diff --git a/vendor/github.com/klauspost/compress/huff0/README.md b/vendor/github.com/klauspost/compress/huff0/README.md
index 0a8448c..8b6e5c6 100644
--- a/vendor/github.com/klauspost/compress/huff0/README.md
+++ b/vendor/github.com/klauspost/compress/huff0/README.md
@@ -12,11 +12,11 @@
 
 * [Godoc documentation](https://godoc.org/github.com/klauspost/compress/huff0)
 
-THIS PACKAGE IS NOT CONSIDERED STABLE AND API OR ENCODING MAY CHANGE IN THE FUTURE.
-
 ## News
 
- * Mar 2018: First implementation released. Consider this beta software for now.
+This is used as part of the [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and decompression package.
+
+This ensures that most functionality is well tested.
 
 # Usage
 
@@ -75,6 +75,8 @@
 Decompressing is done by calling the [`Decompress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress1X) 
 or [`Decompress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress4X) function.
 
+For concurrently decompressing content with a fixed table a stateless [`Decoder`](https://godoc.org/github.com/klauspost/compress/huff0#Decoder) can be requested which will remain correct as long as the scratch is unchanged. The capacity of the provided slice indicates the expected output size.
+
 You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back
 your input was likely corrupted. 
 
@@ -84,4 +86,4 @@
 # Contributing
 
 Contributions are always welcome. Be aware that adding public functions will require good justification and breaking 
-changes will likely not be accepted. If in doubt open an issue before writing the PR.
\ No newline at end of file
+changes will likely not be accepted. If in doubt open an issue before writing the PR.
diff --git a/vendor/github.com/klauspost/compress/huff0/bitreader.go b/vendor/github.com/klauspost/compress/huff0/bitreader.go
index 7d0903c..a4979e8 100644
--- a/vendor/github.com/klauspost/compress/huff0/bitreader.go
+++ b/vendor/github.com/klauspost/compress/huff0/bitreader.go
@@ -6,6 +6,7 @@
 package huff0
 
 import (
+	"encoding/binary"
 	"errors"
 	"io"
 )
@@ -34,29 +35,16 @@
 	}
 	b.bitsRead = 64
 	b.value = 0
-	b.fill()
-	b.fill()
+	if len(in) >= 8 {
+		b.fillFastStart()
+	} else {
+		b.fill()
+		b.fill()
+	}
 	b.bitsRead += 8 - uint8(highBit32(uint32(v)))
 	return nil
 }
 
-// getBits will return n bits. n can be 0.
-func (b *bitReader) getBits(n uint8) uint16 {
-	if n == 0 || b.bitsRead >= 64 {
-		return 0
-	}
-	return b.getBitsFast(n)
-}
-
-// getBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReader) getBitsFast(n uint8) uint16 {
-	const regMask = 64 - 1
-	v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
-	b.bitsRead += n
-	return v
-}
-
 // peekBitsFast requires that at least one bit is requested every time.
 // There are no checks if the buffer is filled.
 func (b *bitReader) peekBitsFast(n uint8) uint16 {
@@ -71,21 +59,36 @@
 	if b.bitsRead < 32 {
 		return
 	}
-	// Do single re-slice to avoid bounds checks.
+
+	// 2 bounds checks.
 	v := b.in[b.off-4 : b.off]
+	v = v[:4]
 	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
 	b.value = (b.value << 32) | uint64(low)
 	b.bitsRead -= 32
 	b.off -= 4
 }
 
