diff --git a/vendor/github.com/klauspost/compress/.gitattributes b/vendor/github.com/klauspost/compress/.gitattributes
new file mode 100644
index 0000000..4024335
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/.gitattributes
@@ -0,0 +1,2 @@
+* -text
+*.bin -text -diff
diff --git a/vendor/github.com/klauspost/compress/.gitignore b/vendor/github.com/klauspost/compress/.gitignore
new file mode 100644
index 0000000..d31b378
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/.gitignore
@@ -0,0 +1,32 @@
+# Compiled Object files, Static and Dynamic libs (Shared Objects)
+*.o
+*.a
+*.so
+
+# Folders
+_obj
+_test
+
+# Architecture specific extensions/prefixes
+*.[568vq]
+[568vq].out
+
+*.cgo1.go
+*.cgo2.c
+_cgo_defun.c
+_cgo_gotypes.go
+_cgo_export.*
+
+_testmain.go
+
+*.exe
+*.test
+*.prof
+/s2/cmd/_s2sx/sfx-exe
+
+# Linux perf files
+perf.data
+perf.data.old
+
+# gdb history
+.gdb_history
diff --git a/vendor/github.com/klauspost/compress/.goreleaser.yml b/vendor/github.com/klauspost/compress/.goreleaser.yml
new file mode 100644
index 0000000..0af08e6
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/.goreleaser.yml
@@ -0,0 +1,141 @@
+# This is an example goreleaser.yaml file with some sane defaults.
+# Make sure to check the documentation at http://goreleaser.com
+before:
+  hooks:
+    - ./gen.sh
+    - go install mvdan.cc/garble@latest
+
+builds:
+  -
+    id: "s2c"
+    binary: s2c
+    main: ./s2/cmd/s2c/main.go
+    flags:
+      - -trimpath
+    env:
+      - CGO_ENABLED=0
+    goos:
+      - aix
+      - linux
+      - freebsd
+      - netbsd
+      - windows
+      - darwin
+    goarch:
+      - 386
+      - amd64
+      - arm
+      - arm64
+      - ppc64
+      - ppc64le
+      - mips64
+      - mips64le
+    goarm:
+      - 7
+    gobinary: garble
+  -
+    id: "s2d"
+    binary: s2d
+    main: ./s2/cmd/s2d/main.go
+    flags:
+      - -trimpath
+    env:
+      - CGO_ENABLED=0
+    goos:
+      - aix
+      - linux
+      - freebsd
+      - netbsd
+      - windows
+      - darwin
+    goarch:
+      - 386
+      - amd64
+      - arm
+      - arm64
+      - ppc64
+      - ppc64le
+      - mips64
+      - mips64le
+    goarm:
+      - 7
+    gobinary: garble
+  -
+    id: "s2sx"
+    binary: s2sx
+    main: ./s2/cmd/_s2sx/main.go
+    flags:
+      - -modfile=s2sx.mod
+      - -trimpath
+    env:
+      - CGO_ENABLED=0
+    goos:
+      - aix
+      - linux
+      - freebsd
+      - netbsd
+      - windows
+      - darwin
+    goarch:
+      - 386
+      - amd64
+      - arm
+      - arm64
+      - ppc64
+      - ppc64le
+      - mips64
+      - mips64le
+    goarm:
+      - 7
+    gobinary: garble
+
+archives:
+  -
+    id: s2-binaries
+    name_template: "s2-{{ .Os }}_{{ .Arch }}_{{ .Version }}"
+    replacements:
+      aix: AIX
+      darwin: OSX
+      linux: Linux
+      windows: Windows
+      386: i386
+      amd64: x86_64
+      freebsd: FreeBSD
+      netbsd: NetBSD
+    format_overrides:
+      - goos: windows
+        format: zip
+    files:
+      - unpack/*
+      - s2/LICENSE
+      - s2/README.md
+checksum:
+  name_template: 'checksums.txt'
+snapshot:
+  name_template: "{{ .Tag }}-next"
+changelog:
+  sort: asc
+  filters:
+    exclude:
+    - '^doc:'
+    - '^docs:'
+    - '^test:'
+    - '^tests:'
+    - '^Update\sREADME.md'
+
+nfpms:
+  -
+    file_name_template: "s2_package_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
+    vendor: Klaus Post
+    homepage: https://github.com/klauspost/compress
+    maintainer: Klaus Post <klauspost@gmail.com>
+    description: S2 Compression Tool
+    license: BSD 3-Clause
+    formats:
+      - deb
+      - rpm
+    replacements:
+      darwin: Darwin
+      linux: Linux
+      freebsd: FreeBSD
+      amd64: x86_64
diff --git a/vendor/github.com/klauspost/compress/LICENSE b/vendor/github.com/klauspost/compress/LICENSE
index 1eb75ef..87d5574 100644
--- a/vendor/github.com/klauspost/compress/LICENSE
+++ b/vendor/github.com/klauspost/compress/LICENSE
@@ -26,3 +26,279 @@
 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.
+
+------------------
+
+Files: gzhttp/*
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2016-2017 The New York Times Company
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
+
+------------------
+
+Files: s2/cmd/internal/readahead/*
+
+The MIT License (MIT)
+
+Copyright (c) 2015 Klaus Post
+
+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.
+
+---------------------
+Files: snappy/*
+Files: internal/snapref/*
+
+Copyright (c) 2011 The Snappy-Go Authors. 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 Google Inc. 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.
+
+-----------------
+
+Files: s2/cmd/internal/filepathx/*
+
+Copyright 2016 The filepathx Authors
+
+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.
diff --git a/vendor/github.com/klauspost/compress/README.md b/vendor/github.com/klauspost/compress/README.md
new file mode 100644
index 0000000..ad5c63a
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/README.md
@@ -0,0 +1,560 @@
+# compress
+
+This package provides various compression algorithms.
+
+* [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and decompression in pure Go.
+* [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) is a high performance replacement for Snappy.
+* Optimized [deflate](https://godoc.org/github.com/klauspost/compress/flate) packages which can be used as a dropin replacement for [gzip](https://godoc.org/github.com/klauspost/compress/gzip), [zip](https://godoc.org/github.com/klauspost/compress/zip) and [zlib](https://godoc.org/github.com/klauspost/compress/zlib).
+* [snappy](https://github.com/klauspost/compress/tree/master/snappy) is a drop-in replacement for `github.com/golang/snappy` offering better compression and concurrent streams.
+* [huff0](https://github.com/klauspost/compress/tree/master/huff0) and [FSE](https://github.com/klauspost/compress/tree/master/fse) implementations for raw entropy encoding.
+* [gzhttp](https://github.com/klauspost/compress/tree/master/gzhttp) Provides client and server wrappers for handling gzipped requests efficiently.
+* [pgzip](https://github.com/klauspost/pgzip) is a separate package that provides a very fast parallel gzip implementation.
+* [fuzz package](https://github.com/klauspost/compress-fuzz) for fuzz testing all compressors/decompressors here.
+
+[![Go Reference](https://pkg.go.dev/badge/klauspost/compress.svg)](https://pkg.go.dev/github.com/klauspost/compress?tab=subdirectories)
+[![Go](https://github.com/klauspost/compress/actions/workflows/go.yml/badge.svg)](https://github.com/klauspost/compress/actions/workflows/go.yml)
+[![Sourcegraph Badge](https://sourcegraph.com/github.com/klauspost/compress/-/badge.svg)](https://sourcegraph.com/github.com/klauspost/compress?badge)
+
+# changelog
+
+* July 13, 2022 (v1.15.8)
+
+	* gzip: fix stack exhaustion bug in Reader.Read https://github.com/klauspost/compress/pull/641
+	* s2: Add Index header trim/restore https://github.com/klauspost/compress/pull/638
+	* zstd: Optimize seqdeq amd64 asm by @greatroar in https://github.com/klauspost/compress/pull/636
+	* zstd: Improve decoder memcopy https://github.com/klauspost/compress/pull/637
+	* huff0: Pass a single bitReader pointer to asm by @greatroar in https://github.com/klauspost/compress/pull/634
+	* zstd: Branchless getBits for amd64 w/o BMI2 by @greatroar in https://github.com/klauspost/compress/pull/640
+	* gzhttp: Remove header before writing https://github.com/klauspost/compress/pull/639
+
+* June 29, 2022 (v1.15.7)
+
+	* s2: Fix absolute forward seeks  https://github.com/klauspost/compress/pull/633
+	* zip: Merge upstream  https://github.com/klauspost/compress/pull/631
+	* zip: Re-add zip64 fix https://github.com/klauspost/compress/pull/624
+	* zstd: translate fseDecoder.buildDtable into asm by @WojciechMula in https://github.com/klauspost/compress/pull/598
+	* flate: Faster histograms  https://github.com/klauspost/compress/pull/620
+	* deflate: Use compound hcode  https://github.com/klauspost/compress/pull/622
+
+* June 3, 2022 (v1.15.6)
+	* s2: Improve coding for long, close matches https://github.com/klauspost/compress/pull/613
+	* s2c: Add Snappy/S2 stream recompression https://github.com/klauspost/compress/pull/611
+	* zstd: Always use configured block size https://github.com/klauspost/compress/pull/605
+	* zstd: Fix incorrect hash table placement for dict encoding in default https://github.com/klauspost/compress/pull/606
+	* zstd: Apply default config to ZipDecompressor without options https://github.com/klauspost/compress/pull/608
+	* gzhttp: Exclude more common archive formats https://github.com/klauspost/compress/pull/612
+	* s2: Add ReaderIgnoreCRC https://github.com/klauspost/compress/pull/609
+	* s2: Remove sanity load on index creation https://github.com/klauspost/compress/pull/607
+	* snappy: Use dedicated function for scoring https://github.com/klauspost/compress/pull/614
+	* s2c+s2d: Use official snappy framed extension https://github.com/klauspost/compress/pull/610
+
+* May 25, 2022 (v1.15.5)
+	* s2: Add concurrent stream decompression https://github.com/klauspost/compress/pull/602
+	* s2: Fix final emit oob read crash on amd64 https://github.com/klauspost/compress/pull/601
+	* huff0: asm implementation of Decompress1X by @WojciechMula https://github.com/klauspost/compress/pull/596
+	* zstd: Use 1 less goroutine for stream decoding https://github.com/klauspost/compress/pull/588
+	* zstd: Copy literal in 16 byte blocks when possible https://github.com/klauspost/compress/pull/592
+	* zstd: Speed up when WithDecoderLowmem(false) https://github.com/klauspost/compress/pull/599
+	* zstd: faster next state update in BMI2 version of decode by @WojciechMula in https://github.com/klauspost/compress/pull/593
+	* huff0: Do not check max size when reading table. https://github.com/klauspost/compress/pull/586
+	* flate: Inplace hashing for level 7-9 by @klauspost in https://github.com/klauspost/compress/pull/590
+
+
+* May 11, 2022 (v1.15.4)
+	* huff0: decompress directly into output by @WojciechMula in [#577](https://github.com/klauspost/compress/pull/577)
+	* inflate: Keep dict on stack [#581](https://github.com/klauspost/compress/pull/581)
+	* zstd: Faster decoding memcopy in asm [#583](https://github.com/klauspost/compress/pull/583)
+	* zstd: Fix ignored crc [#580](https://github.com/klauspost/compress/pull/580)
+
+* May 5, 2022 (v1.15.3)
+	* zstd: Allow to ignore checksum checking by @WojciechMula [#572](https://github.com/klauspost/compress/pull/572)
+	* s2: Fix incorrect seek for io.SeekEnd in [#575](https://github.com/klauspost/compress/pull/575)
+
+* Apr 26, 2022 (v1.15.2)
+	* zstd: Add x86-64 assembly for decompression on streams and blocks. Contributed by [@WojciechMula](https://github.com/WojciechMula). Typically 2x faster.  [#528](https://github.com/klauspost/compress/pull/528) [#531](https://github.com/klauspost/compress/pull/531) [#545](https://github.com/klauspost/compress/pull/545) [#537](https://github.com/klauspost/compress/pull/537)
+	* zstd: Add options to ZipDecompressor and fixes [#539](https://github.com/klauspost/compress/pull/539)
+	* s2: Use sorted search for index [#555](https://github.com/klauspost/compress/pull/555)
+	* Minimum version is Go 1.16, added CI test on 1.18.
+
+* Mar 11, 2022 (v1.15.1)
+	* huff0: Add x86 assembly of Decode4X by @WojciechMula in [#512](https://github.com/klauspost/compress/pull/512)
+	* zstd: Reuse zip decoders in [#514](https://github.com/klauspost/compress/pull/514)
+	* zstd: Detect extra block data and report as corrupted in [#520](https://github.com/klauspost/compress/pull/520)
+	* zstd: Handle zero sized frame content size stricter in [#521](https://github.com/klauspost/compress/pull/521)
+	* zstd: Add stricter block size checks in [#523](https://github.com/klauspost/compress/pull/523)
+
+* Mar 3, 2022 (v1.15.0)
+	* zstd: Refactor decoder by @klauspost in [#498](https://github.com/klauspost/compress/pull/498)
+	* zstd: Add stream encoding without goroutines by @klauspost in [#505](https://github.com/klauspost/compress/pull/505)
+	* huff0: Prevent single blocks exceeding 16 bits by @klauspost in[#507](https://github.com/klauspost/compress/pull/507)
+	* flate: Inline literal emission by @klauspost in [#509](https://github.com/klauspost/compress/pull/509)
+	* gzhttp: Add zstd to transport by @klauspost in [#400](https://github.com/klauspost/compress/pull/400)
+	* gzhttp: Make content-type optional by @klauspost in [#510](https://github.com/klauspost/compress/pull/510)
+
+<details>
+	<summary>See  Details</summary>
+Both compression and decompression now supports "synchronous" stream operations. This means that whenever "concurrency" is set to 1, they will operate without spawning goroutines.
+
+Stream decompression is now faster on asynchronous, since the goroutine allocation much more effectively splits the workload. On typical streams this will typically use 2 cores fully for decompression. When a stream has finished decoding no goroutines will be left over, so decoders can now safely be pooled and still be garbage collected.
+
+While the release has been extensively tested, it is recommended to testing when upgrading.
+</details>
+
+* Feb 22, 2022 (v1.14.4)
+	* flate: Fix rare huffman only (-2) corruption. [#503](https://github.com/klauspost/compress/pull/503)
+	* zip: Update deprecated CreateHeaderRaw to correctly call CreateRaw by @saracen in [#502](https://github.com/klauspost/compress/pull/502)
+	* zip: don't read data descriptor early by @saracen in [#501](https://github.com/klauspost/compress/pull/501)  #501
+	* huff0: Use static decompression buffer up to 30% faster by @klauspost in [#499](https://github.com/klauspost/compress/pull/499) [#500](https://github.com/klauspost/compress/pull/500)
+
+* Feb 17, 2022 (v1.14.3)
+	* flate: Improve fastest levels compression speed ~10% more throughput. [#482](https://github.com/klauspost/compress/pull/482) [#489](https://github.com/klauspost/compress/pull/489) [#490](https://github.com/klauspost/compress/pull/490) [#491](https://github.com/klauspost/compress/pull/491) [#494](https://github.com/klauspost/compress/pull/494)  [#478](https://github.com/klauspost/compress/pull/478)
+	* flate: Faster decompression speed, ~5-10%. [#483](https://github.com/klauspost/compress/pull/483)
+	* s2: Faster compression with Go v1.18 and amd64 microarch level 3+. [#484](https://github.com/klauspost/compress/pull/484) [#486](https://github.com/klauspost/compress/pull/486)
+
+* Jan 25, 2022 (v1.14.2)
+	* zstd: improve header decoder by @dsnet  [#476](https://github.com/klauspost/compress/pull/476)
+	* zstd: Add bigger default blocks  [#469](https://github.com/klauspost/compress/pull/469)
+	* zstd: Remove unused decompression buffer [#470](https://github.com/klauspost/compress/pull/470)
+	* zstd: Fix logically dead code by @ningmingxiao [#472](https://github.com/klauspost/compress/pull/472)
+	* flate: Improve level 7-9 [#471](https://github.com/klauspost/compress/pull/471) [#473](https://github.com/klauspost/compress/pull/473)
+	* zstd: Add noasm tag for xxhash [#475](https://github.com/klauspost/compress/pull/475)
+
+* Jan 11, 2022 (v1.14.1)
+	* s2: Add stream index in [#462](https://github.com/klauspost/compress/pull/462)
+	* flate: Speed and efficiency improvements in [#439](https://github.com/klauspost/compress/pull/439) [#461](https://github.com/klauspost/compress/pull/461) [#455](https://github.com/klauspost/compress/pull/455) [#452](https://github.com/klauspost/compress/pull/452) [#458](https://github.com/klauspost/compress/pull/458)
+	* zstd: Performance improvement in [#420]( https://github.com/klauspost/compress/pull/420) [#456](https://github.com/klauspost/compress/pull/456) [#437](https://github.com/klauspost/compress/pull/437) [#467](https://github.com/klauspost/compress/pull/467) [#468](https://github.com/klauspost/compress/pull/468)
+	* zstd: add arm64 xxhash assembly in [#464](https://github.com/klauspost/compress/pull/464)
+	* Add garbled for binaries for s2 in [#445](https://github.com/klauspost/compress/pull/445)
+
+<details>
+	<summary>See changes to v1.13.x</summary>
+	
+* Aug 30, 2021 (v1.13.5)
+	* gz/zlib/flate: Alias stdlib errors [#425](https://github.com/klauspost/compress/pull/425)
+	* s2: Add block support to commandline tools [#413](https://github.com/klauspost/compress/pull/413)
+	* zstd: pooledZipWriter should return Writers to the same pool [#426](https://github.com/klauspost/compress/pull/426)
+	* Removed golang/snappy as external dependency for tests [#421](https://github.com/klauspost/compress/pull/421)
+
+* Aug 12, 2021 (v1.13.4)
+	* Add [snappy replacement package](https://github.com/klauspost/compress/tree/master/snappy).
+	* zstd: Fix incorrect encoding in "best" mode [#415](https://github.com/klauspost/compress/pull/415)
+
+* Aug 3, 2021 (v1.13.3) 
+	* zstd: Improve Best compression [#404](https://github.com/klauspost/compress/pull/404)
+	* zstd: Fix WriteTo error forwarding [#411](https://github.com/klauspost/compress/pull/411)
+	* gzhttp: Return http.HandlerFunc instead of http.Handler. Unlikely breaking change. [#406](https://github.com/klauspost/compress/pull/406)
+	* s2sx: Fix max size error [#399](https://github.com/klauspost/compress/pull/399)
+	* zstd: Add optional stream content size on reset [#401](https://github.com/klauspost/compress/pull/401)
+	* zstd: use SpeedBestCompression for level >= 10 [#410](https://github.com/klauspost/compress/pull/410)
+
+* Jun 14, 2021 (v1.13.1)
+	* s2: Add full Snappy output support  [#396](https://github.com/klauspost/compress/pull/396)
+	* zstd: Add configurable [Decoder window](https://pkg.go.dev/github.com/klauspost/compress/zstd#WithDecoderMaxWindow) size [#394](https://github.com/klauspost/compress/pull/394)
+	* gzhttp: Add header to skip compression  [#389](https://github.com/klauspost/compress/pull/389)
+	* s2: Improve speed with bigger output margin  [#395](https://github.com/klauspost/compress/pull/395)
+
+* Jun 3, 2021 (v1.13.0)
+	* Added [gzhttp](https://github.com/klauspost/compress/tree/master/gzhttp#gzip-handler) which allows wrapping HTTP servers and clients with GZIP compressors.
+	* zstd: Detect short invalid signatures [#382](https://github.com/klauspost/compress/pull/382)
+	* zstd: Spawn decoder goroutine only if needed. [#380](https://github.com/klauspost/compress/pull/380)
+</details>
+
+
+<details>
+	<summary>See changes to v1.12.x</summary>
+	
+* May 25, 2021 (v1.12.3)
+	* deflate: Better/faster Huffman encoding [#374](https://github.com/klauspost/compress/pull/374)
+	* deflate: Allocate less for history. [#375](https://github.com/klauspost/compress/pull/375)
+	* zstd: Forward read errors [#373](https://github.com/klauspost/compress/pull/373) 
+
+* Apr 27, 2021 (v1.12.2)
+	* zstd: Improve better/best compression [#360](https://github.com/klauspost/compress/pull/360) [#364](https://github.com/klauspost/compress/pull/364) [#365](https://github.com/klauspost/compress/pull/365)
+	* zstd: Add helpers to compress/decompress zstd inside zip files [#363](https://github.com/klauspost/compress/pull/363)
+	* deflate: Improve level 5+6 compression [#367](https://github.com/klauspost/compress/pull/367)
+	* s2: Improve better/best compression [#358](https://github.com/klauspost/compress/pull/358) [#359](https://github.com/klauspost/compress/pull/358)
+	* s2: Load after checking src limit on amd64. [#362](https://github.com/klauspost/compress/pull/362)
+	* s2sx: Limit max executable size [#368](https://github.com/klauspost/compress/pull/368) 
+
+* Apr 14, 2021 (v1.12.1)
+	* snappy package removed. Upstream added as dependency.
+	* s2: Better compression in "best" mode [#353](https://github.com/klauspost/compress/pull/353)
+	* s2sx: Add stdin input and detect pre-compressed from signature [#352](https://github.com/klauspost/compress/pull/352)
+	* s2c/s2d: Add http as possible input [#348](https://github.com/klauspost/compress/pull/348)
+	* s2c/s2d/s2sx: Always truncate when writing files [#352](https://github.com/klauspost/compress/pull/352)
+	* zstd: Reduce memory usage further when using [WithLowerEncoderMem](https://pkg.go.dev/github.com/klauspost/compress/zstd#WithLowerEncoderMem) [#346](https://github.com/klauspost/compress/pull/346)
+	* s2: Fix potential problem with amd64 assembly and profilers [#349](https://github.com/klauspost/compress/pull/349)
+</details>
+
+<details>
+	<summary>See changes to v1.11.x</summary>
+	
+* Mar 26, 2021 (v1.11.13)
+	* zstd: Big speedup on small dictionary encodes [#344](https://github.com/klauspost/compress/pull/344) [#345](https://github.com/klauspost/compress/pull/345)
+	* zstd: Add [WithLowerEncoderMem](https://pkg.go.dev/github.com/klauspost/compress/zstd#WithLowerEncoderMem) encoder option [#336](https://github.com/klauspost/compress/pull/336)
+	* deflate: Improve entropy compression [#338](https://github.com/klauspost/compress/pull/338)
+	* s2: Clean up and minor performance improvement in best [#341](https://github.com/klauspost/compress/pull/341)
+
+* Mar 5, 2021 (v1.11.12)
+	* s2: Add `s2sx` binary that creates [self extracting archives](https://github.com/klauspost/compress/tree/master/s2#s2sx-self-extracting-archives).
+	* s2: Speed up decompression on non-assembly platforms [#328](https://github.com/klauspost/compress/pull/328)
+
+* Mar 1, 2021 (v1.11.9)
+	* s2: Add ARM64 decompression assembly. Around 2x output speed. [#324](https://github.com/klauspost/compress/pull/324)
+	* s2: Improve "better" speed and efficiency. [#325](https://github.com/klauspost/compress/pull/325)
+	* s2: Fix binaries.
+
+* Feb 25, 2021 (v1.11.8)
+	* s2: Fixed occational out-of-bounds write on amd64. Upgrade recommended.
+	* s2: Add AMD64 assembly for better mode. 25-50% faster. [#315](https://github.com/klauspost/compress/pull/315)
+	* s2: Less upfront decoder allocation. [#322](https://github.com/klauspost/compress/pull/322)
+	* zstd: Faster "compression" of incompressible data. [#314](https://github.com/klauspost/compress/pull/314)
+	* zip: Fix zip64 headers. [#313](https://github.com/klauspost/compress/pull/313)
+  
+* Jan 14, 2021 (v1.11.7)
+	* Use Bytes() interface to get bytes across packages. [#309](https://github.com/klauspost/compress/pull/309)
+	* s2: Add 'best' compression option.  [#310](https://github.com/klauspost/compress/pull/310)
+	* s2: Add ReaderMaxBlockSize, changes `s2.NewReader` signature to include varargs. [#311](https://github.com/klauspost/compress/pull/311)
+	* s2: Fix crash on small better buffers. [#308](https://github.com/klauspost/compress/pull/308)
+	* s2: Clean up decoder. [#312](https://github.com/klauspost/compress/pull/312)
+
+* Jan 7, 2021 (v1.11.6)
+	* zstd: Make decoder allocations smaller [#306](https://github.com/klauspost/compress/pull/306)
+	* zstd: Free Decoder resources when Reset is called with a nil io.Reader  [#305](https://github.com/klauspost/compress/pull/305)
+
+* Dec 20, 2020 (v1.11.4)
+	* zstd: Add Best compression mode [#304](https://github.com/klauspost/compress/pull/304)
+	* Add header decoder [#299](https://github.com/klauspost/compress/pull/299)
+	* s2: Add uncompressed stream option [#297](https://github.com/klauspost/compress/pull/297)
+	* Simplify/speed up small blocks with known max size. [#300](https://github.com/klauspost/compress/pull/300)
+	* zstd: Always reset literal dict encoder [#303](https://github.com/klauspost/compress/pull/303)
+
+* Nov 15, 2020 (v1.11.3)
+	* inflate: 10-15% faster decompression  [#293](https://github.com/klauspost/compress/pull/293)
+	* zstd: Tweak DecodeAll default allocation [#295](https://github.com/klauspost/compress/pull/295)
+
+* Oct 11, 2020 (v1.11.2)
+	* s2: Fix out of bounds read in "better" block compression [#291](https://github.com/klauspost/compress/pull/291)
+
+* Oct 1, 2020 (v1.11.1)
+	* zstd: Set allLitEntropy true in default configuration [#286](https://github.com/klauspost/compress/pull/286)
+
+* Sept 8, 2020 (v1.11.0)
+	* zstd: Add experimental compression [dictionaries](https://github.com/klauspost/compress/tree/master/zstd#dictionaries) [#281](https://github.com/klauspost/compress/pull/281)
+	* zstd: Fix mixed Write and ReadFrom calls [#282](https://github.com/klauspost/compress/pull/282)
+	* inflate/gz: Limit variable shifts, ~5% faster decompression [#274](https://github.com/klauspost/compress/pull/274)
+</details>
+
+<details>
+	<summary>See changes to v1.10.x</summary>
+ 
+* July 8, 2020 (v1.10.11) 
+	* zstd: Fix extra block when compressing with ReadFrom. [#278](https://github.com/klauspost/compress/pull/278)
+	* huff0: Also populate compression table when reading decoding table. [#275](https://github.com/klauspost/compress/pull/275)
+	
+* June 23, 2020 (v1.10.10) 
+	* zstd: Skip entropy compression in fastest mode when no matches. [#270](https://github.com/klauspost/compress/pull/270)
+	
+* June 16, 2020 (v1.10.9): 
+	* zstd: API change for specifying dictionaries. See [#268](https://github.com/klauspost/compress/pull/268)
+	* zip: update CreateHeaderRaw to handle zip64 fields. [#266](https://github.com/klauspost/compress/pull/266)
+	* Fuzzit tests removed. The service has been purchased and is no longer available.
+	
+* June 5, 2020 (v1.10.8): 
+	* 1.15x faster zstd block decompression. [#265](https://github.com/klauspost/compress/pull/265)
+	
+* June 1, 2020 (v1.10.7): 
+	* Added zstd decompression [dictionary support](https://github.com/klauspost/compress/tree/master/zstd#dictionaries)
+	* Increase zstd decompression speed up to 1.19x.  [#259](https://github.com/klauspost/compress/pull/259)
+	* Remove internal reset call in zstd compression and reduce allocations. [#263](https://github.com/klauspost/compress/pull/263)
+	
+* May 21, 2020: (v1.10.6) 
+	* zstd: Reduce allocations while decoding. [#258](https://github.com/klauspost/compress/pull/258), [#252](https://github.com/klauspost/compress/pull/252)
+	* zstd: Stricter decompression checks.
+	
+* April 12, 2020: (v1.10.5)
+	* s2-commands: Flush output when receiving SIGINT. [#239](https://github.com/klauspost/compress/pull/239)
+	
+* Apr 8, 2020: (v1.10.4) 
+	* zstd: Minor/special case optimizations. [#251](https://github.com/klauspost/compress/pull/251),  [#250](https://github.com/klauspost/compress/pull/250),  [#249](https://github.com/klauspost/compress/pull/249),  [#247](https://github.com/klauspost/compress/pull/247)
+* Mar 11, 2020: (v1.10.3) 
+	* s2: Use S2 encoder in pure Go mode for Snappy output as well. [#245](https://github.com/klauspost/compress/pull/245)
+	* s2: Fix pure Go block encoder. [#244](https://github.com/klauspost/compress/pull/244)
+	* zstd: Added "better compression" mode. [#240](https://github.com/klauspost/compress/pull/240)
+	* zstd: Improve speed of fastest compression mode by 5-10% [#241](https://github.com/klauspost/compress/pull/241)
+	* zstd: Skip creating encoders when not needed. [#238](https://github.com/klauspost/compress/pull/238)
+	
+* Feb 27, 2020: (v1.10.2) 
+	* Close to 50% speedup in inflate (gzip/zip decompression). [#236](https://github.com/klauspost/compress/pull/236) [#234](https://github.com/klauspost/compress/pull/234) [#232](https://github.com/klauspost/compress/pull/232)
+	* Reduce deflate level 1-6 memory usage up to 59%. [#227](https://github.com/klauspost/compress/pull/227)
+	
+* Feb 18, 2020: (v1.10.1)
+	* Fix zstd crash when resetting multiple times without sending data. [#226](https://github.com/klauspost/compress/pull/226)
+	* deflate: Fix dictionary use on level 1-6. [#224](https://github.com/klauspost/compress/pull/224)
+	* Remove deflate writer reference when closing. [#224](https://github.com/klauspost/compress/pull/224)
+	
+* Feb 4, 2020: (v1.10.0) 
+	* Add optional dictionary to [stateless deflate](https://pkg.go.dev/github.com/klauspost/compress/flate?tab=doc#StatelessDeflate). Breaking change, send `nil` for previous behaviour. [#216](https://github.com/klauspost/compress/pull/216)
+	* Fix buffer overflow on repeated small block deflate.  [#218](https://github.com/klauspost/compress/pull/218)
+	* Allow copying content from an existing ZIP file without decompressing+compressing. [#214](https://github.com/klauspost/compress/pull/214)
+	* Added [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) AMD64 assembler and various optimizations. Stream speed >10GB/s.  [#186](https://github.com/klauspost/compress/pull/186)
+
+</details>
+
+<details>
+	<summary>See changes prior to v1.10.0</summary>
+
+* Jan 20,2020 (v1.9.8) Optimize gzip/deflate with better size estimates and faster table generation. [#207](https://github.com/klauspost/compress/pull/207) by [luyu6056](https://github.com/luyu6056),  [#206](https://github.com/klauspost/compress/pull/206).
+* Jan 11, 2020: S2 Encode/Decode will use provided buffer if capacity is big enough. [#204](https://github.com/klauspost/compress/pull/204) 
+* Jan 5, 2020: (v1.9.7) Fix another zstd regression in v1.9.5 - v1.9.6 removed.
+* Jan 4, 2020: (v1.9.6) Regression in v1.9.5 fixed causing corrupt zstd encodes in rare cases.
+* Jan 4, 2020: Faster IO in [s2c + s2d commandline tools](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) compression/decompression. [#192](https://github.com/klauspost/compress/pull/192)
+* Dec 29, 2019: Removed v1.9.5 since fuzz tests showed a compatibility problem with the reference zstandard decoder.
+* Dec 29, 2019: (v1.9.5) zstd: 10-20% faster block compression. [#199](https://github.com/klauspost/compress/pull/199)
+* Dec 29, 2019: [zip](https://godoc.org/github.com/klauspost/compress/zip) package updated with latest Go features
+* Dec 29, 2019: zstd: Single segment flag condintions tweaked. [#197](https://github.com/klauspost/compress/pull/197)
+* Dec 18, 2019: s2: Faster compression when ReadFrom is used. [#198](https://github.com/klauspost/compress/pull/198)
+* Dec 10, 2019: s2: Fix repeat length output when just above at 16MB limit.
+* Dec 10, 2019: zstd: Add function to get decoder as io.ReadCloser. [#191](https://github.com/klauspost/compress/pull/191)
+* Dec 3, 2019: (v1.9.4) S2: limit max repeat length. [#188](https://github.com/klauspost/compress/pull/188)
+* Dec 3, 2019: Add [WithNoEntropyCompression](https://godoc.org/github.com/klauspost/compress/zstd#WithNoEntropyCompression) to zstd [#187](https://github.com/klauspost/compress/pull/187)
+* Dec 3, 2019: Reduce memory use for tests. Check for leaked goroutines.
+* Nov 28, 2019 (v1.9.3) Less allocations in stateless deflate.
+* Nov 28, 2019: 5-20% Faster huff0 decode. Impacts zstd as well. [#184](https://github.com/klauspost/compress/pull/184)
+* Nov 12, 2019 (v1.9.2) Added [Stateless Compression](#stateless-compression) for gzip/deflate.
+* Nov 12, 2019: Fixed zstd decompression of large single blocks. [#180](https://github.com/klauspost/compress/pull/180)
+* Nov 11, 2019: Set default  [s2c](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) block size to 4MB.
+* Nov 11, 2019: Reduce inflate memory use by 1KB.
+* Nov 10, 2019: Less allocations in deflate bit writer.
+* Nov 10, 2019: Fix inconsistent error returned by zstd decoder.
+* Oct 28, 2019 (v1.9.1) ztsd: Fix crash when compressing blocks. [#174](https://github.com/klauspost/compress/pull/174)
+* Oct 24, 2019 (v1.9.0) zstd: Fix rare data corruption [#173](https://github.com/klauspost/compress/pull/173)
+* Oct 24, 2019 zstd: Fix huff0 out of buffer write [#171](https://github.com/klauspost/compress/pull/171) and always return errors [#172](https://github.com/klauspost/compress/pull/172) 
+* Oct 10, 2019: Big deflate rewrite, 30-40% faster with better compression [#105](https://github.com/klauspost/compress/pull/105)
+
+</details>
+
+<details>
+	<summary>See changes prior to v1.9.0</summary>
+
+* Oct 10, 2019: (v1.8.6) zstd: Allow partial reads to get flushed data. [#169](https://github.com/klauspost/compress/pull/169)
+* Oct 3, 2019: Fix inconsistent results on broken zstd streams.
+* Sep 25, 2019: Added `-rm` (remove source files) and `-q` (no output except errors) to `s2c` and `s2d` [commands](https://github.com/klauspost/compress/tree/master/s2#commandline-tools)
+* Sep 16, 2019: (v1.8.4) Add `s2c` and `s2d` [commandline tools](https://github.com/klauspost/compress/tree/master/s2#commandline-tools).
+* Sep 10, 2019: (v1.8.3) Fix s2 decoder [Skip](https://godoc.org/github.com/klauspost/compress/s2#Reader.Skip).
+* Sep 7, 2019: zstd: Added [WithWindowSize](https://godoc.org/github.com/klauspost/compress/zstd#WithWindowSize), contributed by [ianwilkes](https://github.com/ianwilkes).
+* Sep 5, 2019: (v1.8.2) Add [WithZeroFrames](https://godoc.org/github.com/klauspost/compress/zstd#WithZeroFrames) which adds full zero payload block encoding option.
+* Sep 5, 2019: Lazy initialization of zstandard predefined en/decoder tables.
+* Aug 26, 2019: (v1.8.1) S2: 1-2% compression increase in "better" compression mode.
+* Aug 26, 2019: zstd: Check maximum size of Huffman 1X compressed literals while decoding.
+* Aug 24, 2019: (v1.8.0) Added [S2 compression](https://github.com/klauspost/compress/tree/master/s2#s2-compression), a high performance replacement for Snappy. 
+* Aug 21, 2019: (v1.7.6) Fixed minor issues found by fuzzer. One could lead to zstd not decompressing.
+* Aug 18, 2019: Add [fuzzit](https://fuzzit.dev/) continuous fuzzing.
+* Aug 14, 2019: zstd: Skip incompressible data 2x faster.  [#147](https://github.com/klauspost/compress/pull/147)
+* Aug 4, 2019 (v1.7.5): Better literal compression. [#146](https://github.com/klauspost/compress/pull/146)
+* Aug 4, 2019: Faster zstd compression. [#143](https://github.com/klauspost/compress/pull/143) [#144](https://github.com/klauspost/compress/pull/144)
+* Aug 4, 2019: Faster zstd decompression. [#145](https://github.com/klauspost/compress/pull/145) [#143](https://github.com/klauspost/compress/pull/143) [#142](https://github.com/klauspost/compress/pull/142)
+* July 15, 2019 (v1.7.4): Fix double EOF block in rare cases on zstd encoder.
+* July 15, 2019 (v1.7.3): Minor speedup/compression increase in default zstd encoder.
+* July 14, 2019: zstd decoder: Fix decompression error on multiple uses with mixed content.
+* July 7, 2019 (v1.7.2): Snappy update, zstd decoder potential race fix.
+* June 17, 2019: zstd decompression bugfix.
+* June 17, 2019: fix 32 bit builds.
+* June 17, 2019: Easier use in modules (less dependencies).
+* June 9, 2019: New stronger "default" [zstd](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression mode. Matches zstd default compression ratio.
+* June 5, 2019: 20-40% throughput in [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and better compression.
+* June 5, 2019: deflate/gzip compression: Reduce memory usage of lower compression levels.
+* June 2, 2019: Added [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression!
+* May 25, 2019: deflate/gzip: 10% faster bit writer, mostly visible in lower levels.
+* Apr 22, 2019: [zstd](https://github.com/klauspost/compress/tree/master/zstd#zstd) decompression added.
+* Aug 1, 2018: Added [huff0 README](https://github.com/klauspost/compress/tree/master/huff0#huff0-entropy-compression).
+* Jul 8, 2018: Added [Performance Update 2018](#performance-update-2018) below.
+* Jun 23, 2018: Merged [Go 1.11 inflate optimizations](https://go-review.googlesource.com/c/go/+/102235). Go 1.9 is now required. Backwards compatible version tagged with [v1.3.0](https://github.com/klauspost/compress/releases/tag/v1.3.0).
+* Apr 2, 2018: Added [huff0](https://godoc.org/github.com/klauspost/compress/huff0) en/decoder. Experimental for now, API may change.
+* Mar 4, 2018: Added [FSE Entropy](https://godoc.org/github.com/klauspost/compress/fse) en/decoder. Experimental for now, API may change.
+* Nov 3, 2017: Add compression [Estimate](https://godoc.org/github.com/klauspost/compress#Estimate) function.
+* May 28, 2017: Reduce allocations when resetting decoder.
+* Apr 02, 2017: Change back to official crc32, since changes were merged in Go 1.7.
+* Jan 14, 2017: Reduce stack pressure due to array copies. See [Issue #18625](https://github.com/golang/go/issues/18625).
+* Oct 25, 2016: Level 2-4 have been rewritten and now offers significantly better performance than before.
+* Oct 20, 2016: Port zlib changes from Go 1.7 to fix zlib writer issue. Please update.
+* Oct 16, 2016: Go 1.7 changes merged. Apples to apples this package is a few percent faster, but has a significantly better balance between speed and compression per level. 
+* Mar 24, 2016: Always attempt Huffman encoding on level 4-7. This improves base 64 encoded data compression.
+* Mar 24, 2016: Small speedup for level 1-3.
+* Feb 19, 2016: Faster bit writer, level -2 is 15% faster, level 1 is 4% faster.
+* Feb 19, 2016: Handle small payloads faster in level 1-3.
+* Feb 19, 2016: Added faster level 2 + 3 compression modes.
+* Feb 19, 2016: [Rebalanced compression levels](https://blog.klauspost.com/rebalancing-deflate-compression-levels/), so there is a more even progresssion in terms of compression. New default level is 5.
+* Feb 14, 2016: Snappy: Merge upstream changes. 
+* Feb 14, 2016: Snappy: Fix aggressive skipping.
+* Feb 14, 2016: Snappy: Update benchmark.
+* Feb 13, 2016: Deflate: Fixed assembler problem that could lead to sub-optimal compression.
+* Feb 12, 2016: Snappy: Added AMD64 SSE 4.2 optimizations to matching, which makes easy to compress material run faster. Typical speedup is around 25%.
+* Feb 9, 2016: Added Snappy package fork. This version is 5-7% faster, much more on hard to compress content.
+* Jan 30, 2016: Optimize level 1 to 3 by not considering static dictionary or storing uncompressed. ~4-5% speedup.
+* Jan 16, 2016: Optimization on deflate level 1,2,3 compression.
+* Jan 8 2016: Merge [CL 18317](https://go-review.googlesource.com/#/c/18317): fix reading, writing of zip64 archives.
+* Dec 8 2015: Make level 1 and -2 deterministic even if write size differs.
+* Dec 8 2015: Split encoding functions, so hashing and matching can potentially be inlined. 1-3% faster on AMD64. 5% faster on other platforms.
+* Dec 8 2015: Fixed rare [one byte out-of bounds read](https://github.com/klauspost/compress/issues/20). Please update!
+* Nov 23 2015: Optimization on token writer. ~2-4% faster. Contributed by [@dsnet](https://github.com/dsnet).
+* Nov 20 2015: Small optimization to bit writer on 64 bit systems.
+* Nov 17 2015: Fixed out-of-bound errors if the underlying Writer returned an error. See [#15](https://github.com/klauspost/compress/issues/15).
+* Nov 12 2015: Added [io.WriterTo](https://golang.org/pkg/io/#WriterTo) support to gzip/inflate.
+* Nov 11 2015: Merged [CL 16669](https://go-review.googlesource.com/#/c/16669/4): archive/zip: enable overriding (de)compressors per file
+* Oct 15 2015: Added skipping on uncompressible data. Random data speed up >5x.
+
+</details>
+
+# deflate usage
+
+The packages are drop-in replacements for standard libraries. Simply replace the import path to use them:
+
+| old import         | new import                              | Documentation
+|--------------------|-----------------------------------------|--------------------|
+| `compress/gzip`    | `github.com/klauspost/compress/gzip`    | [gzip](https://pkg.go.dev/github.com/klauspost/compress/gzip?tab=doc)
+| `compress/zlib`    | `github.com/klauspost/compress/zlib`    | [zlib](https://pkg.go.dev/github.com/klauspost/compress/zlib?tab=doc)
+| `archive/zip`      | `github.com/klauspost/compress/zip`     | [zip](https://pkg.go.dev/github.com/klauspost/compress/zip?tab=doc)
+| `compress/flate`   | `github.com/klauspost/compress/flate`   | [flate](https://pkg.go.dev/github.com/klauspost/compress/flate?tab=doc)
+
+* Optimized [deflate](https://godoc.org/github.com/klauspost/compress/flate) packages which can be used as a dropin replacement for [gzip](https://godoc.org/github.com/klauspost/compress/gzip), [zip](https://godoc.org/github.com/klauspost/compress/zip) and [zlib](https://godoc.org/github.com/klauspost/compress/zlib).
+
+You may also be interested in [pgzip](https://github.com/klauspost/pgzip), which is a drop in replacement for gzip, which support multithreaded compression on big files and the optimized [crc32](https://github.com/klauspost/crc32) package used by these packages.
+
+The packages contains the same as the standard library, so you can use the godoc for that: [gzip](http://golang.org/pkg/compress/gzip/), [zip](http://golang.org/pkg/archive/zip/),  [zlib](http://golang.org/pkg/compress/zlib/), [flate](http://golang.org/pkg/compress/flate/).
+
+Currently there is only minor speedup on decompression (mostly CRC32 calculation).
+
+Memory usage is typically 1MB for a Writer. stdlib is in the same range. 
+If you expect to have a lot of concurrently allocated Writers consider using 
+the stateless compress described below.
+
+For compression performance, see: [this spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing).
+
+# Stateless compression
+
+This package offers stateless compression as a special option for gzip/deflate. 
+It will do compression but without maintaining any state between Write calls.
+
+This means there will be no memory kept between Write calls, but compression and speed will be suboptimal.
+
+This is only relevant in cases where you expect to run many thousands of compressors concurrently, 
+but with very little activity. This is *not* intended for regular web servers serving individual requests.  
+
+Because of this, the size of actual Write calls will affect output size.
+
+In gzip, specify level `-3` / `gzip.StatelessCompression` to enable.
+
+For direct deflate use, NewStatelessWriter and StatelessDeflate are available. See [documentation](https://godoc.org/github.com/klauspost/compress/flate#NewStatelessWriter)
+
+A `bufio.Writer` can of course be used to control write sizes. For example, to use a 4KB buffer:
+
+```
+	// replace 'ioutil.Discard' with your output.
+	gzw, err := gzip.NewWriterLevel(ioutil.Discard, gzip.StatelessCompression)
+	if err != nil {
+		return err
+	}
+	defer gzw.Close()
+
+	w := bufio.NewWriterSize(gzw, 4096)
+	defer w.Flush()
+	
+	// Write to 'w' 
+```
+
+This will only use up to 4KB in memory when the writer is idle. 
+
+Compression is almost always worse than the fastest compression level 
+and each write will allocate (a little) memory. 
+
+# Performance Update 2018
+
+It has been a while since we have been looking at the speed of this package compared to the standard library, so I thought I would re-do my tests and give some overall recommendations based on the current state. All benchmarks have been performed with Go 1.10 on my Desktop Intel(R) Core(TM) i7-2600 CPU @3.40GHz. Since I last ran the tests, I have gotten more RAM, which means tests with big files are no longer limited by my SSD.
+
+The raw results are in my [updated spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing). Due to cgo changes and upstream updates i could not get the cgo version of gzip to compile. Instead I included the [zstd](https://github.com/datadog/zstd) cgo implementation. If I get cgo gzip to work again, I might replace the results in the sheet.
+
+The columns to take note of are: *MB/s* - the throughput. *Reduction* - the data size reduction in percent of the original. *Rel Speed* relative speed compared to the standard library at the same level. *Smaller* - how many percent smaller is the compressed output compared to stdlib. Negative means the output was bigger. *Loss* means the loss (or gain) in compression as a percentage difference of the input.
+
+The `gzstd` (standard library gzip) and `gzkp` (this package gzip) only uses one CPU core. [`pgzip`](https://github.com/klauspost/pgzip), [`bgzf`](https://github.com/biogo/hts/tree/master/bgzf) uses all 4 cores. [`zstd`](https://github.com/DataDog/zstd) uses one core, and is a beast (but not Go, yet).
+
+
+## Overall differences.
+
+There appears to be a roughly 5-10% speed advantage over the standard library when comparing at similar compression levels.
+
+The biggest difference you will see is the result of [re-balancing](https://blog.klauspost.com/rebalancing-deflate-compression-levels/) the compression levels. I wanted by library to give a smoother transition between the compression levels than the standard library.
+
+This package attempts to provide a more smooth transition, where "1" is taking a lot of shortcuts, "5" is the reasonable trade-off and "9" is the "give me the best compression", and the values in between gives something reasonable in between. The standard library has big differences in levels 1-4, but levels 5-9 having no significant gains - often spending a lot more time than can be justified by the achieved compression.
+
+There are links to all the test data in the [spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing) in the top left field on each tab.
+
+## Web Content
+
+This test set aims to emulate typical use in a web server. The test-set is 4GB data in 53k files, and is a mixture of (mostly) HTML, JS, CSS.
+
+Since level 1 and 9 are close to being the same code, they are quite close. But looking at the levels in-between the differences are quite big.
+
+Looking at level 6, this package is 88% faster, but will output about 6% more data. For a web server, this means you can serve 88% more data, but have to pay for 6% more bandwidth. You can draw your own conclusions on what would be the most expensive for your case.
+
+## Object files
+
+This test is for typical data files stored on a server. In this case it is a collection of Go precompiled objects. They are very compressible.
+
+The picture is similar to the web content, but with small differences since this is very compressible. Levels 2-3 offer good speed, but is sacrificing quite a bit of compression. 
+
+The standard library seems suboptimal on level 3 and 4 - offering both worse compression and speed than level 6 & 7 of this package respectively.
+
+## Highly Compressible File
+
+This is a JSON file with very high redundancy. The reduction starts at 95% on level 1, so in real life terms we are dealing with something like a highly redundant stream of data, etc.
+
+It is definitely visible that we are dealing with specialized content here, so the results are very scattered. This package does not do very well at levels 1-4, but picks up significantly at level 5 and levels 7 and 8 offering great speed for the achieved compression.
+
+So if you know you content is extremely compressible you might want to go slightly higher than the defaults. The standard library has a huge gap between levels 3 and 4 in terms of speed (2.75x slowdown), so it offers little "middle ground".
+
+## Medium-High Compressible
+
+This is a pretty common test corpus: [enwik9](http://mattmahoney.net/dc/textdata.html). It contains the first 10^9 bytes of the English Wikipedia dump on Mar. 3, 2006. This is a very good test of typical text based compression and more data heavy streams.
+
+We see a similar picture here as in "Web Content". On equal levels some compression is sacrificed for more speed. Level 5 seems to be the best trade-off between speed and size, beating stdlib level 3 in both.
+
+## Medium Compressible
+
+I will combine two test sets, one [10GB file set](http://mattmahoney.net/dc/10gb.html) and a VM disk image (~8GB). Both contain different data types and represent a typical backup scenario.
+
+The most notable thing is how quickly the standard library drops to very low compression speeds around level 5-6 without any big gains in compression. Since this type of data is fairly common, this does not seem like good behavior.
+
+
+## Un-compressible Content
+
+This is mainly a test of how good the algorithms are at detecting un-compressible input. The standard library only offers this feature with very conservative settings at level 1. Obviously there is no reason for the algorithms to try to compress input that cannot be compressed.  The only downside is that it might skip some compressible data on false detections.
+
+
+## Huffman only compression
+
+This compression library adds a special compression level, named `HuffmanOnly`, which allows near linear time compression. This is done by completely disabling matching of previous data, and only reduce the number of bits to represent each character. 
+
+This means that often used characters, like 'e' and ' ' (space) in text use the fewest bits to represent, and rare characters like '¤' takes more bits to represent. For more information see [wikipedia](https://en.wikipedia.org/wiki/Huffman_coding) or this nice [video](https://youtu.be/ZdooBTdW5bM).
+
+Since this type of compression has much less variance, the compression speed is mostly unaffected by the input data, and is usually more than *180MB/s* for a single core.
+
+The downside is that the compression ratio is usually considerably worse than even the fastest conventional compression. The compression ratio can never be better than 8:1 (12.5%). 
+
+The linear time compression can be used as a "better than nothing" mode, where you cannot risk the encoder to slow down on some content. For comparison, the size of the "Twain" text is *233460 bytes* (+29% vs. level 1) and encode speed is 144MB/s (4.5x level 1). So in this case you trade a 30% size increase for a 4 times speedup.
+
+For more information see my blog post on [Fast Linear Time Compression](http://blog.klauspost.com/constant-time-gzipzip-compression/).
+
+This is implemented on Go 1.7 as "Huffman Only" mode, though not exposed for gzip.
+
+# Other packages
+
+Here are other packages of good quality and pure Go (no cgo wrappers or autoconverted code):
+
+* [github.com/pierrec/lz4](https://github.com/pierrec/lz4) - strong multithreaded LZ4 compression.
+* [github.com/cosnicolaou/pbzip2](https://github.com/cosnicolaou/pbzip2) - multithreaded bzip2 decompression.
+* [github.com/dsnet/compress](https://github.com/dsnet/compress) - brotli decompression, bzip2 writer.
+
+# license
+
+This code is licensed under the same conditions as the original Go code. See LICENSE file.
diff --git a/vendor/github.com/klauspost/compress/compressible.go b/vendor/github.com/klauspost/compress/compressible.go
new file mode 100644
index 0000000..ea5a692
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/compressible.go
@@ -0,0 +1,85 @@
+package compress
+
+import "math"
+
+// Estimate returns a normalized compressibility estimate of block b.
+// Values close to zero are likely uncompressible.
+// Values above 0.1 are likely to be compressible.
+// Values above 0.5 are very compressible.
+// Very small lengths will return 0.
+func Estimate(b []byte) float64 {
+	if len(b) < 16 {
+		return 0
+	}
+
+	// Correctly predicted order 1
+	hits := 0
+	lastMatch := false
+	var o1 [256]byte
+	var hist [256]int
+	c1 := byte(0)
+	for _, c := range b {
+		if c == o1[c1] {
+			// We only count a hit if there was two correct predictions in a row.
+			if lastMatch {
+				hits++
+			}
+			lastMatch = true
+		} else {
+			lastMatch = false
+		}
+		o1[c1] = c
+		c1 = c
+		hist[c]++
+	}
+
+	// Use x^0.6 to give better spread
+	prediction := math.Pow(float64(hits)/float64(len(b)), 0.6)
+
+	// Calculate histogram distribution
+	variance := float64(0)
+	avg := float64(len(b)) / 256
+
+	for _, v := range hist {
+		Δ := float64(v) - avg
+		variance += Δ * Δ
+	}
+
+	stddev := math.Sqrt(float64(variance)) / float64(len(b))
+	exp := math.Sqrt(1 / float64(len(b)))
+
+	// Subtract expected stddev
+	stddev -= exp
+	if stddev < 0 {
+		stddev = 0
+	}
+	stddev *= 1 + exp
+
+	// Use x^0.4 to give better spread
+	entropy := math.Pow(stddev, 0.4)
+
+	// 50/50 weight between prediction and histogram distribution
+	return math.Pow((prediction+entropy)/2, 0.9)
+}
+
+// ShannonEntropyBits returns the number of bits minimum required to represent
+// an entropy encoding of the input bytes.
+// https://en.wiktionary.org/wiki/Shannon_entropy
+func ShannonEntropyBits(b []byte) int {
+	if len(b) == 0 {
+		return 0
+	}
+	var hist [256]int
+	for _, c := range b {
+		hist[c]++
+	}
+	shannon := float64(0)
+	invTotal := 1.0 / float64(len(b))
+	for _, v := range hist[:] {
+		if v > 0 {
+			n := float64(v)
+			shannon += math.Ceil(-math.Log2(n*invTotal) * n)
+		}
+	}
+	return int(math.Ceil(shannon))
+}
diff --git a/vendor/github.com/klauspost/compress/gen.sh b/vendor/github.com/klauspost/compress/gen.sh
new file mode 100644
index 0000000..aff9422
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/gen.sh
@@ -0,0 +1,4 @@
+#!/bin/sh
+
+cd s2/cmd/_s2sx/ || exit 1
+go generate .
diff --git a/vendor/github.com/klauspost/compress/huff0/bitreader.go b/vendor/github.com/klauspost/compress/huff0/bitreader.go
index a4979e8..504a7be 100644
--- a/vendor/github.com/klauspost/compress/huff0/bitreader.go
+++ b/vendor/github.com/klauspost/compress/huff0/bitreader.go
@@ -8,118 +8,13 @@
 import (
 	"encoding/binary"
 	"errors"
+	"fmt"
 	"io"
 )
 
 // bitReader reads a bitstream in reverse.
 // The last set bit indicates the start of the stream and is used
 // for aligning the input.
-type bitReader 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 *bitReader) 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.bitsRead += 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 *bitReader) peekBitsFast(n uint8) uint16 {
-	const regMask = 64 - 1
-	v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
-	return v
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReader) 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 = (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:]
-		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
-		return
-	}
-	for b.off > 0 {
-		b.value = (b.value << 8) | uint64(b.in[b.off-1])
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReader) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReader) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	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]
@@ -213,10 +108,17 @@
 	return b.off == 0 && b.bitsRead >= 64
 }
 
+func (b *bitReaderBytes) remaining() uint {
+	return b.off*8 + uint(64-b.bitsRead)
+}
+
 // 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.remaining() > 0 {
+		return fmt.Errorf("corrupt input: %d bits remain on stream", b.remaining())
+	}
 	if b.bitsRead > 64 {
 		return io.ErrUnexpectedEOF
 	}
@@ -313,15 +215,17 @@
 	}
 }
 
-// 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
+func (b *bitReaderShifted) remaining() uint {
+	return b.off*8 + uint(64-b.bitsRead)
 }
 
 // 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.remaining() > 0 {
+		return fmt.Errorf("corrupt input: %d bits remain on stream", b.remaining())
+	}
 	if b.bitsRead > 64 {
 		return io.ErrUnexpectedEOF
 	}
diff --git a/vendor/github.com/klauspost/compress/huff0/bitwriter.go b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
index 6bce4e8..ec71f7a 100644
--- a/vendor/github.com/klauspost/compress/huff0/bitwriter.go
+++ b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
@@ -5,8 +5,6 @@
 
 package huff0
 
-import "fmt"
-
 // bitWriter will write bits.
 // First bit will be LSB of the first byte of output.
 type bitWriter struct {
@@ -23,14 +21,6 @@
 	0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
 	0xFFFF, 0xFFFF} /* up to 16 bits */
 
-// addBits16NC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16NC(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63)
-	b.nBits += bits
-}
-
 // addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated.
 // It will not check if there is space for them, so the caller must ensure that it has flushed recently.
 func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
@@ -70,104 +60,6 @@
 	b.nBits += encA.nBits + encB.nBits
 }
 
-// addBits16ZeroNC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-// This is fastest if bits can be zero.
-func (b *bitWriter) addBits16ZeroNC(value uint16, bits uint8) {
-	if bits == 0 {
-		return
-	}
-	value <<= (16 - bits) & 15
-	value >>= (16 - bits) & 15
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// flush will flush all pending full bytes.
-// There will be at least 56 bits available for writing when this has been called.
-// Using flush32 is faster, but leaves less space for writing.
-func (b *bitWriter) flush() {
-	v := b.nBits >> 3
-	switch v {
-	case 0:
-		return
-	case 1:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-		)
-		b.bitContainer >>= 1 << 3
-	case 2:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-		)
-		b.bitContainer >>= 2 << 3
-	case 3:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-		)
-		b.bitContainer >>= 3 << 3
-	case 4:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-		)
-		b.bitContainer >>= 4 << 3
-	case 5:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-		)
-		b.bitContainer >>= 5 << 3
-	case 6:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-		)
-		b.bitContainer >>= 6 << 3
-	case 7:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-		)
-		b.bitContainer >>= 7 << 3
-	case 8:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-			byte(b.bitContainer>>56),
-		)
-		b.bitContainer = 0
-		b.nBits = 0
-		return
-	default:
-		panic(fmt.Errorf("bits (%d) > 64", b.nBits))
-	}
-	b.nBits &= 7
-}
-
 // flush32 will flush out, so there are at least 32 bits available for writing.
 func (b *bitWriter) flush32() {
 	if b.nBits < 32 {
@@ -201,10 +93,3 @@
 	b.flushAlign()
 	return nil
 }
-
-// reset and continue writing by appending to out.
-func (b *bitWriter) reset(out []byte) {
-	b.bitContainer = 0
-	b.nBits = 0
-	b.out = out
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/bytereader.go b/vendor/github.com/klauspost/compress/huff0/bytereader.go
index 50bcdf6..4dcab8d 100644
--- a/vendor/github.com/klauspost/compress/huff0/bytereader.go
+++ b/vendor/github.com/klauspost/compress/huff0/bytereader.go
@@ -20,11 +20,6 @@
 	b.off = 0
 }
 
-// advance the stream b n bytes.
-func (b *byteReader) advance(n uint) {
-	b.off += int(n)
-}
-
 // Int32 returns a little endian int32 starting at current offset.
 func (b byteReader) Int32() int32 {
 	v3 := int32(b.b[b.off+3])
@@ -43,11 +38,6 @@
 	return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0
 }
 
-// unread returns the unread portion of the input.
-func (b byteReader) unread() []byte {
-	return b.b[b.off:]
-}
-
 // remain will return the number of bytes remaining.
 func (b byteReader) remain() int {
 	return len(b.b) - b.off
diff --git a/vendor/github.com/klauspost/compress/huff0/compress.go b/vendor/github.com/klauspost/compress/huff0/compress.go
index 0823c92..4d14542 100644
--- a/vendor/github.com/klauspost/compress/huff0/compress.go
+++ b/vendor/github.com/klauspost/compress/huff0/compress.go
@@ -2,6 +2,7 @@
 
 import (
 	"fmt"
+	"math"
 	"runtime"
 	"sync"
 )
@@ -161,6 +162,70 @@
 	return s.Out, false, nil
 }
 
+// EstimateSizes will estimate the data sizes
+func EstimateSizes(in []byte, s *Scratch) (tableSz, dataSz, reuseSz int, err error) {
+	s, err = s.prepare(in)
+	if err != nil {
+		return 0, 0, 0, err
+	}
+
+	// Create histogram, if none was provided.
+	tableSz, dataSz, reuseSz = -1, -1, -1
+	maxCount := s.maxCount
+	var canReuse = false
+	if maxCount == 0 {
+		maxCount, canReuse = s.countSimple(in)
+	} else {
+		canReuse = s.canUseTable(s.prevTable)
+	}
+
+	// We want the output size to be less than this:
+	wantSize := len(in)
+	if s.WantLogLess > 0 {
+		wantSize -= wantSize >> s.WantLogLess
+	}
+
+	// Reset for next run.
+	s.clearCount = true
+	s.maxCount = 0
+	if maxCount >= len(in) {
+		if maxCount > len(in) {
+			return 0, 0, 0, fmt.Errorf("maxCount (%d) > length (%d)", maxCount, len(in))
+		}
+		if len(in) == 1 {
+			return 0, 0, 0, ErrIncompressible
+		}
+		// One symbol, use RLE
+		return 0, 0, 0, ErrUseRLE
+	}
+	if maxCount == 1 || maxCount < (len(in)>>7) {
+		// Each symbol present maximum once or too well distributed.
+		return 0, 0, 0, ErrIncompressible
+	}
+
+	// Calculate new table.
+	err = s.buildCTable()
+	if err != nil {
+		return 0, 0, 0, err
+	}
+
+	if false && !s.canUseTable(s.cTable) {
+		panic("invalid table generated")
+	}
+
+	tableSz, err = s.cTable.estTableSize(s)
+	if err != nil {
+		return 0, 0, 0, err
+	}
+	if canReuse {
+		reuseSz = s.prevTable.estimateSize(s.count[:s.symbolLen])
+	}
+	dataSz = s.cTable.estimateSize(s.count[:s.symbolLen])
+
+	// Restore
+	return tableSz, dataSz, reuseSz, nil
+}
+
 func (s *Scratch) compress1X(src []byte) ([]byte, error) {
 	return s.compress1xDo(s.Out, src)
 }
@@ -225,6 +290,10 @@
 		if err != nil {
 			return nil, err
 		}
+		if len(s.Out)-idx > math.MaxUint16 {
+			// We cannot store the size in the jump table
+			return nil, ErrIncompressible
+		}
 		// Write compressed length as little endian before block.
 		if i < 3 {
 			// Last length is not written.
@@ -268,6 +337,10 @@
 			return nil, errs[i]
 		}
 		o := s.tmpOut[i]
+		if len(o) > math.MaxUint16 {
+			// We cannot store the size in the jump table
+			return nil, ErrIncompressible
+		}
 		// Write compressed length as little endian before block.
 		if i < 3 {
 			// Last length is not written.
@@ -331,6 +404,7 @@
 	return true
 }
 
+//lint:ignore U1000 used for debugging
 func (s *Scratch) validateTable(c cTable) bool {
 	if len(c) < int(s.symbolLen) {
 		return false
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress.go b/vendor/github.com/klauspost/compress/huff0/decompress.go
index 41703bb..c0c48bd 100644
--- a/vendor/github.com/klauspost/compress/huff0/decompress.go
+++ b/vendor/github.com/klauspost/compress/huff0/decompress.go
@@ -4,13 +4,13 @@
 	"errors"
 	"fmt"
 	"io"
+	"sync"
 
 	"github.com/klauspost/compress/fse"
 )
 
 type dTable struct {
 	single []dEntrySingle
-	double []dEntryDouble
 }
 
 // single-symbols decoding
@@ -18,13 +18,6 @@
 	entry uint16
 }
 
-// double-symbols decoding
-type dEntryDouble struct {
-	seq   uint16
-	nBits uint8
-	len   uint8
-}
-
 // Uses special code for all tables that are < 8 bits.
 const use8BitTables = true
 
@@ -34,7 +27,7 @@
 // If no Scratch is provided a new one is allocated.
 // 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)
+	s, err = s.prepare(nil)
 	if err != nil {
 		return s, nil, err
 	}
@@ -216,6 +209,7 @@
 	return &Decoder{
 		dt:             s.dt,
 		actualTableLog: s.actualTableLog,
+		bufs:           &s.decPool,
 	}
 }
 
@@ -223,103 +217,15 @@
 type Decoder struct {
 	dt             dTable
 	actualTableLog uint8
+	bufs           *sync.Pool
 }
 
