From 692d69efbf3df83b261aa3b0561e09cbff8e9e16 Mon Sep 17 00:00:00 2001 From: Cyril Tovena Date: Thu, 10 Dec 2020 10:55:27 +0100 Subject: [PATCH 1/3] Add zstd and flate compressions algorithms. zstd has the highest ratio of all while flate is very simalar to gzip. I'm adding those mainly for posterity and to keep them benchmarked. This also update gzip and lz4 library which both have been improved. Signed-off-by: Cyril Tovena --- go.mod | 4 +- go.sum | 12 +- pkg/chunkenc/interface.go | 8 + pkg/chunkenc/memchunk_test.go | 11 +- pkg/chunkenc/pool.go | 106 +- .../klauspost/compress/flate/deflate.go | 63 +- .../klauspost/compress/flate/fast_encoder.go | 37 +- .../klauspost/compress/flate/gen_inflate.go | 294 +++++ .../compress/flate/huffman_bit_writer.go | 82 +- .../klauspost/compress/flate/huffman_code.go | 80 +- .../compress/flate/huffman_sortByFreq.go | 178 +++ .../compress/flate/huffman_sortByLiteral.go | 201 +++ .../klauspost/compress/flate/inflate.go | 171 ++- .../klauspost/compress/flate/inflate_gen.go | 1002 ++++++++++++++ .../klauspost/compress/flate/level1.go | 25 +- .../klauspost/compress/flate/level2.go | 32 +- .../klauspost/compress/flate/level3.go | 58 +- .../klauspost/compress/flate/level4.go | 34 +- .../klauspost/compress/flate/level5.go | 47 +- .../klauspost/compress/flate/level6.go | 37 +- .../klauspost/compress/flate/regmask_amd64.go | 37 + .../klauspost/compress/flate/regmask_other.go | 39 + .../klauspost/compress/flate/stateless.go | 67 +- .../klauspost/compress/flate/token.go | 46 +- .../klauspost/compress/fse/README.md | 79 ++ .../klauspost/compress/fse/bitreader.go | 122 ++ .../klauspost/compress/fse/bitwriter.go | 168 +++ .../klauspost/compress/fse/bytereader.go | 47 + .../klauspost/compress/fse/compress.go | 684 ++++++++++ .../klauspost/compress/fse/decompress.go | 374 ++++++ .../github.com/klauspost/compress/fse/fse.go | 144 ++ .../klauspost/compress/gzip/gzip.go | 8 +- .../klauspost/compress/huff0/.gitignore | 1 + .../klauspost/compress/huff0/README.md | 87 ++ .../klauspost/compress/huff0/bitreader.go | 329 +++++ .../klauspost/compress/huff0/bitwriter.go | 210 +++ .../klauspost/compress/huff0/bytereader.go | 54 + .../klauspost/compress/huff0/compress.go | 657 ++++++++++ .../klauspost/compress/huff0/decompress.go | 1164 +++++++++++++++++ .../klauspost/compress/huff0/huff0.go | 273 ++++ .../klauspost/compress/snappy/.gitignore | 16 + .../klauspost/compress/snappy/AUTHORS | 15 + .../klauspost/compress/snappy/CONTRIBUTORS | 37 + .../klauspost/compress/snappy/LICENSE | 27 + .../klauspost/compress/snappy/README | 107 ++ .../klauspost/compress/snappy/decode.go | 237 ++++ .../klauspost/compress/snappy/decode_amd64.go | 14 + .../klauspost/compress/snappy/decode_amd64.s | 482 +++++++ .../klauspost/compress/snappy/decode_other.go | 115 ++ .../klauspost/compress/snappy/encode.go | 285 ++++ .../klauspost/compress/snappy/encode_amd64.go | 29 + .../klauspost/compress/snappy/encode_amd64.s | 730 +++++++++++ .../klauspost/compress/snappy/encode_other.go | 238 ++++ .../klauspost/compress/snappy/runbench.cmd | 2 + .../klauspost/compress/snappy/snappy.go | 98 ++ .../klauspost/compress/zstd/README.md | 406 ++++++ .../klauspost/compress/zstd/bitreader.go | 136 ++ .../klauspost/compress/zstd/bitwriter.go | 169 +++ .../klauspost/compress/zstd/blockdec.go | 739 +++++++++++ .../klauspost/compress/zstd/blockenc.go | 854 ++++++++++++ .../compress/zstd/blocktype_string.go | 85 ++ .../klauspost/compress/zstd/bytebuf.go | 127 ++ .../klauspost/compress/zstd/bytereader.go | 88 ++ .../klauspost/compress/zstd/decoder.go | 550 ++++++++ .../compress/zstd/decoder_options.go | 84 ++ .../klauspost/compress/zstd/dict.go | 122 ++ .../klauspost/compress/zstd/enc_base.go | 155 +++ .../klauspost/compress/zstd/enc_better.go | 595 +++++++++ .../klauspost/compress/zstd/enc_dfast.go | 713 ++++++++++ .../klauspost/compress/zstd/enc_fast.go | 661 ++++++++++ .../klauspost/compress/zstd/encoder.go | 570 ++++++++ .../compress/zstd/encoder_options.go | 282 ++++ .../klauspost/compress/zstd/framedec.go | 494 +++++++ .../klauspost/compress/zstd/frameenc.go | 137 ++ .../klauspost/compress/zstd/fse_decoder.go | 385 ++++++ .../klauspost/compress/zstd/fse_encoder.go | 726 ++++++++++ .../klauspost/compress/zstd/fse_predefined.go | 158 +++ .../klauspost/compress/zstd/hash.go | 77 ++ .../klauspost/compress/zstd/history.go | 89 ++ .../compress/zstd/internal/xxhash/LICENSE.txt | 22 + .../compress/zstd/internal/xxhash/README.md | 58 + .../compress/zstd/internal/xxhash/xxhash.go | 238 ++++ .../zstd/internal/xxhash/xxhash_amd64.go | 13 + .../zstd/internal/xxhash/xxhash_amd64.s | 215 +++ .../zstd/internal/xxhash/xxhash_other.go | 76 ++ .../zstd/internal/xxhash/xxhash_safe.go | 11 + .../klauspost/compress/zstd/seqdec.go | 485 +++++++ .../klauspost/compress/zstd/seqenc.go | 115 ++ .../klauspost/compress/zstd/snappy.go | 436 ++++++ .../klauspost/compress/zstd/zstd.go | 144 ++ vendor/github.com/pierrec/lz4/v4/go.mod | 7 - vendor/github.com/pierrec/lz4/v4/go.sum | 84 +- .../pierrec/lz4/v4/internal/lz4block/block.go | 180 ++- .../lz4/v4/internal/lz4block/blocks.go | 37 +- .../lz4/v4/internal/lz4block/decode_amd64.s | 2 + .../lz4/v4/internal/lz4block/decode_arm.s | 201 +++ .../{decode_amd64.go => decode_asm.go} | 1 + .../lz4/v4/internal/lz4block/decode_other.go | 28 +- .../lz4/v4/internal/lz4stream/block.go | 331 +++++ .../lz4/v4/internal/lz4stream/frame.go | 277 +--- .../lz4/v4/internal/xxh32/xxh32zero.go | 57 +- .../lz4/v4/internal/xxh32/xxh32zero_arm.go | 11 + .../lz4/v4/internal/xxh32/xxh32zero_arm.s | 259 ++++ .../lz4/v4/internal/xxh32/xxh32zero_other.go | 10 + vendor/github.com/pierrec/lz4/v4/lz4.go | 84 +- vendor/github.com/pierrec/lz4/v4/options.go | 52 +- vendor/github.com/pierrec/lz4/v4/reader.go | 180 ++- vendor/github.com/pierrec/lz4/v4/writer.go | 27 +- vendor/modules.txt | 9 +- 109 files changed, 20666 insertions(+), 890 deletions(-) create mode 100644 vendor/github.com/klauspost/compress/flate/gen_inflate.go create mode 100644 vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go create mode 100644 vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go create mode 100644 vendor/github.com/klauspost/compress/flate/inflate_gen.go create mode 100644 vendor/github.com/klauspost/compress/flate/regmask_amd64.go create mode 100644 vendor/github.com/klauspost/compress/flate/regmask_other.go create mode 100644 vendor/github.com/klauspost/compress/fse/README.md create mode 100644 vendor/github.com/klauspost/compress/fse/bitreader.go create mode 100644 vendor/github.com/klauspost/compress/fse/bitwriter.go create mode 100644 vendor/github.com/klauspost/compress/fse/bytereader.go create mode 100644 vendor/github.com/klauspost/compress/fse/compress.go create mode 100644 vendor/github.com/klauspost/compress/fse/decompress.go create mode 100644 vendor/github.com/klauspost/compress/fse/fse.go create mode 100644 vendor/github.com/klauspost/compress/huff0/.gitignore create mode 100644 vendor/github.com/klauspost/compress/huff0/README.md create mode 100644 vendor/github.com/klauspost/compress/huff0/bitreader.go create mode 100644 vendor/github.com/klauspost/compress/huff0/bitwriter.go create mode 100644 vendor/github.com/klauspost/compress/huff0/bytereader.go create mode 100644 vendor/github.com/klauspost/compress/huff0/compress.go create mode 100644 vendor/github.com/klauspost/compress/huff0/decompress.go create mode 100644 vendor/github.com/klauspost/compress/huff0/huff0.go create mode 100644 vendor/github.com/klauspost/compress/snappy/.gitignore create mode 100644 vendor/github.com/klauspost/compress/snappy/AUTHORS create mode 100644 vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS create mode 100644 vendor/github.com/klauspost/compress/snappy/LICENSE create mode 100644 vendor/github.com/klauspost/compress/snappy/README create mode 100644 vendor/github.com/klauspost/compress/snappy/decode.go create mode 100644 vendor/github.com/klauspost/compress/snappy/decode_amd64.go create mode 100644 vendor/github.com/klauspost/compress/snappy/decode_amd64.s create mode 100644 vendor/github.com/klauspost/compress/snappy/decode_other.go create mode 100644 vendor/github.com/klauspost/compress/snappy/encode.go create mode 100644 vendor/github.com/klauspost/compress/snappy/encode_amd64.go create mode 100644 vendor/github.com/klauspost/compress/snappy/encode_amd64.s create mode 100644 vendor/github.com/klauspost/compress/snappy/encode_other.go create mode 100644 vendor/github.com/klauspost/compress/snappy/runbench.cmd create mode 100644 vendor/github.com/klauspost/compress/snappy/snappy.go create mode 100644 vendor/github.com/klauspost/compress/zstd/README.md create mode 100644 vendor/github.com/klauspost/compress/zstd/bitreader.go create mode 100644 vendor/github.com/klauspost/compress/zstd/bitwriter.go create mode 100644 vendor/github.com/klauspost/compress/zstd/blockdec.go create mode 100644 vendor/github.com/klauspost/compress/zstd/blockenc.go create mode 100644 vendor/github.com/klauspost/compress/zstd/blocktype_string.go create mode 100644 vendor/github.com/klauspost/compress/zstd/bytebuf.go create mode 100644 vendor/github.com/klauspost/compress/zstd/bytereader.go create mode 100644 vendor/github.com/klauspost/compress/zstd/decoder.go create mode 100644 vendor/github.com/klauspost/compress/zstd/decoder_options.go create mode 100644 vendor/github.com/klauspost/compress/zstd/dict.go create mode 100644 vendor/github.com/klauspost/compress/zstd/enc_base.go create mode 100644 vendor/github.com/klauspost/compress/zstd/enc_better.go create mode 100644 vendor/github.com/klauspost/compress/zstd/enc_dfast.go create mode 100644 vendor/github.com/klauspost/compress/zstd/enc_fast.go create mode 100644 vendor/github.com/klauspost/compress/zstd/encoder.go create mode 100644 vendor/github.com/klauspost/compress/zstd/encoder_options.go create mode 100644 vendor/github.com/klauspost/compress/zstd/framedec.go create mode 100644 vendor/github.com/klauspost/compress/zstd/frameenc.go create mode 100644 vendor/github.com/klauspost/compress/zstd/fse_decoder.go create mode 100644 vendor/github.com/klauspost/compress/zstd/fse_encoder.go create mode 100644 vendor/github.com/klauspost/compress/zstd/fse_predefined.go create mode 100644 vendor/github.com/klauspost/compress/zstd/hash.go create mode 100644 vendor/github.com/klauspost/compress/zstd/history.go create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go create mode 100644 vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go create mode 100644 vendor/github.com/klauspost/compress/zstd/seqdec.go create mode 100644 vendor/github.com/klauspost/compress/zstd/seqenc.go create mode 100644 vendor/github.com/klauspost/compress/zstd/snappy.go create mode 100644 vendor/github.com/klauspost/compress/zstd/zstd.go create mode 100644 vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s rename vendor/github.com/pierrec/lz4/v4/internal/lz4block/{decode_amd64.go => decode_asm.go} (85%) create mode 100644 vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go create mode 100644 vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go create mode 100644 vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s create mode 100644 vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go diff --git a/go.mod b/go.mod index cf12ead0d178..c95f38cedc52 100644 --- a/go.mod +++ b/go.mod @@ -37,13 +37,13 @@ require ( github.com/jmespath/go-jmespath v0.4.0 github.com/joncrlsn/dque v2.2.1-0.20200515025108-956d14155fa2+incompatible github.com/json-iterator/go v1.1.10 - github.com/klauspost/compress v1.9.5 github.com/mitchellh/mapstructure v1.3.3 + github.com/klauspost/compress v1.11.3 github.com/moby/term v0.0.0-20200915141129-7f0af18e79f2 // indirect github.com/mwitkow/go-conntrack v0.0.0-20190716064945-2f068394615f github.com/opentracing/opentracing-go v1.2.0 // github.com/pierrec/lz4 v2.0.5+incompatible - github.com/pierrec/lz4/v4 v4.0.2-0.20200813132121-22f5d580d5c4 + github.com/pierrec/lz4/v4 v4.1.1 github.com/pkg/errors v0.9.1 github.com/prometheus/client_golang v1.8.0 github.com/prometheus/client_model v0.2.0 diff --git a/go.sum b/go.sum index d48822ae7064..1d807014f407 100644 --- a/go.sum +++ b/go.sum @@ -48,7 +48,6 @@ cloud.google.com/go/storage v1.6.0/go.mod h1:N7U0C8pVQ/+NIKOBQyamJIeKQKkZ+mxpohl cloud.google.com/go/storage v1.8.0/go.mod h1:Wv1Oy7z6Yz3DshWRJFhqM/UCfaWIRTdp0RXyy7KQOVs= cloud.google.com/go/storage v1.10.0 h1:STgFzyU5/8miMl0//zKh2aQeTyeaUH3WN9bSUiJ09bA= cloud.google.com/go/storage v1.10.0/go.mod h1:FLPqc6j+Ki4BU591ie1oL6qBQGu2Bl/tZ9ullr3+Kg0= -code.cloudfoundry.org/bytefmt v0.0.0-20200131002437-cf55d5288a48/go.mod h1:wN/zk7mhREp/oviagqUXY3EwuHhWyOvAdsn5Y4CzOrc= collectd.org v0.3.0/go.mod h1:A/8DzQBkF6abtvrT2j/AU/4tiBgJWYyh0y/oB/4MlWE= contrib.go.opencensus.io/exporter/ocagent v0.6.0/go.mod h1:zmKjrJcdo0aYcVS7bmEeSEBLPA9YJp5bjrofdU3pIXs= dmitri.shuralyov.com/gpu/mtl v0.0.0-20190408044501-666a987793e9/go.mod h1:H6x//7gZCb22OMCxBHrMx7a5I7Hp++hsVxbQ4BYO7hU= @@ -825,7 +824,6 @@ github.com/julienschmidt/httprouter v1.3.0 h1:U0609e9tgbseu3rBINet9P48AI/D3oJs4d github.com/julienschmidt/httprouter v1.3.0/go.mod h1:JR6WtHb+2LUe8TCKY3cZOxFyyO8IZAc4RVcycCCAKdM= github.com/jung-kurt/gofpdf v1.0.3-0.20190309125859-24315acbbda5/go.mod h1:7Id9E/uU8ce6rXgefFLlgrJj/GYY22cpxn+r32jIOes= github.com/jwilder/encoding v0.0.0-20170811194829-b4e1701a28ef/go.mod h1:Ct9fl0F6iIOGgxJ5npU/IUOhOhqlVrGjyIZc8/MagT0= -github.com/k0kubun/go-ansi v0.0.0-20180517002512-3bf9e2903213/go.mod h1:vNUNkEQ1e29fT/6vq2aBdFsgNPmy8qMdSay1npru+Sw= github.com/kardianos/osext v0.0.0-20190222173326-2bc1f35cddc0/go.mod h1:1NbS8ALrpOvjt0rHPNLyCIeMtbizbir8U//inJ+zuB8= github.com/karrick/godirwalk v1.8.0/go.mod h1:H5KPZjojv4lE+QYImBI8xVtrBRgYrIVsaRPx4tDPEn4= github.com/karrick/godirwalk v1.10.3/go.mod h1:RoGL9dQei4vP9ilrpETWE8CLOZ1kiN0LhBygSwrAsHA= @@ -835,6 +833,8 @@ github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+o github.com/klauspost/compress v1.4.0/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A= github.com/klauspost/compress v1.9.5 h1:U+CaK85mrNNb4k8BNOfgJtJ/gr6kswUCFj6miSzVC6M= github.com/klauspost/compress v1.9.5/go.mod h1:RyIbtBH6LamlWaDj8nUwkbUhJ87Yi3uG0guNDohfE1A= +github.com/klauspost/compress v1.11.3 h1:dB4Bn0tN3wdCzQxnS8r06kV74qN/TAfaIS0bVE8h3jc= +github.com/klauspost/compress v1.11.3/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs= github.com/klauspost/cpuid v0.0.0-20170728055534-ae7887de9fa5/go.mod h1:Pj4uuM528wm8OyEC2QMXAi2YiTZ96dNQPGgoMS4s3ek= github.com/klauspost/cpuid v1.2.3/go.mod h1:Pj4uuM528wm8OyEC2QMXAi2YiTZ96dNQPGgoMS4s3ek= github.com/klauspost/cpuid v1.3.1 h1:5JNjFYYQrZeKRJ0734q51WCEEn2huer72Dc7K+R/b6s= @@ -911,7 +911,6 @@ github.com/mattn/go-runewidth v0.0.2/go.mod h1:LwmH8dsx7+W8Uxz3IHJYH5QSwggIsqBzp github.com/mattn/go-runewidth v0.0.3/go.mod h1:LwmH8dsx7+W8Uxz3IHJYH5QSwggIsqBzpuz5H//U1FU= github.com/mattn/go-runewidth v0.0.4/go.mod h1:LwmH8dsx7+W8Uxz3IHJYH5QSwggIsqBzpuz5H//U1FU= github.com/mattn/go-runewidth v0.0.6/go.mod h1:H031xJmbD/WCDINGzjvQ9THkh0rPKHF+m2gUSrubnMI= -github.com/mattn/go-runewidth v0.0.9/go.mod h1:H031xJmbD/WCDINGzjvQ9THkh0rPKHF+m2gUSrubnMI= github.com/mattn/go-sqlite3 v1.10.0/go.mod h1:FPy6KqzDD04eiIsT53CuJW3U88zkxoIYsOqkbpncsNc= github.com/mattn/go-sqlite3 v1.11.0/go.mod h1:FPy6KqzDD04eiIsT53CuJW3U88zkxoIYsOqkbpncsNc= github.com/mattn/go-tty v0.0.0-20180907095812-13ff1204f104/go.mod h1:XPvLUNfbS4fJH25nqRHfWLMa1ONC8Amw+mIA639KxkE= @@ -948,7 +947,6 @@ github.com/minio/sha256-simd v0.1.1 h1:5QHSlgo3nt5yKOJrC7W8w7X+NFl8cMPZm96iu8kKU github.com/minio/sha256-simd v0.1.1/go.mod h1:B5e1o+1/KgNmWrSQK08Y6Z1Vb5pwIktudl0J58iy0KM= github.com/mitchellh/cli v1.0.0/go.mod h1:hNIlj7HEI86fIcpObd7a0FcrxTWetlwJDGcceTlRvqc= github.com/mitchellh/cli v1.1.0/go.mod h1:xcISNoH86gajksDmfB23e/pu+B+GeFRMYmoHXxx3xhI= -github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db/go.mod h1:l0dey0ia/Uv7NcFFVbCLtqEBQbrT4OCwCSKTEv6enCw= github.com/mitchellh/go-homedir v1.0.0 h1:vKb8ShqSby24Yrqr/yDYkuFz8d0WUjys40rvnGC8aR0= github.com/mitchellh/go-homedir v1.0.0/go.mod h1:SfyaCUpYCn1Vlf4IUYiD9fPX4A5wJrkLzIz1N1q0pr0= github.com/mitchellh/go-homedir v1.1.0 h1:lukF9ziXFxDFPkA1vsr5zpc1XuPDn/wFntq5mG+4E0Y= @@ -1069,12 +1067,11 @@ github.com/performancecopilot/speed v3.0.0+incompatible/go.mod h1:/CLtqpZ5gBg1M9 github.com/peterbourgon/diskv v2.0.1+incompatible/go.mod h1:uqqh8zWWbv1HBMNONnaR/tNboyR3/BZd58JJSHlUSCU= github.com/peterh/liner v1.0.1-0.20180619022028-8c1271fcf47f/go.mod h1:xIteQHvHuaLYG9IFj6mSxM0fCKrs34IrEQUhOYuGPHc= github.com/philhofer/fwd v1.0.0/go.mod h1:gk3iGcWd9+svBvR0sR+KPcfE+RNWozjowpeBVG3ZVNU= -github.com/pierrec/cmdflag v0.0.2/go.mod h1:a3zKGZ3cdQUfxjd0RGMLZr8xI3nvpJOB+m6o/1X5BmU= github.com/pierrec/lz4 v1.0.2-0.20190131084431-473cd7ce01a1/go.mod h1:3/3N9NVKO0jef7pBehbT1qWhCMrIgbYNnFAZCqQ5LRc= github.com/pierrec/lz4 v2.0.5+incompatible h1:2xWsjqPFWcplujydGg4WmhC/6fZqK42wMM8aXeqhl0I= github.com/pierrec/lz4 v2.0.5+incompatible/go.mod h1:pdkljMzZIN41W+lC3N2tnIh5sFi+IEE17M5jbnwPHcY= -github.com/pierrec/lz4/v4 v4.0.2-0.20200813132121-22f5d580d5c4 h1:qId52nKbgRXnPooGimqIBgQCECmjgQedP8YsrOFPOOo= -github.com/pierrec/lz4/v4 v4.0.2-0.20200813132121-22f5d580d5c4/go.mod h1:vvUajMAuienWCEdMnA5Zb5mp0VIa9M8VvKcVEOkoAh8= +github.com/pierrec/lz4/v4 v4.1.1 h1:cS6aGkNLJr4u+UwaA21yp+gbWN3WJWtKo1axmPDObMA= +github.com/pierrec/lz4/v4 v4.1.1/go.mod h1:gZWDp/Ze/IJXGXf23ltt2EXimqmTUXEy0GFuRQyBid4= github.com/pkg/errors v0.8.0/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0= github.com/pkg/errors v0.8.1/go.mod h1:bwawxfHBFNV+L2hUp1rHADufV3IMtnDRdf1r5NINEl0= github.com/pkg/errors v0.9.1 h1:FEBLx1zS214owpjy7qsBeixbURkuhQAwrK5UwLGTwt4= @@ -1192,7 +1189,6 @@ github.com/samuel/go-zookeeper v0.0.0-20200724154423-2164a8ac840e/go.mod h1:gi+0 github.com/santhosh-tekuri/jsonschema v1.2.4/go.mod h1:TEAUOeZSmIxTTuHatJzrvARHiuO9LYd+cIxzgEHCQI4= github.com/satori/go.uuid v1.2.0 h1:0uYX9dsZ2yD7q2RtLRtPSdGDWzjeM3TbMJP9utgA0ww= github.com/satori/go.uuid v1.2.0/go.mod h1:dA0hQrYB0VpLJoorglMZABFdXlWrHn1NEOzdhQKdks0= -github.com/schollz/progressbar/v3 v3.3.4/go.mod h1:Rp5lZwpgtYmlvmGo1FyDwXMqagyRBQYSDwzlP9QDu84= github.com/sean-/seed v0.0.0-20170313163322-e2103e2c3529 h1:nn5Wsu0esKSJiIVhscUtVbo7ada43DJhG55ua/hjS5I= github.com/sean-/seed v0.0.0-20170313163322-e2103e2c3529/go.mod h1:DxrIzT+xaE7yg65j358z/aeFdxmN0P9QXhEzd20vsDc= github.com/segmentio/fasthash v0.0.0-20180216231524-a72b379d632e/go.mod h1:tm/wZFQ8e24NYaBGIlnO2WGCAi67re4HHuOm0sftE/M= diff --git a/pkg/chunkenc/interface.go b/pkg/chunkenc/interface.go index 6b077289c954..0102818d9e0a 100644 --- a/pkg/chunkenc/interface.go +++ b/pkg/chunkenc/interface.go @@ -36,6 +36,8 @@ const ( EncLZ4_256k EncLZ4_1M EncLZ4_4M + EncFlate + EncZstd ) var supportedEncoding = []Encoding{ @@ -46,6 +48,8 @@ var supportedEncoding = []Encoding{ EncLZ4_256k, EncLZ4_1M, EncLZ4_4M, + EncFlate, + EncZstd, } func (e Encoding) String() string { @@ -66,6 +70,10 @@ func (e Encoding) String() string { return "lz4" case EncSnappy: return "snappy" + case EncFlate: + return "flate" + case EncZstd: + return "zstd" default: return "unknown" } diff --git a/pkg/chunkenc/memchunk_test.go b/pkg/chunkenc/memchunk_test.go index eececd98b7fd..dfbb270830c0 100644 --- a/pkg/chunkenc/memchunk_test.go +++ b/pkg/chunkenc/memchunk_test.go @@ -33,6 +33,8 @@ var testEncoding = []Encoding{ EncLZ4_1M, EncLZ4_4M, EncSnappy, + EncFlate, + EncZstd, } var ( @@ -619,6 +621,13 @@ func BenchmarkWrite(b *testing.B) { } +type nomatchPipeline struct{} + +func (nomatchPipeline) Process(line []byte) ([]byte, log.LabelsResult, bool) { return line, nil, false } +func (nomatchPipeline) ProcessString(line string) (string, log.LabelsResult, bool) { + return line, nil, false +} + func BenchmarkRead(b *testing.B) { for _, enc := range testEncoding { b.Run(enc.String(), func(b *testing.B) { @@ -629,7 +638,7 @@ func BenchmarkRead(b *testing.B) { for n := 0; n < b.N; n++ { for _, c := range chunks { // use forward iterator for benchmark -- backward iterator does extra allocations by keeping entries in memory - iterator, err := c.Iterator(context.Background(), time.Unix(0, 0), time.Now(), logproto.FORWARD, noopStreamPipeline) + iterator, err := c.Iterator(context.Background(), time.Unix(0, 0), time.Now(), logproto.FORWARD, nomatchPipeline{}) if err != nil { panic(err) } diff --git a/pkg/chunkenc/pool.go b/pkg/chunkenc/pool.go index 60912e3a9076..8d37e2ffcee8 100644 --- a/pkg/chunkenc/pool.go +++ b/pkg/chunkenc/pool.go @@ -7,7 +7,9 @@ import ( "sync" "github.com/golang/snappy" + "github.com/klauspost/compress/flate" "github.com/klauspost/compress/gzip" + "github.com/klauspost/compress/zstd" "github.com/pierrec/lz4/v4" "github.com/prometheus/prometheus/pkg/pool" ) @@ -32,7 +34,8 @@ var ( Lz4_256k = LZ4Pool{bufferSize: 1 << 18} // Lz4_256k uses 256k buffer Lz4_1M = LZ4Pool{bufferSize: 1 << 20} // Lz4_1M uses 1M buffer Lz4_4M = LZ4Pool{bufferSize: 1 << 22} // Lz4_4M uses 4M buffer - + Flate = FlatePool{} + Zstd = ZstdPool{} // Snappy is the snappy compression pool Snappy SnappyPool // Noop is the no compression pool @@ -74,6 +77,10 @@ func getReaderPool(enc Encoding) ReaderPool { return &Snappy case EncNone: return &Noop + case EncFlate: + return &Flate + case EncZstd: + return &Zstd default: panic("unknown encoding") } @@ -132,6 +139,103 @@ func (pool *GzipPool) PutWriter(writer io.WriteCloser) { pool.writers.Put(writer) } +// FlatePool is a flate compression pool +type FlatePool struct { + readers sync.Pool + writers sync.Pool + level int +} + +// GetReader gets or creates a new CompressionReader and reset it to read from src +func (pool *FlatePool) GetReader(src io.Reader) io.Reader { + if r := pool.readers.Get(); r != nil { + reader := r.(flate.Resetter) + err := reader.Reset(src, nil) + if err != nil { + panic(err) + } + return reader.(io.Reader) + } + return flate.NewReader(src) +} + +// PutReader places back in the pool a CompressionReader +func (pool *FlatePool) PutReader(reader io.Reader) { + pool.readers.Put(reader) +} + +// GetWriter gets or creates a new CompressionWriter and reset it to write to dst +func (pool *FlatePool) GetWriter(dst io.Writer) io.WriteCloser { + if w := pool.writers.Get(); w != nil { + writer := w.(*flate.Writer) + writer.Reset(dst) + return writer + } + + level := pool.level + if level == 0 { + level = flate.DefaultCompression + } + w, err := flate.NewWriter(dst, level) + if err != nil { + panic(err) // never happens, error is only returned on wrong compression level. + } + return w +} + +// PutWriter places back in the pool a CompressionWriter +func (pool *FlatePool) PutWriter(writer io.WriteCloser) { + pool.writers.Put(writer) +} + +// GzipPool is a gun zip compression pool +type ZstdPool struct { + readers sync.Pool + writers sync.Pool +} + +// GetReader gets or creates a new CompressionReader and reset it to read from src +func (pool *ZstdPool) GetReader(src io.Reader) io.Reader { + if r := pool.readers.Get(); r != nil { + reader := r.(*zstd.Decoder) + err := reader.Reset(src) + if err != nil { + panic(err) + } + return reader + } + reader, err := zstd.NewReader(src) + if err != nil { + panic(err) + } + return reader +} + +// PutReader places back in the pool a CompressionReader +func (pool *ZstdPool) PutReader(reader io.Reader) { + pool.readers.Put(reader) +} + +// GetWriter gets or creates a new CompressionWriter and reset it to write to dst +func (pool *ZstdPool) GetWriter(dst io.Writer) io.WriteCloser { + if w := pool.writers.Get(); w != nil { + writer := w.(*zstd.Encoder) + writer.Reset(dst) + return writer + } + + w, err := zstd.NewWriter(dst) + if err != nil { + panic(err) // never happens, error is only returned on wrong compression level. + } + return w +} + +// PutWriter places back in the pool a CompressionWriter +func (pool *ZstdPool) PutWriter(writer io.WriteCloser) { + pool.writers.Put(writer) +} + type LZ4Pool struct { readers sync.Pool writers sync.Pool diff --git a/vendor/github.com/klauspost/compress/flate/deflate.go b/vendor/github.com/klauspost/compress/flate/deflate.go index 20c94f596843..25dbe3e15f49 100644 --- a/vendor/github.com/klauspost/compress/flate/deflate.go +++ b/vendor/github.com/klauspost/compress/flate/deflate.go @@ -48,6 +48,8 @@ const ( maxHashOffset = 1 << 24 skipNever = math.MaxInt32 + + debugDeflate = false ) type compressionLevel struct { @@ -59,15 +61,13 @@ type compressionLevel struct { // See https://blog.klauspost.com/rebalancing-deflate-compression-levels/ var levels = []compressionLevel{ {}, // 0 - // Level 1-4 uses specialized algorithm - values not used + // Level 1-6 uses specialized algorithm - values not used {0, 0, 0, 0, 0, 1}, {0, 0, 0, 0, 0, 2}, {0, 0, 0, 0, 0, 3}, {0, 0, 0, 0, 0, 4}, - // For levels 5-6 we don't bother trying with lazy matches. - // Lazy matching is at least 30% slower, with 1.5% increase. - {6, 0, 12, 8, 12, 5}, - {8, 0, 24, 16, 16, 6}, + {0, 0, 0, 0, 0, 5}, + {0, 0, 0, 0, 0, 6}, // Levels 7-9 use increasingly more lazy matching // and increasingly stringent conditions for "good enough". {8, 8, 24, 16, skipNever, 7}, @@ -80,9 +80,7 @@ type advancedState struct { // deflate state length int offset int - hash uint32 maxInsertIndex int - ii uint16 // position of last match, intended to overflow to reset. // Input hash chains // hashHead[hashValue] contains the largest inputIndex with the specified hash value @@ -97,6 +95,9 @@ type advancedState struct { // input window: unprocessed data is window[index:windowEnd] index int hashMatch [maxMatchLength + minMatchLength]uint32 + + hash uint32 + ii uint16 // position of last match, intended to overflow to reset. } type compressor struct { @@ -107,18 +108,19 @@ type compressor struct { // compression algorithm fill func(*compressor, []byte) int // copy data to window step func(*compressor) // process window - sync bool // requesting flush - window []byte - windowEnd int - blockStart int // window index where current tokens start - byteAvailable bool // if true, still need to process window[index-1]. - err error + window []byte + windowEnd int + blockStart int // window index where current tokens start + err error // queued output tokens tokens tokens fast fastEnc state *advancedState + + sync bool // requesting flush + byteAvailable bool // if true, still need to process window[index-1]. } func (d *compressor) fillDeflate(b []byte) int { @@ -203,9 +205,8 @@ func (d *compressor) writeBlockSkip(tok *tokens, index int, eof bool) error { // This is much faster than doing a full encode. // Should only be used after a start/reset. func (d *compressor) fillWindow(b []byte) { - // Do not fill window if we are in store-only mode, - // use constant or Snappy compression. - if d.level == 0 { + // Do not fill window if we are in store-only or huffman mode. + if d.level <= 0 { return } if d.fast != nil { @@ -368,7 +369,7 @@ func (d *compressor) deflateLazy() { // Sanity enables additional runtime tests. // It's intended to be used during development // to supplement the currently ad-hoc unit tests. - const sanity = false + const sanity = debugDeflate if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync { return @@ -644,7 +645,7 @@ func (d *compressor) init(w io.Writer, level int) (err error) { d.fill = (*compressor).fillBlock d.step = (*compressor).store case level == ConstantCompression: - d.w.logReusePenalty = uint(4) + d.w.logNewTablePenalty = 4 d.window = make([]byte, maxStoreBlockSize) d.fill = (*compressor).fillBlock d.step = (*compressor).storeHuff @@ -652,13 +653,13 @@ func (d *compressor) init(w io.Writer, level int) (err error) { level = 5 fallthrough case level >= 1 && level <= 6: - d.w.logReusePenalty = uint(level + 1) + d.w.logNewTablePenalty = 6 d.fast = newFastEnc(level) d.window = make([]byte, maxStoreBlockSize) d.fill = (*compressor).fillBlock d.step = (*compressor).storeFast case 7 <= level && level <= 9: - d.w.logReusePenalty = uint(level) + d.w.logNewTablePenalty = 10 d.state = &advancedState{} d.compressionLevel = levels[level] d.initDeflate() @@ -667,6 +668,7 @@ func (d *compressor) init(w io.Writer, level int) (err error) { default: return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level) } + d.level = level return nil } @@ -720,6 +722,7 @@ func (d *compressor) close() error { return d.w.err } d.w.flush() + d.w.reset(nil) return d.w.err } @@ -750,8 +753,7 @@ func NewWriter(w io.Writer, level int) (*Writer, error) { // can only be decompressed by a Reader initialized with the // same dictionary. func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) { - dw := &dictWriter{w} - zw, err := NewWriter(dw, level) + zw, err := NewWriter(w, level) if err != nil { return nil, err } @@ -760,14 +762,6 @@ func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) { return zw, err } -type dictWriter struct { - w io.Writer -} - -func (w *dictWriter) Write(b []byte) (n int, err error) { - return w.w.Write(b) -} - // A Writer takes data written to it and writes the compressed // form of that data to an underlying writer (see NewWriter). type Writer struct { @@ -805,11 +799,12 @@ func (w *Writer) Close() error { // the result of NewWriter or NewWriterDict called with dst // and w's level and dictionary. func (w *Writer) Reset(dst io.Writer) { - if dw, ok := w.d.w.writer.(*dictWriter); ok { + if len(w.dict) > 0 { // w was created with NewWriterDict - dw.w = dst - w.d.reset(dw) - w.d.fillWindow(w.dict) + w.d.reset(dst) + if dst != nil { + w.d.fillWindow(w.dict) + } } else { // w was created with NewWriter w.d.reset(dst) diff --git a/vendor/github.com/klauspost/compress/flate/fast_encoder.go b/vendor/github.com/klauspost/compress/flate/fast_encoder.go index b0a470f92e0e..4a73e1bdd30a 100644 --- a/vendor/github.com/klauspost/compress/flate/fast_encoder.go +++ b/vendor/github.com/klauspost/compress/flate/fast_encoder.go @@ -35,17 +35,17 @@ func newFastEnc(level int) fastEnc { } const ( - tableBits = 16 // Bits used in the table + tableBits = 15 // Bits used in the table tableSize = 1 << tableBits // Size of the table tableShift = 32 - tableBits // Right-shift to get the tableBits most significant bits of a uint32. baseMatchOffset = 1 // The smallest match offset baseMatchLength = 3 // The smallest match length per the RFC section 3.2.5 maxMatchOffset = 1 << 15 // The largest match offset - bTableBits = 18 // Bits used in the big tables - bTableSize = 1 << bTableBits // Size of the table - allocHistory = maxMatchOffset * 10 // Size to preallocate for history. - bufferReset = (1 << 31) - allocHistory - maxStoreBlockSize // Reset the buffer offset when reaching this. + bTableBits = 17 // Bits used in the big tables + bTableSize = 1 << bTableBits // Size of the table + allocHistory = maxStoreBlockSize * 10 // Size to preallocate for history. + bufferReset = (1 << 31) - allocHistory - maxStoreBlockSize - 1 // Reset the buffer offset when reaching this. ) const ( @@ -92,7 +92,6 @@ func hash(u uint32) uint32 { } type tableEntry struct { - val uint32 offset int32 } @@ -128,7 +127,7 @@ func (e *fastGen) addBlock(src []byte) int32 { // 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 hash4u(u uint32, h uint8) uint32 { - return (u * prime4bytes) >> ((32 - h) & 31) + return (u * prime4bytes) >> ((32 - h) & reg8SizeMask32) } type tableEntryPrev struct { @@ -139,25 +138,25 @@ type tableEntryPrev struct { // 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) + return (uint32(u) * prime4bytes) >> ((32 - h) & reg8SizeMask32) } // 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)) + return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & reg8SizeMask64)) } // 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)) + return uint32((u * prime8bytes) >> ((64 - h) & reg8SizeMask64)) } // 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)) + return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & reg8SizeMask64)) } // matchlen will return the match length between offsets and t in src. @@ -210,16 +209,14 @@ func (e *fastGen) matchlenLong(s, t int32, src []byte) int32 { // Reset the encoding table. func (e *fastGen) Reset() { - if cap(e.hist) < int(maxMatchOffset*8) { - l := maxMatchOffset * 8 - // Make it at least 1MB. - if l < 1<<20 { - l = 1 << 20 - } - e.hist = make([]byte, 0, l) + if cap(e.hist) < allocHistory { + e.hist = make([]byte, 0, allocHistory) + } + // We offset current position so everything will be out of reach. + // If we are above the buffer reset it will be cleared anyway since len(hist) == 0. + if e.cur <= bufferReset { + e.cur += maxMatchOffset + int32(len(e.hist)) } - // We offset current position so everything will be out of reach - e.cur += maxMatchOffset + int32(len(e.hist)) e.hist = e.hist[:0] } diff --git a/vendor/github.com/klauspost/compress/flate/gen_inflate.go b/vendor/github.com/klauspost/compress/flate/gen_inflate.go new file mode 100644 index 000000000000..35fc072a3efa --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/gen_inflate.go @@ -0,0 +1,294 @@ +// +build generate + +//go:generate go run $GOFILE && gofmt -w inflate_gen.go + +package main + +import ( + "os" + "strings" +) + +func main() { + f, err := os.Create("inflate_gen.go") + if err != nil { + panic(err) + } + defer f.Close() + types := []string{"*bytes.Buffer", "*bytes.Reader", "*bufio.Reader", "*strings.Reader"} + names := []string{"BytesBuffer", "BytesReader", "BufioReader", "StringsReader"} + imports := []string{"bytes", "bufio", "io", "strings", "math/bits"} + f.WriteString(`// Code generated by go generate gen_inflate.go. DO NOT EDIT. + +package flate + +import ( +`) + + for _, imp := range imports { + f.WriteString("\t\"" + imp + "\"\n") + } + f.WriteString(")\n\n") + + template := ` + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) $FUNCNAME$() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.($TYPE$) + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + f.dict.writeByte(byte(v)) + if f.dict.availWrite() == 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).$FUNCNAME$ + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for f.nb < n { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + length += int(f.b & uint32(1<<(n®SizeMaskUint32)-1)) + f.b >>= n & regSizeMaskUint32 + f.nb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + + var dist uint32 + if f.hd == nil { + for f.nb < 5 { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hd.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for f.nb < nb { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + f.nb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + default: + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(f.dict.histSize()) { + if debugDecode { + fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).$FUNCNAME$ // We need to continue this work + f.stepState = stateDict + return + } + goto readLiteral + } +} + +` + for i, t := range types { + s := strings.Replace(template, "$FUNCNAME$", "huffman"+names[i], -1) + s = strings.Replace(s, "$TYPE$", t, -1) + f.WriteString(s) + } + f.WriteString("func (f *decompressor) huffmanBlockDecoder() func() {\n") + f.WriteString("\tswitch f.r.(type) {\n") + for i, t := range types { + f.WriteString("\t\tcase " + t + ":\n") + f.WriteString("\t\t\treturn f.huffman" + names[i] + "\n") + } + f.WriteString("\t\tdefault:\n") + f.WriteString("\t\t\treturn f.huffmanBlockGeneric") + f.WriteString("\t}\n}\n") +} diff --git a/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go index dd74ffb87232..208d66711df8 100644 --- a/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go +++ b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go @@ -93,12 +93,12 @@ type huffmanBitWriter struct { err error lastHeader int // Set between 0 (reused block can be up to 2x the size) - logReusePenalty uint - lastHuffMan bool - bytes [256]byte - literalFreq [lengthCodesStart + 32]uint16 - offsetFreq [32]uint16 - codegenFreq [codegenCodeCount]uint16 + logNewTablePenalty uint + lastHuffMan bool + bytes [256]byte + literalFreq [lengthCodesStart + 32]uint16 + offsetFreq [32]uint16 + codegenFreq [codegenCodeCount]uint16 // codegen must have an extra space for the final symbol. codegen [literalCount + offsetCodeCount + 1]uint8 @@ -119,7 +119,7 @@ type huffmanBitWriter struct { // If lastHuffMan is set, a table for outputting literals has been generated and offsets are invalid. // // An incoming block estimates the output size of a new table using a 'fresh' by calculating the -// optimal size and adding a penalty in 'logReusePenalty'. +// optimal size and adding a penalty in 'logNewTablePenalty'. // A Huffman table is not optimal, which is why we add a penalty, and generating a new table // is slower both for compression and decompression. @@ -177,6 +177,11 @@ func (w *huffmanBitWriter) flush() { w.nbits = 0 return } + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } n := w.nbytes for w.nbits != 0 { w.bytes[n] = byte(w.bits) @@ -201,7 +206,7 @@ func (w *huffmanBitWriter) write(b []byte) { } func (w *huffmanBitWriter) writeBits(b int32, nb uint16) { - w.bits |= uint64(b) << (w.nbits & 63) + w.bits |= uint64(b) << (w.nbits & reg16SizeMask64) w.nbits += nb if w.nbits >= 48 { w.writeOutBits() @@ -349,6 +354,13 @@ func (w *huffmanBitWriter) headerSize() (size, numCodegens int) { int(w.codegenFreq[18])*7, numCodegens } +// dynamicSize returns the size of dynamically encoded data in bits. +func (w *huffmanBitWriter) dynamicReuseSize(litEnc, offEnc *huffmanEncoder) (size int) { + size = litEnc.bitLength(w.literalFreq[:]) + + offEnc.bitLength(w.offsetFreq[:]) + return size +} + // dynamicSize returns the size of dynamically encoded data in bits. func (w *huffmanBitWriter) dynamicSize(litEnc, offEnc *huffmanEncoder, extraBits int) (size, numCodegens int) { header, numCodegens := w.headerSize() @@ -451,12 +463,12 @@ func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, n i := 0 for { - var codeWord int = int(w.codegen[i]) + var codeWord = uint32(w.codegen[i]) i++ if codeWord == badCode { break } - w.writeCode(w.codegenEncoding.codes[uint32(codeWord)]) + w.writeCode(w.codegenEncoding.codes[codeWord]) switch codeWord { case 16: @@ -472,6 +484,9 @@ func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, n } } +// writeStoredHeader will write a stored header. +// If the stored block is only used for EOF, +// it is replaced with a fixed huffman block. func (w *huffmanBitWriter) writeStoredHeader(length int, isEof bool) { if w.err != nil { return @@ -481,6 +496,16 @@ func (w *huffmanBitWriter) writeStoredHeader(length int, isEof bool) { w.writeCode(w.literalEncoding.codes[endBlockMarker]) w.lastHeader = 0 } + + // To write EOF, use a fixed encoding block. 10 bits instead of 5 bytes. + if length == 0 && isEof { + w.writeFixedHeader(isEof) + // EOB: 7 bits, value: 0 + w.writeBits(0, 7) + w.flush() + return + } + var flag int32 if isEof { flag = 1 @@ -587,8 +612,8 @@ func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []b tokens.AddEOB() } - // We cannot reuse pure huffman table. - if w.lastHuffMan && w.lastHeader > 0 { + // We cannot reuse pure huffman table, and must mark as EOF. + if (w.lastHuffMan || eof) && w.lastHeader > 0 { // We will not try to reuse. w.writeCode(w.literalEncoding.codes[endBlockMarker]) w.lastHeader = 0 @@ -602,14 +627,14 @@ func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []b var size int // Check if we should reuse. if w.lastHeader > 0 { - // Estimate size for using a new table + // Estimate size for using a new table. + // Use the previous header size as the best estimate. newSize := w.lastHeader + tokens.EstimatedBits() + newSize += newSize >> w.logNewTablePenalty // The estimated size is calculated as an optimal table. // We add a penalty to make it more realistic and re-use a bit more. - newSize += newSize >> (w.logReusePenalty & 31) - extra := w.extraBitSize() - reuseSize, _ := w.dynamicSize(w.literalEncoding, w.offsetEncoding, extra) + reuseSize := w.dynamicReuseSize(w.literalEncoding, w.offsetEncoding) + w.extraBitSize() // Check if a new table is better. if newSize < reuseSize { @@ -734,7 +759,7 @@ func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) } else { // inlined c := lengths[lengthCode&31] - w.bits |= uint64(c.code) << (w.nbits & 63) + w.bits |= uint64(c.code) << (w.nbits & reg16SizeMask64) w.nbits += c.len if w.nbits >= 48 { w.writeOutBits() @@ -754,7 +779,7 @@ func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) } else { // inlined c := offs[offsetCode&31] - w.bits |= uint64(c.code) << (w.nbits & 63) + w.bits |= uint64(c.code) << (w.nbits & reg16SizeMask64) w.nbits += c.len if w.nbits >= 48 { w.writeOutBits() @@ -801,21 +826,30 @@ func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) { } // Add everything as literals - estBits := histogramSize(input, w.literalFreq[:], !eof && !sync) + 15 + // We have to estimate the header size. + // Assume header is around 70 bytes: + // https://stackoverflow.com/a/25454430 + const guessHeaderSizeBits = 70 * 8 + estBits, estExtra := histogramSize(input, w.literalFreq[:], !eof && !sync) + estBits += w.lastHeader + 15 + if w.lastHeader == 0 { + estBits += guessHeaderSizeBits + } + estBits += estBits >> w.logNewTablePenalty // Store bytes, if we don't get a reasonable improvement. ssize, storable := w.storedSize(input) - if storable && ssize < (estBits+estBits>>4) { + if storable && ssize < estBits { w.writeStoredHeader(len(input), eof) w.writeBytes(input) return } if w.lastHeader > 0 { - size, _ := w.dynamicSize(w.literalEncoding, huffOffset, w.lastHeader) - estBits += estBits >> (w.logReusePenalty) + reuseSize := w.literalEncoding.bitLength(w.literalFreq[:256]) + estBits += estExtra - if estBits < size { + if estBits < reuseSize { // We owe an EOB w.writeCode(w.literalEncoding.codes[endBlockMarker]) w.lastHeader = 0 @@ -844,7 +878,7 @@ func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) { for _, t := range input { // Bitwriting inlined, ~30% speedup c := encoding[t] - w.bits |= uint64(c.code) << ((w.nbits) & 63) + w.bits |= uint64(c.code) << ((w.nbits) & reg16SizeMask64) w.nbits += c.len if w.nbits >= 48 { bits := w.bits diff --git a/vendor/github.com/klauspost/compress/flate/huffman_code.go b/vendor/github.com/klauspost/compress/flate/huffman_code.go index 1810c6898d0b..4c39a3018711 100644 --- a/vendor/github.com/klauspost/compress/flate/huffman_code.go +++ b/vendor/github.com/klauspost/compress/flate/huffman_code.go @@ -7,7 +7,6 @@ package flate import ( "math" "math/bits" - "sort" ) const ( @@ -25,8 +24,6 @@ type huffmanEncoder struct { codes []hcode freqcache []literalNode bitCount [17]int32 - lns byLiteral // stored to avoid repeated allocation in generate - lfs byFreq // stored to avoid repeated allocation in generate } type literalNode struct { @@ -112,8 +109,8 @@ func generateFixedOffsetEncoding() *huffmanEncoder { return h } -var fixedLiteralEncoding *huffmanEncoder = generateFixedLiteralEncoding() -var fixedOffsetEncoding *huffmanEncoder = generateFixedOffsetEncoding() +var fixedLiteralEncoding = generateFixedLiteralEncoding() +var fixedOffsetEncoding = generateFixedOffsetEncoding() func (h *huffmanEncoder) bitLength(freq []uint16) int { var total int @@ -270,7 +267,7 @@ func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalN // assigned in literal order (not frequency order). chunk := list[len(list)-int(bits):] - h.lns.sort(chunk) + sortByLiteral(chunk) for _, node := range chunk { h.codes[node.literal] = hcode{code: reverseBits(code, uint8(n)), len: uint16(n)} code++ @@ -315,7 +312,7 @@ func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) { } return } - h.lfs.sort(list) + sortByFreq(list) // Get the number of literals for each bit count bitCount := h.bitCounts(list, maxBits) @@ -323,59 +320,44 @@ func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) { h.assignEncodingAndSize(bitCount, list) } -type byLiteral []literalNode - -func (s *byLiteral) sort(a []literalNode) { - *s = byLiteral(a) - sort.Sort(s) -} - -func (s byLiteral) Len() int { return len(s) } - -func (s byLiteral) Less(i, j int) bool { - return s[i].literal < s[j].literal -} - -func (s byLiteral) Swap(i, j int) { s[i], s[j] = s[j], s[i] } - -type byFreq []literalNode - -func (s *byFreq) sort(a []literalNode) { - *s = byFreq(a) - sort.Sort(s) -} - -func (s byFreq) Len() int { return len(s) } - -func (s byFreq) Less(i, j int) bool { - if s[i].freq == s[j].freq { - return s[i].literal < s[j].literal +func atLeastOne(v float32) float32 { + if v < 1 { + return 1 } - return s[i].freq < s[j].freq + return v } -func (s byFreq) Swap(i, j int) { s[i], s[j] = s[j], s[i] } - // histogramSize accumulates a histogram of b in h. // An estimated size in bits is returned. // Unassigned values are assigned '1' in the histogram. // len(h) must be >= 256, and h's elements must be all zeroes. -func histogramSize(b []byte, h []uint16, fill bool) int { +func histogramSize(b []byte, h []uint16, fill bool) (int, int) { h = h[:256] for _, t := range b { h[t]++ } - invTotal := 1.0 / float64(len(b)) - shannon := 0.0 - single := math.Ceil(-math.Log2(invTotal)) - for i, v := range h[:] { - if v > 0 { - n := float64(v) - shannon += math.Ceil(-math.Log2(n*invTotal) * n) - } else if fill { - shannon += single - h[i] = 1 + invTotal := 1.0 / float32(len(b)) + shannon := float32(0.0) + var extra float32 + if fill { + oneBits := atLeastOne(-mFastLog2(invTotal)) + for i, v := range h[:] { + if v > 0 { + n := float32(v) + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n + } else { + h[i] = 1 + extra += oneBits + } + } + } else { + for _, v := range h[:] { + if v > 0 { + n := float32(v) + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n + } } } - return int(shannon + 0.99) + + return int(shannon + 0.99), int(extra + 0.99) } diff --git a/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go b/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go new file mode 100644 index 000000000000..207780299007 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go @@ -0,0 +1,178 @@ +// Copyright 2009 The 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 flate + +// Sort sorts data. +// It makes one call to data.Len to determine n, and O(n*log(n)) calls to +// data.Less and data.Swap. The sort is not guaranteed to be stable. +func sortByFreq(data []literalNode) { + n := len(data) + quickSortByFreq(data, 0, n, maxDepth(n)) +} + +func quickSortByFreq(data []literalNode, a, b, maxDepth int) { + for b-a > 12 { // Use ShellSort for slices <= 12 elements + if maxDepth == 0 { + heapSort(data, a, b) + return + } + maxDepth-- + mlo, mhi := doPivotByFreq(data, a, b) + // Avoiding recursion on the larger subproblem guarantees + // a stack depth of at most lg(b-a). + if mlo-a < b-mhi { + quickSortByFreq(data, a, mlo, maxDepth) + a = mhi // i.e., quickSortByFreq(data, mhi, b) + } else { + quickSortByFreq(data, mhi, b, maxDepth) + b = mlo // i.e., quickSortByFreq(data, a, mlo) + } + } + if b-a > 1 { + // Do ShellSort pass with gap 6 + // It could be written in this simplified form cause b-a <= 12 + for i := a + 6; i < b; i++ { + if data[i].freq == data[i-6].freq && data[i].literal < data[i-6].literal || data[i].freq < data[i-6].freq { + data[i], data[i-6] = data[i-6], data[i] + } + } + insertionSortByFreq(data, a, b) + } +} + +// siftDownByFreq implements the heap property on data[lo, hi). +// first is an offset into the array where the root of the heap lies. +func siftDownByFreq(data []literalNode, lo, hi, first int) { + root := lo + for { + child := 2*root + 1 + if child >= hi { + break + } + if child+1 < hi && (data[first+child].freq == data[first+child+1].freq && data[first+child].literal < data[first+child+1].literal || data[first+child].freq < data[first+child+1].freq) { + child++ + } + if data[first+root].freq == data[first+child].freq && data[first+root].literal > data[first+child].literal || data[first+root].freq > data[first+child].freq { + return + } + data[first+root], data[first+child] = data[first+child], data[first+root] + root = child + } +} +func doPivotByFreq(data []literalNode, lo, hi int) (midlo, midhi int) { + m := int(uint(lo+hi) >> 1) // Written like this to avoid integer overflow. + if hi-lo > 40 { + // Tukey's ``Ninther,'' median of three medians of three. + s := (hi - lo) / 8 + medianOfThreeSortByFreq(data, lo, lo+s, lo+2*s) + medianOfThreeSortByFreq(data, m, m-s, m+s) + medianOfThreeSortByFreq(data, hi-1, hi-1-s, hi-1-2*s) + } + medianOfThreeSortByFreq(data, lo, m, hi-1) + + // Invariants are: + // data[lo] = pivot (set up by ChoosePivot) + // data[lo < i < a] < pivot + // data[a <= i < b] <= pivot + // data[b <= i < c] unexamined + // data[c <= i < hi-1] > pivot + // data[hi-1] >= pivot + pivot := lo + a, c := lo+1, hi-1 + + for ; a < c && (data[a].freq == data[pivot].freq && data[a].literal < data[pivot].literal || data[a].freq < data[pivot].freq); a++ { + } + b := a + for { + for ; b < c && (data[pivot].freq == data[b].freq && data[pivot].literal > data[b].literal || data[pivot].freq > data[b].freq); b++ { // data[b] <= pivot + } + for ; b < c && (data[pivot].freq == data[c-1].freq && data[pivot].literal < data[c-1].literal || data[pivot].freq < data[c-1].freq); c-- { // data[c-1] > pivot + } + if b >= c { + break + } + // data[b] > pivot; data[c-1] <= pivot + data[b], data[c-1] = data[c-1], data[b] + b++ + c-- + } + // If hi-c<3 then there are duplicates (by property of median of nine). + // Let's be a bit more conservative, and set border to 5. + protect := hi-c < 5 + if !protect && hi-c < (hi-lo)/4 { + // Lets test some points for equality to pivot + dups := 0 + if data[pivot].freq == data[hi-1].freq && data[pivot].literal > data[hi-1].literal || data[pivot].freq > data[hi-1].freq { // data[hi-1] = pivot + data[c], data[hi-1] = data[hi-1], data[c] + c++ + dups++ + } + if data[b-1].freq == data[pivot].freq && data[b-1].literal > data[pivot].literal || data[b-1].freq > data[pivot].freq { // data[b-1] = pivot + b-- + dups++ + } + // m-lo = (hi-lo)/2 > 6 + // b-lo > (hi-lo)*3/4-1 > 8 + // ==> m < b ==> data[m] <= pivot + if data[m].freq == data[pivot].freq && data[m].literal > data[pivot].literal || data[m].freq > data[pivot].freq { // data[m] = pivot + data[m], data[b-1] = data[b-1], data[m] + b-- + dups++ + } + // if at least 2 points are equal to pivot, assume skewed distribution + protect = dups > 1 + } + if protect { + // Protect against a lot of duplicates + // Add invariant: + // data[a <= i < b] unexamined + // data[b <= i < c] = pivot + for { + for ; a < b && (data[b-1].freq == data[pivot].freq && data[b-1].literal > data[pivot].literal || data[b-1].freq > data[pivot].freq); b-- { // data[b] == pivot + } + for ; a < b && (data[a].freq == data[pivot].freq && data[a].literal < data[pivot].literal || data[a].freq < data[pivot].freq); a++ { // data[a] < pivot + } + if a >= b { + break + } + // data[a] == pivot; data[b-1] < pivot + data[a], data[b-1] = data[b-1], data[a] + a++ + b-- + } + } + // Swap pivot into middle + data[pivot], data[b-1] = data[b-1], data[pivot] + return b - 1, c +} + +// Insertion sort +func insertionSortByFreq(data []literalNode, a, b int) { + for i := a + 1; i < b; i++ { + for j := i; j > a && (data[j].freq == data[j-1].freq && data[j].literal < data[j-1].literal || data[j].freq < data[j-1].freq); j-- { + data[j], data[j-1] = data[j-1], data[j] + } + } +} + +// quickSortByFreq, loosely following Bentley and McIlroy, +// ``Engineering a Sort Function,'' SP&E November 1993. + +// medianOfThreeSortByFreq moves the median of the three values data[m0], data[m1], data[m2] into data[m1]. +func medianOfThreeSortByFreq(data []literalNode, m1, m0, m2 int) { + // sort 3 elements + if data[m1].freq == data[m0].freq && data[m1].literal < data[m0].literal || data[m1].freq < data[m0].freq { + data[m1], data[m0] = data[m0], data[m1] + } + // data[m0] <= data[m1] + if data[m2].freq == data[m1].freq && data[m2].literal < data[m1].literal || data[m2].freq < data[m1].freq { + data[m2], data[m1] = data[m1], data[m2] + // data[m0] <= data[m2] && data[m1] < data[m2] + if data[m1].freq == data[m0].freq && data[m1].literal < data[m0].literal || data[m1].freq < data[m0].freq { + data[m1], data[m0] = data[m0], data[m1] + } + } + // now data[m0] <= data[m1] <= data[m2] +} diff --git a/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go b/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go new file mode 100644 index 000000000000..93f1aea109e1 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go @@ -0,0 +1,201 @@ +// Copyright 2009 The 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 flate + +// Sort sorts data. +// It makes one call to data.Len to determine n, and O(n*log(n)) calls to +// data.Less and data.Swap. The sort is not guaranteed to be stable. +func sortByLiteral(data []literalNode) { + n := len(data) + quickSort(data, 0, n, maxDepth(n)) +} + +func quickSort(data []literalNode, a, b, maxDepth int) { + for b-a > 12 { // Use ShellSort for slices <= 12 elements + if maxDepth == 0 { + heapSort(data, a, b) + return + } + maxDepth-- + mlo, mhi := doPivot(data, a, b) + // Avoiding recursion on the larger subproblem guarantees + // a stack depth of at most lg(b-a). + if mlo-a < b-mhi { + quickSort(data, a, mlo, maxDepth) + a = mhi // i.e., quickSort(data, mhi, b) + } else { + quickSort(data, mhi, b, maxDepth) + b = mlo // i.e., quickSort(data, a, mlo) + } + } + if b-a > 1 { + // Do ShellSort pass with gap 6 + // It could be written in this simplified form cause b-a <= 12 + for i := a + 6; i < b; i++ { + if data[i].literal < data[i-6].literal { + data[i], data[i-6] = data[i-6], data[i] + } + } + insertionSort(data, a, b) + } +} +func heapSort(data []literalNode, a, b int) { + first := a + lo := 0 + hi := b - a + + // Build heap with greatest element at top. + for i := (hi - 1) / 2; i >= 0; i-- { + siftDown(data, i, hi, first) + } + + // Pop elements, largest first, into end of data. + for i := hi - 1; i >= 0; i-- { + data[first], data[first+i] = data[first+i], data[first] + siftDown(data, lo, i, first) + } +} + +// siftDown implements the heap property on data[lo, hi). +// first is an offset into the array where the root of the heap lies. +func siftDown(data []literalNode, lo, hi, first int) { + root := lo + for { + child := 2*root + 1 + if child >= hi { + break + } + if child+1 < hi && data[first+child].literal < data[first+child+1].literal { + child++ + } + if data[first+root].literal > data[first+child].literal { + return + } + data[first+root], data[first+child] = data[first+child], data[first+root] + root = child + } +} +func doPivot(data []literalNode, lo, hi int) (midlo, midhi int) { + m := int(uint(lo+hi) >> 1) // Written like this to avoid integer overflow. + if hi-lo > 40 { + // Tukey's ``Ninther,'' median of three medians of three. + s := (hi - lo) / 8 + medianOfThree(data, lo, lo+s, lo+2*s) + medianOfThree(data, m, m-s, m+s) + medianOfThree(data, hi-1, hi-1-s, hi-1-2*s) + } + medianOfThree(data, lo, m, hi-1) + + // Invariants are: + // data[lo] = pivot (set up by ChoosePivot) + // data[lo < i < a] < pivot + // data[a <= i < b] <= pivot + // data[b <= i < c] unexamined + // data[c <= i < hi-1] > pivot + // data[hi-1] >= pivot + pivot := lo + a, c := lo+1, hi-1 + + for ; a < c && data[a].literal < data[pivot].literal; a++ { + } + b := a + for { + for ; b < c && data[pivot].literal > data[b].literal; b++ { // data[b] <= pivot + } + for ; b < c && data[pivot].literal < data[c-1].literal; c-- { // data[c-1] > pivot + } + if b >= c { + break + } + // data[b] > pivot; data[c-1] <= pivot + data[b], data[c-1] = data[c-1], data[b] + b++ + c-- + } + // If hi-c<3 then there are duplicates (by property of median of nine). + // Let's be a bit more conservative, and set border to 5. + protect := hi-c < 5 + if !protect && hi-c < (hi-lo)/4 { + // Lets test some points for equality to pivot + dups := 0 + if data[pivot].literal > data[hi-1].literal { // data[hi-1] = pivot + data[c], data[hi-1] = data[hi-1], data[c] + c++ + dups++ + } + if data[b-1].literal > data[pivot].literal { // data[b-1] = pivot + b-- + dups++ + } + // m-lo = (hi-lo)/2 > 6 + // b-lo > (hi-lo)*3/4-1 > 8 + // ==> m < b ==> data[m] <= pivot + if data[m].literal > data[pivot].literal { // data[m] = pivot + data[m], data[b-1] = data[b-1], data[m] + b-- + dups++ + } + // if at least 2 points are equal to pivot, assume skewed distribution + protect = dups > 1 + } + if protect { + // Protect against a lot of duplicates + // Add invariant: + // data[a <= i < b] unexamined + // data[b <= i < c] = pivot + for { + for ; a < b && data[b-1].literal > data[pivot].literal; b-- { // data[b] == pivot + } + for ; a < b && data[a].literal < data[pivot].literal; a++ { // data[a] < pivot + } + if a >= b { + break + } + // data[a] == pivot; data[b-1] < pivot + data[a], data[b-1] = data[b-1], data[a] + a++ + b-- + } + } + // Swap pivot into middle + data[pivot], data[b-1] = data[b-1], data[pivot] + return b - 1, c +} + +// Insertion sort +func insertionSort(data []literalNode, a, b int) { + for i := a + 1; i < b; i++ { + for j := i; j > a && data[j].literal < data[j-1].literal; j-- { + data[j], data[j-1] = data[j-1], data[j] + } + } +} + +// maxDepth returns a threshold at which quicksort should switch +// to heapsort. It returns 2*ceil(lg(n+1)). +func maxDepth(n int) int { + var depth int + for i := n; i > 0; i >>= 1 { + depth++ + } + return depth * 2 +} + +// medianOfThree moves the median of the three values data[m0], data[m1], data[m2] into data[m1]. +func medianOfThree(data []literalNode, m1, m0, m2 int) { + // sort 3 elements + if data[m1].literal < data[m0].literal { + data[m1], data[m0] = data[m0], data[m1] + } + // data[m0] <= data[m1] + if data[m2].literal < data[m1].literal { + data[m2], data[m1] = data[m1], data[m2] + // data[m0] <= data[m2] && data[m1] < data[m2] + if data[m1].literal < data[m0].literal { + data[m1], data[m0] = data[m0], data[m1] + } + } + // now data[m0] <= data[m1] <= data[m2] +} diff --git a/vendor/github.com/klauspost/compress/flate/inflate.go b/vendor/github.com/klauspost/compress/flate/inflate.go index 6dc5b5d06e30..16bc51408e07 100644 --- a/vendor/github.com/klauspost/compress/flate/inflate.go +++ b/vendor/github.com/klauspost/compress/flate/inflate.go @@ -29,6 +29,13 @@ const ( debugDecode = false ) +// Value of length - 3 and extra bits. +type lengthExtra struct { + length, extra uint8 +} + +var decCodeToLen = [32]lengthExtra{{length: 0x0, extra: 0x0}, {length: 0x1, extra: 0x0}, {length: 0x2, extra: 0x0}, {length: 0x3, extra: 0x0}, {length: 0x4, extra: 0x0}, {length: 0x5, extra: 0x0}, {length: 0x6, extra: 0x0}, {length: 0x7, extra: 0x0}, {length: 0x8, extra: 0x1}, {length: 0xa, extra: 0x1}, {length: 0xc, extra: 0x1}, {length: 0xe, extra: 0x1}, {length: 0x10, extra: 0x2}, {length: 0x14, extra: 0x2}, {length: 0x18, extra: 0x2}, {length: 0x1c, extra: 0x2}, {length: 0x20, extra: 0x3}, {length: 0x28, extra: 0x3}, {length: 0x30, extra: 0x3}, {length: 0x38, extra: 0x3}, {length: 0x40, extra: 0x4}, {length: 0x50, extra: 0x4}, {length: 0x60, extra: 0x4}, {length: 0x70, extra: 0x4}, {length: 0x80, extra: 0x5}, {length: 0xa0, extra: 0x5}, {length: 0xc0, extra: 0x5}, {length: 0xe0, extra: 0x5}, {length: 0xff, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}} + // Initialize the fixedHuffmanDecoder only once upon first use. var fixedOnce sync.Once var fixedHuffmanDecoder huffmanDecoder @@ -106,7 +113,7 @@ const ( ) type huffmanDecoder struct { - min int // the minimum code length + maxRead int // the maximum number of bits we can read and not overread chunks *[huffmanNumChunks]uint16 // chunks as described above links [][]uint16 // overflow links linkMask uint32 // mask the width of the link table @@ -126,12 +133,12 @@ func (h *huffmanDecoder) init(lengths []int) bool { if h.chunks == nil { h.chunks = &[huffmanNumChunks]uint16{} } - if h.min != 0 { + if h.maxRead != 0 { *h = huffmanDecoder{chunks: h.chunks, links: h.links} } // Count number of codes of each length, - // compute min and max length. + // compute maxRead and max length. var count [maxCodeLen]int var min, max int for _, n := range lengths { @@ -178,7 +185,7 @@ func (h *huffmanDecoder) init(lengths []int) bool { return false } - h.min = min + h.maxRead = min chunks := h.chunks[:] for i := range chunks { chunks[i] = 0 @@ -295,10 +302,6 @@ type decompressor struct { r Reader roffset int64 - // Input bits, in top of b. - b uint32 - nb uint - // Huffman decoders for literal/length, distance. h1, h2 huffmanDecoder @@ -309,19 +312,24 @@ type decompressor struct { // Output history, buffer. dict dictDecoder - // Temporary buffer (avoids repeated allocation). - buf [4]byte - // Next step in the decompression, // and decompression state. step func(*decompressor) stepState int - final bool err error toRead []byte hl, hd *huffmanDecoder copyLen int copyDist int + + // Temporary buffer (avoids repeated allocation). + buf [4]byte + + // Input bits, in top of b. + b uint32 + + nb uint + final bool } func (f *decompressor) nextBlock() { @@ -342,7 +350,7 @@ func (f *decompressor) nextBlock() { // compressed, fixed Huffman tables f.hl = &fixedHuffmanDecoder f.hd = nil - f.huffmanBlock() + f.huffmanBlockDecoder()() case 2: // compressed, dynamic Huffman tables if f.err = f.readHuffman(); f.err != nil { @@ -350,7 +358,7 @@ func (f *decompressor) nextBlock() { } f.hl = &f.h1 f.hd = &f.h2 - f.huffmanBlock() + f.huffmanBlockDecoder()() default: // 3 is reserved. if debugDecode { @@ -521,8 +529,8 @@ func (f *decompressor) readHuffman() error { return err } } - rep += int(f.b & uint32(1<>= nb + rep += int(f.b & uint32(1<<(nb®SizeMaskUint32)-1)) + f.b >>= nb & regSizeMaskUint32 f.nb -= nb if i+rep > n { if debugDecode { @@ -543,12 +551,18 @@ func (f *decompressor) readHuffman() error { return CorruptInputError(f.roffset) } - // As an optimization, we can initialize the min bits to read at a time + // As an optimization, we can initialize the maxRead bits to read at a time // for the HLIT tree to the length of the EOB marker since we know that // every block must terminate with one. This preserves the property that // we never read any extra bytes after the end of the DEFLATE stream. - if f.h1.min < f.bits[endBlockMarker] { - f.h1.min = f.bits[endBlockMarker] + if f.h1.maxRead < f.bits[endBlockMarker] { + f.h1.maxRead = f.bits[endBlockMarker] + } + if !f.final { + // If not the final block, the smallest block possible is + // a predefined table, BTYPE=01, with a single EOB marker. + // This will take up 3 + 7 bits. + f.h1.maxRead += 10 } return nil @@ -558,7 +572,7 @@ func (f *decompressor) readHuffman() error { // hl and hd are the Huffman states for the lit/length values // and the distance values, respectively. If hd == nil, using the // fixed distance encoding associated with fixed Huffman blocks. -func (f *decompressor) huffmanBlock() { +func (f *decompressor) huffmanBlockGeneric() { const ( stateInit = iota // Zero value must be stateInit stateDict @@ -574,19 +588,64 @@ func (f *decompressor) huffmanBlock() { readLiteral: // Read literal and/or (length, distance) according to RFC section 3.2.3. { - v, err := f.huffSym(f.hl) - if err != nil { - f.err = err - return + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := f.r.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + v = int(chunk >> huffmanValueShift) + break + } + } } + var n uint // number of bits extra var length int + var err error switch { case v < 256: f.dict.writeByte(byte(v)) if f.dict.availWrite() == 0 { f.toRead = f.dict.readFlush() - f.step = (*decompressor).huffmanBlock + f.step = (*decompressor).huffmanBlockGeneric f.stepState = stateInit return } @@ -633,12 +692,12 @@ readLiteral: return } } - length += int(f.b & uint32(1<>= n + length += int(f.b & uint32(1<<(n®SizeMaskUint32)-1)) + f.b >>= n & regSizeMaskUint32 f.nb -= n } - var dist int + var dist uint32 if f.hd == nil { for f.nb < 5 { if err = f.moreBits(); err != nil { @@ -649,17 +708,19 @@ readLiteral: return } } - dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3))) f.b >>= 5 f.nb -= 5 } else { - if dist, err = f.huffSym(f.hd); err != nil { + sym, err := f.huffSym(f.hd) + if err != nil { if debugDecode { fmt.Println("huffsym:", err) } f.err = err return } + dist = uint32(sym) } switch { @@ -668,7 +729,7 @@ readLiteral: case dist < maxNumDist: nb := uint(dist-2) >> 1 // have 1 bit in bottom of dist, need nb more. - extra := (dist & 1) << nb + extra := (dist & 1) << (nb & regSizeMaskUint32) for f.nb < nb { if err = f.moreBits(); err != nil { if debugDecode { @@ -678,10 +739,10 @@ readLiteral: return } } - extra |= int(f.b & uint32(1<>= nb + extra |= f.b & uint32(1<<(nb®SizeMaskUint32)-1) + f.b >>= nb & regSizeMaskUint32 f.nb -= nb - dist = 1<<(nb+1) + 1 + extra + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra default: if debugDecode { fmt.Println("dist too big:", dist, maxNumDist) @@ -691,7 +752,7 @@ readLiteral: } // No check on length; encoding can be prescient. - if dist > f.dict.histSize() { + if dist > uint32(f.dict.histSize()) { if debugDecode { fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) } @@ -699,7 +760,7 @@ readLiteral: return } - f.copyLen, f.copyDist = length, dist + f.copyLen, f.copyDist = length, int(dist) goto copyHistory } @@ -714,7 +775,7 @@ copyHistory: if f.dict.availWrite() == 0 || f.copyLen > 0 { f.toRead = f.dict.readFlush() - f.step = (*decompressor).huffmanBlock // We need to continue this work + f.step = (*decompressor).huffmanBlockGeneric // We need to continue this work f.stepState = stateDict return } @@ -726,21 +787,33 @@ copyHistory: func (f *decompressor) dataBlock() { // Uncompressed. // Discard current half-byte. - f.nb = 0 - f.b = 0 + left := (f.nb) & 7 + f.nb -= left + f.b >>= left + + offBytes := f.nb >> 3 + // Unfilled values will be overwritten. + f.buf[0] = uint8(f.b) + f.buf[1] = uint8(f.b >> 8) + f.buf[2] = uint8(f.b >> 16) + f.buf[3] = uint8(f.b >> 24) + + f.roffset += int64(offBytes) + f.nb, f.b = 0, 0 // Length then ones-complement of length. - nr, err := io.ReadFull(f.r, f.buf[0:4]) + nr, err := io.ReadFull(f.r, f.buf[offBytes:4]) f.roffset += int64(nr) if err != nil { f.err = noEOF(err) return } - n := int(f.buf[0]) | int(f.buf[1])<<8 - nn := int(f.buf[2]) | int(f.buf[3])<<8 - if uint16(nn) != uint16(^n) { + n := uint16(f.buf[0]) | uint16(f.buf[1])<<8 + nn := uint16(f.buf[2]) | uint16(f.buf[3])<<8 + if nn != ^n { if debugDecode { - fmt.Println("uint16(nn) != uint16(^n)", nn, ^n) + ncomp := ^n + fmt.Println("uint16(nn) != uint16(^n)", nn, ncomp) } f.err = CorruptInputError(f.roffset) return @@ -752,7 +825,7 @@ func (f *decompressor) dataBlock() { return } - f.copyLen = n + f.copyLen = int(n) f.copyData() } @@ -805,7 +878,7 @@ func (f *decompressor) moreBits() error { return noEOF(err) } f.roffset++ - f.b |= uint32(c) << f.nb + f.b |= uint32(c) << (f.nb & regSizeMaskUint32) f.nb += 8 return nil } @@ -816,7 +889,7 @@ func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { // with single element, huffSym must error on these two edge cases. In both // cases, the chunks slice will be 0 for the invalid sequence, leading it // satisfy the n == 0 check below. - n := uint(h.min) + n := uint(h.maxRead) // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, // but is smart enough to keep local variables in registers, so use nb and b, // inline call to moreBits and reassign b,nb back to f on return. @@ -830,7 +903,7 @@ func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { return 0, noEOF(err) } f.roffset++ - b |= uint32(c) << (nb & 31) + b |= uint32(c) << (nb & regSizeMaskUint32) nb += 8 } chunk := h.chunks[b&(huffmanNumChunks-1)] @@ -849,7 +922,7 @@ func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { f.err = CorruptInputError(f.roffset) return 0, f.err } - f.b = b >> (n & 31) + f.b = b >> (n & regSizeMaskUint32) f.nb = nb - n return int(chunk >> huffmanValueShift), nil } diff --git a/vendor/github.com/klauspost/compress/flate/inflate_gen.go b/vendor/github.com/klauspost/compress/flate/inflate_gen.go new file mode 100644 index 000000000000..cc6db27925ce --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/inflate_gen.go @@ -0,0 +1,1002 @@ +// Code generated by go generate gen_inflate.go. DO NOT EDIT. + +package flate + +import ( + "bufio" + "bytes" + "fmt" + "math/bits" + "strings" +) + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanBytesBuffer() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*bytes.Buffer) + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + f.dict.writeByte(byte(v)) + if f.dict.availWrite() == 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanBytesBuffer + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for f.nb < n { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + length += int(f.b & uint32(1<<(n®SizeMaskUint32)-1)) + f.b >>= n & regSizeMaskUint32 + f.nb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + + var dist uint32 + if f.hd == nil { + for f.nb < 5 { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hd.