+func (b *bitReader) advance(n uint8) {
+	b.bitsRead += n
+}
+
+// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read.
+func (b *bitReader) fillFastStart() {
+	// Do single re-slice to avoid bounds checks.
+	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
+	b.bitsRead = 0
+	b.off -= 8
+}
+
 // fill() will make sure at least 32 bits are available.
 func (b *bitReader) fill() {
 	if b.bitsRead < 32 {
 		return
 	}
 	if b.off > 4 {
-		v := b.in[b.off-4 : b.off]
+		v := b.in[b.off-4:]
+		v = v[:4]
 		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
 		b.value = (b.value << 32) | uint64(low)
 		b.bitsRead -= 32
@@ -113,3 +116,214 @@
 	}
 	return nil
 }
+
+// bitReader reads a bitstream in reverse.
+// The last set bit indicates the start of the stream and is used
+// for aligning the input.
+type bitReaderBytes struct {
+	in       []byte
+	off      uint // next byte to read is at in[off - 1]
+	value    uint64
+	bitsRead uint8
+}
+
+// init initializes and resets the bit reader.
+func (b *bitReaderBytes) init(in []byte) error {
+	if len(in) < 1 {
+		return errors.New("corrupt stream: too short")
+	}
+	b.in = in
+	b.off = uint(len(in))
+	// The highest bit of the last byte indicates where to start
+	v := in[len(in)-1]
+	if v == 0 {
+		return errors.New("corrupt stream, did not find end of stream")
+	}
+	b.bitsRead = 64
+	b.value = 0
+	if len(in) >= 8 {
+		b.fillFastStart()
+	} else {
+		b.fill()
+		b.fill()
+	}
+	b.advance(8 - uint8(highBit32(uint32(v))))
+	return nil
+}
+
+// peekBitsFast requires that at least one bit is requested every time.
+// There are no checks if the buffer is filled.
+func (b *bitReaderBytes) peekByteFast() uint8 {
+	got := uint8(b.value >> 56)
+	return got
+}
+
+func (b *bitReaderBytes) advance(n uint8) {
+	b.bitsRead += n
+	b.value <<= n & 63
+}
+
+// fillFast() will make sure at least 32 bits are available.
+// There must be at least 4 bytes available.
+func (b *bitReaderBytes) fillFast() {
+	if b.bitsRead < 32 {
+		return
+	}
+
+	// 2 bounds checks.
+	v := b.in[b.off-4 : b.off]
+	v = v[:4]
+	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+	b.value |= uint64(low) << (b.bitsRead - 32)
+	b.bitsRead -= 32
+	b.off -= 4
+}
+
+// fillFastStart() assumes the bitReaderBytes is empty and there is at least 8 bytes to read.
+func (b *bitReaderBytes) fillFastStart() {
+	// Do single re-slice to avoid bounds checks.
+	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
+	b.bitsRead = 0
+	b.off -= 8
+}
+
+// fill() will make sure at least 32 bits are available.
+func (b *bitReaderBytes) fill() {
+	if b.bitsRead < 32 {
+		return
+	}
+	if b.off > 4 {
+		v := b.in[b.off-4:]
+		v = v[:4]
+		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+		b.value |= uint64(low) << (b.bitsRead - 32)
+		b.bitsRead -= 32
+		b.off -= 4
+		return
+	}
+	for b.off > 0 {
+		b.value |= uint64(b.in[b.off-1]) << (b.bitsRead - 8)
+		b.bitsRead -= 8
+		b.off--
+	}
+}
+
+// finished returns true if all bits have been read from the bit stream.
+func (b *bitReaderBytes) finished() bool {
+	return b.off == 0 && b.bitsRead >= 64
+}
+
+// close the bitstream and returns an error if out-of-buffer reads occurred.
+func (b *bitReaderBytes) close() error {
+	// Release reference.
+	b.in = nil
+	if b.bitsRead > 64 {
+		return io.ErrUnexpectedEOF
+	}
+	return nil
+}
+
+// bitReaderShifted reads a bitstream in reverse.
+// The last set bit indicates the start of the stream and is used
+// for aligning the input.
+type bitReaderShifted struct {
+	in       []byte
+	off      uint // next byte to read is at in[off - 1]
+	value    uint64
+	bitsRead uint8
+}
+
+// init initializes and resets the bit reader.
+func (b *bitReaderShifted) init(in []byte) error {
+	if len(in) < 1 {
+		return errors.New("corrupt stream: too short")
+	}
+	b.in = in
+	b.off = uint(len(in))
+	// The highest bit of the last byte indicates where to start
+	v := in[len(in)-1]
+	if v == 0 {
+		return errors.New("corrupt stream, did not find end of stream")
+	}
+	b.bitsRead = 64
+	b.value = 0
+	if len(in) >= 8 {
+		b.fillFastStart()
+	} else {
+		b.fill()
+		b.fill()
+	}
+	b.advance(8 - uint8(highBit32(uint32(v))))
+	return nil
+}
+
+// peekBitsFast requires that at least one bit is requested every time.
+// There are no checks if the buffer is filled.
+func (b *bitReaderShifted) peekBitsFast(n uint8) uint16 {
+	return uint16(b.value >> ((64 - n) & 63))
+}
+
+func (b *bitReaderShifted) advance(n uint8) {
+	b.bitsRead += n
+	b.value <<= n & 63
+}
+
+// fillFast() will make sure at least 32 bits are available.
+// There must be at least 4 bytes available.
+func (b *bitReaderShifted) fillFast() {
+	if b.bitsRead < 32 {
+		return
+	}
+
+	// 2 bounds checks.
+	v := b.in[b.off-4 : b.off]
+	v = v[:4]
+	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+	b.value |= uint64(low) << ((b.bitsRead - 32) & 63)
+	b.bitsRead -= 32
+	b.off -= 4
+}
+
+// fillFastStart() assumes the bitReaderShifted is empty and there is at least 8 bytes to read.
+func (b *bitReaderShifted) fillFastStart() {
+	// Do single re-slice to avoid bounds checks.
+	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
+	b.bitsRead = 0
+	b.off -= 8
+}
+
+// fill() will make sure at least 32 bits are available.
+func (b *bitReaderShifted) fill() {
+	if b.bitsRead < 32 {
+		return
+	}
+	if b.off > 4 {
+		v := b.in[b.off-4:]
+		v = v[:4]
+		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+		b.value |= uint64(low) << ((b.bitsRead - 32) & 63)
+		b.bitsRead -= 32
+		b.off -= 4
+		return
+	}
+	for b.off > 0 {
+		b.value |= uint64(b.in[b.off-1]) << ((b.bitsRead - 8) & 63)
+		b.bitsRead -= 8
+		b.off--
+	}
+}
+
+// finished returns true if all bits have been read from the bit stream.
+func (b *bitReaderShifted) finished() bool {
+	return b.off == 0 && b.bitsRead >= 64
+}
+
+// close the bitstream and returns an error if out-of-buffer reads occurred.
+func (b *bitReaderShifted) close() error {
+	// Release reference.
+	b.in = nil
+	if b.bitsRead > 64 {
+		return io.ErrUnexpectedEOF
+	}
+	return nil
+}
diff --git a/vendor/github.com/klauspost/compress/huff0/bitwriter.go b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
index bda4021..6bce4e8 100644
--- a/vendor/github.com/klauspost/compress/huff0/bitwriter.go
+++ b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
@@ -43,6 +43,11 @@
 func (b *bitWriter) encSymbol(ct cTable, symbol byte) {
 	enc := ct[symbol]
 	b.bitContainer |= uint64(enc.val) << (b.nBits & 63)
+	if false {
+		if enc.nBits == 0 {
+			panic("nbits 0")
+		}
+	}
 	b.nBits += enc.nBits
 }
 
