commit 69f9277bd7271210f89516dc8cecf587903df8e0 Author: Marko Seidel Date: Fri Nov 13 15:34:37 2020 +0100 archium - go/src (20201113) diff --git a/github.com/klauspost/compress/.gitattributes b/github.com/klauspost/compress/.gitattributes new file mode 100644 index 0000000..e1c269d --- /dev/null +++ b/github.com/klauspost/compress/.gitattributes @@ -0,0 +1 @@ +*.bin -text -diff diff --git a/github.com/klauspost/compress/.gitignore b/github.com/klauspost/compress/.gitignore new file mode 100644 index 0000000..daf913b --- /dev/null +++ b/github.com/klauspost/compress/.gitignore @@ -0,0 +1,24 @@ +# Compiled Object files, Static and Dynamic libs (Shared Objects) +*.o +*.a +*.so + +# Folders +_obj +_test + +# Architecture specific extensions/prefixes +*.[568vq] +[568vq].out + +*.cgo1.go +*.cgo2.c +_cgo_defun.c +_cgo_gotypes.go +_cgo_export.* + +_testmain.go + +*.exe +*.test +*.prof diff --git a/github.com/klauspost/compress/.goreleaser.yml b/github.com/klauspost/compress/.goreleaser.yml new file mode 100644 index 0000000..bf49ec8 --- /dev/null +++ b/github.com/klauspost/compress/.goreleaser.yml @@ -0,0 +1,98 @@ +# This is an example goreleaser.yaml file with some sane defaults. +# Make sure to check the documentation at http://goreleaser.com +before: + hooks: +builds: + - + id: "s2c" + binary: s2c + main: ./s2/cmd/s2c/main.go + env: + - CGO_ENABLED=0 + goos: + - linux + - freebsd + - netbsd + - windows + goarch: + - 386 + - amd64 + - arm + - arm64 + - ppc64 + - ppc64le + - mips64 + - mips64le + goarm: + - 7 + - + id: "s2d" + binary: s2d + main: ./s2/cmd/s2d/main.go + env: + - CGO_ENABLED=0 + goos: + - linux + - freebsd + - netbsd + - windows + goarch: + - 386 + - amd64 + - arm + - arm64 + - ppc64 + - ppc64le + - mips64 + - mips64le + goarm: + - 7 + +archives: + - + id: s2-binaries + name_template: "s2-cmds_{{ .Version }}_{{ .Os }}_{{ .Arch }}" + replacements: + darwin: Darwin + linux: Linux + windows: Windows + 386: i386 + amd64: x86_64 + freebsd: FreeBSD + netbsd: NetBSD + format_overrides: + - goos: windows + format: zip + files: + - s2/LICENSE + - s2/README.md +checksum: + name_template: 'checksums.txt' +snapshot: + name_template: "s2-cmds_{{ .Tag }}-next" +changelog: + sort: asc + filters: + exclude: + - '^doc:' + - '^docs:' + - '^test:' + - '^tests:' + - '^Update\sREADME.md' + +nfpms: + - + name_template: "s2-cmds_{{ .Version }}_{{ .Os }}_{{ .Arch }}" + vendor: Klaus Post + homepage: https://github.com/klauspost/compress + maintainer: Klaus Post + description: S2 Compression Tool + license: BSD 3-Clause + formats: + - deb + - rpm + replacements: + darwin: Darwin + linux: Linux + freebsd: FreeBSD + amd64: x86_64 diff --git a/github.com/klauspost/compress/.travis.yml b/github.com/klauspost/compress/.travis.yml new file mode 100644 index 0000000..e5cd2a5 --- /dev/null +++ b/github.com/klauspost/compress/.travis.yml @@ -0,0 +1,45 @@ +language: go + +os: + - linux + - osx + +go: + - 1.12.x + - 1.13.x + - 1.14.x + - master + +env: + - GO111MODULE=off + +install: + - go get ./... + - go get github.com/klauspost/compress-fuzz + +script: + - diff <(gofmt -d .) <(printf "") + - IS_GO112=`go version | cut -d ' ' -f3 | grep 1.12`; if [ ! -z "$IS_GO112" ]; then echo 'Skipping vet on Go 1.12...'; else go vet ./...; fi + - go test -cpu=2 ./... + - go test -cpu=2 -tags=noasm ./... + - go test -cpu=1,4 -short -race ./... + - go build github.com/klauspost/compress/s2/cmd/s2c && go build github.com/klauspost/compress/s2/cmd/s2d && s2c s2c && s2d s2c.s2 && rm s2c && rm s2d && rm s2c.s2 + +jobs: + allow_failures: + - go: 'master' + fast_finish: true + include: + - stage: 386 linux test + go: 1.14.x + script: + - GOOS=linux GOARCH=386 go test -short ./... + +deploy: +- provider: script + skip_cleanup: true + script: curl -sL https://git.io/goreleaser | VERSION=v0.127.0 bash || true + on: + tags: true + condition: $TRAVIS_OS_NAME = linux + go: 1.14.x diff --git a/github.com/klauspost/compress/LICENSE b/github.com/klauspost/compress/LICENSE new file mode 100644 index 0000000..1eb75ef --- /dev/null +++ b/github.com/klauspost/compress/LICENSE @@ -0,0 +1,28 @@ +Copyright (c) 2012 The Go Authors. All rights reserved. +Copyright (c) 2019 Klaus Post. 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/github.com/klauspost/compress/README.md b/github.com/klauspost/compress/README.md new file mode 100644 index 0000000..d196091 --- /dev/null +++ b/github.com/klauspost/compress/README.md @@ -0,0 +1,300 @@ +# compress + +This package provides various compression algorithms. + +* [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and decompression in pure Go. +* [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) is a high performance replacement for Snappy. +* Optimized [deflate](https://godoc.org/github.com/klauspost/compress/flate) packages which can be used as a dropin replacement for [gzip](https://godoc.org/github.com/klauspost/compress/gzip), [zip](https://godoc.org/github.com/klauspost/compress/zip) and [zlib](https://godoc.org/github.com/klauspost/compress/zlib). +* [huff0](https://github.com/klauspost/compress/tree/master/huff0) and [FSE](https://github.com/klauspost/compress/tree/master/fse) implementations for raw entropy encoding. +* [pgzip](https://github.com/klauspost/pgzip) is a separate package that provides a very fast parallel gzip implementation. +* [fuzz package](https://github.com/klauspost/compress-fuzz) for fuzz testing all compressors/decompressors here. + +[![Documentation](https://godoc.org/github.com/klauspost/compress?status.svg)](https://pkg.go.dev/github.com/klauspost/compress?tab=subdirectories) +[![Build Status](https://travis-ci.org/klauspost/compress.svg?branch=master)](https://travis-ci.org/klauspost/compress) +[![Sourcegraph Badge](https://sourcegraph.com/github.com/klauspost/compress/-/badge.svg)](https://sourcegraph.com/github.com/klauspost/compress?badge) + +# changelog + +* June 23, 2020 (v1.10.10) zstd: Skip entropy compression in fastest mode when no matches. [#270](https://github.com/klauspost/compress/pull/270) +* June 16, 2020 (v1.10.9): zstd: API change for specifying dictionaries. See [#268](https://github.com/klauspost/compress/pull/268) +* June 16, 2020: zip: update CreateHeaderRaw