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for f.nb < nb { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + f.nb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + default: + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(f.dict.histSize()) { + if debugDecode { + fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanBytesBuffer // We need to continue this work + f.stepState = stateDict + return + } + goto readLiteral + } +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanBytesReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*bytes.Reader) + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + f.dict.writeByte(byte(v)) + if f.dict.availWrite() == 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanBytesReader + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for f.nb < n { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + length += int(f.b & uint32(1<<(n®SizeMaskUint32)-1)) + f.b >>= n & regSizeMaskUint32 + f.nb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + + var dist uint32 + if f.hd == nil { + for f.nb < 5 { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hd.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for f.nb < nb { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + f.nb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + default: + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(f.dict.histSize()) { + if debugDecode { + fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanBytesReader // We need to continue this work + f.stepState = stateDict + return + } + goto readLiteral + } +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanBufioReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*bufio.Reader) + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + f.dict.writeByte(byte(v)) + if f.dict.availWrite() == 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanBufioReader + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for f.nb < n { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + length += int(f.b & uint32(1<<(n®SizeMaskUint32)-1)) + f.b >>= n & regSizeMaskUint32 + f.nb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + + var dist uint32 + if f.hd == nil { + for f.nb < 5 { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hd.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for f.nb < nb { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + f.nb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + default: + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(f.dict.histSize()) { + if debugDecode { + fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanBufioReader // We need to continue this work + f.stepState = stateDict + return + } + goto readLiteral + } +} + +// Decode a single Huffman block from f. +// hl and hd are the Huffman states for the lit/length values +// and the distance values, respectively. If hd == nil, using the +// fixed distance encoding associated with fixed Huffman blocks. +func (f *decompressor) huffmanStringsReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(*strings.Reader) + + switch f.stepState { + case stateInit: + goto readLiteral + case stateDict: + goto copyHistory + } + +readLiteral: + // Read literal and/or (length, distance) according to RFC section 3.2.3. + { + var v int + { + // Inlined v, err := f.huffSym(f.hl) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hl.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + f.dict.writeByte(byte(v)) + if f.dict.availWrite() == 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanStringsReader + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for f.nb < n { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + length += int(f.b & uint32(1<<(n®SizeMaskUint32)-1)) + f.b >>= n & regSizeMaskUint32 + f.nb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + + var dist uint32 + if f.hd == nil { + for f.nb < 5 { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + } + dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := fr.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & regSizeMaskUint32) + nb += 8 + } + chunk := f.hd.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= nb { + if n == 0 { + f.b = b + f.nb = nb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + f.b = b >> (n & regSizeMaskUint32) + f.nb = nb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for f.nb < nb { + c, err := fr.ReadByte() + if err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + f.nb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + default: + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(f.dict.histSize()) { + if debugDecode { + fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).huffmanStringsReader // We need to continue this work + f.stepState = stateDict + return + } + goto readLiteral + } +} + +func (f *decompressor) huffmanBlockDecoder() func() { + switch f.r.(type) { + case *bytes.Buffer: + return f.huffmanBytesBuffer + case *bytes.Reader: + return f.huffmanBytesReader + case *bufio.Reader: + return f.huffmanBufioReader + case *strings.Reader: + return f.huffmanStringsReader + default: + return f.huffmanBlockGeneric + } +} diff --git a/vendor/github.com/klauspost/compress/flate/level1.go b/vendor/github.com/klauspost/compress/flate/level1.go index 20de8f11f4f3..1e5eea3968aa 100644 --- a/vendor/github.com/klauspost/compress/flate/level1.go +++ b/vendor/github.com/klauspost/compress/flate/level1.go @@ -1,5 +1,7 @@ package flate +import "fmt" + // fastGen maintains the table for matches, // and the previous byte block for level 2. // This is the generic implementation. @@ -14,6 +16,9 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } // Protect against e.cur wraparound. for e.cur >= bufferReset { @@ -76,12 +81,12 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { } now := load6432(src, nextS) - e.table[nextHash] = tableEntry{offset: s + e.cur, val: cv} + e.table[nextHash] = tableEntry{offset: s + e.cur} nextHash = hash(uint32(now)) offset := s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == candidate.val { - e.table[nextHash] = tableEntry{offset: nextS + e.cur, val: uint32(now)} + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} break } @@ -91,11 +96,11 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { nextS++ candidate = e.table[nextHash] now >>= 8 - e.table[nextHash] = tableEntry{offset: s + e.cur, val: cv} + e.table[nextHash] = tableEntry{offset: s + e.cur} offset = s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == candidate.val { - e.table[nextHash] = tableEntry{offset: nextS + e.cur, val: uint32(now)} + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} break } cv = uint32(now) @@ -134,7 +139,7 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { // Index first pair after match end. if int(s+l+4) < len(src) { cv := load3232(src, s) - e.table[hash(cv)] = tableEntry{offset: s + e.cur, val: cv} + e.table[hash(cv)] = tableEntry{offset: s + e.cur} } goto emitRemainder } @@ -148,14 +153,14 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { x := load6432(src, s-2) o := e.cur + s - 2 prevHash := hash(uint32(x)) - e.table[prevHash] = tableEntry{offset: o, val: uint32(x)} + e.table[prevHash] = tableEntry{offset: o} x >>= 16 currHash := hash(uint32(x)) candidate = e.table[currHash] - e.table[currHash] = tableEntry{offset: o + 2, val: uint32(x)} + e.table[currHash] = tableEntry{offset: o + 2} offset := s - (candidate.offset - e.cur) - if offset > maxMatchOffset || uint32(x) != candidate.val { + if offset > maxMatchOffset || uint32(x) != load3232(src, candidate.offset-e.cur) { cv = uint32(x >> 8) s++ break diff --git a/vendor/github.com/klauspost/compress/flate/level2.go b/vendor/github.com/klauspost/compress/flate/level2.go index 7c824431e647..5b986a1944ea 100644 --- a/vendor/github.com/klauspost/compress/flate/level2.go +++ b/vendor/github.com/klauspost/compress/flate/level2.go @@ -1,5 +1,7 @@ package flate +import "fmt" + // fastGen maintains the table for matches, // and the previous byte block for level 2. // This is the generic implementation. @@ -16,6 +18,10 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { minNonLiteralBlockSize = 1 + 1 + inputMargin ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + // Protect against e.cur wraparound. for e.cur >= bufferReset { if len(e.hist) == 0 { @@ -77,12 +83,12 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { } candidate = e.table[nextHash] now := load6432(src, nextS) - e.table[nextHash] = tableEntry{offset: s + e.cur, val: cv} + e.table[nextHash] = tableEntry{offset: s + e.cur} nextHash = hash4u(uint32(now), bTableBits) offset := s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == candidate.val { - e.table[nextHash] = tableEntry{offset: nextS + e.cur, val: uint32(now)} + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} break } @@ -92,10 +98,10 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { nextS++ candidate = e.table[nextHash] now >>= 8 - e.table[nextHash] = tableEntry{offset: s + e.cur, val: cv} + e.table[nextHash] = tableEntry{offset: s + e.cur} offset = s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == candidate.val { + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { break } cv = uint32(now) @@ -142,7 +148,7 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { // Index first pair after match end. if int(s+l+4) < len(src) { cv := load3232(src, s) - e.table[hash4u(cv, bTableBits)] = tableEntry{offset: s + e.cur, val: cv} + e.table[hash4u(cv, bTableBits)] = tableEntry{offset: s + e.cur} } goto emitRemainder } @@ -151,15 +157,15 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { for i := s - l + 2; i < s-5; i += 7 { x := load6432(src, int32(i)) nextHash := hash4u(uint32(x), bTableBits) - e.table[nextHash] = tableEntry{offset: e.cur + i, val: uint32(x)} + e.table[nextHash] = tableEntry{offset: e.cur + i} // Skip one x >>= 16 nextHash = hash4u(uint32(x), bTableBits) - e.table[nextHash] = tableEntry{offset: e.cur + i + 2, val: uint32(x)} + e.table[nextHash] = tableEntry{offset: e.cur + i + 2} // Skip one x >>= 16 nextHash = hash4u(uint32(x), bTableBits) - e.table[nextHash] = tableEntry{offset: e.cur + i + 4, val: uint32(x)} + e.table[nextHash] = tableEntry{offset: e.cur + i + 4} } // We could immediately start working at s now, but to improve @@ -172,14 +178,14 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { o := e.cur + s - 2 prevHash := hash4u(uint32(x), bTableBits) prevHash2 := hash4u(uint32(x>>8), bTableBits) - e.table[prevHash] = tableEntry{offset: o, val: uint32(x)} - e.table[prevHash2] = tableEntry{offset: o + 1, val: uint32(x >> 8)} + e.table[prevHash] = tableEntry{offset: o} + e.table[prevHash2] = tableEntry{offset: o + 1} currHash := hash4u(uint32(x>>16), bTableBits) candidate = e.table[currHash] - e.table[currHash] = tableEntry{offset: o + 2, val: uint32(x >> 16)} + e.table[currHash] = tableEntry{offset: o + 2} offset := s - (candidate.offset - e.cur) - if offset > maxMatchOffset || uint32(x>>16) != candidate.val { + if offset > maxMatchOffset || uint32(x>>16) != load3232(src, candidate.offset-e.cur) { cv = uint32(x >> 24) s++ break diff --git a/vendor/github.com/klauspost/compress/flate/level3.go b/vendor/github.com/klauspost/compress/flate/level3.go index 4153d24c95fa..c22b4244a5c0 100644 --- a/vendor/github.com/klauspost/compress/flate/level3.go +++ b/vendor/github.com/klauspost/compress/flate/level3.go @@ -1,5 +1,7 @@ package flate +import "fmt" + // fastEncL3 type fastEncL3 struct { fastGen @@ -13,6 +15,10 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { minNonLiteralBlockSize = 1 + 1 + inputMargin ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } + // Protect against e.cur wraparound. for e.cur >= bufferReset { if len(e.hist) == 0 { @@ -75,22 +81,26 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { } candidates := e.table[nextHash] now := load3232(src, nextS) - e.table[nextHash] = tableEntryPrev{Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur, val: cv}} + + // Safe offset distance until s + 4... + minOffset := e.cur + s - (maxMatchOffset - 4) + e.table[nextHash] = tableEntryPrev{Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur}} // Check both candidates candidate = candidates.Cur - offset := s - (candidate.offset - e.cur) - if cv == candidate.val { - if offset > maxMatchOffset { - cv = now - // Previous will also be invalid, we have nothing. - continue - } - o2 := s - (candidates.Prev.offset - e.cur) - if cv != candidates.Prev.val || o2 > maxMatchOffset { + if candidate.offset < minOffset { + cv = now + // Previous will also be invalid, we have nothing. + continue + } + + if cv == load3232(src, candidate.offset-e.cur) { + if candidates.Prev.offset < minOffset || cv != load3232(src, candidates.Prev.offset-e.cur) { break } // Both match and are valid, pick longest. + offset := s - (candidate.offset - e.cur) + o2 := s - (candidates.Prev.offset - e.cur) l1, l2 := matchLen(src[s+4:], src[s-offset+4:]), matchLen(src[s+4:], src[s-o2+4:]) if l2 > l1 { candidate = candidates.Prev @@ -100,11 +110,8 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { // We only check if value mismatches. // Offset will always be invalid in other cases. candidate = candidates.Prev - if cv == candidate.val { - offset := s - (candidate.offset - e.cur) - if offset <= maxMatchOffset { - break - } + if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { + break } } cv = now @@ -152,7 +159,7 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { nextHash := hash(cv) e.table[nextHash] = tableEntryPrev{ Prev: e.table[nextHash].Cur, - Cur: tableEntry{offset: e.cur + t, val: cv}, + Cur: tableEntry{offset: e.cur + t}, } } goto emitRemainder @@ -164,21 +171,21 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { prevHash := hash(uint32(x)) e.table[prevHash] = tableEntryPrev{ Prev: e.table[prevHash].Cur, - Cur: tableEntry{offset: e.cur + s - 3, val: uint32(x)}, + Cur: tableEntry{offset: e.cur + s - 3}, } x >>= 8 prevHash = hash(uint32(x)) e.table[prevHash] = tableEntryPrev{ Prev: e.table[prevHash].Cur, - Cur: tableEntry{offset: e.cur + s - 2, val: uint32(x)}, + Cur: tableEntry{offset: e.cur + s - 2}, } x >>= 8 prevHash = hash(uint32(x)) e.table[prevHash] = tableEntryPrev{ Prev: e.table[prevHash].Cur, - Cur: tableEntry{offset: e.cur + s - 1, val: uint32(x)}, + Cur: tableEntry{offset: e.cur + s - 1}, } x >>= 8 currHash := hash(uint32(x)) @@ -186,21 +193,18 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { cv = uint32(x) e.table[currHash] = tableEntryPrev{ Prev: candidates.Cur, - Cur: tableEntry{offset: s + e.cur, val: cv}, + Cur: tableEntry{offset: s + e.cur}, } // Check both candidates candidate = candidates.Cur - if cv == candidate.val { - offset := s - (candidate.offset - e.cur) - if offset <= maxMatchOffset { - continue - } - } else { + minOffset := e.cur + s - (maxMatchOffset - 4) + + if candidate.offset > minOffset && cv != load3232(src, candidate.offset-e.cur) { // We only check if value mismatches. // Offset will always be invalid in other cases. candidate = candidates.Prev - if cv == candidate.val { + if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { offset := s - (candidate.offset - e.cur) if offset <= maxMatchOffset { continue diff --git a/vendor/github.com/klauspost/compress/flate/level4.go b/vendor/github.com/klauspost/compress/flate/level4.go index c689ac771b82..e62f0c02b1e7 100644 --- a/vendor/github.com/klauspost/compress/flate/level4.go +++ b/vendor/github.com/klauspost/compress/flate/level4.go @@ -13,7 +13,9 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin ) - + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } // Protect against e.cur wraparound. for e.cur >= bufferReset { if len(e.hist) == 0 { @@ -90,24 +92,24 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { sCandidate := e.table[nextHashS] lCandidate := e.bTable[nextHashL] next := load6432(src, nextS) - entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + entry := tableEntry{offset: s + e.cur} e.table[nextHashS] = entry e.bTable[nextHashL] = entry t = lCandidate.offset - e.cur - if s-t < maxMatchOffset && uint32(cv) == lCandidate.val { + if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.offset-e.cur) { // We got a long match. Use that. break } t = sCandidate.offset - e.cur - if s-t < maxMatchOffset && uint32(cv) == sCandidate.val { + if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) { // Found a 4 match... lCandidate = e.bTable[hash7(next, tableBits)] // If the next long is a candidate, check if we should use that instead... lOff := nextS - (lCandidate.offset - e.cur) - if lOff < maxMatchOffset && lCandidate.val == uint32(next) { + if lOff < maxMatchOffset && load3232(src, lCandidate.offset-e.cur) == uint32(next) { l1, l2 := matchLen(src[s+4:], src[t+4:]), matchLen(src[nextS+4:], src[nextS-lOff+4:]) if l2 > l1 { s = nextS @@ -135,7 +137,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { if nextEmit < s { emitLiteral(dst, src[nextEmit:s]) } - if false { + if debugDeflate { if t >= s { panic("s-t") } @@ -158,8 +160,8 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { // Index first pair after match end. if int(s+8) < len(src) { cv := load6432(src, s) - e.table[hash4x64(cv, tableBits)] = tableEntry{offset: s + e.cur, val: uint32(cv)} - e.bTable[hash7(cv, tableBits)] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.table[hash4x64(cv, tableBits)] = tableEntry{offset: s + e.cur} + e.bTable[hash7(cv, tableBits)] = tableEntry{offset: s + e.cur} } goto emitRemainder } @@ -169,20 +171,20 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { i := nextS if i < s-1 { cv := load6432(src, i) - t := tableEntry{offset: i + e.cur, val: uint32(cv)} - t2 := tableEntry{val: uint32(cv >> 8), offset: t.offset + 1} + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} e.bTable[hash7(cv, tableBits)] = t e.bTable[hash7(cv>>8, tableBits)] = t2 - e.table[hash4u(t2.val, tableBits)] = t2 + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 i += 3 for ; i < s-1; i += 3 { cv := load6432(src, i) - t := tableEntry{offset: i + e.cur, val: uint32(cv)} - t2 := tableEntry{val: uint32(cv >> 8), offset: t.offset + 1} + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} e.bTable[hash7(cv, tableBits)] = t e.bTable[hash7(cv>>8, tableBits)] = t2 - e.table[hash4u(t2.val, tableBits)] = t2 + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 } } } @@ -193,8 +195,8 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { o := e.cur + s - 1 prevHashS := hash4x64(x, tableBits) prevHashL := hash7(x, tableBits) - e.table[prevHashS] = tableEntry{offset: o, val: uint32(x)} - e.bTable[prevHashL] = tableEntry{offset: o, val: uint32(x)} + e.table[prevHashS] = tableEntry{offset: o} + e.bTable[prevHashL] = tableEntry{offset: o} cv = x >> 8 } diff --git a/vendor/github.com/klauspost/compress/flate/level5.go b/vendor/github.com/klauspost/compress/flate/level5.go index 14a2356126aa..d513f1ffd37c 100644 --- a/vendor/github.com/klauspost/compress/flate/level5.go +++ b/vendor/github.com/klauspost/compress/flate/level5.go @@ -13,6 +13,9 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } // Protect against e.cur wraparound. for e.cur >= bufferReset { @@ -97,7 +100,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { sCandidate := e.table[nextHashS] lCandidate := e.bTable[nextHashL] next := load6432(src, nextS) - entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + entry := tableEntry{offset: s + e.cur} e.table[nextHashS] = entry eLong := &e.bTable[nextHashL] eLong.Cur, eLong.Prev = entry, eLong.Cur @@ -107,14 +110,14 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { t = lCandidate.Cur.offset - e.cur if s-t < maxMatchOffset { - if uint32(cv) == lCandidate.Cur.val { + if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) { // Store the next match - e.table[nextHashS] = tableEntry{offset: nextS + e.cur, val: uint32(next)} + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} eLong := &e.bTable[nextHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur, val: uint32(next)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur t2 := lCandidate.Prev.offset - e.cur - if s-t2 < maxMatchOffset && uint32(cv) == lCandidate.Prev.val { + if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { l = e.matchlen(s+4, t+4, src) + 4 ml1 := e.matchlen(s+4, t2+4, src) + 4 if ml1 > l { @@ -126,30 +129,30 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { break } t = lCandidate.Prev.offset - e.cur - if s-t < maxMatchOffset && uint32(cv) == lCandidate.Prev.val { + if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { // Store the next match - e.table[nextHashS] = tableEntry{offset: nextS + e.cur, val: uint32(next)} + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} eLong := &e.bTable[nextHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur, val: uint32(next)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur break } } t = sCandidate.offset - e.cur - if s-t < maxMatchOffset && uint32(cv) == sCandidate.val { + if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) { // Found a 4 match... l = e.matchlen(s+4, t+4, src) + 4 lCandidate = e.bTable[nextHashL] // Store the next match - e.table[nextHashS] = tableEntry{offset: nextS + e.cur, val: uint32(next)} + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} eLong := &e.bTable[nextHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur, val: uint32(next)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur // If the next long is a candidate, use that... t2 := lCandidate.Cur.offset - e.cur if nextS-t2 < maxMatchOffset { - if lCandidate.Cur.val == uint32(next) { + if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) { ml := e.matchlen(nextS+4, t2+4, src) + 4 if ml > l { t = t2 @@ -160,7 +163,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { } // If the previous long is a candidate, use that... t2 = lCandidate.Prev.offset - e.cur - if nextS-t2 < maxMatchOffset && lCandidate.Prev.val == uint32(next) { + if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) { ml := e.matchlen(nextS+4, t2+4, src) + 4 if ml > l { t = t2 @@ -194,7 +197,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { if nextEmit < s { emitLiteral(dst, src[nextEmit:s]) } - if false { + if debugDeflate { if t >= s { panic(fmt.Sprintln("s-t", s, t)) } @@ -223,31 +226,31 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { i := s - l + 1 if i < s-1 { cv := load6432(src, i) - t := tableEntry{offset: i + e.cur, val: uint32(cv)} + t := tableEntry{offset: i + e.cur} e.table[hash4x64(cv, tableBits)] = t eLong := &e.bTable[hash7(cv, tableBits)] eLong.Cur, eLong.Prev = t, eLong.Cur // Do an long at i+1 cv >>= 8 - t = tableEntry{offset: t.offset + 1, val: uint32(cv)} + t = tableEntry{offset: t.offset + 1} eLong = &e.bTable[hash7(cv, tableBits)] eLong.Cur, eLong.Prev = t, eLong.Cur // We only have enough bits for a short entry at i+2 cv >>= 8 - t = tableEntry{offset: t.offset + 1, val: uint32(cv)} + t = tableEntry{offset: t.offset + 1} e.table[hash4x64(cv, tableBits)] = t // Skip one - otherwise we risk hitting 's' i += 4 for ; i < s-1; i += hashEvery { cv := load6432(src, i) - t := tableEntry{offset: i + e.cur, val: uint32(cv)} - t2 := tableEntry{offset: t.offset + 1, val: uint32(cv >> 8)} + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} eLong := &e.bTable[hash7(cv, tableBits)] eLong.Cur, eLong.Prev = t, eLong.Cur - e.table[hash4u(t2.val, tableBits)] = t2 + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 } } } @@ -258,9 +261,9 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { o := e.cur + s - 1 prevHashS := hash4x64(x, tableBits) prevHashL := hash7(x, tableBits) - e.table[prevHashS] = tableEntry{offset: o, val: uint32(x)} + e.table[prevHashS] = tableEntry{offset: o} eLong := &e.bTable[prevHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: o, val: uint32(x)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: o}, eLong.Cur cv = x >> 8 } diff --git a/vendor/github.com/klauspost/compress/flate/level6.go b/vendor/github.com/klauspost/compress/flate/level6.go index cad0c7df7fc3..a52c80ea456c 100644 --- a/vendor/github.com/klauspost/compress/flate/level6.go +++ b/vendor/github.com/klauspost/compress/flate/level6.go @@ -13,6 +13,9 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin ) + if debugDeflate && e.cur < 0 { + panic(fmt.Sprint("e.cur < 0: ", e.cur)) + } // Protect against e.cur wraparound. for e.cur >= bufferReset { @@ -98,7 +101,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { sCandidate := e.table[nextHashS] lCandidate := e.bTable[nextHashL] next := load6432(src, nextS) - entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + entry := tableEntry{offset: s + e.cur} e.table[nextHashS] = entry eLong := &e.bTable[nextHashL] eLong.Cur, eLong.Prev = entry, eLong.Cur @@ -109,17 +112,17 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { t = lCandidate.Cur.offset - e.cur if s-t < maxMatchOffset { - if uint32(cv) == lCandidate.Cur.val { + if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) { // Long candidate matches at least 4 bytes. // Store the next match - e.table[nextHashS] = tableEntry{offset: nextS + e.cur, val: uint32(next)} + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} eLong := &e.bTable[nextHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur, val: uint32(next)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur // Check the previous long candidate as well. t2 := lCandidate.Prev.offset - e.cur - if s-t2 < maxMatchOffset && uint32(cv) == lCandidate.Prev.val { + if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { l = e.matchlen(s+4, t+4, src) + 4 ml1 := e.matchlen(s+4, t2+4, src) + 4 if ml1 > l { @@ -132,17 +135,17 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { } // Current value did not match, but check if previous long value does. t = lCandidate.Prev.offset - e.cur - if s-t < maxMatchOffset && uint32(cv) == lCandidate.Prev.val { + if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) { // Store the next match - e.table[nextHashS] = tableEntry{offset: nextS + e.cur, val: uint32(next)} + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} eLong := &e.bTable[nextHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur, val: uint32(next)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur break } } t = sCandidate.offset - e.cur - if s-t < maxMatchOffset && uint32(cv) == sCandidate.val { + if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) { // Found a 4 match... l = e.matchlen(s+4, t+4, src) + 4 @@ -150,9 +153,9 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { lCandidate = e.bTable[nextHashL] // Store the next match - e.table[nextHashS] = tableEntry{offset: nextS + e.cur, val: uint32(next)} + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} eLong := &e.bTable[nextHashL] - eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur, val: uint32(next)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur // Check repeat at s + repOff const repOff = 1 @@ -171,7 +174,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { // If the next long is a candidate, use that... t2 = lCandidate.Cur.offset - e.cur if nextS-t2 < maxMatchOffset { - if lCandidate.Cur.val == uint32(next) { + if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) { ml := e.matchlen(nextS+4, t2+4, src) + 4 if ml > l { t = t2 @@ -182,7 +185,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { } // If the previous long is a candidate, use that... t2 = lCandidate.Prev.offset - e.cur - if nextS-t2 < maxMatchOffset && lCandidate.Prev.val == uint32(next) { + if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) { ml := e.matchlen(nextS+4, t2+4, src) + 4 if ml > l { t = t2 @@ -241,9 +244,9 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { // Index after match end. for i := nextS + 1; i < int32(len(src))-8; i += 2 { cv := load6432(src, i) - e.table[hash4x64(cv, tableBits)] = tableEntry{offset: i + e.cur, val: uint32(cv)} + e.table[hash4x64(cv, tableBits)] = tableEntry{offset: i + e.cur} eLong := &e.bTable[hash7(cv, tableBits)] - eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur, val: uint32(cv)}, eLong.Cur + eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur}, eLong.Cur } goto emitRemainder } @@ -252,8 +255,8 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { if true { for i := nextS + 1; i < s-1; i += 2 { cv := load6432(src, i) - t := tableEntry{offset: i + e.cur, val: uint32(cv)} - t2 := tableEntry{offset: t.offset + 1, val: uint32(cv >> 8)} + t := tableEntry{offset: i + e.cur} + t2 := tableEntry{offset: t.offset + 1} eLong := &e.bTable[hash7(cv, tableBits)] eLong2 := &e.bTable[hash7(cv>>8, tableBits)] e.table[hash4x64(cv, tableBits)] = t diff --git a/vendor/github.com/klauspost/compress/flate/regmask_amd64.go b/vendor/github.com/klauspost/compress/flate/regmask_amd64.go new file mode 100644 index 000000000000..6ed28061b2b8 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/regmask_amd64.go @@ -0,0 +1,37 @@ +package flate + +const ( + // Masks for shifts with register sizes of the shift value. + // This can be used to work around the x86 design of shifting by mod register size. + // It can be used when a variable shift is always smaller than the register size. + + // reg8SizeMaskX - shift value is 8 bits, shifted is X + reg8SizeMask8 = 7 + reg8SizeMask16 = 15 + reg8SizeMask32 = 31 + reg8SizeMask64 = 63 + + // reg16SizeMaskX - shift value is 16 bits, shifted is X + reg16SizeMask8 = reg8SizeMask8 + reg16SizeMask16 = reg8SizeMask16 + reg16SizeMask32 = reg8SizeMask32 + reg16SizeMask64 = reg8SizeMask64 + + // reg32SizeMaskX - shift value is 32 bits, shifted is X + reg32SizeMask8 = reg8SizeMask8 + reg32SizeMask16 = reg8SizeMask16 + reg32SizeMask32 = reg8SizeMask32 + reg32SizeMask64 = reg8SizeMask64 + + // reg64SizeMaskX - shift value is 64 bits, shifted is X + reg64SizeMask8 = reg8SizeMask8 + reg64SizeMask16 = reg8SizeMask16 + reg64SizeMask32 = reg8SizeMask32 + reg64SizeMask64 = reg8SizeMask64 + + // regSizeMaskUintX - shift value is uint, shifted is X + regSizeMaskUint8 = reg8SizeMask8 + regSizeMaskUint16 = reg8SizeMask16 + regSizeMaskUint32 = reg8SizeMask32 + regSizeMaskUint64 = reg8SizeMask64 +) diff --git a/vendor/github.com/klauspost/compress/flate/regmask_other.go b/vendor/github.com/klauspost/compress/flate/regmask_other.go new file mode 100644 index 000000000000..f477a5d6e5a6 --- /dev/null +++ b/vendor/github.com/klauspost/compress/flate/regmask_other.go @@ -0,0 +1,39 @@ +//+build !amd64 + +package flate + +const ( + // Masks for shifts with register sizes of the shift value. + // This can be used to work around the x86 design of shifting by mod register size. + // It can be used when a variable shift is always smaller than the register size. + + // reg8SizeMaskX - shift value is 8 bits, shifted is X + reg8SizeMask8 = 0xff + reg8SizeMask16 = 0xff + reg8SizeMask32 = 0xff + reg8SizeMask64 = 0xff + + // reg16SizeMaskX - shift value is 16 bits, shifted is X + reg16SizeMask8 = 0xffff + reg16SizeMask16 = 0xffff + reg16SizeMask32 = 0xffff + reg16SizeMask64 = 0xffff + + // reg32SizeMaskX - shift value is 32 bits, shifted is X + reg32SizeMask8 = 0xffffffff + reg32SizeMask16 = 0xffffffff + reg32SizeMask32 = 0xffffffff + reg32SizeMask64 = 0xffffffff + + // reg64SizeMaskX - shift value is 64 bits, shifted is X + reg64SizeMask8 = 0xffffffffffffffff + reg64SizeMask16 = 0xffffffffffffffff + reg64SizeMask32 = 0xffffffffffffffff + reg64SizeMask64 = 0xffffffffffffffff + + // regSizeMaskUintX - shift value is uint, shifted is X + regSizeMaskUint8 = ^uint(0) + regSizeMaskUint16 = ^uint(0) + regSizeMaskUint32 = ^uint(0) + regSizeMaskUint64 = ^uint(0) +) diff --git a/vendor/github.com/klauspost/compress/flate/stateless.go b/vendor/github.com/klauspost/compress/flate/stateless.go index a4705119757d..53e899124639 100644 --- a/vendor/github.com/klauspost/compress/flate/stateless.go +++ b/vendor/github.com/klauspost/compress/flate/stateless.go @@ -8,6 +8,8 @@ import ( const ( maxStatelessBlock = math.MaxInt16 + // dictionary will be taken from maxStatelessBlock, so limit it. + maxStatelessDict = 8 << 10 slTableBits = 13 slTableSize = 1 << slTableBits @@ -25,11 +27,11 @@ func (s *statelessWriter) Close() error { } s.closed = true // Emit EOF block - return StatelessDeflate(s.dst, nil, true) + return StatelessDeflate(s.dst, nil, true, nil) } func (s *statelessWriter) Write(p []byte) (n int, err error) { - err = StatelessDeflate(s.dst, p, false) + err = StatelessDeflate(s.dst, p, false, nil) if err != nil { return 0, err } @@ -59,7 +61,10 @@ var bitWriterPool = sync.Pool{ // StatelessDeflate allows to compress directly to a Writer without retaining state. // When returning everything will be flushed. -func StatelessDeflate(out io.Writer, in []byte, eof bool) error { +// Up to 8KB of an optional dictionary can be given which is presumed to presumed to precede the block. +// Longer dictionaries will be truncated and will still produce valid output. +// Sending nil dictionary is perfectly fine. +func StatelessDeflate(out io.Writer, in []byte, eof bool, dict []byte) error { var dst tokens bw := bitWriterPool.Get().(*huffmanBitWriter) bw.reset(out) @@ -76,35 +81,53 @@ func StatelessDeflate(out io.Writer, in []byte, eof bool) error { return bw.err } + // Truncate dict + if len(dict) > maxStatelessDict { + dict = dict[len(dict)-maxStatelessDict:] + } + for len(in) > 0 { todo := in - if len(todo) > maxStatelessBlock { - todo = todo[:maxStatelessBlock] + if len(todo) > maxStatelessBlock-len(dict) { + todo = todo[:maxStatelessBlock-len(dict)] } in = in[len(todo):] + uncompressed := todo + if len(dict) > 0 { + // combine dict and source + bufLen := len(todo) + len(dict) + combined := make([]byte, bufLen) + copy(combined, dict) + copy(combined[len(dict):], todo) + todo = combined + } // Compress - statelessEnc(&dst, todo) + statelessEnc(&dst, todo, int16(len(dict))) isEof := eof && len(in) == 0 if dst.n == 0 { - bw.writeStoredHeader(len(todo), isEof) + bw.writeStoredHeader(len(uncompressed), isEof) if bw.err != nil { return bw.err } - bw.writeBytes(todo) - } else if int(dst.n) > len(todo)-len(todo)>>4 { + bw.writeBytes(uncompressed) + } else if int(dst.n) > len(uncompressed)-len(uncompressed)>>4 { // If we removed less than 1/16th, huffman compress the block. - bw.writeBlockHuff(isEof, todo, false) + bw.writeBlockHuff(isEof, uncompressed, len(in) == 0) } else { - bw.writeBlockDynamic(&dst, isEof, todo, false) + bw.writeBlockDynamic(&dst, isEof, uncompressed, len(in) == 0) + } + if len(in) > 0 { + // Retain a dict if we have more + dict = todo[len(todo)-maxStatelessDict:] + dst.Reset() } if bw.err != nil { return bw.err } - dst.Reset() } if !eof { - // Align. + // Align, only a stored block can do that. bw.writeStoredHeader(0, false) } bw.flush() @@ -130,7 +153,7 @@ func load6416(b []byte, i int16) uint64 { uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56 } -func statelessEnc(dst *tokens, src []byte) { +func statelessEnc(dst *tokens, src []byte, startAt int16) { const ( inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin @@ -144,15 +167,23 @@ func statelessEnc(dst *tokens, src []byte) { // This check isn't in the Snappy implementation, but there, the caller // instead of the callee handles this case. - if len(src) < minNonLiteralBlockSize { + if len(src)-int(startAt) < minNonLiteralBlockSize { // We do not fill the token table. // This will be picked up by caller. - dst.n = uint16(len(src)) + dst.n = 0 return } + // Index until startAt + if startAt > 0 { + cv := load3232(src, 0) + for i := int16(0); i < startAt; i++ { + table[hashSL(cv)] = tableEntry{offset: i} + cv = (cv >> 8) | (uint32(src[i+4]) << 24) + } + } - s := int16(1) - nextEmit := int16(0) + s := startAt + 1 + nextEmit := startAt // 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. diff --git a/vendor/github.com/klauspost/compress/flate/token.go b/vendor/github.com/klauspost/compress/flate/token.go index b3df0d8941e1..f9abf606d67c 100644 --- a/vendor/github.com/klauspost/compress/flate/token.go +++ b/vendor/github.com/klauspost/compress/flate/token.go @@ -184,9 +184,7 @@ func (t *tokens) indexTokens(in []token) { t.Reset() for _, tok := range in { if tok < matchType { - t.tokens[t.n] = tok - t.litHist[tok]++ - t.n++ + t.AddLiteral(tok.literal()) continue } t.AddMatch(uint32(tok.length()), tok.offset()) @@ -211,50 +209,60 @@ func (t *tokens) AddLiteral(lit byte) { t.nLits++ } +// from https://stackoverflow.com/a/28730362 +func mFastLog2(val float32) float32 { + ux := int32(math.Float32bits(val)) + log2 := (float32)(((ux >> 23) & 255) - 128) + ux &= -0x7f800001 + ux += 127 << 23 + uval := math.Float32frombits(uint32(ux)) + log2 += ((-0.34484843)*uval+2.02466578)*uval - 0.67487759 + return log2 +} + // EstimatedBits will return an minimum size estimated by an *optimal* // compression of the block. // The size of the block func (t *tokens) EstimatedBits() int { - shannon := float64(0) + shannon := float32(0) bits := int(0) nMatches := 0 if t.nLits > 0 { - invTotal := 1.0 / float64(t.nLits) + invTotal := 1.0 / float32(t.nLits) for _, v := range t.litHist[:] { if v > 0 { - n := float64(v) - shannon += math.Ceil(-math.Log2(n*invTotal) * n) + n := float32(v) + shannon += -mFastLog2(n*invTotal) * n } } // Just add 15 for EOB shannon += 15 - for _, v := range t.extraHist[1 : literalCount-256] { + for i, v := range t.extraHist[1 : literalCount-256] { if v > 0 { - n := float64(v) - shannon += math.Ceil(-math.Log2(n*invTotal) * n) - bits += int(lengthExtraBits[v&31]) * int(v) + n := float32(v) + shannon += -mFastLog2(n*invTotal) * n + bits += int(lengthExtraBits[i&31]) * int(v) nMatches += int(v) } } } if nMatches > 0 { - invTotal := 1.0 / float64(nMatches) - for _, v := range t.offHist[:offsetCodeCount] { + invTotal := 1.0 / float32(nMatches) + for i, v := range t.offHist[:offsetCodeCount] { if v > 0 { - n := float64(v) - shannon += math.Ceil(-math.Log2(n*invTotal) * n) - bits += int(offsetExtraBits[v&31]) * int(n) + n := float32(v) + shannon += -mFastLog2(n*invTotal) * n + bits += int(offsetExtraBits[i&31]) * int(v) } } } - return int(shannon) + bits } // AddMatch adds a match to the tokens. // This function is very sensitive to inlining and right on the border. func (t *tokens) AddMatch(xlength uint32, xoffset uint32) { - if debugDecode { + if debugDeflate { if xlength >= maxMatchLength+baseMatchLength { panic(fmt.Errorf("invalid length: %v", xlength)) } @@ -273,7 +281,7 @@ func (t *tokens) AddMatch(xlength uint32, xoffset uint32) { // AddMatchLong adds a match to the tokens, potentially longer than max match length. // Length should NOT have the base subtracted, only offset should. func (t *tokens) AddMatchLong(xlength int32, xoffset uint32) { - if debugDecode { + if debugDeflate { if xoffset >= maxMatchOffset+baseMatchOffset { panic(fmt.Errorf("invalid offset: %v", xoffset)) } diff --git a/vendor/github.com/klauspost/compress/fse/README.md b/vendor/github.com/klauspost/compress/fse/README.md new file mode 100644 index 000000000000..27d8ed56fcc4 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/README.md @@ -0,0 +1,79 @@ +# Finite State Entropy + +This package provides Finite State Entropy encoding and decoding. + +Finite State Entropy (also referenced as [tANS](https://en.wikipedia.org/wiki/Asymmetric_numeral_systems#tANS)) +encoding provides a fast near-optimal symbol encoding/decoding +for byte blocks as implemented in [zstandard](https://github.com/facebook/zstd). + +This can be used for compressing input with a lot of similar input values to the smallest number of bytes. +This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, +but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. + +* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/fse) + +## News + + * Feb 2018: First implementation released. Consider this beta software for now. + +# Usage + +This package provides a low level interface that allows to compress single independent blocks. + +Each block is separate, and there is no built in integrity checks. +This means that the caller should keep track of block sizes and also do checksums if needed. + +Compressing a block is done via the [`Compress`](https://godoc.org/github.com/klauspost/compress/fse#Compress) function. +You must provide input and will receive the output and maybe an error. + +These error values can be returned: + +| Error | Description | +|---------------------|-----------------------------------------------------------------------------| +| `` | Everything ok, output is returned | +| `ErrIncompressible` | Returned when input is judged to be too hard to compress | +| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | +| `(error)` | An internal error occurred. | + +As can be seen above there are errors that will be returned even under normal operation so it is important to handle these. + +To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/fse#Scratch) object +that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same +object can be used for both. + +Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this +you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. + +Decompressing is done by calling the [`Decompress`](https://godoc.org/github.com/klauspost/compress/fse#Decompress) function. +You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back +your input was likely corrupted. + +It is important to note that a successful decoding does *not* mean your output matches your original input. +There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. + +For more detailed usage, see examples in the [godoc documentation](https://godoc.org/github.com/klauspost/compress/fse#pkg-examples). + +# Performance + +A lot of factors are affecting speed. Block sizes and compressibility of the material are primary factors. +All compression functions are currently only running on the calling goroutine so only one core will be used per block. + +The compressor is significantly faster if symbols are kept as small as possible. The highest byte value of the input +is used to reduce some of the processing, so if all your input is above byte value 64 for instance, it may be +beneficial to transpose all your input values down by 64. + +With moderate block sizes around 64k speed are typically 200MB/s per core for compression and +around 300MB/s decompression speed. + +The same hardware typically does Huffman (deflate) encoding at 125MB/s and decompression at 100MB/s. + +# Plans + +At one point, more internals will be exposed to facilitate more "expert" usage of the components. + +A streaming interface is also likely to be implemented. Likely compatible with [FSE stream format](https://github.com/Cyan4973/FiniteStateEntropy/blob/dev/programs/fileio.c#L261). + +# Contributing + +Contributions are always welcome. Be aware that adding public functions will require good justification and breaking +changes will likely not be accepted. If in doubt open an issue before writing the PR. \ No newline at end of file diff --git a/vendor/github.com/klauspost/compress/fse/bitreader.go b/vendor/github.com/klauspost/compress/fse/bitreader.go new file mode 100644 index 000000000000..f65eb3909cf4 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/bitreader.go @@ -0,0 +1,122 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +import ( + "encoding/binary" + "errors" + "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(highBits(uint32(v))) + return nil +} + +// getBits will return n bits. n can be 0. +func (b *bitReader) getBits(n uint8) uint16 { + if n == 0 || b.bitsRead >= 64 { + return 0 + } + return b.getBitsFast(n) +} + +// getBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReader) getBitsFast(n uint8) uint16 { + const regMask = 64 - 1 + v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask)) + b.bitsRead += n + return v +} + +// 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:] + 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 +} + +// 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-- + } +} + +// 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 +} + +// finished returns true if all bits have been read from the bit stream. +func (b *bitReader) finished() bool { + return b.bitsRead >= 64 && b.off == 0 +} + +// 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 +} diff --git a/vendor/github.com/klauspost/compress/fse/bitwriter.go b/vendor/github.com/klauspost/compress/fse/bitwriter.go new file mode 100644 index 000000000000..43e463611b15 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/bitwriter.go @@ -0,0 +1,168 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +import "fmt" + +// bitWriter will write bits. +// First bit will be LSB of the first byte of output. +type bitWriter struct { + bitContainer uint64 + nBits uint8 + out []byte +} + +// bitMask16 is bitmasks. Has extra to avoid bounds check. +var bitMask16 = [32]uint16{ + 0, 1, 3, 7, 0xF, 0x1F, + 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, + 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF, + 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) { + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// 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: + 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 { + return + } + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24)) + b.nBits -= 32 + b.bitContainer >>= 32 +} + +// flushAlign will flush remaining full bytes and align to next byte boundary. +func (b *bitWriter) flushAlign() { + nbBytes := (b.nBits + 7) >> 3 + for i := uint8(0); i < nbBytes; i++ { + b.out = append(b.out, byte(b.bitContainer>>(i*8))) + } + b.nBits = 0 + b.bitContainer = 0 +} + +// close will write the alignment bit and write the final byte(s) +// to the output. +func (b *bitWriter) close() error { + // End mark + b.addBits16Clean(1, 1) + // flush until next byte. + 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/fse/bytereader.go b/vendor/github.com/klauspost/compress/fse/bytereader.go new file mode 100644 index 000000000000..abade2d60527 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/bytereader.go @@ -0,0 +1,47 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +// byteReader provides a byte reader that reads +// little endian values from a byte stream. +// The input stream is manually advanced. +// The reader performs no bounds checks. +type byteReader struct { + b []byte + 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) +} + +// Uint32 returns a little endian uint32 starting at current offset. +func (b byteReader) Uint32() uint32 { + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := uint32(b2[3]) + v2 := uint32(b2[2]) + v1 := uint32(b2[1]) + v0 := uint32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// 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/fse/compress.go b/vendor/github.com/klauspost/compress/fse/compress.go new file mode 100644 index 000000000000..b69237c9b8f5 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/compress.go @@ -0,0 +1,684 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package fse + +import ( + "errors" + "fmt" +) + +// Compress the input bytes. Input must be < 2GB. +// Provide a Scratch buffer to avoid memory allocations. +// Note that the output is also kept in the scratch buffer. +// If input is too hard to compress, ErrIncompressible is returned. +// If input is a single byte value repeated ErrUseRLE is returned. +func Compress(in []byte, s *Scratch) ([]byte, error) { + if len(in) <= 1 { + return nil, ErrIncompressible + } + if len(in) > (2<<30)-1 { + return nil, errors.New("input too big, must be < 2GB") + } + s, err := s.prepare(in) + if err != nil { + return nil, err + } + + // Create histogram, if none was provided. + maxCount := s.maxCount + if maxCount == 0 { + maxCount = s.countSimple(in) + } + // Reset for next run. + s.clearCount = true + s.maxCount = 0 + if maxCount == len(in) { + // One symbol, use RLE + return nil, ErrUseRLE + } + if maxCount == 1 || maxCount < (len(in)>>7) { + // Each symbol present maximum once or too well distributed. + return nil, ErrIncompressible + } + s.optimalTableLog() + err = s.normalizeCount() + if err != nil { + return nil, err + } + err = s.writeCount() + if err != nil { + return nil, err + } + + if false { + err = s.validateNorm() + if err != nil { + return nil, err + } + } + + err = s.buildCTable() + if err != nil { + return nil, err + } + err = s.compress(in) + if err != nil { + return nil, err + } + s.Out = s.bw.out + // Check if we compressed. + if len(s.Out) >= len(in) { + return nil, ErrIncompressible + } + return s.Out, nil +} + +// cState contains the compression state of a stream. +type cState struct { + bw *bitWriter + stateTable []uint16 + state uint16 +} + +// init will initialize the compression state to the first symbol of the stream. +func (c *cState) init(bw *bitWriter, ct *cTable, tableLog uint8, first symbolTransform) { + c.bw = bw + c.stateTable = ct.stateTable + + nbBitsOut := (first.deltaNbBits + (1 << 15)) >> 16 + im := int32((nbBitsOut << 16) - first.deltaNbBits) + lu := (im >> nbBitsOut) + first.deltaFindState + c.state = c.stateTable[lu] + return +} + +// 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)) + symbolTT.deltaFindState + c.bw.addBits16NC(c.state, uint8(nbBitsOut)) + c.state = c.stateTable[dstState] +} + +// encode the output symbol provided and write it to the bitstream. +func (c *cState) encodeZero(symbolTT symbolTransform) { + nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16 + dstState := int32(c.state>>(nbBitsOut&15)) + symbolTT.deltaFindState + c.bw.addBits16ZeroNC(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() + c.bw.addBits16NC(c.state, tableLog) + c.bw.flush() +} + +// compress is the main compression loop that will encode the input from the last byte to the first. +func (s *Scratch) compress(src []byte) error { + if len(src) <= 2 { + return errors.New("compress: src too small") + } + tt := s.ct.symbolTT[:256] + s.bw.reset(s.Out) + + // Our two states each encodes every second byte. + // Last byte encoded (first byte decoded) will always be encoded by c1. + var c1, c2 cState + + // Encode so remaining size is divisible by 4. + ip := len(src) + if ip&1 == 1 { + c1.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-1]]) + c2.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-2]]) + c1.encodeZero(tt[src[ip-3]]) + ip -= 3 + } else { + c2.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-1]]) + c1.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-2]]) + ip -= 2 + } + if ip&2 != 0 { + c2.encodeZero(tt[src[ip-1]]) + c1.encodeZero(tt[src[ip-2]]) + ip -= 2 + } + + // Main compression loop. + switch { + case !s.zeroBits && s.actualTableLog <= 8: + // We can encode 4 symbols without requiring a flush. + // We do not need to check if any output is 0 bits. + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encode(tt[v0]) + c1.encode(tt[v1]) + c2.encode(tt[v2]) + c1.encode(tt[v3]) + ip -= 4 + } + case !s.zeroBits: + // We do not need to check if any output is 0 bits. + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encode(tt[v0]) + c1.encode(tt[v1]) + s.bw.flush32() + c2.encode(tt[v2]) + c1.encode(tt[v3]) + ip -= 4 + } + case s.actualTableLog <= 8: + // We can encode 4 symbols without requiring a flush + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encodeZero(tt[v0]) + c1.encodeZero(tt[v1]) + c2.encodeZero(tt[v2]) + c1.encodeZero(tt[v3]) + ip -= 4 + } + default: + for ip >= 4 { + s.bw.flush32() + v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1] + c2.encodeZero(tt[v0]) + c1.encodeZero(tt[v1]) + s.bw.flush32() + c2.encodeZero(tt[v2]) + c1.encodeZero(tt[v3]) + ip -= 4 + } + } + + // Flush final state. + // Used to initialize state when decoding. + c2.flush(s.actualTableLog) + c1.flush(s.actualTableLog) + + return s.bw.close() +} + +// writeCount will write the normalized histogram count to header. +// This is read back by readNCount. +func (s *Scratch) writeCount() error { + var ( + tableLog = s.actualTableLog + tableSize = 1 << tableLog + previous0 bool + charnum uint16 + + maxHeaderSize = ((int(s.symbolLen) * int(tableLog)) >> 3) + 3 + + // Write Table Size + bitStream = uint32(tableLog - minTablelog) + bitCount = uint(4) + remaining = int16(tableSize + 1) /* +1 for extra accuracy */ + threshold = int16(tableSize) + nbBits = uint(tableLog + 1) + ) + if cap(s.Out) < maxHeaderSize { + s.Out = make([]byte, 0, s.br.remain()+maxHeaderSize) + } + outP := uint(0) + out := s.Out[:maxHeaderSize] + + // stops at 1 + for remaining > 1 { + if previous0 { + start := charnum + for s.norm[charnum] == 0 { + charnum++ + } + for charnum >= start+24 { + start += 24 + bitStream += uint32(0xFFFF) << bitCount + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + } + for charnum >= start+3 { + start += 3 + bitStream += 3 << bitCount + bitCount += 2 + } + bitStream += uint32(charnum-start) << bitCount + bitCount += 2 + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + count := s.norm[charnum] + charnum++ + max := (2*threshold - 1) - remaining + if count < 0 { + remaining += count + } else { + remaining -= count + } + count++ // +1 for extra accuracy + if count >= threshold { + count += max // [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ + } + bitStream += uint32(count) << bitCount + bitCount += nbBits + if count < max { + bitCount-- + } + + previous0 = count == 1 + if remaining < 1 { + return errors.New("internal error: remaining<1") + } + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += (bitCount + 7) / 8 + + if uint16(charnum) > s.symbolLen { + return errors.New("internal error: charnum > s.symbolLen") + } + s.Out = out[:outP] + return nil +} + +// symbolTransform contains the state transform for a symbol. +type symbolTransform struct { + deltaFindState int32 + deltaNbBits uint32 +} + +// String prints values as a human readable string. +func (s symbolTransform) String() string { + return fmt.Sprintf("dnbits: %08x, fs:%d", s.deltaNbBits, s.deltaFindState) +} + +// cTable contains tables used for compression. +type cTable struct { + tableSymbol []byte + stateTable []uint16 + symbolTT []symbolTransform +} + +// allocCtable will allocate tables needed for compression. +// If existing tables a re big enough, they are simply re-used. +func (s *Scratch) allocCtable() { + tableSize := 1 << s.actualTableLog + // get tableSymbol that is big enough. + if cap(s.ct.tableSymbol) < int(tableSize) { + s.ct.tableSymbol = make([]byte, tableSize) + } + s.ct.tableSymbol = s.ct.tableSymbol[:tableSize] + + ctSize := tableSize + if cap(s.ct.stateTable) < ctSize { + s.ct.stateTable = make([]uint16, ctSize) + } + s.ct.stateTable = s.ct.stateTable[:ctSize] + + if cap(s.ct.symbolTT) < 256 { + s.ct.symbolTT = make([]symbolTransform, 256) + } + s.ct.symbolTT = s.ct.symbolTT[:256] +} + +// buildCTable will populate the compression table so it is ready to be used. +func (s *Scratch) buildCTable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + var cumul [maxSymbolValue + 2]int16 + + s.allocCtable() + tableSymbol := s.ct.tableSymbol[:tableSize] + // symbol start positions + { + cumul[0] = 0 + for ui, v := range s.norm[:s.symbolLen-1] { + u := byte(ui) // one less than reference + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = u + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + } + // Encode last symbol separately to avoid overflowing u + u := int(s.symbolLen - 1) + v := s.norm[s.symbolLen-1] + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = byte(u) + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + if uint32(cumul[s.symbolLen]) != tableSize { + return fmt.Errorf("internal error: expected cumul[s.symbolLen] (%d) == tableSize (%d)", cumul[s.symbolLen], tableSize) + } + cumul[s.symbolLen] = int16(tableSize) + 1 + } + // Spread symbols + s.zeroBits = false + { + step := tableStep(tableSize) + tableMask := tableSize - 1 + var position uint32 + // if any symbol > largeLimit, we may have 0 bits output. + largeLimit := int16(1 << (s.actualTableLog - 1)) + for ui, v := range s.norm[:s.symbolLen] { + symbol := byte(ui) + if v > largeLimit { + s.zeroBits = true + } + for nbOccurrences := int16(0); nbOccurrences < v; nbOccurrences++ { + tableSymbol[position] = symbol + position = (position + step) & tableMask + for position > highThreshold { + position = (position + step) & tableMask + } /* Low proba area */ + } + } + + // Check if we have gone through all positions + if position != 0 { + return errors.New("position!