmh....
+    op = oend - litSize;
+
+    (void)ctx;
+    if (litSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    errorCode = HUF_decompress(op, litSize, ip+2, srcSize-2);
+    if (FSE_isError(errorCode)) return ERROR(GENERIC);
+    return litSize;
+}
+
+
+size_t ZSTDv01_decodeLiteralsBlock(void* ctx,
+                                void* dst, size_t maxDstSize,
+                          const BYTE** litStart, size_t* litSize,
+                          const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    blockProperties_t litbp;
+
+    size_t litcSize = ZSTDv01_getcBlockSize(src, srcSize, &litbp);
+    if (ZSTDv01_isError(litcSize)) return litcSize;
+    if (litcSize > srcSize - ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    ip += ZSTD_blockHeaderSize;
+
+    switch(litbp.blockType)
+    {
+    case bt_raw:
+        *litStart = ip;
+        ip += litcSize;
+        *litSize = litcSize;
+        break;
+    case bt_rle:
+        {
+            size_t rleSize = litbp.origSize;
+            if (rleSize>maxDstSize) return ERROR(dstSize_tooSmall);
+            if (!srcSize) return ERROR(srcSize_wrong);
+            memset(oend - rleSize, *ip, rleSize);
+            *litStart = oend - rleSize;
+            *litSize = rleSize;
+            ip++;
+            break;
+        }
+    case bt_compressed:
+        {
+            size_t decodedLitSize = ZSTD_decompressLiterals(ctx, dst, maxDstSize, ip, litcSize);
+            if (ZSTDv01_isError(decodedLitSize)) return decodedLitSize;
+            *litStart = oend - decodedLitSize;
+            *litSize = decodedLitSize;
+            ip += litcSize;
+            break;
+        }
+    case bt_end:
+    default:
+        return ERROR(GENERIC);
+    }
+
+    return ip-istart;
+}
+
+
+size_t ZSTDv01_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = ZSTD_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++); break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    FSE_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) litLength += add;
+        else
+        {
+            if (dumps<=(de-3))
+            {
+                litLength = ZSTD_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+                dumps += 3;
+            }
+        }
+    }
+
+    /* Offset */
+    {
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));
+        if (ZSTD_32bits()) FSE_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = ((size_t)1 << (nbBits & ((sizeof(offset)*8)-1))) + FSE_readBits(&(seqState->DStream), nbBits);
+        if (ZSTD_32bits()) FSE_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        U32 add = dumps<de ? *dumps++ : 0;
+        if (add < 255) matchLength += add;
+        else
+        {
+            if (dumps<=(de-3))
+            {
+                matchLength = ZSTD_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+                dumps += 3;
+            }
+        }
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* substracted */
+    const BYTE* const ostart = op;
+    const size_t litLength = sequence.litLength;
+    BYTE* const endMatch = op + litLength + sequence.matchLength;    /* risk : address space overflow (32-bits) */
+    const BYTE* const litEnd = *litPtr + litLength;
+
+    /* check */
+    if (endMatch > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);
+    if (sequence.matchLength > (size_t)(*litPtr-op))  return ERROR(dstSize_tooSmall);    /* overwrite literal segment */
+
+    /* copy Literals */
+    if (((size_t)(*litPtr - op) < 8) || ((size_t)(oend-litEnd) < 8) || (op+litLength > oend-8))
+        memmove(op, *litPtr, litLength);   /* overwrite risk */
+    else
+        ZSTD_wildcopy(op, *litPtr, litLength);
+    op += litLength;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* check : last match must be at a minimum distance of 8 from end of dest buffer */
+    if (oend-op < 8) return ERROR(dstSize_tooSmall);
+
+    /* copy Match */
+    {
+        const U32 overlapRisk = (((size_t)(litEnd - endMatch)) < 12);
+        const BYTE* match = op - sequence.offset;            /* possible underflow at op - offset ? */
+        size_t qutt = 12;
+        U64 saved[2];
+
+        /* check */
+        if (match < base) return ERROR(corruption_detected);
+        if (sequence.offset > (size_t)base) return ERROR(corruption_detected);
+
+        /* save beginning of literal sequence, in case of write overlap */
+        if (overlapRisk)
+        {
+            if ((endMatch + qutt) > oend) qutt = oend-endMatch;
+            memcpy(saved, endMatch, qutt);
+        }
+
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        } else { ZSTD_copy8(op, match); }
+        op += 8; match += 8;
+
+        if (endMatch > oend-(16-MINMATCH))
+        {
+            if (op < oend-8)
+            {
+                ZSTD_wildcopy(op, match, (oend-8) - op);
+                match += (oend-8) - op;
+                op = oend-8;
+            }
+            while (op<endMatch) *op++ = *match++;
+        }
+        else
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+
+        /* restore, in case of overlap */
+        if (overlapRisk) memcpy(endMatch, saved, qutt);
+    }
+
+    return endMatch-ostart;
+}
+
+typedef struct ZSTDv01_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+} dctx_t;
+
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize,
+                         const BYTE* litStart, size_t litSize)
+{
+    dctx_t* dctx = (dctx_t*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = litStart;
+    const BYTE* const litEnd = litStart + litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTDv01_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTDv01_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 1;
+        errorCode = FSE_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (FSE_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (FSE_reloadDStream(&(seqState.DStream)) <= FSE_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTDv01_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !FSE_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (op != litPtr) memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed, srcSize is trusted */
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* litPtr = NULL;
+    size_t litSize = 0;
+    size_t errorCode;
+
+    /* Decode literals sub-block */
+    errorCode = ZSTDv01_decodeLiteralsBlock(ctx, dst, maxDstSize, &litPtr, &litSize, src, srcSize);
+    if (ZSTDv01_isError(errorCode)) return errorCode;
+    ip += errorCode;
+    srcSize -= errorCode;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize, litPtr, litSize);
+}
+
+
+size_t ZSTDv01_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    size_t errorCode=0;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = ZSTD_readBE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t blockSize = ZSTDv01_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv01_isError(blockSize)) return blockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (blockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            errorCode = ZSTD_decompressBlock(ctx, op, oend-op, ip, blockSize);
+            break;
+        case bt_raw :
+            errorCode = ZSTD_copyUncompressedBlock(op, oend-op, ip, blockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        if (blockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv01_isError(errorCode)) return errorCode;
+        op += errorCode;
+        ip += blockSize;
+        remainingSize -= blockSize;
+    }
+
+    return op-ostart;
+}
+
+size_t ZSTDv01_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    dctx_t ctx;
+    ctx.base = dst;
+    return ZSTDv01_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+size_t ZSTDv01_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = ZSTD_readBE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t blockSize = ZSTDv01_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv01_isError(blockSize)) return blockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (blockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        if (blockSize == 0) break;   /* bt_end */
+
+        ip += blockSize;
+        remainingSize -= blockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+size_t ZSTDv01_resetDCtx(ZSTDv01_Dctx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+ZSTDv01_Dctx* ZSTDv01_createDCtx(void)
+{
+    ZSTDv01_Dctx* dctx = (ZSTDv01_Dctx*)malloc(sizeof(ZSTDv01_Dctx));
+    if (dctx==NULL) return NULL;
+    ZSTDv01_resetDCtx(dctx);
+    return dctx;
+}
+
+size_t ZSTDv01_freeDCtx(ZSTDv01_Dctx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+size_t ZSTDv01_nextSrcSizeToDecompress(ZSTDv01_Dctx* dctx)
+{
+    return ((dctx_t*)dctx)->expected;
+}
+
+size_t ZSTDv01_decompressContinue(ZSTDv01_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    dctx_t* ctx = (dctx_t*)dctx;
+
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = ZSTD_readBE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTDv01_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTDv01_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_v01.h b/vendor/github.com/DataDog/zstd/zstd_v01.h
new file mode 100644
index 0000000..42f0897
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v01.h
@@ -0,0 +1,89 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V01_H_28739879432
+#define ZSTD_V01_H_28739879432
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv01_decompress() : decompress ZSTD frames compliant with v0.1.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv01_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+/**
+ZSTDv01_getFrameSrcSize() : get the source length of a ZSTD frame compliant with v0.1.x format
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv01_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+/**
+ZSTDv01_isError() : tells if the result of ZSTDv01_decompress() is an error
+*/
+unsigned ZSTDv01_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv01_Dctx_s ZSTDv01_Dctx;
+ZSTDv01_Dctx* ZSTDv01_createDCtx(void);
+size_t ZSTDv01_freeDCtx(ZSTDv01_Dctx* dctx);
+
+size_t ZSTDv01_decompressDCtx(void* ctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+/* *************************************
+*  Streaming functions
+***************************************/
+size_t ZSTDv01_resetDCtx(ZSTDv01_Dctx* dctx);
+
+size_t ZSTDv01_nextSrcSizeToDecompress(ZSTDv01_Dctx* dctx);
+size_t ZSTDv01_decompressContinue(ZSTDv01_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv01_magicNumber   0xFD2FB51E   /* Big Endian version */
+#define ZSTDv01_magicNumberLE 0x1EB52FFD   /* Little Endian version */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V01_H_28739879432 */
diff --git a/vendor/github.com/DataDog/zstd/zstd_v02.c b/vendor/github.com/DataDog/zstd/zstd_v02.c
new file mode 100644
index 0000000..8bc0ece
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v02.c
@@ -0,0 +1,3483 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v02.h"
+#include "error_private.h"
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets generating assembly depending on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif // MEM_FORCE_MEMORY_ACCESS
+
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r=0;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   Error codes and messages
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#elif defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define PREFIX(name) ZSTD_error_##name
+
+#define ERROR(name) (size_t)-PREFIX(name)
+
+#define ERROR_LIST(ITEM) \
+        ITEM(PREFIX(No_Error)) ITEM(PREFIX(GENERIC)) \
+        ITEM(PREFIX(dstSize_tooSmall)) ITEM(PREFIX(srcSize_wrong)) \
+        ITEM(PREFIX(prefix_unknown)) ITEM(PREFIX(corruption_detected)) \
+        ITEM(PREFIX(tableLog_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooSmall)) \
+        ITEM(PREFIX(maxCode))
+
+#define ERROR_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { ERROR_LIST(ERROR_GENERATE_ENUM) } ERR_codes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+#define ERROR_CONVERTTOSTRING(STRING) #STRING,
+#define ERROR_GENERATE_STRING(EXPR) ERROR_CONVERTTOSTRING(EXPR)
+static const char* ERR_strings[] = { ERROR_LIST(ERROR_GENERATE_STRING) };
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    static const char* codeError = "Unspecified error code";
+    if (ERR_isError(code)) return ERR_strings[-(int)(code)];
+    return codeError;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
+/*
+Constructor and Destructor of type FSE_CTable
+    Note that its size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/******************************************
+*  FSE advanced API
+******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+/******************************************
+*  FSE symbol decompression API
+******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/******************************************
+*  FSE unsafe API
+******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/******************************************
+*  Implementation of inline functions
+******************************************/
+
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Static allocation macros
+******************************************/
+/* Huff0 buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/******************************************
+*  Advanced functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* quad-symbols decoder */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*
+    zstd - standard compression library
+    Header File
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Version
+***************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    2    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
+
+#if defined (__cplusplus)
+}
+#endif
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Streaming functions
+***************************************/
+
+typedef struct ZSTD_DCtx_s ZSTD_DCtx;
+
+/*
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTD_magicNumber 0xFD2FB522   /* v0.2 (current)*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt+1;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)ptr;
+    FSE_DTableHeader DTableH;
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));   /* memcpy(), to avoid strict aliasing warnings */
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return (short)(a<0 ? -a : a);
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;   /* because dt is unsigned */
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/******************************************
+*  Helper functions
+******************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*********************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)ptr;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    void* ptr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)ptr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        {if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* quad-symbol decoding           */
+/**********************************/
+typedef struct { BYTE nbBits; BYTE nbBytes; } HUF_DDescX6;
+typedef union { BYTE byte[4]; U32 sequence; } HUF_DSeqX6;
+
+/* recursive, up to level 3; may benefit from <template>-like strategy to nest each level inline */
+static void HUF_fillDTableX6LevelN(HUF_DDescX6* DDescription, HUF_DSeqX6* DSequence, int sizeLog,
+                           const rankVal_t rankValOrigin, const U32 consumed, const int minWeight, const U32 maxWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize, const U32* rankStart,
+                           const U32 nbBitsBaseline, HUF_DSeqX6 baseSeq, HUF_DDescX6 DDesc)
+{
+    const int scaleLog = nbBitsBaseline - sizeLog;   /* note : targetLog >= (nbBitsBaseline-1), hence scaleLog <= 1 */
+    const int minBits  = nbBitsBaseline - maxWeight;
+    const U32 level = DDesc.nbBytes;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 symbolStartPos, s;
+
+    /* local rankVal, will be modified */
+    memcpy(rankVal, rankValOrigin[consumed], sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i;
+        const U32 skipSize = rankVal[minWeight];
+        for (i = 0; i < skipSize; i++)
+        {
+            DSequence[i] = baseSeq;
+            DDescription[i] = DDesc;
+        }
+    }
+
+    /* fill DTable */
+    DDesc.nbBytes++;
+    symbolStartPos = rankStart[minWeight];
+    for (s=symbolStartPos; s<sortedListSize; s++)
+    {
+        const BYTE symbol = sortedSymbols[s].symbol;
+        const U32  weight = sortedSymbols[s].weight;   /* >= 1 (sorted) */
+        const int  nbBits = nbBitsBaseline - weight;   /* >= 1 (by construction) */
+        const int  totalBits = consumed+nbBits;
+        const U32  start  = rankVal[weight];
+        const U32  length = 1 << (sizeLog-nbBits);
+        baseSeq.byte[level] = symbol;
+        DDesc.nbBits = (BYTE)totalBits;
+
+        if ((level<3) && (sizeLog-totalBits >= minBits))   /* enough room for another symbol */
+        {
+            int nextMinWeight = totalBits + scaleLog;
+            if (nextMinWeight < 1) nextMinWeight = 1;
+            HUF_fillDTableX6LevelN(DDescription+start, DSequence+start, sizeLog-nbBits,
+                           rankValOrigin, totalBits, nextMinWeight, maxWeight,
+                           sortedSymbols, sortedListSize, rankStart,
+                           nbBitsBaseline, baseSeq, DDesc);   /* recursive (max : level 3) */
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            for (i = start; i < end; i++)
+            {
+                DDescription[i] = DDesc;
+                DSequence[i] = baseSeq;
+            }
+        }
+        rankVal[weight] += length;
+    }
+}
+
+
+/* note : same preparation as X4 */
+static size_t HUF_readDTableX6 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    rankVal_t rankVal;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+
+    /* fill tables */
+    {
+        void* ptr = DTable+1;
+        HUF_DDescX6* DDescription = (HUF_DDescX6*)(ptr);
+        void* dSeqStart = DTable + 1 + ((size_t)1<<(memLog-1));
+        HUF_DSeqX6* DSequence = (HUF_DSeqX6*)(dSeqStart);
+        HUF_DSeqX6 DSeq;
+        HUF_DDescX6 DDesc;
+        DSeq.sequence = 0;
+        DDesc.nbBits = 0;
+        DDesc.nbBytes = 0;
+        HUF_fillDTableX6LevelN(DDescription, DSequence, memLog,
+                       (const U32 (*)[HUF_ABSOLUTEMAX_TABLELOG + 1])rankVal, 0, 1, maxW,
+                       sortedSymbol, sizeOfSort, rankStart0,
+                       tableLog+1, DSeq, DDesc);
+    }
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX6(void* op, BIT_DStream_t* DStream, const HUF_DDescX6* dd, const HUF_DSeqX6* ds, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, ds+val, sizeof(HUF_DSeqX6));
+    BIT_skipBits(DStream, dd[val].nbBits);
+    return dd[val].nbBytes;
+}
+
+static U32 HUF_decodeLastSymbolsX6(void* op, const U32 maxL, BIT_DStream_t* DStream,
+                                  const HUF_DDescX6* dd, const HUF_DSeqX6* ds, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    U32 length = dd[val].nbBytes;
+    if (length <= maxL)
+    {
+        memcpy(op, ds+val, length);
+        BIT_skipBits(DStream, dd[val].nbBits);
+        return length;
+    }
+    memcpy(op, ds+val, maxL);
+    if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+    {
+        BIT_skipBits(DStream, dd[val].nbBits);
+        if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+            DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }
+    return maxL;
+}
+
+
+#define HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX6(ptr, DStreamPtr, dd, ds, dtLog)
+
+#define HUF_DECODE_SYMBOLX6_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX6_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX6_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX6(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const U32* DTable, const U32 dtLog)
+{
+    const void* ddPtr = DTable+1;
+    const HUF_DDescX6* dd = (const HUF_DDescX6*)(ddPtr);
+    const void* dsPtr = DTable + 1 + ((size_t)1<<(dtLog-1));
+    const HUF_DSeqX6* ds = (const HUF_DSeqX6*)(dsPtr);
+    BYTE* const pStart = p;
+
+    /* up to 16 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-16))
+    {
+        HUF_DECODE_SYMBOLX6_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);
+    }
+
+    /* closer to the end, up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);
+
+    while (p <= pEnd-4)
+        HUF_DECODE_SYMBOLX6_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    while (p < pEnd)
+        p += HUF_decodeLastSymbolsX6(p, (U32)(pEnd-p), bitDPtr, dd, ds, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X6_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const U32 dtLog = DTable[0];
+        const void* ddPtr = DTable+1;
+        const HUF_DDescX6* dd = (const HUF_DDescX6*)(ddPtr);
+        const void* dsPtr = DTable + 1 + ((size_t)1<<(dtLog-1));
+        const HUF_DSeqX6* ds = (const HUF_DSeqX6*)(dsPtr);
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-64 symbols per loop (4-16 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (op3 <= opStart4) && (endSignal==BIT_DStream_unfinished) && (op4<=(oend-16)) ; )
+        {
+            HUF_DECODE_SYMBOLX6_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX6_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX6_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX6_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX6_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX6(op1, &bitD1, opStart2, DTable, dtLog);
+        HUF_decodeStreamX6(op2, &bitD2, opStart3, DTable, dtLog);
+        HUF_decodeStreamX6(op3, &bitD3, opStart4, DTable, dtLog);
+        HUF_decodeStreamX6(op4, &bitD4, oend,     DTable, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X6 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX6(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX6 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X6_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, HUF_decompress4X6 };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+    if (Dtime[2] < Dtime[algoNb]) algoNb = 2;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+*  MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*/
+#define ZSTD_MEMORY_USAGE 17
+
+/*!
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0, fastest), or in memory heap (1, requires malloc())
+ * Note that compression context is fairly large, as a consequence heap memory is recommended.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif /* ZSTD_HEAPMODE */
+
+/*!
+*  LEGACY_SUPPORT :
+*  decompressor can decode older formats (starting from Zstd 0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/* *******************************************************
+*  Constants
+*********************************************************/
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff   31
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do COPY8(op, ip) while (op < oend);
+}
+
+
+/* **************************************
+*  Local structures
+****************************************/
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} seqStore_t;
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+struct ZSTD_DCtx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+};   /* typedef'd to ZSTD_Dctx within "zstd_static.h" */
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize )*/
+static size_t ZSTD_decodeLiteralsBlock(void* ctx,
+                          const void* src, size_t srcSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* const istart = (const BYTE* const)src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    default:
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;
+        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        U32 add = *dumps++;
+        if (add < 255) litLength += add;
+        else
+        {
+            litLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const size_t offsetPrefix[MaxOff+1] = {  /* note : size_t faster than U32 */
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        U32 add = *dumps++;
+        if (add < 255) matchLength += add;
+        else
+        {
+            matchLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* substracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    BYTE* const oMatchEnd = op + sequence.litLength + sequence.matchLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+
+    /* checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    {
+        const BYTE* match = op - sequence.offset;
+
+        /* check */
+        if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
+        //if (match > op) return ERROR(corruption_detected);   /* address space overflow test (is clang optimizer removing this test ?) */
+        if (match < base) return ERROR(corruption_detected);
+
+        /* close range match, overlap */
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        }
+        else
+        {
+            ZSTD_copy8(op, match);
+        }
+        op += 8; match += 8;
+
+        if (oMatchEnd > oend-(16-MINMATCH))
+        {
+            if (op < oend_8)
+            {
+                ZSTD_wildcopy(op, match, oend_8 - op);
+                match += oend_8 - op;
+                op = oend_8;
+            }
+            while (op < oMatchEnd) *op++ = *match++;
+        }
+        else
+        {
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        }
+    }
+
+    return oMatchEnd - ostart;
+}
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 1;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (op != litPtr) memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(ctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyUncompressedBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    ZSTD_DCtx ctx;
+    ctx.base = dst;
+    return ZSTD_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+static size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTD_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
+
+
+/* wrapper layer */
+
+unsigned ZSTDv02_isError(size_t code)
+{
+    return ZSTD_isError(code);
+}
+
+size_t ZSTDv02_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize)
+{
+    return ZSTD_decompress(dst, maxOriginalSize, src, compressedSize);
+}
+
+size_t ZSTDv02_findFrameCompressedSize(const void *src, size_t compressedSize)
+{
+    return ZSTD_findFrameCompressedSize(src, compressedSize);
+}
+
+ZSTDv02_Dctx* ZSTDv02_createDCtx(void)
+{
+    return (ZSTDv02_Dctx*)ZSTD_createDCtx();
+}
+
+size_t ZSTDv02_freeDCtx(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_freeDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_resetDCtx(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_resetDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_nextSrcSizeToDecompress(ZSTDv02_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv02_decompressContinue(ZSTDv02_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue((ZSTD_DCtx*)dctx, dst, maxDstSize, src, srcSize);
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_v02.h b/vendor/github.com/DataDog/zstd/zstd_v02.h
new file mode 100644
index 0000000..0dde7a6
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v02.h
@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V02_H_4174539423
+#define ZSTD_V02_H_4174539423
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv02_decompress() : decompress ZSTD frames compliant with v0.2.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv02_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+/**
+ZSTDv02_getFrameSrcSize() : get the source length of a ZSTD frame compliant with v0.2.x format
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv02_isError())
+*/
+size_t ZSTDv02_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+/**
+ZSTDv02_isError() : tells if the result of ZSTDv02_decompress() is an error
+*/
+unsigned ZSTDv02_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv02_Dctx_s ZSTDv02_Dctx;
+ZSTDv02_Dctx* ZSTDv02_createDCtx(void);
+size_t ZSTDv02_freeDCtx(ZSTDv02_Dctx* dctx);
+
+size_t ZSTDv02_decompressDCtx(void* ctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+/* *************************************
+*  Streaming functions
+***************************************/
+size_t ZSTDv02_resetDCtx(ZSTDv02_Dctx* dctx);
+
+size_t ZSTDv02_nextSrcSizeToDecompress(ZSTDv02_Dctx* dctx);
+size_t ZSTDv02_decompressContinue(ZSTDv02_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv02_magicNumber 0xFD2FB522   /* v0.2 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V02_H_4174539423 */
diff --git a/vendor/github.com/DataDog/zstd/zstd_v03.c b/vendor/github.com/DataDog/zstd/zstd_v03.c
new file mode 100644
index 0000000..54445af
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v03.c
@@ -0,0 +1,3124 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include "zstd_v03.h"
+#include "error_private.h"
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets generating assembly depending on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+
+#endif // MEM_FORCE_MEMORY_ACCESS
+
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r=0;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24;
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16;
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8;
+            default:;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   Error codes and messages
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef ERROR_H_MODULE
+#define ERROR_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define ERR_STATIC static inline
+#elif defined(_MSC_VER)
+#  define ERR_STATIC static __inline
+#elif defined(__GNUC__)
+#  define ERR_STATIC static __attribute__((unused))
+#else
+#  define ERR_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/******************************************
+*  Error Management
+******************************************/
+#define PREFIX(name) ZSTD_error_##name
+
+#define ERROR(name) (size_t)-PREFIX(name)
+
+#define ERROR_LIST(ITEM) \
+        ITEM(PREFIX(No_Error)) ITEM(PREFIX(GENERIC)) \
+        ITEM(PREFIX(dstSize_tooSmall)) ITEM(PREFIX(srcSize_wrong)) \
+        ITEM(PREFIX(prefix_unknown)) ITEM(PREFIX(corruption_detected)) \
+        ITEM(PREFIX(tableLog_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooLarge)) ITEM(PREFIX(maxSymbolValue_tooSmall)) \
+        ITEM(PREFIX(maxCode))
+
+#define ERROR_GENERATE_ENUM(ENUM) ENUM,
+typedef enum { ERROR_LIST(ERROR_GENERATE_ENUM) } ERR_codes;  /* enum is exposed, to detect & handle specific errors; compare function result to -enum value */
+
+#define ERROR_CONVERTTOSTRING(STRING) #STRING,
+#define ERROR_GENERATE_STRING(EXPR) ERROR_CONVERTTOSTRING(EXPR)
+static const char* ERR_strings[] = { ERROR_LIST(ERROR_GENERATE_STRING) };
+
+ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
+
+ERR_STATIC const char* ERR_getErrorName(size_t code)
+{
+    static const char* codeError = "Unspecified error code";
+    if (ERR_isError(code)) return ERR_strings[-(int)(code)];
+    return codeError;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ERROR_H_MODULE */
+/*
+Constructor and Destructor of type FSE_CTable
+    Note that its size depends on 'tableLog' and 'maxSymbolValue' */
+typedef unsigned FSE_CTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/******************************************
+*  Static allocation
+******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* You can statically allocate FSE CTable/DTable as a table of unsigned using below macro */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/******************************************
+*  FSE advanced API
+******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+/******************************************
+*  FSE symbol decompression API
+******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/******************************************
+*  FSE unsafe API
+******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/******************************************
+*  Implementation of inline functions
+******************************************/
+
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Static allocation macros
+******************************************/
+/* Huff0 buffer bounds */
+#define HUF_CTABLEBOUND 129
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/******************************************
+*  Advanced functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/*
+    zstd - standard compression library
+    Header File
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Version
+***************************************/
+#define ZSTD_VERSION_MAJOR    0    /* for breaking interface changes  */
+#define ZSTD_VERSION_MINOR    2    /* for new (non-breaking) interface capabilities */
+#define ZSTD_VERSION_RELEASE  2    /* for tweaks, bug-fixes, or development */
+#define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTD_CCtx_s ZSTD_CCtx;   /* incomplete type */
+
+#if defined (__cplusplus)
+}
+#endif
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Streaming functions
+***************************************/
+
+typedef struct ZSTD_DCtx_s ZSTD_DCtx;
+
+/*
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTD_magicNumber 0xFD2FB523   /* v0.3 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/****************************************************************
+*  Tuning parameters
+****************************************************************/
+/* MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/* FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/****************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+
+
+/****************************************************************
+*  Byte symbol type
+****************************************************************/
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/****************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/****************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+#define FSE_DECODE_TYPE FSE_decode_t
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+static size_t FSE_buildDTable
+(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* ptr = dt+1;
+    FSE_DTableHeader DTableH;
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*)ptr;
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return a<0 ? -a : a;
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const cell = (FSE_decode_t*)(ptr) + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)(ptr) + 1;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+
+    /* select fast mode (static) */
+    if (DTableH.fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/****************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+/****************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+/****************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/******************************************
+*  Helper functions
+******************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*********************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* ptr = DTable+1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)(ptr);
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize = ip[0];
+    void* ptr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)ptr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+
+        const void* ptr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)ptr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+*  MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*/
+#define ZSTD_MEMORY_USAGE 17
+
+/*!
+ * HEAPMODE :
+ * Select how default compression functions will allocate memory for their hash table,
+ * in memory stack (0, fastest), or in memory heap (1, requires malloc())
+ * Note that compression context is fairly large, as a consequence heap memory is recommended.
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif /* ZSTD_HEAPMODE */
+
+/*!
+*  LEGACY_SUPPORT :
+*  decompressor can decode older formats (starting from Zstd 0.1+)
+*/
+#ifndef ZSTD_LEGACY_SUPPORT
+#  define ZSTD_LEGACY_SUPPORT 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef __AVX2__
+#  include <immintrin.h>   /* AVX2 intrinsics */
+#endif
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#else
+#  define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#endif
+
+
+/* *******************************************************
+*  Constants
+*********************************************************/
+#define HASH_LOG (ZSTD_MEMORY_USAGE - 2)
+#define HASH_TABLESIZE (1 << HASH_LOG)
+#define HASH_MASK (HASH_TABLESIZE - 1)
+
+#define KNUTH 2654435761
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define WORKPLACESIZE (BLOCKSIZE*3)
+#define MINMATCH 4
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits )-1)
+#define MaxLL  ((1<<LLbits )-1)
+#define MaxOff   31
+#define LitFSELog  11
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MAX(a,b) ((a)<(b)?(b):(a))
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define LITERAL_NOENTROPY 63
+#define COMMAND_NOENTROPY 7   /* to remove */
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize = 4;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do COPY8(op, ip) while (op < oend);
+}
+
+
+/* **************************************
+*  Local structures
+****************************************/
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+} seqStore_t;
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+struct ZSTD_DCtx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    void* previousDstEnd;
+    void* base;
+    size_t expected;
+    blockType_t bType;
+    U32 phase;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+};   /* typedef'd to ZSTD_Dctx within "zstd_static.h" */
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyUncompressedBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize )*/
+static size_t ZSTD_decodeLiteralsBlock(void* ctx,
+                          const void* src, size_t srcSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* const istart = (const BYTE* const)src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    default:
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;
+        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL and MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    seqState->prevOffset = seq->offset;
+    if (litLength == MaxLL)
+    {
+        U32 add = *dumps++;
+        if (add < 255) litLength += add;
+        else
+        {
+            litLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const size_t offsetPrefix[MaxOff+1] = {  /* note : size_t faster than U32 */
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML)
+    {
+        U32 add = *dumps++;
+        if (add < 255) matchLength += add;
+        else
+        {
+            matchLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            dumps += 3;
+        }
+        if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                BYTE* const base, BYTE* const oend)
+{
+    static const int dec32table[] = {0, 1, 2, 1, 4, 4, 4, 4};   /* added */
+    static const int dec64table[] = {8, 8, 8, 7, 8, 9,10,11};   /* substracted */
+    const BYTE* const ostart = op;
+    BYTE* const oLitEnd = op + sequence.litLength;
+    BYTE* const oMatchEnd = op + sequence.litLength + sequence.matchLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+
+    /* checks */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* overRead beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    {
+        const BYTE* match = op - sequence.offset;
+
+        /* check */
+        if (sequence.offset > (size_t)op) return ERROR(corruption_detected);   /* address space overflow test (this test seems kept by clang optimizer) */
+        //if (match > op) return ERROR(corruption_detected);   /* address space overflow test (is clang optimizer removing this test ?) */
+        if (match < base) return ERROR(corruption_detected);
+
+        /* close range match, overlap */
+        if (sequence.offset < 8)
+        {
+            const int dec64 = dec64table[sequence.offset];
+            op[0] = match[0];
+            op[1] = match[1];
+            op[2] = match[2];
+            op[3] = match[3];
+            match += dec32table[sequence.offset];
+            ZSTD_copy4(op+4, match);
+            match -= dec64;
+        }
+        else
+        {
+            ZSTD_copy8(op, match);
+        }
+        op += 8; match += 8;
+
+        if (oMatchEnd > oend-(16-MINMATCH))
+        {
+            if (op < oend_8)
+            {
+                ZSTD_wildcopy(op, match, oend_8 - op);
+                match += oend_8 - op;
+                op = oend_8;
+            }
+            while (op < oMatchEnd) *op++ = *match++;
+        }
+        else
+        {
+            ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        }
+    }
+
+    return oMatchEnd - ostart;
+}
+
+static size_t ZSTD_decompressSequences(
+                               void* ctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)ctx;
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    BYTE* const base = (BYTE*) (dctx->base);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = sequence.offset = 4;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (nbSeq>0) ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, sequence, &litPtr, litEnd, base, oend);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* requested too much : data is corrupted */
+        if (nbSeq<0) return ERROR(corruption_detected);   /* requested too many sequences : data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (op != litPtr) memmove(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static size_t ZSTD_decompressBlock(
+                            void* ctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(ctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(ctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompressDCtx(void* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyUncompressedBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    ZSTD_DCtx ctx;
+    ctx.base = dst;
+    return ZSTD_decompressDCtx(&ctx, dst, maxDstSize, src, srcSize);
+}
+
+static size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    U32 magicNumber;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize; remainingSize -= ZSTD_frameHeaderSize;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+
+/*******************************
+*  Streaming Decompression API
+*******************************/
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize;
+    dctx->phase = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    if (dst != ctx->previousDstEnd)  /* not contiguous */
+        ctx->base = dst;
+
+    /* Decompress : frame header */
+    if (ctx->phase == 0)
+    {
+        /* Check frame magic header */
+        U32 magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_magicNumber) return ERROR(prefix_unknown);
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        return 0;
+    }
+
+    /* Decompress : block header */
+    if (ctx->phase == 1)
+    {
+        blockProperties_t bp;
+        size_t blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+        if (ZSTD_isError(blockSize)) return blockSize;
+        if (bp.blockType == bt_end)
+        {
+            ctx->expected = 0;
+            ctx->phase = 0;
+        }
+        else
+        {
+            ctx->expected = blockSize;
+            ctx->bType = bp.blockType;
+            ctx->phase = 2;
+        }
+
+        return 0;
+    }
+
+    /* Decompress : block content */
+    {
+        size_t rSize;
+        switch(ctx->bType)
+        {
+        case bt_compressed:
+            rSize = ZSTD_decompressBlock(ctx, dst, maxDstSize, src, srcSize);
+            break;
+        case bt_raw :
+            rSize = ZSTD_copyUncompressedBlock(dst, maxDstSize, src, srcSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet handled */
+            break;
+        case bt_end :   /* should never happen (filtered at phase 1) */
+            rSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);
+        }
+        ctx->phase = 1;
+        ctx->expected = ZSTD_blockHeaderSize;
+        ctx->previousDstEnd = (void*)( ((char*)dst) + rSize);
+        return rSize;
+    }
+
+}
+
+
+/* wrapper layer */
+
+unsigned ZSTDv03_isError(size_t code)
+{
+    return ZSTD_isError(code);
+}
+
+size_t ZSTDv03_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize)
+{
+    return ZSTD_decompress(dst, maxOriginalSize, src, compressedSize);
+}
+
+size_t ZSTDv03_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    return ZSTD_findFrameCompressedSize(src, srcSize);
+}
+
+ZSTDv03_Dctx* ZSTDv03_createDCtx(void)
+{
+    return (ZSTDv03_Dctx*)ZSTD_createDCtx();
+}
+
+size_t ZSTDv03_freeDCtx(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_freeDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_resetDCtx(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_resetDCtx((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_nextSrcSizeToDecompress(ZSTDv03_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress((ZSTD_DCtx*)dctx);
+}
+
+size_t ZSTDv03_decompressContinue(ZSTDv03_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue((ZSTD_DCtx*)dctx, dst, maxDstSize, src, srcSize);
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_v03.h b/vendor/github.com/DataDog/zstd/zstd_v03.h
new file mode 100644
index 0000000..b4449e2
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v03.h
@@ -0,0 +1,88 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V03_H_298734209782
+#define ZSTD_V03_H_298734209782
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv03_decompress() : decompress ZSTD frames compliant with v0.3.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv03_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+/**
+ZSTDv03_getFrameSrcSize() : get the source length of a ZSTD frame compliant with v0.3.x format
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv03_isError())
+*/
+size_t ZSTDv03_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+    /**
+ZSTDv03_isError() : tells if the result of ZSTDv03_decompress() is an error
+*/
+unsigned ZSTDv03_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv03_Dctx_s ZSTDv03_Dctx;
+ZSTDv03_Dctx* ZSTDv03_createDCtx(void);
+size_t ZSTDv03_freeDCtx(ZSTDv03_Dctx* dctx);
+
+size_t ZSTDv03_decompressDCtx(void* ctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+/* *************************************
+*  Streaming functions
+***************************************/
+size_t ZSTDv03_resetDCtx(ZSTDv03_Dctx* dctx);
+
+size_t ZSTDv03_nextSrcSizeToDecompress(ZSTDv03_Dctx* dctx);
+size_t ZSTDv03_decompressContinue(ZSTDv03_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv03_magicNumber 0xFD2FB523   /* v0.3 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V03_H_298734209782 */
diff --git a/vendor/github.com/DataDog/zstd/zstd_v04.c b/vendor/github.com/DataDog/zstd/zstd_v04.c
new file mode 100644
index 0000000..fb6d1d4
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v04.c
@@ -0,0 +1,3677 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v04.h"
+#include "error_private.h"
+
+
+/* ******************************************************************
+   mem.h
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/******************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/******************************************
+*  Compiler-specific
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/****************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-*************************************
+*  Debug
+***************************************/
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
+#  include <assert.h>
+#else
+#  ifndef assert
+#    define assert(condition) ((void)0)
+#  endif
+#endif
+
+#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
+
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
+#  include <stdio.h>
+extern int g_debuglog_enable;
+/* recommended values for ZSTD_DEBUG display levels :
+ * 1 : no display, enables assert() only
+ * 2 : reserved for currently active debug path
+ * 3 : events once per object lifetime (CCtx, CDict, etc.)
+ * 4 : events once per frame
+ * 5 : events once per block
+ * 6 : events once per sequence (*very* verbose) */
+#  define RAWLOG(l, ...) {                                      \
+                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
+                    fprintf(stderr, __VA_ARGS__);               \
+            }   }
+#  define DEBUGLOG(l, ...) {                                    \
+                if ((g_debuglog_enable) & (l<=ZSTD_DEBUG)) {    \
+                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
+                    fprintf(stderr, " \n");                     \
+            }   }
+#else
+#  define RAWLOG(l, ...)      {}    /* disabled */
+#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#endif
+
+
+/****************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets generating assembly depending on alignment.
+ *            But in some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; } __attribute__((packed)) unalign;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif // MEM_FORCE_MEMORY_ACCESS
+
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian())
+    {
+        MEM_write16(memPtr, val);
+    }
+    else
+    {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+    {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+*/
+#ifndef ZSTD_STATIC_H
+#define ZSTD_STATIC_H
+
+/* The objects defined into this file shall be considered experimental.
+ * They are not considered stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risks of future changes.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Types
+***************************************/
+#define ZSTD_WINDOWLOG_MAX 26
+#define ZSTD_WINDOWLOG_MIN 18
+#define ZSTD_WINDOWLOG_ABSOLUTEMIN 11
+#define ZSTD_CONTENTLOG_MAX (ZSTD_WINDOWLOG_MAX+1)
+#define ZSTD_CONTENTLOG_MIN 4
+#define ZSTD_HASHLOG_MAX 28
+#define ZSTD_HASHLOG_MIN 4
+#define ZSTD_SEARCHLOG_MAX (ZSTD_CONTENTLOG_MAX-1)
+#define ZSTD_SEARCHLOG_MIN 1
+#define ZSTD_SEARCHLENGTH_MAX 7
+#define ZSTD_SEARCHLENGTH_MIN 4
+
+/** from faster to stronger */
+typedef enum { ZSTD_fast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2 } ZSTD_strategy;
+
+typedef struct
+{
+    U64 srcSize;       /* optional : tells how much bytes are present in the frame. Use 0 if not known. */
+    U32 windowLog;     /* largest match distance : larger == more compression, more memory needed during decompression */
+    U32 contentLog;    /* full search segment : larger == more compression, slower, more memory (useless for fast) */
+    U32 hashLog;       /* dispatch table : larger == more memory, faster */
+    U32 searchLog;     /* nb of searches : larger == more compression, slower */
+    U32 searchLength;  /* size of matches : larger == faster decompression, sometimes less compression */
+    ZSTD_strategy strategy;
+} ZSTD_parameters;
+
+typedef ZSTDv04_Dctx ZSTD_DCtx;
+
+/* *************************************
+*  Advanced functions
+***************************************/
+/** ZSTD_decompress_usingDict
+*   Same as ZSTD_decompressDCtx, using a Dictionary content as prefix
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTD_decompressDCtx() */
+static size_t ZSTD_decompress_usingDict(ZSTD_DCtx* ctx,
+                                             void* dst, size_t maxDstSize,
+                                       const void* src, size_t srcSize,
+                                       const void* dict,size_t dictSize);
+
+
+/* **************************************
+*  Streaming functions (direct mode)
+****************************************/
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx);
+static size_t ZSTD_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize);
+static void   ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* src, size_t srcSize);
+
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+
+/**
+  Streaming decompression, bufferless mode
+
+  A ZSTD_DCtx object is required to track streaming operations.
+  Use ZSTD_createDCtx() / ZSTD_freeDCtx() to manage it.
+  A ZSTD_DCtx object can be re-used multiple times. Use ZSTD_resetDCtx() to return to fresh status.
+
+  First operation is to retrieve frame parameters, using ZSTD_getFrameParams().
+  This function doesn't consume its input. It needs enough input data to properly decode the frame header.
+  Objective is to retrieve *params.windowlog, to know minimum amount of memory required during decoding.
+  Result : 0 when successful, it means the ZSTD_parameters structure has been filled.
+           >0 : means there is not enough data into src. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTD_isError() (For example, if it's not a ZSTD header)
+
+  Then, you can optionally insert a dictionary.
+  This operation must mimic the compressor behavior, otherwise decompression will fail or be corrupted.
+
+  Then it's possible to start decompression.
+  Use ZSTD_nextSrcSizeToDecompress() and ZSTD_decompressContinue() alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTD_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTD_decompressContinue() is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTD_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+
+#endif  /* ZSTD_STATIC_H */
+
+
+/*
+    zstd_internal - common functions to include
+    Header File for include
+*/
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/* *************************************
+*  Common constants
+***************************************/
+#define ZSTD_MAGICNUMBER 0xFD2FB524   /* v0.4 */
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+static const size_t ZSTD_blockHeaderSize = 3;
+static const size_t ZSTD_frameHeaderSize_min = 5;
+#define ZSTD_frameHeaderSize_max 5         /* define, for static allocation */
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define IS_RAW BIT0
+#define IS_RLE BIT1
+
+#define MINMATCH 4
+#define REPCODE_STARTVALUE 4
+
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxML  ((1<<MLbits) - 1)
+#define MaxLL  ((1<<LLbits) - 1)
+#define MaxOff ((1<<Offbits)- 1)
+#define MLFSELog   10
+#define LLFSELog   10
+#define OffFSELog   9
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define MIN_SEQUENCES_SIZE (2 /*seqNb*/ + 2 /*dumps*/ + 3 /*seqTables*/ + 1 /*bitStream*/)
+#define MIN_CBLOCK_SIZE (3 /*litCSize*/ + MIN_SEQUENCES_SIZE)
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+
+/* ******************************************
+*  Shared functions to include for inlining
+********************************************/
+static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTD_wildcopy : custom version of memcpy(), can copy up to 7-8 bytes too many */
+static void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file
+****************************************************************** */
+#ifndef FSE_H
+#define FSE_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/* *****************************************
+*  FSE simple functions
+******************************************/
+static size_t FSE_decompress(void* dst,  size_t maxDstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+FSE_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'maxDstSize'.
+    return : size of regenerated data (<= maxDstSize)
+             or an error code, which can be tested using FSE_isError()
+
+    ** Important ** : FSE_decompress() doesn't decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+
+
+/* *****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+static unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+
+
+
+/* *****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSE_compress() does the following:
+1. count symbol occurrence from source[] into table count[]
+2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
+3. save normalized counters to memory buffer using writeNCount()
+4. build encoding table 'CTable' from normalized counters
+5. encode the data stream using encoding table 'CTable'
+
+FSE_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*!
+FSE_readNCount():
+   Read compactly saved 'normalizedCounter' from 'rBuffer'.
+   return : size read from 'rBuffer'
+            or an errorCode, which can be tested using FSE_isError()
+            maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+static  size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*!
+Constructor and Destructor of type FSE_DTable
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+
+/*!
+FSE_buildDTable():
+   Builds 'dt', which must be already allocated, using FSE_createDTable()
+   return : 0,
+            or an errorCode, which can be tested using FSE_isError() */
+static size_t FSE_buildDTable ( FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*!
+FSE_decompress_usingDTable():
+   Decompress compressed source 'cSrc' of size 'cSrcSize' using 'dt'
+   into 'dst' which must be already allocated.
+   return : size of regenerated data (necessarily <= maxDstSize)
+            or an errorCode, which can be tested using FSE_isError() */
+static  size_t FSE_decompress_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.
+This is performed by the function FSE_buildDTable().
+The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSE_isError().
+
+'FSE_DTable' can then be used to decompress 'cSrc', with FSE_decompress_usingDTable().
+'cSrcSize' must be strictly correct, otherwise decompression will fail.
+FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=maxDstSize).
+If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSE_H */
+
+
+/* ******************************************************************
+   bitstream
+   Part of NewGen Entropy library
+   header file (to include)
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+/**********************************************
+*  bitStream decompression API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BIT_DStream_t;
+
+typedef enum { BIT_DStream_unfinished = 0,
+               BIT_DStream_endOfBuffer = 1,
+               BIT_DStream_completed = 2,
+               BIT_DStream_overflow = 3 } BIT_DStream_status;  /* result of BIT_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD);
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* bitD);
+
+
+
+
+/******************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/****************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r=0;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+/**********************************************************
+* bitStream decoding
+**********************************************************/
+
+/*!BIT_initDStream
+*  Initialize a BIT_DStream_t.
+*  @bitD : a pointer to an already allocated BIT_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t))   /* normal case */
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+    }
+    else
+    {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BIT_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BIT_lookBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BIT_lookBitsFast :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_lookBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBits(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BIT_readBitsFast :
+*  unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BIT_lookBitsFast(bitD, nbBits);
+    BIT_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BIT_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer))
+    {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BIT_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start)
+    {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
+        return BIT_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BIT_DStream_status result = BIT_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start)
+        {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BIT_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BIT_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BIT_endOfDStream(const BIT_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+
+
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSE_STATIC_H
+#define FSE_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSE_NCOUNTBOUND 512
+#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
+#define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
+/* build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
+/* build a fake FSE_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSE_DState_t;
+
+
+static void     FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);
+
+static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+
+static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSE_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSE_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    DStatePtr->state = BIT_readBits(bitD, DTableH.tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)
+{
+    const FSE_decode_t DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BIT_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSE_STATIC_H */
+
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSE_MAX_MEMORY_USAGE 14
+#define FSE_DEFAULT_MEMORY_USAGE 13
+
+/*!FSE_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSE_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSE_FUNCTION_TYPE BYTE
+#define FSE_FUNCTION_EXTENSION
+#define FSE_DECODE_TYPE FSE_decode_t
+
+
+#endif   /* !FSE_COMMONDEFS_ONLY */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Dependencies
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSE_MAX_TABLELOG  (FSE_MAX_MEMORY_USAGE-2)
+#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)
+#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)
+#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)
+#define FSE_MIN_TABLELOG 5
+
+#define FSE_TABLELOG_ABSOLUTE_MAX 15
+#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX
+#error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+
+/*-**************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+static U32 FSE_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+
+static size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSE_DTableHeader DTableH;
+    void* const tdPtr = dt+1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    FSE_DECODE_TYPE* const tableDecode = (FSE_DECODE_TYPE*) (tdPtr);
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSE_tableStep(tableSize);
+    U16 symbolNext[FSE_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSE_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        if (normalizedCounter[s]==-1)
+        {
+            tableDecode[highThreshold--].symbol = (FSE_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        }
+        else
+        {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+        }
+    }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++)
+        {
+            tableDecode[position].symbol = (FSE_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }
+    }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++)
+        {
+            FSE_FUNCTION_TYPE symbol = (FSE_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+        }
+    }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+/******************************************
+*  FSE helper functions
+******************************************/
+static unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+
+
+/****************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSE_abs(short a)
+{
+    return a<0 ? -a : a;
+}
+
+static size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr))
+    {
+        if (previous0)
+        {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF)
+            {
+                n0+=24;
+                if (ip < iend-5)
+                {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                }
+                else
+                {
+                    bitStream >>= 16;
+                    bitCount+=16;
+                }
+            }
+            while ((bitStream & 3) == 3)
+            {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+            {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max)
+            {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            }
+            else
+            {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSE_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold)
+            {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            {
+                if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4))
+                {
+                    ip += bitCount>>3;
+                    bitCount &= 7;
+                }
+                else
+                {
+                    bitCount -= (int)(8 * (iend - 4 - ip));
+                    ip = iend - 4;
+                }
+                bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+            }
+        }
+    }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*********************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+static size_t FSE_buildDTable_rle (FSE_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const cell = (FSE_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSE_DTableHeader* const DTableH = (FSE_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSE_decode_t* const dinfo = (FSE_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++)
+    {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSE_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSE_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BIT_DStream_t bitD;
+    FSE_DState_t state1;
+    FSE_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSE_isError(errorCode)) return errorCode;
+
+    FSE_initDState(&state1, &bitD, dt);
+    FSE_initDState(&state2, &bitD, dt);
+
+#define FSE_GETSYMBOL(statePtr) fast ? FSE_decodeSymbolFast(statePtr, &bitD) : FSE_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BIT_reloadDStream(&bitD)==BIT_DStream_unfinished) && (op<olimit) ; op+=4)
+    {
+        op[0] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[1] = FSE_GETSYMBOL(&state2);
+
+        if (FSE_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BIT_reloadDStream(&bitD) > BIT_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSE_GETSYMBOL(&state1);
+
+        if (FSE_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BIT_reloadDStream(&bitD);
+
+        op[3] = FSE_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BIT_reloadDStream(&bitD) >= FSE_DStream_partiallyFilled; Ends at exactly BIT_DStream_completed */
+    while (1)
+    {
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state1);
+
+        if ( (BIT_reloadDStream(&bitD)>BIT_DStream_completed) || (op==omax) || (BIT_endOfDStream(&bitD) && (fast || FSE_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSE_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BIT_endOfDStream(&bitD) && FSE_endOfDState(&state1) && FSE_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+static size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSE_DTable* dt)
+{
+    FSE_DTableHeader DTableH;
+    U32 fastMode;
+
+    memcpy(&DTableH, dt, sizeof(DTableH));
+    fastMode = DTableH.fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSE_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+static size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSE_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    errorCode = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSE_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSE_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef HUFF0_H
+#define HUFF0_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Dependency
+******************************************/
+#include <stddef.h>    /* size_t */
+
+
+/* ****************************************
+*  Huff0 simple functions
+******************************************/
+static size_t HUF_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+HUF_decompress():
+    Decompress Huff0 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    'dstSize' must be the exact size of original (uncompressed) data.
+    Note : in contrast with FSE, HUF_decompress can regenerate RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data, because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUF_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+static unsigned    HUF_isError(size_t code);        /* tells if a return value is an error code */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFF0_H */
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef HUFF0_STATIC_H
+#define HUFF0_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Static allocation macros
+******************************************/
+/* static allocation of Huff0's DTable */
+#define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))  /* nb Cells; use unsigned short for X2, unsigned int for X4 */
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUF_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUF_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+/* ****************************************
+*  Huff0 detailed API
+******************************************/
+/*!