to handle zip64 fields. [#266](https://github.com/klauspost/compress/pull/266) +* June 16, 2020: Fuzzit tests removed. The service has been purchased and is no longer available. +* June 5, 2020 (v1.10.8): 1.15x faster zstd block decompression. [#265](https://github.com/klauspost/compress/pull/265) +* June 1, 2020 (v1.10.7): Added zstd decompression [dictionary support](https://github.com/klauspost/compress/tree/master/zstd#dictionaries) +* June 1, 2020: Increase zstd decompression speed up to 1.19x. [#259](https://github.com/klauspost/compress/pull/259) +* June 1, 2020: Remove internal reset call in zstd compression and reduce allocations. [#263](https://github.com/klauspost/compress/pull/263) +* May 21, 2020: (v1.10.6) zstd: Reduce allocations while decoding. [#258](https://github.com/klauspost/compress/pull/258), [#252](https://github.com/klauspost/compress/pull/252) +* May 21, 2020: zstd: Stricter decompression checks. +* April 12, 2020: (v1.10.5) s2-commands: Flush output when receiving SIGINT. [#239](https://github.com/klauspost/compress/pull/239) +* Apr 8, 2020: (v1.10.4) zstd: Minor/special case optimizations. [#251](https://github.com/klauspost/compress/pull/251), [#250](https://github.com/klauspost/compress/pull/250), [#249](https://github.com/klauspost/compress/pull/249), [#247](https://github.com/klauspost/compress/pull/247) +* Mar 11, 2020: (v1.10.3) s2: Use S2 encoder in pure Go mode for Snappy output as well. [#245](https://github.com/klauspost/compress/pull/245) +* Mar 10, 2020: s2: Fix pure Go block encoder. [#244](https://github.com/klauspost/compress/pull/244) +* Mar 9, 2020: zstd: Added "better compression" mode. [#240](https://github.com/klauspost/compress/pull/240) +* Mar 9, 2020: zstd: Improve speed of fastest compression mode by 5-10% [#241](https://github.com/klauspost/compress/pull/241) +* Feb 28, 2020: zstd: Skip creating encoders when not needed. [#238](https://github.com/klauspost/compress/pull/238) +* Feb 27, 2020: (v1.10.2) Close to 50% speedup in inflate (gzip/zip decompression). [#236](https://github.com/klauspost/compress/pull/236) [#234](https://github.com/klauspost/compress/pull/234) [#232](https://github.com/klauspost/compress/pull/232) +* Feb 23, 2020: Reduce deflate level 1-6 memory usage up to 59%. [#227](https://github.com/klauspost/compress/pull/227) +* Feb 18, 2020: (v1.10.1) Fix zstd crash when resetting multiple times without sending data. [#226](https://github.com/klauspost/compress/pull/226) +* Feb 16, 2020: deflate: Fix dictionary use on level 1-6. [#224](https://github.com/klauspost/compress/pull/224) +* Feb 16, 2020: Remove deflate writer reference when closing. [#224](https://github.com/klauspost/compress/pull/224) +* Feb 4, 2020: (v1.10.0) Add optional dictionary to [stateless deflate](https://pkg.go.dev/github.com/klauspost/compress/flate?tab=doc#StatelessDeflate). Breaking change, send `nil` for previous behaviour. [#216](https://github.com/klauspost/compress/pull/216) +* Feb 3, 2020: Fix buffer overflow on repeated small block deflate. [#218](https://github.com/klauspost/compress/pull/218) +* Jan 31, 2020: Allow copying content from an existing ZIP file without decompressing+compressing. [#214](https://github.com/klauspost/compress/pull/214) +* Jan 28, 2020: Added [S2](https://github.com/klauspost/compress/tree/master/s2#s2-compression) AMD64 assembler and various optimizations. Stream speed >10GB/s. [#186](https://github.com/klauspost/compress/pull/186) + +
+ See changes prior to v1.10.0 + +* Jan 20,2020 (v1.9.8) Optimize gzip/deflate with better size estimates and faster table generation. [#207](https://github.com/klauspost/compress/pull/207) by [luyu6056](https://github.com/luyu6056), [#206](https://github.com/klauspost/compress/pull/206). +* Jan 11, 2020: S2 Encode/Decode will use provided buffer if capacity is big enough. [#204](https://github.com/klauspost/compress/pull/204) +* Jan 5, 2020: (v1.9.7) Fix another zstd regression in v1.9.5 - v1.9.6 removed. +* Jan 4, 2020: (v1.9.6) Regression in v1.9.5 fixed causing corrupt zstd encodes in rare cases. +* Jan 4, 2020: Faster IO in [s2c + s2d commandline tools](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) compression/decompression. [#192](https://github.com/klauspost/compress/pull/192) +* Dec 29, 2019: Removed v1.9.5 since fuzz tests showed a compatibility problem with the reference zstandard decoder. +* Dec 29, 2019: (v1.9.5) zstd: 10-20% faster block compression. [#199](https://github.com/klauspost/compress/pull/199) +* Dec 29, 2019: [zip](https://godoc.org/github.com/klauspost/compress/zip) package updated with latest Go features +* Dec 29, 2019: zstd: Single segment flag condintions tweaked. [#197](https://github.com/klauspost/compress/pull/197) +* Dec 18, 2019: s2: Faster compression when ReadFrom is used. [#198](https://github.com/klauspost/compress/pull/198) +* Dec 10, 2019: s2: Fix repeat length output when just above at 16MB limit. +* Dec 10, 2019: zstd: Add function to get decoder as io.ReadCloser. [#191](https://github.com/klauspost/compress/pull/191) +* Dec 3, 2019: (v1.9.4) S2: limit max repeat length. [#188](https://github.com/klauspost/compress/pull/188) +* Dec 3, 2019: Add [WithNoEntropyCompression](https://godoc.org/github.com/klauspost/compress/zstd#WithNoEntropyCompression) to zstd [#187](https://github.com/klauspost/compress/pull/187) +* Dec 3, 2019: Reduce memory use for tests. Check for leaked goroutines. +* Nov 28, 2019 (v1.9.3) Less allocations in stateless deflate. +* Nov 28, 2019: 5-20% Faster huff0 decode. Impacts zstd as well. [#184](https://github.com/klauspost/compress/pull/184) +* Nov 12, 2019 (v1.9.2) Added [Stateless Compression](#stateless-compression) for gzip/deflate. +* Nov 12, 2019: Fixed zstd decompression of large single blocks. [#180](https://github.com/klauspost/compress/pull/180) +* Nov 11, 2019: Set default [s2c](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) block size to 4MB. +* Nov 11, 2019: Reduce inflate memory use by 1KB. +* Nov 10, 2019: Less allocations in deflate bit writer. +* Nov 10, 2019: Fix inconsistent error returned by zstd decoder. +* Oct 28, 2019 (v1.9.1) ztsd: Fix crash when compressing blocks. [#174](https://github.com/klauspost/compress/pull/174) +* Oct 24, 2019 (v1.9.0) zstd: Fix rare data corruption [#173](https://github.com/klauspost/compress/pull/173) +* Oct 24, 2019 zstd: Fix huff0 out of buffer write [#171](https://github.com/klauspost/compress/pull/171) and always return errors [#172](https://github.com/klauspost/compress/pull/172) +* Oct 10, 2019: Big deflate rewrite, 30-40% faster with better compression [#105](https://github.com/klauspost/compress/pull/105) + +