@@ -54,6 +59,14 @@
 	sh := b.nBits & 63
 	combined := uint64(encA.val) | (uint64(encB.val) << (encA.nBits & 63))
 	b.bitContainer |= combined << sh
+	if false {
+		if encA.nBits == 0 {
+			panic("nbitsA 0")
+		}
+		if encB.nBits == 0 {
+			panic("nbitsB 0")
+		}
+	}
 	b.nBits += encA.nBits + encB.nBits
 }
 
diff --git a/vendor/github.com/klauspost/compress/huff0/compress.go b/vendor/github.com/klauspost/compress/huff0/compress.go
index 0843cb0..0823c92 100644
--- a/vendor/github.com/klauspost/compress/huff0/compress.go
+++ b/vendor/github.com/klauspost/compress/huff0/compress.go
@@ -77,8 +77,11 @@
 		// Each symbol present maximum once or too well distributed.
 		return nil, false, ErrIncompressible
 	}
-
-	if s.Reuse == ReusePolicyPrefer && canReuse {
+	if s.Reuse == ReusePolicyMust && !canReuse {
+		// We must reuse, but we can't.
+		return nil, false, ErrIncompressible
+	}
+	if (s.Reuse == ReusePolicyPrefer || s.Reuse == ReusePolicyMust) && canReuse {
 		keepTable := s.cTable
 		keepTL := s.actualTableLog
 		s.cTable = s.prevTable
@@ -90,6 +93,9 @@
 			s.OutData = s.Out
 			return s.Out, true, nil
 		}
+		if s.Reuse == ReusePolicyMust {
+			return nil, false, ErrIncompressible
+		}
 		// Do not attempt to re-use later.
 		s.prevTable = s.prevTable[:0]
 	}
@@ -397,7 +403,7 @@
 	var startNode = int16(s.symbolLen)
 	nonNullRank := s.symbolLen - 1
 
-	nodeNb := int16(startNode)
+	nodeNb := startNode
 	huffNode := s.nodes[1 : huffNodesLen+1]
 
 	// This overlays the slice above, but allows "-1" index lookups.
@@ -530,7 +536,6 @@
 		}
 		nodes[pos&huffNodesMask] = nodeElt{count: c, symbol: byte(n)}
 	}
-	return
 }
 
 func (s *Scratch) setMaxHeight(lastNonNull int) uint8 {
@@ -574,7 +579,7 @@
 
 		// Get pos of last (smallest) symbol per rank
 		{
-			currentNbBits := uint8(maxNbBits)
+			currentNbBits := maxNbBits
 			for pos := int(n); pos >= 0; pos-- {
 				if huffNode[pos].nbBits >= currentNbBits {
 					continue
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress.go b/vendor/github.com/klauspost/compress/huff0/decompress.go
index 97ae66a..41703bb 100644
--- a/vendor/github.com/klauspost/compress/huff0/decompress.go
+++ b/vendor/github.com/klauspost/compress/huff0/decompress.go
@@ -25,11 +25,14 @@
 	len   uint8
 }
 
+// Uses special code for all tables that are < 8 bits.
+const use8BitTables = true
+
 // ReadTable will read a table from the input.
 // The size of the input may be larger than the table definition.
 // Any content remaining after the table definition will be returned.
 // If no Scratch is provided a new one is allocated.
-// The returned Scratch can be used for decoding input using this table.
+// The returned Scratch can be used for encoding or decoding input using this table.
 func ReadTable(in []byte, s *Scratch) (s2 *Scratch, remain []byte, err error) {
 	s, err = s.prepare(in)
 	if err != nil {
@@ -55,8 +58,8 @@
 		s.symbolLen = uint16(oSize)
 		in = in[iSize:]
 	} else {
-		if len(in) <= int(iSize) {
-			return s, nil, errors.New("input too small for table")
+		if len(in) < int(iSize) {
+			return s, nil, fmt.Errorf("input too small for table, want %d bytes, have %d", iSize, len(in))
 		}
 		// FSE compressed weights
 		s.fse.DecompressLimit = 255
@@ -83,6 +86,7 @@
 		}
 		v2 := v & 15
 		rankStats[v2]++
+		// (1 << (v2-1)) is slower since the compiler cannot prove that v2 isn't 0.
 		weightTotal += (1 << v2) >> 1
 	}
 	if weightTotal == 0 {
@@ -134,20 +138,40 @@
 	if len(s.dt.single) != tSize {
 		s.dt.single = make([]dEntrySingle, tSize)
 	}
+	cTable := s.prevTable
+	if cap(cTable) < maxSymbolValue+1 {
+		cTable = make([]cTableEntry, 0, maxSymbolValue+1)
+	}
+	cTable = cTable[:maxSymbolValue+1]
+	s.prevTable = cTable[:s.symbolLen]
+	s.prevTableLog = s.actualTableLog
+
 	for n, w := range s.huffWeight[:s.symbolLen] {
 		if w == 0 {
+			cTable[n] = cTableEntry{
+				val:   0,
+				nBits: 0,
+			}
 			continue
 		}
 		length := (uint32(1) << w) >> 1
 		d := dEntrySingle{
 			entry: uint16(s.actualTableLog+1-w) | (uint16(n) << 8),
 		}
-		single := s.dt.single[rankStats[w] : rankStats[w]+length]
+
+		rank := &rankStats[w]
+		cTable[n] = cTableEntry{
+			val:   uint16(*rank >> (w - 1)),
+			nBits: uint8(d.entry),
+		}
+
+		single := s.dt.single[*rank : *rank+length]
 		for i := range single {
 			single[i] = d
 		}
-		rankStats[w] += length
+		*rank += length
 	}
+
 	return s, in, nil
 }
 