=0") + } + } + + // Build table + table := s.ct.stateTable + { + tsi := int(tableSize) + for u, v := range tableSymbol { + // TableU16 : sorted by symbol order; gives next state value + table[cumul[v]] = uint16(tsi + u) + cumul[v]++ + } + } + + // Build Symbol Transformation Table + { + total := int16(0) + symbolTT := s.ct.symbolTT[:s.symbolLen] + tableLog := s.actualTableLog + tl := (uint32(tableLog) << 16) - (1 << tableLog) + for i, v := range s.norm[:s.symbolLen] { + switch v { + case 0: + case -1, 1: + symbolTT[i].deltaNbBits = tl + symbolTT[i].deltaFindState = int32(total - 1) + total++ + default: + maxBitsOut := uint32(tableLog) - highBits(uint32(v-1)) + minStatePlus := uint32(v) << maxBitsOut + symbolTT[i].deltaNbBits = (maxBitsOut << 16) - minStatePlus + symbolTT[i].deltaFindState = int32(total - v) + total += v + } + } + if total != int16(tableSize) { + return fmt.Errorf("total mismatch %d (got) != %d (want)", total, tableSize) + } + } + return nil +} + +// countSimple will create a simple histogram in s.count. +// Returns the biggest count. +// Does not update s.clearCount. +func (s *Scratch) countSimple(in []byte) (max int) { + for _, v := range in { + s.count[v]++ + } + m := uint32(0) + for i, v := range s.count[:] { + if v > m { + m = v + } + if v > 0 { + s.symbolLen = uint16(i) + 1 + } + } + return int(m) +} + +// minTableLog provides the minimum logSize to safely represent a distribution. +func (s *Scratch) minTableLog() uint8 { + minBitsSrc := highBits(uint32(s.br.remain()-1)) + 1 + minBitsSymbols := highBits(uint32(s.symbolLen-1)) + 2 + if minBitsSrc < minBitsSymbols { + return uint8(minBitsSrc) + } + return uint8(minBitsSymbols) +} + +// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog +func (s *Scratch) optimalTableLog() { + tableLog := s.TableLog + minBits := s.minTableLog() + maxBitsSrc := uint8(highBits(uint32(s.br.remain()-1))) - 2 + if maxBitsSrc < tableLog { + // Accuracy can be reduced + tableLog = maxBitsSrc + } + if minBits > tableLog { + tableLog = minBits + } + // Need a minimum to safely represent all symbol values + if tableLog < minTablelog { + tableLog = minTablelog + } + if tableLog > maxTableLog { + tableLog = maxTableLog + } + s.actualTableLog = tableLog +} + +var rtbTable = [...]uint32{0, 473195, 504333, 520860, 550000, 700000, 750000, 830000} + +// normalizeCount will normalize the count of the symbols so +// the total is equal to the table size. +func (s *Scratch) normalizeCount() error { + var ( + tableLog = s.actualTableLog + scale = 62 - uint64(tableLog) + step = (1 << 62) / uint64(s.br.remain()) + vStep = uint64(1) << (scale - 20) + stillToDistribute = int16(1 << tableLog) + largest int + largestP int16 + lowThreshold = (uint32)(s.br.remain() >> tableLog) + ) + + for i, cnt := range s.count[:s.symbolLen] { + // already handled + // if (count[s] == s.length) return 0; /* rle special case */ + + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + stillToDistribute-- + } else { + proba := (int16)((uint64(cnt) * step) >> scale) + if proba < 8 { + restToBeat := vStep * uint64(rtbTable[proba]) + v := uint64(cnt)*step - (uint64(proba) << scale) + if v > restToBeat { + proba++ + } + } + if proba > largestP { + largestP = proba + largest = i + } + s.norm[i] = proba + stillToDistribute -= proba + } + } + + if -stillToDistribute >= (s.norm[largest] >> 1) { + // corner case, need another normalization method + return s.normalizeCount2() + } + s.norm[largest] += stillToDistribute + return nil +} + +// Secondary normalization method. +// To be used when primary method fails. +func (s *Scratch) normalizeCount2() error { + const notYetAssigned = -2 + var ( + distributed uint32 + total = uint32(s.br.remain()) + tableLog = s.actualTableLog + lowThreshold = uint32(total >> tableLog) + lowOne = uint32((total * 3) >> (tableLog + 1)) + ) + for i, cnt := range s.count[:s.symbolLen] { + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + distributed++ + total -= cnt + continue + } + if cnt <= lowOne { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + s.norm[i] = notYetAssigned + } + toDistribute := (1 << tableLog) - distributed + + if (total / toDistribute) > lowOne { + // risk of rounding to zero + lowOne = uint32((total * 3) / (toDistribute * 2)) + for i, cnt := range s.count[:s.symbolLen] { + if (s.norm[i] == notYetAssigned) && (cnt <= lowOne) { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + } + toDistribute = (1 << tableLog) - distributed + } + if distributed == uint32(s.symbolLen)+1 { + // all values are pretty poor; + // probably incompressible data (should have already been detected); + // find max, then give all remaining points to max + var maxV int + var maxC uint32 + for i, cnt := range s.count[:s.symbolLen] { + if cnt > maxC { + maxV = i + maxC = cnt + } + } + s.norm[maxV] += int16(toDistribute) + return nil + } + + if total == 0 { + // all of the symbols were low enough for the lowOne or lowThreshold + for i := uint32(0); toDistribute > 0; i = (i + 1) % (uint32(s.symbolLen)) { + if s.norm[i] > 0 { + toDistribute-- + s.norm[i]++ + } + } + return nil + } + + var ( + vStepLog = 62 - uint64(tableLog) + mid = uint64((1 << (vStepLog - 1)) - 1) + rStep = (((1 << vStepLog) * uint64(toDistribute)) + mid) / uint64(total) // scale on remaining + tmpTotal = mid + ) + for i, cnt := range s.count[:s.symbolLen] { + if s.norm[i] == notYetAssigned { + var ( + end = tmpTotal + uint64(cnt)*rStep + sStart = uint32(tmpTotal >> vStepLog) + sEnd = uint32(end >> vStepLog) + weight = sEnd - sStart + ) + if weight < 1 { + return errors.New("weight < 1") + } + s.norm[i] = int16(weight) + tmpTotal = end + } + } + return nil +} + +// validateNorm validates the normalized histogram table. +func (s *Scratch) validateNorm() (err error) { + var total int + for _, v := range s.norm[:s.symbolLen] { + if v >= 0 { + total += int(v) + } else { + total -= int(v) + } + } + defer func() { + if err == nil { + return + } + fmt.Printf("selected TableLog: %d, Symbol length: %d\n", s.actualTableLog, s.symbolLen) + for i, v := range s.norm[:s.symbolLen] { + fmt.Printf("%3d: %5d -> %4d \n", i, s.count[i], v) + } + }() + if total != (1 << s.actualTableLog) { + return fmt.Errorf("warning: Total == %d != %d", total, 1< tablelogAbsoluteMax { + return errors.New("tableLog too large") + } + bitStream >>= 4 + bitCount := uint(4) + + s.actualTableLog = uint8(nbBits) + remaining := int32((1 << nbBits) + 1) + threshold := int32(1 << nbBits) + gotTotal := int32(0) + nbBits++ + + for remaining > 1 { + if previous0 { + n0 := charnum + for (bitStream & 0xFFFF) == 0xFFFF { + n0 += 24 + if b.off < iend-5 { + b.advance(2) + bitStream = b.Uint32() >> bitCount + } else { + bitStream >>= 16 + bitCount += 16 + } + } + for (bitStream & 3) == 3 { + n0 += 3 + bitStream >>= 2 + bitCount += 2 + } + n0 += uint16(bitStream & 3) + bitCount += 2 + if n0 > maxSymbolValue { + return errors.New("maxSymbolValue too small") + } + for charnum < n0 { + s.norm[charnum&0xff] = 0 + charnum++ + } + + if b.off <= iend-7 || b.off+int(bitCount>>3) <= iend-4 { + b.advance(bitCount >> 3) + bitCount &= 7 + bitStream = b.Uint32() >> bitCount + } else { + bitStream >>= 2 + } + } + + max := (2*(threshold) - 1) - (remaining) + var count int32 + + if (int32(bitStream) & (threshold - 1)) < max { + count = int32(bitStream) & (threshold - 1) + bitCount += nbBits - 1 + } else { + count = int32(bitStream) & (2*threshold - 1) + if count >= threshold { + count -= max + } + bitCount += nbBits + } + + count-- // extra accuracy + if count < 0 { + // -1 means +1 + remaining += count + gotTotal -= count + } else { + remaining -= count + gotTotal += count + } + s.norm[charnum&0xff] = int16(count) + charnum++ + previous0 = count == 0 + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + if b.off <= iend-7 || b.off+int(bitCount>>3) <= iend-4 { + b.advance(bitCount >> 3) + bitCount &= 7 + } else { + bitCount -= (uint)(8 * (len(b.b) - 4 - b.off)) + b.off = len(b.b) - 4 + } + bitStream = b.Uint32() >> (bitCount & 31) + } + s.symbolLen = charnum + + if s.symbolLen <= 1 { + return fmt.Errorf("symbolLen (%d) too small", s.symbolLen) + } + if s.symbolLen > maxSymbolValue+1 { + return fmt.Errorf("symbolLen (%d) too big", s.symbolLen) + } + if remaining != 1 { + return fmt.Errorf("corruption detected (remaining %d != 1)", remaining) + } + if bitCount > 32 { + return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount) + } + if gotTotal != 1<> 3) + return nil +} + +// 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. +type decSymbol struct { + newState uint16 + symbol uint8 + nbBits uint8 +} + +// allocDtable will allocate decoding tables if they are not big enough. +func (s *Scratch) allocDtable() { + tableSize := 1 << s.actualTableLog + if cap(s.decTable) < int(tableSize) { + s.decTable = make([]decSymbol, tableSize) + } + s.decTable = s.decTable[:tableSize] + + if cap(s.ct.tableSymbol) < 256 { + s.ct.tableSymbol = make([]byte, 256) + } + s.ct.tableSymbol = s.ct.tableSymbol[:256] + + if cap(s.ct.stateTable) < 256 { + s.ct.stateTable = make([]uint16, 256) + } + s.ct.stateTable = s.ct.stateTable[:256] +} + +// buildDtable will build the decoding table. +func (s *Scratch) buildDtable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + s.allocDtable() + symbolNext := s.ct.stateTable[:256] + + // Init, lay down lowprob symbols + s.zeroBits = false + { + largeLimit := int16(1 << (s.actualTableLog - 1)) + for i, v := range s.norm[:s.symbolLen] { + if v == -1 { + s.decTable[highThreshold].symbol = uint8(i) + highThreshold-- + symbolNext[i] = 1 + } else { + if v >= largeLimit { + s.zeroBits = true + } + 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.decTable[position].symbol = 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.decTable { + symbol := v.symbol + nextState := symbolNext[symbol] + symbolNext[symbol] = nextState + 1 + nBits := s.actualTableLog - byte(highBits(uint32(nextState))) + s.decTable[u].nbBits = 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.decTable[u].newState = newState + } + } + return nil +} + +// decompress will decompress the bitstream. +// If the buffer is over-read an error is returned. +func (s *Scratch) decompress() error { + br := &s.bits + br.init(s.br.unread()) + + var s1, s2 decoder + // Initialize and decode first state and symbol. + s1.init(br, s.decTable, s.actualTableLog) + s2.init(br, s.decTable, s.actualTableLog) + + // Use temp table to avoid bound checks/append penalty. + var tmp = s.ct.tableSymbol[:256] + var off uint8 + + // Main part + if !s.zeroBits { + for br.off >= 8 { + br.fillFast() + tmp[off+0] = s1.nextFast() + tmp[off+1] = s2.nextFast() + br.fillFast() + tmp[off+2] = s1.nextFast() + tmp[off+3] = s2.nextFast() + off += 4 + // When off is 0, we have overflowed and should write. + if off == 0 { + s.Out = append(s.Out, tmp...) + if len(s.Out) >= s.DecompressLimit { + return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit) + } + } + } + } else { + for br.off >= 8 { + br.fillFast() + tmp[off+0] = s1.next() + tmp[off+1] = s2.next() + br.fillFast() + tmp[off+2] = s1.next() + tmp[off+3] = s2.next() + off += 4 + if off == 0 { + s.Out = append(s.Out, tmp...) + // When off is 0, we have overflowed and should write. + if len(s.Out) >= s.DecompressLimit { + return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit) + } + } + } + } + s.Out = append(s.Out, tmp[:off]...) + + // Final bits, a bit more expensive check + for { + if s1.finished() { + s.Out = append(s.Out, s1.final(), s2.final()) + break + } + br.fill() + s.Out = append(s.Out, s1.next()) + if s2.finished() { + s.Out = append(s.Out, s2.final(), s1.final()) + break + } + s.Out = append(s.Out, s2.next()) + if len(s.Out) >= s.DecompressLimit { + return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit) + } + } + return br.close() +} + +// decoder keeps track of the current state and updates it from the bitstream. +type decoder struct { + state uint16 + br *bitReader + dt []decSymbol +} + +// init will initialize the decoder and read the first state from the stream. +func (d *decoder) init(in *bitReader, dt []decSymbol, tableLog uint8) { + d.dt = dt + d.br = in + d.state = uint16(in.getBits(tableLog)) +} + +// next returns the next symbol and sets the next state. +// At least tablelog bits must be available in the bit reader. +func (d *decoder) next() uint8 { + n := &d.dt[d.state] + lowBits := d.br.getBits(n.nbBits) + d.state = n.newState + lowBits + return n.symbol +} + +// finished returns true if all bits have been read from the bitstream +// and the next state would require reading bits from the input. +func (d *decoder) finished() bool { + return d.br.finished() && d.dt[d.state].nbBits > 0 +} + +// final returns the current state symbol without decoding the next. +func (d *decoder) final() uint8 { + return d.dt[d.state].symbol +} + +// 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 (d *decoder) nextFast() uint8 { + n := d.dt[d.state] + lowBits := d.br.getBitsFast(n.nbBits) + d.state = n.newState + lowBits + return n.symbol +} diff --git a/vendor/github.com/klauspost/compress/fse/fse.go b/vendor/github.com/klauspost/compress/fse/fse.go new file mode 100644 index 000000000000..535cbadfdea1 --- /dev/null +++ b/vendor/github.com/klauspost/compress/fse/fse.go @@ -0,0 +1,144 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +// Package fse provides Finite State Entropy encoding and decoding. +// +// Finite State Entropy encoding provides a fast near-optimal symbol encoding/decoding +// for byte blocks as implemented in zstd. +// +// See https://github.com/klauspost/compress/tree/master/fse for more information. +package fse + +import ( + "errors" + "fmt" + "math/bits" +) + +const ( + /*!MEMORY_USAGE : + * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.) + * Increasing memory usage improves compression ratio + * Reduced memory usage can improve speed, due to cache effect + * Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */ + maxMemoryUsage = 14 + defaultMemoryUsage = 13 + + maxTableLog = maxMemoryUsage - 2 + maxTablesize = 1 << maxTableLog + defaultTablelog = defaultMemoryUsage - 2 + minTablelog = 5 + maxSymbolValue = 255 +) + +var ( + // ErrIncompressible is returned when input is judged to be too hard to compress. + ErrIncompressible = errors.New("input is not compressible") + + // ErrUseRLE is returned from the compressor when the input is a single byte value repeated. + ErrUseRLE = errors.New("input is single value repeated") +) + +// Scratch provides temporary storage for compression and decompression. +type Scratch struct { + // Private + count [maxSymbolValue + 1]uint32 + norm [maxSymbolValue + 1]int16 + br byteReader + bits bitReader + bw bitWriter + ct cTable // Compression tables. + decTable []decSymbol // Decompression table. + maxCount int // count of the most probable symbol + + // Per block parameters. + // These can be used to override compression parameters of the block. + // Do not touch, unless you know what you are doing. + + // Out is output buffer. + // If the scratch is re-used before the caller is done processing the output, + // set this field to nil. + // Otherwise the output buffer will be re-used for next Compression/Decompression step + // and allocation will be avoided. + Out []byte + + // DecompressLimit limits the maximum decoded size acceptable. + // If > 0 decompression will stop when approximately this many bytes + // has been decoded. + // If 0, maximum size will be 2GB. + DecompressLimit int + + symbolLen uint16 // Length of active part of the symbol table. + actualTableLog uint8 // Selected tablelog. + zeroBits bool // no bits has prob > 50%. + clearCount bool // clear count + + // MaxSymbolValue will override the maximum symbol value of the next block. + MaxSymbolValue uint8 + + // TableLog will attempt to override the tablelog for the next block. + TableLog uint8 +} + +// Histogram allows to populate the histogram and skip that step in the compression, +// It otherwise allows to inspect the histogram when compression is done. +// 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 *Scratch) Histogram() []uint32 { + return s.count[:] +} + +// HistogramFinished can be called to indicate that the histogram has been populated. +// maxSymbol is the index of the highest set symbol of the next data segment. +// maxCount is the number of entries in the most populated entry. +// These are accepted at face value. +func (s *Scratch) HistogramFinished(maxSymbol uint8, maxCount int) { + s.maxCount = maxCount + s.symbolLen = uint16(maxSymbol) + 1 + s.clearCount = maxCount != 0 +} + +// prepare will prepare and allocate scratch tables used for both compression and decompression. +func (s *Scratch) prepare(in []byte) (*Scratch, error) { + if s == nil { + s = &Scratch{} + } + if s.MaxSymbolValue == 0 { + s.MaxSymbolValue = 255 + } + if s.TableLog == 0 { + s.TableLog = defaultTablelog + } + if s.TableLog > maxTableLog { + return nil, fmt.Errorf("tableLog (%d) > maxTableLog (%d)", s.TableLog, maxTableLog) + } + if cap(s.Out) == 0 { + s.Out = make([]byte, 0, len(in)) + } + if s.clearCount && s.maxCount == 0 { + for i := range s.count { + s.count[i] = 0 + } + s.clearCount = false + } + s.br.init(in) + if s.DecompressLimit == 0 { + // Max size 2GB. + s.DecompressLimit = (2 << 30) - 1 + } + + return s, nil +} + +// tableStep returns the next table index. +func tableStep(tableSize uint32) uint32 { + return (tableSize >> 1) + (tableSize >> 3) + 3 +} + +func highBits(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} diff --git a/vendor/github.com/klauspost/compress/gzip/gzip.go b/vendor/github.com/klauspost/compress/gzip/gzip.go index ed0cc148f8c7..26203851bdfa 100644 --- a/vendor/github.com/klauspost/compress/gzip/gzip.go +++ b/vendor/github.com/klauspost/compress/gzip/gzip.go @@ -37,13 +37,13 @@ type Writer struct { Header // written at first call to Write, Flush, or Close w io.Writer level int - wroteHeader bool + err error compressor *flate.Writer digest uint32 // CRC-32, IEEE polynomial (section 8) size uint32 // Uncompressed size (section 2.3.1) + wroteHeader bool closed bool buf [10]byte - err error } // NewWriter returns a new Writer. @@ -207,7 +207,7 @@ func (z *Writer) Write(p []byte) (int, error) { z.size += uint32(len(p)) z.digest = crc32.Update(z.digest, crc32.IEEETable, p) if z.level == StatelessCompression { - return len(p), flate.StatelessDeflate(z.w, p, false) + return len(p), flate.StatelessDeflate(z.w, p, false, nil) } n, z.err = z.compressor.Write(p) return n, z.err @@ -255,7 +255,7 @@ func (z *Writer) Close() error { } } if z.level == StatelessCompression { - z.err = flate.StatelessDeflate(z.w, nil, true) + z.err = flate.StatelessDeflate(z.w, nil, true, nil) } else { z.err = z.compressor.Close() } diff --git a/vendor/github.com/klauspost/compress/huff0/.gitignore b/vendor/github.com/klauspost/compress/huff0/.gitignore new file mode 100644 index 000000000000..b3d262958f85 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/.gitignore @@ -0,0 +1 @@ +/huff0-fuzz.zip diff --git a/vendor/github.com/klauspost/compress/huff0/README.md b/vendor/github.com/klauspost/compress/huff0/README.md new file mode 100644 index 000000000000..ec4f9098866c --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/README.md @@ -0,0 +1,87 @@ +# Huff0 entropy compression + +This package provides Huff0 encoding and decoding as used in zstd. + +[Huff0](https://github.com/Cyan4973/FiniteStateEntropy#new-generation-entropy-coders), +a Huffman codec designed for modern CPU, featuring OoO (Out of Order) operations on multiple ALU +(Arithmetic Logic Unit), achieving extremely fast compression and decompression speeds. + +This can be used for compressing input with a lot of similar input values to the smallest number of bytes. +This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, +but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. + +* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/huff0) + +## News + + * Mar 2018: First implementation released. Consider this beta software for now. + +# Usage + +This package provides a low level interface that allows to compress single independent blocks. + +Each block is separate, and there is no built in integrity checks. +This means that the caller should keep track of block sizes and also do checksums if needed. + +Compressing a block is done via the [`Compress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress1X) and +[`Compress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress4X) functions. +You must provide input and will receive the output and maybe an error. + +These error values can be returned: + +| Error | Description | +|---------------------|-----------------------------------------------------------------------------| +| `` | Everything ok, output is returned | +| `ErrIncompressible` | Returned when input is judged to be too hard to compress | +| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | +| `ErrTooBig` | Returned if the input block exceeds the maximum allowed size (128 Kib) | +| `(error)` | An internal error occurred. | + + +As can be seen above some of there are errors that will be returned even under normal operation so it is important to handle these. + +To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object +that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same +object can be used for both. + +Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this +you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. + +The `Scratch` object will retain state that allows to re-use previous tables for encoding and decoding. + +## Tables and re-use + +Huff0 allows for reusing tables from the previous block to save space if that is expected to give better/faster results. + +The Scratch object allows you to set a [`ReusePolicy`](https://godoc.org/github.com/klauspost/compress/huff0#ReusePolicy) +that controls this behaviour. See the documentation for details. This can be altered between each block. + +Do however note that this information is *not* stored in the output block and it is up to the users of the package to +record whether [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable) should be called, +based on the boolean reported back from the CompressXX call. + +If you want to store the table separate from the data, you can access them as `OutData` and `OutTable` on the +[`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object. + +## Decompressing + +The first part of decoding is to initialize the decoding table through [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable). +This will initialize the decoding tables. +You can supply the complete block to `ReadTable` and it will return the data part of the block +which can be given to the decompressor. + +Decompressing is done by calling the [`Decompress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress1X) +or [`Decompress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress4X) function. + +For concurrently decompressing content with a fixed table a stateless [`Decoder`](https://godoc.org/github.com/klauspost/compress/huff0#Decoder) can be requested which will remain correct as long as the scratch is unchanged. The capacity of the provided slice indicates the expected output size. + +You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back +your input was likely corrupted. + +It is important to note that a successful decoding does *not* mean your output matches your original input. +There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. + +# Contributing + +Contributions are always welcome. Be aware that adding public functions will require good justification and breaking +changes will likely not be accepted. If in doubt open an issue before writing the PR. diff --git a/vendor/github.com/klauspost/compress/huff0/bitreader.go b/vendor/github.com/klauspost/compress/huff0/bitreader.go new file mode 100644 index 000000000000..a4979e8868a5 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/bitreader.go @@ -0,0 +1,329 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package huff0 + +import ( + "encoding/binary" + "errors" + "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] + value uint64 + bitsRead uint8 +} + +// init initializes and resets the bit reader. +func (b *bitReaderBytes) init(in []byte) error { + if len(in) < 1 { + return errors.New("corrupt stream: too short") + } + b.in = in + b.off = uint(len(in)) + // The highest bit of the last byte indicates where to start + v := in[len(in)-1] + if v == 0 { + return errors.New("corrupt stream, did not find end of stream") + } + b.bitsRead = 64 + b.value = 0 + if len(in) >= 8 { + b.fillFastStart() + } else { + b.fill() + b.fill() + } + b.advance(8 - uint8(highBit32(uint32(v)))) + return nil +} + +// peekBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReaderBytes) peekByteFast() uint8 { + got := uint8(b.value >> 56) + return got +} + +func (b *bitReaderBytes) advance(n uint8) { + b.bitsRead += n + b.value <<= n & 63 +} + +// fillFast() will make sure at least 32 bits are available. +// There must be at least 4 bytes available. +func (b *bitReaderBytes) fillFast() { + if b.bitsRead < 32 { + return + } + + // 2 bounds checks. + v := b.in[b.off-4 : b.off] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << (b.bitsRead - 32) + b.bitsRead -= 32 + b.off -= 4 +} + +// fillFastStart() assumes the bitReaderBytes is empty and there is at least 8 bytes to read. +func (b *bitReaderBytes) fillFastStart() { + // Do single re-slice to avoid bounds checks. + b.value = binary.LittleEndian.Uint64(b.in[b.off-8:]) + b.bitsRead = 0 + b.off -= 8 +} + +// fill() will make sure at least 32 bits are available. +func (b *bitReaderBytes) fill() { + if b.bitsRead < 32 { + return + } + if b.off > 4 { + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << (b.bitsRead - 32) + b.bitsRead -= 32 + b.off -= 4 + return + } + for b.off > 0 { + b.value |= uint64(b.in[b.off-1]) << (b.bitsRead - 8) + b.bitsRead -= 8 + b.off-- + } +} + +// finished returns true if all bits have been read from the bit stream. +func (b *bitReaderBytes) finished() bool { + return b.off == 0 && b.bitsRead >= 64 +} + +// close the bitstream and returns an error if out-of-buffer reads occurred. +func (b *bitReaderBytes) close() error { + // Release reference. + b.in = nil + if b.bitsRead > 64 { + return io.ErrUnexpectedEOF + } + return nil +} + +// bitReaderShifted reads a bitstream in reverse. +// The last set bit indicates the start of the stream and is used +// for aligning the input. +type bitReaderShifted struct { + in []byte + off uint // next byte to read is at in[off - 1] + value uint64 + bitsRead uint8 +} + +// init initializes and resets the bit reader. +func (b *bitReaderShifted) init(in []byte) error { + if len(in) < 1 { + return errors.New("corrupt stream: too short") + } + b.in = in + b.off = uint(len(in)) + // The highest bit of the last byte indicates where to start + v := in[len(in)-1] + if v == 0 { + return errors.New("corrupt stream, did not find end of stream") + } + b.bitsRead = 64 + b.value = 0 + if len(in) >= 8 { + b.fillFastStart() + } else { + b.fill() + b.fill() + } + b.advance(8 - uint8(highBit32(uint32(v)))) + return nil +} + +// peekBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReaderShifted) peekBitsFast(n uint8) uint16 { + return uint16(b.value >> ((64 - n) & 63)) +} + +func (b *bitReaderShifted) advance(n uint8) { + b.bitsRead += n + b.value <<= n & 63 +} + +// fillFast() will make sure at least 32 bits are available. +// There must be at least 4 bytes available. +func (b *bitReaderShifted) fillFast() { + if b.bitsRead < 32 { + return + } + + // 2 bounds checks. + v := b.in[b.off-4 : b.off] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << ((b.bitsRead - 32) & 63) + b.bitsRead -= 32 + b.off -= 4 +} + +// fillFastStart() assumes the bitReaderShifted is empty and there is at least 8 bytes to read. +func (b *bitReaderShifted) fillFastStart() { + // Do single re-slice to avoid bounds checks. + b.value = binary.LittleEndian.Uint64(b.in[b.off-8:]) + b.bitsRead = 0 + b.off -= 8 +} + +// fill() will make sure at least 32 bits are available. +func (b *bitReaderShifted) fill() { + if b.bitsRead < 32 { + return + } + if b.off > 4 { + v := b.in[b.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + b.value |= uint64(low) << ((b.bitsRead - 32) & 63) + b.bitsRead -= 32 + b.off -= 4 + return + } + for b.off > 0 { + b.value |= uint64(b.in[b.off-1]) << ((b.bitsRead - 8) & 63) + b.bitsRead -= 8 + b.off-- + } +} + +// finished returns true if all bits have been read from the bit stream. +func (b *bitReaderShifted) finished() bool { + return b.off == 0 && b.bitsRead >= 64 +} + +// close the bitstream and returns an error if out-of-buffer reads occurred. +func (b *bitReaderShifted) close() error { + // Release reference. + b.in = nil + if b.bitsRead > 64 { + return io.ErrUnexpectedEOF + } + return nil +} diff --git a/vendor/github.com/klauspost/compress/huff0/bitwriter.go b/vendor/github.com/klauspost/compress/huff0/bitwriter.go new file mode 100644 index 000000000000..6bce4e87d4ff --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/bitwriter.go @@ -0,0 +1,210 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package huff0 + +import "fmt" + +// bitWriter will write bits. +// First bit will be LSB of the first byte of output. +type bitWriter struct { + bitContainer uint64 + nBits uint8 + out []byte +} + +// bitMask16 is bitmasks. Has extra to avoid bounds check. +var bitMask16 = [32]uint16{ + 0, 1, 3, 7, 0xF, 0x1F, + 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, + 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF, + 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) { + b.bitContainer |= uint64(value) << (b.nBits & 63) + b.nBits += bits +} + +// encSymbol 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) encSymbol(ct cTable, symbol byte) { + enc := ct[symbol] + b.bitContainer |= uint64(enc.val) << (b.nBits & 63) + if false { + if enc.nBits == 0 { + panic("nbits 0") + } + } + b.nBits += enc.nBits +} + +// encTwoSymbols will add up to 32 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) encTwoSymbols(ct cTable, av, bv byte) { + encA := ct[av] + encB := ct[bv] + sh := b.nBits & 63 + combined := uint64(encA.val) | (uint64(encB.val) << (encA.nBits & 63)) + b.bitContainer |= combined << sh + if false { + if encA.nBits == 0 { + panic("nbitsA 0") + } + if encB.nBits == 0 { + panic("nbitsB 0") + } + } + b.nBits += encA.nBits + encB.nBits +} + +// 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 { + return + } + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24)) + b.nBits -= 32 + b.bitContainer >>= 32 +} + +// flushAlign will flush remaining full bytes and align to next byte boundary. +func (b *bitWriter) flushAlign() { + nbBytes := (b.nBits + 7) >> 3 + for i := uint8(0); i < nbBytes; i++ { + b.out = append(b.out, byte(b.bitContainer>>(i*8))) + } + b.nBits = 0 + b.bitContainer = 0 +} + +// close will write the alignment bit and write the final byte(s) +// to the output. +func (b *bitWriter) close() error { + // End mark + b.addBits16Clean(1, 1) + // flush until next byte. + 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 new file mode 100644 index 000000000000..50bcdf6ea99c --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/bytereader.go @@ -0,0 +1,54 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package huff0 + +// byteReader provides a byte reader that reads +// little endian values from a byte stream. +// The input stream is manually advanced. +// The reader performs no bounds checks. +type byteReader struct { + b []byte + 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) +} + +// Int32 returns a little endian int32 starting at current offset. +func (b byteReader) Int32() int32 { + v3 := int32(b.b[b.off+3]) + v2 := int32(b.b[b.off+2]) + v1 := int32(b.b[b.off+1]) + v0 := int32(b.b[b.off]) + return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0 +} + +// Uint32 returns a little endian uint32 starting at current offset. +func (b byteReader) Uint32() uint32 { + v3 := uint32(b.b[b.off+3]) + v2 := uint32(b.b[b.off+2]) + v1 := uint32(b.b[b.off+1]) + v0 := uint32(b.b[b.off]) + 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 new file mode 100644 index 000000000000..f9ed5f8306ec --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/compress.go @@ -0,0 +1,657 @@ +package huff0 + +import ( + "fmt" + "runtime" + "sync" +) + +// Compress1X will compress the input. +// The output can be decoded using Decompress1X. +// Supply a Scratch object. The scratch object contains state about re-use, +// So when sharing across independent encodes, be sure to set the re-use policy. +func Compress1X(in []byte, s *Scratch) (out []byte, reUsed bool, err error) { + s, err = s.prepare(in) + if err != nil { + return nil, false, err + } + return compress(in, s, s.compress1X) +} + +// Compress4X will compress the input. The input is split into 4 independent blocks +// and compressed similar to Compress1X. +// The output can be decoded using Decompress4X. +// Supply a Scratch object. The scratch object contains state about re-use, +// So when sharing across independent encodes, be sure to set the re-use policy. +func Compress4X(in []byte, s *Scratch) (out []byte, reUsed bool, err error) { + s, err = s.prepare(in) + if err != nil { + return nil, false, err + } + if false { + // TODO: compress4Xp only slightly faster. + const parallelThreshold = 8 << 10 + if len(in) < parallelThreshold || runtime.GOMAXPROCS(0) == 1 { + return compress(in, s, s.compress4X) + } + return compress(in, s, s.compress4Xp) + } + return compress(in, s, s.compress4X) +} + +func compress(in []byte, s *Scratch, compressor func(src []byte) ([]byte, error)) (out []byte, reUsed bool, err error) { + // Nuke previous table if we cannot reuse anyway. + if s.Reuse == ReusePolicyNone { + s.prevTable = s.prevTable[:0] + } + + // Create histogram, if none was provided. + 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 nil, false, fmt.Errorf("maxCount (%d) > length (%d)", maxCount, len(in)) + } + if len(in) == 1 { + return nil, false, ErrIncompressible + } + // One symbol, use RLE + return nil, false, ErrUseRLE + } + if maxCount == 1 || maxCount < (len(in)>>7) { + // Each symbol present maximum once or too well distributed. + return nil, false, ErrIncompressible + } + if s.Reuse == ReusePolicyMust && !canReuse { + // We must reuse, but we can't. + return nil, false, ErrIncompressible + } + if (s.Reuse == ReusePolicyPrefer || s.Reuse == ReusePolicyMust) && canReuse { + keepTable := s.cTable + keepTL := s.actualTableLog + s.cTable = s.prevTable + s.actualTableLog = s.prevTableLog + s.Out, err = compressor(in) + s.cTable = keepTable + s.actualTableLog = keepTL + if err == nil && len(s.Out) < wantSize { + s.OutData = s.Out + return s.Out, true, nil + } + if s.Reuse == ReusePolicyMust { + return nil, false, ErrIncompressible + } + // Do not attempt to re-use later. + s.prevTable = s.prevTable[:0] + } + + // Calculate new table. + err = s.buildCTable() + if err != nil { + return nil, false, err + } + + if false && !s.canUseTable(s.cTable) { + panic("invalid table generated") + } + + if s.Reuse == ReusePolicyAllow && canReuse { + hSize := len(s.Out) + oldSize := s.prevTable.estimateSize(s.count[:s.symbolLen]) + newSize := s.cTable.estimateSize(s.count[:s.symbolLen]) + if oldSize <= hSize+newSize || hSize+12 >= wantSize { + // Retain cTable even if we re-use. + keepTable := s.cTable + keepTL := s.actualTableLog + + s.cTable = s.prevTable + s.actualTableLog = s.prevTableLog + s.Out, err = compressor(in) + + // Restore ctable. + s.cTable = keepTable + s.actualTableLog = keepTL + if err != nil { + return nil, false, err + } + if len(s.Out) >= wantSize { + return nil, false, ErrIncompressible + } + s.OutData = s.Out + return s.Out, true, nil + } + } + + // Use new table + err = s.cTable.write(s) + if err != nil { + s.OutTable = nil + return nil, false, err + } + s.OutTable = s.Out + + // Compress using new table + s.Out, err = compressor(in) + if err != nil { + s.OutTable = nil + return nil, false, err + } + if len(s.Out) >= wantSize { + s.OutTable = nil + return nil, false, ErrIncompressible + } + // Move current table into previous. + s.prevTable, s.prevTableLog, s.cTable = s.cTable, s.actualTableLog, s.prevTable[:0] + s.OutData = s.Out[len(s.OutTable):] + return s.Out, false, nil +} + +func (s *Scratch) compress1X(src []byte) ([]byte, error) { + return s.compress1xDo(s.Out, src) +} + +func (s *Scratch) compress1xDo(dst, src []byte) ([]byte, error) { + var bw = bitWriter{out: dst} + + // N is length divisible by 4. + n := len(src) + n -= n & 3 + cTable := s.cTable[:256] + + // Encode last bytes. + for i := len(src) & 3; i > 0; i-- { + bw.encSymbol(cTable, src[n+i-1]) + } + n -= 4 + if s.actualTableLog <= 8 { + for ; n >= 0; n -= 4 { + tmp := src[n : n+4] + // tmp should be len 4 + bw.flush32() + bw.encTwoSymbols(cTable, tmp[3], tmp[2]) + bw.encTwoSymbols(cTable, tmp[1], tmp[0]) + } + } else { + for ; n >= 0; n -= 4 { + tmp := src[n : n+4] + // tmp should be len 4 + bw.flush32() + bw.encTwoSymbols(cTable, tmp[3], tmp[2]) + bw.flush32() + bw.encTwoSymbols(cTable, tmp[1], tmp[0]) + } + } + err := bw.close() + return bw.out, err +} + +var sixZeros [6]byte + +func (s *Scratch) compress4X(src []byte) ([]byte, error) { + if len(src) < 12 { + return nil, ErrIncompressible + } + segmentSize := (len(src) + 3) / 4 + + // Add placeholder for output length + offsetIdx := len(s.Out) + s.Out = append(s.Out, sixZeros[:]...) + + for i := 0; i < 4; i++ { + toDo := src + if len(toDo) > segmentSize { + toDo = toDo[:segmentSize] + } + src = src[len(toDo):] + + var err error + idx := len(s.Out) + s.Out, err = s.compress1xDo(s.Out, toDo) + if err != nil { + return nil, err + } + // Write compressed length as little endian before block. + if i < 3 { + // Last length is not written. + length := len(s.Out) - idx + s.Out[i*2+offsetIdx] = byte(length) + s.Out[i*2+offsetIdx+1] = byte(length >> 8) + } + } + + return s.Out, nil +} + +// compress4Xp will compress 4 streams using separate goroutines. +func (s *Scratch) compress4Xp(src []byte) ([]byte, error) { + if len(src) < 12 { + return nil, ErrIncompressible + } + // Add placeholder for output length + s.Out = s.Out[:6] + + segmentSize := (len(src) + 3) / 4 + var wg sync.WaitGroup + var errs [4]error + wg.Add(4) + for i := 0; i < 4; i++ { + toDo := src + if len(toDo) > segmentSize { + toDo = toDo[:segmentSize] + } + src = src[len(toDo):] + + // Separate goroutine for each block. + go func(i int) { + s.tmpOut[i], errs[i] = s.compress1xDo(s.tmpOut[i][:0], toDo) + wg.Done() + }(i) + } + wg.Wait() + for i := 0; i < 4; i++ { + if errs[i] != nil { + return nil, errs[i] + } + o := s.tmpOut[i] + // Write compressed length as little endian before block. + if i < 3 { + // Last length is not written. + s.Out[i*2] = byte(len(o)) + s.Out[i*2+1] = byte(len(o) >> 8) + } + + // Write output. + s.Out = append(s.Out, o...) + } + return s.Out, nil +} + +// countSimple will create a simple histogram in s.count. +// Returns the biggest count. +// Does not update s.clearCount. +func (s *Scratch) countSimple(in []byte) (max int, reuse bool) { + reuse = true + for _, v := range in { + s.count[v]++ + } + m := uint32(0) + if len(s.prevTable) > 0 { + for i, v := range s.count[:] { + if v > m { + m = v + } + if v > 0 { + s.symbolLen = uint16(i) + 1 + if i >= len(s.prevTable) { + reuse = false + } else { + if s.prevTable[i].nBits == 0 { + reuse = false + } + } + } + } + return int(m), reuse + } + for i, v := range s.count[:] { + if v > m { + m = v + } + if v > 0 { + s.symbolLen = uint16(i) + 1 + } + } + return int(m), false +} + +func (s *Scratch) canUseTable(c cTable) bool { + if len(c) < int(s.symbolLen) { + return false + } + for i, v := range s.count[:s.symbolLen] { + if v != 0 && c[i].nBits == 0 { + return false + } + } + return true +} + +func (s *Scratch) validateTable(c cTable) bool { + if len(c) < int(s.symbolLen) { + return false + } + for i, v := range s.count[:s.symbolLen] { + if v != 0 { + if c[i].nBits == 0 { + return false + } + if c[i].nBits > s.actualTableLog { + return false + } + } + } + return true +} + +// minTableLog provides the minimum logSize to safely represent a distribution. +func (s *Scratch) minTableLog() uint8 { + minBitsSrc := highBit32(uint32(s.br.remain())) + 1 + minBitsSymbols := highBit32(uint32(s.symbolLen-1)) + 2 + if minBitsSrc < minBitsSymbols { + return uint8(minBitsSrc) + } + return uint8(minBitsSymbols) +} + +// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog +func (s *Scratch) optimalTableLog() { + tableLog := s.TableLog + minBits := s.minTableLog() + maxBitsSrc := uint8(highBit32(uint32(s.br.remain()-1))) - 1 + if maxBitsSrc < tableLog { + // Accuracy can be reduced + tableLog = maxBitsSrc + } + if minBits > tableLog { + tableLog = minBits + } + // Need a minimum to safely represent all symbol values + if tableLog < minTablelog { + tableLog = minTablelog + } + if tableLog > tableLogMax { + tableLog = tableLogMax + } + s.actualTableLog = tableLog +} + +type cTableEntry struct { + val uint16 + nBits uint8 + // We have 8 bits extra +} + +const huffNodesMask = huffNodesLen - 1 + +func (s *Scratch) buildCTable() error { + s.optimalTableLog() + s.huffSort() + if cap(s.cTable) < maxSymbolValue+1 { + s.cTable = make([]cTableEntry, s.symbolLen, maxSymbolValue+1) + } else { + s.cTable = s.cTable[:s.symbolLen] + for i := range s.cTable { + s.cTable[i] = cTableEntry{} + } + } + + var startNode = int16(s.symbolLen) + nonNullRank := s.symbolLen - 1 + + nodeNb := int16(startNode) + huffNode := s.nodes[1 : huffNodesLen+1] + + // This overlays the slice above, but allows "-1" index lookups. + // Different from reference implementation. + huffNode0 := s.nodes[0 : huffNodesLen+1] + + for huffNode[nonNullRank].count == 0 { + nonNullRank-- + } + + lowS := int16(nonNullRank) + nodeRoot := nodeNb + lowS - 1 + lowN := nodeNb + huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count + huffNode[lowS].parent, huffNode[lowS-1].parent = uint16(nodeNb), uint16(nodeNb) + nodeNb++ + lowS -= 2 + for n := nodeNb; n <= nodeRoot; n++ { + huffNode[n].count = 1 << 30 + } + // fake entry, strong barrier + huffNode0[0].count = 1 << 31 + + // create parents + for nodeNb <= nodeRoot { + var n1, n2 int16 + if huffNode0[lowS+1].count < huffNode0[lowN+1].count { + n1 = lowS + lowS-- + } else { + n1 = lowN + lowN++ + } + if huffNode0[lowS+1].count < huffNode0[lowN+1].count { + n2 = lowS + lowS-- + } else { + n2 = lowN + lowN++ + } + + huffNode[nodeNb].count = huffNode0[n1+1].count + huffNode0[n2+1].count + huffNode0[n1+1].parent, huffNode0[n2+1].parent = uint16(nodeNb), uint16(nodeNb) + nodeNb++ + } + + // distribute weights (unlimited tree height) + huffNode[nodeRoot].nbBits = 0 + for n := nodeRoot - 1; n >= startNode; n-- { + huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1 + } + for n := uint16(0); n <= nonNullRank; n++ { + huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1 + } + s.actualTableLog = s.setMaxHeight(int(nonNullRank)) + maxNbBits := s.actualTableLog + + // fill result into tree (val, nbBits) + if maxNbBits > tableLogMax { + return fmt.Errorf("internal error: maxNbBits (%d) > tableLogMax (%d)", maxNbBits, tableLogMax) + } + var nbPerRank [tableLogMax + 1]uint16 + var valPerRank [16]uint16 + for _, v := range huffNode[:nonNullRank+1] { + nbPerRank[v.nbBits]++ + } + // determine stating value per rank + { + min := uint16(0) + for n := maxNbBits; n > 0; n-- { + // get starting value within each rank + valPerRank[n] = min + min += nbPerRank[n] + min >>= 1 + } + } + + // push nbBits per symbol, symbol order + for _, v := range huffNode[:nonNullRank+1] { + s.cTable[v.symbol].nBits = v.nbBits + } + + // assign value within rank, symbol order + t := s.cTable[:s.symbolLen] + for n, val := range t { + nbits := val.nBits & 15 + v := valPerRank[nbits] + t[n].val = v + valPerRank[nbits] = v + 1 + } + + return nil +} + +// huffSort will sort symbols, decreasing order. +func (s *Scratch) huffSort() { + type rankPos struct { + base uint32 + current uint32 + } + + // Clear nodes + nodes := s.nodes[:huffNodesLen+1] + s.nodes = nodes + nodes = nodes[1 : huffNodesLen+1] + + // Sort into buckets based on length of symbol count. + var rank [32]rankPos + for _, v := range s.count[:s.symbolLen] { + r := highBit32(v+1) & 31 + rank[r].base++ + } + // maxBitLength is log2(BlockSizeMax) + 1 + const maxBitLength = 18 + 1 + for n := maxBitLength; n > 0; n-- { + rank[n-1].base += rank[n].base + } + for n := range rank[:maxBitLength] { + rank[n].current = rank[n].base + } + for n, c := range s.count[:s.symbolLen] { + r := (highBit32(c+1) + 1) & 31 + pos := rank[r].current + rank[r].current++ + prev := nodes[(pos-1)&huffNodesMask] + for pos > rank[r].base && c > prev.count { + nodes[pos&huffNodesMask] = prev + pos-- + prev = nodes[(pos-1)&huffNodesMask] + } + nodes[pos&huffNodesMask] = nodeElt{count: c, symbol: byte(n)} + } + return +} + +func (s *Scratch) setMaxHeight(lastNonNull int) uint8 { + maxNbBits := s.actualTableLog + huffNode := s.nodes[1 : huffNodesLen+1] + //huffNode = huffNode[: huffNodesLen] + + largestBits := huffNode[lastNonNull].nbBits + + // early exit : no elt > maxNbBits + if largestBits <= maxNbBits { + return largestBits + } + totalCost := int(0) + baseCost := int(1) << (largestBits - maxNbBits) + n := uint32(lastNonNull) + + for huffNode[n].nbBits > maxNbBits { + totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits)) + huffNode[n].nbBits = maxNbBits + n-- + } + // n stops at huffNode[n].nbBits <= maxNbBits + + for huffNode[n].nbBits == maxNbBits { + n-- + } + // n end at index of smallest symbol using < maxNbBits + + // renorm totalCost + totalCost >>= largestBits - maxNbBits /* note : totalCost is necessarily a multiple of baseCost */ + + // repay normalized cost + { + const noSymbol = 0xF0F0F0F0 + var rankLast [tableLogMax + 2]uint32 + + for i := range rankLast[:] { + rankLast[i] = noSymbol + } + + // Get pos of last (smallest) symbol per rank + { + currentNbBits := uint8(maxNbBits) + for pos := int(n); pos >= 0; pos-- { + if huffNode[pos].nbBits >= currentNbBits { + continue + } + currentNbBits = huffNode[pos].nbBits // < maxNbBits + rankLast[maxNbBits-currentNbBits] = uint32(pos) + } + } + + for totalCost > 0 { + nBitsToDecrease := uint8(highBit32(uint32(totalCost))) + 1 + + for ; nBitsToDecrease > 1; nBitsToDecrease-- { + highPos := rankLast[nBitsToDecrease] + lowPos := rankLast[nBitsToDecrease-1] + if highPos == noSymbol { + continue + } + if lowPos == noSymbol { + break + } + highTotal := huffNode[highPos].count + lowTotal := 2 * huffNode[lowPos].count + if highTotal <= lowTotal { + break + } + } + // only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) + // HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary + // FIXME: try to remove + for (nBitsToDecrease <= tableLogMax) && (rankLast[nBitsToDecrease] == noSymbol) { + nBitsToDecrease++ + } + totalCost -= 1 << (nBitsToDecrease - 1) + if rankLast[nBitsToDecrease-1] == noSymbol { + // this rank is no longer empty + rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease] + } + huffNode[rankLast[nBitsToDecrease]].nbBits++ + if rankLast[nBitsToDecrease] == 0 { + /* special case, reached largest symbol */ + rankLast[nBitsToDecrease] = noSymbol + } else { + rankLast[nBitsToDecrease]-- + if huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease { + rankLast[nBitsToDecrease] = noSymbol /* this rank is now empty */ + } + } + } + + for totalCost < 0 { /* Sometimes, cost correction overshoot */ + if rankLast[1] == noSymbol { /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */ + for huffNode[n].nbBits == maxNbBits { + n-- + } + huffNode[n+1].nbBits-- + rankLast[1] = n + 1 + totalCost++ + continue + } + huffNode[rankLast[1]+1].nbBits-- + rankLast[1]++ + totalCost++ + } + } + return maxNbBits +} + +type nodeElt struct { + count uint32 + parent uint16 + symbol byte + nbBits uint8 +} diff --git a/vendor/github.com/klauspost/compress/huff0/decompress.go b/vendor/github.com/klauspost/compress/huff0/decompress.go new file mode 100644 index 000000000000..41703bba4d65 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/decompress.go @@ -0,0 +1,1164 @@ +package huff0 + +import ( + "errors" + "fmt" + "io" + + "github.com/klauspost/compress/fse" +) + +type dTable struct { + single []dEntrySingle + double []dEntryDouble +} + +// single-symbols decoding +type dEntrySingle struct { + 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 + +// ReadTable will read a table from the input. +// The size of the input may be larger than the table definition. +// Any content remaining after the table definition will be returned. +// If no Scratch is provided a new one is allocated. +// The returned Scratch can be used for encoding or decoding input using this table. +func ReadTable(in []byte, s *Scratch) (s2 *Scratch, remain []byte, err error) { + s, err = s.prepare(in) + if err != nil { + return s, nil, err + } + if len(in) <= 1 { + return s, nil, errors.New("input too small for table") + } + iSize := in[0] + in = in[1:] + if iSize >= 128 { + // Uncompressed + oSize := iSize - 127 + iSize = (oSize + 1) / 2 + if int(iSize) > len(in) { + return s, nil, errors.New("input too small for table") + } + for n := uint8(0); n < oSize; n += 2 { + v := in[n/2] + s.huffWeight[n] = v >> 4 + s.huffWeight[n+1] = v & 15 + } + s.symbolLen = uint16(oSize) + in = in[iSize:] + } else { + if len(in) < int(iSize) { + return s, nil, fmt.Errorf("input too small for table, want %d bytes, have %d", iSize, len(in)) + } + // FSE compressed weights + s.fse.DecompressLimit = 255 + hw := s.huffWeight[:] + s.fse.Out = hw + b, err := fse.Decompress(in[:iSize], s.fse) + s.fse.Out = nil + if err != nil { + return s, nil, err + } + if len(b) > 255 { + return s, nil, errors.New("corrupt input: output table too large") + } + s.symbolLen = uint16(len(b)) + in = in[iSize:] + } + + // collect weight stats + var rankStats [16]uint32 + weightTotal := uint32(0) + for _, v := range s.huffWeight[:s.symbolLen] { + if v > tableLogMax { + return s, nil, errors.New("corrupt input: weight too large") + } + v2 := v & 15 + rankStats[v2]++ + // (1 << (v2-1)) is slower since the compiler cannot prove that v2 isn't 0. + weightTotal += (1 << v2) >> 1 + } + if weightTotal == 0 { + return s, nil, errors.New("corrupt input: weights zero") + } + + // get last non-null symbol weight (implied, total must be 2^n) + { + tableLog := highBit32(weightTotal) + 1 + if tableLog > tableLogMax { + return s, nil, errors.New("corrupt input: tableLog too big") + } + s.actualTableLog = uint8(tableLog) + // determine last weight + { + total := uint32(1) << tableLog + rest := total - weightTotal + verif := uint32(1) << highBit32(rest) + lastWeight := highBit32(rest) + 1 + if verif != rest { + // last value must be a clean power of 2 + return s, nil, errors.New("corrupt input: last value not power of two") + } + s.huffWeight[s.symbolLen] = uint8(lastWeight) + s.symbolLen++ + rankStats[lastWeight]++ + } + } + + if (rankStats[1] < 2) || (rankStats[1]&1 != 0) { + // by construction : at least 2 elts of rank 1, must be even + return s, nil, errors.New("corrupt input: min elt size, even check failed ") + } + + // TODO: Choose between single/double symbol decoding + + // Calculate starting value for each rank + { + var nextRankStart uint32 + for n := uint8(1); n < s.actualTableLog+1; n++ { + current := nextRankStart + nextRankStart += rankStats[n] << (n - 1) + rankStats[n] = current + } + } + + // fill DTable (always full size) + tSize := 1 << tableLogMax + if len(s.dt.single) != tSize { + s.dt.single = make([]dEntrySingle, tSize) + } + cTable := s.prevTable + if cap(cTable) < maxSymbolValue+1 { + cTable = make([]cTableEntry, 0, maxSymbolValue+1) + } + cTable = cTable[:maxSymbolValue+1] + s.prevTable = cTable[:s.symbolLen] + s.prevTableLog = s.actualTableLog + + for n, w := range s.huffWeight[:s.symbolLen] { + if w == 0 { + cTable[n] = cTableEntry{ + val: 0, + nBits: 0, + } + continue + } + length := (uint32(1) << w) >> 1 + d := dEntrySingle{ + entry: uint16(s.actualTableLog+1-w) | (uint16(n) << 8), + } + + rank := &rankStats[w] + cTable[n] = cTableEntry{ + val: uint16(*rank >> (w - 1)), + nBits: uint8(d.entry), + } + + single := s.dt.single[*rank : *rank+length] + for i := range single { + single[i] = d + } + *rank += length + } + + return s, in, nil +} + +// Decompress1X will decompress a 1X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// Before this is called, the table must be initialized with ReadTable unless +// the encoder re-used the table. +// deprecated: Use the stateless Decoder() to get a concurrent version. +func (s *Scratch) Decompress1X(in []byte) (out []byte, err error) { + if cap(s.Out) < s.MaxDecodedSize { + s.Out = make([]byte, s.MaxDecodedSize) + } + s.Out = s.Out[:0:s.MaxDecodedSize] + s.Out, err = s.Decoder().Decompress1X(s.Out, in) + return s.Out, err +} + +// Decompress4X will decompress a 4X encoded stream. +// Before this is called, the table must be initialized with ReadTable unless +// the encoder re-used the table. +// The length of the supplied input must match the end of a block exactly. +// The destination size of the uncompressed data must be known and provided. +// deprecated: Use the stateless Decoder() to get a concurrent version. +func (s *Scratch) Decompress4X(in []byte, dstSize int) (out []byte, err error) { + if dstSize > s.MaxDecodedSize { + return nil, ErrMaxDecodedSizeExceeded + } + if cap(s.Out) < dstSize { + s.Out = make([]byte, s.MaxDecodedSize) + } + s.Out = s.Out[:0:dstSize] + s.Out, err = s.Decoder().Decompress4X(s.Out, in) + return s.Out, err +} + +// Decoder will return a stateless decoder that can be used by multiple +// decompressors concurrently. +// Before this is called, the table must be initialized with ReadTable. +// The Decoder is still linked to the scratch buffer so that cannot be reused. +// However, it is safe to discard the scratch. +func (s *Scratch) Decoder() *Decoder { + return &Decoder{ + dt: s.dt, + actualTableLog: s.actualTableLog, + } +} + +// Decoder provides stateless decoding. +type Decoder struct { + dt dTable + actualTableLog uint8 +} + +// Decompress1X will decompress a 1X encoded stream. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) { + if len(d.dt.single) == 0 { + return nil, errors.New("no table loaded") + } + if use8BitTables && d.actualTableLog <= 8 { + return d.decompress1X8Bit(dst, src) + } + var br bitReaderShifted + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:0] + + // Avoid bounds check by always having full sized table. + const tlSize = 1 << tableLogMax + const tlMask = tlSize - 1 + dt := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + var buf [256]byte + var off uint8 + + for br.off >= 8 { + br.fillFast() + v := dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + // Refill + br.fillFast() + + v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + + if len(dst)+int(off) > maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:off]...) + + // br < 8, so uint8 is fine + bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead + for bitsLeft > 0 { + br.fill() + if false && br.bitsRead >= 32 { + if br.off >= 4 { + v := br.in[br.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + br.value = (br.value << 32) | uint64(low) + br.bitsRead -= 32 + br.off -= 4 + } else { + for br.off > 0 { + br.value = (br.value << 8) | uint64(br.in[br.off-1]) + br.bitsRead -= 8 + br.off-- + } + } + } + if len(dst) >= maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= nBits + dst = append(dst, uint8(v.entry>>8)) + } + return dst, br.close() +} + +// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) decompress1X8Bit(dst, src []byte) ([]byte, error) { + if d.actualTableLog == 8 { + return d.decompress1X8BitExactly(dst, src) + } + var br bitReaderBytes + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:0] + + // Avoid bounds check by always having full sized table. + dt := d.dt.single[:256] + + // Use temp table to avoid bound checks/append penalty. + var buf [256]byte + var off uint8 + + shift := (8 - d.actualTableLog) & 7 + + //fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog) + for br.off >= 4 { + br.fillFast() + v := dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + + if len(dst)+int(off) > maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:off]...) + + // br < 4, so uint8 is fine + bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead)) + for bitsLeft > 0 { + if br.bitsRead >= 64-8 { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + if len(dst) >= maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + v := dt[br.peekByteFast()>>shift] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= int8(nBits) + dst = append(dst, uint8(v.entry>>8)) + } + return dst, br.close() +} + +// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8. +// The cap of the output buffer will be the maximum decompressed size. +// The length of the supplied input must match the end of a block exactly. +func (d *Decoder) decompress1X8BitExactly(dst, src []byte) ([]byte, error) { + var br bitReaderBytes + err := br.init(src) + if err != nil { + return dst, err + } + maxDecodedSize := cap(dst) + dst = dst[:0] + + // Avoid bounds check by always having full sized table. + dt := d.dt.single[:256] + + // Use temp table to avoid bound checks/append penalty. + var buf [256]byte + var off uint8 + + const shift = 0 + + //fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog) + for br.off >= 4 { + br.fillFast() + v := dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+0] = uint8(v.entry >> 8) + + v = dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+1] = uint8(v.entry >> 8) + + v = dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+2] = uint8(v.entry >> 8) + + v = dt[br.peekByteFast()>>shift] + br.advance(uint8(v.entry)) + buf[off+3] = uint8(v.entry >> 8) + + off += 4 + if off == 0 { + if len(dst)+256 > maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:]...) + } + } + + if len(dst)+int(off) > maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + dst = append(dst, buf[:off]...) + + // br < 4, so uint8 is fine + bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead)) + for bitsLeft > 0 { + if br.bitsRead >= 64-8 { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + if len(dst) >= maxDecodedSize { + br.close() + return nil, ErrMaxDecodedSizeExceeded + } + v := dt[br.peekByteFast()>>shift] + nBits := uint8(v.entry) + br.advance(nBits) + bitsLeft -= int8(nBits) + dst = append(dst, uint8(v.entry>>8)) + } + return dst, br.close() +} + +// Decompress4X will decompress a 4X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// The *capacity* of the dst slice must match the destination size of +// the uncompressed data exactly. +func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) { + if len(d.dt.single) == 0 { + return nil, errors.New("no table loaded") + } + if len(src) < 6+(4*1) { + return nil, errors.New("input too small") + } + if use8BitTables && d.actualTableLog <= 8 { + return d.decompress4X8bit(dst, src) + } + + var br [4]bitReaderShifted + start := 6 + for i := 0; i < 3; i++ { + length := int(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) + } + + var br [4]bitReaderBytes + start := 6 + for i := 0; i < 3; i++ { + length := int(src[i*2]) | (int(src[i*2+1]) << 8) + if start+length >= len(src) { + return nil, errors.New("truncated input (or invalid offset)") + } + err := br[i].init(src[start : start+length]) + if err != nil { + return nil, err + } + start += length + } + err := br[3].init(src[start:]) + if err != nil { + return nil, err + } + + // destination, offset to match first output + dstSize := cap(dst) + dst = dst[:dstSize] + out := dst + dstEvery := (dstSize + 3) / 4 + + shift := (8 - d.actualTableLog) & 7 + + const tlSize = 1 << 8 + const tlMask = tlSize - 1 + single := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + var buf [256]byte + var off uint8 + var decoded int + + // Decode 4 values from each decoder/loop. + const bufoff = 256 / 4 + for { + if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 { + break + } + + { + // Interleave 2 decodes. + const stream = 0 + const stream2 = 1 + br[stream].fillFast() + br[stream2].fillFast() + + v := single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 := single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+1] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+1] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+2] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+3] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+3] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + } + + { + const stream = 2 + const stream2 = 3 + br[stream].fillFast() + br[stream2].fillFast() + + v := single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 := single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+1] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+1] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+2] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+3] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+3] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + } + + off += 4 + + if off == bufoff { + if bufoff > dstEvery { + return nil, errors.New("corruption detected: stream overrun 1") + } + copy(out, buf[:bufoff]) + copy(out[dstEvery:], buf[bufoff:bufoff*2]) + copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3]) + copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4]) + off = 0 + out = out[bufoff:] + decoded += 256 + // There must at least be 3 buffers left. + if len(out) < dstEvery*3 { + return nil, errors.New("corruption detected: stream overrun 2") + } + } + } + if off > 0 { + ioff := int(off) + if len(out) < dstEvery*3+ioff { + return nil, errors.New("corruption detected: stream overrun 3") + } + copy(out, buf[:off]) + copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2]) + copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3]) + copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4]) + decoded += int(off) * 4 + out = out[off:] + } + + // Decode remaining. + for i := range br { + offset := dstEvery * i + br := &br[i] + bitsLeft := int(br.off*8) + int(64-br.bitsRead) + for bitsLeft > 0 { + if br.finished() { + return nil, io.ErrUnexpectedEOF + } + if br.bitsRead >= 56 { + if br.off >= 4 { + v := br.in[br.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + br.value |= uint64(low) << (br.bitsRead - 32) + br.bitsRead -= 32 + br.off -= 4 + } else { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + } + // end inline... + if offset >= len(out) { + return nil, errors.New("corruption detected: stream overrun 4") + } + + // Read value and increment offset. + v := single[br.peekByteFast()>>shift].entry + nBits := uint8(v) + br.advance(nBits) + bitsLeft -= int(nBits) + out[offset] = uint8(v >> 8) + offset++ + } + decoded += offset - dstEvery*i + err = br.close() + if err != nil { + return nil, err + } + } + if dstSize != decoded { + return nil, errors.New("corruption detected: short output block") + } + return dst, nil +} + +// Decompress4X will decompress a 4X encoded stream. +// The length of the supplied input must match the end of a block exactly. +// The *capacity* of the dst slice must match the destination size of +// the uncompressed data exactly. +func (d *Decoder) decompress4X8bitExactly(dst, src []byte) ([]byte, error) { + var br [4]bitReaderBytes + start := 6 + for i := 0; i < 3; i++ { + length := int(src[i*2]) | (int(src[i*2+1]) << 8) + if start+length >= len(src) { + return nil, errors.New("truncated input (or invalid offset)") + } + err := br[i].init(src[start : start+length]) + if err != nil { + return nil, err + } + start += length + } + err := br[3].init(src[start:]) + if err != nil { + return nil, err + } + + // destination, offset to match first output + dstSize := cap(dst) + dst = dst[:dstSize] + out := dst + dstEvery := (dstSize + 3) / 4 + + const shift = 0 + const tlSize = 1 << 8 + const tlMask = tlSize - 1 + single := d.dt.single[:tlSize] + + // Use temp table to avoid bound checks/append penalty. + var buf [256]byte + var off uint8 + var decoded int + + // Decode 4 values from each decoder/loop. + const bufoff = 256 / 4 + for { + if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 { + break + } + + { + // Interleave 2 decodes. + const stream = 0 + const stream2 = 1 + br[stream].fillFast() + br[stream2].fillFast() + + v := single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 := single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+1] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+1] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+2] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+3] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+3] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + } + + { + const stream = 2 + const stream2 = 3 + br[stream].fillFast() + br[stream2].fillFast() + + v := single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 := single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+1] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+1] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+2] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+2] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + + v = single[br[stream].peekByteFast()>>shift].entry + buf[off+bufoff*stream+3] = uint8(v >> 8) + br[stream].advance(uint8(v)) + + v2 = single[br[stream2].peekByteFast()>>shift].entry + buf[off+bufoff*stream2+3] = uint8(v2 >> 8) + br[stream2].advance(uint8(v2)) + } + + off += 4 + + if off == bufoff { + if bufoff > dstEvery { + return nil, errors.New("corruption detected: stream overrun 1") + } + copy(out, buf[:bufoff]) + copy(out[dstEvery:], buf[bufoff:bufoff*2]) + copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3]) + copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4]) + off = 0 + out = out[bufoff:] + decoded += 256 + // There must at least be 3 buffers left. + if len(out) < dstEvery*3 { + return nil, errors.New("corruption detected: stream overrun 2") + } + } + } + if off > 0 { + ioff := int(off) + if len(out) < dstEvery*3+ioff { + return nil, errors.New("corruption detected: stream overrun 3") + } + copy(out, buf[:off]) + copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2]) + copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3]) + copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4]) + decoded += int(off) * 4 + out = out[off:] + } + + // Decode remaining. + for i := range br { + offset := dstEvery * i + br := &br[i] + bitsLeft := int(br.off*8) + int(64-br.bitsRead) + for bitsLeft > 0 { + if br.finished() { + return nil, io.ErrUnexpectedEOF + } + if br.bitsRead >= 56 { + if br.off >= 4 { + v := br.in[br.off-4:] + v = v[:4] + low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) + br.value |= uint64(low) << (br.bitsRead - 32) + br.bitsRead -= 32 + br.off -= 4 + } else { + for br.off > 0 { + br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8) + br.bitsRead -= 8 + br.off-- + } + } + } + // end inline... + if offset >= len(out) { + return nil, errors.New("corruption detected: stream overrun 4") + } + + // Read value and increment offset. + v := single[br.peekByteFast()>>shift].entry + nBits := uint8(v) + br.advance(nBits) + bitsLeft -= int(nBits) + out[offset] = uint8(v >> 8) + offset++ + } + decoded += offset - dstEvery*i + err = br.close() + if err != nil { + return nil, err + } + } + if dstSize != decoded { + return nil, errors.New("corruption detected: short output block") + } + return dst, nil +} + +// matches will compare a decoding table to a coding table. +// Errors are written to the writer. +// Nothing will be written if table is ok. +func (s *Scratch) matches(ct cTable, w io.Writer) { + if s == nil || len(s.dt.single) == 0 { + return + } + dt := s.dt.single[:1<>8) == byte(sym) { + fmt.Fprintf(w, "symbol %x has decoder, but no encoder\n", sym) + errs++ + break + } + } + if errs == 0 { + broken-- + } + continue + } + // Unused bits in input + ub := tablelog - enc.nBits + top := enc.val << ub + // decoder looks at top bits. + dec := dt[top] + if uint8(dec.entry) != enc.nBits { + fmt.Fprintf(w, "symbol 0x%x bit size mismatch (enc: %d, dec:%d).\n", sym, enc.nBits, uint8(dec.entry)) + errs++ + } + if uint8(dec.entry>>8) != uint8(sym) { + fmt.Fprintf(w, "symbol 0x%x decoder output mismatch (enc: %d, dec:%d).\n", sym, sym, uint8(dec.entry>>8)) + errs++ + } + if errs > 0 { + fmt.Fprintf(w, "%d errros in base, stopping\n", errs) + continue + } + // Ensure that all combinations are covered. + for i := uint16(0); i < (1 << ub); i++ { + vval := top | i + dec := dt[vval] + if uint8(dec.entry) != enc.nBits { + fmt.Fprintf(w, "symbol 0x%x bit size mismatch (enc: %d, dec:%d).\n", vval, enc.nBits, uint8(dec.entry)) + errs++ + } + if uint8(dec.entry>>8) != uint8(sym) { + fmt.Fprintf(w, "symbol 0x%x decoder output mismatch (enc: %d, dec:%d).