+HUF_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUF_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUF_decompressSXn_usingDTable
+
+*/
+static size_t HUF_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+static size_t HUF_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+static size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+static size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUFF0_STATIC_H */
+
+
+
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUF_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_DEFAULT_TABLELOG  HUF_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUF_MAX_SYMBOL_VALUE 255
+#if (HUF_MAX_TABLELOG > HUF_ABSOLUTEMAX_TABLELOG)
+#  error "HUF_MAX_TABLELOG is too large !"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+static unsigned HUF_isError(size_t code) { return ERR_isError(code); }
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+
+/*-*******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUF_readStats
+    Read compact Huffman tree, saved by HUF_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  /* special header */
+    {
+        if (iSize >= (242))   /* RLE */
+        {
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else   /* Incompressible */
+        {
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2)
+            {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+            }
+        }
+    }
+    else  /* header compressed with FSE (normal case) */
+    {
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSE_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUF_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++)
+    {
+        if (huffWeight[n] >= HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BIT_highbit32(weightTotal) + 1;
+    if (tableLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BIT_highbit32(rest);
+        U32 lastWeight = BIT_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/**************************/
+/* single-symbol decoding */
+/**************************/
+
+static size_t HUF_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(huffWeight, HUF_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof DTable, as allocated, from used size of DTable, in case of DTable re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++)
+    {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++)
+    {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUF_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BIT_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4))
+    {
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+
+static size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUF_DEltX2* const dt = ((const HUF_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUF_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUF_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/***************************/
+/* double-symbols decoding */
+/***************************/
+
+static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUF_DEltX4 DElt;
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1)
+    {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)   /* note : sortedSymbols already skipped */
+    {
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUF_ABSOLUTEMAX_TABLELOG][HUF_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUF_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++)
+    {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits)   /* enough room for a second symbol */
+        {
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        }
+        else
+        {
+            U32 i;
+            const U32 end = start + length;
+            HUF_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+static size_t HUF_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUF_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUF_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUF_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUF_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUF_DEltX4* const dt = ((HUF_DEltX4*)dtPtr) + 1;
+
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUF_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUF_readStats(weightList, HUF_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUF_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--)
+        { if (!maxW) return ERROR(GENERIC); }  /* necessarily finds a solution before maxW==0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++)
+        {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++)
+        {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++)
+        {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++)
+            {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+            }
+        }
+    }
+
+    HUF_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BIT_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BIT_skipBits(DStream, dt[val].nbBits);
+    else
+    {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8))
+        {
+            BIT_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+        }
+    }
+    return 1;
+}
+
+
+#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUF_MAX_TABLELOG<=12)) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd-7))
+    {
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-2))
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+static size_t HUF_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUF_DEltX4* const dt = ((const HUF_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BIT_DStream_t bitD1;
+        BIT_DStream_t bitD2;
+        BIT_DStream_t bitD3;
+        BIT_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BIT_initDStream(&bitD1, istart1, length1);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD2, istart2, length2);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD3, istart3, length3);
+        if (HUF_isError(errorCode)) return errorCode;
+        errorCode = BIT_initDStream(&bitD4, istart4, length4);
+        if (HUF_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; )
+        {
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+static size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUF_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/**********************************/
+/* Generic decompression selector */
+/**********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+static size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUF_decompress4X2, HUF_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUF_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUF_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+
+
+
+#endif   /* ZSTD_CCOMMON_H_MODULE */
+
+
+/*
+    zstd - decompression module fo v0.4 legacy format
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTD_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTD_HEAPMODE
+#  define ZSTD_HEAPMODE 1
+#endif
+
+
+/* *******************************************************
+*  Includes
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug : printf */
+
+
+/* *******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/* *************************************
+*  Local types
+***************************************/
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/* *************************************
+*  Error Management
+***************************************/
+
+/*! ZSTD_isError
+*   tells if a return value is an error code */
+static unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
+
+
+/* *************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock } ZSTD_dStage;
+
+struct ZSTDv04_Dctx_s
+{
+    U32 LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
+    U32 OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    U32 MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTD_parameters params;
+    blockType_t bType;
+    ZSTD_dStage stage;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + 8 /* margin for wildcopy */];
+    BYTE headerBuffer[ZSTD_frameHeaderSize_max];
+};  /* typedef'd to ZSTD_DCtx within "zstd_static.h" */
+
+static size_t ZSTD_resetDCtx(ZSTD_DCtx* dctx)
+{
+    dctx->expected = ZSTD_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    return 0;
+}
+
+static ZSTD_DCtx* ZSTD_createDCtx(void)
+{
+    ZSTD_DCtx* dctx = (ZSTD_DCtx*)malloc(sizeof(ZSTD_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTD_resetDCtx(dctx);
+    return dctx;
+}
+
+static size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
+{
+    free(dctx);
+    return 0;
+}
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+/** ZSTD_decodeFrameHeader_Part1
+*   decode the 1st part of the Frame Header, which tells Frame Header size.
+*   srcSize must be == ZSTD_frameHeaderSize_min
+*   @return : the full size of the Frame Header */
+static size_t ZSTD_decodeFrameHeader_Part1(ZSTD_DCtx* zc, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    zc->headerSize = ZSTD_frameHeaderSize_min;
+    return zc->headerSize;
+}
+
+
+static size_t ZSTD_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize < ZSTD_frameHeaderSize_min) return ZSTD_frameHeaderSize_max;
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    memset(params, 0, sizeof(*params));
+    params->windowLog = (((const BYTE*)src)[4] & 15) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
+    if ((((const BYTE*)src)[4] >> 4) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits */
+    return 0;
+}
+
+/** ZSTD_decodeFrameHeader_Part2
+*   decode the full Frame Header
+*   srcSize must be the size provided by ZSTD_decodeFrameHeader_Part1
+*   @return : 0, or an error code, which can be tested using ZSTD_isError() */
+static size_t ZSTD_decodeFrameHeader_Part2(ZSTD_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t result;
+    if (srcSize != zc->headerSize) return ERROR(srcSize_wrong);
+    result = ZSTD_getFrameParams(&(zc->params), src, srcSize);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+static size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3) return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+static size_t ZSTD_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/** ZSTD_decompressLiterals
+    @return : nb of bytes read from src, or an error code*/
+static size_t ZSTD_decompressLiterals(void* dst, size_t* maxDstSizePtr,
+                                const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    const size_t litSize = (MEM_readLE32(src) & 0x1FFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+    const size_t litCSize = (MEM_readLE32(ip+2) & 0xFFFFFF) >> 5;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+
+    if (litSize > *maxDstSizePtr) return ERROR(corruption_detected);
+    if (litCSize + 5 > srcSize) return ERROR(corruption_detected);
+
+    if (HUF_isError(HUF_decompress(dst, litSize, ip+5, litCSize))) return ERROR(corruption_detected);
+
+    *maxDstSizePtr = litSize;
+    return litCSize + 5;
+}
+
+
+/** ZSTD_decodeLiteralsBlock
+    @return : nb of bytes read from src (< srcSize ) */
+static size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(*istart & 3)
+    {
+    /* compressed */
+    case 0:
+        {
+            size_t litSize = BLOCKSIZE;
+            const size_t readSize = ZSTD_decompressLiterals(dctx->litBuffer, &litSize, src, srcSize);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, 8);
+            return readSize;   /* works if it's an error too */
+        }
+    case IS_RAW:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > srcSize-11)   /* risk of reading too far with wildcopy */
+            {
+                if (litSize > srcSize-3) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, 8);
+                return litSize+3;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+3;
+            dctx->litSize = litSize;
+            return litSize+3;        }
+    case IS_RLE:
+        {
+            const size_t litSize = (MEM_readLE32(istart) & 0xFFFFFF) >> 2;   /* no buffer issue : srcSize >= MIN_CBLOCK_SIZE */
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[3], litSize + 8);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return 4;
+        }
+    default:
+        return ERROR(corruption_detected);   /* forbidden nominal case */
+    }
+}
+
+
+static size_t ZSTD_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb,
+                         const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < 5) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = MEM_readLE16(ip); ip+=2;
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2)
+    {
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    }
+    else
+    {
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL >= MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case bt_rle :
+            LLlog = 0;
+            FSE_buildDTable_rle(DTableLL, *ip++); break;
+        case bt_raw :
+            LLlog = LLbits;
+            FSE_buildDTable_raw(DTableLL, LLbits); break;
+        default :
+            {   U32 max = MaxLL;
+                headerSize = FSE_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case bt_rle :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case bt_raw :
+            Offlog = Offbits;
+            FSE_buildDTable_raw(DTableOffb, Offbits); break;
+        default :
+            {   U32 max = MaxOff;
+                headerSize = FSE_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case bt_rle :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSE_buildDTable_rle(DTableML, *ip++); break;
+        case bt_raw :
+            MLlog = MLbits;
+            FSE_buildDTable_raw(DTableML, MLbits); break;
+        default :
+            {   U32 max = MaxML;
+                headerSize = FSE_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSE_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSELog) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSE_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t offset;
+    size_t matchLength;
+} seq_t;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    FSE_DState_t stateLL;
+    FSE_DState_t stateOffb;
+    FSE_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    if (litLength == MaxLL) {
+        U32 add = *dumps++;
+        if (add < 255) litLength += add;
+        else {
+            litLength = dumps[0] + (dumps[1]<<8) + (dumps[2]<<16);
+            dumps += 3;
+        }
+        if (dumps > de) { litLength = MaxLL+255; }  /* late correction, to avoid using uninitialized memory */
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {   static const U32 offsetPrefix[MaxOff+1] = {
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode, nbBits;
+        offsetCode = FSE_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));   /* <= maxOff, by table construction */
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BIT_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* cmove */
+        if (offsetCode | !litLength) seqState->prevOffset = seq->offset;   /* cmove */
+    }
+
+    /* MatchLength */
+    matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML) {
+        U32 add = *dumps++;
+        if (add < 255) matchLength += add;
+        else {
+            matchLength = dumps[0] + (dumps[1]<<8) + (dumps[2]<<16);
+            dumps += 3;
+        }
+        if (dumps > de) { matchLength = MaxML+255; }  /* late correction, to avoid using uninitialized memory */
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+}
+
+
+static size_t ZSTD_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    static const int dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+    static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* risk read beyond lit buffer */
+
+    /* copy Literals */
+    ZSTD_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base))
+    {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd)
+        {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {
+            size_t length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+        }
+    }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        const int sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH))
+    {
+        if (op < oend_8)
+        {
+            ZSTD_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    }
+    else
+    {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTD_decompressSequences(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTD_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, iend-ip);
+    if (ZSTD_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = 4;
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = 4;
+        errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; )
+        {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTD_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if ( !BIT_endOfDStream(&(seqState.DStream)) ) return ERROR(corruption_detected);   /* DStream should be entirely and exactly consumed; otherwise data is corrupted */
+
+        /* last literal segment */
+        {
+            size_t lastLLSize = litEnd - litPtr;
+            if (litPtr > litEnd) return ERROR(corruption_detected);
+            if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+            if (op != litPtr) memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd)   /* not contiguous */
+    {
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                            void* dst, size_t maxDstSize,
+                      const void* src, size_t srcSize)
+{
+    /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    /* Decode literals sub-block */
+    size_t litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
+    if (ZSTD_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTD_decompressSequences(dctx, dst, maxDstSize, ip, srcSize);
+}
+
+
+static size_t ZSTD_decompress_usingDict(ZSTD_DCtx* ctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+
+    /* init */
+    ZSTD_resetDCtx(ctx);
+    if (dict)
+    {
+        ZSTD_decompress_insertDictionary(ctx, dict, dictSize);
+        ctx->dictEnd = ctx->previousDstEnd;
+        ctx->vBase = (const char*)dst - ((const char*)(ctx->previousDstEnd) - (const char*)(ctx->base));
+        ctx->base = dst;
+    }
+    else
+    {
+        ctx->vBase = ctx->base = ctx->dictEnd = dst;
+    }
+
+    /* Frame Header */
+    {
+        size_t frameHeaderSize;
+        if (srcSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        frameHeaderSize = ZSTD_decodeFrameHeader_Part1(ctx, src, ZSTD_frameHeaderSize_min);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        frameHeaderSize = ZSTD_decodeFrameHeader_Part2(ctx, src, frameHeaderSize);
+        if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTD_decompressBlock_internal(ctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTD_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+static size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    if (MEM_readLE32(src) != ZSTD_MAGICNUMBER) return ERROR(prefix_unknown);
+    ip += ZSTD_frameHeaderSize_min; remainingSize -= ZSTD_frameHeaderSize_min;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTD_blockHeaderSize;
+        remainingSize -= ZSTD_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+/* ******************************
+*  Streaming Decompression API
+********************************/
+static size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+static size_t ZSTD_decompressContinue(ZSTD_DCtx* ctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != ctx->expected) return ERROR(srcSize_wrong);
+    ZSTD_checkContinuity(ctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (ctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        /* get frame header size */
+        if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        ctx->headerSize = ZSTD_decodeFrameHeader_Part1(ctx, src, ZSTD_frameHeaderSize_min);
+        if (ZSTD_isError(ctx->headerSize)) return ctx->headerSize;
+        memcpy(ctx->headerBuffer, src, ZSTD_frameHeaderSize_min);
+        if (ctx->headerSize > ZSTD_frameHeaderSize_min) return ERROR(GENERIC);   /* impossible */
+        ctx->expected = 0;   /* not necessary to copy more */
+        /* fallthrough */
+    case ZSTDds_decodeFrameHeader:
+        /* get frame header */
+        {   size_t const result = ZSTD_decodeFrameHeader_Part2(ctx, ctx->headerBuffer, ctx->headerSize);
+            if (ZSTD_isError(result)) return result;
+            ctx->expected = ZSTD_blockHeaderSize;
+            ctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        /* Decode block header */
+        {   blockProperties_t bp;
+            size_t const blockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
+            if (ZSTD_isError(blockSize)) return blockSize;
+            if (bp.blockType == bt_end)
+            {
+                ctx->expected = 0;
+                ctx->stage = ZSTDds_getFrameHeaderSize;
+            }
+            else
+            {
+                ctx->expected = blockSize;
+                ctx->bType = bp.blockType;
+                ctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {
+            /* Decompress : block content */
+            size_t rSize;
+            switch(ctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTD_decompressBlock_internal(ctx, dst, maxDstSize, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTD_copyRawBlock(dst, maxDstSize, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);
+            }
+            ctx->stage = ZSTDds_decodeBlockHeader;
+            ctx->expected = ZSTD_blockHeaderSize;
+            ctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTD_decompress_insertDictionary(ZSTD_DCtx* ctx, const void* dict, size_t dictSize)
+{
+    ctx->dictEnd = ctx->previousDstEnd;
+    ctx->vBase = (const char*)dict - ((const char*)(ctx->previousDstEnd) - (const char*)(ctx->base));
+    ctx->base = dict;
+    ctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stdlib.h>
+
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFF_DCtx object is required to track streaming operation.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation.
+*  ZBUFF_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst .
+*  return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  output : 128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_readHeader, ZBUFFds_loadHeader, ZBUFFds_decodeHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFF_dStage;
+
+/* *** Resource management *** */
+
+#define ZSTD_frameHeaderSize_max 5   /* too magical, should come from reference */
+struct ZBUFFv04_DCtx_s {
+    ZSTD_DCtx* zc;
+    ZSTD_parameters params;
+    char* inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char* outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t hPos;
+    const char* dict;
+    size_t dictSize;
+    ZBUFF_dStage stage;
+    unsigned char headerBuffer[ZSTD_frameHeaderSize_max];
+};   /* typedef'd to ZBUFF_DCtx within "zstd_buffered.h" */
+
+typedef ZBUFFv04_DCtx ZBUFF_DCtx;
+
+
+static ZBUFF_DCtx* ZBUFF_createDCtx(void)
+{
+    ZBUFF_DCtx* zbc = (ZBUFF_DCtx*)malloc(sizeof(ZBUFF_DCtx));
+    if (zbc==NULL) return NULL;
+    memset(zbc, 0, sizeof(*zbc));
+    zbc->zc = ZSTD_createDCtx();
+    zbc->stage = ZBUFFds_init;
+    return zbc;
+}
+
+static size_t ZBUFF_freeDCtx(ZBUFF_DCtx* zbc)
+{
+    if (zbc==NULL) return 0;   /* support free on null */
+    ZSTD_freeDCtx(zbc->zc);
+    free(zbc->inBuff);
+    free(zbc->outBuff);
+    free(zbc);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+static size_t ZBUFF_decompressInit(ZBUFF_DCtx* zbc)
+{
+    zbc->stage = ZBUFFds_readHeader;
+    zbc->hPos = zbc->inPos = zbc->outStart = zbc->outEnd = zbc->dictSize = 0;
+    return ZSTD_resetDCtx(zbc->zc);
+}
+
+
+static size_t ZBUFF_decompressWithDictionary(ZBUFF_DCtx* zbc, const void* src, size_t srcSize)
+{
+    zbc->dict = (const char*)src;
+    zbc->dictSize = srcSize;
+    return 0;
+}
+
+static size_t ZBUFF_limitCopy(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    size_t length = MIN(maxDstSize, srcSize);
+    memcpy(dst, src, length);
+    return length;
+}
+
+/* *** Decompression *** */
+
+static size_t ZBUFF_decompressContinue(ZBUFF_DCtx* zbc, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* ip = istart;
+    const char* const iend = istart + *srcSizePtr;
+    char* const ostart = (char*)dst;
+    char* op = ostart;
+    char* const oend = ostart + *maxDstSizePtr;
+    U32 notDone = 1;
+
+    DEBUGLOG(5, "ZBUFF_decompressContinue");
+    while (notDone)
+    {
+        switch(zbc->stage)
+        {
+
+        case ZBUFFds_init :
+            DEBUGLOG(5, "ZBUFF_decompressContinue: stage==ZBUFFds_init => ERROR(init_missing)");
+            return ERROR(init_missing);
+
+        case ZBUFFds_readHeader :
+            /* read header from src */
+            {   size_t const headerSize = ZSTD_getFrameParams(&(zbc->params), src, *srcSizePtr);
+                if (ZSTD_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    memcpy(zbc->headerBuffer+zbc->hPos, src, *srcSizePtr);
+                    zbc->hPos += *srcSizePtr;
+                    *maxDstSizePtr = 0;
+                    zbc->stage = ZBUFFds_loadHeader;
+                    return headerSize - zbc->hPos;
+                }
+                zbc->stage = ZBUFFds_decodeHeader;
+                break;
+            }
+
+        case ZBUFFds_loadHeader:
+            /* complete header from src */
+            {   size_t headerSize = ZBUFF_limitCopy(
+                    zbc->headerBuffer + zbc->hPos, ZSTD_frameHeaderSize_max - zbc->hPos,
+                    src, *srcSizePtr);
+                zbc->hPos += headerSize;
+                ip += headerSize;
+                headerSize = ZSTD_getFrameParams(&(zbc->params), zbc->headerBuffer, zbc->hPos);
+                if (ZSTD_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    *maxDstSizePtr = 0;
+                    return headerSize - zbc->hPos;
+            }   }
+            /* intentional fallthrough */
+
+        case ZBUFFds_decodeHeader:
+                /* apply header to create / resize buffers */
+                {   size_t const neededOutSize = (size_t)1 << zbc->params.windowLog;
+                    size_t const neededInSize = BLOCKSIZE;   /* a block is never > BLOCKSIZE */
+                    if (zbc->inBuffSize < neededInSize) {
+                        free(zbc->inBuff);
+                        zbc->inBuffSize = neededInSize;
+                        zbc->inBuff = (char*)malloc(neededInSize);
+                        if (zbc->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    if (zbc->outBuffSize < neededOutSize) {
+                        free(zbc->outBuff);
+                        zbc->outBuffSize = neededOutSize;
+                        zbc->outBuff = (char*)malloc(neededOutSize);
+                        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
+                }   }
+                if (zbc->dictSize)
+                    ZSTD_decompress_insertDictionary(zbc->zc, zbc->dict, zbc->dictSize);
+                if (zbc->hPos) {
+                    /* some data already loaded into headerBuffer : transfer into inBuff */
+                    memcpy(zbc->inBuff, zbc->headerBuffer, zbc->hPos);
+                    zbc->inPos = zbc->hPos;
+                    zbc->hPos = 0;
+                    zbc->stage = ZBUFFds_load;
+                    break;
+                }
+                zbc->stage = ZBUFFds_read;
+		/* fall-through */
+        case ZBUFFds_read:
+            {
+                size_t neededInSize = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+                if (neededInSize==0)   /* end of frame */
+                {
+                    zbc->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize)
+                {
+                    /* directly decode from src */
+                    size_t decodedSize = ZSTD_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        ip, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbc->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {
+                size_t neededInSize = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+                size_t toLoad = neededInSize - zbc->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbc->inBuffSize - zbc->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFF_limitCopy(zbc->inBuff + zbc->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbc->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+                {
+                    size_t decodedSize = ZSTD_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        zbc->inBuff, neededInSize);
+                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    zbc->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbc->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFds_flush;
+                    /* ZBUFFds_flush follows */
+                }
+            }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {
+                size_t toFlushSize = zbc->outEnd - zbc->outStart;
+                size_t flushedSize = ZBUFF_limitCopy(op, oend-op, zbc->outBuff + zbc->outStart, toFlushSize);
+                op += flushedSize;
+                zbc->outStart += flushedSize;
+                if (flushedSize == toFlushSize)
+                {
+                    zbc->stage = ZBUFFds_read;
+                    if (zbc->outStart + BLOCKSIZE > zbc->outBuffSize)
+                        zbc->outStart = zbc->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+        }
+    }
+
+    *srcSizePtr = ip-istart;
+    *maxDstSizePtr = op-ostart;
+
+    {
+        size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zbc->zc);
+        if (nextSrcSizeHint > 3) nextSrcSizeHint+= 3;   /* get the next block header while at it */
+        nextSrcSizeHint -= zbc->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv04_isError(size_t errorCode) { return ERR_isError(errorCode); }
+const char* ZBUFFv04_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+size_t ZBUFFv04_recommendedDInSize()  { return BLOCKSIZE + 3; }
+size_t ZBUFFv04_recommendedDOutSize() { return BLOCKSIZE; }
+
+
+
+/*- ========================================================================= -*/
+
+/* final wrapping stage */
+
+size_t ZSTDv04_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompress_usingDict(dctx, dst, maxDstSize, src, srcSize, NULL, 0);
+}
+
+size_t ZSTDv04_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+#if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE==1)
+    size_t regenSize;
+    ZSTD_DCtx* dctx = ZSTD_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv04_decompressDCtx(dctx, dst, maxDstSize, src, srcSize);
+    ZSTD_freeDCtx(dctx);
+    return regenSize;
+#else
+    ZSTD_DCtx dctx;
+    return ZSTDv04_decompressDCtx(&dctx, dst, maxDstSize, src, srcSize);
+#endif
+}
+
+size_t ZSTDv04_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    return ZSTD_findFrameCompressedSize(src, srcSize);
+}
+
+size_t ZSTDv04_resetDCtx(ZSTDv04_Dctx* dctx) { return ZSTD_resetDCtx(dctx); }
+
+size_t ZSTDv04_nextSrcSizeToDecompress(ZSTDv04_Dctx* dctx)
+{
+    return ZSTD_nextSrcSizeToDecompress(dctx);
+}
+
+size_t ZSTDv04_decompressContinue(ZSTDv04_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTD_decompressContinue(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+
+ZBUFFv04_DCtx* ZBUFFv04_createDCtx(void) { return ZBUFF_createDCtx(); }
+size_t ZBUFFv04_freeDCtx(ZBUFFv04_DCtx* dctx) { return ZBUFF_freeDCtx(dctx); }
+
+size_t ZBUFFv04_decompressInit(ZBUFFv04_DCtx* dctx) { return ZBUFF_decompressInit(dctx); }
+size_t ZBUFFv04_decompressWithDictionary(ZBUFFv04_DCtx* dctx, const void* src, size_t srcSize)
+{ return ZBUFF_decompressWithDictionary(dctx, src, srcSize); }
+
+size_t ZBUFFv04_decompressContinue(ZBUFFv04_DCtx* dctx, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    DEBUGLOG(5, "ZBUFFv04_decompressContinue");
+    return ZBUFF_decompressContinue(dctx, dst, maxDstSizePtr, src, srcSizePtr);
+}
+
+ZSTD_DCtx* ZSTDv04_createDCtx(void) { return ZSTD_createDCtx(); }
+size_t ZSTDv04_freeDCtx(ZSTD_DCtx* dctx) { return ZSTD_freeDCtx(dctx); }
+
+size_t ZSTDv04_getFrameParams(ZSTD_parameters* params, const void* src, size_t srcSize)
+{
+    return ZSTD_getFrameParams(params, src, srcSize);
+}
diff --git a/vendor/github.com/DataDog/zstd/zstd_v04.h b/vendor/github.com/DataDog/zstd/zstd_v04.h
new file mode 100644
index 0000000..6391631
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v04.h
@@ -0,0 +1,137 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_V04_H_91868324769238
+#define ZSTD_V04_H_91868324769238
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* *************************************
+*  Includes
+***************************************/
+#include <stddef.h>   /* size_t */
+
+
+/* *************************************
+*  Simple one-step function
+***************************************/
+/**
+ZSTDv04_decompress() : decompress ZSTD frames compliant with v0.4.x format
+    compressedSize : is the exact source size
+    maxOriginalSize : is the size of the 'dst' buffer, which must be already allocated.
+                      It must be equal or larger than originalSize, otherwise decompression will fail.
+    return : the number of bytes decompressed into destination buffer (originalSize)
+             or an errorCode if it fails (which can be tested using ZSTDv01_isError())
+*/
+size_t ZSTDv04_decompress( void* dst, size_t maxOriginalSize,
+                     const void* src, size_t compressedSize);
+
+/**
+ZSTDv04_getFrameSrcSize() : get the source length of a ZSTD frame compliant with v0.4.x format
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv04_isError())
+*/
+size_t ZSTDv04_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+/**
+ZSTDv04_isError() : tells if the result of ZSTDv04_decompress() is an error
+*/
+unsigned ZSTDv04_isError(size_t code);
+
+
+/* *************************************
+*  Advanced functions
+***************************************/
+typedef struct ZSTDv04_Dctx_s ZSTDv04_Dctx;
+ZSTDv04_Dctx* ZSTDv04_createDCtx(void);
+size_t ZSTDv04_freeDCtx(ZSTDv04_Dctx* dctx);
+
+size_t ZSTDv04_decompressDCtx(ZSTDv04_Dctx* dctx,
+                              void* dst, size_t maxOriginalSize,
+                        const void* src, size_t compressedSize);
+
+
+/* *************************************
+*  Direct Streaming
+***************************************/
+size_t ZSTDv04_resetDCtx(ZSTDv04_Dctx* dctx);
+
+size_t ZSTDv04_nextSrcSizeToDecompress(ZSTDv04_Dctx* dctx);
+size_t ZSTDv04_decompressContinue(ZSTDv04_Dctx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize);
+/**
+  Use above functions alternatively.
+  ZSTD_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTD_decompressContinue().
+  ZSTD_decompressContinue() will use previous data blocks to improve compression if they are located prior to current block.
+  Result is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTD_decompressContinue() has decoded some header.
+*/
+
+
+/* *************************************
+*  Buffered Streaming
+***************************************/
+typedef struct ZBUFFv04_DCtx_s ZBUFFv04_DCtx;
+ZBUFFv04_DCtx* ZBUFFv04_createDCtx(void);
+size_t         ZBUFFv04_freeDCtx(ZBUFFv04_DCtx* dctx);
+
+size_t ZBUFFv04_decompressInit(ZBUFFv04_DCtx* dctx);
+size_t ZBUFFv04_decompressWithDictionary(ZBUFFv04_DCtx* dctx, const void* dict, size_t dictSize);
+
+size_t ZBUFFv04_decompressContinue(ZBUFFv04_DCtx* dctx, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr);
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFF_DCtx object is required to track streaming operation.
+*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
+*  Use ZBUFF_decompressInit() to start a new decompression operation.
+*  ZBUFF_DCtx objects can be reused multiple times.
+*
+*  Optionally, a reference to a static dictionary can be set, using ZBUFF_decompressWithDictionary()
+*  It must be the same content as the one set during compression phase.
+*  Dictionary content must remain accessible during the decompression process.
+*
+*  Use ZBUFF_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFF_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFF_recommendedDInSize / ZBUFF_recommendedDOutSize
+*  output : ZBUFF_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : ZBUFF_recommendedDInSize==128Kb+3; just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+unsigned ZBUFFv04_isError(size_t errorCode);
+const char* ZBUFFv04_getErrorName(size_t errorCode);
+
+
+/** The below functions provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are not compulsory, they just tend to offer better latency */
+size_t ZBUFFv04_recommendedDInSize(void);
+size_t ZBUFFv04_recommendedDOutSize(void);
+
+
+/* *************************************
+*  Prefix - version detection
+***************************************/
+#define ZSTDv04_magicNumber 0xFD2FB524   /* v0.4 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_V04_H_91868324769238 */
diff --git a/vendor/github.com/DataDog/zstd/zstd_v05.c b/vendor/github.com/DataDog/zstd/zstd_v05.c
new file mode 100644
index 0000000..a5e1b1f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v05.c
@@ -0,0 +1,4011 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v05.h"
+#include "error_private.h"
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Dependencies
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS :
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets depending on alignment.
+ *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(void*)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(void*)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard, by lying on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign*)memPtr)->u32 = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign*)memPtr)->u64 = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write32(void* memPtr, U32 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+MEM_STATIC void MEM_write64(void* memPtr, U64 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* MEM_FORCE_MEMORY_ACCESS */
+
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
+    }
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U64)((U64)p[0] + ((U64)p[1]<<8) + ((U64)p[2]<<16) + ((U64)p[3]<<24)
+                     + ((U64)p[4]<<32) + ((U64)p[5]<<40) + ((U64)p[6]<<48) + ((U64)p[7]<<56));
+    }
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net
+*/
+#ifndef ZSTD_STATIC_H
+#define ZSTD_STATIC_H
+
+/* The prototypes defined within this file are considered experimental.
+ * They should not be used in the context DLL as they may change in the future.
+ * Prefer static linking if you need them, to control breaking version changes issues.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-*************************************
+*  Types
+***************************************/
+#define ZSTDv05_WINDOWLOG_ABSOLUTEMIN 11
+
+
+/*-*************************************
+*  Advanced functions
+***************************************/
+/*- Advanced Decompression functions -*/
+
+/*! ZSTDv05_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv05_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv05_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference, which will not be modified, and 1 to run the decompression operation */
+size_t ZSTDv05_decompress_usingPreparedDCtx(
+                                             ZSTDv05_DCtx* dctx, const ZSTDv05_DCtx* preparedDCtx,
+                                             void* dst, size_t dstCapacity,
+                                       const void* src, size_t srcSize);
+
+
+/* **************************************
+*  Streaming functions (direct mode)
+****************************************/
+size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx);
+
+/*
+  Streaming decompression, direct mode (bufferless)
+
+  A ZSTDv05_DCtx object is required to track streaming operations.
+  Use ZSTDv05_createDCtx() / ZSTDv05_freeDCtx() to manage it.
+  A ZSTDv05_DCtx object can be re-used multiple times.
+
+  First typical operation is to retrieve frame parameters, using ZSTDv05_getFrameParams().
+  This operation is independent, and just needs enough input data to properly decode the frame header.
+  Objective is to retrieve *params.windowlog, to know minimum amount of memory required during decoding.
+  Result : 0 when successful, it means the ZSTDv05_parameters structure has been filled.
+           >0 : means there is not enough data into src. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv05_isError()
+
+  Start decompression, with ZSTDv05_decompressBegin() or ZSTDv05_decompressBegin_usingDict()
+  Alternatively, you can copy a prepared context, using ZSTDv05_copyDCtx()
+
+  Then use ZSTDv05_nextSrcSizeToDecompress() and ZSTDv05_decompressContinue() alternatively.
+  ZSTDv05_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv05_decompressContinue().
+  ZSTDv05_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTDv05_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTDv05_decompressContinue() is the number of bytes regenerated within 'dst'.
+  It can be zero, which is not an error; it just means ZSTDv05_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTDv05_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    User will have to take in charge required information to regenerate data, such as block sizes.
+
+    A few rules to respect :
+    - Uncompressed block size must be <= 128 KB
+    - Compressing or decompressing requires a context structure
+      + Use ZSTDv05_createCCtx() and ZSTDv05_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv05_compressBegin()
+      + decompression : ZSTDv05_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - When a block is considered not compressible enough, ZSTDv05_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv05_decompressBlock() doesn't accept uncompressed data as input !!
+*/
+
+size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv05_STATIC_H */
+
+
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+*/
+#ifndef ZSTD_CCOMMON_H_MODULE
+#define ZSTD_CCOMMON_H_MODULE
+
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv05_DICT_MAGIC  0xEC30A435
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BLOCKSIZE (128 KB)                 /* define, for static allocation */
+
+static const size_t ZSTDv05_blockHeaderSize = 3;
+static const size_t ZSTDv05_frameHeaderSize_min = 5;
+#define ZSTDv05_frameHeaderSize_max 5         /* define, for static allocation */
+
+#define BITv057 128
+#define BITv056  64
+#define BITv055  32
+#define BITv054  16
+#define BITv051   2
+#define BITv050   1
+
+#define IS_HUFv05 0
+#define IS_PCH 1
+#define IS_RAW 2
+#define IS_RLE 3
+
+#define MINMATCH 4
+#define REPCODE_STARTVALUE 1
+
+#define Litbits  8
+#define MLbits   7
+#define LLbits   6
+#define Offbits  5
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  ((1<<MLbits) - 1)
+#define MaxLL  ((1<<LLbits) - 1)
+#define MaxOff ((1<<Offbits)- 1)
+#define MLFSEv05Log   10
+#define LLFSEv05Log   10
+#define OffFSEv05Log   9
+#define MaxSeq MAX(MaxLL, MaxML)
+
+#define FSEv05_ENCODING_RAW     0
+#define FSEv05_ENCODING_RLE     1
+#define FSEv05_ENCODING_STATIC  2
+#define FSEv05_ENCODING_DYNAMIC 3
+
+
+#define HufLog 12
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define WILDCOPY_OVERLENGTH 8
+
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv05_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
+#define COPY8(d,s) { ZSTDv05_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv05_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+MEM_STATIC void ZSTDv05_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* offCode;
+    BYTE* litStart;
+    BYTE* lit;
+    BYTE* litLengthStart;
+    BYTE* litLength;
+    BYTE* matchLengthStart;
+    BYTE* matchLength;
+    BYTE* dumpsStart;
+    BYTE* dumps;
+    /* opt */
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+} seqStore_t;
+
+
+
+#endif   /* ZSTDv05_CCOMMON_H_MODULE */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   header file
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv05_H
+#define FSEv05_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Includes
+******************************************/
+#include <stddef.h>    /* size_t, ptrdiff_t */
+
+
+/*-****************************************
+*  FSEv05 simple functions
+******************************************/
+size_t FSEv05_decompress(void* dst,  size_t maxDstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+FSEv05_decompress():
+    Decompress FSEv05 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'maxDstSize'.
+    return : size of regenerated data (<= maxDstSize)
+             or an error code, which can be tested using FSEv05_isError()
+
+    ** Important ** : FSEv05_decompress() doesn't decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+
+
+/* *****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+unsigned    FSEv05_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv05_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+
+
+/* *****************************************
+*  FSEv05 detailed API
+******************************************/
+/* *** DECOMPRESSION *** */
+
+/*!
+FSEv05_readNCount():
+   Read compactly saved 'normalizedCounter' from 'rBuffer'.
+   return : size read from 'rBuffer'
+            or an errorCode, which can be tested using FSEv05_isError()
+            maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv05_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*!
+Constructor and Destructor of type FSEv05_DTable
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv05_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv05_DTable* FSEv05_createDTable(unsigned tableLog);
+void        FSEv05_freeDTable(FSEv05_DTable* dt);
+
+/*!
+FSEv05_buildDTable():
+   Builds 'dt', which must be already allocated, using FSEv05_createDTable()
+   @return : 0,
+             or an errorCode, which can be tested using FSEv05_isError() */
+size_t FSEv05_buildDTable (FSEv05_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*!
+FSEv05_decompress_usingDTable():
+   Decompress compressed source @cSrc of size @cSrcSize using `dt`
+   into `dst` which must be already allocated.
+   @return : size of regenerated data (necessarily <= @dstCapacity)
+             or an errorCode, which can be tested using FSEv05_isError() */
+size_t FSEv05_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv05_DTable* dt);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv05_H */
+/* ******************************************************************
+   bitstream
+   Part of FSEv05 library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITv05STREAM_H_MODULE
+#define BITv05STREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which highly benefit from being inlined.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv05_DStream_t;
+
+typedef enum { BITv05_DStream_unfinished = 0,
+               BITv05_DStream_endOfBuffer = 1,
+               BITv05_DStream_completed = 2,
+               BITv05_DStream_overflow = 3 } BITv05_DStream_status;  /* result of BITv05_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv05_initDStream(BITv05_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv05_readBits(BITv05_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv05_DStream_status BITv05_reloadDStream(BITv05_DStream_t* bitD);
+MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* bitD);
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Helper functions
+****************************************************************/
+MEM_STATIC unsigned BITv05_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r=0;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*!BITv05_initDStream
+*  Initialize a BITv05_DStream_t.
+*  @bitD : a pointer to an already allocated BITv05_DStream_t structure
+*  @srcBuffer must point at the beginning of a bitStream
+*  @srcSize must be the exact size of the bitStream
+*  @result : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv05_initDStream(BITv05_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(size_t)) {  /* normal case */
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(size_t);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BITv05_highbit32(contain32);
+    } else {
+        U32 contain32;
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[6]) << (sizeof(size_t)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[5]) << (sizeof(size_t)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[4]) << (sizeof(size_t)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(bitD->start))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        contain32 = ((const BYTE*)srcBuffer)[srcSize-1];
+        if (contain32 == 0) return ERROR(GENERIC);   /* endMark not present */
+        bitD->bitsConsumed = 8 - BITv05_highbit32(contain32);
+        bitD->bitsConsumed += (U32)(sizeof(size_t) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+MEM_STATIC size_t BITv05_lookBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv05_lookBitsFast :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BITv05_lookBitsFast(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    const U32 bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv05_skipBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv05_readBits(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BITv05_lookBits(bitD, nbBits);
+    BITv05_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*!BITv05_readBitsFast :
+*  unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BITv05_readBitsFast(BITv05_DStream_t* bitD, U32 nbBits)
+{
+    size_t value = BITv05_lookBitsFast(bitD, nbBits);
+    BITv05_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv05_DStream_status BITv05_reloadDStream(BITv05_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BITv05_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv05_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv05_DStream_endOfBuffer;
+        return BITv05_DStream_completed;
+    }
+    {
+        U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv05_DStream_status result = BITv05_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv05_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv05_endOfDStream
+*   @return Tells if DStream has reached its exact end
+*/
+MEM_STATIC unsigned BITv05_endOfDStream(const BITv05_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITv05STREAM_H_MODULE */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv05_STATIC_H
+#define FSEv05_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* It is possible to statically allocate FSEv05 CTable/DTable as a table of unsigned using below macros */
+#define FSEv05_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSEv05 advanced API
+*******************************************/
+size_t FSEv05_buildDTable_raw (FSEv05_DTable* dt, unsigned nbBits);
+/* build a fake FSEv05_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv05_buildDTable_rle (FSEv05_DTable* dt, unsigned char symbolValue);
+/* build a fake FSEv05_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSEv05 symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv05_DState_t;
+
+
+static void     FSEv05_initDState(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD, const FSEv05_DTable* dt);
+
+static unsigned char FSEv05_decodeSymbol(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD);
+
+static unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr);
+
+
+
+/* *****************************************
+*  FSEv05 unsafe API
+*******************************************/
+static unsigned char FSEv05_decodeSymbolFast(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+/* decompression */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv05_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv05_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv05_initDState(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD, const FSEv05_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv05_DTableHeader* const DTableH = (const FSEv05_DTableHeader*)ptr;
+    DStatePtr->state = BITv05_readBits(bitD, DTableH->tableLog);
+    BITv05_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv05_peakSymbol(FSEv05_DState_t* DStatePtr)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC BYTE FSEv05_decodeSymbol(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32  nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BITv05_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC BYTE FSEv05_decodeSymbolFast(FSEv05_DState_t* DStatePtr, BITv05_DStream_t* bitD)
+{
+    const FSEv05_decode_t DInfo = ((const FSEv05_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    const U32 nbBits = DInfo.nbBits;
+    BYTE symbol = DInfo.symbol;
+    size_t lowBits = BITv05_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+MEM_STATIC unsigned FSEv05_endOfDState(const FSEv05_DState_t* DStatePtr)
+{
+    return DStatePtr->state == 0;
+}
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv05_STATIC_H */
+/* ******************************************************************
+   FSEv05 : Finite State Entropy coder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+#ifndef FSEv05_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv05_MAX_MEMORY_USAGE 14
+#define FSEv05_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv05_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv05_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv05_FUNCTION_TYPE BYTE
+#define FSEv05_FUNCTION_EXTENSION
+#define FSEv05_DECODE_TYPE FSEv05_decode_t
+
+
+#endif   /* !FSEv05_COMMONDEFS_ONLY */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv05_MAX_TABLELOG  (FSEv05_MAX_MEMORY_USAGE-2)
+#define FSEv05_MAX_TABLESIZE (1U<<FSEv05_MAX_TABLELOG)
+#define FSEv05_MAXTABLESIZE_MASK (FSEv05_MAX_TABLESIZE-1)
+#define FSEv05_DEFAULT_TABLELOG (FSEv05_DEFAULT_MEMORY_USAGE-2)
+#define FSEv05_MIN_TABLELOG 5
+
+#define FSEv05_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv05_MAX_TABLELOG > FSEv05_TABLELOG_ABSOLUTE_MAX
+#error "FSEv05_MAX_TABLELOG > FSEv05_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv05_STATIC_ASSERT(c) { enum { FSEv05_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv05_DTABLE_SIZE_U32(FSEv05_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv05_FUNCTION_EXTENSION
+#  error "FSEv05_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv05_FUNCTION_TYPE
+#  error "FSEv05_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv05_CAT(X,Y) X##Y
+#define FSEv05_FUNCTION_NAME(X,Y) FSEv05_CAT(X,Y)
+#define FSEv05_TYPE_NAME(X,Y) FSEv05_CAT(X,Y)
+
+
+/* Function templates */
+static U32 FSEv05_tableStep(U32 tableSize) { return (tableSize>>1) + (tableSize>>3) + 3; }
+
+
+
+FSEv05_DTable* FSEv05_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv05_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv05_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv05_DTable*)malloc( FSEv05_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv05_freeDTable (FSEv05_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv05_buildDTable(FSEv05_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSEv05_DTableHeader DTableH;
+    void* const tdPtr = dt+1;   /* because dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv05_DECODE_TYPE* const tableDecode = (FSEv05_DECODE_TYPE*) (tdPtr);
+    const U32 tableSize = 1 << tableLog;
+    const U32 tableMask = tableSize-1;
+    const U32 step = FSEv05_tableStep(tableSize);
+    U16 symbolNext[FSEv05_MAX_SYMBOL_VALUE+1];
+    U32 position = 0;
+    U32 highThreshold = tableSize-1;
+    const S16 largeLimit= (S16)(1 << (tableLog-1));
+    U32 noLarge = 1;
+    U32 s;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv05_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv05_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    DTableH.tableLog = (U16)tableLog;
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (normalizedCounter[s]==-1) {
+            tableDecode[highThreshold--].symbol = (FSEv05_FUNCTION_TYPE)s;
+            symbolNext[s] = 1;
+        } else {
+            if (normalizedCounter[s] >= largeLimit) noLarge=0;
+            symbolNext[s] = normalizedCounter[s];
+    }   }
+
+    /* Spread symbols */
+    for (s=0; s<=maxSymbolValue; s++) {
+        int i;
+        for (i=0; i<normalizedCounter[s]; i++) {
+            tableDecode[position].symbol = (FSEv05_FUNCTION_TYPE)s;
+            position = (position + step) & tableMask;
+            while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+    }   }
+
+    if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+
+    /* Build Decoding table */
+    {
+        U32 i;
+        for (i=0; i<tableSize; i++) {
+            FSEv05_FUNCTION_TYPE symbol = (FSEv05_FUNCTION_TYPE)(tableDecode[i].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[i].nbBits = (BYTE) (tableLog - BITv05_highbit32 ((U32)nextState) );
+            tableDecode[i].newState = (U16) ( (nextState << tableDecode[i].nbBits) - tableSize);
+    }   }
+
+    DTableH.fastMode = (U16)noLarge;
+    memcpy(dt, &DTableH, sizeof(DTableH));
+    return 0;
+}
+
+
+#ifndef FSEv05_COMMONDEFS_ONLY
+/*-****************************************
+*  FSEv05 helper functions
+******************************************/
+unsigned FSEv05_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSEv05 NCount encoding-decoding
+****************************************************************/
+static short FSEv05_abs(short a) { return a<0 ? -a : a; }
+
+
+size_t FSEv05_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv05_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv05_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {
+            const short max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv05_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv05_buildDTable_rle (FSEv05_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv05_DTableHeader* const DTableH = (FSEv05_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv05_decode_t* const cell = (FSEv05_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv05_buildDTable_raw (FSEv05_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv05_DTableHeader* const DTableH = (FSEv05_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv05_decode_t* const dinfo = (FSEv05_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<=maxSymbolValue; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv05_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv05_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv05_DStream_t bitD;
+    FSEv05_DState_t state1;
+    FSEv05_DState_t state2;
+    size_t errorCode;
+
+    /* Init */
+    errorCode = BITv05_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+    if (FSEv05_isError(errorCode)) return errorCode;
+
+    FSEv05_initDState(&state1, &bitD, dt);
+    FSEv05_initDState(&state2, &bitD, dt);
+
+#define FSEv05_GETSYMBOL(statePtr) fast ? FSEv05_decodeSymbolFast(statePtr, &bitD) : FSEv05_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv05_reloadDStream(&bitD)==BITv05_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv05_GETSYMBOL(&state1);
+
+        if (FSEv05_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv05_reloadDStream(&bitD);
+
+        op[1] = FSEv05_GETSYMBOL(&state2);
+
+        if (FSEv05_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv05_reloadDStream(&bitD) > BITv05_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv05_GETSYMBOL(&state1);
+
+        if (FSEv05_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv05_reloadDStream(&bitD);
+
+        op[3] = FSEv05_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv05_reloadDStream(&bitD) >= FSEv05_DStream_partiallyFilled; Ends at exactly BITv05_DStream_completed */
+    while (1) {
+        if ( (BITv05_reloadDStream(&bitD)>BITv05_DStream_completed) || (op==omax) || (BITv05_endOfDStream(&bitD) && (fast || FSEv05_endOfDState(&state1))) )
+            break;
+
+        *op++ = FSEv05_GETSYMBOL(&state1);
+
+        if ( (BITv05_reloadDStream(&bitD)>BITv05_DStream_completed) || (op==omax) || (BITv05_endOfDStream(&bitD) && (fast || FSEv05_endOfDState(&state2))) )
+            break;
+
+        *op++ = FSEv05_GETSYMBOL(&state2);
+    }
+
+    /* end ? */
+    if (BITv05_endOfDStream(&bitD) && FSEv05_endOfDState(&state1) && FSEv05_endOfDState(&state2))
+        return op-ostart;
+
+    if (op==omax) return ERROR(dstSize_tooSmall);   /* dst buffer is full, but cSrc unfinished */
+
+    return ERROR(corruption_detected);
+}
+
+
+size_t FSEv05_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv05_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv05_DTableHeader* DTableH = (const FSEv05_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv05_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv05_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv05_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv05_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv05_MAX_SYMBOL_VALUE;
+    size_t errorCode;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSEv05 decoding mode */
+    errorCode = FSEv05_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSEv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    errorCode = FSEv05_buildDTable (dt, counting, maxSymbolValue, tableLog);
+    if (FSEv05_isError(errorCode)) return errorCode;
+
+    /* always return, even if it is an error code */
+    return FSEv05_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);
+}
+
+
+
+#endif   /* FSEv05_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFF0_H
+#define HUFF0_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Huff0 simple functions
+******************************************/
+size_t HUFv05_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*!
+HUFv05_decompress():
+    Decompress Huff0 data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    @dstSize : must be the **exact** size of original (uncompressed) data.
+    Note : in contrast with FSEv05, HUFv05_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUFv05_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+/* Error Management */
+unsigned    HUFv05_isError(size_t code);        /* tells if a return value is an error code */
+const char* HUFv05_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUF0_H */
+/* ******************************************************************
+   Huff0 : Huffman codec, part of New Generation Entropy library
+   header file, for static linking only
+   Copyright (C) 2013-2016, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUF0_STATIC_H
+#define HUF0_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* static allocation of Huff0's DTable */
+#define HUFv05_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUFv05_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUFv05_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv05_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUFv05_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv05_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUFv05_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv05_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv05_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+/* ****************************************
+*  Huff0 detailed API
+******************************************/
+/*!
+HUFv05_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv05_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv05_decompressSXn_usingDTable
+*/
+size_t HUFv05_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+size_t HUFv05_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+size_t HUFv05_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv05_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+/* single stream variants */
+
+size_t HUFv05_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv05_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv05_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv05_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUF0_STATIC_H */
+/* ******************************************************************
+   Huff0 : Huffman coder, part of New Generation Entropy library
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSEv05+Huff0 source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUFv05_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUFv05_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv05_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUFv05_ABSOLUTEMAX_TABLELOG */
+#define HUFv05_DEFAULT_TABLELOG  HUFv05_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUFv05_MAX_SYMBOL_VALUE 255
+#if (HUFv05_MAX_TABLELOG > HUFv05_ABSOLUTEMAX_TABLELOG)
+#  error "HUFv05_MAX_TABLELOG is too large !"