+ +
+ See changes prior to v1.9.0 + +* Oct 10, 2019: (v1.8.6) zstd: Allow partial reads to get flushed data. [#169](https://github.com/klauspost/compress/pull/169) +* Oct 3, 2019: Fix inconsistent results on broken zstd streams. +* Sep 25, 2019: Added `-rm` (remove source files) and `-q` (no output except errors) to `s2c` and `s2d` [commands](https://github.com/klauspost/compress/tree/master/s2#commandline-tools) +* Sep 16, 2019: (v1.8.4) Add `s2c` and `s2d` [commandline tools](https://github.com/klauspost/compress/tree/master/s2#commandline-tools). +* Sep 10, 2019: (v1.8.3) Fix s2 decoder [Skip](https://godoc.org/github.com/klauspost/compress/s2#Reader.Skip). +* Sep 7, 2019: zstd: Added [WithWindowSize](https://godoc.org/github.com/klauspost/compress/zstd#WithWindowSize), contributed by [ianwilkes](https://github.com/ianwilkes). +* Sep 5, 2019: (v1.8.2) Add [WithZeroFrames](https://godoc.org/github.com/klauspost/compress/zstd#WithZeroFrames) which adds full zero payload block encoding option. +* Sep 5, 2019: Lazy initialization of zstandard predefined en/decoder tables. +* Aug 26, 2019: (v1.8.1) S2: 1-2% compression increase in "better" compression mode. +* Aug 26, 2019: zstd: Check maximum size of Huffman 1X compressed literals while decoding. +* Aug 24, 2019: (v1.8.0) Added [S2 compression](https://github.com/klauspost/compress/tree/master/s2#s2-compression), a high performance replacement for Snappy. +* Aug 21, 2019: (v1.7.6) Fixed minor issues found by fuzzer. One could lead to zstd not decompressing. +* Aug 18, 2019: Add [fuzzit](https://fuzzit.dev/) continuous fuzzing. +* Aug 14, 2019: zstd: Skip incompressible data 2x faster. [#147](https://github.com/klauspost/compress/pull/147) +* Aug 4, 2019 (v1.7.5): Better literal compression. [#146](https://github.com/klauspost/compress/pull/146) +* Aug 4, 2019: Faster zstd compression. [#143](https://github.com/klauspost/compress/pull/143) [#144](https://github.com/klauspost/compress/pull/144) +* Aug 4, 2019: Faster zstd decompression. [#145](https://github.com/klauspost/compress/pull/145) [#143](https://github.com/klauspost/compress/pull/143) [#142](https://github.com/klauspost/compress/pull/142) +* July 15, 2019 (v1.7.4): Fix double EOF block in rare cases on zstd encoder. +* July 15, 2019 (v1.7.3): Minor speedup/compression increase in default zstd encoder. +* July 14, 2019: zstd decoder: Fix decompression error on multiple uses with mixed content. +* July 7, 2019 (v1.7.2): Snappy update, zstd decoder potential race fix. +* June 17, 2019: zstd decompression bugfix. +* June 17, 2019: fix 32 bit builds. +* June 17, 2019: Easier use in modules (less dependencies). +* June 9, 2019: New stronger "default" [zstd](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression mode. Matches zstd default compression ratio. +* June 5, 2019: 20-40% throughput in [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and better compression. +* June 5, 2019: deflate/gzip compression: Reduce memory usage of lower compression levels. +* June 2, 2019: Added [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression! +* May 25, 2019: deflate/gzip: 10% faster bit writer, mostly visible in lower levels. +* Apr 22, 2019: [zstd](https://github.com/klauspost/compress/tree/master/zstd#zstd) decompression added. +* Aug 1, 2018: Added [huff0 README](https://github.com/klauspost/compress/tree/master/huff0#huff0-entropy-compression). +* Jul 8, 2018: Added [Performance Update 2018](#performance-update-2018) below. +* Jun 23, 2018: Merged [Go 1.11 inflate optimizations](https://go-review.googlesource.com/c/go/+/102235). Go 1.9 is now required. Backwards compatible version tagged with [v1.3.0](https://github.com/klauspost/compress/releases/tag/v1.3.0). +* Apr 2, 2018: Added [huff0](https://godoc.org/github.com/klauspost/compress/huff0) en/decoder. Experimental for now, API may change. +* Mar 4, 2018: Added [FSE Entropy](https://godoc.org/github.com/klauspost/compress/fse) en/decoder. Experimental for now, API may change. +* Nov 3, 2017: Add compression [Estimate](https://godoc.org/github.com/klauspost/compress#Estimate) function. +* May 28, 2017: Reduce allocations when resetting decoder. +* Apr 02, 2017: Change back to official crc32, since changes were merged in Go 1.7. +* Jan 14, 2017: Reduce stack pressure due to array copies. See [Issue #18625](https://github.com/golang/go/issues/18625). +* Oct 25, 2016: Level 2-4 have been rewritten and now offers significantly better performance than before. +* Oct 20, 2016: Port zlib changes from Go 1.7 to fix zlib writer issue. Please update. +* Oct 16, 2016: Go 1.7 changes merged. Apples to apples this package is a few percent faster, but has a significantly better balance between speed and compression per level. +* Mar 24, 2016: Always attempt Huffman encoding on level 4-7. This improves base 64 encoded data compression. +* Mar 24, 2016: Small speedup for level 1-3. +* Feb 19, 2016: Faster bit writer, level -2 is 15% faster, level 1 is 4% faster. +* Feb 19, 2016: Handle small payloads faster in level 1-3. +* Feb 19, 2016: Added faster level 2 + 3 compression modes. +* Feb 19, 2016: [Rebalanced compression levels](https://blog.klauspost.com/rebalancing-deflate-compression-levels/), so there is a more even progresssion in terms of compression. New default level is 5. +* Feb 14, 2016: Snappy: Merge upstream changes. +* Feb 14, 2016: Snappy: Fix aggressive skipping. +* Feb 14, 2016: Snappy: Update benchmark. +* Feb 13, 2016: Deflate: Fixed assembler problem that could lead to sub-optimal compression. +* Feb 12, 2016: Snappy: Added AMD64 SSE 4.2 optimizations to matching, which makes easy to compress material run faster. Typical speedup is around 25%. +* Feb 9, 2016: Added Snappy package fork. This version is 5-7% faster, much more on hard to compress content. +* Jan 30, 2016: Optimize level 1 to 3 by not considering static dictionary or storing uncompressed. ~4-5% speedup. +* Jan 16, 2016: Optimization on deflate level 1,2,3 compression. +* Jan 8 2016: Merge [CL 18317](https://go-review.googlesource.com/#/c/18317): fix reading, writing of zip64 archives. +* Dec 8 2015: Make level 1 and -2 deterministic even if write size differs. +* Dec 8 2015: Split encoding functions, so hashing and matching can potentially be inlined. 