-// 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")
+func (d *Decoder) buffer() *[4][256]byte {
+	buf, ok := d.bufs.Get().(*[4][256]byte)
+	if ok {
+		return buf
 	}
-	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()
+	return &[4][256]byte{}
 }
 
 // decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8.
@@ -341,41 +247,258 @@
 	dt := d.dt.single[:256]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
+	bufs := d.buffer()
+	buf := &bufs[0]
 	var off uint8
 
-	shift := (8 - d.actualTableLog) & 7
+	switch d.actualTableLog {
+	case 8:
+		const shift = 8 - 8
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
 
-	//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[uint8(br.value>>(56+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+1] = uint8(v.entry >> 8)
+			v = dt[uint8(br.value>>(56+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+2] = uint8(v.entry >> 8)
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = 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
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					br.close()
+					d.bufs.Put(bufs)
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
 			}
-			dst = append(dst, buf[:]...)
 		}
+	case 7:
+		const shift = 8 - 7
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+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()
+					d.bufs.Put(bufs)
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 6:
+		const shift = 8 - 6
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 5:
+		const shift = 8 - 5
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 4:
+		const shift = 8 - 4
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 3:
+		const shift = 8 - 3
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 2:
+		const shift = 8 - 2
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	case 1:
+		const shift = 8 - 1
+		for br.off >= 4 {
+			br.fillFast()
+			v := dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+0] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+1] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+2] = uint8(v.entry >> 8)
+
+			v = dt[uint8(br.value>>(56+shift))]
+			br.advance(uint8(v.entry))
+			buf[off+3] = uint8(v.entry >> 8)
+
+			off += 4
+			if off == 0 {
+				if len(dst)+256 > maxDecodedSize {
+					d.bufs.Put(bufs)
+					br.close()
+					return nil, ErrMaxDecodedSizeExceeded
+				}
+				dst = append(dst, buf[:]...)
+			}
+		}
+	default:
+		d.bufs.Put(bufs)
+		return nil, fmt.Errorf("invalid tablelog: %d", d.actualTableLog)
 	}
 
 	if len(dst)+int(off) > maxDecodedSize {
+		d.bufs.Put(bufs)
 		br.close()
 		return nil, ErrMaxDecodedSizeExceeded
 	}
@@ -383,6 +506,8 @@
 
 	// br < 4, so uint8 is fine
 	bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead))
+	shift := (8 - d.actualTableLog) & 7
+
 	for bitsLeft > 0 {
 		if br.bitsRead >= 64-8 {
 			for br.off > 0 {
@@ -393,6 +518,7 @@
 		}
 		if len(dst) >= maxDecodedSize {
 			br.close()
+			d.bufs.Put(bufs)
 			return nil, ErrMaxDecodedSizeExceeded
 		}
 		v := dt[br.peekByteFast()>>shift]
@@ -401,6 +527,7 @@
 		bitsLeft -= int8(nBits)
 		dst = append(dst, uint8(v.entry>>8))
 	}
+	d.bufs.Put(bufs)
 	return dst, br.close()
 }
 
@@ -420,33 +547,35 @@
 	dt := d.dt.single[:256]
 
 	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
+	bufs := d.buffer()
+	buf := &bufs[0]
 	var off uint8
 
-	const shift = 0
+	const shift = 56
 
 	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
 	for br.off >= 4 {
 		br.fillFast()
-		v := dt[br.peekByteFast()>>shift]
+		v := dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+0] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
+		v = dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+1] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
+		v = dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+2] = uint8(v.entry >> 8)
 
-		v = dt[br.peekByteFast()>>shift]
+		v = dt[uint8(br.value>>shift)]
 		br.advance(uint8(v.entry))
 		buf[off+3] = uint8(v.entry >> 8)
 
 		off += 4
 		if off == 0 {
 			if len(dst)+256 > maxDecodedSize {
+				d.bufs.Put(bufs)
 				br.close()
 				return nil, ErrMaxDecodedSizeExceeded
 			}
@@ -455,6 +584,7 @@
 	}
 
 	if len(dst)+int(off) > maxDecodedSize {
+		d.bufs.Put(bufs)
 		br.close()
 		return nil, ErrMaxDecodedSizeExceeded
 	}
@@ -471,15 +601,17 @@
 			}
 		}
 		if len(dst) >= maxDecodedSize {
+			d.bufs.Put(bufs)
 			br.close()
 			return nil, ErrMaxDecodedSizeExceeded
 		}
-		v := dt[br.peekByteFast()>>shift]
+		v := dt[br.peekByteFast()]
 		nBits := uint8(v.entry)
 		br.advance(nBits)
 		bitsLeft -= int8(nBits)
 		dst = append(dst, uint8(v.entry>>8))
 	}
+	d.bufs.Put(bufs)
 	return dst, br.close()
 }
 
@@ -487,196 +619,6 @@
 // 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(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 tlSize = 1 << tableLogMax
-	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 2 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
-		}
-
-		{
-			const stream = 0
-			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			val := br[stream].peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream] = uint8(v.entry >> 8)
-
-			val2 := br[stream2].peekBitsFast(d.actualTableLog)
-			v2 := single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
-
-			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[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
-			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			val := br[stream].peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream] = uint8(v.entry >> 8)
-
-			val2 := br[stream2].peekBitsFast(d.actualTableLog)
-			v2 := single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
-
-			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[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
-
-		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 := br.off*8 + uint(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--
-					}
-				}
-			}
-			// end inline...
-			if offset >= len(out) {
-				return nil, errors.New("corruption detected: stream overrun 4")
-			}
-
-			// 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
-		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) decompress4X8bit(dst, src []byte) ([]byte, error) {
 	if d.actualTableLog == 8 {
 		return d.decompress4X8bitExactly(dst, src)
@@ -706,19 +648,18 @@
 	out := dst
 	dstEvery := (dstSize + 3) / 4
 
-	shift := (8 - d.actualTableLog) & 7
+	shift := (56 + (8 - d.actualTableLog)) & 63
 
 	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
+	buf := d.buffer()
 	var off uint8
 	var decoded int
 
 	// Decode 4 values from each decoder/loop.
-	const bufoff = 256 / 4
+	const bufoff = 256
 	for {
 		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
 			break
@@ -728,96 +669,109 @@
 			// Interleave 2 decodes.
 			const stream = 0
 			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			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))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		{
 			const stream = 2
 			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			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))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		off += 4
 
-		if off == bufoff {
+		if off == 0 {
 			if bufoff > dstEvery {
+				d.bufs.Put(buf)
 				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
+			copy(out, buf[0][:])
+			copy(out[dstEvery:], buf[1][:])
+			copy(out[dstEvery*2:], buf[2][:])
+			copy(out[dstEvery*3:], buf[3][:])
 			out = out[bufoff:]
-			decoded += 256
+			decoded += bufoff * 4
 			// There must at least be 3 buffers left.
 			if len(out) < dstEvery*3 {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 2")
 			}
 		}
@@ -825,23 +779,31 @@
 	if off > 0 {
 		ioff := int(off)
 		if len(out) < dstEvery*3+ioff {
+			d.bufs.Put(buf)
 			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])
+		copy(out, buf[0][:off])
+		copy(out[dstEvery:], buf[1][:off])
+		copy(out[dstEvery*2:], buf[2][:off])
+		copy(out[dstEvery*3:], buf[3][:off])
 		decoded += int(off) * 4
 		out = out[off:]
 	}
 
 	// Decode remaining.
+	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
 	for i := range br {
 		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
 		br := &br[i]
-		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
+		bitsLeft := br.remaining()
 		for bitsLeft > 0 {
 			if br.finished() {
+				d.bufs.Put(buf)
 				return nil, io.ErrUnexpectedEOF
 			}
 			if br.bitsRead >= 56 {
@@ -861,24 +823,31 @@
 				}
 			}
 			// end inline...
-			if offset >= len(out) {
+			if offset >= endsAt {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 4")
 			}
 
 			// Read value and increment offset.
-			v := single[br.peekByteFast()>>shift].entry
+			v := single[uint8(br.value>>shift)].entry
 			nBits := uint8(v)
 			br.advance(nBits)
-			bitsLeft -= int(nBits)
+			bitsLeft -= uint(nBits)
 			out[offset] = uint8(v >> 8)
 			offset++
 		}
+		if offset != endsAt {
+			d.bufs.Put(buf)
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
 		decoded += offset - dstEvery*i
 		err = br.close()
 		if err != nil {
+			d.bufs.Put(buf)
 			return nil, err
 		}
 	}
+	d.bufs.Put(buf)
 	if dstSize != decoded {
 		return nil, errors.New("corruption detected: short output block")
 	}
@@ -914,18 +883,17 @@
 	out := dst
 	dstEvery := (dstSize + 3) / 4
 
-	const shift = 0
+	const shift = 56
 	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
+	buf := d.buffer()
 	var off uint8
 	var decoded int
 
 	// Decode 4 values from each decoder/loop.
-	const bufoff = 256 / 4
+	const bufoff = 256
 	for {
 		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
 			break
@@ -935,96 +903,109 @@
 			// Interleave 2 decodes.
 			const stream = 0
 			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			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))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		{
 			const stream = 2
 			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
+			br1 := &br[stream]
+			br2 := &br[stream2]
+			br1.fillFast()
+			br2.fillFast()
 
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v := single[uint8(br1.value>>shift)].entry
+			v2 := single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off] = uint8(v >> 8)
+			buf[stream2][off] = uint8(v2 >> 8)
 
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+1] = uint8(v >> 8)
+			buf[stream2][off+1] = uint8(v2 >> 8)
 
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+2] = uint8(v >> 8)
+			buf[stream2][off+2] = uint8(v2 >> 8)
 
-			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))
+			v = single[uint8(br1.value>>shift)].entry
+			v2 = single[uint8(br2.value>>shift)].entry
+			br1.bitsRead += uint8(v)
+			br1.value <<= v & 63
+			br2.bitsRead += uint8(v2)
+			br2.value <<= v2 & 63
+			buf[stream][off+3] = uint8(v >> 8)
+			buf[stream2][off+3] = uint8(v2 >> 8)
 		}
 
 		off += 4
 
-		if off == bufoff {
+		if off == 0 {
 			if bufoff > dstEvery {
+				d.bufs.Put(buf)
 				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
+			copy(out, buf[0][:])
+			copy(out[dstEvery:], buf[1][:])
+			copy(out[dstEvery*2:], buf[2][:])
+			copy(out[dstEvery*3:], buf[3][:])
 			out = out[bufoff:]
-			decoded += 256
+			decoded += bufoff * 4
 			// There must at least be 3 buffers left.
 			if len(out) < dstEvery*3 {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 2")
 			}
 		}
@@ -1034,21 +1015,27 @@
 		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])
+		copy(out, buf[0][:off])
+		copy(out[dstEvery:], buf[1][:off])
+		copy(out[dstEvery*2:], buf[2][:off])
+		copy(out[dstEvery*3:], buf[3][:off])
 		decoded += int(off) * 4
 		out = out[off:]
 	}
 
 	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
 	for i := range br {
 		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
 		br := &br[i]
-		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
+		bitsLeft := br.remaining()
 		for bitsLeft > 0 {
 			if br.finished() {
+				d.bufs.Put(buf)
 				return nil, io.ErrUnexpectedEOF
 			}
 			if br.bitsRead >= 56 {
@@ -1068,24 +1055,32 @@
 				}
 			}
 			// end inline...
-			if offset >= len(out) {
+			if offset >= endsAt {
+				d.bufs.Put(buf)
 				return nil, errors.New("corruption detected: stream overrun 4")
 			}
 
 			// Read value and increment offset.
-			v := single[br.peekByteFast()>>shift].entry
+			v := single[br.peekByteFast()].entry
 			nBits := uint8(v)
 			br.advance(nBits)
-			bitsLeft -= int(nBits)
+			bitsLeft -= uint(nBits)
 			out[offset] = uint8(v >> 8)
 			offset++
 		}
+		if offset != endsAt {
+			d.bufs.Put(buf)
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
+
 		decoded += offset - dstEvery*i
 		err = br.close()
 		if err != nil {
+			d.bufs.Put(buf)
 			return nil, err
 		}
 	}
+	d.bufs.Put(buf)
 	if dstSize != decoded {
 		return nil, errors.New("corruption detected: short output block")
 	}
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go
new file mode 100644
index 0000000..9f3e9f7
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.go
@@ -0,0 +1,222 @@
+//go:build amd64 && !appengine && !noasm && gc
+// +build amd64,!appengine,!noasm,gc
+
+// This file contains the specialisation of Decoder.Decompress4X
+// and Decoder.Decompress1X that use an asm implementation of thir main loops.
+package huff0
+
+import (
+	"errors"
+	"fmt"
+
+	"github.com/klauspost/compress/internal/cpuinfo"
+)
+
+// decompress4x_main_loop_x86 is an x86 assembler implementation
+// of Decompress4X when tablelog > 8.
+//go:noescape
+func decompress4x_main_loop_amd64(ctx *decompress4xContext)
+
+// decompress4x_8b_loop_x86 is an x86 assembler implementation
+// of Decompress4X when tablelog <= 8 which decodes 4 entries
+// per loop.
+//go:noescape
+func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext)
+
+// fallback8BitSize is the size where using Go version is faster.
+const fallback8BitSize = 800
+
+type decompress4xContext struct {
+	pbr      *[4]bitReaderShifted
+	peekBits uint8
+	out      *byte
+	dstEvery int
+	tbl      *dEntrySingle
+	decoded  int
+	limit    *byte
+}
+
+// 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")
+	}
+
+	use8BitTables := d.actualTableLog <= 8
+	if cap(dst) < fallback8BitSize && use8BitTables {
+		return d.decompress4X8bit(dst, src)
+	}
+
+	var br [4]bitReaderShifted
+	// Decode "jump table"
+	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 tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+	single := d.dt.single[:tlSize]
+
+	var decoded int
+
+	if len(out) > 4*4 && !(br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4) {
+		ctx := decompress4xContext{
+			pbr:      &br,
+			peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast()
+			out:      &out[0],
+			dstEvery: dstEvery,
+			tbl:      &single[0],
+			limit:    &out[dstEvery-4], // Always stop decoding when first buffer gets here to avoid writing OOB on last.
+		}
+		if use8BitTables {
+			decompress4x_8b_main_loop_amd64(&ctx)
+		} else {
+			decompress4x_main_loop_amd64(&ctx)
+		}
+
+		decoded = ctx.decoded
+		out = out[decoded/4:]
+	}
+
+	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
+	for i := range br {
+		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
+		br := &br[i]
+		bitsLeft := br.remaining()
+		for bitsLeft > 0 {
+			br.fill()
+			if offset >= endsAt {
+				return nil, errors.New("corruption detected: stream overrun 4")
+			}
+
+			// 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++
+		}
+		if offset != endsAt {
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
+		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_main_loop_x86 is an x86 assembler implementation
+// of Decompress1X when tablelog > 8.
+//go:noescape
+func decompress1x_main_loop_amd64(ctx *decompress1xContext)
+
+// decompress4x_main_loop_x86 is an x86 with BMI2 assembler implementation
+// of Decompress1X when tablelog > 8.
+//go:noescape
+func decompress1x_main_loop_bmi2(ctx *decompress1xContext)
+
+type decompress1xContext struct {
+	pbr      *bitReaderShifted
+	peekBits uint8
+	out      *byte
+	outCap   int
+	tbl      *dEntrySingle
+	decoded  int
+}
+
+// Error reported by asm implementations
+const error_max_decoded_size_exeeded = -1
+
+// 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")
+	}
+	var br bitReaderShifted
+	err := br.init(src)
+	if err != nil {
+		return dst, err
+	}
+	maxDecodedSize := cap(dst)
+	dst = dst[:maxDecodedSize]
+
+	const tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+
+	if maxDecodedSize >= 4 {
+		ctx := decompress1xContext{
+			pbr:      &br,
+			out:      &dst[0],
+			outCap:   maxDecodedSize,
+			peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast()
+			tbl:      &d.dt.single[0],
+		}
+
+		if cpuinfo.HasBMI2() {
+			decompress1x_main_loop_bmi2(&ctx)
+		} else {
+			decompress1x_main_loop_amd64(&ctx)
+		}
+		if ctx.decoded == error_max_decoded_size_exeeded {
+			return nil, ErrMaxDecodedSizeExceeded
+		}
+
+		dst = dst[:ctx.decoded]
+	}
+
+	// br < 8, so uint8 is fine
+	bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead
+	for bitsLeft > 0 {
+		br.fill()
+		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()
+}
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s
new file mode 100644
index 0000000..dd1a5ae
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/huff0/decompress_amd64.s
@@ -0,0 +1,847 @@
+// Code generated by command: go run gen.go -out ../decompress_amd64.s -pkg=huff0. DO NOT EDIT.
+
+//go:build amd64 && !appengine && !noasm && gc
+// +build amd64,!appengine,!noasm,gc
+
+// func decompress4x_main_loop_amd64(ctx *decompress4xContext)
+TEXT ·decompress4x_main_loop_amd64(SB), $0-8
+	XORQ DX, DX
+
+	// Preload values
+	MOVQ    ctx+0(FP), AX
+	MOVBQZX 8(AX), DI
+	MOVQ    16(AX), SI
+	MOVQ    48(AX), BX
+	MOVQ    24(AX), R9
+	MOVQ    32(AX), R10
+	MOVQ    (AX), R11
+
+	// Main loop
+main_loop:
+	MOVQ  SI, R8
+	CMPQ  R8, BX
+	SETGE DL
+
+	// br0.fillFast32()
+	MOVQ    32(R11), R12
+	MOVBQZX 40(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill0
+	MOVQ    24(R11), AX
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, AX
+	MOVQ    (R11), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R14*1), R14
+	MOVQ R13, CX
+	SHLQ CL, R14
+	MOVQ AX, 24(R11)
+	ORQ  R14, R12
+
+	// exhausted = exhausted || (br0.off < 4)
+	CMPQ  AX, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill0:
+	// val0 := br0.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br0.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br0.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R12, R14
+	SHRQ CL, R14
+
+	// v1 := table[val1&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br0.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (R8)
+
+	// update the bitreader structure
+	MOVQ R12, 32(R11)
+	MOVB R13, 40(R11)
+	ADDQ R9, R8
+
+	// br1.fillFast32()
+	MOVQ    80(R11), R12
+	MOVBQZX 88(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill1
+	MOVQ    72(R11), AX
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, AX
+	MOVQ    48(R11), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R14*1), R14
+	MOVQ R13, CX
+	SHLQ CL, R14
+	MOVQ AX, 72(R11)
+	ORQ  R14, R12
+
+	// exhausted = exhausted || (br1.off < 4)
+	CMPQ  AX, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill1:
+	// val0 := br1.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br1.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br1.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R12, R14
+	SHRQ CL, R14
+
+	// v1 := table[val1&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br1.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (R8)
+
+	// update the bitreader structure
+	MOVQ R12, 80(R11)
+	MOVB R13, 88(R11)
+	ADDQ R9, R8
+
+	// br2.fillFast32()
+	MOVQ    128(R11), R12
+	MOVBQZX 136(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill2
+	MOVQ    120(R11), AX
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, AX
+	MOVQ    96(R11), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R14*1), R14
+	MOVQ R13, CX
+	SHLQ CL, R14
+	MOVQ AX, 120(R11)
+	ORQ  R14, R12
+
+	// exhausted = exhausted || (br2.off < 4)
+	CMPQ  AX, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill2:
+	// val0 := br2.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br2.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br2.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R12, R14
+	SHRQ CL, R14
+
+	// v1 := table[val1&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br2.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (R8)
+
+	// update the bitreader structure
+	MOVQ R12, 128(R11)
+	MOVB R13, 136(R11)
+	ADDQ R9, R8
+
+	// br3.fillFast32()
+	MOVQ    176(R11), R12
+	MOVBQZX 184(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill3
+	MOVQ    168(R11), AX
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, AX
+	MOVQ    144(R11), R14
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (AX)(R14*1), R14
+	MOVQ R13, CX
+	SHLQ CL, R14
+	MOVQ AX, 168(R11)
+	ORQ  R14, R12
+
+	// exhausted = exhausted || (br3.off < 4)
+	CMPQ  AX, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill3:
+	// val0 := br3.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br3.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br3.peekTopBits(peekBits)
+	MOVQ DI, CX
+	MOVQ R12, R14
+	SHRQ CL, R14
+
+	// v1 := table[val1&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br3.advance(uint8(v1.entry))
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// these two writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	MOVW AX, (R8)
+
+	// update the bitreader structure
+	MOVQ  R12, 176(R11)
+	MOVB  R13, 184(R11)
+	ADDQ  $0x02, SI
+	TESTB DL, DL
+	JZ    main_loop
+	MOVQ  ctx+0(FP), AX
+	SUBQ  16(AX), SI
+	SHLQ  $0x02, SI
+	MOVQ  SI, 40(AX)
+	RET
+
+// func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext)
+TEXT ·decompress4x_8b_main_loop_amd64(SB), $0-8
+	XORQ DX, DX
+
+	// Preload values
+	MOVQ    ctx+0(FP), CX
+	MOVBQZX 8(CX), DI
+	MOVQ    16(CX), BX
+	MOVQ    48(CX), SI
+	MOVQ    24(CX), R9
+	MOVQ    32(CX), R10
+	MOVQ    (CX), R11
+
+	// Main loop
+main_loop:
+	MOVQ  BX, R8
+	CMPQ  R8, SI
+	SETGE DL
+
+	// br0.fillFast32()
+	MOVQ    32(R11), R12
+	MOVBQZX 40(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill0
+	MOVQ    24(R11), R14
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, R14
+	MOVQ    (R11), R15
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R14)(R15*1), R15
+	MOVQ R13, CX
+	SHLQ CL, R15
+	MOVQ R14, 24(R11)
+	ORQ  R15, R12
+
+	// exhausted = exhausted || (br0.off < 4)
+	CMPQ  R14, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill0:
+	// val0 := br0.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br0.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br0.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v1 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br0.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// val2 := br0.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v2 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br0.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val3 := br0.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v3 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br0.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (R8)
+
+	// update the bitreader structure
+	MOVQ R12, 32(R11)
+	MOVB R13, 40(R11)
+	ADDQ R9, R8
+
+	// br1.fillFast32()
+	MOVQ    80(R11), R12
+	MOVBQZX 88(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill1
+	MOVQ    72(R11), R14
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, R14
+	MOVQ    48(R11), R15
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R14)(R15*1), R15
+	MOVQ R13, CX
+	SHLQ CL, R15
+	MOVQ R14, 72(R11)
+	ORQ  R15, R12
+
+	// exhausted = exhausted || (br1.off < 4)
+	CMPQ  R14, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill1:
+	// val0 := br1.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br1.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br1.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v1 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br1.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// val2 := br1.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v2 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br1.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val3 := br1.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v3 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br1.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (R8)
+
+	// update the bitreader structure
+	MOVQ R12, 80(R11)
+	MOVB R13, 88(R11)
+	ADDQ R9, R8
+
+	// br2.fillFast32()
+	MOVQ    128(R11), R12
+	MOVBQZX 136(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill2
+	MOVQ    120(R11), R14
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, R14
+	MOVQ    96(R11), R15
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R14)(R15*1), R15
+	MOVQ R13, CX
+	SHLQ CL, R15
+	MOVQ R14, 120(R11)
+	ORQ  R15, R12
+
+	// exhausted = exhausted || (br2.off < 4)
+	CMPQ  R14, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill2:
+	// val0 := br2.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br2.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br2.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v1 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br2.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// val2 := br2.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v2 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br2.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val3 := br2.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v3 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br2.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (R8)
+
+	// update the bitreader structure
+	MOVQ R12, 128(R11)
+	MOVB R13, 136(R11)
+	ADDQ R9, R8
+
+	// br3.fillFast32()
+	MOVQ    176(R11), R12
+	MOVBQZX 184(R11), R13
+	CMPQ    R13, $0x20
+	JBE     skip_fill3
+	MOVQ    168(R11), R14
+	SUBQ    $0x20, R13
+	SUBQ    $0x04, R14
+	MOVQ    144(R11), R15
+
+	// b.value |= uint64(low) << (b.bitsRead & 63)
+	MOVL (R14)(R15*1), R15
+	MOVQ R13, CX
+	SHLQ CL, R15
+	MOVQ R14, 168(R11)
+	ORQ  R15, R12
+
+	// exhausted = exhausted || (br3.off < 4)
+	CMPQ  R14, $0x04
+	SETLT AL
+	ORB   AL, DL
+
+skip_fill3:
+	// val0 := br3.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v0 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br3.advance(uint8(v0.entry)
+	MOVB CH, AL
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val1 := br3.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v1 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br3.advance(uint8(v1.entry)
+	MOVB   CH, AH
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// val2 := br3.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v2 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br3.advance(uint8(v2.entry)
+	MOVB CH, AH
+	SHLQ CL, R12
+	ADDB CL, R13
+
+	// val3 := br3.peekTopBits(peekBits)
+	MOVQ R12, R14
+	MOVQ DI, CX
+	SHRQ CL, R14
+
+	// v3 := table[val0&mask]
+	MOVW (R10)(R14*2), CX
+
+	// br3.advance(uint8(v3.entry)
+	MOVB   CH, AL
+	SHLQ   CL, R12
+	ADDB   CL, R13
+	BSWAPL AX
+
+	// these four writes get coalesced
+	// out[id * dstEvery + 0] = uint8(v0.entry >> 8)
+	// out[id * dstEvery + 1] = uint8(v1.entry >> 8)
+	// out[id * dstEvery + 3] = uint8(v2.entry >> 8)
+	// out[id * dstEvery + 4] = uint8(v3.entry >> 8)
+	MOVL AX, (R8)
+
+	// update the bitreader structure
+	MOVQ  R12, 176(R11)
+	MOVB  R13, 184(R11)
+	ADDQ  $0x04, BX
+	TESTB DL, DL
+	JZ    main_loop
+	MOVQ  ctx+0(FP), AX
+	SUBQ  16(AX), BX
+	SHLQ  $0x02, BX
+	MOVQ  BX, 40(AX)
+	RET
+
+// func decompress1x_main_loop_amd64(ctx *decompress1xContext)
+TEXT ·decompress1x_main_loop_amd64(SB), $0-8
+	MOVQ    ctx+0(FP), CX
+	MOVQ    16(CX), DX
+	MOVQ    24(CX), BX
+	CMPQ    BX, $0x04
+	JB      error_max_decoded_size_exeeded
+	LEAQ    (DX)(BX*1), BX
+	MOVQ    (CX), SI
+	MOVQ    (SI), R8
+	MOVQ    24(SI), R9
+	MOVQ    32(SI), R10
+	MOVBQZX 40(SI), R11
+	MOVQ    32(CX), SI
+	MOVBQZX 8(CX), DI
+	JMP     loop_condition
+
+main_loop:
+	// Check if we have room for 4 bytes in the output buffer
+	LEAQ 4(DX), CX
+	CMPQ CX, BX
+	JGE  error_max_decoded_size_exeeded
+
+	// Decode 4 values
+	CMPQ R11, $0x20
+	JL   bitReader_fillFast_1_end
+	SUBQ $0x20, R11
+	SUBQ $0x04, R9
+	MOVL (R8)(R9*1), R12
+	MOVQ R11, CX
+	SHLQ CL, R12
+	ORQ  R12, R10
+
+bitReader_fillFast_1_end:
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	BSWAPL  AX
+	CMPQ    R11, $0x20
+	JL      bitReader_fillFast_2_end
+	SUBQ    $0x20, R11
+	SUBQ    $0x04, R9
+	MOVL    (R8)(R9*1), R12
+	MOVQ    R11, CX
+	SHLQ    CL, R12
+	ORQ     R12, R10
+
+bitReader_fillFast_2_end:
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	MOVQ    DI, CX
+	MOVQ    R10, R12
+	SHRQ    CL, R12
+	MOVW    (SI)(R12*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLQ    CL, R10
+	BSWAPL  AX
+
+	// Store the decoded values
+	MOVL AX, (DX)
+	ADDQ $0x04, DX
+
+loop_condition:
+	CMPQ R9, $0x08
+	JGE  main_loop
+
+	// Update ctx structure
+	MOVQ ctx+0(FP), AX
+	SUBQ 16(AX), DX
+	MOVQ DX, 40(AX)
+	MOVQ (AX), AX
+	MOVQ R9, 24(AX)
+	MOVQ R10, 32(AX)
+	MOVB R11, 40(AX)
+	RET
+
+	// Report error
+error_max_decoded_size_exeeded:
+	MOVQ ctx+0(FP), AX
+	MOVQ $-1, CX
+	MOVQ CX, 40(AX)
+	RET
+
+// func decompress1x_main_loop_bmi2(ctx *decompress1xContext)
+// Requires: BMI2
+TEXT ·decompress1x_main_loop_bmi2(SB), $0-8
+	MOVQ    ctx+0(FP), CX
+	MOVQ    16(CX), DX
+	MOVQ    24(CX), BX
+	CMPQ    BX, $0x04
+	JB      error_max_decoded_size_exeeded
+	LEAQ    (DX)(BX*1), BX
+	MOVQ    (CX), SI
+	MOVQ    (SI), R8
+	MOVQ    24(SI), R9
+	MOVQ    32(SI), R10
+	MOVBQZX 40(SI), R11
+	MOVQ    32(CX), SI
+	MOVBQZX 8(CX), DI
+	JMP     loop_condition
+
+main_loop:
+	// Check if we have room for 4 bytes in the output buffer
+	LEAQ 4(DX), CX
+	CMPQ CX, BX
+	JGE  error_max_decoded_size_exeeded
+
+	// Decode 4 values
+	CMPQ  R11, $0x20
+	JL    bitReader_fillFast_1_end
+	SUBQ  $0x20, R11
+	SUBQ  $0x04, R9
+	MOVL  (R8)(R9*1), CX
+	SHLXQ R11, CX, CX
+	ORQ   CX, R10
+
+bitReader_fillFast_1_end:
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	BSWAPL  AX
+	CMPQ    R11, $0x20
+	JL      bitReader_fillFast_2_end
+	SUBQ    $0x20, R11
+	SUBQ    $0x04, R9
+	MOVL    (R8)(R9*1), CX
+	SHLXQ   R11, CX, CX
+	ORQ     CX, R10
+
+bitReader_fillFast_2_end:
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AH
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	SHRXQ   DI, R10, CX
+	MOVW    (SI)(CX*2), CX
+	MOVB    CH, AL
+	MOVBQZX CL, CX
+	ADDQ    CX, R11
+	SHLXQ   CX, R10, R10
+	BSWAPL  AX
+
+	// Store the decoded values
+	MOVL AX, (DX)
+	ADDQ $0x04, DX
+
+loop_condition:
+	CMPQ R9, $0x08
+	JGE  main_loop
+
+	// Update ctx structure
+	MOVQ ctx+0(FP), AX
+	SUBQ 16(AX), DX
+	MOVQ DX, 40(AX)
+	MOVQ (AX), AX
+	MOVQ R9, 24(AX)
+	MOVQ R10, 32(AX)
+	MOVB R11, 40(AX)
+	RET
+
+	// Report error
+error_max_decoded_size_exeeded:
+	MOVQ ctx+0(FP), AX
+	MOVQ $-1, CX
+	MOVQ CX, 40(AX)
+	RET
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress_generic.go b/vendor/github.com/klauspost/compress/huff0/decompress_generic.go
new file mode 100644
index 0000000..4f6f37c
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/huff0/decompress_generic.go
@@ -0,0 +1,295 @@
+//go:build !amd64 || appengine || !gc || noasm
+// +build !amd64 appengine !gc noasm
+
+// This file contains a generic implementation of Decoder.Decompress4X.
+package huff0
+
+import (
+	"errors"
+	"fmt"
+)
+
+// 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
+	// Decode "jump table"
+	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 tlSize = 1 << tableLogMax
+	const tlMask = tlSize - 1
+	single := d.dt.single[:tlSize]
+
+	// Use temp table to avoid bound checks/append penalty.
+	buf := d.buffer()
+	var off uint8
+	var decoded int
+
+	// Decode 2 values from each decoder/loop.
+	const bufoff = 256
+	for {
+		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
+			break
+		}
+
+		{
+			const stream = 0
+			const stream2 = 1
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			val := br[stream].peekBitsFast(d.actualTableLog)
+			val2 := br[stream2].peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask]
+			v2 := single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off] = uint8(v.entry >> 8)
+			buf[stream2][off] = uint8(v2.entry >> 8)
+
+			val = br[stream].peekBitsFast(d.actualTableLog)
+			val2 = br[stream2].peekBitsFast(d.actualTableLog)
+			v = single[val&tlMask]
+			v2 = single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off+1] = uint8(v.entry >> 8)
+			buf[stream2][off+1] = uint8(v2.entry >> 8)
+		}
+
+		{
+			const stream = 2
+			const stream2 = 3
+			br[stream].fillFast()
+			br[stream2].fillFast()
+
+			val := br[stream].peekBitsFast(d.actualTableLog)
+			val2 := br[stream2].peekBitsFast(d.actualTableLog)
+			v := single[val&tlMask]
+			v2 := single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off] = uint8(v.entry >> 8)
+			buf[stream2][off] = uint8(v2.entry >> 8)
+
+			val = br[stream].peekBitsFast(d.actualTableLog)
+			val2 = br[stream2].peekBitsFast(d.actualTableLog)
+			v = single[val&tlMask]
+			v2 = single[val2&tlMask]
+			br[stream].advance(uint8(v.entry))
+			br[stream2].advance(uint8(v2.entry))
+			buf[stream][off+1] = uint8(v.entry >> 8)
+			buf[stream2][off+1] = uint8(v2.entry >> 8)
+		}
+
+		off += 2
+
+		if off == 0 {
+			if bufoff > dstEvery {
+				d.bufs.Put(buf)
+				return nil, errors.New("corruption detected: stream overrun 1")
+			}
+			copy(out, buf[0][:])
+			copy(out[dstEvery:], buf[1][:])
+			copy(out[dstEvery*2:], buf[2][:])
+			copy(out[dstEvery*3:], buf[3][:])
+			out = out[bufoff:]
+			decoded += bufoff * 4
+			// There must at least be 3 buffers left.
+			if len(out) < dstEvery*3 {
+				d.bufs.Put(buf)
+				return nil, errors.New("corruption detected: stream overrun 2")
+			}
+		}
+	}
+	if off > 0 {
+		ioff := int(off)
+		if len(out) < dstEvery*3+ioff {
+			d.bufs.Put(buf)
+			return nil, errors.New("corruption detected: stream overrun 3")
+		}
+		copy(out, buf[0][:off])
+		copy(out[dstEvery:], buf[1][:off])
+		copy(out[dstEvery*2:], buf[2][:off])
+		copy(out[dstEvery*3:], buf[3][:off])
+		decoded += int(off) * 4
+		out = out[off:]
+	}
+
+	// Decode remaining.
+	remainBytes := dstEvery - (decoded / 4)
+	for i := range br {
+		offset := dstEvery * i
+		endsAt := offset + remainBytes
+		if endsAt > len(out) {
+			endsAt = len(out)
+		}
+		br := &br[i]
+		bitsLeft := br.remaining()
+		for bitsLeft > 0 {
+			br.fill()
+			if offset >= endsAt {
+				d.bufs.Put(buf)
+				return nil, errors.New("corruption detected: stream overrun 4")
+			}
+
+			// 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++
+		}
+		if offset != endsAt {
+			d.bufs.Put(buf)
+			return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
+		}
+		decoded += offset - dstEvery*i
+		err = br.close()
+		if err != nil {
+			return nil, err
+		}
+	}
+	d.bufs.Put(buf)
+	if dstSize != decoded {
+		return nil, errors.New("corruption detected: short output block")
+	}
+	return dst, nil
+}
+
+// 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.
+	bufs := d.buffer()
+	buf := &bufs[0]
+	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()
+				d.bufs.Put(bufs)
+				return nil, ErrMaxDecodedSizeExceeded
+			}
+			dst = append(dst, buf[:]...)
+		}
+	}
+
+	if len(dst)+int(off) > maxDecodedSize {
+		d.bufs.Put(bufs)
+		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 {
+			d.bufs.Put(bufs)
+			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))
+	}
+	d.bufs.Put(bufs)
+	return dst, br.close()
+}
diff --git a/vendor/github.com/klauspost/compress/huff0/huff0.go b/vendor/github.com/klauspost/compress/huff0/huff0.go
index 7ec2022..e8ad17a 100644
--- a/vendor/github.com/klauspost/compress/huff0/huff0.go
+++ b/vendor/github.com/klauspost/compress/huff0/huff0.go
@@ -8,6 +8,7 @@
 	"fmt"
 	"math"
 	"math/bits"
+	"sync"
 
 	"github.com/klauspost/compress/fse"
 )
@@ -116,6 +117,7 @@
 	nodes          []nodeElt
 	tmpOut         [4][]byte
 	fse            *fse.Scratch
+	decPool        sync.Pool // *[4][256]byte buffers.
 	huffWeight     [maxSymbolValue + 1]byte
 }
 
@@ -245,6 +247,68 @@
 	return nil
 }
 
+func (c cTable) estTableSize(s *Scratch) (sz int, err error) {
+	var (
+		// precomputed conversion table
+		bitsToWeight [tableLogMax + 1]byte
+		huffLog      = s.actualTableLog
+		// last weight is not saved.
+		maxSymbolValue = uint8(s.symbolLen - 1)
+		huffWeight     = s.huffWeight[:256]
+	)
+	const (
+		maxFSETableLog = 6
+	)
+	// convert to weight
+	bitsToWeight[0] = 0
+	for n := uint8(1); n < huffLog+1; n++ {
+		bitsToWeight[n] = huffLog + 1 - n
+	}
+
+	// Acquire histogram for FSE.
+	hist := s.fse.Histogram()
+	hist = hist[:256]
+	for i := range hist[:16] {
+		hist[i] = 0
+	}
+	for n := uint8(0); n < maxSymbolValue; n++ {
+		v := bitsToWeight[c[n].nBits] & 15
+		huffWeight[n] = v
+		hist[v]++
+	}
+
+	// FSE compress if feasible.
+	if maxSymbolValue >= 2 {
+		huffMaxCnt := uint32(0)
+		huffMax := uint8(0)
+		for i, v := range hist[:16] {
+			if v == 0 {
+				continue
+			}
+			huffMax = byte(i)
+			if v > huffMaxCnt {
+				huffMaxCnt = v
+			}
+		}
+		s.fse.HistogramFinished(huffMax, int(huffMaxCnt))
+		s.fse.TableLog = maxFSETableLog
+		b, err := fse.Compress(huffWeight[:maxSymbolValue], s.fse)
+		if err == nil && len(b) < int(s.symbolLen>>1) {
+			sz += 1 + len(b)
+			return sz, nil
+		}
+		// Unable to compress (RLE/uncompressible)
+	}
+	// write raw values as 4-bits (max : 15)
+	if maxSymbolValue > (256 - 128) {
+		// should not happen : likely means source cannot be compressed
+		return 0, ErrIncompressible
+	}
+	// special case, pack weights 4 bits/weight.
+	sz += 1 + int(maxSymbolValue/2)
+	return sz, nil
+}
+
 // estimateSize returns the estimated size in bytes of the input represented in the
 // histogram supplied.
 func (c cTable) estimateSize(hist []uint32) int {
diff --git a/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go
new file mode 100644
index 0000000..3954c51
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go
@@ -0,0 +1,34 @@
+// Package cpuinfo gives runtime info about the current CPU.
+//
+// This is a very limited module meant for use internally
+// in this project. For more versatile solution check
+// https://github.com/klauspost/cpuid.
+package cpuinfo
+
+// HasBMI1 checks whether an x86 CPU supports the BMI1 extension.
+func HasBMI1() bool {
+	return hasBMI1
+}
+
+// HasBMI2 checks whether an x86 CPU supports the BMI2 extension.
+func HasBMI2() bool {
+	return hasBMI2
+}
+
+// DisableBMI2 will disable BMI2, for testing purposes.
+// Call returned function to restore previous state.
+func DisableBMI2() func() {
+	old := hasBMI2
+	hasBMI2 = false
+	return func() {
+		hasBMI2 = old
+	}
+}
+
+// HasBMI checks whether an x86 CPU supports both BMI1 and BMI2 extensions.
+func HasBMI() bool {
+	return HasBMI1() && HasBMI2()
+}
+
+var hasBMI1 bool
+var hasBMI2 bool
diff --git a/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go
new file mode 100644
index 0000000..e802579
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go
@@ -0,0 +1,11 @@
+//go:build amd64 && !appengine && !noasm && gc
+// +build amd64,!appengine,!noasm,gc
+
+package cpuinfo
+
+// go:noescape
+func x86extensions() (bmi1, bmi2 bool)
+
+func init() {
+	hasBMI1, hasBMI2 = x86extensions()
+}
diff --git a/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s
new file mode 100644
index 0000000..4465fbe
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s
@@ -0,0 +1,36 @@
+// +build !appengine
+// +build gc
+// +build !noasm
+
+#include "textflag.h"
+#include "funcdata.h"
+#include "go_asm.h"
+
+TEXT ·x86extensions(SB), NOSPLIT, $0
+	// 1. determine max EAX value
+	XORQ AX, AX
+	CPUID
+
+	CMPQ AX, $7
+	JB   unsupported
+
+	// 2. EAX = 7, ECX = 0 --- see Table 3-8 "Information Returned by CPUID Instruction"
+	MOVQ $7, AX
+	MOVQ $0, CX
+	CPUID
+
+	BTQ   $3, BX // bit 3 = BMI1
+	SETCS AL
+
+	BTQ   $8, BX // bit 8 = BMI2
+	SETCS AH
+
+	MOVB AL, bmi1+0(FP)
+	MOVB AH, bmi2+1(FP)
+	RET
+
+unsupported:
+	XORQ AX, AX
+	MOVB AL, bmi1+0(FP)
+	MOVB AL, bmi2+1(FP)
+	RET
diff --git a/vendor/github.com/klauspost/compress/internal/snapref/LICENSE b/vendor/github.com/klauspost/compress/internal/snapref/LICENSE
new file mode 100644
index 0000000..6050c10
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/snapref/LICENSE
@@ -0,0 +1,27 @@
+Copyright (c) 2011 The Snappy-Go Authors. 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 Google Inc. 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/klauspost/compress/internal/snapref/decode.go b/vendor/github.com/klauspost/compress/internal/snapref/decode.go
new file mode 100644
index 0000000..40796a4
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/snapref/decode.go
@@ -0,0 +1,264 @@
+// Copyright 2011 The Snappy-Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package snapref
+
+import (
+	"encoding/binary"
+	"errors"
+	"io"
+)
+
+var (
+	// ErrCorrupt reports that the input is invalid.
+	ErrCorrupt = errors.New("snappy: corrupt input")
+	// ErrTooLarge reports that the uncompressed length is too large.
+	ErrTooLarge = errors.New("snappy: decoded block is too large")
+	// ErrUnsupported reports that the input isn't supported.
+	ErrUnsupported = errors.New("snappy: unsupported input")
+
+	errUnsupportedLiteralLength = errors.New("snappy: unsupported literal length")
+)
+
+// DecodedLen returns the length of the decoded block.
+func DecodedLen(src []byte) (int, error) {
+	v, _, err := decodedLen(src)
+	return v, err
+}
+
+// decodedLen returns the length of the decoded block and the number of bytes
+// that the length header occupied.
+func decodedLen(src []byte) (blockLen, headerLen int, err error) {
+	v, n := binary.Uvarint(src)
+	if n <= 0 || v > 0xffffffff {
+		return 0, 0, ErrCorrupt
+	}
+
+	const wordSize = 32 << (^uint(0) >> 32 & 1)
+	if wordSize == 32 && v > 0x7fffffff {
+		return 0, 0, ErrTooLarge
+	}
+	return int(v), n, nil
+}
+
+const (
+	decodeErrCodeCorrupt                  = 1
+	decodeErrCodeUnsupportedLiteralLength = 2
+)
+
+// Decode returns the decoded form of src. The returned slice may be a sub-
+// slice of dst if dst was large enough to hold the entire decoded block.
+// Otherwise, a newly allocated slice will be returned.
+//
+// The dst and src must not overlap. It is valid to pass a nil dst.
+//
+// Decode handles the Snappy block format, not the Snappy stream format.
+func Decode(dst, src []byte) ([]byte, error) {
+	dLen, s, err := decodedLen(src)
+	if err != nil {
+		return nil, err
+	}
+	if dLen <= len(dst) {
+		dst = dst[:dLen]
+	} else {
+		dst = make([]byte, dLen)
+	}
+	switch decode(dst, src[s:]) {
+	case 0:
+		return dst, nil
+	case decodeErrCodeUnsupportedLiteralLength:
+		return nil, errUnsupportedLiteralLength
+	}
+	return nil, ErrCorrupt
+}
+
+// NewReader returns a new Reader that decompresses from r, using the framing
+// format described at
+// https://github.com/google/snappy/blob/master/framing_format.txt
+func NewReader(r io.Reader) *Reader {
+	return &Reader{
+		r:       r,
+		decoded: make([]byte, maxBlockSize),
+		buf:     make([]byte, maxEncodedLenOfMaxBlockSize+checksumSize),
+	}
+}
+
+// Reader is an io.Reader that can read Snappy-compressed bytes.
+//
+// Reader handles the Snappy stream format, not the Snappy block format.
+type Reader struct {
+	r       io.Reader
+	err     error
+	decoded []byte
+	buf     []byte
+	// decoded[i:j] contains decoded bytes that have not yet been passed on.
+	i, j       int
+	readHeader bool
+}
+
+// Reset discards any buffered data, resets all state, and switches the Snappy
+// reader to read from r. This permits reusing a Reader rather than allocating
+// a new one.
+func (r *Reader) Reset(reader io.Reader) {
+	r.r = reader
+	r.err = nil
+	r.i = 0
+	r.j = 0
+	r.readHeader = false
+}
+
+func (r *Reader) readFull(p []byte, allowEOF bool) (ok bool) {
+	if _, r.err = io.ReadFull(r.r, p); r.err != nil {
+		if r.err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) {
+			r.err = ErrCorrupt
+		}
+		return false
+	}
+	return true
+}
+
+func (r *Reader) fill() error {
+	for r.i >= r.j {
+		if !r.readFull(r.buf[:4], true) {
+			return r.err
+		}
+		chunkType := r.buf[0]
+		if !r.readHeader {
+			if chunkType != chunkTypeStreamIdentifier {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			r.readHeader = true
+		}
+		chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16
+		if chunkLen > len(r.buf) {
+			r.err = ErrUnsupported
+			return r.err
+		}
+
+		// The chunk types are specified at
+		// https://github.com/google/snappy/blob/master/framing_format.txt
+		switch chunkType {
+		case chunkTypeCompressedData:
+			// Section 4.2. Compressed data (chunk type 0x00).
+			if chunkLen < checksumSize {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			buf := r.buf[:chunkLen]
+			if !r.readFull(buf, false) {
+				return r.err
+			}
+			checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24
+			buf = buf[checksumSize:]
+
+			n, err := DecodedLen(buf)
+			if err != nil {
+				r.err = err
+				return r.err
+			}
+			if n > len(r.decoded) {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			if _, err := Decode(r.decoded, buf); err != nil {
+				r.err = err
+				return r.err
+			}
+			if crc(r.decoded[:n]) != checksum {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			r.i, r.j = 0, n
+			continue
+
+		case chunkTypeUncompressedData:
+			// Section 4.3. Uncompressed data (chunk type 0x01).
+			if chunkLen < checksumSize {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			buf := r.buf[:checksumSize]
+			if !r.readFull(buf, false) {
+				return r.err
+			}
+			checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24
+			// Read directly into r.decoded instead of via r.buf.
+			n := chunkLen - checksumSize
+			if n > len(r.decoded) {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			if !r.readFull(r.decoded[:n], false) {
+				return r.err
+			}
+			if crc(r.decoded[:n]) != checksum {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			r.i, r.j = 0, n
+			continue
+
+		case chunkTypeStreamIdentifier:
+			// Section 4.1. Stream identifier (chunk type 0xff).
+			if chunkLen != len(magicBody) {
+				r.err = ErrCorrupt
+				return r.err
+			}
+			if !r.readFull(r.buf[:len(magicBody)], false) {
+				return r.err
+			}
+			for i := 0; i < len(magicBody); i++ {
+				if r.buf[i] != magicBody[i] {
+					r.err = ErrCorrupt
+					return r.err
+				}
+			}
+			continue
+		}
+
+		if chunkType <= 0x7f {
+			// Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f).
+			r.err = ErrUnsupported
+			return r.err
+		}
+		// Section 4.4 Padding (chunk type 0xfe).
+		// Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd).
+		if !r.readFull(r.buf[:chunkLen], false) {
+			return r.err
+		}
+	}
+
+	return nil
+}
+
+// Read satisfies the io.Reader interface.
+func (r *Reader) Read(p []byte) (int, error) {
+	if r.err != nil {
+		return 0, r.err
+	}
+
+	if err := r.fill(); err != nil {
+		return 0, err
+	}
+
+	n := copy(p, r.decoded[r.i:r.j])
+	r.i += n
+	return n, nil
+}
+
+// ReadByte satisfies the io.ByteReader interface.
+func (r *Reader) ReadByte() (byte, error) {
+	if r.err != nil {
+		return 0, r.err
+	}
+
+	if err := r.fill(); err != nil {
+		return 0, err
+	}
+
+	c := r.decoded[r.i]
+	r.i++
+	return c, nil
+}
diff --git a/vendor/github.com/klauspost/compress/internal/snapref/decode_other.go b/vendor/github.com/klauspost/compress/internal/snapref/decode_other.go
new file mode 100644
index 0000000..77395a6
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/snapref/decode_other.go
@@ -0,0 +1,113 @@
+// Copyright 2016 The Snappy-Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package snapref
+
+// decode writes the decoding of src to dst. It assumes that the varint-encoded
+// length of the decompressed bytes has already been read, and that len(dst)
+// equals that length.
+//
+// It returns 0 on success or a decodeErrCodeXxx error code on failure.
+func decode(dst, src []byte) int {
+	var d, s, offset, length int
+	for s < len(src) {
+		switch src[s] & 0x03 {
+		case tagLiteral:
+			x := uint32(src[s] >> 2)
+			switch {
+			case x < 60:
+				s++
+			case x == 60:
+				s += 2
+				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+					return decodeErrCodeCorrupt
+				}
+				x = uint32(src[s-1])
+			case x == 61:
+				s += 3
+				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+					return decodeErrCodeCorrupt
+				}
+				x = uint32(src[s-2]) | uint32(src[s-1])<<8
+			case x == 62:
+				s += 4
+				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+					return decodeErrCodeCorrupt
+				}
+				x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16
+			case x == 63:
+				s += 5
+				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+					return decodeErrCodeCorrupt
+				}
+				x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24
+			}
+			length = int(x) + 1
+			if length <= 0 {
+				return decodeErrCodeUnsupportedLiteralLength
+			}
+			if length > len(dst)-d || length > len(src)-s {
+				return decodeErrCodeCorrupt
+			}
+			copy(dst[d:], src[s:s+length])
+			d += length
+			s += length
+			continue
+
+		case tagCopy1:
+			s += 2
+			if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+				return decodeErrCodeCorrupt
+			}
+			length = 4 + int(src[s-2])>>2&0x7
+			offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1]))
+
+		case tagCopy2:
+			s += 3
+			if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+				return decodeErrCodeCorrupt
+			}
+			length = 1 + int(src[s-3])>>2
+			offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8)
+
+		case tagCopy4:
+			s += 5
+			if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
+				return decodeErrCodeCorrupt
+			}
+			length = 1 + int(src[s-5])>>2
+			offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24)
+		}
+
+		if offset <= 0 || d < offset || length > len(dst)-d {
+			return decodeErrCodeCorrupt
+		}
+		// Copy from an earlier sub-slice of dst to a later sub-slice.
+		// If no overlap, use the built-in copy:
+		if offset >= length {
+			copy(dst[d:d+length], dst[d-offset:])
+			d += length
+			continue
+		}
+
+		// Unlike the built-in copy function, this byte-by-byte copy always runs
+		// forwards, even if the slices overlap. Conceptually, this is:
+		//
+		// d += forwardCopy(dst[d:d+length], dst[d-offset:])
+		//
+		// We align the slices into a and b and show the compiler they are the same size.
+		// This allows the loop to run without bounds checks.
+		a := dst[d : d+length]
+		b := dst[d-offset:]
+		b = b[:len(a)]
+		for i := range a {
+			a[i] = b[i]
+		}
+		d += length
+	}
+	if d != len(dst) {
+		return decodeErrCodeCorrupt
+	}
+	return 0
+}
diff --git a/vendor/github.com/klauspost/compress/internal/snapref/encode.go b/vendor/github.com/klauspost/compress/internal/snapref/encode.go
new file mode 100644
index 0000000..13c6040
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/snapref/encode.go
@@ -0,0 +1,289 @@
+// Copyright 2011 The Snappy-Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package snapref
+
+import (
+	"encoding/binary"
+	"errors"
+	"io"
+)
+
+// Encode returns the encoded form of src. The returned slice may be a sub-
+// slice of dst if dst was large enough to hold the entire encoded block.
+// Otherwise, a newly allocated slice will be returned.
+//
+// The dst and src must not overlap. It is valid to pass a nil dst.
+//
+// Encode handles the Snappy block format, not the Snappy stream format.
+func Encode(dst, src []byte) []byte {
+	if n := MaxEncodedLen(len(src)); n < 0 {
+		panic(ErrTooLarge)
+	} else if len(dst) < n {
+		dst = make([]byte, n)
+	}
+
+	// The block starts with the varint-encoded length of the decompressed bytes.
+	d := binary.PutUvarint(dst, uint64(len(src)))
+
+	for len(src) > 0 {
+		p := src
+		src = nil
+		if len(p) > maxBlockSize {
+			p, src = p[:maxBlockSize], p[maxBlockSize:]
+		}
+		if len(p) < minNonLiteralBlockSize {
+			d += emitLiteral(dst[d:], p)
+		} else {
+			d += encodeBlock(dst[d:], p)
+		}
+	}
+	return dst[:d]
+}
+
+// inputMargin is the minimum number of extra input bytes to keep, inside
+// encodeBlock's inner loop. On some architectures, this margin lets us
+// implement a fast path for emitLiteral, where the copy of short (<= 16 byte)
+// literals can be implemented as a single load to and store from a 16-byte
+// register. That literal's actual length can be as short as 1 byte, so this
+// can copy up to 15 bytes too much, but that's OK as subsequent iterations of
+// the encoding loop will fix up the copy overrun, and this inputMargin ensures
+// that we don't overrun the dst and src buffers.
+const inputMargin = 16 - 1
+
+// minNonLiteralBlockSize is the minimum size of the input to encodeBlock that
+// could be encoded with a copy tag. This is the minimum with respect to the
+// algorithm used by encodeBlock, not a minimum enforced by the file format.
+//
+// The encoded output must start with at least a 1 byte literal, as there are
+// no previous bytes to copy. A minimal (1 byte) copy after that, generated
+// from an emitCopy call in encodeBlock's main loop, would require at least
+// another inputMargin bytes, for the reason above: we want any emitLiteral
+// calls inside encodeBlock's main loop to use the fast path if possible, which
+// requires being able to overrun by inputMargin bytes. Thus,
+// minNonLiteralBlockSize equals 1 + 1 + inputMargin.
+//
+// The C++ code doesn't use this exact threshold, but it could, as discussed at
+// https://groups.google.com/d/topic/snappy-compression/oGbhsdIJSJ8/discussion
+// The difference between Go (2+inputMargin) and C++ (inputMargin) is purely an
+// optimization. It should not affect the encoded form. This is tested by
+// TestSameEncodingAsCppShortCopies.
+const minNonLiteralBlockSize = 1 + 1 + inputMargin
+
+// MaxEncodedLen returns the maximum length of a snappy block, given its
+// uncompressed length.
+//
+// It will return a negative value if srcLen is too large to encode.
+func MaxEncodedLen(srcLen int) int {
+	n := uint64(srcLen)
+	if n > 0xffffffff {
+		return -1
+	}
+	// Compressed data can be defined as:
+	//    compressed := item* literal*
+	//    item       := literal* copy
+	//
+	// The trailing literal sequence has a space blowup of at most 62/60
+	// since a literal of length 60 needs one tag byte + one extra byte
+	// for length information.
+	//
+	// Item blowup is trickier to measure. Suppose the "copy" op copies
+	// 4 bytes of data. Because of a special check in the encoding code,
+	// we produce a 4-byte copy only if the offset is < 65536. Therefore
+	// the copy op takes 3 bytes to encode, and this type of item leads
+	// to at most the 62/60 blowup for representing literals.
+	//
+	// Suppose the "copy" op copies 5 bytes of data. If the offset is big
+	// enough, it will take 5 bytes to encode the copy op. Therefore the
+	// worst case here is a one-byte literal followed by a five-byte copy.
+	// That is, 6 bytes of input turn into 7 bytes of "compressed" data.
+	//
+	// This last factor dominates the blowup, so the final estimate is:
+	n = 32 + n + n/6
+	if n > 0xffffffff {
+		return -1
+	}
+	return int(n)
+}
+
+var errClosed = errors.New("snappy: Writer is closed")
+
+// NewWriter returns a new Writer that compresses to w.
+//
+// The Writer returned does not buffer writes. There is no need to Flush or
+// Close such a Writer.
+//
+// Deprecated: the Writer returned is not suitable for many small writes, only
+// for few large writes. Use NewBufferedWriter instead, which is efficient
+// regardless of the frequency and shape of the writes, and remember to Close
+// that Writer when done.
+func NewWriter(w io.Writer) *Writer {
+	return &Writer{
+		w:    w,
+		obuf: make([]byte, obufLen),
+	}
+}
+
+// NewBufferedWriter returns a new Writer that compresses to w, using the
+// framing format described at
+// https://github.com/google/snappy/blob/master/framing_format.txt
+//
+// The Writer returned buffers writes. Users must call Close to guarantee all
+// data has been forwarded to the underlying io.Writer. They may also call
+// Flush zero or more times before calling Close.
+func NewBufferedWriter(w io.Writer) *Writer {
+	return &Writer{
+		w:    w,
+		ibuf: make([]byte, 0, maxBlockSize),
+		obuf: make([]byte, obufLen),
+	}
+}
+
+// Writer is an io.Writer that can write Snappy-compressed bytes.
+//
+// Writer handles the Snappy stream format, not the Snappy block format.
+type Writer struct {
+	w   io.Writer
+	err error
+
+	// ibuf is a buffer for the incoming (uncompressed) bytes.
+	//
+	// Its use is optional. For backwards compatibility, Writers created by the
+	// NewWriter function have ibuf == nil, do not buffer incoming bytes, and
+	// therefore do not need to be Flush'ed or Close'd.
+	ibuf []byte
+
+	// obuf is a buffer for the outgoing (compressed) bytes.
+	obuf []byte
+
+	// wroteStreamHeader is whether we have written the stream header.
+	wroteStreamHeader bool
+}
+
+// Reset discards the writer's state and switches the Snappy writer to write to
+// w. This permits reusing a Writer rather than allocating a new one.
+func (w *Writer) Reset(writer io.Writer) {
+	w.w = writer
+	w.err = nil
+	if w.ibuf != nil {
+		w.ibuf = w.ibuf[:0]
+	}
+	w.wroteStreamHeader = false
+}
+
+// Write satisfies the io.Writer interface.
+func (w *Writer) Write(p []byte) (nRet int, errRet error) {
+	if w.ibuf == nil {
+		// Do not buffer incoming bytes. This does not perform or compress well
+		// if the caller of Writer.Write writes many small slices. This
+		// behavior is therefore deprecated, but still supported for backwards
+		// compatibility with code that doesn't explicitly Flush or Close.
+		return w.write(p)
+	}
+
+	// The remainder of this method is based on bufio.Writer.Write from the
+	// standard library.
+
+	for len(p) > (cap(w.ibuf)-len(w.ibuf)) && w.err == nil {
+		var n int
+		if len(w.ibuf) == 0 {
+			// Large write, empty buffer.
+			// Write directly from p to avoid copy.
+			n, _ = w.write(p)
+		} else {
+			n = copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p)
+			w.ibuf = w.ibuf[:len(w.ibuf)+n]
+			w.Flush()
+		}
+		nRet += n
+		p = p[n:]
+	}
+	if w.err != nil {
+		return nRet, w.err
+	}
+	n := copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p)
+	w.ibuf = w.ibuf[:len(w.ibuf)+n]
+	nRet += n
+	return nRet, nil
+}
+
+func (w *Writer) write(p []byte) (nRet int, errRet error) {
+	if w.err != nil {
+		return 0, w.err
+	}
+	for len(p) > 0 {
+		obufStart := len(magicChunk)
+		if !w.wroteStreamHeader {
+			w.wroteStreamHeader = true
+			copy(w.obuf, magicChunk)
+			obufStart = 0
+		}
+
+		var uncompressed []byte
+		if len(p) > maxBlockSize {
+			uncompressed, p = p[:maxBlockSize], p[maxBlockSize:]
+		} else {
+			uncompressed, p = p, nil
+		}
+		checksum := crc(uncompressed)
+
+		// Compress the buffer, discarding the result if the improvement
+		// isn't at least 12.5%.
+		compressed := Encode(w.obuf[obufHeaderLen:], uncompressed)
+		chunkType := uint8(chunkTypeCompressedData)
+		chunkLen := 4 + len(compressed)
+		obufEnd := obufHeaderLen + len(compressed)
+		if len(compressed) >= len(uncompressed)-len(uncompressed)/8 {
+			chunkType = chunkTypeUncompressedData
+			chunkLen = 4 + len(uncompressed)
+			obufEnd = obufHeaderLen
+		}
+
+		// Fill in the per-chunk header that comes before the body.
+		w.obuf[len(magicChunk)+0] = chunkType
+		w.obuf[len(magicChunk)+1] = uint8(chunkLen >> 0)
+		w.obuf[len(magicChunk)+2] = uint8(chunkLen >> 8)
+		w.obuf[len(magicChunk)+3] = uint8(chunkLen >> 16)
+		w.obuf[len(magicChunk)+4] = uint8(checksum >> 0)
+		w.obuf[len(magicChunk)+5] = uint8(checksum >> 8)
+		w.obuf[len(magicChunk)+6] = uint8(checksum >> 16)
+		w.obuf[len(magicChunk)+7] = uint8(checksum >> 24)
+
+		if _, err := w.w.Write(w.obuf[obufStart:obufEnd]); err != nil {
+			w.err = err
+			return nRet, err
+		}
+		if chunkType == chunkTypeUncompressedData {
+			if _, err := w.w.Write(uncompressed); err != nil {
+				w.err = err
+				return nRet, err
+			}
+		}
+		nRet += len(uncompressed)
+	}
+	return nRet, nil
+}
+
+// Flush flushes the Writer to its underlying io.Writer.
+func (w *Writer) Flush() error {
+	if w.err != nil {
+		return w.err
+	}
+	if len(w.ibuf) == 0 {
+		return nil
+	}
+	w.write(w.ibuf)
+	w.ibuf = w.ibuf[:0]
+	return w.err
+}
+
+// Close calls Flush and then closes the Writer.
+func (w *Writer) Close() error {
+	w.Flush()
+	ret := w.err
+	if w.err == nil {
+		w.err = errClosed
+	}
+	return ret
+}
diff --git a/vendor/github.com/klauspost/compress/internal/snapref/encode_other.go b/vendor/github.com/klauspost/compress/internal/snapref/encode_other.go
new file mode 100644
index 0000000..511bba6
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/snapref/encode_other.go
@@ -0,0 +1,236 @@
+// Copyright 2016 The Snappy-Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package snapref
+
+func load32(b []byte, i int) uint32 {
+	b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line.
+	return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
+}
+
+func load64(b []byte, i int) uint64 {
+	b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line.
+	return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
+		uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
+}
+
+// emitLiteral writes a literal chunk and returns the number of bytes written.
+//
+// It assumes that:
+//	dst is long enough to hold the encoded bytes
+//	1 <= len(lit) && len(lit) <= 65536
+func emitLiteral(dst, lit []byte) int {
+	i, n := 0, uint(len(lit)-1)
+	switch {
+	case n < 60:
+		dst[0] = uint8(n)<<2 | tagLiteral
+		i = 1
+	case n < 1<<8:
+		dst[0] = 60<<2 | tagLiteral
+		dst[1] = uint8(n)
+		i = 2
+	default:
+		dst[0] = 61<<2 | tagLiteral
+		dst[1] = uint8(n)
+		dst[2] = uint8(n >> 8)
+		i = 3
+	}
+	return i + copy(dst[i:], lit)
+}
+
+// emitCopy writes a copy chunk and returns the number of bytes written.
+//
+// It assumes that:
+//	dst is long enough to hold the encoded bytes
+//	1 <= offset && offset <= 65535
+//	4 <= length && length <= 65535
+func emitCopy(dst []byte, offset, length int) int {
+	i := 0
+	// The maximum length for a single tagCopy1 or tagCopy2 op is 64 bytes. The
+	// threshold for this loop is a little higher (at 68 = 64 + 4), and the
+	// length emitted down below is is a little lower (at 60 = 64 - 4), because
+	// it's shorter to encode a length 67 copy as a length 60 tagCopy2 followed
+	// by a length 7 tagCopy1 (which encodes as 3+2 bytes) than to encode it as
+	// a length 64 tagCopy2 followed by a length 3 tagCopy2 (which encodes as
+	// 3+3 bytes). The magic 4 in the 64±4 is because the minimum length for a
+	// tagCopy1 op is 4 bytes, which is why a length 3 copy has to be an
+	// encodes-as-3-bytes tagCopy2 instead of an encodes-as-2-bytes tagCopy1.
+	for length >= 68 {
+		// Emit a length 64 copy, encoded as 3 bytes.
+		dst[i+0] = 63<<2 | tagCopy2
+		dst[i+1] = uint8(offset)
+		dst[i+2] = uint8(offset >> 8)
+		i += 3
+		length -= 64
+	}
+	if length > 64 {
+		// Emit a length 60 copy, encoded as 3 bytes.
+		dst[i+0] = 59<<2 | tagCopy2
+		dst[i+1] = uint8(offset)
+		dst[i+2] = uint8(offset >> 8)
+		i += 3
+		length -= 60
+	}
+	if length >= 12 || offset >= 2048 {
+		// Emit the remaining copy, encoded as 3 bytes.
+		dst[i+0] = uint8(length-1)<<2 | tagCopy2
+		dst[i+1] = uint8(offset)
+		dst[i+2] = uint8(offset >> 8)
+		return i + 3
+	}
+	// Emit the remaining copy, encoded as 2 bytes.
+	dst[i+0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1
+	dst[i+1] = uint8(offset)
+	return i + 2
+}
+
+// extendMatch returns the largest k such that k <= len(src) and that
+// src[i:i+k-j] and src[j:k] have the same contents.
+//
+// It assumes that:
+//	0 <= i && i < j && j <= len(src)
+func extendMatch(src []byte, i, j int) int {
+	for ; j < len(src) && src[i] == src[j]; i, j = i+1, j+1 {
+	}
+	return j
+}
+
+func hash(u, shift uint32) uint32 {
+	return (u * 0x1e35a7bd) >> shift
+}
+
+// encodeBlock encodes a non-empty src to a guaranteed-large-enough dst. It
+// assumes that the varint-encoded length of the decompressed bytes has already
+// been written.
+//
+// It also assumes that:
+//	len(dst) >= MaxEncodedLen(len(src)) &&
+// 	minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
+func encodeBlock(dst, src []byte) (d int) {
+	// Initialize the hash table. Its size ranges from 1<<8 to 1<<14 inclusive.
+	// The table element type is uint16, as s < sLimit and sLimit < len(src)
+	// and len(src) <= maxBlockSize and maxBlockSize == 65536.
+	const (
+		maxTableSize = 1 << 14
+		// tableMask is redundant, but helps the compiler eliminate bounds
+		// checks.
+		tableMask = maxTableSize - 1
+	)
+	shift := uint32(32 - 8)
+	for tableSize := 1 << 8; tableSize < maxTableSize && tableSize < len(src); tableSize *= 2 {
+		shift--
+	}
+	// In Go, all array elements are zero-initialized, so there is no advantage
+	// to a smaller tableSize per se. However, it matches the C++ algorithm,
+	// and in the asm versions of this code, we can get away with zeroing only
+	// the first tableSize elements.
+	var table [maxTableSize]uint16
+
+	// sLimit is when to stop looking for offset/length copies. The inputMargin
+	// lets us use a fast path for emitLiteral in the main loop, while we are
+	// looking for copies.
+	sLimit := len(src) - inputMargin
+
+	// nextEmit is where in src the next emitLiteral should start from.
+	nextEmit := 0
+
+	// The encoded form must start with a literal, as there are no previous
+	// bytes to copy, so we start looking for hash matches at s == 1.
+	s := 1
+	nextHash := hash(load32(src, s), shift)
+
+	for {
+		// Copied from the C++ snappy implementation:
+		//
+		// Heuristic match skipping: If 32 bytes are scanned with no matches
+		// found, start looking only at every other byte. If 32 more bytes are
+		// scanned (or skipped), look at every third byte, etc.. When a match
+		// is found, immediately go back to looking at every byte. This is a
+		// small loss (~5% performance, ~0.1% density) for compressible data
+		// due to more bookkeeping, but for non-compressible data (such as
+		// JPEG) it's a huge win since the compressor quickly "realizes" the
+		// data is incompressible and doesn't bother looking for matches
+		// everywhere.
+		//
+		// The "skip" variable keeps track of how many bytes there are since
+		// the last match; dividing it by 32 (ie. right-shifting by five) gives
+		// the number of bytes to move ahead for each iteration.
+		skip := 32
+
+		nextS := s
+		candidate := 0
+		for {
+			s = nextS
+			bytesBetweenHashLookups := skip >> 5
+			nextS = s + bytesBetweenHashLookups
+			skip += bytesBetweenHashLookups
+			if nextS > sLimit {
+				goto emitRemainder
+			}
+			candidate = int(table[nextHash&tableMask])
+			table[nextHash&tableMask] = uint16(s)
+			nextHash = hash(load32(src, nextS), shift)
+			if load32(src, s) == load32(src, candidate) {
+				break
+			}
+		}
+
+		// A 4-byte match has been found. We'll later see if more than 4 bytes
+		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
+		// them as literal bytes.
+		d += emitLiteral(dst[d:], src[nextEmit:s])
+
+		// Call emitCopy, and then see if another emitCopy could be our next
+		// move. Repeat until we find no match for the input immediately after
+		// what was consumed by the last emitCopy call.
+		//
+		// If we exit this loop normally then we need to call emitLiteral next,
+		// though we don't yet know how big the literal will be. We handle that
+		// by proceeding to the next iteration of the main loop. We also can
+		// exit this loop via goto if we get close to exhausting the input.
+		for {
+			// Invariant: we have a 4-byte match at s, and no need to emit any
+			// literal bytes prior to s.
+			base := s
+
+			// Extend the 4-byte match as long as possible.
+			//
+			// This is an inlined version of:
+			//	s = extendMatch(src, candidate+4, s+4)
+			s += 4
+			for i := candidate + 4; s < len(src) && src[i] == src[s]; i, s = i+1, s+1 {
+			}
+
+			d += emitCopy(dst[d:], base-candidate, s-base)
+			nextEmit = s
+			if s >= sLimit {
+				goto emitRemainder
+			}
+
+			// We could immediately start working at s now, but to improve
+			// compression we first update the hash table at s-1 and at s. If
+			// another emitCopy is not our next move, also calculate nextHash
+			// at s+1. At least on GOARCH=amd64, these three hash calculations
+			// are faster as one load64 call (with some shifts) instead of
+			// three load32 calls.
+			x := load64(src, s-1)
+			prevHash := hash(uint32(x>>0), shift)
+			table[prevHash&tableMask] = uint16(s - 1)
+			currHash := hash(uint32(x>>8), shift)
+			candidate = int(table[currHash&tableMask])
+			table[currHash&tableMask] = uint16(s)
+			if uint32(x>>8) != load32(src, candidate) {
+				nextHash = hash(uint32(x>>16), shift)
+				s++
+				break
+			}
+		}
+	}
+
+emitRemainder:
+	if nextEmit < len(src) {
+		d += emitLiteral(dst[d:], src[nextEmit:])
+	}
+	return d
+}
diff --git a/vendor/github.com/klauspost/compress/internal/snapref/snappy.go b/vendor/github.com/klauspost/compress/internal/snapref/snappy.go
new file mode 100644
index 0000000..34d01f4
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/internal/snapref/snappy.go
@@ -0,0 +1,98 @@
+// Copyright 2011 The Snappy-Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package snapref implements the Snappy compression format. It aims for very
+// high speeds and reasonable compression.
+//
+// There are actually two Snappy formats: block and stream. They are related,
+// but different: trying to decompress block-compressed data as a Snappy stream
+// will fail, and vice versa. The block format is the Decode and Encode
+// functions and the stream format is the Reader and Writer types.
+//
+// The block format, the more common case, is used when the complete size (the
+// number of bytes) of the original data is known upfront, at the time
+// compression starts. The stream format, also known as the framing format, is
+// for when that isn't always true.
+//
+// The canonical, C++ implementation is at https://github.com/google/snappy and
+// it only implements the block format.
+package snapref
+
+import (
+	"hash/crc32"
+)
+
+/*
+Each encoded block begins with the varint-encoded length of the decoded data,
+followed by a sequence of chunks. Chunks begin and end on byte boundaries. The
+first byte of each chunk is broken into its 2 least and 6 most significant bits
+called l and m: l ranges in [0, 4) and m ranges in [0, 64). l is the chunk tag.
+Zero means a literal tag. All other values mean a copy tag.
+
+For literal tags:
+  - If m < 60, the next 1 + m bytes are literal bytes.
+  - Otherwise, let n be the little-endian unsigned integer denoted by the next
+    m - 59 bytes. The next 1 + n bytes after that are literal bytes.
+
+For copy tags, length bytes are copied from offset bytes ago, in the style of
+Lempel-Ziv compression algorithms. In particular:
+  - For l == 1, the offset ranges in [0, 1<<11) and the length in [4, 12).
+    The length is 4 + the low 3 bits of m. The high 3 bits of m form bits 8-10
+    of the offset. The next byte is bits 0-7 of the offset.
+  - For l == 2, the offset ranges in [0, 1<<16) and the length in [1, 65).
+    The length is 1 + m. The offset is the little-endian unsigned integer
+    denoted by the next 2 bytes.
+  - For l == 3, this tag is a legacy format that is no longer issued by most
+    encoders. Nonetheless, the offset ranges in [0, 1<<32) and the length in
+    [1, 65). The length is 1 + m. The offset is the little-endian unsigned
+    integer denoted by the next 4 bytes.
+*/
+const (
+	tagLiteral = 0x00
+	tagCopy1   = 0x01
+	tagCopy2   = 0x02
+	tagCopy4   = 0x03
+)
+
+const (
+	checksumSize    = 4
+	chunkHeaderSize = 4
+	magicChunk      = "\xff\x06\x00\x00" + magicBody
+	magicBody       = "sNaPpY"
+
+	// maxBlockSize is the maximum size of the input to encodeBlock. It is not
+	// part of the wire format per se, but some parts of the encoder assume
+	// that an offset fits into a uint16.
+	//
+	// Also, for the framing format (Writer type instead of Encode function),
+	// https://github.com/google/snappy/blob/master/framing_format.txt says
+	// that "the uncompressed data in a chunk must be no longer than 65536
+	// bytes".
+	maxBlockSize = 65536
+
+	// maxEncodedLenOfMaxBlockSize equals MaxEncodedLen(maxBlockSize), but is
+	// hard coded to be a const instead of a variable, so that obufLen can also
+	// be a const. Their equivalence is confirmed by
+	// TestMaxEncodedLenOfMaxBlockSize.
+	maxEncodedLenOfMaxBlockSize = 76490
+
+	obufHeaderLen = len(magicChunk) + checksumSize + chunkHeaderSize
+	obufLen       = obufHeaderLen + maxEncodedLenOfMaxBlockSize
+)
+
+const (
+	chunkTypeCompressedData   = 0x00
+	chunkTypeUncompressedData = 0x01
+	chunkTypePadding          = 0xfe
+	chunkTypeStreamIdentifier = 0xff
+)
+
+var crcTable = crc32.MakeTable(crc32.Castagnoli)
+
+// crc implements the checksum specified in section 3 of
+// https://github.com/google/snappy/blob/master/framing_format.txt
+func crc(b []byte) uint32 {
+	c := crc32.Update(0, crcTable, b)
+	return uint32(c>>15|c<<17) + 0xa282ead8
+}
diff --git a/vendor/github.com/klauspost/compress/s2sx.mod b/vendor/github.com/klauspost/compress/s2sx.mod
new file mode 100644
index 0000000..2263853
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/s2sx.mod
@@ -0,0 +1,4 @@
+module github.com/klauspost/compress
+
+go 1.16
+
diff --git a/vendor/github.com/klauspost/compress/s2sx.sum b/vendor/github.com/klauspost/compress/s2sx.sum
new file mode 100644
index 0000000..e69de29
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/s2sx.sum
diff --git a/vendor/github.com/klauspost/compress/zstd/README.md b/vendor/github.com/klauspost/compress/zstd/README.md
index e7d7eb0..beb7fa8 100644
--- a/vendor/github.com/klauspost/compress/zstd/README.md
+++ b/vendor/github.com/klauspost/compress/zstd/README.md
@@ -16,8 +16,7 @@
 