@@ -155,69 +179,14 @@
 // The length of the supplied input must match the end of a block exactly.
 // Before this is called, the table must be initialized with ReadTable unless
 // the encoder re-used the table.
+// deprecated: Use the stateless Decoder() to get a concurrent version.
 func (s *Scratch) Decompress1X(in []byte) (out []byte, err error) {
-	if len(s.dt.single) == 0 {
-		return nil, errors.New("no table loaded")
+	if cap(s.Out) < s.MaxDecodedSize {
+		s.Out = make([]byte, s.MaxDecodedSize)
 	}
-	var br bitReader
-	err = br.init(in)
-	if err != nil {
-		return nil, err
-	}
-	s.Out = s.Out[:0]
-
-	decode := func() byte {
-		val := br.peekBitsFast(s.actualTableLog) /* note : actualTableLog >= 1 */
-		v := s.dt.single[val]
-		br.bitsRead += uint8(v.entry)
-		return uint8(v.entry >> 8)
-	}
-	hasDec := func(v dEntrySingle) byte {
-		br.bitsRead += uint8(v.entry)
-		return uint8(v.entry >> 8)
-	}
-
-	// Avoid bounds check by always having full sized table.
-	const tlSize = 1 << tableLogMax
-	const tlMask = tlSize - 1
-	dt := s.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var tmp = s.huffWeight[:256]
-	var off uint8
-
-	for br.off >= 8 {
-		br.fillFast()
-		tmp[off+0] = hasDec(dt[br.peekBitsFast(s.actualTableLog)&tlMask])
-		tmp[off+1] = hasDec(dt[br.peekBitsFast(s.actualTableLog)&tlMask])
-		br.fillFast()
-		tmp[off+2] = hasDec(dt[br.peekBitsFast(s.actualTableLog)&tlMask])
-		tmp[off+3] = hasDec(dt[br.peekBitsFast(s.actualTableLog)&tlMask])
-		off += 4
-		if off == 0 {
-			if len(s.Out)+256 > s.MaxDecodedSize {
-				br.close()
-				return nil, ErrMaxDecodedSizeExceeded
-			}
-			s.Out = append(s.Out, tmp...)
-		}
-	}
-
-	if len(s.Out)+int(off) > s.MaxDecodedSize {
-		br.close()
-		return nil, ErrMaxDecodedSizeExceeded
-	}
-	s.Out = append(s.Out, tmp[:off]...)
-
-	for !br.finished() {
-		br.fill()
-		if len(s.Out) >= s.MaxDecodedSize {
-			br.close()
-			return nil, ErrMaxDecodedSizeExceeded
-		}
-		s.Out = append(s.Out, decode())
-	}
-	return s.Out, br.close()
+	s.Out = s.Out[:0:s.MaxDecodedSize]
+	s.Out, err = s.Decoder().Decompress1X(s.Out, in)
+	return s.Out, err
 }
 
 // Decompress4X will decompress a 4X encoded stream.
@@ -225,123 +194,402 @@
 // the encoder re-used the table.
 // The length of the supplied input must match the end of a block exactly.
 // The destination size of the uncompressed data must be known and provided.
+// deprecated: Use the stateless Decoder() to get a concurrent version.
 func (s *Scratch) Decompress4X(in []byte, dstSize int) (out []byte, err error) {
-	if len(s.dt.single) == 0 {
-		return nil, errors.New("no table loaded")
-	}
-	if len(in) < 6+(4*1) {
-		return nil, errors.New("input too small")
-	}
 	if dstSize > s.MaxDecodedSize {
 		return nil, ErrMaxDecodedSizeExceeded
 	}
-	// TODO: We do not detect when we overrun a buffer, except if the last one does.
+	if cap(s.Out) < dstSize {
+		s.Out = make([]byte, s.MaxDecodedSize)
+	}
+	s.Out = s.Out[:0:dstSize]
+	s.Out, err = s.Decoder().Decompress4X(s.Out, in)
+	return s.Out, err
+}
 