1-3% faster on AMD64. 5% faster on other platforms. +* Dec 8 2015: Fixed rare [one byte out-of bounds read](https://github.com/klauspost/compress/issues/20). Please update! +* Nov 23 2015: Optimization on token writer. ~2-4% faster. Contributed by [@dsnet](https://github.com/dsnet). +* Nov 20 2015: Small optimization to bit writer on 64 bit systems. +* Nov 17 2015: Fixed out-of-bound errors if the underlying Writer returned an error. See [#15](https://github.com/klauspost/compress/issues/15). +* Nov 12 2015: Added [io.WriterTo](https://golang.org/pkg/io/#WriterTo) support to gzip/inflate. +* Nov 11 2015: Merged [CL 16669](https://go-review.googlesource.com/#/c/16669/4): archive/zip: enable overriding (de)compressors per file +* Oct 15 2015: Added skipping on uncompressible data. Random data speed up >5x. + +
+ +# deflate usage + +* [High Throughput Benchmark](http://blog.klauspost.com/go-gzipdeflate-benchmarks/). +* [Small Payload/Webserver Benchmarks](http://blog.klauspost.com/gzip-performance-for-go-webservers/). +* [Linear Time Compression](http://blog.klauspost.com/constant-time-gzipzip-compression/). +* [Re-balancing Deflate Compression Levels](https://blog.klauspost.com/rebalancing-deflate-compression-levels/) + +The packages are drop-in replacements for standard libraries. Simply replace the import path to use them: + +| old import | new import | Documentation +|--------------------|-----------------------------------------|--------------------| +| `compress/gzip` | `github.com/klauspost/compress/gzip` | [gzip](https://pkg.go.dev/github.com/klauspost/compress/gzip?tab=doc) +| `compress/zlib` | `github.com/klauspost/compress/zlib` | [zlib](https://pkg.go.dev/github.com/klauspost/compress/zlib?tab=doc) +| `archive/zip` | `github.com/klauspost/compress/zip` | [zip](https://pkg.go.dev/github.com/klauspost/compress/zip?tab=doc) +| `compress/flate` | `github.com/klauspost/compress/flate` | [flate](https://pkg.go.dev/github.com/klauspost/compress/flate?tab=doc) + +* Optimized [deflate](https://godoc.org/github.com/klauspost/compress/flate) packages which can be used as a dropin replacement for [gzip](https://godoc.org/github.com/klauspost/compress/gzip), [zip](https://godoc.org/github.com/klauspost/compress/zip) and [zlib](https://godoc.org/github.com/klauspost/compress/zlib). + +You may also be interested in [pgzip](https://github.com/klauspost/pgzip), which is a drop in replacement for gzip, which support multithreaded compression on big files and the optimized [crc32](https://github.com/klauspost/crc32) package used by these packages. + +The packages contains the same as the standard library, so you can use the godoc for that: [gzip](http://golang.org/pkg/compress/gzip/), [zip](http://golang.org/pkg/archive/zip/), [zlib](http://golang.org/pkg/compress/zlib/), [flate](http://golang.org/pkg/compress/flate/). + +Currently there is only minor speedup on decompression (mostly CRC32 calculation). + +# Stateless compression + +This package offers stateless compression as a special option for gzip/deflate. +It will do compression but without maintaining any state between Write calls. + +This means there will be no memory kept between Write calls, but compression and speed will be suboptimal. + +This is only relevant in cases where you expect to run many thousands of compressors concurrently, +but with very little activity. This is *not* intended for regular web servers serving individual requests. + +Because of this, the size of actual Write calls will affect output size. + +In gzip, specify level `-3` / `gzip.StatelessCompression` to enable. + +For direct deflate use, NewStatelessWriter and StatelessDeflate are available. See [documentation](https://godoc.org/github.com/klauspost/compress/flate#NewStatelessWriter) + +A `bufio.Writer` can of course be used to control write sizes. For example, to use a 4KB buffer: + +``` + // replace 'ioutil.Discard' with your output. + gzw, err := gzip.NewWriterLevel(ioutil.Discard, gzip.StatelessCompression) + if err != nil { + return err + } + defer gzw.Close() + + w := bufio.NewWriterSize(gzw, 4096) + defer w.Flush() + + // Write to 'w' +``` + +This will only use up to 4KB in memory when the writer is idle. + +Compression is almost always worse than the fastest compression level +and each write will allocate (a little) memory. + +# Performance Update 2018 + +It has been a while since we have been looking at the speed of this package compared to the standard library, so I thought I would re-do my tests and give some overall recommendations based on the current state. All benchmarks have been performed with Go 1.10 on my Desktop Intel(R) Core(TM) i7-2600 CPU @3.40GHz. Since I last ran the tests, I have gotten more RAM, which means tests with big files are no longer limited by my SSD. + +The raw results are in my [updated spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing). Due to cgo changes and upstream updates i could not get the cgo version of gzip to compile. Instead I included the [zstd](https://github.com/datadog/zstd) cgo implementation. If I get cgo gzip to work again, I might replace the results in the sheet. + +The columns to take note of are: *MB/s* - the throughput. *Reduction* - the data size reduction in percent of the original. *Rel Speed* relative speed compared to the standard library at the same level. *Smaller* - how many percent smaller is the compressed output compared to stdlib. Negative means the output was bigger. *Loss* means the loss (or gain) in compression as a percentage difference of the input. + +The `gzstd` (standard library gzip) and `gzkp` (this package gzip) only uses one CPU