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+unsigned HUFv05_isError(size_t code) { return ERR_isError(code); }
+const char* HUFv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+#define HUFv05_STATIC_ASSERT(c) { enum { HUFv05_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* *******************************************************
+*  Huff0 : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUFv05_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv05_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+/*! HUFv05_readStats
+    Read compact Huffman tree, saved by HUFv05_writeCTable
+    @huffWeight : destination buffer
+    @return : size read from `src`
+*/
+static size_t HUFv05_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    U32 tableLog;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+    U32 n;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static int l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            for (n=0; n<oSize; n+=2) {
+                huffWeight[n]   = ip[n/2] >> 4;
+                huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }
+    else  {   /* header compressed with FSEv05 (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv05_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv05_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv05_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    for (n=0; n<oSize; n++) {
+        if (huffWeight[n] >= HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        rankStats[huffWeight[n]]++;
+        weightTotal += (1 << huffWeight[n]) >> 1;
+    }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    tableLog = BITv05_highbit32(weightTotal) + 1;
+    if (tableLog > HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+    {   /* determine last weight */
+        U32 total = 1 << tableLog;
+        U32 rest = total - weightTotal;
+        U32 verif = 1 << BITv05_highbit32(rest);
+        U32 lastWeight = BITv05_highbit32(rest) + 1;
+        if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+        huffWeight[oSize] = (BYTE)lastWeight;
+        rankStats[lastWeight]++;
+    }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    *tableLogPtr = tableLog;
+    return iSize+1;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+size_t HUFv05_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv05_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUFv05_DEltX2* const dt = (HUFv05_DEltX2*)dtPtr;
+
+    HUFv05_STATIC_ASSERT(sizeof(HUFv05_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv05_readStats(huffWeight, HUFv05_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv05_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof allocated DTable, from used size of DTable, in case of re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<=tableLog; n++) {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++) {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUFv05_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+static BYTE HUFv05_decodeSymbolX2(BITv05_DStream_t* Dstream, const HUFv05_DEltX2* dt, const U32 dtLog)
+{
+        const size_t val = BITv05_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+        const BYTE c = dt[val].byte;
+        BITv05_skipBits(Dstream, dt[val].nbBits);
+        return c;
+}
+
+#define HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv05_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv05_MAX_TABLELOG<=12)) \
+        HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv05_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv05_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv05_decodeStreamX2(BYTE* p, BITv05_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv05_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv05_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p < pEnd))
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv05_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+size_t HUFv05_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const U32 dtLog = DTable[0];
+    const void* dtPtr = DTable;
+    const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr)+1;
+    BITv05_DStream_t bitD;
+
+    if (dstSize <= cSrcSize) return ERROR(dstSize_tooSmall);
+    { size_t const errorCode = BITv05_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv05_isError(errorCode)) return errorCode; }
+
+    HUFv05_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv05_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv05_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX2(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUFv05_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv05_decompress1X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+size_t HUFv05_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+    const void* const dtPtr = DTable;
+    const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
+    const U32 dtLog = DTable[0];
+    size_t errorCode;
+
+    /* Init */
+    BITv05_DStream_t bitD1;
+    BITv05_DStream_t bitD2;
+    BITv05_DStream_t bitD3;
+    BITv05_DStream_t bitD4;
+    const size_t length1 = MEM_readLE16(istart);
+    const size_t length2 = MEM_readLE16(istart+2);
+    const size_t length3 = MEM_readLE16(istart+4);
+    size_t length4;
+    const BYTE* const istart1 = istart + 6;  /* jumpTable */
+    const BYTE* const istart2 = istart1 + length1;
+    const BYTE* const istart3 = istart2 + length2;
+    const BYTE* const istart4 = istart3 + length3;
+    const size_t segmentSize = (dstSize+3) / 4;
+    BYTE* const opStart2 = ostart + segmentSize;
+    BYTE* const opStart3 = opStart2 + segmentSize;
+    BYTE* const opStart4 = opStart3 + segmentSize;
+    BYTE* op1 = ostart;
+    BYTE* op2 = opStart2;
+    BYTE* op3 = opStart3;
+    BYTE* op4 = opStart4;
+    U32 endSignal;
+
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    length4 = cSrcSize - (length1 + length2 + length3 + 6);
+    if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+    errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+    if (HUFv05_isError(errorCode)) return errorCode;
+
+    /* 16-32 symbols per loop (4-8 symbols per stream) */
+    endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+    for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+        HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+        HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+        HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+        HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+        HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
+        HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
+        HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
+        HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
+        HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+        HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+        HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+        HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+        HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
+        HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
+        HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
+        HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
+        endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+    }
+
+    /* check corruption */
+    if (op1 > opStart2) return ERROR(corruption_detected);
+    if (op2 > opStart3) return ERROR(corruption_detected);
+    if (op3 > opStart4) return ERROR(corruption_detected);
+    /* note : op4 supposed already verified within main loop */
+
+    /* finish bitStreams one by one */
+    HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+    HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+    HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+    HUFv05_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+    /* check */
+    endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+    if (!endSignal) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+
+size_t HUFv05_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX2(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+    size_t errorCode;
+
+    errorCode = HUFv05_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv05_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+static void HUFv05_fillDTableX4Level2(HUFv05_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv05_DEltX4 DElt;
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+    U32 s;
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }
+}
+
+typedef U32 rankVal_t[HUFv05_ABSOLUTEMAX_TABLELOG][HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUFv05_fillDTableX4(HUFv05_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv05_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv05_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            U32 i;
+            const U32 end = start + length;
+            HUFv05_DEltX4 DElt;
+
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits   = (BYTE)(nbBits);
+            DElt.length   = 1;
+            for (i = start; i < end; i++)
+                DTable[i] = DElt;
+        }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv05_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv05_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUFv05_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUFv05_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUFv05_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUFv05_DEltX4* const dt = ((HUFv05_DEltX4*)dtPtr) + 1;
+
+    HUFv05_STATIC_ASSERT(sizeof(HUFv05_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUFv05_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv05_readStats(weightList, HUFv05_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv05_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {
+        U32 w, nextRankStart = 0;
+        for (w=1; w<=maxW; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {
+        U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 w = weightList[s];
+            U32 r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {
+        const U32 minBits = tableLog+1 - maxW;
+        U32 nextRankVal = 0;
+        U32 w, consumed;
+        const int rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+        U32* rankVal0 = rankVal[0];
+        for (w=1; w<=maxW; w++) {
+            U32 current = nextRankVal;
+            nextRankVal += rankStats[w] << (w+rescale);
+            rankVal0[w] = current;
+        }
+        for (consumed = minBits; consumed <= memLog - minBits; consumed++) {
+            U32* rankValPtr = rankVal[consumed];
+            for (w = 1; w <= maxW; w++) {
+                rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }
+
+    HUFv05_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUFv05_decodeSymbolX4(void* op, BITv05_DStream_t* DStream, const HUFv05_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv05_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv05_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv05_decodeLastSymbolX4(void* op, BITv05_DStream_t* DStream, const HUFv05_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv05_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv05_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv05_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv05_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv05_MAX_TABLELOG<=12)) \
+        ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv05_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv05_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv05_decodeStreamX4(BYTE* p, BITv05_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv05_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv05_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv05_reloadDStream(bitDPtr) == BITv05_DStream_unfinished) && (p <= pEnd-2))
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv05_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv05_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+size_t HUFv05_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+
+    const U32 dtLog = DTable[0];
+    const void* const dtPtr = DTable;
+    const HUFv05_DEltX4* const dt = ((const HUFv05_DEltX4*)dtPtr) +1;
+    size_t errorCode;
+
+    /* Init */
+    BITv05_DStream_t bitD;
+    errorCode = BITv05_initDStream(&bitD, istart, cSrcSize);
+    if (HUFv05_isError(errorCode)) return errorCode;
+
+    /* finish bitStreams one by one */
+    HUFv05_decodeStreamX4(ostart, &bitD, oend,     dt, dtLog);
+
+    /* check */
+    if (!BITv05_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv05_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX4(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv05_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv05_decompress1X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+size_t HUFv05_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv05_DEltX4* const dt = ((const HUFv05_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv05_DStream_t bitD1;
+        BITv05_DStream_t bitD2;
+        BITv05_DStream_t bitD3;
+        BITv05_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+        if (HUFv05_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv05_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv05_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv05_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv05_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv05_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv05_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv05_CREATE_STATIC_DTABLEX4(DTable, HUFv05_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv05_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv05_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv05_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv05_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUFv05_decompress4X2, HUFv05_decompress4X4, NULL };
+    /* estimate decompression time */
+    U32 Q;
+    const U32 D256 = (U32)(dstSize >> 8);
+    U32 Dtime[3];
+    U32 algoNb = 0;
+    int n;
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize >= dstSize) return ERROR(corruption_detected);   /* invalid, or not compressed, but not compressed already dealt with */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    for (n=0; n<3; n++)
+        Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    if (Dtime[1] < Dtime[0]) algoNb = 1;
+
+    return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+
+    //return HUFv05_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUFv05_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUFv05_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv05_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv05_HEAPMODE
+#  define ZSTDv05_HEAPMODE 1
+#endif
+
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include <stdlib.h>      /* calloc */
+#include <string.h>      /* memcpy, memmove */
+#include <stdio.h>       /* debug only : printf */
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/*-*************************************
+*  Local types
+***************************************/
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+
+/* *******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv05_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/* *************************************
+*  Error Management
+***************************************/
+/*! ZSTDv05_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv05_isError(size_t code) { return ERR_isError(code); }
+
+
+/*! ZSTDv05_getErrorName() :
+*   provides error code string (useful for debugging) */
+const char* ZSTDv05_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* *************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDv05ds_getFrameHeaderSize, ZSTDv05ds_decodeFrameHeader,
+               ZSTDv05ds_decodeBlockHeader, ZSTDv05ds_decompressBlock } ZSTDv05_dStage;
+
+struct ZSTDv05_DCtx_s
+{
+    FSEv05_DTable LLTable[FSEv05_DTABLE_SIZE_U32(LLFSEv05Log)];
+    FSEv05_DTable OffTable[FSEv05_DTABLE_SIZE_U32(OffFSEv05Log)];
+    FSEv05_DTable MLTable[FSEv05_DTABLE_SIZE_U32(MLFSEv05Log)];
+    unsigned   hufTableX4[HUFv05_DTABLE_SIZE(HufLog)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTDv05_parameters params;
+    blockType_t bType;   /* used in ZSTDv05_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv05_dStage stage;
+    U32 flagStaticTables;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[BLOCKSIZE + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv05_frameHeaderSize_max];
+};  /* typedef'd to ZSTDv05_DCtx within "zstd_static.h" */
+
+size_t ZSTDv05_sizeofDCtx (void) { return sizeof(ZSTDv05_DCtx); }
+
+size_t ZSTDv05_decompressBegin(ZSTDv05_DCtx* dctx)
+{
+    dctx->expected = ZSTDv05_frameHeaderSize_min;
+    dctx->stage = ZSTDv05ds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTableX4[0] = HufLog;
+    dctx->flagStaticTables = 0;
+    return 0;
+}
+
+ZSTDv05_DCtx* ZSTDv05_createDCtx(void)
+{
+    ZSTDv05_DCtx* dctx = (ZSTDv05_DCtx*)malloc(sizeof(ZSTDv05_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTDv05_decompressBegin(dctx);
+    return dctx;
+}
+
+size_t ZSTDv05_freeDCtx(ZSTDv05_DCtx* dctx)
+{
+    free(dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv05_copyDCtx(ZSTDv05_DCtx* dstDCtx, const ZSTDv05_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv05_DCtx) - (BLOCKSIZE+WILDCOPY_OVERLENGTH + ZSTDv05_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/* *************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv05_MAGICNUMBER (defined within zstd_internal.h)
+      - 1 byte  - Window Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+/* Block format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+
+/** ZSTDv05_decodeFrameHeader_Part1() :
+*   decode the 1st part of the Frame Header, which tells Frame Header size.
+*   srcSize must be == ZSTDv05_frameHeaderSize_min.
+*   @return : the full size of the Frame Header */
+static size_t ZSTDv05_decodeFrameHeader_Part1(ZSTDv05_DCtx* zc, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize != ZSTDv05_frameHeaderSize_min)
+        return ERROR(srcSize_wrong);
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    zc->headerSize = ZSTDv05_frameHeaderSize_min;
+    return zc->headerSize;
+}
+
+
+size_t ZSTDv05_getFrameParams(ZSTDv05_parameters* params, const void* src, size_t srcSize)
+{
+    U32 magicNumber;
+    if (srcSize < ZSTDv05_frameHeaderSize_min) return ZSTDv05_frameHeaderSize_max;
+    magicNumber = MEM_readLE32(src);
+    if (magicNumber != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    memset(params, 0, sizeof(*params));
+    params->windowLog = (((const BYTE*)src)[4] & 15) + ZSTDv05_WINDOWLOG_ABSOLUTEMIN;
+    if ((((const BYTE*)src)[4] >> 4) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits */
+    return 0;
+}
+
+/** ZSTDv05_decodeFrameHeader_Part2() :
+*   decode the full Frame Header.
+*   srcSize must be the size provided by ZSTDv05_decodeFrameHeader_Part1().
+*   @return : 0, or an error code, which can be tested using ZSTDv05_isError() */
+static size_t ZSTDv05_decodeFrameHeader_Part2(ZSTDv05_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t result;
+    if (srcSize != zc->headerSize)
+        return ERROR(srcSize_wrong);
+    result = ZSTDv05_getFrameParams(&(zc->params), src, srcSize);
+    if ((MEM_32bits()) && (zc->params.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+size_t ZSTDv05_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    BYTE headerFlags;
+    U32 cSize;
+
+    if (srcSize < 3)
+        return ERROR(srcSize_wrong);
+
+    headerFlags = *in;
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+
+    bpPtr->blockType = (blockType_t)(headerFlags >> 6);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv05_copyRawBlock(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    if (srcSize > maxDstSize) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv05_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+size_t ZSTDv05_decodeLiteralsBlock(ZSTDv05_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(istart[0]>> 6)
+    {
+    case IS_HUFv05:
+        {
+            size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv05_isError(singleStream ?
+                            HUFv05_decompress1X2(dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv05_decompress   (dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_PCH:
+        {
+            size_t errorCode;
+            size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (!dctx->flagStaticTables)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            errorCode = HUFv05_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTableX4);
+            if (HUFv05_isError(errorCode)) return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_RAW:
+        {
+            size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case IS_RLE:
+        {
+            size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > BLOCKSIZE) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+size_t ZSTDv05_decodeSeqHeaders(int* nbSeq, const BYTE** dumpsPtr, size_t* dumpsLengthPtr,
+                         FSEv05_DTable* DTableLL, FSEv05_DTable* DTableML, FSEv05_DTable* DTableOffb,
+                         const void* src, size_t srcSize, U32 flagStaticTable)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* ip = istart;
+    const BYTE* const iend = istart + srcSize;
+    U32 LLtype, Offtype, MLtype;
+    U32 LLlog, Offlog, MLlog;
+    size_t dumpsLength;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE)
+        return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    *nbSeq = *ip++;
+    if (*nbSeq==0) return 1;
+    if (*nbSeq >= 128) {
+        if (ip >= iend) return ERROR(srcSize_wrong);
+        *nbSeq = ((nbSeq[0]-128)<<8) + *ip++;
+    }
+
+    if (ip >= iend) return ERROR(srcSize_wrong);
+    LLtype  = *ip >> 6;
+    Offtype = (*ip >> 4) & 3;
+    MLtype  = (*ip >> 2) & 3;
+    if (*ip & 2) {
+        if (ip+3 > iend) return ERROR(srcSize_wrong);
+        dumpsLength  = ip[2];
+        dumpsLength += ip[1] << 8;
+        ip += 3;
+    } else {
+        if (ip+2 > iend) return ERROR(srcSize_wrong);
+        dumpsLength  = ip[1];
+        dumpsLength += (ip[0] & 1) << 8;
+        ip += 2;
+    }
+    *dumpsPtr = ip;
+    ip += dumpsLength;
+    *dumpsLengthPtr = dumpsLength;
+
+    /* check */
+    if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+    /* sequences */
+    {
+        S16 norm[MaxML+1];    /* assumption : MaxML >= MaxLL >= MaxOff */
+        size_t headerSize;
+
+        /* Build DTables */
+        switch(LLtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            LLlog = 0;
+            FSEv05_buildDTable_rle(DTableLL, *ip++);
+            break;
+        case FSEv05_ENCODING_RAW :
+            LLlog = LLbits;
+            FSEv05_buildDTable_raw(DTableLL, LLbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   U32 max = MaxLL;
+                headerSize = FSEv05_readNCount(norm, &max, &LLlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (LLlog > LLFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableLL, norm, max, LLlog);
+        }   }
+
+        switch(Offtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            Offlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong);   /* min : "raw", hence no header, but at least xxLog bits */
+            FSEv05_buildDTable_rle(DTableOffb, *ip++ & MaxOff); /* if *ip > MaxOff, data is corrupted */
+            break;
+        case FSEv05_ENCODING_RAW :
+            Offlog = Offbits;
+            FSEv05_buildDTable_raw(DTableOffb, Offbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   U32 max = MaxOff;
+                headerSize = FSEv05_readNCount(norm, &max, &Offlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (Offlog > OffFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableOffb, norm, max, Offlog);
+        }   }
+
+        switch(MLtype)
+        {
+        case FSEv05_ENCODING_RLE :
+            MLlog = 0;
+            if (ip > iend-2) return ERROR(srcSize_wrong); /* min : "raw", hence no header, but at least xxLog bits */
+            FSEv05_buildDTable_rle(DTableML, *ip++);
+            break;
+        case FSEv05_ENCODING_RAW :
+            MLlog = MLbits;
+            FSEv05_buildDTable_raw(DTableML, MLbits);
+            break;
+        case FSEv05_ENCODING_STATIC:
+            if (!flagStaticTable) return ERROR(corruption_detected);
+            break;
+        case FSEv05_ENCODING_DYNAMIC :
+        default :   /* impossible */
+            {   U32 max = MaxML;
+                headerSize = FSEv05_readNCount(norm, &max, &MLlog, ip, iend-ip);
+                if (FSEv05_isError(headerSize)) return ERROR(GENERIC);
+                if (MLlog > MLFSEv05Log) return ERROR(corruption_detected);
+                ip += headerSize;
+                FSEv05_buildDTable(DTableML, norm, max, MLlog);
+    }   }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv05_DStream_t DStream;
+    FSEv05_DState_t stateLL;
+    FSEv05_DState_t stateOffb;
+    FSEv05_DState_t stateML;
+    size_t prevOffset;
+    const BYTE* dumps;
+    const BYTE* dumpsEnd;
+} seqState_t;
+
+
+
+static void ZSTDv05_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    size_t litLength;
+    size_t prevOffset;
+    size_t offset;
+    size_t matchLength;
+    const BYTE* dumps = seqState->dumps;
+    const BYTE* const de = seqState->dumpsEnd;
+
+    /* Literal length */
+    litLength = FSEv05_peakSymbol(&(seqState->stateLL));
+    prevOffset = litLength ? seq->offset : seqState->prevOffset;
+    if (litLength == MaxLL) {
+        U32 add = *dumps++;
+        if (add < 255) litLength += add;
+        else {
+            litLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no risk : dumps is always followed by seq tables > 1 byte */
+            if (litLength&1) litLength>>=1, dumps += 3;
+            else litLength = (U16)(litLength)>>1, dumps += 2;
+        }
+        if (dumps > de) { litLength = MaxLL+255; }  /* late correction, to avoid using uninitialized memory */
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+
+    /* Offset */
+    {
+        static const U32 offsetPrefix[MaxOff+1] = {
+                1 /*fake*/, 1, 2, 4, 8, 16, 32, 64, 128, 256,
+                512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072, 262144,
+                524288, 1048576, 2097152, 4194304, 8388608, 16777216, 33554432, /*fake*/ 1, 1, 1, 1, 1 };
+        U32 offsetCode = FSEv05_peakSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+        U32 nbBits = offsetCode - 1;
+        if (offsetCode==0) nbBits = 0;   /* cmove */
+        offset = offsetPrefix[offsetCode] + BITv05_readBits(&(seqState->DStream), nbBits);
+        if (MEM_32bits()) BITv05_reloadDStream(&(seqState->DStream));
+        if (offsetCode==0) offset = prevOffset;   /* repcode, cmove */
+        if (offsetCode | !litLength) seqState->prevOffset = seq->offset;   /* cmove */
+        FSEv05_decodeSymbol(&(seqState->stateOffb), &(seqState->DStream));    /* update */
+    }
+
+    /* Literal length update */
+    FSEv05_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));   /* update */
+    if (MEM_32bits()) BITv05_reloadDStream(&(seqState->DStream));
+
+    /* MatchLength */
+    matchLength = FSEv05_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
+    if (matchLength == MaxML) {
+        U32 add = *dumps++;
+        if (add < 255) matchLength += add;
+        else {
+            matchLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            if (matchLength&1) matchLength>>=1, dumps += 3;
+            else matchLength = (U16)(matchLength)>>1, dumps += 2;
+        }
+        if (dumps > de) { matchLength = MaxML+255; }  /* late correction, to avoid using uninitialized memory */
+        if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
+    }
+    matchLength += MINMATCH;
+
+    /* save result */
+    seq->litLength = litLength;
+    seq->offset = offset;
+    seq->matchLength = matchLength;
+    seqState->dumps = dumps;
+
+#if 0   /* debug */
+    {
+        static U64 totalDecoded = 0;
+        printf("pos %6u : %3u literals & match %3u bytes at distance %6u \n",
+           (U32)(totalDecoded), (U32)litLength, (U32)matchLength, (U32)offset);
+        totalDecoded += litLength + matchLength;
+    }
+#endif
+}
+
+
+static size_t ZSTDv05_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    static const int dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+    static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
+    BYTE* const oLitEnd = op + sequence.litLength;
+    const size_t sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const litEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (litEnd > litLimit) return ERROR(corruption_detected);   /* risk read beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv05_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = litEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase))
+            return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {
+            size_t length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        const int sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv05_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv05_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_8) {
+            ZSTDv05_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd)
+            *op++ = *match++;
+    } else {
+        ZSTDv05_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv05_decompressSequences(
+                               ZSTDv05_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t errorCode, dumpsLength;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    int nbSeq;
+    const BYTE* dumps;
+    U32* DTableLL = dctx->LLTable;
+    U32* DTableML = dctx->MLTable;
+    U32* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+
+    /* Build Decoding Tables */
+    errorCode = ZSTDv05_decodeSeqHeaders(&nbSeq, &dumps, &dumpsLength,
+                                      DTableLL, DTableML, DTableOffb,
+                                      ip, seqSize, dctx->flagStaticTables);
+    if (ZSTDv05_isError(errorCode)) return errorCode;
+    ip += errorCode;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = REPCODE_STARTVALUE;
+        seqState.dumps = dumps;
+        seqState.dumpsEnd = dumps + dumpsLength;
+        seqState.prevOffset = REPCODE_STARTVALUE;
+        errorCode = BITv05_initDStream(&(seqState.DStream), ip, iend-ip);
+        if (ERR_isError(errorCode)) return ERROR(corruption_detected);
+        FSEv05_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv05_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv05_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv05_reloadDStream(&(seqState.DStream)) <= BITv05_DStream_completed) && nbSeq ; ) {
+            size_t oneSeqSize;
+            nbSeq--;
+            ZSTDv05_decodeSequence(&sequence, &seqState);
+            oneSeqSize = ZSTDv05_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTDv05_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+    }
+
+    /* last literal segment */
+    {
+        size_t lastLLSize = litEnd - litPtr;
+        if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+        memcpy(op, litPtr, lastLLSize);
+        op += lastLLSize;
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv05_checkContinuity(ZSTDv05_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv05_decompressBlock_internal(ZSTDv05_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    size_t litCSize;
+
+    if (srcSize >= BLOCKSIZE) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    litCSize = ZSTDv05_decodeLiteralsBlock(dctx, src, srcSize);
+    if (ZSTDv05_isError(litCSize)) return litCSize;
+    ip += litCSize;
+    srcSize -= litCSize;
+
+    return ZSTDv05_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv05_decompressBlock(ZSTDv05_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+/*! ZSTDv05_decompress_continueDCtx
+*   dctx must have been properly initialized */
+static size_t ZSTDv05_decompress_continueDCtx(ZSTDv05_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + maxDstSize;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    {
+        size_t frameHeaderSize;
+        if (srcSize < ZSTDv05_frameHeaderSize_min+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
+        frameHeaderSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
+        if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv05_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        frameHeaderSize = ZSTDv05_decodeFrameHeader_Part2(dctx, src, frameHeaderSize);
+        if (ZSTDv05_isError(frameHeaderSize)) return frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t decodedSize=0;
+        size_t cBlockSize = ZSTDv05_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv05_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv05_blockHeaderSize;
+        remainingSize -= ZSTDv05_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv05_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv05_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv05_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+size_t ZSTDv05_decompress_usingPreparedDCtx(ZSTDv05_DCtx* dctx, const ZSTDv05_DCtx* refDCtx,
+                                         void* dst, size_t maxDstSize,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv05_copyDCtx(dctx, refDCtx);
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompress_continueDCtx(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+size_t ZSTDv05_decompress_usingDict(ZSTDv05_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv05_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv05_checkContinuity(dctx, dst);
+    return ZSTDv05_decompress_continueDCtx(dctx, dst, maxDstSize, src, srcSize);
+}
+
+
+size_t ZSTDv05_decompressDCtx(ZSTDv05_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return ZSTDv05_decompress_usingDict(dctx, dst, maxDstSize, src, srcSize, NULL, 0);
+}
+
+size_t ZSTDv05_decompress(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv05_HEAPMODE) && (ZSTDv05_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv05_DCtx* dctx = ZSTDv05_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv05_decompressDCtx(dctx, dst, maxDstSize, src, srcSize);
+    ZSTDv05_freeDCtx(dctx);
+    return regenSize;
+#else
+    ZSTDv05_DCtx dctx;
+    return ZSTDv05_decompressDCtx(&dctx, dst, maxDstSize, src, srcSize);
+#endif
+}
+
+size_t ZSTDv05_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties;
+
+    /* Frame Header */
+    if (srcSize < ZSTDv05_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    if (MEM_readLE32(src) != ZSTDv05_MAGICNUMBER) return ERROR(prefix_unknown);
+    ip += ZSTDv05_frameHeaderSize_min; remainingSize -= ZSTDv05_frameHeaderSize_min;
+
+    /* Loop on each block */
+    while (1)
+    {
+        size_t cBlockSize = ZSTDv05_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv05_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv05_blockHeaderSize;
+        remainingSize -= ZSTDv05_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+/* ******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv05_nextSrcSizeToDecompress(ZSTDv05_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+size_t ZSTDv05_decompressContinue(ZSTDv05_DCtx* dctx, void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    ZSTDv05_checkContinuity(dctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (dctx->stage)
+    {
+    case ZSTDv05ds_getFrameHeaderSize :
+        /* get frame header size */
+        if (srcSize != ZSTDv05_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        dctx->headerSize = ZSTDv05_decodeFrameHeader_Part1(dctx, src, ZSTDv05_frameHeaderSize_min);
+        if (ZSTDv05_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv05_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv05_frameHeaderSize_min) return ERROR(GENERIC); /* should never happen */
+        dctx->expected = 0;   /* not necessary to copy more */
+        /* fallthrough */
+    case ZSTDv05ds_decodeFrameHeader:
+        /* get frame header */
+        {   size_t const result = ZSTDv05_decodeFrameHeader_Part2(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv05_isError(result)) return result;
+            dctx->expected = ZSTDv05_blockHeaderSize;
+            dctx->stage = ZSTDv05ds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDv05ds_decodeBlockHeader:
+        {
+            /* Decode block header */
+            blockProperties_t bp;
+            size_t blockSize = ZSTDv05_getcBlockSize(src, ZSTDv05_blockHeaderSize, &bp);
+            if (ZSTDv05_isError(blockSize)) return blockSize;
+            if (bp.blockType == bt_end) {
+                dctx->expected = 0;
+                dctx->stage = ZSTDv05ds_getFrameHeaderSize;
+            }
+            else {
+                dctx->expected = blockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDv05ds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDv05ds_decompressBlock:
+        {
+            /* Decompress : block content */
+            size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv05_decompressBlock_internal(dctx, dst, maxDstSize, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv05_copyRawBlock(dst, maxDstSize, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDv05ds_decodeBlockHeader;
+            dctx->expected = ZSTDv05_blockHeaderSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTDv05_refDictContent(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+static size_t ZSTDv05_loadEntropy(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t hSize, offcodeHeaderSize, matchlengthHeaderSize, errorCode, litlengthHeaderSize;
+    short offcodeNCount[MaxOff+1];
+    U32 offcodeMaxValue=MaxOff, offcodeLog;
+    short matchlengthNCount[MaxML+1];
+    unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+    short litlengthNCount[MaxLL+1];
+    unsigned litlengthMaxValue = MaxLL, litlengthLog;
+
+    hSize = HUFv05_readDTableX4(dctx->hufTableX4, dict, dictSize);
+    if (HUFv05_isError(hSize)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + hSize;
+    dictSize -= hSize;
+
+    offcodeHeaderSize = FSEv05_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dict, dictSize);
+    if (FSEv05_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+    if (offcodeLog > OffFSEv05Log) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + offcodeHeaderSize;
+    dictSize -= offcodeHeaderSize;
+
+    matchlengthHeaderSize = FSEv05_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dict, dictSize);
+    if (FSEv05_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+    if (matchlengthLog > MLFSEv05Log) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + matchlengthHeaderSize;
+    dictSize -= matchlengthHeaderSize;
+
+    litlengthHeaderSize = FSEv05_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dict, dictSize);
+    if (litlengthLog > LLFSEv05Log) return ERROR(dictionary_corrupted);
+    if (FSEv05_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+    errorCode = FSEv05_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+    if (FSEv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+
+    dctx->flagStaticTables = 1;
+    return hSize + offcodeHeaderSize + matchlengthHeaderSize + litlengthHeaderSize;
+}
+
+static size_t ZSTDv05_decompress_insertDictionary(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t eSize;
+    U32 magic = MEM_readLE32(dict);
+    if (magic != ZSTDv05_DICT_MAGIC) {
+        /* pure content mode */
+        ZSTDv05_refDictContent(dctx, dict, dictSize);
+        return 0;
+    }
+    /* load entropy tables */
+    dict = (const char*)dict + 4;
+    dictSize -= 4;
+    eSize = ZSTDv05_loadEntropy(dctx, dict, dictSize);
+    if (ZSTDv05_isError(eSize)) return ERROR(dictionary_corrupted);
+
+    /* reference dictionary content */
+    dict = (const char*)dict + eSize;
+    dictSize -= eSize;
+    ZSTDv05_refDictContent(dctx, dict, dictSize);
+
+    return 0;
+}
+
+
+size_t ZSTDv05_decompressBegin_usingDict(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t errorCode;
+    errorCode = ZSTDv05_decompressBegin(dctx);
+    if (ZSTDv05_isError(errorCode)) return errorCode;
+
+    if (dict && dictSize) {
+        errorCode = ZSTDv05_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv05_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd source repository : https://github.com/Cyan4973/zstd
+    - ztsd public forum : https://groups.google.com/forum/#!forum/lz4c
+*/
+
+/* The objects defined into this file should be considered experimental.
+ * They are not labelled stable, as their prototype may change in the future.
+ * You can use them for tests, provide feedback, or if you can endure risk of future changes.
+ */
+
+
+
+/* *************************************
+*  Constants
+***************************************/
+static size_t ZBUFFv05_blockHeaderSize = 3;
+
+
+
+/* *** Compression *** */
+
+static size_t ZBUFFv05_limitCopy(void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    size_t length = MIN(maxDstSize, srcSize);
+    memcpy(dst, src, length);
+    return length;
+}
+
+
+
+
+/** ************************************************
+*  Streaming decompression
+*
+*  A ZBUFFv05_DCtx object is required to track streaming operation.
+*  Use ZBUFFv05_createDCtx() and ZBUFFv05_freeDCtx() to create/release resources.
+*  Use ZBUFFv05_decompressInit() to start a new decompression operation.
+*  ZBUFFv05_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFFv05_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *maxDstSizePtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *maxDstSizePtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
+*  The content of dst will be overwritten (up to *maxDstSizePtr) at each function call, so save its content if it matters or change dst .
+*  return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFFv05_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory)
+*  output : 128 KB block size is the internal unit, it ensures it's always possible to write a full block when it's decoded.
+*  input : just follow indications from ZBUFFv05_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* **************************************************/
+
+typedef enum { ZBUFFv05ds_init, ZBUFFv05ds_readHeader, ZBUFFv05ds_loadHeader, ZBUFFv05ds_decodeHeader,
+               ZBUFFv05ds_read, ZBUFFv05ds_load, ZBUFFv05ds_flush } ZBUFFv05_dStage;
+
+/* *** Resource management *** */
+
+#define ZSTDv05_frameHeaderSize_max 5   /* too magical, should come from reference */
+struct ZBUFFv05_DCtx_s {
+    ZSTDv05_DCtx* zc;
+    ZSTDv05_parameters params;
+    char* inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char* outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t hPos;
+    ZBUFFv05_dStage stage;
+    unsigned char headerBuffer[ZSTDv05_frameHeaderSize_max];
+};   /* typedef'd to ZBUFFv05_DCtx within "zstd_buffered.h" */
+
+
+ZBUFFv05_DCtx* ZBUFFv05_createDCtx(void)
+{
+    ZBUFFv05_DCtx* zbc = (ZBUFFv05_DCtx*)malloc(sizeof(ZBUFFv05_DCtx));
+    if (zbc==NULL) return NULL;
+    memset(zbc, 0, sizeof(*zbc));
+    zbc->zc = ZSTDv05_createDCtx();
+    zbc->stage = ZBUFFv05ds_init;
+    return zbc;
+}
+
+size_t ZBUFFv05_freeDCtx(ZBUFFv05_DCtx* zbc)
+{
+    if (zbc==NULL) return 0;   /* support free on null */
+    ZSTDv05_freeDCtx(zbc->zc);
+    free(zbc->inBuff);
+    free(zbc->outBuff);
+    free(zbc);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv05_decompressInitDictionary(ZBUFFv05_DCtx* zbc, const void* dict, size_t dictSize)
+{
+    zbc->stage = ZBUFFv05ds_readHeader;
+    zbc->hPos = zbc->inPos = zbc->outStart = zbc->outEnd = 0;
+    return ZSTDv05_decompressBegin_usingDict(zbc->zc, dict, dictSize);
+}
+
+size_t ZBUFFv05_decompressInit(ZBUFFv05_DCtx* zbc)
+{
+    return ZBUFFv05_decompressInitDictionary(zbc, NULL, 0);
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv05_decompressContinue(ZBUFFv05_DCtx* zbc, void* dst, size_t* maxDstSizePtr, const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* ip = istart;
+    const char* const iend = istart + *srcSizePtr;
+    char* const ostart = (char*)dst;
+    char* op = ostart;
+    char* const oend = ostart + *maxDstSizePtr;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbc->stage)
+        {
+        case ZBUFFv05ds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFv05ds_readHeader :
+            /* read header from src */
+            {
+                size_t headerSize = ZSTDv05_getFrameParams(&(zbc->params), src, *srcSizePtr);
+                if (ZSTDv05_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    memcpy(zbc->headerBuffer+zbc->hPos, src, *srcSizePtr);
+                    zbc->hPos += *srcSizePtr;
+                    *maxDstSizePtr = 0;
+                    zbc->stage = ZBUFFv05ds_loadHeader;
+                    return headerSize - zbc->hPos;
+                }
+                zbc->stage = ZBUFFv05ds_decodeHeader;
+                break;
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_loadHeader:
+            /* complete header from src */
+            {
+                size_t headerSize = ZBUFFv05_limitCopy(
+                    zbc->headerBuffer + zbc->hPos, ZSTDv05_frameHeaderSize_max - zbc->hPos,
+                    src, *srcSizePtr);
+                zbc->hPos += headerSize;
+                ip += headerSize;
+                headerSize = ZSTDv05_getFrameParams(&(zbc->params), zbc->headerBuffer, zbc->hPos);
+                if (ZSTDv05_isError(headerSize)) return headerSize;
+                if (headerSize) {
+                    /* not enough input to decode header : tell how many bytes would be necessary */
+                    *maxDstSizePtr = 0;
+                    return headerSize - zbc->hPos;
+                }
+                // zbc->stage = ZBUFFv05ds_decodeHeader; break;   /* useless : stage follows */
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_decodeHeader:
+                /* apply header to create / resize buffers */
+                {
+                    size_t neededOutSize = (size_t)1 << zbc->params.windowLog;
+                    size_t neededInSize = BLOCKSIZE;   /* a block is never > BLOCKSIZE */
+                    if (zbc->inBuffSize < neededInSize) {
+                        free(zbc->inBuff);
+                        zbc->inBuffSize = neededInSize;
+                        zbc->inBuff = (char*)malloc(neededInSize);
+                        if (zbc->inBuff == NULL) return ERROR(memory_allocation);
+                    }
+                    if (zbc->outBuffSize < neededOutSize) {
+                        free(zbc->outBuff);
+                        zbc->outBuffSize = neededOutSize;
+                        zbc->outBuff = (char*)malloc(neededOutSize);
+                        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
+                }   }
+                if (zbc->hPos) {
+                    /* some data already loaded into headerBuffer : transfer into inBuff */
+                    memcpy(zbc->inBuff, zbc->headerBuffer, zbc->hPos);
+                    zbc->inPos = zbc->hPos;
+                    zbc->hPos = 0;
+                    zbc->stage = ZBUFFv05ds_load;
+                    break;
+                }
+                zbc->stage = ZBUFFv05ds_read;
+		/* fall-through */
+        case ZBUFFv05ds_read:
+            {
+                size_t neededInSize = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+                if (neededInSize==0) {  /* end of frame */
+                    zbc->stage = ZBUFFv05ds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {
+                    /* directly decode from src */
+                    size_t decodedSize = ZSTDv05_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        ip, neededInSize);
+                    if (ZSTDv05_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFv05ds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbc->stage = ZBUFFv05ds_load;
+            }
+	    /* fall-through */
+        case ZBUFFv05ds_load:
+            {
+                size_t neededInSize = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+                size_t toLoad = neededInSize - zbc->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbc->inBuffSize - zbc->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv05_limitCopy(zbc->inBuff + zbc->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbc->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+                {
+                    size_t decodedSize = ZSTDv05_decompressContinue(zbc->zc,
+                        zbc->outBuff + zbc->outStart, zbc->outBuffSize - zbc->outStart,
+                        zbc->inBuff, neededInSize);
+                    if (ZSTDv05_isError(decodedSize)) return decodedSize;
+                    zbc->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbc->stage = ZBUFFv05ds_read; break; }   /* this was just a header */
+                    zbc->outEnd = zbc->outStart +  decodedSize;
+                    zbc->stage = ZBUFFv05ds_flush;
+                    // break; /* ZBUFFv05ds_flush follows */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFv05ds_flush:
+            {
+                size_t toFlushSize = zbc->outEnd - zbc->outStart;
+                size_t flushedSize = ZBUFFv05_limitCopy(op, oend-op, zbc->outBuff + zbc->outStart, toFlushSize);
+                op += flushedSize;
+                zbc->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbc->stage = ZBUFFv05ds_read;
+                    if (zbc->outStart + BLOCKSIZE > zbc->outBuffSize)
+                        zbc->outStart = zbc->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    *srcSizePtr = ip-istart;
+    *maxDstSizePtr = op-ostart;
+
+    {   size_t nextSrcSizeHint = ZSTDv05_nextSrcSizeToDecompress(zbc->zc);
+        if (nextSrcSizeHint > ZBUFFv05_blockHeaderSize) nextSrcSizeHint+= ZBUFFv05_blockHeaderSize;   /* get next block header too */
+        nextSrcSizeHint -= zbc->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv05_isError(size_t errorCode) { return ERR_isError(errorCode); }
+const char* ZBUFFv05_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+size_t ZBUFFv05_recommendedDInSize(void)  { return BLOCKSIZE + ZBUFFv05_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv05_recommendedDOutSize(void) { return BLOCKSIZE; }
diff --git a/vendor/github.com/DataDog/zstd/zstd_v05.h b/vendor/github.com/DataDog/zstd/zstd_v05.h
new file mode 100644
index 0000000..b68fd57
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v05.h
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTDv05_H
+#define ZSTDv05_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stddef.h>   /* size_t */
+#include "mem.h"      /* U64, U32 */
+
+
+/* *************************************
+*  Simple functions
+***************************************/
+/*! ZSTDv05_decompress() :
+    `compressedSize` : is the _exact_ size of the compressed blob, otherwise decompression will fail.
+    `dstCapacity` must be large enough, equal or larger than originalSize.
+    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+              or an errorCode if it fails (which can be tested using ZSTDv05_isError()) */
+size_t ZSTDv05_decompress( void* dst, size_t dstCapacity,
+                     const void* src, size_t compressedSize);
+
+/**
+ZSTDv05_getFrameSrcSize() : get the source length of a ZSTD frame
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv05_isError())
+*/
+size_t ZSTDv05_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+/* *************************************
+*  Helper functions
+***************************************/
+/* Error Management */
+unsigned    ZSTDv05_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+const char* ZSTDv05_getErrorName(size_t code);     /*!< provides readable string for an error code */
+
+
+/* *************************************
+*  Explicit memory management
+***************************************/
+/** Decompression context */
+typedef struct ZSTDv05_DCtx_s ZSTDv05_DCtx;
+ZSTDv05_DCtx* ZSTDv05_createDCtx(void);
+size_t ZSTDv05_freeDCtx(ZSTDv05_DCtx* dctx);      /*!< @return : errorCode */
+
+/** ZSTDv05_decompressDCtx() :
+*   Same as ZSTDv05_decompress(), but requires an already allocated ZSTDv05_DCtx (see ZSTDv05_createDCtx()) */
+size_t ZSTDv05_decompressDCtx(ZSTDv05_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-***********************
+*  Simple Dictionary API
+*************************/
+/*! ZSTDv05_decompress_usingDict() :
+*   Decompression using a pre-defined Dictionary content (see dictBuilder).
+*   Dictionary must be identical to the one used during compression, otherwise regenerated data will be corrupted.
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTDv05_decompressDCtx() */
+size_t ZSTDv05_decompress_usingDict(ZSTDv05_DCtx* dctx,
+                                            void* dst, size_t dstCapacity,
+                                      const void* src, size_t srcSize,
+                                      const void* dict,size_t dictSize);
+
+/*-************************
+*  Advanced Streaming API
+***************************/
+typedef enum { ZSTDv05_fast, ZSTDv05_greedy, ZSTDv05_lazy, ZSTDv05_lazy2, ZSTDv05_btlazy2, ZSTDv05_opt, ZSTDv05_btopt } ZSTDv05_strategy;
+typedef struct {
+    U64 srcSize;
+    U32 windowLog;     /* the only useful information to retrieve */
+    U32 contentLog; U32 hashLog; U32 searchLog; U32 searchLength; U32 targetLength; ZSTDv05_strategy strategy;
+} ZSTDv05_parameters;
+size_t ZSTDv05_getFrameParams(ZSTDv05_parameters* params, const void* src, size_t srcSize);
+
+size_t ZSTDv05_decompressBegin_usingDict(ZSTDv05_DCtx* dctx, const void* dict, size_t dictSize);
+void   ZSTDv05_copyDCtx(ZSTDv05_DCtx* dstDCtx, const ZSTDv05_DCtx* srcDCtx);
+size_t ZSTDv05_nextSrcSizeToDecompress(ZSTDv05_DCtx* dctx);
+size_t ZSTDv05_decompressContinue(ZSTDv05_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-***********************
+*  ZBUFF API
+*************************/
+typedef struct ZBUFFv05_DCtx_s ZBUFFv05_DCtx;
+ZBUFFv05_DCtx* ZBUFFv05_createDCtx(void);
+size_t         ZBUFFv05_freeDCtx(ZBUFFv05_DCtx* dctx);
+
+size_t ZBUFFv05_decompressInit(ZBUFFv05_DCtx* dctx);
+size_t ZBUFFv05_decompressInitDictionary(ZBUFFv05_DCtx* dctx, const void* dict, size_t dictSize);
+
+size_t ZBUFFv05_decompressContinue(ZBUFFv05_DCtx* dctx,
+                                            void* dst, size_t* dstCapacityPtr,
+                                      const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression
+*
+*  A ZBUFFv05_DCtx object is required to track streaming operations.
+*  Use ZBUFFv05_createDCtx() and ZBUFFv05_freeDCtx() to create/release resources.
+*  Use ZBUFFv05_decompressInit() to start a new decompression operation,
+*   or ZBUFFv05_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv05_DCtx objects can be reused multiple times.
+*
+*  Use ZBUFFv05_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency)
+*            or 0 when a frame is completely decoded
+*            or an error code, which can be tested using ZBUFFv05_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv05_recommendedDInSize() / ZBUFFv05_recommendedDOutSize()
+*  output : ZBUFFv05_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv05_recommendedDInSize==128Kb+3; just follow indications from ZBUFFv05_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+unsigned ZBUFFv05_isError(size_t errorCode);
+const char* ZBUFFv05_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, and tend to offer better latency */
+size_t ZBUFFv05_recommendedDInSize(void);
+size_t ZBUFFv05_recommendedDOutSize(void);
+
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ZSTDv05_MAGICNUMBER 0xFD2FB525   /* v0.5 */
+
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv0505_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_v06.c b/vendor/github.com/DataDog/zstd/zstd_v06.c
new file mode 100644
index 0000000..8b068b3
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v06.c
@@ -0,0 +1,4124 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include "zstd_v06.h"
+#include <stddef.h>    /* size_t, ptrdiff_t */
+#include <string.h>    /* memcpy */
+#include <stdlib.h>    /* malloc, free, qsort */
+#include "error_private.h"
+
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS :
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets depending on alignment.
+ *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard, by lying on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+
+#endif /* MEM_FORCE_MEMORY_ACCESS */
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+
+/*
+    zstd - standard compression library
+    Header File for static linking only
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net
+*/
+#ifndef ZSTDv06_STATIC_H
+#define ZSTDv06_STATIC_H
+
+/* The prototypes defined within this file are considered experimental.
+ * They should not be used in the context DLL as they may change in the future.
+ * Prefer static linking if you need them, to control breaking version changes issues.
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*- Advanced Decompression functions -*/
+
+/*! ZSTDv06_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv06_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv06_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
+ZSTDLIBv06_API size_t ZSTDv06_decompress_usingPreparedDCtx(
+                                           ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* preparedDCtx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize);
+
+
+
+#define ZSTDv06_FRAMEHEADERSIZE_MAX 13    /* for static allocation */
+static const size_t ZSTDv06_frameHeaderSize_min = 5;
+static const size_t ZSTDv06_frameHeaderSize_max = ZSTDv06_FRAMEHEADERSIZE_MAX;
+
+ZSTDLIBv06_API size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx);
+
+/*
+  Streaming decompression, direct mode (bufferless)
+
+  A ZSTDv06_DCtx object is required to track streaming operations.
+  Use ZSTDv06_createDCtx() / ZSTDv06_freeDCtx() to manage it.
+  A ZSTDv06_DCtx object can be re-used multiple times.
+
+  First optional operation is to retrieve frame parameters, using ZSTDv06_getFrameParams(), which doesn't consume the input.
+  It can provide the minimum size of rolling buffer required to properly decompress data,
+  and optionally the final size of uncompressed content.
+  (Note : content size is an optional info that may not be present. 0 means : content size unknown)
+  Frame parameters are extracted from the beginning of compressed frame.
+  The amount of data to read is variable, from ZSTDv06_frameHeaderSize_min to ZSTDv06_frameHeaderSize_max (so if `srcSize` >= ZSTDv06_frameHeaderSize_max, it will always work)
+  If `srcSize` is too small for operation to succeed, function will return the minimum size it requires to produce a result.
+  Result : 0 when successful, it means the ZSTDv06_frameParams structure has been filled.
+          >0 : means there is not enough data into `src`. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv06_isError()
+
+  Start decompression, with ZSTDv06_decompressBegin() or ZSTDv06_decompressBegin_usingDict().
+  Alternatively, you can copy a prepared context, using ZSTDv06_copyDCtx().
+
+  Then use ZSTDv06_nextSrcSizeToDecompress() and ZSTDv06_decompressContinue() alternatively.
+  ZSTDv06_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv06_decompressContinue().
+  ZSTDv06_decompressContinue() requires this exact amount of bytes, or it will fail.
+  ZSTDv06_decompressContinue() needs previous data blocks during decompression, up to (1 << windowlog).
+  They should preferably be located contiguously, prior to current block. Alternatively, a round buffer is also possible.
+
+  @result of ZSTDv06_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity)
+  It can be zero, which is not an error; it just means ZSTDv06_decompressContinue() has decoded some header.
+
+  A frame is fully decoded when ZSTDv06_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Uncompressed block size must be <= ZSTDv06_BLOCKSIZE_MAX (128 KB)
+    - Compressing or decompressing requires a context structure
+      + Use ZSTDv06_createCCtx() and ZSTDv06_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv06_compressBegin()
+      + decompression : ZSTDv06_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - When a block is considered not compressible enough, ZSTDv06_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv06_decompressBlock() doesn't accept uncompressed data as input !!
+*/
+
+#define ZSTDv06_BLOCKSIZE_MAX (128 * 1024)   /* define, for static allocation */
+ZSTDLIBv06_API size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv06_STATIC_H */
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://www.zstd.net
+*/
+#ifndef ZSTDv06_CCOMMON_H_MODULE
+#define ZSTDv06_CCOMMON_H_MODULE
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv06_DICT_MAGIC  0xEC30A436
+
+#define ZSTDv06_REP_NUM    3
+#define ZSTDv06_REP_INIT   ZSTDv06_REP_NUM
+#define ZSTDv06_REP_MOVE   (ZSTDv06_REP_NUM-1)
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTDv06_WINDOWLOG_ABSOLUTEMIN 12
+static const size_t ZSTDv06_fcs_fieldSize[4] = { 0, 1, 2, 8 };
+
+#define ZSTDv06_BLOCKHEADERSIZE 3   /* because C standard does not allow a static const value to be defined using another static const value .... :( */
+static const size_t ZSTDv06_blockHeaderSize = ZSTDv06_BLOCKHEADERSIZE;
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define HufLog 12
+
+#define IS_HUF 0
+#define IS_PCH 1
+#define IS_RAW 2
+#define IS_RLE 3
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+#define EQUAL_READ32 4
+#define REPCODE_STARTVALUE 1
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  52
+#define MaxLL  35
+#define MaxOff 28
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+
+#define FSEv06_ENCODING_RAW     0
+#define FSEv06_ENCODING_RLE     1
+#define FSEv06_ENCODING_STATIC  2
+#define FSEv06_ENCODING_DYNAMIC 3
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+static const U32 LL_defaultNormLog = 6;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+static const U32 ML_defaultNormLog = 6;
+
+static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const U32 OF_defaultNormLog = 5;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv06_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTDv06_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv06_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTDv06_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTDv06_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTDv06_REP_INIT];
+} ZSTDv06_optimal_t;
+
+typedef struct { U32  unused; } ZSTDv06_stats_t;
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* litStart;
+    BYTE* lit;
+    U16*  litLengthStart;
+    U16*  litLength;
+    BYTE* llCodeStart;
+    U16*  matchLengthStart;
+    U16*  matchLength;
+    BYTE* mlCodeStart;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+    /* opt */
+    ZSTDv06_optimal_t* priceTable;
+    ZSTDv06_match_t* matchTable;
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  matchSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+    U32  log2matchLengthSum;
+    U32  log2matchSum;
+    U32  log2litLengthSum;
+    U32  log2litSum;
+    U32  log2offCodeSum;
+    U32  factor;
+    U32  cachedPrice;
+    U32  cachedLitLength;
+    const BYTE* cachedLiterals;
+    ZSTDv06_stats_t stats;
+} seqStore_t;
+
+void ZSTDv06_seqToCodes(const seqStore_t* seqStorePtr, size_t const nbSeq);
+
+
+#endif   /* ZSTDv06_CCOMMON_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef FSEv06_H
+#define FSEv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+/*! FSEv06_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSEv06_isError() .
+
+    ** Important ** : FSEv06_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+size_t FSEv06_decompress(void* dst,  size_t dstCapacity,
+                const void* cSrc, size_t cSrcSize);
+
+
+/*-*****************************************
+*  Tool functions
+******************************************/
+size_t FSEv06_compressBound(size_t size);       /* maximum compressed size */
+
+/* Error Management */
+unsigned    FSEv06_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv06_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+
+FSEv06_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSEv06_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSEv06_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSEv06_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv06_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv06_DTable* FSEv06_createDTable(unsigned tableLog);
+void        FSEv06_freeDTable(FSEv06_DTable* dt);
+
+/*! FSEv06_buildDTable():
+    Builds 'dt', which must be already allocated, using FSEv06_createDTable().
+    return : 0, or an errorCode, which can be tested using FSEv06_isError() */
+size_t FSEv06_buildDTable (FSEv06_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSEv06_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSEv06_isError() */
+size_t FSEv06_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv06_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSEv06_readNCount() if it was saved using FSEv06_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSEv06_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSEv06_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError().
+
+The next step is to build the decompression tables 'FSEv06_DTable' from 'normalizedCounter'.
+This is performed by the function FSEv06_buildDTable().
+The space required by 'FSEv06_DTable' must be already allocated using FSEv06_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError().
+
+`FSEv06_DTable` can then be used to decompress `cSrc`, with FSEv06_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSEv06_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSEv06_isError(). (ex: dst buffer too small)
+*/
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv06_H */
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv06_DStream_t;
+
+typedef enum { BITv06_DStream_unfinished = 0,
+               BITv06_DStream_endOfBuffer = 1,
+               BITv06_DStream_completed = 2,
+               BITv06_DStream_overflow = 3 } BITv06_DStream_status;  /* result of BITv06_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv06_readBits(BITv06_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv06_DStream_status BITv06_reloadDStream(BITv06_DStream_t* bitD);
+MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* bitD);
+
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BITv06_highbit32 ( U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r=0;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    unsigned r;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    r = DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+    return r;
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*! BITv06_initDStream() :
+*   Initialize a BITv06_DStream_t.
+*   `bitD` : a pointer to an already allocated BITv06_DStream_t structure.
+*   `srcSize` must be the *exact* size of the bitStream, in bytes.
+*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv06_initDStream(BITv06_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          if (lastByte == 0) return ERROR(GENERIC);   /* endMark not present */
+          bitD->bitsConsumed = 8 - BITv06_highbit32(lastByte); }
+    } else {
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          if (lastByte == 0) return ERROR(GENERIC);   /* endMark not present */
+          bitD->bitsConsumed = 8 - BITv06_highbit32(lastByte); }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+ MEM_STATIC size_t BITv06_lookBits(const BITv06_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv06_lookBitsFast() :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BITv06_lookBitsFast(const BITv06_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv06_skipBits(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv06_readBits(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv06_lookBits(bitD, nbBits);
+    BITv06_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BITv06_readBitsFast() :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BITv06_readBitsFast(BITv06_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv06_lookBitsFast(bitD, nbBits);
+    BITv06_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv06_DStream_status BITv06_reloadDStream(BITv06_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should never happen */
+        return BITv06_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv06_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv06_DStream_endOfBuffer;
+        return BITv06_DStream_completed;
+    }
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv06_DStream_status result = BITv06_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv06_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv06_endOfDStream() :
+*   @return Tells if DStream has exactly reached its end (all bits consumed).