\n", vval, sym, uint8(dec.entry>>8)) + errs++ + } + if errs > 20 { + fmt.Fprintf(w, "%d errros, stopping\n", errs) + break + } + } + if errs == 0 { + ok++ + broken-- + } + } + if broken > 0 { + fmt.Fprintf(w, "%d broken, %d ok\n", broken, ok) + } +} diff --git a/vendor/github.com/klauspost/compress/huff0/huff0.go b/vendor/github.com/klauspost/compress/huff0/huff0.go new file mode 100644 index 000000000000..7ec2022b6506 --- /dev/null +++ b/vendor/github.com/klauspost/compress/huff0/huff0.go @@ -0,0 +1,273 @@ +// Package huff0 provides fast huffman encoding as used in zstd. +// +// See README.md at https://github.com/klauspost/compress/tree/master/huff0 for details. +package huff0 + +import ( + "errors" + "fmt" + "math" + "math/bits" + + "github.com/klauspost/compress/fse" +) + +const ( + maxSymbolValue = 255 + + // zstandard limits tablelog to 11, see: + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#huffman-tree-description + tableLogMax = 11 + tableLogDefault = 11 + minTablelog = 5 + huffNodesLen = 512 + + // BlockSizeMax is maximum input size for a single block uncompressed. + BlockSizeMax = 1<<18 - 1 +) + +var ( + // ErrIncompressible is returned when input is judged to be too hard to compress. + ErrIncompressible = errors.New("input is not compressible") + + // ErrUseRLE is returned from the compressor when the input is a single byte value repeated. + ErrUseRLE = errors.New("input is single value repeated") + + // ErrTooBig is return if input is too large for a single block. + ErrTooBig = errors.New("input too big") + + // ErrMaxDecodedSizeExceeded is return if input is too large for a single block. + ErrMaxDecodedSizeExceeded = errors.New("maximum output size exceeded") +) + +type ReusePolicy uint8 + +const ( + // ReusePolicyAllow will allow reuse if it produces smaller output. + ReusePolicyAllow ReusePolicy = iota + + // ReusePolicyPrefer will re-use aggressively if possible. + // This will not check if a new table will produce smaller output, + // except if the current table is impossible to use or + // compressed output is bigger than input. + ReusePolicyPrefer + + // ReusePolicyNone will disable re-use of tables. + // This is slightly faster than ReusePolicyAllow but may produce larger output. + ReusePolicyNone + + // ReusePolicyMust must allow reuse and produce smaller output. + ReusePolicyMust +) + +type Scratch struct { + count [maxSymbolValue + 1]uint32 + + // Per block parameters. + // These can be used to override compression parameters of the block. + // Do not touch, unless you know what you are doing. + + // Out is output buffer. + // If the scratch is re-used before the caller is done processing the output, + // set this field to nil. + // Otherwise the output buffer will be re-used for next Compression/Decompression step + // and allocation will be avoided. + Out []byte + + // OutTable will contain the table data only, if a new table has been generated. + // Slice of the returned data. + OutTable []byte + + // OutData will contain the compressed data. + // Slice of the returned data. + OutData []byte + + // MaxDecodedSize will set the maximum allowed output size. + // This value will automatically be set to BlockSizeMax if not set. + // Decoders will return ErrMaxDecodedSizeExceeded is this limit is exceeded. + MaxDecodedSize int + + br byteReader + + // MaxSymbolValue will override the maximum symbol value of the next block. + MaxSymbolValue uint8 + + // TableLog will attempt to override the tablelog for the next block. + // Must be <= 11 and >= 5. + TableLog uint8 + + // Reuse will specify the reuse policy + Reuse ReusePolicy + + // WantLogLess allows to specify a log 2 reduction that should at least be achieved, + // otherwise the block will be returned as incompressible. + // The reduction should then at least be (input size >> WantLogLess) + // If WantLogLess == 0 any improvement will do. + WantLogLess uint8 + + symbolLen uint16 // Length of active part of the symbol table. + maxCount int // count of the most probable symbol + clearCount bool // clear count + actualTableLog uint8 // Selected tablelog. + prevTableLog uint8 // Tablelog for previous table + prevTable cTable // Table used for previous compression. + cTable cTable // compression table + dt dTable // decompression table + nodes []nodeElt + tmpOut [4][]byte + fse *fse.Scratch + huffWeight [maxSymbolValue + 1]byte +} + +// TransferCTable will transfer the previously used compression table. +func (s *Scratch) TransferCTable(src *Scratch) { + if cap(s.prevTable) < len(src.prevTable) { + s.prevTable = make(cTable, 0, maxSymbolValue+1) + } + s.prevTable = s.prevTable[:len(src.prevTable)] + copy(s.prevTable, src.prevTable) + s.prevTableLog = src.prevTableLog +} + +func (s *Scratch) prepare(in []byte) (*Scratch, error) { + if len(in) > BlockSizeMax { + return nil, ErrTooBig + } + if s == nil { + s = &Scratch{} + } + if s.MaxSymbolValue == 0 { + s.MaxSymbolValue = maxSymbolValue + } + if s.TableLog == 0 { + s.TableLog = tableLogDefault + } + if s.TableLog > tableLogMax || s.TableLog < minTablelog { + return nil, fmt.Errorf(" invalid tableLog %d (%d -> %d)", s.TableLog, minTablelog, tableLogMax) + } + if s.MaxDecodedSize <= 0 || s.MaxDecodedSize > BlockSizeMax { + s.MaxDecodedSize = BlockSizeMax + } + if s.clearCount && s.maxCount == 0 { + for i := range s.count { + s.count[i] = 0 + } + s.clearCount = false + } + if cap(s.Out) == 0 { + s.Out = make([]byte, 0, len(in)) + } + s.Out = s.Out[:0] + + s.OutTable = nil + s.OutData = nil + if cap(s.nodes) < huffNodesLen+1 { + s.nodes = make([]nodeElt, 0, huffNodesLen+1) + } + s.nodes = s.nodes[:0] + if s.fse == nil { + s.fse = &fse.Scratch{} + } + s.br.init(in) + + return s, nil +} + +type cTable []cTableEntry + +func (c cTable) write(s *Scratch) 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) { + s.Out = append(s.Out, uint8(len(b))) + s.Out = append(s.Out, b...) + return 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 ErrIncompressible + } + op := s.Out + // special case, pack weights 4 bits/weight. + op = append(op, 128|(maxSymbolValue-1)) + // be sure it doesn't cause msan issue in final combination + huffWeight[maxSymbolValue] = 0 + for n := uint16(0); n < uint16(maxSymbolValue); n += 2 { + op = append(op, (huffWeight[n]<<4)|huffWeight[n+1]) + } + s.Out = op + return nil +} + +// estimateSize returns the estimated size in bytes of the input represented in the +// histogram supplied. +func (c cTable) estimateSize(hist []uint32) int { + nbBits := uint32(7) + for i, v := range c[:len(hist)] { + nbBits += uint32(v.nBits) * hist[i] + } + return int(nbBits >> 3) +} + +// minSize returns the minimum possible size considering the shannon limit. +func (s *Scratch) minSize(total int) int { + nbBits := float64(7) + fTotal := float64(total) + for _, v := range s.count[:s.symbolLen] { + n := float64(v) + if n > 0 { + nbBits += math.Log2(fTotal/n) * n + } + } + return int(nbBits) >> 3 +} + +func highBit32(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} diff --git a/vendor/github.com/klauspost/compress/snappy/.gitignore b/vendor/github.com/klauspost/compress/snappy/.gitignore new file mode 100644 index 000000000000..042091d9b3b0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/.gitignore @@ -0,0 +1,16 @@ +cmd/snappytool/snappytool +testdata/bench + +# These explicitly listed benchmark data files are for an obsolete version of +# snappy_test.go. +testdata/alice29.txt +testdata/asyoulik.txt +testdata/fireworks.jpeg +testdata/geo.protodata +testdata/html +testdata/html_x_4 +testdata/kppkn.gtb +testdata/lcet10.txt +testdata/paper-100k.pdf +testdata/plrabn12.txt +testdata/urls.10K diff --git a/vendor/github.com/klauspost/compress/snappy/AUTHORS b/vendor/github.com/klauspost/compress/snappy/AUTHORS new file mode 100644 index 000000000000..bcfa19520af9 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/AUTHORS @@ -0,0 +1,15 @@ +# This is the official list of Snappy-Go authors for copyright purposes. +# This file is distinct from the CONTRIBUTORS files. +# See the latter for an explanation. + +# Names should be added to this file as +# Name or Organization +# The email address is not required for organizations. + +# Please keep the list sorted. + +Damian Gryski +Google Inc. +Jan Mercl <0xjnml@gmail.com> +Rodolfo Carvalho +Sebastien Binet diff --git a/vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS b/vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS new file mode 100644 index 000000000000..931ae31606f8 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/CONTRIBUTORS @@ -0,0 +1,37 @@ +# This is the official list of people who can contribute +# (and typically have contributed) code to the Snappy-Go repository. +# The AUTHORS file lists the copyright holders; this file +# lists people. For example, Google employees are listed here +# but not in AUTHORS, because Google holds the copyright. +# +# The submission process automatically checks to make sure +# that people submitting code are listed in this file (by email address). +# +# Names should be added to this file only after verifying that +# the individual or the individual's organization has agreed to +# the appropriate Contributor License Agreement, found here: +# +# http://code.google.com/legal/individual-cla-v1.0.html +# http://code.google.com/legal/corporate-cla-v1.0.html +# +# The agreement for individuals can be filled out on the web. +# +# When adding J Random Contributor's name to this file, +# either J's name or J's organization's name should be +# added to the AUTHORS file, depending on whether the +# individual or corporate CLA was used. + +# Names should be added to this file like so: +# Name + +# Please keep the list sorted. + +Damian Gryski +Jan Mercl <0xjnml@gmail.com> +Kai Backman +Marc-Antoine Ruel +Nigel Tao +Rob Pike +Rodolfo Carvalho +Russ Cox +Sebastien Binet diff --git a/vendor/github.com/klauspost/compress/snappy/LICENSE b/vendor/github.com/klauspost/compress/snappy/LICENSE new file mode 100644 index 000000000000..6050c10f4c8b --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/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/snappy/README b/vendor/github.com/klauspost/compress/snappy/README new file mode 100644 index 000000000000..cea12879a0ea --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/README @@ -0,0 +1,107 @@ +The Snappy compression format in the Go programming language. + +To download and install from source: +$ go get github.com/golang/snappy + +Unless otherwise noted, the Snappy-Go source files are distributed +under the BSD-style license found in the LICENSE file. + + + +Benchmarks. + +The golang/snappy benchmarks include compressing (Z) and decompressing (U) ten +or so files, the same set used by the C++ Snappy code (github.com/google/snappy +and note the "google", not "golang"). On an "Intel(R) Core(TM) i7-3770 CPU @ +3.40GHz", Go's GOARCH=amd64 numbers as of 2016-05-29: + +"go test -test.bench=." + +_UFlat0-8 2.19GB/s ± 0% html +_UFlat1-8 1.41GB/s ± 0% urls +_UFlat2-8 23.5GB/s ± 2% jpg +_UFlat3-8 1.91GB/s ± 0% jpg_200 +_UFlat4-8 14.0GB/s ± 1% pdf +_UFlat5-8 1.97GB/s ± 0% html4 +_UFlat6-8 814MB/s ± 0% txt1 +_UFlat7-8 785MB/s ± 0% txt2 +_UFlat8-8 857MB/s ± 0% txt3 +_UFlat9-8 719MB/s ± 1% txt4 +_UFlat10-8 2.84GB/s ± 0% pb +_UFlat11-8 1.05GB/s ± 0% gaviota + +_ZFlat0-8 1.04GB/s ± 0% html +_ZFlat1-8 534MB/s ± 0% urls +_ZFlat2-8 15.7GB/s ± 1% jpg +_ZFlat3-8 740MB/s ± 3% jpg_200 +_ZFlat4-8 9.20GB/s ± 1% pdf +_ZFlat5-8 991MB/s ± 0% html4 +_ZFlat6-8 379MB/s ± 0% txt1 +_ZFlat7-8 352MB/s ± 0% txt2 +_ZFlat8-8 396MB/s ± 1% txt3 +_ZFlat9-8 327MB/s ± 1% txt4 +_ZFlat10-8 1.33GB/s ± 1% pb +_ZFlat11-8 605MB/s ± 1% gaviota + + + +"go test -test.bench=. -tags=noasm" + +_UFlat0-8 621MB/s ± 2% html +_UFlat1-8 494MB/s ± 1% urls +_UFlat2-8 23.2GB/s ± 1% jpg +_UFlat3-8 1.12GB/s ± 1% jpg_200 +_UFlat4-8 4.35GB/s ± 1% pdf +_UFlat5-8 609MB/s ± 0% html4 +_UFlat6-8 296MB/s ± 0% txt1 +_UFlat7-8 288MB/s ± 0% txt2 +_UFlat8-8 309MB/s ± 1% txt3 +_UFlat9-8 280MB/s ± 1% txt4 +_UFlat10-8 753MB/s ± 0% pb +_UFlat11-8 400MB/s ± 0% gaviota + +_ZFlat0-8 409MB/s ± 1% html +_ZFlat1-8 250MB/s ± 1% urls +_ZFlat2-8 12.3GB/s ± 1% jpg +_ZFlat3-8 132MB/s ± 0% jpg_200 +_ZFlat4-8 2.92GB/s ± 0% pdf +_ZFlat5-8 405MB/s ± 1% html4 +_ZFlat6-8 179MB/s ± 1% txt1 +_ZFlat7-8 170MB/s ± 1% txt2 +_ZFlat8-8 189MB/s ± 1% txt3 +_ZFlat9-8 164MB/s ± 1% txt4 +_ZFlat10-8 479MB/s ± 1% pb +_ZFlat11-8 270MB/s ± 1% gaviota + + + +For comparison (Go's encoded output is byte-for-byte identical to C++'s), here +are the numbers from C++ Snappy's + +make CXXFLAGS="-O2 -DNDEBUG -g" clean snappy_unittest.log && cat snappy_unittest.log + +BM_UFlat/0 2.4GB/s html +BM_UFlat/1 1.4GB/s urls +BM_UFlat/2 21.8GB/s jpg +BM_UFlat/3 1.5GB/s jpg_200 +BM_UFlat/4 13.3GB/s pdf +BM_UFlat/5 2.1GB/s html4 +BM_UFlat/6 1.0GB/s txt1 +BM_UFlat/7 959.4MB/s txt2 +BM_UFlat/8 1.0GB/s txt3 +BM_UFlat/9 864.5MB/s txt4 +BM_UFlat/10 2.9GB/s pb +BM_UFlat/11 1.2GB/s gaviota + +BM_ZFlat/0 944.3MB/s html (22.31 %) +BM_ZFlat/1 501.6MB/s urls (47.78 %) +BM_ZFlat/2 14.3GB/s jpg (99.95 %) +BM_ZFlat/3 538.3MB/s jpg_200 (73.00 %) +BM_ZFlat/4 8.3GB/s pdf (83.30 %) +BM_ZFlat/5 903.5MB/s html4 (22.52 %) +BM_ZFlat/6 336.0MB/s txt1 (57.88 %) +BM_ZFlat/7 312.3MB/s txt2 (61.91 %) +BM_ZFlat/8 353.1MB/s txt3 (54.99 %) +BM_ZFlat/9 289.9MB/s txt4 (66.26 %) +BM_ZFlat/10 1.2GB/s pb (19.68 %) +BM_ZFlat/11 527.4MB/s gaviota (37.72 %) diff --git a/vendor/github.com/klauspost/compress/snappy/decode.go b/vendor/github.com/klauspost/compress/snappy/decode.go new file mode 100644 index 000000000000..72efb0353ddf --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/decode.go @@ -0,0 +1,237 @@ +// 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 snappy + +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. +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. +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 +} + +// Read satisfies the io.Reader interface. +func (r *Reader) Read(p []byte) (int, error) { + if r.err != nil { + return 0, r.err + } + for { + if r.i < r.j { + n := copy(p, r.decoded[r.i:r.j]) + r.i += n + return n, nil + } + if !r.readFull(r.buf[:4], true) { + return 0, r.err + } + chunkType := r.buf[0] + if !r.readHeader { + if chunkType != chunkTypeStreamIdentifier { + r.err = ErrCorrupt + return 0, 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 0, 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 0, r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return 0, 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 0, r.err + } + if n > len(r.decoded) { + r.err = ErrCorrupt + return 0, r.err + } + if _, err := Decode(r.decoded, buf); err != nil { + r.err = err + return 0, r.err + } + if crc(r.decoded[:n]) != checksum { + r.err = ErrCorrupt + return 0, 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 0, r.err + } + buf := r.buf[:checksumSize] + if !r.readFull(buf, false) { + return 0, 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 0, r.err + } + if !r.readFull(r.decoded[:n], false) { + return 0, r.err + } + if crc(r.decoded[:n]) != checksum { + r.err = ErrCorrupt + return 0, 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 0, r.err + } + if !r.readFull(r.buf[:len(magicBody)], false) { + return 0, r.err + } + for i := 0; i < len(magicBody); i++ { + if r.buf[i] != magicBody[i] { + r.err = ErrCorrupt + return 0, r.err + } + } + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + r.err = ErrUnsupported + return 0, 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 0, r.err + } + } +} diff --git a/vendor/github.com/klauspost/compress/snappy/decode_amd64.go b/vendor/github.com/klauspost/compress/snappy/decode_amd64.go new file mode 100644 index 000000000000..fcd192b849ed --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/decode_amd64.go @@ -0,0 +1,14 @@ +// 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. + +// +build !appengine +// +build gc +// +build !noasm + +package snappy + +// decode has the same semantics as in decode_other.go. +// +//go:noescape +func decode(dst, src []byte) int diff --git a/vendor/github.com/klauspost/compress/snappy/decode_amd64.s b/vendor/github.com/klauspost/compress/snappy/decode_amd64.s new file mode 100644 index 000000000000..1c66e37234dc --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/decode_amd64.s @@ -0,0 +1,482 @@ +// Copyright 2016 The 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. + +// +build !appengine +// +build gc +// +build !noasm + +#include "textflag.h" + +// The asm code generally follows the pure Go code in decode_other.go, except +// where marked with a "!!!". + +// func decode(dst, src []byte) int +// +// All local variables fit into registers. The non-zero stack size is only to +// spill registers and push args when issuing a CALL. The register allocation: +// - AX scratch +// - BX scratch +// - CX length or x +// - DX offset +// - SI &src[s] +// - DI &dst[d] +// + R8 dst_base +// + R9 dst_len +// + R10 dst_base + dst_len +// + R11 src_base +// + R12 src_len +// + R13 src_base + src_len +// - R14 used by doCopy +// - R15 used by doCopy +// +// The registers R8-R13 (marked with a "+") are set at the start of the +// function, and after a CALL returns, and are not otherwise modified. +// +// The d variable is implicitly DI - R8, and len(dst)-d is R10 - DI. +// The s variable is implicitly SI - R11, and len(src)-s is R13 - SI. +TEXT ·decode(SB), NOSPLIT, $48-56 + // Initialize SI, DI and R8-R13. + MOVQ dst_base+0(FP), R8 + MOVQ dst_len+8(FP), R9 + MOVQ R8, DI + MOVQ R8, R10 + ADDQ R9, R10 + MOVQ src_base+24(FP), R11 + MOVQ src_len+32(FP), R12 + MOVQ R11, SI + MOVQ R11, R13 + ADDQ R12, R13 + +loop: + // for s < len(src) + CMPQ SI, R13 + JEQ end + + // CX = uint32(src[s]) + // + // switch src[s] & 0x03 + MOVBLZX (SI), CX + MOVL CX, BX + ANDL $3, BX + CMPL BX, $1 + JAE tagCopy + + // ---------------------------------------- + // The code below handles literal tags. + + // case tagLiteral: + // x := uint32(src[s] >> 2) + // switch + SHRL $2, CX + CMPL CX, $60 + JAE tagLit60Plus + + // case x < 60: + // s++ + INCQ SI + +doLit: + // This is the end of the inner "switch", when we have a literal tag. + // + // We assume that CX == x and x fits in a uint32, where x is the variable + // used in the pure Go decode_other.go code. + + // length = int(x) + 1 + // + // Unlike the pure Go code, we don't need to check if length <= 0 because + // CX can hold 64 bits, so the increment cannot overflow. + INCQ CX + + // Prepare to check if copying length bytes will run past the end of dst or + // src. + // + // AX = len(dst) - d + // BX = len(src) - s + MOVQ R10, AX + SUBQ DI, AX + MOVQ R13, BX + SUBQ SI, BX + + // !!! Try a faster technique for short (16 or fewer bytes) copies. + // + // if length > 16 || len(dst)-d < 16 || len(src)-s < 16 { + // goto callMemmove // Fall back on calling runtime·memmove. + // } + // + // The C++ snappy code calls this TryFastAppend. It also checks len(src)-s + // against 21 instead of 16, because it cannot assume that all of its input + // is contiguous in memory and so it needs to leave enough source bytes to + // read the next tag without refilling buffers, but Go's Decode assumes + // contiguousness (the src argument is a []byte). + CMPQ CX, $16 + JGT callMemmove + CMPQ AX, $16 + JLT callMemmove + CMPQ BX, $16 + JLT callMemmove + + // !!! Implement the copy from src to dst as a 16-byte load and store. + // (Decode's documentation says that dst and src must not overlap.) + // + // This always copies 16 bytes, instead of only length bytes, but that's + // OK. If the input is a valid Snappy encoding then subsequent iterations + // will fix up the overrun. Otherwise, Decode returns a nil []byte (and a + // non-nil error), so the overrun will be ignored. + // + // Note that on amd64, it is legal and cheap to issue unaligned 8-byte or + // 16-byte loads and stores. This technique probably wouldn't be as + // effective on architectures that are fussier about alignment. + MOVOU 0(SI), X0 + MOVOU X0, 0(DI) + + // d += length + // s += length + ADDQ CX, DI + ADDQ CX, SI + JMP loop + +callMemmove: + // if length > len(dst)-d || length > len(src)-s { etc } + CMPQ CX, AX + JGT errCorrupt + CMPQ CX, BX + JGT errCorrupt + + // copy(dst[d:], src[s:s+length]) + // + // This means calling runtime·memmove(&dst[d], &src[s], length), so we push + // DI, SI and CX as arguments. Coincidentally, we also need to spill those + // three registers to the stack, to save local variables across the CALL. + MOVQ DI, 0(SP) + MOVQ SI, 8(SP) + MOVQ CX, 16(SP) + MOVQ DI, 24(SP) + MOVQ SI, 32(SP) + MOVQ CX, 40(SP) + CALL runtime·memmove(SB) + + // Restore local variables: unspill registers from the stack and + // re-calculate R8-R13. + MOVQ 24(SP), DI + MOVQ 32(SP), SI + MOVQ 40(SP), CX + MOVQ dst_base+0(FP), R8 + MOVQ dst_len+8(FP), R9 + MOVQ R8, R10 + ADDQ R9, R10 + MOVQ src_base+24(FP), R11 + MOVQ src_len+32(FP), R12 + MOVQ R11, R13 + ADDQ R12, R13 + + // d += length + // s += length + ADDQ CX, DI + ADDQ CX, SI + JMP loop + +tagLit60Plus: + // !!! This fragment does the + // + // s += x - 58; if uint(s) > uint(len(src)) { etc } + // + // checks. In the asm version, we code it once instead of once per switch case. + ADDQ CX, SI + SUBQ $58, SI + CMPQ SI, R13 + JA errCorrupt + + // case x == 60: + CMPL CX, $61 + JEQ tagLit61 + JA tagLit62Plus + + // x = uint32(src[s-1]) + MOVBLZX -1(SI), CX + JMP doLit + +tagLit61: + // case x == 61: + // x = uint32(src[s-2]) | uint32(src[s-1])<<8 + MOVWLZX -2(SI), CX + JMP doLit + +tagLit62Plus: + CMPL CX, $62 + JA tagLit63 + + // case x == 62: + // x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + MOVWLZX -3(SI), CX + MOVBLZX -1(SI), BX + SHLL $16, BX + ORL BX, CX + JMP doLit + +tagLit63: + // case x == 63: + // x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + MOVL -4(SI), CX + JMP doLit + +// The code above handles literal tags. +// ---------------------------------------- +// The code below handles copy tags. + +tagCopy4: + // case tagCopy4: + // s += 5 + ADDQ $5, SI + + // if uint(s) > uint(len(src)) { etc } + CMPQ SI, R13 + JA errCorrupt + + // length = 1 + int(src[s-5])>>2 + SHRQ $2, CX + INCQ CX + + // offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + MOVLQZX -4(SI), DX + JMP doCopy + +tagCopy2: + // case tagCopy2: + // s += 3 + ADDQ $3, SI + + // if uint(s) > uint(len(src)) { etc } + CMPQ SI, R13 + JA errCorrupt + + // length = 1 + int(src[s-3])>>2 + SHRQ $2, CX + INCQ CX + + // offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + MOVWQZX -2(SI), DX + JMP doCopy + +tagCopy: + // We have a copy tag. We assume that: + // - BX == src[s] & 0x03 + // - CX == src[s] + CMPQ BX, $2 + JEQ tagCopy2 + JA tagCopy4 + + // case tagCopy1: + // s += 2 + ADDQ $2, SI + + // if uint(s) > uint(len(src)) { etc } + CMPQ SI, R13 + JA errCorrupt + + // offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + MOVQ CX, DX + ANDQ $0xe0, DX + SHLQ $3, DX + MOVBQZX -1(SI), BX + ORQ BX, DX + + // length = 4 + int(src[s-2])>>2&0x7 + SHRQ $2, CX + ANDQ $7, CX + ADDQ $4, CX + +doCopy: + // This is the end of the outer "switch", when we have a copy tag. + // + // We assume that: + // - CX == length && CX > 0 + // - DX == offset + + // if offset <= 0 { etc } + CMPQ DX, $0 + JLE errCorrupt + + // if d < offset { etc } + MOVQ DI, BX + SUBQ R8, BX + CMPQ BX, DX + JLT errCorrupt + + // if length > len(dst)-d { etc } + MOVQ R10, BX + SUBQ DI, BX + CMPQ CX, BX + JGT errCorrupt + + // forwardCopy(dst[d:d+length], dst[d-offset:]); d += length + // + // Set: + // - R14 = len(dst)-d + // - R15 = &dst[d-offset] + MOVQ R10, R14 + SUBQ DI, R14 + MOVQ DI, R15 + SUBQ DX, R15 + + // !!! Try a faster technique for short (16 or fewer bytes) forward copies. + // + // First, try using two 8-byte load/stores, similar to the doLit technique + // above. Even if dst[d:d+length] and dst[d-offset:] can overlap, this is + // still OK if offset >= 8. Note that this has to be two 8-byte load/stores + // and not one 16-byte load/store, and the first store has to be before the + // second load, due to the overlap if offset is in the range [8, 16). + // + // if length > 16 || offset < 8 || len(dst)-d < 16 { + // goto slowForwardCopy + // } + // copy 16 bytes + // d += length + CMPQ CX, $16 + JGT slowForwardCopy + CMPQ DX, $8 + JLT slowForwardCopy + CMPQ R14, $16 + JLT slowForwardCopy + MOVQ 0(R15), AX + MOVQ AX, 0(DI) + MOVQ 8(R15), BX + MOVQ BX, 8(DI) + ADDQ CX, DI + JMP loop + +slowForwardCopy: + // !!! If the forward copy is longer than 16 bytes, or if offset < 8, we + // can still try 8-byte load stores, provided we can overrun up to 10 extra + // bytes. As above, the overrun will be fixed up by subsequent iterations + // of the outermost loop. + // + // The C++ snappy code calls this technique IncrementalCopyFastPath. Its + // commentary says: + // + // ---- + // + // The main part of this loop is a simple copy of eight bytes at a time + // until we've copied (at least) the requested amount of bytes. However, + // if d and d-offset are less than eight bytes apart (indicating a + // repeating pattern of length < 8), we first need to expand the pattern in + // order to get the correct results. For instance, if the buffer looks like + // this, with the eight-byte and patterns marked as + // intervals: + // + // abxxxxxxxxxxxx + // [------] d-offset + // [------] d + // + // a single eight-byte copy from to will repeat the pattern + // once, after which we can move two bytes without moving : + // + // ababxxxxxxxxxx + // [------] d-offset + // [------] d + // + // and repeat the exercise until the two no longer overlap. + // + // This allows us to do very well in the special case of one single byte + // repeated many times, without taking a big hit for more general cases. + // + // The worst case of extra writing past the end of the match occurs when + // offset == 1 and length == 1; the last copy will read from byte positions + // [0..7] and write to [4..11], whereas it was only supposed to write to + // position 1. Thus, ten excess bytes. + // + // ---- + // + // That "10 byte overrun" worst case is confirmed by Go's + // TestSlowForwardCopyOverrun, which also tests the fixUpSlowForwardCopy + // and finishSlowForwardCopy algorithm. + // + // if length > len(dst)-d-10 { + // goto verySlowForwardCopy + // } + SUBQ $10, R14 + CMPQ CX, R14 + JGT verySlowForwardCopy + +makeOffsetAtLeast8: + // !!! As above, expand the pattern so that offset >= 8 and we can use + // 8-byte load/stores. + // + // for offset < 8 { + // copy 8 bytes from dst[d-offset:] to dst[d:] + // length -= offset + // d += offset + // offset += offset + // // The two previous lines together means that d-offset, and therefore + // // R15, is unchanged. + // } + CMPQ DX, $8 + JGE fixUpSlowForwardCopy + MOVQ (R15), BX + MOVQ BX, (DI) + SUBQ DX, CX + ADDQ DX, DI + ADDQ DX, DX + JMP makeOffsetAtLeast8 + +fixUpSlowForwardCopy: + // !!! Add length (which might be negative now) to d (implied by DI being + // &dst[d]) so that d ends up at the right place when we jump back to the + // top of the loop. Before we do that, though, we save DI to AX so that, if + // length is positive, copying the remaining length bytes will write to the + // right place. + MOVQ DI, AX + ADDQ CX, DI + +finishSlowForwardCopy: + // !!! Repeat 8-byte load/stores until length <= 0. Ending with a negative + // length means that we overrun, but as above, that will be fixed up by + // subsequent iterations of the outermost loop. + CMPQ CX, $0 + JLE loop + MOVQ (R15), BX + MOVQ BX, (AX) + ADDQ $8, R15 + ADDQ $8, AX + SUBQ $8, CX + JMP finishSlowForwardCopy + +verySlowForwardCopy: + // verySlowForwardCopy is a simple implementation of forward copy. In C + // parlance, this is a do/while loop instead of a while loop, since we know + // that length > 0. In Go syntax: + // + // for { + // dst[d] = dst[d - offset] + // d++ + // length-- + // if length == 0 { + // break + // } + // } + MOVB (R15), BX + MOVB BX, (DI) + INCQ R15 + INCQ DI + DECQ CX + JNZ verySlowForwardCopy + JMP loop + +// The code above handles copy tags. +// ---------------------------------------- + +end: + // This is the end of the "for s < len(src)". + // + // if d != len(dst) { etc } + CMPQ DI, R10 + JNE errCorrupt + + // return 0 + MOVQ $0, ret+48(FP) + RET + +errCorrupt: + // return decodeErrCodeCorrupt + MOVQ $1, ret+48(FP) + RET diff --git a/vendor/github.com/klauspost/compress/snappy/decode_other.go b/vendor/github.com/klauspost/compress/snappy/decode_other.go new file mode 100644 index 000000000000..94a96c5d7b85 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/decode_other.go @@ -0,0 +1,115 @@ +// 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. + +// +build !amd64 appengine !gc noasm + +package snappy + +// 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/snappy/encode.go b/vendor/github.com/klauspost/compress/snappy/encode.go new file mode 100644 index 000000000000..8d393e904bb3 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/encode.go @@ -0,0 +1,285 @@ +// 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 snappy + +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. +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. +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/snappy/encode_amd64.go b/vendor/github.com/klauspost/compress/snappy/encode_amd64.go new file mode 100644 index 000000000000..150d91bc8be5 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/encode_amd64.go @@ -0,0 +1,29 @@ +// 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. + +// +build !appengine +// +build gc +// +build !noasm + +package snappy + +// emitLiteral has the same semantics as in encode_other.go. +// +//go:noescape +func emitLiteral(dst, lit []byte) int + +// emitCopy has the same semantics as in encode_other.go. +// +//go:noescape +func emitCopy(dst []byte, offset, length int) int + +// extendMatch has the same semantics as in encode_other.go. +// +//go:noescape +func extendMatch(src []byte, i, j int) int + +// encodeBlock has the same semantics as in encode_other.go. +// +//go:noescape +func encodeBlock(dst, src []byte) (d int) diff --git a/vendor/github.com/klauspost/compress/snappy/encode_amd64.s b/vendor/github.com/klauspost/compress/snappy/encode_amd64.s new file mode 100644 index 000000000000..adfd979fe277 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/encode_amd64.s @@ -0,0 +1,730 @@ +// Copyright 2016 The 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. + +// +build !appengine +// +build gc +// +build !noasm + +#include "textflag.h" + +// The XXX lines assemble on Go 1.4, 1.5 and 1.7, but not 1.6, due to a +// Go toolchain regression. See https://github.com/golang/go/issues/15426 and +// https://github.com/golang/snappy/issues/29 +// +// As a workaround, the package was built with a known good assembler, and +// those instructions were disassembled by "objdump -d" to yield the +// 4e 0f b7 7c 5c 78 movzwq 0x78(%rsp,%r11,2),%r15 +// style comments, in AT&T asm syntax. Note that rsp here is a physical +// register, not Go/asm's SP pseudo-register (see https://golang.org/doc/asm). +// The instructions were then encoded as "BYTE $0x.." sequences, which assemble +// fine on Go 1.6. + +// The asm code generally follows the pure Go code in encode_other.go, except +// where marked with a "!!!". + +// ---------------------------------------------------------------------------- + +// func emitLiteral(dst, lit []byte) int +// +// All local variables fit into registers. The register allocation: +// - AX len(lit) +// - BX n +// - DX return value +// - DI &dst[i] +// - R10 &lit[0] +// +// The 24 bytes of stack space is to call runtime·memmove. +// +// The unusual register allocation of local variables, such as R10 for the +// source pointer, matches the allocation used at the call site in encodeBlock, +// which makes it easier to manually inline this function. +TEXT ·emitLiteral(SB), NOSPLIT, $24-56 + MOVQ dst_base+0(FP), DI + MOVQ lit_base+24(FP), R10 + MOVQ lit_len+32(FP), AX + MOVQ AX, DX + MOVL AX, BX + SUBL $1, BX + + CMPL BX, $60 + JLT oneByte + CMPL BX, $256 + JLT twoBytes + +threeBytes: + MOVB $0xf4, 0(DI) + MOVW BX, 1(DI) + ADDQ $3, DI + ADDQ $3, DX + JMP memmove + +twoBytes: + MOVB $0xf0, 0(DI) + MOVB BX, 1(DI) + ADDQ $2, DI + ADDQ $2, DX + JMP memmove + +oneByte: + SHLB $2, BX + MOVB BX, 0(DI) + ADDQ $1, DI + ADDQ $1, DX + +memmove: + MOVQ DX, ret+48(FP) + + // copy(dst[i:], lit) + // + // This means calling runtime·memmove(&dst[i], &lit[0], len(lit)), so we push + // DI, R10 and AX as arguments. + MOVQ DI, 0(SP) + MOVQ R10, 8(SP) + MOVQ AX, 16(SP) + CALL runtime·memmove(SB) + RET + +// ---------------------------------------------------------------------------- + +// func emitCopy(dst []byte, offset, length int) int +// +// All local variables fit into registers. The register allocation: +// - AX length +// - SI &dst[0] +// - DI &dst[i] +// - R11 offset +// +// The unusual register allocation of local variables, such as R11 for the +// offset, matches the allocation used at the call site in encodeBlock, which +// makes it easier to manually inline this function. +TEXT ·emitCopy(SB), NOSPLIT, $0-48 + MOVQ dst_base+0(FP), DI + MOVQ DI, SI + MOVQ offset+24(FP), R11 + MOVQ length+32(FP), AX + +loop0: + // for length >= 68 { etc } + CMPL AX, $68 + JLT step1 + + // Emit a length 64 copy, encoded as 3 bytes. + MOVB $0xfe, 0(DI) + MOVW R11, 1(DI) + ADDQ $3, DI + SUBL $64, AX + JMP loop0 + +step1: + // if length > 64 { etc } + CMPL AX, $64 + JLE step2 + + // Emit a length 60 copy, encoded as 3 bytes. + MOVB $0xee, 0(DI) + MOVW R11, 1(DI) + ADDQ $3, DI + SUBL $60, AX + +step2: + // if length >= 12 || offset >= 2048 { goto step3 } + CMPL AX, $12 + JGE step3 + CMPL R11, $2048 + JGE step3 + + // Emit the remaining copy, encoded as 2 bytes. + MOVB R11, 1(DI) + SHRL $8, R11 + SHLB $5, R11 + SUBB $4, AX + SHLB $2, AX + ORB AX, R11 + ORB $1, R11 + MOVB R11, 0(DI) + ADDQ $2, DI + + // Return the number of bytes written. + SUBQ SI, DI + MOVQ DI, ret+40(FP) + RET + +step3: + // Emit the remaining copy, encoded as 3 bytes. + SUBL $1, AX + SHLB $2, AX + ORB $2, AX + MOVB AX, 0(DI) + MOVW R11, 1(DI) + ADDQ $3, DI + + // Return the number of bytes written. + SUBQ SI, DI + MOVQ DI, ret+40(FP) + RET + +// ---------------------------------------------------------------------------- + +// func extendMatch(src []byte, i, j int) int +// +// All local variables fit into registers. The register allocation: +// - DX &src[0] +// - SI &src[j] +// - R13 &src[len(src) - 8] +// - R14 &src[len(src)] +// - R15 &src[i] +// +// The unusual register allocation of local variables, such as R15 for a source +// pointer, matches the allocation used at the call site in encodeBlock, which +// makes it easier to manually inline this function. +TEXT ·extendMatch(SB), NOSPLIT, $0-48 + MOVQ src_base+0(FP), DX + MOVQ src_len+8(FP), R14 + MOVQ i+24(FP), R15 + MOVQ j+32(FP), SI + ADDQ DX, R14 + ADDQ DX, R15 + ADDQ DX, SI + MOVQ R14, R13 + SUBQ $8, R13 + +cmp8: + // As long as we are 8 or more bytes before the end of src, we can load and + // compare 8 bytes at a time. If those 8 bytes are equal, repeat. + CMPQ SI, R13 + JA cmp1 + MOVQ (R15), AX + MOVQ (SI), BX + CMPQ AX, BX + JNE bsf + ADDQ $8, R15 + ADDQ $8, SI + JMP cmp8 + +bsf: + // If those 8 bytes were not equal, XOR the two 8 byte values, and return + // the index of the first byte that differs. The BSF instruction finds the + // least significant 1 bit, the amd64 architecture is little-endian, and + // the shift by 3 converts a bit index to a byte index. + XORQ AX, BX + BSFQ BX, BX + SHRQ $3, BX + ADDQ BX, SI + + // Convert from &src[ret] to ret. + SUBQ DX, SI + MOVQ SI, ret+40(FP) + RET + +cmp1: + // In src's tail, compare 1 byte at a time. + CMPQ SI, R14 + JAE extendMatchEnd + MOVB (R15), AX + MOVB (SI), BX + CMPB AX, BX + JNE extendMatchEnd + ADDQ $1, R15 + ADDQ $1, SI + JMP cmp1 + +extendMatchEnd: + // Convert from &src[ret] to ret. + SUBQ DX, SI + MOVQ SI, ret+40(FP) + RET + +// ---------------------------------------------------------------------------- + +// func encodeBlock(dst, src []byte) (d int) +// +// All local variables fit into registers, other than "var table". The register +// allocation: +// - AX . . +// - BX . . +// - CX 56 shift (note that amd64 shifts by non-immediates must use CX). +// - DX 64 &src[0], tableSize +// - SI 72 &src[s] +// - DI 80 &dst[d] +// - R9 88 sLimit +// - R10 . &src[nextEmit] +// - R11 96 prevHash, currHash, nextHash, offset +// - R12 104 &src[base], skip +// - R13 . &src[nextS], &src[len(src) - 8] +// - R14 . len(src), bytesBetweenHashLookups, &src[len(src)], x +// - R15 112 candidate +// +// The second column (56, 64, etc) is the stack offset to spill the registers +// when calling other functions. We could pack this slightly tighter, but it's +// simpler to have a dedicated spill map independent of the function called. +// +// "var table [maxTableSize]uint16" takes up 32768 bytes of stack space. An +// extra 56 bytes, to call other functions, and an extra 64 bytes, to spill +// local variables (registers) during calls gives 32768 + 56 + 64 = 32888. +TEXT ·encodeBlock(SB), 0, $32888-56 + MOVQ dst_base+0(FP), DI + MOVQ src_base+24(FP), SI + MOVQ src_len+32(FP), R14 + + // shift, tableSize := uint32(32-8), 1<<8 + MOVQ $24, CX + MOVQ $256, DX + +calcShift: + // for ; tableSize < maxTableSize && tableSize < len(src); tableSize *= 2 { + // shift-- + // } + CMPQ DX, $16384 + JGE varTable + CMPQ DX, R14 + JGE varTable + SUBQ $1, CX + SHLQ $1, DX + JMP calcShift + +varTable: + // var table [maxTableSize]uint16 + // + // In the asm code, unlike the Go code, we can zero-initialize only the + // first tableSize elements. Each uint16 element is 2 bytes and each MOVOU + // writes 16 bytes, so we can do only tableSize/8 writes instead of the + // 2048 writes that would zero-initialize all of table's 32768 bytes. + SHRQ $3, DX + LEAQ table-32768(SP), BX + PXOR X0, X0 + +memclr: + MOVOU X0, 0(BX) + ADDQ $16, BX + SUBQ $1, DX + JNZ memclr + + // !!! DX = &src[0] + MOVQ SI, DX + + // sLimit := len(src) - inputMargin + MOVQ R14, R9 + SUBQ $15, R9 + + // !!! Pre-emptively spill CX, DX and R9 to the stack. Their values don't + // change for the rest of the function. + MOVQ CX, 56(SP) + MOVQ DX, 64(SP) + MOVQ R9, 88(SP) + + // nextEmit := 0 + MOVQ DX, R10 + + // s := 1 + ADDQ $1, SI + + // nextHash := hash(load32(src, s), shift) + MOVL 0(SI), R11 + IMULL $0x1e35a7bd, R11 + SHRL CX, R11 + +outer: + // for { etc } + + // skip := 32 + MOVQ $32, R12 + + // nextS := s + MOVQ SI, R13 + + // candidate := 0 + MOVQ $0, R15 + +inner0: + // for { etc } + + // s := nextS + MOVQ R13, SI + + // bytesBetweenHashLookups := skip >> 5 + MOVQ R12, R14 + SHRQ $5, R14 + + // nextS = s + bytesBetweenHashLookups + ADDQ R14, R13 + + // skip += bytesBetweenHashLookups + ADDQ R14, R12 + + // if nextS > sLimit { goto emitRemainder } + MOVQ R13, AX + SUBQ DX, AX + CMPQ AX, R9 + JA emitRemainder + + // candidate = int(table[nextHash]) + // XXX: MOVWQZX table-32768(SP)(R11*2), R15 + // XXX: 4e 0f b7 7c 5c 78 movzwq 0x78(%rsp,%r11,2),%r15 + BYTE $0x4e + BYTE $0x0f + BYTE $0xb7 + BYTE $0x7c + BYTE $0x5c + BYTE $0x78 + + // table[nextHash] = uint16(s) + MOVQ SI, AX + SUBQ DX, AX + + // XXX: MOVW AX, table-32768(SP)(R11*2) + // XXX: 66 42 89 44 5c 78 mov %ax,0x78(%rsp,%r11,2) + BYTE $0x66 + BYTE $0x42 + BYTE $0x89 + BYTE $0x44 + BYTE $0x5c + BYTE $0x78 + + // nextHash = hash(load32(src, nextS), shift) + MOVL 0(R13), R11 + IMULL $0x1e35a7bd, R11 + SHRL CX, R11 + + // if load32(src, s) != load32(src, candidate) { continue } break + MOVL 0(SI), AX + MOVL (DX)(R15*1), BX + CMPL AX, BX + JNE inner0 + +fourByteMatch: + // As per the encode_other.go code: + // + // A 4-byte match has been found. We'll later see etc. + + // !!! Jump to a fast path for short (<= 16 byte) literals. See the comment + // on inputMargin in encode.go. + MOVQ SI, AX + SUBQ R10, AX + CMPQ AX, $16 + JLE emitLiteralFastPath + + // ---------------------------------------- + // Begin inline of the emitLiteral call. + // + // d += emitLiteral(dst[d:], src[nextEmit:s]) + + MOVL AX, BX + SUBL $1, BX + + CMPL BX, $60 + JLT inlineEmitLiteralOneByte + CMPL BX, $256 + JLT inlineEmitLiteralTwoBytes + +inlineEmitLiteralThreeBytes: + MOVB $0xf4, 0(DI) + MOVW BX, 1(DI) + ADDQ $3, DI + JMP inlineEmitLiteralMemmove + +inlineEmitLiteralTwoBytes: + MOVB $0xf0, 0(DI) + MOVB BX, 1(DI) + ADDQ $2, DI + JMP inlineEmitLiteralMemmove + +inlineEmitLiteralOneByte: + SHLB $2, BX + MOVB BX, 0(DI) + ADDQ $1, DI + +inlineEmitLiteralMemmove: + // Spill local variables (registers) onto the stack; call; unspill. + // + // copy(dst[i:], lit) + // + // This means calling runtime·memmove(&dst[i], &lit[0], len(lit)), so we push + // DI, R10 and AX as arguments. + MOVQ DI, 0(SP) + MOVQ R10, 8(SP) + MOVQ AX, 16(SP) + ADDQ AX, DI // Finish the "d +=" part of "d += emitLiteral(etc)". + MOVQ SI, 72(SP) + MOVQ DI, 80(SP) + MOVQ R15, 112(SP) + CALL runtime·memmove(SB) + MOVQ 56(SP), CX + MOVQ 64(SP), DX + MOVQ 72(SP), SI + MOVQ 80(SP), DI + MOVQ 88(SP), R9 + MOVQ 112(SP), R15 + JMP inner1 + +inlineEmitLiteralEnd: + // End inline of the emitLiteral call. + // ---------------------------------------- + +emitLiteralFastPath: + // !!! Emit the 1-byte encoding "uint8(len(lit)-1)<<2". + MOVB AX, BX + SUBB $1, BX + SHLB $2, BX + MOVB BX, (DI) + ADDQ $1, DI + + // !!! Implement the copy from lit to dst as a 16-byte load and store. + // (Encode's documentation says that dst and src must not overlap.) + // + // This always copies 16 bytes, instead of only len(lit) bytes, but that's + // OK. Subsequent iterations will fix up the overrun. + // + // Note that on amd64, it is legal and cheap to issue unaligned 8-byte or + // 16-byte loads and stores. This technique probably wouldn't be as + // effective on architectures that are fussier about alignment. + MOVOU 0(R10), X0 + MOVOU X0, 0(DI) + ADDQ AX, DI + +inner1: + // for { etc } + + // base := s + MOVQ SI, R12 + + // !!! offset := base - candidate + MOVQ R12, R11 + SUBQ R15, R11 + SUBQ DX, R11 + + // ---------------------------------------- + // Begin inline of the extendMatch call. + // + // s = extendMatch(src, candidate+4, s+4) + + // !!! R14 = &src[len(src)] + MOVQ src_len+32(FP), R14 + ADDQ DX, R14 + + // !!! R13 = &src[len(src) - 8] + MOVQ R14, R13 + SUBQ $8, R13 + + // !!! R15 = &src[candidate + 4] + ADDQ $4, R15 + ADDQ DX, R15 + + // !!! s += 4 + ADDQ $4, SI + +inlineExtendMatchCmp8: + // As long as we are 8 or more bytes before the end of src, we can load and + // compare 8 bytes at a time. If those 8 bytes are equal, repeat. + CMPQ SI, R13 + JA inlineExtendMatchCmp1 + MOVQ (R15), AX + MOVQ (SI), BX + CMPQ AX, BX + JNE inlineExtendMatchBSF + ADDQ $8, R15 + ADDQ $8, SI + JMP inlineExtendMatchCmp8 + +inlineExtendMatchBSF: + // If those 8 bytes were not equal, XOR the two 8 byte values, and return + // the index of the first byte that differs. The BSF instruction finds the + // least significant 1 bit, the amd64 architecture is little-endian, and + // the shift by 3 converts a bit index to a byte index. + XORQ AX, BX + BSFQ BX, BX + SHRQ $3, BX + ADDQ BX, SI + JMP inlineExtendMatchEnd + +inlineExtendMatchCmp1: + // In src's tail, compare 1 byte at a time. + CMPQ SI, R14 + JAE inlineExtendMatchEnd + MOVB (R15), AX + MOVB (SI), BX + CMPB AX, BX + JNE inlineExtendMatchEnd + ADDQ $1, R15 + ADDQ $1, SI + JMP inlineExtendMatchCmp1 + +inlineExtendMatchEnd: + // End inline of the extendMatch call. + // ---------------------------------------- + + // ---------------------------------------- + // Begin inline of the emitCopy call. + // + // d += emitCopy(dst[d:], base-candidate, s-base) + + // !!! length := s - base + MOVQ SI, AX + SUBQ R12, AX + +inlineEmitCopyLoop0: + // for length >= 68 { etc } + CMPL AX, $68 + JLT inlineEmitCopyStep1 + + // Emit a length 64 copy, encoded as 3 bytes. + MOVB $0xfe, 0(DI) + MOVW R11, 1(DI) + ADDQ $3, DI + SUBL $64, AX + JMP inlineEmitCopyLoop0 + +inlineEmitCopyStep1: + // if length > 64 { etc } + CMPL AX, $64 + JLE inlineEmitCopyStep2 + + // Emit a length 60 copy, encoded as 3 bytes. + MOVB $0xee, 0(DI) + MOVW R11, 1(DI) + ADDQ $3, DI + SUBL $60, AX + +inlineEmitCopyStep2: + // if length >= 12 || offset >= 2048 { goto inlineEmitCopyStep3 } + CMPL AX, $12 + JGE inlineEmitCopyStep3 + CMPL R11, $2048 + JGE inlineEmitCopyStep3 + + // Emit the remaining copy, encoded as 2 bytes. + MOVB R11, 1(DI) + SHRL $8, R11 + SHLB $5, R11 + SUBB $4, AX + SHLB $2, AX + ORB AX, R11 + ORB $1, R11 + MOVB R11, 0(DI) + ADDQ $2, DI + JMP inlineEmitCopyEnd + +inlineEmitCopyStep3: + // Emit the remaining copy, encoded as 3 bytes. + SUBL $1, AX + SHLB $2, AX + ORB $2, AX + MOVB AX, 0(DI) + MOVW R11, 1(DI) + ADDQ $3, DI + +inlineEmitCopyEnd: + // End inline of the emitCopy call. + // ---------------------------------------- + + // nextEmit = s + MOVQ SI, R10 + + // if s >= sLimit { goto emitRemainder } + MOVQ SI, AX + SUBQ DX, AX + CMPQ AX, R9 + JAE emitRemainder + + // As per the encode_other.go code: + // + // We could immediately etc. + + // x := load64(src, s-1) + MOVQ -1(SI), R14 + + // prevHash := hash(uint32(x>>0), shift) + MOVL R14, R11 + IMULL $0x1e35a7bd, R11 + SHRL CX, R11 + + // table[prevHash] = uint16(s-1) + MOVQ SI, AX + SUBQ DX, AX + SUBQ $1, AX + + // XXX: MOVW AX, table-32768(SP)(R11*2) + // XXX: 66 42 89 44 5c 78 mov %ax,0x78(%rsp,%r11,2) + BYTE $0x66 + BYTE $0x42 + BYTE $0x89 + BYTE $0x44 + BYTE $0x5c + BYTE $0x78 + + // currHash := hash(uint32(x>>8), shift) + SHRQ $8, R14 + MOVL R14, R11 + IMULL $0x1e35a7bd, R11 + SHRL CX, R11 + + // candidate = int(table[currHash]) + // XXX: MOVWQZX table-32768(SP)(R11*2), R15 + // XXX: 4e 0f b7 7c 5c 78 movzwq 0x78(%rsp,%r11,2),%r15 + BYTE $0x4e + BYTE $0x0f + BYTE $0xb7 + BYTE $0x7c + BYTE $0x5c + BYTE $0x78 + + // table[currHash] = uint16(s) + ADDQ $1, AX + + // XXX: MOVW AX, table-32768(SP)(R11*2) + // XXX: 66 42 89 44 5c 78 mov %ax,0x78(%rsp,%r11,2) + BYTE $0x66 + BYTE $0x42 + BYTE $0x89 + BYTE $0x44 + BYTE $0x5c + BYTE $0x78 + + // if uint32(x>>8) == load32(src, candidate) { continue } + MOVL (DX)(R15*1), BX + CMPL R14, BX + JEQ inner1 + + // nextHash = hash(uint32(x>>16), shift) + SHRQ $8, R14 + MOVL R14, R11 + IMULL $0x1e35a7bd, R11 + SHRL CX, R11 + + // s++ + ADDQ $1, SI + + // break out of the inner1 for loop, i.e. continue the outer loop. + JMP outer + +emitRemainder: + // if nextEmit < len(src) { etc } + MOVQ src_len+32(FP), AX + ADDQ DX, AX + CMPQ R10, AX + JEQ encodeBlockEnd + + // d += emitLiteral(dst[d:], src[nextEmit:]) + // + // Push args. + MOVQ DI, 0(SP) + MOVQ $0, 8(SP) // Unnecessary, as the callee ignores it, but conservative. + MOVQ $0, 16(SP) // Unnecessary, as the callee ignores it, but conservative. + MOVQ R10, 24(SP) + SUBQ R10, AX + MOVQ AX, 32(SP) + MOVQ AX, 40(SP) // Unnecessary, as the callee ignores it, but conservative. + + // Spill local variables (registers) onto the stack; call; unspill. + MOVQ DI, 80(SP) + CALL ·emitLiteral(SB) + MOVQ 80(SP), DI + + // Finish the "d +=" part of "d += emitLiteral(etc)". + ADDQ 48(SP), DI + +encodeBlockEnd: + MOVQ dst_base+0(FP), AX + SUBQ AX, DI + MOVQ DI, d+48(FP) + RET diff --git a/vendor/github.com/klauspost/compress/snappy/encode_other.go b/vendor/github.com/klauspost/compress/snappy/encode_other.go new file mode 100644 index 000000000000..dbcae905e6e0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/encode_other.go @@ -0,0 +1,238 @@ +// 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. + +// +build !amd64 appengine !gc noasm + +package snappy + +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/snappy/runbench.cmd b/vendor/github.com/klauspost/compress/snappy/runbench.cmd new file mode 100644 index 000000000000..d24eb4b47c37 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/runbench.cmd @@ -0,0 +1,2 @@ +del old.txt +go test -bench=. >>old.txt && go test -bench=. >>old.txt && go test -bench=. >>old.txt && benchstat -delta-test=ttest old.txt new.txt diff --git a/vendor/github.com/klauspost/compress/snappy/snappy.go b/vendor/github.com/klauspost/compress/snappy/snappy.go new file mode 100644 index 000000000000..74a36689e878 --- /dev/null +++ b/vendor/github.com/klauspost/compress/snappy/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 snappy 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 snappy + +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/zstd/README.md b/vendor/github.com/klauspost/compress/zstd/README.md new file mode 100644 index 000000000000..08e553f7530d --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/README.md @@ -0,0 +1,406 @@ +# zstd + +[Zstandard](https://facebook.github.io/zstd/) is a real-time compression algorithm, providing high compression ratios. +It offers a very wide range of compression / speed trade-off, while being backed by a very fast decoder. +A high performance compression algorithm is implemented. For now focused on speed. + +This package provides [compression](#Compressor) to and [decompression](#Decompressor) of Zstandard content. + +This package is pure Go and without use of "unsafe". + +The `zstd` package is provided as open source software using a Go standard license. + +Currently the package is heavily optimized for 64 bit processors and will be significantly slower on 32 bit processors. + +## Installation + +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 + + +## Compressor + +### Status: + +STABLE - there may always be subtle bugs, a wide variety of content has been tested and the library is actively +used by several projects. This library is being continuously [fuzz-tested](https://github.com/klauspost/compress-fuzz), +kindly supplied by [fuzzit.dev](https://fuzzit.dev/). + +There may still be specific combinations of data types/size/settings that could lead to edge cases, +so as always, testing is recommended. + +For now, a high speed (fastest) and medium-fast (default) compressor has been implemented. + +The "Fastest" compression ratio is roughly equivalent to zstd level 1. +The "Default" compression ratio is roughly equivalent to zstd level 3 (default). + +In terms of speed, it is typically 2x as fast as the stdlib deflate/gzip in its fastest mode. +The compression ratio compared to stdlib is around level 3, but usually 3x as fast. + +Compared to cgo zstd, the speed is around level 3 (default), but compression slightly worse, between level 1&2. + + +### Usage + +An Encoder can be used for either compressing a stream via the +`io.WriteCloser` interface supported by the Encoder or as multiple independent +tasks via the `EncodeAll` function. +Smaller encodes are encouraged to use the EncodeAll function. +Use `NewWriter` to create a new instance that can be used for both. + +To create a writer with default options, do like this: + +```Go +// Compress input to output. +func Compress(in io.Reader, out io.Writer) error { + enc, err := zstd.NewWriter(out) + if err != nil { + return err + } + _, err = io.Copy(enc, in) + if err != nil { + enc.Close() + return err + } + return enc.Close() +} +``` + +Now you can encode by writing data to `enc`. The output will be finished writing when `Close()` is called. +Even if your encode fails, you should still call `Close()` to release any resources that may be held up. + +The above is fine for big encodes. However, whenever possible try to *reuse* the writer. + +To reuse the encoder, you can use the `Reset(io.Writer)` function to change to another output. +This will allow the encoder to reuse all resources and avoid wasteful allocations. + +Currently stream encoding has 'light' concurrency, meaning up to 2 goroutines can be working on part +of a stream. This is independent of the `WithEncoderConcurrency(n)`, but that is likely to change +in the future. So if you want to limit concurrency for future updates, specify the concurrency +you would like. + +You can specify your desired compression level using `WithEncoderLevel()` option. Currently only pre-defined +compression settings can be specified. + +#### Future Compatibility Guarantees + +This will be an evolving project. When using this package it is important to note that both the compression efficiency and speed may change. + +The goal will be to keep the default efficiency at the default zstd (level 3). +However the encoding should never be assumed to remain the same, +and you should not use hashes of compressed output for similarity checks. + +The Encoder can be assumed to produce the same output from the exact same code version. +However, the may be modes in the future that break this, +although they will not be enabled without an explicit option. + +This encoder is not designed to (and will probably never) output the exact same bitstream as the reference encoder. + +Also note, that the cgo decompressor currently does not [report all errors on invalid input](https://github.com/DataDog/zstd/issues/59), +[omits error checks](https://github.com/DataDog/zstd/issues/61), [ignores checksums](https://github.com/DataDog/zstd/issues/43) +and seems to ignore concatenated streams, even though [it is part of the spec](https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frames). + +#### Blocks + +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. + +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`. + +Especially when encoding blocks you should take special care to reuse the encoder. +This will effectively make it run without allocations after a warmup period. +To make it run completely without allocations, supply a destination buffer with space for all content. + +```Go +import "github.com/klauspost/compress/zstd" + +// Create a writer that caches compressors. +// For this operation type we supply a nil Reader. +var encoder, _ = zstd.NewWriter(nil) + +// Compress a buffer. +// If you have a destination buffer, the allocation in the call can also be eliminated. +func Compress(src []byte) []byte { + return encoder.EncodeAll(src, make([]byte, 0, len(src))) +} +``` + +You can control the maximum number of concurrent encodes using the `WithEncoderConcurrency(n)` +option when creating the writer. + +Using the Encoder for both a stream and individual blocks concurrently is safe. + +### Performance + +I have collected some speed examples to compare speed and compression against other compressors. + +* `file` is the input file. +* `out` is the compressor used. `zskp` is this package. `zstd` is the Datadog cgo library. `gzstd/gzkp` is gzip standard and this library. +* `level` is the compression level used. For `zskp` level 1 is "fastest", level 2 is "default". +* `insize`/`outsize` is the input/output size. +* `millis` is the number of milliseconds used for compression. +* `mb/s` is megabytes (2^20 bytes) per second. + +``` +Silesia Corpus: +http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip + +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 65177448 1899 106.44 + +cgo zstd: +silesia.tar zstd 1 211947520 73605392 543 371.56 +silesia.tar zstd 3 211947520 66793289 864 233.68 +silesia.tar zstd 6 211947520 62916450 1913 105.66 + +gzip, stdlib/this package: +silesia.tar gzstd 1 211947520 80007735 1654 122.21 +silesia.tar gzkp 1 211947520 80369488 1168 173.06 + +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 185792019 9324 195.48 +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 gzstd 1 1911399616 357382641 10251 177.82 +gob-stream gzkp 1 1911399616 362156523 5695 320.08 + +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 294540704 11725 81.34 +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 gzstd 1 1000000000 382578136 9604 99.30 +enwik9 gzkp 1 1000000000 383825945 6544 145.73 + +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 537511906 29252 204.54 +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 gzstd 1 6273951764 1164400847 29948 199.79 +github-june-2days-2019.json gzkp 1 6273951764 1128755542 19236 311.03 + +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 3224594213 71751 113.75 +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 gzstd 1 8558382592 3926257486 57722 141.40 +rawstudio-mint14.tar gzkp 1 8558382592 3970463184 41749 195.49 + +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 538490114 19880 159.53 +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 gzstd 1 3325605752 928656485 23876 132.83 +nyc-taxi-data-10M.csv gzkp 1 3325605752 924718719 16388 193.53 +``` + +## Decompressor + +Staus: STABLE - there may still be subtle bugs, but a wide variety of content has been tested. + +This library is being continuously [fuzz-tested](https://github.com/klauspost/compress-fuzz), +kindly supplied by [fuzzit.dev](https://fuzzit.dev/). +The main purpose of the fuzz testing is to ensure that it is not possible to crash the decoder, +or run it past its limits with ANY input provided. + +### Usage + +The package has been designed for two main usages, big streams of data and smaller in-memory buffers. +There are two main usages of the package for these. Both of them are accessed by creating a `Decoder`. + +For streaming use a simple setup could look like this: + +```Go +import "github.com/klauspost/compress/zstd" + +func Decompress(in io.Reader, out io.Writer) error { + d, err := zstd.NewReader(in) + if err != nil { + return err + } + defer d.Close() + + // Copy content... + _, err = io.Copy(out, d) + return err +} +``` + +It is important to use the "Close" function when you no longer need the Reader to stop running goroutines. +See "Allocation-less operation" below. + +For decoding buffers, it could look something like this: + +```Go +import "github.com/klauspost/compress/zstd" + +// Create a reader that caches decompressors. +// For this operation type we supply a nil Reader. +var decoder, _ = zstd.NewReader(nil) + +// Decompress a buffer. We don't supply a destination buffer, +// so it will be allocated by the decoder. +func Decompress(src []byte) ([]byte, error) { + return decoder.DecodeAll(src, nil) +} +``` + +Both of these cases should provide the functionality needed. +The decoder can be used for *concurrent* decompression of multiple buffers. +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. + +### Dictionaries + +Data compressed with [dictionaries](https://github.com/facebook/zstd#the-case-for-small-data-compression) can be decompressed. + +Dictionaries are added individually to Decoders. +Dictionaries are generated by the `zstd --train` command and contains an initial state for the decoder. +To add a dictionary use the `WithDecoderDicts(dicts ...[]byte)` option with the dictionary data. +Several dictionaries can be added at once. + +The dictionary will be used automatically for the data that specifies them. +A re-used Decoder will still contain the dictionaries registered. + +When registering multiple dictionaries with the same ID, the last one will be used. + +It is possible to use dictionaries when compressing data. + +To enable a dictionary use `WithEncoderDict(dict []byte)`. Here only one dictionary will be used +and it will likely be used even if it doesn't improve compression. + +The used dictionary must be used to decompress the content. + +For any real gains, the dictionary should be built with similar data. +If an unsuitable dictionary is used the output may be slightly larger than using no dictionary. +Use the [zstd commandline tool](https://github.com/facebook/zstd/releases) to build a dictionary from sample data. +For information see [zstd dictionary information](https://github.com/facebook/zstd#the-case-for-small-data-compression). + +For now there is a fixed startup performance penalty for compressing content with dictionaries. +This will likely be improved over time. Just be aware to test performance when implementing. + +### Allocation-less operation + +The decoder has been designed to operate without allocations after a warmup. + +This means that you should *store* the decoder for best performance. +To re-use a stream decoder, use the `Reset(r io.Reader) error` to switch to another stream. +A decoder can safely be re-used even if the previous stream failed. + +To release the resources, you must call the `Close()` function on a decoder. +After this it can *no longer be reused*, but all running goroutines will be stopped. +So you *must* use this if you will no longer need the Reader. + +For decompressing smaller buffers a single decoder can be used. +When decoding buffers, you can supply a destination slice with length 0 and your expected capacity. +In this case no unneeded allocations should be made. + +### Concurrency + +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 + +* 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. + +So effectively this also means the decoder will "read ahead" and prepare data to always be available for output. + +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. + + +### 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. + +``` +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 + +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 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 +``` + +This reflects the performance around May 2020, but this may be out of date. + +# 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 new file mode 100644 index 000000000000..854458537154 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bitreader.go @@ -0,0 +1,136 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "errors" + "io" + "math/bits" +) + +// 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 // Maybe use [16]byte, but shifting is awkward. + 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(highBits(uint32(v))) + return nil +} + +// getBits will return n bits. n can be 0. +func (b *bitReader) getBits(n uint8) int { + if n == 0 /*|| b.bitsRead >= 64 */ { + return 0 + } + return b.getBitsFast(n) +} + +// getBitsFast requires that at least one bit is requested every time. +// There are no checks if the buffer is filled. +func (b *bitReader) getBitsFast(n uint8) int { + const regMask = 64 - 1 + v := uint32((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask)) + b.bitsRead += n + return int(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:] + 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 +} + +// 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 +} + +// overread returns true if more bits have been requested than is on the stream. +func (b *bitReader) overread() bool { + return b.bitsRead > 64 +} + +// remain returns the number of bits remaining. +func (b *bitReader) remain() uint { + return b.off*8 + 64 - uint(b.bitsRead) +} + +// 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 +} + +func highBits(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} diff --git a/vendor/github.com/klauspost/compress/zstd/bitwriter.go b/vendor/github.com/klauspost/compress/zstd/bitwriter.go new file mode 100644 index 000000000000..303ae90f9447 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bitwriter.go @@ -0,0 +1,169 @@ +// Copyright 2018 Klaus Post. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. +// Based on work Copyright (c) 2013, Yann Collet, released under BSD License. + +package zstd + +import "fmt" + +// bitWriter will write bits. +// First bit will be LSB of the first byte of output. +type bitWriter struct { + bitContainer uint64 + nBits uint8 + out []byte +} + +// bitMask16 is bitmasks. Has extra to avoid bounds check. +var bitMask16 = [32]uint16{ + 0, 1, 3, 7, 0xF, 0x1F, + 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, + 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, + 0xFFFF, 0xFFFF} /* up to 16 bits */ + +var bitMask32 = [32]uint32{ + 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF, + 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, + 0x1ffff, 0x3ffff, 0x7FFFF, 0xfFFFF, 0x1fFFFF, 0x3fFFFF, 0x7fFFFF, 0xffFFFF, + 0x1ffFFFF, 0x3ffFFFF, 0x7ffFFFF, 0xfffFFFF, 0x1fffFFFF, 0x3fffFFFF, 0x7fffFFFF, +} // up to 32 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 +} + +// addBits32NC will add up to 32 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) { + b.bitContainer |= uint64(value&bitMask32[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) { + 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: + 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 { + return + } + b.out = append(b.out, + byte(b.bitContainer), + byte(b.bitContainer>>8), + byte(b.bitContainer>>16), + byte(b.bitContainer>>24)) + b.nBits -= 32 + b.bitContainer >>= 32 +} + +// flushAlign will flush remaining full bytes and align to next byte boundary. +func (b *bitWriter) flushAlign() { + nbBytes := (b.nBits + 7) >> 3 + for i := uint8(0); i < nbBytes; i++ { + b.out = append(b.out, byte(b.bitContainer>>(i*8))) + } + b.nBits = 0 + b.bitContainer = 0 +} + +// close will write the alignment bit and write the final byte(s) +// to the output. +func (b *bitWriter) close() error { + // End mark + b.addBits16Clean(1, 1) + // flush until next byte. + 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/zstd/blockdec.go b/vendor/github.com/klauspost/compress/zstd/blockdec.go new file mode 100644 index 000000000000..4733ea876a1d --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/blockdec.go @@ -0,0 +1,739 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "io" + "sync" + + "github.com/klauspost/compress/huff0" + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +type blockType uint8 + +//go:generate stringer -type=blockType,literalsBlockType,seqCompMode,tableIndex + +const ( + blockTypeRaw blockType = iota + blockTypeRLE + blockTypeCompressed + blockTypeReserved +) + +type literalsBlockType uint8 + +const ( + literalsBlockRaw literalsBlockType = iota + literalsBlockRLE + literalsBlockCompressed + literalsBlockTreeless +) + +const ( + // maxCompressedBlockSize is the biggest allowed compressed block size (128KB) + maxCompressedBlockSize = 128 << 10 + + // 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 + + // We support slightly less than the reference decoder to be able to + // use ints on 32 bit archs. + maxOffsetBits = 30 +) + +var ( + huffDecoderPool = sync.Pool{New: func() interface{} { + return &huff0.Scratch{} + }} + + fseDecoderPool = sync.Pool{New: func() interface{} { + return &fseDecoder{} + }} +) + +type blockDec struct { + // Raw source data of the block. + data []byte + dataStorage []byte + + // Destination of the decoded data. + dst []byte + + // Buffer for literals data. + literalBuf []byte + + // Window size of the block. + WindowSize uint64 + + history chan *history + input chan struct{} + result chan decodeOutput + sequenceBuf []seq + err error + decWG sync.WaitGroup + + // Frame to use for singlethreaded decoding. + // Should not be used by the decoder itself since parent may be another frame. + localFrame *frameDec + + // Block is RLE, this is the size. + RLESize uint32 + tmp [4]byte + + Type blockType + + // Is this the last block of a frame? + Last bool + + // Use less memory + lowMem bool +} + +func (b *blockDec) String() string { + if b == nil { + return "" + } + return fmt.Sprintf("Steam Size: %d, Type: %v, Last: %t, Window: %d", len(b.data), b.Type, b.Last, b.WindowSize) +} + +func newBlockDec(lowMem bool) *blockDec { + b := blockDec{ + lowMem: lowMem, + result: make(chan decodeOutput, 1), + input: make(chan struct{}, 1), + history: make(chan *history, 1), + } + b.decWG.Add(1) + go b.startDecoder() + return &b +} + +// reset will reset the block. +// 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 + } + 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 + switch b.Type { + case blockTypeReserved: + return ErrReservedBlockType + case blockTypeRLE: + b.RLESize = uint32(cSize) + if b.lowMem { + maxSize = cSize + } + cSize = 1 + case blockTypeCompressed: + if debug { + println("Data size on stream:", cSize) + } + b.RLESize = 0 + maxSize = maxCompressedBlockSize + if windowSize < maxCompressedBlockSize && b.lowMem { + maxSize = int(windowSize) + } + if cSize > maxCompressedBlockSize || uint64(cSize) > b.WindowSize { + if debug { + printf("compressed block too big: csize:%d block: %+v\n", uint64(cSize), b) + } + return ErrCompressedSizeTooBig + } + case blockTypeRaw: + b.RLESize = 0 + // We do not need a destination for raw blocks. + maxSize = -1 + default: + panic("Invalid block type") + } + + // Read block data. + if cap(b.dataStorage) < cSize { + if b.lowMem { + b.dataStorage = make([]byte, 0, cSize) + } else { + b.dataStorage = make([]byte, 0, maxBlockSize) + } + } + 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 { + println("Reading block:", err, "(", cSize, ")", len(b.data)) + printf("%T", br) + } + return err + } + return nil +} + +// sendEOF will make the decoder send EOF on this frame. +func (b *blockDec) sendErr(err error) { + 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. +func (b *blockDec) decodeBuf(hist *history) error { + 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) + } + } + b.dst = b.dst[:b.RLESize] + v := b.data[0] + for i := range b.dst { + b.dst[i] = v + } + hist.appendKeep(b.dst) + return nil + case blockTypeRaw: + hist.appendKeep(b.data) + return nil + case blockTypeCompressed: + saved := b.dst + b.dst = hist.b + hist.b = nil + err := b.decodeCompressed(hist) + if debug { + println("Decompressed to total", len(b.dst), "bytes, hash:", xxhash.Sum64(b.dst), "error:", err) + } + hist.b = b.dst + b.dst = saved + return err + case blockTypeReserved: + // Used for returning errors. + return b.err + default: + panic("Invalid block type") + } +} + +// 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 + } + }() + } + // There must be at least one byte for Literals_Block_Type and one for Sequences_Section_Header + if len(in) < 2 { + return ErrBlockTooSmall + } + litType := literalsBlockType(in[0] & 3) + var litRegenSize int + var litCompSize int + sizeFormat := (in[0] >> 2) & 3 + var fourStreams bool + switch litType { + case literalsBlockRaw, literalsBlockRLE: + switch sizeFormat { + case 0, 2: + // Regenerated_Size uses 5 bits (0-31). Literals_Section_Header uses 1 byte. + litRegenSize = int(in[0] >> 3) + in = in[1:] + case 1: + // Regenerated_Size uses 12 bits (0-4095). Literals_Section_Header uses 2 bytes. + litRegenSize = int(in[0]>>4) + (int(in[1]) << 4) + in = in[2:] + case 3: + // 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 + } + litRegenSize = int(in[0]>>4) + (int(in[1]) << 4) + (int(in[2]) << 12) + in = in[3:] + } + case literalsBlockCompressed, literalsBlockTreeless: + switch sizeFormat { + case 0, 1: + // 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 + } + n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + litRegenSize = int(n & 1023) + litCompSize = int(n >> 10) + fourStreams = sizeFormat == 1 + in = in[3:] + case 2: + fourStreams = true + if len(in) < 4 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in)) + return ErrBlockTooSmall + } + n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20) + litRegenSize = int(n & 16383) + litCompSize = int(n >> 14) + in = in[4:] + case 3: + fourStreams = true + if len(in) < 5 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in)) + return 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) + litCompSize = int(n >> 18) + in = in[5:] + } + } + if debug { + println("literals type:", litType, "litRegenSize:", litRegenSize, "litCompSize:", litCompSize, "sizeFormat:", sizeFormat, "4X:", fourStreams) + } + var literals []byte + var huff *huff0.Scratch + switch litType { + case literalsBlockRaw: + if len(in) < litRegenSize { + println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litRegenSize) + return ErrBlockTooSmall + } + literals = in[:litRegenSize] + in = in[litRegenSize:] + //printf("Found %d uncompressed literals\n", litRegenSize) + case literalsBlockRLE: + if len(in) < 1 { + println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", 1) + return ErrBlockTooSmall + } + if cap(b.literalBuf) < litRegenSize { + if b.lowMem { + b.literalBuf = make([]byte, litRegenSize) + } else { + if litRegenSize > maxCompressedLiteralSize { + // Exceptional + b.literalBuf = make([]byte, litRegenSize) + } else { + b.literalBuf = make([]byte, litRegenSize, maxCompressedLiteralSize) + + } + } + } + literals = b.literalBuf[:litRegenSize] + v := in[0] + for i := range literals { + literals[i] = v + } + in = in[1:] + if debug { + 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 + } + // Store compressed literals, so we defer decoding until we get history. + literals = in[:litCompSize] + in = in[litCompSize:] + if debug { + 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 + } + 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) + } else { + b.literalBuf = make([]byte, 0, maxCompressedLiteralSize) + } + } + if huff == nil { + huff = &huff0.Scratch{} + } + huff, literals, err = huff0.ReadTable(literals, huff) + if err != nil { + println("reading huffman table:", err) + return err + } + // Use our out buffer. + if fourStreams { + literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals) + } else { + literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals) + } + if err != nil { + println("decoding compressed literals:", err) + return 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)) + } + if debug { + printf("Decompressed %d literals into %d bytes\n", litCompSize, litRegenSize) + } + } + + // Decode Sequences + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#sequences-section + if len(in) < 1 { + return ErrBlockTooSmall + } + seqHeader := in[0] + nSeqs := 0 + switch { + case seqHeader == 0: + in = in[1:] + case seqHeader < 128: + nSeqs = int(seqHeader) + in = in[1:] + case seqHeader < 255: + if len(in) < 2 { + return ErrBlockTooSmall + } + nSeqs = int(seqHeader-128)<<8 | int(in[1]) + in = in[2:] + case seqHeader == 255: + if len(in) < 3 { + return ErrBlockTooSmall + } + 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) + } + } else { + // Reuse buffer + b.sequenceBuf = b.sequenceBuf[:nSeqs] + } + var seqs = &sequenceDecs{} + if nSeqs > 0 { + if len(in) < 1 { + return ErrBlockTooSmall + } + br := byteReader{b: in, off: 0} + compMode := br.Uint8() + br.advance(1) + if debug { + printf("Compression modes: 0b%b", compMode) + } + for i := uint(0); i < 3; i++ { + mode := seqCompMode((compMode >> (6 - i*2)) & 3) + if debug { + println("Table", tableIndex(i), "is", mode) + } + var seq *sequenceDec + switch tableIndex(i) { + case tableLiteralLengths: + seq = &seqs.litLengths + case tableOffsets: + seq = &seqs.offsets + case tableMatchLengths: + seq = &seqs.matchLengths + default: + panic("unknown table") + } + switch mode { + case compModePredefined: + seq.fse = &fsePredef[i] + case compModeRLE: + if br.remain() < 1 { + return ErrBlockTooSmall + } + v := br.Uint8() + br.advance(1) + dec := 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 { + 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 err != nil { + println("Read table error:", err) + return err + } + err = dec.transform(symbolTableX[i]) + if err != nil { + println("Transform table error:", err) + return err + } + if debug { + println("Read table ok", "symbolLen:", dec.symbolLen) + } + seq.fse = dec + case compModeRepeat: + seq.repeat = true + } + if br.overread() { + return io.ErrUnexpectedEOF + } + } + 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 literials 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 nSeqs == 0 { + // Decompressed content is defined entirely as Literals Section content. + b.dst = append(b.dst, literals...) + if delayedHistory { + hist.append(literals) + } + return nil + } + + seqs, err := seqs.mergeHistory(&hist.decoders) + if err != nil { + return err + } + 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. + + hbytes := hist.b + if len(hbytes) > hist.windowSize { + hbytes = hbytes[len(hbytes)-hist.windowSize:] + // We do not need history any more. + 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) + if err != nil { + return err + } + if !br.finished() { + return fmt.Errorf("%d extra bits on block, should be 0", br.remain()) + } + + err = br.close() + if err != nil { + printf("Closing sequences: %v, %+v\n", err, *br) + } + 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 + + 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 + } + hist.append(b.dst) + hist.recentOffsets = seqs.prevOffset + if debug { + println("Finished block with literals:", len(literals), "and", nSeqs, "sequences.") + } + + return nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/blockenc.go b/vendor/github.com/klauspost/compress/zstd/blockenc.go new file mode 100644 index 000000000000..083fbb502f40 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/blockenc.go @@ -0,0 +1,854 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "math" + "math/bits" + + "github.com/klauspost/compress/huff0" +) + +type blockEnc struct { + size int + literals []byte + sequences []seq + coders seqCoders + litEnc *huff0.Scratch + dictLitEnc *huff0.Scratch + wr bitWriter + + extraLits int + last bool + + output []byte + recentOffsets [3]uint32 + prevRecentOffsets [3]uint32 +} + +// init should be used once the block has been created. +// If called more than once, the effect is the same as calling reset. +func (b *blockEnc) init() { + if cap(b.literals) < maxCompressedLiteralSize { + b.literals = make([]byte, 0, maxCompressedLiteralSize) + } + const defSeqs = 200 + b.literals = b.literals[:0] + if cap(b.sequences) < defSeqs { + b.sequences = make([]seq, 0, defSeqs) + } + if cap(b.output) < maxCompressedBlockSize { + b.output = make([]byte, 0, maxCompressedBlockSize) + } + if b.coders.mlEnc == nil { + b.coders.mlEnc = &fseEncoder{} + b.coders.mlPrev = &fseEncoder{} + b.coders.ofEnc = &fseEncoder{} + b.coders.ofPrev = &fseEncoder{} + b.coders.llEnc = &fseEncoder{} + b.coders.llPrev = &fseEncoder{} + } + b.litEnc = &huff0.Scratch{WantLogLess: 4} + b.reset(nil) +} + +// initNewEncode can be used to reset offsets and encoders to the initial state. +func (b *blockEnc) initNewEncode() { + b.recentOffsets = [3]uint32{1, 4, 8} + b.litEnc.Reuse = huff0.ReusePolicyNone + b.coders.setPrev(nil, nil, nil) +} + +// reset will reset the block for a new encode, but in the same stream, +// meaning that state will be carried over, but the block content is reset. +// If a previous block is provided, the recent offsets are carried over. +func (b *blockEnc) reset(prev *blockEnc) { + b.extraLits = 0 + b.literals = b.literals[:0] + b.size = 0 + b.sequences = b.sequences[:0] + b.output = b.output[:0] + b.last = false + if prev != nil { + b.recentOffsets = prev.prevRecentOffsets + } +} + +// reset will reset the block for a new encode, but in the same stream, +// meaning that state will be carried over, but the block content is reset. +// If a previous block is provided, the recent offsets are carried over. +func (b *blockEnc) swapEncoders(prev *blockEnc) { + b.coders.swap(&prev.coders) + b.litEnc, prev.litEnc = prev.litEnc, b.litEnc +} + +// blockHeader contains the information for a block header. +type blockHeader uint32 + +// setLast sets the 'last' indicator on a block. +func (h *blockHeader) setLast(b bool) { + if b { + *h = *h | 1 + } else { + const mask = (1 << 24) - 2 + *h = *h & mask + } +} + +// setSize will store the compressed size of a block. +func (h *blockHeader) setSize(v uint32) { + const mask = 7 + *h = (*h)&mask | blockHeader(v<<3) +} + +// setType sets the block type. +func (h *blockHeader) setType(t blockType) { + const mask = 1 | (((1 << 24) - 1) ^ 7) + *h = (*h & mask) | blockHeader(t<<1) +} + +// appendTo will append the block header to a slice. +func (h blockHeader) appendTo(b []byte) []byte { + return append(b, uint8(h), uint8(h>>8), uint8(h>>16)) +} + +// String returns a string representation of the block. +func (h blockHeader) String() string { + return fmt.Sprintf("Type: %d, Size: %d, Last:%t", (h>>1)&3, h>>3, h&1 == 1) +} + +// literalsHeader contains literals header information. +type literalsHeader uint64 + +// setType can be used to set the type of literal block. +func (h *literalsHeader) setType(t literalsBlockType) { + const mask = math.MaxUint64 - 3 + *h = (*h & mask) | literalsHeader(t) +} + +// setSize can be used to set a single size, for uncompressed and RLE content. +func (h *literalsHeader) setSize(regenLen int) { + inBits := bits.Len32(uint32(regenLen)) + // Only retain 2 bits + const mask = 3 + lh := uint64(*h & mask) + switch { + case inBits < 5: + lh |= (uint64(regenLen) << 3) | (1 << 60) + if debug { + got := int(lh>>3) & 0xff + if got != regenLen { + panic(fmt.Sprint("litRegenSize = ", regenLen, "(want) != ", got, "(got)")) + } + } + case inBits < 12: + lh |= (1 << 2) | (uint64(regenLen) << 4) | (2 << 60) + case inBits < 20: + lh |= (3 << 2) | (uint64(regenLen) << 4) | (3 << 60) + default: + panic(fmt.Errorf("internal error: block too big (%d)", regenLen)) + } + *h = literalsHeader(lh) +} + +// setSizes will set the size of a compressed literals section and the input length. +func (h *literalsHeader) setSizes(compLen, inLen int, single bool) { + compBits, inBits := bits.Len32(uint32(compLen)), bits.Len32(uint32(inLen)) + // Only retain 2 bits + const mask = 3 + lh := uint64(*h & mask) + switch { + case compBits <= 10 && inBits <= 10: + if !single { + lh |= 1 << 2 + } + lh |= (uint64(inLen) << 4) | (uint64(compLen) << (10 + 4)) | (3 << 60) + if debug { + const mmask = (1 << 24) - 1 + n := (lh >> 4) & mmask + if int(n&1023) != inLen { + panic(fmt.Sprint("regensize:", int(n&1023), "!=", inLen, inBits)) + } + if int(n>>10) != compLen { + panic(fmt.Sprint("compsize:", int(n>>10), "!=", compLen, compBits)) + } + } + case compBits <= 14 && inBits <= 14: + lh |= (2 << 2) | (uint64(inLen) << 4) | (uint64(compLen) << (14 + 4)) | (4 << 60) + if single { + panic("single stream used with more than 10 bits length.") + } + case compBits <= 18 && inBits <= 18: + lh |= (3 << 2) | (uint64(inLen) << 4) | (uint64(compLen) << (18 + 4)) | (5 << 60) + if single { + panic("single stream used with more than 10 bits length.") + } + default: + panic("internal error: block too big") + } + *h = literalsHeader(lh) +} + +// appendTo will append the literals header to a byte slice. +func (h literalsHeader) appendTo(b []byte) []byte { + size := uint8(h >> 60) + switch size { + case 1: + b = append(b, uint8(h)) + case 2: + b = append(b, uint8(h), uint8(h>>8)) + case 3: + b = append(b, uint8(h), uint8(h>>8), uint8(h>>16)) + case 4: + b = append(b, uint8(h), uint8(h>>8), uint8(h>>16), uint8(h>>24)) + case 5: + b = append(b, uint8(h), uint8(h>>8), uint8(h>>16), uint8(h>>24), uint8(h>>32)) + default: + panic(fmt.Errorf("internal error: literalsHeader has invalid size (%d)", size)) + } + return b +} + +// size returns the output size with currently set values. +func (h literalsHeader) size() int { + return int(h >> 60) +} + +func (h literalsHeader) String() string { + return fmt.Sprintf("Type: %d, SizeFormat: %d, Size: 0x%d, Bytes:%d", literalsBlockType(h&3), (h>>2)&3, h&((1<<60)-1)>>4, h>>60) +} + +// pushOffsets will push the recent offsets to the backup store. +func (b *blockEnc) pushOffsets() { + b.prevRecentOffsets = b.recentOffsets +} + +// pushOffsets will push the recent offsets to the backup store. +func (b *blockEnc) popOffsets() { + b.recentOffsets = b.prevRecentOffsets +} + +// matchOffset will adjust recent offsets and return the adjusted one, +// if it matches a previous offset. +func (b *blockEnc) matchOffset(offset, lits uint32) uint32 { + // Check if offset is one of the recent offsets. + // Adjusts the output offset accordingly. + // Gives a tiny bit of compression, typically around 1%. + if true { + if lits > 0 { + switch offset { + case b.recentOffsets[0]: + offset = 1 + case b.recentOffsets[1]: + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 2 + case b.recentOffsets[2]: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 3 + default: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset += 3 + } + } else { + switch offset { + case b.recentOffsets[1]: + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 1 + case b.recentOffsets[2]: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 2 + case b.recentOffsets[0] - 1: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset = 3 + default: + b.recentOffsets[2] = b.recentOffsets[1] + b.recentOffsets[1] = b.recentOffsets[0] + b.recentOffsets[0] = offset + offset += 3 + } + } + } else { + offset += 3 + } + return offset +} + +// encodeRaw can be used to set the output to a raw representation of supplied bytes. +func (b *blockEnc) encodeRaw(a []byte) { + var bh blockHeader + bh.setLast(b.last) + bh.setSize(uint32(len(a))) + bh.setType(blockTypeRaw) + b.output = bh.appendTo(b.output[:0]) + b.output = append(b.output, a...) + if debug { + println("Adding RAW block, length", len(a), "last:", b.last) + } +} + +// encodeRaw can be used to set the output to a raw representation of supplied bytes. +func (b *blockEnc) encodeRawTo(dst, src []byte) []byte { + var bh blockHeader + bh.setLast(b.last) + bh.setSize(uint32(len(src))) + bh.setType(blockTypeRaw) + dst = bh.appendTo(dst) + dst = append(dst, src...) + if debug { + println("Adding RAW block, length", len(src), "last:", b.last) + } + return dst +} + +// encodeLits can be used if the block is only litLen. +func (b *blockEnc) encodeLits(lits []byte, raw bool) error { + var bh blockHeader + bh.setLast(b.last) + bh.setSize(uint32(len(lits))) + + // Don't compress extremely small blocks + if len(lits) < 8 || (len(lits) < 32 && b.dictLitEnc == nil) || raw { + if debug { + println("Adding RAW block, length", len(lits), "last:", b.last) + } + bh.setType(blockTypeRaw) + b.output = bh.appendTo(b.output) + b.output = append(b.output, lits...) + return nil + } + + var ( + out []byte + reUsed, single bool + err error + ) + if b.dictLitEnc != nil { + b.litEnc.TransferCTable(b.dictLitEnc) + b.litEnc.Reuse = huff0.ReusePolicyAllow + b.dictLitEnc = nil + } + if len(lits) >= 1024 { + // Use 4 Streams. + out, reUsed, err = huff0.Compress4X(lits, b.litEnc) + } else if len(lits) > 32 { + // Use 1 stream + single = true + out, reUsed, err = huff0.Compress1X(lits, b.litEnc) + } else { + err = huff0.ErrIncompressible + } + + switch err { + case huff0.ErrIncompressible: + if debug { + println("Adding RAW block, length", len(lits), "last:", b.last) + } + bh.setType(blockTypeRaw) + b.output = bh.appendTo(b.output) + b.output = append(b.output, lits...) + return nil + case huff0.ErrUseRLE: + if debug { + println("Adding RLE block, length", len(lits)) + } + bh.setType(blockTypeRLE) + b.output = bh.appendTo(b.output) + b.output = append(b.output, lits[0]) + return nil + default: + return err + case nil: + } + // Compressed... + // Now, allow reuse + b.litEnc.Reuse = huff0.ReusePolicyAllow + bh.setType(blockTypeCompressed) + var lh literalsHeader + if reUsed { + if debug { + println("Reused tree, compressed to", len(out)) + } + lh.setType(literalsBlockTreeless) + } else { + if debug { + println("New tree, compressed to", len(out), "tree size:", len(b.litEnc.OutTable)) + } + lh.setType(literalsBlockCompressed) + } + // Set sizes + lh.setSizes(len(out), len(lits), single) + bh.setSize(uint32(len(out) + lh.size() + 1)) + + // Write block headers. + b.output = bh.appendTo(b.output) + b.output = lh.appendTo(b.output) + // Add compressed data. + b.output = append(b.output, out...) + // No sequences. + b.output = append(b.output, 0) + return nil +} + +// fuzzFseEncoder can be used to fuzz the FSE encoder. +func fuzzFseEncoder(data []byte) int { + if len(data) > maxSequences || len(data) < 2 { + return 0 + } + enc := fseEncoder{} + hist := enc.Histogram()[:256] + maxSym := uint8(0) + for i, v := range data { + v = v & 63 + data[i] = v + hist[v]++ + if v > maxSym { + maxSym = v + } + } + if maxSym == 0 { + // All 0 + return 0 + } + maxCount := func(a []uint32) int { + var max uint32 + for _, v := range a { + if v > max { + max = v + } + } + return int(max) + } + cnt := maxCount(hist[:maxSym]) + if cnt == len(data) { + // RLE + return 0 + } + enc.HistogramFinished(maxSym, cnt) + err := enc.normalizeCount(len(data)) + if err != nil { + return 0 + } + _, err = enc.writeCount(nil) + if err != nil { + panic(err) + } + return 1 +} + +// encode will encode the block and append the output in b.output. +// Previous offset codes must be pushed if more blocks are expected. +func (b *blockEnc) encode(org []byte, raw, rawAllLits bool) error { + if len(b.sequences) == 0 { + return b.encodeLits(b.literals, rawAllLits) + } + // We want some difference to at least account for the headers. + saved := b.size - len(b.literals) - (b.size >> 5) + if saved < 16 { + if org == nil { + return errIncompressible + } + b.popOffsets() + return b.encodeLits(org, rawAllLits) + } + + var bh blockHeader + var lh literalsHeader + bh.setLast(b.last) + bh.setType(blockTypeCompressed) + // Store offset of the block header. Needed when we know the size. + bhOffset := len(b.output) + b.output = bh.appendTo(b.output) + + var ( + out []byte + reUsed, single bool + err error + ) + if b.dictLitEnc != nil { + b.litEnc.TransferCTable(b.dictLitEnc) + b.litEnc.Reuse = huff0.ReusePolicyAllow + b.dictLitEnc = nil + } + if len(b.literals) >= 1024 && !raw { + // Use 4 Streams. + out, reUsed, err = huff0.Compress4X(b.literals, b.litEnc) + } else if len(b.literals) > 32 && !raw { + // Use 1 stream + single = true + out, reUsed, err = huff0.Compress1X(b.literals, b.litEnc) + } else { + err = huff0.ErrIncompressible + } + + switch err { + case huff0.ErrIncompressible: + lh.setType(literalsBlockRaw) + lh.setSize(len(b.literals)) + b.output = lh.appendTo(b.output) + b.output = append(b.output, b.literals...) + if debug { + println("Adding literals RAW, length", len(b.literals)) + } + case huff0.ErrUseRLE: + lh.setType(literalsBlockRLE) + lh.setSize(len(b.literals)) + b.output = lh.appendTo(b.output) + b.output = append(b.output, b.literals[0]) + if debug { + println("Adding literals RLE") + } + default: + if debug { + println("Adding literals ERROR:", err) + } + return err + case nil: + // Compressed litLen... + if reUsed { + if debug { + println("reused tree") + } + lh.setType(literalsBlockTreeless) + } else { + if debug { + println("new tree, size:", len(b.litEnc.OutTable)) + } + lh.setType(literalsBlockCompressed) + if debug { + _, _, err := huff0.ReadTable(out, nil) + if err != nil { + panic(err) + } + } + } + lh.setSizes(len(out), len(b.literals), single) + if debug { + 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 { + println("Adding literals compressed") + } + } + // Sequence compression + + // Write the number of sequences + switch { + case len(b.sequences) < 128: + b.output = append(b.output, uint8(len(b.sequences))) + case len(b.sequences) < 0x7f00: // TODO: this could be wrong + n := len(b.sequences) + b.output = append(b.output, 128+uint8(n>>8), uint8(n)) + default: + n := len(b.sequences) - 0x7f00 + b.output = append(b.output, 255, uint8(n), uint8(n>>8)) + } + if debug { + println("Encoding", len(b.sequences), "sequences") + } + b.genCodes() + llEnc := b.coders.llEnc + ofEnc := b.coders.ofEnc + mlEnc := b.coders.mlEnc + err = llEnc.normalizeCount(len(b.sequences)) + if err != nil { + return err + } + err = ofEnc.normalizeCount(len(b.sequences)) + if err != nil { + return err + } + err = mlEnc.normalizeCount(len(b.sequences)) + if err != nil { + return err + } + + // Choose the best compression mode for each type. + // Will evaluate the new vs predefined and previous. + chooseComp := func(cur, prev, preDef *fseEncoder) (*fseEncoder, seqCompMode) { + // See if predefined/previous is better + hist := cur.count[:cur.symbolLen] + nSize := cur.approxSize(hist) + cur.maxHeaderSize() + predefSize := preDef.approxSize(hist) + prevSize := prev.approxSize(hist) + + // Add a small penalty for new encoders. + // Don't bother with extremely small (<2 byte gains). + nSize = nSize + (nSize+2*8*16)>>4 + switch { + case predefSize <= prevSize && predefSize <= nSize || forcePreDef: + if debug { + println("Using predefined", predefSize>>3, "<=", nSize>>3) + } + return preDef, compModePredefined + case prevSize <= nSize: + if debug { + println("Using previous", prevSize>>3, "<=", nSize>>3) + } + return prev, compModeRepeat + default: + if debug { + 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]) + } + return cur, compModeFSE + } + } + + // Write compression mode + var mode uint8 + if llEnc.useRLE { + mode |= uint8(compModeRLE) << 6 + llEnc.setRLE(b.sequences[0].llCode) + if debug { + println("llEnc.useRLE") + } + } else { + var m seqCompMode + llEnc, m = chooseComp(llEnc, b.coders.llPrev, &fsePredefEnc[tableLiteralLengths]) + mode |= uint8(m) << 6 + } + if ofEnc.useRLE { + mode |= uint8(compModeRLE) << 4 + ofEnc.setRLE(b.sequences[0].ofCode) + if debug { + println("ofEnc.useRLE") + } + } else { + var m seqCompMode + ofEnc, m = chooseComp(ofEnc, b.coders.ofPrev, &fsePredefEnc[tableOffsets]) + mode |= uint8(m) << 4 + } + + if mlEnc.useRLE { + mode |= uint8(compModeRLE) << 2 + mlEnc.setRLE(b.sequences[0].mlCode) + if debug { + println("mlEnc.useRLE, code: ", b.sequences[0].mlCode, "value", b.sequences[0].matchLen) + } + } else { + var m seqCompMode + mlEnc, m = chooseComp(mlEnc, b.coders.mlPrev, &fsePredefEnc[tableMatchLengths]) + mode |= uint8(m) << 2 + } + b.output = append(b.output, mode) + if debug { + printf("Compression modes: 0b%b", mode) + } + b.output, err = llEnc.writeCount(b.output) + if err != nil { + return err + } + start := len(b.output) + b.output, err = ofEnc.writeCount(b.output) + if err != nil { + return err + } + if false { + println("block:", b.output[start:], "tablelog", ofEnc.actualTableLog, "maxcount:", ofEnc.maxCount) + fmt.Printf("selected TableLog: %d, Symbol length: %d\n", ofEnc.actualTableLog, ofEnc.symbolLen) + for i, v := range ofEnc.norm[:ofEnc.symbolLen] { + fmt.Printf("%3d: %5d -> %4d \n", i, ofEnc.count[i], v) + } + } + b.output, err = mlEnc.writeCount(b.output) + if err != nil { + return err + } + + // Maybe in block? + wr := &b.wr + wr.reset(b.output) + + var ll, of, ml cState + + // Current sequence + seq := len(b.sequences) - 1 + s := b.sequences[seq] + llEnc.setBits(llBitsTable[:]) + mlEnc.setBits(mlBitsTable[:]) + ofEnc.setBits(nil) + + llTT, ofTT, mlTT := llEnc.ct.symbolTT[:256], ofEnc.ct.symbolTT[:256], mlEnc.ct.symbolTT[:256] + + // We have 3 bounds checks here (and in the loop). + // Since we are iterating backwards it is kinda hard to avoid. + llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode] + ll.init(wr, &llEnc.ct, llB) + of.init(wr, &ofEnc.ct, ofB) + wr.flush32() + ml.init(wr, &mlEnc.ct, mlB) + + // Each of these lookups also generates a bounds check. + 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, "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() + + // 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) + + if debugSequences { + println("Encoded seq", seq, s) + } + + seq-- + } + } 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-- + } + } + ml.flush(mlEnc.actualTableLog) + of.flush(ofEnc.actualTableLog) + ll.flush(llEnc.actualTableLog) + err = wr.close() + if err != nil { + return err + } + b.output = wr.out + + if len(b.output)-3-bhOffset >= b.size { + // Maybe even add a bigger margin. + b.litEnc.Reuse = huff0.ReusePolicyNone + return errIncompressible + } + + // Size is output minus block header. + bh.setSize(uint32(len(b.output)-bhOffset) - 3) + if debug { + println("Rewriting block header", bh) + } + _ = bh.appendTo(b.output[bhOffset:bhOffset]) + b.coders.setPrev(llEnc, mlEnc, ofEnc) + return nil +} + +var errIncompressible = errors.New("incompressible") + +func (b *blockEnc) genCodes() { + if len(b.sequences) == 0 { + // 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] + for i := range llH { + llH[i] = 0 + } + for i := range ofH { + ofH[i] = 0 + } + for i := range mlH { + mlH[i] = 0 + } + + var llMax, ofMax, mlMax uint8 + for i, seq := range b.sequences { + v := llCode(seq.litLen) + seq.llCode = v + llH[v]++ + if v > llMax { + llMax = v + } + + v = ofCode(seq.offset) + seq.ofCode = v + ofH[v]++ + if v > ofMax { + ofMax = v + } + + v = mlCode(seq.matchLen) + seq.mlCode = v + mlH[v]++ + if v > mlMax { + mlMax = v + if debugAsserts && mlMax > maxMatchLengthSymbol { + panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d), matchlen: %d", mlMax, seq.matchLen)) + } + } + b.sequences[i] = seq + } + maxCount := func(a []uint32) int { + var max uint32 + for _, v := range a { + if v > max { + max = v + } + } + return int(max) + } + if debugAsserts && mlMax > maxMatchLengthSymbol { + panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d)", mlMax)) + } + if debugAsserts && ofMax > maxOffsetBits { + panic(fmt.Errorf("ofMax > maxOffsetBits (%d)", ofMax)) + } + if debugAsserts && llMax > maxLiteralLengthSymbol { + panic(fmt.Errorf("llMax > maxLiteralLengthSymbol (%d)", llMax)) + } + + b.coders.mlEnc.HistogramFinished(mlMax, maxCount(mlH[:mlMax+1])) + b.coders.ofEnc.HistogramFinished(ofMax, maxCount(ofH[:ofMax+1])) + b.coders.llEnc.HistogramFinished(llMax, maxCount(llH[:llMax+1])) +} diff --git a/vendor/github.com/klauspost/compress/zstd/blocktype_string.go b/vendor/github.com/klauspost/compress/zstd/blocktype_string.go new file mode 100644 index 000000000000..01a01e486e18 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/blocktype_string.go @@ -0,0 +1,85 @@ +// Code generated by "stringer -type=blockType,literalsBlockType,seqCompMode,tableIndex"; DO NOT EDIT. + +package zstd + +import "strconv" + +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[blockTypeRaw-0] + _ = x[blockTypeRLE-1] + _ = x[blockTypeCompressed-2] + _ = x[blockTypeReserved-3] +} + +const _blockType_name = "blockTypeRawblockTypeRLEblockTypeCompressedblockTypeReserved" + +var _blockType_index = [...]uint8{0, 12, 24, 43, 60} + +func (i blockType) String() string { + if i >= blockType(len(_blockType_index)-1) { + return "blockType(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _blockType_name[_blockType_index[i]:_blockType_index[i+1]] +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[literalsBlockRaw-0] + _ = x[literalsBlockRLE-1] + _ = x[literalsBlockCompressed-2] + _ = x[literalsBlockTreeless-3] +} + +const _literalsBlockType_name = "literalsBlockRawliteralsBlockRLEliteralsBlockCompressedliteralsBlockTreeless" + +var _literalsBlockType_index = [...]uint8{0, 16, 32, 55, 76} + +func (i literalsBlockType) String() string { + if i >= literalsBlockType(len(_literalsBlockType_index)-1) { + return "literalsBlockType(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _literalsBlockType_name[_literalsBlockType_index[i]:_literalsBlockType_index[i+1]] +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[compModePredefined-0] + _ = x[compModeRLE-1] + _ = x[compModeFSE-2] + _ = x[compModeRepeat-3] +} + +const _seqCompMode_name = "compModePredefinedcompModeRLEcompModeFSEcompModeRepeat" + +var _seqCompMode_index = [...]uint8{0, 18, 29, 40, 54} + +func (i seqCompMode) String() string { + if i >= seqCompMode(len(_seqCompMode_index)-1) { + return "seqCompMode(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _seqCompMode_name[_seqCompMode_index[i]:_seqCompMode_index[i+1]] +} +func _() { + // An "invalid array index" compiler error signifies that the constant values have changed. + // Re-run the stringer command to generate them again. + var x [1]struct{} + _ = x[tableLiteralLengths-0] + _ = x[tableOffsets-1] + _ = x[tableMatchLengths-2] +} + +const _tableIndex_name = "tableLiteralLengthstableOffsetstableMatchLengths" + +var _tableIndex_index = [...]uint8{0, 19, 31, 48} + +func (i tableIndex) String() string { + if i >= tableIndex(len(_tableIndex_index)-1) { + return "tableIndex(" + strconv.FormatInt(int64(i), 10) + ")" + } + return _tableIndex_name[_tableIndex_index[i]:_tableIndex_index[i+1]] +} diff --git a/vendor/github.com/klauspost/compress/zstd/bytebuf.go b/vendor/github.com/klauspost/compress/zstd/bytebuf.go new file mode 100644 index 000000000000..658ef78380e1 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bytebuf.go @@ -0,0 +1,127 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "fmt" + "io" + "io/ioutil" +) + +type byteBuffer interface { + // Read up to 8 bytes. + // Returns nil if no more input is available. + readSmall(n int) []byte + + // Read >8 bytes. + // MAY use the destination slice. + readBig(n int, dst []byte) ([]byte, error) + + // Read a single byte. + readByte() (byte, error) + + // Skip n bytes. + skipN(n int) error +} + +// in-memory buffer +type byteBuf []byte + +func (b *byteBuf) readSmall(n int) []byte { + if debugAsserts && n > 8 { + panic(fmt.Errorf("small read > 8 (%d). use readBig", n)) + } + bb := *b + if len(bb) < n { + return nil + } + r := bb[:n] + *b = bb[n:] + return r +} + +func (b *byteBuf) readBig(n int, dst []byte) ([]byte, error) { + bb := *b + if len(bb) < n { + return nil, io.ErrUnexpectedEOF + } + r := bb[:n] + *b = bb[n:] + return r, nil +} + +func (b *byteBuf) remain() []byte { + return *b +} + +func (b *byteBuf) readByte() (byte, error) { + bb := *b + if len(bb) < 1 { + return 0, nil + } + r := bb[0] + *b = bb[1:] + return r, nil +} + +func (b *byteBuf) skipN(n int) error { + bb := *b + if len(bb) < n { + return io.ErrUnexpectedEOF + } + *b = bb[n:] + return nil +} + +// wrapper around a reader. +type readerWrapper struct { + r io.Reader + tmp [8]byte +} + +func (r *readerWrapper) readSmall(n int) []byte { + 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 { + println("readSmall: got", n2, "want", n, "err", err) + } + return nil + } + return r.tmp[:n] +} + +func (r *readerWrapper) readBig(n int, dst []byte) ([]byte, error) { + if cap(dst) < n { + dst = make([]byte, n) + } + n2, err := io.ReadFull(r.r, dst[:n]) + if err == io.EOF && n > 0 { + err = io.ErrUnexpectedEOF + } + return dst[:n2], err +} + +func (r *readerWrapper) readByte() (byte, error) { + n2, err := r.r.Read(r.tmp[:1]) + if err != nil { + return 0, err + } + if n2 != 1 { + return 0, io.ErrUnexpectedEOF + } + 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) { + 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 new file mode 100644 index 000000000000..2c4fca17fa1d --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/bytereader.go @@ -0,0 +1,88 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +// byteReader provides a byte reader that reads +// little endian values from a byte stream. +// The input stream is manually advanced. +// The reader performs no bounds checks. +type byteReader struct { + b []byte + 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) +} + +// overread returns whether we have advanced too far. +func (b *byteReader) overread() bool { + return b.off > len(b.b) +} + +// Int32 returns a little endian int32 starting at current offset. +func (b byteReader) Int32() int32 { + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := int32(b2[3]) + v2 := int32(b2[2]) + v1 := int32(b2[1]) + v0 := int32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// Uint8 returns the next byte +func (b *byteReader) Uint8() uint8 { + v := b.b[b.off] + return v +} + +// Uint32 returns a little endian uint32 starting at current offset. +func (b byteReader) Uint32() uint32 { + if r := b.remain(); r < 4 { + // Very rare + v := uint32(0) + for i := 1; i <= r; i++ { + v = (v << 8) | uint32(b.b[len(b.b)-i]) + } + return v + } + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := uint32(b2[3]) + v2 := uint32(b2[2]) + v1 := uint32(b2[1]) + v0 := uint32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// Uint32NC returns a little endian uint32 starting at current offset. +// The caller must be sure if there are at least 4 bytes left. +func (b byteReader) Uint32NC() uint32 { + b2 := b.b[b.off:] + b2 = b2[:4] + v3 := uint32(b2[3]) + v2 := uint32(b2[2]) + v1 := uint32(b2[1]) + v0 := uint32(b2[0]) + return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24) +} + +// 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/zstd/decoder.go b/vendor/github.com/klauspost/compress/zstd/decoder.go new file mode 100644 index 000000000000..cdda0de58b44 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/decoder.go @@ -0,0 +1,550 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "bytes" + "errors" + "io" + "sync" +) + +// Decoder provides decoding of zstandard streams. +// The decoder has been designed to operate without allocations after a warmup. +// This means that you should store the decoder for best performance. +// To re-use a stream decoder, use the Reset(r io.Reader) error to switch to another stream. +// A decoder can safely be re-used even if the previous stream failed. +// To release the resources, you must call the Close() function on a decoder. +type Decoder struct { + o decoderOptions + + // 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 + + // Custom dictionaries. + // Always uses copies. + dicts map[uint32]dict + + // streamWg is the waitgroup for all streams + streamWg sync.WaitGroup +} + +// decoderState is used for maintaining state when the decoder +// is used for streaming. +type decoderState struct { + // current block being written to stream. + decodeOutput + + // output in order to be written to stream. + output chan decodeOutput + + // cancel remaining output. + cancel chan struct{} + + flushed bool +} + +var ( + // Check the interfaces we want to support. + _ = io.WriterTo(&Decoder{}) + _ = io.Reader(&Decoder{}) +) + +// NewReader creates a new decoder. +// A nil Reader can be provided in which case Reset can be used to start a decode. +// +// A Decoder can be used in two modes: +// +// 1) As a stream, or +// 2) For stateless decoding using DecodeAll. +// +// Only a single stream can be decoded concurrently, but the same decoder +// can run multiple concurrent stateless decodes. It is even possible to +// use stateless decodes while a stream is being decoded. +// +// The Reset function can be used to initiate a new stream, which is will considerably +// reduce the allocations normally caused by NewReader. +func NewReader(r io.Reader, opts ...DOption) (*Decoder, error) { + initPredefined() + var d Decoder + d.o.setDefault() + for _, o := range opts { + err := o(&d.o) + if err != nil { + return nil, err + } + } + d.current.output = make(chan decodeOutput, d.o.concurrent) + d.current.flushed = true + + // Transfer option dicts. + d.dicts = make(map[uint32]dict, len(d.o.dicts)) + for _, dc := range d.o.dicts { + d.dicts[dc.id] = dc + } + d.o.dicts = nil + + // Create decoders + d.decoders = make(chan *blockDec, d.o.concurrent) + for i := 0; i < d.o.concurrent; i++ { + dec := newBlockDec(d.o.lowMem) + dec.localFrame = newFrameDec(d.o) + d.decoders <- dec + } + + if r == nil { + return &d, nil + } + return &d, d.Reset(r) +} + +// Read bytes from the decompressed stream into p. +// 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, errors.New("no input has been initialized") + } + var n int + for { + if len(d.current.b) > 0 { + filled := copy(p, d.current.b) + p = p[filled:] + d.current.b = d.current.b[filled:] + n += filled + } + if len(p) == 0 { + break + } + if len(d.current.b) == 0 { + // We have an error and no more data + if d.current.err != nil { + break + } + if !d.nextBlock(n == 0) { + return n, nil + } + } + } + if len(d.current.b) > 0 { + if debug { + println("returning", n, "still bytes left:", len(d.current.b)) + } + // Only return error at end of block + return n, nil + } + if d.current.err != nil { + d.drainOutput() + } + if debug { + println("returning", n, d.current.err, len(d.decoders)) + } + return n, d.current.err +} + +// Reset will reset the decoder the supplied stream after the current has finished processing. +// Note that this functionality cannot be used after Close has been called. +func (d *Decoder) Reset(r io.Reader) error { + if d.current.err == ErrDecoderClosed { + return d.current.err + } + if r == nil { + return errors.New("nil Reader sent as input") + } + + if d.stream == nil { + d.stream = make(chan decodeStream, 1) + d.streamWg.Add(1) + go d.startStreamDecoder(d.stream) + } + + d.drainOutput() + + // If bytes buffer and < 1MB, do sync decoding anyway. + if bb, ok := r.(*bytes.Buffer); ok && bb.Len() < 1<<20 { + if debug { + println("*bytes.Buffer detected, doing sync decode, len:", bb.Len()) + } + b := bb.Bytes() + var dst []byte + if cap(d.current.b) > 0 { + dst = d.current.b + } + + dst, err := d.DecodeAll(b, dst[:0]) + if err == nil { + err = io.EOF + } + d.current.b = dst + d.current.err = err + d.current.flushed = true + if debug { + println("sync decode to", len(dst), "bytes, err:", err) + } + return nil + } + + // Remove current block. + 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, + } + 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) + d.current.cancel = nil + } + if d.current.d != nil { + if debug { + printf("re-adding current decoder %p, decoders: %d", d.current.d, len(d.decoders)) + } + d.decoders <- d.current.d + d.current.d = nil + d.current.b = nil + } + if d.current.output == nil || d.current.flushed { + println("current already flushed") + return + } + for { + select { + case v := <-d.current.output: + if v.d != nil { + if debug { + printf("re-adding decoder %p", v.d) + } + d.decoders <- v.d + } + if v.err == errEndOfStream { + println("current flushed") + d.current.flushed = true + return + } + } + } +} + +// 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, errors.New("no input has been initialized") + } + 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 { + d.current.err = err2 + break + } + } + if d.current.err != nil { + break + } + d.nextBlock(true) + } + err := d.current.err + if err != nil { + d.drainOutput() + } + if err == io.EOF { + err = nil + } + return n, err +} + +// DecodeAll allows stateless decoding of a blob of bytes. +// Output will be appended to dst, so if the destination size is known +// you can pre-allocate the destination slice to avoid allocations. +// 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 { + return dst, ErrDecoderClosed + } + + // Grab a block decoder and frame decoder. + block := <-d.decoders + frame := block.localFrame + defer func() { + if debug { + printf("re-adding decoder: %p", block) + } + frame.rawInput = nil + frame.bBuf = nil + d.decoders <- block + }() + frame.bBuf = input + + for { + frame.history.reset() + err := frame.reset(&frame.bBuf) + if err == io.EOF { + if debug { + println("frame reset return EOF") + } + return dst, nil + } + if frame.DictionaryID != nil { + dict, ok := d.dicts[*frame.DictionaryID] + if !ok { + return nil, ErrUnknownDictionary + } + frame.history.setDict(&dict) + } + if err != nil { + return dst, err + } + 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 cap(dst)-len(dst) < int(frame.FrameContentSize) { + dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize)) + 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. + size := len(input) * 2 + // Cap to 1 MB. + if size > 1<<20 { + size = 1 << 20 + } + if uint64(size) > d.o.maxDecodedSize { + size = int(d.o.maxDecodedSize) + } + dst = make([]byte, 0, size) + } + + dst, err = frame.runDecoder(dst, block) + if err != nil { + return dst, err + } + if len(frame.bBuf) == 0 { + if debug { + println("frame dbuf empty") + } + break + } + } + return dst, nil +} + +// nextBlock returns the next block. +// If an error occurs d.err will be set. +// Optionally the function can block for new output. +// 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.d != nil { + if debug { + 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. +// It is NOT possible to reuse the decoder after this. +func (d *Decoder) Close() { + if d.current.err == ErrDecoderClosed { + return + } + d.drainOutput() + if d.stream != nil { + close(d.stream) + d.streamWg.Wait() + d.stream = nil + } + if d.decoders != nil { + close(d.decoders) + for dec := range d.decoders { + dec.Close() + } + d.decoders = nil + } + if d.current.d != nil { + d.current.d.Close() + d.current.d = nil + } + d.current.err = ErrDecoderClosed +} + +// IOReadCloser returns the decoder as an io.ReadCloser for convenience. +// Any changes to the decoder will be reflected, so the returned ReadCloser +// can be reused along with the decoder. +// io.WriterTo is also supported by the returned ReadCloser. +func (d *Decoder) IOReadCloser() io.ReadCloser { + return closeWrapper{d: d} +} + +// closeWrapper wraps a function call as a closer. +type closeWrapper struct { + d *Decoder +} + +// WriteTo forwards WriteTo calls to the decoder. +func (c closeWrapper) WriteTo(w io.Writer) (n int64, err error) { + return c.d.WriteTo(w) +} + +// Read forwards read calls to the decoder. +func (c closeWrapper) Read(p []byte) (n int, err error) { + return c.d.Read(p) +} + +// Close closes the decoder. +func (c closeWrapper) Close() error { + c.d.Close() + return nil +} + +type decodeOutput struct { + d *blockDec + b []byte + 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{} +} + +// 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) { + defer d.streamWg.Done() + frame := newFrameDec(d.o) + for stream := range inStream { + if debug { + println("got new stream") + } + br := readerWrapper{r: stream.r} + decodeStream: + for { + frame.history.reset() + err := frame.reset(&br) + if debug && 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 { + stream.output <- decodeOutput{ + err: err, + } + 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} + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/decoder_options.go b/vendor/github.com/klauspost/compress/zstd/decoder_options.go new file mode 100644 index 000000000000..284d384492b1 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/decoder_options.go @@ -0,0 +1,84 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "runtime" +) + +// DOption is an option for creating a decoder. +type DOption func(*decoderOptions) error + +// options retains accumulated state of multiple options. +type decoderOptions struct { + lowMem bool + concurrent int + maxDecodedSize uint64 + dicts []dict +} + +func (o *decoderOptions) setDefault() { + *o = decoderOptions{ + // use less ram: true for now, but may change. + lowMem: true, + concurrent: runtime.GOMAXPROCS(0), + } + o.maxDecodedSize = 1 << 63 +} + +// WithDecoderLowmem will set whether to use a lower amount of memory, +// but possibly have to allocate more while running. +func WithDecoderLowmem(b bool) DOption { + 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. +func WithDecoderConcurrency(n int) DOption { + return func(o *decoderOptions) error { + if n <= 0 { + return fmt.Errorf("Concurrency must be at least 1") + } + o.concurrent = n + return nil + } +} + +// 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. +func WithDecoderMaxMemory(n uint64) DOption { + return func(o *decoderOptions) error { + if n == 0 { + return errors.New("WithDecoderMaxMemory must be at least 1") + } + if n > 1<<63 { + return fmt.Errorf("WithDecoderMaxmemory must be less than 1 << 63") + } + o.maxDecodedSize = n + return nil + } +} + +// WithDecoderDicts allows to register one or more dictionaries for the decoder. +// If several dictionaries with the same ID is provided the last one will be used. +func WithDecoderDicts(dicts ...[]byte) DOption { + return func(o *decoderOptions) error { + for _, b := range dicts { + d, err := loadDict(b) + if err != nil { + return err + } + o.dicts = append(o.dicts, *d) + } + return nil + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/dict.go b/vendor/github.com/klauspost/compress/zstd/dict.go new file mode 100644 index 000000000000..fa25a18d8644 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/dict.go @@ -0,0 +1,122 @@ +package zstd + +import ( + "bytes" + "encoding/binary" + "errors" + "fmt" + "io" + + "github.com/klauspost/compress/huff0" +) + +type dict struct { + id uint32 + + litEnc *huff0.Scratch + llDec, ofDec, mlDec sequenceDec + //llEnc, ofEnc, mlEnc []*fseEncoder + offsets [3]int + content []byte +} + +var dictMagic = [4]byte{0x37, 0xa4, 0x30, 0xec} + +// ID returns the dictionary id or 0 if d is nil. +func (d *dict) ID() uint32 { + if d == nil { + return 0 + } + return d.id +} + +// DictContentSize returns the dictionary content size or 0 if d is nil. +func (d *dict) DictContentSize() int { + if d == nil { + return 0 + } + return len(d.content) +} + +// Load a dictionary as described in +// https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format +func loadDict(b []byte) (*dict, error) { + // Check static field size. + if len(b) <= 8+(3*4) { + return nil, io.ErrUnexpectedEOF + } + d := dict{ + llDec: sequenceDec{fse: &fseDecoder{}}, + ofDec: sequenceDec{fse: &fseDecoder{}}, + mlDec: sequenceDec{fse: &fseDecoder{}}, + } + if !bytes.Equal(b[:4], dictMagic[:]) { + return nil, ErrMagicMismatch + } + d.id = binary.LittleEndian.Uint32(b[4:8]) + if d.id == 0 { + return nil, errors.New("dictionaries cannot have ID 0") + } + + // Read literal table + var err error + d.litEnc, b, err = huff0.ReadTable(b[8:], nil) + if err != nil { + return nil, err + } + d.litEnc.Reuse = huff0.ReusePolicyMust + + br := byteReader{ + b: b, + off: 0, + } + readDec := func(i tableIndex, dec *fseDecoder) error { + if err := dec.readNCount(&br, uint16(maxTableSymbol[i])); err != nil { + return err + } + if br.overread() { + return io.ErrUnexpectedEOF + } + err = dec.transform(symbolTableX[i]) + if err != nil { + println("Transform table error:", err) + return err + } + if debug { + println("Read table ok", "symbolLen:", dec.symbolLen) + } + // Set decoders as predefined so they aren't reused. + dec.preDefined = true + return nil + } + + if err := readDec(tableOffsets, d.ofDec.fse); err != nil { + return nil, err + } + if err := readDec(tableMatchLengths, d.mlDec.fse); err != nil { + return nil, err + } + if err := readDec(tableLiteralLengths, d.llDec.fse); err != nil { + return nil, err + } + if br.remain() < 12 { + return nil, io.ErrUnexpectedEOF + } + + d.offsets[0] = int(br.Uint32()) + br.advance(4) + d.offsets[1] = int(br.Uint32()) + br.advance(4) + d.offsets[2] = int(br.Uint32()) + br.advance(4) + if d.offsets[0] <= 0 || d.offsets[1] <= 0 || d.offsets[2] <= 0 { + return nil, errors.New("invalid offset in dictionary") + } + d.content = make([]byte, br.remain()) + copy(d.content, br.unread()) + if d.offsets[0] > len(d.content) || d.offsets[1] > len(d.content) || d.offsets[2] > len(d.content) { + return nil, fmt.Errorf("initial offset bigger than dictionary content size %d, offsets: %v", len(d.content), d.offsets) + } + + return &d, nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_base.go b/vendor/github.com/klauspost/compress/zstd/enc_base.go new file mode 100644 index 000000000000..b1b7c6e6a72d --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_base.go @@ -0,0 +1,155 @@ +package zstd + +import ( + "fmt" + "math/bits" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +type fastBase struct { + // cur is the offset at the start of hist + cur int32 + // maximum offset. Should be at least 2x block size. + maxMatchOff int32 + hist []byte + crc *xxhash.Digest + tmp [8]byte + blk *blockEnc + lastDictID uint32 +} + +// CRC returns the underlying CRC writer. +func (e *fastBase) CRC() *xxhash.Digest { + return e.crc +} + +// AppendCRC will append the CRC to the destination slice and return it. +func (e *fastBase) AppendCRC(dst []byte) []byte { + crc := e.crc.Sum(e.tmp[:0]) + dst = append(dst, crc[7], crc[6], crc[5], crc[4]) + return dst +} + +// 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) { + b := int32(1) << uint(bits.Len(uint(size))) + // Keep minimum window. + if b < 1024 { + b = 1024 + } + return b + } + return e.maxMatchOff +} + +// Block returns the current block. +func (e *fastBase) Block() *blockEnc { + return e.blk +} + +func (e *fastBase) addBlock(src []byte) int32 { + if debugAsserts && e.cur > bufferReset { + panic(fmt.Sprintf("ecur (%d) > buffer reset (%d)", e.cur, bufferReset)) + } + // check if we have space already + if len(e.hist)+len(src) > cap(e.hist) { + if cap(e.hist) == 0 { + l := e.maxMatchOff * 2 + // Make it at least 1MB. + if l < 1<<20 { + l = 1 << 20 + } + e.hist = make([]byte, 0, l) + } else { + if cap(e.hist) < int(e.maxMatchOff*2) { + panic("unexpected buffer size") + } + // Move down + offset := int32(len(e.hist)) - e.maxMatchOff + copy(e.hist[0:e.maxMatchOff], e.hist[offset:]) + e.cur += offset + e.hist = e.hist[:e.maxMatchOff] + } + } + s := int32(len(e.hist)) + e.hist = append(e.hist, src...) + return s +} + +// useBlock will replace the block with the provided one, +// but transfer recent offsets from the previous. +func (e *fastBase) UseBlock(enc *blockEnc) { + enc.reset(e.blk) + 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 { + err := fmt.Sprintf("s (%d) < 0", s) + panic(err) + } + if t < 0 { + err := fmt.Sprintf("s (%d) < 0", s) + panic(err) + } + if s-t > e.maxMatchOff { + err := fmt.Sprintf("s (%d) - t (%d) > maxMatchOff (%d)", s, t, e.maxMatchOff) + panic(err) + } + if len(src)-int(s) > maxCompressedBlockSize { + panic(fmt.Sprintf("len(src)-s (%d) > maxCompressedBlockSize (%d)", len(src)-int(s), maxCompressedBlockSize)) + } + } + + // Extend the match to be as long as possible. + return int32(matchLen(src[s:], src[t:])) +} + +// Reset the encoding table. +func (e *fastBase) resetBase(d *dict, singleBlock bool) { + if e.blk == nil { + e.blk = &blockEnc{} + e.blk.init() + } else { + e.blk.reset(nil) + } + e.blk.initNewEncode() + if e.crc == nil { + e.crc = xxhash.New() + } else { + e.crc.Reset() + } + if (!singleBlock || d.DictContentSize() > 0) && cap(e.hist) < int(e.maxMatchOff*2)+d.DictContentSize() { + l := e.maxMatchOff*2 + int32(d.DictContentSize()) + // Make it at least 1MB. + if l < 1<<20 { + l = 1 << 20 + } + e.hist = make([]byte, 0, l) + } + // We offset current position so everything will be out of reach. + // If above reset line, history will be purged. + if e.cur < bufferReset { + e.cur += e.maxMatchOff + int32(len(e.hist)) + } + e.hist = e.hist[:0] + if d != nil { + // Set offsets (currently not used) + for i, off := range d.offsets { + e.blk.recentOffsets[i] = uint32(off) + e.blk.prevRecentOffsets[i] = e.blk.recentOffsets[i] + } + // Transfer litenc. + e.blk.dictLitEnc = d.litEnc + e.hist = append(e.hist, d.content...) + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_better.go b/vendor/github.com/klauspost/compress/zstd/enc_better.go new file mode 100644 index 000000000000..94a5343d00ec --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_better.go @@ -0,0 +1,595 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import "fmt" + +const ( + betterLongTableBits = 19 // Bits used in the long match table + betterLongTableSize = 1 << betterLongTableBits // Size of the table + + // 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. + betterShortTableBits = 13 // Bits used in the short match table + betterShortTableSize = 1 << betterShortTableBits // Size of the table +) + +type prevEntry struct { + offset int32 + prev int32 +} + +// betterFastEncoder 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. +// When we find a long match we choose between the two values and select the longest. +// When we find a short match, after checking the long, we check if we can find a long at n+1 +// and that it is longer (lazy matching). +type betterFastEncoder struct { + fastBase + table [betterShortTableSize]tableEntry + longTable [betterLongTableSize]prevEntry + dictTable []tableEntry + dictLongTable []prevEntry +} + +// Encode improves compression... +func (e *betterFastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = prevEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + v2 := e.longTable[i].prev + if v < minOff { + v = 0 + v2 = 0 + } else { + v = v - e.cur + e.maxMatchOff + if v2 < minOff { + v2 = 0 + } else { + v2 = v2 - e.cur + e.maxMatchOff + } + } + e.longTable[i] = prevEntry{ + offset: v, + prev: v2, + } + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 9 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debug { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + var matched int32 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHashS := hash5(cv, betterShortTableBits) + nextHashL := hash8(cv, betterLongTableBits) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + off := s + e.cur + e.longTable[nextHashL] = prevEntry{offset: off, prev: candidateL.offset} + e.table[nextHashS] = tableEntry{offset: off, val: uint32(cv)} + + if canRepeat { + if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Index match start+1 (long) -> s - 1 + index0 := s + repOff + s += lenght + repOff + + nextEmit = s + if s >= sLimit { + if debug { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + // Index skipped... + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hash8(cv0, betterLongTableBits) + 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)} + index0 += 2 + } + cv = load6432(src, s) + continue + } + const repOff2 = 1 + + // We deviate from the reference encoder and also check offset 2. + // Still slower and not much better, so disabled. + // repIndex = s - offset2 + repOff2 + if false && repIndex >= 0 && load6432(src, repIndex) == load6432(src, s+repOff) { + // Consider history as well. + var seq seq + lenght := 8 + e.matchlen(s+8+repOff2, repIndex+8, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 2 + seq.offset = 2 + if debugSequences { + println("repeat sequence 2", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + index0 := s + repOff2 + s += lenght + repOff2 + nextEmit = s + if s >= sLimit { + if debug { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + + // Index skipped... + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hash8(cv0, betterLongTableBits) + 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)} + index0 += 2 + } + cv = load6432(src, s) + // Swap offsets + offset1, offset2 = offset2, offset1 + continue + } + } + // Find the offsets of our two matches. + coffsetL := candidateL.offset - e.cur + coffsetLP := candidateL.prev - e.cur + + // Check if we have a long match. + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matched = e.matchlen(s+8, coffsetL+8, src) + 8 + t = coffsetL + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + + if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) { + // Found a long match, at least 8 bytes. + prevMatch := e.matchlen(s+8, coffsetLP+8, src) + 8 + if prevMatch > matched { + matched = prevMatch + t = coffsetLP + } + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + } + break + } + + // Check if we have a long match on prev. + if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) { + // Found a long match, at least 8 bytes. + matched = e.matchlen(s+8, coffsetLP+8, src) + 8 + t = coffsetLP + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + coffsetS := candidateS.offset - e.cur + + // Check if we have a short match. + if s-coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + matched = e.matchlen(s+4, coffsetS+4, src) + 4 + + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hash8(cv, betterLongTableBits) + candidateL = e.longTable[nextHashL] + coffsetL = candidateL.offset - e.cur + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = prevEntry{offset: s + checkAt + e.cur, prev: candidateL.offset} + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8 + if matchedNext > matched { + t = coffsetL + s += checkAt + matched = matchedNext + if debugMatches { + println("long match (after short)") + } + break + } + } + + // Check prev long... + coffsetL = candidateL.prev - e.cur + if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) { + // Found a long match, at least 8 bytes. + matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8 + if matchedNext > matched { + t = coffsetL + s += checkAt + matched = matchedNext + if debugMatches { + println("prev long match (after short)") + } + break + } + } + t = coffsetS + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the n-byte match as long as possible. + l := matched + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) -> s - 1 + index0 := s - l + 1 + for index0 < s-1 { + cv0 := load6432(src, index0) + cv1 := cv0 >> 8 + h0 := hash8(cv0, betterLongTableBits) + 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)} + index0 += 2 + } + + cv = load6432(src, s) + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashS := hash5(cv, betterShortTableBits) + nextHashL := hash8(cv, betterLongTableBits) + + // 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) + + e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset} + e.table[nextHashS] = tableEntry{offset: s + e.cur, val: uint32(cv)} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debug { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *betterFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + e.Encode(blk, src) +} + +// ResetDict will reset and set a dictionary if not nil +func (e *betterFastEncoder) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d == nil { + return + } + // Init or copy dict table + if len(e.dictTable) != len(e.table) || d.id != e.lastDictID { + if len(e.dictTable) != len(e.table) { + e.dictTable = make([]tableEntry, len(e.table)) + } + end := int32(len(d.content)) - 8 + e.maxMatchOff + for i := e.maxMatchOff; i < end; i += 4 { + 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 + e.dictTable[nextHash] = tableEntry{ + val: uint32(cv), + offset: i, + } + e.dictTable[nextHash1] = tableEntry{ + val: uint32(cv >> 8), + offset: i + 1, + } + e.dictTable[nextHash2] = tableEntry{ + val: uint32(cv >> 16), + offset: i + 2, + } + e.dictTable[nextHash3] = tableEntry{ + val: uint32(cv >> 24), + offset: i + 3, + } + } + e.lastDictID = d.id + } + + // Init or copy dict table + if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID { + if len(e.dictLongTable) != len(e.longTable) { + e.dictLongTable = make([]prevEntry, len(e.longTable)) + } + if len(d.content) >= 8 { + cv := load6432(d.content, 0) + h := hash8(cv, betterLongTableBits) + e.dictLongTable[h] = prevEntry{ + offset: e.maxMatchOff, + prev: e.dictLongTable[h].offset, + } + + end := int32(len(d.content)) - 8 + e.maxMatchOff + 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) + e.dictLongTable[h] = prevEntry{ + offset: i, + prev: e.dictLongTable[h].offset, + } + off++ + } + } + e.lastDictID = d.id + } + // Reset table to initial state + copy(e.longTable[:], e.dictLongTable) + + e.cur = e.maxMatchOff + // Reset table to initial state + copy(e.table[:], e.dictTable) +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_dfast.go b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go new file mode 100644 index 000000000000..19eebf66e502 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go @@ -0,0 +1,713 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import "fmt" + +const ( + 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. + + 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. +) + +type doubleFastEncoder struct { + fastEncoder + longTable [dFastLongTableSize]tableEntry + dictLongTable []tableEntry +} + +// Encode mimmics functionality in zstd_dfast.c +func (e *doubleFastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = tableEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + for i := range e.longTable[:] { + v := e.longTable[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.longTable[i].offset = v + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debug { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + // We allow the encoder to optionally turn off repeat offsets across blocks + canRepeat := len(blk.sequences) > 2 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHashS := hash5(cv, dFastShortTableBits) + nextHashL := hash8(cv, dFastLongTableBits) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + + if canRepeat { + if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + + seq.matchLen = uint32(lenght - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += lenght + repOff + nextEmit = s + if s >= sLimit { + if debug { + println("repeat ended", s, lenght) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + } + // Find the offsets of our two matches. + coffsetL := s - (candidateL.offset - e.cur) + coffsetS := s - (candidateS.offset - e.cur) + + // Check if we have a long match. + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + // Check if we have a short match. + if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hash8(cv, dFastLongTableBits) + candidateL = e.longTable[nextHashL] + coffsetL = s - (candidateL.offset - e.cur) + checkAt + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)} + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + s += checkAt + if debugMatches { + println("long match (after short)") + } + break + } + + t = candidateS.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // Extend the 4-byte match as long as possible. + l := e.matchlen(s+4, t+4, src) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) and start+2 (short) + index0 := s - l + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load6432(src, index0) + 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 + 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 + + cv = load6432(src, s) + + if !canRepeat { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashS := hash5(cv, dFastShortTableBits) + nextHashL := hash8(cv, dFastLongTableBits) + + // 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) + + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debug { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *doubleFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + const ( + // Input margin is the number of bytes we read (8) + // and the maximum we will read ahead (2) + inputMargin = 8 + 2 + minNonLiteralBlockSize = 16 + ) + + // Protect against e.cur wraparound. + if e.cur >= bufferReset { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.longTable[:] { + e.longTable[i] = tableEntry{} + } + e.cur = e.maxMatchOff + } + + s := int32(0) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 1. + const stepSize = 1 + + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debug { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + var t int32 + for { + + nextHashS := hash5(cv, dFastShortTableBits) + nextHashL := hash8(cv, dFastLongTableBits) + candidateL := e.longTable[nextHashL] + candidateS := e.table[nextHashS] + + const repOff = 1 + repIndex := s - offset1 + repOff + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + + if len(blk.sequences) > 2 { + if load3232(src, repIndex) == uint32(cv>>(repOff*8)) { + // Consider history as well. + var seq seq + //length := 4 + e.matchlen(s+4+repOff, repIndex+4, src) + length := 4 + int32(matchLen(src[s+4+repOff:], src[repIndex+4:])) + + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + repOff + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + repOff + nextEmit = s + if s >= sLimit { + if debug { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + } + // Find the offsets of our two matches. + coffsetL := s - (candidateL.offset - e.cur) + coffsetS := s - (candidateS.offset - e.cur) + + // Check if we have a long match. + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d). cur: %d", s, t, e.cur)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugMatches { + println("long match") + } + break + } + + // Check if we have a short match. + if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val { + // found a regular match + // See if we can find a long match at s+1 + const checkAt = 1 + cv := load6432(src, s+checkAt) + nextHashL = hash8(cv, dFastLongTableBits) + candidateL = e.longTable[nextHashL] + coffsetL = s - (candidateL.offset - e.cur) + checkAt + + // We can store it, since we have at least a 4 byte match. + e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)} + if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val { + // Found a long match, likely at least 8 bytes. + // Reference encoder checks all 8 bytes, we only check 4, + // but the likelihood of both the first 4 bytes and the hash matching should be enough. + t = candidateL.offset - e.cur + s += checkAt + if debugMatches { + println("long match (after short)") + } + break + } + + t = candidateS.