 Install using `go get -u github.com/klauspost/compress`. The package is located in `github.com/klauspost/compress/zstd`.
 
-Godoc Documentation: https://godoc.org/github.com/klauspost/compress/zstd
-
+[![Go Reference](https://pkg.go.dev/badge/github.com/klauspost/compress/zstd.svg)](https://pkg.go.dev/github.com/klauspost/compress/zstd)
 
 ## Compressor
 
@@ -79,6 +78,9 @@
 in the future. So if you want to limit concurrency for future updates, specify the concurrency
 you would like.
 
+If you would like stream encoding to be done without spawning async goroutines, use `WithEncoderConcurrency(1)`
+which will compress input as each block is completed, blocking on writes until each has completed.
+
 You can specify your desired compression level using `WithEncoderLevel()` option. Currently only pre-defined 
 compression settings can be specified.
 
@@ -105,7 +107,8 @@
 For compressing small blocks, the returned encoder has a function called `EncodeAll(src, dst []byte) []byte`.
 
 `EncodeAll` will encode all input in src and append it to dst.
-This function can be called concurrently, but each call will only run on a single goroutine.
+This function can be called concurrently. 
+Each call will only run on a same goroutine as the caller.
 
 Encoded blocks can be concatenated and the result will be the combined input stream.
 Data compressed with EncodeAll can be decoded with the Decoder, using either a stream or `DecodeAll`.
@@ -150,10 +153,10 @@
 
 This package:
 file    out     level   insize      outsize     millis  mb/s
-silesia.tar zskp    1   211947520   73101992    643     313.87
-silesia.tar zskp    2   211947520   67504318    969     208.38
-silesia.tar zskp    3   211947520   64595893    2007    100.68
-silesia.tar zskp    4   211947520   60995370    7691    26.28
+silesia.tar zskp    1   211947520   73821326    634     318.47
+silesia.tar zskp    2   211947520   67655404    1508    133.96
+silesia.tar zskp    3   211947520   64746933    3000    67.37
+silesia.tar zskp    4   211947520   60073508    16926   11.94
 
 cgo zstd:
 silesia.tar zstd    1   211947520   73605392    543     371.56
@@ -162,84 +165,94 @@
 silesia.tar zstd    9   211947520   60212393    5063    39.92
 
 gzip, stdlib/this package:
-silesia.tar gzstd   1   211947520   80007735    1654    122.21
-silesia.tar gzkp    1   211947520   80369488    1168    173.06
+silesia.tar gzstd   1   211947520   80007735    1498    134.87
+silesia.tar gzkp    1   211947520   80088272    1009    200.31
 
 GOB stream of binary data. Highly compressible.
 https://files.klauspost.com/compress/gob-stream.7z
 
 file        out     level   insize  outsize     millis  mb/s
-gob-stream  zskp    1   1911399616  235022249   3088    590.30
-gob-stream  zskp    2   1911399616  205669791   3786    481.34
-gob-stream  zskp    3   1911399616  175034659   9636    189.17
-gob-stream  zskp    4   1911399616  167273881   29337   62.13
+gob-stream  zskp    1   1911399616  233948096   3230    564.34
+gob-stream  zskp    2   1911399616  203997694   4997    364.73
+gob-stream  zskp    3   1911399616  173526523   13435   135.68
+gob-stream  zskp    4   1911399616  162195235   47559   38.33
+
 gob-stream  zstd    1   1911399616  249810424   2637    691.26
 gob-stream  zstd    3   1911399616  208192146   3490    522.31
 gob-stream  zstd    6   1911399616  193632038   6687    272.56
 gob-stream  zstd    9   1911399616  177620386   16175   112.70
-gob-stream  gzstd   1   1911399616  357382641   10251   177.82
-gob-stream  gzkp    1   1911399616  362156523   5695    320.08
+
+gob-stream  gzstd   1   1911399616  357382013   9046    201.49
+gob-stream  gzkp    1   1911399616  359136669   4885    373.08
 
 The test data for the Large Text Compression Benchmark is the first
 10^9 bytes of the English Wikipedia dump on Mar. 3, 2006.
 http://mattmahoney.net/dc/textdata.html
 
 file    out level   insize      outsize     millis  mb/s
-enwik9  zskp    1   1000000000  343848582   3609    264.18
-enwik9  zskp    2   1000000000  317276632   5746    165.97
-enwik9  zskp    3   1000000000  292243069   12162   78.41
-enwik9  zskp    4   1000000000  275241169   36430   26.18
+enwik9  zskp    1   1000000000  343833605   3687    258.64
+enwik9  zskp    2   1000000000  317001237   7672    124.29
+enwik9  zskp    3   1000000000  291915823   15923   59.89
+enwik9  zskp    4   1000000000  261710291   77697   12.27
+
 enwik9  zstd    1   1000000000  358072021   3110    306.65
 enwik9  zstd    3   1000000000  313734672   4784    199.35
 enwik9  zstd    6   1000000000  295138875   10290   92.68
 enwik9  zstd    9   1000000000  278348700   28549   33.40
-enwik9  gzstd   1   1000000000  382578136   9604    99.30
-enwik9  gzkp    1   1000000000  383825945   6544    145.73
+
+enwik9  gzstd   1   1000000000  382578136   8608    110.78
+enwik9  gzkp    1   1000000000  382781160   5628    169.45
 
 Highly compressible JSON file.
 https://files.klauspost.com/compress/github-june-2days-2019.json.zst
 
 file                        out level   insize      outsize     millis  mb/s
-github-june-2days-2019.json zskp    1   6273951764  699045015   10620   563.40
-github-june-2days-2019.json zskp    2   6273951764  617881763   11687   511.96
-github-june-2days-2019.json zskp    3   6273951764  524340691   34043   175.75
-github-june-2days-2019.json zskp    4   6273951764  503314661   93811   63.78
+github-june-2days-2019.json zskp    1   6273951764  697439532   9789    611.17
+github-june-2days-2019.json zskp    2   6273951764  610876538   18553   322.49
+github-june-2days-2019.json zskp    3   6273951764  517662858   44186   135.41
+github-june-2days-2019.json zskp    4   6273951764  464617114   165373  36.18
+
 github-june-2days-2019.json zstd    1   6273951764  766284037   8450    708.00
 github-june-2days-2019.json zstd    3   6273951764  661889476   10927   547.57
 github-june-2days-2019.json zstd    6   6273951764  642756859   22996   260.18
 github-june-2days-2019.json zstd    9   6273951764  601974523   52413   114.16
-github-june-2days-2019.json gzstd   1   6273951764  1164400847  29948   199.79
-github-june-2days-2019.json gzkp    1   6273951764  1128755542  19236   311.03
+
+github-june-2days-2019.json gzstd   1   6273951764  1164397768  26793   223.32
+github-june-2days-2019.json gzkp    1   6273951764  1120631856  17693   338.16
 
 VM Image, Linux mint with a few installed applications:
 https://files.klauspost.com/compress/rawstudio-mint14.7z
 
 file                    out level   insize      outsize     millis  mb/s
-rawstudio-mint14.tar    zskp    1   8558382592  3667489370  20210   403.84
-rawstudio-mint14.tar    zskp    2   8558382592  3364592300  31873   256.07
-rawstudio-mint14.tar    zskp    3   8558382592  3158085214  77675   105.08
-rawstudio-mint14.tar    zskp    4   8558382592  3020370044  404956  20.16
+rawstudio-mint14.tar    zskp    1   8558382592  3718400221  18206   448.29
+rawstudio-mint14.tar    zskp    2   8558382592  3326118337  37074   220.15
+rawstudio-mint14.tar    zskp    3   8558382592  3163842361  87306   93.49
+rawstudio-mint14.tar    zskp    4   8558382592  2970480650  783862  10.41
+
 rawstudio-mint14.tar    zstd    1   8558382592  3609250104  17136   476.27
 rawstudio-mint14.tar    zstd    3   8558382592  3341679997  29262   278.92
 rawstudio-mint14.tar    zstd    6   8558382592  3235846406  77904   104.77
 rawstudio-mint14.tar    zstd    9   8558382592  3160778861  140946  57.91
-rawstudio-mint14.tar    gzstd   1   8558382592  3926257486  57722   141.40
-rawstudio-mint14.tar    gzkp    1   8558382592  3970463184  41749   195.49
+
+rawstudio-mint14.tar    gzstd   1   8558382592  3926234992  51345   158.96
+rawstudio-mint14.tar    gzkp    1   8558382592  3960117298  36722   222.26
 
 CSV data:
 https://files.klauspost.com/compress/nyc-taxi-data-10M.csv.zst
 
 file                    out level   insize      outsize     millis  mb/s
-nyc-taxi-data-10M.csv   zskp    1   3325605752  641339945   8925    355.35
-nyc-taxi-data-10M.csv   zskp    2   3325605752  591748091   11268   281.44
-nyc-taxi-data-10M.csv   zskp    3   3325605752  530289687   25239   125.66
-nyc-taxi-data-10M.csv   zskp    4   3325605752  490907191   65939   48.10
+nyc-taxi-data-10M.csv   zskp    1   3325605752  641319332   9462    335.17
+nyc-taxi-data-10M.csv   zskp    2   3325605752  588976126   17570   180.50
+nyc-taxi-data-10M.csv   zskp    3   3325605752  529329260   32432   97.79
+nyc-taxi-data-10M.csv   zskp    4   3325605752  474949772   138025  22.98
+
 nyc-taxi-data-10M.csv   zstd    1   3325605752  687399637   8233    385.18
 nyc-taxi-data-10M.csv   zstd    3   3325605752  598514411   10065   315.07
 nyc-taxi-data-10M.csv   zstd    6   3325605752  570522953   20038   158.27
 nyc-taxi-data-10M.csv   zstd    9   3325605752  517554797   64565   49.12
-nyc-taxi-data-10M.csv   gzstd   1   3325605752  928656485   23876   132.83
-nyc-taxi-data-10M.csv   gzkp    1   3325605752  924718719   16388   193.53
+
+nyc-taxi-data-10M.csv   gzstd   1   3325605752  928654908   21270   149.11
+nyc-taxi-data-10M.csv   gzkp    1   3325605752  922273214   13929   227.68
 ```
 
 ## Decompressor
@@ -274,8 +287,13 @@
 }
 ```
 
-It is important to use the "Close" function when you no longer need the Reader to stop running goroutines. 
-See "Allocation-less operation" below.
+It is important to use the "Close" function when you no longer need the Reader to stop running goroutines, 
+when running with default settings.
+Goroutines will exit once an error has been returned, including `io.EOF` at the end of a stream.
+
+Streams are decoded concurrently in 4 asynchronous stages to give the best possible throughput.
+However, if you prefer synchronous decompression, use `WithDecoderConcurrency(1)` which will decompress data 
+as it is being requested only.
 
 For decoding buffers, it could look something like this:
 
@@ -284,7 +302,7 @@
 
 // Create a reader that caches decompressors.
 // For this operation type we supply a nil Reader.
-var decoder, _ = zstd.NewReader(nil)
+var decoder, _ = zstd.NewReader(nil, WithDecoderConcurrency(0))
 
 // Decompress a buffer. We don't supply a destination buffer,
 // so it will be allocated by the decoder.
@@ -294,9 +312,12 @@
 ```
 
 Both of these cases should provide the functionality needed. 
-The decoder can be used for *concurrent* decompression of multiple buffers. 
+The decoder can be used for *concurrent* decompression of multiple buffers.
+By default 4 decompressors will be created. 
+
 It will only allow a certain number of concurrent operations to run. 
-To tweak that yourself use the `WithDecoderConcurrency(n)` option when creating the decoder.   
+To tweak that yourself use the `WithDecoderConcurrency(n)` option when creating the decoder.
+It is possible to use `WithDecoderConcurrency(0)` to create GOMAXPROCS decoders.
 
 ### Dictionaries
 
@@ -348,70 +369,71 @@
 The buffer decoder does everything on the same goroutine and does nothing concurrently.
 It can however decode several buffers concurrently. Use `WithDecoderConcurrency(n)` to limit that.
 
-The stream decoder operates on
+The stream decoder will create goroutines that:
 
-* One goroutine reads input and splits the input to several block decoders.
-* A number of decoders will decode blocks.
-* A goroutine coordinates these blocks and sends history from one to the next.
+1) Reads input and splits the input into blocks.
+2) Decompression of literals.
+3) Decompression of sequences.
+4) Reconstruction of output stream.
 
 So effectively this also means the decoder will "read ahead" and prepare data to always be available for output.
 
+The concurrency level will, for streams, determine how many blocks ahead the compression will start.
+
 Since "blocks" are quite dependent on the output of the previous block stream decoding will only have limited concurrency.
 
-In practice this means that concurrency is often limited to utilizing about 2 cores effectively.
- 
- 
+In practice this means that concurrency is often limited to utilizing about 3 cores effectively.
+  
 ### Benchmarks
 
-These are some examples of performance compared to [datadog cgo library](https://github.com/DataDog/zstd).
-
 The first two are streaming decodes and the last are smaller inputs. 
- 
+
+Running on AMD Ryzen 9 3950X 16-Core Processor. AMD64 assembly used.
+
 ```
-BenchmarkDecoderSilesia-8                          3     385000067 ns/op     550.51 MB/s        5498 B/op          8 allocs/op
-BenchmarkDecoderSilesiaCgo-8                       6     197666567 ns/op    1072.25 MB/s      270672 B/op          8 allocs/op
+BenchmarkDecoderSilesia-32    	                   5	 206878840 ns/op	1024.50 MB/s	   49808 B/op	      43 allocs/op
+BenchmarkDecoderEnwik9-32                          1	1271809000 ns/op	 786.28 MB/s	   72048 B/op	      52 allocs/op
 
-BenchmarkDecoderEnwik9-8                           1    2027001600 ns/op     493.34 MB/s       10496 B/op         18 allocs/op
-BenchmarkDecoderEnwik9Cgo-8                        2     979499200 ns/op    1020.93 MB/s      270672 B/op          8 allocs/op
+Concurrent blocks, performance:
 