+*/
+MEM_STATIC unsigned BITv06_endOfDStream(const BITv06_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy coder
+   header file for static linking (only)
+   Copyright (C) 2013-2015, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+   - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef FSEv06_STATIC_H
+#define FSEv06_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSEv06_NCOUNTBOUND 512
+#define FSEv06_BLOCKBOUND(size) (size + (size>>7))
+#define FSEv06_COMPRESSBOUND(size) (FSEv06_NCOUNTBOUND + FSEv06_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSEv06_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+size_t FSEv06_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+/* same as FSEv06_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+size_t FSEv06_buildDTable_raw (FSEv06_DTable* dt, unsigned nbBits);
+/* build a fake FSEv06_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv06_buildDTable_rle (FSEv06_DTable* dt, unsigned char symbolValue);
+/* build a fake FSEv06_DTable, designed to always generate the same symbolValue */
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv06_DState_t;
+
+
+static void     FSEv06_initDState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD, const FSEv06_DTable* dt);
+
+static unsigned char FSEv06_decodeSymbol(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD);
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* *****************************************
+*  Implementation of inlined functions
+*******************************************/
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv06_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv06_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv06_initDState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD, const FSEv06_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv06_DTableHeader* const DTableH = (const FSEv06_DTableHeader*)ptr;
+    DStatePtr->state = BITv06_readBits(bitD, DTableH->tableLog);
+    BITv06_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv06_peekSymbol(const FSEv06_DState_t* DStatePtr)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSEv06_updateState(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BITv06_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSEv06_decodeSymbol(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv06_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSEv06_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSEv06_decodeSymbolFast(FSEv06_DState_t* DStatePtr, BITv06_DStream_t* bitD)
+{
+    FSEv06_decode_t const DInfo = ((const FSEv06_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv06_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+
+
+#ifndef FSEv06_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv06_MAX_MEMORY_USAGE 14
+#define FSEv06_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv06_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv06_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv06_FUNCTION_TYPE BYTE
+#define FSEv06_FUNCTION_EXTENSION
+#define FSEv06_DECODE_TYPE FSEv06_decode_t
+
+
+#endif   /* !FSEv06_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv06_MAX_TABLELOG  (FSEv06_MAX_MEMORY_USAGE-2)
+#define FSEv06_MAX_TABLESIZE (1U<<FSEv06_MAX_TABLELOG)
+#define FSEv06_MAXTABLESIZE_MASK (FSEv06_MAX_TABLESIZE-1)
+#define FSEv06_DEFAULT_TABLELOG (FSEv06_DEFAULT_MEMORY_USAGE-2)
+#define FSEv06_MIN_TABLELOG 5
+
+#define FSEv06_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv06_MAX_TABLELOG > FSEv06_TABLELOG_ABSOLUTE_MAX
+#error "FSEv06_MAX_TABLELOG > FSEv06_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSEv06_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv06_STATIC_H */
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+
+/*-****************************************
+*  FSE Error Management
+******************************************/
+unsigned FSEv06_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  HUF Error Management
+****************************************************************/
+unsigned HUFv06_isError(size_t code) { return ERR_isError(code); }
+
+const char* HUFv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSEv06_abs(short a) { return a<0 ? -a : a; }
+
+size_t FSEv06_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv06_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv06_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {   short const max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv06_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) && (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv06_isError ERR_isError
+#define FSEv06_STATIC_ASSERT(c) { enum { FSEv06_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv06_DTABLE_SIZE_U32(FSEv06_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv06_FUNCTION_EXTENSION
+#  error "FSEv06_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv06_FUNCTION_TYPE
+#  error "FSEv06_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv06_CAT(X,Y) X##Y
+#define FSEv06_FUNCTION_NAME(X,Y) FSEv06_CAT(X,Y)
+#define FSEv06_TYPE_NAME(X,Y) FSEv06_CAT(X,Y)
+
+
+/* Function templates */
+FSEv06_DTable* FSEv06_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv06_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv06_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv06_DTable*)malloc( FSEv06_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv06_freeDTable (FSEv06_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv06_buildDTable(FSEv06_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv06_DECODE_TYPE* const tableDecode = (FSEv06_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSEv06_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv06_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv06_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSEv06_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSEv06_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSEv06_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSEv06_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSEv06_FUNCTION_TYPE const symbol = (FSEv06_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BITv06_highbit32 ((U32)nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+
+#ifndef FSEv06_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv06_buildDTable_rle (FSEv06_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv06_DTableHeader* const DTableH = (FSEv06_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv06_decode_t* const cell = (FSEv06_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv06_buildDTable_raw (FSEv06_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv06_DTableHeader* const DTableH = (FSEv06_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv06_decode_t* const dinfo = (FSEv06_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv06_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv06_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv06_DStream_t bitD;
+    FSEv06_DState_t state1;
+    FSEv06_DState_t state2;
+
+    /* Init */
+    { size_t const errorCode = BITv06_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+      if (FSEv06_isError(errorCode)) return errorCode; }
+
+    FSEv06_initDState(&state1, &bitD, dt);
+    FSEv06_initDState(&state2, &bitD, dt);
+
+#define FSEv06_GETSYMBOL(statePtr) fast ? FSEv06_decodeSymbolFast(statePtr, &bitD) : FSEv06_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv06_reloadDStream(&bitD)==BITv06_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv06_GETSYMBOL(&state1);
+
+        if (FSEv06_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv06_reloadDStream(&bitD);
+
+        op[1] = FSEv06_GETSYMBOL(&state2);
+
+        if (FSEv06_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv06_reloadDStream(&bitD) > BITv06_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv06_GETSYMBOL(&state1);
+
+        if (FSEv06_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv06_reloadDStream(&bitD);
+
+        op[3] = FSEv06_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv06_reloadDStream(&bitD) >= FSEv06_DStream_partiallyFilled; Ends at exactly BITv06_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv06_GETSYMBOL(&state1);
+
+        if (BITv06_reloadDStream(&bitD)==BITv06_DStream_overflow) {
+            *op++ = FSEv06_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv06_GETSYMBOL(&state2);
+
+        if (BITv06_reloadDStream(&bitD)==BITv06_DStream_overflow) {
+            *op++ = FSEv06_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSEv06_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv06_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv06_DTableHeader* DTableH = (const FSEv06_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv06_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv06_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv06_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv06_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv06_MAX_SYMBOL_VALUE;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    {   size_t const NCountLength = FSEv06_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+        if (FSEv06_isError(NCountLength)) return NCountLength;
+        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    { size_t const errorCode = FSEv06_buildDTable (dt, counting, maxSymbolValue, tableLog);
+      if (FSEv06_isError(errorCode)) return errorCode; }
+
+    return FSEv06_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+}
+
+
+
+#endif   /* FSEv06_COMMONDEFS_ONLY */
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv06_H
+#define HUFv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  HUF simple functions
+******************************************/
+size_t HUFv06_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+/*
+HUFv06_decompress() :
+    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstSize'.
+    `dstSize` : must be the **exact** size of original (uncompressed) data.
+    Note : in contrast with FSE, HUFv06_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize)
+              or an error code, which can be tested using HUFv06_isError()
+*/
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+size_t HUFv06_compressBound(size_t size);       /**< maximum compressed size */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFv06_H */
+/* ******************************************************************
+   Huffman codec, part of New Generation Entropy library
+   header file, for static linking only
+   Copyright (C) 2013-2016, Yann Collet
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv06_STATIC_H
+#define HUFv06_STATIC_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUFv06_CTABLEBOUND 129
+#define HUFv06_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUFv06_COMPRESSBOUND(size) (HUFv06_CTABLEBOUND + HUFv06_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
+
+/* static allocation of HUF's DTable */
+#define HUFv06_DTABLE_SIZE(maxTableLog)   (1 + (1<<maxTableLog))
+#define HUFv06_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        unsigned short DTable[HUFv06_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv06_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        unsigned int DTable[HUFv06_DTABLE_SIZE(maxTableLog)] = { maxTableLog }
+#define HUFv06_CREATE_STATIC_DTABLEX6(DTable, maxTableLog) \
+        unsigned int DTable[HUFv06_DTABLE_SIZE(maxTableLog) * 3 / 2] = { maxTableLog }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv06_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv06_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbols decoder */
+
+
+
+/*!
+HUFv06_decompress() does the following:
+1. select the decompression algorithm (X2, X4, X6) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv06_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv06_decompressSXn_usingDTable
+*/
+size_t HUFv06_readDTableX2 (unsigned short* DTable, const void* src, size_t srcSize);
+size_t HUFv06_readDTableX4 (unsigned* DTable, const void* src, size_t srcSize);
+
+size_t HUFv06_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv06_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+/* single stream variants */
+size_t HUFv06_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv06_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv06_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned short* DTable);
+size_t HUFv06_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const unsigned* DTable);
+
+
+
+/* **************************************************************
+*  Constants
+****************************************************************/
+#define HUFv06_ABSOLUTEMAX_TABLELOG  16   /* absolute limit of HUFv06_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv06_MAX_TABLELOG  12           /* max configured tableLog (for static allocation); can be modified up to HUFv06_ABSOLUTEMAX_TABLELOG */
+#define HUFv06_DEFAULT_TABLELOG  HUFv06_MAX_TABLELOG   /* tableLog by default, when not specified */
+#define HUFv06_MAX_SYMBOL_VALUE 255
+#if (HUFv06_MAX_TABLELOG > HUFv06_ABSOLUTEMAX_TABLELOG)
+#  error "HUFv06_MAX_TABLELOG is too large !"
+#endif
+
+
+
+/*! HUFv06_readStats() :
+    Read compact Huffman tree, saved by HUFv06_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src`
+*/
+MEM_STATIC size_t HUFv06_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                            U32* nbSymbolsPtr, U32* tableLogPtr,
+                            const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static U32 l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            {   U32 n;
+                for (n=0; n<oSize; n+=2) {
+                    huffWeight[n]   = ip[n/2] >> 4;
+                    huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv06_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv06_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv06_ABSOLUTEMAX_TABLELOG + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BITv06_highbit32(weightTotal) + 1;
+        if (tableLog > HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BITv06_highbit32(rest);
+            U32 const lastWeight = BITv06_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* HUFv06_STATIC_H */
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUFv06_STATIC_ASSERT(c) { enum { HUFv06_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+
+/* *******************************************************
+*  HUF : Huffman block decompression
+*********************************************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUFv06_DEltX2;   /* single-symbol decoding */
+
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv06_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+size_t HUFv06_readDTableX2 (U16* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv06_MAX_SYMBOL_VALUE + 1];
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    size_t iSize;
+    U32 nbSymbols = 0;
+    U32 n;
+    U32 nextRankStart;
+    void* const dtPtr = DTable + 1;
+    HUFv06_DEltX2* const dt = (HUFv06_DEltX2*)dtPtr;
+
+    HUFv06_STATIC_ASSERT(sizeof(HUFv06_DEltX2) == sizeof(U16));   /* if compilation fails here, assertion is false */
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv06_readStats(huffWeight, HUFv06_MAX_SYMBOL_VALUE + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv06_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > DTable[0]) return ERROR(tableLog_tooLarge);   /* DTable is too small */
+    DTable[0] = (U16)tableLog;   /* maybe should separate sizeof allocated DTable, from used size of DTable, in case of re-use */
+
+    /* Prepare ranks */
+    nextRankStart = 0;
+    for (n=1; n<tableLog+1; n++) {
+        U32 current = nextRankStart;
+        nextRankStart += (rankVal[n] << (n-1));
+        rankVal[n] = current;
+    }
+
+    /* fill DTable */
+    for (n=0; n<nbSymbols; n++) {
+        const U32 w = huffWeight[n];
+        const U32 length = (1 << w) >> 1;
+        U32 i;
+        HUFv06_DEltX2 D;
+        D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+        for (i = rankVal[w]; i < rankVal[w] + length; i++)
+            dt[i] = D;
+        rankVal[w] += length;
+    }
+
+    return iSize;
+}
+
+
+static BYTE HUFv06_decodeSymbolX2(BITv06_DStream_t* Dstream, const HUFv06_DEltX2* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    const BYTE c = dt[val].byte;
+    BITv06_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv06_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv06_MAX_TABLELOG<=12)) \
+        HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv06_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv06_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv06_decodeStreamX2(BYTE* p, BITv06_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv06_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv06_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p < pEnd))
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv06_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+size_t HUFv06_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const U32 dtLog = DTable[0];
+    const void* dtPtr = DTable;
+    const HUFv06_DEltX2* const dt = ((const HUFv06_DEltX2*)dtPtr)+1;
+    BITv06_DStream_t bitD;
+
+    { size_t const errorCode = BITv06_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv06_isError(errorCode)) return errorCode; }
+
+    HUFv06_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv06_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv06_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX2(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const errorCode = HUFv06_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv06_decompress1X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+size_t HUFv06_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U16* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv06_DEltX2* const dt = ((const HUFv06_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv06_DStream_t bitD1;
+        BITv06_DStream_t bitD2;
+        BITv06_DStream_t bitD3;
+        BITv06_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv06_initDStream(&bitD1, istart1, length1);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD2, istart2, length2);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD3, istart3, length3);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD4, istart4, length4);
+        if (HUFv06_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv06_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv06_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv06_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv06_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv06_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv06_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv06_endOfDStream(&bitD1) & BITv06_endOfDStream(&bitD2) & BITv06_endOfDStream(&bitD3) & BITv06_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv06_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX2(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const errorCode = HUFv06_readDTableX2 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(errorCode)) return errorCode;
+    if (errorCode >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += errorCode;
+    cSrcSize -= errorCode;
+
+    return HUFv06_decompress4X2_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+
+static void HUFv06_fillDTableX4Level2(HUFv06_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv06_DEltX4 DElt;
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    { U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }}
+}
+
+typedef U32 rankVal_t[HUFv06_ABSOLUTEMAX_TABLELOG][HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+
+static void HUFv06_fillDTableX4(HUFv06_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv06_ABSOLUTEMAX_TABLELOG + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv06_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUFv06_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 u;
+                const U32 end = start + length;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv06_readDTableX4 (U32* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv06_MAX_SYMBOL_VALUE + 1];
+    sortedSymbol_t sortedSymbol[HUFv06_MAX_SYMBOL_VALUE + 1];
+    U32 rankStats[HUFv06_ABSOLUTEMAX_TABLELOG + 1] = { 0 };
+    U32 rankStart0[HUFv06_ABSOLUTEMAX_TABLELOG + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    const U32 memLog = DTable[0];
+    size_t iSize;
+    void* dtPtr = DTable;
+    HUFv06_DEltX4* const dt = ((HUFv06_DEltX4*)dtPtr) + 1;
+
+    HUFv06_STATIC_ASSERT(sizeof(HUFv06_DEltX4) == sizeof(U32));   /* if compilation fails here, assertion is false */
+    if (memLog > HUFv06_ABSOLUTEMAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv06_readStats(weightList, HUFv06_MAX_SYMBOL_VALUE + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv06_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > memLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (memLog-tableLog) - 1;   /* tableLog <= memLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < memLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUFv06_fillDTableX4(dt, memLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    return iSize;
+}
+
+
+static U32 HUFv06_decodeSymbolX4(void* op, BITv06_DStream_t* DStream, const HUFv06_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv06_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv06_decodeLastSymbolX4(void* op, BITv06_DStream_t* DStream, const HUFv06_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv06_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv06_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv06_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv06_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv06_MAX_TABLELOG<=12)) \
+        ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv06_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv06_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv06_decodeStreamX4(BYTE* p, BITv06_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv06_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv06_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv06_reloadDStream(bitDPtr) == BITv06_DStream_unfinished) && (p <= pEnd-2))
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv06_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv06_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+size_t HUFv06_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    const BYTE* const istart = (const BYTE*) cSrc;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+
+    const U32 dtLog = DTable[0];
+    const void* const dtPtr = DTable;
+    const HUFv06_DEltX4* const dt = ((const HUFv06_DEltX4*)dtPtr) +1;
+
+    /* Init */
+    BITv06_DStream_t bitD;
+    { size_t const errorCode = BITv06_initDStream(&bitD, istart, cSrcSize);
+      if (HUFv06_isError(errorCode)) return errorCode; }
+
+    /* decode */
+    HUFv06_decodeStreamX4(ostart, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv06_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv06_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX4(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv06_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv06_decompress1X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+size_t HUFv06_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const U32* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv06_DEltX4* const dt = ((const HUFv06_DEltX4*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+
+        /* Init */
+        BITv06_DStream_t bitD1;
+        BITv06_DStream_t bitD2;
+        BITv06_DStream_t bitD3;
+        BITv06_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv06_initDStream(&bitD1, istart1, length1);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD2, istart2, length2);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD3, istart3, length3);
+        if (HUFv06_isError(errorCode)) return errorCode;
+        errorCode = BITv06_initDStream(&bitD4, istart4, length4);
+        if (HUFv06_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv06_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv06_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv06_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv06_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv06_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv06_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv06_reloadDStream(&bitD1) | BITv06_reloadDStream(&bitD2) | BITv06_reloadDStream(&bitD3) | BITv06_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv06_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv06_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv06_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv06_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv06_endOfDStream(&bitD1) & BITv06_endOfDStream(&bitD2) & BITv06_endOfDStream(&bitD3) & BITv06_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv06_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv06_CREATE_STATIC_DTABLEX4(DTable, HUFv06_MAX_TABLELOG);
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv06_readDTableX4 (DTable, cSrc, cSrcSize);
+    if (HUFv06_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize;
+    cSrcSize -= hSize;
+
+    return HUFv06_decompress4X4_usingDTable (dst, dstSize, ip, cSrcSize, DTable);
+}
+
+
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv06_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[3] = { HUFv06_decompress4X2, HUFv06_decompress4X4, NULL };
+    U32 Dtime[3];   /* decompression time estimation */
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    /* decoder timing evaluation */
+    {   U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 n; for (n=0; n<3; n++)
+            Dtime[n] = algoTime[Q][n].tableTime + (algoTime[Q][n].decode256Time * D256);
+    }
+
+    Dtime[1] += Dtime[1] >> 4; Dtime[2] += Dtime[2] >> 3; /* advantage to algorithms using less memory, for cache eviction */
+
+    {   U32 algoNb = 0;
+        if (Dtime[1] < Dtime[0]) algoNb = 1;
+        // if (Dtime[2] < Dtime[algoNb]) algoNb = 2;   /* current speed of HUFv06_decompress4X6 is not good */
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+
+    //return HUFv06_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUFv06_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+    //return HUFv06_decompress4X6(dst, dstSize, cSrc, cSrcSize);   /* multi-streams quad-symbols decoding */
+}
+/*
+    Common functions of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net/
+*/
+
+
+/*-****************************************
+*  Version
+******************************************/
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTDv06_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv06_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTDv06_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZSTDv06_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  ZBUFF Error Management
+****************************************************************/
+unsigned ZBUFFv06_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZBUFFv06_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv06_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv06_HEAPMODE
+#  define ZSTDv06_HEAPMODE 1
+#endif
+
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#endif
+
+
+/*-*************************************
+*  Macros
+***************************************/
+#define ZSTDv06_isError ERR_isError   /* for inlining */
+#define FSEv06_isError  ERR_isError
+#define HUFv06_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv06_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock } ZSTDv06_dStage;
+
+struct ZSTDv06_DCtx_s
+{
+    FSEv06_DTable LLTable[FSEv06_DTABLE_SIZE_U32(LLFSELog)];
+    FSEv06_DTable OffTable[FSEv06_DTABLE_SIZE_U32(OffFSELog)];
+    FSEv06_DTable MLTable[FSEv06_DTABLE_SIZE_U32(MLFSELog)];
+    unsigned   hufTableX4[HUFv06_DTABLE_SIZE(HufLog)];
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    size_t headerSize;
+    ZSTDv06_frameParams fParams;
+    blockType_t bType;   /* used in ZSTDv06_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv06_dStage stage;
+    U32 flagRepeatTable;
+    const BYTE* litPtr;
+    size_t litSize;
+    BYTE litBuffer[ZSTDv06_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTDv06_DCtx within "zstd_static.h" */
+
+size_t ZSTDv06_sizeofDCtx (void) { return sizeof(ZSTDv06_DCtx); }   /* non published interface */
+
+size_t ZSTDv06_decompressBegin(ZSTDv06_DCtx* dctx)
+{
+    dctx->expected = ZSTDv06_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTableX4[0] = HufLog;
+    dctx->flagRepeatTable = 0;
+    return 0;
+}
+
+ZSTDv06_DCtx* ZSTDv06_createDCtx(void)
+{
+    ZSTDv06_DCtx* dctx = (ZSTDv06_DCtx*)malloc(sizeof(ZSTDv06_DCtx));
+    if (dctx==NULL) return NULL;
+    ZSTDv06_decompressBegin(dctx);
+    return dctx;
+}
+
+size_t ZSTDv06_freeDCtx(ZSTDv06_DCtx* dctx)
+{
+    free(dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv06_copyDCtx(ZSTDv06_DCtx* dstDCtx, const ZSTDv06_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv06_DCtx) - (ZSTDv06_BLOCKSIZE_MAX+WILDCOPY_OVERLENGTH + ZSTDv06_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv06_MAGICNUMBER (defined within zstd_static.h)
+      - 1 byte  - Frame Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+
+/* Frame descriptor
+
+   1 byte, using :
+   bit 0-3 : windowLog - ZSTDv06_WINDOWLOG_ABSOLUTEMIN   (see zstd_internal.h)
+   bit 4   : minmatch 4(0) or 3(1)
+   bit 5   : reserved (must be zero)
+   bit 6-7 : Frame content size : unknown, 1 byte, 2 bytes, 8 bytes
+
+   Optional : content size (0, 1, 2 or 8 bytes)
+   0 : unknown
+   1 : 0-255 bytes
+   2 : 256 - 65535+256
+   8 : up to 16 exa
+*/
+
+
+/* Compressed Block, format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+/** ZSTDv06_frameHeaderSize() :
+*   srcSize must be >= ZSTDv06_frameHeaderSize_min.
+*   @return : size of the Frame Header */
+static size_t ZSTDv06_frameHeaderSize(const void* src, size_t srcSize)
+{
+    if (srcSize < ZSTDv06_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    { U32 const fcsId = (((const BYTE*)src)[4]) >> 6;
+      return ZSTDv06_frameHeaderSize_min + ZSTDv06_fcs_fieldSize[fcsId]; }
+}
+
+
+/** ZSTDv06_getFrameParams() :
+*   decode Frame Header, or provide expected `srcSize`.
+*   @return : 0, `fparamsPtr` is correctly filled,
+*            >0, `srcSize` is too small, result is expected `srcSize`,
+*             or an error code, which can be tested using ZSTDv06_isError() */
+size_t ZSTDv06_getFrameParams(ZSTDv06_frameParams* fparamsPtr, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize < ZSTDv06_frameHeaderSize_min) return ZSTDv06_frameHeaderSize_min;
+    if (MEM_readLE32(src) != ZSTDv06_MAGICNUMBER) return ERROR(prefix_unknown);
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    { size_t const fhsize = ZSTDv06_frameHeaderSize(src, srcSize);
+      if (srcSize < fhsize) return fhsize; }
+
+    memset(fparamsPtr, 0, sizeof(*fparamsPtr));
+    {   BYTE const frameDesc = ip[4];
+        fparamsPtr->windowLog = (frameDesc & 0xF) + ZSTDv06_WINDOWLOG_ABSOLUTEMIN;
+        if ((frameDesc & 0x20) != 0) return ERROR(frameParameter_unsupported);   /* reserved 1 bit */
+        switch(frameDesc >> 6)  /* fcsId */
+        {
+            default:   /* impossible */
+            case 0 : fparamsPtr->frameContentSize = 0; break;
+            case 1 : fparamsPtr->frameContentSize = ip[5]; break;
+            case 2 : fparamsPtr->frameContentSize = MEM_readLE16(ip+5)+256; break;
+            case 3 : fparamsPtr->frameContentSize = MEM_readLE64(ip+5); break;
+    }   }
+    return 0;
+}
+
+
+/** ZSTDv06_decodeFrameHeader() :
+*   `srcSize` must be the size provided by ZSTDv06_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTDv06_isError() */
+static size_t ZSTDv06_decodeFrameHeader(ZSTDv06_DCtx* zc, const void* src, size_t srcSize)
+{
+    size_t const result = ZSTDv06_getFrameParams(&(zc->fParams), src, srcSize);
+    if ((MEM_32bits()) && (zc->fParams.windowLog > 25)) return ERROR(frameParameter_unsupported);
+    return result;
+}
+
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTDv06_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+size_t ZSTDv06_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    U32 cSize;
+
+    if (srcSize < ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    bpPtr->blockType = (blockType_t)((*in) >> 6);
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv06_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv06_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+size_t ZSTDv06_decodeLiteralsBlock(ZSTDv06_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    /* any compressed block with literals segment must be at least this size */
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch(istart[0]>> 6)
+    {
+    case IS_HUF:
+        {   size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for lhSize, + cSize (+nbSeq) */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > ZSTDv06_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv06_isError(singleStream ?
+                            HUFv06_decompress1X2(dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv06_decompress   (dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_PCH:
+        {   size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (!dctx->flagRepeatTable)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            {   size_t const errorCode = HUFv06_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTableX4);
+                if (HUFv06_isError(errorCode)) return ERROR(corruption_detected);
+            }
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case IS_RAW:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case IS_RLE:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > ZSTDv06_BLOCKSIZE_MAX) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+/*! ZSTDv06_buildSeqTable() :
+    @return : nb bytes read from src,
+              or an error code if it fails, testable with ZSTDv06_isError()
+*/
+size_t ZSTDv06_buildSeqTable(FSEv06_DTable* DTable, U32 type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case FSEv06_ENCODING_RLE :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        FSEv06_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        return 1;
+    case FSEv06_ENCODING_RAW :
+        FSEv06_buildDTable(DTable, defaultNorm, max, defaultLog);
+        return 0;
+    case FSEv06_ENCODING_STATIC:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    default :   /* impossible */
+    case FSEv06_ENCODING_DYNAMIC :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSEv06_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSEv06_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            FSEv06_buildDTable(DTable, norm, max, tableLog);
+            return headerSize;
+    }   }
+}
+
+
+size_t ZSTDv06_decodeSeqHeaders(int* nbSeqPtr,
+                             FSEv06_DTable* DTableLL, FSEv06_DTable* DTableML, FSEv06_DTable* DTableOffb, U32 flagRepeatTable,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    {   U32 const LLtype  = *ip >> 6;
+        U32 const Offtype = (*ip >> 4) & 3;
+        U32 const MLtype  = (*ip >> 2) & 3;
+        ip++;
+
+        /* check */
+        if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+        /* Build DTables */
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+        }
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableOffb, Offtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+        }
+        {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+            if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
+            ip += bhSize;
+    }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv06_DStream_t DStream;
+    FSEv06_DState_t stateLL;
+    FSEv06_DState_t stateOffb;
+    FSEv06_DState_t stateML;
+    size_t prevOffset[ZSTDv06_REP_INIT];
+} seqState_t;
+
+
+
+static void ZSTDv06_decodeSequence(seq_t* seq, seqState_t* seqState)
+{
+    /* Literal length */
+    U32 const llCode = FSEv06_peekSymbol(&(seqState->stateLL));
+    U32 const mlCode = FSEv06_peekSymbol(&(seqState->stateML));
+    U32 const ofCode = FSEv06_peekSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10,   11,    12,    13,    14,    15,
+                            16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26,   27,    28,    29,    30,    31,
+                            32, 34, 36, 38, 40, 44, 48, 56, 64, 80, 96, 0x80, 0x100, 0x200, 0x400, 0x800,
+                            0x1000, 0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       3,       7,     0xF,     0x1F,     0x3F,     0x7F,
+                 0xFF,   0x1FF,   0x3FF,   0x7FF,   0xFFF,   0x1FFF,   0x3FFF,   0x7FFF,
+                 0xFFFF, 0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,
+                 0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF, /*fake*/ 1, 1 };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BITv06_readBits(&(seqState->DStream), ofBits);   /* <=  26 bits */
+            if (MEM_32bits()) BITv06_reloadDStream(&(seqState->DStream));
+        }
+
+        if (offset < ZSTDv06_REP_NUM) {
+            if (llCode == 0 && offset <= 1) offset = 1-offset;
+
+            if (offset != 0) {
+                size_t temp = seqState->prevOffset[offset];
+                if (offset != 1) {
+                    seqState->prevOffset[2] = seqState->prevOffset[1];
+                }
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            offset -= ZSTDv06_REP_MOVE;
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq->offset = offset;
+    }
+
+    seq->matchLength = ML_base[mlCode] + MINMATCH + ((mlCode>31) ? BITv06_readBits(&(seqState->DStream), mlBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BITv06_reloadDStream(&(seqState->DStream));
+
+    seq->litLength = LL_base[llCode] + ((llCode>15) ? BITv06_readBits(&(seqState->DStream), llBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BITv06_reloadDStream(&(seqState->DStream));
+
+    /* ANS state update */
+    FSEv06_updateState(&(seqState->stateLL), &(seqState->DStream));   /* <=  9 bits */
+    FSEv06_updateState(&(seqState->stateML), &(seqState->DStream));   /* <=  9 bits */
+    if (MEM_32bits()) BITv06_reloadDStream(&(seqState->DStream));     /* <= 18 bits */
+    FSEv06_updateState(&(seqState->stateOffb), &(seqState->DStream)); /* <=  8 bits */
+}
+
+
+size_t ZSTDv06_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_8 = oend-8;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if (oLitEnd > oend_8) return ERROR(dstSize_tooSmall);   /* last match must start at a minimum distance of 8 from oend */
+    if (oMatchEnd > oend) return ERROR(dstSize_tooSmall);   /* overwrite beyond dst buffer */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv06_wildcopy(op, *litPtr, sequence.litLength);   /* note : oLitEnd <= oend-8 : no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_8 || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_8 */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv06_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv06_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_8) {
+            ZSTDv06_wildcopy(op, match, oend_8 - op);
+            match += oend_8 - op;
+            op = oend_8;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTDv06_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv06_decompressSequences(
+                               ZSTDv06_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    FSEv06_DTable* DTableLL = dctx->LLTable;
+    FSEv06_DTable* DTableML = dctx->MLTable;
+    FSEv06_DTable* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTDv06_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->flagRepeatTable, ip, seqSize);
+        if (ZSTDv06_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        dctx->flagRepeatTable = 0;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seq_t sequence;
+        seqState_t seqState;
+
+        memset(&sequence, 0, sizeof(sequence));
+        sequence.offset = REPCODE_STARTVALUE;
+        { U32 i; for (i=0; i<ZSTDv06_REP_INIT; i++) seqState.prevOffset[i] = REPCODE_STARTVALUE; }
+        { size_t const errorCode = BITv06_initDStream(&(seqState.DStream), ip, iend-ip);
+          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
+        FSEv06_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv06_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv06_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv06_reloadDStream(&(seqState.DStream)) <= BITv06_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            ZSTDv06_decodeSequence(&sequence, &seqState);
+
+#if 0  /* debug */
+            static BYTE* start = NULL;
+            if (start==NULL) start = op;
+            size_t pos = (size_t)(op-start);
+            if ((pos >= 5810037) && (pos < 5810400))
+                printf("Dpos %6u :%5u literals & match %3u bytes at distance %6u \n",
+                       pos, (U32)sequence.litLength, (U32)sequence.matchLength, (U32)sequence.offset);
+#endif
+
+            {   size_t const oneSeqSize = ZSTDv06_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                if (ZSTDv06_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (op+lastLLSize > oend) return ERROR(dstSize_tooSmall);
+        memcpy(op, litPtr, lastLLSize);
+        op += lastLLSize;
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv06_checkContinuity(ZSTDv06_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv06_decompressBlock_internal(ZSTDv06_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize >= ZSTDv06_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    {   size_t const litCSize = ZSTDv06_decodeLiteralsBlock(dctx, src, srcSize);
+        if (ZSTDv06_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+    return ZSTDv06_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv06_decompressBlock(ZSTDv06_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+/*! ZSTDv06_decompressFrame() :
+*   `dctx` must be properly initialized */
+static size_t ZSTDv06_decompressFrame(ZSTDv06_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstCapacity;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties = { bt_compressed, 0 };
+
+    /* check */
+    if (srcSize < ZSTDv06_frameHeaderSize_min+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+        if (ZSTDv06_decodeFrameHeader(dctx, src, frameHeaderSize)) return ERROR(corruption_detected);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize=0;
+        size_t const cBlockSize = ZSTDv06_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv06_blockHeaderSize;
+        remainingSize -= ZSTDv06_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv06_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv06_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            return ERROR(GENERIC);   /* not yet supported */
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        if (ZSTDv06_isError(decodedSize)) return decodedSize;
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+size_t ZSTDv06_decompress_usingPreparedDCtx(ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* refDCtx,
+                                         void* dst, size_t dstCapacity,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv06_copyDCtx(dctx, refDCtx);
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv06_decompress_usingDict(ZSTDv06_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv06_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv06_checkContinuity(dctx, dst);
+    return ZSTDv06_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv06_decompressDCtx(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTDv06_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTDv06_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv06_HEAPMODE) && (ZSTDv06_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv06_DCtx* dctx = ZSTDv06_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv06_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTDv06_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTDv06_DCtx dctx;
+    return ZSTDv06_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+size_t ZSTDv06_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+    blockProperties_t blockProperties = { bt_compressed, 0 };
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(frameHeaderSize)) return frameHeaderSize;
+        if (MEM_readLE32(src) != ZSTDv06_MAGICNUMBER) return ERROR(prefix_unknown);
+        if (srcSize < frameHeaderSize+ZSTDv06_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t const cBlockSize = ZSTDv06_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv06_blockHeaderSize;
+        remainingSize -= ZSTDv06_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        if (cBlockSize == 0) break;   /* bt_end */
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+/*_******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv06_nextSrcSizeToDecompress(ZSTDv06_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+size_t ZSTDv06_decompressContinue(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (dstCapacity) ZSTDv06_checkContinuity(dctx, dst);
+
+    /* Decompress : frame header; part 1 */
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        if (srcSize != ZSTDv06_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        dctx->headerSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+        if (ZSTDv06_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv06_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv06_frameHeaderSize_min) {
+            dctx->expected = dctx->headerSize - ZSTDv06_frameHeaderSize_min;
+            dctx->stage = ZSTDds_decodeFrameHeader;
+            return 0;
+        }
+        dctx->expected = 0;   /* not necessary to copy more */
+	/* fall-through */
+    case ZSTDds_decodeFrameHeader:
+        {   size_t result;
+            memcpy(dctx->headerBuffer + ZSTDv06_frameHeaderSize_min, src, dctx->expected);
+            result = ZSTDv06_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv06_isError(result)) return result;
+            dctx->expected = ZSTDv06_blockHeaderSize;
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTDv06_getcBlockSize(src, ZSTDv06_blockHeaderSize, &bp);
+            if (ZSTDv06_isError(cBlockSize)) return cBlockSize;
+            if (bp.blockType == bt_end) {
+                dctx->expected = 0;
+                dctx->stage = ZSTDds_getFrameHeaderSize;
+            } else {
+                dctx->expected = cBlockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv06_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv06_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            dctx->expected = ZSTDv06_blockHeaderSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            return rSize;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static void ZSTDv06_refDictContent(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+}
+
+static size_t ZSTDv06_loadEntropy(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t hSize, offcodeHeaderSize, matchlengthHeaderSize, litlengthHeaderSize;
+
+    hSize = HUFv06_readDTableX4(dctx->hufTableX4, dict, dictSize);
+    if (HUFv06_isError(hSize)) return ERROR(dictionary_corrupted);
+    dict = (const char*)dict + hSize;
+    dictSize -= hSize;
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        offcodeHeaderSize = FSEv06_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dict, dictSize);
+        if (FSEv06_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dict = (const char*)dict + offcodeHeaderSize;
+        dictSize -= offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        matchlengthHeaderSize = FSEv06_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dict, dictSize);
+        if (FSEv06_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dict = (const char*)dict + matchlengthHeaderSize;
+        dictSize -= matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        litlengthHeaderSize = FSEv06_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dict, dictSize);
+        if (FSEv06_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv06_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+          if (FSEv06_isError(errorCode)) return ERROR(dictionary_corrupted); }
+    }
+
+    dctx->flagRepeatTable = 1;
+    return hSize + offcodeHeaderSize + matchlengthHeaderSize + litlengthHeaderSize;
+}
+
+static size_t ZSTDv06_decompress_insertDictionary(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    size_t eSize;
+    U32 const magic = MEM_readLE32(dict);
+    if (magic != ZSTDv06_DICT_MAGIC) {
+        /* pure content mode */
+        ZSTDv06_refDictContent(dctx, dict, dictSize);
+        return 0;
+    }
+    /* load entropy tables */
+    dict = (const char*)dict + 4;
+    dictSize -= 4;
+    eSize = ZSTDv06_loadEntropy(dctx, dict, dictSize);
+    if (ZSTDv06_isError(eSize)) return ERROR(dictionary_corrupted);
+
+    /* reference dictionary content */
+    dict = (const char*)dict + eSize;
+    dictSize -= eSize;
+    ZSTDv06_refDictContent(dctx, dict, dictSize);
+
+    return 0;
+}
+
+
+size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    { size_t const errorCode = ZSTDv06_decompressBegin(dctx);
+      if (ZSTDv06_isError(errorCode)) return errorCode; }
+
+    if (dict && dictSize) {
+        size_t const errorCode = ZSTDv06_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv06_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net/
+*/
+
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv06_DCtx object is required to track streaming operations.
+*  Use ZBUFFv06_createDCtx() and ZBUFFv06_freeDCtx() to create/release resources.
+*  Use ZBUFFv06_decompressInit() to start a new decompression operation,
+*   or ZBUFFv06_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv06_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv06_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv06_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv06_recommendedDInSize() and ZBUFFv06_recommendedDOutSize()
+*  output : ZBUFFv06_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv06_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv06_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFFv06_dStage;
+
+/* *** Resource management *** */
+struct ZBUFFv06_DCtx_s {
+    ZSTDv06_DCtx* zd;
+    ZSTDv06_frameParams fParams;
+    ZBUFFv06_dStage stage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t blockSize;
+    BYTE headerBuffer[ZSTDv06_FRAMEHEADERSIZE_MAX];
+    size_t lhSize;
+};   /* typedef'd to ZBUFFv06_DCtx within "zstd_buffered.h" */
+
+
+ZBUFFv06_DCtx* ZBUFFv06_createDCtx(void)
+{
+    ZBUFFv06_DCtx* zbd = (ZBUFFv06_DCtx*)malloc(sizeof(ZBUFFv06_DCtx));
+    if (zbd==NULL) return NULL;
+    memset(zbd, 0, sizeof(*zbd));
+    zbd->zd = ZSTDv06_createDCtx();
+    zbd->stage = ZBUFFds_init;
+    return zbd;
+}
+
+size_t ZBUFFv06_freeDCtx(ZBUFFv06_DCtx* zbd)
+{
+    if (zbd==NULL) return 0;   /* support free on null */
+    ZSTDv06_freeDCtx(zbd->zd);
+    free(zbd->inBuff);
+    free(zbd->outBuff);
+    free(zbd);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv06_decompressInitDictionary(ZBUFFv06_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    zbd->stage = ZBUFFds_loadHeader;
+    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
+    return ZSTDv06_decompressBegin_usingDict(zbd->zd, dict, dictSize);
+}
+
+size_t ZBUFFv06_decompressInit(ZBUFFv06_DCtx* zbd)
+{
+    return ZBUFFv06_decompressInitDictionary(zbd, NULL, 0);
+}
+
+
+
+MEM_STATIC size_t ZBUFFv06_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t length = MIN(dstCapacity, srcSize);
+    memcpy(dst, src, length);
+    return length;
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbd->stage)
+        {
+        case ZBUFFds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFds_loadHeader :
+            {   size_t const hSize = ZSTDv06_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
+                if (hSize != 0) {
+                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
+                    if (ZSTDv06_isError(hSize)) return hSize;
+                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
+                        memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        zbd->lhSize += iend-ip; ip = iend; notDone = 0;
+                        *dstCapacityPtr = 0;
+                        return (hSize - zbd->lhSize) + ZSTDv06_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* Consume header */
+            {   size_t const h1Size = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTDv06_frameHeaderSize_min */
+                size_t const h1Result = ZSTDv06_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
+                if (ZSTDv06_isError(h1Result)) return h1Result;
+                if (h1Size < zbd->lhSize) {   /* long header */
+                    size_t const h2Size = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                    size_t const h2Result = ZSTDv06_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
+                    if (ZSTDv06_isError(h2Result)) return h2Result;
+            }   }
+
+            /* Frame header instruct buffer sizes */
+            {   size_t const blockSize = MIN(1 << zbd->fParams.windowLog, ZSTDv06_BLOCKSIZE_MAX);
+                zbd->blockSize = blockSize;
+                if (zbd->inBuffSize < blockSize) {
+                    free(zbd->inBuff);
+                    zbd->inBuffSize = blockSize;
+                    zbd->inBuff = (char*)malloc(blockSize);
+                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
+                }
+                {   size_t const neededOutSize = ((size_t)1 << zbd->fParams.windowLog) + blockSize + WILDCOPY_OVERLENGTH * 2;
+                    if (zbd->outBuffSize < neededOutSize) {
+                        free(zbd->outBuff);
+                        zbd->outBuffSize = neededOutSize;
+                        zbd->outBuff = (char*)malloc(neededOutSize);
+                        if (zbd->outBuff == NULL) return ERROR(memory_allocation);
+            }   }   }
+            zbd->stage = ZBUFFds_read;
+	    /* fall-through */
+        case ZBUFFds_read:
+            {   size_t const neededInSize = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                if (neededInSize==0) {  /* end of frame */
+                    zbd->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    size_t const decodedSize = ZSTDv06_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        ip, neededInSize);
+                    if (ZSTDv06_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize) break;   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbd->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {   size_t const neededInSize = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv06_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbd->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {   size_t const decodedSize = ZSTDv06_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        zbd->inBuff, neededInSize);
+                    if (ZSTDv06_isError(decodedSize)) return decodedSize;
+                    zbd->inPos = 0;   /* input is consumed */
+                    if (!decodedSize) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    // break; /* ZBUFFds_flush follows */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
+                size_t const flushedSize = ZBUFFv06_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
+                op += flushedSize;
+                zbd->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbd->stage = ZBUFFds_read;
+                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
+                        zbd->outStart = zbd->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    *srcSizePtr = ip-istart;
+    *dstCapacityPtr = op-ostart;
+    {   size_t nextSrcSizeHint = ZSTDv06_nextSrcSizeToDecompress(zbd->zd);
+        if (nextSrcSizeHint > ZSTDv06_blockHeaderSize) nextSrcSizeHint+= ZSTDv06_blockHeaderSize;   /* get following block header too */
+        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFFv06_recommendedDInSize(void)  { return ZSTDv06_BLOCKSIZE_MAX + ZSTDv06_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv06_recommendedDOutSize(void) { return ZSTDv06_BLOCKSIZE_MAX; }
diff --git a/vendor/github.com/DataDog/zstd/zstd_v06.h b/vendor/github.com/DataDog/zstd/zstd_v06.h
new file mode 100644
index 0000000..fb4eb37
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v06.h
@@ -0,0 +1,167 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTDv06_H
+#define ZSTDv06_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*======  Dependency  ======*/
+#include <stddef.h>   /* size_t */
+
+
+/*======  Export for Windows  ======*/
+/*!
+*  ZSTDv06_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*/
+#if defined(_WIN32) && defined(ZSTDv06_DLL_EXPORT) && (ZSTDv06_DLL_EXPORT==1)
+#  define ZSTDLIBv06_API __declspec(dllexport)
+#else
+#  define ZSTDLIBv06_API
+#endif
+
+
+/* *************************************
+*  Simple functions
+***************************************/
+/*! ZSTDv06_decompress() :
+    `compressedSize` : is the _exact_ size of the compressed blob, otherwise decompression will fail.
+    `dstCapacity` must be large enough, equal or larger than originalSize.
+    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+              or an errorCode if it fails (which can be tested using ZSTDv06_isError()) */
+ZSTDLIBv06_API size_t ZSTDv06_decompress( void* dst, size_t dstCapacity,
+                                    const void* src, size_t compressedSize);
+
+/**
+ZSTDv06_getFrameSrcSize() : get the source length of a ZSTD frame
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv06_isError())
+*/
+size_t ZSTDv06_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+/* *************************************
+*  Helper functions
+***************************************/
+ZSTDLIBv06_API size_t      ZSTDv06_compressBound(size_t srcSize); /*!< maximum compressed size (worst case scenario) */
+
+/* Error Management */
+ZSTDLIBv06_API unsigned    ZSTDv06_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+ZSTDLIBv06_API const char* ZSTDv06_getErrorName(size_t code);     /*!< provides readable string for an error code */
+
+
+/* *************************************
+*  Explicit memory management
+***************************************/
+/** Decompression context */
+typedef struct ZSTDv06_DCtx_s ZSTDv06_DCtx;
+ZSTDLIBv06_API ZSTDv06_DCtx* ZSTDv06_createDCtx(void);
+ZSTDLIBv06_API size_t     ZSTDv06_freeDCtx(ZSTDv06_DCtx* dctx);      /*!< @return : errorCode */
+
+/** ZSTDv06_decompressDCtx() :
+*   Same as ZSTDv06_decompress(), but requires an already allocated ZSTDv06_DCtx (see ZSTDv06_createDCtx()) */
+ZSTDLIBv06_API size_t ZSTDv06_decompressDCtx(ZSTDv06_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-***********************
+*  Dictionary API
+*************************/
+/*! ZSTDv06_decompress_usingDict() :
+*   Decompression using a pre-defined Dictionary content (see dictBuilder).
+*   Dictionary must be identical to the one used during compression, otherwise regenerated data will be corrupted.
+*   Note : dict can be NULL, in which case, it's equivalent to ZSTDv06_decompressDCtx() */
+ZSTDLIBv06_API size_t ZSTDv06_decompress_usingDict(ZSTDv06_DCtx* dctx,
+                                                   void* dst, size_t dstCapacity,
+                                             const void* src, size_t srcSize,
+                                             const void* dict,size_t dictSize);
+
+
+/*-************************
+*  Advanced Streaming API
+***************************/
+struct ZSTDv06_frameParams_s { unsigned long long frameContentSize; unsigned windowLog; };
+typedef struct ZSTDv06_frameParams_s ZSTDv06_frameParams;
+
+ZSTDLIBv06_API size_t ZSTDv06_getFrameParams(ZSTDv06_frameParams* fparamsPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+ZSTDLIBv06_API size_t ZSTDv06_decompressBegin_usingDict(ZSTDv06_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIBv06_API void   ZSTDv06_copyDCtx(ZSTDv06_DCtx* dctx, const ZSTDv06_DCtx* preparedDCtx);
+
+ZSTDLIBv06_API size_t ZSTDv06_nextSrcSizeToDecompress(ZSTDv06_DCtx* dctx);
+ZSTDLIBv06_API size_t ZSTDv06_decompressContinue(ZSTDv06_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+
+/* *************************************
+*  ZBUFF API
+***************************************/
+
+typedef struct ZBUFFv06_DCtx_s ZBUFFv06_DCtx;
+ZSTDLIBv06_API ZBUFFv06_DCtx* ZBUFFv06_createDCtx(void);
+ZSTDLIBv06_API size_t         ZBUFFv06_freeDCtx(ZBUFFv06_DCtx* dctx);
+
+ZSTDLIBv06_API size_t ZBUFFv06_decompressInit(ZBUFFv06_DCtx* dctx);
+ZSTDLIBv06_API size_t ZBUFFv06_decompressInitDictionary(ZBUFFv06_DCtx* dctx, const void* dict, size_t dictSize);
+
+ZSTDLIBv06_API size_t ZBUFFv06_decompressContinue(ZBUFFv06_DCtx* dctx,
+                                                  void* dst, size_t* dstCapacityPtr,
+                                            const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv06_DCtx object is required to track streaming operations.
+*  Use ZBUFFv06_createDCtx() and ZBUFFv06_freeDCtx() to create/release resources.
+*  Use ZBUFFv06_decompressInit() to start a new decompression operation,
+*   or ZBUFFv06_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv06_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv06_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv06_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv06_recommendedDInSize() and ZBUFFv06_recommendedDOutSize()
+*  output : ZBUFFv06_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv06_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv06_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+ZSTDLIBv06_API unsigned ZBUFFv06_isError(size_t errorCode);
+ZSTDLIBv06_API const char* ZBUFFv06_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, they tend to offer better latency */
+ZSTDLIBv06_API size_t ZBUFFv06_recommendedDInSize(void);
+ZSTDLIBv06_API size_t ZBUFFv06_recommendedDOutSize(void);
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ZSTDv06_MAGICNUMBER 0xFD2FB526   /* v0.6 */
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv06_BUFFERED_H */
diff --git a/vendor/github.com/DataDog/zstd/zstd_v07.c b/vendor/github.com/DataDog/zstd/zstd_v07.c
new file mode 100644
index 0000000..70b170f
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v07.c
@@ -0,0 +1,4502 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/*- Dependencies -*/
+#include <stddef.h>     /* size_t, ptrdiff_t */
+#include <string.h>     /* memcpy */
+#include <stdlib.h>     /* malloc, free, qsort */
+
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY    /* XXH64_state_t */
+#endif
+#include "xxhash.h"                  /* XXH64_* */
+#include "zstd_v07.h"
+
+#define FSEv07_STATIC_LINKING_ONLY   /* FSEv07_MIN_TABLELOG */
+#define HUFv07_STATIC_LINKING_ONLY   /* HUFv07_TABLELOG_ABSOLUTEMAX */
+#define ZSTDv07_STATIC_LINKING_ONLY
+
+#include "error_private.h"
+
+
+#ifdef ZSTDv07_STATIC_LINKING_ONLY
+
+/* ====================================================================================
+ * The definitions in this section are considered experimental.
+ * They should never be used with a dynamic library, as they may change in the future.
+ * They are provided for advanced usages.
+ * Use them only in association with static linking.
+ * ==================================================================================== */
+
+/*--- Constants ---*/
+#define ZSTDv07_MAGIC_SKIPPABLE_START  0x184D2A50U
+
+#define ZSTDv07_WINDOWLOG_MAX_32  25
+#define ZSTDv07_WINDOWLOG_MAX_64  27
+#define ZSTDv07_WINDOWLOG_MAX    ((U32)(MEM_32bits() ? ZSTDv07_WINDOWLOG_MAX_32 : ZSTDv07_WINDOWLOG_MAX_64))
+#define ZSTDv07_WINDOWLOG_MIN     18
+#define ZSTDv07_CHAINLOG_MAX     (ZSTDv07_WINDOWLOG_MAX+1)
+#define ZSTDv07_CHAINLOG_MIN       4
+#define ZSTDv07_HASHLOG_MAX       ZSTDv07_WINDOWLOG_MAX
+#define ZSTDv07_HASHLOG_MIN       12
+#define ZSTDv07_HASHLOG3_MAX      17
+#define ZSTDv07_SEARCHLOG_MAX    (ZSTDv07_WINDOWLOG_MAX-1)
+#define ZSTDv07_SEARCHLOG_MIN      1
+#define ZSTDv07_SEARCHLENGTH_MAX   7
+#define ZSTDv07_SEARCHLENGTH_MIN   3
+#define ZSTDv07_TARGETLENGTH_MIN   4
+#define ZSTDv07_TARGETLENGTH_MAX 999
+
+#define ZSTDv07_FRAMEHEADERSIZE_MAX 18    /* for static allocation */
+static const size_t ZSTDv07_frameHeaderSize_min = 5;
+static const size_t ZSTDv07_frameHeaderSize_max = ZSTDv07_FRAMEHEADERSIZE_MAX;
+static const size_t ZSTDv07_skippableHeaderSize = 8;  /* magic number + skippable frame length */
+
+
+/* custom memory allocation functions */
+typedef void* (*ZSTDv07_allocFunction) (void* opaque, size_t size);
+typedef void  (*ZSTDv07_freeFunction) (void* opaque, void* address);
+typedef struct { ZSTDv07_allocFunction customAlloc; ZSTDv07_freeFunction customFree; void* opaque; } ZSTDv07_customMem;
+
+
+/*--- Advanced Decompression functions ---*/
+
+/*! ZSTDv07_estimateDCtxSize() :
+ *  Gives the potential amount of memory allocated to create a ZSTDv07_DCtx */
+ZSTDLIBv07_API size_t ZSTDv07_estimateDCtxSize(void);
+
+/*! ZSTDv07_createDCtx_advanced() :
+ *  Create a ZSTD decompression context using external alloc and free functions */
+ZSTDLIBv07_API ZSTDv07_DCtx* ZSTDv07_createDCtx_advanced(ZSTDv07_customMem customMem);
+
+/*! ZSTDv07_sizeofDCtx() :
+ *  Gives the amount of memory used by a given ZSTDv07_DCtx */
+ZSTDLIBv07_API size_t ZSTDv07_sizeofDCtx(const ZSTDv07_DCtx* dctx);
+
+
+/* ******************************************************************
+*  Buffer-less streaming functions (synchronous mode)
+********************************************************************/
+
+ZSTDLIBv07_API size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZSTDv07_decompressBegin_usingDict(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIBv07_API void   ZSTDv07_copyDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* preparedDCtx);
+
+ZSTDLIBv07_API size_t ZSTDv07_nextSrcSizeToDecompress(ZSTDv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZSTDv07_decompressContinue(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+/*
+  Buffer-less streaming decompression (synchronous mode)
+
+  A ZSTDv07_DCtx object is required to track streaming operations.