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + if debugMatches { + println("short match") + } + break + } + + // No match found, move forward in input. + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + + // A 4-byte match has been found. Update recent offsets. + // We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + // Extend the 4-byte match as long as possible. + //l := e.matchlen(s+4, t+4, src) + 4 + l := int32(matchLen(src[s+4:], src[t+4:])) + 4 + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + + // Write our sequence + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + + // Index match start+1 (long) and start+2 (short) + index0 := s - l + 1 + // Index match end-2 (long) and end-1 (short) + index1 := s - 2 + + cv0 := load6432(src, index0) + 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 + 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 + + cv = load6432(src, s) + + if len(blk.sequences) <= 2 { + continue + } + + // Check offset 2 + for { + o2 := s - offset2 + if load3232(src, o2) != uint32(cv) { + // Do regular search + break + } + + // Store this, since we have it. + nextHashS := hash5(cv1>>8, dFastShortTableBits) + nextHashL := hash8(cv, dFastLongTableBits) + + // 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) + l := 4 + int32(matchLen(src[s+4:], src[o2+4:])) + + entry := tableEntry{offset: s + e.cur, val: uint32(cv)} + e.longTable[nextHashL] = entry + e.table[nextHashS] = entry + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + // Finished + break encodeLoop + } + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + if debug { + 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. + if e.cur < bufferReset { + e.cur += int32(len(src)) + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *doubleFastEncoder) Reset(d *dict, singleBlock bool) { + e.fastEncoder.Reset(d, singleBlock) + if d == nil { + return + } + + // Init or copy dict table + if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID { + if len(e.dictLongTable) != len(e.longTable) { + e.dictLongTable = make([]tableEntry, len(e.longTable)) + } + if len(d.content) >= 8 { + cv := load6432(d.content, 0) + e.dictLongTable[hash8(cv, dFastLongTableBits)] = 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{ + val: uint32(cv), + offset: i, + } + } + } + e.lastDictID = d.id + } + // Reset table to initial state + e.cur = e.maxMatchOff + copy(e.longTable[:], e.dictLongTable) +} diff --git a/vendor/github.com/klauspost/compress/zstd/enc_fast.go b/vendor/github.com/klauspost/compress/zstd/enc_fast.go new file mode 100644 index 000000000000..0b301df43908 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/enc_fast.go @@ -0,0 +1,661 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "fmt" + "math" + "math/bits" +) + +const ( + tableBits = 15 // Bits used in the table + tableSize = 1 << tableBits // Size of the table + tableMask = tableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks. + maxMatchLength = 131074 +) + +type tableEntry struct { + val uint32 + offset int32 +} + +type fastEncoder struct { + fastBase + table [tableSize]tableEntry + dictTable []tableEntry +} + +// Encode mimmics functionality in zstd_fast.c +func (e *fastEncoder) Encode(blk *blockEnc, src []byte) { + const ( + inputMargin = 8 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + + // Protect against e.cur wraparound. + for e.cur >= bufferReset { + if len(e.hist) == 0 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = e.maxMatchOff + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff + for i := range e.table[:] { + v := e.table[i].offset + if v < minOff { + v = 0 + } else { + v = v - e.cur + e.maxMatchOff + } + e.table[i].offset = v + } + e.cur = e.maxMatchOff + break + } + + s := e.addBlock(src) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + // Override src + src = e.hist + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 2. + const stepSize = 2 + + // TEMPLATE + const hashLog = tableBits + // seems global, but would be nice to tweak. + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debug { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + // t will contain the match offset when we find one. + // When existing the search loop, we have already checked 4 bytes. + var t int32 + + // We will not use repeat offsets across blocks. + // By not using them for the first 3 matches + canRepeat := len(blk.sequences) > 2 + + for { + if debugAsserts && canRepeat && offset1 == 0 { + panic("offset0 was 0") + } + + nextHash := hash6(cv, hashLog) + nextHash2 := hash6(cv>>8, hashLog) + candidate := e.table[nextHash] + candidate2 := e.table[nextHash2] + repIndex := s - offset1 + 2 + + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} + + if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) { + // 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 + } + } + } + + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + 2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + sMin := s - e.maxMatchOff + if sMin < 0 { + sMin = 0 + } + for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + 2 + nextEmit = s + if s >= sLimit { + if debug { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + coffset0 := s - (candidate.offset - e.cur) + coffset1 := s - (candidate2.offset - e.cur) + 1 + if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { + // found a regular match + t = candidate.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + break + } + + if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { + // found a regular match + t = candidate2.offset - e.cur + s++ + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + break + } + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + // A 4-byte match has been found. We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && canRepeat && int(offset1) > len(src) { + panic("invalid offset") + } + + // 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 + } + } + } + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength { + s-- + t-- + l++ + } + + // Write our sequence. + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + // Don't use repeat offsets + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + + // Check offset 2 + 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 + } + } + } + + // Store this, since we have it. + nextHash := hash6(cv, hashLog) + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + break encodeLoop + } + // Prepare next loop. + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + blk.recentOffsets[0] = uint32(offset1) + blk.recentOffsets[1] = uint32(offset2) + if debug { + println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits) + } +} + +// EncodeNoHist will encode a block with no history and no following blocks. +// Most notable difference is that src will not be copied for history and +// we do not need to check for max match length. +func (e *fastEncoder) EncodeNoHist(blk *blockEnc, src []byte) { + const ( + inputMargin = 8 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + if debug { + if len(src) > maxBlockSize { + panic("src too big") + } + } + + // Protect against e.cur wraparound. + if e.cur >= bufferReset { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = e.maxMatchOff + } + + s := int32(0) + blk.size = len(src) + if len(src) < minNonLiteralBlockSize { + blk.extraLits = len(src) + blk.literals = blk.literals[:len(src)] + copy(blk.literals, src) + return + } + + sLimit := int32(len(src)) - inputMargin + // stepSize is the number of bytes to skip on every main loop iteration. + // It should be >= 2. + const stepSize = 2 + + // TEMPLATE + const hashLog = tableBits + // seems global, but would be nice to tweak. + const kSearchStrength = 8 + + // nextEmit is where in src the next emitLiteral should start from. + nextEmit := s + cv := load6432(src, s) + + // Relative offsets + offset1 := int32(blk.recentOffsets[0]) + offset2 := int32(blk.recentOffsets[1]) + + addLiterals := func(s *seq, until int32) { + if until == nextEmit { + return + } + blk.literals = append(blk.literals, src[nextEmit:until]...) + s.litLen = uint32(until - nextEmit) + } + if debug { + println("recent offsets:", blk.recentOffsets) + } + +encodeLoop: + for { + // t will contain the match offset when we find one. + // When existing the search loop, we have already checked 4 bytes. + var t int32 + + // We will not use repeat offsets across blocks. + // By not using them for the first 3 matches + + for { + nextHash := hash6(cv, hashLog) + nextHash2 := hash6(cv>>8, hashLog) + candidate := e.table[nextHash] + candidate2 := e.table[nextHash2] + repIndex := s - offset1 + 2 + + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)} + + 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 + } + } + } + + seq.matchLen = uint32(length - zstdMinMatch) + + // We might be able to match backwards. + // Extend as long as we can. + start := s + 2 + // We end the search early, so we don't risk 0 literals + // and have to do special offset treatment. + startLimit := nextEmit + 1 + + sMin := s - e.maxMatchOff + if sMin < 0 { + sMin = 0 + } + for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] { + repIndex-- + start-- + seq.matchLen++ + } + addLiterals(&seq, start) + + // rep 0 + seq.offset = 1 + if debugSequences { + println("repeat sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + s += length + 2 + nextEmit = s + if s >= sLimit { + if debug { + println("repeat ended", s, length) + + } + break encodeLoop + } + cv = load6432(src, s) + continue + } + coffset0 := s - (candidate.offset - e.cur) + coffset1 := s - (candidate2.offset - e.cur) + 1 + if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val { + // found a regular match + t = candidate.offset - e.cur + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic(fmt.Sprintf("t (%d) < 0, candidate.offset: %d, e.cur: %d, coffset0: %d, e.maxMatchOff: %d", t, candidate.offset, e.cur, coffset0, e.maxMatchOff)) + } + break + } + + if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val { + // found a regular match + t = candidate2.offset - e.cur + s++ + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + if debugAsserts && s-t > e.maxMatchOff { + panic("s - t >e.maxMatchOff") + } + if debugAsserts && t < 0 { + panic("t<0") + } + break + } + s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1)) + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + } + // A 4-byte match has been found. We'll later see if more than 4 bytes. + offset2 = offset1 + offset1 = s - t + + if debugAsserts && s <= t { + panic(fmt.Sprintf("s (%d) <= t (%d)", s, t)) + } + + if debugAsserts && t < 0 { + 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 + } + } + } + + // Extend backwards + tMin := s - e.maxMatchOff + if tMin < 0 { + tMin = 0 + } + for t > tMin && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + + // Write our sequence. + var seq seq + seq.litLen = uint32(s - nextEmit) + seq.matchLen = uint32(l - zstdMinMatch) + if seq.litLen > 0 { + blk.literals = append(blk.literals, src[nextEmit:s]...) + } + // Don't use repeat offsets + seq.offset = uint32(s-t) + 3 + s += l + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + nextEmit = s + if s >= sLimit { + break encodeLoop + } + cv = load6432(src, s) + + // Check offset 2 + 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 + } + } + } + + // Store this, since we have it. + nextHash := hash6(cv, hashLog) + e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)} + seq.matchLen = uint32(l) - zstdMinMatch + seq.litLen = 0 + // Since litlen is always 0, this is offset 1. + seq.offset = 1 + s += l + nextEmit = s + if debugSequences { + println("sequence", seq, "next s:", s) + } + blk.sequences = append(blk.sequences, seq) + + // Swap offset 1 and 2. + offset1, offset2 = offset2, offset1 + if s >= sLimit { + break encodeLoop + } + // Prepare next loop. + cv = load6432(src, s) + } + } + + if int(nextEmit) < len(src) { + blk.literals = append(blk.literals, src[nextEmit:]...) + blk.extraLits = len(src) - int(nextEmit) + } + if debug { + 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. + if e.cur < bufferReset { + e.cur += int32(len(src)) + } +} + +// ResetDict will reset and set a dictionary if not nil +func (e *fastEncoder) Reset(d *dict, singleBlock bool) { + e.resetBase(d, singleBlock) + if d == nil { + return + } + + // Init or copy dict table + if len(e.dictTable) != len(e.table) || d.id != e.lastDictID { + if len(e.dictTable) != len(e.table) { + e.dictTable = make([]tableEntry, len(e.table)) + } + if true { + end := e.maxMatchOff + int32(len(d.content)) - 8 + for i := e.maxMatchOff; i < end; i += 3 { + 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 + e.dictTable[nextHash] = tableEntry{ + val: uint32(cv), + offset: i, + } + e.dictTable[nextHash1] = tableEntry{ + val: uint32(cv >> 8), + offset: i + 1, + } + e.dictTable[nextHash2] = tableEntry{ + val: uint32(cv >> 16), + offset: i + 2, + } + } + } + e.lastDictID = d.id + } + + e.cur = e.maxMatchOff + // Reset table to initial state + copy(e.table[:], e.dictTable) +} diff --git a/vendor/github.com/klauspost/compress/zstd/encoder.go b/vendor/github.com/klauspost/compress/zstd/encoder.go new file mode 100644 index 000000000000..f5759211dac5 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/encoder.go @@ -0,0 +1,570 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "crypto/rand" + "fmt" + "io" + rdebug "runtime/debug" + "sync" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +// Encoder provides encoding to Zstandard. +// An Encoder can be used for either compressing a stream via the +// io.WriteCloser interface supported by the Encoder or as multiple independent +// tasks via the EncodeAll function. +// Smaller encodes are encouraged to use the EncodeAll function. +// Use NewWriter to create a new instance. +type Encoder struct { + o encoderOptions + encoders chan encoder + state encoderState + init sync.Once +} + +type encoder interface { + Encode(blk *blockEnc, src []byte) + EncodeNoHist(blk *blockEnc, src []byte) + Block() *blockEnc + CRC() *xxhash.Digest + AppendCRC([]byte) []byte + WindowSize(size int) int32 + UseBlock(*blockEnc) + Reset(d *dict, singleBlock bool) +} + +type encoderState struct { + w io.Writer + filling []byte + current []byte + previous []byte + encoder encoder + writing *blockEnc + err error + writeErr error + nWritten int64 + headerWritten bool + eofWritten bool + fullFrameWritten bool + + // This waitgroup indicates an encode is running. + wg sync.WaitGroup + // This waitgroup indicates we have a block encoding/writing. + wWg sync.WaitGroup +} + +// NewWriter will create a new Zstandard encoder. +// If the encoder will be used for encoding blocks a nil writer can be used. +func NewWriter(w io.Writer, opts ...EOption) (*Encoder, error) { + initPredefined() + var e Encoder + e.o.setDefault() + for _, o := range opts { + err := o(&e.o) + if err != nil { + return nil, err + } + } + if w != nil { + e.Reset(w) + } + return &e, nil +} + +func (e *Encoder) initialize() { + if e.o.concurrent == 0 { + e.o.setDefault() + } + e.encoders = make(chan encoder, e.o.concurrent) + for i := 0; i < e.o.concurrent; i++ { + enc := e.o.encoder() + e.encoders <- enc + } +} + +// Reset will re-initialize the writer and new writes will encode to the supplied writer +// as a new, independent stream. +func (e *Encoder) Reset(w io.Writer) { + s := &e.state + s.wg.Wait() + s.wWg.Wait() + 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 s.encoder == nil { + s.encoder = e.o.encoder() + } + if s.writing == nil { + s.writing = &blockEnc{} + 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 + s.fullFrameWritten = false + s.w = w + s.err = nil + s.nWritten = 0 + s.writeErr = nil +} + +// Write data to the encoder. +// Input data will be buffered and as the buffer fills up +// content will be compressed and written to the output. +// When done writing, use Close to flush the remaining output +// and write CRC if requested. +func (e *Encoder) Write(p []byte) (n int, err error) { + s := &e.state + for len(p) > 0 { + if len(p)+len(s.filling) < e.o.blockSize { + if e.o.crc { + _, _ = s.encoder.CRC().Write(p) + } + s.filling = append(s.filling, p...) + return n + len(p), nil + } + add := p + if len(p)+len(s.filling) > e.o.blockSize { + add = add[:e.o.blockSize-len(s.filling)] + } + if e.o.crc { + _, _ = s.encoder.CRC().Write(add) + } + s.filling = append(s.filling, add...) + p = p[len(add):] + n += len(add) + if len(s.filling) < e.o.blockSize { + return n, nil + } + err := e.nextBlock(false) + if err != nil { + return n, err + } + if debugAsserts && len(s.filling) > 0 { + panic(len(s.filling)) + } + } + return n, nil +} + +// nextBlock will synchronize and start compressing input in e.state.filling. +// If an error has occurred during encoding it will be returned. +func (e *Encoder) nextBlock(final bool) error { + s := &e.state + // Wait for current block. + s.wg.Wait() + if s.err != nil { + return s.err + } + if len(s.filling) > e.o.blockSize { + return fmt.Errorf("block > maxStoreBlockSize") + } + if !s.headerWritten { + // If we have a single block encode, do a sync compression. + if final && len(s.filling) > 0 { + s.current = e.EncodeAll(s.filling, s.current[:0]) + var n2 int + n2, s.err = s.w.Write(s.current) + if s.err != nil { + return s.err + } + s.nWritten += int64(n2) + s.current = s.current[:0] + s.filling = s.filling[:0] + s.headerWritten = true + s.fullFrameWritten = true + s.eofWritten = true + return nil + } + + var tmp [maxHeaderSize]byte + fh := frameHeader{ + ContentSize: 0, + WindowSize: uint32(s.encoder.WindowSize(0)), + SingleSegment: false, + Checksum: e.o.crc, + DictID: e.o.dict.ID(), + } + + dst, err := fh.appendTo(tmp[:0]) + if err != nil { + return err + } + s.headerWritten = true + s.wWg.Wait() + var n2 int + n2, s.err = s.w.Write(dst) + if s.err != nil { + return s.err + } + s.nWritten += int64(n2) + } + if s.eofWritten { + // Ensure we only write it once. + final = false + } + + if len(s.filling) == 0 { + // Final block, but no data. + if final { + enc := s.encoder + blk := enc.Block() + blk.reset(nil) + blk.last = true + blk.encodeRaw(nil) + s.wWg.Wait() + _, s.err = s.w.Write(blk.output) + s.nWritten += int64(len(blk.output)) + s.eofWritten = true + } + return s.err + } + + // Move blocks forward. + s.filling, s.current, s.previous = s.previous[:0], s.filling, s.current + s.wg.Add(1) + go func(src []byte) { + if debug { + println("Adding block,", len(src), "bytes, final:", final) + } + defer func() { + if r := recover(); r != nil { + s.err = fmt.Errorf("panic while encoding: %v", r) + rdebug.PrintStack() + } + s.wg.Done() + }() + enc := s.encoder + blk := enc.Block() + enc.Encode(blk, src) + blk.last = final + if final { + s.eofWritten = true + } + // Wait for pending writes. + s.wWg.Wait() + if s.writeErr != nil { + s.err = s.writeErr + return + } + // Transfer encoders from previous write block. + blk.swapEncoders(s.writing) + // Transfer recent offsets to next. + enc.UseBlock(s.writing) + s.writing = blk + s.wWg.Add(1) + go func() { + defer func() { + if r := recover(); r != nil { + s.writeErr = fmt.Errorf("panic while encoding/writing: %v", r) + rdebug.PrintStack() + } + s.wWg.Done() + }() + 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 debug { + 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.writeErr = err + return + } + _, s.writeErr = s.w.Write(blk.output) + s.nWritten += int64(len(blk.output)) + }() + }(s.current) + return nil +} + +// ReadFrom reads data from r until EOF or error. +// The return value n is the number of bytes read. +// Any error except io.EOF encountered during the read is also returned. +// +// The Copy function uses ReaderFrom if available. +func (e *Encoder) ReadFrom(r io.Reader) (n int64, err error) { + if debug { + println("Using ReadFrom") + } + + // Flush any current writes. + if len(e.state.filling) > 0 { + if err := e.nextBlock(false); err != nil { + return 0, err + } + } + e.state.filling = e.state.filling[:e.o.blockSize] + src := e.state.filling + for { + n2, err := r.Read(src) + if e.o.crc { + _, _ = e.state.encoder.CRC().Write(src[:n2]) + } + // src is now the unfilled part... + src = src[n2:] + n += int64(n2) + switch err { + case io.EOF: + e.state.filling = e.state.filling[:len(e.state.filling)-len(src)] + if debug { + println("ReadFrom: got EOF final block:", len(e.state.filling)) + } + return n, nil + default: + if debug { + println("ReadFrom: got error:", err) + } + e.state.err = err + return n, err + case nil: + } + if len(src) > 0 { + if debug { + println("ReadFrom: got space left in source:", len(src)) + } + continue + } + err = e.nextBlock(false) + if err != nil { + return n, err + } + e.state.filling = e.state.filling[:e.o.blockSize] + src = e.state.filling + } +} + +// Flush will send the currently written data to output +// and block until everything has been written. +// This should only be used on rare occasions where pushing the currently queued data is critical. +func (e *Encoder) Flush() error { + s := &e.state + if len(s.filling) > 0 { + err := e.nextBlock(false) + if err != nil { + return err + } + } + s.wg.Wait() + s.wWg.Wait() + if s.err != nil { + return s.err + } + return s.writeErr +} + +// Close will flush the final output and close the stream. +// The function will block until everything has been written. +// The Encoder can still be re-used after calling this. +func (e *Encoder) Close() error { + s := &e.state + if s.encoder == nil { + return nil + } + err := e.nextBlock(true) + if err != nil { + return err + } + if e.state.fullFrameWritten { + return s.err + } + s.wg.Wait() + s.wWg.Wait() + + if s.err != nil { + return s.err + } + if s.writeErr != nil { + return s.writeErr + } + + // Write CRC + if e.o.crc && s.err == nil { + // heap alloc. + var tmp [4]byte + _, s.err = s.w.Write(s.encoder.AppendCRC(tmp[:0])) + s.nWritten += 4 + } + + // Add padding with content from crypto/rand.Reader + if s.err == nil && e.o.pad > 0 { + add := calcSkippableFrame(s.nWritten, int64(e.o.pad)) + frame, err := skippableFrame(s.filling[:0], add, rand.Reader) + if err != nil { + return err + } + _, s.err = s.w.Write(frame) + } + return s.err +} + +// 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. +// If empty input is given, nothing is returned, unless WithZeroFrames is specified. +// 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. +func (e *Encoder) EncodeAll(src, dst []byte) []byte { + if len(src) == 0 { + if e.o.fullZero { + // Add frame header. + fh := frameHeader{ + ContentSize: 0, + WindowSize: MinWindowSize, + SingleSegment: true, + // Adding a checksum would be a waste of space. + Checksum: false, + DictID: 0, + } + dst, _ = fh.appendTo(dst) + + // Write raw block as last one only. + var blk blockHeader + blk.setSize(0) + blk.setType(blockTypeRaw) + blk.setLast(true) + dst = blk.appendTo(dst) + } + return dst + } + e.init.Do(e.initialize) + enc := <-e.encoders + defer func() { + // Release encoder reference to last block. + // 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 + if e.o.single != nil { + single = *e.o.single + } + fh := frameHeader{ + ContentSize: uint64(len(src)), + WindowSize: uint32(enc.WindowSize(len(src))), + SingleSegment: single, + Checksum: e.o.crc, + DictID: e.o.dict.ID(), + } + + // If less than 1MB, allocate a buffer up front. + if len(dst) == 0 && cap(dst) == 0 && len(src) < 1<<20 { + dst = make([]byte, 0, len(src)) + } + dst, err := fh.appendTo(dst) + if err != nil { + panic(err) + } + + // If we can do everything in one block, prefer that. + if len(src) <= maxCompressedBlockSize { + enc.Reset(e.o.dict, true) + // Slightly faster with no history and everything in one block. + if e.o.crc { + _, _ = enc.CRC().Write(src) + } + blk := enc.Block() + blk.last = true + if e.o.dict == nil { + enc.EncodeNoHist(blk, src) + } else { + enc.Encode(blk, src) + } + + // If we got the exact same number of literals as input, + // assume the literals cannot be compressed. + err := errIncompressible + oldout := blk.output + if len(blk.literals) != len(src) || len(src) != e.o.blockSize { + // Output directly to dst + blk.output = dst + err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy) + } + + switch err { + case errIncompressible: + if debug { + println("Storing incompressible block as raw") + } + dst = blk.encodeRawTo(dst, src) + case nil: + dst = blk.output + default: + panic(err) + } + blk.output = oldout + } else { + enc.Reset(e.o.dict, false) + blk := enc.Block() + for len(src) > 0 { + todo := src + if len(todo) > e.o.blockSize { + todo = todo[:e.o.blockSize] + } + src = src[len(todo):] + if e.o.crc { + _, _ = enc.CRC().Write(todo) + } + blk.pushOffsets() + enc.Encode(blk, todo) + if len(src) == 0 { + blk.last = true + } + err := errIncompressible + // If we got the exact same number of literals as input, + // assume the literals cannot be compressed. + if len(blk.literals) != len(todo) || len(todo) != e.o.blockSize { + err = blk.encode(todo, e.o.noEntropy, !e.o.allLitEntropy) + } + + switch err { + case errIncompressible: + if debug { + println("Storing incompressible block as raw") + } + dst = blk.encodeRawTo(dst, todo) + blk.popOffsets() + case nil: + dst = append(dst, blk.output...) + default: + panic(err) + } + blk.reset(nil) + } + } + if e.o.crc { + dst = enc.AppendCRC(dst) + } + // Add padding with content from crypto/rand.Reader + if e.o.pad > 0 { + add := calcSkippableFrame(int64(len(dst)), int64(e.o.pad)) + dst, err = skippableFrame(dst, add, rand.Reader) + if err != nil { + panic(err) + } + } + return dst +} diff --git a/vendor/github.com/klauspost/compress/zstd/encoder_options.go b/vendor/github.com/klauspost/compress/zstd/encoder_options.go new file mode 100644 index 000000000000..1209915bcce7 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/encoder_options.go @@ -0,0 +1,282 @@ +package zstd + +import ( + "errors" + "fmt" + "runtime" + "strings" +) + +// EOption is an option for creating a encoder. +type EOption func(*encoderOptions) error + +// options retains accumulated state of multiple options. +type encoderOptions struct { + concurrent int + level EncoderLevel + single *bool + pad int + blockSize int + windowSize int + crc bool + fullZero bool + noEntropy bool + allLitEntropy bool + customWindow bool + customALEntropy bool + dict *dict +} + +func (o *encoderOptions) setDefault() { + *o = encoderOptions{ + // use less ram: true for now, but may change. + concurrent: runtime.GOMAXPROCS(0), + crc: true, + single: nil, + blockSize: 1 << 16, + windowSize: 8 << 20, + level: SpeedDefault, + allLitEntropy: true, + } +} + +// encoder returns an encoder with the selected options. +func (o encoderOptions) encoder() encoder { + switch o.level { + case SpeedDefault: + return &doubleFastEncoder{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize)}}} + case SpeedBetterCompression: + return &betterFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize)}} + case SpeedFastest: + return &fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize)}} + } + panic("unknown compression level") +} + +// WithEncoderCRC will add CRC value to output. +// Output will be 4 bytes larger. +func WithEncoderCRC(b bool) EOption { + return func(o *encoderOptions) error { o.crc = b; return nil } +} + +// WithEncoderConcurrency 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. +func WithEncoderConcurrency(n int) EOption { + return func(o *encoderOptions) error { + if n <= 0 { + return fmt.Errorf("concurrency must be at least 1") + } + o.concurrent = n + return nil + } +} + +// WithWindowSize will set the maximum allowed back-reference distance. +// The value must be a power of two between MinWindowSize and MaxWindowSize. +// A larger value will enable better compression but allocate more memory and, +// for above-default values, take considerably longer. +// The default value is determined by the compression level. +func WithWindowSize(n int) EOption { + return func(o *encoderOptions) error { + switch { + case n < MinWindowSize: + return fmt.Errorf("window size must be at least %d", MinWindowSize) + case n > MaxWindowSize: + return fmt.Errorf("window size must be at most %d", MaxWindowSize) + case (n & (n - 1)) != 0: + return errors.New("window size must be a power of 2") + } + + o.windowSize = n + o.customWindow = true + if o.blockSize > o.windowSize { + o.blockSize = o.windowSize + } + return nil + } +} + +// WithEncoderPadding will add padding to all output so the size will be a multiple of n. +// This can be used to obfuscate the exact output size or make blocks of a certain size. +// The contents will be a skippable frame, so it will be invisible by the decoder. +// n must be > 0 and <= 1GB, 1<<30 bytes. +// The padded area will be filled with data from crypto/rand.Reader. +// If `EncodeAll` is used with data already in the destination, the total size will be multiple of this. +func WithEncoderPadding(n int) EOption { + return func(o *encoderOptions) error { + if n <= 0 { + return fmt.Errorf("padding must be at least 1") + } + // No need to waste our time. + if n == 1 { + o.pad = 0 + } + if n > 1<<30 { + return fmt.Errorf("padding must less than 1GB (1<<30 bytes) ") + } + o.pad = n + return nil + } +} + +// EncoderLevel predefines encoder compression levels. +// Only use the constants made available, since the actual mapping +// of these values are very likely to change and your compression could change +// unpredictably when upgrading the library. +type EncoderLevel int + +const ( + speedNotSet EncoderLevel = iota + + // SpeedFastest will choose the fastest reasonable compression. + // This is roughly equivalent to the fastest Zstandard mode. + SpeedFastest + + // SpeedDefault is the default "pretty fast" compression option. + // This is roughly equivalent to the default Zstandard mode (level 3). + SpeedDefault + + // SpeedBetterCompression will yield better compression than the default. + // Currently it is about zstd level 7-8 with ~ 2x-3x the default CPU usage. + // By using this, notice that CPU usage may go up in the future. + SpeedBetterCompression + + // speedLast should be kept as the last actual compression option. + // The is not for external usage, but is used to keep track of the valid options. + speedLast + + // SpeedBestCompression will choose the best available compression option. + // For now this is not implemented. + SpeedBestCompression = SpeedBetterCompression +) + +// EncoderLevelFromString will convert a string representation of an encoding level back +// to a compression level. The compare is not case sensitive. +// If the string wasn't recognized, (false, SpeedDefault) will be returned. +func EncoderLevelFromString(s string) (bool, EncoderLevel) { + for l := EncoderLevel(speedNotSet + 1); l < speedLast; l++ { + if strings.EqualFold(s, l.String()) { + return true, l + } + } + return false, SpeedDefault +} + +// EncoderLevelFromZstd will return an encoder level that closest matches the compression +// ratio of a specific zstd compression level. +// Many input values will provide the same compression level. +func EncoderLevelFromZstd(level int) EncoderLevel { + switch { + case level < 3: + return SpeedFastest + case level >= 3 && level < 6: + return SpeedDefault + case level > 5: + return SpeedBetterCompression + } + return SpeedDefault +} + +// String provides a string representation of the compression level. +func (e EncoderLevel) String() string { + switch e { + case SpeedFastest: + return "fastest" + case SpeedDefault: + return "default" + case SpeedBetterCompression: + return "better" + default: + return "invalid" + } +} + +// WithEncoderLevel specifies a predefined compression level. +func WithEncoderLevel(l EncoderLevel) EOption { + return func(o *encoderOptions) error { + switch { + case l <= speedNotSet || l >= speedLast: + return fmt.Errorf("unknown encoder level") + } + o.level = l + if !o.customWindow { + switch o.level { + case SpeedFastest: + o.windowSize = 4 << 20 + case SpeedDefault: + o.windowSize = 8 << 20 + case SpeedBetterCompression: + o.windowSize = 16 << 20 + } + } + if !o.customALEntropy { + o.allLitEntropy = l > SpeedFastest + } + + return nil + } +} + +// WithZeroFrames will encode 0 length input as full frames. +// This can be needed for compatibility with zstandard usage, +// but is not needed for this package. +func WithZeroFrames(b bool) EOption { + return func(o *encoderOptions) error { + o.fullZero = b + return nil + } +} + +// WithAllLitEntropyCompression will apply entropy compression if no matches are found. +// Disabling this will skip incompressible data faster, but in cases with no matches but +// skewed character distribution compression is lost. +// Default value depends on the compression level selected. +func WithAllLitEntropyCompression(b bool) EOption { + return func(o *encoderOptions) error { + o.customALEntropy = true + o.allLitEntropy = b + return nil + } +} + +// WithNoEntropyCompression will always skip entropy compression of literals. +// This can be useful if content has matches, but unlikely to benefit from entropy +// compression. Usually the slight speed improvement is not worth enabling this. +func WithNoEntropyCompression(b bool) EOption { + return func(o *encoderOptions) error { + o.noEntropy = b + return nil + } +} + +// WithSingleSegment will set the "single segment" flag when EncodeAll is used. +// If this flag is set, data must be regenerated within a single continuous memory segment. +// In this case, Window_Descriptor byte is skipped, but Frame_Content_Size is necessarily present. +// As a consequence, the decoder must allocate a memory segment of size equal or larger than size of your content. +// In order to preserve the decoder from unreasonable memory requirements, +// 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. +// This setting has no effect on streamed encodes. +func WithSingleSegment(b bool) EOption { + return func(o *encoderOptions) error { + o.single = &b + return nil + } +} + +// WithEncoderDict allows to register a dictionary that will be used for the encode. +// The encoder *may* choose to use no dictionary instead for certain payloads. +func WithEncoderDict(dict []byte) EOption { + return func(o *encoderOptions) error { + d, err := loadDict(dict) + if err != nil { + return err + } + o.dict = d + return nil + } +} diff --git a/vendor/github.com/klauspost/compress/zstd/framedec.go b/vendor/github.com/klauspost/compress/zstd/framedec.go new file mode 100644 index 000000000000..fc4a566d39aa --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/framedec.go @@ -0,0 +1,494 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "bytes" + "encoding/hex" + "errors" + "hash" + "io" + "sync" + + "github.com/klauspost/compress/zstd/internal/xxhash" +) + +type frameDec struct { + o decoderOptions + crc hash.Hash64 + offset int64 + + 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 + + rawInput byteBuffer + + // Byte buffer that can be reused for small input blocks. + 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 = 1 << 10 + MaxWindowSize = 1 << 29 +) + +var ( + frameMagic = []byte{0x28, 0xb5, 0x2f, 0xfd} + skippableFrameMagic = []byte{0x2a, 0x4d, 0x18} +) + +func newFrameDec(o decoderOptions) *frameDec { + d := frameDec{ + o: o, + maxWindowSize: MaxWindowSize, + } + if d.maxWindowSize > o.maxDecodedSize { + d.maxWindowSize = o.maxDecodedSize + } + return &d +} + +// reset will read the frame header and prepare for block decoding. +// If nothing can be read from the input, io.EOF will be returned. +// Any other error indicated that the stream contained data, but +// there was a problem. +func (d *frameDec) reset(br byteBuffer) error { + d.HasCheckSum = false + d.WindowSize = 0 + var b []byte + for { + b = br.readSmall(4) + if b == nil { + return io.EOF + } + if !bytes.Equal(b[1:4], skippableFrameMagic) || b[0]&0xf0 != 0x50 { + if debug { + println("Not skippable", hex.EncodeToString(b), 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 + } + 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)) + if err != nil { + if debug { + println("Reading discarded frame", err) + } + return err + } + } + if !bytes.Equal(b, frameMagic) { + println("Got magic numbers: ", b, "want:", frameMagic) + return ErrMagicMismatch + } + + // Read Frame_Header_Descriptor + fhd, err := br.readByte() + if err != nil { + println("Reading Frame_Header_Descriptor", err) + return err + } + d.SingleSegment = fhd&(1<<5) != 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 + d.WindowSize = 0 + if !d.SingleSegment { + wd, err := br.readByte() + if err != nil { + println("Reading Window_Descriptor", err) + return err + } + printf("raw: %x, mantissa: %d, exponent: %d\n", wd, wd&7, wd>>3) + windowLog := 10 + (wd >> 3) + windowBase := uint64(1) << windowLog + windowAdd := (windowBase / 8) * uint64(wd&0x7) + d.WindowSize = windowBase + windowAdd + } + + // Read Dictionary_ID + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary_id + d.DictionaryID = nil + if size := fhd & 3; size != 0 { + if size == 3 { + size = 4 + } + b = br.readSmall(int(size)) + if b == nil { + if debug { + println("Reading Dictionary_ID", io.ErrUnexpectedEOF) + } + return io.ErrUnexpectedEOF + } + var id uint32 + switch size { + case 1: + id = uint32(b[0]) + case 2: + id = uint32(b[0]) | (uint32(b[1]) << 8) + case 4: + id = uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + } + if debug { + println("Dict size", size, "ID:", id) + } + if id > 0 { + // ID 0 means "sorry, no dictionary anyway". + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary-format + d.DictionaryID = &id + } + } + + // Read Frame_Content_Size + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frame_content_size + var fcsSize int + v := fhd >> 6 + switch v { + case 0: + if d.SingleSegment { + fcsSize = 1 + } + default: + fcsSize = 1 << v + } + d.FrameContentSize = 0 + if fcsSize > 0 { + b := br.readSmall(fcsSize) + if b == nil { + println("Reading Frame content", io.ErrUnexpectedEOF) + return io.ErrUnexpectedEOF + } + switch fcsSize { + case 1: + d.FrameContentSize = uint64(b[0]) + case 2: + // When FCS_Field_Size is 2, the offset of 256 is added. + d.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) + 256 + case 4: + d.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) | (uint64(b[2]) << 16) | (uint64(b[3]) << 24) + case 8: + d1 := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) + 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) + } + } + // Move this to shared. + d.HasCheckSum = fhd&(1<<2) != 0 + if d.HasCheckSum { + if d.crc == nil { + d.crc = xxhash.New() + } + d.crc.Reset() + } + + 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.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) + return ErrWindowSizeTooSmall + } + d.history.windowSize = int(d.WindowSize) + if d.o.lowMem && d.history.windowSize < maxBlockSize { + d.history.maxSize = d.history.windowSize * 2 + } else { + d.history.maxSize = d.history.windowSize + maxBlockSize + } + // history contains input - maybe we do something + d.rawInput = br + return nil +} + +// 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) + } + 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) + 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 + } + var tmp [4]byte + got := d.crc.Sum64() + // Flip to match file order. + tmp[0] = byte(got >> 0) + tmp[1] = byte(got >> 8) + 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 { + println("CRC Check Failed:", tmp[:], "!=", want) + } + return ErrCRCMismatch + } + if debug { + 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 + } + // 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 + } +} + +// runDecoder will create a sync decoder that will decode a block of data. +func (d *frameDec) runDecoder(dst []byte, dec *blockDec) ([]byte, error) { + saved := d.history.b + + // We use the history for output to avoid copying it. + d.history.b = dst + // Store input length, so we only check new data. + crcStart := len(dst) + var err error + for { + err = dec.reset(d.rawInput, d.WindowSize) + if err != nil { + break + } + if debug { + println("next block:", dec) + } + err = dec.decodeBuf(&d.history) + if err != nil || dec.Last { + 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) + err = ErrFrameSizeExceeded + break + } + } + 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() + } + } + } + } + d.history.b = saved + return dst, err +} diff --git a/vendor/github.com/klauspost/compress/zstd/frameenc.go b/vendor/github.com/klauspost/compress/zstd/frameenc.go new file mode 100644 index 000000000000..4ef7f5a3e3d5 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/frameenc.go @@ -0,0 +1,137 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "fmt" + "io" + "math" + "math/bits" +) + +type frameHeader struct { + ContentSize uint64 + WindowSize uint32 + SingleSegment bool + Checksum bool + DictID uint32 +} + +const maxHeaderSize = 14 + +func (f frameHeader) appendTo(dst []byte) ([]byte, error) { + dst = append(dst, frameMagic...) + var fhd uint8 + if f.Checksum { + fhd |= 1 << 2 + } + if f.SingleSegment { + fhd |= 1 << 5 + } + + var dictIDContent []byte + if f.DictID > 0 { + var tmp [4]byte + if f.DictID < 256 { + fhd |= 1 + tmp[0] = uint8(f.DictID) + dictIDContent = tmp[:1] + } else if f.DictID < 1<<16 { + fhd |= 2 + binary.LittleEndian.PutUint16(tmp[:2], uint16(f.DictID)) + dictIDContent = tmp[:2] + } else { + fhd |= 3 + binary.LittleEndian.PutUint32(tmp[:4], f.DictID) + dictIDContent = tmp[:4] + } + } + var fcs uint8 + if f.ContentSize >= 256 { + fcs++ + } + if f.ContentSize >= 65536+256 { + fcs++ + } + if f.ContentSize >= 0xffffffff { + fcs++ + } + + fhd |= fcs << 6 + + dst = append(dst, fhd) + if !f.SingleSegment { + const winLogMin = 10 + windowLog := (bits.Len32(f.WindowSize-1) - winLogMin) << 3 + dst = append(dst, uint8(windowLog)) + } + if f.DictID > 0 { + dst = append(dst, dictIDContent...) + } + switch fcs { + case 0: + if f.SingleSegment { + dst = append(dst, uint8(f.ContentSize)) + } + // Unless SingleSegment is set, framessizes < 256 are nto stored. + case 1: + f.ContentSize -= 256 + dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8)) + case 2: + dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8), uint8(f.ContentSize>>16), uint8(f.ContentSize>>24)) + case 3: + dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8), uint8(f.ContentSize>>16), uint8(f.ContentSize>>24), + uint8(f.ContentSize>>32), uint8(f.ContentSize>>40), uint8(f.ContentSize>>48), uint8(f.ContentSize>>56)) + default: + panic("invalid fcs") + } + return dst, nil +} + +const skippableFrameHeader = 4 + 4 + +// calcSkippableFrame will return a total size to be added for written +// to be divisible by multiple. +// The value will always be > skippableFrameHeader. +// The function will panic if written < 0 or wantMultiple <= 0. +func calcSkippableFrame(written, wantMultiple int64) int { + if wantMultiple <= 0 { + panic("wantMultiple <= 0") + } + if written < 0 { + panic("written < 0") + } + leftOver := written % wantMultiple + if leftOver == 0 { + return 0 + } + toAdd := wantMultiple - leftOver + for toAdd < skippableFrameHeader { + toAdd += wantMultiple + } + return int(toAdd) +} + +// skippableFrame will add a skippable frame with a total size of bytes. +// total should be >= skippableFrameHeader and < math.MaxUint32. +func skippableFrame(dst []byte, total int, r io.Reader) ([]byte, error) { + if total == 0 { + return dst, nil + } + if total < skippableFrameHeader { + return dst, fmt.Errorf("requested skippable frame (%d) < 8", total) + } + if int64(total) > math.MaxUint32 { + return dst, fmt.Errorf("requested skippable frame (%d) > max uint32", total) + } + dst = append(dst, 0x50, 0x2a, 0x4d, 0x18) + f := uint32(total - skippableFrameHeader) + dst = append(dst, uint8(f), uint8(f>>8), uint8(f>>16), uint8(f>>24)) + start := len(dst) + dst = append(dst, make([]byte, f)...) + _, err := io.ReadFull(r, dst[start:]) + return dst, err +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go new file mode 100644 index 000000000000..e6d3d49b39c0 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go @@ -0,0 +1,385 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" +) + +const ( + tablelogAbsoluteMax = 9 +) + +const ( + /*!MEMORY_USAGE : + * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.) + * Increasing memory usage improves compression ratio + * Reduced memory usage can improve speed, due to cache effect + * Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */ + maxMemoryUsage = tablelogAbsoluteMax + 2 + + maxTableLog = maxMemoryUsage - 2 + maxTablesize = 1 << maxTableLog + maxTableMask = (1 << maxTableLog) - 1 + minTablelog = 5 + maxSymbolValue = 255 +) + +// fseDecoder provides temporary storage for compression and decompression. +type fseDecoder struct { + 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 +} + +// tableStep returns the next table index. +func tableStep(tableSize uint32) uint32 { + return (tableSize >> 1) + (tableSize >> 3) + 3 +} + +// readNCount will read the symbol distribution so decoding tables can be constructed. +func (s *fseDecoder) readNCount(b *byteReader, maxSymbol uint16) error { + var ( + charnum uint16 + previous0 bool + ) + if b.remain() < 4 { + return errors.New("input too small") + } + bitStream := b.Uint32NC() + nbBits := uint((bitStream & 0xF) + minTablelog) // extract tableLog + if nbBits > tablelogAbsoluteMax { + println("Invalid tablelog:", nbBits) + return errors.New("tableLog too large") + } + bitStream >>= 4 + bitCount := uint(4) + + s.actualTableLog = uint8(nbBits) + remaining := int32((1 << nbBits) + 1) + threshold := int32(1 << nbBits) + gotTotal := int32(0) + nbBits++ + + for remaining > 1 && charnum <= maxSymbol { + if previous0 { + //println("prev0") + n0 := charnum + for (bitStream & 0xFFFF) == 0xFFFF { + //println("24 x 0") + n0 += 24 + if r := b.remain(); r > 5 { + b.advance(2) + // The check above should make sure we can read 32 bits + bitStream = b.Uint32NC() >> bitCount + } else { + // end of bit stream + bitStream >>= 16 + bitCount += 16 + } + } + //printf("bitstream: %d, 0b%b", bitStream&3, bitStream) + for (bitStream & 3) == 3 { + n0 += 3 + bitStream >>= 2 + bitCount += 2 + } + n0 += uint16(bitStream & 3) + bitCount += 2 + + if n0 > maxSymbolValue { + return errors.New("maxSymbolValue too small") + } + //println("inserting ", n0-charnum, "zeroes from idx", charnum, "ending before", n0) + for charnum < n0 { + s.norm[uint8(charnum)] = 0 + charnum++ + } + + if r := b.remain(); r >= 7 || r-int(bitCount>>3) >= 4 { + b.advance(bitCount >> 3) + bitCount &= 7 + // The check above should make sure we can read 32 bits + bitStream = b.Uint32NC() >> bitCount + } else { + bitStream >>= 2 + } + } + + max := (2*threshold - 1) - remaining + var count int32 + + if int32(bitStream)&(threshold-1) < max { + count = int32(bitStream) & (threshold - 1) + if debugAsserts && nbBits < 1 { + panic("nbBits underflow") + } + bitCount += nbBits - 1 + } else { + count = int32(bitStream) & (2*threshold - 1) + if count >= threshold { + count -= max + } + bitCount += nbBits + } + + // extra accuracy + count-- + if count < 0 { + // -1 means +1 + remaining += count + gotTotal -= count + } else { + remaining -= count + gotTotal += count + } + s.norm[charnum&0xff] = int16(count) + charnum++ + previous0 = count == 0 + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + + if r := b.remain(); r >= 7 || r-int(bitCount>>3) >= 4 { + b.advance(bitCount >> 3) + bitCount &= 7 + // The check above should make sure we can read 32 bits + bitStream = b.Uint32NC() >> (bitCount & 31) + } else { + bitCount -= (uint)(8 * (len(b.b) - 4 - b.off)) + b.off = len(b.b) - 4 + bitStream = b.Uint32() >> (bitCount & 31) + } + } + s.symbolLen = charnum + if s.symbolLen <= 1 { + return fmt.Errorf("symbolLen (%d) too small", s.symbolLen) + } + if s.symbolLen > maxSymbolValue+1 { + return fmt.Errorf("symbolLen (%d) too big", s.symbolLen) + } + if remaining != 1 { + return fmt.Errorf("corruption detected (remaining %d != 1)", remaining) + } + if bitCount > 32 { + return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount) + } + if gotTotal != 1<> 3) + // println(s.norm[:s.symbolLen], s.symbolLen) + return s.buildDtable() +} + +// 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. +// Using a composite uint64 is faster than a struct with separate members. +type decSymbol uint64 + +func newDecSymbol(nbits, addBits uint8, newState uint16, baseline uint32) decSymbol { + return decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32) +} + +func (d decSymbol) nbBits() uint8 { + return uint8(d) +} + +func (d decSymbol) addBits() uint8 { + return uint8(d >> 8) +} + +func (d decSymbol) newState() uint16 { + 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) +} + +func (d *decSymbol) setAddBits(addBits uint8) { + const mask = 0xffffffffffff00ff + *d = (*d & mask) | (decSymbol(addBits) << 8) +} + +func (d *decSymbol) setNewState(state uint16) { + const mask = 0xffffffff0000ffff + *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) +} + +// decSymbolValue returns the transformed decSymbol for the given symbol. +func decSymbolValue(symb uint8, t []baseOffset) (decSymbol, error) { + if int(symb) >= len(t) { + return 0, fmt.Errorf("rle symbol %d >= max %d", symb, len(t)) + } + lu := t[symb] + return newDecSymbol(0, lu.addBits, 0, lu.baseLine), nil +} + +// setRLE will set the decoder til RLE mode. +func (s *fseDecoder) setRLE(symbol decSymbol) { + s.actualTableLog = 0 + s.maxBits = symbol.addBits() + 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. +func (s *fseDecoder) transform(t []baseOffset) error { + tableSize := uint16(1 << s.actualTableLog) + s.maxBits = 0 + for i, v := range s.dt[:tableSize] { + add := v.addBits() + if int(add) >= len(t) { + return fmt.Errorf("invalid decoding table entry %d, symbol %d >= max (%d)", i, v.addBits(), len(t)) + } + lu := t[add] + if lu.addBits > s.maxBits { + s.maxBits = lu.addBits + } + v.setExt(lu.addBits, lu.baseLine) + s.dt[i] = v + } + return nil +} + +type fseState struct { + dt []decSymbol + state decSymbol +} + +// Initialize and decodeAsync first state and symbol. +func (s *fseState) init(br *bitReader, tableLog uint8, dt []decSymbol) { + s.dt = dt + br.fill() + 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_encoder.go b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go new file mode 100644 index 000000000000..aa9eba88b805 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go @@ -0,0 +1,726 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "math" +) + +const ( + // For encoding we only support up to + maxEncTableLog = 8 + maxEncTablesize = 1 << maxTableLog + maxEncTableMask = (1 << maxTableLog) - 1 + minEncTablelog = 5 + maxEncSymbolValue = maxMatchLengthSymbol +) + +// Scratch provides temporary storage for compression and decompression. +type fseEncoder struct { + symbolLen uint16 // Length of active part of the symbol table. + actualTableLog uint8 // Selected tablelog. + ct cTable // Compression tables. + maxCount int // count of the most probable symbol + zeroBits bool // no bits has prob > 50%. + clearCount bool // clear count + useRLE bool // This encoder is for RLE + preDefined bool // This encoder is predefined. + reUsed bool // Set to know when the encoder has been reused. + rleVal uint8 // RLE Symbol + maxBits uint8 // Maximum output bits after transform. + + // TODO: Technically zstd should be fine with 64 bytes. + count [256]uint32 + norm [256]int16 +} + +// cTable contains tables used for compression. +type cTable struct { + tableSymbol []byte + stateTable []uint16 + symbolTT []symbolTransform +} + +// symbolTransform contains the state transform for a symbol. +type symbolTransform struct { + deltaNbBits uint32 + deltaFindState int16 + outBits uint8 +} + +// String prints values as a human readable string. +func (s symbolTransform) String() string { + return fmt.Sprintf("{deltabits: %08x, findstate:%d outbits:%d}", s.deltaNbBits, s.deltaFindState, s.outBits) +} + +// Histogram allows to populate the histogram and skip that step in the compression, +// It otherwise allows to inspect the histogram when compression is done. +// 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[:] +} + +// HistogramFinished can be called to indicate that the histogram has been populated. +// maxSymbol is the index of the highest set symbol of the next data segment. +// maxCount is the number of entries in the most populated entry. +// These are accepted at face value. +func (s *fseEncoder) HistogramFinished(maxSymbol uint8, maxCount int) { + s.maxCount = maxCount + s.symbolLen = uint16(maxSymbol) + 1 + 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() { + tableSize := 1 << s.actualTableLog + // get tableSymbol that is big enough. + if cap(s.ct.tableSymbol) < int(tableSize) { + s.ct.tableSymbol = make([]byte, tableSize) + } + s.ct.tableSymbol = s.ct.tableSymbol[:tableSize] + + ctSize := tableSize + if cap(s.ct.stateTable) < ctSize { + s.ct.stateTable = make([]uint16, ctSize) + } + s.ct.stateTable = s.ct.stateTable[:ctSize] + + if cap(s.ct.symbolTT) < 256 { + s.ct.symbolTT = make([]symbolTransform, 256) + } + s.ct.symbolTT = s.ct.symbolTT[:256] +} + +// buildCTable will populate the compression table so it is ready to be used. +func (s *fseEncoder) buildCTable() error { + tableSize := uint32(1 << s.actualTableLog) + highThreshold := tableSize - 1 + var cumul [256]int16 + + s.allocCtable() + tableSymbol := s.ct.tableSymbol[:tableSize] + // symbol start positions + { + cumul[0] = 0 + for ui, v := range s.norm[:s.symbolLen-1] { + u := byte(ui) // one less than reference + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = u + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + } + // Encode last symbol separately to avoid overflowing u + u := int(s.symbolLen - 1) + v := s.norm[s.symbolLen-1] + if v == -1 { + // Low proba symbol + cumul[u+1] = cumul[u] + 1 + tableSymbol[highThreshold] = byte(u) + highThreshold-- + } else { + cumul[u+1] = cumul[u] + v + } + if uint32(cumul[s.symbolLen]) != tableSize { + return fmt.Errorf("internal error: expected cumul[s.symbolLen] (%d) == tableSize (%d)", cumul[s.symbolLen], tableSize) + } + cumul[s.symbolLen] = int16(tableSize) + 1 + } + // Spread symbols + s.zeroBits = false + { + step := tableStep(tableSize) + tableMask := tableSize - 1 + var position uint32 + // if any symbol > largeLimit, we may have 0 bits output. + largeLimit := int16(1 << (s.actualTableLog - 1)) + for ui, v := range s.norm[:s.symbolLen] { + symbol := byte(ui) + if v > largeLimit { + s.zeroBits = true + } + for nbOccurrences := int16(0); nbOccurrences < v; nbOccurrences++ { + tableSymbol[position] = symbol + position = (position + step) & tableMask + for position > highThreshold { + position = (position + step) & tableMask + } /* Low proba area */ + } + } + + // Check if we have gone through all positions + if position != 0 { + return errors.New("position!=0") + } + } + + // Build table + table := s.ct.stateTable + { + tsi := int(tableSize) + for u, v := range tableSymbol { + // TableU16 : sorted by symbol order; gives next state value + table[cumul[v]] = uint16(tsi + u) + cumul[v]++ + } + } + + // Build Symbol Transformation Table + { + total := int16(0) + symbolTT := s.ct.symbolTT[:s.symbolLen] + tableLog := s.actualTableLog + tl := (uint32(tableLog) << 16) - (1 << tableLog) + for i, v := range s.norm[:s.symbolLen] { + switch v { + case 0: + case -1, 1: + symbolTT[i].deltaNbBits = tl + symbolTT[i].deltaFindState = int16(total - 1) + total++ + default: + maxBitsOut := uint32(tableLog) - highBit(uint32(v-1)) + minStatePlus := uint32(v) << maxBitsOut + symbolTT[i].deltaNbBits = (maxBitsOut << 16) - minStatePlus + symbolTT[i].deltaFindState = int16(total - v) + total += v + } + } + if total != int16(tableSize) { + return fmt.Errorf("total mismatch %d (got) != %d (want)", total, tableSize) + } + } + return nil +} + +var rtbTable = [...]uint32{0, 473195, 504333, 520860, 550000, 700000, 750000, 830000} + +func (s *fseEncoder) setRLE(val byte) { + s.allocCtable() + s.actualTableLog = 0 + s.ct.stateTable = s.ct.stateTable[:1] + s.ct.symbolTT[val] = symbolTransform{ + deltaFindState: 0, + deltaNbBits: 0, + } + if debug { + println("setRLE: val", val, "symbolTT", s.ct.symbolTT[val]) + } + s.rleVal = val + s.useRLE = true +} + +// setBits will set output bits for the transform. +// if nil is provided, the number of bits is equal to the index. +func (s *fseEncoder) setBits(transform []byte) { + if s.reUsed || s.preDefined { + return + } + if s.useRLE { + if transform == nil { + s.ct.symbolTT[s.rleVal].outBits = s.rleVal + s.maxBits = s.rleVal + return + } + s.maxBits = transform[s.rleVal] + s.ct.symbolTT[s.rleVal].outBits = s.maxBits + return + } + if transform == nil { + for i := range s.ct.symbolTT[:s.symbolLen] { + s.ct.symbolTT[i].outBits = uint8(i) + } + s.maxBits = uint8(s.symbolLen - 1) + return + } + s.maxBits = 0 + for i, v := range transform[:s.symbolLen] { + s.ct.symbolTT[i].outBits = v + if v > s.maxBits { + // We could assume bits always going up, but we play safe. + s.maxBits = v + } + } +} + +// normalizeCount will normalize the count of the symbols so +// the total is equal to the table size. +// If successful, compression tables will also be made ready. +func (s *fseEncoder) normalizeCount(length int) error { + if s.reUsed { + return nil + } + s.optimalTableLog(length) + var ( + tableLog = s.actualTableLog + scale = 62 - uint64(tableLog) + step = (1 << 62) / uint64(length) + vStep = uint64(1) << (scale - 20) + stillToDistribute = int16(1 << tableLog) + largest int + largestP int16 + lowThreshold = (uint32)(length >> tableLog) + ) + if s.maxCount == length { + s.useRLE = true + return nil + } + s.useRLE = false + for i, cnt := range s.count[:s.symbolLen] { + // already handled + // if (count[s] == s.length) return 0; /* rle special case */ + + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + stillToDistribute-- + } else { + proba := (int16)((uint64(cnt) * step) >> scale) + if proba < 8 { + restToBeat := vStep * uint64(rtbTable[proba]) + v := uint64(cnt)*step - (uint64(proba) << scale) + if v > restToBeat { + proba++ + } + } + if proba > largestP { + largestP = proba + largest = i + } + s.norm[i] = proba + stillToDistribute -= proba + } + } + + if -stillToDistribute >= (s.norm[largest] >> 1) { + // corner case, need another normalization method + err := s.normalizeCount2(length) + if err != nil { + return err + } + if debugAsserts { + err = s.validateNorm() + if err != nil { + return err + } + } + return s.buildCTable() + } + s.norm[largest] += stillToDistribute + if debugAsserts { + err := s.validateNorm() + if err != nil { + return err + } + } + return s.buildCTable() +} + +// Secondary normalization method. +// To be used when primary method fails. +func (s *fseEncoder) normalizeCount2(length int) error { + const notYetAssigned = -2 + var ( + distributed uint32 + total = uint32(length) + tableLog = s.actualTableLog + lowThreshold = uint32(total >> tableLog) + lowOne = uint32((total * 3) >> (tableLog + 1)) + ) + for i, cnt := range s.count[:s.symbolLen] { + if cnt == 0 { + s.norm[i] = 0 + continue + } + if cnt <= lowThreshold { + s.norm[i] = -1 + distributed++ + total -= cnt + continue + } + if cnt <= lowOne { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + s.norm[i] = notYetAssigned + } + toDistribute := (1 << tableLog) - distributed + + if (total / toDistribute) > lowOne { + // risk of rounding to zero + lowOne = uint32((total * 3) / (toDistribute * 2)) + for i, cnt := range s.count[:s.symbolLen] { + if (s.norm[i] == notYetAssigned) && (cnt <= lowOne) { + s.norm[i] = 1 + distributed++ + total -= cnt + continue + } + } + toDistribute = (1 << tableLog) - distributed + } + if distributed == uint32(s.symbolLen)+1 { + // all values are pretty poor; + // probably incompressible data (should have already been detected); + // find max, then give all remaining points to max + var maxV int + var maxC uint32 + for i, cnt := range s.count[:s.symbolLen] { + if cnt > maxC { + maxV = i + maxC = cnt + } + } + s.norm[maxV] += int16(toDistribute) + return nil + } + + if total == 0 { + // all of the symbols were low enough for the lowOne or lowThreshold + for i := uint32(0); toDistribute > 0; i = (i + 1) % (uint32(s.symbolLen)) { + if s.norm[i] > 0 { + toDistribute-- + s.norm[i]++ + } + } + return nil + } + + var ( + vStepLog = 62 - uint64(tableLog) + mid = uint64((1 << (vStepLog - 1)) - 1) + rStep = (((1 << vStepLog) * uint64(toDistribute)) + mid) / uint64(total) // scale on remaining + tmpTotal = mid + ) + for i, cnt := range s.count[:s.symbolLen] { + if s.norm[i] == notYetAssigned { + var ( + end = tmpTotal + uint64(cnt)*rStep + sStart = uint32(tmpTotal >> vStepLog) + sEnd = uint32(end >> vStepLog) + weight = sEnd - sStart + ) + if weight < 1 { + return errors.New("weight < 1") + } + s.norm[i] = int16(weight) + tmpTotal = end + } + } + return nil +} + +// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog +func (s *fseEncoder) optimalTableLog(length int) { + tableLog := uint8(maxEncTableLog) + minBitsSrc := highBit(uint32(length)) + 1 + minBitsSymbols := highBit(uint32(s.symbolLen-1)) + 2 + minBits := uint8(minBitsSymbols) + if minBitsSrc < minBitsSymbols { + minBits = uint8(minBitsSrc) + } + + maxBitsSrc := uint8(highBit(uint32(length-1))) - 2 + if maxBitsSrc < tableLog { + // Accuracy can be reduced + tableLog = maxBitsSrc + } + if minBits > tableLog { + tableLog = minBits + } + // Need a minimum to safely represent all symbol values + if tableLog < minEncTablelog { + tableLog = minEncTablelog + } + if tableLog > maxEncTableLog { + tableLog = maxEncTableLog + } + s.actualTableLog = tableLog +} + +// validateNorm validates the normalized histogram table. +func (s *fseEncoder) validateNorm() (err error) { + var total int + for _, v := range s.norm[:s.symbolLen] { + if v >= 0 { + total += int(v) + } else { + total -= int(v) + } + } + defer func() { + if err == nil { + return + } + fmt.Printf("selected TableLog: %d, Symbol length: %d\n", s.actualTableLog, s.symbolLen) + for i, v := range s.norm[:s.symbolLen] { + fmt.Printf("%3d: %5d -> %4d \n", i, s.count[i], v) + } + }() + if total != (1 << s.actualTableLog) { + return fmt.Errorf("warning: Total == %d != %d", total, 1<> 3) + 3 + 2 + + // Write Table Size + bitStream = uint32(tableLog - minEncTablelog) + bitCount = uint(4) + remaining = int16(tableSize + 1) /* +1 for extra accuracy */ + threshold = int16(tableSize) + nbBits = uint(tableLog + 1) + outP = len(out) + ) + if cap(out) < outP+maxHeaderSize { + out = append(out, make([]byte, maxHeaderSize*3)...) + out = out[:len(out)-maxHeaderSize*3] + } + out = out[:outP+maxHeaderSize] + + // stops at 1 + for remaining > 1 { + if previous0 { + start := charnum + for s.norm[charnum] == 0 { + charnum++ + } + for charnum >= start+24 { + start += 24 + bitStream += uint32(0xFFFF) << bitCount + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + } + for charnum >= start+3 { + start += 3 + bitStream += 3 << bitCount + bitCount += 2 + } + bitStream += uint32(charnum-start) << bitCount + bitCount += 2 + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + count := s.norm[charnum] + charnum++ + max := (2*threshold - 1) - remaining + if count < 0 { + remaining += count + } else { + remaining -= count + } + count++ // +1 for extra accuracy + if count >= threshold { + count += max // [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ + } + bitStream += uint32(count) << bitCount + bitCount += nbBits + if count < max { + bitCount-- + } + + previous0 = count == 1 + if remaining < 1 { + return nil, errors.New("internal error: remaining < 1") + } + for remaining < threshold { + nbBits-- + threshold >>= 1 + } + + if bitCount > 16 { + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += 2 + bitStream >>= 16 + bitCount -= 16 + } + } + + if outP+2 > len(out) { + return nil, fmt.Errorf("internal error: %d > %d, maxheader: %d, sl: %d, tl: %d, normcount: %v", outP+2, len(out), maxHeaderSize, s.symbolLen, int(tableLog), s.norm[:s.symbolLen]) + } + out[outP] = byte(bitStream) + out[outP+1] = byte(bitStream >> 8) + outP += int((bitCount + 7) / 8) + + if charnum > s.symbolLen { + return nil, errors.New("internal error: charnum > s.symbolLen") + } + return out[:outP], nil +} + +// Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits) +// note 1 : assume symbolValue is valid (<= maxSymbolValue) +// note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits * +func (s *fseEncoder) bitCost(symbolValue uint8, accuracyLog uint32) uint32 { + minNbBits := s.ct.symbolTT[symbolValue].deltaNbBits >> 16 + threshold := (minNbBits + 1) << 16 + if debugAsserts { + if !(s.actualTableLog < 16) { + panic("!s.actualTableLog < 16") + } + // ensure enough room for renormalization double shift + if !