-	var br [4]bitReader
+// Decoder will return a stateless decoder that can be used by multiple
+// decompressors concurrently.
+// Before this is called, the table must be initialized with ReadTable.
+// The Decoder is still linked to the scratch buffer so that cannot be reused.
+// However, it is safe to discard the scratch.
+func (s *Scratch) Decoder() *Decoder {
+	return &Decoder{
+		dt:             s.dt,
+		actualTableLog: s.actualTableLog,
+	}
+}
+
+// Decoder provides stateless decoding.
+type Decoder struct {
+	dt             dTable
+	actualTableLog uint8
+}
+
+// Decompress1X will decompress a 1X encoded stream.
+// The cap of the output buffer will be the maximum decompressed size.
+// The length of the supplied input must match the end of a block exactly.
+func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
+	if len(d.dt.single) == 0 {
+		return nil, errors.New("no table loaded")
+	}
+	if use8BitTables && d.actualTableLog <= 8 {
+		return d.decompress1X8Bit(dst, src)
+	}
+	var br bitReaderShifted
+	err := br.init(src)
+	if err != nil {
+		return dst, err
+	}
+	maxDecodedSize := cap(dst)
+	dst = dst[:0]
+
+	// Avoid bounds check by always having full sized table.
+	const tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+	dt := d.dt.single[:tlSize]
+
+	// Use temp table to avoid bound checks/append penalty.
+	var buf [256]byte
+	var off uint8
+
+	for br.off >= 8 {
+		br.fillFast()
+		v := dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+0] = uint8(v.entry >> 8)
+
+		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+1] = uint8(v.entry >> 8)
+
+		// Refill
+		br.fillFast()
+
+		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+2] = uint8(v.entry >> 8)
+
+		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
+		br.advance(uint8(v.entry))
+		buf[off+3] = uint8(v.entry >> 8)
+
+		off += 4
+		if off == 0 {
+			if len(dst)+256 > maxDecodedSize {
+				br.close()
+				return nil, ErrMaxDecodedSizeExceeded
+			}
+			dst = append(dst, buf[:]...)
+		}
+	}
+
+	if len(dst)+int(off) > maxDecodedSize {
+		br.close()
+		return nil, ErrMaxDecodedSizeExceeded
+	}
+	dst = append(dst, buf[:off]...)
+
+	// br < 8, so uint8 is fine
+	bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead
+	for bitsLeft > 0 {
+		br.fill()
+		if false && br.bitsRead >= 32 {
+			if br.off >= 4 {
+				v := br.in[br.off-4:]
+				v = v[:4]
+				low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+				br.value = (br.value << 32) | uint64(low)
+				br.bitsRead -= 32
+				br.off -= 4
+			} else {
+				for br.off > 0 {
+					br.value = (br.value << 8) | uint64(br.in[br.off-1])
+					br.bitsRead -= 8
+					br.off--
+				}
+			}
+		}
+		if len(dst) >= maxDecodedSize {
+			br.close()
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+		v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask]
+		nBits := uint8(v.entry)
+		br.advance(nBits)
+		bitsLeft -= nBits
+		dst = append(dst, uint8(v.entry>>8))
+	}
+	return dst, br.close()
+}
+
+// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8.
+// The cap of the output buffer will be the maximum decompressed size.
+// The length of the supplied input must match the end of a block exactly.
+func (d *Decoder) decompress1X8Bit(dst, src []byte) ([]byte, error) {
+	if d.actualTableLog == 8 {
+		return d.decompress1X8BitExactly(dst, src)
+	}
+	var br bitReaderBytes
+	err := br.init(src)
+	if err != nil {
+		return dst, err
+	}
+	maxDecodedSize := cap(dst)
+	dst = dst[:0]
+
+	// Avoid bounds check by always having full sized table.
+	dt := d.dt.single[:256]
+
+	// Use temp table to avoid bound checks/append penalty.
+	var buf [256]byte
+	var off uint8
+
+	shift := (8 - d.actualTableLog) & 7
+
+	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
+	for br.off >= 4 {
+		br.fillFast()
+		v := dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+0] = uint8(v.entry >> 8)
+
+		v = dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+1] = uint8(v.entry >> 8)
+
+		v = dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+2] = uint8(v.entry >> 8)
+
+		v = dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+3] = uint8(v.entry >> 8)
+
+		off += 4
+		if off == 0 {
+			if len(dst)+256 > maxDecodedSize {
+				br.close()
+				return nil, ErrMaxDecodedSizeExceeded
+			}
+			dst = append(dst, buf[:]...)
+		}
+	}
+
+	if len(dst)+int(off) > maxDecodedSize {
+		br.close()
+		return nil, ErrMaxDecodedSizeExceeded
+	}
+	dst = append(dst, buf[:off]...)
+
+	// br < 4, so uint8 is fine
+	bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead))
+	for bitsLeft > 0 {
+		if br.bitsRead >= 64-8 {
+			for br.off > 0 {
+				br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
+				br.bitsRead -= 8
+				br.off--
+			}
+		}
+		if len(dst) >= maxDecodedSize {
+			br.close()
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+		v := dt[br.peekByteFast()>>shift]
+		nBits := uint8(v.entry)
+		br.advance(nBits)
+		bitsLeft -= int8(nBits)
+		dst = append(dst, uint8(v.entry>>8))
+	}
+	return dst, br.close()
+}
+
+// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8.
+// The cap of the output buffer will be the maximum decompressed size.
+// The length of the supplied input must match the end of a block exactly.
+func (d *Decoder) decompress1X8BitExactly(dst, src []byte) ([]byte, error) {
+	var br bitReaderBytes
+	err := br.init(src)
+	if err != nil {
+		return dst, err
+	}
+	maxDecodedSize := cap(dst)
+	dst = dst[:0]
+
+	// Avoid bounds check by always having full sized table.
+	dt := d.dt.single[:256]
+
+	// Use temp table to avoid bound checks/append penalty.
+	var buf [256]byte
+	var off uint8
+
+	const shift = 0
+
+	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
+	for br.off >= 4 {
+		br.fillFast()
+		v := dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+0] = uint8(v.entry >> 8)
+
+		v = dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+1] = uint8(v.entry >> 8)
+
+		v = dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+2] = uint8(v.entry >> 8)
+
+		v = dt[br.peekByteFast()>>shift]
+		br.advance(uint8(v.entry))
+		buf[off+3] = uint8(v.entry >> 8)
+
+		off += 4
+		if off == 0 {
+			if len(dst)+256 > maxDecodedSize {
+				br.close()
+				return nil, ErrMaxDecodedSizeExceeded
+			}
+			dst = append(dst, buf[:]...)
+		}
+	}
+
+	if len(dst)+int(off) > maxDecodedSize {
+		br.close()
+		return nil, ErrMaxDecodedSizeExceeded
+	}
+	dst = append(dst, buf[:off]...)
+
+	// br < 4, so uint8 is fine
+	bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead))
+	for bitsLeft > 0 {
+		if br.bitsRead >= 64-8 {
+			for br.off > 0 {
+				br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
+				br.bitsRead -= 8
+				br.off--
+			}
+		}
+		if len(dst) >= maxDecodedSize {
+			br.close()
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+		v := dt[br.peekByteFast()>>shift]
+		nBits := uint8(v.entry)
+		br.advance(nBits)
+		bitsLeft -= int8(nBits)
+		dst = append(dst, uint8(v.entry>>8))
+	}
+	return dst, br.close()
+}
+
+// Decompress4X will decompress a 4X encoded stream.
+// The length of the supplied input must match the end of a block exactly.
+// The *capacity* of the dst slice must match the destination size of
+// the uncompressed data exactly.
+func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
+	if len(d.dt.single) == 0 {
+		return nil, errors.New("no table loaded")
+	}
+	if len(src) < 6+(4*1) {
+		return nil, errors.New("input too small")
+	}
+	if use8BitTables && d.actualTableLog <= 8 {
+		return d.decompress4X8bit(dst, src)
+	}
+
+	var br [4]bitReaderShifted
 	start := 6
 	for i := 0; i < 3; i++ {
-		length := int(in[i*2]) | (int(in[i*2+1]) << 8)
-		if start+length >= len(in) {
+		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
+		if start+length >= len(src) {
 			return nil, errors.New("truncated input (or invalid offset)")
 		}
-		err = br[i].init(in[start : start+length])
+		err := br[i].init(src[start : start+length])
 		if err != nil {
 			return nil, err
 		}
 		start += length
 	}
-	err = br[3].init(in[start:])
+	err := br[3].init(src[start:])
 	if err != nil {
 		return nil, err
 	}
 
-	// Prepare output
-	if cap(s.Out) < dstSize {
-		s.Out = make([]byte, 0, dstSize)
-	}
-	s.Out = s.Out[:dstSize]
 	// destination, offset to match first output
-	dstOut := s.Out
+	dstSize := cap(dst)
+	dst = dst[:dstSize]
+	out := dst
 	dstEvery := (dstSize + 3) / 4
 
 	const tlSize = 1 << tableLogMax
 	const tlMask = tlSize - 1
-	single := s.dt.single[:tlSize]
-
-	decode := func(br *bitReader) byte {
-		val := br.peekBitsFast(s.actualTableLog) /* note : actualTableLog >= 1 */
-		v := single[val&tlMask]
-		br.bitsRead += uint8(v.entry)
-		return uint8(v.entry >> 8)
-	}
+	single := d.dt.single[:tlSize]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var tmp = s.huffWeight[:256]
+	var buf [256]byte
 	var off uint8
 	var decoded int
 