+  Use ZSTDv07_createDCtx() / ZSTDv07_freeDCtx() to manage it.
+  A ZSTDv07_DCtx object can be re-used multiple times.
+
+  First optional operation is to retrieve frame parameters, using ZSTDv07_getFrameParams(), which doesn't consume the input.
+  It can provide the minimum size of rolling buffer required to properly decompress data (`windowSize`),
+  and optionally the final size of uncompressed content.
+  (Note : content size is an optional info that may not be present. 0 means : content size unknown)
+  Frame parameters are extracted from the beginning of compressed frame.
+  The amount of data to read is variable, from ZSTDv07_frameHeaderSize_min to ZSTDv07_frameHeaderSize_max (so if `srcSize` >= ZSTDv07_frameHeaderSize_max, it will always work)
+  If `srcSize` is too small for operation to succeed, function will return the minimum size it requires to produce a result.
+  Result : 0 when successful, it means the ZSTDv07_frameParams structure has been filled.
+          >0 : means there is not enough data into `src`. Provides the expected size to successfully decode header.
+           errorCode, which can be tested using ZSTDv07_isError()
+
+  Start decompression, with ZSTDv07_decompressBegin() or ZSTDv07_decompressBegin_usingDict().
+  Alternatively, you can copy a prepared context, using ZSTDv07_copyDCtx().
+
+  Then use ZSTDv07_nextSrcSizeToDecompress() and ZSTDv07_decompressContinue() alternatively.
+  ZSTDv07_nextSrcSizeToDecompress() tells how much bytes to provide as 'srcSize' to ZSTDv07_decompressContinue().
+  ZSTDv07_decompressContinue() requires this exact amount of bytes, or it will fail.
+
+  @result of ZSTDv07_decompressContinue() is the number of bytes regenerated within 'dst' (necessarily <= dstCapacity).
+  It can be zero, which is not an error; it just means ZSTDv07_decompressContinue() has decoded some header.
+
+  ZSTDv07_decompressContinue() needs previous data blocks during decompression, up to `windowSize`.
+  They should preferably be located contiguously, prior to current block.
+  Alternatively, a round buffer of sufficient size is also possible. Sufficient size is determined by frame parameters.
+  ZSTDv07_decompressContinue() is very sensitive to contiguity,
+  if 2 blocks don't follow each other, make sure that either the compressor breaks contiguity at the same place,
+    or that previous contiguous segment is large enough to properly handle maximum back-reference.
+
+  A frame is fully decoded when ZSTDv07_nextSrcSizeToDecompress() returns zero.
+  Context can then be reset to start a new decompression.
+
+
+  == Special case : skippable frames ==
+
+  Skippable frames allow the integration of user-defined data into a flow of concatenated frames.
+  Skippable frames will be ignored (skipped) by a decompressor. The format of skippable frame is following:
+  a) Skippable frame ID - 4 Bytes, Little endian format, any value from 0x184D2A50 to 0x184D2A5F
+  b) Frame Size - 4 Bytes, Little endian format, unsigned 32-bits
+  c) Frame Content - any content (User Data) of length equal to Frame Size
+  For skippable frames ZSTDv07_decompressContinue() always returns 0.
+  For skippable frames ZSTDv07_getFrameParams() returns fparamsPtr->windowLog==0 what means that a frame is skippable.
+  It also returns Frame Size as fparamsPtr->frameContentSize.
+*/
+
+
+/* **************************************
+*  Block functions
+****************************************/
+/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+    Frame metadata cost is typically ~18 bytes, which can be non-negligible for very small blocks (< 100 bytes).
+    User will have to take in charge required information to regenerate data, such as compressed and content sizes.
+
+    A few rules to respect :
+    - Compressing and decompressing require a context structure
+      + Use ZSTDv07_createCCtx() and ZSTDv07_createDCtx()
+    - It is necessary to init context before starting
+      + compression : ZSTDv07_compressBegin()
+      + decompression : ZSTDv07_decompressBegin()
+      + variants _usingDict() are also allowed
+      + copyCCtx() and copyDCtx() work too
+    - Block size is limited, it must be <= ZSTDv07_getBlockSizeMax()
+      + If you need to compress more, cut data into multiple blocks
+      + Consider using the regular ZSTDv07_compress() instead, as frame metadata costs become negligible when source size is large.
+    - When a block is considered not compressible enough, ZSTDv07_compressBlock() result will be zero.
+      In which case, nothing is produced into `dst`.
+      + User must test for such outcome and deal directly with uncompressed data
+      + ZSTDv07_decompressBlock() doesn't accept uncompressed data as input !!!
+      + In case of multiple successive blocks, decoder must be informed of uncompressed block existence to follow proper history.
+        Use ZSTDv07_insertBlock() in such a case.
+*/
+
+#define ZSTDv07_BLOCKSIZE_ABSOLUTEMAX (128 * 1024)   /* define, for static allocation */
+ZSTDLIBv07_API size_t ZSTDv07_decompressBlock(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert block into `dctx` history. Useful for uncompressed blocks */
+
+
+#endif   /* ZSTDv07_STATIC_LINKING_ONLY */
+
+
+/* ******************************************************************
+   mem.h
+   low-level memory access routines
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+#ifndef MEM_H_MODULE
+#define MEM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-****************************************
+*  Compiler specifics
+******************************************/
+#if defined(_MSC_VER)   /* Visual Studio */
+#   include <stdlib.h>  /* _byteswap_ulong */
+#   include <intrin.h>  /* _byteswap_* */
+#endif
+#if defined(__GNUC__)
+#  define MEM_STATIC static __attribute__((unused))
+#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#  define MEM_STATIC static inline
+#elif defined(_MSC_VER)
+#  define MEM_STATIC static __inline
+#else
+#  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
+#endif
+
+
+/*-**************************************************************
+*  Basic Types
+*****************************************************************/
+#if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include <stdint.h>
+  typedef  uint8_t BYTE;
+  typedef uint16_t U16;
+  typedef  int16_t S16;
+  typedef uint32_t U32;
+  typedef  int32_t S32;
+  typedef uint64_t U64;
+  typedef  int64_t S64;
+#else
+  typedef unsigned char       BYTE;
+  typedef unsigned short      U16;
+  typedef   signed short      S16;
+  typedef unsigned int        U32;
+  typedef   signed int        S32;
+  typedef unsigned long long  U64;
+  typedef   signed long long  S64;
+#endif
+
+
+/*-**************************************************************
+*  Memory I/O
+*****************************************************************/
+/* MEM_FORCE_MEMORY_ACCESS :
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
+ * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
+ * The below switch allow to select different access method for improved performance.
+ * Method 0 (default) : use `memcpy()`. Safe and portable.
+ * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
+ * Method 2 : direct access. This method is portable but violate C standard.
+ *            It can generate buggy code on targets depending on alignment.
+ *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
+ * Prefer these methods in priority order (0 > 1 > 2)
+ */
+#ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+#  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
+#    define MEM_FORCE_MEMORY_ACCESS 2
+#  elif (defined(__INTEL_COMPILER) && !defined(WIN32)) || \
+  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#    define MEM_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
+MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
+
+MEM_STATIC unsigned MEM_isLittleEndian(void)
+{
+    const union { U32 u; BYTE c[4]; } one = { 1 };   /* don't use static : performance detrimental  */
+    return one.c[0];
+}
+
+#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
+
+/* violates C standard, by lying on structure alignment.
+Only use if no other choice to achieve best performance on target platform */
+MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
+MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
+MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
+
+#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
+
+/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
+/* currently only defined for gcc and icc */
+typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
+
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
+
+#else
+
+/* default method, safe and standard.
+   can sometimes prove slower */
+
+MEM_STATIC U16 MEM_read16(const void* memPtr)
+{
+    U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U32 MEM_read32(const void* memPtr)
+{
+    U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC U64 MEM_read64(const void* memPtr)
+{
+    U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
+}
+
+MEM_STATIC void MEM_write16(void* memPtr, U16 value)
+{
+    memcpy(memPtr, &value, sizeof(value));
+}
+
+#endif /* MEM_FORCE_MEMORY_ACCESS */
+
+MEM_STATIC U32 MEM_swap32(U32 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_ulong(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap32(in);
+#else
+    return  ((in << 24) & 0xff000000 ) |
+            ((in <<  8) & 0x00ff0000 ) |
+            ((in >>  8) & 0x0000ff00 ) |
+            ((in >> 24) & 0x000000ff );
+#endif
+}
+
+MEM_STATIC U64 MEM_swap64(U64 in)
+{
+#if defined(_MSC_VER)     /* Visual Studio */
+    return _byteswap_uint64(in);
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+    return __builtin_bswap64(in);
+#else
+    return  ((in << 56) & 0xff00000000000000ULL) |
+            ((in << 40) & 0x00ff000000000000ULL) |
+            ((in << 24) & 0x0000ff0000000000ULL) |
+            ((in << 8)  & 0x000000ff00000000ULL) |
+            ((in >> 8)  & 0x00000000ff000000ULL) |
+            ((in >> 24) & 0x0000000000ff0000ULL) |
+            ((in >> 40) & 0x000000000000ff00ULL) |
+            ((in >> 56) & 0x00000000000000ffULL);
+#endif
+}
+
+
+/*=== Little endian r/w ===*/
+
+MEM_STATIC U16 MEM_readLE16(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read16(memPtr);
+    else {
+        const BYTE* p = (const BYTE*)memPtr;
+        return (U16)(p[0] + (p[1]<<8));
+    }
+}
+
+MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
+{
+    if (MEM_isLittleEndian()) {
+        MEM_write16(memPtr, val);
+    } else {
+        BYTE* p = (BYTE*)memPtr;
+        p[0] = (BYTE)val;
+        p[1] = (BYTE)(val>>8);
+    }
+}
+
+MEM_STATIC U32 MEM_readLE32(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read32(memPtr);
+    else
+        return MEM_swap32(MEM_read32(memPtr));
+}
+
+
+MEM_STATIC U64 MEM_readLE64(const void* memPtr)
+{
+    if (MEM_isLittleEndian())
+        return MEM_read64(memPtr);
+    else
+        return MEM_swap64(MEM_read64(memPtr));
+}
+
+MEM_STATIC size_t MEM_readLEST(const void* memPtr)
+{
+    if (MEM_32bits())
+        return (size_t)MEM_readLE32(memPtr);
+    else
+        return (size_t)MEM_readLE64(memPtr);
+}
+
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* MEM_H_MODULE */
+/* ******************************************************************
+   bitstream
+   Part of FSE library
+   header file (to include)
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef BITSTREAM_H_MODULE
+#define BITSTREAM_H_MODULE
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*
+*  This API consists of small unitary functions, which must be inlined for best performance.
+*  Since link-time-optimization is not available for all compilers,
+*  these functions are defined into a .h to be included.
+*/
+
+
+/*=========================================
+*  Target specific
+=========================================*/
+#if defined(__BMI__) && defined(__GNUC__)
+#  include <immintrin.h>   /* support for bextr (experimental) */
+#endif
+
+/*-********************************************
+*  bitStream decoding API (read backward)
+**********************************************/
+typedef struct
+{
+    size_t   bitContainer;
+    unsigned bitsConsumed;
+    const char* ptr;
+    const char* start;
+} BITv07_DStream_t;
+
+typedef enum { BITv07_DStream_unfinished = 0,
+               BITv07_DStream_endOfBuffer = 1,
+               BITv07_DStream_completed = 2,
+               BITv07_DStream_overflow = 3 } BITv07_DStream_status;  /* result of BITv07_reloadDStream() */
+               /* 1,2,4,8 would be better for bitmap combinations, but slows down performance a bit ... :( */
+
+MEM_STATIC size_t   BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuffer, size_t srcSize);
+MEM_STATIC size_t   BITv07_readBits(BITv07_DStream_t* bitD, unsigned nbBits);
+MEM_STATIC BITv07_DStream_status BITv07_reloadDStream(BITv07_DStream_t* bitD);
+MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* bitD);
+
+
+
+/*-****************************************
+*  unsafe API
+******************************************/
+MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, unsigned nbBits);
+/* faster, but works only if nbBits >= 1 */
+
+
+
+/*-**************************************************************
+*  Internal functions
+****************************************************************/
+MEM_STATIC unsigned BITv07_highbit32 (U32 val)
+{
+#   if defined(_MSC_VER)   /* Visual */
+    unsigned long r=0;
+    _BitScanReverse ( &r, val );
+    return (unsigned) r;
+#   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
+    return 31 - __builtin_clz (val);
+#   else   /* Software version */
+    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    U32 v = val;
+    v |= v >> 1;
+    v |= v >> 2;
+    v |= v >> 4;
+    v |= v >> 8;
+    v |= v >> 16;
+    return DeBruijnClz[ (U32) (v * 0x07C4ACDDU) >> 27];
+#   endif
+}
+
+
+
+/*-********************************************************
+* bitStream decoding
+**********************************************************/
+/*! BITv07_initDStream() :
+*   Initialize a BITv07_DStream_t.
+*   `bitD` : a pointer to an already allocated BITv07_DStream_t structure.
+*   `srcSize` must be the *exact* size of the bitStream, in bytes.
+*   @return : size of stream (== srcSize) or an errorCode if a problem is detected
+*/
+MEM_STATIC size_t BITv07_initDStream(BITv07_DStream_t* bitD, const void* srcBuffer, size_t srcSize)
+{
+    if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
+
+    if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BITv07_highbit32(lastByte) : 0;
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+    } else {
+        bitD->start = (const char*)srcBuffer;
+        bitD->ptr   = bitD->start;
+        bitD->bitContainer = *(const BYTE*)(bitD->start);
+        switch(srcSize)
+        {
+            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);/* fall-through */
+            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);/* fall-through */
+            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);/* fall-through */
+            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24; /* fall-through */
+            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16; /* fall-through */
+            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8; /* fall-through */
+            default: break;
+        }
+        { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
+          bitD->bitsConsumed = lastByte ? 8 - BITv07_highbit32(lastByte) : 0;
+          if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
+        bitD->bitsConsumed += (U32)(sizeof(bitD->bitContainer) - srcSize)*8;
+    }
+
+    return srcSize;
+}
+
+
+ MEM_STATIC size_t BITv07_lookBits(const BITv07_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+}
+
+/*! BITv07_lookBitsFast() :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BITv07_lookBitsFast(const BITv07_DStream_t* bitD, U32 nbBits)
+{
+    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
+    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+}
+
+MEM_STATIC void BITv07_skipBits(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    bitD->bitsConsumed += nbBits;
+}
+
+MEM_STATIC size_t BITv07_readBits(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv07_lookBits(bitD, nbBits);
+    BITv07_skipBits(bitD, nbBits);
+    return value;
+}
+
+/*! BITv07_readBitsFast() :
+*   unsafe version; only works only if nbBits >= 1 */
+MEM_STATIC size_t BITv07_readBitsFast(BITv07_DStream_t* bitD, U32 nbBits)
+{
+    size_t const value = BITv07_lookBitsFast(bitD, nbBits);
+    BITv07_skipBits(bitD, nbBits);
+    return value;
+}
+
+MEM_STATIC BITv07_DStream_status BITv07_reloadDStream(BITv07_DStream_t* bitD)
+{
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should not happen => corruption detected */
+        return BITv07_DStream_overflow;
+
+    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+        bitD->ptr -= bitD->bitsConsumed >> 3;
+        bitD->bitsConsumed &= 7;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);
+        return BITv07_DStream_unfinished;
+    }
+    if (bitD->ptr == bitD->start) {
+        if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BITv07_DStream_endOfBuffer;
+        return BITv07_DStream_completed;
+    }
+    {   U32 nbBytes = bitD->bitsConsumed >> 3;
+        BITv07_DStream_status result = BITv07_DStream_unfinished;
+        if (bitD->ptr - nbBytes < bitD->start) {
+            nbBytes = (U32)(bitD->ptr - bitD->start);  /* ptr > start */
+            result = BITv07_DStream_endOfBuffer;
+        }
+        bitD->ptr -= nbBytes;
+        bitD->bitsConsumed -= nbBytes*8;
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        return result;
+    }
+}
+
+/*! BITv07_endOfDStream() :
+*   @return Tells if DStream has exactly reached its end (all bits consumed).
+*/
+MEM_STATIC unsigned BITv07_endOfDStream(const BITv07_DStream_t* DStream)
+{
+    return ((DStream->ptr == DStream->start) && (DStream->bitsConsumed == sizeof(DStream->bitContainer)*8));
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* BITSTREAM_H_MODULE */
+/* ******************************************************************
+   FSE : Finite State Entropy codec
+   Public Prototypes declaration
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef FSEv07_H
+#define FSEv07_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/*-****************************************
+*  FSE simple functions
+******************************************/
+
+/*! FSEv07_decompress():
+    Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated destination buffer 'dst', of size 'dstCapacity'.
+    @return : size of regenerated data (<= maxDstSize),
+              or an error code, which can be tested using FSEv07_isError() .
+
+    ** Important ** : FSEv07_decompress() does not decompress non-compressible nor RLE data !!!
+    Why ? : making this distinction requires a header.
+    Header management is intentionally delegated to the user layer, which can better manage special cases.
+*/
+size_t FSEv07_decompress(void* dst,  size_t dstCapacity,
+                const void* cSrc, size_t cSrcSize);
+
+
+/* Error Management */
+unsigned    FSEv07_isError(size_t code);        /* tells if a return value is an error code */
+const char* FSEv07_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+
+
+/*-*****************************************
+*  FSE detailed API
+******************************************/
+/*!
+FSEv07_decompress() does the following:
+1. read normalized counters with readNCount()
+2. build decoding table 'DTable' from normalized counters
+3. decode the data stream using decoding table 'DTable'
+
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and provide normalized distribution using external method.
+*/
+
+
+/* *** DECOMPRESSION *** */
+
+/*! FSEv07_readNCount():
+    Read compactly saved 'normalizedCounter' from 'rBuffer'.
+    @return : size read from 'rBuffer',
+              or an errorCode, which can be tested using FSEv07_isError().
+              maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
+size_t FSEv07_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+
+/*! Constructor and Destructor of FSEv07_DTable.
+    Note that its size depends on 'tableLog' */
+typedef unsigned FSEv07_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
+FSEv07_DTable* FSEv07_createDTable(unsigned tableLog);
+void        FSEv07_freeDTable(FSEv07_DTable* dt);
+
+/*! FSEv07_buildDTable():
+    Builds 'dt', which must be already allocated, using FSEv07_createDTable().
+    return : 0, or an errorCode, which can be tested using FSEv07_isError() */
+size_t FSEv07_buildDTable (FSEv07_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+
+/*! FSEv07_decompress_usingDTable():
+    Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
+    into `dst` which must be already allocated.
+    @return : size of regenerated data (necessarily <= `dstCapacity`),
+              or an errorCode, which can be tested using FSEv07_isError() */
+size_t FSEv07_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSEv07_DTable* dt);
+
+/*!
+Tutorial :
+----------
+(Note : these functions only decompress FSE-compressed blocks.
+ If block is uncompressed, use memcpy() instead
+ If block is a single repeated byte, use memset() instead )
+
+The first step is to obtain the normalized frequencies of symbols.
+This can be performed by FSEv07_readNCount() if it was saved using FSEv07_writeNCount().
+'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.
+In practice, that means it's necessary to know 'maxSymbolValue' beforehand,
+or size the table to handle worst case situations (typically 256).
+FSEv07_readNCount() will provide 'tableLog' and 'maxSymbolValue'.
+The result of FSEv07_readNCount() is the number of bytes read from 'rBuffer'.
+Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError().
+
+The next step is to build the decompression tables 'FSEv07_DTable' from 'normalizedCounter'.
+This is performed by the function FSEv07_buildDTable().
+The space required by 'FSEv07_DTable' must be already allocated using FSEv07_createDTable().
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError().
+
+`FSEv07_DTable` can then be used to decompress `cSrc`, with FSEv07_decompress_usingDTable().
+`cSrcSize` must be strictly correct, otherwise decompression will fail.
+FSEv07_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).
+If there is an error, the function will return an error code, which can be tested using FSEv07_isError(). (ex: dst buffer too small)
+*/
+
+
+#ifdef FSEv07_STATIC_LINKING_ONLY
+
+
+/* *****************************************
+*  Static allocation
+*******************************************/
+/* FSE buffer bounds */
+#define FSEv07_NCOUNTBOUND 512
+#define FSEv07_BLOCKBOUND(size) (size + (size>>7))
+
+/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+#define FSEv07_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
+
+
+/* *****************************************
+*  FSE advanced API
+*******************************************/
+size_t FSEv07_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+/**< same as FSEv07_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+unsigned FSEv07_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
+/**< same as FSEv07_optimalTableLog(), which used `minus==2` */
+
+size_t FSEv07_buildDTable_raw (FSEv07_DTable* dt, unsigned nbBits);
+/**< build a fake FSEv07_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+
+size_t FSEv07_buildDTable_rle (FSEv07_DTable* dt, unsigned char symbolValue);
+/**< build a fake FSEv07_DTable, designed to always generate the same symbolValue */
+
+
+
+/* *****************************************
+*  FSE symbol decompression API
+*******************************************/
+typedef struct
+{
+    size_t      state;
+    const void* table;   /* precise table may vary, depending on U16 */
+} FSEv07_DState_t;
+
+
+static void     FSEv07_initDState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD, const FSEv07_DTable* dt);
+
+static unsigned char FSEv07_decodeSymbol(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD);
+
+
+
+/* *****************************************
+*  FSE unsafe API
+*******************************************/
+static unsigned char FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD);
+/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */
+
+
+/* ======    Decompression    ====== */
+
+typedef struct {
+    U16 tableLog;
+    U16 fastMode;
+} FSEv07_DTableHeader;   /* sizeof U32 */
+
+typedef struct
+{
+    unsigned short newState;
+    unsigned char  symbol;
+    unsigned char  nbBits;
+} FSEv07_decode_t;   /* size == U32 */
+
+MEM_STATIC void FSEv07_initDState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD, const FSEv07_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv07_DTableHeader* const DTableH = (const FSEv07_DTableHeader*)ptr;
+    DStatePtr->state = BITv07_readBits(bitD, DTableH->tableLog);
+    BITv07_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+MEM_STATIC BYTE FSEv07_peekSymbol(const FSEv07_DState_t* DStatePtr)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    return DInfo.symbol;
+}
+
+MEM_STATIC void FSEv07_updateState(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BITv07_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.newState + lowBits;
+}
+
+MEM_STATIC BYTE FSEv07_decodeSymbol(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv07_readBits(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+/*! FSEv07_decodeSymbolFast() :
+    unsafe, only works if no symbol has a probability > 50% */
+MEM_STATIC BYTE FSEv07_decodeSymbolFast(FSEv07_DState_t* DStatePtr, BITv07_DStream_t* bitD)
+{
+    FSEv07_decode_t const DInfo = ((const FSEv07_decode_t*)(DStatePtr->table))[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    BYTE const symbol = DInfo.symbol;
+    size_t const lowBits = BITv07_readBitsFast(bitD, nbBits);
+
+    DStatePtr->state = DInfo.newState + lowBits;
+    return symbol;
+}
+
+
+
+#ifndef FSEv07_COMMONDEFS_ONLY
+
+/* **************************************************************
+*  Tuning parameters
+****************************************************************/
+/*!MEMORY_USAGE :
+*  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
+*  Increasing memory usage improves compression ratio
+*  Reduced memory usage can improve speed, due to cache effect
+*  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
+#define FSEv07_MAX_MEMORY_USAGE 14
+#define FSEv07_DEFAULT_MEMORY_USAGE 13
+
+/*!FSEv07_MAX_SYMBOL_VALUE :
+*  Maximum symbol value authorized.
+*  Required for proper stack allocation */
+#define FSEv07_MAX_SYMBOL_VALUE 255
+
+
+/* **************************************************************
+*  template functions type & suffix
+****************************************************************/
+#define FSEv07_FUNCTION_TYPE BYTE
+#define FSEv07_FUNCTION_EXTENSION
+#define FSEv07_DECODE_TYPE FSEv07_decode_t
+
+
+#endif   /* !FSEv07_COMMONDEFS_ONLY */
+
+
+/* ***************************************************************
+*  Constants
+*****************************************************************/
+#define FSEv07_MAX_TABLELOG  (FSEv07_MAX_MEMORY_USAGE-2)
+#define FSEv07_MAX_TABLESIZE (1U<<FSEv07_MAX_TABLELOG)
+#define FSEv07_MAXTABLESIZE_MASK (FSEv07_MAX_TABLESIZE-1)
+#define FSEv07_DEFAULT_TABLELOG (FSEv07_DEFAULT_MEMORY_USAGE-2)
+#define FSEv07_MIN_TABLELOG 5
+
+#define FSEv07_TABLELOG_ABSOLUTE_MAX 15
+#if FSEv07_MAX_TABLELOG > FSEv07_TABLELOG_ABSOLUTE_MAX
+#  error "FSEv07_MAX_TABLELOG > FSEv07_TABLELOG_ABSOLUTE_MAX is not supported"
+#endif
+
+#define FSEv07_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)
+
+
+#endif /* FSEv07_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* FSEv07_H */
+/* ******************************************************************
+   Huffman coder, part of New Generation Entropy library
+   header file
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+   You can contact the author at :
+   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+****************************************************************** */
+#ifndef HUFv07_H_298734234
+#define HUFv07_H_298734234
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+
+/* *** simple functions *** */
+/**
+HUFv07_decompress() :
+    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+    into already allocated buffer 'dst', of minimum size 'dstSize'.
+    `dstSize` : **must** be the ***exact*** size of original (uncompressed) data.
+    Note : in contrast with FSE, HUFv07_decompress can regenerate
+           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+           because it knows size to regenerate.
+    @return : size of regenerated data (== dstSize),
+              or an error code, which can be tested using HUFv07_isError()
+*/
+size_t HUFv07_decompress(void* dst,  size_t dstSize,
+                const void* cSrc, size_t cSrcSize);
+
+
+/* ****************************************
+*  Tool functions
+******************************************/
+#define HUFv07_BLOCKSIZE_MAX (128 * 1024)
+
+/* Error Management */
+unsigned    HUFv07_isError(size_t code);        /**< tells if a return value is an error code */
+const char* HUFv07_getErrorName(size_t code);   /**< provides error code string (useful for debugging) */
+
+
+/* *** Advanced function *** */
+
+
+#ifdef HUFv07_STATIC_LINKING_ONLY
+
+
+/* *** Constants *** */
+#define HUFv07_TABLELOG_ABSOLUTEMAX  16   /* absolute limit of HUFv07_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUFv07_TABLELOG_MAX  12           /* max configured tableLog (for static allocation); can be modified up to HUFv07_ABSOLUTEMAX_TABLELOG */
+#define HUFv07_TABLELOG_DEFAULT  11       /* tableLog by default, when not specified */
+#define HUFv07_SYMBOLVALUE_MAX 255
+#if (HUFv07_TABLELOG_MAX > HUFv07_TABLELOG_ABSOLUTEMAX)
+#  error "HUFv07_TABLELOG_MAX is too large !"
+#endif
+
+
+/* ****************************************
+*  Static allocation
+******************************************/
+/* HUF buffer bounds */
+#define HUFv07_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+
+/* static allocation of HUF's DTable */
+typedef U32 HUFv07_DTable;
+#define HUFv07_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
+#define HUFv07_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+        HUFv07_DTable DTable[HUFv07_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1)*0x1000001) }
+#define HUFv07_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+        HUFv07_DTable DTable[HUFv07_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog)*0x1000001) }
+
+
+/* ****************************************
+*  Advanced decompression functions
+******************************************/
+size_t HUFv07_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUFv07_decompress4X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
+size_t HUFv07_decompress4X_hufOnly(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
+size_t HUFv07_decompress4X2_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress4X4_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+size_t HUFv07_decompress1X2_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUFv07_decompress1X4_DCtx(HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+
+/* ****************************************
+*  HUF detailed API
+******************************************/
+/*!
+The following API allows targeting specific sub-functions for advanced tasks.
+For example, it's possible to compress several blocks using the same 'CTable',
+or to save and regenerate 'CTable' using external methods.
+*/
+/* FSEv07_count() : find it within "fse.h" */
+
+/*! HUFv07_readStats() :
+    Read compact Huffman tree, saved by HUFv07_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUFv07_readCTable() and HUFv07_readDTableXn() . */
+size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize);
+
+
+/*
+HUFv07_decompress() does the following:
+1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
+2. build Huffman table from save, using HUFv07_readDTableXn()
+3. decode 1 or 4 segments in parallel using HUFv07_decompressSXn_usingDTable
+*/
+
+/** HUFv07_selectDecoder() :
+*   Tells which decoder is likely to decode faster,
+*   based on a set of pre-determined metrics.
+*   @return : 0==HUFv07_decompress4X2, 1==HUFv07_decompress4X4 .
+*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+U32 HUFv07_selectDecoder (size_t dstSize, size_t cSrcSize);
+
+size_t HUFv07_readDTableX2 (HUFv07_DTable* DTable, const void* src, size_t srcSize);
+size_t HUFv07_readDTableX4 (HUFv07_DTable* DTable, const void* src, size_t srcSize);
+
+size_t HUFv07_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+
+
+/* single stream variants */
+size_t HUFv07_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+size_t HUFv07_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+
+size_t HUFv07_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+size_t HUFv07_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUFv07_DTable* DTable);
+
+
+#endif /* HUFv07_STATIC_LINKING_ONLY */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* HUFv07_H_298734234 */
+/*
+   Common functions of New Generation Entropy library
+   Copyright (C) 2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+*************************************************************************** */
+
+
+
+/*-****************************************
+*  FSE Error Management
+******************************************/
+unsigned FSEv07_isError(size_t code) { return ERR_isError(code); }
+
+const char* FSEv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/* **************************************************************
+*  HUF Error Management
+****************************************************************/
+unsigned HUFv07_isError(size_t code) { return ERR_isError(code); }
+
+const char* HUFv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+static short FSEv07_abs(short a) { return (short)(a<0 ? -a : a); }
+
+size_t FSEv07_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
+                 const void* headerBuffer, size_t hbSize)
+{
+    const BYTE* const istart = (const BYTE*) headerBuffer;
+    const BYTE* const iend = istart + hbSize;
+    const BYTE* ip = istart;
+    int nbBits;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    if (hbSize < 4) return ERROR(srcSize_wrong);
+    bitStream = MEM_readLE32(ip);
+    nbBits = (bitStream & 0xF) + FSEv07_MIN_TABLELOG;   /* extract tableLog */
+    if (nbBits > FSEv07_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
+    bitStream >>= 4;
+    bitCount = 4;
+    *tableLogPtr = nbBits;
+    remaining = (1<<nbBits)+1;
+    threshold = 1<<nbBits;
+    nbBits++;
+
+    while ((remaining>1) && (charnum<=*maxSVPtr)) {
+        if (previous0) {
+            unsigned n0 = charnum;
+            while ((bitStream & 0xFFFF) == 0xFFFF) {
+                n0+=24;
+                if (ip < iend-5) {
+                    ip+=2;
+                    bitStream = MEM_readLE32(ip) >> bitCount;
+                } else {
+                    bitStream >>= 16;
+                    bitCount+=16;
+            }   }
+            while ((bitStream & 3) == 3) {
+                n0+=3;
+                bitStream>>=2;
+                bitCount+=2;
+            }
+            n0 += bitStream & 3;
+            bitCount += 2;
+            if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
+            while (charnum < n0) normalizedCounter[charnum++] = 0;
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+                bitStream = MEM_readLE32(ip) >> bitCount;
+            }
+            else
+                bitStream >>= 2;
+        }
+        {   short const max = (short)((2*threshold-1)-remaining);
+            short count;
+
+            if ((bitStream & (threshold-1)) < (U32)max) {
+                count = (short)(bitStream & (threshold-1));
+                bitCount   += nbBits-1;
+            } else {
+                count = (short)(bitStream & (2*threshold-1));
+                if (count >= threshold) count -= max;
+                bitCount   += nbBits;
+            }
+
+            count--;   /* extra accuracy */
+            remaining -= FSEv07_abs(count);
+            normalizedCounter[charnum++] = count;
+            previous0 = !count;
+            while (remaining < threshold) {
+                nbBits--;
+                threshold >>= 1;
+            }
+
+            if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                ip += bitCount>>3;
+                bitCount &= 7;
+            } else {
+                bitCount -= (int)(8 * (iend - 4 - ip));
+                ip = iend - 4;
+            }
+            bitStream = MEM_readLE32(ip) >> (bitCount & 31);
+    }   }   /* while ((remaining>1) && (charnum<=*maxSVPtr)) */
+    if (remaining != 1) return ERROR(GENERIC);
+    *maxSVPtr = charnum-1;
+
+    ip += (bitCount+7)>>3;
+    if ((size_t)(ip-istart) > hbSize) return ERROR(srcSize_wrong);
+    return ip-istart;
+}
+
+
+/*! HUFv07_readStats() :
+    Read compact Huffman tree, saved by HUFv07_writeCTable().
+    `huffWeight` is destination buffer.
+    @return : size read from `src` , or an error Code .
+    Note : Needed by HUFv07_readCTable() and HUFv07_readDTableXn() .
+*/
+size_t HUFv07_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
+                     U32* nbSymbolsPtr, U32* tableLogPtr,
+                     const void* src, size_t srcSize)
+{
+    U32 weightTotal;
+    const BYTE* ip = (const BYTE*) src;
+    size_t iSize;
+    size_t oSize;
+
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
+    //memset(huffWeight, 0, hwSize);   /* is not necessary, even though some analyzer complain ... */
+
+    if (iSize >= 128)  { /* special header */
+        if (iSize >= (242)) {  /* RLE */
+            static U32 l[14] = { 1, 2, 3, 4, 7, 8, 15, 16, 31, 32, 63, 64, 127, 128 };
+            oSize = l[iSize-242];
+            memset(huffWeight, 1, hwSize);
+            iSize = 0;
+        }
+        else {   /* Incompressible */
+            oSize = iSize - 127;
+            iSize = ((oSize+1)/2);
+            if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+            if (oSize >= hwSize) return ERROR(corruption_detected);
+            ip += 1;
+            {   U32 n;
+                for (n=0; n<oSize; n+=2) {
+                    huffWeight[n]   = ip[n/2] >> 4;
+                    huffWeight[n+1] = ip[n/2] & 15;
+    }   }   }   }
+    else  {   /* header compressed with FSE (normal case) */
+        if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
+        oSize = FSEv07_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        if (FSEv07_isError(oSize)) return oSize;
+    }
+
+    /* collect weight stats */
+    memset(rankStats, 0, (HUFv07_TABLELOG_ABSOLUTEMAX + 1) * sizeof(U32));
+    weightTotal = 0;
+    {   U32 n; for (n=0; n<oSize; n++) {
+            if (huffWeight[n] >= HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+            rankStats[huffWeight[n]]++;
+            weightTotal += (1 << huffWeight[n]) >> 1;
+    }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
+
+    /* get last non-null symbol weight (implied, total must be 2^n) */
+    {   U32 const tableLog = BITv07_highbit32(weightTotal) + 1;
+        if (tableLog > HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+        *tableLogPtr = tableLog;
+        /* determine last weight */
+        {   U32 const total = 1 << tableLog;
+            U32 const rest = total - weightTotal;
+            U32 const verif = 1 << BITv07_highbit32(rest);
+            U32 const lastWeight = BITv07_highbit32(rest) + 1;
+            if (verif != rest) return ERROR(corruption_detected);    /* last value must be a clean power of 2 */
+            huffWeight[oSize] = (BYTE)lastWeight;
+            rankStats[lastWeight]++;
+    }   }
+
+    /* check tree construction validity */
+    if ((rankStats[1] < 2) || (rankStats[1] & 1)) return ERROR(corruption_detected);   /* by construction : at least 2 elts of rank 1, must be even */
+
+    /* results */
+    *nbSymbolsPtr = (U32)(oSize+1);
+    return iSize+1;
+}
+/* ******************************************************************
+   FSE : Finite State Entropy decoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSEv07_isError ERR_isError
+#define FSEv07_STATIC_ASSERT(c) { enum { FSEv07_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Complex types
+****************************************************************/
+typedef U32 DTable_max_t[FSEv07_DTABLE_SIZE_U32(FSEv07_MAX_TABLELOG)];
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSEv07_FUNCTION_EXTENSION
+#  error "FSEv07_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSEv07_FUNCTION_TYPE
+#  error "FSEv07_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSEv07_CAT(X,Y) X##Y
+#define FSEv07_FUNCTION_NAME(X,Y) FSEv07_CAT(X,Y)
+#define FSEv07_TYPE_NAME(X,Y) FSEv07_CAT(X,Y)
+
+
+/* Function templates */
+FSEv07_DTable* FSEv07_createDTable (unsigned tableLog)
+{
+    if (tableLog > FSEv07_TABLELOG_ABSOLUTE_MAX) tableLog = FSEv07_TABLELOG_ABSOLUTE_MAX;
+    return (FSEv07_DTable*)malloc( FSEv07_DTABLE_SIZE_U32(tableLog) * sizeof (U32) );
+}
+
+void FSEv07_freeDTable (FSEv07_DTable* dt)
+{
+    free(dt);
+}
+
+size_t FSEv07_buildDTable(FSEv07_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    void* const tdPtr = dt+1;   /* because *dt is unsigned, 32-bits aligned on 32-bits */
+    FSEv07_DECODE_TYPE* const tableDecode = (FSEv07_DECODE_TYPE*) (tdPtr);
+    U16 symbolNext[FSEv07_MAX_SYMBOL_VALUE+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    if (maxSymbolValue > FSEv07_MAX_SYMBOL_VALUE) return ERROR(maxSymbolValue_tooLarge);
+    if (tableLog > FSEv07_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+
+    /* Init, lay down lowprob symbols */
+    {   FSEv07_DTableHeader DTableH;
+        DTableH.tableLog = (U16)tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].symbol = (FSEv07_FUNCTION_TYPE)s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    symbolNext[s] = normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSEv07_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].symbol = (FSEv07_FUNCTION_TYPE)s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            FSEv07_FUNCTION_TYPE const symbol = (FSEv07_FUNCTION_TYPE)(tableDecode[u].symbol);
+            U16 nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BITv07_highbit32 ((U32)nextState) );
+            tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+    }   }
+
+    return 0;
+}
+
+
+
+#ifndef FSEv07_COMMONDEFS_ONLY
+
+/*-*******************************************************
+*  Decompression (Byte symbols)
+*********************************************************/
+size_t FSEv07_buildDTable_rle (FSEv07_DTable* dt, BYTE symbolValue)
+{
+    void* ptr = dt;
+    FSEv07_DTableHeader* const DTableH = (FSEv07_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv07_decode_t* const cell = (FSEv07_decode_t*)dPtr;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->newState = 0;
+    cell->symbol = symbolValue;
+    cell->nbBits = 0;
+
+    return 0;
+}
+
+
+size_t FSEv07_buildDTable_raw (FSEv07_DTable* dt, unsigned nbBits)
+{
+    void* ptr = dt;
+    FSEv07_DTableHeader* const DTableH = (FSEv07_DTableHeader*)ptr;
+    void* dPtr = dt + 1;
+    FSEv07_decode_t* const dinfo = (FSEv07_decode_t*)dPtr;
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSV1 = tableMask+1;
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);         /* min size */
+
+    /* Build Decoding Table */
+    DTableH->tableLog = (U16)nbBits;
+    DTableH->fastMode = 1;
+    for (s=0; s<maxSV1; s++) {
+        dinfo[s].newState = 0;
+        dinfo[s].symbol = (BYTE)s;
+        dinfo[s].nbBits = (BYTE)nbBits;
+    }
+
+    return 0;
+}
+
+FORCE_INLINE size_t FSEv07_decompress_usingDTable_generic(
+          void* dst, size_t maxDstSize,
+    const void* cSrc, size_t cSrcSize,
+    const FSEv07_DTable* dt, const unsigned fast)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const omax = op + maxDstSize;
+    BYTE* const olimit = omax-3;
+
+    BITv07_DStream_t bitD;
+    FSEv07_DState_t state1;
+    FSEv07_DState_t state2;
+
+    /* Init */
+    { size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);   /* replaced last arg by maxCompressed Size */
+      if (FSEv07_isError(errorCode)) return errorCode; }
+
+    FSEv07_initDState(&state1, &bitD, dt);
+    FSEv07_initDState(&state2, &bitD, dt);
+
+#define FSEv07_GETSYMBOL(statePtr) fast ? FSEv07_decodeSymbolFast(statePtr, &bitD) : FSEv07_decodeSymbol(statePtr, &bitD)
+
+    /* 4 symbols per loop */
+    for ( ; (BITv07_reloadDStream(&bitD)==BITv07_DStream_unfinished) && (op<olimit) ; op+=4) {
+        op[0] = FSEv07_GETSYMBOL(&state1);
+
+        if (FSEv07_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv07_reloadDStream(&bitD);
+
+        op[1] = FSEv07_GETSYMBOL(&state2);
+
+        if (FSEv07_MAX_TABLELOG*4+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            { if (BITv07_reloadDStream(&bitD) > BITv07_DStream_unfinished) { op+=2; break; } }
+
+        op[2] = FSEv07_GETSYMBOL(&state1);
+
+        if (FSEv07_MAX_TABLELOG*2+7 > sizeof(bitD.bitContainer)*8)    /* This test must be static */
+            BITv07_reloadDStream(&bitD);
+
+        op[3] = FSEv07_GETSYMBOL(&state2);
+    }
+
+    /* tail */
+    /* note : BITv07_reloadDStream(&bitD) >= FSEv07_DStream_partiallyFilled; Ends at exactly BITv07_DStream_completed */
+    while (1) {
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv07_GETSYMBOL(&state1);
+
+        if (BITv07_reloadDStream(&bitD)==BITv07_DStream_overflow) {
+            *op++ = FSEv07_GETSYMBOL(&state2);
+            break;
+        }
+
+        if (op>(omax-2)) return ERROR(dstSize_tooSmall);
+
+        *op++ = FSEv07_GETSYMBOL(&state2);
+
+        if (BITv07_reloadDStream(&bitD)==BITv07_DStream_overflow) {
+            *op++ = FSEv07_GETSYMBOL(&state1);
+            break;
+    }   }
+
+    return op-ostart;
+}
+
+
+size_t FSEv07_decompress_usingDTable(void* dst, size_t originalSize,
+                            const void* cSrc, size_t cSrcSize,
+                            const FSEv07_DTable* dt)
+{
+    const void* ptr = dt;
+    const FSEv07_DTableHeader* DTableH = (const FSEv07_DTableHeader*)ptr;
+    const U32 fastMode = DTableH->fastMode;
+
+    /* select fast mode (static) */
+    if (fastMode) return FSEv07_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 1);
+    return FSEv07_decompress_usingDTable_generic(dst, originalSize, cSrc, cSrcSize, dt, 0);
+}
+
+
+size_t FSEv07_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* const istart = (const BYTE*)cSrc;
+    const BYTE* ip = istart;
+    short counting[FSEv07_MAX_SYMBOL_VALUE+1];
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    unsigned tableLog;
+    unsigned maxSymbolValue = FSEv07_MAX_SYMBOL_VALUE;
+
+    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
+
+    /* normal FSE decoding mode */
+    {   size_t const NCountLength = FSEv07_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+        if (FSEv07_isError(NCountLength)) return NCountLength;
+        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
+        ip += NCountLength;
+        cSrcSize -= NCountLength;
+    }
+
+    { size_t const errorCode = FSEv07_buildDTable (dt, counting, maxSymbolValue, tableLog);
+      if (FSEv07_isError(errorCode)) return errorCode; }
+
+    return FSEv07_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+}
+
+
+
+#endif   /* FSEv07_COMMONDEFS_ONLY */
+
+/* ******************************************************************
+   Huffman decoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+/* inline is defined */
+#elif defined(_MSC_VER)
+#  define inline __inline
+#else
+#  define inline /* disable inline */
+#endif
+
+
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUFv07_STATIC_ASSERT(c) { enum { HUFv07_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/*-***************************/
+/*  generic DTableDesc       */
+/*-***************************/
+
+typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
+
+static DTableDesc HUFv07_getDTableDesc(const HUFv07_DTable* table)
+{
+    DTableDesc dtd;
+    memcpy(&dtd, table, sizeof(dtd));
+    return dtd;
+}
+
+
+/*-***************************/
+/*  single-symbol decoding   */
+/*-***************************/
+
+typedef struct { BYTE byte; BYTE nbBits; } HUFv07_DEltX2;   /* single-symbol decoding */
+
+size_t HUFv07_readDTableX2 (HUFv07_DTable* DTable, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUFv07_SYMBOLVALUE_MAX + 1];
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+    size_t iSize;
+    void* const dtPtr = DTable + 1;
+    HUFv07_DEltX2* const dt = (HUFv07_DEltX2*)dtPtr;
+
+    HUFv07_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUFv07_DTable));
+    //memset(huffWeight, 0, sizeof(huffWeight));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv07_readStats(huffWeight, HUFv07_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv07_isError(iSize)) return iSize;
+
+    /* Table header */
+    {   DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, huffman tree cannot fit in */
+        dtd.tableType = 0;
+        dtd.tableLog = (BYTE)tableLog;
+        memcpy(DTable, &dtd, sizeof(dtd));
+    }
+
+    /* Prepare ranks */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<tableLog+1; n++) {
+            U32 current = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = current;
+    }   }
+
+    /* fill DTable */
+    {   U32 n;
+        for (n=0; n<nbSymbols; n++) {
+            U32 const w = huffWeight[n];
+            U32 const length = (1 << w) >> 1;
+            U32 i;
+            HUFv07_DEltX2 D;
+            D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
+            for (i = rankVal[w]; i < rankVal[w] + length; i++)
+                dt[i] = D;
+            rankVal[w] += length;
+    }   }
+
+    return iSize;
+}
+
+
+static BYTE HUFv07_decodeSymbolX2(BITv07_DStream_t* Dstream, const HUFv07_DEltX2* dt, const U32 dtLog)
+{
+    size_t const val = BITv07_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
+    BYTE const c = dt[val].byte;
+    BITv07_skipBits(Dstream, dt[val].nbBits);
+    return c;
+}
+
+#define HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    *ptr++ = HUFv07_decodeSymbolX2(DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv07_TABLELOG_MAX<=12)) \
+        HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+#define HUFv07_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        HUFv07_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+
+static inline size_t HUFv07_decodeStreamX2(BYTE* p, BITv07_DStream_t* const bitDPtr, BYTE* const pEnd, const HUFv07_DEltX2* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 4 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p <= pEnd-4)) {
+        HUFv07_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+    }
+
+    /* closer to the end */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p < pEnd))
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    /* no more data to retrieve from bitstream, hence no need to reload */
+    while (p < pEnd)
+        HUFv07_DECODE_SYMBOLX2_0(p, bitDPtr);
+
+    return pEnd-pStart;
+}
+
+static size_t HUFv07_decompress1X2_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + dstSize;
+    const void* dtPtr = DTable + 1;
+    const HUFv07_DEltX2* const dt = (const HUFv07_DEltX2*)dtPtr;
+    BITv07_DStream_t bitD;
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    U32 const dtLog = dtd.tableLog;
+
+    { size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);
+      if (HUFv07_isError(errorCode)) return errorCode; }
+
+    HUFv07_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+
+    /* check */
+    if (!BITv07_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    return dstSize;
+}
+
+size_t HUFv07_decompress1X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUFv07_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress1X2_DCtx (HUFv07_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX2 (DCtx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress1X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+}
+
+size_t HUFv07_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX2(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+static size_t HUFv07_decompress4X2_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    /* Check */
+    if (cSrcSize < 10) return ERROR(corruption_detected);  /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable + 1;
+        const HUFv07_DEltX2* const dt = (const HUFv07_DEltX2*)dtPtr;
+
+        /* Init */
+        BITv07_DStream_t bitD1;
+        BITv07_DStream_t bitD2;
+        BITv07_DStream_t bitD3;
+        BITv07_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        { size_t const errorCode = BITv07_initDStream(&bitD1, istart1, length1);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD2, istart2, length2);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD3, istart3, length3);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD4, istart4, length4);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv07_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv07_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv07_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv07_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv07_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv07_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        endSignal = BITv07_endOfDStream(&bitD1) & BITv07_endOfDStream(&bitD2) & BITv07_endOfDStream(&bitD3) & BITv07_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv07_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUFv07_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+
+size_t HUFv07_decompress4X2_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX2 (dctx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress4X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, dctx);
+}
+
+size_t HUFv07_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX2(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* *************************/
+/* double-symbols decoding */
+/* *************************/
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUFv07_DEltX4;  /* double-symbols decoding */
+
+typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+
+static void HUFv07_fillDTableX4Level2(HUFv07_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+                           const U32* rankValOrigin, const int minWeight,
+                           const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
+                           U32 nbBitsBaseline, U16 baseSeq)
+{
+    HUFv07_DEltX4 DElt;
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+
+    /* get pre-calculated rankVal */
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill skipped values */
+    if (minWeight>1) {
+        U32 i, skipSize = rankVal[minWeight];
+        MEM_writeLE16(&(DElt.sequence), baseSeq);
+        DElt.nbBits   = (BYTE)(consumed);
+        DElt.length   = 1;
+        for (i = 0; i < skipSize; i++)
+            DTable[i] = DElt;
+    }
+
+    /* fill DTable */
+    { U32 s; for (s=0; s<sortedListSize; s++) {   /* note : sortedSymbols already skipped */
+        const U32 symbol = sortedSymbols[s].symbol;
+        const U32 weight = sortedSymbols[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 length = 1 << (sizeLog-nbBits);
+        const U32 start = rankVal[weight];
+        U32 i = start;
+        const U32 end = start + length;
+
+        MEM_writeLE16(&(DElt.sequence), (U16)(baseSeq + (symbol << 8)));
+        DElt.nbBits = (BYTE)(nbBits + consumed);
+        DElt.length = 2;
+        do { DTable[i++] = DElt; } while (i<end);   /* since length >= 1 */
+
+        rankVal[weight] += length;
+    }}
+}
+
+typedef U32 rankVal_t[HUFv07_TABLELOG_ABSOLUTEMAX][HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+
+static void HUFv07_fillDTableX4(HUFv07_DEltX4* DTable, const U32 targetLog,
+                           const sortedSymbol_t* sortedList, const U32 sortedListSize,
+                           const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
+                           const U32 nbBitsBaseline)
+{
+    U32 rankVal[HUFv07_TABLELOG_ABSOLUTEMAX + 1];
+    const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
+    const U32 minBits  = nbBitsBaseline - maxWeight;
+    U32 s;
+
+    memcpy(rankVal, rankValOrigin, sizeof(rankVal));
+
+    /* fill DTable */
+    for (s=0; s<sortedListSize; s++) {
+        const U16 symbol = sortedList[s].symbol;
+        const U32 weight = sortedList[s].weight;
+        const U32 nbBits = nbBitsBaseline - weight;
+        const U32 start = rankVal[weight];
+        const U32 length = 1 << (targetLog-nbBits);
+
+        if (targetLog-nbBits >= minBits) {   /* enough room for a second symbol */
+            U32 sortedRank;
+            int minWeight = nbBits + scaleLog;
+            if (minWeight < 1) minWeight = 1;
+            sortedRank = rankStart[minWeight];
+            HUFv07_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+                           rankValOrigin[nbBits], minWeight,
+                           sortedList+sortedRank, sortedListSize-sortedRank,
+                           nbBitsBaseline, symbol);
+        } else {
+            HUFv07_DEltX4 DElt;
+            MEM_writeLE16(&(DElt.sequence), symbol);
+            DElt.nbBits = (BYTE)(nbBits);
+            DElt.length = 1;
+            {   U32 u;
+                const U32 end = start + length;
+                for (u = start; u < end; u++) DTable[u] = DElt;
+        }   }
+        rankVal[weight] += length;
+    }
+}
+
+size_t HUFv07_readDTableX4 (HUFv07_DTable* DTable, const void* src, size_t srcSize)
+{
+    BYTE weightList[HUFv07_SYMBOLVALUE_MAX + 1];
+    sortedSymbol_t sortedSymbol[HUFv07_SYMBOLVALUE_MAX + 1];
+    U32 rankStats[HUFv07_TABLELOG_ABSOLUTEMAX + 1] = { 0 };
+    U32 rankStart0[HUFv07_TABLELOG_ABSOLUTEMAX + 2] = { 0 };
+    U32* const rankStart = rankStart0+1;
+    rankVal_t rankVal;
+    U32 tableLog, maxW, sizeOfSort, nbSymbols;
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    U32 const maxTableLog = dtd.maxTableLog;
+    size_t iSize;
+    void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
+    HUFv07_DEltX4* const dt = (HUFv07_DEltX4*)dtPtr;
+
+    HUFv07_STATIC_ASSERT(sizeof(HUFv07_DEltX4) == sizeof(HUFv07_DTable));   /* if compilation fails here, assertion is false */
+    if (maxTableLog > HUFv07_TABLELOG_ABSOLUTEMAX) return ERROR(tableLog_tooLarge);
+    //memset(weightList, 0, sizeof(weightList));   /* is not necessary, even though some analyzer complain ... */
+
+    iSize = HUFv07_readStats(weightList, HUFv07_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
+    if (HUFv07_isError(iSize)) return iSize;
+
+    /* check result */
+    if (tableLog > maxTableLog) return ERROR(tableLog_tooLarge);   /* DTable can't fit code depth */
+
+    /* find maxWeight */
+    for (maxW = tableLog; rankStats[maxW]==0; maxW--) {}  /* necessarily finds a solution before 0 */
+
+    /* Get start index of each weight */
+    {   U32 w, nextRankStart = 0;
+        for (w=1; w<maxW+1; w++) {
+            U32 current = nextRankStart;
+            nextRankStart += rankStats[w];
+            rankStart[w] = current;
+        }
+        rankStart[0] = nextRankStart;   /* put all 0w symbols at the end of sorted list*/
+        sizeOfSort = nextRankStart;
+    }
+
+    /* sort symbols by weight */
+    {   U32 s;
+        for (s=0; s<nbSymbols; s++) {
+            U32 const w = weightList[s];
+            U32 const r = rankStart[w]++;
+            sortedSymbol[r].symbol = (BYTE)s;
+            sortedSymbol[r].weight = (BYTE)w;
+        }
+        rankStart[0] = 0;   /* forget 0w symbols; this is beginning of weight(1) */
+    }
+
+    /* Build rankVal */
+    {   U32* const rankVal0 = rankVal[0];
+        {   int const rescale = (maxTableLog-tableLog) - 1;   /* tableLog <= maxTableLog */
+            U32 nextRankVal = 0;
+            U32 w;
+            for (w=1; w<maxW+1; w++) {
+                U32 current = nextRankVal;
+                nextRankVal += rankStats[w] << (w+rescale);
+                rankVal0[w] = current;
+        }   }
+        {   U32 const minBits = tableLog+1 - maxW;
+            U32 consumed;
+            for (consumed = minBits; consumed < maxTableLog - minBits + 1; consumed++) {
+                U32* const rankValPtr = rankVal[consumed];
+                U32 w;
+                for (w = 1; w < maxW+1; w++) {
+                    rankValPtr[w] = rankVal0[w] >> consumed;
+    }   }   }   }
+
+    HUFv07_fillDTableX4(dt, maxTableLog,
+                   sortedSymbol, sizeOfSort,
+                   rankStart0, rankVal, maxW,
+                   tableLog+1);
+
+    dtd.tableLog = (BYTE)maxTableLog;
+    dtd.tableType = 1;
+    memcpy(DTable, &dtd, sizeof(dtd));
+    return iSize;
+}
+
+
+static U32 HUFv07_decodeSymbolX4(void* op, BITv07_DStream_t* DStream, const HUFv07_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv07_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 2);
+    BITv07_skipBits(DStream, dt[val].nbBits);
+    return dt[val].length;
+}
+
+static U32 HUFv07_decodeLastSymbolX4(void* op, BITv07_DStream_t* DStream, const HUFv07_DEltX4* dt, const U32 dtLog)
+{
+    const size_t val = BITv07_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
+    memcpy(op, dt+val, 1);
+    if (dt[val].length==1) BITv07_skipBits(DStream, dt[val].nbBits);
+    else {
+        if (DStream->bitsConsumed < (sizeof(DStream->bitContainer)*8)) {
+            BITv07_skipBits(DStream, dt[val].nbBits);
+            if (DStream->bitsConsumed > (sizeof(DStream->bitContainer)*8))
+                DStream->bitsConsumed = (sizeof(DStream->bitContainer)*8);   /* ugly hack; works only because it's the last symbol. Note : can't easily extract nbBits from just this symbol */
+    }   }
+    return 1;
+}
+
+
+#define HUFv07_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
+    ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+    if (MEM_64bits() || (HUFv07_TABLELOG_MAX<=12)) \
+        ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+#define HUFv07_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+    if (MEM_64bits()) \
+        ptr += HUFv07_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+
+static inline size_t HUFv07_decodeStreamX4(BYTE* p, BITv07_DStream_t* bitDPtr, BYTE* const pEnd, const HUFv07_DEltX4* const dt, const U32 dtLog)
+{
+    BYTE* const pStart = p;
+
+    /* up to 8 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p < pEnd-7)) {
+        HUFv07_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_1(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_2(p, bitDPtr);
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);
+    }
+
+    /* closer to end : up to 2 symbols at a time */
+    while ((BITv07_reloadDStream(bitDPtr) == BITv07_DStream_unfinished) && (p <= pEnd-2))
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);
+
+    while (p <= pEnd-2)
+        HUFv07_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+
+    if (p < pEnd)
+        p += HUFv07_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+
+    return p-pStart;
+}
+
+
+static size_t HUFv07_decompress1X4_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    BITv07_DStream_t bitD;
+
+    /* Init */
+    {   size_t const errorCode = BITv07_initDStream(&bitD, cSrc, cSrcSize);
+        if (HUFv07_isError(errorCode)) return errorCode;
+    }
+
+    /* decode */
+    {   BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
+        const HUFv07_DEltX4* const dt = (const HUFv07_DEltX4*)dtPtr;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        HUFv07_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
+    }
+
+    /* check */
+    if (!BITv07_endOfDStream(&bitD)) return ERROR(corruption_detected);
+
+    /* decoded size */
+    return dstSize;
+}
+
+size_t HUFv07_decompress1X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUFv07_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress1X4_DCtx (HUFv07_DTable* DCtx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUFv07_readDTableX4 (DCtx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress1X4_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
+}
+
+size_t HUFv07_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX4(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+static size_t HUFv07_decompress4X4_usingDTable_internal(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+
+    {   const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable+1;
+        const HUFv07_DEltX4* const dt = (const HUFv07_DEltX4*)dtPtr;
+
+        /* Init */
+        BITv07_DStream_t bitD1;
+        BITv07_DStream_t bitD2;
+        BITv07_DStream_t bitD3;
+        BITv07_DStream_t bitD4;
+        size_t const length1 = MEM_readLE16(istart);
+        size_t const length2 = MEM_readLE16(istart+2);
+        size_t const length3 = MEM_readLE16(istart+4);
+        size_t const length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        size_t const segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
+        DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+        U32 const dtLog = dtd.tableLog;
+
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        { size_t const errorCode = BITv07_initDStream(&bitD1, istart1, length1);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD2, istart2, length2);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD3, istart3, length3);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+        { size_t const errorCode = BITv07_initDStream(&bitD4, istart4, length4);
+          if (HUFv07_isError(errorCode)) return errorCode; }
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
+        endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        for ( ; (endSignal==BITv07_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv07_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_1(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_1(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_1(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_1(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_2(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_2(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_2(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_2(op4, &bitD4);
+            HUFv07_DECODE_SYMBOLX4_0(op1, &bitD1);
+            HUFv07_DECODE_SYMBOLX4_0(op2, &bitD2);
+            HUFv07_DECODE_SYMBOLX4_0(op3, &bitD3);
+            HUFv07_DECODE_SYMBOLX4_0(op4, &bitD4);
+
+            endSignal = BITv07_reloadDStream(&bitD1) | BITv07_reloadDStream(&bitD2) | BITv07_reloadDStream(&bitD3) | BITv07_reloadDStream(&bitD4);
+        }
+
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
+
+        /* finish bitStreams one by one */
+        HUFv07_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv07_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv07_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv07_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+
+        /* check */
+        { U32 const endCheck = BITv07_endOfDStream(&bitD1) & BITv07_endOfDStream(&bitD2) & BITv07_endOfDStream(&bitD3) & BITv07_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
+
+        /* decoded size */
+        return dstSize;
+    }
+}
+
+
+size_t HUFv07_decompress4X4_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUFv07_DTable* DTable)
+{
+    DTableDesc dtd = HUFv07_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUFv07_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+}
+
+
+size_t HUFv07_decompress4X4_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUFv07_readDTableX4 (dctx, cSrc, cSrcSize);
+    if (HUFv07_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUFv07_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx);
+}
+
+size_t HUFv07_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUFv07_CREATE_STATIC_DTABLEX4(DTable, HUFv07_TABLELOG_MAX);
+    return HUFv07_decompress4X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+
+/* ********************************/
+/* Generic decompression selector */
+/* ********************************/
+
+size_t HUFv07_decompress1X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUFv07_DTable* DTable)
+{
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    return dtd.tableType ? HUFv07_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
+                           HUFv07_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+}
+
+size_t HUFv07_decompress4X_usingDTable(void* dst, size_t maxDstSize,
+                                    const void* cSrc, size_t cSrcSize,
+                                    const HUFv07_DTable* DTable)
+{
+    DTableDesc const dtd = HUFv07_getDTableDesc(DTable);
+    return dtd.tableType ? HUFv07_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
+                           HUFv07_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+}
+
+
+typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
+static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
+{
+    /* single, double, quad */
+    {{0,0}, {1,1}, {2,2}},  /* Q==0 : impossible */
+    {{0,0}, {1,1}, {2,2}},  /* Q==1 : impossible */
+    {{  38,130}, {1313, 74}, {2151, 38}},   /* Q == 2 : 12-18% */
+    {{ 448,128}, {1353, 74}, {2238, 41}},   /* Q == 3 : 18-25% */
+    {{ 556,128}, {1353, 74}, {2238, 47}},   /* Q == 4 : 25-32% */
+    {{ 714,128}, {1418, 74}, {2436, 53}},   /* Q == 5 : 32-38% */
+    {{ 883,128}, {1437, 74}, {2464, 61}},   /* Q == 6 : 38-44% */
+    {{ 897,128}, {1515, 75}, {2622, 68}},   /* Q == 7 : 44-50% */
+    {{ 926,128}, {1613, 75}, {2730, 75}},   /* Q == 8 : 50-56% */
+    {{ 947,128}, {1729, 77}, {3359, 77}},   /* Q == 9 : 56-62% */
+    {{1107,128}, {2083, 81}, {4006, 84}},   /* Q ==10 : 62-69% */
+    {{1177,128}, {2379, 87}, {4785, 88}},   /* Q ==11 : 69-75% */
+    {{1242,128}, {2415, 93}, {5155, 84}},   /* Q ==12 : 75-81% */
+    {{1349,128}, {2644,106}, {5260,106}},   /* Q ==13 : 81-87% */
+    {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
+    {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
+};
+
+/** HUFv07_selectDecoder() :
+*   Tells which decoder is likely to decode faster,
+*   based on a set of pre-determined metrics.