(uint8(accuracyLog) < 31-s.actualTableLog) { + panic("!uint8(accuracyLog) < 31-s.actualTableLog") + } + } + tableSize := uint32(1) << s.actualTableLog + deltaFromThreshold := threshold - (s.ct.symbolTT[symbolValue].deltaNbBits + tableSize) + // linear interpolation (very approximate) + normalizedDeltaFromThreshold := (deltaFromThreshold << accuracyLog) >> s.actualTableLog + bitMultiplier := uint32(1) << accuracyLog + if debugAsserts { + if s.ct.symbolTT[symbolValue].deltaNbBits+tableSize > threshold { + panic("s.ct.symbolTT[symbolValue].deltaNbBits+tableSize > threshold") + } + if normalizedDeltaFromThreshold > bitMultiplier { + panic("normalizedDeltaFromThreshold > bitMultiplier") + } + } + return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold +} + +// Returns the cost in bits of encoding the distribution in count using ctable. +// Histogram should only be up to the last non-zero symbol. +// Returns an -1 if ctable cannot represent all the symbols in count. +func (s *fseEncoder) approxSize(hist []uint32) uint32 { + if int(s.symbolLen) < len(hist) { + // More symbols than we have. + return math.MaxUint32 + } + if s.useRLE { + // We will never reuse RLE encoders. + return math.MaxUint32 + } + const kAccuracyLog = 8 + badCost := (uint32(s.actualTableLog) + 1) << kAccuracyLog + var cost uint32 + for i, v := range hist { + if v == 0 { + continue + } + if s.norm[i] == 0 { + return math.MaxUint32 + } + bitCost := s.bitCost(uint8(i), kAccuracyLog) + if bitCost > badCost { + return math.MaxUint32 + } + cost += v * bitCost + } + return cost >> kAccuracyLog +} + +// maxHeaderSize returns the maximum header size in bits. +// This is not exact size, but we want a penalty for new tables anyway. +func (s *fseEncoder) maxHeaderSize() uint32 { + if s.preDefined { + return 0 + } + if s.useRLE { + return 8 + } + return (((uint32(s.symbolLen) * uint32(s.actualTableLog)) >> 3) + 3) * 8 +} + +// cState contains the compression state of a stream. +type cState struct { + bw *bitWriter + stateTable []uint16 + state uint16 +} + +// init will initialize the compression state to the first symbol of the stream. +func (c *cState) init(bw *bitWriter, ct *cTable, first symbolTransform) { + c.bw = bw + c.stateTable = ct.stateTable + if len(c.stateTable) == 1 { + // RLE + c.stateTable[0] = uint16(0) + c.state = 0 + return + } + nbBitsOut := (first.deltaNbBits + (1 << 15)) >> 16 + im := int32((nbBitsOut << 16) - first.deltaNbBits) + lu := (im >> nbBitsOut) + int32(first.deltaFindState) + c.state = c.stateTable[lu] + return +} + +// 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() + c.bw.addBits16NC(c.state, tableLog) +} diff --git a/vendor/github.com/klauspost/compress/zstd/fse_predefined.go b/vendor/github.com/klauspost/compress/zstd/fse_predefined.go new file mode 100644 index 000000000000..6c17dc17f4fd --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/fse_predefined.go @@ -0,0 +1,158 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "fmt" + "math" + "sync" +) + +var ( + // fsePredef are the predefined fse tables as defined here: + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions + // These values are already transformed. + fsePredef [3]fseDecoder + + // fsePredefEnc are the predefined encoder based on fse tables as defined here: + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions + // These values are already transformed. + fsePredefEnc [3]fseEncoder + + // symbolTableX contain the transformations needed for each type as defined in + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets + symbolTableX [3][]baseOffset + + // maxTableSymbol is the biggest supported symbol for each table type + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets + maxTableSymbol = [3]uint8{tableLiteralLengths: maxLiteralLengthSymbol, tableOffsets: maxOffsetLengthSymbol, tableMatchLengths: maxMatchLengthSymbol} + + // bitTables is the bits table for each table. + bitTables = [3][]byte{tableLiteralLengths: llBitsTable[:], tableOffsets: nil, tableMatchLengths: mlBitsTable[:]} +) + +type tableIndex uint8 + +const ( + // indexes for fsePredef and symbolTableX + tableLiteralLengths tableIndex = 0 + tableOffsets tableIndex = 1 + tableMatchLengths tableIndex = 2 + + maxLiteralLengthSymbol = 35 + maxOffsetLengthSymbol = 30 + maxMatchLengthSymbol = 52 +) + +// baseOffset is used for calculating transformations. +type baseOffset struct { + baseLine uint32 + addBits uint8 +} + +// fillBase will precalculate base offsets with the given bit distributions. +func fillBase(dst []baseOffset, base uint32, bits ...uint8) { + if len(bits) != len(dst) { + panic(fmt.Sprintf("len(dst) (%d) != len(bits) (%d)", len(dst), len(bits))) + } + for i, bit := range bits { + if base > math.MaxInt32 { + panic(fmt.Sprintf("invalid decoding table, base overflows int32")) + } + + dst[i] = baseOffset{ + baseLine: base, + addBits: bit, + } + base += 1 << bit + } +} + +var predef sync.Once + +func initPredefined() { + predef.Do(func() { + // Literals length codes + tmp := make([]baseOffset, 36) + for i := range tmp[:16] { + tmp[i] = baseOffset{ + baseLine: uint32(i), + addBits: 0, + } + } + fillBase(tmp[16:], 16, 1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16) + symbolTableX[tableLiteralLengths] = tmp + + // Match length codes + tmp = make([]baseOffset, 53) + for i := range tmp[:32] { + tmp[i] = baseOffset{ + // The transformation adds the 3 length. + baseLine: uint32(i) + 3, + addBits: 0, + } + } + fillBase(tmp[32:], 35, 1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16) + symbolTableX[tableMatchLengths] = tmp + + // Offset codes + tmp = make([]baseOffset, maxOffsetBits+1) + tmp[1] = baseOffset{ + baseLine: 1, + addBits: 1, + } + fillBase(tmp[2:], 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30) + symbolTableX[tableOffsets] = tmp + + // Fill predefined tables and transform them. + // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions + for i := range fsePredef[:] { + f := &fsePredef[i] + switch tableIndex(i) { + case tableLiteralLengths: + // https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L243 + f.actualTableLog = 6 + copy(f.norm[:], []int16{4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1, + 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1, + -1, -1, -1, -1}) + f.symbolLen = 36 + case tableOffsets: + // https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L281 + f.actualTableLog = 5 + copy(f.norm[:], []int16{ + 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, -1, -1, -1}) + f.symbolLen = 29 + case tableMatchLengths: + //https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L304 + f.actualTableLog = 6 + copy(f.norm[:], []int16{ + 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, + -1, -1, -1, -1, -1}) + f.symbolLen = 53 + } + if err := f.buildDtable(); err != nil { + panic(fmt.Errorf("building table %v: %v", tableIndex(i), err)) + } + if err := f.transform(symbolTableX[i]); err != nil { + panic(fmt.Errorf("building table %v: %v", tableIndex(i), err)) + } + f.preDefined = true + + // Create encoder as well + enc := &fsePredefEnc[i] + copy(enc.norm[:], f.norm[:]) + enc.symbolLen = f.symbolLen + enc.actualTableLog = f.actualTableLog + if err := enc.buildCTable(); err != nil { + panic(fmt.Errorf("building encoding table %v: %v", tableIndex(i), err)) + } + enc.setBits(bitTables[i]) + enc.preDefined = true + } + }) +} diff --git a/vendor/github.com/klauspost/compress/zstd/hash.go b/vendor/github.com/klauspost/compress/zstd/hash.go new file mode 100644 index 000000000000..4a752067fc90 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/hash.go @@ -0,0 +1,77 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +const ( + prime3bytes = 506832829 + prime4bytes = 2654435761 + prime5bytes = 889523592379 + prime6bytes = 227718039650203 + prime7bytes = 58295818150454627 + 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 { + switch mls { + case 5: + return hash5(u, hashLog) + case 6: + return hash6(u, hashLog) + case 7: + return hash7(u, hashLog) + case 8: + return hash8(u, hashLog) + default: + return hash4x64(u, hashLog) + } +} + +// 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 new file mode 100644 index 000000000000..f783e32d251b --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/history.go @@ -0,0 +1,89 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "github.com/klauspost/compress/huff0" +) + +// history contains the information transferred between blocks. +type history struct { + b []byte + huffTree *huff0.Scratch + recentOffsets [3]int + decoders sequenceDecs + windowSize int + maxSize int + 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.error = false + h.recentOffsets = [3]int{1, 4, 8} + if f := h.decoders.litLengths.fse; f != nil && !f.preDefined { + fseDecoderPool.Put(f) + } + if f := h.decoders.offsets.fse; f != nil && !f.preDefined { + fseDecoderPool.Put(f) + } + if f := h.decoders.matchLengths.fse; f != nil && !f.preDefined { + fseDecoderPool.Put(f) + } + h.decoders = sequenceDecs{} + if h.huffTree != nil { + if h.dict == nil || h.dict.litEnc != h.huffTree { + huffDecoderPool.Put(h.huffTree) + } + } + h.huffTree = nil + h.dict = nil + //printf("history created: %+v (l: %d, c: %d)", *h, len(h.b), cap(h.b)) +} + +func (h *history) setDict(dict *dict) { + if dict == nil { + return + } + h.dict = dict + h.decoders.litLengths = dict.llDec + h.decoders.offsets = dict.ofDec + h.decoders.matchLengths = dict.mlDec + h.recentOffsets = dict.offsets + h.huffTree = dict.litEnc +} + +// append bytes to history. +// This function will make sure there is space for it, +// if the buffer has been allocated with enough extra space. +func (h *history) append(b []byte) { + if len(b) >= h.windowSize { + // Discard all history by simply overwriting + h.b = h.b[:h.windowSize] + copy(h.b, b[len(b)-h.windowSize:]) + return + } + + // If there is space, append it. + if len(b) < cap(h.b)-len(h.b) { + h.b = append(h.b, b...) + 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(b) + len(h.b) - h.windowSize + copy(h.b, h.b[discard:]) + h.b = h.b[:h.windowSize] + copy(h.b[h.windowSize-len(b):], b) +} + +// 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/LICENSE.txt b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt new file mode 100644 index 000000000000..24b53065f40b --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt @@ -0,0 +1,22 @@ +Copyright (c) 2016 Caleb Spare + +MIT License + +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/zstd/internal/xxhash/README.md b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md new file mode 100644 index 000000000000..69aa3bb587c8 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md @@ -0,0 +1,58 @@ +# xxhash + +VENDORED: Go to [github.com/cespare/xxhash](https://github.com/cespare/xxhash) for original package. + + +[![GoDoc](https://godoc.org/github.com/cespare/xxhash?status.svg)](https://godoc.org/github.com/cespare/xxhash) +[![Build Status](https://travis-ci.org/cespare/xxhash.svg?branch=master)](https://travis-ci.org/cespare/xxhash) + +xxhash is a Go implementation of the 64-bit +[xxHash](http://cyan4973.github.io/xxHash/) algorithm, XXH64. This is a +high-quality hashing algorithm that is much faster than anything in the Go +standard library. + +This package provides a straightforward API: + +``` +func Sum64(b []byte) uint64 +func Sum64String(s string) uint64 +type Digest struct{ ... } + func New() *Digest +``` + +The `Digest` type implements hash.Hash64. Its key methods are: + +``` +func (*Digest) Write([]byte) (int, error) +func (*Digest) WriteString(string) (int, error) +func (*Digest) Sum64() uint64 +``` + +This implementation provides a fast pure-Go implementation and an even faster +assembly implementation for amd64. + +## Benchmarks + +Here are some quick benchmarks comparing the pure-Go and assembly +implementations of Sum64. + +| input size | purego | asm | +| --- | --- | --- | +| 5 B | 979.66 MB/s | 1291.17 MB/s | +| 100 B | 7475.26 MB/s | 7973.40 MB/s | +| 4 KB | 17573.46 MB/s | 17602.65 MB/s | +| 10 MB | 17131.46 MB/s | 17142.16 MB/s | + +These numbers were generated on Ubuntu 18.04 with an Intel i7-8700K CPU using +the following commands under Go 1.11.2: + +``` +$ go test -tags purego -benchtime 10s -bench '/xxhash,direct,bytes' +$ go test -benchtime 10s -bench '/xxhash,direct,bytes' +``` + +## Projects using this package + +- [InfluxDB](https://github.com/influxdata/influxdb) +- [Prometheus](https://github.com/prometheus/prometheus) +- [FreeCache](https://github.com/coocood/freecache) diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go new file mode 100644 index 000000000000..426b9cac7869 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go @@ -0,0 +1,238 @@ +// Package xxhash implements the 64-bit variant of xxHash (XXH64) as described +// at http://cyan4973.github.io/xxHash/. +// THIS IS VENDORED: Go to github.com/cespare/xxhash for original package. + +package xxhash + +import ( + "encoding/binary" + "errors" + "math/bits" +) + +const ( + prime1 uint64 = 11400714785074694791 + prime2 uint64 = 14029467366897019727 + prime3 uint64 = 1609587929392839161 + prime4 uint64 = 9650029242287828579 + prime5 uint64 = 2870177450012600261 +) + +// NOTE(caleb): I'm using both consts and vars of the primes. Using consts where +// possible in the Go code is worth a small (but measurable) performance boost +// by avoiding some MOVQs. Vars are needed for the asm and also are useful for +// convenience in the Go code in a few places where we need to intentionally +// avoid constant arithmetic (e.g., v1 := prime1 + prime2 fails because the +// result overflows a uint64). +var ( + prime1v = prime1 + prime2v = prime2 + prime3v = prime3 + prime4v = prime4 + prime5v = prime5 +) + +// Digest implements hash.Hash64. +type Digest struct { + v1 uint64 + v2 uint64 + v3 uint64 + v4 uint64 + total uint64 + mem [32]byte + n int // how much of mem is used +} + +// New creates a new Digest that computes the 64-bit xxHash algorithm. +func New() *Digest { + var d Digest + d.Reset() + return &d +} + +// Reset clears the Digest's state so that it can be reused. +func (d *Digest) Reset() { + d.v1 = prime1v + prime2 + d.v2 = prime2 + d.v3 = 0 + d.v4 = -prime1v + d.total = 0 + d.n = 0 +} + +// Size always returns 8 bytes. +func (d *Digest) Size() int { return 8 } + +// BlockSize always returns 32 bytes. +func (d *Digest) BlockSize() int { return 32 } + +// Write adds more data to d. It always returns len(b), nil. +func (d *Digest) Write(b []byte) (n int, err error) { + n = len(b) + d.total += uint64(n) + + if d.n+n < 32 { + // This new data doesn't even fill the current block. + copy(d.mem[d.n:], b) + d.n += n + return + } + + if d.n > 0 { + // Finish off the partial block. + copy(d.mem[d.n:], b) + d.v1 = round(d.v1, u64(d.mem[0:8])) + d.v2 = round(d.v2, u64(d.mem[8:16])) + d.v3 = round(d.v3, u64(d.mem[16:24])) + d.v4 = round(d.v4, u64(d.mem[24:32])) + b = b[32-d.n:] + d.n = 0 + } + + if len(b) >= 32 { + // One or more full blocks left. + nw := writeBlocks(d, b) + b = b[nw:] + } + + // Store any remaining partial block. + copy(d.mem[:], b) + d.n = len(b) + + return +} + +// Sum appends the current hash to b and returns the resulting slice. +func (d *Digest) Sum(b []byte) []byte { + s := d.Sum64() + return append( + b, + byte(s>>56), + byte(s>>48), + byte(s>>40), + byte(s>>32), + byte(s>>24), + byte(s>>16), + byte(s>>8), + byte(s), + ) +} + +// Sum64 returns the current hash. +func (d *Digest) Sum64() uint64 { + var h uint64 + + if d.total >= 32 { + v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4 + h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + h = mergeRound(h, v1) + h = mergeRound(h, v2) + h = mergeRound(h, v3) + h = mergeRound(h, v4) + } else { + h = d.v3 + prime5 + } + + h += d.total + + i, end := 0, d.n + for ; i+8 <= end; i += 8 { + k1 := round(0, u64(d.mem[i:i+8])) + h ^= k1 + h = rol27(h)*prime1 + prime4 + } + if i+4 <= end { + h ^= uint64(u32(d.mem[i:i+4])) * prime1 + h = rol23(h)*prime2 + prime3 + i += 4 + } + for i < end { + h ^= uint64(d.mem[i]) * prime5 + h = rol11(h) * prime1 + i++ + } + + h ^= h >> 33 + h *= prime2 + h ^= h >> 29 + h *= prime3 + h ^= h >> 32 + + return h +} + +const ( + magic = "xxh\x06" + marshaledSize = len(magic) + 8*5 + 32 +) + +// MarshalBinary implements the encoding.BinaryMarshaler interface. +func (d *Digest) MarshalBinary() ([]byte, error) { + b := make([]byte, 0, marshaledSize) + b = append(b, magic...) + b = appendUint64(b, d.v1) + b = appendUint64(b, d.v2) + b = appendUint64(b, d.v3) + b = appendUint64(b, d.v4) + b = appendUint64(b, d.total) + b = append(b, d.mem[:d.n]...) + b = b[:len(b)+len(d.mem)-d.n] + return b, nil +} + +// UnmarshalBinary implements the encoding.BinaryUnmarshaler interface. +func (d *Digest) UnmarshalBinary(b []byte) error { + if len(b) < len(magic) || string(b[:len(magic)]) != magic { + return errors.New("xxhash: invalid hash state identifier") + } + if len(b) != marshaledSize { + return errors.New("xxhash: invalid hash state size") + } + b = b[len(magic):] + b, d.v1 = consumeUint64(b) + b, d.v2 = consumeUint64(b) + b, d.v3 = consumeUint64(b) + 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 +} + +func appendUint64(b []byte, x uint64) []byte { + var a [8]byte + binary.LittleEndian.PutUint64(a[:], x) + return append(b, a[:]...) +} + +func consumeUint64(b []byte) ([]byte, uint64) { + x := u64(b) + return b[8:], x +} + +func u64(b []byte) uint64 { return binary.LittleEndian.Uint64(b) } +func u32(b []byte) uint32 { return binary.LittleEndian.Uint32(b) } + +func round(acc, input uint64) uint64 { + acc += input * prime2 + acc = rol31(acc) + acc *= prime1 + return acc +} + +func mergeRound(acc, val uint64) uint64 { + val = round(0, val) + acc ^= val + acc = acc*prime1 + prime4 + return acc +} + +func rol1(x uint64) uint64 { return bits.RotateLeft64(x, 1) } +func rol7(x uint64) uint64 { return bits.RotateLeft64(x, 7) } +func rol11(x uint64) uint64 { return bits.RotateLeft64(x, 11) } +func rol12(x uint64) uint64 { return bits.RotateLeft64(x, 12) } +func rol18(x uint64) uint64 { return bits.RotateLeft64(x, 18) } +func rol23(x uint64) uint64 { return bits.RotateLeft64(x, 23) } +func rol27(x uint64) uint64 { return bits.RotateLeft64(x, 27) } +func rol31(x uint64) uint64 { return bits.RotateLeft64(x, 31) } 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 new file mode 100644 index 000000000000..35318d7c46cb --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go @@ -0,0 +1,13 @@ +// +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 new file mode 100644 index 000000000000..2c9c5357a14b --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s @@ -0,0 +1,215 @@ +// +build !appengine +// +build gc +// +build !purego + +#include "textflag.h" + +// Register allocation: +// AX h +// CX pointer to advance through b +// DX n +// BX loop end +// R8 v1, k1 +// R9 v2 +// R10 v3 +// R11 v4 +// R12 tmp +// R13 prime1v +// R14 prime2v +// R15 prime4v + +// round reads from and advances the buffer pointer in CX. +// It assumes that R13 has prime1v and R14 has prime2v. +#define round(r) \ + MOVQ (CX), R12 \ + ADDQ $8, CX \ + 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. +#define mergeRound(acc, val) \ + IMULQ R14, val \ + ROLQ $31, val \ + IMULQ R13, val \ + XORQ val, acc \ + IMULQ R13, acc \ + ADDQ R15, 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 + + // Load slice. + MOVQ b_base+0(FP), CX + MOVQ b_len+8(FP), DX + LEAQ (CX)(DX*1), BX + + // The first loop limit will be len(b)-32. + SUBQ $32, BX + + // Check whether we have at least one block. + CMPQ DX, $32 + JLT noBlocks + + // Set up initial state (v1, v2, v3, v4). + MOVQ R13, R8 + ADDQ R14, R8 + MOVQ R14, R9 + XORQ R10, R10 + XORQ R11, R11 + SUBQ R13, R11 + + // Loop until CX > BX. +blockLoop: + round(R8) + round(R9) + round(R10) + round(R11) + + CMPQ CX, BX + JLE blockLoop + + MOVQ R8, AX + ROLQ $1, AX + MOVQ R9, R12 + ROLQ $7, R12 + ADDQ R12, AX + MOVQ R10, R12 + ROLQ $12, R12 + ADDQ R12, AX + MOVQ R11, R12 + ROLQ $18, R12 + ADDQ R12, AX + + mergeRound(AX, R8) + mergeRound(AX, R9) + mergeRound(AX, R10) + mergeRound(AX, R11) + + JMP afterBlocks + +noBlocks: + MOVQ ·prime5v(SB), AX + +afterBlocks: + ADDQ DX, AX + + // Right now BX has len(b)-32, and we want to loop until CX > len(b)-8. + ADDQ $24, BX + + CMPQ CX, BX + JG fourByte + +wordLoop: + // Calculate k1. + MOVQ (CX), R8 + ADDQ $8, CX + IMULQ R14, R8 + ROLQ $31, R8 + IMULQ R13, R8 + + XORQ R8, AX + ROLQ $27, AX + IMULQ R13, AX + ADDQ R15, AX + + CMPQ CX, BX + JLE wordLoop + +fourByte: + ADDQ $4, BX + CMPQ CX, BX + JG singles + + MOVL (CX), R8 + ADDQ $4, CX + IMULQ R13, R8 + XORQ R8, AX + + ROLQ $23, AX + IMULQ R14, AX + ADDQ ·prime3v(SB), AX + +singles: + ADDQ $4, BX + CMPQ CX, BX + JGE finalize + +singlesLoop: + MOVBQZX (CX), R12 + ADDQ $1, CX + IMULQ ·prime5v(SB), R12 + XORQ R12, AX + + ROLQ $11, AX + IMULQ R13, AX + + CMPQ CX, BX + JL singlesLoop + +finalize: + MOVQ AX, R12 + SHRQ $33, R12 + XORQ R12, AX + IMULQ R14, AX + MOVQ AX, R12 + SHRQ $29, R12 + XORQ R12, AX + IMULQ ·prime3v(SB), AX + MOVQ AX, R12 + SHRQ $32, R12 + XORQ R12, AX + + MOVQ AX, ret+24(FP) + RET + +// writeBlocks uses the same registers as above except that it uses AX to store +// the d pointer. + +// func writeBlocks(d *Digest, b []byte) int +TEXT ·writeBlocks(SB), NOSPLIT, $0-40 + // Load fixed primes needed for round. + MOVQ ·prime1v(SB), R13 + MOVQ ·prime2v(SB), R14 + + // Load slice. + MOVQ arg1_base+8(FP), CX + MOVQ arg1_len+16(FP), DX + LEAQ (CX)(DX*1), BX + SUBQ $32, BX + + // Load vN from d. + MOVQ arg+0(FP), AX + MOVQ 0(AX), R8 // v1 + MOVQ 8(AX), R9 // v2 + MOVQ 16(AX), R10 // v3 + MOVQ 24(AX), R11 // v4 + + // We don't need to check the loop condition here; this function is + // always called with at least one block of data to process. +blockLoop: + round(R8) + round(R9) + round(R10) + round(R11) + + CMPQ CX, BX + JLE blockLoop + + // Copy vN back to d. + MOVQ R8, 0(AX) + MOVQ R9, 8(AX) + 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) + + RET 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 new file mode 100644 index 000000000000..4a5a821603e5 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go @@ -0,0 +1,76 @@ +// +build !amd64 appengine !gc purego + +package xxhash + +// Sum64 computes the 64-bit xxHash digest of b. +func Sum64(b []byte) uint64 { + // A simpler version would be + // d := New() + // d.Write(b) + // return d.Sum64() + // but this is faster, particularly for small inputs. + + n := len(b) + var h uint64 + + if n >= 32 { + v1 := prime1v + prime2 + v2 := prime2 + v3 := uint64(0) + v4 := -prime1v + for len(b) >= 32 { + v1 = round(v1, u64(b[0:8:len(b)])) + v2 = round(v2, u64(b[8:16:len(b)])) + v3 = round(v3, u64(b[16:24:len(b)])) + v4 = round(v4, u64(b[24:32:len(b)])) + b = b[32:len(b):len(b)] + } + h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4) + h = mergeRound(h, v1) + h = mergeRound(h, v2) + h = mergeRound(h, v3) + h = mergeRound(h, v4) + } else { + h = prime5 + } + + h += uint64(n) + + i, end := 0, len(b) + for ; i+8 <= end; i += 8 { + k1 := round(0, u64(b[i:i+8:len(b)])) + h ^= k1 + h = rol27(h)*prime1 + prime4 + } + if i+4 <= end { + h ^= uint64(u32(b[i:i+4:len(b)])) * prime1 + h = rol23(h)*prime2 + prime3 + i += 4 + } + for ; i < end; i++ { + h ^= uint64(b[i]) * prime5 + h = rol11(h) * prime1 + } + + h ^= h >> 33 + h *= prime2 + h ^= h >> 29 + h *= prime3 + h ^= h >> 32 + + return h +} + +func writeBlocks(d *Digest, b []byte) int { + v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4 + n := len(b) + for len(b) >= 32 { + v1 = round(v1, u64(b[0:8:len(b)])) + v2 = round(v2, u64(b[8:16:len(b)])) + v3 = round(v3, u64(b[16:24:len(b)])) + v4 = round(v4, u64(b[24:32:len(b)])) + b = b[32:len(b):len(b)] + } + d.v1, d.v2, d.v3, d.v4 = v1, v2, v3, v4 + return n - len(b) +} diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go new file mode 100644 index 000000000000..6f3b0cb10264 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go @@ -0,0 +1,11 @@ +package xxhash + +// Sum64String computes the 64-bit xxHash digest of s. +func Sum64String(s string) uint64 { + return Sum64([]byte(s)) +} + +// WriteString adds more data to d. It always returns len(s), nil. +func (d *Digest) WriteString(s string) (n int, err error) { + return d.Write([]byte(s)) +} diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec.go b/vendor/github.com/klauspost/compress/zstd/seqdec.go new file mode 100644 index 000000000000..b5c8ef133296 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqdec.go @@ -0,0 +1,485 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "errors" + "fmt" + "io" +) + +type seq struct { + litLen uint32 + matchLen uint32 + offset uint32 + + // Codes are stored here for the encoder + // so they only have to be looked up once. + llCode, mlCode, ofCode uint8 +} + +func (s seq) String() string { + if s.offset <= 3 { + if s.offset == 0 { + return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset: INVALID (0)") + } + return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset, " (repeat)") + } + return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset-3, " (new)") +} + +type seqCompMode uint8 + +const ( + compModePredefined seqCompMode = iota + compModeRLE + compModeFSE + compModeRepeat +) + +type sequenceDec struct { + // decoder keeps track of the current state and updates it from the bitstream. + fse *fseDecoder + state fseState + repeat bool +} + +// init the state of the decoder with input from stream. +func (s *sequenceDec) init(br *bitReader) error { + if s.fse == nil { + return errors.New("sequence decoder not defined") + } + s.state.init(br, s.fse.actualTableLog, s.fse.dt[:1<= 0; i-- { + if br.overread() { + printf("reading sequence %d, exceeded available data\n", seqs-i) + return io.ErrUnexpectedEOF + } + 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("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 { + ll, mo, ml = s.next(br, llState, mlState, ofState) + br.fill() + } + + if debugSequences { + println("Seq", seqs-i-1, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml) + } + + 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) + if size-startSize > maxBlockSize { + return fmt.Errorf("output (%d) bigger than max block size", size) + } + if size > cap(s.out) { + // Not enough size, will be extremely rarely triggered, + // but could be if destination slice is too small for sync operations. + // We add maxBlockSize to the capacity. + s.out = append(s.out, make([]byte, maxBlockSize)...) + s.out = s.out[:len(s.out)-maxBlockSize] + } + 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]...) + 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 len(s.dict) == 0 { + return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(s.out)+len(hist)) + } + + // we may be in dictionary. + dictO := len(s.dict) - (mo - (len(s.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)) + } + 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]...) + mo = 0 + ml = 0 + } + } + + // Copy from history. + // TODO: Blocks without history could be made to ignore this completely. + if v := mo - len(s.out); v > 0 { + // v is the start position in history from end. + start := len(s.hist) - v + if ml > v { + // Some goes into current block. + // Copy remainder of history + out = append(out, s.hist[start:]...) + mo -= v + ml -= v + } else { + out = append(out, s.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 { + // No overlap + out = append(out, s.out[start:start+ml]...) + } else { + // Overlapping copy + // Extend destination slice and copy one byte at the time. + out = out[:len(out)+ml] + src := out[start : start+ml] + // Destination is the space we just added. + dst := out[len(out)-ml:] + dst = dst[:len(src)] + for i := range src { + dst[i] = src[i] + } + } + } + s.out = out + if i == 0 { + // 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.getBitsFast(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] + } + } + + // Add final literals + s.out = append(s.out, s.literals...) + return nil +} + +// 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) +} + +var bitMask [16]uint16 + +func init() { + for i := range bitMask[:] { + bitMask[i] = uint16((1 << uint(i)) - 1) + } +} + +// 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() + ml, mlB := mlState.final() + mo, moB := ofState.final() + + // extra bits are stored in reverse order. + br.fill() + if s.maxBits <= 32 { + mo += br.getBits(moB) + ml += br.getBits(mlB) + ll += br.getBits(llB) + } else { + mo += br.getBits(moB) + br.fill() + // matchlength+literal length, max 32 bits + ml += br.getBits(mlB) + ll += br.getBits(llB) + + } + mo = s.adjustOffset(mo, ll, moB) + return +} + +func (s *sequenceDecs) adjustOffset(offset, litLen int, offsetB uint8) int { + if offsetB > 1 { + s.prevOffset[2] = s.prevOffset[1] + s.prevOffset[1] = s.prevOffset[0] + s.prevOffset[0] = offset + return offset + } + + if litLen == 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. + offset++ + } + + if offset == 0 { + return s.prevOffset[0] + } + var temp int + if offset == 3 { + temp = s.prevOffset[0] - 1 + } else { + temp = s.prevOffset[offset] + } + + if temp == 0 { + // 0 is not valid; input is corrupted; force offset to 1 + println("temp was 0") + temp = 1 + } + + if offset != 1 { + s.prevOffset[2] = s.prevOffset[1] + } + s.prevOffset[1] = s.prevOffset[0] + 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/seqenc.go b/vendor/github.com/klauspost/compress/zstd/seqenc.go new file mode 100644 index 000000000000..36bcc3cc02eb --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/seqenc.go @@ -0,0 +1,115 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import "math/bits" + +type seqCoders struct { + llEnc, ofEnc, mlEnc *fseEncoder + llPrev, ofPrev, mlPrev *fseEncoder +} + +// swap coders with another (block). +func (s *seqCoders) swap(other *seqCoders) { + *s, *other = *other, *s +} + +// setPrev will update the previous encoders to the actually used ones +// and make sure a fresh one is in the main slot. +func (s *seqCoders) setPrev(ll, ml, of *fseEncoder) { + compareSwap := func(used *fseEncoder, current, prev **fseEncoder) { + // We used the new one, more current to history and reuse the previous history + if *current == used { + *prev, *current = *current, *prev + c := *current + p := *prev + c.reUsed = false + p.reUsed = true + return + } + if used == *prev { + return + } + // Ensure we cannot reuse by accident + prevEnc := *prev + prevEnc.symbolLen = 0 + return + } + compareSwap(ll, &s.llEnc, &s.llPrev) + compareSwap(ml, &s.mlEnc, &s.mlPrev) + compareSwap(of, &s.ofEnc, &s.ofPrev) +} + +func highBit(val uint32) (n uint32) { + return uint32(bits.Len32(val) - 1) +} + +var llCodeTable = [64]byte{0, 1, 2, 3, 4, 5, 6, 7, + 8, 9, 10, 11, 12, 13, 14, 15, + 16, 16, 17, 17, 18, 18, 19, 19, + 20, 20, 20, 20, 21, 21, 21, 21, + 22, 22, 22, 22, 22, 22, 22, 22, + 23, 23, 23, 23, 23, 23, 23, 23, + 24, 24, 24, 24, 24, 24, 24, 24, + 24, 24, 24, 24, 24, 24, 24, 24} + +// Up to 6 bits +const maxLLCode = 35 + +// llBitsTable translates from ll code to number of bits. +var llBitsTable = [maxLLCode + 1]byte{ + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 1, 1, 1, 1, 2, 2, 3, 3, + 4, 6, 7, 8, 9, 10, 11, 12, + 13, 14, 15, 16} + +// llCode returns the code that represents the literal length requested. +func llCode(litLength uint32) uint8 { + const llDeltaCode = 19 + if litLength <= 63 { + // Compiler insists on bounds check (Go 1.12) + return llCodeTable[litLength&63] + } + return uint8(highBit(litLength)) + llDeltaCode +} + +var mlCodeTable = [128]byte{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, + 32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37, + 38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39, + 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, + 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, + 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, + 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42} + +// Up to 6 bits +const maxMLCode = 52 + +// mlBitsTable translates from ml code to number of bits. +var mlBitsTable = [maxMLCode + 1]byte{ + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 1, 1, 1, 1, 2, 2, 3, 3, + 4, 4, 5, 7, 8, 9, 10, 11, + 12, 13, 14, 15, 16} + +// note : mlBase = matchLength - MINMATCH; +// because it's the format it's stored in seqStore->sequences +func mlCode(mlBase uint32) uint8 { + const mlDeltaCode = 36 + if mlBase <= 127 { + // Compiler insists on bounds check (Go 1.12) + return mlCodeTable[mlBase&127] + } + return uint8(highBit(mlBase)) + mlDeltaCode +} + +func ofCode(offset uint32) uint8 { + // A valid offset will always be > 0. + return uint8(bits.Len32(offset) - 1) +} diff --git a/vendor/github.com/klauspost/compress/zstd/snappy.go b/vendor/github.com/klauspost/compress/zstd/snappy.go new file mode 100644 index 000000000000..841fd95acce9 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/snappy.go @@ -0,0 +1,436 @@ +// Copyright 2019+ Klaus Post. All rights reserved. +// License information can be found in the LICENSE file. +// Based on work by Yann Collet, released under BSD License. + +package zstd + +import ( + "encoding/binary" + "errors" + "hash/crc32" + "io" + + "github.com/klauspost/compress/huff0" + "github.com/klauspost/compress/snappy" +) + +const ( + snappyTagLiteral = 0x00 + snappyTagCopy1 = 0x01 + snappyTagCopy2 = 0x02 + snappyTagCopy4 = 0x03 +) + +const ( + snappyChecksumSize = 4 + snappyMagicBody = "sNaPpY" + + // snappyMaxBlockSize 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". + snappyMaxBlockSize = 65536 + + // snappyMaxEncodedLenOfMaxBlockSize equals MaxEncodedLen(snappyMaxBlockSize), 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. + snappyMaxEncodedLenOfMaxBlockSize = 76490 +) + +const ( + chunkTypeCompressedData = 0x00 + chunkTypeUncompressedData = 0x01 + chunkTypePadding = 0xfe + chunkTypeStreamIdentifier = 0xff +) + +var ( + // ErrSnappyCorrupt reports that the input is invalid. + ErrSnappyCorrupt = errors.New("snappy: corrupt input") + // ErrSnappyTooLarge reports that the uncompressed length is too large. + ErrSnappyTooLarge = errors.New("snappy: decoded block is too large") + // ErrSnappyUnsupported reports that the input isn't supported. + ErrSnappyUnsupported = errors.New("snappy: unsupported input") + + errUnsupportedLiteralLength = errors.New("snappy: unsupported literal length") +) + +// SnappyConverter can read SnappyConverter-compressed streams and convert them to zstd. +// Conversion is done by converting the stream directly from Snappy without intermediate +// full decoding. +// Therefore the compression ratio is much less than what can be done by a full decompression +// and compression, and a faulty Snappy stream may lead to a faulty Zstandard stream without +// any errors being generated. +// No CRC value is being generated and not all CRC values of the Snappy stream are checked. +// However, it provides really fast recompression of Snappy streams. +// The converter can be reused to avoid allocations, even after errors. +type SnappyConverter struct { + r io.Reader + err error + buf []byte + block *blockEnc +} + +// Convert the Snappy stream supplied in 'in' and write the zStandard stream to 'w'. +// If any error is detected on the Snappy stream it is returned. +// The number of bytes written is returned. +func (r *SnappyConverter) Convert(in io.Reader, w io.Writer) (int64, error) { + initPredefined() + r.err = nil + r.r = in + if r.block == nil { + r.block = &blockEnc{} + r.block.init() + } + r.block.initNewEncode() + if len(r.buf) != snappyMaxEncodedLenOfMaxBlockSize+snappyChecksumSize { + r.buf = make([]byte, snappyMaxEncodedLenOfMaxBlockSize+snappyChecksumSize) + } + r.block.litEnc.Reuse = huff0.ReusePolicyNone + var written int64 + var readHeader bool + { + var header []byte + var n int + header, r.err = frameHeader{WindowSize: snappyMaxBlockSize}.appendTo(r.buf[:0]) + + n, r.err = w.Write(header) + if r.err != nil { + return written, r.err + } + written += int64(n) + } + + for { + if !r.readFull(r.buf[:4], true) { + // Add empty last block + r.block.reset(nil) + r.block.last = true + err := r.block.encodeLits(r.block.literals, false) + if err != nil { + return written, err + } + n, err := w.Write(r.block.output) + if err != nil { + return written, err + } + written += int64(n) + + return written, r.err + } + chunkType := r.buf[0] + if !readHeader { + if chunkType != chunkTypeStreamIdentifier { + println("chunkType != chunkTypeStreamIdentifier", chunkType) + r.err = ErrSnappyCorrupt + return written, r.err + } + readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + if chunkLen > len(r.buf) { + println("chunkLen > len(r.buf)", chunkType) + r.err = ErrSnappyUnsupported + return written, 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 < snappyChecksumSize { + println("chunkLen < snappyChecksumSize", chunkLen, snappyChecksumSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return written, r.err + } + //checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[snappyChecksumSize:] + + n, hdr, err := snappyDecodedLen(buf) + if err != nil { + r.err = err + return written, r.err + } + buf = buf[hdr:] + if n > snappyMaxBlockSize { + println("n > snappyMaxBlockSize", n, snappyMaxBlockSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + r.block.reset(nil) + r.block.pushOffsets() + if err := decodeSnappy(r.block, buf); err != nil { + r.err = err + return written, r.err + } + if r.block.size+r.block.extraLits != n { + printf("invalid size, want %d, got %d\n", n, r.block.size+r.block.extraLits) + r.err = ErrSnappyCorrupt + return written, r.err + } + err = r.block.encode(nil, false, false) + switch err { + case errIncompressible: + r.block.popOffsets() + r.block.reset(nil) + r.block.literals, err = snappy.Decode(r.block.literals[:n], r.buf[snappyChecksumSize:chunkLen]) + if err != nil { + println("snappy.Decode:", err) + return written, err + } + err = r.block.encodeLits(r.block.literals, false) + if err != nil { + return written, err + } + case nil: + default: + return written, err + } + + n, r.err = w.Write(r.block.output) + if r.err != nil { + return written, err + } + written += int64(n) + continue + case chunkTypeUncompressedData: + if debug { + println("Uncompressed, chunklen", chunkLen) + } + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < snappyChecksumSize { + println("chunkLen < snappyChecksumSize", chunkLen, snappyChecksumSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + r.block.reset(nil) + buf := r.buf[:snappyChecksumSize] + if !r.readFull(buf, false) { + return written, 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 - snappyChecksumSize + if n > snappyMaxBlockSize { + println("n > snappyMaxBlockSize", n, snappyMaxBlockSize) + r.err = ErrSnappyCorrupt + return written, r.err + } + r.block.literals = r.block.literals[:n] + if !r.readFull(r.block.literals, false) { + return written, r.err + } + if snappyCRC(r.block.literals) != checksum { + println("literals crc mismatch") + r.err = ErrSnappyCorrupt + return written, r.err + } + err := r.block.encodeLits(r.block.literals, false) + if err != nil { + return written, err + } + n, r.err = w.Write(r.block.output) + if r.err != nil { + return written, err + } + written += int64(n) + continue + + case chunkTypeStreamIdentifier: + if debug { + println("stream id", chunkLen, len(snappyMagicBody)) + } + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(snappyMagicBody) { + println("chunkLen != len(snappyMagicBody)", chunkLen, len(snappyMagicBody)) + r.err = ErrSnappyCorrupt + return written, r.err + } + if !r.readFull(r.buf[:len(snappyMagicBody)], false) { + return written, r.err + } + for i := 0; i < len(snappyMagicBody); i++ { + if r.buf[i] != snappyMagicBody[i] { + println("r.buf[i] != snappyMagicBody[i]", r.buf[i], snappyMagicBody[i], i) + r.err = ErrSnappyCorrupt + return written, r.err + } + } + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + println("chunkType <= 0x7f") + r.err = ErrSnappyUnsupported + return written, 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 written, r.err + } + } +} + +// decodeSnappy writes the decoding of src to dst. It assumes that the varint-encoded +// length of the decompressed bytes has already been read. +func decodeSnappy(blk *blockEnc, src []byte) error { + //decodeRef(make([]byte, snappyMaxBlockSize), src) + var s, length int + lits := blk.extraLits + var offset uint32 + for s < len(src) { + switch src[s] & 0x03 { + case snappyTagLiteral: + 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. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + 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. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + 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. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + 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. + println("uint(s) > uint(len(src)", s, src) + return ErrSnappyCorrupt + } + x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + if x > snappyMaxBlockSize { + println("x > snappyMaxBlockSize", x, snappyMaxBlockSize) + return ErrSnappyCorrupt + } + length = int(x) + 1 + if length <= 0 { + println("length <= 0 ", length) + + return errUnsupportedLiteralLength + } + //if length > snappyMaxBlockSize-d || uint32(length) > len(src)-s { + // return ErrSnappyCorrupt + //} + + blk.literals = append(blk.literals, src[s:s+length]...) + //println(length, "litLen") + lits += length + s += length + continue + + case snappyTagCopy1: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, len(src)) + return ErrSnappyCorrupt + } + length = 4 + int(src[s-2])>>2&0x7 + offset = uint32(src[s-2])&0xe0<<3 | uint32(src[s-1]) + + case snappyTagCopy2: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, len(src)) + return ErrSnappyCorrupt + } + length = 1 + int(src[s-3])>>2 + offset = uint32(src[s-2]) | uint32(src[s-1])<<8 + + case snappyTagCopy4: + s += 5 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + println("uint(s) > uint(len(src)", s, len(src)) + return ErrSnappyCorrupt + } + length = 1 + int(src[s-5])>>2 + offset = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24 + } + + if offset <= 0 || blk.size+lits < int(offset) /*|| length > len(blk)-d */ { + println("offset <= 0 || blk.size+lits < int(offset)", offset, blk.size+lits, int(offset), blk.size, lits) + + return ErrSnappyCorrupt + } + + // Check if offset is one of the recent offsets. + // Adjusts the output offset accordingly. + // Gives a tiny bit of compression, typically around 1%. + if false { + offset = blk.matchOffset(offset, uint32(lits)) + } else { + offset += 3 + } + + blk.sequences = append(blk.sequences, seq{ + litLen: uint32(lits), + offset: offset, + matchLen: uint32(length) - zstdMinMatch, + }) + blk.size += length + lits + lits = 0 + } + blk.extraLits = lits + return nil +} + +func (r *SnappyConverter) 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 = ErrSnappyCorrupt + } + return false + } + return true +} + +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 snappyCRC(b []byte) uint32 { + c := crc32.Update(0, crcTable, b) + return uint32(c>>15|c<<17) + 0xa282ead8 +} + +// snappyDecodedLen returns the length of the decoded block and the number of bytes +// that the length header occupied. +func snappyDecodedLen(src []byte) (blockLen, headerLen int, err error) { + v, n := binary.Uvarint(src) + if n <= 0 || v > 0xffffffff { + return 0, 0, ErrSnappyCorrupt + } + + const wordSize = 32 << (^uint(0) >> 32 & 1) + if wordSize == 32 && v > 0x7fffffff { + return 0, 0, ErrSnappyTooLarge + } + return int(v), n, nil +} diff --git a/vendor/github.com/klauspost/compress/zstd/zstd.go b/vendor/github.com/klauspost/compress/zstd/zstd.go new file mode 100644 index 000000000000..0807719c8b90 --- /dev/null +++ b/vendor/github.com/klauspost/compress/zstd/zstd.go @@ -0,0 +1,144 @@ +// Package zstd provides decompression of zstandard files. +// +// For advanced usage and examples, go to the README: https://github.com/klauspost/compress/tree/master/zstd#zstd +package zstd + +import ( + "errors" + "log" + "math" + "math/bits" +) + +// enable debug printing +const debug = false + +// Enable extra assertions. +const debugAsserts = debug || false + +// print sequence details +const debugSequences = false + +// print detailed matching information +const debugMatches = false + +// force encoder to use predefined tables. +const forcePreDef = false + +// zstdMinMatch is the minimum zstd match length. +const zstdMinMatch = 3 + +// Reset the buffer offset when reaching this. +const bufferReset = math.MaxInt32 - MaxWindowSize + +var ( + // ErrReservedBlockType is returned when a reserved block type is found. + // Typically this indicates wrong or corrupted input. + ErrReservedBlockType = errors.New("invalid input: reserved block type encountered") + + // ErrCompressedSizeTooBig is returned when a block is bigger than allowed. + // Typically this indicates wrong or corrupted input. + ErrCompressedSizeTooBig = errors.New("invalid input: compressed size too big") + + // ErrBlockTooSmall is returned when a block is too small to be decoded. + // Typically returned on invalid input. + ErrBlockTooSmall = errors.New("block too small") + + // 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") + + // ErrWindowSizeExceeded is returned when a reference exceeds the valid window size. + // Typically this indicates wrong or corrupted input. + ErrWindowSizeExceeded = errors.New("window size exceeded") + + // ErrWindowSizeTooSmall is returned when no window size is specified. + // Typically this indicates wrong or corrupted input. + ErrWindowSizeTooSmall = errors.New("invalid input: window size was too small") + + // ErrDecoderSizeExceeded is returned if decompressed size exceeds the configured limit. + ErrDecoderSizeExceeded = errors.New("decompressed size exceeds configured limit") + + // ErrUnknownDictionary is returned if the dictionary ID is unknown. + // For the time being dictionaries are not supported. + ErrUnknownDictionary = errors.New("unknown dictionary") + + // ErrFrameSizeExceeded is returned if the stated frame size is exceeded. + // This is only returned if SingleSegment is specified on the frame. + ErrFrameSizeExceeded = errors.New("frame size exceeded") + + // ErrCRCMismatch is returned if CRC mismatches. + ErrCRCMismatch = errors.New("CRC check failed") + + // ErrDecoderClosed will be returned if the Decoder was used after + // Close has been called. + ErrDecoderClosed = errors.New("decoder used after Close") +) + +func println(a ...interface{}) { + if debug { + log.Println(a...) + } +} + +func printf(format string, a ...interface{}) { + if debug { + 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. +func matchLen(a, b []byte) int { + b = b[:len(a)] + for i := 0; i < len(a)-7; i += 8 { + if diff := load64(a, i) ^ load64(b, i); diff != 0 { + return i + (bits.TrailingZeros64(diff) >> 3) + } + } + + checked := (len(a) >> 3) << 3 + a = a[checked:] + b = b[checked:] + for i := range a { + if a[i] != b[i] { + return i + checked + } + } + return len(a) + checked +} + +func load3232(b []byte, i int32) uint32 { + // Help the compiler eliminate bounds checks on the read so it can be done in a single read. + b = b[i:] + b = b[:4] + return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 +} + +func load6432(b []byte, i int32) uint64 { + // Help the compiler eliminate bounds checks on the read so it can be done in a single read. + b = b[i:] + b = b[:8] + 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 +} + +func load64(b []byte, i int) uint64 { + // Help the compiler eliminate bounds checks on the read so it can be done in a single read. + b = b[i:] + b = b[:8] + 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 +} diff --git a/vendor/github.com/pierrec/lz4/v4/go.mod b/vendor/github.com/pierrec/lz4/v4/go.mod index 96d69d005952..42229b29672b 100644 --- a/vendor/github.com/pierrec/lz4/v4/go.mod +++ b/vendor/github.com/pierrec/lz4/v4/go.mod @@ -1,10 +1,3 @@ module github.com/pierrec/lz4/v4 go 1.14 - -require ( - code.cloudfoundry.org/bytefmt v0.0.0-20200131002437-cf55d5288a48 - github.com/onsi/ginkgo v1.14.0 // indirect - github.com/pierrec/cmdflag v0.0.2 - github.com/schollz/progressbar/v3 v3.3.4 -) diff --git a/vendor/github.com/pierrec/lz4/v4/go.sum b/vendor/github.com/pierrec/lz4/v4/go.sum index 25db567f7400..6973bd668a8f 100644 --- a/vendor/github.com/pierrec/lz4/v4/go.sum +++ b/vendor/github.com/pierrec/lz4/v4/go.sum @@ -1,81 +1,3 @@ -code.cloudfoundry.org/bytefmt v0.0.0-20200131002437-cf55d5288a48 h1:/EMHruHCFXR9xClkGV/t0rmHrdhX4+trQUcBqjwc9xE= -code.cloudfoundry.org/bytefmt v0.0.0-20200131002437-cf55d5288a48/go.mod h1:wN/zk7mhREp/oviagqUXY3EwuHhWyOvAdsn5Y4CzOrc= -github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38= -github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c= -github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38= -github.com/fsnotify/fsnotify v1.4.7/go.mod h1:jwhsz4b93w/PPRr/qN1Yymfu8t87LnFCMoQvtojpjFo= -github.com/fsnotify/fsnotify v1.4.9 h1:hsms1Qyu0jgnwNXIxa+/V/PDsU6CfLf6CNO8H7IWoS4= -github.com/fsnotify/fsnotify v1.4.9/go.mod h1:znqG4EE+3YCdAaPaxE2ZRY/06pZUdp0tY4IgpuI1SZQ= -github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U= -github.com/golang/protobuf v1.4.0-rc.1/go.mod h1:ceaxUfeHdC40wWswd/P6IGgMaK3YpKi5j83Wpe3EHw8= -github.com/golang/protobuf v1.4.0-rc.1.0.20200221234624-67d41d38c208/go.mod h1:xKAWHe0F5eneWXFV3EuXVDTCmh+JuBKY0li0aMyXATA= -github.com/golang/protobuf v1.4.0-rc.2/go.mod h1:LlEzMj4AhA7rCAGe4KMBDvJI+AwstrUpVNzEA03Pprs= -github.com/golang/protobuf v1.4.0-rc.4.0.20200313231945-b860323f09d0/go.mod h1:WU3c8KckQ9AFe+yFwt9sWVRKCVIyN9cPHBJSNnbL67w= -github.com/golang/protobuf v1.4.0/go.mod h1:jodUvKwWbYaEsadDk5Fwe5c77LiNKVO9IDvqG2KuDX0= -github.com/golang/protobuf v1.4.2 h1:+Z5KGCizgyZCbGh1KZqA0fcLLkwbsjIzS4aV2v7wJX0= -github.com/golang/protobuf v1.4.2/go.mod h1:oDoupMAO8OvCJWAcko0GGGIgR6R6ocIYbsSw735rRwI= -github.com/google/go-cmp v0.3.0/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU= -github.com/google/go-cmp v0.3.1/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU= -github.com/google/go-cmp v0.4.0 h1:xsAVV57WRhGj6kEIi8ReJzQlHHqcBYCElAvkovg3B/4= -github.com/google/go-cmp v0.4.0/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= -github.com/hpcloud/tail v1.0.0/go.mod h1:ab1qPbhIpdTxEkNHXyeSf5vhxWSCs/tWer42PpOxQnU= -github.com/k0kubun/go-ansi v0.0.0-20180517002512-3bf9e2903213/go.mod h1:vNUNkEQ1e29fT/6vq2aBdFsgNPmy8qMdSay1npru+Sw= -github.com/mattn/go-isatty v0.0.12/go.mod h1:cbi8OIDigv2wuxKPP5vlRcQ1OAZbq2CE4Kysco4FUpU= -github.com/mattn/go-runewidth v0.0.9 h1:Lm995f3rfxdpd6TSmuVCHVb/QhupuXlYr8sCI/QdE+0= -github.com/mattn/go-runewidth v0.0.9/go.mod h1:H031xJmbD/WCDINGzjvQ9THkh0rPKHF+m2gUSrubnMI= -github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db h1:62I3jR2EmQ4l5rM/4FEfDWcRD+abF5XlKShorW5LRoQ= -github.com/mitchellh/colorstring v0.0.0-20190213212951-d06e56a500db/go.mod h1:l0dey0ia/Uv7NcFFVbCLtqEBQbrT4OCwCSKTEv6enCw= -github.com/nxadm/tail v1.4.4 h1:DQuhQpB1tVlglWS2hLQ5OV6B5r8aGxSrPc5Qo6uTN78= -github.com/nxadm/tail v1.4.4/go.mod h1:kenIhsEOeOJmVchQTgglprH7qJGnHDVpk1VPCcaMI8A= -github.com/onsi/ginkgo v1.6.0/go.mod h1:lLunBs/Ym6LB5Z9jYTR76FiuTmxDTDusOGeTQH+WWjE= -github.com/onsi/ginkgo v1.12.1/go.mod h1:zj2OWP4+oCPe1qIXoGWkgMRwljMUYCdkwsT2108oapk= -github.com/onsi/ginkgo v1.14.0 h1:2mOpI4JVVPBN+WQRa0WKH2eXR+Ey+uK4n7Zj0aYpIQA= -github.com/onsi/ginkgo v1.14.0/go.mod h1:iSB4RoI2tjJc9BBv4NKIKWKya62Rps+oPG/Lv9klQyY= -github.com/onsi/gomega v1.7.1/go.mod h1:XdKZgCCFLUoM/7CFJVPcG8C1xQ1AJ0vpAezJrB7JYyY= -github.com/onsi/gomega v1.10.1 h1:o0+MgICZLuZ7xjH7Vx6zS/zcu93/BEp1VwkIW1mEXCE= -github.com/onsi/gomega v1.10.1/go.mod h1:iN09h71vgCQne3DLsj+A5owkum+a2tYe+TOCB1ybHNo= -github.com/pierrec/cmdflag v0.0.2 h1:ybjGJnPr/aURn2IKWjO49znx9N0DL6YfGsIxN0PYuVY= -github.com/pierrec/cmdflag v0.0.2/go.mod h1:a3zKGZ3cdQUfxjd0RGMLZr8xI3nvpJOB+m6o/1X5BmU= -github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM= -github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4= -github.com/schollz/progressbar/v3 v3.3.4 h1:nMinx+JaEm/zJz4cEyClQeAw5rsYSB5th3xv+5lV6Vg= -github.com/schollz/progressbar/v3 v3.3.4/go.mod h1:Rp5lZwpgtYmlvmGo1FyDwXMqagyRBQYSDwzlP9QDu84= -github.com/stretchr/objx v0.1.0/go.mod h1:HFkY916IF+rwdDfMAkV7OtwuqBVzrE8GR6GFx+wExME= -github.com/stretchr/testify v1.3.0 h1:TivCn/peBQ7UY8ooIcPgZFpTNSz0Q2U6UrFlUfqbe0Q= -github.com/stretchr/testify v1.3.0/go.mod h1:M5WIy9Dh21IEIfnGCwXGc5bZfKNJtfHm1UVUgZn+9EI= -golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w= -golang.org/x/net v0.0.0-20180906233101-161cd47e91fd/go.mod h1:mL1N/T3taQHkDXs73rZJwtUhF3w3ftmwwsq0BUmARs4= -golang.org/x/net v0.0.0-20200520004742-59133d7f0dd7 h1:AeiKBIuRw3UomYXSbLy0Mc2dDLfdtbT/IVn4keq83P0= -golang.org/x/net v0.0.0-20200520004742-59133d7f0dd7/go.mod h1:qpuaurCH72eLCgpAm/N6yyVIVM9cpaDIP3A8BGJEC5A= -golang.org/x/sync v0.0.0-20180314180146-1d60e4601c6f/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM= -golang.org/x/sys v0.0.0-20180909124046-d0be0721c37e/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY= -golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY= -golang.org/x/sys v0.0.0-20190904154756-749cb33beabd/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/sys v0.0.0-20191005200804-aed5e4c7ecf9/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/sys v0.0.0-20191120155948-bd437916bb0e/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/sys v0.0.0-20200116001909-b77594299b42/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/sys v0.0.0-20200323222414-85ca7c5b95cd/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/sys v0.0.0-20200519105757-fe76b779f299/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/sys v0.0.0-20200615200032-f1bc736245b1 h1:ogLJMz+qpzav7lGMh10LMvAkM/fAoGlaiiHYiFYdm80= -golang.org/x/sys v0.0.0-20200615200032-f1bc736245b1/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= -golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ= -golang.org/x/text v0.3.2 h1:tW2bmiBqwgJj/UpqtC8EpXEZVYOwU0yG4iWbprSVAcs= -golang.org/x/text v0.3.2/go.mod h1:bEr9sfX3Q8Zfm5fL9x+3itogRgK3+ptLWKqgva+5dAk= -golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ= -golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543 h1:E7g+9GITq07hpfrRu66IVDexMakfv52eLZ2CXBWiKr4= -golang.org/x/xerrors v0.0.0-20191204190536-9bdfabe68543/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0= -google.golang.org/protobuf v0.0.0-20200109180630-ec00e32a8dfd/go.mod h1:DFci5gLYBciE7Vtevhsrf46CRTquxDuWsQurQQe4oz8= -google.golang.org/protobuf v0.0.0-20200221191635-4d8936d0db64/go.mod h1:kwYJMbMJ01Woi6D6+Kah6886xMZcty6N08ah7+eCXa0= -google.golang.org/protobuf v0.0.0-20200228230310-ab0ca4ff8a60/go.mod h1:cfTl7dwQJ+fmap5saPgwCLgHXTUD7jkjRqWcaiX5VyM= -google.golang.org/protobuf v1.20.1-0.20200309200217-e05f789c0967/go.mod h1:A+miEFZTKqfCUM6K7xSMQL9OKL/b6hQv+e19PK+JZNE= -google.golang.org/protobuf v1.21.0/go.mod h1:47Nbq4nVaFHyn7ilMalzfO3qCViNmqZ2kzikPIcrTAo= -google.golang.org/protobuf v1.23.0 h1:4MY060fB1DLGMB/7MBTLnwQUY6+F09GEiz6SsrNqyzM= -google.golang.org/protobuf v1.23.0/go.mod h1:EGpADcykh3NcUnDUJcl1+ZksZNG86OlYog2l/sGQquU= -gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405 h1:yhCVgyC4o1eVCa2tZl7eS0r+SDo693bJlVdllGtEeKM= -gopkg.in/check.v1 v0.0.0-20161208181325-20d25e280405/go.mod h1:Co6ibVJAznAaIkqp8huTwlJQCZ016jof/cbN4VW5Yz0= -gopkg.in/fsnotify.v1 v1.4.7/go.mod h1:Tz8NjZHkW78fSQdbUxIjBTcgA1z1m8ZHf0WmKUhAMys= -gopkg.in/tomb.v1 v1.0.0-20141024135613-dd632973f1e7 h1:uRGJdciOHaEIrze2W8Q3AKkepLTh2hOroT7a+7czfdQ= -gopkg.in/tomb.v1 v1.0.0-20141024135613-dd632973f1e7/go.mod h1:dt/ZhP58zS4L8KSrWDmTeBkI65Dw0HsyUHuEVlX15mw= -gopkg.in/yaml.v2 v2.2.4/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI= -gopkg.in/yaml.v2 v2.3.0 h1:clyUAQHOM3G0M3f5vQj7LuJrETvjVot3Z5el9nffUtU= -gopkg.in/yaml.v2 v2.3.0/go.mod h1:hI93XBmqTisBFMUTm0b8Fm+jr3Dg1NNxqwp+5A1VGuI= +github.com/pierrec/lz4 v1.0.1 h1:w6GMGWSsCI04fTM8wQRdnW74MuJISakuUU0onU0TYB4= +github.com/pierrec/lz4 v2.6.0+incompatible h1:Ix9yFKn1nSPBLFl/yZknTp8TU5G4Ps0JDmguYK6iH1A= +github.com/pierrec/lz4 v2.6.0+incompatible/go.mod h1:pdkljMzZIN41W+lC3N2tnIh5sFi+IEE17M5jbnwPHcY= diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go index 39015acbdd19..f38264943084 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/block.go @@ -25,23 +25,6 @@ const ( mfLimit = 10 + minMatch // The last match cannot start within the last 14 bytes. ) -// Pool of hash tables for CompressBlock. -var HashTablePool = hashTablePool{sync.Pool{New: func() interface{} { return new([htSize]int) }}} - -type hashTablePool struct { - sync.Pool -} - -func (p *hashTablePool) Get() *[htSize]int { - return p.Pool.Get().(*[htSize]int) -} - -// Zero out the table to avoid non-deterministic outputs (see issue#65). -func (p *hashTablePool) Put(t *[htSize]int) { - *t = [htSize]int{} - p.Pool.Put(t) -} - func recoverBlock(e *error) { if r := recover(); r != nil && *e == nil { *e = lz4errors.ErrInvalidSourceShortBuffer @@ -68,8 +51,54 @@ func UncompressBlock(src, dst []byte) (int, error) { return 0, lz4errors.ErrInvalidSourceShortBuffer } -func CompressBlock(src, dst []byte, hashTable []int) (_ int, err error) { - defer recoverBlock(&err) +type Compressor struct { + // Offsets are at most 64kiB, so we can store only the lower 16 bits of + // match positions: effectively, an offset from some 64kiB block boundary. + // + // When we retrieve such an offset, we interpret it as relative to the last + // block boundary si &^ 0xffff, or the one before, (si &^ 0xffff) - 0x10000, + // depending on which of these is inside the current window. If a table + // entry was generated more than 64kiB back in the input, we find out by + // inspecting the input stream. + table [htSize]uint16 + + needsReset bool +} + +// Get returns the position of a presumptive match for the hash h. +// The match may be a false positive due to a hash collision or an old entry. +// If si < winSize, the return value may be negative. +func (c *Compressor) get(h uint32, si int) int { + h &= htSize - 1 + i := int(c.table[h]) + i += si &^ winMask + if i >= si { + // Try previous 64kiB block (negative when in first block). + i -= winSize + } + return i +} + +func (c *Compressor) put(h uint32, si int) { + h &= htSize - 1 + c.table[h] = uint16(si) +} + +var compressorPool = sync.Pool{New: func() interface{} { return new(Compressor) }} + +func CompressBlock(src, dst []byte) (int, error) { + c := compressorPool.Get().(*Compressor) + n, err := c.CompressBlock(src, dst) + compressorPool.Put(c) + return n, err +} + +func (c *Compressor) CompressBlock(src, dst []byte) (int, error) { + if c.needsReset { + // Zero out reused table to avoid non-deterministic output (issue #65). + c.table = [htSize]uint16{} + } + c.needsReset = true // Only false on first call. // Return 0, nil only if the destination buffer size is < CompressBlockBound. isNotCompressible := len(dst) < CompressBlockBound(len(src)) @@ -87,15 +116,6 @@ func CompressBlock(src, dst []byte, hashTable []int) (_ int, err error) { goto lastLiterals } - if cap(hashTable) < htSize { - poolTable := HashTablePool.Get() - defer HashTablePool.Put(poolTable) - hashTable = poolTable[:] - } else { - hashTable = hashTable[:htSize] - } - _ = hashTable[htSize-1] - // Fast scan strategy: the hash table only stores the last 4 bytes sequences. for si < sn { // Hash the next 6 bytes (sequence)... @@ -105,33 +125,30 @@ func CompressBlock(src, dst []byte, hashTable []int) (_ int, err error) { // We check a match at s, s+1 and s+2 and pick the first one we get. // Checking 3 only requires us to load the source one. - ref := hashTable[h] - ref2 := hashTable[h2] - hashTable[h] = si - hashTable[h2] = si + 1 + ref := c.get(h, si) + ref2 := c.get(h2, si) + c.put(h, si) + c.put(h2, si+1) + offset := si - ref - // If offset <= 0 we got an old entry in the hash table. - if offset <= 0 || offset >= winSize || // Out of window. - uint32(match) != binary.LittleEndian.Uint32(src[ref:]) { // Hash collision on different matches. + if offset <= 0 || offset >= winSize || uint32(match) != binary.LittleEndian.Uint32(src[ref:]) { // No match. Start calculating another hash. // The processor can usually do this out-of-order. h = blockHash(match >> 16) - ref = hashTable[h] + ref3 := c.get(h, si+2) // Check the second match at si+1 si += 1 offset = si - ref2 - if offset <= 0 || offset >= winSize || - uint32(match>>8) != binary.LittleEndian.Uint32(src[ref2:]) { + if offset <= 0 || offset >= winSize || uint32(match>>8) != binary.LittleEndian.Uint32(src[ref2:]) { // No match. Check the third match at si+2 si += 1 - offset = si - ref - hashTable[h] = si + offset = si - ref3 + c.put(h, si) - if offset <= 0 || offset >= winSize || - uint32(match>>16) != binary.LittleEndian.Uint32(src[ref:]) { + if offset <= 0 || offset >= winSize || uint32(match>>16) != binary.LittleEndian.Uint32(src[ref3:]) { // Skip one extra byte (at si+3) before we check 3 matches again. si += 2 + (si-anchor)>>adaptSkipLog continue @@ -192,20 +209,28 @@ func CompressBlock(src, dst []byte, hashTable []int) (_ int, err error) { di++ // Literals. + if di+lLen > len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } copy(dst[di:di+lLen], src[anchor:anchor+lLen]) di += lLen + 2 anchor = si // Encode offset. - _ = dst[di] // Bound check elimination. + if di > len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } dst[di-2], dst[di-1] = byte(offset), byte(offset>>8) // Encode match length part 2. if mLen >= 0xF { - for mLen -= 0xF; mLen >= 0xFF; mLen -= 0xFF { + for mLen -= 0xF; mLen >= 0xFF && di < len(dst); mLen -= 0xFF { dst[di] = 0xFF di++ } + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } dst[di] = byte(mLen) di++ } @@ -215,7 +240,7 @@ func CompressBlock(src, dst []byte, hashTable []int) (_ int, err error) { } // Hash match end-2 h = blockHash(binary.LittleEndian.Uint64(src[si-2:])) - hashTable[h] = si - 2 + c.put(h, si-2) } lastLiterals: @@ -225,16 +250,22 @@ lastLiterals: } // Last literals. + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } lLen := len(src) - anchor if lLen < 0xF { dst[di] = byte(lLen << 4) } else { dst[di] = 0xF0 di++ - for lLen -= 0xF; lLen >= 0xFF; lLen -= 0xFF { + for lLen -= 0xF; lLen >= 0xFF && di < len(dst); lLen -= 0xFF { dst[di] = 0xFF di++ } + if di >= len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } dst[di] = byte(lLen) } di++ @@ -244,6 +275,9 @@ lastLiterals: // Incompressible. return 0, nil } + if di+len(src)-anchor > len(dst) { + return 0, lz4errors.ErrInvalidSourceShortBuffer + } di += copy(dst[di:di+len(src)-anchor], src[anchor:]) return di, nil } @@ -254,7 +288,30 @@ func blockHashHC(x uint32) uint32 { return x * hasher >> (32 - winSizeLog) } -func CompressBlockHC(src, dst []byte, depth CompressionLevel, hashTable, chainTable []int) (_ int, err error) { +type CompressorHC struct { + // hashTable: stores the last position found for a given hash + // chainTable: stores previous positions for a given hash + hashTable, chainTable [htSize]int + needsReset bool +} + +var compressorHCPool = sync.Pool{New: func() interface{} { return new(CompressorHC) }} + +func CompressBlockHC(src, dst []byte, depth CompressionLevel) (int, error) { + c := compressorHCPool.Get().(*CompressorHC) + n, err := c.CompressBlock(src, dst, depth) + compressorHCPool.Put(c) + return n, err +} + +func (c *CompressorHC) CompressBlock(src, dst []byte, depth CompressionLevel) (_ int, err error) { + if c.needsReset { + // Zero out reused table to avoid non-deterministic output (issue #65). + c.hashTable = [htSize]int{} + c.chainTable = [htSize]int{} + } + c.needsReset = true // Only false on first call. + defer recoverBlock(&err) // Return 0, nil only if the destination buffer size is < CompressBlockBound. @@ -271,25 +328,6 @@ func CompressBlockHC(src, dst []byte, depth CompressionLevel, hashTable, chainTa goto lastLiterals } - // hashTable: stores the last position found for a given hash - // chainTable: stores previous positions for a given hash - if cap(hashTable) < htSize { - poolTable := HashTablePool.Get() - defer HashTablePool.Put(poolTable) - hashTable = poolTable[:] - } else { - hashTable = hashTable[:htSize] - } - _ = hashTable[htSize-1] - if cap(chainTable) < htSize { - poolTable := HashTablePool.Get() - defer HashTablePool.Put(poolTable) - chainTable = poolTable[:] - } else { - chainTable = chainTable[:htSize] - } - _ = chainTable[htSize-1] - if depth == 0 { depth = winSize } @@ -302,7 +340,7 @@ func CompressBlockHC(src, dst []byte, depth CompressionLevel, hashTable, chainTa // Follow the chain until out of window and give the longest match. mLen := 0 offset := 0 - for next, try := hashTable[h], depth; try > 0 && next > 0 && si-next < winSize; next, try = chainTable[next&winMask], try-1 { + for next, try := c.hashTable[h], depth; try > 0 && next > 0 && si-next < winSize; next, try = c.chainTable[next&winMask], try-1 { // The first (mLen==0) or next byte (mLen>=minMatch) at current match length // must match to improve on the match length. if src[next+mLen] != src[si+mLen] { @@ -329,8 +367,8 @@ func CompressBlockHC(src, dst []byte, depth CompressionLevel, hashTable, chainTa offset = si - next // Try another previous position with the same hash. } - chainTable[si&winMask] = hashTable[h] - hashTable[h] = si + c.chainTable[si&winMask] = c.hashTable[h] + c.hashTable[h] = si // No match found. if mLen == 0 { @@ -349,8 +387,8 @@ func CompressBlockHC(src, dst []byte, depth CompressionLevel, hashTable, chainTa match >>= 8 match |= uint32(src[si+3]) << 24 h := blockHashHC(match) - chainTable[si&winMask] = hashTable[h] - hashTable[h] = si + c.chainTable[si&winMask] = c.hashTable[h] + c.hashTable[h] = si si++ } diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go index 098a5cf9b968..e6cf88d71c03 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/blocks.go @@ -8,6 +8,8 @@ const ( Block256Kb Block1Mb Block4Mb + Block8Mb = 2 * Block4Mb + legacyBlockSize = Block8Mb + Block8Mb/255 + 16 // CompressBound(Block8Mb) ) var ( @@ -15,6 +17,7 @@ var ( BlockPool256K = sync.Pool{New: func() interface{} { return make([]byte, Block256Kb) }} BlockPool1M = sync.Pool{New: func() interface{} { return make([]byte, Block1Mb) }} BlockPool4M = sync.Pool{New: func() interface{} { return make([]byte, Block4Mb) }} + BlockPool8M = sync.Pool{New: func() interface{} { return make([]byte, legacyBlockSize) }} ) func Index(b uint32) BlockSizeIndex { @@ -27,6 +30,8 @@ func Index(b uint32) BlockSizeIndex { return 6 case Block4Mb: return 7 + case Block8Mb: // only valid in legacy mode + return 3 } return 0 } @@ -56,29 +61,25 @@ func (b BlockSizeIndex) Get() []byte { buf = BlockPool1M.Get() case 7: buf = BlockPool4M.Get() + case 3: + buf = BlockPool8M.Get() } return buf.