-Concurrent performance:
-
-BenchmarkDecoder_DecodeAllParallel/kppkn.gtb.zst-16                28915         42469 ns/op    4340.07 MB/s         114 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/geo.protodata.zst-16           116505          9965 ns/op    11900.16 MB/s         16 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/plrabn12.txt.zst-16              8952        134272 ns/op    3588.70 MB/s         915 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/lcet10.txt.zst-16               11820        102538 ns/op    4161.90 MB/s         594 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/asyoulik.txt.zst-16             34782         34184 ns/op    3661.88 MB/s          60 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/alice29.txt.zst-16              27712         43447 ns/op    3500.58 MB/s          99 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/html_x_4.zst-16                 62826         18750 ns/op    21845.10 MB/s        104 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/paper-100k.pdf.zst-16          631545          1794 ns/op    57078.74 MB/s          2 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/fireworks.jpeg.zst-16         1690140           712 ns/op    172938.13 MB/s         1 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/urls.10K.zst-16                 10432        113593 ns/op    6180.73 MB/s        1143 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/html.zst-16                    113206         10671 ns/op    9596.27 MB/s          15 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/comp-data.bin.zst-16          1530615           779 ns/op    5229.49 MB/s           0 B/op          0 allocs/op
-
-BenchmarkDecoder_DecodeAllParallelCgo/kppkn.gtb.zst-16             65217         16192 ns/op    11383.34 MB/s         46 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/geo.protodata.zst-16        292671          4039 ns/op    29363.19 MB/s          6 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/plrabn12.txt.zst-16          26314         46021 ns/op    10470.43 MB/s        293 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/lcet10.txt.zst-16            33897         34900 ns/op    12227.96 MB/s        205 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/asyoulik.txt.zst-16         104348         11433 ns/op    10949.01 MB/s         20 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/alice29.txt.zst-16           75949         15510 ns/op    9805.60 MB/s          32 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/html_x_4.zst-16             173910          6756 ns/op    60624.29 MB/s         37 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/paper-100k.pdf.zst-16       923076          1339 ns/op    76474.87 MB/s          1 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/fireworks.jpeg.zst-16       922920          1351 ns/op    91102.57 MB/s          2 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/urls.10K.zst-16              27649         43618 ns/op    16096.19 MB/s        407 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/html.zst-16                 279073          4160 ns/op    24614.18 MB/s          6 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/comp-data.bin.zst-16        749938          1579 ns/op    2581.71 MB/s           0 B/op          0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/kppkn.gtb.zst-32         	   67356	     17857 ns/op	10321.96 MB/s	        22.48 pct	     102 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/geo.protodata.zst-32     	  266656	      4421 ns/op	26823.21 MB/s	        11.89 pct	      19 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/plrabn12.txt.zst-32      	   20992	     56842 ns/op	8477.17 MB/s	        39.90 pct	     754 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/lcet10.txt.zst-32        	   27456	     43932 ns/op	9714.01 MB/s	        33.27 pct	     524 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/asyoulik.txt.zst-32      	   78432	     15047 ns/op	8319.15 MB/s	        40.34 pct	      66 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/alice29.txt.zst-32       	   65800	     18436 ns/op	8249.63 MB/s	        37.75 pct	      88 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/html_x_4.zst-32          	  102993	     11523 ns/op	35546.09 MB/s	         3.637 pct	     143 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/paper-100k.pdf.zst-32    	 1000000	      1070 ns/op	95720.98 MB/s	        80.53 pct	       3 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/fireworks.jpeg.zst-32    	  749802	      1752 ns/op	70272.35 MB/s	       100.0 pct	       5 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/urls.10K.zst-32          	   22640	     52934 ns/op	13263.37 MB/s	        26.25 pct	    1014 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/html.zst-32              	  226412	      5232 ns/op	19572.27 MB/s	        14.49 pct	      20 B/op	       0 allocs/op
+BenchmarkDecoder_DecodeAllParallel/comp-data.bin.zst-32     	  923041	      1276 ns/op	3194.71 MB/s	        31.26 pct	       0 B/op	       0 allocs/op
 ```
 
-This reflects the performance around May 2020, but this may be out of date.
+This reflects the performance around May 2022, but this may be out of date.
+
+## Zstd inside ZIP files
+
+It is possible to use zstandard to compress individual files inside zip archives.
+While this isn't widely supported it can be useful for internal files.
+
+To support the compression and decompression of these files you must register a compressor and decompressor.
+
+It is highly recommended registering the (de)compressors on individual zip Reader/Writer and NOT
+use the global registration functions. The main reason for this is that 2 registrations from 
+different packages will result in a panic.
+
+It is a good idea to only have a single compressor and decompressor, since they can be used for multiple zip
+files concurrently, and using a single instance will allow reusing some resources.
+
+See [this example](https://pkg.go.dev/github.com/klauspost/compress/zstd#example-ZipCompressor) for 
+how to compress and decompress files inside zip archives.
 
 # Contributions
 
 Contributions are always welcome. 
 For new features/fixes, remember to add tests and for performance enhancements include benchmarks.
 
-For sending files for reproducing errors use a service like [goobox](https://goobox.io/#/upload) or similar to share your files.
-
 For general feedback and experience reports, feel free to open an issue or write me on [Twitter](https://twitter.com/sh0dan).
 
 This package includes the excellent [`github.com/cespare/xxhash`](https://github.com/cespare/xxhash) package Copyright (c) 2016 Caleb Spare.
diff --git a/vendor/github.com/klauspost/compress/zstd/bitreader.go b/vendor/github.com/klauspost/compress/zstd/bitreader.go
index 8544585..97299d4 100644
--- a/vendor/github.com/klauspost/compress/zstd/bitreader.go
+++ b/vendor/github.com/klauspost/compress/zstd/bitreader.go
@@ -7,6 +7,7 @@
 import (
 	"encoding/binary"
 	"errors"
+	"fmt"
 	"io"
 	"math/bits"
 )
@@ -50,16 +51,16 @@
 	if n == 0 /*|| b.bitsRead >= 64 */ {
 		return 0
 	}
-	return b.getBitsFast(n)
+	return int(b.get32BitsFast(n))
 }
 
-// getBitsFast requires that at least one bit is requested every time.
+// get32BitsFast 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) int {
+func (b *bitReader) get32BitsFast(n uint8) uint32 {
 	const regMask = 64 - 1
 	v := uint32((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
 	b.bitsRead += n
-	return int(v)
+	return v
 }
 
 // fillFast() will make sure at least 32 bits are available.
@@ -125,6 +126,9 @@
 func (b *bitReader) close() error {
 	// Release reference.
 	b.in = nil
+	if !b.finished() {
+		return fmt.Errorf("%d extra bits on block, should be 0", b.remain())
+	}
 	if b.bitsRead > 64 {
 		return io.ErrUnexpectedEOF
 	}
diff --git a/vendor/github.com/klauspost/compress/zstd/bitwriter.go b/vendor/github.com/klauspost/compress/zstd/bitwriter.go
index 303ae90..78b3c61 100644
--- a/vendor/github.com/klauspost/compress/zstd/bitwriter.go
+++ b/vendor/github.com/klauspost/compress/zstd/bitwriter.go
@@ -5,8 +5,6 @@
 
 package zstd
 
-import "fmt"
-
 // bitWriter will write bits.
 // First bit will be LSB of the first byte of output.
 type bitWriter struct {
@@ -38,7 +36,7 @@
 	b.nBits += bits
 }
 
-// addBits32NC will add up to 32 bits.
+// addBits32NC will add up to 31 bits.
 // It will not check if there is space for them,
 // so the caller must ensure that it has flushed recently.
 func (b *bitWriter) addBits32NC(value uint32, bits uint8) {
@@ -46,6 +44,26 @@
 	b.nBits += bits
 }
 
+// addBits64NC will add up to 64 bits.
+// There must be space for 32 bits.
+func (b *bitWriter) addBits64NC(value uint64, bits uint8) {
+	if bits <= 31 {
+		b.addBits32Clean(uint32(value), bits)
+		return
+	}
+	b.addBits32Clean(uint32(value), 32)
+	b.flush32()
+	b.addBits32Clean(uint32(value>>32), bits-32)
+}
+
+// addBits32Clean will add up to 32 bits.
+// It will not check if there is space for them.
+// The input must not contain more bits than specified.
+func (b *bitWriter) addBits32Clean(value uint32, bits uint8) {
+	b.bitContainer |= uint64(value) << (b.nBits & 63)
+	b.nBits += bits
+}
+
 // addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated.
 // It will not check if there is space for them, so the caller must ensure that it has flushed recently.
 func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
@@ -53,80 +71,6 @@
 	b.nBits += bits
 }
 
-// flush will flush all pending full bytes.
-// There will be at least 56 bits available for writing when this has been called.
-// Using flush32 is faster, but leaves less space for writing.
-func (b *bitWriter) flush() {
-	v := b.nBits >> 3
-	switch v {
-	case 0:
-	case 1:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-		)
-	case 2:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-		)
-	case 3:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-		)
-	case 4:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-		)
-	case 5:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-		)
-	case 6:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-		)
-	case 7:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-		)
-	case 8:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-			byte(b.bitContainer>>56),
-		)
-	default:
-		panic(fmt.Errorf("bits (%d) > 64", b.nBits))
-	}
-	b.bitContainer >>= v << 3
-	b.nBits &= 7
-}
-
 // flush32 will flush out, so there are at least 32 bits available for writing.
 func (b *bitWriter) flush32() {
 	if b.nBits < 32 {
diff --git a/vendor/github.com/klauspost/compress/zstd/blockdec.go b/vendor/github.com/klauspost/compress/zstd/blockdec.go
index b51d922..7eed729 100644
--- a/vendor/github.com/klauspost/compress/zstd/blockdec.go
+++ b/vendor/github.com/klauspost/compress/zstd/blockdec.go
@@ -5,9 +5,14 @@
 package zstd
 
 import (
+	"bytes"
+	"encoding/binary"
 	"errors"
 	"fmt"
 	"io"
+	"io/ioutil"
+	"os"
+	"path/filepath"
 	"sync"
 
 	"github.com/klauspost/compress/huff0"
@@ -38,14 +43,14 @@
 	// maxCompressedBlockSize is the biggest allowed compressed block size (128KB)
 	maxCompressedBlockSize = 128 << 10
 
+	compressedBlockOverAlloc    = 16
+	maxCompressedBlockSizeAlloc = 128<<10 + compressedBlockOverAlloc
+
 	// Maximum possible block size (all Raw+Uncompressed).
 	maxBlockSize = (1 << 21) - 1
 
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#literals_section_header
-	maxCompressedLiteralSize = 1 << 18
-	maxRLELiteralSize        = 1 << 20
-	maxMatchLen              = 131074
-	maxSequences             = 0x7f00 + 0xffff
+	maxMatchLen  = 131074
+	maxSequences = 0x7f00 + 0xffff
 
 	// We support slightly less than the reference decoder to be able to
 	// use ints on 32 bit archs.
@@ -76,20 +81,27 @@
 	// Window size of the block.
 	WindowSize uint64
 
-	history     chan *history
-	input       chan struct{}
-	result      chan decodeOutput
-	sequenceBuf []seq
-	err         error
-	decWG       sync.WaitGroup
+	err error
+
+	// Check against this crc
+	checkCRC []byte
 
 	// Frame to use for singlethreaded decoding.
 	// Should not be used by the decoder itself since parent may be another frame.
 	localFrame *frameDec
 
+	sequence []seqVals
+
+	async struct {
+		newHist  *history
+		literals []byte
+		seqData  []byte
+		seqSize  int // Size of uncompressed sequences
+		fcs      uint64
+	}
+
 	// Block is RLE, this is the size.
 	RLESize uint32
-	tmp     [4]byte
 
 	Type blockType
 
@@ -109,13 +121,8 @@
 
 func newBlockDec(lowMem bool) *blockDec {
 	b := blockDec{
-		lowMem:  lowMem,
-		result:  make(chan decodeOutput, 1),
-		input:   make(chan struct{}, 1),
-		history: make(chan *history, 1),
+		lowMem: lowMem,
 	}
-	b.decWG.Add(1)
-	go b.startDecoder()
 	return &b
 }
 
@@ -123,44 +130,60 @@
 // Input must be a start of a block and will be at the end of the block when returned.
 func (b *blockDec) reset(br byteBuffer, windowSize uint64) error {
 	b.WindowSize = windowSize
-	tmp := br.readSmall(3)
-	if tmp == nil {
-		if debug {
-			println("Reading block header:", io.ErrUnexpectedEOF)
-		}
-		return io.ErrUnexpectedEOF
+	tmp, err := br.readSmall(3)
+	if err != nil {
+		println("Reading block header:", err)
+		return err
 	}
 	bh := uint32(tmp[0]) | (uint32(tmp[1]) << 8) | (uint32(tmp[2]) << 16)
 	b.Last = bh&1 != 0
 	b.Type = blockType((bh >> 1) & 3)
 	// find size.
 	cSize := int(bh >> 3)
-	maxSize := maxBlockSize
+	maxSize := maxCompressedBlockSizeAlloc
 	switch b.Type {
 	case blockTypeReserved:
 		return ErrReservedBlockType
 	case blockTypeRLE:
+		if cSize > maxCompressedBlockSize || cSize > int(b.WindowSize) {
+			if debugDecoder {
+				printf("rle block too big: csize:%d block: %+v\n", uint64(cSize), b)
+			}
+			return ErrWindowSizeExceeded
+		}
 		b.RLESize = uint32(cSize)
 		if b.lowMem {
 			maxSize = cSize
 		}
 		cSize = 1
 	case blockTypeCompressed:
-		if debug {
+		if debugDecoder {
 			println("Data size on stream:", cSize)
 		}
 		b.RLESize = 0
-		maxSize = maxCompressedBlockSize
+		maxSize = maxCompressedBlockSizeAlloc
 		if windowSize < maxCompressedBlockSize && b.lowMem {
-			maxSize = int(windowSize)
+			maxSize = int(windowSize) + compressedBlockOverAlloc
 		}
 		if cSize > maxCompressedBlockSize || uint64(cSize) > b.WindowSize {
-			if debug {
+			if debugDecoder {
 				printf("compressed block too big: csize:%d block: %+v\n", uint64(cSize), b)
 			}
 			return ErrCompressedSizeTooBig
 		}
+		// Empty compressed blocks must at least be 2 bytes
+		// for Literals_Block_Type and one for Sequences_Section_Header.
+		if cSize < 2 {
+			return ErrBlockTooSmall
+		}
 	case blockTypeRaw:
+		if cSize > maxCompressedBlockSize || cSize > int(b.WindowSize) {
+			if debugDecoder {
+				printf("rle block too big: csize:%d block: %+v\n", uint64(cSize), b)
+			}
+			return ErrWindowSizeExceeded
+		}
+
 		b.RLESize = 0
 		// We do not need a destination for raw blocks.
 		maxSize = -1
@@ -170,19 +193,18 @@
 
 	// Read block data.
 	if cap(b.dataStorage) < cSize {
-		if b.lowMem {
-			b.dataStorage = make([]byte, 0, cSize)
+		if b.lowMem || cSize > maxCompressedBlockSize {
+			b.dataStorage = make([]byte, 0, cSize+compressedBlockOverAlloc)
 		} else {
-			b.dataStorage = make([]byte, 0, maxBlockSize)
+			b.dataStorage = make([]byte, 0, maxCompressedBlockSizeAlloc)
 		}
 	}
 	if cap(b.dst) <= maxSize {
 		b.dst = make([]byte, 0, maxSize+1)
 	}
-	var err error
 	b.data, err = br.readBig(cSize, b.dataStorage)
 	if err != nil {
-		if debug {
+		if debugDecoder {
 			println("Reading block:", err, "(", cSize, ")", len(b.data))
 			printf("%T", br)
 		}
@@ -196,85 +218,14 @@
 	b.Last = true
 	b.Type = blockTypeReserved
 	b.err = err
-	b.input <- struct{}{}
 }
 
 // Close will release resources.
 // Closed blockDec cannot be reset.
 func (b *blockDec) Close() {
-	close(b.input)
-	close(b.history)
-	close(b.result)
-	b.decWG.Wait()
 }
 
-// decodeAsync will prepare decoding the block when it receives input.
-// This will separate output and history.
-func (b *blockDec) startDecoder() {
-	defer b.decWG.Done()
-	for range b.input {
-		//println("blockDec: Got block input")
-		switch b.Type {
-		case blockTypeRLE:
-			if cap(b.dst) < int(b.RLESize) {
-				if b.lowMem {
-					b.dst = make([]byte, b.RLESize)
-				} else {
-					b.dst = make([]byte, maxBlockSize)
-				}
-			}
-			o := decodeOutput{
-				d:   b,
-				b:   b.dst[:b.RLESize],
-				err: nil,
-			}
-			v := b.data[0]
-			for i := range o.b {
-				o.b[i] = v
-			}
-			hist := <-b.history
-			hist.append(o.b)
-			b.result <- o
-		case blockTypeRaw:
-			o := decodeOutput{
-				d:   b,
-				b:   b.data,
-				err: nil,
-			}
-			hist := <-b.history
-			hist.append(o.b)
-			b.result <- o
-		case blockTypeCompressed:
-			b.dst = b.dst[:0]
-			err := b.decodeCompressed(nil)
-			o := decodeOutput{
-				d:   b,
-				b:   b.dst,
-				err: err,
-			}
-			if debug {
-				println("Decompressed to", len(b.dst), "bytes, error:", err)
-			}
-			b.result <- o
-		case blockTypeReserved:
-			// Used for returning errors.
-			<-b.history
-			b.result <- decodeOutput{
-				d:   b,
-				b:   nil,
-				err: b.err,
-			}
-		default:
-			panic("Invalid block type")
-		}
-		if debug {
-			println("blockDec: Finished block")
-		}
-	}
-}
-
-// decodeAsync will prepare decoding the block when it receives the history.
-// If history is provided, it will not fetch it from the channel.
+// decodeBuf
 func (b *blockDec) decodeBuf(hist *history) error {
 	switch b.Type {
 	case blockTypeRLE:
@@ -297,14 +248,23 @@
 		return nil
 	case blockTypeCompressed:
 		saved := b.dst
-		b.dst = hist.b
-		hist.b = nil
+		// Append directly to history
+		if hist.ignoreBuffer == 0 {
+			b.dst = hist.b
+			hist.b = nil
+		} else {
+			b.dst = b.dst[:0]
+		}
 		err := b.decodeCompressed(hist)
-		if debug {
+		if debugDecoder {
 			println("Decompressed to total", len(b.dst), "bytes, hash:", xxhash.Sum64(b.dst), "error:", err)
 		}
-		hist.b = b.dst
-		b.dst = saved
+		if hist.ignoreBuffer == 0 {
+			hist.b = b.dst
+			b.dst = saved
+		} else {
+			hist.appendKeep(b.dst)
+		}
 		return err
 	case blockTypeReserved:
 		// Used for returning errors.
@@ -314,30 +274,18 @@
 	}
 }
 
-// decodeCompressed will start decompressing a block.
-// If no history is supplied the decoder will decodeAsync as much as possible
-// before fetching from blockDec.history
-func (b *blockDec) decodeCompressed(hist *history) error {
-	in := b.data
-	delayedHistory := hist == nil
-
-	if delayedHistory {
-		// We must always grab history.
-		defer func() {
-			if hist == nil {
-				<-b.history
-			}
-		}()
-	}
+func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err error) {
 	// There must be at least one byte for Literals_Block_Type and one for Sequences_Section_Header
 	if len(in) < 2 {
-		return ErrBlockTooSmall
+		return in, ErrBlockTooSmall
 	}
+
 	litType := literalsBlockType(in[0] & 3)
 	var litRegenSize int
 	var litCompSize int
 	sizeFormat := (in[0] >> 2) & 3
 	var fourStreams bool
+	var literals []byte
 	switch litType {
 	case literalsBlockRaw, literalsBlockRLE:
 		switch sizeFormat {
@@ -353,7 +301,7 @@
 			//  Regenerated_Size uses 20 bits (0-1048575). Literals_Section_Header uses 3 bytes.
 			if len(in) < 3 {
 				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
+				return in, ErrBlockTooSmall
 			}
 			litRegenSize = int(in[0]>>4) + (int(in[1]) << 4) + (int(in[2]) << 12)
 			in = in[3:]
@@ -364,7 +312,7 @@
 			// Both Regenerated_Size and Compressed_Size use 10 bits (0-1023).
 			if len(in) < 3 {
 				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
+				return in, ErrBlockTooSmall
 			}
 			n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12)
 			litRegenSize = int(n & 1023)
@@ -375,7 +323,7 @@
 			fourStreams = true
 			if len(in) < 4 {
 				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
+				return in, ErrBlockTooSmall
 			}
 			n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20)
 			litRegenSize = int(n & 16383)
@@ -385,7 +333,7 @@
 			fourStreams = true
 			if len(in) < 5 {
 				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
+				return in, ErrBlockTooSmall
 			}
 			n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20) + (uint64(in[4]) << 28)
 			litRegenSize = int(n & 262143)
@@ -393,16 +341,18 @@
 			in = in[5:]
 		}
 	}
-	if debug {
+	if debugDecoder {
 		println("literals type:", litType, "litRegenSize:", litRegenSize, "litCompSize:", litCompSize, "sizeFormat:", sizeFormat, "4X:", fourStreams)
 	}
-	var literals []byte
-	var huff *huff0.Scratch
+	if litRegenSize > int(b.WindowSize) || litRegenSize > maxCompressedBlockSize {
+		return in, ErrWindowSizeExceeded
+	}
+
 	switch litType {
 	case literalsBlockRaw:
 		if len(in) < litRegenSize {
 			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litRegenSize)
-			return ErrBlockTooSmall
+			return in, ErrBlockTooSmall
 		}
 		literals = in[:litRegenSize]
 		in = in[litRegenSize:]
@@ -410,19 +360,13 @@
 	case literalsBlockRLE:
 		if len(in) < 1 {
 			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", 1)
-			return ErrBlockTooSmall
+			return in, ErrBlockTooSmall
 		}
 		if cap(b.literalBuf) < litRegenSize {
 			if b.lowMem {
-				b.literalBuf = make([]byte, litRegenSize)
+				b.literalBuf = make([]byte, litRegenSize, litRegenSize+compressedBlockOverAlloc)
 			} else {
-				if litRegenSize > maxCompressedLiteralSize {
-					// Exceptional
-					b.literalBuf = make([]byte, litRegenSize)
-				} else {
-					b.literalBuf = make([]byte, litRegenSize, maxCompressedLiteralSize)
-
-				}
+				b.literalBuf = make([]byte, litRegenSize, maxCompressedBlockSize+compressedBlockOverAlloc)
 			}
 		}
 		literals = b.literalBuf[:litRegenSize]
@@ -431,45 +375,79 @@
 			literals[i] = v
 		}
 		in = in[1:]
-		if debug {
+		if debugDecoder {
 			printf("Found %d RLE compressed literals\n", litRegenSize)
 		}
 	case literalsBlockTreeless:
 		if len(in) < litCompSize {
 			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize)
-			return ErrBlockTooSmall
+			return in, ErrBlockTooSmall
 		}
 		// Store compressed literals, so we defer decoding until we get history.
 		literals = in[:litCompSize]
 		in = in[litCompSize:]
-		if debug {
+		if debugDecoder {
 			printf("Found %d compressed literals\n", litCompSize)
 		}
-	case literalsBlockCompressed:
-		if len(in) < litCompSize {
-			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize)
-			return ErrBlockTooSmall
+		huff := hist.huffTree
+		if huff == nil {
+			return in, errors.New("literal block was treeless, but no history was defined")
 		}
-		literals = in[:litCompSize]
-		in = in[litCompSize:]
-		huff = huffDecoderPool.Get().(*huff0.Scratch)
-		var err error
 		// Ensure we have space to store it.
 		if cap(b.literalBuf) < litRegenSize {
 			if b.lowMem {
-				b.literalBuf = make([]byte, 0, litRegenSize)
+				b.literalBuf = make([]byte, 0, litRegenSize+compressedBlockOverAlloc)
 			} else {
-				b.literalBuf = make([]byte, 0, maxCompressedLiteralSize)
+				b.literalBuf = make([]byte, 0, maxCompressedBlockSize+compressedBlockOverAlloc)
 			}
 		}
-		if huff == nil {
-			huff = &huff0.Scratch{}
+		var err error
+		// Use our out buffer.
+		huff.MaxDecodedSize = litRegenSize
+		if fourStreams {
+			literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals)
+		} else {
+			literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals)
 		}
+		// Make sure we don't leak our literals buffer
+		if err != nil {
+			println("decompressing literals:", err)
+			return in, err
+		}
+		if len(literals) != litRegenSize {
+			return in, fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
+		}
+
+	case literalsBlockCompressed:
+		if len(in) < litCompSize {
+			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize)
+			return in, ErrBlockTooSmall
+		}
+		literals = in[:litCompSize]
+		in = in[litCompSize:]
+		// Ensure we have space to store it.
+		if cap(b.literalBuf) < litRegenSize {
+			if b.lowMem {
+				b.literalBuf = make([]byte, 0, litRegenSize+compressedBlockOverAlloc)
+			} else {
+				b.literalBuf = make([]byte, 0, maxCompressedBlockSize+compressedBlockOverAlloc)
+			}
+		}
+		huff := hist.huffTree
+		if huff == nil || (hist.dict != nil && huff == hist.dict.litEnc) {
+			huff = huffDecoderPool.Get().(*huff0.Scratch)
+			if huff == nil {
+				huff = &huff0.Scratch{}
+			}
+		}
+		var err error
 		huff, literals, err = huff0.ReadTable(literals, huff)
 		if err != nil {
 			println("reading huffman table:", err)
-			return err
+			return in, err
 		}
+		hist.huffTree = huff
+		huff.MaxDecodedSize = litRegenSize
 		// Use our out buffer.
 		if fourStreams {
 			literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals)
@@ -478,27 +456,63 @@
 		}
 		if err != nil {
 			println("decoding compressed literals:", err)
-			return err
+			return in, err
 		}
 		// Make sure we don't leak our literals buffer
 		if len(literals) != litRegenSize {
-			return fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
+			return in, fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
 		}
-		if debug {
+		// Re-cap to get extra size.
+		literals = b.literalBuf[:len(literals)]
+		if debugDecoder {
 			printf("Decompressed %d literals into %d bytes\n", litCompSize, litRegenSize)
 		}
 	}
+	hist.decoders.literals = literals
+	return in, nil
+}
 
+// decodeCompressed will start decompressing a block.
+func (b *blockDec) decodeCompressed(hist *history) error {
+	in := b.data
+	in, err := b.decodeLiterals(in, hist)
+	if err != nil {
+		return err
+	}
+	err = b.prepareSequences(in, hist)
+	if err != nil {
+		return err
+	}
+	if hist.decoders.nSeqs == 0 {
+		b.dst = append(b.dst, hist.decoders.literals...)
+		return nil
+	}
+	before := len(hist.decoders.out)
+	err = hist.decoders.decodeSync(hist.b[hist.ignoreBuffer:])
+	if err != nil {
+		return err
+	}
+	if hist.decoders.maxSyncLen > 0 {
+		hist.decoders.maxSyncLen += uint64(before)
+		hist.decoders.maxSyncLen -= uint64(len(hist.decoders.out))
+	}
+	b.dst = hist.decoders.out
+	hist.recentOffsets = hist.decoders.prevOffset
+	return nil
+}
+
+func (b *blockDec) prepareSequences(in []byte, hist *history) (err error) {
+	if debugDecoder {
+		printf("prepareSequences: %d byte(s) input\n", len(in))
+	}
 	// Decode Sequences
 	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#sequences-section
 	if len(in) < 1 {
 		return ErrBlockTooSmall
 	}
+	var nSeqs int
 	seqHeader := in[0]
-	nSeqs := 0
 	switch {
-	case seqHeader == 0:
-		in = in[1:]
 	case seqHeader < 128:
 		nSeqs = int(seqHeader)
 		in = in[1:]
@@ -515,19 +529,16 @@
 		nSeqs = 0x7f00 + int(in[1]) + (int(in[2]) << 8)
 		in = in[3:]
 	}
-	// Allocate sequences
-	if cap(b.sequenceBuf) < nSeqs {
-		if b.lowMem {
-			b.sequenceBuf = make([]seq, nSeqs)
-		} else {
-			// Allocate max
-			b.sequenceBuf = make([]seq, nSeqs, maxSequences)
+	if nSeqs == 0 && len(in) != 0 {
+		// When no sequences, there should not be any more data...
+		if debugDecoder {
+			printf("prepareSequences: 0 sequences, but %d byte(s) left on stream\n", len(in))
 		}
-	} else {
-		// Reuse buffer
-		b.sequenceBuf = b.sequenceBuf[:nSeqs]
+		return ErrUnexpectedBlockSize
 	}
-	var seqs = &sequenceDecs{}
+
+	var seqs = &hist.decoders
+	seqs.nSeqs = nSeqs
 	if nSeqs > 0 {
 		if len(in) < 1 {
 			return ErrBlockTooSmall
@@ -535,12 +546,12 @@
 		br := byteReader{b: in, off: 0}
 		compMode := br.Uint8()
 		br.advance(1)
-		if debug {
+		if debugDecoder {
 			printf("Compression modes: 0b%b", compMode)
 		}
 		for i := uint(0); i < 3; i++ {
 			mode := seqCompMode((compMode >> (6 - i*2)) & 3)
-			if debug {
+			if debugDecoder {
 				println("Table", tableIndex(i), "is", mode)
 			}
 			var seq *sequenceDec
@@ -556,6 +567,9 @@
 			}
 			switch mode {
 			case compModePredefined:
+				if seq.fse != nil && !seq.fse.preDefined {
+					fseDecoderPool.Put(seq.fse)
+				}
 				seq.fse = &fsePredef[i]
 			case compModeRLE:
 				if br.remain() < 1 {
@@ -563,34 +577,36 @@
 				}
 				v := br.Uint8()
 				br.advance(1)
-				dec := fseDecoderPool.Get().(*fseDecoder)
+				if seq.fse == nil || seq.fse.preDefined {
+					seq.fse = fseDecoderPool.Get().(*fseDecoder)
+				}
 				symb, err := decSymbolValue(v, symbolTableX[i])
 				if err != nil {
 					printf("RLE Transform table (%v) error: %v", tableIndex(i), err)
 					return err
 				}
-				dec.setRLE(symb)
-				seq.fse = dec
-				if debug {
+				seq.fse.setRLE(symb)
+				if debugDecoder {
 					printf("RLE set to %+v, code: %v", symb, v)
 				}
 			case compModeFSE:
 				println("Reading table for", tableIndex(i))
-				dec := fseDecoderPool.Get().(*fseDecoder)
-				err := dec.readNCount(&br, uint16(maxTableSymbol[i]))
+				if seq.fse == nil || seq.fse.preDefined {
+					seq.fse = fseDecoderPool.Get().(*fseDecoder)
+				}
+				err := seq.fse.readNCount(&br, uint16(maxTableSymbol[i]))
 				if err != nil {
 					println("Read table error:", err)
 					return err
 				}
-				err = dec.transform(symbolTableX[i])
+				err = seq.fse.transform(symbolTableX[i])
 				if err != nil {
 					println("Transform table error:", err)
 					return err
 				}
-				if debug {
-					println("Read table ok", "symbolLen:", dec.symbolLen)
+				if debugDecoder {
+					println("Read table ok", "symbolLen:", seq.fse.symbolLen)
 				}
-				seq.fse = dec
 			case compModeRepeat:
 				seq.repeat = true
 			}
@@ -600,140 +616,106 @@
 		}
 		in = br.unread()
 	}
-
-	// Wait for history.
-	// All time spent after this is critical since it is strictly sequential.
-	if hist == nil {
-		hist = <-b.history
-		if hist.error {
-			return ErrDecoderClosed
-		}
-	}
-
-	// Decode treeless literal block.
-	if litType == literalsBlockTreeless {
-		// TODO: We could send the history early WITHOUT the stream history.
-		//   This would allow decoding treeless literals before the byte history is available.
-		//   Silencia stats: Treeless 4393, with: 32775, total: 37168, 11% treeless.
-		//   So not much obvious gain here.
-
-		if hist.huffTree == nil {
-			return errors.New("literal block was treeless, but no history was defined")
-		}
-		// Ensure we have space to store it.
-		if cap(b.literalBuf) < litRegenSize {
-			if b.lowMem {
-				b.literalBuf = make([]byte, 0, litRegenSize)
-			} else {
-				b.literalBuf = make([]byte, 0, maxCompressedLiteralSize)
-			}
-		}
-		var err error
-		// Use our out buffer.
-		huff = hist.huffTree
-		if fourStreams {
-			literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals)
-		} else {
-			literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals)
-		}
-		// Make sure we don't leak our literals buffer
-		if err != nil {
-			println("decompressing literals:", err)
-			return err
-		}
-		if len(literals) != litRegenSize {
-			return fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
-		}
-	} else {
-		if hist.huffTree != nil && huff != nil {
-			if hist.dict == nil || hist.dict.litEnc != hist.huffTree {
-				huffDecoderPool.Put(hist.huffTree)
-			}
-			hist.huffTree = nil
-		}
-	}
-	if huff != nil {
-		hist.huffTree = huff
-	}
-	if debug {
-		println("Final literals:", len(literals), "hash:", xxhash.Sum64(literals), "and", nSeqs, "sequences.")
+	if debugDecoder {
+		println("Literals:", len(seqs.literals), "hash:", xxhash.Sum64(seqs.literals), "and", seqs.nSeqs, "sequences.")
 	}
 
 	if nSeqs == 0 {
-		// Decompressed content is defined entirely as Literals Section content.
-		b.dst = append(b.dst, literals...)
-		if delayedHistory {
-			hist.append(literals)
+		if len(b.sequence) > 0 {
+			b.sequence = b.sequence[:0]
 		}
 		return nil
 	}
-
-	seqs, err := seqs.mergeHistory(&hist.decoders)
-	if err != nil {
-		return err
+	br := seqs.br
+	if br == nil {
+		br = &bitReader{}
 	}
-	if debug {
-		println("History merged ok")
-	}
-	br := &bitReader{}
 	if err := br.init(in); err != nil {
 		return err
 	}
 
-	// TODO: Investigate if sending history without decoders are faster.
-	//   This would allow the sequences to be decoded async and only have to construct stream history.
-	//   If only recent offsets were not transferred, this would be an obvious win.
-	// 	 Also, if first 3 sequences don't reference recent offsets, all sequences can be decoded.
+	if err := seqs.initialize(br, hist, b.dst); err != nil {
+		println("initializing sequences:", err)
+		return err
+	}
+	// Extract blocks...
+	if false && hist.dict == nil {
+		fatalErr := func(err error) {
+			if err != nil {
+				panic(err)
+			}
+		}
+		fn := fmt.Sprintf("n-%d-lits-%d-prev-%d-%d-%d-win-%d.blk", hist.decoders.nSeqs, len(hist.decoders.literals), hist.recentOffsets[0], hist.recentOffsets[1], hist.recentOffsets[2], hist.windowSize)
+		var buf bytes.Buffer
+		fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.litLengths.fse))
+		fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.matchLengths.fse))
+		fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.offsets.fse))
+		buf.Write(in)
+		ioutil.WriteFile(filepath.Join("testdata", "seqs", fn), buf.Bytes(), os.ModePerm)
+	}
 
+	return nil
+}
+
+func (b *blockDec) decodeSequences(hist *history) error {
+	if cap(b.sequence) < hist.decoders.nSeqs {
+		if b.lowMem {
+			b.sequence = make([]seqVals, 0, hist.decoders.nSeqs)
+		} else {
+			b.sequence = make([]seqVals, 0, 0x7F00+0xffff)
+		}
+	}
+	b.sequence = b.sequence[:hist.decoders.nSeqs]
+	if hist.decoders.nSeqs == 0 {
+		hist.decoders.seqSize = len(hist.decoders.literals)
+		return nil
+	}
+	hist.decoders.windowSize = hist.windowSize
+	hist.decoders.prevOffset = hist.recentOffsets
+
+	err := hist.decoders.decode(b.sequence)
+	hist.recentOffsets = hist.decoders.prevOffset
+	return err
+}
+
+func (b *blockDec) executeSequences(hist *history) error {
 	hbytes := hist.b
 	if len(hbytes) > hist.windowSize {
 		hbytes = hbytes[len(hbytes)-hist.windowSize:]
-		// We do not need history any more.
+		// We do not need history anymore.
 		if hist.dict != nil {
 			hist.dict.content = nil
 		}
 	}
-
-	if err := seqs.initialize(br, hist, literals, b.dst); err != nil {
-		println("initializing sequences:", err)
-		return err
-	}
-
-	err = seqs.decode(nSeqs, br, hbytes)
+	hist.decoders.windowSize = hist.windowSize
+	hist.decoders.out = b.dst[:0]
+	err := hist.decoders.execute(b.sequence, hbytes)
 	if err != nil {
 		return err
 	}
-	if !br.finished() {
-		return fmt.Errorf("%d extra bits on block, should be 0", br.remain())
-	}
+	return b.updateHistory(hist)
+}
 
-	err = br.close()
-	if err != nil {
-		printf("Closing sequences: %v, %+v\n", err, *br)
-	}
+func (b *blockDec) updateHistory(hist *history) error {
 	if len(b.data) > maxCompressedBlockSize {
 		return fmt.Errorf("compressed block size too large (%d)", len(b.data))
 	}
 	// Set output and release references.
-	b.dst = seqs.out
-	seqs.out, seqs.literals, seqs.hist = nil, nil, nil
+	b.dst = hist.decoders.out
+	hist.recentOffsets = hist.decoders.prevOffset
 
-	if !delayedHistory {
-		// If we don't have delayed history, no need to update.
-		hist.recentOffsets = seqs.prevOffset
-		return nil
-	}
 	if b.Last {
 		// if last block we don't care about history.
 		println("Last block, no history returned")
 		hist.b = hist.b[:0]
 		return nil
+	} else {
+		hist.append(b.dst)
+		if debugDecoder {
+			println("Finished block with ", len(b.sequence), "sequences. Added", len(b.dst), "to history, now length", len(hist.b))
+		}
 	}
-	hist.append(b.dst)
-	hist.recentOffsets = seqs.prevOffset
-	if debug {
-		println("Finished block with literals:", len(literals), "and", nSeqs, "sequences.")
-	}
+	hist.decoders.out, hist.decoders.literals = nil, nil
 
 	return nil
 }
diff --git a/vendor/github.com/klauspost/compress/zstd/blockenc.go b/vendor/github.com/klauspost/compress/zstd/blockenc.go
index e1be092..12e8f6f 100644
--- a/vendor/github.com/klauspost/compress/zstd/blockenc.go
+++ b/vendor/github.com/klauspost/compress/zstd/blockenc.go
@@ -51,7 +51,7 @@
 		if cap(b.literals) < maxCompressedBlockSize {
 			b.literals = make([]byte, 0, maxCompressedBlockSize)
 		}
-		const defSeqs = 200
+		const defSeqs = 2000
 		if cap(b.sequences) < defSeqs {
 			b.sequences = make([]seq, 0, defSeqs)
 		}
@@ -156,7 +156,7 @@
 	switch {
 	case inBits < 5:
 		lh |= (uint64(regenLen) << 3) | (1 << 60)
-		if debug {
+		if debugEncoder {
 			got := int(lh>>3) & 0xff
 			if got != regenLen {
 				panic(fmt.Sprint("litRegenSize = ", regenLen, "(want) != ", got, "(got)"))
@@ -184,7 +184,7 @@
 			lh |= 1 << 2
 		}
 		lh |= (uint64(inLen) << 4) | (uint64(compLen) << (10 + 4)) | (3 << 60)
-		if debug {
+		if debugEncoder {
 			const mmask = (1 << 24) - 1
 			n := (lh >> 4) & mmask
 			if int(n&1023) != inLen {
@@ -312,7 +312,7 @@
 	bh.setType(blockTypeRaw)
 	b.output = bh.appendTo(b.output[:0])
 	b.output = append(b.output, a...)
-	if debug {
+	if debugEncoder {
 		println("Adding RAW block, length", len(a), "last:", b.last)
 	}
 }
@@ -325,7 +325,7 @@
 	bh.setType(blockTypeRaw)
 	dst = bh.appendTo(dst)
 	dst = append(dst, src...)
-	if debug {
+	if debugEncoder {
 		println("Adding RAW block, length", len(src), "last:", b.last)
 	}
 	return dst
@@ -339,7 +339,7 @@
 
 	// Don't compress extremely small blocks
 	if len(lits) < 8 || (len(lits) < 32 && b.dictLitEnc == nil) || raw {
-		if debug {
+		if debugEncoder {
 			println("Adding RAW block, length", len(lits), "last:", b.last)
 		}
 		bh.setType(blockTypeRaw)
@@ -371,7 +371,7 @@
 
 	switch err {
 	case huff0.ErrIncompressible:
-		if debug {
+		if debugEncoder {
 			println("Adding RAW block, length", len(lits), "last:", b.last)
 		}
 		bh.setType(blockTypeRaw)
@@ -379,7 +379,7 @@
 		b.output = append(b.output, lits...)
 		return nil
 	case huff0.ErrUseRLE:
-		if debug {
+		if debugEncoder {
 			println("Adding RLE block, length", len(lits))
 		}
 		bh.setType(blockTypeRLE)
@@ -396,12 +396,12 @@
 	bh.setType(blockTypeCompressed)
 	var lh literalsHeader
 	if reUsed {
-		if debug {
+		if debugEncoder {
 			println("Reused tree, compressed to", len(out))
 		}
 		lh.setType(literalsBlockTreeless)
 	} else {
-		if debug {
+		if debugEncoder {
 			println("New tree, compressed to", len(out), "tree size:", len(b.litEnc.OutTable))
 		}
 		lh.setType(literalsBlockCompressed)
@@ -426,7 +426,7 @@
 		return 0
 	}
 	enc := fseEncoder{}
-	hist := enc.Histogram()[:256]
+	hist := enc.Histogram()
 	maxSym := uint8(0)
 	for i, v := range data {
 		v = v & 63
@@ -517,7 +517,7 @@
 		lh.setSize(len(b.literals))
 		b.output = lh.appendTo(b.output)
 		b.output = append(b.output, b.literals...)
-		if debug {
+		if debugEncoder {
 			println("Adding literals RAW, length", len(b.literals))
 		}
 	case huff0.ErrUseRLE:
@@ -525,22 +525,22 @@
 		lh.setSize(len(b.literals))
 		b.output = lh.appendTo(b.output)
 		b.output = append(b.output, b.literals[0])
-		if debug {
+		if debugEncoder {
 			println("Adding literals RLE")
 		}
 	case nil:
 		// Compressed litLen...
 		if reUsed {
-			if debug {
+			if debugEncoder {
 				println("reused tree")
 			}
 			lh.setType(literalsBlockTreeless)
 		} else {
-			if debug {
+			if debugEncoder {
 				println("new tree, size:", len(b.litEnc.OutTable))
 			}
 			lh.setType(literalsBlockCompressed)
-			if debug {
+			if debugEncoder {
 				_, _, err := huff0.ReadTable(out, nil)
 				if err != nil {
 					panic(err)
@@ -548,18 +548,18 @@
 			}
 		}
 		lh.setSizes(len(out), len(b.literals), single)
-		if debug {
+		if debugEncoder {
 			printf("Compressed %d literals to %d bytes", len(b.literals), len(out))
 			println("Adding literal header:", lh)
 		}
 		b.output = lh.appendTo(b.output)
 		b.output = append(b.output, out...)
 		b.litEnc.Reuse = huff0.ReusePolicyAllow
-		if debug {
+		if debugEncoder {
 			println("Adding literals compressed")
 		}
 	default:
-		if debug {
+		if debugEncoder {
 			println("Adding literals ERROR:", err)
 		}
 		return err
@@ -577,7 +577,7 @@
 		n := len(b.sequences) - 0x7f00
 		b.output = append(b.output, 255, uint8(n), uint8(n>>8))
 	}
-	if debug {
+	if debugEncoder {
 		println("Encoding", len(b.sequences), "sequences")
 	}
 	b.genCodes()
@@ -611,17 +611,17 @@
 		nSize = nSize + (nSize+2*8*16)>>4
 		switch {
 		case predefSize <= prevSize && predefSize <= nSize || forcePreDef:
-			if debug {
+			if debugEncoder {
 				println("Using predefined", predefSize>>3, "<=", nSize>>3)
 			}
 			return preDef, compModePredefined
 		case prevSize <= nSize:
-			if debug {
+			if debugEncoder {
 				println("Using previous", prevSize>>3, "<=", nSize>>3)
 			}
 			return prev, compModeRepeat
 		default:
-			if debug {
+			if debugEncoder {
 				println("Using new, predef", predefSize>>3, ". previous:", prevSize>>3, ">", nSize>>3, "header max:", cur.maxHeaderSize()>>3, "bytes")
 				println("tl:", cur.actualTableLog, "symbolLen:", cur.symbolLen, "norm:", cur.norm[:cur.symbolLen], "hist", cur.count[:cur.symbolLen])
 			}
@@ -634,7 +634,7 @@
 	if llEnc.useRLE {
 		mode |= uint8(compModeRLE) << 6
 		llEnc.setRLE(b.sequences[0].llCode)
-		if debug {
+		if debugEncoder {
 			println("llEnc.useRLE")
 		}
 	} else {
@@ -645,7 +645,7 @@
 	if ofEnc.useRLE {
 		mode |= uint8(compModeRLE) << 4
 		ofEnc.setRLE(b.sequences[0].ofCode)
-		if debug {
+		if debugEncoder {
 			println("ofEnc.useRLE")
 		}
 	} else {
@@ -657,7 +657,7 @@
 	if mlEnc.useRLE {
 		mode |= uint8(compModeRLE) << 2
 		mlEnc.setRLE(b.sequences[0].mlCode)
-		if debug {
+		if debugEncoder {
 			println("mlEnc.useRLE, code: ", b.sequences[0].mlCode, "value", b.sequences[0].matchLen)
 		}
 	} else {
@@ -666,7 +666,7 @@
 		mode |= uint8(m) << 2
 	}
 	b.output = append(b.output, mode)
-	if debug {
+	if debugEncoder {
 		printf("Compression modes: 0b%b", mode)
 	}
 	b.output, err = llEnc.writeCount(b.output)
@@ -722,52 +722,53 @@
 		println("Encoded seq", seq, s, "codes:", s.llCode, s.mlCode, s.ofCode, "states:", ll.state, ml.state, of.state, "bits:", llB, mlB, ofB)
 	}
 	seq--
-	if llEnc.maxBits+mlEnc.maxBits+ofEnc.maxBits <= 32 {
-		// No need to flush (common)
-		for seq >= 0 {
-			s = b.sequences[seq]
-			wr.flush32()
-			llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode]
-			// tabelog max is 8 for all.
-			of.encode(ofB)
-			ml.encode(mlB)
-			ll.encode(llB)
-			wr.flush32()
+	// Store sequences in reverse...
+	for seq >= 0 {
+		s = b.sequences[seq]
 
-			// We checked that all can stay within 32 bits
-			wr.addBits32NC(s.litLen, llB.outBits)
-			wr.addBits32NC(s.matchLen, mlB.outBits)
-			wr.addBits32NC(s.offset, ofB.outBits)
+		ofB := ofTT[s.ofCode]
+		wr.flush32() // tablelog max is below 8 for each, so it will fill max 24 bits.
+		//of.encode(ofB)
+		nbBitsOut := (uint32(of.state) + ofB.deltaNbBits) >> 16
+		dstState := int32(of.state>>(nbBitsOut&15)) + int32(ofB.deltaFindState)
+		wr.addBits16NC(of.state, uint8(nbBitsOut))
+		of.state = of.stateTable[dstState]
 
-			if debugSequences {
-				println("Encoded seq", seq, s)
-			}
+		// Accumulate extra bits.
+		outBits := ofB.outBits & 31
+		extraBits := uint64(s.offset & bitMask32[outBits])
+		extraBitsN := outBits
 
-			seq--
+		mlB := mlTT[s.mlCode]
+		//ml.encode(mlB)
+		nbBitsOut = (uint32(ml.state) + mlB.deltaNbBits) >> 16
+		dstState = int32(ml.state>>(nbBitsOut&15)) + int32(mlB.deltaFindState)
+		wr.addBits16NC(ml.state, uint8(nbBitsOut))
+		ml.state = ml.stateTable[dstState]
+
+		outBits = mlB.outBits & 31
+		extraBits = extraBits<<outBits | uint64(s.matchLen&bitMask32[outBits])
+		extraBitsN += outBits
+
+		llB := llTT[s.llCode]
+		//ll.encode(llB)
+		nbBitsOut = (uint32(ll.state) + llB.deltaNbBits) >> 16
+		dstState = int32(ll.state>>(nbBitsOut&15)) + int32(llB.deltaFindState)
+		wr.addBits16NC(ll.state, uint8(nbBitsOut))
+		ll.state = ll.stateTable[dstState]
+
+		outBits = llB.outBits & 31
+		extraBits = extraBits<<outBits | uint64(s.litLen&bitMask32[outBits])
+		extraBitsN += outBits
+
+		wr.flush32()
+		wr.addBits64NC(extraBits, extraBitsN)
+
+		if debugSequences {
+			println("Encoded seq", seq, s)
 		}
-	} else {
-		for seq >= 0 {
-			s = b.sequences[seq]
-			wr.flush32()
-			llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode]
-			// tabelog max is below 8 for each.
-			of.encode(ofB)
-			ml.encode(mlB)
-			ll.encode(llB)
-			wr.flush32()
 
-			// ml+ll = max 32 bits total
-			wr.addBits32NC(s.litLen, llB.outBits)
-			wr.addBits32NC(s.matchLen, mlB.outBits)
-			wr.flush32()
-			wr.addBits32NC(s.offset, ofB.outBits)
-
-			if debugSequences {
-				println("Encoded seq", seq, s)
-			}
-
-			seq--
-		}
+		seq--
 	}
 	ml.flush(mlEnc.actualTableLog)
 	of.flush(ofEnc.actualTableLog)
@@ -786,7 +787,7 @@
 
 	// Size is output minus block header.
 	bh.setSize(uint32(len(b.output)-bhOffset) - 3)
-	if debug {
+	if debugEncoder {
 		println("Rewriting block header", bh)
 	}
 	_ = bh.appendTo(b.output[bhOffset:bhOffset])
@@ -801,14 +802,13 @@
 		// nothing to do
 		return
 	}
-
 	if len(b.sequences) > math.MaxUint16 {
 		panic("can only encode up to 64K sequences")
 	}
 	// No bounds checks after here:
-	llH := b.coders.llEnc.Histogram()[:256]
-	ofH := b.coders.ofEnc.Histogram()[:256]
-	mlH := b.coders.mlEnc.Histogram()[:256]
+	llH := b.coders.llEnc.Histogram()
+	ofH := b.coders.ofEnc.Histogram()
+	mlH := b.coders.mlEnc.Histogram()
 	for i := range llH {
 		llH[i] = 0
 	}
@@ -820,7 +820,8 @@
 	}
 
 	var llMax, ofMax, mlMax uint8
-	for i, seq := range b.sequences {
+	for i := range b.sequences {
+		seq := &b.sequences[i]
 		v := llCode(seq.litLen)
 		seq.llCode = v
 		llH[v]++
@@ -844,7 +845,6 @@
 				panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d), matchlen: %d", mlMax, seq.matchLen))
 			}
 		}
-		b.sequences[i] = seq
 	}
 	maxCount := func(a []uint32) int {
 		var max uint32
diff --git a/vendor/github.com/klauspost/compress/zstd/bytebuf.go b/vendor/github.com/klauspost/compress/zstd/bytebuf.go
index 658ef78..2ad0207 100644
--- a/vendor/github.com/klauspost/compress/zstd/bytebuf.go
+++ b/vendor/github.com/klauspost/compress/zstd/bytebuf.go
@@ -12,8 +12,8 @@
 
 type byteBuffer interface {
 	// Read up to 8 bytes.
-	// Returns nil if no more input is available.
-	readSmall(n int) []byte
+	// Returns io.ErrUnexpectedEOF if this cannot be satisfied.
+	readSmall(n int) ([]byte, error)
 
 	// Read >8 bytes.
 	// MAY use the destination slice.
@@ -23,23 +23,23 @@
 	readByte() (byte, error)
 
 	// Skip n bytes.
-	skipN(n int) error
+	skipN(n int64) error
 }
 
 // in-memory buffer
 type byteBuf []byte
 
-func (b *byteBuf) readSmall(n int) []byte {
+func (b *byteBuf) readSmall(n int) ([]byte, error) {
 	if debugAsserts && n > 8 {
 		panic(fmt.Errorf("small read > 8 (%d). use readBig", n))
 	}
 	bb := *b
 	if len(bb) < n {
-		return nil
+		return nil, io.ErrUnexpectedEOF
 	}
 	r := bb[:n]
 	*b = bb[n:]
-	return r
+	return r, nil
 }
 
 func (b *byteBuf) readBig(n int, dst []byte) ([]byte, error) {
@@ -52,10 +52,6 @@
 	return r, nil
 }
 
-func (b *byteBuf) remain() []byte {
-	return *b
-}
-
 func (b *byteBuf) readByte() (byte, error) {
 	bb := *b
 	if len(bb) < 1 {
@@ -66,9 +62,12 @@
 	return r, nil
 }
 
-func (b *byteBuf) skipN(n int) error {
+func (b *byteBuf) skipN(n int64) error {
 	bb := *b
-	if len(bb) < n {
+	if n < 0 {
+		return fmt.Errorf("negative skip (%d) requested", n)
+	}
+	if int64(len(bb)) < n {
 		return io.ErrUnexpectedEOF
 	}
 	*b = bb[n:]
@@ -81,19 +80,22 @@
 	tmp [8]byte
 }
 
-func (r *readerWrapper) readSmall(n int) []byte {
+func (r *readerWrapper) readSmall(n int) ([]byte, error) {
 	if debugAsserts && n > 8 {
 		panic(fmt.Errorf("small read > 8 (%d). use readBig", n))
 	}
 	n2, err := io.ReadFull(r.r, r.tmp[:n])
 	// We only really care about the actual bytes read.
-	if n2 != n {
-		if debug {
+	if err != nil {
+		if err == io.EOF {
+			return nil, io.ErrUnexpectedEOF
+		}
+		if debugDecoder {
 			println("readSmall: got", n2, "want", n, "err", err)
 		}
-		return nil
+		return nil, err
 	}
-	return r.tmp[:n]
+	return r.tmp[:n], nil
 }
 
 func (r *readerWrapper) readBig(n int, dst []byte) ([]byte, error) {
@@ -110,6 +112,9 @@
 func (r *readerWrapper) readByte() (byte, error) {
 	n2, err := r.r.Read(r.tmp[:1])
 	if err != nil {
+		if err == io.EOF {
+			err = io.ErrUnexpectedEOF
+		}
 		return 0, err
 	}
 	if n2 != 1 {
@@ -118,9 +123,9 @@
 	return r.tmp[0], nil
 }
 
-func (r *readerWrapper) skipN(n int) error {
-	n2, err := io.CopyN(ioutil.Discard, r.r, int64(n))
-	if n2 != int64(n) {
+func (r *readerWrapper) skipN(n int64) error {
+	n2, err := io.CopyN(ioutil.Discard, r.r, n)
+	if n2 != n {
 		err = io.ErrUnexpectedEOF
 	}
 	return err
diff --git a/vendor/github.com/klauspost/compress/zstd/bytereader.go b/vendor/github.com/klauspost/compress/zstd/bytereader.go
index 2c4fca1..0e59a24 100644
--- a/vendor/github.com/klauspost/compress/zstd/bytereader.go
+++ b/vendor/github.com/klauspost/compress/zstd/bytereader.go
@@ -13,12 +13,6 @@
 	off int
 }
 
-// init will initialize the reader and set the input.
-func (b *byteReader) init(in []byte) {
-	b.b = in
-	b.off = 0
-}
-
 // advance the stream b n bytes.
 func (b *byteReader) advance(n uint) {
 	b.off += int(n)
diff --git a/vendor/github.com/klauspost/compress/zstd/decodeheader.go b/vendor/github.com/klauspost/compress/zstd/decodeheader.go
index 69736e8..5022e71 100644
--- a/vendor/github.com/klauspost/compress/zstd/decodeheader.go
+++ b/vendor/github.com/klauspost/compress/zstd/decodeheader.go
@@ -5,6 +5,7 @@
 
 import (
 	"bytes"
+	"encoding/binary"
 	"errors"
 	"io"
 )
@@ -15,18 +16,50 @@
 
 // Header contains information about the first frame and block within that.
 type Header struct {
-	// Window Size the window of data to keep while decoding.
-	// Will only be set if HasFCS is false.
-	WindowSize uint64
+	// SingleSegment specifies whether the data is to be decompressed into a
+	// single contiguous memory segment.
+	// It implies that WindowSize is invalid and that FrameContentSize is valid.
+	SingleSegment bool
 
-	// Frame content size.
-	// Expected size of the entire frame.
-	FrameContentSize uint64
+	// WindowSize is the window of data to keep while decoding.
+	// Will only be set if SingleSegment is false.
+	WindowSize uint64
 
 	// Dictionary ID.
 	// If 0, no dictionary.
 	DictionaryID uint32
 
+	// HasFCS specifies whether FrameContentSize has a valid value.
+	HasFCS bool
+
+	// FrameContentSize is the expected uncompressed size of the entire frame.
+	FrameContentSize uint64
+
+	// Skippable will be true if the frame is meant to be skipped.
+	// This implies that FirstBlock.OK is false.
+	Skippable bool
+
+	// SkippableID is the user-specific ID for the skippable frame.
+	// Valid values are between 0 to 15, inclusive.
+	SkippableID int
+
+	// SkippableSize is the length of the user data to skip following
+	// the header.
+	SkippableSize uint32
+
+	// HeaderSize is the raw size of the frame header.
+	//
+	// For normal frames, it includes the size of the magic number and
+	// the size of the header (per section 3.1.1.1).
+	// It does not include the size for any data blocks (section 3.1.1.2) nor
+	// the size for the trailing content checksum.
+	//
+	// For skippable frames, this counts the size of the magic number
+	// along with the size of the size field of the payload.
+	// It does not include the size of the skippable payload itself.
+	// The total frame size is the HeaderSize plus the SkippableSize.
+	HeaderSize int
+
 	// First block information.
 	FirstBlock struct {
 		// OK will be set if first block could be decoded.
@@ -51,17 +84,9 @@
 		CompressedSize int
 	}
 
-	// Skippable will be true if the frame is meant to be skipped.
-	// No other information will be populated.
-	Skippable bool
-
 	// If set there is a checksum present for the block content.
+	// The checksum field at the end is always 4 bytes long.
 	HasCheckSum bool
-
-	// If this is true FrameContentSize will have a valid value
-	HasFCS bool
-
-	SingleSegment bool
 }
 
 // Decode the header from the beginning of the stream.
@@ -71,39 +96,46 @@
 // If there isn't enough input, io.ErrUnexpectedEOF is returned.
 // The FirstBlock.OK will indicate if enough information was available to decode the first block header.
 func (h *Header) Decode(in []byte) error {
+	*h = Header{}
 	if len(in) < 4 {
 		return io.ErrUnexpectedEOF
 	}
+	h.HeaderSize += 4
 	b, in := in[:4], in[4:]
 	if !bytes.Equal(b, frameMagic) {
 		if !bytes.Equal(b[1:4], skippableFrameMagic) || b[0]&0xf0 != 0x50 {
 			return ErrMagicMismatch
 		}
-		*h = Header{Skippable: true}
+		if len(in) < 4 {
+			return io.ErrUnexpectedEOF
+		}
+		h.HeaderSize += 4
+		h.Skippable = true
+		h.SkippableID = int(b[0] & 0xf)
+		h.SkippableSize = binary.LittleEndian.Uint32(in)
 		return nil
 	}
-	if len(in) < 1 {
-		return io.ErrUnexpectedEOF
-	}
-
-	// Clear output
-	*h = Header{}
-	fhd, in := in[0], in[1:]
-	h.SingleSegment = fhd&(1<<5) != 0
-	h.HasCheckSum = fhd&(1<<2) != 0
-
-	if fhd&(1<<3) != 0 {
-		return errors.New("reserved bit set on frame header")
-	}
 
 	// Read Window_Descriptor
 	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#window_descriptor
+	if len(in) < 1 {
+		return io.ErrUnexpectedEOF
+	}
+	fhd, in := in[0], in[1:]
+	h.HeaderSize++
+	h.SingleSegment = fhd&(1<<5) != 0
+	h.HasCheckSum = fhd&(1<<2) != 0
+	if fhd&(1<<3) != 0 {
+		return errors.New("reserved bit set on frame header")
+	}
+
 	if !h.SingleSegment {
 		if len(in) < 1 {
 			return io.ErrUnexpectedEOF
 		}
 		var wd byte
 		wd, in = in[0], in[1:]
+		h.HeaderSize++
 		windowLog := 10 + (wd >> 3)
 		windowBase := uint64(1) << windowLog
 		windowAdd := (windowBase / 8) * uint64(wd&0x7)
@@ -120,9 +152,7 @@
 			return io.ErrUnexpectedEOF
 		}
 		b, in = in[:size], in[size:]
-		if b == nil {
-			return io.ErrUnexpectedEOF
-		}
+		h.HeaderSize += int(size)
 		switch size {
 		case 1:
 			h.DictionaryID = uint32(b[0])
@@ -152,9 +182,7 @@
 			return io.ErrUnexpectedEOF
 		}
 		b, in = in[:fcsSize], in[fcsSize:]
-		if b == nil {
-			return io.ErrUnexpectedEOF
-		}
+		h.HeaderSize += int(fcsSize)
 		switch fcsSize {
 		case 1:
 			h.FrameContentSize = uint64(b[0])
diff --git a/vendor/github.com/klauspost/compress/zstd/decoder.go b/vendor/github.com/klauspost/compress/zstd/decoder.go
index f593e46..d212f47 100644
--- a/vendor/github.com/klauspost/compress/zstd/decoder.go
+++ b/vendor/github.com/klauspost/compress/zstd/decoder.go
@@ -5,9 +5,13 @@
 package zstd
 
 import (
-	"errors"
+	"bytes"
+	"context"
+	"encoding/binary"
 	"io"
 	"sync"
+
+	"github.com/klauspost/compress/zstd/internal/xxhash"
 )
 
 // Decoder provides decoding of zstandard streams.
@@ -22,12 +26,19 @@
 	// Unreferenced decoders, ready for use.
 	decoders chan *blockDec
 
-	// Streams ready to be decoded.
-	stream chan decodeStream
-
 	// Current read position used for Reader functionality.
 	current decoderState
 
+	// sync stream decoding
+	syncStream struct {
+		decodedFrame uint64
+		br           readerWrapper
+		enabled      bool
+		inFrame      bool
+	}
+
+	frame *frameDec
+
 	// Custom dictionaries.
 	// Always uses copies.
 	dicts map[uint32]dict
@@ -46,7 +57,10 @@
 	output chan decodeOutput
 
 	// cancel remaining output.
-	cancel chan struct{}
+	cancel context.CancelFunc
+
+	// crc of current frame
+	crc *xxhash.Digest
 
 	flushed bool
 }
@@ -81,7 +95,7 @@
 			return nil, err
 		}
 	}
-	d.current.output = make(chan decodeOutput, d.o.concurrent)
+	d.current.crc = xxhash.New()
 	d.current.flushed = true
 
 	if r == nil {
@@ -113,9 +127,6 @@
 // Returns the number of bytes written and any error that occurred.
 // When the stream is done, io.EOF will be returned.
 func (d *Decoder) Read(p []byte) (int, error) {
-	if d.stream == nil {
-		return 0, ErrDecoderNilInput
-	}
 	var n int
 	for {
 		if len(d.current.b) > 0 {
@@ -133,12 +144,12 @@
 				break
 			}
 			if !d.nextBlock(n == 0) {
-				return n, nil
+				return n, d.current.err
 			}
 		}
 	}
 	if len(d.current.b) > 0 {
-		if debug {
+		if debugDecoder {
 			println("returning", n, "still bytes left:", len(d.current.b))
 		}
 		// Only return error at end of block
@@ -147,7 +158,7 @@
 	if d.current.err != nil {
 		d.drainOutput()
 	}
-	if debug {
+	if debugDecoder {
 		println("returning", n, d.current.err, len(d.decoders))
 	}
 	return n, d.current.err
@@ -165,22 +176,20 @@
 
 	d.drainOutput()
 
+	d.syncStream.br.r = nil
 	if r == nil {
 		d.current.err = ErrDecoderNilInput
+		if len(d.current.b) > 0 {
+			d.current.b = d.current.b[:0]
+		}
 		d.current.flushed = true
 		return nil
 	}
 
-	if d.stream == nil {
-		d.stream = make(chan decodeStream, 1)
-		d.streamWg.Add(1)
-		go d.startStreamDecoder(d.stream)
-	}
-
-	// If bytes buffer and < 1MB, do sync decoding anyway.
-	if bb, ok := r.(byter); ok && bb.Len() < 1<<20 {
+	// If bytes buffer and < 5MB, do sync decoding anyway.
+	if bb, ok := r.(byter); ok && bb.Len() < 5<<20 {
 		bb2 := bb
-		if debug {
+		if debugDecoder {
 			println("*bytes.Buffer detected, doing sync decode, len:", bb.Len())
 		}
 		b := bb2.Bytes()
@@ -196,36 +205,48 @@
 		d.current.b = dst
 		d.current.err = err
 		d.current.flushed = true
-		if debug {
+		if debugDecoder {
 			println("sync decode to", len(dst), "bytes, err:", err)
 		}
 		return nil
 	}
-
 	// Remove current block.
+	d.stashDecoder()
 	d.current.decodeOutput = decodeOutput{}
 	d.current.err = nil
-	d.current.cancel = make(chan struct{})
 	d.current.flushed = false
 	d.current.d = nil
 
-	d.stream <- decodeStream{
-		r:      r,
-		output: d.current.output,
-		cancel: d.current.cancel,
+	// Ensure no-one else is still running...
+	d.streamWg.Wait()
+	if d.frame == nil {
+		d.frame = newFrameDec(d.o)
 	}
+
+	if d.o.concurrent == 1 {
+		return d.startSyncDecoder(r)
+	}
+
+	d.current.output = make(chan decodeOutput, d.o.concurrent)
+	ctx, cancel := context.WithCancel(context.Background())
+	d.current.cancel = cancel
+	d.streamWg.Add(1)
+	go d.startStreamDecoder(ctx, r, d.current.output)
+
 	return nil
 }
 
 // drainOutput will drain the output until errEndOfStream is sent.
 func (d *Decoder) drainOutput() {
 	if d.current.cancel != nil {
-		println("cancelling current")
-		close(d.current.cancel)
+		if debugDecoder {
+			println("cancelling current")
+		}
+		d.current.cancel()
 		d.current.cancel = nil
 	}
 	if d.current.d != nil {
-		if debug {
+		if debugDecoder {
 			printf("re-adding current decoder %p, decoders: %d", d.current.d, len(d.decoders))
 		}
 		d.decoders <- d.current.d
@@ -238,34 +259,29 @@
 	}
 	for v := range d.current.output {
 		if v.d != nil {
-			if debug {
+			if debugDecoder {
 				printf("re-adding decoder %p", v.d)
 			}
 			d.decoders <- v.d
 		}
-		if v.err == errEndOfStream {
-			println("current flushed")
-			d.current.flushed = true
-			return
-		}
 	}
+	d.current.output = nil
+	d.current.flushed = true
 }
 
 // WriteTo writes data to w until there's no more data to write or when an error occurs.
 // The return value n is the number of bytes written.
 // Any error encountered during the write is also returned.
 func (d *Decoder) WriteTo(w io.Writer) (int64, error) {
-	if d.stream == nil {
-		return 0, ErrDecoderNilInput
-	}
 	var n int64
 	for {
 		if len(d.current.b) > 0 {
 			n2, err2 := w.Write(d.current.b)
 			n += int64(n2)
-			if err2 != nil && d.current.err == nil {
+			if err2 != nil && (d.current.err == nil || d.current.err == io.EOF) {
 				d.current.err = err2
-				break
+			} else if n2 != len(d.current.b) {
+				d.current.err = io.ErrShortWrite
 			}
 		}
 		if d.current.err != nil {
@@ -289,7 +305,7 @@
 // DecodeAll can be used concurrently.
 // The Decoder concurrency limits will be respected.
 func (d *Decoder) DecodeAll(input, dst []byte) ([]byte, error) {
-	if d.current.err == ErrDecoderClosed {