core. [`pgzip`](https://github.com/klauspost/pgzip), [`bgzf`](https://github.com/biogo/hts/tree/master/bgzf) uses all 4 cores. [`zstd`](https://github.com/DataDog/zstd) uses one core, and is a beast (but not Go, yet). + + +## Overall differences. + +There appears to be a roughly 5-10% speed advantage over the standard library when comparing at similar compression levels. + +The biggest difference you will see is the result of [re-balancing](https://blog.klauspost.com/rebalancing-deflate-compression-levels/) the compression levels. I wanted by library to give a smoother transition between the compression levels than the standard library. + +This package attempts to provide a more smooth transition, where "1" is taking a lot of shortcuts, "5" is the reasonable trade-off and "9" is the "give me the best compression", and the values in between gives something reasonable in between. The standard library has big differences in levels 1-4, but levels 5-9 having no significant gains - often spending a lot more time than can be justified by the achieved compression. + +There are links to all the test data in the [spreadsheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing) in the top left field on each tab. + +## Web Content + +This test set aims to emulate typical use in a web server. The test-set is 4GB data in 53k files, and is a mixture of (mostly) HTML, JS, CSS. + +Since level 1 and 9 are close to being the same code, they are quite close. But looking at the levels in-between the differences are quite big. + +Looking at level 6, this package is 88% faster, but will output about 6% more data. For a web server, this means you can serve 88% more data, but have to pay for 6% more bandwidth. You can draw your own conclusions on what would be the most expensive for your case. + +## Object files + +This test is for typical data files stored on a server. In this case it is a collection of Go precompiled objects. They are very compressible. + +The picture is similar to the web content, but with small differences since this is very compressible. Levels 2-3 offer good speed, but is sacrificing quite a bit of compression. + +The standard library seems suboptimal on level 3 and 4 - offering both worse compression and speed than level 6 & 7 of this package respectively. + +## Highly Compressible File + +This is a JSON file with very high redundancy. The reduction starts at 95% on level 1, so in real life terms we are dealing with something like a highly redundant stream of data, etc. + +It is definitely visible that we are dealing with specialized content here, so the results are very scattered. This package does not do very well at levels 1-4, but picks up significantly at level 5 and levels 7 and 8 offering great speed for the achieved compression. + +So if you know you content is extremely compressible you might want to go slightly higher than the defaults. The standard library has a huge gap between levels 3 and 4 in terms of speed (2.75x slowdown), so it offers little "middle ground". + +## Medium-High Compressible + +This is a pretty common test corpus: [enwik9](http://mattmahoney.net/dc/textdata.html). It contains the first 10^9 bytes of the English Wikipedia dump on Mar. 3, 2006. This is a very good test of typical text based compression and more data heavy streams. + +We see a similar picture here as in "Web Content". On equal levels some compression is sacrificed for more speed. Level 5 seems to be the best trade-off between speed and size, beating stdlib level 3 in both. + +## Medium Compressible + +I will combine two test sets, one [10GB file set](http://mattmahoney.net/dc/10gb.html) and a VM disk image (~8GB). Both contain different data types and represent a typical backup scenario. + +The most notable thing is how quickly the standard library drops to very low compression speeds around level 5-6 without any big gains in compression. Since this type of data is fairly common, this does not seem like good behavior. + + +## Un-compressible Content + +This is mainly a test of how good the algorithms are at detecting un-compressible input. The standard library only offers this feature with very conservative settings at level 1. Obviously there is no reason for the algorithms to try to compress input that cannot be compressed. The only downside is that it might skip some compressible data on false detections. + + +# linear time compression (huffman only) + +This compression library adds a special compression level, named `HuffmanOnly`, which allows near linear time compression. This is done by completely disabling matching of previous data, and only reduce the number of bits to represent each character. + +This means that often used characters, like 'e' and ' ' (space) in text use the fewest bits to represent, and rare characters like '¤' takes more bits to represent. For more information see [wikipedia](https://en.wikipedia.org/wiki/Huffman_coding) or this nice [video](https://youtu.be/ZdooBTdW5bM). + +Since this type of compression has much less variance, the compression speed is mostly unaffected by the input data, and is usually more than *180MB/s* for a single core. + +The downside is that the compression ratio is usually considerably worse than even the fastest conventional compression. The compression ratio can never be better than 8:1 (12.5%). + +The linear time compression can be used as a "better than nothing" mode, where you cannot risk the encoder to slow down on some content. For comparison, the size of the "Twain" text is *233460 bytes* (+29% vs. level 1) and encode speed is 144MB/s (4.5x level 1). So in this case you trade a 30% size increase for a 4 times speedup. + +For more information see my blog post on [Fast Linear Time Compression](http://blog.klauspost.com/constant-time-gzipzip-compression/). + +This is implemented on Go 1.7 as "Huffman Only" mode, though not exposed for gzip. + + +# snappy package + +The standard snappy package has now been improved. This repo contains a copy of the snappy repo. + +I would advise to use the standard package: https://github.com/golang/snappy + + +# license + +This code is licensed under the same conditions as the original Go code. See LICENSE file. diff --git