([]byte) } -func (b BlockSizeIndex) Put(buf []byte) { +func Put(buf []byte) { // Safeguard: do not allow invalid buffers. - switch c := uint32(cap(buf)); b { - case 4: - if c == Block64Kb { - BlockPool64K.Put(buf[:c]) - } - case 5: - if c == Block256Kb { - BlockPool256K.Put(buf[:c]) - } - case 6: - if c == Block1Mb { - BlockPool1M.Put(buf[:c]) - } - case 7: - if c == Block4Mb { - BlockPool4M.Put(buf[:c]) - } + switch c := cap(buf); uint32(c) { + case Block64Kb: + BlockPool64K.Put(buf[:c]) + case Block256Kb: + BlockPool256K.Put(buf[:c]) + case Block1Mb: + BlockPool1M.Put(buf[:c]) + case Block4Mb: + BlockPool4M.Put(buf[:c]) + case legacyBlockSize: + BlockPool8M.Put(buf[:c]) } } diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s index a473a9bfc145..be79faa3fe85 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.s @@ -26,6 +26,8 @@ TEXT ·decodeBlock(SB), NOSPLIT, $64-56 MOVQ src_base+24(FP), SI MOVQ src_len+32(FP), R9 + CMPQ R9, $0 + JE err_corrupt ADDQ SI, R9 // shortcut ends diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s new file mode 100644 index 000000000000..ec94b7b3c316 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_arm.s @@ -0,0 +1,201 @@ +// +build gc +// +build !noasm + +#include "textflag.h" + +// Register allocation. +#define dst R0 +#define dstorig R1 +#define src R2 +#define dstend R3 +#define srcend R4 +#define match R5 // Match address. +#define token R6 +#define len R7 // Literal and match lengths. +#define offset R6 // Match offset; overlaps with token. +#define tmp1 R8 +#define tmp2 R9 +#define tmp3 R12 + +#define minMatch $4 + +// func decodeBlock(dst, src []byte) int +TEXT ·decodeBlock(SB), NOFRAME|NOSPLIT, $-4-28 + MOVW dst_base +0(FP), dst + MOVW dst_len +4(FP), dstend + MOVW src_base+12(FP), src + MOVW src_len +16(FP), srcend + + CMP $0, srcend + BEQ shortSrc + + ADD dst, dstend + ADD src, srcend + + MOVW dst, dstorig + +loop: + // Read token. Extract literal length. + MOVBU.P 1(src), token + MOVW token >> 4, len + CMP $15, len + BNE readLitlenDone + +readLitlenLoop: + CMP src, srcend + BEQ shortSrc + MOVBU.P 1(src), tmp1 + ADD tmp1, len + CMP $255, tmp1 + BEQ readLitlenLoop + +readLitlenDone: + CMP $0, len + BEQ copyLiteralDone + + // Bounds check dst+len and src+len. + ADD dst, len, tmp1 + CMP dstend, tmp1 + //BHI shortDst // Uncomment for distinct error codes. + ADD src, len, tmp2 + CMP.LS srcend, tmp2 + BHI shortSrc + + // Copy literal. + CMP $4, len + BLO copyLiteralFinish + + // Copy 0-3 bytes until src is aligned. + TST $1, src + MOVBU.NE.P 1(src), tmp1 + MOVB.NE.P tmp1, 1(dst) + SUB.NE $1, len + + TST $2, src + MOVHU.NE.P 2(src), tmp2 + MOVB.NE.P tmp2, 1(dst) + MOVW.NE tmp2 >> 8, tmp1 + MOVB.NE.P tmp1, 1(dst) + SUB.NE $2, len + + B copyLiteralLoopCond + +copyLiteralLoop: + // Aligned load, unaligned write. + MOVW.P 4(src), tmp1 + MOVW tmp1 >> 8, tmp2 + MOVB tmp2, 1(dst) + MOVW tmp1 >> 16, tmp3 + MOVB tmp3, 2(dst) + MOVW tmp1 >> 24, tmp2 + MOVB tmp2, 3(dst) + MOVB.P tmp1, 4(dst) +copyLiteralLoopCond: + // Loop until len-4 < 0. + SUB.S $4, len + BPL copyLiteralLoop + + // Restore len, which is now negative. + ADD $4, len + +copyLiteralFinish: + // Copy remaining 0-3 bytes. + TST $2, len + MOVHU.NE.P 2(src), tmp2 + MOVB.NE.P tmp2, 1(dst) + MOVW.NE tmp2 >> 8, tmp1 + MOVB.NE.P tmp1, 1(dst) + TST $1, len + MOVBU.NE.P 1(src), tmp1 + MOVB.NE.P tmp1, 1(dst) + +copyLiteralDone: + CMP src, srcend + BEQ end + + // Initial part of match length. + // This frees up the token register for reuse as offset. + AND $15, token, len + + // Read offset. + ADD $2, src + CMP srcend, src + BHI shortSrc + MOVBU -2(src), offset + MOVBU -1(src), tmp1 + ORR tmp1 << 8, offset + CMP $0, offset + BEQ corrupt + + // Read rest of match length. + CMP $15, len + BNE readMatchlenDone + +readMatchlenLoop: + CMP src, srcend + BEQ shortSrc + MOVBU.P 1(src), tmp1 + ADD tmp1, len + CMP $255, tmp1 + BEQ readMatchlenLoop + +readMatchlenDone: + ADD minMatch, len + + // Bounds check dst+len and match = dst-offset. + ADD dst, len, tmp1 + CMP dstend, tmp1 + //BHI shortDst // Uncomment for distinct error codes. + SUB offset, dst, match + CMP.LS match, dstorig + BHI corrupt + + // If the offset is at least four (len is, because of minMatch), + // do a four-way unrolled byte copy loop. Using MOVD instead of four + // byte loads is much faster, but to remain portable we'd have to + // align match first, which in turn is too expensive. + CMP $4, offset + BLO copyMatch + + SUB $4, len +copyMatch4: + MOVBU.P 4(match), tmp1 + MOVB.P tmp1, 4(dst) + MOVBU -3(match), tmp2 + MOVB tmp2, -3(dst) + MOVBU -2(match), tmp3 + MOVB tmp3, -2(dst) + MOVBU -1(match), tmp1 + MOVB tmp1, -1(dst) + SUB.S $4, len + BPL copyMatch4 + + // Restore len, which is now negative. + ADD.S $4, len + BEQ copyMatchDone + +copyMatch: + // Simple byte-at-a-time copy. + SUB.S $1, len + MOVBU.P 1(match), tmp2 + MOVB.P tmp2, 1(dst) + BNE copyMatch + +copyMatchDone: + CMP src, srcend + BNE loop + +end: + SUB dstorig, dst, tmp1 + MOVW tmp1, ret+24(FP) + RET + + // The three error cases have distinct labels so we can put different + // return codes here when debugging, or if the error returns need to + // be changed. +shortDst: +shortSrc: +corrupt: + MOVW $-1, tmp1 + MOVW tmp1, ret+24(FP) + RET diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_asm.go similarity index 85% rename from vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.go rename to vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_asm.go index 816e8991bd92..e26f8cd613ed 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_amd64.go +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_asm.go @@ -1,3 +1,4 @@ +// +build amd64 arm // +build !appengine // +build gc // +build !noasm diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go index 60a2ed5ebd1e..9065653a9f28 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4block/decode_other.go @@ -1,4 +1,4 @@ -// +build !amd64 appengine !gc noasm +// +build !amd64,!arm appengine !gc noasm package lz4block @@ -10,16 +10,16 @@ func decodeBlock(dst, src []byte) (ret int) { } }() - var si, di int + var si, di uint for { // Literals and match lengths (token). - b := int(src[si]) + b := uint(src[si]) si++ // Literals. if lLen := b >> 4; lLen > 0 { switch { - case lLen < 0xF && si+16 < len(src): + case lLen < 0xF && si+16 < uint(len(src)): // Shortcut 1 // if we have enough room in src and dst, and the literals length // is small enough (0..14) then copy all 16 bytes, even if not all @@ -32,13 +32,13 @@ func decodeBlock(dst, src []byte) (ret int) { // if the match length (4..18) fits within the literals, then copy // all 18 bytes, even if not all are part of the literals. mLen += 4 - if offset := int(src[si]) | int(src[si+1])<<8; mLen <= offset { + if offset := uint(src[si]) | uint(src[si+1])<<8; mLen <= offset { i := di - offset end := i + 18 - if end > len(dst) { + if end > uint(len(dst)) { // The remaining buffer may not hold 18 bytes. // See https://github.com/pierrec/lz4/issues/51. - end = len(dst) + end = uint(len(dst)) } copy(dst[di:], dst[i:end]) si += 2 @@ -51,7 +51,7 @@ func decodeBlock(dst, src []byte) (ret int) { lLen += 0xFF si++ } - lLen += int(src[si]) + lLen += uint(src[si]) si++ fallthrough default: @@ -60,11 +60,13 @@ func decodeBlock(dst, src []byte) (ret int) { di += lLen } } - if si >= len(src) { - return di + if si == uint(len(src)) { + return int(di) + } else if si > uint(len(src)) { + return hasError } - offset := int(src[si]) | int(src[si+1])<<8 + offset := uint(src[si]) | uint(src[si+1])<<8 if offset == 0 { return hasError } @@ -77,7 +79,7 @@ func decodeBlock(dst, src []byte) (ret int) { mLen += 0xFF si++ } - mLen += int(src[si]) + mLen += uint(src[si]) si++ } mLen += minMatch @@ -93,6 +95,6 @@ func decodeBlock(dst, src []byte) (ret int) { di += bytesToCopy mLen -= bytesToCopy } - di += copy(dst[di:di+mLen], expanded[:mLen]) + di += uint(copy(dst[di:di+mLen], expanded[:mLen])) } } diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go new file mode 100644 index 000000000000..279a8cc49330 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/block.go @@ -0,0 +1,331 @@ +package lz4stream + +import ( + "encoding/binary" + "fmt" + "io" + "sync" + + "github.com/pierrec/lz4/v4/internal/lz4block" + "github.com/pierrec/lz4/v4/internal/lz4errors" + "github.com/pierrec/lz4/v4/internal/xxh32" +) + +type Blocks struct { + Block *FrameDataBlock + Blocks chan chan *FrameDataBlock + mu sync.Mutex + err error +} + +func (b *Blocks) initW(f *Frame, dst io.Writer, num int) { + if num == 1 { + b.Blocks = nil + b.Block = NewFrameDataBlock(f) + return + } + b.Block = nil + if cap(b.Blocks) != num { + b.Blocks = make(chan chan *FrameDataBlock, num) + } + // goroutine managing concurrent block compression goroutines. + go func() { + // Process next block compression item. + for c := range b.Blocks { + // Read the next compressed block result. + // Waiting here ensures that the blocks are output in the order they were sent. + // The incoming channel is always closed as it indicates to the caller that + // the block has been processed. + block := <-c + if block == nil { + // Notify the block compression routine that we are done with its result. + // This is used when a sentinel block is sent to terminate the compression. + close(c) + return + } + // Do not attempt to write the block upon any previous failure. + if b.err == nil { + // Write the block. + if err := block.Write(f, dst); err != nil { + // Keep the first error. + b.err = err + // All pending compression goroutines need to shut down, so we need to keep going. + } + } + close(c) + } + }() +} + +func (b *Blocks) close(f *Frame, num int) error { + if num == 1 { + if b.Block != nil { + b.Block.Close(f) + } + err := b.err + b.err = nil + return err + } + if b.Blocks == nil { + // Not initialized yet. + return nil + } + c := make(chan *FrameDataBlock) + b.Blocks <- c + c <- nil + <-c + err := b.err + b.err = nil + return err +} + +// ErrorR returns any error set while uncompressing a stream. +func (b *Blocks) ErrorR() error { + b.mu.Lock() + defer b.mu.Unlock() + return b.err +} + +// initR returns a channel that streams the uncompressed blocks if in concurrent +// mode and no error. When the channel is closed, check for any error with b.ErrorR. +// +// If not in concurrent mode, the uncompressed block is b.Block and the returned error +// needs to be checked. +func (b *Blocks) initR(f *Frame, num int, src io.Reader) (chan []byte, error) { + size := f.Descriptor.Flags.BlockSizeIndex() + if num == 1 { + b.Blocks = nil + b.Block = NewFrameDataBlock(f) + return nil, nil + } + b.Block = nil + blocks := make(chan chan []byte, num) + // data receives the uncompressed blocks. + data := make(chan []byte) + // Read blocks from the source sequentially + // and uncompress them concurrently. + + // In legacy mode, accrue the uncompress sizes in cum. + var cum uint32 + go func() { + var cumx uint32 + var err error + for b.ErrorR() == nil { + block := NewFrameDataBlock(f) + cumx, err = block.Read(f, src, 0) + if err != nil { + break + } + // Recheck for an error as reading may be slow and uncompressing is expensive. + if b.ErrorR() != nil { + break + } + c := make(chan []byte) + blocks <- c + go func() { + data, err := block.Uncompress(f, size.Get(), false) + if err != nil { + b.closeR(err) + } else { + c <- data + } + }() + } + // End the collection loop and the data channel. + c := make(chan []byte) + blocks <- c + c <- nil // signal the collection loop that we are done + <-c // wait for the collect loop to complete + if f.isLegacy() && cum == cumx { + err = io.EOF + } + b.closeR(err) + close(data) + }() + // Collect the uncompressed blocks and make them available + // on the returned channel. + go func(leg bool) { + defer close(blocks) + for c := range blocks { + buf := <-c + if buf == nil { + // Signal to end the loop. + close(c) + return + } + // Perform checksum now as the blocks are received in order. + if f.Descriptor.Flags.ContentChecksum() { + _, _ = f.checksum.Write(buf) + } + if leg { + cum += uint32(len(buf)) + } + data <- buf + close(c) + } + }(f.isLegacy()) + return data, nil +} + +// closeR safely sets the error on b if not already set. +func (b *Blocks) closeR(err error) { + b.mu.Lock() + if b.err == nil { + b.err = err + } + b.mu.Unlock() +} + +func NewFrameDataBlock(f *Frame) *FrameDataBlock { + buf := f.Descriptor.Flags.BlockSizeIndex().Get() + return &FrameDataBlock{Data: buf, data: buf} +} + +type FrameDataBlock struct { + Size DataBlockSize + Data []byte // compressed or uncompressed data (.data or .src) + Checksum uint32 + data []byte // buffer for compressed data + src []byte // uncompressed data + err error // used in concurrent mode +} + +func (b *FrameDataBlock) Close(f *Frame) { + b.Size = 0 + b.Checksum = 0 + b.err = nil + if b.data != nil { + // Block was not already closed. + lz4block.Put(b.data) + b.Data = nil + b.data = nil + b.src = nil + } +} + +// Block compression errors are ignored since the buffer is sized appropriately. +func (b *FrameDataBlock) Compress(f *Frame, src []byte, level lz4block.CompressionLevel) *FrameDataBlock { + data := b.data + if f.isLegacy() { + data = data[:cap(data)] + } else { + data = data[:len(src)] // trigger the incompressible flag in CompressBlock + } + var n int + switch level { + case lz4block.Fast: + n, _ = lz4block.CompressBlock(src, data) + default: + n, _ = lz4block.CompressBlockHC(src, data, level) + } + if n == 0 { + b.Size.UncompressedSet(true) + b.Data = src + } else { + b.Size.UncompressedSet(false) + b.Data = data[:n] + } + b.Size.sizeSet(len(b.Data)) + b.src = src // keep track of the source for content checksum + + if f.Descriptor.Flags.BlockChecksum() { + b.Checksum = xxh32.ChecksumZero(src) + } + return b +} + +func (b *FrameDataBlock) Write(f *Frame, dst io.Writer) error { + // Write is called in the same order as blocks are compressed, + // so content checksum must be done here. + if f.Descriptor.Flags.ContentChecksum() { + _, _ = f.checksum.Write(b.src) + } + buf := f.buf[:] + binary.LittleEndian.PutUint32(buf, uint32(b.Size)) + if _, err := dst.Write(buf[:4]); err != nil { + return err + } + + if _, err := dst.Write(b.Data); err != nil { + return err + } + + if b.Checksum == 0 { + return nil + } + binary.LittleEndian.PutUint32(buf, b.Checksum) + _, err := dst.Write(buf[:4]) + return err +} + +// Read updates b with the next block data, size and checksum if available. +func (b *FrameDataBlock) Read(f *Frame, src io.Reader, cum uint32) (uint32, error) { + x, err := f.readUint32(src) + if err != nil { + return 0, err + } + if f.isLegacy() { + switch x { + case frameMagicLegacy: + // Concatenated legacy frame. + return b.Read(f, src, cum) + case cum: + // Only works in non concurrent mode, for concurrent mode + // it is handled separately. + // Linux kernel format appends the total uncompressed size at the end. + return 0, io.EOF + } + } else if x == 0 { + // Marker for end of stream. + return 0, io.EOF + } + b.Size = DataBlockSize(x) + + size := b.Size.size() + if size > cap(b.data) { + return x, lz4errors.ErrOptionInvalidBlockSize + } + b.data = b.data[:size] + if _, err := io.ReadFull(src, b.data); err != nil { + return x, err + } + if f.Descriptor.Flags.BlockChecksum() { + sum, err := f.readUint32(src) + if err != nil { + return 0, err + } + b.Checksum = sum + } + return x, nil +} + +func (b *FrameDataBlock) Uncompress(f *Frame, dst []byte, sum bool) ([]byte, error) { + if b.Size.Uncompressed() { + n := copy(dst, b.data) + dst = dst[:n] + } else { + n, err := lz4block.UncompressBlock(b.data, dst) + if err != nil { + return nil, err + } + dst = dst[:n] + } + if f.Descriptor.Flags.BlockChecksum() { + if c := xxh32.ChecksumZero(dst); c != b.Checksum { + err := fmt.Errorf("%w: got %x; expected %x", lz4errors.ErrInvalidBlockChecksum, c, b.Checksum) + return nil, err + } + } + if sum && f.Descriptor.Flags.ContentChecksum() { + _, _ = f.checksum.Write(dst) + } + return dst, nil +} + +func (f *Frame) readUint32(r io.Reader) (x uint32, err error) { + if _, err = io.ReadFull(r, f.buf[:4]); err != nil { + return + } + x = binary.LittleEndian.Uint32(f.buf[:4]) + return +} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go index a58d896c2005..cfbd5674d9d1 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go +++ b/vendor/github.com/pierrec/lz4/v4/internal/lz4stream/frame.go @@ -15,8 +15,9 @@ import ( //go:generate go run gen.go const ( - frameMagic uint32 = 0x184D2204 - frameSkipMagic uint32 = 0x184D2A50 + frameMagic uint32 = 0x184D2204 + frameSkipMagic uint32 = 0x184D2A50 + frameMagicLegacy uint32 = 0x184C2102 ) func NewFrame() *Frame { @@ -38,21 +39,30 @@ func (f *Frame) Reset(num int) { f.Magic = 0 f.Descriptor.Checksum = 0 f.Descriptor.ContentSize = 0 - _ = f.Blocks.closeW(f, num) + _ = f.Blocks.close(f, num) f.Checksum = 0 } -func (f *Frame) InitW(dst io.Writer, num int) { - f.Magic = frameMagic - f.Descriptor.initW() +func (f *Frame) InitW(dst io.Writer, num int, legacy bool) { + if legacy { + f.Magic = frameMagicLegacy + idx := lz4block.Index(lz4block.Block8Mb) + f.Descriptor.Flags.BlockSizeIndexSet(idx) + } else { + f.Magic = frameMagic + f.Descriptor.initW() + } f.Blocks.initW(f, dst, num) f.checksum.Reset() } func (f *Frame) CloseW(dst io.Writer, num int) error { - if err := f.Blocks.closeW(f, num); err != nil { + if err := f.Blocks.close(f, num); err != nil { return err } + if f.isLegacy() { + return nil + } buf := f.buf[:0] // End mark (data block size of uint32(0)). buf = append(buf, 0, 0, 0, 0) @@ -63,41 +73,47 @@ func (f *Frame) CloseW(dst io.Writer, num int) error { return err } -func (f *Frame) InitR(src io.Reader) error { +func (f *Frame) isLegacy() bool { + return f.Magic == frameMagicLegacy +} + +func (f *Frame) InitR(src io.Reader, num int) (chan []byte, error) { if f.Magic > 0 { // Header already read. - return nil + return nil, nil } newFrame: var err error if f.Magic, err = f.readUint32(src); err != nil { - return err + return nil, err } switch m := f.Magic; { - case m == frameMagic: + case m == frameMagic || m == frameMagicLegacy: // All 16 values of frameSkipMagic are valid. case m>>8 == frameSkipMagic>>8: - var skip uint32 - if err := binary.Read(src, binary.LittleEndian, &skip); err != nil { - return err + skip, err := f.readUint32(src) + if err != nil { + return nil, err } if _, err := io.CopyN(ioutil.Discard, src, int64(skip)); err != nil { - return err + return nil, err } goto newFrame default: - return lz4errors.ErrInvalidFrame + return nil, lz4errors.ErrInvalidFrame } if err := f.Descriptor.initR(f, src); err != nil { - return err + return nil, err } - f.Blocks.initR(f) f.checksum.Reset() - return nil + return f.Blocks.initR(f, num, src) } func (f *Frame) CloseR(src io.Reader) (err error) { + if f.isLegacy() { + return nil + } if !f.Descriptor.Flags.ContentChecksum() { return nil } @@ -127,23 +143,31 @@ func (fd *FrameDescriptor) Write(f *Frame, dst io.Writer) error { return nil } - buf := f.buf[:4+2] + buf := f.buf[:4] // Write the magic number here even though it belongs to the Frame. binary.LittleEndian.PutUint32(buf, f.Magic) - binary.LittleEndian.PutUint16(buf[4:], uint16(fd.Flags)) + if !f.isLegacy() { + buf = buf[:4+2] + binary.LittleEndian.PutUint16(buf[4:], uint16(fd.Flags)) - if fd.Flags.Size() { - buf = buf[:4+2+8] - binary.LittleEndian.PutUint64(buf[4+2:], fd.ContentSize) + if fd.Flags.Size() { + buf = buf[:4+2+8] + binary.LittleEndian.PutUint64(buf[4+2:], fd.ContentSize) + } + fd.Checksum = descriptorChecksum(buf[4:]) + buf = append(buf, fd.Checksum) } - fd.Checksum = descriptorChecksum(buf[4:]) - buf = append(buf, fd.Checksum) _, err := dst.Write(buf) return err } func (fd *FrameDescriptor) initR(f *Frame, src io.Reader) error { + if f.isLegacy() { + idx := lz4block.Index(lz4block.Block8Mb) + f.Descriptor.Flags.BlockSizeIndexSet(idx) + return nil + } // Read the flags and the checksum, hoping that there is not content size. buf := f.buf[:3] if _, err := io.ReadFull(src, buf); err != nil { @@ -174,204 +198,3 @@ func (fd *FrameDescriptor) initR(f *Frame, src io.Reader) error { func descriptorChecksum(buf []byte) byte { return byte(xxh32.ChecksumZero(buf) >> 8) } - -type Blocks struct { - Block *FrameDataBlock - Blocks chan chan *FrameDataBlock - err error -} - -func (b *Blocks) initW(f *Frame, dst io.Writer, num int) { - size := f.Descriptor.Flags.BlockSizeIndex() - if num == 1 { - b.Blocks = nil - b.Block = NewFrameDataBlock(size) - return - } - b.Block = nil - if cap(b.Blocks) != num { - b.Blocks = make(chan chan *FrameDataBlock, num) - } - // goroutine managing concurrent block compression goroutines. - go func() { - // Process next block compression item. - for c := range b.Blocks { - // Read the next compressed block result. - // Waiting here ensures that the blocks are output in the order they were sent. - // The incoming channel is always closed as it indicates to the caller that - // the block has been processed. - block := <-c - if block == nil { - // Notify the block compression routine that we are done with its result. - // This is used when a sentinel block is sent to terminate the compression. - close(c) - return - } - // Do not attempt to write the block upon any previous failure. - if b.err == nil { - // Write the block. - if err := block.Write(f, dst); err != nil && b.err == nil { - // Keep the first error. - b.err = err - // All pending compression goroutines need to shut down, so we need to keep going. - } - } - close(c) - } - }() -} - -func (b *Blocks) closeW(f *Frame, num int) error { - if num == 1 { - if b.Block == nil { - // Not initialized yet. - return nil - } - b.Block.CloseW(f) - return nil - } - if b.Blocks == nil { - // Not initialized yet. - return nil - } - c := make(chan *FrameDataBlock) - b.Blocks <- c - c <- nil - <-c - err := b.err - b.err = nil - return err -} - -func (b *Blocks) initR(f *Frame) { - size := f.Descriptor.Flags.BlockSizeIndex() - b.Block = NewFrameDataBlock(size) -} - -func NewFrameDataBlock(size lz4block.BlockSizeIndex) *FrameDataBlock { - buf := size.Get() - return &FrameDataBlock{Data: buf, data: buf} -} - -type FrameDataBlock struct { - Size DataBlockSize - Data []byte // compressed or uncompressed data (.data or .src) - Checksum uint32 - data []byte // buffer for compressed data - src []byte // uncompressed data -} - -func (b *FrameDataBlock) CloseW(f *Frame) { - if b.data != nil { - // Block was not already closed. - size := f.Descriptor.Flags.BlockSizeIndex() - size.Put(b.data) - b.Data = nil - b.data = nil - b.src = nil - } -} - -// Block compression errors are ignored since the buffer is sized appropriately. -func (b *FrameDataBlock) Compress(f *Frame, src []byte, level lz4block.CompressionLevel) *FrameDataBlock { - data := b.data[:len(src)] // trigger the incompressible flag in CompressBlock - var n int - switch level { - case lz4block.Fast: - n, _ = lz4block.CompressBlock(src, data, nil) - default: - n, _ = lz4block.CompressBlockHC(src, data, level, nil, nil) - } - if n == 0 { - b.Size.UncompressedSet(true) - b.Data = src - } else { - b.Size.UncompressedSet(false) - b.Data = data[:n] - } - b.Size.sizeSet(len(b.Data)) - b.src = src // keep track of the source for content checksum - - if f.Descriptor.Flags.BlockChecksum() { - b.Checksum = xxh32.ChecksumZero(src) - } - return b -} - -func (b *FrameDataBlock) Write(f *Frame, dst io.Writer) error { - if f.Descriptor.Flags.ContentChecksum() { - _, _ = f.checksum.Write(b.src) - } - buf := f.buf[:] - binary.LittleEndian.PutUint32(buf, uint32(b.Size)) - if _, err := dst.Write(buf[:4]); err != nil { - return err - } - - if _, err := dst.Write(b.Data); err != nil { - return err - } - - if b.Checksum == 0 { - return nil - } - binary.LittleEndian.PutUint32(buf, b.Checksum) - _, err := dst.Write(buf[:4]) - return err -} - -func (b *FrameDataBlock) Uncompress(f *Frame, src io.Reader, dst []byte) (int, error) { - x, err := f.readUint32(src) - if err != nil { - return 0, err - } - b.Size = DataBlockSize(x) - if b.Size == 0 { - // End of frame reached. - return 0, io.EOF - } - - isCompressed := !b.Size.Uncompressed() - size := b.Size.size() - var data []byte - if isCompressed { - // Data is first copied into b.Data and then it will get uncompressed into dst. - data = b.Data - } else { - // Data is directly copied into dst as it is not compressed. - data = dst - } - data = data[:size] - if _, err := io.ReadFull(src, data); err != nil { - return 0, err - } - if isCompressed { - n, err := lz4block.UncompressBlock(data, dst) - if err != nil { - return 0, err - } - data = dst[:n] - } - - if f.Descriptor.Flags.BlockChecksum() { - var err error - if b.Checksum, err = f.readUint32(src); err != nil { - return 0, err - } - if c := xxh32.ChecksumZero(data); c != b.Checksum { - return 0, fmt.Errorf("%w: got %x; expected %x", lz4errors.ErrInvalidBlockChecksum, c, b.Checksum) - } - } - if f.Descriptor.Flags.ContentChecksum() { - _, _ = f.checksum.Write(data) - } - return len(data), nil -} - -func (f *Frame) readUint32(r io.Reader) (x uint32, err error) { - if _, err = io.ReadFull(r, f.buf[:4]); err != nil { - return - } - x = binary.LittleEndian.Uint32(f.buf[:4]) - return -} diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go index b33b44bf0022..8d3206a87c55 100644 --- a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero.go @@ -20,10 +20,7 @@ const ( // XXHZero represents an xxhash32 object with seed 0. type XXHZero struct { - v1 uint32 - v2 uint32 - v3 uint32 - v4 uint32 + v [4]uint32 totalLen uint64 buf [16]byte bufused int @@ -38,10 +35,10 @@ func (xxh XXHZero) Sum(b []byte) []byte { // Reset resets the Hash to its initial state. func (xxh *XXHZero) Reset() { - xxh.v1 = prime1plus2 - xxh.v2 = prime2 - xxh.v3 = 0 - xxh.v4 = prime1minus + xxh.v[0] = prime1plus2 + xxh.v[1] = prime2 + xxh.v[2] = 0 + xxh.v[3] = prime1minus xxh.totalLen = 0 xxh.bufused = 0 } @@ -74,42 +71,48 @@ func (xxh *XXHZero) Write(input []byte) (int, error) { return n, nil } - p := 0 - // Causes compiler to work directly from registers instead of stack: - v1, v2, v3, v4 := xxh.v1, xxh.v2, xxh.v3, xxh.v4 - if m > 0 { + var buf *[16]byte + if m != 0 { // some data left from previous update - copy(xxh.buf[m:], input) + buf = &xxh.buf + c := copy(buf[m:], input) + n -= c + input = input[c:] + } + update(&xxh.v, buf, input) + xxh.bufused = copy(xxh.buf[:], input[n-n%16:]) + + return n, nil +} + +// Portable version of update. This updates v by processing all of buf +// (if not nil) and all full 16-byte blocks of input. +func updateGo(v *[4]uint32, buf *[16]byte, input []byte) { + // Causes compiler to work directly from registers instead of stack: + v1, v2, v3, v4 := v[0], v[1], v[2], v[3] - // fast rotl(13) - buf := xxh.buf[:16] // BCE hint. + if buf != nil { v1 = rol13(v1+binary.LittleEndian.Uint32(buf[:])*prime2) * prime1 v2 = rol13(v2+binary.LittleEndian.Uint32(buf[4:])*prime2) * prime1 v3 = rol13(v3+binary.LittleEndian.Uint32(buf[8:])*prime2) * prime1 v4 = rol13(v4+binary.LittleEndian.Uint32(buf[12:])*prime2) * prime1 - p = r } - for n := n - 16; p <= n; p += 16 { - sub := input[p:][:16] //BCE hint for compiler + for ; len(input) >= 16; input = input[16:] { + sub := input[:16] //BCE hint for compiler v1 = rol13(v1+binary.LittleEndian.Uint32(sub[:])*prime2) * prime1 v2 = rol13(v2+binary.LittleEndian.Uint32(sub[4:])*prime2) * prime1 v3 = rol13(v3+binary.LittleEndian.Uint32(sub[8:])*prime2) * prime1 v4 = rol13(v4+binary.LittleEndian.Uint32(sub[12:])*prime2) * prime1 } - xxh.v1, xxh.v2, xxh.v3, xxh.v4 = v1, v2, v3, v4 - - copy(xxh.buf[:], input[p:]) - xxh.bufused = len(input) - p - - return n, nil + v[0], v[1], v[2], v[3] = v1, v2, v3, v4 } // Sum32 returns the 32 bits Hash value. func (xxh *XXHZero) Sum32() uint32 { h32 := uint32(xxh.totalLen) if h32 >= 16 { - h32 += rol1(xxh.v1) + rol7(xxh.v2) + rol12(xxh.v3) + rol18(xxh.v4) + h32 += rol1(xxh.v[0]) + rol7(xxh.v[1]) + rol12(xxh.v[2]) + rol18(xxh.v[3]) } else { h32 += prime5 } @@ -135,8 +138,8 @@ func (xxh *XXHZero) Sum32() uint32 { return h32 } -// ChecksumZero returns the 32bits Hash value. -func ChecksumZero(input []byte) uint32 { +// Portable version of ChecksumZero. +func checksumZeroGo(input []byte) uint32 { n := len(input) h32 := uint32(n) diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go new file mode 100644 index 000000000000..0978b2665bd0 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.go @@ -0,0 +1,11 @@ +// +build !noasm + +package xxh32 + +// ChecksumZero returns the 32-bit hash of input. +// +//go:noescape +func ChecksumZero(input []byte) uint32 + +//go:noescape +func update(v *[4]uint32, buf *[16]byte, input []byte) diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s new file mode 100644 index 000000000000..0e9f146a36aa --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_arm.s @@ -0,0 +1,259 @@ +// +build !noasm + +#include "textflag.h" + +#define prime1 $2654435761 +#define prime2 $2246822519 +#define prime3 $3266489917 +#define prime4 $668265263 +#define prime5 $374761393 + +#define prime1plus2 $606290984 +#define prime1minus $1640531535 + +// Register allocation. +#define p R0 +#define n R1 +#define h R2 +#define v1 R2 // Alias for h. +#define v2 R3 +#define v3 R4 +#define v4 R5 +#define x1 R6 +#define x2 R7 +#define x3 R8 +#define x4 R9 + +// We need the primes in registers. The 16-byte loop only uses prime{1,2}. +#define prime1r R11 +#define prime2r R12 +#define prime3r R3 // The rest can alias v{2-4}. +#define prime4r R4 +#define prime5r R5 + +// Update round macros. These read from and increment p. + +#define round16aligned \ + MOVM.IA.W (p), [x1, x2, x3, x4] \ + \ + MULA x1, prime2r, v1, v1 \ + MULA x2, prime2r, v2, v2 \ + MULA x3, prime2r, v3, v3 \ + MULA x4, prime2r, v4, v4 \ + \ + MOVW v1 @> 19, v1 \ + MOVW v2 @> 19, v2 \ + MOVW v3 @> 19, v3 \ + MOVW v4 @> 19, v4 \ + \ + MUL prime1r, v1 \ + MUL prime1r, v2 \ + MUL prime1r, v3 \ + MUL prime1r, v4 \ + +#define round16unaligned \ + MOVBU.P 16(p), x1 \ + MOVBU -15(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -14(p), x3 \ + MOVBU -13(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v1, v1 \ + MOVW v1 @> 19, v1 \ + MUL prime1r, v1 \ + \ + MOVBU -12(p), x1 \ + MOVBU -11(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -10(p), x3 \ + MOVBU -9(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v2, v2 \ + MOVW v2 @> 19, v2 \ + MUL prime1r, v2 \ + \ + MOVBU -8(p), x1 \ + MOVBU -7(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -6(p), x3 \ + MOVBU -5(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v3, v3 \ + MOVW v3 @> 19, v3 \ + MUL prime1r, v3 \ + \ + MOVBU -4(p), x1 \ + MOVBU -3(p), x2 \ + ORR x2 << 8, x1 \ + MOVBU -2(p), x3 \ + MOVBU -1(p), x4 \ + ORR x4 << 8, x3 \ + ORR x3 << 16, x1 \ + \ + MULA x1, prime2r, v4, v4 \ + MOVW v4 @> 19, v4 \ + MUL prime1r, v4 \ + + +// func ChecksumZero([]byte) uint32 +TEXT ·ChecksumZero(SB), NOFRAME|NOSPLIT, $-4-16 + MOVW input_base+0(FP), p + MOVW input_len+4(FP), n + + MOVW prime1, prime1r + MOVW prime2, prime2r + + // Set up h for n < 16. It's tempting to say {ADD prime5, n, h} + // here, but that's a pseudo-op that generates a load through R11. + MOVW prime5, prime5r + ADD prime5r, n, h + CMP $0, n + BEQ end + + // We let n go negative so we can do comparisons with SUB.S + // instead of separate CMP. + SUB.S $16, n + BMI loop16done + + MOVW prime1plus2, v1 + MOVW prime2, v2 + MOVW $0, v3 + MOVW prime1minus, v4 + + TST $3, p + BNE loop16unaligned + +loop16aligned: + SUB.S $16, n + round16aligned + BPL loop16aligned + B loop16finish + +loop16unaligned: + SUB.S $16, n + round16unaligned + BPL loop16unaligned + +loop16finish: + MOVW v1 @> 31, h + ADD v2 @> 25, h + ADD v3 @> 20, h + ADD v4 @> 14, h + + // h += len(input) with v2 as temporary. + MOVW input_len+4(FP), v2 + ADD v2, h + +loop16done: + ADD $16, n // Restore number of bytes left. + + SUB.S $4, n + MOVW prime3, prime3r + BMI loop4done + MOVW prime4, prime4r + + TST $3, p + BNE loop4unaligned + +loop4aligned: + SUB.S $4, n + + MOVW.P 4(p), x1 + MULA prime3r, x1, h, h + MOVW h @> 15, h + MUL prime4r, h + + BPL loop4aligned + B loop4done + +loop4unaligned: + SUB.S $4, n + + MOVBU.P 4(p), x1 + MOVBU -3(p), x2 + ORR x2 << 8, x1 + MOVBU -2(p), x3 + ORR x3 << 16, x1 + MOVBU -1(p), x4 + ORR x4 << 24, x1 + + MULA prime3r, x1, h, h + MOVW h @> 15, h + MUL prime4r, h + + BPL loop4unaligned + +loop4done: + ADD.S $4, n // Restore number of bytes left. + BEQ end + + MOVW prime5, prime5r + +loop1: + SUB.S $1, n + + MOVBU.P 1(p), x1 + MULA prime5r, x1, h, h + MOVW h @> 21, h + MUL prime1r, h + + BNE loop1 + +end: + MOVW prime3, prime3r + EOR h >> 15, h + MUL prime2r, h + EOR h >> 13, h + MUL prime3r, h + EOR h >> 16, h + + MOVW h, ret+12(FP) + RET + + +// func update(v *[4]uint64, buf *[16]byte, p []byte) +TEXT ·update(SB), NOFRAME|NOSPLIT, $-4-20 + MOVW v+0(FP), p + MOVM.IA (p), [v1, v2, v3, v4] + + MOVW prime1, prime1r + MOVW prime2, prime2r + + // Process buf, if not nil. + MOVW buf+4(FP), p + CMP $0, p + BEQ noBuffered + + round16aligned + +noBuffered: + MOVW input_base +8(FP), p + MOVW input_len +12(FP), n + + SUB.S $16, n + BMI end + + TST $3, p + BNE loop16unaligned + +loop16aligned: + SUB.S $16, n + round16aligned + BPL loop16aligned + B end + +loop16unaligned: + SUB.S $16, n + round16unaligned + BPL loop16unaligned + +end: + MOVW v+0(FP), p + MOVM.IA [v1, v2, v3, v4], (p) + RET diff --git a/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go new file mode 100644 index 000000000000..c96b59b8c3f2 --- /dev/null +++ b/vendor/github.com/pierrec/lz4/v4/internal/xxh32/xxh32zero_other.go @@ -0,0 +1,10 @@ +// +build !arm noasm + +package xxh32 + +// ChecksumZero returns the 32-bit hash of input. +func ChecksumZero(input []byte) uint32 { return checksumZeroGo(input) } + +func update(v *[4]uint32, buf *[16]byte, input []byte) { + updateGo(v, buf, input) +} diff --git a/vendor/github.com/pierrec/lz4/v4/lz4.go b/vendor/github.com/pierrec/lz4/v4/lz4.go index 2ba28eb3d4a1..c585d4064f5c 100644 --- a/vendor/github.com/pierrec/lz4/v4/lz4.go +++ b/vendor/github.com/pierrec/lz4/v4/lz4.go @@ -1,13 +1,11 @@ -// Package lz4 implements reading and writing lz4 compressed data (a frame), -// as specified in http://fastcompression.blogspot.fr/2013/04/lz4-streaming-format-final.html. +// Package lz4 implements reading and writing lz4 compressed data. // -// Although the block level compression and decompression functions are exposed and are fully compatible -// with the lz4 block format definition, they are low level and should not be used directly. -// For a complete description of an lz4 compressed block, see: -// http://fastcompression.blogspot.fr/2011/05/lz4-explained.html +// The package supports both the LZ4 stream format, +// as specified in http://fastcompression.blogspot.fr/2013/04/lz4-streaming-format-final.html, +// and the LZ4 block format, defined at +// http://fastcompression.blogspot.fr/2011/05/lz4-explained.html. // // See https://github.com/lz4/lz4 for the reference C implementation. -// package lz4 import ( @@ -40,6 +38,29 @@ func UncompressBlock(src, dst []byte) (int, error) { return lz4block.UncompressBlock(src, dst) } +// A Compressor compresses data into the LZ4 block format. +// It uses a fast compression algorithm. +// +// A Compressor is not safe for concurrent use by multiple goroutines. +// +// Use a Writer to compress into the LZ4 stream format. +type Compressor struct{ c lz4block.Compressor } + +// CompressBlock compresses the source buffer src into the destination dst. +// +// If compression is successful, the first return value is the size of the +// compressed data, which is always >0. +// +// If dst has length at least CompressBlockBound(len(src)), compression always +// succeeds. Otherwise, the first return value is zero. The error return is +// non-nil if the compressed data does not fit in dst, but it might fit in a +// larger buffer that is still smaller than CompressBlockBound(len(src)). The +// return value (0, nil) means the data is likely incompressible and a buffer +// of length CompressBlockBound(len(src)) should be passed in. +func (c *Compressor) CompressBlock(src, dst []byte) (int, error) { + return c.c.CompressBlock(src, dst) +} + // CompressBlock compresses the source buffer into the destination one. // This is the fast version of LZ4 compression and also the default one. // @@ -53,23 +74,50 @@ func UncompressBlock(src, dst []byte) (int, error) { // the compressed size is 0 and no error, then the data is incompressible. // // An error is returned if the destination buffer is too small. -func CompressBlock(src, dst []byte, hashTable []int) (int, error) { - return lz4block.CompressBlock(src, dst, hashTable) + +// CompressBlock is equivalent to Compressor.CompressBlock. +// The final argument is ignored and should be set to nil. +// +// This function is deprecated. Use a Compressor instead. +func CompressBlock(src, dst []byte, _ []int) (int, error) { + return lz4block.CompressBlock(src, dst) } -// CompressBlockHC compresses the source buffer src into the destination dst -// with max search depth (use 0 or negative value for no max). +// A CompressorHC compresses data into the LZ4 block format. +// Its compression ratio is potentially better than that of a Compressor, +// but it is also slower and requires more memory. // -// CompressBlockHC compression ratio is better than CompressBlock but it is also slower. +// A Compressor is not safe for concurrent use by multiple goroutines. // -// The size of the compressed data is returned. +// Use a Writer to compress into the LZ4 stream format. +type CompressorHC struct { + // Level is the maximum search depth for compression. + // Values <= 0 mean no maximum. + Level CompressionLevel + c lz4block.CompressorHC +} + +// CompressBlock compresses the source buffer src into the destination dst. // -// If the destination buffer size is lower than CompressBlockBound and -// the compressed size is 0 and no error, then the data is incompressible. +// If compression is successful, the first return value is the size of the +// compressed data, which is always >0. // -// An error is returned if the destination buffer is too small. -func CompressBlockHC(src, dst []byte, depth CompressionLevel, hashTable, chainTable []int) (int, error) { - return lz4block.CompressBlockHC(src, dst, lz4block.CompressionLevel(depth), hashTable, chainTable) +// If dst has length at least CompressBlockBound(len(src)), compression always +// succeeds. Otherwise, the first return value is zero. The error return is +// non-nil if the compressed data does not fit in dst, but it might fit in a +// larger buffer that is still smaller than CompressBlockBound(len(src)). The +// return value (0, nil) means the data is likely incompressible and a buffer +// of length CompressBlockBound(len(src)) should be passed in. +func (c *CompressorHC) CompressBlock(src, dst []byte) (int, error) { + return c.c.CompressBlock(src, dst, lz4block.CompressionLevel(c.Level)) +} + +// CompressBlockHC is equivalent to CompressorHC.CompressBlock. +// The final two arguments are ignored and should be set to nil. +// +// This function is deprecated. Use a CompressorHC instead. +func CompressBlockHC(src, dst []byte, depth CompressionLevel, _, _ []int) (int, error) { + return lz4block.CompressBlockHC(src, dst, lz4block.CompressionLevel(depth)) } const ( diff --git a/vendor/github.com/pierrec/lz4/v4/options.go b/vendor/github.com/pierrec/lz4/v4/options.go index 813e75cff115..4e1b6703b57d 100644 --- a/vendor/github.com/pierrec/lz4/v4/options.go +++ b/vendor/github.com/pierrec/lz4/v4/options.go @@ -2,11 +2,10 @@ package lz4 import ( "fmt" - "reflect" - "runtime" - "github.com/pierrec/lz4/v4/internal/lz4block" "github.com/pierrec/lz4/v4/internal/lz4errors" + "reflect" + "runtime" ) //go:generate go run golang.org/x/tools/cmd/stringer -type=BlockSize,CompressionLevel -output options_gen.go @@ -82,7 +81,7 @@ func ChecksumOption(flag bool) Option { return func(a applier) error { switch w := a.(type) { case nil: - s := fmt.Sprintf("BlockChecksumOption(%v)", flag) + s := fmt.Sprintf("ChecksumOption(%v)", flag) return lz4errors.Error(s) case *Writer: w.frame.Descriptor.Flags.ContentChecksumSet(flag) @@ -110,22 +109,21 @@ func SizeOption(size uint64) Option { } // ConcurrencyOption sets the number of go routines used for compression. -// If n<0, then the output of runtime.GOMAXPROCS(0) is used. +// If n <= 0, then the output of runtime.GOMAXPROCS(0) is used. func ConcurrencyOption(n int) Option { + if n <= 0 { + n = runtime.GOMAXPROCS(0) + } return func(a applier) error { - switch w := a.(type) { + switch rw := a.(type) { case nil: s := fmt.Sprintf("ConcurrencyOption(%d)", n) return lz4errors.Error(s) case *Writer: - switch n { - case 0, 1: - default: - if n < 0 { - n = runtime.GOMAXPROCS(0) - } - } - w.num = n + rw.num = n + return nil + case *Reader: + rw.num = n return nil } return lz4errors.ErrOptionNotApplicable @@ -183,9 +181,33 @@ func OnBlockDoneOption(handler func(size int)) Option { return lz4errors.Error(s) case *Writer: rw.handler = handler + return nil case *Reader: rw.handler = handler + return nil } - return nil + return lz4errors.ErrOptionNotApplicable + } +} + +// LegacyOption provides support for writing LZ4 frames in the legacy format. +// +// See https://github.com/lz4/lz4/blob/dev/doc/lz4_Frame_format.md#legacy-frame. +// +// NB. compressed Linux kernel images use a tweaked LZ4 legacy format where +// the compressed stream is followed by the original (uncompressed) size of +// the kernel (https://events.static.linuxfound.org/sites/events/files/lcjpcojp13_klee.pdf). +// This is also supported as a special case. +func LegacyOption(legacy bool) Option { + return func(a applier) error { + switch rw := a.(type) { + case nil: + s := fmt.Sprintf("LegacyOption(%v)", legacy) + return lz4errors.Error(s) + case *Writer: + rw.legacy = legacy + return nil + } + return lz4errors.ErrOptionNotApplicable } } diff --git a/vendor/github.com/pierrec/lz4/v4/reader.go b/vendor/github.com/pierrec/lz4/v4/reader.go index 214da2f29033..403aaf697a3b 100644 --- a/vendor/github.com/pierrec/lz4/v4/reader.go +++ b/vendor/github.com/pierrec/lz4/v4/reader.go @@ -3,6 +3,7 @@ package lz4 import ( "io" + "github.com/pierrec/lz4/v4/internal/lz4block" "github.com/pierrec/lz4/v4/internal/lz4errors" "github.com/pierrec/lz4/v4/internal/lz4stream" ) @@ -17,9 +18,13 @@ var readerStates = []aState{ // NewReader returns a new LZ4 frame decoder. func NewReader(r io.Reader) *Reader { + return newReader(r, false) +} + +func newReader(r io.Reader, legacy bool) *Reader { zr := &Reader{frame: lz4stream.NewFrame()} zr.state.init(readerStates) - _ = zr.Apply(defaultOnBlockDone) + _ = zr.Apply(DefaultConcurrency, defaultOnBlockDone) zr.Reset(r) return zr } @@ -28,10 +33,13 @@ func NewReader(r io.Reader) *Reader { type Reader struct { state _State src io.Reader // source reader + num int // concurrency level frame *lz4stream.Frame // frame being read - data []byte // pending data + data []byte // block buffer allocated in non concurrent mode + reads chan []byte // pending data idx int // size of pending data handler func(int) + cum uint32 } func (*Reader) private() {} @@ -64,8 +72,21 @@ func (r *Reader) Size() int { return 0 } +func (r *Reader) isNotConcurrent() bool { + return r.num == 1 +} + func (r *Reader) init() error { - return r.frame.InitR(r.src) + data, err := r.frame.InitR(r.src, r.num) + if err != nil { + return err + } + r.reads = data + r.idx = 0 + size := r.frame.Descriptor.Flags.BlockSizeIndex() + r.data = size.Get() + r.cum = 0 + return nil } func (r *Reader) Read(buf []byte) (n int, err error) { @@ -79,65 +100,78 @@ func (r *Reader) Read(buf []byte) (n int, err error) { if err = r.init(); r.state.next(err) { return } - size := r.frame.Descriptor.Flags.BlockSizeIndex() - r.data = size.Get() default: return 0, r.state.fail() } - if len(buf) == 0 { - return - } - - var bn int - if r.idx > 0 { - // Some left over data, use it. - goto fillbuf - } - // No uncompressed data yet. - r.data = r.data[:cap(r.data)] - for len(buf) >= len(r.data) { - // Input buffer large enough and no pending data: uncompress directly into it. - switch bn, err = r.frame.Blocks.Block.Uncompress(r.frame, r.src, buf); err { - case nil: - r.handler(bn) - n += bn - buf = buf[bn:] - case io.EOF: - goto close - default: - return + for len(buf) > 0 { + var bn int + if r.idx == 0 { + if r.isNotConcurrent() { + bn, err = r.read(buf) + } else { + lz4block.Put(r.data) + r.data = <-r.reads + if len(r.data) == 0 { + // No uncompressed data: something went wrong or we are done. + err = r.frame.Blocks.ErrorR() + } + } + switch err { + case nil: + case io.EOF: + if er := r.frame.CloseR(r.src); er != nil { + err = er + } + lz4block.Put(r.data) + r.data = nil + return + default: + return + } } + if bn == 0 { + // Fill buf with buffered data. + bn = copy(buf, r.data[r.idx:]) + r.idx += bn + if r.idx == len(r.data) { + // All data read, get ready for the next Read. + r.idx = 0 + } + } + buf = buf[bn:] + n += bn + r.handler(bn) } - if n > 0 { - // Some data was read, done for now. - return + return +} + +// read uncompresses the next block as follow: +// - if buf has enough room, the block is uncompressed into it directly +// and the lenght of used space is returned +// - else, the uncompress data is stored in r.data and 0 is returned +func (r *Reader) read(buf []byte) (int, error) { + block := r.frame.Blocks.Block + _, err := block.Read(r.frame, r.src, r.cum) + if err != nil { + return 0, err } - // Read the next block. - switch bn, err = r.frame.Blocks.Block.Uncompress(r.frame, r.src, r.data); err { - case nil: - r.handler(bn) - r.data = r.data[:bn] - goto fillbuf - case io.EOF: - default: - return + var direct bool + dst := r.data[:cap(r.data)] + if len(buf) >= len(dst) { + // Uncompress directly into buf. + direct = true + dst = buf } -close: - if er := r.frame.CloseR(r.src); er != nil { - err = er + dst, err = block.Uncompress(r.frame, dst, true) + if err != nil { + return 0, err } - r.frame.Descriptor.Flags.BlockSizeIndex().Put(r.data) - r.Reset(nil) - return -fillbuf: - bn = copy(buf, r.data[r.idx:]) - n += bn - r.idx += bn - if r.idx == len(r.data) { - // All data read, get ready for the next Read. - r.idx = 0 + r.cum += uint32(len(dst)) + if direct { + return len(dst), nil } - return + r.data = dst + return 0, nil } // Reset clears the state of the Reader r such that it is equivalent to its @@ -146,11 +180,14 @@ fillbuf: // // w.Close must be called before Reset. func (r *Reader) Reset(reader io.Reader) { - r.frame.Reset(1) - r.src = reader - r.data = nil - r.idx = 0 + if r.data != nil { + lz4block.Put(r.data) + r.data = nil + } + r.frame.Reset(r.num) r.state.reset() + r.src = reader + r.reads = nil } // WriteTo efficiently uncompresses the data from the Reader underlying source to w. @@ -167,13 +204,28 @@ func (r *Reader) WriteTo(w io.Writer) (n int64, err error) { } defer r.state.nextd(&err) - var bn int - block := r.frame.Blocks.Block - size := r.frame.Descriptor.Flags.BlockSizeIndex() - data := size.Get() - defer size.Put(data) + var data []byte + if r.isNotConcurrent() { + size := r.frame.Descriptor.Flags.BlockSizeIndex() + data = size.Get() + defer lz4block.Put(data) + } for { - switch bn, err = block.Uncompress(r.frame, r.src, data); err { + var bn int + var dst []byte + if r.isNotConcurrent() { + bn, err = r.read(data) + dst = data[:bn] + } else { + lz4block.Put(dst) + dst = <-r.reads + bn = len(dst) + if bn == 0 { + // No uncompressed data: something went wrong or we are done. + err = r.frame.Blocks.ErrorR() + } + } + switch err { case nil: case io.EOF: err = r.frame.CloseR(r.src) @@ -182,7 +234,7 @@ func (r *Reader) WriteTo(w io.Writer) (n int64, err error) { return } r.handler(bn) - bn, err = w.Write(data[:bn]) + bn, err = w.Write(dst) n += int64(bn) if err != nil { return diff --git a/vendor/github.com/pierrec/lz4/v4/writer.go b/vendor/github.com/pierrec/lz4/v4/writer.go index e641be7c6a24..44a43d251b0d 100644 --- a/vendor/github.com/pierrec/lz4/v4/writer.go +++ b/vendor/github.com/pierrec/lz4/v4/writer.go @@ -35,6 +35,7 @@ type Writer struct { data []byte // pending data idx int // size of pending data handler func(int) + legacy bool } func (*Writer) private() {} @@ -63,9 +64,11 @@ func (w *Writer) isNotConcurrent() bool { // init sets up the Writer when in newState. It does not change the Writer state. func (w *Writer) init() error { - w.frame.InitW(w.src, w.num) - size := w.frame.Descriptor.Flags.BlockSizeIndex() - w.data = size.Get() + w.frame.InitW(w.src, w.num, w.legacy) + if true || !w.isNotConcurrent() { + size := w.frame.Descriptor.Flags.BlockSizeIndex() + w.data = size.Get() + } w.idx = 0 return w.frame.Descriptor.Write(w.frame, w.src) } @@ -126,20 +129,19 @@ func (w *Writer) write(data []byte, safe bool) error { w.handler(len(block.Data)) return err } - size := w.frame.Descriptor.Flags.BlockSizeIndex() c := make(chan *lz4stream.FrameDataBlock) w.frame.Blocks.Blocks <- c - go func(c chan *lz4stream.FrameDataBlock, data []byte, size lz4block.BlockSizeIndex, safe bool) { - b := lz4stream.NewFrameDataBlock(size) + go func(c chan *lz4stream.FrameDataBlock, data []byte, safe bool) { + b := lz4stream.NewFrameDataBlock(w.frame) c <- b.Compress(w.frame, data, w.level) <-c w.handler(len(b.Data)) - b.CloseW(w.frame) + b.Close(w.frame) if safe { // safe to put it back as the last usage of it was FrameDataBlock.Write() called before c is closed - size.Put(data) + lz4block.Put(data) } - }(c, data, size, safe) + }(c, data, safe) return nil } @@ -165,8 +167,7 @@ func (w *Writer) Close() (err error) { err = w.frame.CloseW(w.src, w.num) // It is now safe to free the buffer. if w.data != nil { - size := w.frame.Descriptor.Flags.BlockSizeIndex() - size.Put(w.data) + lz4block.Put(w.data) w.data = nil } return @@ -204,13 +205,13 @@ func (w *Writer) ReadFrom(r io.Reader) (n int64, err error) { data := size.Get() if w.isNotConcurrent() { // Keep the same buffer for the whole process. - defer size.Put(data) + defer lz4block.Put(data) } for !done { rn, err = io.ReadFull(r, data) switch err { case nil: - case io.EOF: + case io.EOF, io.ErrUnexpectedEOF: // read may be partial done = true default: return diff --git a/vendor/modules.txt b/vendor/modules.txt index a0993a17742e..e0ec461255b5 100644 --- a/vendor/modules.txt +++ b/vendor/modules.txt @@ -579,10 +579,15 @@ github.com/jstemmer/go-junit-report/formatter github.com/jstemmer/go-junit-report/parser # github.com/julienschmidt/httprouter v1.3.0 github.com/julienschmidt/httprouter -# github.com/klauspost/compress v1.9.5 +# github.com/klauspost/compress v1.11.3 ## explicit github.com/klauspost/compress/flate +github.com/klauspost/compress/fse github.com/klauspost/compress/gzip +github.com/klauspost/compress/huff0 +github.com/klauspost/compress/snappy +github.com/klauspost/compress/zstd +github.com/klauspost/compress/zstd/internal/xxhash # github.com/klauspost/cpuid v1.3.1 github.com/klauspost/cpuid # github.com/konsorten/go-windows-terminal-sequences v1.0.3 @@ -666,7 +671,7 @@ github.com/opentracing-contrib/go-stdlib/nethttp github.com/opentracing/opentracing-go github.com/opentracing/opentracing-go/ext github.com/opentracing/opentracing-go/log -# github.com/pierrec/lz4/v4 v4.0.2-0.20200813132121-22f5d580d5c4 +# github.com/pierrec/lz4/v4 v4.1.1 ## explicit github.com/pierrec/lz4/v4 github.com/pierrec/lz4/v4/internal/lz4block From d659dcc6140777548dc8f001786eb13869848f84 Mon Sep 17 00:00:00 2001 From: Cyril Tovena Date: Fri, 11 Dec 2020 10:11:25 +0100 Subject: [PATCH 2/3] Add missing vendor change. Signed-off-by: Cyril Tovena --- .../klauspost/compress/fse/README.md | 156 ++++++++-------- .../klauspost/compress/huff0/README.md | 174 +++++++++--------- 2 files changed, 165 insertions(+), 165 deletions(-) diff --git a/vendor/github.com/klauspost/compress/fse/README.md b/vendor/github.com/klauspost/compress/fse/README.md index 27d8ed56fcc4..ea7324da671f 100644 --- a/vendor/github.com/klauspost/compress/fse/README.md +++ b/vendor/github.com/klauspost/compress/fse/README.md @@ -1,79 +1,79 @@ -# Finite State Entropy - -This package provides Finite State Entropy encoding and decoding. - -Finite State Entropy (also referenced as [tANS](https://en.wikipedia.org/wiki/Asymmetric_numeral_systems#tANS)) -encoding provides a fast near-optimal symbol encoding/decoding -for byte blocks as implemented in [zstandard](https://github.com/facebook/zstd). - -This can be used for compressing input with a lot of similar input values to the smallest number of bytes. -This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, -but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. - -* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/fse) - -## News - - * Feb 2018: First implementation released. Consider this beta software for now. - -# Usage - -This package provides a low level interface that allows to compress single independent blocks. - -Each block is separate, and there is no built in integrity checks. -This means that the caller should keep track of block sizes and also do checksums if needed. - -Compressing a block is done via the [`Compress`](https://godoc.org/github.com/klauspost/compress/fse#Compress) function. -You must provide input and will receive the output and maybe an error. - -These error values can be returned: - -| Error | Description | -|---------------------|-----------------------------------------------------------------------------| -| `` | Everything ok, output is returned | -| `ErrIncompressible` | Returned when input is judged to be too hard to compress | -| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | -| `(error)` | An internal error occurred. | - -As can be seen above there are errors that will be returned even under normal operation so it is important to handle these. - -To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/fse#Scratch) object -that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same -object can be used for both. - -Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this -you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. - -Decompressing is done by calling the [`Decompress`](https://godoc.org/github.com/klauspost/compress/fse#Decompress) function. -You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back -your input was likely corrupted. - -It is important to note that a successful decoding does *not* mean your output matches your original input. -There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. - -For more detailed usage, see examples in the [godoc documentation](https://godoc.org/github.com/klauspost/compress/fse#pkg-examples). - -# Performance - -A lot of factors are affecting speed. Block sizes and compressibility of the material are primary factors. -All compression functions are currently only running on the calling goroutine so only one core will be used per block. - -The compressor is significantly faster if symbols are kept as small as possible. The highest byte value of the input -is used to reduce some of the processing, so if all your input is above byte value 64 for instance, it may be -beneficial to transpose all your input values down by 64. - -With moderate block sizes around 64k speed are typically 200MB/s per core for compression and -around 300MB/s decompression speed. - -The same hardware typically does Huffman (deflate) encoding at 125MB/s and decompression at 100MB/s. - -# Plans - -At one point, more internals will be exposed to facilitate more "expert" usage of the components. - -A streaming interface is also likely to be implemented. Likely compatible with [FSE stream format](https://github.com/Cyan4973/FiniteStateEntropy/blob/dev/programs/fileio.c#L261). - -# Contributing - -Contributions are always welcome. Be aware that adding public functions will require good justification and breaking +# Finite State Entropy + +This package provides Finite State Entropy encoding and decoding. + +Finite State Entropy (also referenced as [tANS](https://en.wikipedia.org/wiki/Asymmetric_numeral_systems#tANS)) +encoding provides a fast near-optimal symbol encoding/decoding +for byte blocks as implemented in [zstandard](https://github.com/facebook/zstd). + +This can be used for compressing input with a lot of similar input values to the smallest number of bytes. +This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, +but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. + +* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/fse) + +## News + + * Feb 2018: First implementation released. Consider this beta software for now. + +# Usage + +This package provides a low level interface that allows to compress single independent blocks. + +Each block is separate, and there is no built in integrity checks. +This means that the caller should keep track of block sizes and also do checksums if needed. + +Compressing a block is done via the [`Compress`](https://godoc.org/github.com/klauspost/compress/fse#Compress) function. +You must provide input and will receive the output and maybe an error. + +These error values can be returned: + +| Error | Description | +|---------------------|-----------------------------------------------------------------------------| +| `` | Everything ok, output is returned | +| `ErrIncompressible` | Returned when input is judged to be too hard to compress | +| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | +| `(error)` | An internal error occurred. | + +As can be seen above there are errors that will be returned even under normal operation so it is important to handle these. + +To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/fse#Scratch) object +that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same +object can be used for both. + +Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this +you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. + +Decompressing is done by calling the [`Decompress`](https://godoc.org/github.com/klauspost/compress/fse#Decompress) function. +You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back +your input was likely corrupted. + +It is important to note that a successful decoding does *not* mean your output matches your original input. +There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. + +For more detailed usage, see examples in the [godoc documentation](https://godoc.org/github.com/klauspost/compress/fse#pkg-examples). + +# Performance + +A lot of factors are affecting speed. Block sizes and compressibility of the material are primary factors. +All compression functions are currently only running on the calling goroutine so only one core will be used per block. + +The compressor is significantly faster if symbols are kept as small as possible. The highest byte value of the input +is used to reduce some of the processing, so if all your input is above byte value 64 for instance, it may be +beneficial to transpose all your input values down by 64. + +With moderate block sizes around 64k speed are typically 200MB/s per core for compression and +around 300MB/s decompression speed. + +The same hardware typically does Huffman (deflate) encoding at 125MB/s and decompression at 100MB/s. + +# Plans + +At one point, more internals will be exposed to facilitate more "expert" usage of the components. + +A streaming interface is also likely to be implemented. Likely compatible with [FSE stream format](https://github.com/Cyan4973/FiniteStateEntropy/blob/dev/programs/fileio.c#L261). + +# Contributing + +Contributions are always welcome. Be aware that adding public functions will require good justification and breaking changes will likely not be accepted. If in doubt open an issue before writing the PR. \ No newline at end of file diff --git a/vendor/github.com/klauspost/compress/huff0/README.md b/vendor/github.com/klauspost/compress/huff0/README.md index ec4f9098866c..e12da4db2fd1 100644 --- a/vendor/github.com/klauspost/compress/huff0/README.md +++ b/vendor/github.com/klauspost/compress/huff0/README.md @@ -1,87 +1,87 @@ -# Huff0 entropy compression - -This package provides Huff0 encoding and decoding as used in zstd. - -[Huff0](https://github.com/Cyan4973/FiniteStateEntropy#new-generation-entropy-coders), -a Huffman codec designed for modern CPU, featuring OoO (Out of Order) operations on multiple ALU -(Arithmetic Logic Unit), achieving extremely fast compression and decompression speeds. - -This can be used for compressing input with a lot of similar input values to the smallest number of bytes. -This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, -but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. - -* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/huff0) - -## News - - * Mar 2018: First implementation released. Consider this beta software for now. - -# Usage - -This package provides a low level interface that allows to compress single independent blocks. - -Each block is separate, and there is no built in integrity checks. -This means that the caller should keep track of block sizes and also do checksums if needed. - -Compressing a block is done via the [`Compress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress1X) and -[`Compress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress4X) functions. -You must provide input and will receive the output and maybe an error. - -These error values can be returned: - -| Error | Description | -|---------------------|-----------------------------------------------------------------------------| -| `` | Everything ok, output is returned | -| `ErrIncompressible` | Returned when input is judged to be too hard to compress | -| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | -| `ErrTooBig` | Returned if the input block exceeds the maximum allowed size (128 Kib) | -| `(error)` | An internal error occurred. | - - -As can be seen above some of there are errors that will be returned even under normal operation so it is important to handle these. - -To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object -that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same -object can be used for both. - -Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this -you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. - -The `Scratch` object will retain state that allows to re-use previous tables for encoding and decoding. - -## Tables and re-use - -Huff0 allows for reusing tables from the previous block to save space if that is expected to give better/faster results. - -The Scratch object allows you to set a [`ReusePolicy`](https://godoc.org/github.com/klauspost/compress/huff0#ReusePolicy) -that controls this behaviour. See the documentation for details. This can be altered between each block. - -Do however note that this information is *not* stored in the output block and it is up to the users of the package to -record whether [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable) should be called, -based on the boolean reported back from the CompressXX call. - -If you want to store the table separate from the data, you can access them as `OutData` and `OutTable` on the -[`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object. - -## Decompressing - -The first part of decoding is to initialize the decoding table through [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable). -This will initialize the decoding tables. -You can supply the complete block to `ReadTable` and it will return the data part of the block -which can be given to the decompressor. - -Decompressing is done by calling the [`Decompress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress1X) -or [`Decompress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress4X) function. - -For concurrently decompressing content with a fixed table a stateless [`Decoder`](https://godoc.org/github.com/klauspost/compress/huff0#Decoder) can be requested which will remain correct as long as the scratch is unchanged. The capacity of the provided slice indicates the expected output size. - -You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back -your input was likely corrupted. - -It is important to note that a successful decoding does *not* mean your output matches your original input. -There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. - -# Contributing - -Contributions are always welcome. Be aware that adding public functions will require good justification and breaking -changes will likely not be accepted. If in doubt open an issue before writing the PR. +# Huff0 entropy compression + +This package provides Huff0 encoding and decoding as used in zstd. + +[Huff0](https://github.com/Cyan4973/FiniteStateEntropy#new-generation-entropy-coders), +a Huffman codec designed for modern CPU, featuring OoO (Out of Order) operations on multiple ALU +(Arithmetic Logic Unit), achieving extremely fast compression and decompression speeds. + +This can be used for compressing input with a lot of similar input values to the smallest number of bytes. +This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders, +but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. + +* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/huff0) + +## News + + * Mar 2018: First implementation released. Consider this beta software for now. + +# Usage + +This package provides a low level interface that allows to compress single independent blocks. + +Each block is separate, and there is no built in integrity checks. +This means that the caller should keep track of block sizes and also do checksums if needed. + +Compressing a block is done via the [`Compress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress1X) and +[`Compress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress4X) functions. +You must provide input and will receive the output and maybe an error. + +These error values can be returned: + +| Error | Description | +|---------------------|-----------------------------------------------------------------------------| +| `` | Everything ok, output is returned | +| `ErrIncompressible` | Returned when input is judged to be too hard to compress | +| `ErrUseRLE` | Returned from the compressor when the input is a single byte value repeated | +| `ErrTooBig` | Returned if the input block exceeds the maximum allowed size (128 Kib) | +| `(error)` | An internal error occurred. | + + +As can be seen above some of there are errors that will be returned even under normal operation so it is important to handle these. + +To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object +that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same +object can be used for both. + +Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this +you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output. + +The `Scratch` object will retain state that allows to re-use previous tables for encoding and decoding. + +## Tables and re-use + +Huff0 allows for reusing tables from the previous block to save space if that is expected to give better/faster results. + +The Scratch object allows you to set a [`ReusePolicy`](https://godoc.org/github.com/klauspost/compress/huff0#ReusePolicy) +that controls this behaviour. See the documentation for details. This can be altered between each block. + +Do however note that this information is *not* stored in the output block and it is up to the users of the package to +record whether [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable) should be called, +based on the boolean reported back from the CompressXX call. + +If you want to store the table separate from the data, you can access them as `OutData` and `OutTable` on the +[`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object. + +## Decompressing + +The first part of decoding is to initialize the decoding table through [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable). +This will initialize the decoding tables. +You can supply the complete block to `ReadTable` and it will return the data part of the block +which can be given to the decompressor. + +Decompressing is done by calling the [`Decompress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress1X) +or [`Decompress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress4X) function. + +For concurrently decompressing content with a fixed table a stateless [`Decoder`](https://godoc.org/github.com/klauspost/compress/huff0#Decoder) can be requested which will remain correct as long as the scratch is unchanged. The capacity of the provided slice indicates the expected output size. + +You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back +your input was likely corrupted. + +It is important to note that a successful decoding does *not* mean your output matches your original input. +There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid. + +# Contributing + +Contributions are always welcome. Be aware that adding public functions will require good justification and breaking +changes will likely not be accepted. If in doubt open an issue before writing the PR. From 4249f45025b683d33b1e776e00a4fb1757fe178e Mon Sep 17 00:00:00 2001 From: Cyril Tovena Date: Thu, 7 Jan 2021 17:46:25 +0100 Subject: [PATCH 3/3] update go.mod. Signed-off-by: Cyril Tovena --- go.mod | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/go.mod b/go.mod index c95f38cedc52..24ff299110b7 100644 --- a/go.mod +++ b/go.mod @@ -37,8 +37,8 @@ require ( github.com/jmespath/go-jmespath v0.4.0 github.com/joncrlsn/dque v2.2.1-0.20200515025108-956d14155fa2+incompatible github.com/json-iterator/go v1.1.10 - github.com/mitchellh/mapstructure v1.3.3 github.com/klauspost/compress v1.11.3 + github.com/mitchellh/mapstructure v1.3.3 github.com/moby/term v0.0.0-20200915141129-7f0af18e79f2 // indirect github.com/mwitkow/go-conntrack v0.0.0-20190716064945-2f068394615f github.com/opentracing/opentracing-go v1.2.0