+	if d.decoders == nil {
 		return dst, ErrDecoderClosed
 	}
 
@@ -297,11 +313,14 @@
 	block := <-d.decoders
 	frame := block.localFrame
 	defer func() {
-		if debug {
+		if debugDecoder {
 			printf("re-adding decoder: %p", block)
 		}
 		frame.rawInput = nil
 		frame.bBuf = nil
+		if frame.history.decoders.br != nil {
+			frame.history.decoders.br.in = nil
+		}
 		d.decoders <- block
 	}()
 	frame.bBuf = input
@@ -309,33 +328,42 @@
 	for {
 		frame.history.reset()
 		err := frame.reset(&frame.bBuf)
-		if err == io.EOF {
-			if debug {
-				println("frame reset return EOF")
+		if err != nil {
+			if err == io.EOF {
+				if debugDecoder {
+					println("frame reset return EOF")
+				}
+				return dst, nil
 			}
-			return dst, nil
+			return dst, err
 		}
 		if frame.DictionaryID != nil {
 			dict, ok := d.dicts[*frame.DictionaryID]
 			if !ok {
 				return nil, ErrUnknownDictionary
 			}
+			if debugDecoder {
+				println("setting dict", frame.DictionaryID)
+			}
 			frame.history.setDict(&dict)
 		}
-		if err != nil {
-			return dst, err
+		if frame.WindowSize > d.o.maxWindowSize {
+			if debugDecoder {
+				println("window size exceeded:", frame.WindowSize, ">", d.o.maxWindowSize)
+			}
+			return dst, ErrWindowSizeExceeded
 		}
-		if frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) {
-			return dst, ErrDecoderSizeExceeded
-		}
-		if frame.FrameContentSize > 0 && frame.FrameContentSize < 1<<30 {
-			// Never preallocate moe than 1 GB up front.
+		if frame.FrameContentSize != fcsUnknown {
+			if frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) {
+				return dst, ErrDecoderSizeExceeded
+			}
 			if cap(dst)-len(dst) < int(frame.FrameContentSize) {
-				dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize))
+				dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize)+compressedBlockOverAlloc)
 				copy(dst2, dst)
 				dst = dst2
 			}
 		}
+
 		if cap(dst) == 0 {
 			// Allocate len(input) * 2 by default if nothing is provided
 			// and we didn't get frame content size.
@@ -355,7 +383,7 @@
 			return dst, err
 		}
 		if len(frame.bBuf) == 0 {
-			if debug {
+			if debugDecoder {
 				println("frame dbuf empty")
 			}
 			break
@@ -370,31 +398,174 @@
 // If non-blocking mode is used the returned boolean will be false
 // if no data was available without blocking.
 func (d *Decoder) nextBlock(blocking bool) (ok bool) {
+	if d.current.err != nil {
+		// Keep error state.
+		return false
+	}
+	d.current.b = d.current.b[:0]
+
+	// SYNC:
+	if d.syncStream.enabled {
+		if !blocking {
+			return false
+		}
+		ok = d.nextBlockSync()
+		if !ok {
+			d.stashDecoder()
+		}
+		return ok
+	}
+
+	//ASYNC:
+	d.stashDecoder()
+	if blocking {
+		d.current.decodeOutput, ok = <-d.current.output
+	} else {
+		select {
+		case d.current.decodeOutput, ok = <-d.current.output:
+		default:
+			return false
+		}
+	}
+	if !ok {
+		// This should not happen, so signal error state...
+		d.current.err = io.ErrUnexpectedEOF
+		return false
+	}
+	next := d.current.decodeOutput
+	if next.d != nil && next.d.async.newHist != nil {
+		d.current.crc.Reset()
+	}
+	if debugDecoder {
+		var tmp [4]byte
+		binary.LittleEndian.PutUint32(tmp[:], uint32(xxhash.Sum64(next.b)))
+		println("got", len(d.current.b), "bytes, error:", d.current.err, "data crc:", tmp)
+	}
+
+	if !d.o.ignoreChecksum && len(next.b) > 0 {
+		n, err := d.current.crc.Write(next.b)
+		if err == nil {
+			if n != len(next.b) {
+				d.current.err = io.ErrShortWrite
+			}
+		}
+	}
+	if next.err == nil && next.d != nil && len(next.d.checkCRC) != 0 {
+		got := d.current.crc.Sum64()
+		var tmp [4]byte
+		binary.LittleEndian.PutUint32(tmp[:], uint32(got))
+		if !d.o.ignoreChecksum && !bytes.Equal(tmp[:], next.d.checkCRC) {
+			if debugDecoder {
+				println("CRC Check Failed:", tmp[:], " (got) !=", next.d.checkCRC, "(on stream)")
+			}
+			d.current.err = ErrCRCMismatch
+		} else {
+			if debugDecoder {
+				println("CRC ok", tmp[:])
+			}
+		}
+	}
+
+	return true
+}
+
+func (d *Decoder) nextBlockSync() (ok bool) {
+	if d.current.d == nil {
+		d.current.d = <-d.decoders
+	}
+	for len(d.current.b) == 0 {
+		if !d.syncStream.inFrame {
+			d.frame.history.reset()
+			d.current.err = d.frame.reset(&d.syncStream.br)
+			if d.current.err != nil {
+				return false
+			}
+			if d.frame.DictionaryID != nil {
+				dict, ok := d.dicts[*d.frame.DictionaryID]
+				if !ok {
+					d.current.err = ErrUnknownDictionary
+					return false
+				} else {
+					d.frame.history.setDict(&dict)
+				}
+			}
+			if d.frame.WindowSize > d.o.maxDecodedSize || d.frame.WindowSize > d.o.maxWindowSize {
+				d.current.err = ErrDecoderSizeExceeded
+				return false
+			}
+
+			d.syncStream.decodedFrame = 0
+			d.syncStream.inFrame = true
+		}
+		d.current.err = d.frame.next(d.current.d)
+		if d.current.err != nil {
+			return false
+		}
+		d.frame.history.ensureBlock()
+		if debugDecoder {
+			println("History trimmed:", len(d.frame.history.b), "decoded already:", d.syncStream.decodedFrame)
+		}
+		histBefore := len(d.frame.history.b)
+		d.current.err = d.current.d.decodeBuf(&d.frame.history)
+
+		if d.current.err != nil {
+			println("error after:", d.current.err)
+			return false
+		}
+		d.current.b = d.frame.history.b[histBefore:]
+		if debugDecoder {
+			println("history after:", len(d.frame.history.b))
+		}
+
+		// Check frame size (before CRC)
+		d.syncStream.decodedFrame += uint64(len(d.current.b))
+		if d.syncStream.decodedFrame > d.frame.FrameContentSize {
+			if debugDecoder {
+				printf("DecodedFrame (%d) > FrameContentSize (%d)\n", d.syncStream.decodedFrame, d.frame.FrameContentSize)
+			}
+			d.current.err = ErrFrameSizeExceeded
+			return false
+		}
+
+		// Check FCS
+		if d.current.d.Last && d.frame.FrameContentSize != fcsUnknown && d.syncStream.decodedFrame != d.frame.FrameContentSize {
+			if debugDecoder {
+				printf("DecodedFrame (%d) != FrameContentSize (%d)\n", d.syncStream.decodedFrame, d.frame.FrameContentSize)
+			}
+			d.current.err = ErrFrameSizeMismatch
+			return false
+		}
+
+		// Update/Check CRC
+		if d.frame.HasCheckSum {
+			if !d.o.ignoreChecksum {
+				d.frame.crc.Write(d.current.b)
+			}
+			if d.current.d.Last {
+				if !d.o.ignoreChecksum {
+					d.current.err = d.frame.checkCRC()
+				} else {
+					d.current.err = d.frame.consumeCRC()
+				}
+				if d.current.err != nil {
+					println("CRC error:", d.current.err)
+					return false
+				}
+			}
+		}
+		d.syncStream.inFrame = !d.current.d.Last
+	}
+	return true
+}
+
+func (d *Decoder) stashDecoder() {
 	if d.current.d != nil {
-		if debug {
+		if debugDecoder {
 			printf("re-adding current decoder %p", d.current.d)
 		}
 		d.decoders <- d.current.d
 		d.current.d = nil
 	}
-	if d.current.err != nil {
-		// Keep error state.
-		return blocking
-	}
-
-	if blocking {
-		d.current.decodeOutput = <-d.current.output
-	} else {
-		select {
-		case d.current.decodeOutput = <-d.current.output:
-		default:
-			return false
-		}
-	}
-	if debug {
-		println("got", len(d.current.b), "bytes, error:", d.current.err)
-	}
-	return true
 }
 
 // Close will release all resources.
@@ -404,10 +575,10 @@
 		return
 	}
 	d.drainOutput()
-	if d.stream != nil {
-		close(d.stream)
+	if d.current.cancel != nil {
+		d.current.cancel()
 		d.streamWg.Wait()
-		d.stream = nil
+		d.current.cancel = nil
 	}
 	if d.decoders != nil {
 		close(d.decoders)
@@ -458,100 +629,296 @@
 	err error
 }
 
-type decodeStream struct {
-	r io.Reader
-
-	// Blocks ready to be written to output.
-	output chan decodeOutput
-
-	// cancel reading from the input
-	cancel chan struct{}
+func (d *Decoder) startSyncDecoder(r io.Reader) error {
+	d.frame.history.reset()
+	d.syncStream.br = readerWrapper{r: r}
+	d.syncStream.inFrame = false
+	d.syncStream.enabled = true
+	d.syncStream.decodedFrame = 0
+	return nil
 }
 
-// errEndOfStream indicates that everything from the stream was read.
-var errEndOfStream = errors.New("end-of-stream")
-
 // Create Decoder:
-// Spawn n block decoders. These accept tasks to decode a block.
-// Create goroutine that handles stream processing, this will send history to decoders as they are available.
-// Decoders update the history as they decode.
-// When a block is returned:
-// 		a) history is sent to the next decoder,
-// 		b) content written to CRC.
-// 		c) return data to WRITER.
-// 		d) wait for next block to return data.
-// Once WRITTEN, the decoders reused by the writer frame decoder for re-use.
-func (d *Decoder) startStreamDecoder(inStream chan decodeStream) {
+// ASYNC:
+// Spawn 3 go routines.
+// 0: Read frames and decode block literals.
+// 1: Decode sequences.
+// 2: Execute sequences, send to output.
+func (d *Decoder) startStreamDecoder(ctx context.Context, r io.Reader, output chan decodeOutput) {
 	defer d.streamWg.Done()
-	frame := newFrameDec(d.o)
-	for stream := range inStream {
-		if debug {
-			println("got new stream")
+	br := readerWrapper{r: r}
+
+	var seqDecode = make(chan *blockDec, d.o.concurrent)
+	var seqExecute = make(chan *blockDec, d.o.concurrent)
+
+	// Async 1: Decode sequences...
+	go func() {
+		var hist history
+		var hasErr bool
+
+		for block := range seqDecode {
+			if hasErr {
+				if block != nil {
+					seqExecute <- block
+				}
+				continue
+			}
+			if block.async.newHist != nil {
+				if debugDecoder {
+					println("Async 1: new history, recent:", block.async.newHist.recentOffsets)
+				}
+				hist.decoders = block.async.newHist.decoders
+				hist.recentOffsets = block.async.newHist.recentOffsets
+				hist.windowSize = block.async.newHist.windowSize
+				if block.async.newHist.dict != nil {
+					hist.setDict(block.async.newHist.dict)
+				}
+			}
+			if block.err != nil || block.Type != blockTypeCompressed {
+				hasErr = block.err != nil
+				seqExecute <- block
+				continue
+			}
+
+			hist.decoders.literals = block.async.literals
+			block.err = block.prepareSequences(block.async.seqData, &hist)
+			if debugDecoder && block.err != nil {
+				println("prepareSequences returned:", block.err)
+			}
+			hasErr = block.err != nil
+			if block.err == nil {
+				block.err = block.decodeSequences(&hist)
+				if debugDecoder && block.err != nil {
+					println("decodeSequences returned:", block.err)
+				}
+				hasErr = block.err != nil
+				//				block.async.sequence = hist.decoders.seq[:hist.decoders.nSeqs]
+				block.async.seqSize = hist.decoders.seqSize
+			}
+			seqExecute <- block
 		}
-		br := readerWrapper{r: stream.r}
-	decodeStream:
-		for {
-			frame.history.reset()
-			err := frame.reset(&br)
-			if debug && err != nil {
-				println("Frame decoder returned", err)
+		close(seqExecute)
+	}()
+
+	var wg sync.WaitGroup
+	wg.Add(1)
+
+	// Async 3: Execute sequences...
+	frameHistCache := d.frame.history.b
+	go func() {
+		var hist history
+		var decodedFrame uint64
+		var fcs uint64
+		var hasErr bool
+		for block := range seqExecute {
+			out := decodeOutput{err: block.err, d: block}
+			if block.err != nil || hasErr {
+				hasErr = true
+				output <- out
+				continue
 			}
-			if err == nil && frame.DictionaryID != nil {
-				dict, ok := d.dicts[*frame.DictionaryID]
-				if !ok {
-					err = ErrUnknownDictionary
+			if block.async.newHist != nil {
+				if debugDecoder {
+					println("Async 2: new history")
+				}
+				hist.windowSize = block.async.newHist.windowSize
+				hist.allocFrameBuffer = block.async.newHist.allocFrameBuffer
+				if block.async.newHist.dict != nil {
+					hist.setDict(block.async.newHist.dict)
+				}
+
+				if cap(hist.b) < hist.allocFrameBuffer {
+					if cap(frameHistCache) >= hist.allocFrameBuffer {
+						hist.b = frameHistCache
+					} else {
+						hist.b = make([]byte, 0, hist.allocFrameBuffer)
+						println("Alloc history sized", hist.allocFrameBuffer)
+					}
+				}
+				hist.b = hist.b[:0]
+				fcs = block.async.fcs
+				decodedFrame = 0
+			}
+			do := decodeOutput{err: block.err, d: block}
+			switch block.Type {
+			case blockTypeRLE:
+				if debugDecoder {
+					println("add rle block length:", block.RLESize)
+				}
+
+				if cap(block.dst) < int(block.RLESize) {
+					if block.lowMem {
+						block.dst = make([]byte, block.RLESize)
+					} else {
+						block.dst = make([]byte, maxBlockSize)
+					}
+				}
+				block.dst = block.dst[:block.RLESize]
+				v := block.data[0]
+				for i := range block.dst {
+					block.dst[i] = v
+				}
+				hist.append(block.dst)
+				do.b = block.dst
+			case blockTypeRaw:
+				if debugDecoder {
+					println("add raw block length:", len(block.data))
+				}
+				hist.append(block.data)
+				do.b = block.data
+			case blockTypeCompressed:
+				if debugDecoder {
+					println("execute with history length:", len(hist.b), "window:", hist.windowSize)
+				}
+				hist.decoders.seqSize = block.async.seqSize
+				hist.decoders.literals = block.async.literals
+				do.err = block.executeSequences(&hist)
+				hasErr = do.err != nil
+				if debugDecoder && hasErr {
+					println("executeSequences returned:", do.err)
+				}
+				do.b = block.dst
+			}
+			if !hasErr {
+				decodedFrame += uint64(len(do.b))
+				if decodedFrame > fcs {
+					println("fcs exceeded", block.Last, fcs, decodedFrame)
+					do.err = ErrFrameSizeExceeded
+					hasErr = true
+				} else if block.Last && fcs != fcsUnknown && decodedFrame != fcs {
+					do.err = ErrFrameSizeMismatch
+					hasErr = true
 				} else {
-					frame.history.setDict(&dict)
+					if debugDecoder {
+						println("fcs ok", block.Last, fcs, decodedFrame)
+					}
 				}
 			}
-			if err != nil {
-				stream.output <- decodeOutput{
-					err: err,
+			output <- do
+		}
+		close(output)
+		frameHistCache = hist.b
+		wg.Done()
+		if debugDecoder {
+			println("decoder goroutines finished")
+		}
+	}()
+
+decodeStream:
+	for {
+		var hist history
+		var hasErr bool
+
+		decodeBlock := func(block *blockDec) {
+			if hasErr {
+				if block != nil {
+					seqDecode <- block
 				}
+				return
+			}
+			if block.err != nil || block.Type != blockTypeCompressed {
+				hasErr = block.err != nil
+				seqDecode <- block
+				return
+			}
+
+			remain, err := block.decodeLiterals(block.data, &hist)
+			block.err = err
+			hasErr = block.err != nil
+			if err == nil {
+				block.async.literals = hist.decoders.literals
+				block.async.seqData = remain
+			} else if debugDecoder {
+				println("decodeLiterals error:", err)
+			}
+			seqDecode <- block
+		}
+		frame := d.frame
+		if debugDecoder {
+			println("New frame...")
+		}
+		var historySent bool
+		frame.history.reset()
+		err := frame.reset(&br)
+		if debugDecoder && err != nil {
+			println("Frame decoder returned", err)
+		}
+		if err == nil && frame.DictionaryID != nil {
+			dict, ok := d.dicts[*frame.DictionaryID]
+			if !ok {
+				err = ErrUnknownDictionary
+			} else {
+				frame.history.setDict(&dict)
+			}
+		}
+		if err == nil && d.frame.WindowSize > d.o.maxWindowSize {
+			err = ErrDecoderSizeExceeded
+		}
+		if err != nil {
+			select {
+			case <-ctx.Done():
+			case dec := <-d.decoders:
+				dec.sendErr(err)
+				decodeBlock(dec)
+			}
+			break decodeStream
+		}
+
+		// Go through all blocks of the frame.
+		for {
+			var dec *blockDec
+			select {
+			case <-ctx.Done():
+				break decodeStream
+			case dec = <-d.decoders:
+				// Once we have a decoder, we MUST return it.
+			}
+			err := frame.next(dec)
+			if !historySent {
+				h := frame.history
+				if debugDecoder {
+					println("Alloc History:", h.allocFrameBuffer)
+				}
+				hist.reset()
+				if h.dict != nil {
+					hist.setDict(h.dict)
+				}
+				dec.async.newHist = &h
+				dec.async.fcs = frame.FrameContentSize
+				historySent = true
+			} else {
+				dec.async.newHist = nil
+			}
+			if debugDecoder && err != nil {
+				println("next block returned error:", err)
+			}
+			dec.err = err
+			dec.checkCRC = nil
+			if dec.Last && frame.HasCheckSum && err == nil {
+				crc, err := frame.rawInput.readSmall(4)
+				if err != nil {
+					println("CRC missing?", err)
+					dec.err = err
+				}
+				var tmp [4]byte
+				copy(tmp[:], crc)
+				dec.checkCRC = tmp[:]
+				if debugDecoder {
+					println("found crc to check:", dec.checkCRC)
+				}
+			}
+			err = dec.err
+			last := dec.Last
+			decodeBlock(dec)
+			if err != nil {
+				break decodeStream
+			}
+			if last {
 				break
 			}
-			if debug {
-				println("starting frame decoder")
-			}
-
-			// This goroutine will forward history between frames.
-			frame.frameDone.Add(1)
-			frame.initAsync()
-
-			go frame.startDecoder(stream.output)
-		decodeFrame:
-			// Go through all blocks of the frame.
-			for {
-				dec := <-d.decoders
-				select {
-				case <-stream.cancel:
-					if !frame.sendErr(dec, io.EOF) {
-						// To not let the decoder dangle, send it back.
-						stream.output <- decodeOutput{d: dec}
-					}
-					break decodeStream
-				default:
-				}
-				err := frame.next(dec)
-				switch err {
-				case io.EOF:
-					// End of current frame, no error
-					println("EOF on next block")
-					break decodeFrame
-				case nil:
-					continue
-				default:
-					println("block decoder returned", err)
-					break decodeStream
-				}
-			}
-			// All blocks have started decoding, check if there are more frames.
-			println("waiting for done")
-			frame.frameDone.Wait()
-			println("done waiting...")
 		}
-		frame.frameDone.Wait()
-		println("Sending EOS")
-		stream.output <- decodeOutput{err: errEndOfStream}
 	}
+	close(seqDecode)
+	wg.Wait()
+	d.frame.history.b = frameHistCache
 }
diff --git a/vendor/github.com/klauspost/compress/zstd/decoder_options.go b/vendor/github.com/klauspost/compress/zstd/decoder_options.go
index c0fd058..c70e6fa 100644
--- a/vendor/github.com/klauspost/compress/zstd/decoder_options.go
+++ b/vendor/github.com/klauspost/compress/zstd/decoder_options.go
@@ -17,16 +17,22 @@
 	lowMem         bool
 	concurrent     int
 	maxDecodedSize uint64
+	maxWindowSize  uint64
 	dicts          []dict
+	ignoreChecksum bool
 }
 
 func (o *decoderOptions) setDefault() {
 	*o = decoderOptions{
 		// use less ram: true for now, but may change.
-		lowMem:     true,
-		concurrent: runtime.GOMAXPROCS(0),
+		lowMem:        true,
+		concurrent:    runtime.GOMAXPROCS(0),
+		maxWindowSize: MaxWindowSize,
 	}
-	o.maxDecodedSize = 1 << 63
+	if o.concurrent > 4 {
+		o.concurrent = 4
+	}
+	o.maxDecodedSize = 64 << 30
 }
 
 // WithDecoderLowmem will set whether to use a lower amount of memory,
@@ -35,16 +41,25 @@
 	return func(o *decoderOptions) error { o.lowMem = b; return nil }
 }
 
-// WithDecoderConcurrency will set the concurrency,
-// meaning the maximum number of decoders to run concurrently.
-// The value supplied must be at least 1.
-// By default this will be set to GOMAXPROCS.
+// WithDecoderConcurrency sets the number of created decoders.
+// When decoding block with DecodeAll, this will limit the number
+// of possible concurrently running decodes.
+// When decoding streams, this will limit the number of
+// inflight blocks.
+// When decoding streams and setting maximum to 1,
+// no async decoding will be done.
+// When a value of 0 is provided GOMAXPROCS will be used.
+// By default this will be set to 4 or GOMAXPROCS, whatever is lower.
 func WithDecoderConcurrency(n int) DOption {
 	return func(o *decoderOptions) error {
-		if n <= 0 {
+		if n < 0 {
 			return errors.New("concurrency must be at least 1")
 		}
-		o.concurrent = n
+		if n == 0 {
+			o.concurrent = runtime.GOMAXPROCS(0)
+		} else {
+			o.concurrent = n
+		}
 		return nil
 	}
 }
@@ -52,8 +67,7 @@
 // WithDecoderMaxMemory allows to set a maximum decoded size for in-memory
 // non-streaming operations or maximum window size for streaming operations.
 // This can be used to control memory usage of potentially hostile content.
-// For streaming operations, the maximum window size is capped at 1<<30 bytes.
-// Maximum and default is 1 << 63 bytes.
+// Maximum is 1 << 63 bytes. Default is 64GiB.
 func WithDecoderMaxMemory(n uint64) DOption {
 	return func(o *decoderOptions) error {
 		if n == 0 {
@@ -81,3 +95,29 @@
 		return nil
 	}
 }
+
+// WithDecoderMaxWindow allows to set a maximum window size for decodes.
+// This allows rejecting packets that will cause big memory usage.
+// The Decoder will likely allocate more memory based on the WithDecoderLowmem setting.
+// If WithDecoderMaxMemory is set to a lower value, that will be used.
+// Default is 512MB, Maximum is ~3.75 TB as per zstandard spec.
+func WithDecoderMaxWindow(size uint64) DOption {
+	return func(o *decoderOptions) error {
+		if size < MinWindowSize {
+			return errors.New("WithMaxWindowSize must be at least 1KB, 1024 bytes")
+		}
+		if size > (1<<41)+7*(1<<38) {
+			return errors.New("WithMaxWindowSize must be less than (1<<41) + 7*(1<<38) ~ 3.75TB")
+		}
+		o.maxWindowSize = size
+		return nil
+	}
+}
+
+// IgnoreChecksum allows to forcibly ignore checksum checking.
+func IgnoreChecksum(b bool) DOption {
+	return func(o *decoderOptions) error {
+		o.ignoreChecksum = b
+		return nil
+	}
+}
diff --git a/vendor/github.com/klauspost/compress/zstd/dict.go b/vendor/github.com/klauspost/compress/zstd/dict.go
index fa25a18..a36ae83 100644
--- a/vendor/github.com/klauspost/compress/zstd/dict.go
+++ b/vendor/github.com/klauspost/compress/zstd/dict.go
@@ -82,7 +82,7 @@
 			println("Transform table error:", err)
 			return err
 		}
-		if debug {
+		if debugDecoder || debugEncoder {
 			println("Read table ok", "symbolLen:", dec.symbolLen)
 		}
 		// Set decoders as predefined so they aren't reused.
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_base.go b/vendor/github.com/klauspost/compress/zstd/enc_base.go
index 60f2986..15ae8ee 100644
--- a/vendor/github.com/klauspost/compress/zstd/enc_base.go
+++ b/vendor/github.com/klauspost/compress/zstd/enc_base.go
@@ -38,8 +38,8 @@
 
 // WindowSize returns the window size of the encoder,
 // or a window size small enough to contain the input size, if > 0.
-func (e *fastBase) WindowSize(size int) int32 {
-	if size > 0 && size < int(e.maxMatchOff) {
+func (e *fastBase) WindowSize(size int64) int32 {
+	if size > 0 && size < int64(e.maxMatchOff) {
 		b := int32(1) << uint(bits.Len(uint(size)))
 		// Keep minimum window.
 		if b < 1024 {
@@ -108,11 +108,6 @@
 	e.blk = enc
 }
 
-func (e *fastBase) matchlenNoHist(s, t int32, src []byte) int32 {
-	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:], src[t:]))
-}
-
 func (e *fastBase) matchlen(s, t int32, src []byte) int32 {
 	if debugAsserts {
 		if s < 0 {
@@ -131,9 +126,24 @@
 			panic(fmt.Sprintf("len(src)-s (%d) > maxCompressedBlockSize (%d)", len(src)-int(s), maxCompressedBlockSize))
 		}
 	}
+	a := src[s:]
+	b := src[t:]
+	b = b[:len(a)]
+	end := int32((len(a) >> 3) << 3)
+	for i := int32(0); i < end; i += 8 {
+		if diff := load6432(a, i) ^ load6432(b, i); diff != 0 {
+			return i + int32(bits.TrailingZeros64(diff)>>3)
+		}
+	}
 
-	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:], src[t:]))
+	a = a[end:]
+	b = b[end:]
+	for i := range a {
+		if a[i] != b[i] {
+			return int32(i) + end
+		}
+	}
+	return int32(len(a)) + end
 }
 
 // Reset the encoding table.
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_best.go b/vendor/github.com/klauspost/compress/zstd/enc_best.go
index dc1eed5..96028ec 100644
--- a/vendor/github.com/klauspost/compress/zstd/enc_best.go
+++ b/vendor/github.com/klauspost/compress/zstd/enc_best.go
@@ -5,22 +5,61 @@
 package zstd
 
 import (
+	"bytes"
 	"fmt"
-	"math/bits"
+
+	"github.com/klauspost/compress"
 )
 
 const (
-	bestLongTableBits = 20                     // Bits used in the long match table
+	bestLongTableBits = 22                     // Bits used in the long match table
 	bestLongTableSize = 1 << bestLongTableBits // Size of the table
+	bestLongLen       = 8                      // Bytes used for table hash
 
 	// Note: Increasing the short table bits or making the hash shorter
 	// can actually lead to compression degradation since it will 'steal' more from the
 	// long match table and match offsets are quite big.
 	// This greatly depends on the type of input.
-	bestShortTableBits = 16                      // Bits used in the short match table
+	bestShortTableBits = 18                      // Bits used in the short match table
 	bestShortTableSize = 1 << bestShortTableBits // Size of the table
+	bestShortLen       = 4                       // Bytes used for table hash
+
 )
 
+type match struct {
+	offset int32
+	s      int32
+	length int32
+	rep    int32
+	est    int32
+}
+
+const highScore = 25000
+
+// estBits will estimate output bits from predefined tables.
+func (m *match) estBits(bitsPerByte int32) {
+	mlc := mlCode(uint32(m.length - zstdMinMatch))
+	var ofc uint8
+	if m.rep < 0 {
+		ofc = ofCode(uint32(m.s-m.offset) + 3)
+	} else {
+		ofc = ofCode(uint32(m.rep))
+	}
+	// Cost, excluding
+	ofTT, mlTT := fsePredefEnc[tableOffsets].ct.symbolTT[ofc], fsePredefEnc[tableMatchLengths].ct.symbolTT[mlc]
+
+	// Add cost of match encoding...
+	m.est = int32(ofTT.outBits + mlTT.outBits)
+	m.est += int32(ofTT.deltaNbBits>>16 + mlTT.deltaNbBits>>16)
+	// Subtract savings compared to literal encoding...
+	m.est -= (m.length * bitsPerByte) >> 10
+	if m.est > 0 {
+		// Unlikely gain..
+		m.length = 0
+		m.est = highScore
+	}
+}
+
 // bestFastEncoder uses 2 tables, one for short matches (5 bytes) and one for long matches.
 // The long match table contains the previous entry with the same hash,
 // effectively making it a "chain" of length 2.
@@ -109,6 +148,14 @@
 		return
 	}
 
+	// Use this to estimate literal cost.
+	// Scaled by 10 bits.
+	bitsPerByte := int32((compress.ShannonEntropyBits(src) * 1024) / len(src))
+	// Huffman can never go < 1 bit/byte
+	if bitsPerByte < 1024 {
+		bitsPerByte = 1024
+	}
+
 	// Override src
 	src = e.hist
 	sLimit := int32(len(src)) - inputMargin
@@ -132,7 +179,7 @@
 	}
 	_ = addLiterals
 
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -145,51 +192,49 @@
 			panic("offset0 was 0")
 		}
 
-		type match struct {
-			offset int32
-			s      int32
-			length int32
-			rep    int32
-		}
-		matchAt := func(offset int32, s int32, first uint32, rep int32) match {
-			if s-offset >= e.maxMatchOff || load3232(src, offset) != first {
-				return match{offset: offset, s: s}
-			}
-			return match{offset: offset, s: s, length: 4 + e.matchlen(s+4, offset+4, src), rep: rep}
-		}
-
 		bestOf := func(a, b match) match {
-			aScore := b.s - a.s + a.length
-			bScore := a.s - b.s + b.length
-			if a.rep < 0 {
-				aScore = aScore - int32(bits.Len32(uint32(a.offset)))/8
-			}
-			if b.rep < 0 {
-				bScore = bScore - int32(bits.Len32(uint32(b.offset)))/8
-			}
-			if aScore >= bScore {
+			if a.est+(a.s-b.s)*bitsPerByte>>10 < b.est+(b.s-a.s)*bitsPerByte>>10 {
 				return a
 			}
 			return b
 		}
 		const goodEnough = 100
 
-		nextHashL := hash8(cv, bestLongTableBits)
-		nextHashS := hash4x64(cv, bestShortTableBits)
+		nextHashL := hashLen(cv, bestLongTableBits, bestLongLen)
+		nextHashS := hashLen(cv, bestShortTableBits, bestShortLen)
 		candidateL := e.longTable[nextHashL]
 		candidateS := e.table[nextHashS]
 
+		matchAt := func(offset int32, s int32, first uint32, rep int32) match {
+			if s-offset >= e.maxMatchOff || load3232(src, offset) != first {
+				return match{s: s, est: highScore}
+			}
+			if debugAsserts {
+				if !bytes.Equal(src[s:s+4], src[offset:offset+4]) {
+					panic(fmt.Sprintf("first match mismatch: %v != %v, first: %08x", src[s:s+4], src[offset:offset+4], first))
+				}
+			}
+			m := match{offset: offset, s: s, length: 4 + e.matchlen(s+4, offset+4, src), rep: rep}
+			m.estBits(bitsPerByte)
+			return m
+		}
+
 		best := bestOf(matchAt(candidateL.offset-e.cur, s, uint32(cv), -1), matchAt(candidateL.prev-e.cur, s, uint32(cv), -1))
 		best = bestOf(best, matchAt(candidateS.offset-e.cur, s, uint32(cv), -1))
 		best = bestOf(best, matchAt(candidateS.prev-e.cur, s, uint32(cv), -1))
+
 		if canRepeat && best.length < goodEnough {
-			best = bestOf(best, matchAt(s-offset1+1, s+1, uint32(cv>>8), 1))
-			best = bestOf(best, matchAt(s-offset2+1, s+1, uint32(cv>>8), 2))
-			best = bestOf(best, matchAt(s-offset3+1, s+1, uint32(cv>>8), 3))
+			cv32 := uint32(cv >> 8)
+			spp := s + 1
+			best = bestOf(best, matchAt(spp-offset1, spp, cv32, 1))
+			best = bestOf(best, matchAt(spp-offset2, spp, cv32, 2))
+			best = bestOf(best, matchAt(spp-offset3, spp, cv32, 3))
 			if best.length > 0 {
-				best = bestOf(best, matchAt(s-offset1+3, s+3, uint32(cv>>24), 1))
-				best = bestOf(best, matchAt(s-offset2+3, s+3, uint32(cv>>24), 2))
-				best = bestOf(best, matchAt(s-offset3+3, s+3, uint32(cv>>24), 3))
+				cv32 = uint32(cv >> 24)
+				spp += 2
+				best = bestOf(best, matchAt(spp-offset1, spp, cv32, 1))
+				best = bestOf(best, matchAt(spp-offset2, spp, cv32, 2))
+				best = bestOf(best, matchAt(spp-offset3, spp, cv32, 3))
 			}
 		}
 		// Load next and check...
@@ -209,22 +254,28 @@
 			}
 
 			s++
-			candidateS = e.table[hash4x64(cv>>8, bestShortTableBits)]
+			candidateS = e.table[hashLen(cv>>8, bestShortTableBits, bestShortLen)]
 			cv = load6432(src, s)
 			cv2 := load6432(src, s+1)
-			candidateL = e.longTable[hash8(cv, bestLongTableBits)]
-			candidateL2 := e.longTable[hash8(cv2, bestLongTableBits)]
+			candidateL = e.longTable[hashLen(cv, bestLongTableBits, bestLongLen)]
+			candidateL2 := e.longTable[hashLen(cv2, bestLongTableBits, bestLongLen)]
 
+			// Short at s+1
 			best = bestOf(best, matchAt(candidateS.offset-e.cur, s, uint32(cv), -1))
+			// Long at s+1, s+2
 			best = bestOf(best, matchAt(candidateL.offset-e.cur, s, uint32(cv), -1))
 			best = bestOf(best, matchAt(candidateL.prev-e.cur, s, uint32(cv), -1))
 			best = bestOf(best, matchAt(candidateL2.offset-e.cur, s+1, uint32(cv2), -1))
 			best = bestOf(best, matchAt(candidateL2.prev-e.cur, s+1, uint32(cv2), -1))
-
+			if false {
+				// Short at s+3.
+				// Too often worse...
+				best = bestOf(best, matchAt(e.table[hashLen(cv2>>8, bestShortTableBits, bestShortLen)].offset-e.cur, s+2, uint32(cv2>>8), -1))
+			}
 			// See if we can find a better match by checking where the current best ends.
 			// Use that offset to see if we can find a better full match.
 			if sAt := best.s + best.length; sAt < sLimit {
-				nextHashL := hash8(load6432(src, sAt), bestLongTableBits)
+				nextHashL := hashLen(load6432(src, sAt), bestLongTableBits, bestLongLen)
 				candidateEnd := e.longTable[nextHashL]
 				if pos := candidateEnd.offset - e.cur - best.length; pos >= 0 {
 					bestEnd := bestOf(best, matchAt(pos, best.s, load3232(src, best.s), -1))
@@ -236,6 +287,12 @@
 			}
 		}
 
+		if debugAsserts {
+			if !bytes.Equal(src[best.s:best.s+best.length], src[best.offset:best.offset+best.length]) {
+				panic(fmt.Sprintf("match mismatch: %v != %v", src[best.s:best.s+best.length], src[best.offset:best.offset+best.length]))
+			}
+		}
+
 		// We have a match, we can store the forward value
 		if best.rep > 0 {
 			s = best.s
@@ -274,7 +331,7 @@
 
 			nextEmit = s
 			if s >= sLimit {
-				if debug {
+				if debugEncoder {
 					println("repeat ended", s, best.length)
 
 				}
@@ -284,8 +341,8 @@
 			off := index0 + e.cur
 			for index0 < s-1 {
 				cv0 := load6432(src, index0)
-				h0 := hash8(cv0, bestLongTableBits)
-				h1 := hash4x64(cv0, bestShortTableBits)
+				h0 := hashLen(cv0, bestLongTableBits, bestLongLen)
+				h1 := hashLen(cv0, bestShortTableBits, bestShortLen)
 				e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
 				e.table[h1] = prevEntry{offset: off, prev: e.table[h1].offset}
 				off++
@@ -311,7 +368,7 @@
 			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
 		}
 
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
+		if debugAsserts && int(offset1) > len(src) {
 			panic("invalid offset")
 		}
 
@@ -352,8 +409,8 @@
 		// every entry
 		for index0 < s-1 {
 			cv0 := load6432(src, index0)
-			h0 := hash8(cv0, bestLongTableBits)
-			h1 := hash4x64(cv0, bestShortTableBits)
+			h0 := hashLen(cv0, bestLongTableBits, bestLongLen)
+			h1 := hashLen(cv0, bestShortTableBits, bestShortLen)
 			off := index0 + e.cur
 			e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
 			e.table[h1] = prevEntry{offset: off, prev: e.table[h1].offset}
@@ -374,8 +431,8 @@
 			}
 
 			// Store this, since we have it.
-			nextHashS := hash4x64(cv, bestShortTableBits)
-			nextHashL := hash8(cv, bestLongTableBits)
+			nextHashS := hashLen(cv, bestShortTableBits, bestShortLen)
+			nextHashL := hashLen(cv, bestLongTableBits, bestLongLen)
 
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
@@ -412,7 +469,7 @@
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
 	blk.recentOffsets[2] = uint32(offset3)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 }
@@ -425,7 +482,7 @@
 	e.Encode(blk, src)
 }
 
-// ResetDict will reset and set a dictionary if not nil
+// Reset will reset and set a dictionary if not nil
 func (e *bestFastEncoder) Reset(d *dict, singleBlock bool) {
 	e.resetBase(d, singleBlock)
 	if d == nil {
@@ -441,10 +498,10 @@
 			const hashLog = bestShortTableBits
 
 			cv := load6432(d.content, i-e.maxMatchOff)
-			nextHash := hash4x64(cv, hashLog)      // 0 -> 4
-			nextHash1 := hash4x64(cv>>8, hashLog)  // 1 -> 5
-			nextHash2 := hash4x64(cv>>16, hashLog) // 2 -> 6
-			nextHash3 := hash4x64(cv>>24, hashLog) // 3 -> 7
+			nextHash := hashLen(cv, hashLog, bestShortLen)      // 0 -> 4
+			nextHash1 := hashLen(cv>>8, hashLog, bestShortLen)  // 1 -> 5
+			nextHash2 := hashLen(cv>>16, hashLog, bestShortLen) // 2 -> 6
+			nextHash3 := hashLen(cv>>24, hashLog, bestShortLen) // 3 -> 7
 			e.dictTable[nextHash] = prevEntry{
 				prev:   e.dictTable[nextHash].offset,
 				offset: i,
@@ -472,7 +529,7 @@
 		}
 		if len(d.content) >= 8 {
 			cv := load6432(d.content, 0)
-			h := hash8(cv, bestLongTableBits)
+			h := hashLen(cv, bestLongTableBits, bestLongLen)
 			e.dictLongTable[h] = prevEntry{
 				offset: e.maxMatchOff,
 				prev:   e.dictLongTable[h].offset,
@@ -482,7 +539,7 @@
 			off := 8 // First to read
 			for i := e.maxMatchOff + 1; i < end; i++ {
 				cv = cv>>8 | (uint64(d.content[off]) << 56)
-				h := hash8(cv, bestLongTableBits)
+				h := hashLen(cv, bestLongTableBits, bestLongLen)
 				e.dictLongTable[h] = prevEntry{
 					offset: i,
 					prev:   e.dictLongTable[h].offset,
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_better.go b/vendor/github.com/klauspost/compress/zstd/enc_better.go
index 6049542..c769f69 100644
--- a/vendor/github.com/klauspost/compress/zstd/enc_better.go
+++ b/vendor/github.com/klauspost/compress/zstd/enc_better.go
@@ -9,6 +9,7 @@
 const (
 	betterLongTableBits = 19                       // Bits used in the long match table
 	betterLongTableSize = 1 << betterLongTableBits // Size of the table
+	betterLongLen       = 8                        // Bytes used for table hash
 
 	// Note: Increasing the short table bits or making the hash shorter
 	// can actually lead to compression degradation since it will 'steal' more from the
@@ -16,6 +17,7 @@
 	// This greatly depends on the type of input.
 	betterShortTableBits = 13                        // Bits used in the short match table
 	betterShortTableSize = 1 << betterShortTableBits // Size of the table
+	betterShortLen       = 5                         // Bytes used for table hash
 
 	betterLongTableShardCnt  = 1 << (betterLongTableBits - dictShardBits)    // Number of shards in the table
 	betterLongTableShardSize = betterLongTableSize / betterLongTableShardCnt // Size of an individual shard
@@ -138,7 +140,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -154,8 +156,8 @@
 				panic("offset0 was 0")
 			}
 
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
+			nextHashL := hashLen(cv, betterLongTableBits, betterLongLen)
+			nextHashS := hashLen(cv, betterShortTableBits, betterShortLen)
 			candidateL := e.longTable[nextHashL]
 			candidateS := e.table[nextHashS]
 
@@ -204,7 +206,7 @@
 
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, lenght)
 
 						}
@@ -214,10 +216,10 @@
 					for index0 < s-1 {
 						cv0 := load6432(src, index0)
 						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
+						h0 := hashLen(cv0, betterLongTableBits, betterLongLen)
 						off := index0 + e.cur
 						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-						e.table[hash5(cv1, betterShortTableBits)] = tableEntry{offset: off + 1, val: uint32(cv1)}
+						e.table[hashLen(cv1, betterShortTableBits, betterShortLen)] = tableEntry{offset: off + 1, val: uint32(cv1)}
 						index0 += 2
 					}
 					cv = load6432(src, s)
@@ -264,7 +266,7 @@
 					s += lenght + repOff2
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, lenght)
 
 						}
@@ -275,10 +277,10 @@
 					for index0 < s-1 {
 						cv0 := load6432(src, index0)
 						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
+						h0 := hashLen(cv0, betterLongTableBits, betterLongLen)
 						off := index0 + e.cur
 						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-						e.table[hash5(cv1, betterShortTableBits)] = tableEntry{offset: off + 1, val: uint32(cv1)}
+						e.table[hashLen(cv1, betterShortTableBits, betterShortLen)] = tableEntry{offset: off + 1, val: uint32(cv1)}
 						index0 += 2
 					}
 					cv = load6432(src, s)
@@ -353,7 +355,7 @@
 				// See if we can find a long match at s+1
 				const checkAt = 1
 				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, betterLongTableBits)
+				nextHashL = hashLen(cv, betterLongTableBits, betterLongLen)
 				candidateL = e.longTable[nextHashL]
 				coffsetL = candidateL.offset - e.cur
 
@@ -413,8 +415,8 @@
 		}
 
 		// Try to find a better match by searching for a long match at the end of the current best match
-		if true && s+matched < sLimit {
-			nextHashL := hash8(load6432(src, s+matched), betterLongTableBits)
+		if s+matched < sLimit {
+			nextHashL := hashLen(load6432(src, s+matched), betterLongTableBits, betterLongLen)
 			cv := load3232(src, s)
 			candidateL := e.longTable[nextHashL]
 			coffsetL := candidateL.offset - e.cur - matched
@@ -495,10 +497,10 @@
 		for index0 < s-1 {
 			cv0 := load6432(src, index0)
 			cv1 := cv0 >> 8
-			h0 := hash8(cv0, betterLongTableBits)
+			h0 := hashLen(cv0, betterLongTableBits, betterLongLen)
 			off := index0 + e.cur
 			e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-			e.table[hash5(cv1, betterShortTableBits)] = tableEntry{offset: off + 1, val: uint32(cv1)}
+			e.table[hashLen(cv1, betterShortTableBits, betterShortLen)] = tableEntry{offset: off + 1, val: uint32(cv1)}
 			index0 += 2
 		}
 
@@ -516,8 +518,8 @@
 			}
 
 			// Store this, since we have it.
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
+			nextHashL := hashLen(cv, betterLongTableBits, betterLongLen)
+			nextHashS := hashLen(cv, betterShortTableBits, betterShortLen)
 
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
@@ -553,7 +555,7 @@
 	}
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 }
@@ -656,7 +658,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -672,8 +674,8 @@
 				panic("offset0 was 0")
 			}
 
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
+			nextHashL := hashLen(cv, betterLongTableBits, betterLongLen)
+			nextHashS := hashLen(cv, betterShortTableBits, betterShortLen)
 			candidateL := e.longTable[nextHashL]
 			candidateS := e.table[nextHashS]
 
@@ -724,7 +726,7 @@
 
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, lenght)
 
 						}
@@ -734,11 +736,11 @@
 					for index0 < s-1 {
 						cv0 := load6432(src, index0)
 						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
+						h0 := hashLen(cv0, betterLongTableBits, betterLongLen)
 						off := index0 + e.cur
 						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
 						e.markLongShardDirty(h0)
-						h1 := hash5(cv1, betterShortTableBits)
+						h1 := hashLen(cv1, betterShortTableBits, betterShortLen)
 						e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)}
 						e.markShortShardDirty(h1)
 						index0 += 2
@@ -787,7 +789,7 @@
 					s += lenght + repOff2
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, lenght)
 
 						}
@@ -798,11 +800,11 @@
 					for index0 < s-1 {
 						cv0 := load6432(src, index0)
 						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
+						h0 := hashLen(cv0, betterLongTableBits, betterLongLen)
 						off := index0 + e.cur
 						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
 						e.markLongShardDirty(h0)
-						h1 := hash5(cv1, betterShortTableBits)
+						h1 := hashLen(cv1, betterShortTableBits, betterShortLen)
 						e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)}
 						e.markShortShardDirty(h1)
 						index0 += 2
@@ -879,7 +881,7 @@
 				// See if we can find a long match at s+1
 				const checkAt = 1
 				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, betterLongTableBits)
+				nextHashL = hashLen(cv, betterLongTableBits, betterLongLen)
 				candidateL = e.longTable[nextHashL]
 				coffsetL = candidateL.offset - e.cur
 
@@ -940,7 +942,7 @@
 		}
 		// Try to find a better match by searching for a long match at the end of the current best match
 		if s+matched < sLimit {
-			nextHashL := hash8(load6432(src, s+matched), betterLongTableBits)
+			nextHashL := hashLen(load6432(src, s+matched), betterLongTableBits, betterLongLen)
 			cv := load3232(src, s)
 			candidateL := e.longTable[nextHashL]
 			coffsetL := candidateL.offset - e.cur - matched
@@ -1021,11 +1023,11 @@
 		for index0 < s-1 {
 			cv0 := load6432(src, index0)
 			cv1 := cv0 >> 8
-			h0 := hash8(cv0, betterLongTableBits)
+			h0 := hashLen(cv0, betterLongTableBits, betterLongLen)
 			off := index0 + e.cur
 			e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
 			e.markLongShardDirty(h0)
-			h1 := hash5(cv1, betterShortTableBits)
+			h1 := hashLen(cv1, betterShortTableBits, betterShortLen)
 			e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)}
 			e.markShortShardDirty(h1)
 			index0 += 2
@@ -1045,8 +1047,8 @@
 			}
 
 			// Store this, since we have it.
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
+			nextHashL := hashLen(cv, betterLongTableBits, betterLongLen)
+			nextHashS := hashLen(cv, betterShortTableBits, betterShortLen)
 
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
@@ -1084,7 +1086,7 @@
 	}
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 }
@@ -1113,10 +1115,10 @@
 			const hashLog = betterShortTableBits
 
 			cv := load6432(d.content, i-e.maxMatchOff)
-			nextHash := hash5(cv, hashLog)      // 0 -> 4
-			nextHash1 := hash5(cv>>8, hashLog)  // 1 -> 5
-			nextHash2 := hash5(cv>>16, hashLog) // 2 -> 6
-			nextHash3 := hash5(cv>>24, hashLog) // 3 -> 7
+			nextHash := hashLen(cv, hashLog, betterShortLen)      // 0 -> 4
+			nextHash1 := hashLen(cv>>8, hashLog, betterShortLen)  // 1 -> 5
+			nextHash2 := hashLen(cv>>16, hashLog, betterShortLen) // 2 -> 6
+			nextHash3 := hashLen(cv>>24, hashLog, betterShortLen) // 3 -> 7
 			e.dictTable[nextHash] = tableEntry{
 				val:    uint32(cv),
 				offset: i,
@@ -1145,7 +1147,7 @@
 		}
 		if len(d.content) >= 8 {
 			cv := load6432(d.content, 0)
-			h := hash8(cv, betterLongTableBits)
+			h := hashLen(cv, betterLongTableBits, betterLongLen)
 			e.dictLongTable[h] = prevEntry{
 				offset: e.maxMatchOff,
 				prev:   e.dictLongTable[h].offset,
@@ -1155,7 +1157,7 @@
 			off := 8 // First to read
 			for i := e.maxMatchOff + 1; i < end; i++ {
 				cv = cv>>8 | (uint64(d.content[off]) << 56)
-				h := hash8(cv, betterLongTableBits)
+				h := hashLen(cv, betterLongTableBits, betterLongLen)
 				e.dictLongTable[h] = prevEntry{
 					offset: i,
 					prev:   e.dictLongTable[h].offset,
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_dfast.go b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go
index 8629d43..7ff0c64 100644
--- a/vendor/github.com/klauspost/compress/zstd/enc_dfast.go
+++ b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go
@@ -10,6 +10,7 @@
 	dFastLongTableBits = 17                      // Bits used in the long match table
 	dFastLongTableSize = 1 << dFastLongTableBits // Size of the table
 	dFastLongTableMask = dFastLongTableSize - 1  // Mask for table indices. Redundant, but can eliminate bounds checks.
+	dFastLongLen       = 8                       // Bytes used for table hash
 
 	dLongTableShardCnt  = 1 << (dFastLongTableBits - dictShardBits) // Number of shards in the table
 	dLongTableShardSize = dFastLongTableSize / tableShardCnt        // Size of an individual shard
@@ -17,6 +18,8 @@
 	dFastShortTableBits = tableBits                // Bits used in the short match table
 	dFastShortTableSize = 1 << dFastShortTableBits // Size of the table
 	dFastShortTableMask = dFastShortTableSize - 1  // Mask for table indices. Redundant, but can eliminate bounds checks.
+	dFastShortLen       = 5                        // Bytes used for table hash
+
 )
 
 type doubleFastEncoder struct {
@@ -109,7 +112,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -124,8 +127,8 @@
 				panic("offset0 was 0")
 			}
 
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
+			nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen)
+			nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen)
 			candidateL := e.longTable[nextHashL]
 			candidateS := e.table[nextHashS]
 
@@ -170,7 +173,7 @@
 					s += lenght + repOff
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, lenght)
 
 						}
@@ -208,7 +211,7 @@
 				// See if we can find a long match at s+1
 				const checkAt = 1
 				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, dFastLongTableBits)
+				nextHashL = hashLen(cv, dFastLongTableBits, dFastLongLen)
 				candidateL = e.longTable[nextHashL]
 				coffsetL = s - (candidateL.offset - e.cur) + checkAt
 
@@ -304,16 +307,16 @@
 		cv1 := load6432(src, index1)
 		te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
 		te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
-		e.longTable[hash8(cv0, dFastLongTableBits)] = te0
-		e.longTable[hash8(cv1, dFastLongTableBits)] = te1
+		e.longTable[hashLen(cv0, dFastLongTableBits, dFastLongLen)] = te0
+		e.longTable[hashLen(cv1, dFastLongTableBits, dFastLongLen)] = te1
 		cv0 >>= 8
 		cv1 >>= 8
 		te0.offset++
 		te1.offset++
 		te0.val = uint32(cv0)
 		te1.val = uint32(cv1)
-		e.table[hash5(cv0, dFastShortTableBits)] = te0
-		e.table[hash5(cv1, dFastShortTableBits)] = te1
+		e.table[hashLen(cv0, dFastShortTableBits, dFastShortLen)] = te0
+		e.table[hashLen(cv1, dFastShortTableBits, dFastShortLen)] = te1
 
 		cv = load6432(src, s)
 
@@ -330,8 +333,8 @@
 			}
 
 			// Store this, since we have it.
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
+			nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen)
+			nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen)
 
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
@@ -368,7 +371,7 @@
 	}
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 }
@@ -427,7 +430,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -436,8 +439,8 @@
 		var t int32
 		for {
 
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
+			nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen)
+			nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen)
 			candidateL := e.longTable[nextHashL]
 			candidateS := e.table[nextHashS]
 
@@ -483,7 +486,7 @@
 					s += length + repOff
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, length)
 
 						}
@@ -521,7 +524,7 @@
 				// See if we can find a long match at s+1
 				const checkAt = 1
 				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, dFastLongTableBits)
+				nextHashL = hashLen(cv, dFastLongTableBits, dFastLongLen)
 				candidateL = e.longTable[nextHashL]
 				coffsetL = s - (candidateL.offset - e.cur) + checkAt
 
@@ -614,16 +617,16 @@
 		cv1 := load6432(src, index1)
 		te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
 		te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
-		e.longTable[hash8(cv0, dFastLongTableBits)] = te0
-		e.longTable[hash8(cv1, dFastLongTableBits)] = te1
+		e.longTable[hashLen(cv0, dFastLongTableBits, dFastLongLen)] = te0
+		e.longTable[hashLen(cv1, dFastLongTableBits, dFastLongLen)] = te1
 		cv0 >>= 8
 		cv1 >>= 8
 		te0.offset++
 		te1.offset++
 		te0.val = uint32(cv0)
 		te1.val = uint32(cv1)
-		e.table[hash5(cv0, dFastShortTableBits)] = te0
-		e.table[hash5(cv1, dFastShortTableBits)] = te1
+		e.table[hashLen(cv0, dFastShortTableBits, dFastShortLen)] = te0
+		e.table[hashLen(cv1, dFastShortTableBits, dFastShortLen)] = te1
 
 		cv = load6432(src, s)
 
@@ -640,8 +643,8 @@
 			}
 
 			// Store this, since we have it.
-			nextHashS := hash5(cv1>>8, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
+			nextHashS := hashLen(cv1>>8, dFastShortTableBits, dFastShortLen)
+			nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen)
 
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
@@ -677,7 +680,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:]...)
 		blk.extraLits = len(src) - int(nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 
@@ -767,7 +770,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -782,8 +785,8 @@
 				panic("offset0 was 0")
 			}
 
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
+			nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen)
+			nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen)
 			candidateL := e.longTable[nextHashL]
 			candidateS := e.table[nextHashS]
 
@@ -830,7 +833,7 @@
 					s += lenght + repOff
 					nextEmit = s
 					if s >= sLimit {
-						if debug {
+						if debugEncoder {
 							println("repeat ended", s, lenght)
 
 						}
@@ -868,7 +871,7 @@
 				// See if we can find a long match at s+1
 				const checkAt = 1
 				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, dFastLongTableBits)
+				nextHashL = hashLen(cv, dFastLongTableBits, dFastLongLen)
 				candidateL = e.longTable[nextHashL]
 				coffsetL = s - (candidateL.offset - e.cur) + checkAt
 
@@ -965,8 +968,8 @@
 		cv1 := load6432(src, index1)
 		te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
 		te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
-		longHash1 := hash8(cv0, dFastLongTableBits)
-		longHash2 := hash8(cv0, dFastLongTableBits)
+		longHash1 := hashLen(cv0, dFastLongTableBits, dFastLongLen)
+		longHash2 := hashLen(cv1, dFastLongTableBits, dFastLongLen)
 		e.longTable[longHash1] = te0
 		e.longTable[longHash2] = te1
 		e.markLongShardDirty(longHash1)
@@ -977,8 +980,8 @@
 		te1.offset++
 		te0.val = uint32(cv0)
 		te1.val = uint32(cv1)
-		hashVal1 := hash5(cv0, dFastShortTableBits)
-		hashVal2 := hash5(cv1, dFastShortTableBits)
+		hashVal1 := hashLen(cv0, dFastShortTableBits, dFastShortLen)
+		hashVal2 := hashLen(cv1, dFastShortTableBits, dFastShortLen)
 		e.table[hashVal1] = te0
 		e.markShardDirty(hashVal1)
 		e.table[hashVal2] = te1
@@ -999,8 +1002,8 @@
 			}
 
 			// Store this, since we have it.
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
+			nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen)
+			nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen)
 
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
@@ -1039,7 +1042,7 @@
 	}
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 	// If we encoded more than 64K mark all dirty.
@@ -1071,14 +1074,14 @@
 		}
 		if len(d.content) >= 8 {
 			cv := load6432(d.content, 0)
-			e.dictLongTable[hash8(cv, dFastLongTableBits)] = tableEntry{
+			e.dictLongTable[hashLen(cv, dFastLongTableBits, dFastLongLen)] = tableEntry{
 				val:    uint32(cv),
 				offset: e.maxMatchOff,
 			}
 			end := int32(len(d.content)) - 8 + e.maxMatchOff
 			for i := e.maxMatchOff + 1; i < end; i++ {
 				cv = cv>>8 | (uint64(d.content[i-e.maxMatchOff+7]) << 56)
-				e.dictLongTable[hash8(cv, dFastLongTableBits)] = tableEntry{
+				e.dictLongTable[hashLen(cv, dFastLongTableBits, dFastLongLen)] = tableEntry{
 					val:    uint32(cv),
 					offset: i,
 				}
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_fast.go b/vendor/github.com/klauspost/compress/zstd/enc_fast.go
index ba4a17e..f51ab52 100644
--- a/vendor/github.com/klauspost/compress/zstd/enc_fast.go
+++ b/vendor/github.com/klauspost/compress/zstd/enc_fast.go
@@ -6,17 +6,16 @@
 
 import (
 	"fmt"
-	"math"
-	"math/bits"
 )
 
 const (
-	tableBits      = 15                               // Bits used in the table
-	tableSize      = 1 << tableBits                   // Size of the table
-	tableShardCnt  = 1 << (tableBits - dictShardBits) // Number of shards in the table
-	tableShardSize = tableSize / tableShardCnt        // Size of an individual shard
-	tableMask      = tableSize - 1                    // Mask for table indices. Redundant, but can eliminate bounds checks.
-	maxMatchLength = 131074
+	tableBits        = 15                               // Bits used in the table
+	tableSize        = 1 << tableBits                   // Size of the table
+	tableShardCnt    = 1 << (tableBits - dictShardBits) // Number of shards in the table
+	tableShardSize   = tableSize / tableShardCnt        // Size of an individual shard
+	tableFastHashLen = 6
+	tableMask        = tableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks.
+	maxMatchLength   = 131074
 )
 
 type tableEntry struct {
@@ -86,7 +85,7 @@
 	// TEMPLATE
 	const hashLog = tableBits
 	// seems global, but would be nice to tweak.
-	const kSearchStrength = 7
+	const kSearchStrength = 6
 
 	// nextEmit is where in src the next emitLiteral should start from.
 	nextEmit := s
@@ -103,7 +102,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -122,8 +121,8 @@
 				panic("offset0 was 0")
 			}
 
-			nextHash := hash6(cv, hashLog)
-			nextHash2 := hash6(cv>>8, hashLog)
+			nextHash := hashLen(cv, hashLog, tableFastHashLen)
+			nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen)
 			candidate := e.table[nextHash]
 			candidate2 := e.table[nextHash2]
 			repIndex := s - offset1 + 2
@@ -135,20 +134,7 @@
 				// Consider history as well.
 				var seq seq
 				var length int32
-				// length = 4 + e.matchlen(s+6, repIndex+4, src)
-				{
-					a := src[s+6:]
-					b := src[repIndex+4:]
-					endI := len(a) & (math.MaxInt32 - 7)
-					length = int32(endI) + 4
-					for i := 0; i < endI; i += 8 {
-						if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-							length = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-							break
-						}
-					}
-				}
-
+				length = 4 + e.matchlen(s+6, repIndex+4, src)
 				seq.matchLen = uint32(length - zstdMinMatch)
 
 				// We might be able to match backwards.
@@ -178,7 +164,7 @@
 				s += length + 2
 				nextEmit = s
 				if s >= sLimit {
-					if debug {
+					if debugEncoder {
 						println("repeat ended", s, length)
 
 					}
@@ -235,20 +221,7 @@
 		}
 
 		// Extend the 4-byte match as long as possible.
-		//l := e.matchlen(s+4, t+4, src) + 4
-		var l int32
-		{
-			a := src[s+4:]
-			b := src[t+4:]
-			endI := len(a) & (math.MaxInt32 - 7)
-			l = int32(endI) + 4
-			for i := 0; i < endI; i += 8 {
-				if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-					l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-					break
-				}
-			}
-		}
+		l := e.matchlen(s+4, t+4, src) + 4
 
 		// Extend backwards
 		tMin := s - e.maxMatchOff
@@ -285,23 +258,10 @@
 		if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) {
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlen(s+4, o2+4, src)
-			var l int32
-			{
-				a := src[s+4:]
-				b := src[o2+4:]
-				endI := len(a) & (math.MaxInt32 - 7)
-				l = int32(endI) + 4
-				for i := 0; i < endI; i += 8 {
-					if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-						l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-						break
-					}
-				}
-			}
+			l := 4 + e.matchlen(s+4, o2+4, src)
 
 			// Store this, since we have it.
-			nextHash := hash6(cv, hashLog)
+			nextHash := hashLen(cv, hashLog, tableFastHashLen)
 			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
 			seq.matchLen = uint32(l) - zstdMinMatch
 			seq.litLen = 0
@@ -330,7 +290,7 @@
 	}
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 }
@@ -343,7 +303,7 @@
 		inputMargin            = 8
 		minNonLiteralBlockSize = 1 + 1 + inputMargin
 	)
-	if debug {
+	if debugEncoder {
 		if len(src) > maxBlockSize {
 			panic("src too big")
 		}
@@ -374,7 +334,7 @@
 	// TEMPLATE
 	const hashLog = tableBits
 	// seems global, but would be nice to tweak.
-	const kSearchStrength = 8
+	const kSearchStrength = 6
 
 	// nextEmit is where in src the next emitLiteral should start from.
 	nextEmit := s
@@ -391,7 +351,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -405,8 +365,8 @@
 		// By not using them for the first 3 matches
 
 		for {
-			nextHash := hash6(cv, hashLog)
-			nextHash2 := hash6(cv>>8, hashLog)
+			nextHash := hashLen(cv, hashLog, tableFastHashLen)
+			nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen)
 			candidate := e.table[nextHash]
 			candidate2 := e.table[nextHash2]
 			repIndex := s - offset1 + 2
@@ -417,21 +377,7 @@
 			if len(blk.sequences) > 2 && load3232(src, repIndex) == uint32(cv>>16) {
 				// Consider history as well.
 				var seq seq
-				// length := 4 + e.matchlen(s+6, repIndex+4, src)
-				// length := 4 + int32(matchLen(src[s+6:], src[repIndex+4:]))
-				var length int32
-				{
-					a := src[s+6:]
-					b := src[repIndex+4:]
-					endI := len(a) & (math.MaxInt32 - 7)
-					length = int32(endI) + 4
-					for i := 0; i < endI; i += 8 {
-						if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-							length = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-							break
-						}
-					}
-				}
+				length := 4 + e.matchlen(s+6, repIndex+4, src)
 
 				seq.matchLen = uint32(length - zstdMinMatch)
 
@@ -462,7 +408,7 @@
 				s += length + 2
 				nextEmit = s
 				if s >= sLimit {
-					if debug {
+					if debugEncoder {
 						println("repeat ended", s, length)
 
 					}
@@ -521,21 +467,7 @@
 			panic(fmt.Sprintf("t (%d) < 0 ", t))
 		}
 		// Extend the 4-byte match as long as possible.
-		//l := e.matchlenNoHist(s+4, t+4, src) + 4
-		// l := int32(matchLen(src[s+4:], src[t+4:])) + 4
-		var l int32
-		{
-			a := src[s+4:]
-			b := src[t+4:]
-			endI := len(a) & (math.MaxInt32 - 7)
-			l = int32(endI) + 4
-			for i := 0; i < endI; i += 8 {
-				if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-					l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-					break
-				}
-			}
-		}
+		l := e.matchlen(s+4, t+4, src) + 4
 
 		// Extend backwards
 		tMin := s - e.maxMatchOff
@@ -572,24 +504,10 @@
 		if o2 := s - offset2; len(blk.sequences) > 2 && load3232(src, o2) == uint32(cv) {
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlenNoHist(s+4, o2+4, src)
-			// l := 4 + int32(matchLen(src[s+4:], src[o2+4:]))
-			var l int32
-			{
-				a := src[s+4:]
-				b := src[o2+4:]
-				endI := len(a) & (math.MaxInt32 - 7)
-				l = int32(endI) + 4
-				for i := 0; i < endI; i += 8 {
-					if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-						l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-						break
-					}
-				}
-			}
+			l := 4 + e.matchlen(s+4, o2+4, src)
 
 			// Store this, since we have it.
-			nextHash := hash6(cv, hashLog)
+			nextHash := hashLen(cv, hashLog, tableFastHashLen)
 			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
 			seq.matchLen = uint32(l) - zstdMinMatch
 			seq.litLen = 0
@@ -616,7 +534,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:]...)
 		blk.extraLits = len(src) - int(nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 	// We do not store history, so we must offset e.cur to avoid false matches for next user.
@@ -696,7 +614,7 @@
 		blk.literals = append(blk.literals, src[nextEmit:until]...)
 		s.litLen = uint32(until - nextEmit)
 	}