+*   @return : 0==HUFv07_decompress4X2, 1==HUFv07_decompress4X4 .
+*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+U32 HUFv07_selectDecoder (size_t dstSize, size_t cSrcSize)
+{
+    /* decoder timing evaluation */
+    U32 const Q = (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 since dstSize > cSrcSize */
+    U32 const D256 = (U32)(dstSize >> 8);
+    U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+    U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+    DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, for cache eviction */
+
+    return DTime1 < DTime0;
+}
+
+
+typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+
+size_t HUFv07_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    static const decompressionAlgo decompress[2] = { HUFv07_decompress4X2, HUFv07_decompress4X4 };
+
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+    }
+
+    //return HUFv07_decompress4X2(dst, dstSize, cSrc, cSrcSize);   /* multi-streams single-symbol decoding */
+    //return HUFv07_decompress4X4(dst, dstSize, cSrc, cSrcSize);   /* multi-streams double-symbols decoding */
+}
+
+size_t HUFv07_decompress4X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUFv07_decompress4X_hufOnly (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if ((cSrcSize >= dstSize) || (cSrcSize <= 1)) return ERROR(corruption_detected);   /* invalid */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+
+size_t HUFv07_decompress1X_DCtx (HUFv07_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize > dstSize) return ERROR(corruption_detected);   /* invalid */
+    if (cSrcSize == dstSize) { memcpy(dst, cSrc, dstSize); return dstSize; }   /* not compressed */
+    if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
+
+    {   U32 const algoNb = HUFv07_selectDecoder(dstSize, cSrcSize);
+        return algoNb ? HUFv07_decompress1X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUFv07_decompress1X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+    }
+}
+/*
+    Common functions of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net/
+*/
+
+
+
+/*-****************************************
+*  ZSTD Error Management
+******************************************/
+/*! ZSTDv07_isError() :
+*   tells if a return value is an error code */
+unsigned ZSTDv07_isError(size_t code) { return ERR_isError(code); }
+
+/*! ZSTDv07_getErrorName() :
+*   provides error code string from function result (useful for debugging) */
+const char* ZSTDv07_getErrorName(size_t code) { return ERR_getErrorName(code); }
+
+
+
+/* **************************************************************
+*  ZBUFF Error Management
+****************************************************************/
+unsigned ZBUFFv07_isError(size_t errorCode) { return ERR_isError(errorCode); }
+
+const char* ZBUFFv07_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
+
+
+
+void* ZSTDv07_defaultAllocFunction(void* opaque, size_t size)
+{
+    void* address = malloc(size);
+    (void)opaque;
+    /* printf("alloc %p, %d opaque=%p \n", address, (int)size, opaque); */
+    return address;
+}
+
+void ZSTDv07_defaultFreeFunction(void* opaque, void* address)
+{
+    (void)opaque;
+    /* if (address) printf("free %p opaque=%p \n", address, opaque); */
+    free(address);
+}
+/*
+    zstd_internal - common functions to include
+    Header File for include
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : https://www.zstd.net
+*/
+#ifndef ZSTDv07_CCOMMON_H_MODULE
+#define ZSTDv07_CCOMMON_H_MODULE
+
+
+/*-*************************************
+*  Common macros
+***************************************/
+#define MIN(a,b) ((a)<(b) ? (a) : (b))
+#define MAX(a,b) ((a)>(b) ? (a) : (b))
+
+
+/*-*************************************
+*  Common constants
+***************************************/
+#define ZSTDv07_OPT_NUM    (1<<12)
+#define ZSTDv07_DICT_MAGIC  0xEC30A437   /* v0.7 */
+
+#define ZSTDv07_REP_NUM    3
+#define ZSTDv07_REP_INIT   ZSTDv07_REP_NUM
+#define ZSTDv07_REP_MOVE   (ZSTDv07_REP_NUM-1)
+static const U32 repStartValue[ZSTDv07_REP_NUM] = { 1, 4, 8 };
+
+#define KB *(1 <<10)
+#define MB *(1 <<20)
+#define GB *(1U<<30)
+
+#define BIT7 128
+#define BIT6  64
+#define BIT5  32
+#define BIT4  16
+#define BIT1   2
+#define BIT0   1
+
+#define ZSTDv07_WINDOWLOG_ABSOLUTEMIN 10
+static const size_t ZSTDv07_fcs_fieldSize[4] = { 0, 2, 4, 8 };
+static const size_t ZSTDv07_did_fieldSize[4] = { 0, 1, 2, 4 };
+
+#define ZSTDv07_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
+static const size_t ZSTDv07_blockHeaderSize = ZSTDv07_BLOCKHEADERSIZE;
+typedef enum { bt_compressed, bt_raw, bt_rle, bt_end } blockType_t;
+
+#define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
+#define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
+
+#define HufLog 12
+typedef enum { lbt_huffman, lbt_repeat, lbt_raw, lbt_rle } litBlockType_t;
+
+#define LONGNBSEQ 0x7F00
+
+#define MINMATCH 3
+#define EQUAL_READ32 4
+
+#define Litbits  8
+#define MaxLit ((1<<Litbits) - 1)
+#define MaxML  52
+#define MaxLL  35
+#define MaxOff 28
+#define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
+#define MLFSELog    9
+#define LLFSELog    9
+#define OffFSELog   8
+
+#define FSEv07_ENCODING_RAW     0
+#define FSEv07_ENCODING_RLE     1
+#define FSEv07_ENCODING_STATIC  2
+#define FSEv07_ENCODING_DYNAMIC 3
+
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+                                     13,14,15,16 };
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+                                            -1,-1,-1,-1 };
+static const U32 LL_defaultNormLog = 6;
+
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+                                     12,13,14,15,16 };
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+                                            -1,-1,-1,-1,-1 };
+static const U32 ML_defaultNormLog = 6;
+
+static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
+                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const U32 OF_defaultNormLog = 5;
+
+
+/*-*******************************************
+*  Shared functions to include for inlining
+*********************************************/
+static void ZSTDv07_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+#define COPY8(d,s) { ZSTDv07_copy8(d,s); d+=8; s+=8; }
+
+/*! ZSTDv07_wildcopy() :
+*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
+#define WILDCOPY_OVERLENGTH 8
+MEM_STATIC void ZSTDv07_wildcopy(void* dst, const void* src, ptrdiff_t length)
+{
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+    do
+        COPY8(op, ip)
+    while (op < oend);
+}
+
+
+/*-*******************************************
+*  Private interfaces
+*********************************************/
+typedef struct ZSTDv07_stats_s ZSTDv07_stats_t;
+
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTDv07_match_t;
+
+typedef struct {
+    U32 price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTDv07_REP_INIT];
+} ZSTDv07_optimal_t;
+
+struct ZSTDv07_stats_s { U32 unused; };
+
+typedef struct {
+    void* buffer;
+    U32*  offsetStart;
+    U32*  offset;
+    BYTE* offCodeStart;
+    BYTE* litStart;
+    BYTE* lit;
+    U16*  litLengthStart;
+    U16*  litLength;
+    BYTE* llCodeStart;
+    U16*  matchLengthStart;
+    U16*  matchLength;
+    BYTE* mlCodeStart;
+    U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
+    U32   longLengthPos;
+    /* opt */
+    ZSTDv07_optimal_t* priceTable;
+    ZSTDv07_match_t* matchTable;
+    U32* matchLengthFreq;
+    U32* litLengthFreq;
+    U32* litFreq;
+    U32* offCodeFreq;
+    U32  matchLengthSum;
+    U32  matchSum;
+    U32  litLengthSum;
+    U32  litSum;
+    U32  offCodeSum;
+    U32  log2matchLengthSum;
+    U32  log2matchSum;
+    U32  log2litLengthSum;
+    U32  log2litSum;
+    U32  log2offCodeSum;
+    U32  factor;
+    U32  cachedPrice;
+    U32  cachedLitLength;
+    const BYTE* cachedLiterals;
+    ZSTDv07_stats_t stats;
+} seqStore_t;
+
+void ZSTDv07_seqToCodes(const seqStore_t* seqStorePtr, size_t const nbSeq);
+
+/* custom memory allocation functions */
+static const ZSTDv07_customMem defaultCustomMem = { ZSTDv07_defaultAllocFunction, ZSTDv07_defaultFreeFunction, NULL };
+
+#endif   /* ZSTDv07_CCOMMON_H_MODULE */
+/*
+    zstd - standard compression library
+    Copyright (C) 2014-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net
+*/
+
+/* ***************************************************************
+*  Tuning parameters
+*****************************************************************/
+/*!
+ * HEAPMODE :
+ * Select how default decompression function ZSTDv07_decompress() will allocate memory,
+ * in memory stack (0), or in memory heap (1, requires malloc())
+ */
+#ifndef ZSTDv07_HEAPMODE
+#  define ZSTDv07_HEAPMODE 1
+#endif
+
+
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#endif
+
+
+/*-*************************************
+*  Macros
+***************************************/
+#define ZSTDv07_isError ERR_isError   /* for inlining */
+#define FSEv07_isError  ERR_isError
+#define HUFv07_isError  ERR_isError
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTDv07_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+*   Context management
+***************************************************************/
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
+               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTDv07_dStage;
+
+struct ZSTDv07_DCtx_s
+{
+    FSEv07_DTable LLTable[FSEv07_DTABLE_SIZE_U32(LLFSELog)];
+    FSEv07_DTable OffTable[FSEv07_DTABLE_SIZE_U32(OffFSELog)];
+    FSEv07_DTable MLTable[FSEv07_DTABLE_SIZE_U32(MLFSELog)];
+    HUFv07_DTable hufTable[HUFv07_DTABLE_SIZE(HufLog)];  /* can accommodate HUFv07_decompress4X */
+    const void* previousDstEnd;
+    const void* base;
+    const void* vBase;
+    const void* dictEnd;
+    size_t expected;
+    U32 rep[3];
+    ZSTDv07_frameParams fParams;
+    blockType_t bType;   /* used in ZSTDv07_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
+    ZSTDv07_dStage stage;
+    U32 litEntropy;
+    U32 fseEntropy;
+    XXH64_state_t xxhState;
+    size_t headerSize;
+    U32 dictID;
+    const BYTE* litPtr;
+    ZSTDv07_customMem customMem;
+    size_t litSize;
+    BYTE litBuffer[ZSTDv07_BLOCKSIZE_ABSOLUTEMAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTDv07_FRAMEHEADERSIZE_MAX];
+};  /* typedef'd to ZSTDv07_DCtx within "zstd_static.h" */
+
+int ZSTDv07_isSkipFrame(ZSTDv07_DCtx* dctx);
+
+size_t ZSTDv07_sizeofDCtx (const ZSTDv07_DCtx* dctx) { return sizeof(*dctx); }
+
+size_t ZSTDv07_estimateDCtxSize(void) { return sizeof(ZSTDv07_DCtx); }
+
+size_t ZSTDv07_decompressBegin(ZSTDv07_DCtx* dctx)
+{
+    dctx->expected = ZSTDv07_frameHeaderSize_min;
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->previousDstEnd = NULL;
+    dctx->base = NULL;
+    dctx->vBase = NULL;
+    dctx->dictEnd = NULL;
+    dctx->hufTable[0] = (HUFv07_DTable)((HufLog)*0x1000001);
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    { int i; for (i=0; i<ZSTDv07_REP_NUM; i++) dctx->rep[i] = repStartValue[i]; }
+    return 0;
+}
+
+ZSTDv07_DCtx* ZSTDv07_createDCtx_advanced(ZSTDv07_customMem customMem)
+{
+    ZSTDv07_DCtx* dctx;
+
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    dctx = (ZSTDv07_DCtx*) customMem.customAlloc(customMem.opaque, sizeof(ZSTDv07_DCtx));
+    if (!dctx) return NULL;
+    memcpy(&dctx->customMem, &customMem, sizeof(ZSTDv07_customMem));
+    ZSTDv07_decompressBegin(dctx);
+    return dctx;
+}
+
+ZSTDv07_DCtx* ZSTDv07_createDCtx(void)
+{
+    return ZSTDv07_createDCtx_advanced(defaultCustomMem);
+}
+
+size_t ZSTDv07_freeDCtx(ZSTDv07_DCtx* dctx)
+{
+    if (dctx==NULL) return 0;   /* support free on NULL */
+    dctx->customMem.customFree(dctx->customMem.opaque, dctx);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+void ZSTDv07_copyDCtx(ZSTDv07_DCtx* dstDCtx, const ZSTDv07_DCtx* srcDCtx)
+{
+    memcpy(dstDCtx, srcDCtx,
+           sizeof(ZSTDv07_DCtx) - (ZSTDv07_BLOCKSIZE_ABSOLUTEMAX+WILDCOPY_OVERLENGTH + ZSTDv07_frameHeaderSize_max));  /* no need to copy workspace */
+}
+
+
+/*-*************************************************************
+*   Decompression section
+***************************************************************/
+
+/* Frame format description
+   Frame Header -  [ Block Header - Block ] - Frame End
+   1) Frame Header
+      - 4 bytes - Magic Number : ZSTDv07_MAGICNUMBER (defined within zstd.h)
+      - 1 byte  - Frame Descriptor
+   2) Block Header
+      - 3 bytes, starting with a 2-bits descriptor
+                 Uncompressed, Compressed, Frame End, unused
+   3) Block
+      See Block Format Description
+   4) Frame End
+      - 3 bytes, compatible with Block Header
+*/
+
+
+/* Frame Header :
+
+   1 byte - FrameHeaderDescription :
+   bit 0-1 : dictID (0, 1, 2 or 4 bytes)
+   bit 2   : checksumFlag
+   bit 3   : reserved (must be zero)
+   bit 4   : reserved (unused, can be any value)
+   bit 5   : Single Segment (if 1, WindowLog byte is not present)
+   bit 6-7 : FrameContentFieldSize (0, 2, 4, or 8)
+             if (SkippedWindowLog && !FrameContentFieldsize) FrameContentFieldsize=1;
+
+   Optional : WindowLog (0 or 1 byte)
+   bit 0-2 : octal Fractional (1/8th)
+   bit 3-7 : Power of 2, with 0 = 1 KB (up to 2 TB)
+
+   Optional : dictID (0, 1, 2 or 4 bytes)
+   Automatic adaptation
+   0 : no dictID
+   1 : 1 - 255
+   2 : 256 - 65535
+   4 : all other values
+
+   Optional : content size (0, 1, 2, 4 or 8 bytes)
+   0 : unknown          (fcfs==0 and swl==0)
+   1 : 0-255 bytes      (fcfs==0 and swl==1)
+   2 : 256 - 65535+256  (fcfs==1)
+   4 : 0 - 4GB-1        (fcfs==2)
+   8 : 0 - 16EB-1       (fcfs==3)
+*/
+
+
+/* Compressed Block, format description
+
+   Block = Literal Section - Sequences Section
+   Prerequisite : size of (compressed) block, maximum size of regenerated data
+
+   1) Literal Section
+
+   1.1) Header : 1-5 bytes
+        flags: 2 bits
+            00 compressed by Huff0
+            01 unused
+            10 is Raw (uncompressed)
+            11 is Rle
+            Note : using 01 => Huff0 with precomputed table ?
+            Note : delta map ? => compressed ?
+
+   1.1.1) Huff0-compressed literal block : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+   1.1.2) Raw (uncompressed) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RAW<<6) + (0<<4) + size
+               12 bits: (IS_RAW<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RAW<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.3) Rle (repeated single byte) literal block header : 1-3 bytes
+        size :  5 bits: (IS_RLE<<6) + (0<<4) + size
+               12 bits: (IS_RLE<<6) + (2<<4) + (size>>8)
+                        size&255
+               20 bits: (IS_RLE<<6) + (3<<4) + (size>>16)
+                        size>>8&255
+                        size&255
+
+   1.1.4) Huff0-compressed literal block, using precomputed CTables : 3-5 bytes
+            srcSize < 1 KB => 3 bytes (2-2-10-10) => single stream
+            srcSize < 1 KB => 3 bytes (2-2-10-10)
+            srcSize < 16KB => 4 bytes (2-2-14-14)
+            else           => 5 bytes (2-2-18-18)
+            big endian convention
+
+        1- CTable available (stored into workspace ?)
+        2- Small input (fast heuristic ? Full comparison ? depend on clevel ?)
+
+
+   1.2) Literal block content
+
+   1.2.1) Huff0 block, using sizes from header
+        See Huff0 format
+
+   1.2.2) Huff0 block, using prepared table
+
+   1.2.3) Raw content
+
+   1.2.4) single byte
+
+
+   2) Sequences section
+      TO DO
+*/
+
+/** ZSTDv07_frameHeaderSize() :
+*   srcSize must be >= ZSTDv07_frameHeaderSize_min.
+*   @return : size of the Frame Header */
+static size_t ZSTDv07_frameHeaderSize(const void* src, size_t srcSize)
+{
+    if (srcSize < ZSTDv07_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    {   BYTE const fhd = ((const BYTE*)src)[4];
+        U32 const dictID= fhd & 3;
+        U32 const directMode = (fhd >> 5) & 1;
+        U32 const fcsId = fhd >> 6;
+        return ZSTDv07_frameHeaderSize_min + !directMode + ZSTDv07_did_fieldSize[dictID] + ZSTDv07_fcs_fieldSize[fcsId]
+                + (directMode && !ZSTDv07_fcs_fieldSize[fcsId]);
+    }
+}
+
+
+/** ZSTDv07_getFrameParams() :
+*   decode Frame Header, or require larger `srcSize`.
+*   @return : 0, `fparamsPtr` is correctly filled,
+*            >0, `srcSize` is too small, result is expected `srcSize`,
+*             or an error code, which can be tested using ZSTDv07_isError() */
+size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize < ZSTDv07_frameHeaderSize_min) return ZSTDv07_frameHeaderSize_min;
+    if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) {
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
+            if (srcSize < ZSTDv07_skippableHeaderSize) return ZSTDv07_skippableHeaderSize; /* magic number + skippable frame length */
+            memset(fparamsPtr, 0, sizeof(*fparamsPtr));
+            fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
+            fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
+            return 0;
+        }
+        return ERROR(prefix_unknown);
+    }
+
+    /* ensure there is enough `srcSize` to fully read/decode frame header */
+    { size_t const fhsize = ZSTDv07_frameHeaderSize(src, srcSize);
+      if (srcSize < fhsize) return fhsize; }
+
+    {   BYTE const fhdByte = ip[4];
+        size_t pos = 5;
+        U32 const dictIDSizeCode = fhdByte&3;
+        U32 const checksumFlag = (fhdByte>>2)&1;
+        U32 const directMode = (fhdByte>>5)&1;
+        U32 const fcsID = fhdByte>>6;
+        U32 const windowSizeMax = 1U << ZSTDv07_WINDOWLOG_MAX;
+        U32 windowSize = 0;
+        U32 dictID = 0;
+        U64 frameContentSize = 0;
+        if ((fhdByte & 0x08) != 0) return ERROR(frameParameter_unsupported);   /* reserved bits, which must be zero */
+        if (!directMode) {
+            BYTE const wlByte = ip[pos++];
+            U32 const windowLog = (wlByte >> 3) + ZSTDv07_WINDOWLOG_ABSOLUTEMIN;
+            if (windowLog > ZSTDv07_WINDOWLOG_MAX) return ERROR(frameParameter_unsupported);
+            windowSize = (1U << windowLog);
+            windowSize += (windowSize >> 3) * (wlByte&7);
+        }
+
+        switch(dictIDSizeCode)
+        {
+            default:   /* impossible */
+            case 0 : break;
+            case 1 : dictID = ip[pos]; pos++; break;
+            case 2 : dictID = MEM_readLE16(ip+pos); pos+=2; break;
+            case 3 : dictID = MEM_readLE32(ip+pos); pos+=4; break;
+        }
+        switch(fcsID)
+        {
+            default:   /* impossible */
+            case 0 : if (directMode) frameContentSize = ip[pos]; break;
+            case 1 : frameContentSize = MEM_readLE16(ip+pos)+256; break;
+            case 2 : frameContentSize = MEM_readLE32(ip+pos); break;
+            case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
+        }
+        if (!windowSize) windowSize = (U32)frameContentSize;
+        if (windowSize > windowSizeMax) return ERROR(frameParameter_unsupported);
+        fparamsPtr->frameContentSize = frameContentSize;
+        fparamsPtr->windowSize = windowSize;
+        fparamsPtr->dictID = dictID;
+        fparamsPtr->checksumFlag = checksumFlag;
+    }
+    return 0;
+}
+
+
+/** ZSTDv07_getDecompressedSize() :
+*   compatible with legacy mode
+*   @return : decompressed size if known, 0 otherwise
+              note : 0 can mean any of the following :
+                   - decompressed size is not provided within frame header
+                   - frame header unknown / not supported
+                   - frame header not completely provided (`srcSize` too small) */
+unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize)
+{
+    {   ZSTDv07_frameParams fparams;
+        size_t const frResult = ZSTDv07_getFrameParams(&fparams, src, srcSize);
+        if (frResult!=0) return 0;
+        return fparams.frameContentSize;
+    }
+}
+
+
+/** ZSTDv07_decodeFrameHeader() :
+*   `srcSize` must be the size provided by ZSTDv07_frameHeaderSize().
+*   @return : 0 if success, or an error code, which can be tested using ZSTDv07_isError() */
+static size_t ZSTDv07_decodeFrameHeader(ZSTDv07_DCtx* dctx, const void* src, size_t srcSize)
+{
+    size_t const result = ZSTDv07_getFrameParams(&(dctx->fParams), src, srcSize);
+    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID)) return ERROR(dictionary_wrong);
+    if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
+    return result;
+}
+
+
+typedef struct
+{
+    blockType_t blockType;
+    U32 origSize;
+} blockProperties_t;
+
+/*! ZSTDv07_getcBlockSize() :
+*   Provides the size of compressed block from block header `src` */
+size_t ZSTDv07_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+{
+    const BYTE* const in = (const BYTE* const)src;
+    U32 cSize;
+
+    if (srcSize < ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    bpPtr->blockType = (blockType_t)((*in) >> 6);
+    cSize = in[2] + (in[1]<<8) + ((in[0] & 7)<<16);
+    bpPtr->origSize = (bpPtr->blockType == bt_rle) ? cSize : 0;
+
+    if (bpPtr->blockType == bt_end) return 0;
+    if (bpPtr->blockType == bt_rle) return 1;
+    return cSize;
+}
+
+
+static size_t ZSTDv07_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy(dst, src, srcSize);
+    return srcSize;
+}
+
+
+/*! ZSTDv07_decodeLiteralsBlock() :
+    @return : nb of bytes read from src (< srcSize ) */
+size_t ZSTDv07_decodeLiteralsBlock(ZSTDv07_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    const BYTE* const istart = (const BYTE*) src;
+
+    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
+
+    switch((litBlockType_t)(istart[0]>> 6))
+    {
+    case lbt_huffman:
+        {   size_t litSize, litCSize, singleStream=0;
+            U32 lhSize = (istart[0] >> 4) & 3;
+            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for lhSize, + cSize (+nbSeq) */
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                /* 2 - 2 - 10 - 10 */
+                lhSize=3;
+                singleStream = istart[0] & 16;
+                litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+                litCSize = ((istart[1] &  3) << 8) + istart[2];
+                break;
+            case 2:
+                /* 2 - 2 - 14 - 14 */
+                lhSize=4;
+                litSize  = ((istart[0] & 15) << 10) + (istart[1] << 2) + (istart[2] >> 6);
+                litCSize = ((istart[2] & 63) <<  8) + istart[3];
+                break;
+            case 3:
+                /* 2 - 2 - 18 - 18 */
+                lhSize=5;
+                litSize  = ((istart[0] & 15) << 14) + (istart[1] << 6) + (istart[2] >> 2);
+                litCSize = ((istart[2] &  3) << 16) + (istart[3] << 8) + istart[4];
+                break;
+            }
+            if (litSize > ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            if (HUFv07_isError(singleStream ?
+                            HUFv07_decompress1X2_DCtx(dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                            HUFv07_decompress4X_hufOnly (dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) ))
+                return ERROR(corruption_detected);
+
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            dctx->litEntropy = 1;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case lbt_repeat:
+        {   size_t litSize, litCSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            if (lhSize != 1)  /* only case supported for now : small litSize, single stream */
+                return ERROR(corruption_detected);
+            if (dctx->litEntropy==0)
+                return ERROR(dictionary_corrupted);
+
+            /* 2 - 2 - 10 - 10 */
+            lhSize=3;
+            litSize  = ((istart[0] & 15) << 6) + (istart[1] >> 2);
+            litCSize = ((istart[1] &  3) << 8) + istart[2];
+            if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
+
+            {   size_t const errorCode = HUFv07_decompress1X4_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTable);
+                if (HUFv07_isError(errorCode)) return ERROR(corruption_detected);
+            }
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+            return litCSize + lhSize;
+        }
+    case lbt_raw:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize=1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                break;
+            }
+
+            if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
+                memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return lhSize+litSize;
+            }
+            /* direct reference into compressed stream */
+            dctx->litPtr = istart+lhSize;
+            dctx->litSize = litSize;
+            return lhSize+litSize;
+        }
+    case lbt_rle:
+        {   size_t litSize;
+            U32 lhSize = ((istart[0]) >> 4) & 3;
+            switch(lhSize)
+            {
+            case 0: case 1: default:   /* note : default is impossible, since lhSize into [0..3] */
+                lhSize = 1;
+                litSize = istart[0] & 31;
+                break;
+            case 2:
+                litSize = ((istart[0] & 15) << 8) + istart[1];
+                break;
+            case 3:
+                litSize = ((istart[0] & 15) << 16) + (istart[1] << 8) + istart[2];
+                if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
+                break;
+            }
+            if (litSize > ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
+            memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+            dctx->litPtr = dctx->litBuffer;
+            dctx->litSize = litSize;
+            return lhSize+1;
+        }
+    default:
+        return ERROR(corruption_detected);   /* impossible */
+    }
+}
+
+
+/*! ZSTDv07_buildSeqTable() :
+    @return : nb bytes read from src,
+              or an error code if it fails, testable with ZSTDv07_isError()
+*/
+size_t ZSTDv07_buildSeqTable(FSEv07_DTable* DTable, U32 type, U32 max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+{
+    switch(type)
+    {
+    case FSEv07_ENCODING_RLE :
+        if (!srcSize) return ERROR(srcSize_wrong);
+        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
+        FSEv07_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        return 1;
+    case FSEv07_ENCODING_RAW :
+        FSEv07_buildDTable(DTable, defaultNorm, max, defaultLog);
+        return 0;
+    case FSEv07_ENCODING_STATIC:
+        if (!flagRepeatTable) return ERROR(corruption_detected);
+        return 0;
+    default :   /* impossible */
+    case FSEv07_ENCODING_DYNAMIC :
+        {   U32 tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSEv07_readNCount(norm, &max, &tableLog, src, srcSize);
+            if (FSEv07_isError(headerSize)) return ERROR(corruption_detected);
+            if (tableLog > maxLog) return ERROR(corruption_detected);
+            FSEv07_buildDTable(DTable, norm, max, tableLog);
+            return headerSize;
+    }   }
+}
+
+
+size_t ZSTDv07_decodeSeqHeaders(int* nbSeqPtr,
+                             FSEv07_DTable* DTableLL, FSEv07_DTable* DTableML, FSEv07_DTable* DTableOffb, U32 flagRepeatTable,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+
+    /* check */
+    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
+
+    /* SeqHead */
+    {   int nbSeq = *ip++;
+        if (!nbSeq) { *nbSeqPtr=0; return 1; }
+        if (nbSeq > 0x7F) {
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
+                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
+                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
+        }
+        *nbSeqPtr = nbSeq;
+    }
+
+    /* FSE table descriptors */
+    {   U32 const LLtype  = *ip >> 6;
+        U32 const OFtype = (*ip >> 4) & 3;
+        U32 const MLtype  = (*ip >> 2) & 3;
+        ip++;
+
+        /* check */
+        if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
+
+        /* Build DTables */
+        {   size_t const llhSize = ZSTDv07_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(llhSize)) return ERROR(corruption_detected);
+            ip += llhSize;
+        }
+        {   size_t const ofhSize = ZSTDv07_buildSeqTable(DTableOffb, OFtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(ofhSize)) return ERROR(corruption_detected);
+            ip += ofhSize;
+        }
+        {   size_t const mlhSize = ZSTDv07_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+            if (ZSTDv07_isError(mlhSize)) return ERROR(corruption_detected);
+            ip += mlhSize;
+    }   }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+} seq_t;
+
+typedef struct {
+    BITv07_DStream_t DStream;
+    FSEv07_DState_t stateLL;
+    FSEv07_DState_t stateOffb;
+    FSEv07_DState_t stateML;
+    size_t prevOffset[ZSTDv07_REP_INIT];
+} seqState_t;
+
+
+static seq_t ZSTDv07_decodeSequence(seqState_t* seqState)
+{
+    seq_t seq;
+
+    U32 const llCode = FSEv07_peekSymbol(&(seqState->stateLL));
+    U32 const mlCode = FSEv07_peekSymbol(&(seqState->stateML));
+    U32 const ofCode = FSEv07_peekSymbol(&(seqState->stateOffb));   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13,   14,    15,    16,    17,    18,
+                            19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29,   30,    31,    32,    33,    34,
+                            35, 37, 39, 41, 43, 47, 51, 59, 67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BITv07_readBits(&(seqState->DStream), ofBits);   /* <=  (ZSTDv07_WINDOWLOG_MAX-1) bits */
+            if (MEM_32bits()) BITv07_reloadDStream(&(seqState->DStream));
+        }
+
+        if (ofCode <= 1) {
+            if ((llCode == 0) & (offset <= 1)) offset = 1-offset;
+            if (offset) {
+                size_t const temp = seqState->prevOffset[offset];
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BITv07_readBits(&(seqState->DStream), mlBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BITv07_reloadDStream(&(seqState->DStream));
+
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BITv07_readBits(&(seqState->DStream), llBits) : 0);   /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BITv07_reloadDStream(&(seqState->DStream));
+
+    /* ANS state update */
+    FSEv07_updateState(&(seqState->stateLL), &(seqState->DStream));   /* <=  9 bits */
+    FSEv07_updateState(&(seqState->stateML), &(seqState->DStream));   /* <=  9 bits */
+    if (MEM_32bits()) BITv07_reloadDStream(&(seqState->DStream));     /* <= 18 bits */
+    FSEv07_updateState(&(seqState->stateOffb), &(seqState->DStream)); /* <=  8 bits */
+
+    return seq;
+}
+
+
+static
+size_t ZSTDv07_execSequence(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend-WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if ((oLitEnd>oend_w) | (oMatchEnd>oend)) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+
+    /* copy Literals */
+    ZSTDv07_wildcopy(op, *litPtr, sequence.litLength);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              while (op < oMatchEnd) *op++ = *match++;
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_w */
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTDv07_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTDv07_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTDv07_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTDv07_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+
+static size_t ZSTDv07_decompressSequences(
+                               ZSTDv07_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    FSEv07_DTable* DTableLL = dctx->LLTable;
+    FSEv07_DTable* DTableML = dctx->MLTable;
+    FSEv07_DTable* DTableOffb = dctx->OffTable;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTDv07_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->fseEntropy, ip, seqSize);
+        if (ZSTDv07_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTDv07_REP_INIT; i++) seqState.prevOffset[i] = dctx->rep[i]; }
+        { size_t const errorCode = BITv07_initDStream(&(seqState.DStream), ip, iend-ip);
+          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
+        FSEv07_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
+        FSEv07_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
+        FSEv07_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+
+        for ( ; (BITv07_reloadDStream(&(seqState.DStream)) <= BITv07_DStream_completed) && nbSeq ; ) {
+            nbSeq--;
+            {   seq_t const sequence = ZSTDv07_decodeSequence(&seqState);
+                size_t const oneSeqSize = ZSTDv07_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
+                if (ZSTDv07_isError(oneSeqSize)) return oneSeqSize;
+                op += oneSeqSize;
+        }   }
+
+        /* check if reached exact end */
+        if (nbSeq) return ERROR(corruption_detected);
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTDv07_REP_INIT; i++) dctx->rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        //if (litPtr > litEnd) return ERROR(corruption_detected);   /* too many literals already used */
+        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
+        memcpy(op, litPtr, lastLLSize);
+        op += lastLLSize;
+    }
+
+    return op-ostart;
+}
+
+
+static void ZSTDv07_checkContinuity(ZSTDv07_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+static size_t ZSTDv07_decompressBlock_internal(ZSTDv07_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+
+    if (srcSize >= ZSTDv07_BLOCKSIZE_ABSOLUTEMAX) return ERROR(srcSize_wrong);
+
+    /* Decode literals sub-block */
+    {   size_t const litCSize = ZSTDv07_decodeLiteralsBlock(dctx, src, srcSize);
+        if (ZSTDv07_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+    return ZSTDv07_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
+}
+
+
+size_t ZSTDv07_decompressBlock(ZSTDv07_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    size_t dSize;
+    ZSTDv07_checkContinuity(dctx, dst);
+    dSize = ZSTDv07_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
+
+
+/** ZSTDv07_insertBlock() :
+    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+ZSTDLIBv07_API size_t ZSTDv07_insertBlock(ZSTDv07_DCtx* dctx, const void* blockStart, size_t blockSize)
+{
+    ZSTDv07_checkContinuity(dctx, blockStart);
+    dctx->previousDstEnd = (const char*)blockStart + blockSize;
+    return blockSize;
+}
+
+
+size_t ZSTDv07_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+{
+    if (length > dstCapacity) return ERROR(dstSize_tooSmall);
+    memset(dst, byte, length);
+    return length;
+}
+
+
+/*! ZSTDv07_decompressFrame() :
+*   `dctx` must be properly initialized */
+static size_t ZSTDv07_decompressFrame(ZSTDv07_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t remainingSize = srcSize;
+
+    /* check */
+    if (srcSize < ZSTDv07_frameHeaderSize_min+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(frameHeaderSize)) return frameHeaderSize;
+        if (srcSize < frameHeaderSize+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+        if (ZSTDv07_decodeFrameHeader(dctx, src, frameHeaderSize)) return ERROR(corruption_detected);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        size_t decodedSize;
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTDv07_getcBlockSize(ip, iend-ip, &blockProperties);
+        if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv07_blockHeaderSize;
+        remainingSize -= ZSTDv07_blockHeaderSize;
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        switch(blockProperties.blockType)
+        {
+        case bt_compressed:
+            decodedSize = ZSTDv07_decompressBlock_internal(dctx, op, oend-op, ip, cBlockSize);
+            break;
+        case bt_raw :
+            decodedSize = ZSTDv07_copyRawBlock(op, oend-op, ip, cBlockSize);
+            break;
+        case bt_rle :
+            decodedSize = ZSTDv07_generateNxBytes(op, oend-op, *ip, blockProperties.origSize);
+            break;
+        case bt_end :
+            /* end of frame */
+            if (remainingSize) return ERROR(srcSize_wrong);
+            decodedSize = 0;
+            break;
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+        if (blockProperties.blockType == bt_end) break;   /* bt_end */
+
+        if (ZSTDv07_isError(decodedSize)) return decodedSize;
+        if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, op, decodedSize);
+        op += decodedSize;
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return op-ostart;
+}
+
+
+/*! ZSTDv07_decompress_usingPreparedDCtx() :
+*   Same as ZSTDv07_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
+*   It avoids reloading the dictionary each time.
+*   `preparedDCtx` must have been properly initialized using ZSTDv07_decompressBegin_usingDict().