a/github.com/klauspost/compress/compressible.go b/github.com/klauspost/compress/compressible.go new file mode 100644 index 0000000..ea5a692 --- /dev/null +++ b/github.com/klauspost/compress/compressible.go @@ -0,0 +1,85 @@ +package compress + +import "math" + +// Estimate returns a normalized compressibility estimate of block b. +// Values close to zero are likely uncompressible. +// Values above 0.1 are likely to be compressible. +// Values above 0.5 are very compressible. +// Very small lengths will return 0. +func Estimate(b []byte) float64 { + if len(b) < 16 { + return 0 + } + + // Correctly predicted order 1 + hits := 0 + lastMatch := false + var o1 [256]byte + var hist [256]int + c1 := byte(0) + for _, c := range b { + if c == o1[c1] { + // We only count a hit if there was two correct predictions in a row. + if lastMatch { + hits++ + } + lastMatch = true + } else { + lastMatch = false + } + o1[c1] = c + c1 = c + hist[c]++ + } + + // Use x^0.6 to give better spread + prediction := math.Pow(float64(hits)/float64(len(b)), 0.6) + + // Calculate histogram distribution + variance := float64(0) + avg := float64(len(b)) / 256 + + for _, v := range hist { + Δ := float64(v) - avg + variance += Δ * Δ + } + + stddev := math.Sqrt(float64(variance)) / float64(len(b)) + exp := math.Sqrt(1 / float64(len(b))) + + // Subtract expected stddev + stddev -= exp + if stddev < 0 { + stddev = 0 + } + stddev *= 1 + exp + + // Use x^0.4 to give better spread + entropy := math.Pow(stddev, 0.4) + + // 50/50 weight between prediction and histogram distribution + return math.Pow((prediction+entropy)/2, 0.9) +} + +// ShannonEntropyBits returns the number of bits minimum required to represent +// an entropy encoding of the input bytes. +// https://en.wiktionary.org/wiki/Shannon_entropy +func ShannonEntropyBits(b []byte) int { + if len(b) == 0 { + return 0 + } + var hist [256]int + for _, c := range b { + hist[c]++ + } + shannon := float64(0) + invTotal := 1.0 / float64(len(b)) + for _, v := range hist[:] { + if v > 0 { + n := float64(v) + shannon += math.Ceil(-math.Log2(n*invTotal) * n) + } + } + return int(math.Ceil(shannon)) +} diff --git a/github.com/klauspost/compress/compressible_test.go b/github.com/klauspost/compress/compressible_test.go new file mode 100644 index 0000000..16817c6 --- /dev/null +++ b/github.com/klauspost/compress/compressible_test.go @@ -0,0 +1,302 @@ +package compress + +import ( + "crypto/rand" + "encoding/base32" + "io/ioutil" + "strconv" + "strings" + "testing" + + "github.com/klauspost/compress/flate" + "github.com/klauspost/compress/gzip" +) + +func BenchmarkEstimate(b *testing.B) { + b.ReportAllocs() + // (predictable, low entropy distibution) + b.Run("zeroes-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + + // (predictable, high entropy distibution) + b.Run("predictable-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + for i := range testData { + testData[i] = byte(float64(i) / float64(len(testData)) * 256) + } + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-500b", func(b *testing.B) { + var testData = make([]byte, 500) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-50k", func(b *testing.B) { + var testData = make([]byte, 50000) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-500k", func(b *testing.B) { + var testData = make([]byte, 500000) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + + // (not predictable, medium entropy distibution) + b.Run("base-32-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + rand.Read(testData) + s := base32.StdEncoding.EncodeToString(testData) + testData = []byte(s) + testData = testData[:5000] + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) + // (medium predictable, medium entropy distibution) + b.Run("text", func(b *testing.B) { + var testData = []byte(`If compression is done per-chunk, care should be taken that it doesn't leave restic backups open to watermarking/fingerprinting attacks. +This is essentially the same problem we discussed related to fingerprinting the CDC deduplication process: +With "naive" CDC, a "known plaintext" file can be verified to exist within the backup if the size of individual blocks can be observed by an attacker, by using CDC on the file in parallel and comparing the resulting amount of chunks and individual chunk lengths. +As discussed earlier, this can be somewhat mitigated by salting the CDC algorithm with a secret value, as done in attic. +With salted CDC, I assume compression would happen on each individual chunk, after splitting the problematic file into chunks. Restic chunks are in the range of 512 KB to 8 MB (but not evenly distributed - right?). +Attacker knows that the CDC algorithm uses a secret salt, so the attacker generates a range of chunks consisting of the first 512 KB to 8 MB of the file, one for each valid chunk length. The attacker is also able to determine the lengths of compressed chunks. +The attacker then compresses that chunk using the compression algorithm. +The attacker compares the lengths of the resulting chunks to the first chunk in the restic backup sets. +IF a matching block length is found, the attacker repeats the exercise with the next chunk, and the next chunk, and the next chunk, ... and the next chunk. +It is my belief that with sufficiently large files, and considering the fact that the CDC algorithm is "biased" (in lack of better of words) towards generating blocks of about 1 MB, this would be sufficient to ascertain whether or not a certain large file exists in the backup. +AS always, a paranoid and highly unscientific stream of consciousness. +Thoughts?`) + testData = append(testData, testData...) + testData = append(testData, testData...) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + Estimate(testData) + } + b.Log(Estimate(testData)) + }) +} + +func BenchmarkSnannonEntropyBits(b *testing.B) { + b.ReportAllocs() + // (predictable, low entropy distibution) + b.Run("zeroes-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + + // (predictable, high entropy distibution) + b.Run("predictable-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + for i := range testData { + testData[i] = byte(float64(i) / float64(len(testData)) * 256) + } + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-500b", func(b *testing.B) { + var testData = make([]byte, 500) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-50k", func(b *testing.B) { + var testData = make([]byte, 50000) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + + // (not predictable, high entropy distibution) + b.Run("random-500k", func(b *testing.B) { + var testData = make([]byte, 500000) + rand.Read(testData) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + + // (not predictable, medium entropy distibution) + b.Run("base-32-5k", func(b *testing.B) { + var testData = make([]byte, 5000) + rand.Read(testData) + s := base32.StdEncoding.EncodeToString(testData) + testData = []byte(s) + testData = testData[:5000] + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) + // (medium predictable, medium entropy distibution) + b.Run("text", func(b *testing.B) { + var testData = []byte(`If compression is done per-chunk, care should be taken that it doesn't leave restic backups open to watermarking/fingerprinting attacks. +This is essentially the same problem we discussed related to fingerprinting the CDC deduplication process: +With "naive" CDC, a "known plaintext" file can be verified to exist within the backup if the size of individual blocks can be observed by an attacker, by using CDC on the file in parallel and comparing the resulting amount of chunks and individual chunk lengths. +As discussed earlier, this can be somewhat mitigated by salting the CDC algorithm with a secret value, as done in attic. +With salted CDC, I assume compression would happen on each individual chunk, after splitting the problematic file into chunks. Restic chunks are in the range of 512 KB to 8 MB (but not evenly distributed - right?). +Attacker knows that the CDC algorithm uses a secret salt, so the attacker generates a range of chunks consisting of the first 512 KB to 8 MB of the file, one for each valid chunk length. The attacker is also able to determine the lengths of compressed chunks. +The attacker then compresses that chunk using the compression algorithm. +The attacker compares the lengths of the resulting chunks to the first chunk in the restic backup sets. +IF a matching block length is found, the attacker repeats the exercise with the next chunk, and the next chunk, and the next chunk, ... and the next chunk. +It is my belief that with sufficiently large files, and considering the fact that the CDC algorithm is "biased" (in lack of better of words) towards generating blocks of about 1 MB, this would be sufficient to ascertain whether or not a certain large file exists in the backup. +AS always, a paranoid and highly unscientific stream of consciousness. +Thoughts?`) + testData = append(testData, testData...) + testData = append(testData, testData...) + b.SetBytes(int64(len(testData))) + b.ResetTimer() + for i := 0; i < b.N; i++ { + ShannonEntropyBits(testData) + } + b.Log(ShannonEntropyBits(testData)) + }) +} + +func BenchmarkCompressAllocations(b *testing.B) { + payload := []byte(strings.Repeat("Tiny payload", 20)) + for j := -2; j <= 9; j++ { + b.Run("level("+strconv.Itoa(j)+")", func(b *testing.B) { + b.Run("flate", func(b *testing.B) { + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + w, err := flate.NewWriter(ioutil.Discard, j) + if err != nil { + b.Fatal(err) + } + w.Write(payload) + w.Close() + } + }) + b.Run("gzip", func(b *testing.B) { + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + w, err := gzip.NewWriterLevel(ioutil.Discard, j) + if err != nil { + b.Fatal(err) + } + w.Write(payload) + w.Close() + } + }) + }) + } +} + +func BenchmarkCompressAllocationsSingle(b *testing.B) { + payload := []byte(strings.Repeat("Tiny payload", 20)) + const level = 2 + b.Run("flate", func(b *testing.B) { + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + w, err := flate.NewWriter(ioutil.Discard, level) + if err != nil { + b.Fatal(err) + } + w.Write(payload) + w.Close() + } + }) + b.Run("gzip", func(b *testing.B) { + b.ReportAllocs() + + for i := 0; i < b.N; i++ { + w, err := gzip.NewWriterLevel(ioutil.Discard, level) + if err != nil { + b.Fatal(err) + } + w.Write(payload) + w.Close() + } + }) +} diff --git a/github.com/klauspost/compress/flate/deflate.go b/github.com/klauspost/compress/flate/deflate.go new file mode 100644 index 0000000..25dbe3e --- /dev/null +++ b/github.com/klauspost/compress/flate/deflate.go @@ -0,0 +1,821 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright (c) 2015 Klaus Post +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "fmt" + "io" + "math" +) + +const ( + NoCompression = 0 + BestSpeed = 1 + BestCompression = 9 + DefaultCompression = -1 + + // HuffmanOnly disables Lempel-Ziv match searching and only performs Huffman + // entropy encoding. This mode is useful in compressing data that has + // already been compressed with an LZ style algorithm (e.g. Snappy or LZ4) + // that lacks an entropy encoder. Compression gains are achieved when + // certain bytes in the input stream occur more frequently than others. + // + // Note that HuffmanOnly produces a compressed output that is + // RFC 1951 compliant. That is, any valid DEFLATE decompressor will + // continue to be able to decompress this output. + HuffmanOnly = -2 + ConstantCompression = HuffmanOnly // compatibility alias. + + logWindowSize = 15 + windowSize = 1 << logWindowSize + windowMask = windowSize - 1 + logMaxOffsetSize = 15 // Standard DEFLATE + minMatchLength = 4 // The smallest match that the compressor looks for + maxMatchLength = 258 // The longest match for the compressor + minOffsetSize = 1 // The shortest offset that makes any sense + + // The maximum number of tokens we put into a single flat block, just too + // stop things from getting too large. + maxFlateBlockTokens = 1 << 14 + maxStoreBlockSize = 65535 + hashBits = 17 // After 17 performance degrades + hashSize = 1 << hashBits + hashMask = (1 << hashBits) - 1 + hashShift = (hashBits + minMatchLength - 1) / minMatchLength + maxHashOffset = 1 << 24 + + skipNever = math.MaxInt32 + + debugDeflate = false +) + +type compressionLevel struct { + good, lazy, nice, chain, fastSkipHashing, level int +} + +// Compression levels have been rebalanced from zlib