 	// Decode 2 values from each decoder/loop.
 	const bufoff = 256 / 4
-bigloop:
 	for {
-		for i := range br {
-			br := &br[i]
-			if br.off < 4 {
-				break bigloop
-			}
-			br.fillFast()
+		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
+			break
 		}
 
 		{
 			const stream = 0
-			val := br[stream].peekBitsFast(s.actualTableLog)
+			const stream2 = 1
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			val := br[stream].peekBitsFast(d.actualTableLog)
 			v := single[val&tlMask]
-			br[stream].bitsRead += uint8(v.entry)
+			br[stream].advance(uint8(v.entry))
+			buf[off+bufoff*stream] = uint8(v.entry >> 8)
 
-			val2 := br[stream].peekBitsFast(s.actualTableLog)
+			val2 := br[stream2].peekBitsFast(d.actualTableLog)
 			v2 := single[val2&tlMask]
-			tmp[off+bufoff*stream+1] = uint8(v2.entry >> 8)
-			tmp[off+bufoff*stream] = uint8(v.entry >> 8)
-			br[stream].bitsRead += uint8(v2.entry)
-		}
+			br[stream2].advance(uint8(v2.entry))
+			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
 
-		{
-			const stream = 1
-			val := br[stream].peekBitsFast(s.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].bitsRead += uint8(v.entry)
+			val = br[stream].peekBitsFast(d.actualTableLog)
+			v = single[val&tlMask]
+			br[stream].advance(uint8(v.entry))
+			buf[off+bufoff*stream+1] = uint8(v.entry >> 8)
 
-			val2 := br[stream].peekBitsFast(s.actualTableLog)
-			v2 := single[val2&tlMask]
-			tmp[off+bufoff*stream+1] = uint8(v2.entry >> 8)
-			tmp[off+bufoff*stream] = uint8(v.entry >> 8)
-			br[stream].bitsRead += uint8(v2.entry)
+			val2 = br[stream2].peekBitsFast(d.actualTableLog)
+			v2 = single[val2&tlMask]
+			br[stream2].advance(uint8(v2.entry))
+			buf[off+bufoff*stream2+1] = uint8(v2.entry >> 8)
 		}
 
 		{
 			const stream = 2
-			val := br[stream].peekBitsFast(s.actualTableLog)
+			const stream2 = 3
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			val := br[stream].peekBitsFast(d.actualTableLog)
 			v := single[val&tlMask]
-			br[stream].bitsRead += uint8(v.entry)
+			br[stream].advance(uint8(v.entry))
+			buf[off+bufoff*stream] = uint8(v.entry >> 8)
 
-			val2 := br[stream].peekBitsFast(s.actualTableLog)
+			val2 := br[stream2].peekBitsFast(d.actualTableLog)
 			v2 := single[val2&tlMask]
-			tmp[off+bufoff*stream+1] = uint8(v2.entry >> 8)
-			tmp[off+bufoff*stream] = uint8(v.entry >> 8)
-			br[stream].bitsRead += uint8(v2.entry)
-		}
+			br[stream2].advance(uint8(v2.entry))
+			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
 
-		{
-			const stream = 3
-			val := br[stream].peekBitsFast(s.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].bitsRead += uint8(v.entry)
+			val = br[stream].peekBitsFast(d.actualTableLog)
+			v = single[val&tlMask]
+			br[stream].advance(uint8(v.entry))
+			buf[off+bufoff*stream+1] = uint8(v.entry >> 8)
 
-			val2 := br[stream].peekBitsFast(s.actualTableLog)
-			v2 := single[val2&tlMask]
-			tmp[off+bufoff*stream+1] = uint8(v2.entry >> 8)
-			tmp[off+bufoff*stream] = uint8(v.entry >> 8)
-			br[stream].bitsRead += uint8(v2.entry)
+			val2 = br[stream2].peekBitsFast(d.actualTableLog)
+			v2 = single[val2&tlMask]
+			br[stream2].advance(uint8(v2.entry))
+			buf[off+bufoff*stream2+1] = uint8(v2.entry >> 8)
 		}
 
 		off += 2
@@ -350,42 +598,67 @@
 			if bufoff > dstEvery {
 				return nil, errors.New("corruption detected: stream overrun 1")
 			}
-			copy(dstOut, tmp[:bufoff])
-			copy(dstOut[dstEvery:], tmp[bufoff:bufoff*2])
-			copy(dstOut[dstEvery*2:], tmp[bufoff*2:bufoff*3])
-			copy(dstOut[dstEvery*3:], tmp[bufoff*3:bufoff*4])
+			copy(out, buf[:bufoff])
+			copy(out[dstEvery:], buf[bufoff:bufoff*2])
+			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
+			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
 			off = 0
-			dstOut = dstOut[bufoff:]
+			out = out[bufoff:]
 			decoded += 256
 			// There must at least be 3 buffers left.
-			if len(dstOut) < dstEvery*3 {
+			if len(out) < dstEvery*3 {
 				return nil, errors.New("corruption detected: stream overrun 2")
 			}
 		}
 	}
 	if off > 0 {
 		ioff := int(off)
-		if len(dstOut) < dstEvery*3+ioff {
+		if len(out) < dstEvery*3+ioff {
 			return nil, errors.New("corruption detected: stream overrun 3")
 		}
-		copy(dstOut, tmp[:off])
-		copy(dstOut[dstEvery:dstEvery+ioff], tmp[bufoff:bufoff*2])
-		copy(dstOut[dstEvery*2:dstEvery*2+ioff], tmp[bufoff*2:bufoff*3])
-		copy(dstOut[dstEvery*3:dstEvery*3+ioff], tmp[bufoff*3:bufoff*4])
+		copy(out, buf[:off])
+		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
+		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
+		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
 		decoded += int(off) * 4
-		dstOut = dstOut[off:]
+		out = out[off:]
 	}
 