-	if debug {
+	if debugEncoder {
 		println("recent offsets:", blk.recentOffsets)
 	}
 
@@ -715,8 +633,8 @@
 				panic("offset0 was 0")
 			}
 
-			nextHash := hash6(cv, hashLog)
-			nextHash2 := hash6(cv>>8, hashLog)
+			nextHash := hashLen(cv, hashLog, tableFastHashLen)
+			nextHash2 := hashLen(cv>>8, hashLog, tableFastHashLen)
 			candidate := e.table[nextHash]
 			candidate2 := e.table[nextHash2]
 			repIndex := s - offset1 + 2
@@ -730,19 +648,7 @@
 				// Consider history as well.
 				var seq seq
 				var length int32
-				// length = 4 + e.matchlen(s+6, repIndex+4, src)
-				{
-					a := src[s+6:]
-					b := src[repIndex+4:]
-					endI := len(a) & (math.MaxInt32 - 7)
-					length = int32(endI) + 4
-					for i := 0; i < endI; i += 8 {
-						if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-							length = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-							break
-						}
-					}
-				}
+				length = 4 + e.matchlen(s+6, repIndex+4, src)
 
 				seq.matchLen = uint32(length - zstdMinMatch)
 
@@ -773,7 +679,7 @@
 				s += length + 2
 				nextEmit = s
 				if s >= sLimit {
-					if debug {
+					if debugEncoder {
 						println("repeat ended", s, length)
 
 					}
@@ -830,20 +736,7 @@
 		}
 
 		// Extend the 4-byte match as long as possible.
-		//l := e.matchlen(s+4, t+4, src) + 4
-		var l int32
-		{
-			a := src[s+4:]
-			b := src[t+4:]
-			endI := len(a) & (math.MaxInt32 - 7)
-			l = int32(endI) + 4
-			for i := 0; i < endI; i += 8 {
-				if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-					l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-					break
-				}
-			}
-		}
+		l := e.matchlen(s+4, t+4, src) + 4
 
 		// Extend backwards
 		tMin := s - e.maxMatchOff
@@ -880,23 +773,10 @@
 		if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) {
 			// We have at least 4 byte match.
 			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlen(s+4, o2+4, src)
-			var l int32
-			{
-				a := src[s+4:]
-				b := src[o2+4:]
-				endI := len(a) & (math.MaxInt32 - 7)
-				l = int32(endI) + 4
-				for i := 0; i < endI; i += 8 {
-					if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-						l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-						break
-					}
-				}
-			}
+			l := 4 + e.matchlen(s+4, o2+4, src)
 
 			// Store this, since we have it.
-			nextHash := hash6(cv, hashLog)
+			nextHash := hashLen(cv, hashLog, tableFastHashLen)
 			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
 			e.markShardDirty(nextHash)
 			seq.matchLen = uint32(l) - zstdMinMatch
@@ -926,7 +806,7 @@
 	}
 	blk.recentOffsets[0] = uint32(offset1)
 	blk.recentOffsets[1] = uint32(offset2)
-	if debug {
+	if debugEncoder {
 		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
 	}
 }
@@ -957,9 +837,9 @@
 				const hashLog = tableBits
 
 				cv := load6432(d.content, i-e.maxMatchOff)
-				nextHash := hash6(cv, hashLog)      // 0 -> 5
-				nextHash1 := hash6(cv>>8, hashLog)  // 1 -> 6
-				nextHash2 := hash6(cv>>16, hashLog) // 2 -> 7
+				nextHash := hashLen(cv, hashLog, tableFastHashLen)      // 0 -> 5
+				nextHash1 := hashLen(cv>>8, hashLog, tableFastHashLen)  // 1 -> 6
+				nextHash2 := hashLen(cv>>16, hashLog, tableFastHashLen) // 2 -> 7
 				e.dictTable[nextHash] = tableEntry{
 					val:    uint32(cv),
 					offset: i,
diff --git a/vendor/github.com/klauspost/compress/zstd/encoder.go b/vendor/github.com/klauspost/compress/zstd/encoder.go
index 4871dd0..7aaaedb 100644
--- a/vendor/github.com/klauspost/compress/zstd/encoder.go
+++ b/vendor/github.com/klauspost/compress/zstd/encoder.go
@@ -33,7 +33,7 @@
 	Block() *blockEnc
 	CRC() *xxhash.Digest
 	AppendCRC([]byte) []byte
-	WindowSize(size int) int32
+	WindowSize(size int64) int32
 	UseBlock(*blockEnc)
 	Reset(d *dict, singleBlock bool)
 }
@@ -48,6 +48,8 @@
 	err              error
 	writeErr         error
 	nWritten         int64
+	nInput           int64
+	frameContentSize int64
 	headerWritten    bool
 	eofWritten       bool
 	fullFrameWritten bool
@@ -96,23 +98,25 @@
 	if cap(s.filling) == 0 {
 		s.filling = make([]byte, 0, e.o.blockSize)
 	}
-	if cap(s.current) == 0 {
-		s.current = make([]byte, 0, e.o.blockSize)
-	}
-	if cap(s.previous) == 0 {
-		s.previous = make([]byte, 0, e.o.blockSize)
+	if e.o.concurrent > 1 {
+		if cap(s.current) == 0 {
+			s.current = make([]byte, 0, e.o.blockSize)
+		}
+		if cap(s.previous) == 0 {
+			s.previous = make([]byte, 0, e.o.blockSize)
+		}
+		s.current = s.current[:0]
+		s.previous = s.previous[:0]
+		if s.writing == nil {
+			s.writing = &blockEnc{lowMem: e.o.lowMem}
+			s.writing.init()
+		}
+		s.writing.initNewEncode()
 	}
 	if s.encoder == nil {
 		s.encoder = e.o.encoder()
 	}
-	if s.writing == nil {
-		s.writing = &blockEnc{lowMem: e.o.lowMem}
-		s.writing.init()
-	}
-	s.writing.initNewEncode()
 	s.filling = s.filling[:0]
-	s.current = s.current[:0]
-	s.previous = s.previous[:0]
 	s.encoder.Reset(e.o.dict, false)
 	s.headerWritten = false
 	s.eofWritten = false
@@ -120,7 +124,21 @@
 	s.w = w
 	s.err = nil
 	s.nWritten = 0
+	s.nInput = 0
 	s.writeErr = nil
+	s.frameContentSize = 0
+}
+
+// ResetContentSize will reset and set a content size for the next stream.
+// If the bytes written does not match the size given an error will be returned
+// when calling Close().
+// This is removed when Reset is called.
+// Sizes <= 0 results in no content size set.
+func (e *Encoder) ResetContentSize(w io.Writer, size int64) {
+	e.Reset(w)
+	if size >= 0 {
+		e.state.frameContentSize = size
+	}
 }
 
 // Write data to the encoder.
@@ -190,6 +208,7 @@
 				return s.err
 			}
 			s.nWritten += int64(n2)
+			s.nInput += int64(len(s.filling))
 			s.current = s.current[:0]
 			s.filling = s.filling[:0]
 			s.headerWritten = true
@@ -200,8 +219,8 @@
 
 		var tmp [maxHeaderSize]byte
 		fh := frameHeader{
-			ContentSize:   0,
-			WindowSize:    uint32(s.encoder.WindowSize(0)),
+			ContentSize:   uint64(s.frameContentSize),
+			WindowSize:    uint32(s.encoder.WindowSize(s.frameContentSize)),
 			SingleSegment: false,
 			Checksum:      e.o.crc,
 			DictID:        e.o.dict.ID(),
@@ -241,11 +260,52 @@
 		return s.err
 	}
 
+	// SYNC:
+	if e.o.concurrent == 1 {
+		src := s.filling
+		s.nInput += int64(len(s.filling))
+		if debugEncoder {
+			println("Adding sync block,", len(src), "bytes, final:", final)
+		}
+		enc := s.encoder
+		blk := enc.Block()
+		blk.reset(nil)
+		enc.Encode(blk, src)
+		blk.last = final
+		if final {
+			s.eofWritten = true
+		}
+
+		err := errIncompressible
+		// If we got the exact same number of literals as input,
+		// assume the literals cannot be compressed.
+		if len(src) != len(blk.literals) || len(src) != e.o.blockSize {
+			err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy)
+		}
+		switch err {
+		case errIncompressible:
+			if debugEncoder {
+				println("Storing incompressible block as raw")
+			}
+			blk.encodeRaw(src)
+			// In fast mode, we do not transfer offsets, so we don't have to deal with changing the.
+		case nil:
+		default:
+			s.err = err
+			return err
+		}
+		_, s.err = s.w.Write(blk.output)
+		s.nWritten += int64(len(blk.output))
+		s.filling = s.filling[:0]
+		return s.err
+	}
+
 	// Move blocks forward.
 	s.filling, s.current, s.previous = s.previous[:0], s.filling, s.current
+	s.nInput += int64(len(s.current))
 	s.wg.Add(1)
 	go func(src []byte) {
-		if debug {
+		if debugEncoder {
 			println("Adding block,", len(src), "bytes, final:", final)
 		}
 		defer func() {
@@ -290,7 +350,7 @@
 			}
 			switch err {
 			case errIncompressible:
-				if debug {
+				if debugEncoder {
 					println("Storing incompressible block as raw")
 				}
 				blk.encodeRaw(src)
@@ -313,7 +373,7 @@
 //
 // The Copy function uses ReaderFrom if available.
 func (e *Encoder) ReadFrom(r io.Reader) (n int64, err error) {
-	if debug {
+	if debugEncoder {
 		println("Using ReadFrom")
 	}
 
@@ -336,20 +396,20 @@
 		switch err {
 		case io.EOF:
 			e.state.filling = e.state.filling[:len(e.state.filling)-len(src)]
-			if debug {
+			if debugEncoder {
 				println("ReadFrom: got EOF final block:", len(e.state.filling))
 			}
 			return n, nil
 		case nil:
 		default:
-			if debug {
+			if debugEncoder {
 				println("ReadFrom: got error:", err)
 			}
 			e.state.err = err
 			return n, err
 		}
 		if len(src) > 0 {
-			if debug {
+			if debugEncoder {
 				println("ReadFrom: got space left in source:", len(src))
 			}
 			continue
@@ -394,6 +454,11 @@
 	if err != nil {
 		return err
 	}
+	if s.frameContentSize > 0 {
+		if s.nInput != s.frameContentSize {
+			return fmt.Errorf("frame content size %d given, but %d bytes was written", s.frameContentSize, s.nInput)
+		}
+	}
 	if e.state.fullFrameWritten {
 		return s.err
 	}
@@ -463,14 +528,14 @@
 		// If a non-single block is needed the encoder will reset again.
 		e.encoders <- enc
 	}()
-	// Use single segments when above minimum window and below 1MB.
-	single := len(src) < 1<<20 && len(src) > MinWindowSize
+	// Use single segments when above minimum window and below window size.
+	single := len(src) <= e.o.windowSize && len(src) > MinWindowSize
 	if e.o.single != nil {
 		single = *e.o.single
 	}
 	fh := frameHeader{
 		ContentSize:   uint64(len(src)),
-		WindowSize:    uint32(enc.WindowSize(len(src))),
+		WindowSize:    uint32(enc.WindowSize(int64(len(src)))),
 		SingleSegment: single,
 		Checksum:      e.o.crc,
 		DictID:        e.o.dict.ID(),
@@ -486,7 +551,7 @@
 	}
 
 	// If we can do everything in one block, prefer that.
-	if len(src) <= maxCompressedBlockSize {
+	if len(src) <= e.o.blockSize {
 		enc.Reset(e.o.dict, true)
 		// Slightly faster with no history and everything in one block.
 		if e.o.crc {
@@ -512,7 +577,7 @@
 
 		switch err {
 		case errIncompressible:
-			if debug {
+			if debugEncoder {
 				println("Storing incompressible block as raw")
 			}
 			dst = blk.encodeRawTo(dst, src)
@@ -548,7 +613,7 @@
 
 			switch err {
 			case errIncompressible:
-				if debug {
+				if debugEncoder {
 					println("Storing incompressible block as raw")
 				}
 				dst = blk.encodeRawTo(dst, todo)
diff --git a/vendor/github.com/klauspost/compress/zstd/encoder_options.go b/vendor/github.com/klauspost/compress/zstd/encoder_options.go
index 16d4ab6..a7c5e1a 100644
--- a/vendor/github.com/klauspost/compress/zstd/encoder_options.go
+++ b/vendor/github.com/klauspost/compress/zstd/encoder_options.go
@@ -24,6 +24,7 @@
 	allLitEntropy   bool
 	customWindow    bool
 	customALEntropy bool
+	customBlockSize bool
 	lowMem          bool
 	dict            *dict
 }
@@ -33,7 +34,7 @@
 		concurrent:    runtime.GOMAXPROCS(0),
 		crc:           true,
 		single:        nil,
-		blockSize:     1 << 16,
+		blockSize:     maxCompressedBlockSize,
 		windowSize:    8 << 20,
 		level:         SpeedDefault,
 		allLitEntropy: true,
@@ -75,6 +76,7 @@
 // WithEncoderConcurrency will set the concurrency,
 // meaning the maximum number of encoders to run concurrently.
 // The value supplied must be at least 1.
+// For streams, setting a value of 1 will disable async compression.
 // By default this will be set to GOMAXPROCS.
 func WithEncoderConcurrency(n int) EOption {
 	return func(o *encoderOptions) error {
@@ -106,6 +108,7 @@
 		o.customWindow = true
 		if o.blockSize > o.windowSize {
 			o.blockSize = o.windowSize
+			o.customBlockSize = true
 		}
 		return nil
 	}
@@ -188,10 +191,9 @@
 		return SpeedDefault
 	case level >= 6 && level < 10:
 		return SpeedBetterCompression
-	case level >= 10:
-		return SpeedBetterCompression
+	default:
+		return SpeedBestCompression
 	}
-	return SpeedDefault
 }
 
 // String provides a string representation of the compression level.
@@ -222,6 +224,9 @@
 			switch o.level {
 			case SpeedFastest:
 				o.windowSize = 4 << 20
+				if !o.customBlockSize {
+					o.blockSize = 1 << 16
+				}
 			case SpeedDefault:
 				o.windowSize = 8 << 20
 			case SpeedBetterCompression:
@@ -278,7 +283,7 @@
 // a decoder is allowed to reject a compressed frame which requests a memory size beyond decoder's authorized range.
 // For broader compatibility, decoders are recommended to support memory sizes of at least 8 MB.
 // This is only a recommendation, each decoder is free to support higher or lower limits, depending on local limitations.
-// If this is not specified, block encodes will automatically choose this based on the input size.
+// If this is not specified, block encodes will automatically choose this based on the input size and the window size.
 // This setting has no effect on streamed encodes.
 func WithSingleSegment(b bool) EOption {
 	return func(o *encoderOptions) error {
diff --git a/vendor/github.com/klauspost/compress/zstd/framedec.go b/vendor/github.com/klauspost/compress/zstd/framedec.go
index 693c5f0..9568a4b 100644
--- a/vendor/github.com/klauspost/compress/zstd/framedec.go
+++ b/vendor/github.com/klauspost/compress/zstd/framedec.go
@@ -8,27 +8,17 @@
 	"bytes"
 	"encoding/hex"
 	"errors"
-	"hash"
 	"io"
-	"sync"
 
 	"github.com/klauspost/compress/zstd/internal/xxhash"
 )
 
 type frameDec struct {
-	o      decoderOptions
-	crc    hash.Hash64
-	offset int64
+	o   decoderOptions
+	crc *xxhash.Digest
 
 	WindowSize uint64
 
-	// maxWindowSize is the maximum windows size to support.
-	// should never be bigger than max-int.
-	maxWindowSize uint64
-
-	// In order queue of blocks being decoded.
-	decoding chan *blockDec
-
 	// Frame history passed between blocks
 	history history
 
@@ -38,20 +28,18 @@
 	bBuf byteBuf
 
 	FrameContentSize uint64
-	frameDone        sync.WaitGroup
 
 	DictionaryID  *uint32
 	HasCheckSum   bool
 	SingleSegment bool
-
-	// asyncRunning indicates whether the async routine processes input on 'decoding'.
-	asyncRunningMu sync.Mutex
-	asyncRunning   bool
 }
 
 const (
-	// The minimum Window_Size is 1 KB.
+	// MinWindowSize is the minimum Window Size, which is 1 KB.
 	MinWindowSize = 1 << 10
+
+	// MaxWindowSize is the maximum encoder window size
+	// and the default decoder maximum window size.
 	MaxWindowSize = 1 << 29
 )
 
@@ -61,12 +49,11 @@
 )
 
 func newFrameDec(o decoderOptions) *frameDec {
-	d := frameDec{
-		o:             o,
-		maxWindowSize: MaxWindowSize,
+	if o.maxWindowSize > o.maxDecodedSize {
+		o.maxWindowSize = o.maxDecodedSize
 	}
-	if d.maxWindowSize > o.maxDecodedSize {
-		d.maxWindowSize = o.maxDecodedSize
+	d := frameDec{
+		o: o,
 	}
 	return &d
 }
@@ -78,44 +65,68 @@
 func (d *frameDec) reset(br byteBuffer) error {
 	d.HasCheckSum = false
 	d.WindowSize = 0
-	var b []byte
+	var signature [4]byte
 	for {
-		b = br.readSmall(4)
-		if b == nil {
+		var err error
+		// Check if we can read more...
+		b, err := br.readSmall(1)
+		switch err {
+		case io.EOF, io.ErrUnexpectedEOF:
 			return io.EOF
+		default:
+			return err
+		case nil:
+			signature[0] = b[0]
 		}
-		if !bytes.Equal(b[1:4], skippableFrameMagic) || b[0]&0xf0 != 0x50 {
-			if debug {
-				println("Not skippable", hex.EncodeToString(b), hex.EncodeToString(skippableFrameMagic))
+		// Read the rest, don't allow io.ErrUnexpectedEOF
+		b, err = br.readSmall(3)
+		switch err {
+		case io.EOF:
+			return io.EOF
+		default:
+			return err
+		case nil:
+			copy(signature[1:], b)
+		}
+
+		if !bytes.Equal(signature[1:4], skippableFrameMagic) || signature[0]&0xf0 != 0x50 {
+			if debugDecoder {
+				println("Not skippable", hex.EncodeToString(signature[:]), hex.EncodeToString(skippableFrameMagic))
 			}
 			// Break if not skippable frame.
 			break
 		}
 		// Read size to skip
-		b = br.readSmall(4)
-		if b == nil {
-			println("Reading Frame Size EOF")
-			return io.ErrUnexpectedEOF
+		b, err = br.readSmall(4)
+		if err != nil {
+			if debugDecoder {
+				println("Reading Frame Size", err)
+			}
+			return err
 		}
 		n := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
 		println("Skipping frame with", n, "bytes.")
-		err := br.skipN(int(n))
+		err = br.skipN(int64(n))
 		if err != nil {
-			if debug {
+			if debugDecoder {
 				println("Reading discarded frame", err)
 			}
 			return err
 		}
 	}
-	if !bytes.Equal(b, frameMagic) {
-		println("Got magic numbers: ", b, "want:", frameMagic)
+	if !bytes.Equal(signature[:], frameMagic) {
+		if debugDecoder {
+			println("Got magic numbers: ", signature, "want:", frameMagic)
+		}
 		return ErrMagicMismatch
 	}
 
 	// Read Frame_Header_Descriptor
 	fhd, err := br.readByte()
 	if err != nil {
-		println("Reading Frame_Header_Descriptor", err)
+		if debugDecoder {
+			println("Reading Frame_Header_Descriptor", err)
+		}
 		return err
 	}
 	d.SingleSegment = fhd&(1<<5) != 0
@@ -130,7 +141,9 @@
 	if !d.SingleSegment {
 		wd, err := br.readByte()
 		if err != nil {
-			println("Reading Window_Descriptor", err)
+			if debugDecoder {
+				println("Reading Window_Descriptor", err)
+			}
 			return err
 		}
 		printf("raw: %x, mantissa: %d, exponent: %d\n", wd, wd&7, wd>>3)
@@ -147,12 +160,11 @@
 		if size == 3 {
 			size = 4
 		}
-		b = br.readSmall(int(size))
-		if b == nil {
-			if debug {
-				println("Reading Dictionary_ID", io.ErrUnexpectedEOF)
-			}
-			return io.ErrUnexpectedEOF
+
+		b, err := br.readSmall(int(size))
+		if err != nil {
+			println("Reading Dictionary_ID", err)
+			return err
 		}
 		var id uint32
 		switch size {
@@ -163,7 +175,7 @@
 		case 4:
 			id = uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
 		}
-		if debug {
+		if debugDecoder {
 			println("Dict size", size, "ID:", id)
 		}
 		if id > 0 {
@@ -185,12 +197,12 @@
 	default:
 		fcsSize = 1 << v
 	}
-	d.FrameContentSize = 0
+	d.FrameContentSize = fcsUnknown
 	if fcsSize > 0 {
-		b := br.readSmall(fcsSize)
-		if b == nil {
-			println("Reading Frame content", io.ErrUnexpectedEOF)
-			return io.ErrUnexpectedEOF
+		b, err := br.readSmall(fcsSize)
+		if err != nil {
+			println("Reading Frame content", err)
+			return err
 		}
 		switch fcsSize {
 		case 1:
@@ -205,10 +217,11 @@
 			d2 := uint32(b[4]) | (uint32(b[5]) << 8) | (uint32(b[6]) << 16) | (uint32(b[7]) << 24)
 			d.FrameContentSize = uint64(d1) | (uint64(d2) << 32)
 		}
-		if debug {
-			println("field size bits:", v, "fcsSize:", fcsSize, "FrameContentSize:", d.FrameContentSize, hex.EncodeToString(b[:fcsSize]), "singleseg:", d.SingleSegment, "window:", d.WindowSize)
+		if debugDecoder {
+			println("Read FCS:", d.FrameContentSize)
 		}
 	}
+
 	// Move this to shared.
 	d.HasCheckSum = fhd&(1<<2) != 0
 	if d.HasCheckSum {
@@ -218,29 +231,47 @@
 		d.crc.Reset()
 	}
 
+	if d.WindowSize > d.o.maxWindowSize {
+		if debugDecoder {
+			printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize)
+		}
+		return ErrWindowSizeExceeded
+	}
+
 	if d.WindowSize == 0 && d.SingleSegment {
 		// We may not need window in this case.
 		d.WindowSize = d.FrameContentSize
 		if d.WindowSize < MinWindowSize {
 			d.WindowSize = MinWindowSize
 		}
+		if d.WindowSize > d.o.maxDecodedSize {
+			if debugDecoder {
+				printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize)
+			}
+			return ErrDecoderSizeExceeded
+		}
 	}
 
-	if d.WindowSize > d.maxWindowSize {
-		printf("window size %d > max %d\n", d.WindowSize, d.maxWindowSize)
-		return ErrWindowSizeExceeded
-	}
 	// The minimum Window_Size is 1 KB.
 	if d.WindowSize < MinWindowSize {
-		println("got window size: ", d.WindowSize)
+		if debugDecoder {
+			println("got window size: ", d.WindowSize)
+		}
 		return ErrWindowSizeTooSmall
 	}
 	d.history.windowSize = int(d.WindowSize)
-	if d.o.lowMem && d.history.windowSize < maxBlockSize {
-		d.history.maxSize = d.history.windowSize * 2
+	if !d.o.lowMem || d.history.windowSize < maxBlockSize {
+		// Alloc 2x window size if not low-mem, or very small window size.
+		d.history.allocFrameBuffer = d.history.windowSize * 2
 	} else {
-		d.history.maxSize = d.history.windowSize + maxBlockSize
+		// Alloc with one additional block
+		d.history.allocFrameBuffer = d.history.windowSize + maxBlockSize
 	}
+
+	if debugDecoder {
+		println("Frame: Dict:", d.DictionaryID, "FrameContentSize:", d.FrameContentSize, "singleseg:", d.SingleSegment, "window:", d.WindowSize, "crc:", d.HasCheckSum)
+	}
+
 	// history contains input - maybe we do something
 	d.rawInput = br
 	return nil
@@ -248,56 +279,37 @@
 
 // next will start decoding the next block from stream.
 func (d *frameDec) next(block *blockDec) error {
-	if debug {
-		printf("decoding new block %p:%p", block, block.data)
+	if debugDecoder {
+		println("decoding new block")
 	}
 	err := block.reset(d.rawInput, d.WindowSize)
 	if err != nil {
 		println("block error:", err)
 		// Signal the frame decoder we have a problem.
-		d.sendErr(block, err)
+		block.sendErr(err)
 		return err
 	}
-	block.input <- struct{}{}
-	if debug {
-		println("next block:", block)
-	}
-	d.asyncRunningMu.Lock()
-	defer d.asyncRunningMu.Unlock()
-	if !d.asyncRunning {
-		return nil
-	}
-	if block.Last {
-		// We indicate the frame is done by sending io.EOF
-		d.decoding <- block
-		return io.EOF
-	}
-	d.decoding <- block
 	return nil
 }
 
-// sendEOF will queue an error block on the frame.
-// This will cause the frame decoder to return when it encounters the block.
-// Returns true if the decoder was added.
-func (d *frameDec) sendErr(block *blockDec, err error) bool {
-	d.asyncRunningMu.Lock()
-	defer d.asyncRunningMu.Unlock()
-	if !d.asyncRunning {
-		return false
-	}
-
-	println("sending error", err.Error())
-	block.sendErr(err)
-	d.decoding <- block
-	return true
-}
-
 // checkCRC will check the checksum if the frame has one.
 // Will return ErrCRCMismatch if crc check failed, otherwise nil.
 func (d *frameDec) checkCRC() error {
 	if !d.HasCheckSum {
 		return nil
 	}
+
+	// We can overwrite upper tmp now
+	want, err := d.rawInput.readSmall(4)
+	if err != nil {
+		println("CRC missing?", err)
+		return err
+	}
+
+	if d.o.ignoreChecksum {
+		return nil
+	}
+
 	var tmp [4]byte
 	got := d.crc.Sum64()
 	// Flip to match file order.
@@ -306,142 +318,29 @@
 	tmp[2] = byte(got >> 16)
 	tmp[3] = byte(got >> 24)
 
-	// We can overwrite upper tmp now
-	want := d.rawInput.readSmall(4)
-	if want == nil {
-		println("CRC missing?")
-		return io.ErrUnexpectedEOF
-	}
-
 	if !bytes.Equal(tmp[:], want) {
-		if debug {
+		if debugDecoder {
 			println("CRC Check Failed:", tmp[:], "!=", want)
 		}
 		return ErrCRCMismatch
 	}
-	if debug {
+	if debugDecoder {
 		println("CRC ok", tmp[:])
 	}
 	return nil
 }
 
-func (d *frameDec) initAsync() {
-	if !d.o.lowMem && !d.SingleSegment {
-		// set max extra size history to 10MB.
-		d.history.maxSize = d.history.windowSize + maxBlockSize*5
+// consumeCRC reads the checksum data if the frame has one.
+func (d *frameDec) consumeCRC() error {
+	if d.HasCheckSum {
+		_, err := d.rawInput.readSmall(4)
+		if err != nil {
+			println("CRC missing?", err)
+			return err
+		}
 	}
-	// re-alloc if more than one extra block size.
-	if d.o.lowMem && cap(d.history.b) > d.history.maxSize+maxBlockSize {
-		d.history.b = make([]byte, 0, d.history.maxSize)
-	}
-	if cap(d.history.b) < d.history.maxSize {
-		d.history.b = make([]byte, 0, d.history.maxSize)
-	}
-	if cap(d.decoding) < d.o.concurrent {
-		d.decoding = make(chan *blockDec, d.o.concurrent)
-	}
-	if debug {
-		h := d.history
-		printf("history init. len: %d, cap: %d", len(h.b), cap(h.b))
-	}
-	d.asyncRunningMu.Lock()
-	d.asyncRunning = true
-	d.asyncRunningMu.Unlock()
-}
 
-// startDecoder will start decoding blocks and write them to the writer.
-// The decoder will stop as soon as an error occurs or at end of frame.
-// When the frame has finished decoding the *bufio.Reader
-// containing the remaining input will be sent on frameDec.frameDone.
-func (d *frameDec) startDecoder(output chan decodeOutput) {
-	written := int64(0)
-
-	defer func() {
-		d.asyncRunningMu.Lock()
-		d.asyncRunning = false
-		d.asyncRunningMu.Unlock()
-
-		// Drain the currently decoding.
-		d.history.error = true
-	flushdone:
-		for {
-			select {
-			case b := <-d.decoding:
-				b.history <- &d.history
-				output <- <-b.result
-			default:
-				break flushdone
-			}
-		}
-		println("frame decoder done, signalling done")
-		d.frameDone.Done()
-	}()
-	// Get decoder for first block.
-	block := <-d.decoding
-	block.history <- &d.history
-	for {
-		var next *blockDec
-		// Get result
-		r := <-block.result
-		if r.err != nil {
-			println("Result contained error", r.err)
-			output <- r
-			return
-		}
-		if debug {
-			println("got result, from ", d.offset, "to", d.offset+int64(len(r.b)))
-			d.offset += int64(len(r.b))
-		}
-		if !block.Last {
-			// Send history to next block
-			select {
-			case next = <-d.decoding:
-				if debug {
-					println("Sending ", len(d.history.b), "bytes as history")
-				}
-				next.history <- &d.history
-			default:
-				// Wait until we have sent the block, so
-				// other decoders can potentially get the decoder.
-				next = nil
-			}
-		}
-
-		// Add checksum, async to decoding.
-		if d.HasCheckSum {
-			n, err := d.crc.Write(r.b)
-			if err != nil {
-				r.err = err
-				if n != len(r.b) {
-					r.err = io.ErrShortWrite
-				}
-				output <- r
-				return
-			}
-		}
-		written += int64(len(r.b))
-		if d.SingleSegment && uint64(written) > d.FrameContentSize {
-			println("runDecoder: single segment and", uint64(written), ">", d.FrameContentSize)
-			r.err = ErrFrameSizeExceeded
-			output <- r
-			return
-		}
-		if block.Last {
-			r.err = d.checkCRC()
-			output <- r
-			return
-		}
-		output <- r
-		if next == nil {
-			// There was no decoder available, we wait for one now that we have sent to the writer.
-			if debug {
-				println("Sending ", len(d.history.b), " bytes as history")
-			}
-			next = <-d.decoding
-			next.history <- &d.history
-		}
-		block = next
-	}
+	return nil
 }
 
 // runDecoder will create a sync decoder that will decode a block of data.
@@ -450,41 +349,67 @@
 
 	// We use the history for output to avoid copying it.
 	d.history.b = dst
+	d.history.ignoreBuffer = len(dst)
 	// Store input length, so we only check new data.
 	crcStart := len(dst)
+	d.history.decoders.maxSyncLen = 0
+	if d.FrameContentSize != fcsUnknown {
+		d.history.decoders.maxSyncLen = d.FrameContentSize + uint64(len(dst))
+		if d.history.decoders.maxSyncLen > d.o.maxDecodedSize {
+			return dst, ErrDecoderSizeExceeded
+		}
+		if uint64(cap(dst)) < d.history.decoders.maxSyncLen {
+			// Alloc for output
+			dst2 := make([]byte, len(dst), d.history.decoders.maxSyncLen+compressedBlockOverAlloc)
+			copy(dst2, dst)
+			dst = dst2
+		}
+	}
 	var err error
 	for {
 		err = dec.reset(d.rawInput, d.WindowSize)
 		if err != nil {
 			break
 		}
-		if debug {
+		if debugDecoder {
 			println("next block:", dec)
 		}
 		err = dec.decodeBuf(&d.history)
-		if err != nil || dec.Last {
+		if err != nil {
 			break
 		}
 		if uint64(len(d.history.b)) > d.o.maxDecodedSize {
 			err = ErrDecoderSizeExceeded
 			break
 		}
-		if d.SingleSegment && uint64(len(d.history.b)) > d.o.maxDecodedSize {
-			println("runDecoder: single segment and", uint64(len(d.history.b)), ">", d.o.maxDecodedSize)
+		if uint64(len(d.history.b)-crcStart) > d.FrameContentSize {
+			println("runDecoder: FrameContentSize exceeded", uint64(len(d.history.b)-crcStart), ">", d.FrameContentSize)
 			err = ErrFrameSizeExceeded
 			break
 		}
+		if dec.Last {
+			break
+		}
+		if debugDecoder {
+			println("runDecoder: FrameContentSize", uint64(len(d.history.b)-crcStart), "<=", d.FrameContentSize)
+		}
 	}
 	dst = d.history.b
 	if err == nil {
-		if d.HasCheckSum {
-			var n int
-			n, err = d.crc.Write(dst[crcStart:])
-			if err == nil {
-				if n != len(dst)-crcStart {
-					err = io.ErrShortWrite
-				} else {
-					err = d.checkCRC()
+		if d.FrameContentSize != fcsUnknown && uint64(len(d.history.b)-crcStart) != d.FrameContentSize {
+			err = ErrFrameSizeMismatch
+		} else if d.HasCheckSum {
+			if d.o.ignoreChecksum {
+				err = d.consumeCRC()
+			} else {
+				var n int
+				n, err = d.crc.Write(dst[crcStart:])
+				if err == nil {
+					if n != len(dst)-crcStart {
+						err = io.ErrShortWrite
+					} else {
+						err = d.checkCRC()
+					}
 				}
 			}
 		}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go
index e6d3d49..2f8860a 100644
--- a/vendor/github.com/klauspost/compress/zstd/fse_decoder.go
+++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go
@@ -5,8 +5,10 @@
 package zstd
 
 import (
+	"encoding/binary"
 	"errors"
 	"fmt"
+	"io"
 )
 
 const (
@@ -178,10 +180,32 @@
 		return fmt.Errorf("corruption detected (total %d != %d)", gotTotal, 1<<s.actualTableLog)
 	}
 	b.advance((bitCount + 7) >> 3)
-	// println(s.norm[:s.symbolLen], s.symbolLen)
 	return s.buildDtable()
 }
 
+func (s *fseDecoder) mustReadFrom(r io.Reader) {
+	fatalErr := func(err error) {
+		if err != nil {
+			panic(err)
+		}
+	}
+	// 	dt             [maxTablesize]decSymbol // Decompression table.
+	//	symbolLen      uint16                  // Length of active part of the symbol table.
+	//	actualTableLog uint8                   // Selected tablelog.
+	//	maxBits        uint8                   // Maximum number of additional bits
+	//	// used for table creation to avoid allocations.
+	//	stateTable [256]uint16
+	//	norm       [maxSymbolValue + 1]int16
+	//	preDefined bool
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.dt))
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.symbolLen))
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.actualTableLog))
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.maxBits))
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.stateTable))
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.norm))
+	fatalErr(binary.Read(r, binary.LittleEndian, &s.preDefined))
+}
+
 // decSymbol contains information about a state entry,
 // Including the state offset base, the output symbol and
 // the number of bits to read for the low part of the destination state.
@@ -204,18 +228,10 @@
 	return uint16(d >> 16)
 }
 
-func (d decSymbol) baseline() uint32 {
-	return uint32(d >> 32)
-}
-
 func (d decSymbol) baselineInt() int {
 	return int(d >> 32)
 }
 
-func (d *decSymbol) set(nbits, addBits uint8, newState uint16, baseline uint32) {
-	*d = decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32)
-}
-
 func (d *decSymbol) setNBits(nBits uint8) {
 	const mask = 0xffffffffffffff00
 	*d = (*d & mask) | decSymbol(nBits)
@@ -231,11 +247,6 @@
 	*d = (*d & mask) | decSymbol(state)<<16
 }
 
-func (d *decSymbol) setBaseline(baseline uint32) {
-	const mask = 0xffffffff
-	*d = (*d & mask) | decSymbol(baseline)<<32
-}
-
 func (d *decSymbol) setExt(addBits uint8, baseline uint32) {
 	const mask = 0xffff00ff
 	*d = (*d & mask) | (decSymbol(addBits) << 8) | (decSymbol(baseline) << 32)
@@ -257,68 +268,6 @@
 	s.dt[0] = symbol
 }
 
-// buildDtable will build the decoding table.
-func (s *fseDecoder) buildDtable() error {
-	tableSize := uint32(1 << s.actualTableLog)
-	highThreshold := tableSize - 1
-	symbolNext := s.stateTable[:256]
-
-	// Init, lay down lowprob symbols
-	{
-		for i, v := range s.norm[:s.symbolLen] {
-			if v == -1 {
-				s.dt[highThreshold].setAddBits(uint8(i))
-				highThreshold--
-				symbolNext[i] = 1
-			} else {
-				symbolNext[i] = uint16(v)
-			}
-		}
-	}
-	// Spread symbols
-	{
-		tableMask := tableSize - 1
-		step := tableStep(tableSize)
-		position := uint32(0)
-		for ss, v := range s.norm[:s.symbolLen] {
-			for i := 0; i < int(v); i++ {
-				s.dt[position].setAddBits(uint8(ss))
-				position = (position + step) & tableMask
-				for position > highThreshold {
-					// lowprob area
-					position = (position + step) & tableMask
-				}
-			}
-		}
-		if position != 0 {
-			// position must reach all cells once, otherwise normalizedCounter is incorrect
-			return errors.New("corrupted input (position != 0)")
-		}
-	}
-
-	// Build Decoding table
-	{
-		tableSize := uint16(1 << s.actualTableLog)
-		for u, v := range s.dt[:tableSize] {
-			symbol := v.addBits()
-			nextState := symbolNext[symbol]
-			symbolNext[symbol] = nextState + 1
-			nBits := s.actualTableLog - byte(highBits(uint32(nextState)))
-			s.dt[u&maxTableMask].setNBits(nBits)
-			newState := (nextState << nBits) - tableSize
-			if newState > tableSize {
-				return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
-			}
-			if newState == uint16(u) && nBits == 0 {
-				// Seems weird that this is possible with nbits > 0.
-				return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
-			}
-			s.dt[u&maxTableMask].setNewState(newState)
-		}
-	}
-	return nil
-}
-
 // transform will transform the decoder table into a table usable for
 // decoding without having to apply the transformation while decoding.
 // The state will contain the base value and the number of bits to read.
@@ -352,34 +301,7 @@
 	s.state = dt[br.getBits(tableLog)]
 }
 
-// next returns the current symbol and sets the next state.
-// At least tablelog bits must be available in the bit reader.
-func (s *fseState) next(br *bitReader) {
-	lowBits := uint16(br.getBits(s.state.nbBits()))
-	s.state = s.dt[s.state.newState()+lowBits]
-}
-
-// finished returns true if all bits have been read from the bitstream
-// and the next state would require reading bits from the input.
-func (s *fseState) finished(br *bitReader) bool {
-	return br.finished() && s.state.nbBits() > 0
-}
-
-// final returns the current state symbol without decoding the next.
-func (s *fseState) final() (int, uint8) {
-	return s.state.baselineInt(), s.state.addBits()
-}
-
 // final returns the current state symbol without decoding the next.
 func (s decSymbol) final() (int, uint8) {
 	return s.baselineInt(), s.addBits()
 }
-
-// nextFast returns the next symbol and sets the next state.
-// This can only be used if no symbols are 0 bits.
-// At least tablelog bits must be available in the bit reader.
-func (s *fseState) nextFast(br *bitReader) (uint32, uint8) {
-	lowBits := uint16(br.getBitsFast(s.state.nbBits()))
-	s.state = s.dt[s.state.newState()+lowBits]
-	return s.state.baseline(), s.state.addBits()
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go
new file mode 100644
index 0000000..c881d28
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go
@@ -0,0 +1,64 @@
+//go:build amd64 && !appengine && !noasm && gc
+// +build amd64,!appengine,!noasm,gc
+
+package zstd
+
+import (
+	"fmt"
+)
+
+type buildDtableAsmContext struct {
+	// inputs
+	stateTable *uint16
+	norm       *int16
+	dt         *uint64
+
+	// outputs --- set by the procedure in the case of error;
+	// for interpretation please see the error handling part below
+	errParam1 uint64
+	errParam2 uint64
+}
+
+// buildDtable_asm is an x86 assembly implementation of fseDecoder.buildDtable.
+// Function returns non-zero exit code on error.
+// go:noescape
+func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int
+
+// please keep in sync with _generate/gen_fse.go
+const (
+	errorCorruptedNormalizedCounter = 1
+	errorNewStateTooBig             = 2
+	errorNewStateNoBits             = 3
+)
+
+// buildDtable will build the decoding table.
+func (s *fseDecoder) buildDtable() error {
+	ctx := buildDtableAsmContext{
+		stateTable: &s.stateTable[0],
+		norm:       &s.norm[0],
+		dt:         (*uint64)(&s.dt[0]),
+	}
+	code := buildDtable_asm(s, &ctx)
+
+	if code != 0 {
+		switch code {
+		case errorCorruptedNormalizedCounter:
+			position := ctx.errParam1
+			return fmt.Errorf("corrupted input (position=%d, expected 0)", position)
+
+		case errorNewStateTooBig:
+			newState := decSymbol(ctx.errParam1)
+			size := ctx.errParam2
+			return fmt.Errorf("newState (%d) outside table size (%d)", newState, size)
+
+		case errorNewStateNoBits:
+			newState := decSymbol(ctx.errParam1)
+			oldState := decSymbol(ctx.errParam2)
+			return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, oldState)
+
+		default:
+			return fmt.Errorf("buildDtable_asm returned unhandled nonzero code = %d", code)
+		}
+	}
+	return nil
+}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s
new file mode 100644
index 0000000..da32b44
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s
@@ -0,0 +1,127 @@
+// Code generated by command: go run gen_fse.go -out ../fse_decoder_amd64.s -pkg=zstd. DO NOT EDIT.
+
+//go:build !appengine && !noasm && gc && !noasm
+// +build !appengine,!noasm,gc,!noasm
+
+// func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int
+TEXT ·buildDtable_asm(SB), $0-24
+	MOVQ ctx+8(FP), CX
+	MOVQ s+0(FP), DI
+
+	// Load values
+	MOVBQZX 4098(DI), DX
+	XORQ    AX, AX
+	BTSQ    DX, AX
+	MOVQ    (CX), BX
+	MOVQ    16(CX), SI
+	LEAQ    -1(AX), R8
+	MOVQ    8(CX), CX
+	MOVWQZX 4096(DI), DI
+
+	// End load values
+	// Init, lay down lowprob symbols
+	XORQ R9, R9
+	JMP  init_main_loop_condition
+
+init_main_loop:
+	MOVWQSX (CX)(R9*2), R10
+	CMPW    R10, $-1
+	JNE     do_not_update_high_threshold
+	MOVB    R9, 1(SI)(R8*8)
+	DECQ    R8
+	MOVQ    $0x0000000000000001, R10
+
+do_not_update_high_threshold:
+	MOVW R10, (BX)(R9*2)
+	INCQ R9
+
+init_main_loop_condition:
+	CMPQ R9, DI
+	JL   init_main_loop
+
+	// Spread symbols
+	// Calculate table step
+	MOVQ AX, R9
+	SHRQ $0x01, R9
+	MOVQ AX, R10
+	SHRQ $0x03, R10
+	LEAQ 3(R9)(R10*1), R9
+
+	// Fill add bits values
+	LEAQ -1(AX), R10
+	XORQ R11, R11
+	XORQ R12, R12
+	JMP  spread_main_loop_condition
+
+spread_main_loop:
+	XORQ    R13, R13
+	MOVWQSX (CX)(R12*2), R14
+	JMP     spread_inner_loop_condition
+
+spread_inner_loop:
+	MOVB R12, 1(SI)(R11*8)
+
+adjust_position:
+	ADDQ R9, R11
+	ANDQ R10, R11
+	CMPQ R11, R8
+	JG   adjust_position
+	INCQ R13
+
+spread_inner_loop_condition:
+	CMPQ R13, R14
+	JL   spread_inner_loop
+	INCQ R12
+
+spread_main_loop_condition:
+	CMPQ  R12, DI
+	JL    spread_main_loop
+	TESTQ R11, R11
+	JZ    spread_check_ok
+	MOVQ  ctx+8(FP), AX
+	MOVQ  R11, 24(AX)
+	MOVQ  $+1, ret+16(FP)
+	RET
+
+spread_check_ok:
+	// Build Decoding table
+	XORQ DI, DI
+
+build_table_main_table:
+	MOVBQZX 1(SI)(DI*8), CX
+	MOVWQZX (BX)(CX*2), R8
+	LEAQ    1(R8), R9
+	MOVW    R9, (BX)(CX*2)
+	MOVQ    R8, R9
+	BSRQ    R9, R9
+	MOVQ    DX, CX
+	SUBQ    R9, CX
+	SHLQ    CL, R8
+	SUBQ    AX, R8
+	MOVB    CL, (SI)(DI*8)
+	MOVW    R8, 2(SI)(DI*8)
+	CMPQ    R8, AX
+	JLE     build_table_check1_ok
+	MOVQ    ctx+8(FP), CX
+	MOVQ    R8, 24(CX)
+	MOVQ    AX, 32(CX)
+	MOVQ    $+2, ret+16(FP)
+	RET
+
+build_table_check1_ok:
+	TESTB CL, CL
+	JNZ   build_table_check2_ok
+	CMPW  R8, DI
+	JNE   build_table_check2_ok
+	MOVQ  ctx+8(FP), AX
+	MOVQ  R8, 24(AX)
+	MOVQ  DI, 32(AX)
+	MOVQ  $+3, ret+16(FP)
+	RET
+
+build_table_check2_ok:
+	INCQ DI
+	CMPQ DI, AX
+	JL   build_table_main_table
+	MOVQ $+0, ret+16(FP)
+	RET
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go
new file mode 100644
index 0000000..332e51f
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go
@@ -0,0 +1,72 @@
+//go:build !amd64 || appengine || !gc || noasm
+// +build !amd64 appengine !gc noasm
+
+package zstd
+
+import (
+	"errors"
+	"fmt"
+)
+
+// buildDtable will build the decoding table.
+func (s *fseDecoder) buildDtable() error {
+	tableSize := uint32(1 << s.actualTableLog)
+	highThreshold := tableSize - 1
+	symbolNext := s.stateTable[:256]
+
+	// Init, lay down lowprob symbols
+	{
+		for i, v := range s.norm[:s.symbolLen] {
+			if v == -1 {
+				s.dt[highThreshold].setAddBits(uint8(i))
+				highThreshold--
+				symbolNext[i] = 1
+			} else {
+				symbolNext[i] = uint16(v)
+			}
+		}
+	}
+
+	// Spread symbols
+	{
+		tableMask := tableSize - 1
+		step := tableStep(tableSize)
+		position := uint32(0)
+		for ss, v := range s.norm[:s.symbolLen] {
+			for i := 0; i < int(v); i++ {
+				s.dt[position].setAddBits(uint8(ss))
+				position = (position + step) & tableMask
+				for position > highThreshold {
+					// lowprob area
+					position = (position + step) & tableMask
+				}
+			}
+		}
+		if position != 0 {
+			// position must reach all cells once, otherwise normalizedCounter is incorrect
+			return errors.New("corrupted input (position != 0)")
+		}
+	}
+
+	// Build Decoding table
+	{
+		tableSize := uint16(1 << s.actualTableLog)
+		for u, v := range s.dt[:tableSize] {
+			symbol := v.addBits()
+			nextState := symbolNext[symbol]
+			symbolNext[symbol] = nextState + 1
+			nBits := s.actualTableLog - byte(highBits(uint32(nextState)))
+			s.dt[u&maxTableMask].setNBits(nBits)
+			newState := (nextState << nBits) - tableSize
+			if newState > tableSize {
+				return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
+			}
+			if newState == uint16(u) && nBits == 0 {
+				// Seems weird that this is possible with nbits > 0.
+				return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
+			}
+			s.dt[u&maxTableMask].setNewState(newState)
+		}
+	}
+	return nil
+}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_encoder.go b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go
index c74681b..ab26326 100644
--- a/vendor/github.com/klauspost/compress/zstd/fse_encoder.go
+++ b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go
@@ -62,9 +62,8 @@
 // To indicate that you have populated the histogram call HistogramFinished
 // with the value of the highest populated symbol, as well as the number of entries
 // in the most populated entry. These are accepted at face value.
-// The returned slice will always be length 256.
-func (s *fseEncoder) Histogram() []uint32 {
-	return s.count[:]
+func (s *fseEncoder) Histogram() *[256]uint32 {
+	return &s.count
 }
 
 // HistogramFinished can be called to indicate that the histogram has been populated.
@@ -77,21 +76,6 @@
 	s.clearCount = maxCount != 0
 }
 
-// prepare will prepare and allocate scratch tables used for both compression and decompression.
-func (s *fseEncoder) prepare() (*fseEncoder, error) {
-	if s == nil {
-		s = &fseEncoder{}
-	}
-	s.useRLE = false
-	if s.clearCount && s.maxCount == 0 {
-		for i := range s.count {
-			s.count[i] = 0
-		}
-		s.clearCount = false
-	}
-	return s, nil
-}
-
 // allocCtable will allocate tables needed for compression.
 // If existing tables a re big enough, they are simply re-used.
 func (s *fseEncoder) allocCtable() {
@@ -229,7 +213,7 @@
 		deltaFindState: 0,
 		deltaNbBits:    0,
 	}
-	if debug {
+	if debugEncoder {
 		println("setRLE: val", val, "symbolTT", s.ct.symbolTT[val])
 	}
 	s.rleVal = val
@@ -710,14 +694,6 @@
 	c.state = c.stateTable[lu]
 }
 
-// encode the output symbol provided and write it to the bitstream.
-func (c *cState) encode(symbolTT symbolTransform) {
-	nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16
-	dstState := int32(c.state>>(nbBitsOut&15)) + int32(symbolTT.deltaFindState)
-	c.bw.addBits16NC(c.state, uint8(nbBitsOut))
-	c.state = c.stateTable[dstState]
-}
-
 // flush will write the tablelog to the output and flush the remaining full bytes.
 func (c *cState) flush(tableLog uint8) {
 	c.bw.flush32()
diff --git a/vendor/github.com/klauspost/compress/zstd/hash.go b/vendor/github.com/klauspost/compress/zstd/hash.go
index 4a75206..5d73c21 100644
--- a/vendor/github.com/klauspost/compress/zstd/hash.go
+++ b/vendor/github.com/klauspost/compress/zstd/hash.go
@@ -13,65 +13,23 @@
 	prime8bytes = 0xcf1bbcdcb7a56463
 )
 
-// hashLen returns a hash of the lowest l bytes of u for a size size of h bytes.
-// l must be >=4 and <=8. Any other value will return hash for 4 bytes.
-// h should always be <32.
-// Preferably h and l should be a constant.
-// FIXME: This does NOT get resolved, if 'mls' is constant,
-//  so this cannot be used.
-func hashLen(u uint64, hashLog, mls uint8) uint32 {
+// hashLen returns a hash of the lowest mls bytes of with length output bits.
+// mls must be >=3 and <=8. Any other value will return hash for 4 bytes.
+// length should always be < 32.
+// Preferably length and mls should be a constant for inlining.
+func hashLen(u uint64, length, mls uint8) uint32 {
 	switch mls {
+	case 3:
+		return (uint32(u<<8) * prime3bytes) >> (32 - length)
 	case 5:
-		return hash5(u, hashLog)
+		return uint32(((u << (64 - 40)) * prime5bytes) >> (64 - length))
 	case 6:
-		return hash6(u, hashLog)
+		return uint32(((u << (64 - 48)) * prime6bytes) >> (64 - length))
 	case 7:
-		return hash7(u, hashLog)
+		return uint32(((u << (64 - 56)) * prime7bytes) >> (64 - length))
 	case 8:
-		return hash8(u, hashLog)
+		return uint32((u * prime8bytes) >> (64 - length))
 	default:
-		return hash4x64(u, hashLog)
+		return (uint32(u) * prime4bytes) >> (32 - length)
 	}
 }
-
-// hash3 returns the hash of the lower 3 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash3(u uint32, h uint8) uint32 {
-	return ((u << (32 - 24)) * prime3bytes) >> ((32 - h) & 31)
-}
-
-// hash4 returns the hash of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash4(u uint32, h uint8) uint32 {
-	return (u * prime4bytes) >> ((32 - h) & 31)
-}
-
-// hash4x64 returns the hash of the lowest 4 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash4x64(u uint64, h uint8) uint32 {
-	return (uint32(u) * prime4bytes) >> ((32 - h) & 31)
-}
-
-// hash5 returns the hash of the lowest 5 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash5(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 40)) * prime5bytes) >> ((64 - h) & 63))
-}
-
-// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash6(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & 63))
-}
-
-// hash7 returns the hash of the lowest 7 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash7(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & 63))
-}
-
-// hash8 returns the hash of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash8(u uint64, h uint8) uint32 {
-	return uint32((u * prime8bytes) >> ((64 - h) & 63))
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/history.go b/vendor/github.com/klauspost/compress/zstd/history.go
index f783e32..28b4015 100644
--- a/vendor/github.com/klauspost/compress/zstd/history.go
+++ b/vendor/github.com/klauspost/compress/zstd/history.go
@@ -10,20 +10,31 @@
 
 // history contains the information transferred between blocks.
 type history struct {
-	b             []byte
-	huffTree      *huff0.Scratch
-	recentOffsets [3]int
+	// Literal decompression
+	huffTree *huff0.Scratch
+
+	// Sequence decompression
 	decoders      sequenceDecs
-	windowSize    int
-	maxSize       int
-	error         bool
-	dict          *dict
+	recentOffsets [3]int
+
+	// History buffer...
+	b []byte
+
+	// ignoreBuffer is meant to ignore a number of bytes
+	// when checking for matches in history
+	ignoreBuffer int
+
+	windowSize       int
+	allocFrameBuffer int // needed?
+	error            bool
+	dict             *dict
 }
 
 // reset will reset the history to initial state of a frame.
 // The history must already have been initialized to the desired size.
 func (h *history) reset() {
 	h.b = h.b[:0]
+	h.ignoreBuffer = 0
 	h.error = false
 	h.recentOffsets = [3]int{1, 4, 8}
 	if f := h.decoders.litLengths.fse; f != nil && !f.preDefined {
@@ -35,7 +46,7 @@
 	if f := h.decoders.matchLengths.fse; f != nil && !f.preDefined {
 		fseDecoderPool.Put(f)
 	}
-	h.decoders = sequenceDecs{}
+	h.decoders = sequenceDecs{br: h.decoders.br}
 	if h.huffTree != nil {
 		if h.dict == nil || h.dict.litEnc != h.huffTree {
 			huffDecoderPool.Put(h.huffTree)
@@ -54,6 +65,7 @@
 	h.decoders.litLengths = dict.llDec
 	h.decoders.offsets = dict.ofDec
 	h.decoders.matchLengths = dict.mlDec
+	h.decoders.dict = dict.content
 	h.recentOffsets = dict.offsets
 	h.huffTree = dict.litEnc
 }
@@ -83,6 +95,24 @@
 	copy(h.b[h.windowSize-len(b):], b)
 }
 
+// ensureBlock will ensure there is space for at least one block...
+func (h *history) ensureBlock() {
+	if cap(h.b) < h.allocFrameBuffer {
+		h.b = make([]byte, 0, h.allocFrameBuffer)
+		return
+	}
+
+	avail := cap(h.b) - len(h.b)
+	if avail >= h.windowSize || avail > maxCompressedBlockSize {
+		return
+	}
+	// Move data down so we only have window size left.
+	// We know we have less than window size in b at this point.
+	discard := len(h.b) - h.windowSize
+	copy(h.b, h.b[discard:])
+	h.b = h.b[:h.windowSize]
+}
+
 // append bytes to history without ever discarding anything.
 func (h *history) appendKeep(b []byte) {
 	h.b = append(h.b, b...)
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go
index 426b9ca..2c112a0 100644
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go
+++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go
@@ -195,7 +195,6 @@
 	b, d.v4 = consumeUint64(b)
 	b, d.total = consumeUint64(b)
 	copy(d.mem[:], b)
-	b = b[len(d.mem):]
 	d.n = int(d.total % uint64(len(d.mem)))
 	return nil
 }
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go
deleted file mode 100644
index 35318d7..0000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go
+++ /dev/null
@@ -1,13 +0,0 @@
-// +build !appengine
-// +build gc
-// +build !purego
-
-package xxhash
-
-// Sum64 computes the 64-bit xxHash digest of b.
-//
-//go:noescape
-func Sum64(b []byte) uint64
-
-//go:noescape
-func writeBlocks(*Digest, []byte) int
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s
index 2c9c535..cea1785 100644
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s
+++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s
@@ -1,12 +1,13 @@
 // +build !appengine
 // +build gc
 // +build !purego
+// +build !noasm
 
 #include "textflag.h"
 
 // Register allocation:
 // AX	h
-// CX	pointer to advance through b
+// SI	pointer to advance through b
 // DX	n
 // BX	loop end
 // R8	v1, k1
@@ -16,39 +17,39 @@
 // R12	tmp
 // R13	prime1v
 // R14	prime2v
-// R15	prime4v
+// DI	prime4v
 
-// round reads from and advances the buffer pointer in CX.
+// round reads from and advances the buffer pointer in SI.
 // It assumes that R13 has prime1v and R14 has prime2v.
 #define round(r) \
-	MOVQ  (CX), R12 \
-	ADDQ  $8, CX    \
+	MOVQ  (SI), R12 \
+	ADDQ  $8, SI    \
 	IMULQ R14, R12  \
 	ADDQ  R12, r    \
 	ROLQ  $31, r    \
 	IMULQ R13, r
 
 // mergeRound applies a merge round on the two registers acc and val.
-// It assumes that R13 has prime1v, R14 has prime2v, and R15 has prime4v.
+// It assumes that R13 has prime1v, R14 has prime2v, and DI has prime4v.
 #define mergeRound(acc, val) \
 	IMULQ R14, val \
 	ROLQ  $31, val \
 	IMULQ R13, val \
 	XORQ  val, acc \
 	IMULQ R13, acc \
-	ADDQ  R15, acc
+	ADDQ  DI, acc
 
 // func Sum64(b []byte) uint64
 TEXT ·Sum64(SB), NOSPLIT, $0-32
 	// Load fixed primes.
 	MOVQ ·prime1v(SB), R13
 	MOVQ ·prime2v(SB), R14
-	MOVQ ·prime4v(SB), R15
+	MOVQ ·prime4v(SB), DI
 
 	// Load slice.
-	MOVQ b_base+0(FP), CX
+	MOVQ b_base+0(FP), SI
 	MOVQ b_len+8(FP), DX
-	LEAQ (CX)(DX*1), BX
+	LEAQ (SI)(DX*1), BX
 
 	// The first loop limit will be len(b)-32.
 	SUBQ $32, BX
@@ -65,14 +66,14 @@
 	XORQ R11, R11
 	SUBQ R13, R11
 
-	// Loop until CX > BX.
+	// Loop until SI > BX.
 blockLoop:
 	round(R8)
 	round(R9)
 	round(R10)
 	round(R11)
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JLE  blockLoop
 
 	MOVQ R8, AX
@@ -100,16 +101,16 @@
 afterBlocks:
 	ADDQ DX, AX
 
-	// Right now BX has len(b)-32, and we want to loop until CX > len(b)-8.
+	// Right now BX has len(b)-32, and we want to loop until SI > len(b)-8.
 	ADDQ $24, BX
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JG   fourByte
 
 wordLoop:
 	// Calculate k1.
-	MOVQ  (CX), R8
-	ADDQ  $8, CX
+	MOVQ  (SI), R8
+	ADDQ  $8, SI
 	IMULQ R14, R8
 	ROLQ  $31, R8
 	IMULQ R13, R8
@@ -117,18 +118,18 @@
 	XORQ  R8, AX
 	ROLQ  $27, AX
 	IMULQ R13, AX
-	ADDQ  R15, AX
+	ADDQ  DI, AX
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JLE  wordLoop
 
 fourByte:
 	ADDQ $4, BX
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JG   singles
 
-	MOVL  (CX), R8
-	ADDQ  $4, CX
+	MOVL  (SI), R8
+	ADDQ  $4, SI
 	IMULQ R13, R8
 	XORQ  R8, AX
 
@@ -138,19 +139,19 @@
 
 singles:
 	ADDQ $4, BX
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JGE  finalize
 
 singlesLoop:
-	MOVBQZX (CX), R12
-	ADDQ    $1, CX
+	MOVBQZX (SI), R12
+	ADDQ    $1, SI
 	IMULQ   ·prime5v(SB), R12
 	XORQ    R12, AX
 
 	ROLQ  $11, AX
 	IMULQ R13, AX
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JL   singlesLoop
 
 finalize:
@@ -179,13 +180,13 @@
 	MOVQ ·prime2v(SB), R14
 
 	// Load slice.
-	MOVQ arg1_base+8(FP), CX
-	MOVQ arg1_len+16(FP), DX
-	LEAQ (CX)(DX*1), BX
+	MOVQ b_base+8(FP), SI
+	MOVQ b_len+16(FP), DX
+	LEAQ (SI)(DX*1), BX
 	SUBQ $32, BX
 
 	// Load vN from d.
-	MOVQ arg+0(FP), AX
+	MOVQ d+0(FP), AX
 	MOVQ 0(AX), R8   // v1
 	MOVQ 8(AX), R9   // v2
 	MOVQ 16(AX), R10 // v3
@@ -199,7 +200,7 @@
 	round(R10)
 	round(R11)
 
-	CMPQ CX, BX
+	CMPQ SI, BX
 	JLE  blockLoop
 
 	// Copy vN back to d.
@@ -208,8 +209,8 @@
 	MOVQ R10, 16(AX)
 	MOVQ R11, 24(AX)
 
-	// The number of bytes written is CX minus the old base pointer.
-	SUBQ arg1_base+8(FP), CX
-	MOVQ CX, ret+32(FP)
+	// The number of bytes written is SI minus the old base pointer.
+	SUBQ b_base+8(FP), SI
+	MOVQ SI, ret+32(FP)
 
 	RET
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_arm64.s b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_arm64.s
new file mode 100644
index 0000000..4d64a17
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_arm64.s
@@ -0,0 +1,186 @@
+// +build gc,!purego,!noasm
+
+#include "textflag.h"
+
+// Register allocation.
+#define digest	R1
+#define h	R2 // Return value.
+#define p	R3 // Input pointer.
+#define len	R4
+#define nblocks	R5 // len / 32.
+#define prime1	R7
+#define prime2	R8
+#define prime3	R9
+#define prime4	R10
+#define prime5	R11
+#define v1	R12
+#define v2	R13
+#define v3	R14
+#define v4	R15
+#define x1	R20
+#define x2	R21
+#define x3	R22
+#define x4	R23
+
+#define round(acc, x) \
+	MADD prime2, acc, x, acc \
+	ROR  $64-31, acc         \
+	MUL  prime1, acc         \
+
+// x = round(0, x).
+#define round0(x) \
+	MUL prime2, x \
+	ROR $64-31, x \
+	MUL prime1, x \
+
+#define mergeRound(x) \
+	round0(x)                 \
+	EOR  x, h                 \
+	MADD h, prime4, prime1, h \
+
+// Update v[1-4] with 32-byte blocks. Assumes len >= 32.
+#define blocksLoop() \
+	LSR     $5, len, nblocks \
+	PCALIGN $16              \
+	loop:                    \
+	LDP.P   32(p), (x1, x2)  \
+	round(v1, x1)            \
+	LDP     -16(p), (x3, x4) \
+	round(v2, x2)            \
+	SUB     $1, nblocks      \
+	round(v3, x3)            \
+	round(v4, x4)            \
+	CBNZ    nblocks, loop    \
+
+// The primes are repeated here to ensure that they're stored
+// in a contiguous array, so we can load them with LDP.
+DATA primes<> +0(SB)/8, $11400714785074694791
+DATA primes<> +8(SB)/8, $14029467366897019727
+DATA primes<>+16(SB)/8, $1609587929392839161
+DATA primes<>+24(SB)/8, $9650029242287828579
+DATA primes<>+32(SB)/8, $2870177450012600261
+GLOBL primes<>(SB), NOPTR+RODATA, $40
+
+// func Sum64(b []byte) uint64
+TEXT ·Sum64(SB), NOFRAME+NOSPLIT, $0-32
+	LDP b_base+0(FP), (p, len)
+
+	LDP  primes<> +0(SB), (prime1, prime2)
+	LDP  primes<>+16(SB), (prime3, prime4)
+	MOVD primes<>+32(SB), prime5
+
+	CMP  $32, len
+	CSEL LO, prime5, ZR, h // if len < 32 { h = prime5 } else { h = 0 }
+	BLO  afterLoop
+
+	ADD  prime1, prime2, v1
+	MOVD prime2, v2
+	MOVD $0, v3
+	NEG  prime1, v4
+
+	blocksLoop()
+
+	ROR $64-1, v1, x1
+	ROR $64-7, v2, x2
+	ADD x1, x2
+	ROR $64-12, v3, x3
+	ROR $64-18, v4, x4
+	ADD x3, x4
+	ADD x2, x4, h
+
+	mergeRound(v1)
+	mergeRound(v2)
+	mergeRound(v3)
+	mergeRound(v4)
+
+afterLoop:
+	ADD len, h
+
+	TBZ   $4, len, try8
+	LDP.P 16(p), (x1, x2)
+
+	round0(x1)
+	ROR  $64-27, h
+	EOR  x1 @> 64-27, h, h
+	MADD h, prime4, prime1, h
+
+	round0(x2)
+	ROR  $64-27, h
+	EOR  x2 @> 64-27, h
+	MADD h, prime4, prime1, h
+
+try8:
+	TBZ    $3, len, try4
+	MOVD.P 8(p), x1
+
+	round0(x1)
+	ROR  $64-27, h
+	EOR  x1 @> 64-27, h
+	MADD h, prime4, prime1, h
+
+try4:
+	TBZ     $2, len, try2
+	MOVWU.P 4(p), x2
+
+	MUL  prime1, x2
+	ROR  $64-23, h
+	EOR  x2 @> 64-23, h
+	MADD h, prime3, prime2, h
+
+try2:
+	TBZ     $1, len, try1
+	MOVHU.P 2(p), x3
+	AND     $255, x3, x1
+	LSR     $8, x3, x2
+
+	MUL prime5, x1
+	ROR $64-11, h
+	EOR x1 @> 64-11, h
+	MUL prime1, h
+
+	MUL prime5, x2
+	ROR $64-11, h
+	EOR x2 @> 64-11, h
+	MUL prime1, h
+
+try1:
+	TBZ   $0, len, end
+	MOVBU (p), x4
+
+	MUL prime5, x4
+	ROR $64-11, h
+	EOR x4 @> 64-11, h
+	MUL prime1, h
+
+end:
+	EOR h >> 33, h
+	MUL prime2, h
+	EOR h >> 29, h
+	MUL prime3, h
+	EOR h >> 32, h
+
+	MOVD h, ret+24(FP)
+	RET
+
+// func writeBlocks(d *Digest, b []byte) int
+//
+// Assumes len(b) >= 32.
+TEXT ·writeBlocks(SB), NOFRAME+NOSPLIT, $0-40
+	LDP primes<>(SB), (prime1, prime2)
+
+	// Load state. Assume v[1-4] are stored contiguously.
+	MOVD d+0(FP), digest
+	LDP  0(digest), (v1, v2)
+	LDP  16(digest), (v3, v4)
+
+	LDP b_base+8(FP), (p, len)
+
+	blocksLoop()
+
+	// Store updated state.
+	STP (v1, v2), 0(digest)
+	STP (v3, v4), 16(digest)
+
+	BIC  $31, len
+	MOVD len, ret+32(FP)
+	RET
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_asm.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_asm.go
new file mode 100644
index 0000000..1a1fac9
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_asm.go
@@ -0,0 +1,16 @@
+//go:build (amd64 || arm64) && !appengine && gc && !purego && !noasm
+// +build amd64 arm64
+// +build !appengine
+// +build gc
+// +build !purego
+// +build !noasm
+
+package xxhash
+
+// Sum64 computes the 64-bit xxHash digest of b.
+//
+//go:noescape
+func Sum64(b []byte) uint64
+
+//go:noescape
+func writeBlocks(d *Digest, b []byte) int
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go
index 4a5a821..209cb4a 100644
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go
+++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go
@@ -1,4 +1,5 @@
-// +build !amd64 appengine !gc purego
+//go:build (!amd64 && !arm64) || appengine || !gc || purego || noasm
+// +build !amd64,!arm64 appengine !gc purego noasm
 
 package xxhash
 
diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec.go b/vendor/github.com/klauspost/compress/zstd/seqdec.go
index 1dd39e6..df04472 100644
--- a/vendor/github.com/klauspost/compress/zstd/seqdec.go
+++ b/vendor/github.com/klauspost/compress/zstd/seqdec.go
@@ -20,6 +20,10 @@
 	llCode, mlCode, ofCode uint8
 }
 
+type seqVals struct {
+	ll, ml, mo int
+}
+
 func (s seq) String() string {
 	if s.offset <= 3 {
 		if s.offset == 0 {
@@ -61,16 +65,19 @@
 	offsets      sequenceDec
 	matchLengths sequenceDec
 	prevOffset   [3]int
-	hist         []byte
 	dict         []byte
 	literals     []byte
 	out          []byte
+	nSeqs        int
+	br           *bitReader
+	seqSize      int
 	windowSize   int
 	maxBits      uint8
+	maxSyncLen   uint64
 }
 
 // initialize all 3 decoders from the stream input.
-func (s *sequenceDecs) initialize(br *bitReader, hist *history, literals, out []byte) error {
+func (s *sequenceDecs) initialize(br *bitReader, hist *history, out []byte) error {
 	if err := s.litLengths.init(br); err != nil {
 		return errors.New("litLengths:" + err.Error())
 	}
@@ -80,8 +87,7 @@
 	if err := s.matchLengths.init(br); err != nil {
 		return errors.New("matchLengths:" + err.Error())
 	}
-	s.literals = literals
-	s.hist = hist.b
+	s.br = br
 	s.prevOffset = hist.recentOffsets
 	s.maxBits = s.litLengths.fse.maxBits + s.offsets.fse.maxBits + s.matchLengths.fse.maxBits
 	s.windowSize = hist.windowSize
@@ -93,12 +99,127 @@
 	return nil
 }
 
+// execute will execute the decoded sequence with the provided history.
+// The sequence must be evaluated before being sent.
+func (s *sequenceDecs) execute(seqs []seqVals, hist []byte) error {
+	if len(s.dict) == 0 {
+		return s.executeSimple(seqs, hist)
+	}
+
+	// Ensure we have enough output size...
+	if len(s.out)+s.seqSize > cap(s.out) {
+		addBytes := s.seqSize + len(s.out)
+		s.out = append(s.out, make([]byte, addBytes)...)
+		s.out = s.out[:len(s.out)-addBytes]
+	}
+
+	if debugDecoder {
+		printf("Execute %d seqs with hist %d, dict %d, literals: %d into %d bytes\n", len(seqs), len(hist), len(s.dict), len(s.literals), s.seqSize)
+	}
+