+*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
+size_t ZSTDv07_decompress_usingPreparedDCtx(ZSTDv07_DCtx* dctx, const ZSTDv07_DCtx* refDCtx,
+                                         void* dst, size_t dstCapacity,
+                                   const void* src, size_t srcSize)
+{
+    ZSTDv07_copyDCtx(dctx, refDCtx);
+    ZSTDv07_checkContinuity(dctx, dst);
+    return ZSTDv07_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv07_decompress_usingDict(ZSTDv07_DCtx* dctx,
+                                 void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict, size_t dictSize)
+{
+    ZSTDv07_decompressBegin_usingDict(dctx, dict, dictSize);
+    ZSTDv07_checkContinuity(dctx, dst);
+    return ZSTDv07_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+}
+
+
+size_t ZSTDv07_decompressDCtx(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return ZSTDv07_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+}
+
+
+size_t ZSTDv07_decompress(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+#if defined(ZSTDv07_HEAPMODE) && (ZSTDv07_HEAPMODE==1)
+    size_t regenSize;
+    ZSTDv07_DCtx* const dctx = ZSTDv07_createDCtx();
+    if (dctx==NULL) return ERROR(memory_allocation);
+    regenSize = ZSTDv07_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
+    ZSTDv07_freeDCtx(dctx);
+    return regenSize;
+#else   /* stack mode */
+    ZSTDv07_DCtx dctx;
+    return ZSTDv07_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
+#endif
+}
+
+size_t ZSTDv07_findFrameCompressedSize(const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    size_t remainingSize = srcSize;
+
+    /* check */
+    if (srcSize < ZSTDv07_frameHeaderSize_min+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+
+    /* Frame Header */
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(frameHeaderSize)) return frameHeaderSize;
+        if (MEM_readLE32(src) != ZSTDv07_MAGICNUMBER) return ERROR(prefix_unknown);
+        if (srcSize < frameHeaderSize+ZSTDv07_blockHeaderSize) return ERROR(srcSize_wrong);
+        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+    }
+
+    /* Loop on each block */
+    while (1) {
+        blockProperties_t blockProperties;
+        size_t const cBlockSize = ZSTDv07_getcBlockSize(ip, remainingSize, &blockProperties);
+        if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+
+        ip += ZSTDv07_blockHeaderSize;
+        remainingSize -= ZSTDv07_blockHeaderSize;
+
+        if (blockProperties.blockType == bt_end) break;
+
+        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+        ip += cBlockSize;
+        remainingSize -= cBlockSize;
+    }
+
+    return ip - (const BYTE*)src;
+}
+
+/*_******************************
+*  Streaming Decompression API
+********************************/
+size_t ZSTDv07_nextSrcSizeToDecompress(ZSTDv07_DCtx* dctx)
+{
+    return dctx->expected;
+}
+
+int ZSTDv07_isSkipFrame(ZSTDv07_DCtx* dctx)
+{
+    return dctx->stage == ZSTDds_skipFrame;
+}
+
+/** ZSTDv07_decompressContinue() :
+*   @return : nb of bytes generated into `dst` (necessarily <= `dstCapacity)
+*             or an error code, which can be tested using ZSTDv07_isError() */
+size_t ZSTDv07_decompressContinue(ZSTDv07_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    /* Sanity check */
+    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);
+    if (dstCapacity) ZSTDv07_checkContinuity(dctx, dst);
+
+    switch (dctx->stage)
+    {
+    case ZSTDds_getFrameHeaderSize :
+        if (srcSize != ZSTDv07_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTDv07_MAGIC_SKIPPABLE_START) {
+            memcpy(dctx->headerBuffer, src, ZSTDv07_frameHeaderSize_min);
+            dctx->expected = ZSTDv07_skippableHeaderSize - ZSTDv07_frameHeaderSize_min; /* magic number + skippable frame length */
+            dctx->stage = ZSTDds_decodeSkippableHeader;
+            return 0;
+        }
+        dctx->headerSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+        if (ZSTDv07_isError(dctx->headerSize)) return dctx->headerSize;
+        memcpy(dctx->headerBuffer, src, ZSTDv07_frameHeaderSize_min);
+        if (dctx->headerSize > ZSTDv07_frameHeaderSize_min) {
+            dctx->expected = dctx->headerSize - ZSTDv07_frameHeaderSize_min;
+            dctx->stage = ZSTDds_decodeFrameHeader;
+            return 0;
+        }
+        dctx->expected = 0;   /* not necessary to copy more */
+	/* fall-through */
+    case ZSTDds_decodeFrameHeader:
+        {   size_t result;
+            memcpy(dctx->headerBuffer + ZSTDv07_frameHeaderSize_min, src, dctx->expected);
+            result = ZSTDv07_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
+            if (ZSTDv07_isError(result)) return result;
+            dctx->expected = ZSTDv07_blockHeaderSize;
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            return 0;
+        }
+    case ZSTDds_decodeBlockHeader:
+        {   blockProperties_t bp;
+            size_t const cBlockSize = ZSTDv07_getcBlockSize(src, ZSTDv07_blockHeaderSize, &bp);
+            if (ZSTDv07_isError(cBlockSize)) return cBlockSize;
+            if (bp.blockType == bt_end) {
+                if (dctx->fParams.checksumFlag) {
+                    U64 const h64 = XXH64_digest(&dctx->xxhState);
+                    U32 const h32 = (U32)(h64>>11) & ((1<<22)-1);
+                    const BYTE* const ip = (const BYTE*)src;
+                    U32 const check32 = ip[2] + (ip[1] << 8) + ((ip[0] & 0x3F) << 16);
+                    if (check32 != h32) return ERROR(checksum_wrong);
+                }
+                dctx->expected = 0;
+                dctx->stage = ZSTDds_getFrameHeaderSize;
+            } else {
+                dctx->expected = cBlockSize;
+                dctx->bType = bp.blockType;
+                dctx->stage = ZSTDds_decompressBlock;
+            }
+            return 0;
+        }
+    case ZSTDds_decompressBlock:
+        {   size_t rSize;
+            switch(dctx->bType)
+            {
+            case bt_compressed:
+                rSize = ZSTDv07_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize);
+                break;
+            case bt_raw :
+                rSize = ZSTDv07_copyRawBlock(dst, dstCapacity, src, srcSize);
+                break;
+            case bt_rle :
+                return ERROR(GENERIC);   /* not yet handled */
+                break;
+            case bt_end :   /* should never happen (filtered at phase 1) */
+                rSize = 0;
+                break;
+            default:
+                return ERROR(GENERIC);   /* impossible */
+            }
+            dctx->stage = ZSTDds_decodeBlockHeader;
+            dctx->expected = ZSTDv07_blockHeaderSize;
+            dctx->previousDstEnd = (char*)dst + rSize;
+            if (ZSTDv07_isError(rSize)) return rSize;
+            if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+            return rSize;
+        }
+    case ZSTDds_decodeSkippableHeader:
+        {   memcpy(dctx->headerBuffer + ZSTDv07_frameHeaderSize_min, src, dctx->expected);
+            dctx->expected = MEM_readLE32(dctx->headerBuffer + 4);
+            dctx->stage = ZSTDds_skipFrame;
+            return 0;
+        }
+    case ZSTDds_skipFrame:
+        {   dctx->expected = 0;
+            dctx->stage = ZSTDds_getFrameHeaderSize;
+            return 0;
+        }
+    default:
+        return ERROR(GENERIC);   /* impossible */
+    }
+}
+
+
+static size_t ZSTDv07_refDictContent(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    dctx->dictEnd = dctx->previousDstEnd;
+    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+    dctx->base = dict;
+    dctx->previousDstEnd = (const char*)dict + dictSize;
+    return 0;
+}
+
+static size_t ZSTDv07_loadEntropy(ZSTDv07_DCtx* dctx, const void* const dict, size_t const dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+
+    {   size_t const hSize = HUFv07_readDTableX4(dctx->hufTable, dict, dictSize);
+        if (HUFv07_isError(hSize)) return ERROR(dictionary_corrupted);
+        dictPtr += hSize;
+    }
+
+    {   short offcodeNCount[MaxOff+1];
+        U32 offcodeMaxValue=MaxOff, offcodeLog;
+        size_t const offcodeHeaderSize = FSEv07_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSEv07_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSEv07_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSEv07_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        { size_t const errorCode = FSEv07_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
+          if (FSEv07_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        dictPtr += litlengthHeaderSize;
+    }
+
+    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    dctx->rep[0] = MEM_readLE32(dictPtr+0); if (dctx->rep[0] == 0 || dctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
+    dctx->rep[1] = MEM_readLE32(dictPtr+4); if (dctx->rep[1] == 0 || dctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
+    dctx->rep[2] = MEM_readLE32(dictPtr+8); if (dctx->rep[2] == 0 || dctx->rep[2] >= dictSize) return ERROR(dictionary_corrupted);
+    dictPtr += 12;
+
+    dctx->litEntropy = dctx->fseEntropy = 1;
+    return dictPtr - (const BYTE*)dict;
+}
+
+static size_t ZSTDv07_decompress_insertDictionary(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return ZSTDv07_refDictContent(dctx, dict, dictSize);
+    {   U32 const magic = MEM_readLE32(dict);
+        if (magic != ZSTDv07_DICT_MAGIC) {
+            return ZSTDv07_refDictContent(dctx, dict, dictSize);   /* pure content mode */
+    }   }
+    dctx->dictID = MEM_readLE32((const char*)dict + 4);
+
+    /* load entropy tables */
+    dict = (const char*)dict + 8;
+    dictSize -= 8;
+    {   size_t const eSize = ZSTDv07_loadEntropy(dctx, dict, dictSize);
+        if (ZSTDv07_isError(eSize)) return ERROR(dictionary_corrupted);
+        dict = (const char*)dict + eSize;
+        dictSize -= eSize;
+    }
+
+    /* reference dictionary content */
+    return ZSTDv07_refDictContent(dctx, dict, dictSize);
+}
+
+
+size_t ZSTDv07_decompressBegin_usingDict(ZSTDv07_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    { size_t const errorCode = ZSTDv07_decompressBegin(dctx);
+      if (ZSTDv07_isError(errorCode)) return errorCode; }
+
+    if (dict && dictSize) {
+        size_t const errorCode = ZSTDv07_decompress_insertDictionary(dctx, dict, dictSize);
+        if (ZSTDv07_isError(errorCode)) return ERROR(dictionary_corrupted);
+    }
+
+    return 0;
+}
+
+
+struct ZSTDv07_DDict_s {
+    void* dict;
+    size_t dictSize;
+    ZSTDv07_DCtx* refContext;
+};  /* typedef'd tp ZSTDv07_CDict within zstd.h */
+
+ZSTDv07_DDict* ZSTDv07_createDDict_advanced(const void* dict, size_t dictSize, ZSTDv07_customMem customMem)
+{
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    {   ZSTDv07_DDict* const ddict = (ZSTDv07_DDict*) customMem.customAlloc(customMem.opaque, sizeof(*ddict));
+        void* const dictContent = customMem.customAlloc(customMem.opaque, dictSize);
+        ZSTDv07_DCtx* const dctx = ZSTDv07_createDCtx_advanced(customMem);
+
+        if (!dictContent || !ddict || !dctx) {
+            customMem.customFree(customMem.opaque, dictContent);
+            customMem.customFree(customMem.opaque, ddict);
+            customMem.customFree(customMem.opaque, dctx);
+            return NULL;
+        }
+
+        memcpy(dictContent, dict, dictSize);
+        {   size_t const errorCode = ZSTDv07_decompressBegin_usingDict(dctx, dictContent, dictSize);
+            if (ZSTDv07_isError(errorCode)) {
+                customMem.customFree(customMem.opaque, dictContent);
+                customMem.customFree(customMem.opaque, ddict);
+                customMem.customFree(customMem.opaque, dctx);
+                return NULL;
+        }   }
+
+        ddict->dict = dictContent;
+        ddict->dictSize = dictSize;
+        ddict->refContext = dctx;
+        return ddict;
+    }
+}
+
+/*! ZSTDv07_createDDict() :
+*   Create a digested dictionary, ready to start decompression without startup delay.
+*   `dict` can be released after `ZSTDv07_DDict` creation */
+ZSTDv07_DDict* ZSTDv07_createDDict(const void* dict, size_t dictSize)
+{
+    ZSTDv07_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTDv07_createDDict_advanced(dict, dictSize, allocator);
+}
+
+size_t ZSTDv07_freeDDict(ZSTDv07_DDict* ddict)
+{
+    ZSTDv07_freeFunction const cFree = ddict->refContext->customMem.customFree;
+    void* const opaque = ddict->refContext->customMem.opaque;
+    ZSTDv07_freeDCtx(ddict->refContext);
+    cFree(opaque, ddict->dict);
+    cFree(opaque, ddict);
+    return 0;
+}
+
+/*! ZSTDv07_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Use dictionary without significant overhead. */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const ZSTDv07_DDict* ddict)
+{
+    return ZSTDv07_decompress_usingPreparedDCtx(dctx, ddict->refContext,
+                                           dst, dstCapacity,
+                                           src, srcSize);
+}
+/*
+    Buffered version of Zstd compression library
+    Copyright (C) 2015-2016, Yann Collet.
+
+    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+    Redistribution and use in source and binary forms, with or without
+    modification, are permitted provided that the following conditions are
+    met:
+    * Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above
+    copyright notice, this list of conditions and the following disclaimer
+    in the documentation and/or other materials provided with the
+    distribution.
+    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+    "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+    OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+    SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+    LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+    DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+    THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+    (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+    OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - zstd homepage : http://www.zstd.net/
+*/
+
+
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv07_DCtx object is required to track streaming operations.
+*  Use ZBUFFv07_createDCtx() and ZBUFFv07_freeDCtx() to create/release resources.
+*  Use ZBUFFv07_decompressInit() to start a new decompression operation,
+*   or ZBUFFv07_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv07_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv07_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of @dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change @dst.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv07_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv07_recommendedDInSize() and ZBUFFv07_recommendedDOutSize()
+*  output : ZBUFFv07_recommendedDOutSize==128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv07_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv07_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
+               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFFv07_dStage;
+
+/* *** Resource management *** */
+struct ZBUFFv07_DCtx_s {
+    ZSTDv07_DCtx* zd;
+    ZSTDv07_frameParams fParams;
+    ZBUFFv07_dStage stage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t blockSize;
+    BYTE headerBuffer[ZSTDv07_FRAMEHEADERSIZE_MAX];
+    size_t lhSize;
+    ZSTDv07_customMem customMem;
+};   /* typedef'd to ZBUFFv07_DCtx within "zstd_buffered.h" */
+
+ZSTDLIBv07_API ZBUFFv07_DCtx* ZBUFFv07_createDCtx_advanced(ZSTDv07_customMem customMem);
+
+ZBUFFv07_DCtx* ZBUFFv07_createDCtx(void)
+{
+    return ZBUFFv07_createDCtx_advanced(defaultCustomMem);
+}
+
+ZBUFFv07_DCtx* ZBUFFv07_createDCtx_advanced(ZSTDv07_customMem customMem)
+{
+    ZBUFFv07_DCtx* zbd;
+
+    if (!customMem.customAlloc && !customMem.customFree)
+        customMem = defaultCustomMem;
+
+    if (!customMem.customAlloc || !customMem.customFree)
+        return NULL;
+
+    zbd = (ZBUFFv07_DCtx*)customMem.customAlloc(customMem.opaque, sizeof(ZBUFFv07_DCtx));
+    if (zbd==NULL) return NULL;
+    memset(zbd, 0, sizeof(ZBUFFv07_DCtx));
+    memcpy(&zbd->customMem, &customMem, sizeof(ZSTDv07_customMem));
+    zbd->zd = ZSTDv07_createDCtx_advanced(customMem);
+    if (zbd->zd == NULL) { ZBUFFv07_freeDCtx(zbd); return NULL; }
+    zbd->stage = ZBUFFds_init;
+    return zbd;
+}
+
+size_t ZBUFFv07_freeDCtx(ZBUFFv07_DCtx* zbd)
+{
+    if (zbd==NULL) return 0;   /* support free on null */
+    ZSTDv07_freeDCtx(zbd->zd);
+    if (zbd->inBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
+    if (zbd->outBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
+    zbd->customMem.customFree(zbd->customMem.opaque, zbd);
+    return 0;
+}
+
+
+/* *** Initialization *** */
+
+size_t ZBUFFv07_decompressInitDictionary(ZBUFFv07_DCtx* zbd, const void* dict, size_t dictSize)
+{
+    zbd->stage = ZBUFFds_loadHeader;
+    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
+    return ZSTDv07_decompressBegin_usingDict(zbd->zd, dict, dictSize);
+}
+
+size_t ZBUFFv07_decompressInit(ZBUFFv07_DCtx* zbd)
+{
+    return ZBUFFv07_decompressInitDictionary(zbd, NULL, 0);
+}
+
+
+/* internal util function */
+MEM_STATIC size_t ZBUFFv07_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    memcpy(dst, src, length);
+    return length;
+}
+
+
+/* *** Decompression *** */
+
+size_t ZBUFFv07_decompressContinue(ZBUFFv07_DCtx* zbd,
+                                void* dst, size_t* dstCapacityPtr,
+                          const void* src, size_t* srcSizePtr)
+{
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+    U32 notDone = 1;
+
+    while (notDone) {
+        switch(zbd->stage)
+        {
+        case ZBUFFds_init :
+            return ERROR(init_missing);
+
+        case ZBUFFds_loadHeader :
+            {   size_t const hSize = ZSTDv07_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
+                if (ZSTDv07_isError(hSize)) return hSize;
+                if (hSize != 0) {
+                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
+                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
+                        memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
+                        zbd->lhSize += iend-ip;
+                        *dstCapacityPtr = 0;
+                        return (hSize - zbd->lhSize) + ZSTDv07_blockHeaderSize;   /* remaining header bytes + next block header */
+                    }
+                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
+                    break;
+            }   }
+
+            /* Consume header */
+            {   size_t const h1Size = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTDv07_frameHeaderSize_min */
+                size_t const h1Result = ZSTDv07_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
+                if (ZSTDv07_isError(h1Result)) return h1Result;
+                if (h1Size < zbd->lhSize) {   /* long header */
+                    size_t const h2Size = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                    size_t const h2Result = ZSTDv07_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
+                    if (ZSTDv07_isError(h2Result)) return h2Result;
+            }   }
+
+            zbd->fParams.windowSize = MAX(zbd->fParams.windowSize, 1U << ZSTDv07_WINDOWLOG_ABSOLUTEMIN);
+
+            /* Frame header instruct buffer sizes */
+            {   size_t const blockSize = MIN(zbd->fParams.windowSize, ZSTDv07_BLOCKSIZE_ABSOLUTEMAX);
+                zbd->blockSize = blockSize;
+                if (zbd->inBuffSize < blockSize) {
+                    zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
+                    zbd->inBuffSize = blockSize;
+                    zbd->inBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, blockSize);
+                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
+                }
+                {   size_t const neededOutSize = zbd->fParams.windowSize + blockSize + WILDCOPY_OVERLENGTH * 2;
+                    if (zbd->outBuffSize < neededOutSize) {
+                        zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
+                        zbd->outBuffSize = neededOutSize;
+                        zbd->outBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, neededOutSize);
+                        if (zbd->outBuff == NULL) return ERROR(memory_allocation);
+            }   }   }
+            zbd->stage = ZBUFFds_read;
+            /* pass-through */
+	    /* fall-through */
+        case ZBUFFds_read:
+            {   size_t const neededInSize = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                if (neededInSize==0) {  /* end of frame */
+                    zbd->stage = ZBUFFds_init;
+                    notDone = 0;
+                    break;
+                }
+                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
+                    const int isSkipFrame = ZSTDv07_isSkipFrame(zbd->zd);
+                    size_t const decodedSize = ZSTDv07_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, (isSkipFrame ? 0 : zbd->outBuffSize - zbd->outStart),
+                        ip, neededInSize);
+                    if (ZSTDv07_isError(decodedSize)) return decodedSize;
+                    ip += neededInSize;
+                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    break;
+                }
+                if (ip==iend) { notDone = 0; break; }   /* no more input */
+                zbd->stage = ZBUFFds_load;
+            }
+	    /* fall-through */
+        case ZBUFFds_load:
+            {   size_t const neededInSize = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
+                size_t loadedSize;
+                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
+                loadedSize = ZBUFFv07_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
+                ip += loadedSize;
+                zbd->inPos += loadedSize;
+                if (loadedSize < toLoad) { notDone = 0; break; }   /* not enough input, wait for more */
+
+                /* decode loaded input */
+                {  const int isSkipFrame = ZSTDv07_isSkipFrame(zbd->zd);
+                   size_t const decodedSize = ZSTDv07_decompressContinue(zbd->zd,
+                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
+                        zbd->inBuff, neededInSize);
+                    if (ZSTDv07_isError(decodedSize)) return decodedSize;
+                    zbd->inPos = 0;   /* input is consumed */
+                    if (!decodedSize && !isSkipFrame) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
+                    zbd->outEnd = zbd->outStart +  decodedSize;
+                    zbd->stage = ZBUFFds_flush;
+                    /* break; */
+                    /* pass-through */
+                }
+	    }
+	    /* fall-through */
+        case ZBUFFds_flush:
+            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
+                size_t const flushedSize = ZBUFFv07_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
+                op += flushedSize;
+                zbd->outStart += flushedSize;
+                if (flushedSize == toFlushSize) {
+                    zbd->stage = ZBUFFds_read;
+                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
+                        zbd->outStart = zbd->outEnd = 0;
+                    break;
+                }
+                /* cannot flush everything */
+                notDone = 0;
+                break;
+            }
+        default: return ERROR(GENERIC);   /* impossible */
+    }   }
+
+    /* result */
+    *srcSizePtr = ip-istart;
+    *dstCapacityPtr = op-ostart;
+    {   size_t nextSrcSizeHint = ZSTDv07_nextSrcSizeToDecompress(zbd->zd);
+        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
+        return nextSrcSizeHint;
+    }
+}
+
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+size_t ZBUFFv07_recommendedDInSize(void)  { return ZSTDv07_BLOCKSIZE_ABSOLUTEMAX + ZSTDv07_blockHeaderSize /* block header size*/ ; }
+size_t ZBUFFv07_recommendedDOutSize(void) { return ZSTDv07_BLOCKSIZE_ABSOLUTEMAX; }
diff --git a/vendor/github.com/DataDog/zstd/zstd_v07.h b/vendor/github.com/DataDog/zstd/zstd_v07.h
new file mode 100644
index 0000000..6591cd3
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstd_v07.h
@@ -0,0 +1,182 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTDv07_H_235446
+#define ZSTDv07_H_235446
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*======  Dependency  ======*/
+#include <stddef.h>   /* size_t */
+
+
+/*======  Export for Windows  ======*/
+/*!
+*  ZSTDv07_DLL_EXPORT :
+*  Enable exporting of functions when building a Windows DLL
+*/
+#if defined(_WIN32) && defined(ZSTDv07_DLL_EXPORT) && (ZSTDv07_DLL_EXPORT==1)
+#  define ZSTDLIBv07_API __declspec(dllexport)
+#else
+#  define ZSTDLIBv07_API
+#endif
+
+
+/* *************************************
+*  Simple API
+***************************************/
+/*! ZSTDv07_getDecompressedSize() :
+*   @return : decompressed size if known, 0 otherwise.
+       note 1 : if `0`, follow up with ZSTDv07_getFrameParams() to know precise failure cause.
+       note 2 : decompressed size could be wrong or intentionally modified !
+                always ensure results fit within application's authorized limits */
+unsigned long long ZSTDv07_getDecompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTDv07_decompress() :
+    `compressedSize` : must be _exact_ size of compressed input, otherwise decompression will fail.
+    `dstCapacity` must be equal or larger than originalSize.
+    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+              or an errorCode if it fails (which can be tested using ZSTDv07_isError()) */
+ZSTDLIBv07_API size_t ZSTDv07_decompress( void* dst, size_t dstCapacity,
+                                    const void* src, size_t compressedSize);
+
+/**
+ZSTDv07_getFrameSrcSize() : get the source length of a ZSTD frame
+    compressedSize : The size of the 'src' buffer, at least as large as the frame pointed to by 'src'
+    return : the number of bytes that would be read to decompress this frame
+             or an errorCode if it fails (which can be tested using ZSTDv07_isError())
+*/
+size_t ZSTDv07_findFrameCompressedSize(const void* src, size_t compressedSize);
+
+/*======  Helper functions  ======*/
+ZSTDLIBv07_API unsigned    ZSTDv07_isError(size_t code);          /*!< tells if a `size_t` function result is an error code */
+ZSTDLIBv07_API const char* ZSTDv07_getErrorName(size_t code);     /*!< provides readable string from an error code */
+
+
+/*-*************************************
+*  Explicit memory management
+***************************************/
+/** Decompression context */
+typedef struct ZSTDv07_DCtx_s ZSTDv07_DCtx;
+ZSTDLIBv07_API ZSTDv07_DCtx* ZSTDv07_createDCtx(void);
+ZSTDLIBv07_API size_t     ZSTDv07_freeDCtx(ZSTDv07_DCtx* dctx);      /*!< @return : errorCode */
+
+/** ZSTDv07_decompressDCtx() :
+*   Same as ZSTDv07_decompress(), requires an allocated ZSTDv07_DCtx (see ZSTDv07_createDCtx()) */
+ZSTDLIBv07_API size_t ZSTDv07_decompressDCtx(ZSTDv07_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+
+/*-************************
+*  Simple dictionary API
+***************************/
+/*! ZSTDv07_decompress_usingDict() :
+*   Decompression using a pre-defined Dictionary content (see dictBuilder).
+*   Dictionary must be identical to the one used during compression.
+*   Note : This function load the dictionary, resulting in a significant startup time */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDict(ZSTDv07_DCtx* dctx,
+                                                   void* dst, size_t dstCapacity,
+                                             const void* src, size_t srcSize,
+                                             const void* dict,size_t dictSize);
+
+
+/*-**************************
+*  Advanced Dictionary API
+****************************/
+/*! ZSTDv07_createDDict() :
+*   Create a digested dictionary, ready to start decompression operation without startup delay.
+*   `dict` can be released after creation */
+typedef struct ZSTDv07_DDict_s ZSTDv07_DDict;
+ZSTDLIBv07_API ZSTDv07_DDict* ZSTDv07_createDDict(const void* dict, size_t dictSize);
+ZSTDLIBv07_API size_t      ZSTDv07_freeDDict(ZSTDv07_DDict* ddict);
+
+/*! ZSTDv07_decompress_usingDDict() :
+*   Decompression using a pre-digested Dictionary
+*   Faster startup than ZSTDv07_decompress_usingDict(), recommended when same dictionary is used multiple times. */
+ZSTDLIBv07_API size_t ZSTDv07_decompress_usingDDict(ZSTDv07_DCtx* dctx,
+                                                    void* dst, size_t dstCapacity,
+                                              const void* src, size_t srcSize,
+                                              const ZSTDv07_DDict* ddict);
+
+typedef struct {
+    unsigned long long frameContentSize;
+    unsigned windowSize;
+    unsigned dictID;
+    unsigned checksumFlag;
+} ZSTDv07_frameParams;
+
+ZSTDLIBv07_API size_t ZSTDv07_getFrameParams(ZSTDv07_frameParams* fparamsPtr, const void* src, size_t srcSize);   /**< doesn't consume input */
+
+
+
+
+/* *************************************
+*  Streaming functions
+***************************************/
+typedef struct ZBUFFv07_DCtx_s ZBUFFv07_DCtx;
+ZSTDLIBv07_API ZBUFFv07_DCtx* ZBUFFv07_createDCtx(void);
+ZSTDLIBv07_API size_t      ZBUFFv07_freeDCtx(ZBUFFv07_DCtx* dctx);
+
+ZSTDLIBv07_API size_t ZBUFFv07_decompressInit(ZBUFFv07_DCtx* dctx);
+ZSTDLIBv07_API size_t ZBUFFv07_decompressInitDictionary(ZBUFFv07_DCtx* dctx, const void* dict, size_t dictSize);
+
+ZSTDLIBv07_API size_t ZBUFFv07_decompressContinue(ZBUFFv07_DCtx* dctx,
+                                            void* dst, size_t* dstCapacityPtr,
+                                      const void* src, size_t* srcSizePtr);
+
+/*-***************************************************************************
+*  Streaming decompression howto
+*
+*  A ZBUFFv07_DCtx object is required to track streaming operations.
+*  Use ZBUFFv07_createDCtx() and ZBUFFv07_freeDCtx() to create/release resources.
+*  Use ZBUFFv07_decompressInit() to start a new decompression operation,
+*   or ZBUFFv07_decompressInitDictionary() if decompression requires a dictionary.
+*  Note that ZBUFFv07_DCtx objects can be re-init multiple times.
+*
+*  Use ZBUFFv07_decompressContinue() repetitively to consume your input.
+*  *srcSizePtr and *dstCapacityPtr can be any size.
+*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
+*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
+*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
+*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to help latency),
+*            or 0 when a frame is completely decoded,
+*            or an error code, which can be tested using ZBUFFv07_isError().
+*
+*  Hint : recommended buffer sizes (not compulsory) : ZBUFFv07_recommendedDInSize() and ZBUFFv07_recommendedDOutSize()
+*  output : ZBUFFv07_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
+*  input  : ZBUFFv07_recommendedDInSize == 128KB + 3;
+*           just follow indications from ZBUFFv07_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
+* *******************************************************************************/
+
+
+/* *************************************
+*  Tool functions
+***************************************/
+ZSTDLIBv07_API unsigned ZBUFFv07_isError(size_t errorCode);
+ZSTDLIBv07_API const char* ZBUFFv07_getErrorName(size_t errorCode);
+
+/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
+*   These sizes are just hints, they tend to offer better latency */
+ZSTDLIBv07_API size_t ZBUFFv07_recommendedDInSize(void);
+ZSTDLIBv07_API size_t ZBUFFv07_recommendedDOutSize(void);
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define ZSTDv07_MAGICNUMBER            0xFD2FB527   /* v0.7 */
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif  /* ZSTDv07_H_235446 */
diff --git a/vendor/github.com/DataDog/zstd/zstdmt_compress.c b/vendor/github.com/DataDog/zstd/zstdmt_compress.c
new file mode 100755
index 0000000..c7a205d
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstdmt_compress.c
@@ -0,0 +1,1831 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* ======   Tuning parameters   ====== */
+#define ZSTDMT_NBWORKERS_MAX 200
+#define ZSTDMT_JOBSIZE_MAX  (MEM_32bits() ? (512 MB) : (2 GB))  /* note : limited by `jobSize` type, which is `unsigned` */
+#define ZSTDMT_OVERLAPLOG_DEFAULT 6
+
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)   /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+/* ======   Dependencies   ====== */
+#include <string.h>      /* memcpy, memset */
+#include <limits.h>      /* INT_MAX */
+#include "pool.h"        /* threadpool */
+#include "threading.h"   /* mutex */
+#include "zstd_compress_internal.h"  /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include "zstd_ldm.h"
+#include "zstdmt_compress.h"
+
+/* Guards code to support resizing the SeqPool.
+ * We will want to resize the SeqPool to save memory in the future.
+ * Until then, comment the code out since it is unused.
+ */
+#define ZSTD_RESIZE_SEQPOOL 0
+
+/* ======   Debug   ====== */
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
+
+#  include <stdio.h>
+#  include <unistd.h>
+#  include <sys/times.h>
+#  define DEBUGLOGRAW(l, ...) if (l<=ZSTD_DEBUG) { fprintf(stderr, __VA_ARGS__); }
+
+#  define DEBUG_PRINTHEX(l,p,n) {            \
+    unsigned debug_u;                        \
+    for (debug_u=0; debug_u<(n); debug_u++)  \
+        DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
+    DEBUGLOGRAW(l, " \n");                   \
+}
+
+static unsigned long long GetCurrentClockTimeMicroseconds(void)
+{
+   static clock_t _ticksPerSecond = 0;
+   if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
+
+   { struct tms junk; clock_t newTicks = (clock_t) times(&junk);
+     return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond); }
+}
+
+#define MUTEX_WAIT_TIME_DLEVEL 6
+#define ZSTD_PTHREAD_MUTEX_LOCK(mutex) {          \
+    if (ZSTD_DEBUG >= MUTEX_WAIT_TIME_DLEVEL) {   \
+        unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
+        ZSTD_pthread_mutex_lock(mutex);           \
+        {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
+            unsigned long long const elapsedTime = (afterTime-beforeTime); \
+            if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \
+                DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
+                   elapsedTime, #mutex);          \
+        }   }                                     \
+    } else {                                      \
+        ZSTD_pthread_mutex_lock(mutex);           \
+    }                                             \
+}
+
+#else
+
+#  define ZSTD_PTHREAD_MUTEX_LOCK(m) ZSTD_pthread_mutex_lock(m)
+#  define DEBUG_PRINTHEX(l,p,n) {}
+
+#endif
+
+
+/* =====   Buffer Pool   ===== */
+/* a single Buffer Pool can be invoked from multiple threads in parallel */
+
+typedef struct buffer_s {
+    void* start;
+    size_t capacity;
+} buffer_t;
+
+static const buffer_t g_nullBuffer = { NULL, 0 };
+
+typedef struct ZSTDMT_bufferPool_s {
+    ZSTD_pthread_mutex_t poolMutex;
+    size_t bufferSize;
+    unsigned totalBuffers;
+    unsigned nbBuffers;
+    ZSTD_customMem cMem;
+    buffer_t bTable[1];   /* variable size */
+} ZSTDMT_bufferPool;
+
+static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    unsigned const maxNbBuffers = 2*nbWorkers + 3;
+    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
+        sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
+    if (bufPool==NULL) return NULL;
+    if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
+        ZSTD_free(bufPool, cMem);
+        return NULL;
+    }
+    bufPool->bufferSize = 64 KB;
+    bufPool->totalBuffers = maxNbBuffers;
+    bufPool->nbBuffers = 0;
+    bufPool->cMem = cMem;
+    return bufPool;
+}
+
+static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    unsigned u;
+    DEBUGLOG(3, "ZSTDMT_freeBufferPool (address:%08X)", (U32)(size_t)bufPool);
+    if (!bufPool) return;   /* compatibility with free on NULL */
+    for (u=0; u<bufPool->totalBuffers; u++) {
+        DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->bTable[u].start);
+        ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
+    ZSTD_free(bufPool, bufPool->cMem);
+}
+
+/* only works at initialization, not during compression */
+static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const poolSize = sizeof(*bufPool)
+                          + (bufPool->totalBuffers - 1) * sizeof(buffer_t);
+    unsigned u;
+    size_t totalBufferSize = 0;
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    for (u=0; u<bufPool->totalBuffers; u++)
+        totalBufferSize += bufPool->bTable[u].capacity;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+
+    return poolSize + totalBufferSize;
+}
+
+/* ZSTDMT_setBufferSize() :
+ * all future buffers provided by this buffer pool will have _at least_ this size
+ * note : it's better for all buffers to have same size,
+ * as they become freely interchangeable, reducing malloc/free usages and memory fragmentation */
+static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const bSize)
+{
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    DEBUGLOG(4, "ZSTDMT_setBufferSize: bSize = %u", (U32)bSize);
+    bufPool->bufferSize = bSize;
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+}
+
+/** ZSTDMT_getBuffer() :
+ *  assumption : bufPool must be valid
+ * @return : a buffer, with start pointer and size
+ *  note: allocation may fail, in this case, start==NULL and size==0 */
+static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
+{
+    size_t const bSize = bufPool->bufferSize;
+    DEBUGLOG(5, "ZSTDMT_getBuffer: bSize = %u", (U32)bufPool->bufferSize);
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers) {   /* try to use an existing buffer */
+        buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)];
+        size_t const availBufferSize = buf.capacity;
+        bufPool->bTable[bufPool->nbBuffers] = g_nullBuffer;
+        if ((availBufferSize >= bSize) & ((availBufferSize>>3) <= bSize)) {
+            /* large enough, but not too much */
+            DEBUGLOG(5, "ZSTDMT_getBuffer: provide buffer %u of size %u",
+                        bufPool->nbBuffers, (U32)buf.capacity);
+            ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+            return buf;
+        }
+        /* size conditions not respected : scratch this buffer, create new one */
+        DEBUGLOG(5, "ZSTDMT_getBuffer: existing buffer does not meet size conditions => freeing");
+        ZSTD_free(buf.start, bufPool->cMem);
+    }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+    /* create new buffer */
+    DEBUGLOG(5, "ZSTDMT_getBuffer: create a new buffer");
+    {   buffer_t buffer;
+        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
+        buffer.start = start;   /* note : start can be NULL if malloc fails ! */
+        buffer.capacity = (start==NULL) ? 0 : bSize;
+        if (start==NULL) {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: buffer allocation failure !!");
+        } else {
+            DEBUGLOG(5, "ZSTDMT_getBuffer: created buffer of size %u", (U32)bSize);
+        }
+        return buffer;
+    }
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+/** ZSTDMT_resizeBuffer() :
+ * assumption : bufPool must be valid
+ * @return : a buffer that is at least the buffer pool buffer size.
+ *           If a reallocation happens, the data in the input buffer is copied.
+ */
+static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
+{
+    size_t const bSize = bufPool->bufferSize;
+    if (buffer.capacity < bSize) {
+        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
+        buffer_t newBuffer;
+        newBuffer.start = start;
+        newBuffer.capacity = start == NULL ? 0 : bSize;
+        if (start != NULL) {
+            assert(newBuffer.capacity >= buffer.capacity);
+            memcpy(newBuffer.start, buffer.start, buffer.capacity);
+            DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
+            return newBuffer;
+        }
+        DEBUGLOG(5, "ZSTDMT_resizeBuffer: buffer allocation failure !!");
+    }
+    return buffer;
+}
+#endif
+
+/* store buffer for later re-use, up to pool capacity */
+static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
+{
+    if (buf.start == NULL) return;   /* compatible with release on NULL */
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer");
+    ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
+    if (bufPool->nbBuffers < bufPool->totalBuffers) {
+        bufPool->bTable[bufPool->nbBuffers++] = buf;  /* stored for later use */
+        DEBUGLOG(5, "ZSTDMT_releaseBuffer: stored buffer of size %u in slot %u",
+                    (U32)buf.capacity, (U32)(bufPool->nbBuffers-1));
+        ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+        return;
+    }
+    ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
+    /* Reached bufferPool capacity (should not happen) */
+    DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
+    ZSTD_free(buf.start, bufPool->cMem);
+}
+
+
+/* =====   Seq Pool Wrapper   ====== */
+
+static rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0};
+
+typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
+
+static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
+{
+    return ZSTDMT_sizeof_bufferPool(seqPool);
+}
+
+static rawSeqStore_t bufferToSeq(buffer_t buffer)
+{
+    rawSeqStore_t seq = {NULL, 0, 0, 0};
+    seq.seq = (rawSeq*)buffer.start;
+    seq.capacity = buffer.capacity / sizeof(rawSeq);
+    return seq;
+}
+
+static buffer_t seqToBuffer(rawSeqStore_t seq)
+{
+    buffer_t buffer;
+    buffer.start = seq.seq;
+    buffer.capacity = seq.capacity * sizeof(rawSeq);
+    return buffer;
+}
+
+static rawSeqStore_t ZSTDMT_getSeq(ZSTDMT_seqPool* seqPool)
+{
+    if (seqPool->bufferSize == 0) {
+        return kNullRawSeqStore;
+    }
+    return bufferToSeq(ZSTDMT_getBuffer(seqPool));
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+static rawSeqStore_t ZSTDMT_resizeSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  return bufferToSeq(ZSTDMT_resizeBuffer(seqPool, seqToBuffer(seq)));
+}
+#endif
+
+static void ZSTDMT_releaseSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  ZSTDMT_releaseBuffer(seqPool, seqToBuffer(seq));
+}
+
+static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
+{
+  ZSTDMT_setBufferSize(seqPool, nbSeq * sizeof(rawSeq));
+}
+
+static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_seqPool* seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    ZSTDMT_setNbSeq(seqPool, 0);
+    return seqPool;
+}
+
+static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
+{
+    ZSTDMT_freeBufferPool(seqPool);
+}
+
+
+
+/* =====   CCtx Pool   ===== */
+/* a single CCtx Pool can be invoked from multiple threads in parallel */
+
+typedef struct {
+    ZSTD_pthread_mutex_t poolMutex;
+    unsigned totalCCtx;
+    unsigned availCCtx;
+    ZSTD_customMem cMem;
+    ZSTD_CCtx* cctx[1];   /* variable size */
+} ZSTDMT_CCtxPool;
+
+/* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
+static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
+{
+    unsigned u;
+    for (u=0; u<pool->totalCCtx; u++)
+        ZSTD_freeCCtx(pool->cctx[u]);  /* note : compatible with free on NULL */
+    ZSTD_pthread_mutex_destroy(&pool->poolMutex);
+    ZSTD_free(pool, pool->cMem);
+}
+
+/* ZSTDMT_createCCtxPool() :
+ * implies nbWorkers >= 1 , checked by caller ZSTDMT_createCCtx() */
+static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbWorkers,
+                                              ZSTD_customMem cMem)
+{
+    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
+        sizeof(ZSTDMT_CCtxPool) + (nbWorkers-1)*sizeof(ZSTD_CCtx*), cMem);
+    assert(nbWorkers > 0);
+    if (!cctxPool) return NULL;
+    if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
+        ZSTD_free(cctxPool, cMem);
+        return NULL;
+    }
+    cctxPool->cMem = cMem;
+    cctxPool->totalCCtx = nbWorkers;
+    cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
+    cctxPool->cctx[0] = ZSTD_createCCtx_advanced(cMem);
+    if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
+    DEBUGLOG(3, "cctxPool created, with %u workers", nbWorkers);
+    return cctxPool;
+}
+
+/* only works during initialization phase, not during compression */
+static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
+{
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    {   unsigned const nbWorkers = cctxPool->totalCCtx;
+        size_t const poolSize = sizeof(*cctxPool)
+                                + (nbWorkers-1) * sizeof(ZSTD_CCtx*);
+        unsigned u;
+        size_t totalCCtxSize = 0;
+        for (u=0; u<nbWorkers; u++) {
+            totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctx[u]);
+        }
+        ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+        assert(nbWorkers > 0);
+        return poolSize + totalCCtxSize;
+    }
+}
+
+static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
+{
+    DEBUGLOG(5, "ZSTDMT_getCCtx");
+    ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
+    if (cctxPool->availCCtx) {
+        cctxPool->availCCtx--;
+        {   ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx];
+            ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+            return cctx;
+    }   }
+    ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
+    DEBUGLOG(5, "create one more CCtx");
+    return ZSTD_createCCtx_advanced(cctxPool->cMem);   /* note : can be NULL, when creation fails ! */
+}
+
+static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return;   /* compatibility with release on NULL */
+    ZSTD_pthread_mutex_lock(&pool->poolMutex);
+    if (pool->availCCtx < pool->totalCCtx)
+        pool->cctx[pool->availCCtx++] = cctx;
+    else {
+        /* pool overflow : should not happen, since totalCCtx==nbWorkers */
+        DEBUGLOG(4, "CCtx pool overflow : free cctx");
+        ZSTD_freeCCtx(cctx);
+    }
+    ZSTD_pthread_mutex_unlock(&pool->poolMutex);
+}
+
+/* ====   Serial State   ==== */
+
+typedef struct {
+    void const* start;
+    size_t size;
+} range_t;
+
+typedef struct {
+    /* All variables in the struct are protected by mutex. */
+    ZSTD_pthread_mutex_t mutex;
+    ZSTD_pthread_cond_t cond;
+    ZSTD_CCtx_params params;
+    ldmState_t ldmState;
+    XXH64_state_t xxhState;
+    unsigned nextJobID;
+    /* Protects ldmWindow.
+     * Must be acquired after the main mutex when acquiring both.
+     */
+    ZSTD_pthread_mutex_t ldmWindowMutex;
+    ZSTD_pthread_cond_t ldmWindowCond;  /* Signaled when ldmWindow is udpated */
+    ZSTD_window_t ldmWindow;  /* A thread-safe copy of ldmState.window */
+} serialState_t;
+
+static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool* seqPool, ZSTD_CCtx_params params)
+{
+    /* Adjust parameters */
+    if (params.ldmParams.enableLdm) {
+        DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
+        params.ldmParams.windowLog = params.cParams.windowLog;
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashEveryLog < 32);
+        serialState->ldmState.hashPower =
+                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+    } else {
+        memset(&params.ldmParams, 0, sizeof(params.ldmParams));
+    }
+    serialState->nextJobID = 0;
+    if (params.fParams.checksumFlag)
+        XXH64_reset(&serialState->xxhState, 0);
+    if (params.ldmParams.enableLdm) {
+        ZSTD_customMem cMem = params.customMem;
+        unsigned const hashLog = params.ldmParams.hashLog;
+        size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
+        unsigned const bucketLog =
+            params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
+        size_t const bucketSize = (size_t)1 << bucketLog;
+        unsigned const prevBucketLog =
+            serialState->params.ldmParams.hashLog -
+            serialState->params.ldmParams.bucketSizeLog;
+        /* Size the seq pool tables */
+        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, params.jobSize));
+        /* Reset the window */
+        ZSTD_window_clear(&serialState->ldmState.window);
+        serialState->ldmWindow = serialState->ldmState.window;
+        /* Resize tables and output space if necessary. */
+        if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
+            ZSTD_free(serialState->ldmState.hashTable, cMem);
+            serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_malloc(hashSize, cMem);
+        }
+        if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
+            ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
+            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_malloc(bucketSize, cMem);
+        }
+        if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
+            return 1;
+        /* Zero the tables */
+        memset(serialState->ldmState.hashTable, 0, hashSize);
+        memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
+    }
+    serialState->params = params;
+    return 0;
+}
+
+static int ZSTDMT_serialState_init(serialState_t* serialState)
+{
+    int initError = 0;
+    memset(serialState, 0, sizeof(*serialState));
+    initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
+    initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->ldmWindowCond, NULL);
+    return initError;
+}
+
+static void ZSTDMT_serialState_free(serialState_t* serialState)
+{
+    ZSTD_customMem cMem = serialState->params.customMem;
+    ZSTD_pthread_mutex_destroy(&serialState->mutex);
+    ZSTD_pthread_cond_destroy(&serialState->cond);
+    ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
+    ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
+    ZSTD_free(serialState->ldmState.hashTable, cMem);
+    ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
+}
+
+static void ZSTDMT_serialState_update(serialState_t* serialState,
+                                      ZSTD_CCtx* jobCCtx, rawSeqStore_t seqStore,
+                                      range_t src, unsigned jobID)
+{
+    /* Wait for our turn */
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    while (serialState->nextJobID < jobID) {
+        ZSTD_pthread_cond_wait(&serialState->cond, &serialState->mutex);
+    }
+    /* A future job may error and skip our job */
+    if (serialState->nextJobID == jobID) {
+        /* It is now our turn, do any processing necessary */
+        if (serialState->params.ldmParams.enableLdm) {
+            size_t error;
+            assert(seqStore.seq != NULL && seqStore.pos == 0 &&
+                   seqStore.size == 0 && seqStore.capacity > 0);
+            ZSTD_window_update(&serialState->ldmState.window, src.start, src.size);
+            error = ZSTD_ldm_generateSequences(
+                &serialState->ldmState, &seqStore,
+                &serialState->params.ldmParams, src.start, src.size);
+            /* We provide a large enough buffer to never fail. */
+            assert(!ZSTD_isError(error)); (void)error;
+            /* Update ldmWindow to match the ldmState.window and signal the main
+             * thread if it is waiting for a buffer.
+             */
+            ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+            serialState->ldmWindow = serialState->ldmState.window;
+            ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+            ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+        }
+        if (serialState->params.fParams.checksumFlag && src.size > 0)
+            XXH64_update(&serialState->xxhState, src.start, src.size);
+    }
+    /* Now it is the next jobs turn */
+    serialState->nextJobID++;
+    ZSTD_pthread_cond_broadcast(&serialState->cond);
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+    if (seqStore.size > 0) {
+        size_t const err = ZSTD_referenceExternalSequences(
+            jobCCtx, seqStore.seq, seqStore.size);
+        assert(serialState->params.ldmParams.enableLdm);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
+
+static void ZSTDMT_serialState_ensureFinished(serialState_t* serialState,
+                                              unsigned jobID, size_t cSize)
+{
+    ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
+    if (serialState->nextJobID <= jobID) {
+        assert(ZSTD_isError(cSize)); (void)cSize;
+        DEBUGLOG(5, "Skipping past job %u because of error", jobID);
+        serialState->nextJobID = jobID + 1;
+        ZSTD_pthread_cond_broadcast(&serialState->cond);
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+        ZSTD_window_clear(&serialState->ldmWindow);
+        ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+        ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+    }
+    ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+}
+
+
+/* ------------------------------------------ */
+/* =====          Worker thread         ===== */
+/* ------------------------------------------ */
+
+static const range_t kNullRange = { NULL, 0 };
+
+typedef struct {
+    size_t   consumed;                   /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx */
+    size_t   cSize;                      /* SHARED - set0 by mtctx, then modified by worker AND read by mtctx, then set0 by mtctx */
+    ZSTD_pthread_mutex_t job_mutex;      /* Thread-safe - used by mtctx and worker */
+    ZSTD_pthread_cond_t job_cond;        /* Thread-safe - used by mtctx and worker */
+    ZSTDMT_CCtxPool* cctxPool;           /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_bufferPool* bufPool;          /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_seqPool* seqPool;             /* Thread-safe - used by mtctx and (all) workers */
+    serialState_t* serial;               /* Thread-safe - used by mtctx and (all) workers */
+    buffer_t dstBuff;                    /* set by worker (or mtctx), then read by worker & mtctx, then modified by mtctx => no barrier */
+    range_t prefix;                      /* set by mtctx, then read by worker & mtctx => no barrier */
+    range_t src;                         /* set by mtctx, then read by worker & mtctx => no barrier */
+    unsigned jobID;                      /* set by mtctx, then read by worker => no barrier */
+    unsigned firstJob;                   /* set by mtctx, then read by worker => no barrier */
+    unsigned lastJob;                    /* set by mtctx, then read by worker => no barrier */
+    ZSTD_CCtx_params params;             /* set by mtctx, then read by worker => no barrier */
+    const ZSTD_CDict* cdict;             /* set by mtctx, then read by worker => no barrier */
+    unsigned long long fullFrameSize;    /* set by mtctx, then read by worker => no barrier */
+    size_t   dstFlushed;                 /* used only by mtctx */
+    unsigned frameChecksumNeeded;        /* used only by mtctx */
+} ZSTDMT_jobDescription;
+
+/* ZSTDMT_compressionJob() is a POOL_function type */
+void ZSTDMT_compressionJob(void* jobDescription)
+{
+    ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
+    ZSTD_CCtx_params jobParams = job->params;   /* do not modify job->params ! copy it, modify the copy */
+    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(job->cctxPool);
+    rawSeqStore_t rawSeqStore = ZSTDMT_getSeq(job->seqPool);
+    buffer_t dstBuff = job->dstBuff;
+
+    /* Don't compute the checksum for chunks, since we compute it externally,
+     * but write it in the header.
+     */
+    if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
+    /* Don't run LDM for the chunks, since we handle it externally */
+    jobParams.ldmParams.enableLdm = 0;
+
+    /* ressources */
+    if (cctx==NULL) {
+        job->cSize = ERROR(memory_allocation);
+        goto _endJob;
+    }
+    if (dstBuff.start == NULL) {   /* streaming job : doesn't provide a dstBuffer */
+        dstBuff = ZSTDMT_getBuffer(job->bufPool);
+        if (dstBuff.start==NULL) {
+            job->cSize = ERROR(memory_allocation);
+            goto _endJob;
+        }
+        job->dstBuff = dstBuff;   /* this value can be read in ZSTDMT_flush, when it copies the whole job */
+    }
+
+    /* init */
+    if (job->cdict) {
+        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, job->cdict, jobParams, job->fullFrameSize);
+        assert(job->firstJob);  /* only allowed for first job */
+        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
+    } else {  /* srcStart points at reloaded section */
+        U64 const pledgedSrcSize = job->firstJob ? job->fullFrameSize : job->src.size;
+        {   size_t const forceWindowError = ZSTD_CCtxParam_setParameter(&jobParams, ZSTD_p_forceMaxWindow, !job->firstJob);
+            if (ZSTD_isError(forceWindowError)) {
+                job->cSize = forceWindowError;
+                goto _endJob;
+        }   }
+        {   size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
+                                        job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
+                                        NULL, /*cdict*/
+                                        jobParams, pledgedSrcSize);
+            if (ZSTD_isError(initError)) {
+                job->cSize = initError;
+                goto _endJob;
+    }   }   }
+
+    /* Perform serial step as early as possible, but after CCtx initialization */
+    ZSTDMT_serialState_update(job->serial, cctx, rawSeqStore, job->src, job->jobID);
+
+    if (!job->firstJob) {  /* flush and overwrite frame header when it's not first job */
+        size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
+        if (ZSTD_isError(hSize)) { job->cSize = hSize; /* save error code */ goto _endJob; }
+        DEBUGLOG(5, "ZSTDMT_compressionJob: flush and overwrite %u bytes of frame header (not first job)", (U32)hSize);
+        ZSTD_invalidateRepCodes(cctx);
+    }
+
+    /* compress */
+    {   size_t const chunkSize = 4*ZSTD_BLOCKSIZE_MAX;
+        int const nbChunks = (int)((job->src.size + (chunkSize-1)) / chunkSize);
+        const BYTE* ip = (const BYTE*) job->src.start;
+        BYTE* const ostart = (BYTE*)dstBuff.start;
+        BYTE* op = ostart;
+        BYTE* oend = op + dstBuff.capacity;
+        int chunkNb;
+        if (sizeof(size_t) > sizeof(int)) assert(job->src.size < ((size_t)INT_MAX) * chunkSize);   /* check overflow */
+        DEBUGLOG(5, "ZSTDMT_compressionJob: compress %u bytes in %i blocks", (U32)job->src.size, nbChunks);
+        assert(job->cSize == 0);
+        for (chunkNb = 1; chunkNb < nbChunks; chunkNb++) {
+            size_t const cSize = ZSTD_compressContinue(cctx, op, oend-op, ip, chunkSize);
+            if (ZSTD_isError(cSize)) { job->cSize = cSize; goto _endJob; }
+            ip += chunkSize;
+            op += cSize; assert(op < oend);
+            /* stats */
+            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+            job->cSize += cSize;
+            job->consumed = chunkSize * chunkNb;
+            DEBUGLOG(5, "ZSTDMT_compressionJob: compress new block : cSize==%u bytes (total: %u)",
+                        (U32)cSize, (U32)job->cSize);
+            ZSTD_pthread_cond_signal(&job->job_cond);   /* warns some more data is ready to be flushed */
+            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+        }
+        /* last block */
+        assert(chunkSize > 0); assert((chunkSize & (chunkSize - 1)) == 0);  /* chunkSize must be power of 2 for mask==(chunkSize-1) to work */
+        if ((nbChunks > 0) | job->lastJob /*must output a "last block" flag*/ ) {
+            size_t const lastBlockSize1 = job->src.size & (chunkSize-1);
+            size_t const lastBlockSize = ((lastBlockSize1==0) & (job->src.size>=chunkSize)) ? chunkSize : lastBlockSize1;
+            size_t const cSize = (job->lastJob) ?
+                 ZSTD_compressEnd     (cctx, op, oend-op, ip, lastBlockSize) :
+                 ZSTD_compressContinue(cctx, op, oend-op, ip, lastBlockSize);
+            if (ZSTD_isError(cSize)) { job->cSize = cSize; goto _endJob; }
+            /* stats */
+            ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+            job->cSize += cSize;
+            ZSTD_pthread_mutex_unlock(&job->job_mutex);
+    }   }
+
+_endJob:
+    ZSTDMT_serialState_ensureFinished(job->serial, job->jobID, job->cSize);
+    if (job->prefix.size > 0)
+        DEBUGLOG(5, "Finished with prefix: %zx", (size_t)job->prefix.start);
+    DEBUGLOG(5, "Finished with source: %zx", (size_t)job->src.start);
+    /* release resources */
+    ZSTDMT_releaseSeq(job->seqPool, rawSeqStore);
+    ZSTDMT_releaseCCtx(job->cctxPool, cctx);
+    /* report */
+    ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
+    job->consumed = job->src.size;
+    ZSTD_pthread_cond_signal(&job->job_cond);
+    ZSTD_pthread_mutex_unlock(&job->job_mutex);
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+typedef struct {
+    range_t prefix;         /* read-only non-owned prefix buffer */
+    buffer_t buffer;
+    size_t filled;
+} inBuff_t;
+
+typedef struct {
+  BYTE* buffer;     /* The round input buffer. All jobs get references
+                     * to pieces of the buffer. ZSTDMT_tryGetInputRange()
+                     * handles handing out job input buffers, and makes
+                     * sure it doesn't overlap with any pieces still in use.
+                     */
+  size_t capacity;  /* The capacity of buffer. */
+  size_t pos;       /* The position of the current inBuff in the round
+                     * buffer. Updated past the end if the inBuff once
+                     * the inBuff is sent to the worker thread.
+                     * pos <= capacity.
+                     */
+} roundBuff_t;
+
+static const roundBuff_t kNullRoundBuff = {NULL, 0, 0};
+
+struct ZSTDMT_CCtx_s {
+    POOL_ctx* factory;
+    ZSTDMT_jobDescription* jobs;
+    ZSTDMT_bufferPool* bufPool;
+    ZSTDMT_CCtxPool* cctxPool;
+    ZSTDMT_seqPool* seqPool;
+    ZSTD_CCtx_params params;
+    size_t targetSectionSize;
+    size_t targetPrefixSize;
+    roundBuff_t roundBuff;
+    inBuff_t inBuff;
+    int jobReady;        /* 1 => one job is already prepared, but pool has shortage of workers. Don't create another one. */
+    serialState_t serial;
+    unsigned singleBlockingThread;
+    unsigned jobIDMask;
+    unsigned doneJobID;
+    unsigned nextJobID;
+    unsigned frameEnded;
+    unsigned allJobsCompleted;
+    unsigned long long frameContentSize;
+    unsigned long long consumed;
+    unsigned long long produced;
+    ZSTD_customMem cMem;
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+};
+
+static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
+{
+    U32 jobNb;
+    if (jobTable == NULL) return;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
+        ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
+    }
+    ZSTD_free(jobTable, cMem);
+}
+
+/* ZSTDMT_allocJobsTable()
+ * allocate and init a job table.