 	// Decode remaining.
 	for i := range br {
 		offset := dstEvery * i
 		br := &br[i]
-		for !br.finished() {
+		bitsLeft := br.off*8 + uint(64-br.bitsRead)
+		for bitsLeft > 0 {
 			br.fill()
-			if offset >= len(dstOut) {
+			if false && br.bitsRead >= 32 {
+				if br.off >= 4 {
+					v := br.in[br.off-4:]
+					v = v[:4]
+					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+					br.value = (br.value << 32) | uint64(low)
+					br.bitsRead -= 32
+					br.off -= 4
+				} else {
+					for br.off > 0 {
+						br.value = (br.value << 8) | uint64(br.in[br.off-1])
+						br.bitsRead -= 8
+						br.off--
+					}
+				}
+			}
+			// end inline...
+			if offset >= len(out) {
 				return nil, errors.New("corruption detected: stream overrun 4")
 			}
-			dstOut[offset] = decode(br)
+
+			// Read value and increment offset.
+			val := br.peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask].entry
+			nBits := uint8(v)
+			br.advance(nBits)
+			bitsLeft -= uint(nBits)
+			out[offset] = uint8(v >> 8)
 			offset++
 		}
 		decoded += offset - dstEvery*i
@@ -397,7 +670,426 @@
 	if dstSize != decoded {
 		return nil, errors.New("corruption detected: short output block")
 	}
-	return s.Out, nil
+	return dst, nil
+}
+
+// Decompress4X will decompress a 4X encoded stream.
+// The length of the supplied input must match the end of a block exactly.
+// The *capacity* of the dst slice must match the destination size of
+// the uncompressed data exactly.
+func (d *Decoder) decompress4X8bit(dst, src []byte) ([]byte, error) {
+	if d.actualTableLog == 8 {
+		return d.decompress4X8bitExactly(dst, src)
+	}
+
+	var br [4]bitReaderBytes
+	start := 6
+	for i := 0; i < 3; i++ {
+		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
+		if start+length >= len(src) {
+			return nil, errors.New("truncated input (or invalid offset)")
+		}
+		err := br[i].init(src[start : start+length])
+		if err != nil {
+			return nil, err
+		}
+		start += length
+	}
+	err := br[3].init(src[start:])
+	if err != nil {
+		return nil, err
+	}
+
+	// destination, offset to match first output
+	dstSize := cap(dst)
+	dst = dst[:dstSize]
+	out := dst
+	dstEvery := (dstSize + 3) / 4
+
+	shift := (8 - d.actualTableLog) & 7
+
+	const tlSize = 1 << 8
+	const tlMask = tlSize - 1
+	single := d.dt.single[:tlSize]
+
+	// Use temp table to avoid bound checks/append penalty.
+	var buf [256]byte
+	var off uint8
+	var decoded int
+
+	// Decode 4 values from each decoder/loop.
+	const bufoff = 256 / 4
+	for {
+		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
+			break
+		}
+
+		{
+			// Interleave 2 decodes.
+			const stream = 0
+			const stream2 = 1
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			v := single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 := single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+1] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+2] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+3] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+		}
+
+		{
+			const stream = 2
+			const stream2 = 3
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			v := single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 := single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+1] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+2] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+3] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+		}
+
+		off += 4
+
+		if off == bufoff {
+			if bufoff > dstEvery {
+				return nil, errors.New("corruption detected: stream overrun 1")
+			}
+			copy(out, buf[:bufoff])
+			copy(out[dstEvery:], buf[bufoff:bufoff*2])
+			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
+			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
+			off = 0
+			out = out[bufoff:]
+			decoded += 256
+			// There must at least be 3 buffers left.
+			if len(out) < dstEvery*3 {
+				return nil, errors.New("corruption detected: stream overrun 2")
+			}
+		}
+	}
+	if off > 0 {
+		ioff := int(off)
+		if len(out) < dstEvery*3+ioff {
+			return nil, errors.New("corruption detected: stream overrun 3")
+		}
+		copy(out, buf[:off])
+		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
+		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
+		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
+		decoded += int(off) * 4
+		out = out[off:]
+	}
+
+	// Decode remaining.
+	for i := range br {
+		offset := dstEvery * i
+		br := &br[i]
+		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
+		for bitsLeft > 0 {
+			if br.finished() {
+				return nil, io.ErrUnexpectedEOF
+			}
+			if br.bitsRead >= 56 {
+				if br.off >= 4 {
+					v := br.in[br.off-4:]
+					v = v[:4]
+					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+					br.value |= uint64(low) << (br.bitsRead - 32)
+					br.bitsRead -= 32
+					br.off -= 4
+				} else {
+					for br.off > 0 {
+						br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
+						br.bitsRead -= 8
+						br.off--
+					}
+				}
+			}
+			// end inline...
+			if offset >= len(out) {
+				return nil, errors.New("corruption detected: stream overrun 4")
+			}
+
+			// Read value and increment offset.
+			v := single[br.peekByteFast()>>shift].entry
+			nBits := uint8(v)
+			br.advance(nBits)
+			bitsLeft -= int(nBits)
+			out[offset] = uint8(v >> 8)
+			offset++
+		}
+		decoded += offset - dstEvery*i
+		err = br.close()
+		if err != nil {
+			return nil, err
+		}
+	}
+	if dstSize != decoded {
+		return nil, errors.New("corruption detected: short output block")
+	}
+	return dst, nil
+}
+
+// Decompress4X will decompress a 4X encoded stream.
+// The length of the supplied input must match the end of a block exactly.
+// The *capacity* of the dst slice must match the destination size of
+// the uncompressed data exactly.
+func (d *Decoder) decompress4X8bitExactly(dst, src []byte) ([]byte, error) {
+	var br [4]bitReaderBytes
+	start := 6
+	for i := 0; i < 3; i++ {
+		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
+		if start+length >= len(src) {
+			return nil, errors.New("truncated input (or invalid offset)")
+		}
+		err := br[i].init(src[start : start+length])
+		if err != nil {
+			return nil, err
+		}
+		start += length
+	}
+	err := br[3].init(src[start:])
+	if err != nil {
+		return nil, err
+	}
+
+	// destination, offset to match first output
+	dstSize := cap(dst)
+	dst = dst[:dstSize]
+	out := dst
+	dstEvery := (dstSize + 3) / 4
+
+	const shift = 0
+	const tlSize = 1 << 8
+	const tlMask = tlSize - 1
+	single := d.dt.single[:tlSize]
+
+	// Use temp table to avoid bound checks/append penalty.
+	var buf [256]byte
+	var off uint8
+	var decoded int
+
+	// Decode 4 values from each decoder/loop.
+	const bufoff = 256 / 4
+	for {
+		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
+			break
+		}
+
+		{
+			// Interleave 2 decodes.
+			const stream = 0
+			const stream2 = 1
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			v := single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 := single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+1] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+2] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+3] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+		}
+
+		{
+			const stream = 2
+			const stream2 = 3
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			v := single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 := single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+1] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+2] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+
+			v = single[br[stream].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream+3] = uint8(v >> 8)
+			br[stream].advance(uint8(v))
+
+			v2 = single[br[stream2].peekByteFast()>>shift].entry
+			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
+			br[stream2].advance(uint8(v2))
+		}
+
+		off += 4
+
+		if off == bufoff {
+			if bufoff > dstEvery {
+				return nil, errors.New("corruption detected: stream overrun 1")
+			}
+			copy(out, buf[:bufoff])
+			copy(out[dstEvery:], buf[bufoff:bufoff*2])
+			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
+			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
+			off = 0
+			out = out[bufoff:]
+			decoded += 256
+			// There must at least be 3 buffers left.
+			if len(out) < dstEvery*3 {
+				return nil, errors.New("corruption detected: stream overrun 2")
+			}
+		}
+	}
+	if off > 0 {
+		ioff := int(off)
+		if len(out) < dstEvery*3+ioff {
+			return nil, errors.New("corruption detected: stream overrun 3")
+		}
+		copy(out, buf[:off])
+		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
+		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
+		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
+		decoded += int(off) * 4
+		out = out[off:]
+	}
+
+	// Decode remaining.
+	for i := range br {
+		offset := dstEvery * i
+		br := &br[i]
+		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
+		for bitsLeft > 0 {
+			if br.finished() {
+				return nil, io.ErrUnexpectedEOF
+			}
+			if br.bitsRead >= 56 {
+				if br.off >= 4 {
+					v := br.in[br.off-4:]
+					v = v[:4]
+					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+					br.value |= uint64(low) << (br.bitsRead - 32)
+					br.bitsRead -= 32
+					br.off -= 4
+				} else {
+					for br.off > 0 {
+						br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
+						br.bitsRead -= 8
+						br.off--
+					}
+				}
+			}
+			// end inline...
+			if offset >= len(out) {
+				return nil, errors.New("corruption detected: stream overrun 4")
+			}
+
+			// Read value and increment offset.
+			v := single[br.peekByteFast()>>shift].entry
+			nBits := uint8(v)
+			br.advance(nBits)
+			bitsLeft -= int(nBits)
+			out[offset] = uint8(v >> 8)
+			offset++
+		}
+		decoded += offset - dstEvery*i
+		err = br.close()
+		if err != nil {
+			return nil, err
+		}
+	}
+	if dstSize != decoded {
+		return nil, errors.New("corruption detected: short output block")
+	}
+	return dst, nil
 }
 