+	var t = len(s.out)
+	out := s.out[:t+s.seqSize]
+
+	for _, seq := range seqs {
+		// Add literals
+		copy(out[t:], s.literals[:seq.ll])
+		t += seq.ll
+		s.literals = s.literals[seq.ll:]
+
+		// Copy from dictionary...
+		if seq.mo > t+len(hist) || seq.mo > s.windowSize {
+			if len(s.dict) == 0 {
+				return fmt.Errorf("match offset (%d) bigger than current history (%d)", seq.mo, t+len(hist))
+			}
+
+			// we may be in dictionary.
+			dictO := len(s.dict) - (seq.mo - (t + len(hist)))
+			if dictO < 0 || dictO >= len(s.dict) {
+				return fmt.Errorf("match offset (%d) bigger than current history+dict (%d)", seq.mo, t+len(hist)+len(s.dict))
+			}
+			end := dictO + seq.ml
+			if end > len(s.dict) {
+				n := len(s.dict) - dictO
+				copy(out[t:], s.dict[dictO:])
+				t += n
+				seq.ml -= n
+			} else {
+				copy(out[t:], s.dict[dictO:end])
+				t += end - dictO
+				continue
+			}
+		}
+
+		// Copy from history.
+		if v := seq.mo - t; v > 0 {
+			// v is the start position in history from end.
+			start := len(hist) - v
+			if seq.ml > v {
+				// Some goes into current block.
+				// Copy remainder of history
+				copy(out[t:], hist[start:])
+				t += v
+				seq.ml -= v
+			} else {
+				copy(out[t:], hist[start:start+seq.ml])
+				t += seq.ml
+				continue
+			}
+		}
+		// We must be in current buffer now
+		if seq.ml > 0 {
+			start := t - seq.mo
+			if seq.ml <= t-start {
+				// No overlap
+				copy(out[t:], out[start:start+seq.ml])
+				t += seq.ml
+				continue
+			} else {
+				// Overlapping copy
+				// Extend destination slice and copy one byte at the time.
+				src := out[start : start+seq.ml]
+				dst := out[t:]
+				dst = dst[:len(src)]
+				t += len(src)
+				// Destination is the space we just added.
+				for i := range src {
+					dst[i] = src[i]
+				}
+			}
+		}
+	}
+
+	// Add final literals
+	copy(out[t:], s.literals)
+	if debugDecoder {
+		t += len(s.literals)
+		if t != len(out) {
+			panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize))
+		}
+	}
+	s.out = out
+
+	return nil
+}
+
 // decode sequences from the stream with the provided history.
-func (s *sequenceDecs) decode(seqs int, br *bitReader, hist []byte) error {
+func (s *sequenceDecs) decodeSync(hist []byte) error {
+	supported, err := s.decodeSyncSimple(hist)
+	if supported {
+		return err
+	}
+
+	br := s.br
+	seqs := s.nSeqs
 	startSize := len(s.out)
 	// Grab full sizes tables, to avoid bounds checks.
 	llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize]
 	llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state
+	out := s.out
+	maxBlockSize := maxCompressedBlockSize
+	if s.windowSize < maxBlockSize {
+		maxBlockSize = s.windowSize
+	}
 
 	for i := seqs - 1; i >= 0; i-- {
 		if br.overread() {
@@ -151,7 +272,7 @@
 
 					if temp == 0 {
 						// 0 is not valid; input is corrupted; force offset to 1
-						println("temp was 0")
+						println("WARNING: temp was 0")
 						temp = 1
 					}
 
@@ -176,51 +297,49 @@
 		if ll > len(s.literals) {
 			return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, len(s.literals))
 		}
-		size := ll + ml + len(s.out)
+		size := ll + ml + len(out)
 		if size-startSize > maxBlockSize {
-			return fmt.Errorf("output (%d) bigger than max block size", size)
+			return fmt.Errorf("output (%d) bigger than max block size (%d)", size-startSize, maxBlockSize)
 		}
-		if size > cap(s.out) {
+		if size > cap(out) {
 			// Not enough size, which can happen under high volume block streaming conditions
 			// but could be if destination slice is too small for sync operations.
 			// over-allocating here can create a large amount of GC pressure so we try to keep
 			// it as contained as possible
-			used := len(s.out) - startSize
+			used := len(out) - startSize
 			addBytes := 256 + ll + ml + used>>2
 			// Clamp to max block size.
 			if used+addBytes > maxBlockSize {
 				addBytes = maxBlockSize - used
 			}
-			s.out = append(s.out, make([]byte, addBytes)...)
-			s.out = s.out[:len(s.out)-addBytes]
+			out = append(out, make([]byte, addBytes)...)
+			out = out[:len(out)-addBytes]
 		}
 		if ml > maxMatchLen {
 			return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
 		}
 
 		// Add literals
-		s.out = append(s.out, s.literals[:ll]...)
+		out = append(out, s.literals[:ll]...)
 		s.literals = s.literals[ll:]
-		out := s.out
 
 		if mo == 0 && ml > 0 {
 			return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
 		}
 
-		if mo > len(s.out)+len(hist) || mo > s.windowSize {
+		if mo > len(out)+len(hist) || mo > s.windowSize {
 			if len(s.dict) == 0 {
-				return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(s.out)+len(hist))
+				return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist)-startSize)
 			}
 
 			// we may be in dictionary.
-			dictO := len(s.dict) - (mo - (len(s.out) + len(hist)))
+			dictO := len(s.dict) - (mo - (len(out) + len(hist)))
 			if dictO < 0 || dictO >= len(s.dict) {
-				return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(s.out)+len(hist))
+				return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist)-startSize)
 			}
 			end := dictO + ml
 			if end > len(s.dict) {
 				out = append(out, s.dict[dictO:]...)
-				mo -= len(s.dict) - dictO
 				ml -= len(s.dict) - dictO
 			} else {
 				out = append(out, s.dict[dictO:end]...)
@@ -231,26 +350,25 @@
 
 		// Copy from history.
 		// TODO: Blocks without history could be made to ignore this completely.
-		if v := mo - len(s.out); v > 0 {
+		if v := mo - len(out); v > 0 {
 			// v is the start position in history from end.
-			start := len(s.hist) - v
+			start := len(hist) - v
 			if ml > v {
 				// Some goes into current block.
 				// Copy remainder of history
-				out = append(out, s.hist[start:]...)
-				mo -= v
+				out = append(out, hist[start:]...)
 				ml -= v
 			} else {
-				out = append(out, s.hist[start:start+ml]...)
+				out = append(out, hist[start:start+ml]...)
 				ml = 0
 			}
 		}
 		// We must be in current buffer now
 		if ml > 0 {
-			start := len(s.out) - mo
-			if ml <= len(s.out)-start {
+			start := len(out) - mo
+			if ml <= len(out)-start {
 				// No overlap
-				out = append(out, s.out[start:start+ml]...)
+				out = append(out, out[start:start+ml]...)
 			} else {
 				// Overlapping copy
 				// Extend destination slice and copy one byte at the time.
@@ -264,7 +382,6 @@
 				}
 			}
 		}
-		s.out = out
 		if i == 0 {
 			// This is the last sequence, so we shouldn't update state.
 			break
@@ -278,7 +395,8 @@
 			mlState = mlTable[mlState.newState()&maxTableMask]
 			ofState = ofTable[ofState.newState()&maxTableMask]
 		} else {
-			bits := br.getBitsFast(nBits)
+			bits := br.get32BitsFast(nBits)
+
 			lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31))
 			llState = llTable[(llState.newState()+lowBits)&maxTableMask]
 
@@ -291,19 +409,14 @@
 		}
 	}
 
-	// Add final literals
-	s.out = append(s.out, s.literals...)
-	return nil
-}
+	// Check if space for literals
+	if size := len(s.literals) + len(s.out) - startSize; size > maxBlockSize {
+		return fmt.Errorf("output (%d) bigger than max block size (%d)", size, maxBlockSize)
+	}
 
-// update states, at least 27 bits must be available.
-func (s *sequenceDecs) update(br *bitReader) {
-	// Max 8 bits
-	s.litLengths.state.next(br)
-	// Max 9 bits
-	s.matchLengths.state.next(br)
-	// Max 8 bits
-	s.offsets.state.next(br)
+	// Add final literals
+	s.out = append(out, s.literals...)
+	return br.close()
 }
 
 var bitMask [16]uint16
@@ -314,87 +427,6 @@
 	}
 }
 
-// update states, at least 27 bits must be available.
-func (s *sequenceDecs) updateAlt(br *bitReader) {
-	// Update all 3 states at once. Approx 20% faster.
-	a, b, c := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state
-
-	nBits := a.nbBits() + b.nbBits() + c.nbBits()
-	if nBits == 0 {
-		s.litLengths.state.state = s.litLengths.state.dt[a.newState()]
-		s.matchLengths.state.state = s.matchLengths.state.dt[b.newState()]
-		s.offsets.state.state = s.offsets.state.dt[c.newState()]
-		return
-	}
-	bits := br.getBitsFast(nBits)
-	lowBits := uint16(bits >> ((c.nbBits() + b.nbBits()) & 31))
-	s.litLengths.state.state = s.litLengths.state.dt[a.newState()+lowBits]
-
-	lowBits = uint16(bits >> (c.nbBits() & 31))
-	lowBits &= bitMask[b.nbBits()&15]
-	s.matchLengths.state.state = s.matchLengths.state.dt[b.newState()+lowBits]
-
-	lowBits = uint16(bits) & bitMask[c.nbBits()&15]
-	s.offsets.state.state = s.offsets.state.dt[c.newState()+lowBits]
-}
-
-// nextFast will return new states when there are at least 4 unused bytes left on the stream when done.
-func (s *sequenceDecs) nextFast(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) {
-	// Final will not read from stream.
-	ll, llB := llState.final()
-	ml, mlB := mlState.final()
-	mo, moB := ofState.final()
-
-	// extra bits are stored in reverse order.
-	br.fillFast()
-	mo += br.getBits(moB)
-	if s.maxBits > 32 {
-		br.fillFast()
-	}
-	ml += br.getBits(mlB)
-	ll += br.getBits(llB)
-
-	if moB > 1 {
-		s.prevOffset[2] = s.prevOffset[1]
-		s.prevOffset[1] = s.prevOffset[0]
-		s.prevOffset[0] = mo
-		return
-	}
-	// mo = s.adjustOffset(mo, ll, moB)
-	// Inlined for rather big speedup
-	if ll == 0 {
-		// There is an exception though, when current sequence's literals_length = 0.
-		// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
-		// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
-		mo++
-	}
-
-	if mo == 0 {
-		mo = s.prevOffset[0]
-		return
-	}
-	var temp int
-	if mo == 3 {
-		temp = s.prevOffset[0] - 1
-	} else {
-		temp = s.prevOffset[mo]
-	}
-
-	if temp == 0 {
-		// 0 is not valid; input is corrupted; force offset to 1
-		println("temp was 0")
-		temp = 1
-	}
-
-	if mo != 1 {
-		s.prevOffset[2] = s.prevOffset[1]
-	}
-	s.prevOffset[1] = s.prevOffset[0]
-	s.prevOffset[0] = temp
-	mo = temp
-	return
-}
-
 func (s *sequenceDecs) next(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) {
 	// Final will not read from stream.
 	ll, llB := llState.final()
@@ -457,36 +489,3 @@
 	s.prevOffset[0] = temp
 	return temp
 }
-
-// mergeHistory will merge history.
-func (s *sequenceDecs) mergeHistory(hist *sequenceDecs) (*sequenceDecs, error) {
-	for i := uint(0); i < 3; i++ {
-		var sNew, sHist *sequenceDec
-		switch i {
-		default:
-			// same as "case 0":
-			sNew = &s.litLengths
-			sHist = &hist.litLengths
-		case 1:
-			sNew = &s.offsets
-			sHist = &hist.offsets
-		case 2:
-			sNew = &s.matchLengths
-			sHist = &hist.matchLengths
-		}
-		if sNew.repeat {
-			if sHist.fse == nil {
-				return nil, fmt.Errorf("sequence stream %d, repeat requested, but no history", i)
-			}
-			continue
-		}
-		if sNew.fse == nil {
-			return nil, fmt.Errorf("sequence stream %d, no fse found", i)
-		}
-		if sHist.fse != nil && !sHist.fse.preDefined {
-			fseDecoderPool.Put(sHist.fse)
-		}
-		sHist.fse = sNew.fse
-	}
-	return hist, nil
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go
new file mode 100644
index 0000000..7598c10
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go
@@ -0,0 +1,368 @@
+//go:build amd64 && !appengine && !noasm && gc
+// +build amd64,!appengine,!noasm,gc
+
+package zstd
+
+import (
+	"fmt"
+
+	"github.com/klauspost/compress/internal/cpuinfo"
+)
+
+type decodeSyncAsmContext struct {
+	llTable     []decSymbol
+	mlTable     []decSymbol
+	ofTable     []decSymbol
+	llState     uint64
+	mlState     uint64
+	ofState     uint64
+	iteration   int
+	litRemain   int
+	out         []byte
+	outPosition int
+	literals    []byte
+	litPosition int
+	history     []byte
+	windowSize  int
+	ll          int // set on error (not for all errors, please refer to _generate/gen.go)
+	ml          int // set on error (not for all errors, please refer to _generate/gen.go)
+	mo          int // set on error (not for all errors, please refer to _generate/gen.go)
+}
+
+// sequenceDecs_decodeSync_amd64 implements the main loop of sequenceDecs.decodeSync in x86 asm.
+//
+// Please refer to seqdec_generic.go for the reference implementation.
+//go:noescape
+func sequenceDecs_decodeSync_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+
+// sequenceDecs_decodeSync_bmi2 implements the main loop of sequenceDecs.decodeSync in x86 asm with BMI2 extensions.
+//go:noescape
+func sequenceDecs_decodeSync_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+
+// sequenceDecs_decodeSync_safe_amd64 does the same as above, but does not write more than output buffer.
+//go:noescape
+func sequenceDecs_decodeSync_safe_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+
+// sequenceDecs_decodeSync_safe_bmi2 does the same as above, but does not write more than output buffer.
+//go:noescape
+func sequenceDecs_decodeSync_safe_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+
+// decode sequences from the stream with the provided history but without a dictionary.
+func (s *sequenceDecs) decodeSyncSimple(hist []byte) (bool, error) {
+	if len(s.dict) > 0 {
+		return false, nil
+	}
+	if s.maxSyncLen == 0 && cap(s.out)-len(s.out) < maxCompressedBlockSize {
+		return false, nil
+	}
+
+	// FIXME: Using unsafe memory copies leads to rare, random crashes
+	// with fuzz testing. It is therefore disabled for now.
+	const useSafe = true
+	/*
+		useSafe := false
+		if s.maxSyncLen == 0 && cap(s.out)-len(s.out) < maxCompressedBlockSizeAlloc {
+			useSafe = true
+		}
+		if s.maxSyncLen > 0 && cap(s.out)-len(s.out)-compressedBlockOverAlloc < int(s.maxSyncLen) {
+			useSafe = true
+		}
+		if cap(s.literals) < len(s.literals)+compressedBlockOverAlloc {
+			useSafe = true
+		}
+	*/
+
+	br := s.br
+
+	maxBlockSize := maxCompressedBlockSize
+	if s.windowSize < maxBlockSize {
+		maxBlockSize = s.windowSize
+	}
+
+	ctx := decodeSyncAsmContext{
+		llTable:     s.litLengths.fse.dt[:maxTablesize],
+		mlTable:     s.matchLengths.fse.dt[:maxTablesize],
+		ofTable:     s.offsets.fse.dt[:maxTablesize],
+		llState:     uint64(s.litLengths.state.state),
+		mlState:     uint64(s.matchLengths.state.state),
+		ofState:     uint64(s.offsets.state.state),
+		iteration:   s.nSeqs - 1,
+		litRemain:   len(s.literals),
+		out:         s.out,
+		outPosition: len(s.out),
+		literals:    s.literals,
+		windowSize:  s.windowSize,
+		history:     hist,
+	}
+
+	s.seqSize = 0
+	startSize := len(s.out)
+
+	var errCode int
+	if cpuinfo.HasBMI2() {
+		if useSafe {
+			errCode = sequenceDecs_decodeSync_safe_bmi2(s, br, &ctx)
+		} else {
+			errCode = sequenceDecs_decodeSync_bmi2(s, br, &ctx)
+		}
+	} else {
+		if useSafe {
+			errCode = sequenceDecs_decodeSync_safe_amd64(s, br, &ctx)
+		} else {
+			errCode = sequenceDecs_decodeSync_amd64(s, br, &ctx)
+		}
+	}
+	switch errCode {
+	case noError:
+		break
+
+	case errorMatchLenOfsMismatch:
+		return true, fmt.Errorf("zero matchoff and matchlen (%d) > 0", ctx.ml)
+
+	case errorMatchLenTooBig:
+		return true, fmt.Errorf("match len (%d) bigger than max allowed length", ctx.ml)
+
+	case errorMatchOffTooBig:
+		return true, fmt.Errorf("match offset (%d) bigger than current history (%d)",
+			ctx.mo, ctx.outPosition+len(hist)-startSize)
+
+	case errorNotEnoughLiterals:
+		return true, fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available",
+			ctx.ll, ctx.litRemain+ctx.ll)
+
+	case errorNotEnoughSpace:
+		size := ctx.outPosition + ctx.ll + ctx.ml
+		if debugDecoder {
+			println("msl:", s.maxSyncLen, "cap", cap(s.out), "bef:", startSize, "sz:", size-startSize, "mbs:", maxBlockSize, "outsz:", cap(s.out)-startSize)
+		}
+		return true, fmt.Errorf("output (%d) bigger than max block size (%d)", size-startSize, maxBlockSize)
+
+	default:
+		return true, fmt.Errorf("sequenceDecs_decode returned erronous code %d", errCode)
+	}
+
+	s.seqSize += ctx.litRemain
+	if s.seqSize > maxBlockSize {
+		return true, fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
+	}
+	err := br.close()
+	if err != nil {
+		printf("Closing sequences: %v, %+v\n", err, *br)
+		return true, err
+	}
+
+	s.literals = s.literals[ctx.litPosition:]
+	t := ctx.outPosition
+	s.out = s.out[:t]
+
+	// Add final literals
+	s.out = append(s.out, s.literals...)
+	if debugDecoder {
+		t += len(s.literals)
+		if t != len(s.out) {
+			panic(fmt.Errorf("length mismatch, want %d, got %d", len(s.out), t))
+		}
+	}
+
+	return true, nil
+}
+
+// --------------------------------------------------------------------------------
+
+type decodeAsmContext struct {
+	llTable   []decSymbol
+	mlTable   []decSymbol
+	ofTable   []decSymbol
+	llState   uint64
+	mlState   uint64
+	ofState   uint64
+	iteration int
+	seqs      []seqVals
+	litRemain int
+}
+
+const noError = 0
+
+// error reported when mo == 0 && ml > 0
+const errorMatchLenOfsMismatch = 1
+
+// error reported when ml > maxMatchLen
+const errorMatchLenTooBig = 2
+
+// error reported when mo > available history or mo > s.windowSize
+const errorMatchOffTooBig = 3
+
+// error reported when the sum of literal lengths exeeceds the literal buffer size
+const errorNotEnoughLiterals = 4
+
+// error reported when capacity of `out` is too small
+const errorNotEnoughSpace = 5
+
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm.
+//
+// Please refer to seqdec_generic.go for the reference implementation.
+//go:noescape
+func sequenceDecs_decode_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm.
+//
+// Please refer to seqdec_generic.go for the reference implementation.
+//go:noescape
+func sequenceDecs_decode_56_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm with BMI2 extensions.
+//go:noescape
+func sequenceDecs_decode_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm with BMI2 extensions.
+//go:noescape
+func sequenceDecs_decode_56_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+
+// decode sequences from the stream without the provided history.
+func (s *sequenceDecs) decode(seqs []seqVals) error {
+	br := s.br
+
+	maxBlockSize := maxCompressedBlockSize
+	if s.windowSize < maxBlockSize {
+		maxBlockSize = s.windowSize
+	}
+
+	ctx := decodeAsmContext{
+		llTable:   s.litLengths.fse.dt[:maxTablesize],
+		mlTable:   s.matchLengths.fse.dt[:maxTablesize],
+		ofTable:   s.offsets.fse.dt[:maxTablesize],
+		llState:   uint64(s.litLengths.state.state),
+		mlState:   uint64(s.matchLengths.state.state),
+		ofState:   uint64(s.offsets.state.state),
+		seqs:      seqs,
+		iteration: len(seqs) - 1,
+		litRemain: len(s.literals),
+	}
+
+	s.seqSize = 0
+	lte56bits := s.maxBits+s.offsets.fse.actualTableLog+s.matchLengths.fse.actualTableLog+s.litLengths.fse.actualTableLog <= 56
+	var errCode int
+	if cpuinfo.HasBMI2() {
+		if lte56bits {
+			errCode = sequenceDecs_decode_56_bmi2(s, br, &ctx)
+		} else {
+			errCode = sequenceDecs_decode_bmi2(s, br, &ctx)
+		}
+	} else {
+		if lte56bits {
+			errCode = sequenceDecs_decode_56_amd64(s, br, &ctx)
+		} else {
+			errCode = sequenceDecs_decode_amd64(s, br, &ctx)
+		}
+	}
+	if errCode != 0 {
+		i := len(seqs) - ctx.iteration - 1
+		switch errCode {
+		case errorMatchLenOfsMismatch:
+			ml := ctx.seqs[i].ml
+			return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
+
+		case errorMatchLenTooBig:
+			ml := ctx.seqs[i].ml
+			return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
+
+		case errorNotEnoughLiterals:
+			ll := ctx.seqs[i].ll
+			return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, ctx.litRemain+ll)
+		}
+
+		return fmt.Errorf("sequenceDecs_decode_amd64 returned erronous code %d", errCode)
+	}
+
+	if ctx.litRemain < 0 {
+		return fmt.Errorf("literal count is too big: total available %d, total requested %d",
+			len(s.literals), len(s.literals)-ctx.litRemain)
+	}
+
+	s.seqSize += ctx.litRemain
+	if s.seqSize > maxBlockSize {
+		return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
+	}
+	err := br.close()
+	if err != nil {
+		printf("Closing sequences: %v, %+v\n", err, *br)
+	}
+	return err
+}
+
+// --------------------------------------------------------------------------------
+
+type executeAsmContext struct {
+	seqs        []seqVals
+	seqIndex    int
+	out         []byte
+	history     []byte
+	literals    []byte
+	outPosition int
+	litPosition int
+	windowSize  int
+}
+
+// sequenceDecs_executeSimple_amd64 implements the main loop of sequenceDecs.executeSimple in x86 asm.
+//
+// Returns false if a match offset is too big.
+//
+// Please refer to seqdec_generic.go for the reference implementation.
+//go:noescape
+func sequenceDecs_executeSimple_amd64(ctx *executeAsmContext) bool
+
+// Same as above, but with safe memcopies
+//go:noescape
+func sequenceDecs_executeSimple_safe_amd64(ctx *executeAsmContext) bool
+
+// executeSimple handles cases when dictionary is not used.
+func (s *sequenceDecs) executeSimple(seqs []seqVals, hist []byte) error {
+	// Ensure we have enough output size...
+	if len(s.out)+s.seqSize+compressedBlockOverAlloc > cap(s.out) {
+		addBytes := s.seqSize + len(s.out) + compressedBlockOverAlloc
+		s.out = append(s.out, make([]byte, addBytes)...)
+		s.out = s.out[:len(s.out)-addBytes]
+	}
+
+	if debugDecoder {
+		printf("Execute %d seqs with literals: %d into %d bytes\n", len(seqs), len(s.literals), s.seqSize)
+	}
+
+	var t = len(s.out)
+	out := s.out[:t+s.seqSize]
+
+	ctx := executeAsmContext{
+		seqs:        seqs,
+		seqIndex:    0,
+		out:         out,
+		history:     hist,
+		outPosition: t,
+		litPosition: 0,
+		literals:    s.literals,
+		windowSize:  s.windowSize,
+	}
+	var ok bool
+	if cap(s.literals) < len(s.literals)+compressedBlockOverAlloc {
+		ok = sequenceDecs_executeSimple_safe_amd64(&ctx)
+	} else {
+		ok = sequenceDecs_executeSimple_amd64(&ctx)
+	}
+	if !ok {
+		return fmt.Errorf("match offset (%d) bigger than current history (%d)",
+			seqs[ctx.seqIndex].mo, ctx.outPosition+len(hist))
+	}
+	s.literals = s.literals[ctx.litPosition:]
+	t = ctx.outPosition
+
+	// Add final literals
+	copy(out[t:], s.literals)
+	if debugDecoder {
+		t += len(s.literals)
+		if t != len(out) {
+			panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize))
+		}
+	}
+	s.out = out
+
+	return nil
+}
diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s
new file mode 100644
index 0000000..27e7677
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s
@@ -0,0 +1,4100 @@
+// Code generated by command: go run gen.go -out ../seqdec_amd64.s -pkg=zstd. DO NOT EDIT.
+
+//go:build !appengine && !noasm && gc && !noasm
+// +build !appengine,!noasm,gc,!noasm
+
+// func sequenceDecs_decode_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+// Requires: CMOV
+TEXT ·sequenceDecs_decode_amd64(SB), $8-32
+	MOVQ    br+8(FP), AX
+	MOVQ    32(AX), DX
+	MOVBQZX 40(AX), BX
+	MOVQ    24(AX), SI
+	MOVQ    (AX), AX
+	ADDQ    SI, AX
+	MOVQ    AX, (SP)
+	MOVQ    ctx+16(FP), AX
+	MOVQ    72(AX), DI
+	MOVQ    80(AX), R8
+	MOVQ    88(AX), R9
+	MOVQ    104(AX), R10
+	MOVQ    s+0(FP), AX
+	MOVQ    144(AX), R11
+	MOVQ    152(AX), R12
+	MOVQ    160(AX), R13
+
+sequenceDecs_decode_amd64_main_loop:
+	MOVQ (SP), R14
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decode_amd64_fill_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R14
+	MOVQ (R14), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decode_amd64_fill_end
+
+sequenceDecs_decode_amd64_fill_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decode_amd64_fill_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decode_amd64_fill_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R14
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R14), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decode_amd64_fill_byte_by_byte
+
+sequenceDecs_decode_amd64_fill_end:
+	// Update offset
+	MOVQ  R9, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R15
+	SHLQ  CL, R15
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decode_amd64_of_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decode_amd64_of_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decode_amd64_of_update_zero
+	NEGQ  CX
+	SHRQ  CL, R15
+	ADDQ  R15, AX
+
+sequenceDecs_decode_amd64_of_update_zero:
+	MOVQ AX, 16(R10)
+
+	// Update match length
+	MOVQ  R8, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R15
+	SHLQ  CL, R15
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decode_amd64_ml_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decode_amd64_ml_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decode_amd64_ml_update_zero
+	NEGQ  CX
+	SHRQ  CL, R15
+	ADDQ  R15, AX
+
+sequenceDecs_decode_amd64_ml_update_zero:
+	MOVQ AX, 8(R10)
+
+	// Fill bitreader to have enough for the remaining
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decode_amd64_fill_2_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R14
+	MOVQ (R14), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decode_amd64_fill_2_end
+
+sequenceDecs_decode_amd64_fill_2_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decode_amd64_fill_2_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decode_amd64_fill_2_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R14
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R14), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decode_amd64_fill_2_byte_by_byte
+
+sequenceDecs_decode_amd64_fill_2_end:
+	// Update literal length
+	MOVQ  DI, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R15
+	SHLQ  CL, R15
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decode_amd64_ll_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decode_amd64_ll_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decode_amd64_ll_update_zero
+	NEGQ  CX
+	SHRQ  CL, R15
+	ADDQ  R15, AX
+
+sequenceDecs_decode_amd64_ll_update_zero:
+	MOVQ AX, (R10)
+
+	// Fill bitreader for state updates
+	MOVQ    R14, (SP)
+	MOVQ    R9, AX
+	SHRQ    $0x08, AX
+	MOVBQZX AL, AX
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decode_amd64_skip_update
+
+	// Update Literal Length State
+	MOVBQZX DI, R14
+	SHRQ    $0x10, DI
+	MOVWQZX DI, DI
+	LEAQ    (BX)(R14*1), CX
+	MOVQ    DX, R15
+	MOVQ    CX, BX
+	ROLQ    CL, R15
+	MOVL    $0x00000001, BP
+	MOVB    R14, CL
+	SHLL    CL, BP
+	DECL    BP
+	ANDQ    BP, R15
+	ADDQ    R15, DI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Match Length State
+	MOVBQZX R8, R14
+	SHRQ    $0x10, R8
+	MOVWQZX R8, R8
+	LEAQ    (BX)(R14*1), CX
+	MOVQ    DX, R15
+	MOVQ    CX, BX
+	ROLQ    CL, R15
+	MOVL    $0x00000001, BP
+	MOVB    R14, CL
+	SHLL    CL, BP
+	DECL    BP
+	ANDQ    BP, R15
+	ADDQ    R15, R8
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Offset State
+	MOVBQZX R9, R14
+	SHRQ    $0x10, R9
+	MOVWQZX R9, R9
+	LEAQ    (BX)(R14*1), CX
+	MOVQ    DX, R15
+	MOVQ    CX, BX
+	ROLQ    CL, R15
+	MOVL    $0x00000001, BP
+	MOVB    R14, CL
+	SHLL    CL, BP
+	DECL    BP
+	ANDQ    BP, R15
+	ADDQ    R15, R9
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R9*8), R9
+
+sequenceDecs_decode_amd64_skip_update:
+	// Adjust offset
+	MOVQ 16(R10), CX
+	CMPQ AX, $0x01
+	JBE  sequenceDecs_decode_amd64_adjust_offsetB_1_or_0
+	MOVQ R12, R13
+	MOVQ R11, R12
+	MOVQ CX, R11
+	JMP  sequenceDecs_decode_amd64_after_adjust
+
+sequenceDecs_decode_amd64_adjust_offsetB_1_or_0:
+	CMPQ (R10), $0x00000000
+	JNE  sequenceDecs_decode_amd64_adjust_offset_maybezero
+	INCQ CX
+	JMP  sequenceDecs_decode_amd64_adjust_offset_nonzero
+
+sequenceDecs_decode_amd64_adjust_offset_maybezero:
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_amd64_adjust_offset_nonzero
+	MOVQ  R11, CX
+	JMP   sequenceDecs_decode_amd64_after_adjust
+
+sequenceDecs_decode_amd64_adjust_offset_nonzero:
+	CMPQ CX, $0x01
+	JB   sequenceDecs_decode_amd64_adjust_zero
+	JEQ  sequenceDecs_decode_amd64_adjust_one
+	CMPQ CX, $0x02
+	JA   sequenceDecs_decode_amd64_adjust_three
+	JMP  sequenceDecs_decode_amd64_adjust_two
+
+sequenceDecs_decode_amd64_adjust_zero:
+	MOVQ R11, AX
+	JMP  sequenceDecs_decode_amd64_adjust_test_temp_valid
+
+sequenceDecs_decode_amd64_adjust_one:
+	MOVQ R12, AX
+	JMP  sequenceDecs_decode_amd64_adjust_test_temp_valid
+
+sequenceDecs_decode_amd64_adjust_two:
+	MOVQ R13, AX
+	JMP  sequenceDecs_decode_amd64_adjust_test_temp_valid
+
+sequenceDecs_decode_amd64_adjust_three:
+	LEAQ -1(R11), AX
+
+sequenceDecs_decode_amd64_adjust_test_temp_valid:
+	TESTQ AX, AX
+	JNZ   sequenceDecs_decode_amd64_adjust_temp_valid
+	MOVQ  $0x00000001, AX
+
+sequenceDecs_decode_amd64_adjust_temp_valid:
+	CMPQ    CX, $0x01
+	CMOVQNE R12, R13
+	MOVQ    R11, R12
+	MOVQ    AX, R11
+	MOVQ    AX, CX
+
+sequenceDecs_decode_amd64_after_adjust:
+	MOVQ CX, 16(R10)
+
+	// Check values
+	MOVQ  8(R10), AX
+	MOVQ  (R10), R14
+	LEAQ  (AX)(R14*1), R15
+	MOVQ  s+0(FP), BP
+	ADDQ  R15, 256(BP)
+	MOVQ  ctx+16(FP), R15
+	SUBQ  R14, 128(R15)
+	JS    error_not_enough_literals
+	CMPQ  AX, $0x00020002
+	JA    sequenceDecs_decode_amd64_error_match_len_too_big
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_amd64_match_len_ofs_ok
+	TESTQ AX, AX
+	JNZ   sequenceDecs_decode_amd64_error_match_len_ofs_mismatch
+
+sequenceDecs_decode_amd64_match_len_ofs_ok:
+	ADDQ $0x18, R10
+	MOVQ ctx+16(FP), AX
+	DECQ 96(AX)
+	JNS  sequenceDecs_decode_amd64_main_loop
+	MOVQ s+0(FP), AX
+	MOVQ R11, 144(AX)
+	MOVQ R12, 152(AX)
+	MOVQ R13, 160(AX)
+	MOVQ br+8(FP), AX
+	MOVQ DX, 32(AX)
+	MOVB BL, 40(AX)
+	MOVQ SI, 24(AX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decode_amd64_error_match_len_ofs_mismatch:
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decode_amd64_error_match_len_too_big:
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_decode_56_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+// Requires: CMOV
+TEXT ·sequenceDecs_decode_56_amd64(SB), $8-32
+	MOVQ    br+8(FP), AX
+	MOVQ    32(AX), DX
+	MOVBQZX 40(AX), BX
+	MOVQ    24(AX), SI
+	MOVQ    (AX), AX
+	ADDQ    SI, AX
+	MOVQ    AX, (SP)
+	MOVQ    ctx+16(FP), AX
+	MOVQ    72(AX), DI
+	MOVQ    80(AX), R8
+	MOVQ    88(AX), R9
+	MOVQ    104(AX), R10
+	MOVQ    s+0(FP), AX
+	MOVQ    144(AX), R11
+	MOVQ    152(AX), R12
+	MOVQ    160(AX), R13
+
+sequenceDecs_decode_56_amd64_main_loop:
+	MOVQ (SP), R14
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decode_56_amd64_fill_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R14
+	MOVQ (R14), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decode_56_amd64_fill_end
+
+sequenceDecs_decode_56_amd64_fill_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decode_56_amd64_fill_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decode_56_amd64_fill_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R14
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R14), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decode_56_amd64_fill_byte_by_byte
+
+sequenceDecs_decode_56_amd64_fill_end:
+	// Update offset
+	MOVQ  R9, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R15
+	SHLQ  CL, R15
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decode_56_amd64_of_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decode_56_amd64_of_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decode_56_amd64_of_update_zero
+	NEGQ  CX
+	SHRQ  CL, R15
+	ADDQ  R15, AX
+
+sequenceDecs_decode_56_amd64_of_update_zero:
+	MOVQ AX, 16(R10)
+
+	// Update match length
+	MOVQ  R8, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R15
+	SHLQ  CL, R15
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decode_56_amd64_ml_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decode_56_amd64_ml_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decode_56_amd64_ml_update_zero
+	NEGQ  CX
+	SHRQ  CL, R15
+	ADDQ  R15, AX
+
+sequenceDecs_decode_56_amd64_ml_update_zero:
+	MOVQ AX, 8(R10)
+
+	// Update literal length
+	MOVQ  DI, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R15
+	SHLQ  CL, R15
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decode_56_amd64_ll_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decode_56_amd64_ll_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decode_56_amd64_ll_update_zero
+	NEGQ  CX
+	SHRQ  CL, R15
+	ADDQ  R15, AX
+
+sequenceDecs_decode_56_amd64_ll_update_zero:
+	MOVQ AX, (R10)
+
+	// Fill bitreader for state updates
+	MOVQ    R14, (SP)
+	MOVQ    R9, AX
+	SHRQ    $0x08, AX
+	MOVBQZX AL, AX
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decode_56_amd64_skip_update
+
+	// Update Literal Length State
+	MOVBQZX DI, R14
+	SHRQ    $0x10, DI
+	MOVWQZX DI, DI
+	LEAQ    (BX)(R14*1), CX
+	MOVQ    DX, R15
+	MOVQ    CX, BX
+	ROLQ    CL, R15
+	MOVL    $0x00000001, BP
+	MOVB    R14, CL
+	SHLL    CL, BP
+	DECL    BP
+	ANDQ    BP, R15
+	ADDQ    R15, DI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Match Length State
+	MOVBQZX R8, R14
+	SHRQ    $0x10, R8
+	MOVWQZX R8, R8
+	LEAQ    (BX)(R14*1), CX
+	MOVQ    DX, R15
+	MOVQ    CX, BX
+	ROLQ    CL, R15
+	MOVL    $0x00000001, BP
+	MOVB    R14, CL
+	SHLL    CL, BP
+	DECL    BP
+	ANDQ    BP, R15
+	ADDQ    R15, R8
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Offset State
+	MOVBQZX R9, R14
+	SHRQ    $0x10, R9
+	MOVWQZX R9, R9
+	LEAQ    (BX)(R14*1), CX
+	MOVQ    DX, R15
+	MOVQ    CX, BX
+	ROLQ    CL, R15
+	MOVL    $0x00000001, BP
+	MOVB    R14, CL
+	SHLL    CL, BP
+	DECL    BP
+	ANDQ    BP, R15
+	ADDQ    R15, R9
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R9*8), R9
+
+sequenceDecs_decode_56_amd64_skip_update:
+	// Adjust offset
+	MOVQ 16(R10), CX
+	CMPQ AX, $0x01
+	JBE  sequenceDecs_decode_56_amd64_adjust_offsetB_1_or_0
+	MOVQ R12, R13
+	MOVQ R11, R12
+	MOVQ CX, R11
+	JMP  sequenceDecs_decode_56_amd64_after_adjust
+
+sequenceDecs_decode_56_amd64_adjust_offsetB_1_or_0:
+	CMPQ (R10), $0x00000000
+	JNE  sequenceDecs_decode_56_amd64_adjust_offset_maybezero
+	INCQ CX
+	JMP  sequenceDecs_decode_56_amd64_adjust_offset_nonzero
+
+sequenceDecs_decode_56_amd64_adjust_offset_maybezero:
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_56_amd64_adjust_offset_nonzero
+	MOVQ  R11, CX
+	JMP   sequenceDecs_decode_56_amd64_after_adjust
+
+sequenceDecs_decode_56_amd64_adjust_offset_nonzero:
+	CMPQ CX, $0x01
+	JB   sequenceDecs_decode_56_amd64_adjust_zero
+	JEQ  sequenceDecs_decode_56_amd64_adjust_one
+	CMPQ CX, $0x02
+	JA   sequenceDecs_decode_56_amd64_adjust_three
+	JMP  sequenceDecs_decode_56_amd64_adjust_two
+
+sequenceDecs_decode_56_amd64_adjust_zero:
+	MOVQ R11, AX
+	JMP  sequenceDecs_decode_56_amd64_adjust_test_temp_valid
+
+sequenceDecs_decode_56_amd64_adjust_one:
+	MOVQ R12, AX
+	JMP  sequenceDecs_decode_56_amd64_adjust_test_temp_valid
+
+sequenceDecs_decode_56_amd64_adjust_two:
+	MOVQ R13, AX
+	JMP  sequenceDecs_decode_56_amd64_adjust_test_temp_valid
+
+sequenceDecs_decode_56_amd64_adjust_three:
+	LEAQ -1(R11), AX
+
+sequenceDecs_decode_56_amd64_adjust_test_temp_valid:
+	TESTQ AX, AX
+	JNZ   sequenceDecs_decode_56_amd64_adjust_temp_valid
+	MOVQ  $0x00000001, AX
+
+sequenceDecs_decode_56_amd64_adjust_temp_valid:
+	CMPQ    CX, $0x01
+	CMOVQNE R12, R13
+	MOVQ    R11, R12
+	MOVQ    AX, R11
+	MOVQ    AX, CX
+
+sequenceDecs_decode_56_amd64_after_adjust:
+	MOVQ CX, 16(R10)
+
+	// Check values
+	MOVQ  8(R10), AX
+	MOVQ  (R10), R14
+	LEAQ  (AX)(R14*1), R15
+	MOVQ  s+0(FP), BP
+	ADDQ  R15, 256(BP)
+	MOVQ  ctx+16(FP), R15
+	SUBQ  R14, 128(R15)
+	JS    error_not_enough_literals
+	CMPQ  AX, $0x00020002
+	JA    sequenceDecs_decode_56_amd64_error_match_len_too_big
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_56_amd64_match_len_ofs_ok
+	TESTQ AX, AX
+	JNZ   sequenceDecs_decode_56_amd64_error_match_len_ofs_mismatch
+
+sequenceDecs_decode_56_amd64_match_len_ofs_ok:
+	ADDQ $0x18, R10
+	MOVQ ctx+16(FP), AX
+	DECQ 96(AX)
+	JNS  sequenceDecs_decode_56_amd64_main_loop
+	MOVQ s+0(FP), AX
+	MOVQ R11, 144(AX)
+	MOVQ R12, 152(AX)
+	MOVQ R13, 160(AX)
+	MOVQ br+8(FP), AX
+	MOVQ DX, 32(AX)
+	MOVB BL, 40(AX)
+	MOVQ SI, 24(AX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decode_56_amd64_error_match_len_ofs_mismatch:
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decode_56_amd64_error_match_len_too_big:
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_decode_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+// Requires: BMI, BMI2, CMOV
+TEXT ·sequenceDecs_decode_bmi2(SB), $8-32
+	MOVQ    br+8(FP), CX
+	MOVQ    32(CX), AX
+	MOVBQZX 40(CX), DX
+	MOVQ    24(CX), BX
+	MOVQ    (CX), CX
+	ADDQ    BX, CX
+	MOVQ    CX, (SP)
+	MOVQ    ctx+16(FP), CX
+	MOVQ    72(CX), SI
+	MOVQ    80(CX), DI
+	MOVQ    88(CX), R8
+	MOVQ    104(CX), R9
+	MOVQ    s+0(FP), CX
+	MOVQ    144(CX), R10
+	MOVQ    152(CX), R11
+	MOVQ    160(CX), R12
+
+sequenceDecs_decode_bmi2_main_loop:
+	MOVQ (SP), R13
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decode_bmi2_fill_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R13
+	MOVQ (R13), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decode_bmi2_fill_end
+
+sequenceDecs_decode_bmi2_fill_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decode_bmi2_fill_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decode_bmi2_fill_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R13), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decode_bmi2_fill_byte_by_byte
+
+sequenceDecs_decode_bmi2_fill_end:
+	// Update offset
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, R8, R14
+	MOVQ   AX, R15
+	LEAQ   (DX)(R14*1), CX
+	ROLQ   CL, R15
+	BZHIQ  R14, R15, R15
+	MOVQ   CX, DX
+	MOVQ   R8, CX
+	SHRQ   $0x20, CX
+	ADDQ   R15, CX
+	MOVQ   CX, 16(R9)
+
+	// Update match length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, DI, R14
+	MOVQ   AX, R15
+	LEAQ   (DX)(R14*1), CX
+	ROLQ   CL, R15
+	BZHIQ  R14, R15, R15
+	MOVQ   CX, DX
+	MOVQ   DI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R15, CX
+	MOVQ   CX, 8(R9)
+
+	// Fill bitreader to have enough for the remaining
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decode_bmi2_fill_2_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R13
+	MOVQ (R13), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decode_bmi2_fill_2_end
+
+sequenceDecs_decode_bmi2_fill_2_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decode_bmi2_fill_2_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decode_bmi2_fill_2_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R13), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decode_bmi2_fill_2_byte_by_byte
+
+sequenceDecs_decode_bmi2_fill_2_end:
+	// Update literal length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, SI, R14
+	MOVQ   AX, R15
+	LEAQ   (DX)(R14*1), CX
+	ROLQ   CL, R15
+	BZHIQ  R14, R15, R15
+	MOVQ   CX, DX
+	MOVQ   SI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R15, CX
+	MOVQ   CX, (R9)
+
+	// Fill bitreader for state updates
+	MOVQ    R13, (SP)
+	MOVQ    $0x00000808, CX
+	BEXTRQ  CX, R8, R13
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decode_bmi2_skip_update
+	LEAQ    (SI)(DI*1), R14
+	ADDQ    R8, R14
+	MOVBQZX R14, R14
+	LEAQ    (DX)(R14*1), CX
+	MOVQ    AX, R15
+	MOVQ    CX, DX
+	ROLQ    CL, R15
+	BZHIQ   R14, R15, R15
+
+	// Update Offset State
+	BZHIQ  R8, R15, CX
+	SHRXQ  R8, R15, R15
+	MOVQ   $0x00001010, R14
+	BEXTRQ R14, R8, R8
+	ADDQ   CX, R8
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Match Length State
+	BZHIQ  DI, R15, CX
+	SHRXQ  DI, R15, R15
+	MOVQ   $0x00001010, R14
+	BEXTRQ R14, DI, DI
+	ADDQ   CX, DI
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Literal Length State
+	BZHIQ  SI, R15, CX
+	MOVQ   $0x00001010, R14
+	BEXTRQ R14, SI, SI
+	ADDQ   CX, SI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(SI*8), SI
+
+sequenceDecs_decode_bmi2_skip_update:
+	// Adjust offset
+	MOVQ 16(R9), CX
+	CMPQ R13, $0x01
+	JBE  sequenceDecs_decode_bmi2_adjust_offsetB_1_or_0
+	MOVQ R11, R12
+	MOVQ R10, R11
+	MOVQ CX, R10
+	JMP  sequenceDecs_decode_bmi2_after_adjust
+
+sequenceDecs_decode_bmi2_adjust_offsetB_1_or_0:
+	CMPQ (R9), $0x00000000
+	JNE  sequenceDecs_decode_bmi2_adjust_offset_maybezero
+	INCQ CX
+	JMP  sequenceDecs_decode_bmi2_adjust_offset_nonzero
+
+sequenceDecs_decode_bmi2_adjust_offset_maybezero:
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_bmi2_adjust_offset_nonzero
+	MOVQ  R10, CX
+	JMP   sequenceDecs_decode_bmi2_after_adjust
+
+sequenceDecs_decode_bmi2_adjust_offset_nonzero:
+	CMPQ CX, $0x01
+	JB   sequenceDecs_decode_bmi2_adjust_zero
+	JEQ  sequenceDecs_decode_bmi2_adjust_one
+	CMPQ CX, $0x02
+	JA   sequenceDecs_decode_bmi2_adjust_three
+	JMP  sequenceDecs_decode_bmi2_adjust_two
+
+sequenceDecs_decode_bmi2_adjust_zero:
+	MOVQ R10, R13
+	JMP  sequenceDecs_decode_bmi2_adjust_test_temp_valid
+
+sequenceDecs_decode_bmi2_adjust_one:
+	MOVQ R11, R13
+	JMP  sequenceDecs_decode_bmi2_adjust_test_temp_valid
+
+sequenceDecs_decode_bmi2_adjust_two:
+	MOVQ R12, R13
+	JMP  sequenceDecs_decode_bmi2_adjust_test_temp_valid
+
+sequenceDecs_decode_bmi2_adjust_three:
+	LEAQ -1(R10), R13
+
+sequenceDecs_decode_bmi2_adjust_test_temp_valid:
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decode_bmi2_adjust_temp_valid
+	MOVQ  $0x00000001, R13
+
+sequenceDecs_decode_bmi2_adjust_temp_valid:
+	CMPQ    CX, $0x01
+	CMOVQNE R11, R12
+	MOVQ    R10, R11
+	MOVQ    R13, R10
+	MOVQ    R13, CX
+
+sequenceDecs_decode_bmi2_after_adjust:
+	MOVQ CX, 16(R9)
+
+	// Check values
+	MOVQ  8(R9), R13
+	MOVQ  (R9), R14
+	LEAQ  (R13)(R14*1), R15
+	MOVQ  s+0(FP), BP
+	ADDQ  R15, 256(BP)
+	MOVQ  ctx+16(FP), R15
+	SUBQ  R14, 128(R15)
+	JS    error_not_enough_literals
+	CMPQ  R13, $0x00020002
+	JA    sequenceDecs_decode_bmi2_error_match_len_too_big
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_bmi2_match_len_ofs_ok
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decode_bmi2_error_match_len_ofs_mismatch
+
+sequenceDecs_decode_bmi2_match_len_ofs_ok:
+	ADDQ $0x18, R9
+	MOVQ ctx+16(FP), CX
+	DECQ 96(CX)
+	JNS  sequenceDecs_decode_bmi2_main_loop
+	MOVQ s+0(FP), CX
+	MOVQ R10, 144(CX)
+	MOVQ R11, 152(CX)
+	MOVQ R12, 160(CX)
+	MOVQ br+8(FP), CX
+	MOVQ AX, 32(CX)
+	MOVB DL, 40(CX)
+	MOVQ BX, 24(CX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decode_bmi2_error_match_len_ofs_mismatch:
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decode_bmi2_error_match_len_too_big:
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_decode_56_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int
+// Requires: BMI, BMI2, CMOV
+TEXT ·sequenceDecs_decode_56_bmi2(SB), $8-32
+	MOVQ    br+8(FP), CX
+	MOVQ    32(CX), AX
+	MOVBQZX 40(CX), DX
+	MOVQ    24(CX), BX
+	MOVQ    (CX), CX
+	ADDQ    BX, CX
+	MOVQ    CX, (SP)
+	MOVQ    ctx+16(FP), CX
+	MOVQ    72(CX), SI
+	MOVQ    80(CX), DI
+	MOVQ    88(CX), R8
+	MOVQ    104(CX), R9
+	MOVQ    s+0(FP), CX
+	MOVQ    144(CX), R10
+	MOVQ    152(CX), R11
+	MOVQ    160(CX), R12
+
+sequenceDecs_decode_56_bmi2_main_loop:
+	MOVQ (SP), R13
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decode_56_bmi2_fill_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R13
+	MOVQ (R13), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decode_56_bmi2_fill_end
+
+sequenceDecs_decode_56_bmi2_fill_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decode_56_bmi2_fill_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decode_56_bmi2_fill_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R13), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decode_56_bmi2_fill_byte_by_byte
+
+sequenceDecs_decode_56_bmi2_fill_end:
+	// Update offset
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, R8, R14
+	MOVQ   AX, R15
+	LEAQ   (DX)(R14*1), CX
+	ROLQ   CL, R15
+	BZHIQ  R14, R15, R15
+	MOVQ   CX, DX
+	MOVQ   R8, CX
+	SHRQ   $0x20, CX
+	ADDQ   R15, CX
+	MOVQ   CX, 16(R9)
+
+	// Update match length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, DI, R14
+	MOVQ   AX, R15
+	LEAQ   (DX)(R14*1), CX
+	ROLQ   CL, R15
+	BZHIQ  R14, R15, R15
+	MOVQ   CX, DX
+	MOVQ   DI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R15, CX
+	MOVQ   CX, 8(R9)
+
+	// Update literal length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, SI, R14
+	MOVQ   AX, R15
+	LEAQ   (DX)(R14*1), CX
+	ROLQ   CL, R15
+	BZHIQ  R14, R15, R15
+	MOVQ   CX, DX
+	MOVQ   SI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R15, CX
+	MOVQ   CX, (R9)
+
+	// Fill bitreader for state updates
+	MOVQ    R13, (SP)
+	MOVQ    $0x00000808, CX
+	BEXTRQ  CX, R8, R13
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decode_56_bmi2_skip_update
+	LEAQ    (SI)(DI*1), R14
+	ADDQ    R8, R14
+	MOVBQZX R14, R14
+	LEAQ    (DX)(R14*1), CX
+	MOVQ    AX, R15
+	MOVQ    CX, DX
+	ROLQ    CL, R15
+	BZHIQ   R14, R15, R15
+
+	// Update Offset State
+	BZHIQ  R8, R15, CX
+	SHRXQ  R8, R15, R15
+	MOVQ   $0x00001010, R14
+	BEXTRQ R14, R8, R8
+	ADDQ   CX, R8
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Match Length State
+	BZHIQ  DI, R15, CX
+	SHRXQ  DI, R15, R15
+	MOVQ   $0x00001010, R14
+	BEXTRQ R14, DI, DI
+	ADDQ   CX, DI
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Literal Length State
+	BZHIQ  SI, R15, CX
+	MOVQ   $0x00001010, R14
+	BEXTRQ R14, SI, SI
+	ADDQ   CX, SI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(SI*8), SI
+
+sequenceDecs_decode_56_bmi2_skip_update:
+	// Adjust offset
+	MOVQ 16(R9), CX
+	CMPQ R13, $0x01
+	JBE  sequenceDecs_decode_56_bmi2_adjust_offsetB_1_or_0
+	MOVQ R11, R12
+	MOVQ R10, R11
+	MOVQ CX, R10
+	JMP  sequenceDecs_decode_56_bmi2_after_adjust
+
+sequenceDecs_decode_56_bmi2_adjust_offsetB_1_or_0:
+	CMPQ (R9), $0x00000000
+	JNE  sequenceDecs_decode_56_bmi2_adjust_offset_maybezero
+	INCQ CX
+	JMP  sequenceDecs_decode_56_bmi2_adjust_offset_nonzero
+
+sequenceDecs_decode_56_bmi2_adjust_offset_maybezero:
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_56_bmi2_adjust_offset_nonzero
+	MOVQ  R10, CX
+	JMP   sequenceDecs_decode_56_bmi2_after_adjust
+
+sequenceDecs_decode_56_bmi2_adjust_offset_nonzero:
+	CMPQ CX, $0x01
+	JB   sequenceDecs_decode_56_bmi2_adjust_zero
+	JEQ  sequenceDecs_decode_56_bmi2_adjust_one
+	CMPQ CX, $0x02
+	JA   sequenceDecs_decode_56_bmi2_adjust_three
+	JMP  sequenceDecs_decode_56_bmi2_adjust_two
+
+sequenceDecs_decode_56_bmi2_adjust_zero:
+	MOVQ R10, R13
+	JMP  sequenceDecs_decode_56_bmi2_adjust_test_temp_valid
+
+sequenceDecs_decode_56_bmi2_adjust_one:
+	MOVQ R11, R13
+	JMP  sequenceDecs_decode_56_bmi2_adjust_test_temp_valid
+
+sequenceDecs_decode_56_bmi2_adjust_two:
+	MOVQ R12, R13
+	JMP  sequenceDecs_decode_56_bmi2_adjust_test_temp_valid
+
+sequenceDecs_decode_56_bmi2_adjust_three:
+	LEAQ -1(R10), R13
+
+sequenceDecs_decode_56_bmi2_adjust_test_temp_valid:
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decode_56_bmi2_adjust_temp_valid
+	MOVQ  $0x00000001, R13
+
+sequenceDecs_decode_56_bmi2_adjust_temp_valid:
+	CMPQ    CX, $0x01
+	CMOVQNE R11, R12
+	MOVQ    R10, R11
+	MOVQ    R13, R10
+	MOVQ    R13, CX
+
+sequenceDecs_decode_56_bmi2_after_adjust:
+	MOVQ CX, 16(R9)
+
+	// Check values
+	MOVQ  8(R9), R13
+	MOVQ  (R9), R14
+	LEAQ  (R13)(R14*1), R15
+	MOVQ  s+0(FP), BP
+	ADDQ  R15, 256(BP)
+	MOVQ  ctx+16(FP), R15
+	SUBQ  R14, 128(R15)
+	JS    error_not_enough_literals
+	CMPQ  R13, $0x00020002
+	JA    sequenceDecs_decode_56_bmi2_error_match_len_too_big
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decode_56_bmi2_match_len_ofs_ok
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decode_56_bmi2_error_match_len_ofs_mismatch
+
+sequenceDecs_decode_56_bmi2_match_len_ofs_ok:
+	ADDQ $0x18, R9
+	MOVQ ctx+16(FP), CX
+	DECQ 96(CX)
+	JNS  sequenceDecs_decode_56_bmi2_main_loop
+	MOVQ s+0(FP), CX
+	MOVQ R10, 144(CX)
+	MOVQ R11, 152(CX)
+	MOVQ R12, 160(CX)
+	MOVQ br+8(FP), CX
+	MOVQ AX, 32(CX)
+	MOVB DL, 40(CX)
+	MOVQ BX, 24(CX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decode_56_bmi2_error_match_len_ofs_mismatch:
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decode_56_bmi2_error_match_len_too_big:
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_executeSimple_amd64(ctx *executeAsmContext) bool
+// Requires: SSE
+TEXT ·sequenceDecs_executeSimple_amd64(SB), $8-9
+	MOVQ  ctx+0(FP), R10
+	MOVQ  8(R10), CX
+	TESTQ CX, CX
+	JZ    empty_seqs
+	MOVQ  (R10), AX
+	MOVQ  24(R10), DX
+	MOVQ  32(R10), BX
+	MOVQ  80(R10), SI
+	MOVQ  104(R10), DI
+	MOVQ  120(R10), R8
+	MOVQ  56(R10), R9
+	MOVQ  64(R10), R10
+	ADDQ  R10, R9
+
+	// seqsBase += 24 * seqIndex
+	LEAQ (DX)(DX*2), R11
+	SHLQ $0x03, R11
+	ADDQ R11, AX
+
+	// outBase += outPosition
+	ADDQ DI, BX
+
+main_loop:
+	MOVQ (AX), R11
+	MOVQ 16(AX), R12
+	MOVQ 8(AX), R13
+
+	// Copy literals
+	TESTQ R11, R11
+	JZ    check_offset
+	XORQ  R14, R14
+
+copy_1:
+	MOVUPS (SI)(R14*1), X0
+	MOVUPS X0, (BX)(R14*1)
+	ADDQ   $0x10, R14
+	CMPQ   R14, R11
+	JB     copy_1
+	ADDQ   R11, SI
+	ADDQ   R11, BX
+	ADDQ   R11, DI
+
+	// Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize)
+check_offset:
+	LEAQ (DI)(R10*1), R11
+	CMPQ R12, R11
+	JG   error_match_off_too_big
+	CMPQ R12, R8
+	JG   error_match_off_too_big
+
+	// Copy match from history
+	MOVQ R12, R11
+	SUBQ DI, R11
+	JLS  copy_match
+	MOVQ R9, R14
+	SUBQ R11, R14
+	CMPQ R13, R11
+	JG   copy_all_from_history
+	MOVQ R13, R11
+	SUBQ $0x10, R11
+	JB   copy_4_small
+
+copy_4_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (BX)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, BX
+	SUBQ   $0x10, R11
+	JAE    copy_4_loop
+	LEAQ   16(R14)(R11*1), R14
+	LEAQ   16(BX)(R11*1), BX
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(BX)
+	JMP    copy_4_end
+
+copy_4_small:
+	CMPQ R13, $0x03
+	JE   copy_4_move_3
+	CMPQ R13, $0x08
+	JB   copy_4_move_4through7
+	JMP  copy_4_move_8through16
+
+copy_4_move_3:
+	MOVW (R14), R11
+	MOVB 2(R14), R12
+	MOVW R11, (BX)
+	MOVB R12, 2(BX)
+	ADDQ R13, R14
+	ADDQ R13, BX
+	JMP  copy_4_end
+
+copy_4_move_4through7:
+	MOVL (R14), R11
+	MOVL -4(R14)(R13*1), R12
+	MOVL R11, (BX)
+	MOVL R12, -4(BX)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, BX
+	JMP  copy_4_end
+
+copy_4_move_8through16:
+	MOVQ (R14), R11
+	MOVQ -8(R14)(R13*1), R12
+	MOVQ R11, (BX)
+	MOVQ R12, -8(BX)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, BX
+
+copy_4_end:
+	ADDQ R13, DI
+	ADDQ $0x18, AX
+	INCQ DX
+	CMPQ DX, CX
+	JB   main_loop
+	JMP  loop_finished
+
+copy_all_from_history:
+	MOVQ R11, R15
+	SUBQ $0x10, R15
+	JB   copy_5_small
+
+copy_5_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (BX)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, BX
+	SUBQ   $0x10, R15
+	JAE    copy_5_loop
+	LEAQ   16(R14)(R15*1), R14
+	LEAQ   16(BX)(R15*1), BX
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(BX)
+	JMP    copy_5_end
+
+copy_5_small:
+	CMPQ R11, $0x03
+	JE   copy_5_move_3
+	JB   copy_5_move_1or2
+	CMPQ R11, $0x08
+	JB   copy_5_move_4through7
+	JMP  copy_5_move_8through16
+
+copy_5_move_1or2:
+	MOVB (R14), R15
+	MOVB -1(R14)(R11*1), BP
+	MOVB R15, (BX)
+	MOVB BP, -1(BX)(R11*1)
+	ADDQ R11, R14
+	ADDQ R11, BX
+	JMP  copy_5_end
+
+copy_5_move_3:
+	MOVW (R14), R15
+	MOVB 2(R14), BP
+	MOVW R15, (BX)
+	MOVB BP, 2(BX)
+	ADDQ R11, R14
+	ADDQ R11, BX
+	JMP  copy_5_end
+
+copy_5_move_4through7:
+	MOVL (R14), R15
+	MOVL -4(R14)(R11*1), BP
+	MOVL R15, (BX)
+	MOVL BP, -4(BX)(R11*1)
+	ADDQ R11, R14
+	ADDQ R11, BX
+	JMP  copy_5_end
+
+copy_5_move_8through16:
+	MOVQ (R14), R15
+	MOVQ -8(R14)(R11*1), BP
+	MOVQ R15, (BX)
+	MOVQ BP, -8(BX)(R11*1)
+	ADDQ R11, R14
+	ADDQ R11, BX
+
+copy_5_end:
+	ADDQ R11, DI
+	SUBQ R11, R13
+
+	// Copy match from the current buffer
+copy_match:
+	MOVQ BX, R11
+	SUBQ R12, R11
+
+	// ml <= mo
+	CMPQ R13, R12
+	JA   copy_overlapping_match
+
+	// Copy non-overlapping match
+	ADDQ R13, DI
+	MOVQ BX, R12
+	ADDQ R13, BX
+
+copy_2:
+	MOVUPS (R11), X0
+	MOVUPS X0, (R12)
+	ADDQ   $0x10, R11
+	ADDQ   $0x10, R12
+	SUBQ   $0x10, R13
+	JHI    copy_2
+	JMP    handle_loop
+
+	// Copy overlapping match
+copy_overlapping_match:
+	ADDQ R13, DI
+
+copy_slow_3:
+	MOVB (R11), R12
+	MOVB R12, (BX)
+	INCQ R11
+	INCQ BX
+	DECQ R13
+	JNZ  copy_slow_3
+
+handle_loop:
+	ADDQ $0x18, AX
+	INCQ DX
+	CMPQ DX, CX
+	JB   main_loop
+
+loop_finished:
+	// Return value
+	MOVB $0x01, ret+8(FP)
+
+	// Update the context
+	MOVQ ctx+0(FP), AX
+	MOVQ DX, 24(AX)
+	MOVQ DI, 104(AX)
+	MOVQ 80(AX), CX
+	SUBQ CX, SI
+	MOVQ SI, 112(AX)
+	RET
+
+error_match_off_too_big:
+	// Return value
+	MOVB $0x00, ret+8(FP)
+
+	// Update the context
+	MOVQ ctx+0(FP), AX
+	MOVQ DX, 24(AX)
+	MOVQ DI, 104(AX)
+	MOVQ 80(AX), CX
+	SUBQ CX, SI
+	MOVQ SI, 112(AX)
+	RET
+
+empty_seqs:
+	// Return value
+	MOVB $0x01, ret+8(FP)
+	RET
+
+// func sequenceDecs_executeSimple_safe_amd64(ctx *executeAsmContext) bool
+// Requires: SSE
+TEXT ·sequenceDecs_executeSimple_safe_amd64(SB), $8-9
+	MOVQ  ctx+0(FP), R10
+	MOVQ  8(R10), CX
+	TESTQ CX, CX
+	JZ    empty_seqs
+	MOVQ  (R10), AX
+	MOVQ  24(R10), DX
+	MOVQ  32(R10), BX
+	MOVQ  80(R10), SI
+	MOVQ  104(R10), DI
+	MOVQ  120(R10), R8
+	MOVQ  56(R10), R9
+	MOVQ  64(R10), R10
+	ADDQ  R10, R9
+
+	// seqsBase += 24 * seqIndex
+	LEAQ (DX)(DX*2), R11
+	SHLQ $0x03, R11
+	ADDQ R11, AX
+
+	// outBase += outPosition
+	ADDQ DI, BX
+
+main_loop:
+	MOVQ (AX), R11
+	MOVQ 16(AX), R12
+	MOVQ 8(AX), R13
+
+	// Copy literals
+	TESTQ R11, R11
+	JZ    check_offset
+	MOVQ  R11, R14
+	SUBQ  $0x10, R14
+	JB    copy_1_small
+
+copy_1_loop:
+	MOVUPS (SI), X0
+	MOVUPS X0, (BX)
+	ADDQ   $0x10, SI
+	ADDQ   $0x10, BX
+	SUBQ   $0x10, R14
+	JAE    copy_1_loop
+	LEAQ   16(SI)(R14*1), SI
+	LEAQ   16(BX)(R14*1), BX
+	MOVUPS -16(SI), X0
+	MOVUPS X0, -16(BX)
+	JMP    copy_1_end
+
+copy_1_small:
+	CMPQ R11, $0x03
+	JE   copy_1_move_3
+	JB   copy_1_move_1or2
+	CMPQ R11, $0x08
+	JB   copy_1_move_4through7
+	JMP  copy_1_move_8through16
+
+copy_1_move_1or2:
+	MOVB (SI), R14
+	MOVB -1(SI)(R11*1), R15
+	MOVB R14, (BX)
+	MOVB R15, -1(BX)(R11*1)
+	ADDQ R11, SI
+	ADDQ R11, BX
+	JMP  copy_1_end
+
+copy_1_move_3:
+	MOVW (SI), R14
+	MOVB 2(SI), R15
+	MOVW R14, (BX)
+	MOVB R15, 2(BX)
+	ADDQ R11, SI
+	ADDQ R11, BX
+	JMP  copy_1_end
+
+copy_1_move_4through7:
+	MOVL (SI), R14
+	MOVL -4(SI)(R11*1), R15
+	MOVL R14, (BX)
+	MOVL R15, -4(BX)(R11*1)
+	ADDQ R11, SI
+	ADDQ R11, BX
+	JMP  copy_1_end
+
+copy_1_move_8through16:
+	MOVQ (SI), R14
+	MOVQ -8(SI)(R11*1), R15
+	MOVQ R14, (BX)
+	MOVQ R15, -8(BX)(R11*1)
+	ADDQ R11, SI
+	ADDQ R11, BX
+
+copy_1_end:
+	ADDQ R11, DI
+
+	// Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize)
+check_offset:
+	LEAQ (DI)(R10*1), R11
+	CMPQ R12, R11
+	JG   error_match_off_too_big
+	CMPQ R12, R8
+	JG   error_match_off_too_big
+
+	// Copy match from history
+	MOVQ R12, R11
+	SUBQ DI, R11
+	JLS  copy_match
+	MOVQ R9, R14
+	SUBQ R11, R14
+	CMPQ R13, R11
+	JG   copy_all_from_history
+	MOVQ R13, R11
+	SUBQ $0x10, R11
+	JB   copy_4_small
+
+copy_4_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (BX)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, BX
+	SUBQ   $0x10, R11
+	JAE    copy_4_loop
+	LEAQ   16(R14)(R11*1), R14
+	LEAQ   16(BX)(R11*1), BX
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(BX)
+	JMP    copy_4_end
+
+copy_4_small:
+	CMPQ R13, $0x03
+	JE   copy_4_move_3
+	CMPQ R13, $0x08
+	JB   copy_4_move_4through7
+	JMP  copy_4_move_8through16
+
+copy_4_move_3:
+	MOVW (R14), R11
+	MOVB 2(R14), R12
+	MOVW R11, (BX)
+	MOVB R12, 2(BX)
+	ADDQ R13, R14
+	ADDQ R13, BX
+	JMP  copy_4_end
+
+copy_4_move_4through7:
+	MOVL (R14), R11
+	MOVL -4(R14)(R13*1), R12
+	MOVL R11, (BX)
+	MOVL R12, -4(BX)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, BX
+	JMP  copy_4_end
+
+copy_4_move_8through16:
+	MOVQ (R14), R11
+	MOVQ -8(R14)(R13*1), R12
+	MOVQ R11, (BX)
+	MOVQ R12, -8(BX)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, BX
+
+copy_4_end:
+	ADDQ R13, DI
+	ADDQ $0x18, AX
+	INCQ DX
+	CMPQ DX, CX
+	JB   main_loop
+	JMP  loop_finished
+
+copy_all_from_history:
+	MOVQ R11, R15
+	SUBQ $0x10, R15
+	JB   copy_5_small
+
+copy_5_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (BX)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, BX
+	SUBQ   $0x10, R15
+	JAE    copy_5_loop
+	LEAQ   16(R14)(R15*1), R14
+	LEAQ   16(BX)(R15*1), BX
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(BX)
+	JMP    copy_5_end
+
+copy_5_small:
+	CMPQ R11, $0x03
+	JE   copy_5_move_3
+	JB   copy_5_move_1or2
+	CMPQ R11, $0x08
+	JB   copy_5_move_4through7
+	JMP  copy_5_move_8through16
+
+copy_5_move_1or2:
+	MOVB (R14), R15
+	MOVB -1(R14)(R11*1), BP
+	MOVB R15, (BX)
+	MOVB BP, -1(BX)(R11*1)
+	ADDQ R11, R14
+	ADDQ R11, BX
+	JMP  copy_5_end
+
+copy_5_move_3:
+	MOVW (R14), R15
+	MOVB 2(R14), BP
+	MOVW R15, (BX)
+	MOVB BP, 2(BX)
+	ADDQ R11, R14
+	ADDQ R11, BX
+	JMP  copy_5_end
+
+copy_5_move_4through7:
+	MOVL (R14), R15
+	MOVL -4(R14)(R11*1), BP
+	MOVL R15, (BX)
+	MOVL BP, -4(BX)(R11*1)
+	ADDQ R11, R14
+	ADDQ R11, BX
+	JMP  copy_5_end
+
+copy_5_move_8through16:
+	MOVQ (R14), R15
+	MOVQ -8(R14)(R11*1), BP
+	MOVQ R15, (BX)
+	MOVQ BP, -8(BX)(R11*1)
+	ADDQ R11, R14
+	ADDQ R11, BX
+
+copy_5_end:
+	ADDQ R11, DI
+	SUBQ R11, R13
+
+	// Copy match from the current buffer
+copy_match:
+	MOVQ BX, R11
+	SUBQ R12, R11
+
+	// ml <= mo
+	CMPQ R13, R12
+	JA   copy_overlapping_match
+
+	// Copy non-overlapping match
+	ADDQ R13, DI
+	MOVQ R13, R12
+	SUBQ $0x10, R12
+	JB   copy_2_small
+
+copy_2_loop:
+	MOVUPS (R11), X0
+	MOVUPS X0, (BX)
+	ADDQ   $0x10, R11
+	ADDQ   $0x10, BX
+	SUBQ   $0x10, R12
+	JAE    copy_2_loop
+	LEAQ   16(R11)(R12*1), R11
+	LEAQ   16(BX)(R12*1), BX
+	MOVUPS -16(R11), X0
+	MOVUPS X0, -16(BX)
+	JMP    copy_2_end
+
+copy_2_small:
+	CMPQ R13, $0x03
+	JE   copy_2_move_3
+	JB   copy_2_move_1or2
+	CMPQ R13, $0x08
+	JB   copy_2_move_4through7
+	JMP  copy_2_move_8through16
+
+copy_2_move_1or2:
+	MOVB (R11), R12
+	MOVB -1(R11)(R13*1), R14
+	MOVB R12, (BX)
+	MOVB R14, -1(BX)(R13*1)
+	ADDQ R13, R11
+	ADDQ R13, BX
+	JMP  copy_2_end
+
+copy_2_move_3:
+	MOVW (R11), R12
+	MOVB 2(R11), R14
+	MOVW R12, (BX)
+	MOVB R14, 2(BX)
+	ADDQ R13, R11
+	ADDQ R13, BX
+	JMP  copy_2_end
+
+copy_2_move_4through7:
+	MOVL (R11), R12
+	MOVL -4(R11)(R13*1), R14
+	MOVL R12, (BX)
+	MOVL R14, -4(BX)(R13*1)
+	ADDQ R13, R11
+	ADDQ R13, BX
+	JMP  copy_2_end
+
+copy_2_move_8through16:
+	MOVQ (R11), R12
+	MOVQ -8(R11)(R13*1), R14
+	MOVQ R12, (BX)
+	MOVQ R14, -8(BX)(R13*1)
+	ADDQ R13, R11
+	ADDQ R13, BX
+
+copy_2_end:
+	JMP handle_loop
+
+	// Copy overlapping match
+copy_overlapping_match:
+	ADDQ R13, DI
+
+copy_slow_3:
+	MOVB (R11), R12
+	MOVB R12, (BX)
+	INCQ R11
+	INCQ BX
+	DECQ R13
+	JNZ  copy_slow_3
+
+handle_loop:
+	ADDQ $0x18, AX
+	INCQ DX
+	CMPQ DX, CX
+	JB   main_loop
+
+loop_finished:
+	// Return value
+	MOVB $0x01, ret+8(FP)
+
+	// Update the context
+	MOVQ ctx+0(FP), AX
+	MOVQ DX, 24(AX)
+	MOVQ DI, 104(AX)
+	MOVQ 80(AX), CX
+	SUBQ CX, SI
+	MOVQ SI, 112(AX)
+	RET
+
+error_match_off_too_big:
+	// Return value
+	MOVB $0x00, ret+8(FP)
+
+	// Update the context
+	MOVQ ctx+0(FP), AX
+	MOVQ DX, 24(AX)
+	MOVQ DI, 104(AX)
+	MOVQ 80(AX), CX
+	SUBQ CX, SI
+	MOVQ SI, 112(AX)
+	RET
+
+empty_seqs:
+	// Return value
+	MOVB $0x01, ret+8(FP)
+	RET
+
+// func sequenceDecs_decodeSync_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+// Requires: CMOV, SSE
+TEXT ·sequenceDecs_decodeSync_amd64(SB), $64-32
+	MOVQ    br+8(FP), AX
+	MOVQ    32(AX), DX
+	MOVBQZX 40(AX), BX
+	MOVQ    24(AX), SI
+	MOVQ    (AX), AX
+	ADDQ    SI, AX
+	MOVQ    AX, (SP)
+	MOVQ    ctx+16(FP), AX
+	MOVQ    72(AX), DI
+	MOVQ    80(AX), R8
+	MOVQ    88(AX), R9
+	XORQ    CX, CX
+	MOVQ    CX, 8(SP)
+	MOVQ    CX, 16(SP)
+	MOVQ    CX, 24(SP)
+	MOVQ    112(AX), R10
+	MOVQ    128(AX), CX
+	MOVQ    CX, 32(SP)
+	MOVQ    144(AX), R11
+	MOVQ    136(AX), R12
+	MOVQ    200(AX), CX
+	MOVQ    CX, 56(SP)
+	MOVQ    176(AX), CX
+	MOVQ    CX, 48(SP)
+	MOVQ    184(AX), AX
+	MOVQ    AX, 40(SP)
+	MOVQ    40(SP), AX
+	ADDQ    AX, 48(SP)
+
+	// Calculate poiter to s.out[cap(s.out)] (a past-end pointer)
+	ADDQ R10, 32(SP)
+
+	// outBase += outPosition
+	ADDQ R12, R10
+
+sequenceDecs_decodeSync_amd64_main_loop:
+	MOVQ (SP), R13
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decodeSync_amd64_fill_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R13
+	MOVQ (R13), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decodeSync_amd64_fill_end
+
+sequenceDecs_decodeSync_amd64_fill_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decodeSync_amd64_fill_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decodeSync_amd64_fill_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R13), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decodeSync_amd64_fill_byte_by_byte