+ * update *nbJobsPtr to next power of 2 value, as size of table */
+static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_customMem cMem)
+{
+    U32 const nbJobsLog2 = ZSTD_highbit32(*nbJobsPtr) + 1;
+    U32 const nbJobs = 1 << nbJobsLog2;
+    U32 jobNb;
+    ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
+                ZSTD_calloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
+    int initError = 0;
+    if (jobTable==NULL) return NULL;
+    *nbJobsPtr = nbJobs;
+    for (jobNb=0; jobNb<nbJobs; jobNb++) {
+        initError |= ZSTD_pthread_mutex_init(&jobTable[jobNb].job_mutex, NULL);
+        initError |= ZSTD_pthread_cond_init(&jobTable[jobNb].job_cond, NULL);
+    }
+    if (initError != 0) {
+        ZSTDMT_freeJobsTable(jobTable, nbJobs, cMem);
+        return NULL;
+    }
+    return jobTable;
+}
+
+/* ZSTDMT_CCtxParam_setNbWorkers():
+ * Internal use only */
+size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
+{
+    if (nbWorkers > ZSTDMT_NBWORKERS_MAX) nbWorkers = ZSTDMT_NBWORKERS_MAX;
+    params->nbWorkers = nbWorkers;
+    params->overlapSizeLog = ZSTDMT_OVERLAPLOG_DEFAULT;
+    params->jobSize = 0;
+    return nbWorkers;
+}
+
+ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_CCtx* mtctx;
+    U32 nbJobs = nbWorkers + 2;
+    int initError;
+    DEBUGLOG(3, "ZSTDMT_createCCtx_advanced (nbWorkers = %u)", nbWorkers);
+
+    if (nbWorkers < 1) return NULL;
+    nbWorkers = MIN(nbWorkers , ZSTDMT_NBWORKERS_MAX);
+    if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
+        /* invalid custom allocator */
+        return NULL;
+
+    mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
+    if (!mtctx) return NULL;
+    ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
+    mtctx->cMem = cMem;
+    mtctx->allJobsCompleted = 1;
+    mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
+    mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
+    assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0);  /* ensure nbJobs is a power of 2 */
+    mtctx->jobIDMask = nbJobs - 1;
+    mtctx->bufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
+    initError = ZSTDMT_serialState_init(&mtctx->serial);
+    mtctx->roundBuff = kNullRoundBuff;
+    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool | !mtctx->seqPool | initError) {
+        ZSTDMT_freeCCtx(mtctx);
+        return NULL;
+    }
+    DEBUGLOG(3, "mt_cctx created, for %u threads", nbWorkers);
+    return mtctx;
+}
+
+ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers)
+{
+    return ZSTDMT_createCCtx_advanced(nbWorkers, ZSTD_defaultCMem);
+}
+
+
+/* ZSTDMT_releaseAllJobResources() :
+ * note : ensure all workers are killed first ! */
+static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
+{
+    unsigned jobID;
+    DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
+    for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
+        DEBUGLOG(4, "job%02u: release dst address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].dstBuff.start);
+        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
+        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+        mtctx->jobs[jobID].cSize = 0;
+    }
+    memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->allJobsCompleted = 1;
+}
+
+static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
+{
+    DEBUGLOG(4, "ZSTDMT_waitForAllJobsCompleted");
+    while (mtctx->doneJobID < mtctx->nextJobID) {
+        unsigned const jobID = mtctx->doneJobID & mtctx->jobIDMask;
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
+        while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
+            DEBUGLOG(5, "waiting for jobCompleted signal from job %u", mtctx->doneJobID);   /* we want to block when waiting for data to flush */
+            ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
+        }
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
+        mtctx->doneJobID++;
+    }
+}
+
+size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx==NULL) return 0;   /* compatible with free on NULL */
+    POOL_free(mtctx->factory);   /* stop and free worker threads */
+    ZSTDMT_releaseAllJobResources(mtctx);  /* release job resources into pools first */
+    ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+    ZSTDMT_freeBufferPool(mtctx->bufPool);
+    ZSTDMT_freeCCtxPool(mtctx->cctxPool);
+    ZSTDMT_freeSeqPool(mtctx->seqPool);
+    ZSTDMT_serialState_free(&mtctx->serial);
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    if (mtctx->roundBuff.buffer)
+        ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
+    ZSTD_free(mtctx, mtctx->cMem);
+    return 0;
+}
+
+size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx == NULL) return 0;   /* supports sizeof NULL */
+    return sizeof(*mtctx)
+            + POOL_sizeof(mtctx->factory)
+            + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
+            + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
+            + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
+            + ZSTDMT_sizeof_seqPool(mtctx->seqPool)
+            + ZSTD_sizeof_CDict(mtctx->cdictLocal)
+            + mtctx->roundBuff.capacity;
+}
+
+/* Internal only */
+size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params,
+                                ZSTDMT_parameter parameter, unsigned value) {
+    DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter");
+    switch(parameter)
+    {
+    case ZSTDMT_p_jobSize :
+        DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter : set jobSize to %u", value);
+        if ( (value > 0)  /* value==0 => automatic job size */
+           & (value < ZSTDMT_JOBSIZE_MIN) )
+            value = ZSTDMT_JOBSIZE_MIN;
+        params->jobSize = value;
+        return value;
+    case ZSTDMT_p_overlapSectionLog :
+        if (value > 9) value = 9;
+        DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);
+        params->overlapSizeLog = (value >= 9) ? 9 : value;
+        return value;
+    default :
+        return ERROR(parameter_unsupported);
+    }
+}
+
+size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value)
+{
+    DEBUGLOG(4, "ZSTDMT_setMTCtxParameter");
+    switch(parameter)
+    {
+    case ZSTDMT_p_jobSize :
+        return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
+    case ZSTDMT_p_overlapSectionLog :
+        return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
+    default :
+        return ERROR(parameter_unsupported);
+    }
+}
+
+/* Sets parameters relevant to the compression job,
+ * initializing others to default values. */
+static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(ZSTD_CCtx_params const params)
+{
+    ZSTD_CCtx_params jobParams;
+    memset(&jobParams, 0, sizeof(jobParams));
+
+    jobParams.cParams = params.cParams;
+    jobParams.fParams = params.fParams;
+    jobParams.compressionLevel = params.compressionLevel;
+    jobParams.disableLiteralCompression = params.disableLiteralCompression;
+
+    return jobParams;
+}
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates only a selected set of compression parameters, to remain compatible with current frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams)
+{
+    U32 const saved_wlog = mtctx->params.cParams.windowLog;   /* Do not modify windowLog while compressing */
+    int const compressionLevel = cctxParams->compressionLevel;
+    DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
+                compressionLevel);
+    mtctx->params.compressionLevel = compressionLevel;
+    {   ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, 0, 0);
+        cParams.windowLog = saved_wlog;
+        mtctx->params.cParams = cParams;
+    }
+}
+
+/* ZSTDMT_getNbWorkers():
+ * @return nb threads currently active in mtctx.
+ * mtctx must be valid */
+unsigned ZSTDMT_getNbWorkers(const ZSTDMT_CCtx* mtctx)
+{
+    assert(mtctx != NULL);
+    return mtctx->params.nbWorkers;
+}
+
+/* ZSTDMT_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads.
+ * Note : mutex will be acquired during statistics collection. */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
+{
+    ZSTD_frameProgression fps;
+    DEBUGLOG(6, "ZSTDMT_getFrameProgression");
+    fps.consumed = mtctx->consumed;
+    fps.produced = mtctx->produced;
+    fps.ingested = mtctx->consumed + mtctx->inBuff.filled;
+    {   unsigned jobNb;
+        unsigned lastJobNb = mtctx->nextJobID + mtctx->jobReady; assert(mtctx->jobReady <= 1);
+        DEBUGLOG(6, "ZSTDMT_getFrameProgression: jobs: from %u to <%u (jobReady:%u)",
+                    mtctx->doneJobID, lastJobNb, mtctx->jobReady)
+        for (jobNb = mtctx->doneJobID ; jobNb < lastJobNb ; jobNb++) {
+            unsigned const wJobID = jobNb & mtctx->jobIDMask;
+            ZSTD_pthread_mutex_lock(&mtctx->jobs[wJobID].job_mutex);
+            {   size_t const cResult = mtctx->jobs[wJobID].cSize;
+                size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
+                fps.consumed += mtctx->jobs[wJobID].consumed;
+                fps.ingested += mtctx->jobs[wJobID].src.size;
+                fps.produced += produced;
+            }
+            ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        }
+    }
+    return fps;
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+static size_t ZSTDMT_computeTargetJobLog(ZSTD_CCtx_params const params)
+{
+    if (params.ldmParams.enableLdm)
+        return MAX(21, params.cParams.chainLog + 4);
+    return MAX(20, params.cParams.windowLog + 2);
+}
+
+static size_t ZSTDMT_computeOverlapLog(ZSTD_CCtx_params const params)
+{
+    unsigned const overlapRLog = (params.overlapSizeLog>9) ? 0 : 9-params.overlapSizeLog;
+    if (params.ldmParams.enableLdm)
+        return (MIN(params.cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2) - overlapRLog);
+    return overlapRLog >= 9 ? 0 : (params.cParams.windowLog - overlapRLog);
+}
+
+static unsigned ZSTDMT_computeNbJobs(ZSTD_CCtx_params params, size_t srcSize, unsigned nbWorkers) {
+    assert(nbWorkers>0);
+    {   size_t const jobSizeTarget = (size_t)1 << ZSTDMT_computeTargetJobLog(params);
+        size_t const jobMaxSize = jobSizeTarget << 2;
+        size_t const passSizeMax = jobMaxSize * nbWorkers;
+        unsigned const multiplier = (unsigned)(srcSize / passSizeMax) + 1;
+        unsigned const nbJobsLarge = multiplier * nbWorkers;
+        unsigned const nbJobsMax = (unsigned)(srcSize / jobSizeTarget) + 1;
+        unsigned const nbJobsSmall = MIN(nbJobsMax, nbWorkers);
+        return (multiplier>1) ? nbJobsLarge : nbJobsSmall;
+}   }
+
+/* ZSTDMT_compress_advanced_internal() :
+ * This is a blocking function : it will only give back control to caller after finishing its compression job.
+ */
+static size_t ZSTDMT_compress_advanced_internal(
+                ZSTDMT_CCtx* mtctx,
+                void* dst, size_t dstCapacity,
+          const void* src, size_t srcSize,
+          const ZSTD_CDict* cdict,
+                ZSTD_CCtx_params params)
+{
+    ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(params);
+    size_t const overlapSize = (size_t)1 << ZSTDMT_computeOverlapLog(params);
+    unsigned const nbJobs = ZSTDMT_computeNbJobs(params, srcSize, params.nbWorkers);
+    size_t const proposedJobSize = (srcSize + (nbJobs-1)) / nbJobs;
+    size_t const avgJobSize = (((proposedJobSize-1) & 0x1FFFF) < 0x7FFF) ? proposedJobSize + 0xFFFF : proposedJobSize;   /* avoid too small last block */
+    const char* const srcStart = (const char*)src;
+    size_t remainingSrcSize = srcSize;
+    unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbJobs : (unsigned)(dstCapacity / ZSTD_compressBound(avgJobSize));  /* presumes avgJobSize >= 256 KB, which should be the case */
+    size_t frameStartPos = 0, dstBufferPos = 0;
+    assert(jobParams.nbWorkers == 0);
+    assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
+
+    params.jobSize = (U32)avgJobSize;
+    DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: nbJobs=%2u (rawSize=%u bytes; fixedSize=%u) ",
+                nbJobs, (U32)proposedJobSize, (U32)avgJobSize);
+
+    if ((nbJobs==1) | (params.nbWorkers<=1)) {   /* fallback to single-thread mode : this is a blocking invocation anyway */
+        ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
+        DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: fallback to single-thread mode");
+        if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
+        return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, jobParams);
+    }
+
+    assert(avgJobSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), required to compress directly into Dst (no additional buffer) */
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgJobSize) );
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+        return ERROR(memory_allocation);
+
+    if (nbJobs > mtctx->jobIDMask+1) {  /* enlarge job table */
+        U32 jobsTableSize = nbJobs;
+        ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
+        mtctx->jobIDMask = 0;
+        mtctx->jobs = ZSTDMT_createJobsTable(&jobsTableSize, mtctx->cMem);
+        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
+        assert((jobsTableSize != 0) && ((jobsTableSize & (jobsTableSize - 1)) == 0));  /* ensure jobsTableSize is a power of 2 */
+        mtctx->jobIDMask = jobsTableSize - 1;
+    }
+
+    {   unsigned u;
+        for (u=0; u<nbJobs; u++) {
+            size_t const jobSize = MIN(remainingSrcSize, avgJobSize);
+            size_t const dstBufferCapacity = ZSTD_compressBound(jobSize);
+            buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
+            buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
+            size_t dictSize = u ? overlapSize : 0;
+
+            mtctx->jobs[u].prefix.start = srcStart + frameStartPos - dictSize;
+            mtctx->jobs[u].prefix.size = dictSize;
+            mtctx->jobs[u].src.start = srcStart + frameStartPos;
+            mtctx->jobs[u].src.size = jobSize; assert(jobSize > 0);  /* avoid job.src.size == 0 */
+            mtctx->jobs[u].consumed = 0;
+            mtctx->jobs[u].cSize = 0;
+            mtctx->jobs[u].cdict = (u==0) ? cdict : NULL;
+            mtctx->jobs[u].fullFrameSize = srcSize;
+            mtctx->jobs[u].params = jobParams;
+            /* do not calculate checksum within sections, but write it in header for first section */
+            mtctx->jobs[u].dstBuff = dstBuffer;
+            mtctx->jobs[u].cctxPool = mtctx->cctxPool;
+            mtctx->jobs[u].bufPool = mtctx->bufPool;
+            mtctx->jobs[u].seqPool = mtctx->seqPool;
+            mtctx->jobs[u].serial = &mtctx->serial;
+            mtctx->jobs[u].jobID = u;
+            mtctx->jobs[u].firstJob = (u==0);
+            mtctx->jobs[u].lastJob = (u==nbJobs-1);
+
+            DEBUGLOG(5, "ZSTDMT_compress_advanced_internal: posting job %u  (%u bytes)", u, (U32)jobSize);
+            DEBUG_PRINTHEX(6, mtctx->jobs[u].prefix.start, 12);
+            POOL_add(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[u]);
+
+            frameStartPos += jobSize;
+            dstBufferPos += dstBufferCapacity;
+            remainingSrcSize -= jobSize;
+    }   }
+
+    /* collect result */
+    {   size_t error = 0, dstPos = 0;
+        unsigned jobID;
+        for (jobID=0; jobID<nbJobs; jobID++) {
+            DEBUGLOG(5, "waiting for job %u ", jobID);
+            ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
+            while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
+                DEBUGLOG(5, "waiting for jobCompleted signal from job %u", jobID);
+                ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
+            }
+            ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
+            DEBUGLOG(5, "ready to write job %u ", jobID);
+
+            {   size_t const cSize = mtctx->jobs[jobID].cSize;
+                if (ZSTD_isError(cSize)) error = cSize;
+                if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
+                if (jobID) {   /* note : job 0 is written directly at dst, which is correct position */
+                    if (!error)
+                        memmove((char*)dst + dstPos, mtctx->jobs[jobID].dstBuff.start, cSize);  /* may overlap when job compressed within dst */
+                    if (jobID >= compressWithinDst) {  /* job compressed into its own buffer, which must be released */
+                        DEBUGLOG(5, "releasing buffer %u>=%u", jobID, compressWithinDst);
+                        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
+                }   }
+                mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+                mtctx->jobs[jobID].cSize = 0;
+                dstPos += cSize ;
+            }
+        }  /* for (jobID=0; jobID<nbJobs; jobID++) */
+
+        DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
+        if (params.fParams.checksumFlag) {
+            U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
+            if (dstPos + 4 > dstCapacity) {
+                error = ERROR(dstSize_tooSmall);
+            } else {
+                DEBUGLOG(4, "writing checksum : %08X \n", checksum);
+                MEM_writeLE32((char*)dst + dstPos, checksum);
+                dstPos += 4;
+        }   }
+
+        if (!error) DEBUGLOG(4, "compressed size : %u  ", (U32)dstPos);
+        return error ? error : dstPos;
+    }
+}
+
+size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const ZSTD_CDict* cdict,
+                               ZSTD_parameters params,
+                               unsigned overlapLog)
+{
+    ZSTD_CCtx_params cctxParams = mtctx->params;
+    cctxParams.cParams = params.cParams;
+    cctxParams.fParams = params.fParams;
+    cctxParams.overlapSizeLog = overlapLog;
+    return ZSTDMT_compress_advanced_internal(mtctx,
+                                             dst, dstCapacity,
+                                             src, srcSize,
+                                             cdict, cctxParams);
+}
+
+
+size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
+                           void* dst, size_t dstCapacity,
+                     const void* src, size_t srcSize,
+                           int compressionLevel)
+{
+    U32 const overlapLog = (compressionLevel >= ZSTD_maxCLevel()) ? 9 : ZSTDMT_OVERLAPLOG_DEFAULT;
+    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
+    params.fParams.contentSizeFlag = 1;
+    return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
+}
+
+
+/* ====================================== */
+/* =======      Streaming API     ======= */
+/* ====================================== */
+
+size_t ZSTDMT_initCStream_internal(
+        ZSTDMT_CCtx* mtctx,
+        const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+        const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
+        unsigned long long pledgedSrcSize)
+{
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal (pledgedSrcSize=%u, nbWorkers=%u, cctxPool=%u, disableLiteralCompression=%i)",
+                (U32)pledgedSrcSize, params.nbWorkers, mtctx->cctxPool->totalCCtx, params.disableLiteralCompression);
+    /* params are supposed to be fully validated at this point */
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
+    assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
+
+    /* init */
+    if (params.jobSize == 0) {
+        params.jobSize = 1U << ZSTDMT_computeTargetJobLog(params);
+    }
+    if (params.jobSize > ZSTDMT_JOBSIZE_MAX) params.jobSize = ZSTDMT_JOBSIZE_MAX;
+
+    mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN);  /* do not trigger multi-threading when srcSize is too small */
+    if (mtctx->singleBlockingThread) {
+        ZSTD_CCtx_params const singleThreadParams = ZSTDMT_initJobCCtxParams(params);
+        DEBUGLOG(5, "ZSTDMT_initCStream_internal: switch to single blocking thread mode");
+        assert(singleThreadParams.nbWorkers == 0);
+        return ZSTD_initCStream_internal(mtctx->cctxPool->cctx[0],
+                                         dict, dictSize, cdict,
+                                         singleThreadParams, pledgedSrcSize);
+    }
+
+    DEBUGLOG(4, "ZSTDMT_initCStream_internal: %u workers", params.nbWorkers);
+
+    if (mtctx->allJobsCompleted == 0) {   /* previous compression not correctly finished */
+        ZSTDMT_waitForAllJobsCompleted(mtctx);
+        ZSTDMT_releaseAllJobResources(mtctx);
+        mtctx->allJobsCompleted = 1;
+    }
+
+    mtctx->params = params;
+    mtctx->frameContentSize = pledgedSrcSize;
+    if (dict) {
+        ZSTD_freeCDict(mtctx->cdictLocal);
+        mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                                    ZSTD_dlm_byCopy, dictContentType, /* note : a loadPrefix becomes an internal CDict */
+                                                    params.cParams, mtctx->cMem);
+        mtctx->cdict = mtctx->cdictLocal;
+        if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
+    } else {
+        ZSTD_freeCDict(mtctx->cdictLocal);
+        mtctx->cdictLocal = NULL;
+        mtctx->cdict = cdict;
+    }
+
+    mtctx->targetPrefixSize = (size_t)1 << ZSTDMT_computeOverlapLog(params);
+    DEBUGLOG(4, "overlapLog=%u => %u KB", params.overlapSizeLog, (U32)(mtctx->targetPrefixSize>>10));
+    mtctx->targetSectionSize = params.jobSize;
+    if (mtctx->targetSectionSize < ZSTDMT_JOBSIZE_MIN) mtctx->targetSectionSize = ZSTDMT_JOBSIZE_MIN;
+    if (mtctx->targetSectionSize < mtctx->targetPrefixSize) mtctx->targetSectionSize = mtctx->targetPrefixSize;  /* job size must be >= overlap size */
+    DEBUGLOG(4, "Job Size : %u KB (note : set to %u)", (U32)(mtctx->targetSectionSize>>10), params.jobSize);
+    DEBUGLOG(4, "inBuff Size : %u KB", (U32)(mtctx->targetSectionSize>>10));
+    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
+    {
+        /* If ldm is enabled we need windowSize space. */
+        size_t const windowSize = mtctx->params.ldmParams.enableLdm ? (1U << mtctx->params.cParams.windowLog) : 0;
+        /* Two buffers of slack, plus extra space for the overlap
+         * This is the minimum slack that LDM works with. One extra because
+         * flush might waste up to targetSectionSize-1 bytes. Another extra
+         * for the overlap (if > 0), then one to fill which doesn't overlap
+         * with the LDM window.
+         */
+        size_t const nbSlackBuffers = 2 + (mtctx->targetPrefixSize > 0);
+        size_t const slackSize = mtctx->targetSectionSize * nbSlackBuffers;
+        /* Compute the total size, and always have enough slack */
+        size_t const nbWorkers = MAX(mtctx->params.nbWorkers, 1);
+        size_t const sectionsSize = mtctx->targetSectionSize * nbWorkers;
+        size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
+        if (mtctx->roundBuff.capacity < capacity) {
+            if (mtctx->roundBuff.buffer)
+                ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
+            mtctx->roundBuff.buffer = (BYTE*)ZSTD_malloc(capacity, mtctx->cMem);
+            if (mtctx->roundBuff.buffer == NULL) {
+                mtctx->roundBuff.capacity = 0;
+                return ERROR(memory_allocation);
+            }
+            mtctx->roundBuff.capacity = capacity;
+        }
+    }
+    DEBUGLOG(4, "roundBuff capacity : %u KB", (U32)(mtctx->roundBuff.capacity>>10));
+    mtctx->roundBuff.pos = 0;
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->inBuff.filled = 0;
+    mtctx->inBuff.prefix = kNullRange;
+    mtctx->doneJobID = 0;
+    mtctx->nextJobID = 0;
+    mtctx->frameEnded = 0;
+    mtctx->allJobsCompleted = 0;
+    mtctx->consumed = 0;
+    mtctx->produced = 0;
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+        return ERROR(memory_allocation);
+    return 0;
+}
+
+size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params,
+                                   unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams = mtctx->params;  /* retrieve sticky params */
+    DEBUGLOG(4, "ZSTDMT_initCStream_advanced (pledgedSrcSize=%u)", (U32)pledgedSrcSize);
+    cctxParams.cParams = params.cParams;
+    cctxParams.fParams = params.fParams;
+    return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, ZSTD_dct_auto, NULL,
+                                       cctxParams, pledgedSrcSize);
+}
+
+size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
+                               const ZSTD_CDict* cdict,
+                                     ZSTD_frameParameters fParams,
+                                     unsigned long long pledgedSrcSize)
+{
+    ZSTD_CCtx_params cctxParams = mtctx->params;
+    if (cdict==NULL) return ERROR(dictionary_wrong);   /* method incompatible with NULL cdict */
+    cctxParams.cParams = ZSTD_getCParamsFromCDict(cdict);
+    cctxParams.fParams = fParams;
+    return ZSTDMT_initCStream_internal(mtctx, NULL, 0 /*dictSize*/, ZSTD_dct_auto, cdict,
+                                       cctxParams, pledgedSrcSize);
+}
+
+
+/* ZSTDMT_resetCStream() :
+ * pledgedSrcSize can be zero == unknown (for the time being)
+ * prefer using ZSTD_CONTENTSIZE_UNKNOWN,
+ * as `0` might mean "empty" in the future */
+size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize)
+{
+    if (!pledgedSrcSize) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, 0, mtctx->params,
+                                       pledgedSrcSize);
+}
+
+size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel) {
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    ZSTD_CCtx_params cctxParams = mtctx->params;   /* retrieve sticky params */
+    DEBUGLOG(4, "ZSTDMT_initCStream (cLevel=%i)", compressionLevel);
+    cctxParams.cParams = params.cParams;
+    cctxParams.fParams = params.fParams;
+    return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
+}
+
+
+/* ZSTDMT_writeLastEmptyBlock()
+ * Write a single empty block with an end-of-frame to finish a frame.
+ * Job must be created from streaming variant.
+ * This function is always successfull if expected conditions are fulfilled.
+ */
+static void ZSTDMT_writeLastEmptyBlock(ZSTDMT_jobDescription* job)
+{
+    assert(job->lastJob == 1);
+    assert(job->src.size == 0);   /* last job is empty -> will be simplified into a last empty block */
+    assert(job->firstJob == 0);   /* cannot be first job, as it also needs to create frame header */
+    assert(job->dstBuff.start == NULL);   /* invoked from streaming variant only (otherwise, dstBuff might be user's output) */
+    job->dstBuff = ZSTDMT_getBuffer(job->bufPool);
+    if (job->dstBuff.start == NULL) {
+      job->cSize = ERROR(memory_allocation);
+      return;
+    }
+    assert(job->dstBuff.capacity >= ZSTD_blockHeaderSize);   /* no buffer should ever be that small */
+    job->src = kNullRange;
+    job->cSize = ZSTD_writeLastEmptyBlock(job->dstBuff.start, job->dstBuff.capacity);
+    assert(!ZSTD_isError(job->cSize));
+    assert(job->consumed == 0);
+}
+
+static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZSTD_EndDirective endOp)
+{
+    unsigned const jobID = mtctx->nextJobID & mtctx->jobIDMask;
+    int const endFrame = (endOp == ZSTD_e_end);
+
+    if (mtctx->nextJobID > mtctx->doneJobID + mtctx->jobIDMask) {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: will not create new job : table is full");
+        assert((mtctx->nextJobID & mtctx->jobIDMask) == (mtctx->doneJobID & mtctx->jobIDMask));
+        return 0;
+    }
+
+    if (!mtctx->jobReady) {
+        BYTE const* src = (BYTE const*)mtctx->inBuff.buffer.start;
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: preparing job %u to compress %u bytes with %u preload ",
+                    mtctx->nextJobID, (U32)srcSize, (U32)mtctx->inBuff.prefix.size);
+        mtctx->jobs[jobID].src.start = src;
+        mtctx->jobs[jobID].src.size = srcSize;
+        assert(mtctx->inBuff.filled >= srcSize);
+        mtctx->jobs[jobID].prefix = mtctx->inBuff.prefix;
+        mtctx->jobs[jobID].consumed = 0;
+        mtctx->jobs[jobID].cSize = 0;
+        mtctx->jobs[jobID].params = mtctx->params;
+        mtctx->jobs[jobID].cdict = mtctx->nextJobID==0 ? mtctx->cdict : NULL;
+        mtctx->jobs[jobID].fullFrameSize = mtctx->frameContentSize;
+        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+        mtctx->jobs[jobID].cctxPool = mtctx->cctxPool;
+        mtctx->jobs[jobID].bufPool = mtctx->bufPool;
+        mtctx->jobs[jobID].seqPool = mtctx->seqPool;
+        mtctx->jobs[jobID].serial = &mtctx->serial;
+        mtctx->jobs[jobID].jobID = mtctx->nextJobID;
+        mtctx->jobs[jobID].firstJob = (mtctx->nextJobID==0);
+        mtctx->jobs[jobID].lastJob = endFrame;
+        mtctx->jobs[jobID].frameChecksumNeeded = endFrame && (mtctx->nextJobID>0) && mtctx->params.fParams.checksumFlag;
+        mtctx->jobs[jobID].dstFlushed = 0;
+
+        /* Update the round buffer pos and clear the input buffer to be reset */
+        mtctx->roundBuff.pos += srcSize;
+        mtctx->inBuff.buffer = g_nullBuffer;
+        mtctx->inBuff.filled = 0;
+        /* Set the prefix */
+        if (!endFrame) {
+            size_t const newPrefixSize = MIN(srcSize, mtctx->targetPrefixSize);
+            mtctx->inBuff.prefix.start = src + srcSize - newPrefixSize;
+            mtctx->inBuff.prefix.size = newPrefixSize;
+        } else {   /* endFrame==1 => no need for another input buffer */
+            mtctx->inBuff.prefix = kNullRange;
+            mtctx->frameEnded = endFrame;
+            if (mtctx->nextJobID == 0) {
+                /* single job exception : checksum is already calculated directly within worker thread */
+                mtctx->params.fParams.checksumFlag = 0;
+        }   }
+
+        if ( (srcSize == 0)
+          && (mtctx->nextJobID>0)/*single job must also write frame header*/ ) {
+            DEBUGLOG(5, "ZSTDMT_createCompressionJob: creating a last empty block to end frame");
+            assert(endOp == ZSTD_e_end);  /* only possible case : need to end the frame with an empty last block */
+            ZSTDMT_writeLastEmptyBlock(mtctx->jobs + jobID);
+            mtctx->nextJobID++;
+            return 0;
+        }
+    }
+
+    DEBUGLOG(5, "ZSTDMT_createCompressionJob: posting job %u : %u bytes  (end:%u, jobNb == %u (mod:%u))",
+                mtctx->nextJobID,
+                (U32)mtctx->jobs[jobID].src.size,
+                mtctx->jobs[jobID].lastJob,
+                mtctx->nextJobID,
+                jobID);
+    if (POOL_tryAdd(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[jobID])) {
+        mtctx->nextJobID++;
+        mtctx->jobReady = 0;
+    } else {
+        DEBUGLOG(5, "ZSTDMT_createCompressionJob: no worker available for job %u", mtctx->nextJobID);
+        mtctx->jobReady = 1;
+    }
+    return 0;
+}
+
+
+/*! ZSTDMT_flushProduced() :
+ * `output` : `pos` will be updated with amount of data flushed .
+ * `blockToFlush` : if >0, the function will block and wait if there is no data available to flush .
+ * @return : amount of data remaining within internal buffer, 0 if no more, 1 if unknown but > 0, or an error code */
+static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, unsigned blockToFlush, ZSTD_EndDirective end)
+{
+    unsigned const wJobID = mtctx->doneJobID & mtctx->jobIDMask;
+    DEBUGLOG(5, "ZSTDMT_flushProduced (blocking:%u , job %u <= %u)",
+                blockToFlush, mtctx->doneJobID, mtctx->nextJobID);
+    assert(output->size >= output->pos);
+
+    ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+    if (  blockToFlush
+      && (mtctx->doneJobID < mtctx->nextJobID) ) {
+        assert(mtctx->jobs[wJobID].dstFlushed <= mtctx->jobs[wJobID].cSize);
+        while (mtctx->jobs[wJobID].dstFlushed == mtctx->jobs[wJobID].cSize) {  /* nothing to flush */
+            if (mtctx->jobs[wJobID].consumed == mtctx->jobs[wJobID].src.size) {
+                DEBUGLOG(5, "job %u is completely consumed (%u == %u) => don't wait for cond, there will be none",
+                            mtctx->doneJobID, (U32)mtctx->jobs[wJobID].consumed, (U32)mtctx->jobs[wJobID].src.size);
+                break;
+            }
+            DEBUGLOG(5, "waiting for something to flush from job %u (currently flushed: %u bytes)",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+            ZSTD_pthread_cond_wait(&mtctx->jobs[wJobID].job_cond, &mtctx->jobs[wJobID].job_mutex);  /* block when nothing to flush but some to come */
+    }   }
+
+    /* try to flush something */
+    {   size_t cSize = mtctx->jobs[wJobID].cSize;                  /* shared */
+        size_t const srcConsumed = mtctx->jobs[wJobID].consumed;   /* shared */
+        size_t const srcSize = mtctx->jobs[wJobID].src.size;        /* read-only, could be done after mutex lock, but no-declaration-after-statement */
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+        if (ZSTD_isError(cSize)) {
+            DEBUGLOG(5, "ZSTDMT_flushProduced: job %u : compression error detected : %s",
+                        mtctx->doneJobID, ZSTD_getErrorName(cSize));
+            ZSTDMT_waitForAllJobsCompleted(mtctx);
+            ZSTDMT_releaseAllJobResources(mtctx);
+            return cSize;
+        }
+        /* add frame checksum if necessary (can only happen once) */
+        assert(srcConsumed <= srcSize);
+        if ( (srcConsumed == srcSize)   /* job completed -> worker no longer active */
+          && mtctx->jobs[wJobID].frameChecksumNeeded ) {
+            U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
+            DEBUGLOG(4, "ZSTDMT_flushProduced: writing checksum : %08X \n", checksum);
+            MEM_writeLE32((char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].cSize, checksum);
+            cSize += 4;
+            mtctx->jobs[wJobID].cSize += 4;  /* can write this shared value, as worker is no longer active */
+            mtctx->jobs[wJobID].frameChecksumNeeded = 0;
+        }
+        if (cSize > 0) {   /* compression is ongoing or completed */
+            size_t const toFlush = MIN(cSize - mtctx->jobs[wJobID].dstFlushed, output->size - output->pos);
+            DEBUGLOG(5, "ZSTDMT_flushProduced: Flushing %u bytes from job %u (completion:%u/%u, generated:%u)",
+                        (U32)toFlush, mtctx->doneJobID, (U32)srcConsumed, (U32)srcSize, (U32)cSize);
+            assert(mtctx->doneJobID < mtctx->nextJobID);
+            assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
+            assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
+            memcpy((char*)output->dst + output->pos,
+                   (const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
+                   toFlush);
+            output->pos += toFlush;
+            mtctx->jobs[wJobID].dstFlushed += toFlush;  /* can write : this value is only used by mtctx */
+
+            if ( (srcConsumed == srcSize)    /* job completed */
+              && (mtctx->jobs[wJobID].dstFlushed == cSize) ) {   /* output buffer fully flushed => free this job position */
+                DEBUGLOG(5, "Job %u completed (%u bytes), moving to next one",
+                        mtctx->doneJobID, (U32)mtctx->jobs[wJobID].dstFlushed);
+                ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[wJobID].dstBuff);
+                mtctx->jobs[wJobID].dstBuff = g_nullBuffer;
+                mtctx->jobs[wJobID].cSize = 0;   /* ensure this job slot is considered "not started" in future check */
+                mtctx->consumed += srcSize;
+                mtctx->produced += cSize;
+                mtctx->doneJobID++;
+        }   }
+
+        /* return value : how many bytes left in buffer ; fake it to 1 when unknown but >0 */
+        if (cSize > mtctx->jobs[wJobID].dstFlushed) return (cSize - mtctx->jobs[wJobID].dstFlushed);
+        if (srcSize > srcConsumed) return 1;   /* current job not completely compressed */
+    }
+    if (mtctx->doneJobID < mtctx->nextJobID) return 1;   /* some more jobs ongoing */
+    if (mtctx->jobReady) return 1;      /* one job is ready to push, just not yet in the list */
+    if (mtctx->inBuff.filled > 0) return 1;   /* input is not empty, and still needs to be converted into a job */
+    mtctx->allJobsCompleted = mtctx->frameEnded;   /* all jobs are entirely flushed => if this one is last one, frame is completed */
+    if (end == ZSTD_e_end) return !mtctx->frameEnded;  /* for ZSTD_e_end, question becomes : is frame completed ? instead of : are internal buffers fully flushed ? */
+    return 0;   /* internal buffers fully flushed */
+}
+
+/**
+ * Returns the range of data used by the earliest job that is not yet complete.
+ * If the data of the first job is broken up into two segments, we cover both
+ * sections.
+ */
+static range_t ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
+{
+    unsigned const firstJobID = mtctx->doneJobID;
+    unsigned const lastJobID = mtctx->nextJobID;
+    unsigned jobID;
+
+    for (jobID = firstJobID; jobID < lastJobID; ++jobID) {
+        unsigned const wJobID = jobID & mtctx->jobIDMask;
+        size_t consumed;
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[wJobID].job_mutex);
+        consumed = mtctx->jobs[wJobID].consumed;
+        ZSTD_pthread_mutex_unlock(&mtctx->jobs[wJobID].job_mutex);
+
+        if (consumed < mtctx->jobs[wJobID].src.size) {
+            range_t range = mtctx->jobs[wJobID].prefix;
+            if (range.size == 0) {
+                /* Empty prefix */
+                range = mtctx->jobs[wJobID].src;
+            }
+            /* Job source in multiple segments not supported yet */
+            assert(range.start <= mtctx->jobs[wJobID].src.start);
+            return range;
+        }
+    }
+    return kNullRange;
+}
+
+/**
+ * Returns non-zero iff buffer and range overlap.
+ */
+static int ZSTDMT_isOverlapped(buffer_t buffer, range_t range)
+{
+    BYTE const* const bufferStart = (BYTE const*)buffer.start;
+    BYTE const* const bufferEnd = bufferStart + buffer.capacity;
+    BYTE const* const rangeStart = (BYTE const*)range.start;
+    BYTE const* const rangeEnd = rangeStart + range.size;
+
+    if (rangeStart == NULL || bufferStart == NULL)
+        return 0;
+    /* Empty ranges cannot overlap */
+    if (bufferStart == bufferEnd || rangeStart == rangeEnd)
+        return 0;
+
+    return bufferStart < rangeEnd && rangeStart < bufferEnd;
+}
+
+static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
+{
+    range_t extDict;
+    range_t prefix;
+
+    extDict.start = window.dictBase + window.lowLimit;
+    extDict.size = window.dictLimit - window.lowLimit;
+
+    prefix.start = window.base + window.dictLimit;
+    prefix.size = window.nextSrc - (window.base + window.dictLimit);
+    DEBUGLOG(5, "extDict [0x%zx, 0x%zx)",
+                (size_t)extDict.start,
+                (size_t)extDict.start + extDict.size);
+    DEBUGLOG(5, "prefix  [0x%zx, 0x%zx)",
+                (size_t)prefix.start,
+                (size_t)prefix.start + prefix.size);
+
+    return ZSTDMT_isOverlapped(buffer, extDict)
+        || ZSTDMT_isOverlapped(buffer, prefix);
+}
+
+static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
+{
+    if (mtctx->params.ldmParams.enableLdm) {
+        ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
+        DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
+                    (size_t)buffer.start,
+                    (size_t)buffer.start + buffer.capacity);
+        ZSTD_PTHREAD_MUTEX_LOCK(mutex);
+        while (ZSTDMT_doesOverlapWindow(buffer, mtctx->serial.ldmWindow)) {
+            DEBUGLOG(6, "Waiting for LDM to finish...");
+            ZSTD_pthread_cond_wait(&mtctx->serial.ldmWindowCond, mutex);
+        }
+        DEBUGLOG(6, "Done waiting for LDM to finish");
+        ZSTD_pthread_mutex_unlock(mutex);
+    }
+}
+
+/**
+ * Attempts to set the inBuff to the next section to fill.
+ * If any part of the new section is still in use we give up.
+ * Returns non-zero if the buffer is filled.
+ */
+static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
+{
+    range_t const inUse = ZSTDMT_getInputDataInUse(mtctx);
+    size_t const spaceLeft = mtctx->roundBuff.capacity - mtctx->roundBuff.pos;
+    size_t const target = mtctx->targetSectionSize;
+    buffer_t buffer;
+
+    assert(mtctx->inBuff.buffer.start == NULL);
+    assert(mtctx->roundBuff.capacity >= target);
+
+    if (spaceLeft < target) {
+        /* ZSTD_invalidateRepCodes() doesn't work for extDict variants.
+         * Simply copy the prefix to the beginning in that case.
+         */
+        BYTE* const start = (BYTE*)mtctx->roundBuff.buffer;
+        size_t const prefixSize = mtctx->inBuff.prefix.size;
+
+        buffer.start = start;
+        buffer.capacity = prefixSize;
+        if (ZSTDMT_isOverlapped(buffer, inUse)) {
+            DEBUGLOG(6, "Waiting for buffer...");
+            return 0;
+        }
+        ZSTDMT_waitForLdmComplete(mtctx, buffer);
+        memmove(start, mtctx->inBuff.prefix.start, prefixSize);
+        mtctx->inBuff.prefix.start = start;
+        mtctx->roundBuff.pos = prefixSize;
+    }
+    buffer.start = mtctx->roundBuff.buffer + mtctx->roundBuff.pos;
+    buffer.capacity = target;
+
+    if (ZSTDMT_isOverlapped(buffer, inUse)) {
+        DEBUGLOG(6, "Waiting for buffer...");
+        return 0;
+    }
+    assert(!ZSTDMT_isOverlapped(buffer, mtctx->inBuff.prefix));
+
+    ZSTDMT_waitForLdmComplete(mtctx, buffer);
+
+    DEBUGLOG(5, "Using prefix range [%zx, %zx)",
+                (size_t)mtctx->inBuff.prefix.start,
+                (size_t)mtctx->inBuff.prefix.start + mtctx->inBuff.prefix.size);
+    DEBUGLOG(5, "Using source range [%zx, %zx)",
+                (size_t)buffer.start,
+                (size_t)buffer.start + buffer.capacity);
+
+
+    mtctx->inBuff.buffer = buffer;
+    mtctx->inBuff.filled = 0;
+    assert(mtctx->roundBuff.pos + buffer.capacity <= mtctx->roundBuff.capacity);
+    return 1;
+}
+
+
+/** ZSTDMT_compressStream_generic() :
+ *  internal use only - exposed to be invoked from zstd_compress.c
+ *  assumption : output and input are valid (pos <= size)
+ * @return : minimum amount of data remaining to flush, 0 if none */
+size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                     ZSTD_outBuffer* output,
+                                     ZSTD_inBuffer* input,
+                                     ZSTD_EndDirective endOp)
+{
+    unsigned forwardInputProgress = 0;
+    DEBUGLOG(5, "ZSTDMT_compressStream_generic (endOp=%u, srcSize=%u)",
+                (U32)endOp, (U32)(input->size - input->pos));
+    assert(output->pos <= output->size);
+    assert(input->pos  <= input->size);
+
+    if (mtctx->singleBlockingThread) {  /* delegate to single-thread (synchronous) */
+        return ZSTD_compressStream_generic(mtctx->cctxPool->cctx[0], output, input, endOp);
+    }
+
+    if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
+        /* current frame being ended. Only flush/end are allowed */
+        return ERROR(stage_wrong);
+    }
+
+    /* single-pass shortcut (note : synchronous-mode) */
+    if ( (mtctx->nextJobID == 0)      /* just started */
+      && (mtctx->inBuff.filled == 0)  /* nothing buffered */
+      && (!mtctx->jobReady)           /* no job already created */
+      && (endOp == ZSTD_e_end)        /* end order */
+      && (output->size - output->pos >= ZSTD_compressBound(input->size - input->pos)) ) { /* enough space in dst */
+        size_t const cSize = ZSTDMT_compress_advanced_internal(mtctx,
+                (char*)output->dst + output->pos, output->size - output->pos,
+                (const char*)input->src + input->pos, input->size - input->pos,
+                mtctx->cdict, mtctx->params);
+        if (ZSTD_isError(cSize)) return cSize;
+        input->pos = input->size;
+        output->pos += cSize;
+        mtctx->allJobsCompleted = 1;
+        mtctx->frameEnded = 1;
+        return 0;
+    }
+
+    /* fill input buffer */
+    if ( (!mtctx->jobReady)
+      && (input->size > input->pos) ) {   /* support NULL input */
+        if (mtctx->inBuff.buffer.start == NULL) {
+            assert(mtctx->inBuff.filled == 0); /* Can't fill an empty buffer */
+            if (!ZSTDMT_tryGetInputRange(mtctx)) {
+                /* It is only possible for this operation to fail if there are
+                 * still compression jobs ongoing.
+                 */
+                assert(mtctx->doneJobID != mtctx->nextJobID);
+            }
+        }
+        if (mtctx->inBuff.buffer.start != NULL) {
+            size_t const toLoad = MIN(input->size - input->pos, mtctx->targetSectionSize - mtctx->inBuff.filled);
+            assert(mtctx->inBuff.buffer.capacity >= mtctx->targetSectionSize);
+            DEBUGLOG(5, "ZSTDMT_compressStream_generic: adding %u bytes on top of %u to buffer of size %u",
+                        (U32)toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
+            memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad);
+            input->pos += toLoad;
+            mtctx->inBuff.filled += toLoad;
+            forwardInputProgress = toLoad>0;
+        }
+        if ((input->pos < input->size) && (endOp == ZSTD_e_end))
+            endOp = ZSTD_e_flush;   /* can't end now : not all input consumed */
+    }
+
+    if ( (mtctx->jobReady)
+      || (mtctx->inBuff.filled >= mtctx->targetSectionSize)  /* filled enough : let's compress */
+      || ((endOp != ZSTD_e_continue) && (mtctx->inBuff.filled > 0))  /* something to flush : let's go */
+      || ((endOp == ZSTD_e_end) && (!mtctx->frameEnded)) ) {   /* must finish the frame with a zero-size block */
+        size_t const jobSize = mtctx->inBuff.filled;
+        assert(mtctx->inBuff.filled <= mtctx->targetSectionSize);
+        CHECK_F( ZSTDMT_createCompressionJob(mtctx, jobSize, endOp) );
+    }
+
+    /* check for potential compressed data ready to be flushed */
+    {   size_t const remainingToFlush = ZSTDMT_flushProduced(mtctx, output, !forwardInputProgress, endOp); /* block if there was no forward input progress */
+        if (input->pos < input->size) return MAX(remainingToFlush, 1);  /* input not consumed : do not end flush yet */
+        return remainingToFlush;
+    }
+}
+
+
+size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    CHECK_F( ZSTDMT_compressStream_generic(mtctx, output, input, ZSTD_e_continue) );
+
+    /* recommended next input size : fill current input buffer */
+    return mtctx->targetSectionSize - mtctx->inBuff.filled;   /* note : could be zero when input buffer is fully filled and no more availability to create new job */
+}
+
+
+static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_EndDirective endFrame)
+{
+    size_t const srcSize = mtctx->inBuff.filled;
+    DEBUGLOG(5, "ZSTDMT_flushStream_internal");
+
+    if ( mtctx->jobReady     /* one job ready for a worker to pick up */
+      || (srcSize > 0)       /* still some data within input buffer */
+      || ((endFrame==ZSTD_e_end) && !mtctx->frameEnded)) {  /* need a last 0-size block to end frame */
+           DEBUGLOG(5, "ZSTDMT_flushStream_internal : create a new job (%u bytes, end:%u)",
+                        (U32)srcSize, (U32)endFrame);
+        CHECK_F( ZSTDMT_createCompressionJob(mtctx, srcSize, endFrame) );
+    }
+
+    /* check if there is any data available to flush */
+    return ZSTDMT_flushProduced(mtctx, output, 1 /* blockToFlush */, endFrame);
+}
+
+
+size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
+{
+    DEBUGLOG(5, "ZSTDMT_flushStream");
+    if (mtctx->singleBlockingThread)
+        return ZSTD_flushStream(mtctx->cctxPool->cctx[0], output);
+    return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_flush);
+}
+
+size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
+{
+    DEBUGLOG(4, "ZSTDMT_endStream");
+    if (mtctx->singleBlockingThread)
+        return ZSTD_endStream(mtctx->cctxPool->cctx[0], output);
+    return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_end);
+}
diff --git a/vendor/github.com/DataDog/zstd/zstdmt_compress.h b/vendor/github.com/DataDog/zstd/zstdmt_compress.h
new file mode 100755
index 0000000..f79e3b4
--- /dev/null
+++ b/vendor/github.com/DataDog/zstd/zstdmt_compress.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ #ifndef ZSTDMT_COMPRESS_H
+ #define ZSTDMT_COMPRESS_H
+
+ #if defined (__cplusplus)
+ extern "C" {
+ #endif
+
+
+/* Note : This is an internal API.
+ *        Some methods are still exposed (ZSTDLIB_API),
+ *        because it used to be the only way to invoke MT compression.
+ *        Now, it's recommended to use ZSTD_compress_generic() instead.
+ *        These methods will stop being exposed in a future version */
+
+/* ===   Dependencies   === */
+#include <stddef.h>                /* size_t */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters */
+#include "zstd.h"            /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
+
+
+/* ===   Memory management   === */
+typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
+ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers);
+ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers,
+                                                    ZSTD_customMem cMem);
+ZSTDLIB_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
+
+ZSTDLIB_API size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
+
+
+/* ===   Simple one-pass compression function   === */
+
+ZSTDLIB_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                       int compressionLevel);
+
+
+
+/* ===   Streaming functions   === */
+
+ZSTDLIB_API size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel);
+ZSTDLIB_API size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize);  /**< if srcSize is not known at reset time, use ZSTD_CONTENTSIZE_UNKNOWN. Note: for compatibility with older programs, 0 means the same as ZSTD_CONTENTSIZE_UNKNOWN, but it will change in the future to mean "empty" */
+
+ZSTDLIB_API size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+
+ZSTDLIB_API size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);   /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);     /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
+
+
+/* ===   Advanced functions and parameters  === */
+
+#ifndef ZSTDMT_JOBSIZE_MIN
+#  define ZSTDMT_JOBSIZE_MIN (1U << 20)   /* 1 MB - Minimum size of each compression job */
+#endif
+
+ZSTDLIB_API size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
+                                           void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     const ZSTD_CDict* cdict,
+                                           ZSTD_parameters params,
+                                           unsigned overlapLog);
+
+ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
+                                        const void* dict, size_t dictSize,   /* dict can be released after init, a local copy is preserved within zcs */
+                                        ZSTD_parameters params,
+                                        unsigned long long pledgedSrcSize);  /* pledgedSrcSize is optional and can be zero == unknown */
+
+ZSTDLIB_API size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
+                                        const ZSTD_CDict* cdict,
+                                        ZSTD_frameParameters fparams,
+                                        unsigned long long pledgedSrcSize);  /* note : zero means empty */
+
+/* ZSTDMT_parameter :
+ * List of parameters that can be set using ZSTDMT_setMTCtxParameter() */
+typedef enum {
+    ZSTDMT_p_jobSize,           /* Each job is compressed in parallel. By default, this value is dynamically determined depending on compression parameters. Can be set explicitly here. */
+    ZSTDMT_p_overlapSectionLog  /* Each job may reload a part of previous job to enhance compressionr ratio; 0 == no overlap, 6(default) == use 1/8th of window, >=9 == use full window. This is a "sticky" parameter : its value will be re-used on next compression job */
+} ZSTDMT_parameter;
+
+/* ZSTDMT_setMTCtxParameter() :
+ * allow setting individual parameters, one at a time, among a list of enums defined in ZSTDMT_parameter.
+ * The function must be called typically after ZSTD_createCCtx() but __before ZSTDMT_init*() !__
+ * Parameters not explicitly reset by ZSTDMT_init*() remain the same in consecutive compression sessions.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value);
+
+
+/*! ZSTDMT_compressStream_generic() :
+ *  Combines ZSTDMT_compressStream() with optional ZSTDMT_flushStream() or ZSTDMT_endStream()
+ *  depending on flush directive.
+ * @return : minimum amount of data still to be flushed
+ *           0 if fully flushed
+ *           or an error code
+ *  note : needs to be init using any ZSTD_initCStream*() variant */
+ZSTDLIB_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
+                                                ZSTD_outBuffer* output,
+                                                ZSTD_inBuffer* input,
+                                                ZSTD_EndDirective endOp);
+
+
+/* ========================================================
+ * ===  Private interface, for use by ZSTD_compress.c   ===
+ * ===  Not exposed in libzstd. Never invoke directly   ===
+ * ======================================================== */
+
+size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, unsigned value);
+
+/* ZSTDMT_CCtxParam_setNbWorkers()
+ * Set nbWorkers, and clamp it.
+ * Also reset jobSize and overlapLog */
+size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers);
+
+/*! ZSTDMT_updateCParams_whileCompressing() :
+ *  Updates only a selected set of compression parameters, to remain compatible with current frame.
+ *  New parameters will be applied to next compression job. */
+void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_params* cctxParams);
+
+/* ZSTDMT_getNbWorkers():
+ * @return nb threads currently active in mtctx.
+ * mtctx must be valid */
+unsigned ZSTDMT_getNbWorkers(const ZSTDMT_CCtx* mtctx);
+
+/* ZSTDMT_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads.
+ */
+ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx);
+
+
+/*! ZSTDMT_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  @return : 0, or an error code */
+size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
+                    const void* dict, size_t dictSize, ZSTD_dictContentType_e dictContentType,
+                    const ZSTD_CDict* cdict,
+                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTDMT_COMPRESS_H */