 // matches will compare a decoding table to a coding table.
diff --git a/vendor/github.com/klauspost/compress/huff0/huff0.go b/vendor/github.com/klauspost/compress/huff0/huff0.go
index 53249df..7ec2022 100644
--- a/vendor/github.com/klauspost/compress/huff0/huff0.go
+++ b/vendor/github.com/klauspost/compress/huff0/huff0.go
@@ -55,6 +55,9 @@
 	// ReusePolicyNone will disable re-use of tables.
 	// This is slightly faster than ReusePolicyAllow but may produce larger output.
 	ReusePolicyNone
+
+	// ReusePolicyMust must allow reuse and produce smaller output.
+	ReusePolicyMust
 )
 
 type Scratch struct {
@@ -79,6 +82,13 @@
 	// Slice of the returned data.
 	OutData []byte
 
+	// MaxDecodedSize will set the maximum allowed output size.
+	// This value will automatically be set to BlockSizeMax if not set.
+	// Decoders will return ErrMaxDecodedSizeExceeded is this limit is exceeded.
+	MaxDecodedSize int
+
+	br byteReader
+
 	// MaxSymbolValue will override the maximum symbol value of the next block.
 	MaxSymbolValue uint8
 
@@ -95,12 +105,6 @@
 	// If WantLogLess == 0 any improvement will do.
 	WantLogLess uint8
 
-	// MaxDecodedSize will set the maximum allowed output size.
-	// This value will automatically be set to BlockSizeMax if not set.
-	// Decoders will return ErrMaxDecodedSizeExceeded is this limit is exceeded.
-	MaxDecodedSize int
-
-	br             byteReader
 	symbolLen      uint16 // Length of active part of the symbol table.
 	maxCount       int    // count of the most probable symbol
 	clearCount     bool   // clear count
@@ -115,6 +119,16 @@
 	huffWeight     [maxSymbolValue + 1]byte
 }
 
+// TransferCTable will transfer the previously used compression table.
+func (s *Scratch) TransferCTable(src *Scratch) {
+	if cap(s.prevTable) < len(src.prevTable) {
+		s.prevTable = make(cTable, 0, maxSymbolValue+1)
+	}
+	s.prevTable = s.prevTable[:len(src.prevTable)]
+	copy(s.prevTable, src.prevTable)
+	s.prevTableLog = src.prevTableLog
+}
+
 func (s *Scratch) prepare(in []byte) (*Scratch, error) {
 	if len(in) > BlockSizeMax {
 		return nil, ErrTooBig