+
+sequenceDecs_decodeSync_amd64_fill_end:
+	// Update offset
+	MOVQ  R9, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R14
+	SHLQ  CL, R14
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decodeSync_amd64_of_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decodeSync_amd64_of_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decodeSync_amd64_of_update_zero
+	NEGQ  CX
+	SHRQ  CL, R14
+	ADDQ  R14, AX
+
+sequenceDecs_decodeSync_amd64_of_update_zero:
+	MOVQ AX, 8(SP)
+
+	// Update match length
+	MOVQ  R8, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R14
+	SHLQ  CL, R14
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decodeSync_amd64_ml_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decodeSync_amd64_ml_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decodeSync_amd64_ml_update_zero
+	NEGQ  CX
+	SHRQ  CL, R14
+	ADDQ  R14, AX
+
+sequenceDecs_decodeSync_amd64_ml_update_zero:
+	MOVQ AX, 16(SP)
+
+	// Fill bitreader to have enough for the remaining
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R13
+	MOVQ (R13), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decodeSync_amd64_fill_2_end
+
+sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decodeSync_amd64_fill_2_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decodeSync_amd64_fill_2_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R13), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte
+
+sequenceDecs_decodeSync_amd64_fill_2_end:
+	// Update literal length
+	MOVQ  DI, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R14
+	SHLQ  CL, R14
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decodeSync_amd64_ll_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decodeSync_amd64_ll_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decodeSync_amd64_ll_update_zero
+	NEGQ  CX
+	SHRQ  CL, R14
+	ADDQ  R14, AX
+
+sequenceDecs_decodeSync_amd64_ll_update_zero:
+	MOVQ AX, 24(SP)
+
+	// Fill bitreader for state updates
+	MOVQ    R13, (SP)
+	MOVQ    R9, AX
+	SHRQ    $0x08, AX
+	MOVBQZX AL, AX
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decodeSync_amd64_skip_update
+
+	// Update Literal Length State
+	MOVBQZX DI, R13
+	SHRQ    $0x10, DI
+	MOVWQZX DI, DI
+	LEAQ    (BX)(R13*1), CX
+	MOVQ    DX, R14
+	MOVQ    CX, BX
+	ROLQ    CL, R14
+	MOVL    $0x00000001, R15
+	MOVB    R13, CL
+	SHLL    CL, R15
+	DECL    R15
+	ANDQ    R15, R14
+	ADDQ    R14, DI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Match Length State
+	MOVBQZX R8, R13
+	SHRQ    $0x10, R8
+	MOVWQZX R8, R8
+	LEAQ    (BX)(R13*1), CX
+	MOVQ    DX, R14
+	MOVQ    CX, BX
+	ROLQ    CL, R14
+	MOVL    $0x00000001, R15
+	MOVB    R13, CL
+	SHLL    CL, R15
+	DECL    R15
+	ANDQ    R15, R14
+	ADDQ    R14, R8
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Offset State
+	MOVBQZX R9, R13
+	SHRQ    $0x10, R9
+	MOVWQZX R9, R9
+	LEAQ    (BX)(R13*1), CX
+	MOVQ    DX, R14
+	MOVQ    CX, BX
+	ROLQ    CL, R14
+	MOVL    $0x00000001, R15
+	MOVB    R13, CL
+	SHLL    CL, R15
+	DECL    R15
+	ANDQ    R15, R14
+	ADDQ    R14, R9
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R9*8), R9
+
+sequenceDecs_decodeSync_amd64_skip_update:
+	// Adjust offset
+	MOVQ   s+0(FP), CX
+	MOVQ   8(SP), R13
+	CMPQ   AX, $0x01
+	JBE    sequenceDecs_decodeSync_amd64_adjust_offsetB_1_or_0
+	MOVUPS 144(CX), X0
+	MOVQ   R13, 144(CX)
+	MOVUPS X0, 152(CX)
+	JMP    sequenceDecs_decodeSync_amd64_after_adjust
+
+sequenceDecs_decodeSync_amd64_adjust_offsetB_1_or_0:
+	CMPQ 24(SP), $0x00000000
+	JNE  sequenceDecs_decodeSync_amd64_adjust_offset_maybezero
+	INCQ R13
+	JMP  sequenceDecs_decodeSync_amd64_adjust_offset_nonzero
+
+sequenceDecs_decodeSync_amd64_adjust_offset_maybezero:
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_amd64_adjust_offset_nonzero
+	MOVQ  144(CX), R13
+	JMP   sequenceDecs_decodeSync_amd64_after_adjust
+
+sequenceDecs_decodeSync_amd64_adjust_offset_nonzero:
+	MOVQ    R13, AX
+	XORQ    R14, R14
+	MOVQ    $-1, R15
+	CMPQ    R13, $0x03
+	CMOVQEQ R14, AX
+	CMOVQEQ R15, R14
+	ADDQ    144(CX)(AX*8), R14
+	JNZ     sequenceDecs_decodeSync_amd64_adjust_temp_valid
+	MOVQ    $0x00000001, R14
+
+sequenceDecs_decodeSync_amd64_adjust_temp_valid:
+	CMPQ R13, $0x01
+	JZ   sequenceDecs_decodeSync_amd64_adjust_skip
+	MOVQ 152(CX), AX
+	MOVQ AX, 160(CX)
+
+sequenceDecs_decodeSync_amd64_adjust_skip:
+	MOVQ 144(CX), AX
+	MOVQ AX, 152(CX)
+	MOVQ R14, 144(CX)
+	MOVQ R14, R13
+
+sequenceDecs_decodeSync_amd64_after_adjust:
+	MOVQ R13, 8(SP)
+
+	// Check values
+	MOVQ  16(SP), AX
+	MOVQ  24(SP), CX
+	LEAQ  (AX)(CX*1), R14
+	MOVQ  s+0(FP), R15
+	ADDQ  R14, 256(R15)
+	MOVQ  ctx+16(FP), R14
+	SUBQ  CX, 104(R14)
+	JS    error_not_enough_literals
+	CMPQ  AX, $0x00020002
+	JA    sequenceDecs_decodeSync_amd64_error_match_len_too_big
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_amd64_match_len_ofs_ok
+	TESTQ AX, AX
+	JNZ   sequenceDecs_decodeSync_amd64_error_match_len_ofs_mismatch
+
+sequenceDecs_decodeSync_amd64_match_len_ofs_ok:
+	MOVQ 24(SP), AX
+	MOVQ 8(SP), CX
+	MOVQ 16(SP), R13
+
+	// Check if we have enough space in s.out
+	LEAQ (AX)(R13*1), R14
+	ADDQ R10, R14
+	CMPQ R14, 32(SP)
+	JA   error_not_enough_space
+
+	// Copy literals
+	TESTQ AX, AX
+	JZ    check_offset
+	XORQ  R14, R14
+
+copy_1:
+	MOVUPS (R11)(R14*1), X0
+	MOVUPS X0, (R10)(R14*1)
+	ADDQ   $0x10, R14
+	CMPQ   R14, AX
+	JB     copy_1
+	ADDQ   AX, R11
+	ADDQ   AX, R10
+	ADDQ   AX, R12
+
+	// Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize)
+check_offset:
+	MOVQ R12, AX
+	ADDQ 40(SP), AX
+	CMPQ CX, AX
+	JG   error_match_off_too_big
+	CMPQ CX, 56(SP)
+	JG   error_match_off_too_big
+
+	// Copy match from history
+	MOVQ CX, AX
+	SUBQ R12, AX
+	JLS  copy_match
+	MOVQ 48(SP), R14
+	SUBQ AX, R14
+	CMPQ R13, AX
+	JG   copy_all_from_history
+	MOVQ R13, AX
+	SUBQ $0x10, AX
+	JB   copy_4_small
+
+copy_4_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R10)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R10
+	SUBQ   $0x10, AX
+	JAE    copy_4_loop
+	LEAQ   16(R14)(AX*1), R14
+	LEAQ   16(R10)(AX*1), R10
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R10)
+	JMP    copy_4_end
+
+copy_4_small:
+	CMPQ R13, $0x03
+	JE   copy_4_move_3
+	CMPQ R13, $0x08
+	JB   copy_4_move_4through7
+	JMP  copy_4_move_8through16
+
+copy_4_move_3:
+	MOVW (R14), AX
+	MOVB 2(R14), CL
+	MOVW AX, (R10)
+	MOVB CL, 2(R10)
+	ADDQ R13, R14
+	ADDQ R13, R10
+	JMP  copy_4_end
+
+copy_4_move_4through7:
+	MOVL (R14), AX
+	MOVL -4(R14)(R13*1), CX
+	MOVL AX, (R10)
+	MOVL CX, -4(R10)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R10
+	JMP  copy_4_end
+
+copy_4_move_8through16:
+	MOVQ (R14), AX
+	MOVQ -8(R14)(R13*1), CX
+	MOVQ AX, (R10)
+	MOVQ CX, -8(R10)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R10
+
+copy_4_end:
+	ADDQ R13, R12
+	JMP  handle_loop
+	JMP loop_finished
+
+copy_all_from_history:
+	MOVQ AX, R15
+	SUBQ $0x10, R15
+	JB   copy_5_small
+
+copy_5_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R10)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R10
+	SUBQ   $0x10, R15
+	JAE    copy_5_loop
+	LEAQ   16(R14)(R15*1), R14
+	LEAQ   16(R10)(R15*1), R10
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R10)
+	JMP    copy_5_end
+
+copy_5_small:
+	CMPQ AX, $0x03
+	JE   copy_5_move_3
+	JB   copy_5_move_1or2
+	CMPQ AX, $0x08
+	JB   copy_5_move_4through7
+	JMP  copy_5_move_8through16
+
+copy_5_move_1or2:
+	MOVB (R14), R15
+	MOVB -1(R14)(AX*1), BP
+	MOVB R15, (R10)
+	MOVB BP, -1(R10)(AX*1)
+	ADDQ AX, R14
+	ADDQ AX, R10
+	JMP  copy_5_end
+
+copy_5_move_3:
+	MOVW (R14), R15
+	MOVB 2(R14), BP
+	MOVW R15, (R10)
+	MOVB BP, 2(R10)
+	ADDQ AX, R14
+	ADDQ AX, R10
+	JMP  copy_5_end
+
+copy_5_move_4through7:
+	MOVL (R14), R15
+	MOVL -4(R14)(AX*1), BP
+	MOVL R15, (R10)
+	MOVL BP, -4(R10)(AX*1)
+	ADDQ AX, R14
+	ADDQ AX, R10
+	JMP  copy_5_end
+
+copy_5_move_8through16:
+	MOVQ (R14), R15
+	MOVQ -8(R14)(AX*1), BP
+	MOVQ R15, (R10)
+	MOVQ BP, -8(R10)(AX*1)
+	ADDQ AX, R14
+	ADDQ AX, R10
+
+copy_5_end:
+	ADDQ AX, R12
+	SUBQ AX, R13
+
+	// Copy match from the current buffer
+copy_match:
+	MOVQ R10, AX
+	SUBQ CX, AX
+
+	// ml <= mo
+	CMPQ R13, CX
+	JA   copy_overlapping_match
+
+	// Copy non-overlapping match
+	ADDQ R13, R12
+	MOVQ R10, CX
+	ADDQ R13, R10
+
+copy_2:
+	MOVUPS (AX), X0
+	MOVUPS X0, (CX)
+	ADDQ   $0x10, AX
+	ADDQ   $0x10, CX
+	SUBQ   $0x10, R13
+	JHI    copy_2
+	JMP    handle_loop
+
+	// Copy overlapping match
+copy_overlapping_match:
+	ADDQ R13, R12
+
+copy_slow_3:
+	MOVB (AX), CL
+	MOVB CL, (R10)
+	INCQ AX
+	INCQ R10
+	DECQ R13
+	JNZ  copy_slow_3
+
+handle_loop:
+	MOVQ ctx+16(FP), AX
+	DECQ 96(AX)
+	JNS  sequenceDecs_decodeSync_amd64_main_loop
+
+loop_finished:
+	MOVQ br+8(FP), AX
+	MOVQ DX, 32(AX)
+	MOVB BL, 40(AX)
+	MOVQ SI, 24(AX)
+
+	// Update the context
+	MOVQ ctx+16(FP), AX
+	MOVQ R12, 136(AX)
+	MOVQ 144(AX), CX
+	SUBQ CX, R11
+	MOVQ R11, 168(AX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decodeSync_amd64_error_match_len_ofs_mismatch:
+	MOVQ 16(SP), AX
+	MOVQ ctx+16(FP), CX
+	MOVQ AX, 216(CX)
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decodeSync_amd64_error_match_len_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+error_match_off_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 8(SP), CX
+	MOVQ CX, 224(AX)
+	MOVQ R12, 136(AX)
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+error_not_enough_space:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ R12, 136(AX)
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_decodeSync_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+// Requires: BMI, BMI2, CMOV, SSE
+TEXT ·sequenceDecs_decodeSync_bmi2(SB), $64-32
+	MOVQ    br+8(FP), CX
+	MOVQ    32(CX), AX
+	MOVBQZX 40(CX), DX
+	MOVQ    24(CX), BX
+	MOVQ    (CX), CX
+	ADDQ    BX, CX
+	MOVQ    CX, (SP)
+	MOVQ    ctx+16(FP), CX
+	MOVQ    72(CX), SI
+	MOVQ    80(CX), DI
+	MOVQ    88(CX), R8
+	XORQ    R9, R9
+	MOVQ    R9, 8(SP)
+	MOVQ    R9, 16(SP)
+	MOVQ    R9, 24(SP)
+	MOVQ    112(CX), R9
+	MOVQ    128(CX), R10
+	MOVQ    R10, 32(SP)
+	MOVQ    144(CX), R10
+	MOVQ    136(CX), R11
+	MOVQ    200(CX), R12
+	MOVQ    R12, 56(SP)
+	MOVQ    176(CX), R12
+	MOVQ    R12, 48(SP)
+	MOVQ    184(CX), CX
+	MOVQ    CX, 40(SP)
+	MOVQ    40(SP), CX
+	ADDQ    CX, 48(SP)
+
+	// Calculate poiter to s.out[cap(s.out)] (a past-end pointer)
+	ADDQ R9, 32(SP)
+
+	// outBase += outPosition
+	ADDQ R11, R9
+
+sequenceDecs_decodeSync_bmi2_main_loop:
+	MOVQ (SP), R12
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decodeSync_bmi2_fill_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R12
+	MOVQ (R12), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decodeSync_bmi2_fill_end
+
+sequenceDecs_decodeSync_bmi2_fill_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decodeSync_bmi2_fill_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decodeSync_bmi2_fill_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R12
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R12), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decodeSync_bmi2_fill_byte_by_byte
+
+sequenceDecs_decodeSync_bmi2_fill_end:
+	// Update offset
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, R8, R13
+	MOVQ   AX, R14
+	LEAQ   (DX)(R13*1), CX
+	ROLQ   CL, R14
+	BZHIQ  R13, R14, R14
+	MOVQ   CX, DX
+	MOVQ   R8, CX
+	SHRQ   $0x20, CX
+	ADDQ   R14, CX
+	MOVQ   CX, 8(SP)
+
+	// Update match length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, DI, R13
+	MOVQ   AX, R14
+	LEAQ   (DX)(R13*1), CX
+	ROLQ   CL, R14
+	BZHIQ  R13, R14, R14
+	MOVQ   CX, DX
+	MOVQ   DI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R14, CX
+	MOVQ   CX, 16(SP)
+
+	// Fill bitreader to have enough for the remaining
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R12
+	MOVQ (R12), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decodeSync_bmi2_fill_2_end
+
+sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decodeSync_bmi2_fill_2_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decodeSync_bmi2_fill_2_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R12
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R12), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte
+
+sequenceDecs_decodeSync_bmi2_fill_2_end:
+	// Update literal length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, SI, R13
+	MOVQ   AX, R14
+	LEAQ   (DX)(R13*1), CX
+	ROLQ   CL, R14
+	BZHIQ  R13, R14, R14
+	MOVQ   CX, DX
+	MOVQ   SI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R14, CX
+	MOVQ   CX, 24(SP)
+
+	// Fill bitreader for state updates
+	MOVQ    R12, (SP)
+	MOVQ    $0x00000808, CX
+	BEXTRQ  CX, R8, R12
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decodeSync_bmi2_skip_update
+	LEAQ    (SI)(DI*1), R13
+	ADDQ    R8, R13
+	MOVBQZX R13, R13
+	LEAQ    (DX)(R13*1), CX
+	MOVQ    AX, R14
+	MOVQ    CX, DX
+	ROLQ    CL, R14
+	BZHIQ   R13, R14, R14
+
+	// Update Offset State
+	BZHIQ  R8, R14, CX
+	SHRXQ  R8, R14, R14
+	MOVQ   $0x00001010, R13
+	BEXTRQ R13, R8, R8
+	ADDQ   CX, R8
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Match Length State
+	BZHIQ  DI, R14, CX
+	SHRXQ  DI, R14, R14
+	MOVQ   $0x00001010, R13
+	BEXTRQ R13, DI, DI
+	ADDQ   CX, DI
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Literal Length State
+	BZHIQ  SI, R14, CX
+	MOVQ   $0x00001010, R13
+	BEXTRQ R13, SI, SI
+	ADDQ   CX, SI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(SI*8), SI
+
+sequenceDecs_decodeSync_bmi2_skip_update:
+	// Adjust offset
+	MOVQ   s+0(FP), CX
+	MOVQ   8(SP), R13
+	CMPQ   R12, $0x01
+	JBE    sequenceDecs_decodeSync_bmi2_adjust_offsetB_1_or_0
+	MOVUPS 144(CX), X0
+	MOVQ   R13, 144(CX)
+	MOVUPS X0, 152(CX)
+	JMP    sequenceDecs_decodeSync_bmi2_after_adjust
+
+sequenceDecs_decodeSync_bmi2_adjust_offsetB_1_or_0:
+	CMPQ 24(SP), $0x00000000
+	JNE  sequenceDecs_decodeSync_bmi2_adjust_offset_maybezero
+	INCQ R13
+	JMP  sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero
+
+sequenceDecs_decodeSync_bmi2_adjust_offset_maybezero:
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero
+	MOVQ  144(CX), R13
+	JMP   sequenceDecs_decodeSync_bmi2_after_adjust
+
+sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero:
+	MOVQ    R13, R12
+	XORQ    R14, R14
+	MOVQ    $-1, R15
+	CMPQ    R13, $0x03
+	CMOVQEQ R14, R12
+	CMOVQEQ R15, R14
+	ADDQ    144(CX)(R12*8), R14
+	JNZ     sequenceDecs_decodeSync_bmi2_adjust_temp_valid
+	MOVQ    $0x00000001, R14
+
+sequenceDecs_decodeSync_bmi2_adjust_temp_valid:
+	CMPQ R13, $0x01
+	JZ   sequenceDecs_decodeSync_bmi2_adjust_skip
+	MOVQ 152(CX), R12
+	MOVQ R12, 160(CX)
+
+sequenceDecs_decodeSync_bmi2_adjust_skip:
+	MOVQ 144(CX), R12
+	MOVQ R12, 152(CX)
+	MOVQ R14, 144(CX)
+	MOVQ R14, R13
+
+sequenceDecs_decodeSync_bmi2_after_adjust:
+	MOVQ R13, 8(SP)
+
+	// Check values
+	MOVQ  16(SP), CX
+	MOVQ  24(SP), R12
+	LEAQ  (CX)(R12*1), R14
+	MOVQ  s+0(FP), R15
+	ADDQ  R14, 256(R15)
+	MOVQ  ctx+16(FP), R14
+	SUBQ  R12, 104(R14)
+	JS    error_not_enough_literals
+	CMPQ  CX, $0x00020002
+	JA    sequenceDecs_decodeSync_bmi2_error_match_len_too_big
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_bmi2_match_len_ofs_ok
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decodeSync_bmi2_error_match_len_ofs_mismatch
+
+sequenceDecs_decodeSync_bmi2_match_len_ofs_ok:
+	MOVQ 24(SP), CX
+	MOVQ 8(SP), R12
+	MOVQ 16(SP), R13
+
+	// Check if we have enough space in s.out
+	LEAQ (CX)(R13*1), R14
+	ADDQ R9, R14
+	CMPQ R14, 32(SP)
+	JA   error_not_enough_space
+
+	// Copy literals
+	TESTQ CX, CX
+	JZ    check_offset
+	XORQ  R14, R14
+
+copy_1:
+	MOVUPS (R10)(R14*1), X0
+	MOVUPS X0, (R9)(R14*1)
+	ADDQ   $0x10, R14
+	CMPQ   R14, CX
+	JB     copy_1
+	ADDQ   CX, R10
+	ADDQ   CX, R9
+	ADDQ   CX, R11
+
+	// Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize)
+check_offset:
+	MOVQ R11, CX
+	ADDQ 40(SP), CX
+	CMPQ R12, CX
+	JG   error_match_off_too_big
+	CMPQ R12, 56(SP)
+	JG   error_match_off_too_big
+
+	// Copy match from history
+	MOVQ R12, CX
+	SUBQ R11, CX
+	JLS  copy_match
+	MOVQ 48(SP), R14
+	SUBQ CX, R14
+	CMPQ R13, CX
+	JG   copy_all_from_history
+	MOVQ R13, CX
+	SUBQ $0x10, CX
+	JB   copy_4_small
+
+copy_4_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R9)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R9
+	SUBQ   $0x10, CX
+	JAE    copy_4_loop
+	LEAQ   16(R14)(CX*1), R14
+	LEAQ   16(R9)(CX*1), R9
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R9)
+	JMP    copy_4_end
+
+copy_4_small:
+	CMPQ R13, $0x03
+	JE   copy_4_move_3
+	CMPQ R13, $0x08
+	JB   copy_4_move_4through7
+	JMP  copy_4_move_8through16
+
+copy_4_move_3:
+	MOVW (R14), CX
+	MOVB 2(R14), R12
+	MOVW CX, (R9)
+	MOVB R12, 2(R9)
+	ADDQ R13, R14
+	ADDQ R13, R9
+	JMP  copy_4_end
+
+copy_4_move_4through7:
+	MOVL (R14), CX
+	MOVL -4(R14)(R13*1), R12
+	MOVL CX, (R9)
+	MOVL R12, -4(R9)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R9
+	JMP  copy_4_end
+
+copy_4_move_8through16:
+	MOVQ (R14), CX
+	MOVQ -8(R14)(R13*1), R12
+	MOVQ CX, (R9)
+	MOVQ R12, -8(R9)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R9
+
+copy_4_end:
+	ADDQ R13, R11
+	JMP  handle_loop
+	JMP loop_finished
+
+copy_all_from_history:
+	MOVQ CX, R15
+	SUBQ $0x10, R15
+	JB   copy_5_small
+
+copy_5_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R9)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R9
+	SUBQ   $0x10, R15
+	JAE    copy_5_loop
+	LEAQ   16(R14)(R15*1), R14
+	LEAQ   16(R9)(R15*1), R9
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R9)
+	JMP    copy_5_end
+
+copy_5_small:
+	CMPQ CX, $0x03
+	JE   copy_5_move_3
+	JB   copy_5_move_1or2
+	CMPQ CX, $0x08
+	JB   copy_5_move_4through7
+	JMP  copy_5_move_8through16
+
+copy_5_move_1or2:
+	MOVB (R14), R15
+	MOVB -1(R14)(CX*1), BP
+	MOVB R15, (R9)
+	MOVB BP, -1(R9)(CX*1)
+	ADDQ CX, R14
+	ADDQ CX, R9
+	JMP  copy_5_end
+
+copy_5_move_3:
+	MOVW (R14), R15
+	MOVB 2(R14), BP
+	MOVW R15, (R9)
+	MOVB BP, 2(R9)
+	ADDQ CX, R14
+	ADDQ CX, R9
+	JMP  copy_5_end
+
+copy_5_move_4through7:
+	MOVL (R14), R15
+	MOVL -4(R14)(CX*1), BP
+	MOVL R15, (R9)
+	MOVL BP, -4(R9)(CX*1)
+	ADDQ CX, R14
+	ADDQ CX, R9
+	JMP  copy_5_end
+
+copy_5_move_8through16:
+	MOVQ (R14), R15
+	MOVQ -8(R14)(CX*1), BP
+	MOVQ R15, (R9)
+	MOVQ BP, -8(R9)(CX*1)
+	ADDQ CX, R14
+	ADDQ CX, R9
+
+copy_5_end:
+	ADDQ CX, R11
+	SUBQ CX, R13
+
+	// Copy match from the current buffer
+copy_match:
+	MOVQ R9, CX
+	SUBQ R12, CX
+
+	// ml <= mo
+	CMPQ R13, R12
+	JA   copy_overlapping_match
+
+	// Copy non-overlapping match
+	ADDQ R13, R11
+	MOVQ R9, R12
+	ADDQ R13, R9
+
+copy_2:
+	MOVUPS (CX), X0
+	MOVUPS X0, (R12)
+	ADDQ   $0x10, CX
+	ADDQ   $0x10, R12
+	SUBQ   $0x10, R13
+	JHI    copy_2
+	JMP    handle_loop
+
+	// Copy overlapping match
+copy_overlapping_match:
+	ADDQ R13, R11
+
+copy_slow_3:
+	MOVB (CX), R12
+	MOVB R12, (R9)
+	INCQ CX
+	INCQ R9
+	DECQ R13
+	JNZ  copy_slow_3
+
+handle_loop:
+	MOVQ ctx+16(FP), CX
+	DECQ 96(CX)
+	JNS  sequenceDecs_decodeSync_bmi2_main_loop
+
+loop_finished:
+	MOVQ br+8(FP), CX
+	MOVQ AX, 32(CX)
+	MOVB DL, 40(CX)
+	MOVQ BX, 24(CX)
+
+	// Update the context
+	MOVQ ctx+16(FP), AX
+	MOVQ R11, 136(AX)
+	MOVQ 144(AX), CX
+	SUBQ CX, R10
+	MOVQ R10, 168(AX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decodeSync_bmi2_error_match_len_ofs_mismatch:
+	MOVQ 16(SP), AX
+	MOVQ ctx+16(FP), CX
+	MOVQ AX, 216(CX)
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decodeSync_bmi2_error_match_len_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+error_match_off_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 8(SP), CX
+	MOVQ CX, 224(AX)
+	MOVQ R11, 136(AX)
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+error_not_enough_space:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ R11, 136(AX)
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_decodeSync_safe_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+// Requires: CMOV, SSE
+TEXT ·sequenceDecs_decodeSync_safe_amd64(SB), $64-32
+	MOVQ    br+8(FP), AX
+	MOVQ    32(AX), DX
+	MOVBQZX 40(AX), BX
+	MOVQ    24(AX), SI
+	MOVQ    (AX), AX
+	ADDQ    SI, AX
+	MOVQ    AX, (SP)
+	MOVQ    ctx+16(FP), AX
+	MOVQ    72(AX), DI
+	MOVQ    80(AX), R8
+	MOVQ    88(AX), R9
+	XORQ    CX, CX
+	MOVQ    CX, 8(SP)
+	MOVQ    CX, 16(SP)
+	MOVQ    CX, 24(SP)
+	MOVQ    112(AX), R10
+	MOVQ    128(AX), CX
+	MOVQ    CX, 32(SP)
+	MOVQ    144(AX), R11
+	MOVQ    136(AX), R12
+	MOVQ    200(AX), CX
+	MOVQ    CX, 56(SP)
+	MOVQ    176(AX), CX
+	MOVQ    CX, 48(SP)
+	MOVQ    184(AX), AX
+	MOVQ    AX, 40(SP)
+	MOVQ    40(SP), AX
+	ADDQ    AX, 48(SP)
+
+	// Calculate poiter to s.out[cap(s.out)] (a past-end pointer)
+	ADDQ R10, 32(SP)
+
+	// outBase += outPosition
+	ADDQ R12, R10
+
+sequenceDecs_decodeSync_safe_amd64_main_loop:
+	MOVQ (SP), R13
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R13
+	MOVQ (R13), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decodeSync_safe_amd64_fill_end
+
+sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decodeSync_safe_amd64_fill_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decodeSync_safe_amd64_fill_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R13), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte
+
+sequenceDecs_decodeSync_safe_amd64_fill_end:
+	// Update offset
+	MOVQ  R9, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R14
+	SHLQ  CL, R14
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decodeSync_safe_amd64_of_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decodeSync_safe_amd64_of_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decodeSync_safe_amd64_of_update_zero
+	NEGQ  CX
+	SHRQ  CL, R14
+	ADDQ  R14, AX
+
+sequenceDecs_decodeSync_safe_amd64_of_update_zero:
+	MOVQ AX, 8(SP)
+
+	// Update match length
+	MOVQ  R8, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R14
+	SHLQ  CL, R14
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decodeSync_safe_amd64_ml_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decodeSync_safe_amd64_ml_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decodeSync_safe_amd64_ml_update_zero
+	NEGQ  CX
+	SHRQ  CL, R14
+	ADDQ  R14, AX
+
+sequenceDecs_decodeSync_safe_amd64_ml_update_zero:
+	MOVQ AX, 16(SP)
+
+	// Fill bitreader to have enough for the remaining
+	CMPQ SI, $0x08
+	JL   sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte
+	MOVQ BX, AX
+	SHRQ $0x03, AX
+	SUBQ AX, R13
+	MOVQ (R13), DX
+	SUBQ AX, SI
+	ANDQ $0x07, BX
+	JMP  sequenceDecs_decodeSync_safe_amd64_fill_2_end
+
+sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte:
+	CMPQ    SI, $0x00
+	JLE     sequenceDecs_decodeSync_safe_amd64_fill_2_end
+	CMPQ    BX, $0x07
+	JLE     sequenceDecs_decodeSync_safe_amd64_fill_2_end
+	SHLQ    $0x08, DX
+	SUBQ    $0x01, R13
+	SUBQ    $0x01, SI
+	SUBQ    $0x08, BX
+	MOVBQZX (R13), AX
+	ORQ     AX, DX
+	JMP     sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte
+
+sequenceDecs_decodeSync_safe_amd64_fill_2_end:
+	// Update literal length
+	MOVQ  DI, AX
+	MOVQ  BX, CX
+	MOVQ  DX, R14
+	SHLQ  CL, R14
+	MOVB  AH, CL
+	SHRQ  $0x20, AX
+	TESTQ CX, CX
+	JZ    sequenceDecs_decodeSync_safe_amd64_ll_update_zero
+	ADDQ  CX, BX
+	CMPQ  BX, $0x40
+	JA    sequenceDecs_decodeSync_safe_amd64_ll_update_zero
+	CMPQ  CX, $0x40
+	JAE   sequenceDecs_decodeSync_safe_amd64_ll_update_zero
+	NEGQ  CX
+	SHRQ  CL, R14
+	ADDQ  R14, AX
+
+sequenceDecs_decodeSync_safe_amd64_ll_update_zero:
+	MOVQ AX, 24(SP)
+
+	// Fill bitreader for state updates
+	MOVQ    R13, (SP)
+	MOVQ    R9, AX
+	SHRQ    $0x08, AX
+	MOVBQZX AL, AX
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decodeSync_safe_amd64_skip_update
+
+	// Update Literal Length State
+	MOVBQZX DI, R13
+	SHRQ    $0x10, DI
+	MOVWQZX DI, DI
+	LEAQ    (BX)(R13*1), CX
+	MOVQ    DX, R14
+	MOVQ    CX, BX
+	ROLQ    CL, R14
+	MOVL    $0x00000001, R15
+	MOVB    R13, CL
+	SHLL    CL, R15
+	DECL    R15
+	ANDQ    R15, R14
+	ADDQ    R14, DI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Match Length State
+	MOVBQZX R8, R13
+	SHRQ    $0x10, R8
+	MOVWQZX R8, R8
+	LEAQ    (BX)(R13*1), CX
+	MOVQ    DX, R14
+	MOVQ    CX, BX
+	ROLQ    CL, R14
+	MOVL    $0x00000001, R15
+	MOVB    R13, CL
+	SHLL    CL, R15
+	DECL    R15
+	ANDQ    R15, R14
+	ADDQ    R14, R8
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Offset State
+	MOVBQZX R9, R13
+	SHRQ    $0x10, R9
+	MOVWQZX R9, R9
+	LEAQ    (BX)(R13*1), CX
+	MOVQ    DX, R14
+	MOVQ    CX, BX
+	ROLQ    CL, R14
+	MOVL    $0x00000001, R15
+	MOVB    R13, CL
+	SHLL    CL, R15
+	DECL    R15
+	ANDQ    R15, R14
+	ADDQ    R14, R9
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R9*8), R9
+
+sequenceDecs_decodeSync_safe_amd64_skip_update:
+	// Adjust offset
+	MOVQ   s+0(FP), CX
+	MOVQ   8(SP), R13
+	CMPQ   AX, $0x01
+	JBE    sequenceDecs_decodeSync_safe_amd64_adjust_offsetB_1_or_0
+	MOVUPS 144(CX), X0
+	MOVQ   R13, 144(CX)
+	MOVUPS X0, 152(CX)
+	JMP    sequenceDecs_decodeSync_safe_amd64_after_adjust
+
+sequenceDecs_decodeSync_safe_amd64_adjust_offsetB_1_or_0:
+	CMPQ 24(SP), $0x00000000
+	JNE  sequenceDecs_decodeSync_safe_amd64_adjust_offset_maybezero
+	INCQ R13
+	JMP  sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero
+
+sequenceDecs_decodeSync_safe_amd64_adjust_offset_maybezero:
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero
+	MOVQ  144(CX), R13
+	JMP   sequenceDecs_decodeSync_safe_amd64_after_adjust
+
+sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero:
+	MOVQ    R13, AX
+	XORQ    R14, R14
+	MOVQ    $-1, R15
+	CMPQ    R13, $0x03
+	CMOVQEQ R14, AX
+	CMOVQEQ R15, R14
+	ADDQ    144(CX)(AX*8), R14
+	JNZ     sequenceDecs_decodeSync_safe_amd64_adjust_temp_valid
+	MOVQ    $0x00000001, R14
+
+sequenceDecs_decodeSync_safe_amd64_adjust_temp_valid:
+	CMPQ R13, $0x01
+	JZ   sequenceDecs_decodeSync_safe_amd64_adjust_skip
+	MOVQ 152(CX), AX
+	MOVQ AX, 160(CX)
+
+sequenceDecs_decodeSync_safe_amd64_adjust_skip:
+	MOVQ 144(CX), AX
+	MOVQ AX, 152(CX)
+	MOVQ R14, 144(CX)
+	MOVQ R14, R13
+
+sequenceDecs_decodeSync_safe_amd64_after_adjust:
+	MOVQ R13, 8(SP)
+
+	// Check values
+	MOVQ  16(SP), AX
+	MOVQ  24(SP), CX
+	LEAQ  (AX)(CX*1), R14
+	MOVQ  s+0(FP), R15
+	ADDQ  R14, 256(R15)
+	MOVQ  ctx+16(FP), R14
+	SUBQ  CX, 104(R14)
+	JS    error_not_enough_literals
+	CMPQ  AX, $0x00020002
+	JA    sequenceDecs_decodeSync_safe_amd64_error_match_len_too_big
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_safe_amd64_match_len_ofs_ok
+	TESTQ AX, AX
+	JNZ   sequenceDecs_decodeSync_safe_amd64_error_match_len_ofs_mismatch
+
+sequenceDecs_decodeSync_safe_amd64_match_len_ofs_ok:
+	MOVQ 24(SP), AX
+	MOVQ 8(SP), CX
+	MOVQ 16(SP), R13
+
+	// Check if we have enough space in s.out
+	LEAQ (AX)(R13*1), R14
+	ADDQ R10, R14
+	CMPQ R14, 32(SP)
+	JA   error_not_enough_space
+
+	// Copy literals
+	TESTQ AX, AX
+	JZ    check_offset
+	MOVQ  AX, R14
+	SUBQ  $0x10, R14
+	JB    copy_1_small
+
+copy_1_loop:
+	MOVUPS (R11), X0
+	MOVUPS X0, (R10)
+	ADDQ   $0x10, R11
+	ADDQ   $0x10, R10
+	SUBQ   $0x10, R14
+	JAE    copy_1_loop
+	LEAQ   16(R11)(R14*1), R11
+	LEAQ   16(R10)(R14*1), R10
+	MOVUPS -16(R11), X0
+	MOVUPS X0, -16(R10)
+	JMP    copy_1_end
+
+copy_1_small:
+	CMPQ AX, $0x03
+	JE   copy_1_move_3
+	JB   copy_1_move_1or2
+	CMPQ AX, $0x08
+	JB   copy_1_move_4through7
+	JMP  copy_1_move_8through16
+
+copy_1_move_1or2:
+	MOVB (R11), R14
+	MOVB -1(R11)(AX*1), R15
+	MOVB R14, (R10)
+	MOVB R15, -1(R10)(AX*1)
+	ADDQ AX, R11
+	ADDQ AX, R10
+	JMP  copy_1_end
+
+copy_1_move_3:
+	MOVW (R11), R14
+	MOVB 2(R11), R15
+	MOVW R14, (R10)
+	MOVB R15, 2(R10)
+	ADDQ AX, R11
+	ADDQ AX, R10
+	JMP  copy_1_end
+
+copy_1_move_4through7:
+	MOVL (R11), R14
+	MOVL -4(R11)(AX*1), R15
+	MOVL R14, (R10)
+	MOVL R15, -4(R10)(AX*1)
+	ADDQ AX, R11
+	ADDQ AX, R10
+	JMP  copy_1_end
+
+copy_1_move_8through16:
+	MOVQ (R11), R14
+	MOVQ -8(R11)(AX*1), R15
+	MOVQ R14, (R10)
+	MOVQ R15, -8(R10)(AX*1)
+	ADDQ AX, R11
+	ADDQ AX, R10
+
+copy_1_end:
+	ADDQ AX, R12
+
+	// Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize)
+check_offset:
+	MOVQ R12, AX
+	ADDQ 40(SP), AX
+	CMPQ CX, AX
+	JG   error_match_off_too_big
+	CMPQ CX, 56(SP)
+	JG   error_match_off_too_big
+
+	// Copy match from history
+	MOVQ CX, AX
+	SUBQ R12, AX
+	JLS  copy_match
+	MOVQ 48(SP), R14
+	SUBQ AX, R14
+	CMPQ R13, AX
+	JG   copy_all_from_history
+	MOVQ R13, AX
+	SUBQ $0x10, AX
+	JB   copy_4_small
+
+copy_4_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R10)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R10
+	SUBQ   $0x10, AX
+	JAE    copy_4_loop
+	LEAQ   16(R14)(AX*1), R14
+	LEAQ   16(R10)(AX*1), R10
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R10)
+	JMP    copy_4_end
+
+copy_4_small:
+	CMPQ R13, $0x03
+	JE   copy_4_move_3
+	CMPQ R13, $0x08
+	JB   copy_4_move_4through7
+	JMP  copy_4_move_8through16
+
+copy_4_move_3:
+	MOVW (R14), AX
+	MOVB 2(R14), CL
+	MOVW AX, (R10)
+	MOVB CL, 2(R10)
+	ADDQ R13, R14
+	ADDQ R13, R10
+	JMP  copy_4_end
+
+copy_4_move_4through7:
+	MOVL (R14), AX
+	MOVL -4(R14)(R13*1), CX
+	MOVL AX, (R10)
+	MOVL CX, -4(R10)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R10
+	JMP  copy_4_end
+
+copy_4_move_8through16:
+	MOVQ (R14), AX
+	MOVQ -8(R14)(R13*1), CX
+	MOVQ AX, (R10)
+	MOVQ CX, -8(R10)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R10
+
+copy_4_end:
+	ADDQ R13, R12
+	JMP  handle_loop
+	JMP loop_finished
+
+copy_all_from_history:
+	MOVQ AX, R15
+	SUBQ $0x10, R15
+	JB   copy_5_small
+
+copy_5_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R10)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R10
+	SUBQ   $0x10, R15
+	JAE    copy_5_loop
+	LEAQ   16(R14)(R15*1), R14
+	LEAQ   16(R10)(R15*1), R10
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R10)
+	JMP    copy_5_end
+
+copy_5_small:
+	CMPQ AX, $0x03
+	JE   copy_5_move_3
+	JB   copy_5_move_1or2
+	CMPQ AX, $0x08
+	JB   copy_5_move_4through7
+	JMP  copy_5_move_8through16
+
+copy_5_move_1or2:
+	MOVB (R14), R15
+	MOVB -1(R14)(AX*1), BP
+	MOVB R15, (R10)
+	MOVB BP, -1(R10)(AX*1)
+	ADDQ AX, R14
+	ADDQ AX, R10
+	JMP  copy_5_end
+
+copy_5_move_3:
+	MOVW (R14), R15
+	MOVB 2(R14), BP
+	MOVW R15, (R10)
+	MOVB BP, 2(R10)
+	ADDQ AX, R14
+	ADDQ AX, R10
+	JMP  copy_5_end
+
+copy_5_move_4through7:
+	MOVL (R14), R15
+	MOVL -4(R14)(AX*1), BP
+	MOVL R15, (R10)
+	MOVL BP, -4(R10)(AX*1)
+	ADDQ AX, R14
+	ADDQ AX, R10
+	JMP  copy_5_end
+
+copy_5_move_8through16:
+	MOVQ (R14), R15
+	MOVQ -8(R14)(AX*1), BP
+	MOVQ R15, (R10)
+	MOVQ BP, -8(R10)(AX*1)
+	ADDQ AX, R14
+	ADDQ AX, R10
+
+copy_5_end:
+	ADDQ AX, R12
+	SUBQ AX, R13
+
+	// Copy match from the current buffer
+copy_match:
+	MOVQ R10, AX
+	SUBQ CX, AX
+
+	// ml <= mo
+	CMPQ R13, CX
+	JA   copy_overlapping_match
+
+	// Copy non-overlapping match
+	ADDQ R13, R12
+	MOVQ R13, CX
+	SUBQ $0x10, CX
+	JB   copy_2_small
+
+copy_2_loop:
+	MOVUPS (AX), X0
+	MOVUPS X0, (R10)
+	ADDQ   $0x10, AX
+	ADDQ   $0x10, R10
+	SUBQ   $0x10, CX
+	JAE    copy_2_loop
+	LEAQ   16(AX)(CX*1), AX
+	LEAQ   16(R10)(CX*1), R10
+	MOVUPS -16(AX), X0
+	MOVUPS X0, -16(R10)
+	JMP    copy_2_end
+
+copy_2_small:
+	CMPQ R13, $0x03
+	JE   copy_2_move_3
+	JB   copy_2_move_1or2
+	CMPQ R13, $0x08
+	JB   copy_2_move_4through7
+	JMP  copy_2_move_8through16
+
+copy_2_move_1or2:
+	MOVB (AX), CL
+	MOVB -1(AX)(R13*1), R14
+	MOVB CL, (R10)
+	MOVB R14, -1(R10)(R13*1)
+	ADDQ R13, AX
+	ADDQ R13, R10
+	JMP  copy_2_end
+
+copy_2_move_3:
+	MOVW (AX), CX
+	MOVB 2(AX), R14
+	MOVW CX, (R10)
+	MOVB R14, 2(R10)
+	ADDQ R13, AX
+	ADDQ R13, R10
+	JMP  copy_2_end
+
+copy_2_move_4through7:
+	MOVL (AX), CX
+	MOVL -4(AX)(R13*1), R14
+	MOVL CX, (R10)
+	MOVL R14, -4(R10)(R13*1)
+	ADDQ R13, AX
+	ADDQ R13, R10
+	JMP  copy_2_end
+
+copy_2_move_8through16:
+	MOVQ (AX), CX
+	MOVQ -8(AX)(R13*1), R14
+	MOVQ CX, (R10)
+	MOVQ R14, -8(R10)(R13*1)
+	ADDQ R13, AX
+	ADDQ R13, R10
+
+copy_2_end:
+	JMP handle_loop
+
+	// Copy overlapping match
+copy_overlapping_match:
+	ADDQ R13, R12
+
+copy_slow_3:
+	MOVB (AX), CL
+	MOVB CL, (R10)
+	INCQ AX
+	INCQ R10
+	DECQ R13
+	JNZ  copy_slow_3
+
+handle_loop:
+	MOVQ ctx+16(FP), AX
+	DECQ 96(AX)
+	JNS  sequenceDecs_decodeSync_safe_amd64_main_loop
+
+loop_finished:
+	MOVQ br+8(FP), AX
+	MOVQ DX, 32(AX)
+	MOVB BL, 40(AX)
+	MOVQ SI, 24(AX)
+
+	// Update the context
+	MOVQ ctx+16(FP), AX
+	MOVQ R12, 136(AX)
+	MOVQ 144(AX), CX
+	SUBQ CX, R11
+	MOVQ R11, 168(AX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decodeSync_safe_amd64_error_match_len_ofs_mismatch:
+	MOVQ 16(SP), AX
+	MOVQ ctx+16(FP), CX
+	MOVQ AX, 216(CX)
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decodeSync_safe_amd64_error_match_len_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+error_match_off_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 8(SP), CX
+	MOVQ CX, 224(AX)
+	MOVQ R12, 136(AX)
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+error_not_enough_space:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ R12, 136(AX)
+	MOVQ $0x00000005, ret+24(FP)
+	RET
+
+// func sequenceDecs_decodeSync_safe_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int
+// Requires: BMI, BMI2, CMOV, SSE
+TEXT ·sequenceDecs_decodeSync_safe_bmi2(SB), $64-32
+	MOVQ    br+8(FP), CX
+	MOVQ    32(CX), AX
+	MOVBQZX 40(CX), DX
+	MOVQ    24(CX), BX
+	MOVQ    (CX), CX
+	ADDQ    BX, CX
+	MOVQ    CX, (SP)
+	MOVQ    ctx+16(FP), CX
+	MOVQ    72(CX), SI
+	MOVQ    80(CX), DI
+	MOVQ    88(CX), R8
+	XORQ    R9, R9
+	MOVQ    R9, 8(SP)
+	MOVQ    R9, 16(SP)
+	MOVQ    R9, 24(SP)
+	MOVQ    112(CX), R9
+	MOVQ    128(CX), R10
+	MOVQ    R10, 32(SP)
+	MOVQ    144(CX), R10
+	MOVQ    136(CX), R11
+	MOVQ    200(CX), R12
+	MOVQ    R12, 56(SP)
+	MOVQ    176(CX), R12
+	MOVQ    R12, 48(SP)
+	MOVQ    184(CX), CX
+	MOVQ    CX, 40(SP)
+	MOVQ    40(SP), CX
+	ADDQ    CX, 48(SP)
+
+	// Calculate poiter to s.out[cap(s.out)] (a past-end pointer)
+	ADDQ R9, 32(SP)
+
+	// outBase += outPosition
+	ADDQ R11, R9
+
+sequenceDecs_decodeSync_safe_bmi2_main_loop:
+	MOVQ (SP), R12
+
+	// Fill bitreader to have enough for the offset and match length.
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R12
+	MOVQ (R12), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decodeSync_safe_bmi2_fill_end
+
+sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decodeSync_safe_bmi2_fill_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decodeSync_safe_bmi2_fill_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R12
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R12), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte
+
+sequenceDecs_decodeSync_safe_bmi2_fill_end:
+	// Update offset
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, R8, R13
+	MOVQ   AX, R14
+	LEAQ   (DX)(R13*1), CX
+	ROLQ   CL, R14
+	BZHIQ  R13, R14, R14
+	MOVQ   CX, DX
+	MOVQ   R8, CX
+	SHRQ   $0x20, CX
+	ADDQ   R14, CX
+	MOVQ   CX, 8(SP)
+
+	// Update match length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, DI, R13
+	MOVQ   AX, R14
+	LEAQ   (DX)(R13*1), CX
+	ROLQ   CL, R14
+	BZHIQ  R13, R14, R14
+	MOVQ   CX, DX
+	MOVQ   DI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R14, CX
+	MOVQ   CX, 16(SP)
+
+	// Fill bitreader to have enough for the remaining
+	CMPQ BX, $0x08
+	JL   sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte
+	MOVQ DX, CX
+	SHRQ $0x03, CX
+	SUBQ CX, R12
+	MOVQ (R12), AX
+	SUBQ CX, BX
+	ANDQ $0x07, DX
+	JMP  sequenceDecs_decodeSync_safe_bmi2_fill_2_end
+
+sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte:
+	CMPQ    BX, $0x00
+	JLE     sequenceDecs_decodeSync_safe_bmi2_fill_2_end
+	CMPQ    DX, $0x07
+	JLE     sequenceDecs_decodeSync_safe_bmi2_fill_2_end
+	SHLQ    $0x08, AX
+	SUBQ    $0x01, R12
+	SUBQ    $0x01, BX
+	SUBQ    $0x08, DX
+	MOVBQZX (R12), CX
+	ORQ     CX, AX
+	JMP     sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte
+
+sequenceDecs_decodeSync_safe_bmi2_fill_2_end:
+	// Update literal length
+	MOVQ   $0x00000808, CX
+	BEXTRQ CX, SI, R13
+	MOVQ   AX, R14
+	LEAQ   (DX)(R13*1), CX
+	ROLQ   CL, R14
+	BZHIQ  R13, R14, R14
+	MOVQ   CX, DX
+	MOVQ   SI, CX
+	SHRQ   $0x20, CX
+	ADDQ   R14, CX
+	MOVQ   CX, 24(SP)
+
+	// Fill bitreader for state updates
+	MOVQ    R12, (SP)
+	MOVQ    $0x00000808, CX
+	BEXTRQ  CX, R8, R12
+	MOVQ    ctx+16(FP), CX
+	CMPQ    96(CX), $0x00
+	JZ      sequenceDecs_decodeSync_safe_bmi2_skip_update
+	LEAQ    (SI)(DI*1), R13
+	ADDQ    R8, R13
+	MOVBQZX R13, R13
+	LEAQ    (DX)(R13*1), CX
+	MOVQ    AX, R14
+	MOVQ    CX, DX
+	ROLQ    CL, R14
+	BZHIQ   R13, R14, R14
+
+	// Update Offset State
+	BZHIQ  R8, R14, CX
+	SHRXQ  R8, R14, R14
+	MOVQ   $0x00001010, R13
+	BEXTRQ R13, R8, R8
+	ADDQ   CX, R8
+
+	// Load ctx.ofTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 48(CX), CX
+	MOVQ (CX)(R8*8), R8
+
+	// Update Match Length State
+	BZHIQ  DI, R14, CX
+	SHRXQ  DI, R14, R14
+	MOVQ   $0x00001010, R13
+	BEXTRQ R13, DI, DI
+	ADDQ   CX, DI
+
+	// Load ctx.mlTable
+	MOVQ ctx+16(FP), CX
+	MOVQ 24(CX), CX
+	MOVQ (CX)(DI*8), DI
+
+	// Update Literal Length State
+	BZHIQ  SI, R14, CX
+	MOVQ   $0x00001010, R13
+	BEXTRQ R13, SI, SI
+	ADDQ   CX, SI
+
+	// Load ctx.llTable
+	MOVQ ctx+16(FP), CX
+	MOVQ (CX), CX
+	MOVQ (CX)(SI*8), SI
+
+sequenceDecs_decodeSync_safe_bmi2_skip_update:
+	// Adjust offset
+	MOVQ   s+0(FP), CX
+	MOVQ   8(SP), R13
+	CMPQ   R12, $0x01
+	JBE    sequenceDecs_decodeSync_safe_bmi2_adjust_offsetB_1_or_0
+	MOVUPS 144(CX), X0
+	MOVQ   R13, 144(CX)
+	MOVUPS X0, 152(CX)
+	JMP    sequenceDecs_decodeSync_safe_bmi2_after_adjust
+
+sequenceDecs_decodeSync_safe_bmi2_adjust_offsetB_1_or_0:
+	CMPQ 24(SP), $0x00000000
+	JNE  sequenceDecs_decodeSync_safe_bmi2_adjust_offset_maybezero
+	INCQ R13
+	JMP  sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero
+
+sequenceDecs_decodeSync_safe_bmi2_adjust_offset_maybezero:
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero
+	MOVQ  144(CX), R13
+	JMP   sequenceDecs_decodeSync_safe_bmi2_after_adjust
+
+sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero:
+	MOVQ    R13, R12
+	XORQ    R14, R14
+	MOVQ    $-1, R15
+	CMPQ    R13, $0x03
+	CMOVQEQ R14, R12
+	CMOVQEQ R15, R14
+	ADDQ    144(CX)(R12*8), R14
+	JNZ     sequenceDecs_decodeSync_safe_bmi2_adjust_temp_valid
+	MOVQ    $0x00000001, R14
+
+sequenceDecs_decodeSync_safe_bmi2_adjust_temp_valid:
+	CMPQ R13, $0x01
+	JZ   sequenceDecs_decodeSync_safe_bmi2_adjust_skip
+	MOVQ 152(CX), R12
+	MOVQ R12, 160(CX)
+
+sequenceDecs_decodeSync_safe_bmi2_adjust_skip:
+	MOVQ 144(CX), R12
+	MOVQ R12, 152(CX)
+	MOVQ R14, 144(CX)
+	MOVQ R14, R13
+
+sequenceDecs_decodeSync_safe_bmi2_after_adjust:
+	MOVQ R13, 8(SP)
+
+	// Check values
+	MOVQ  16(SP), CX
+	MOVQ  24(SP), R12
+	LEAQ  (CX)(R12*1), R14
+	MOVQ  s+0(FP), R15
+	ADDQ  R14, 256(R15)
+	MOVQ  ctx+16(FP), R14
+	SUBQ  R12, 104(R14)
+	JS    error_not_enough_literals
+	CMPQ  CX, $0x00020002
+	JA    sequenceDecs_decodeSync_safe_bmi2_error_match_len_too_big
+	TESTQ R13, R13
+	JNZ   sequenceDecs_decodeSync_safe_bmi2_match_len_ofs_ok
+	TESTQ CX, CX
+	JNZ   sequenceDecs_decodeSync_safe_bmi2_error_match_len_ofs_mismatch
+
+sequenceDecs_decodeSync_safe_bmi2_match_len_ofs_ok:
+	MOVQ 24(SP), CX
+	MOVQ 8(SP), R12
+	MOVQ 16(SP), R13
+
+	// Check if we have enough space in s.out
+	LEAQ (CX)(R13*1), R14
+	ADDQ R9, R14
+	CMPQ R14, 32(SP)
+	JA   error_not_enough_space
+
+	// Copy literals
+	TESTQ CX, CX
+	JZ    check_offset
+	MOVQ  CX, R14
+	SUBQ  $0x10, R14
+	JB    copy_1_small
+
+copy_1_loop:
+	MOVUPS (R10), X0
+	MOVUPS X0, (R9)
+	ADDQ   $0x10, R10
+	ADDQ   $0x10, R9
+	SUBQ   $0x10, R14
+	JAE    copy_1_loop
+	LEAQ   16(R10)(R14*1), R10
+	LEAQ   16(R9)(R14*1), R9
+	MOVUPS -16(R10), X0
+	MOVUPS X0, -16(R9)
+	JMP    copy_1_end
+
+copy_1_small:
+	CMPQ CX, $0x03
+	JE   copy_1_move_3
+	JB   copy_1_move_1or2
+	CMPQ CX, $0x08
+	JB   copy_1_move_4through7
+	JMP  copy_1_move_8through16
+
+copy_1_move_1or2:
+	MOVB (R10), R14
+	MOVB -1(R10)(CX*1), R15
+	MOVB R14, (R9)
+	MOVB R15, -1(R9)(CX*1)
+	ADDQ CX, R10
+	ADDQ CX, R9
+	JMP  copy_1_end
+
+copy_1_move_3:
+	MOVW (R10), R14
+	MOVB 2(R10), R15
+	MOVW R14, (R9)
+	MOVB R15, 2(R9)
+	ADDQ CX, R10
+	ADDQ CX, R9
+	JMP  copy_1_end
+
+copy_1_move_4through7:
+	MOVL (R10), R14
+	MOVL -4(R10)(CX*1), R15
+	MOVL R14, (R9)
+	MOVL R15, -4(R9)(CX*1)
+	ADDQ CX, R10
+	ADDQ CX, R9
+	JMP  copy_1_end
+
+copy_1_move_8through16:
+	MOVQ (R10), R14
+	MOVQ -8(R10)(CX*1), R15
+	MOVQ R14, (R9)
+	MOVQ R15, -8(R9)(CX*1)
+	ADDQ CX, R10
+	ADDQ CX, R9
+
+copy_1_end:
+	ADDQ CX, R11
+
+	// Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize)
+check_offset:
+	MOVQ R11, CX
+	ADDQ 40(SP), CX
+	CMPQ R12, CX
+	JG   error_match_off_too_big
+	CMPQ R12, 56(SP)
+	JG   error_match_off_too_big
+
+	// Copy match from history
+	MOVQ R12, CX
+	SUBQ R11, CX
+	JLS  copy_match
+	MOVQ 48(SP), R14
+	SUBQ CX, R14
+	CMPQ R13, CX
+	JG   copy_all_from_history
+	MOVQ R13, CX
+	SUBQ $0x10, CX
+	JB   copy_4_small
+
+copy_4_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R9)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R9
+	SUBQ   $0x10, CX
+	JAE    copy_4_loop
+	LEAQ   16(R14)(CX*1), R14
+	LEAQ   16(R9)(CX*1), R9
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R9)
+	JMP    copy_4_end
+
+copy_4_small:
+	CMPQ R13, $0x03
+	JE   copy_4_move_3
+	CMPQ R13, $0x08
+	JB   copy_4_move_4through7
+	JMP  copy_4_move_8through16
+
+copy_4_move_3:
+	MOVW (R14), CX
+	MOVB 2(R14), R12
+	MOVW CX, (R9)
+	MOVB R12, 2(R9)
+	ADDQ R13, R14
+	ADDQ R13, R9
+	JMP  copy_4_end
+
+copy_4_move_4through7:
+	MOVL (R14), CX
+	MOVL -4(R14)(R13*1), R12
+	MOVL CX, (R9)
+	MOVL R12, -4(R9)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R9
+	JMP  copy_4_end
+
+copy_4_move_8through16:
+	MOVQ (R14), CX
+	MOVQ -8(R14)(R13*1), R12
+	MOVQ CX, (R9)
+	MOVQ R12, -8(R9)(R13*1)
+	ADDQ R13, R14
+	ADDQ R13, R9
+
+copy_4_end:
+	ADDQ R13, R11
+	JMP  handle_loop
+	JMP loop_finished
+
+copy_all_from_history:
+	MOVQ CX, R15
+	SUBQ $0x10, R15
+	JB   copy_5_small
+
+copy_5_loop:
+	MOVUPS (R14), X0
+	MOVUPS X0, (R9)
+	ADDQ   $0x10, R14
+	ADDQ   $0x10, R9
+	SUBQ   $0x10, R15
+	JAE    copy_5_loop
+	LEAQ   16(R14)(R15*1), R14
+	LEAQ   16(R9)(R15*1), R9
+	MOVUPS -16(R14), X0
+	MOVUPS X0, -16(R9)
+	JMP    copy_5_end
+
+copy_5_small:
+	CMPQ CX, $0x03
+	JE   copy_5_move_3
+	JB   copy_5_move_1or2
+	CMPQ CX, $0x08
+	JB   copy_5_move_4through7
+	JMP  copy_5_move_8through16
+
+copy_5_move_1or2:
+	MOVB (R14), R15
+	MOVB -1(R14)(CX*1), BP
+	MOVB R15, (R9)
+	MOVB BP, -1(R9)(CX*1)
+	ADDQ CX, R14
+	ADDQ CX, R9
+	JMP  copy_5_end
+
+copy_5_move_3:
+	MOVW (R14), R15
+	MOVB 2(R14), BP
+	MOVW R15, (R9)
+	MOVB BP, 2(R9)
+	ADDQ CX, R14
+	ADDQ CX, R9
+	JMP  copy_5_end
+
+copy_5_move_4through7:
+	MOVL (R14), R15
+	MOVL -4(R14)(CX*1), BP
+	MOVL R15, (R9)
+	MOVL BP, -4(R9)(CX*1)
+	ADDQ CX, R14
+	ADDQ CX, R9
+	JMP  copy_5_end
+
+copy_5_move_8through16:
+	MOVQ (R14), R15
+	MOVQ -8(R14)(CX*1), BP
+	MOVQ R15, (R9)
+	MOVQ BP, -8(R9)(CX*1)
+	ADDQ CX, R14
+	ADDQ CX, R9
+
+copy_5_end:
+	ADDQ CX, R11
+	SUBQ CX, R13
+
+	// Copy match from the current buffer
+copy_match:
+	MOVQ R9, CX
+	SUBQ R12, CX
+
+	// ml <= mo
+	CMPQ R13, R12
+	JA   copy_overlapping_match
+
+	// Copy non-overlapping match
+	ADDQ R13, R11
+	MOVQ R13, R12
+	SUBQ $0x10, R12
+	JB   copy_2_small
+
+copy_2_loop:
+	MOVUPS (CX), X0
+	MOVUPS X0, (R9)
+	ADDQ   $0x10, CX
+	ADDQ   $0x10, R9
+	SUBQ   $0x10, R12
+	JAE    copy_2_loop
+	LEAQ   16(CX)(R12*1), CX
+	LEAQ   16(R9)(R12*1), R9
+	MOVUPS -16(CX), X0
+	MOVUPS X0, -16(R9)
+	JMP    copy_2_end
+
+copy_2_small:
+	CMPQ R13, $0x03
+	JE   copy_2_move_3
+	JB   copy_2_move_1or2
+	CMPQ R13, $0x08
+	JB   copy_2_move_4through7
+	JMP  copy_2_move_8through16
+
+copy_2_move_1or2:
+	MOVB (CX), R12
+	MOVB -1(CX)(R13*1), R14
+	MOVB R12, (R9)
+	MOVB R14, -1(R9)(R13*1)
+	ADDQ R13, CX
+	ADDQ R13, R9
+	JMP  copy_2_end
+
+copy_2_move_3:
+	MOVW (CX), R12
+	MOVB 2(CX), R14
+	MOVW R12, (R9)
+	MOVB R14, 2(R9)
+	ADDQ R13, CX
+	ADDQ R13, R9
+	JMP  copy_2_end
+
+copy_2_move_4through7:
+	MOVL (CX), R12
+	MOVL -4(CX)(R13*1), R14
+	MOVL R12, (R9)
+	MOVL R14, -4(R9)(R13*1)
+	ADDQ R13, CX
+	ADDQ R13, R9
+	JMP  copy_2_end
+
+copy_2_move_8through16:
+	MOVQ (CX), R12
+	MOVQ -8(CX)(R13*1), R14
+	MOVQ R12, (R9)
+	MOVQ R14, -8(R9)(R13*1)
+	ADDQ R13, CX
+	ADDQ R13, R9
+
+copy_2_end:
+	JMP handle_loop
+
+	// Copy overlapping match
+copy_overlapping_match:
+	ADDQ R13, R11
+
+copy_slow_3:
+	MOVB (CX), R12
+	MOVB R12, (R9)
+	INCQ CX
+	INCQ R9
+	DECQ R13
+	JNZ  copy_slow_3
+
+handle_loop:
+	MOVQ ctx+16(FP), CX
+	DECQ 96(CX)
+	JNS  sequenceDecs_decodeSync_safe_bmi2_main_loop
+
+loop_finished:
+	MOVQ br+8(FP), CX
+	MOVQ AX, 32(CX)
+	MOVB DL, 40(CX)
+	MOVQ BX, 24(CX)
+
+	// Update the context
+	MOVQ ctx+16(FP), AX
+	MOVQ R11, 136(AX)
+	MOVQ 144(AX), CX
+	SUBQ CX, R10
+	MOVQ R10, 168(AX)
+
+	// Return success
+	MOVQ $0x00000000, ret+24(FP)
+	RET
+
+	// Return with match length error
+sequenceDecs_decodeSync_safe_bmi2_error_match_len_ofs_mismatch:
+	MOVQ 16(SP), AX
+	MOVQ ctx+16(FP), CX
+	MOVQ AX, 216(CX)
+	MOVQ $0x00000001, ret+24(FP)
+	RET
+
+	// Return with match too long error
+sequenceDecs_decodeSync_safe_bmi2_error_match_len_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ $0x00000002, ret+24(FP)
+	RET
+
+	// Return with match offset too long error
+error_match_off_too_big:
+	MOVQ ctx+16(FP), AX
+	MOVQ 8(SP), CX
+	MOVQ CX, 224(AX)
+	MOVQ R11, 136(AX)
+	MOVQ $0x00000003, ret+24(FP)
+	RET
+
+	// Return with not enough literals error
+error_not_enough_literals:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ $0x00000004, ret+24(FP)
+	RET
+
+	// Return with not enough output space error
+error_not_enough_space:
+	MOVQ ctx+16(FP), AX
+	MOVQ 24(SP), CX
+	MOVQ CX, 208(AX)
+	MOVQ 16(SP), CX
+	MOVQ CX, 216(AX)
+	MOVQ R11, 136(AX)
+	MOVQ $0x00000005, ret+24(FP)
+	RET
diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec_generic.go b/vendor/github.com/klauspost/compress/zstd/seqdec_generic.go
new file mode 100644
index 0000000..c3452bc
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/zstd/seqdec_generic.go
@@ -0,0 +1,237 @@
+//go:build !amd64 || appengine || !gc || noasm
+// +build !amd64 appengine !gc noasm
+
+package zstd
+
+import (
+	"fmt"
+	"io"
+)
+
+// decode sequences from the stream with the provided history but without dictionary.
+func (s *sequenceDecs) decodeSyncSimple(hist []byte) (bool, error) {
+	return false, nil
+}
+
+// decode sequences from the stream without the provided history.
+func (s *sequenceDecs) decode(seqs []seqVals) error {
+	br := s.br
+
+	// Grab full sizes tables, to avoid bounds checks.
+	llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize]
+	llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state
+	s.seqSize = 0
+	litRemain := len(s.literals)
+
+	maxBlockSize := maxCompressedBlockSize
+	if s.windowSize < maxBlockSize {
+		maxBlockSize = s.windowSize
+	}
+	for i := range seqs {
+		var ll, mo, ml int
+		if br.off > 4+((maxOffsetBits+16+16)>>3) {
+			// inlined function:
+			// ll, mo, ml = s.nextFast(br, llState, mlState, ofState)
+
+			// Final will not read from stream.
+			var llB, mlB, moB uint8
+			ll, llB = llState.final()
+			ml, mlB = mlState.final()
+			mo, moB = ofState.final()
+
+			// extra bits are stored in reverse order.
+			br.fillFast()
+			mo += br.getBits(moB)
+			if s.maxBits > 32 {
+				br.fillFast()
+			}
+			ml += br.getBits(mlB)
+			ll += br.getBits(llB)
+
+			if moB > 1 {
+				s.prevOffset[2] = s.prevOffset[1]
+				s.prevOffset[1] = s.prevOffset[0]
+				s.prevOffset[0] = mo
+			} else {
+				// mo = s.adjustOffset(mo, ll, moB)
+				// Inlined for rather big speedup
+				if ll == 0 {
+					// There is an exception though, when current sequence's literals_length = 0.
+					// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
+					// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
+					mo++
+				}
+
+				if mo == 0 {
+					mo = s.prevOffset[0]
+				} else {
+					var temp int
+					if mo == 3 {
+						temp = s.prevOffset[0] - 1
+					} else {
+						temp = s.prevOffset[mo]
+					}
+
+					if temp == 0 {
+						// 0 is not valid; input is corrupted; force offset to 1
+						println("WARNING: temp was 0")
+						temp = 1
+					}
+
+					if mo != 1 {
+						s.prevOffset[2] = s.prevOffset[1]
+					}
+					s.prevOffset[1] = s.prevOffset[0]
+					s.prevOffset[0] = temp
+					mo = temp
+				}
+			}
+			br.fillFast()
+		} else {
+			if br.overread() {
+				if debugDecoder {
+					printf("reading sequence %d, exceeded available data\n", i)
+				}
+				return io.ErrUnexpectedEOF
+			}
+			ll, mo, ml = s.next(br, llState, mlState, ofState)
+			br.fill()
+		}
+
+		if debugSequences {
+			println("Seq", i, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml)
+		}
+		// Evaluate.
+		// We might be doing this async, so do it early.
+		if mo == 0 && ml > 0 {
+			return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
+		}
+		if ml > maxMatchLen {
+			return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
+		}
+		s.seqSize += ll + ml
+		if s.seqSize > maxBlockSize {
+			return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
+		}
+		litRemain -= ll
+		if litRemain < 0 {
+			return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, litRemain+ll)
+		}
+		seqs[i] = seqVals{
+			ll: ll,
+			ml: ml,
+			mo: mo,
+		}
+		if i == len(seqs)-1 {
+			// This is the last sequence, so we shouldn't update state.
+			break
+		}
+
+		// Manually inlined, ~ 5-20% faster
+		// Update all 3 states at once. Approx 20% faster.
+		nBits := llState.nbBits() + mlState.nbBits() + ofState.nbBits()
+		if nBits == 0 {
+			llState = llTable[llState.newState()&maxTableMask]
+			mlState = mlTable[mlState.newState()&maxTableMask]
+			ofState = ofTable[ofState.newState()&maxTableMask]
+		} else {
+			bits := br.get32BitsFast(nBits)
+			lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31))
+			llState = llTable[(llState.newState()+lowBits)&maxTableMask]
+
+			lowBits = uint16(bits >> (ofState.nbBits() & 31))
+			lowBits &= bitMask[mlState.nbBits()&15]
+			mlState = mlTable[(mlState.newState()+lowBits)&maxTableMask]
+
+			lowBits = uint16(bits) & bitMask[ofState.nbBits()&15]
+			ofState = ofTable[(ofState.newState()+lowBits)&maxTableMask]
+		}
+	}
+	s.seqSize += litRemain
+	if s.seqSize > maxBlockSize {
+		return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
+	}
+	err := br.close()
+	if err != nil {
+		printf("Closing sequences: %v, %+v\n", err, *br)
+	}
+	return err
+}
+
+// executeSimple handles cases when a dictionary is not used.
+func (s *sequenceDecs) executeSimple(seqs []seqVals, hist []byte) error {
+	// Ensure we have enough output size...
+	if len(s.out)+s.seqSize > cap(s.out) {
+		addBytes := s.seqSize + len(s.out)
+		s.out = append(s.out, make([]byte, addBytes)...)
+		s.out = s.out[:len(s.out)-addBytes]
+	}
+
+	if debugDecoder {
+		printf("Execute %d seqs with literals: %d into %d bytes\n", len(seqs), len(s.literals), s.seqSize)
+	}
+
+	var t = len(s.out)
+	out := s.out[:t+s.seqSize]
+
+	for _, seq := range seqs {
+		// Add literals
+		copy(out[t:], s.literals[:seq.ll])
+		t += seq.ll
+		s.literals = s.literals[seq.ll:]
+
+		// Malformed input
+		if seq.mo > t+len(hist) || seq.mo > s.windowSize {
+			return fmt.Errorf("match offset (%d) bigger than current history (%d)", seq.mo, t+len(hist))
+		}
+
+		// Copy from history.
+		if v := seq.mo - t; v > 0 {
+			// v is the start position in history from end.
+			start := len(hist) - v
+			if seq.ml > v {
+				// Some goes into the current block.
+				// Copy remainder of history
+				copy(out[t:], hist[start:])
+				t += v
+				seq.ml -= v
+			} else {
+				copy(out[t:], hist[start:start+seq.ml])
+				t += seq.ml
+				continue
+			}
+		}
+
+		// We must be in the current buffer now
+		if seq.ml > 0 {
+			start := t - seq.mo
+			if seq.ml <= t-start {
+				// No overlap
+				copy(out[t:], out[start:start+seq.ml])
+				t += seq.ml
+			} else {
+				// Overlapping copy
+				// Extend destination slice and copy one byte at the time.
+				src := out[start : start+seq.ml]
+				dst := out[t:]
+				dst = dst[:len(src)]
+				t += len(src)
+				// Destination is the space we just added.
+				for i := range src {
+					dst[i] = src[i]
+				}
+			}
+		}
+	}
+	// Add final literals
+	copy(out[t:], s.literals)
+	if debugDecoder {
+		t += len(s.literals)
+		if t != len(out) {
+			panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize))
+		}
+	}
+	s.out = out
+
+	return nil
+}
diff --git a/vendor/github.com/klauspost/compress/zstd/snappy.go b/vendor/github.com/klauspost/compress/zstd/snappy.go
index 9d9d1d5..9e1baad 100644
--- a/vendor/github.com/klauspost/compress/zstd/snappy.go
+++ b/vendor/github.com/klauspost/compress/zstd/snappy.go
@@ -10,8 +10,8 @@
 	"hash/crc32"
 	"io"
 
-	"github.com/golang/snappy"
 	"github.com/klauspost/compress/huff0"
+	snappy "github.com/klauspost/compress/internal/snapref"
 )
 
 const (
@@ -203,7 +203,7 @@
 			written += int64(n)
 			continue
 		case chunkTypeUncompressedData:
-			if debug {
+			if debugEncoder {
 				println("Uncompressed, chunklen", chunkLen)
 			}
 			// Section 4.3. Uncompressed data (chunk type 0x01).
@@ -246,7 +246,7 @@
 			continue
 
 		case chunkTypeStreamIdentifier:
-			if debug {
+			if debugEncoder {
 				println("stream id", chunkLen, len(snappyMagicBody))
 			}
 			// Section 4.1. Stream identifier (chunk type 0xff).
diff --git a/vendor/github.com/klauspost/compress/zstd/zip.go b/vendor/github.com/klauspost/compress/zstd/zip.go
index e35a0a2..29c15c8 100644
--- a/vendor/github.com/klauspost/compress/zstd/zip.go
+++ b/vendor/github.com/klauspost/compress/zstd/zip.go
@@ -13,40 +13,63 @@
 // See https://www.winzip.com/win/en/comp_info.html
 const ZipMethodWinZip = 93
 
-// ZipMethodPKWare is the method number used by PKWARE to indicate Zstandard compression.
-// See https://pkware.cachefly.net/webdocs/APPNOTE/APPNOTE-6.3.7.TXT
+// ZipMethodPKWare is the original method number used by PKWARE to indicate Zstandard compression.
+// Deprecated: This has been deprecated by PKWARE, use ZipMethodWinZip instead for compression.
+// See https://pkware.cachefly.net/webdocs/APPNOTE/APPNOTE-6.3.9.TXT
 const ZipMethodPKWare = 20
 
-var zipReaderPool sync.Pool
-
-// newZipReader cannot be used since we would leak goroutines...
-func newZipReader(r io.Reader) io.ReadCloser {
-	dec, ok := zipReaderPool.Get().(*Decoder)
-	if ok {
-		dec.Reset(r)
-	} else {
-		d, err := NewReader(r, WithDecoderConcurrency(1), WithDecoderLowmem(true))
-		if err != nil {
-			panic(err)
-		}
-		dec = d
+// zipReaderPool is the default reader pool.
+var zipReaderPool = sync.Pool{New: func() interface{} {
+	z, err := NewReader(nil, WithDecoderLowmem(true), WithDecoderMaxWindow(128<<20), WithDecoderConcurrency(1))
+	if err != nil {
+		panic(err)
 	}
-	return &pooledZipReader{dec: dec}
+	return z
+}}
+
+// newZipReader creates a pooled zip decompressor.
+func newZipReader(opts ...DOption) func(r io.Reader) io.ReadCloser {
+	pool := &zipReaderPool
+	if len(opts) > 0 {
+		opts = append([]DOption{WithDecoderLowmem(true), WithDecoderMaxWindow(128 << 20)}, opts...)
+		// Force concurrency 1
+		opts = append(opts, WithDecoderConcurrency(1))
+		// Create our own pool
+		pool = &sync.Pool{}
+	}
+	return func(r io.Reader) io.ReadCloser {
+		dec, ok := pool.Get().(*Decoder)
+		if ok {
+			dec.Reset(r)
+		} else {
+			d, err := NewReader(r, opts...)
+			if err != nil {
+				panic(err)
+			}
+			dec = d
+		}
+		return &pooledZipReader{dec: dec, pool: pool}
+	}
 }
 
 type pooledZipReader struct {
-	mu  sync.Mutex // guards Close and Read
-	dec *Decoder
+	mu   sync.Mutex // guards Close and Read
+	pool *sync.Pool
+	dec  *Decoder
 }
 
 func (r *pooledZipReader) Read(p []byte) (n int, err error) {
 	r.mu.Lock()
 	defer r.mu.Unlock()
 	if r.dec == nil {
-		return 0, errors.New("Read after Close")
+		return 0, errors.New("read after close or EOF")
 	}
 	dec, err := r.dec.Read(p)
-
+	if err == io.EOF {
+		r.dec.Reset(nil)
+		r.pool.Put(r.dec)
+		r.dec = nil
+	}
 	return dec, err
 }
 
@@ -56,15 +79,16 @@
 	var err error
 	if r.dec != nil {
 		err = r.dec.Reset(nil)
-		zipReaderPool.Put(r.dec)
+		r.pool.Put(r.dec)
 		r.dec = nil
 	}
 	return err
 }
 
 type pooledZipWriter struct {
-	mu  sync.Mutex // guards Close and Read
-	enc *Encoder
+	mu   sync.Mutex // guards Close and Read
+	enc  *Encoder
+	pool *sync.Pool
 }
 
 func (w *pooledZipWriter) Write(p []byte) (n int, err error) {
@@ -82,7 +106,7 @@
 	var err error
 	if w.enc != nil {
 		err = w.enc.Close()
-		zipReaderPool.Put(w.enc)
+		w.pool.Put(w.enc)
 		w.enc = nil
 	}
 	return err
@@ -103,18 +127,15 @@
 				return nil, err
 			}
 		}
-		return &pooledZipWriter{enc: enc}, nil
+		return &pooledZipWriter{enc: enc, pool: &pool}, nil
 	}
 }
 
 // ZipDecompressor returns a decompressor that can be registered with zip libraries.
 // See ZipCompressor for example.
-func ZipDecompressor() func(r io.Reader) io.ReadCloser {
-	return func(r io.Reader) io.ReadCloser {
-		d, err := NewReader(r, WithDecoderConcurrency(1), WithDecoderLowmem(true))
-		if err != nil {
-			panic(err)
-		}
-		return d.IOReadCloser()
-	}
+// Options can be specified. WithDecoderConcurrency(1) is forced,
+// and by default a 128MB maximum decompression window is specified.
+// The window size can be overridden if required.
+func ZipDecompressor(opts ...DOption) func(r io.Reader) io.ReadCloser {
+	return newZipReader(opts...)
 }
diff --git a/vendor/github.com/klauspost/compress/zstd/zstd.go b/vendor/github.com/klauspost/compress/zstd/zstd.go
index 1ba308c..3eb3f1c 100644
--- a/vendor/github.com/klauspost/compress/zstd/zstd.go
+++ b/vendor/github.com/klauspost/compress/zstd/zstd.go
@@ -15,6 +15,12 @@
 // enable debug printing
 const debug = false
 
+// enable encoding debug printing
+const debugEncoder = debug
+
+// enable decoding debug printing
+const debugDecoder = debug
+
 // Enable extra assertions.
 const debugAsserts = debug || false
 
@@ -33,6 +39,9 @@
 // Reset the buffer offset when reaching this.
 const bufferReset = math.MaxInt32 - MaxWindowSize
 
+// fcsUnknown is used for unknown frame content size.
+const fcsUnknown = math.MaxUint64
+
 var (
 	// ErrReservedBlockType is returned when a reserved block type is found.
 	// Typically this indicates wrong or corrupted input.
@@ -46,6 +55,10 @@
 	// Typically returned on invalid input.
 	ErrBlockTooSmall = errors.New("block too small")
 
+	// ErrUnexpectedBlockSize is returned when a block has unexpected size.
+	// Typically returned on invalid input.
+	ErrUnexpectedBlockSize = errors.New("unexpected block size")
+
 	// ErrMagicMismatch is returned when a "magic" number isn't what is expected.
 	// Typically this indicates wrong or corrupted input.
 	ErrMagicMismatch = errors.New("invalid input: magic number mismatch")
@@ -69,6 +82,10 @@
 	// This is only returned if SingleSegment is specified on the frame.
 	ErrFrameSizeExceeded = errors.New("frame size exceeded")
 
+	// ErrFrameSizeMismatch is returned if the stated frame size does not match the expected size.
+	// This is only returned if SingleSegment is specified on the frame.
+	ErrFrameSizeMismatch = errors.New("frame size does not match size on stream")
+
 	// ErrCRCMismatch is returned if CRC mismatches.
 	ErrCRCMismatch = errors.New("CRC check failed")
 
@@ -82,28 +99,17 @@
 )
 
 func println(a ...interface{}) {
-	if debug {
+	if debug || debugDecoder || debugEncoder {
 		log.Println(a...)
 	}
 }
 
 func printf(format string, a ...interface{}) {
-	if debug {
+	if debug || debugDecoder || debugEncoder {
 		log.Printf(format, a...)
 	}
 }
 
-// matchLenFast does matching, but will not match the last up to 7 bytes.
-func matchLenFast(a, b []byte) int {
-	endI := len(a) & (math.MaxInt32 - 7)
-	for i := 0; i < endI; i += 8 {
-		if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-			return i + bits.TrailingZeros64(diff)>>3
-		}
-	}
-	return endI
-}
-
 // matchLen returns the maximum length.
 // a must be the shortest of the two.
 // The function also returns whether all bytes matched.