deflate defaults +// to give a bigger spread in speed and compression. +// See https://blog.klauspost.com/rebalancing-deflate-compression-levels/ +var levels = []compressionLevel{ + {}, // 0 + // 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}, + {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}, + {10, 16, 24, 64, skipNever, 8}, + {32, 258, 258, 4096, skipNever, 9}, +} + +// advancedState contains state for the advanced levels, with bigger hash tables, etc. +type advancedState struct { + // deflate state + length int + offset int + maxInsertIndex int + + // Input hash chains + // hashHead[hashValue] contains the largest inputIndex with the specified hash value + // If hashHead[hashValue] is within the current window, then + // hashPrev[hashHead[hashValue] & windowMask] contains the previous index + // with the same hash value. + chainHead int + hashHead [hashSize]uint32 + hashPrev [windowSize]uint32 + hashOffset int + + // 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 { + compressionLevel + + w *huffmanBitWriter + + // compression algorithm + fill func(*compressor, []byte) int // copy data to window + step func(*compressor) // process window + + 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 { + s := d.state + if s.index >= 2*windowSize-(minMatchLength+maxMatchLength) { + // shift the window by windowSize + copy(d.window[:], d.window[windowSize:2*windowSize]) + s.index -= windowSize + d.windowEnd -= windowSize + if d.blockStart >= windowSize { + d.blockStart -= windowSize + } else { + d.blockStart = math.MaxInt32 + } + s.hashOffset += windowSize + if s.hashOffset > maxHashOffset { + delta := s.hashOffset - 1 + s.hashOffset -= delta + s.chainHead -= delta + // Iterate over slices instead of arrays to avoid copying + // the entire table onto the stack (Issue #18625). + for i, v := range s.hashPrev[:] { + if int(v) > delta { + s.hashPrev[i] = uint32(int(v) - delta) + } else { + s.hashPrev[i] = 0 + } + } + for i, v := range s.hashHead[:] { + if int(v) > delta { + s.hashHead[i] = uint32(int(v) - delta) + } else { + s.hashHead[i] = 0 + } + } + } + } + n := copy(d.window[d.windowEnd:], b) + d.windowEnd += n + return n +} + +func (d *compressor) writeBlock(tok *tokens, index int, eof bool) error { + if index > 0 || eof { + var window []byte + if d.blockStart <= index { + window = d.window[d.blockStart:index] + } + d.blockStart = index + d.w.writeBlock(tok, eof, window) + return d.w.err + } + return nil +} + +// writeBlockSkip writes the current block and uses the number of tokens +// to determine if the block should be stored on no matches, or +// only huffman encoded. +func (d *compressor) writeBlockSkip(tok *tokens, index int, eof bool) error { + if index > 0 || eof { + if d.blockStart <= index { + window := d.window[d.blockStart:index] + // If we removed less than a 64th of all literals + // we huffman compress the block. + if int(tok.n) > len(window)-int(tok.n>>6) { + d.w.writeBlockHuff(eof, window, d.sync) + } else { + // Write a dynamic huffman block. + d.w.writeBlockDynamic(tok, eof, window, d.sync) + } + } else { + d.w.writeBlock(tok, eof, nil) + } + d.blockStart = index + return d.w.err + } + return nil +} + +// fillWindow will fill the current window with the supplied +// dictionary and calculate all hashes. +// 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 or huffman mode. + if d.level <= 0 { + return + } + if d.fast != nil { + // encode the last data, but discard the result + if len(b) > maxMatchOffset { + b = b[len(b)-maxMatchOffset:] + } + d.fast.Encode(&d.tokens, b) + d.tokens.Reset() + return + } + s := d.state + // If we are given too much, cut it. + if len(b) > windowSize { + b = b[len(b)-windowSize:] + } + // Add all to window. + n := copy(d.window[d.windowEnd:], b) + + // Calculate 256 hashes at the time (more L1 cache hits) + loops := (n + 256 - minMatchLength) / 256 + for j := 0; j < loops; j++ { + startindex := j * 256 + end := startindex + 256 + minMatchLength - 1 + if end > n { + end = n + } + tocheck := d.window[startindex:end] + dstSize := len(tocheck) - minMatchLength + 1 + + if dstSize <= 0 { + continue + } + + dst := s.hashMatch[:dstSize] + bulkHash4(tocheck, dst) + var newH uint32 + for i, val := range dst { + di := i + startindex + newH = val & hashMask + // Get previous value with the same hash. + // Our chain should point to the previous value. + s.hashPrev[di&windowMask] = s.hashHead[newH] + // Set the head of the hash chain to us. + s.hashHead[newH] = uint32(di + s.hashOffset) + } + s.hash = newH + } + // Update window information. + d.windowEnd += n + s.index = n +} + +// Try to find a match starting at index whose length is greater than prevSize. +// We only look at chainCount possibilities before giving up. +// pos = s.index, prevHead = s.chainHead-s.hashOffset, prevLength=minMatchLength-1, lookahead +func (d *compressor) findMatch(pos int, prevHead int, prevLength int, lookahead int) (length, offset int, ok bool) { + minMatchLook := maxMatchLength + if lookahead < minMatchLook { + minMatchLook = lookahead + } + + win := d.window[0 : pos+minMatchLook] + + // We quit when we get a match that's at least nice long + nice := len(win) - pos + if d.nice < nice { + nice = d.nice + } + + // If we've got a match that's good enough, only look in 1/4 the chain. + tries := d.chain + length = prevLength + if length >= d.good { + tries >>= 2 + } + + wEnd := win[pos+length] + wPos := win[pos:] + minIndex := pos - windowSize + + for i := prevHead; tries > 0; tries-- { + if wEnd == win[i+length] { + n := matchLen(win[i:i+minMatchLook], wPos) + + if n > length && (n > minMatchLength || pos-i <= 4096) { + length = n + offset = pos - i + ok = true + if n >= nice { + // The match is good enough that we don't try to find a better one. + break + } + wEnd = win[pos+n] + } + } + if i == minIndex { + // hashPrev[i & windowMask] has already been overwritten, so stop now. + break + } + i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset + if i < minIndex || i < 0 { + break + } + } + return +} + +func (d *compressor) writeStoredBlock(buf []byte) error { + if d.w.writeStoredHeader(len(buf), false); d.w.err != nil { + return d.w.err + } + d.w.writeBytes(buf) + return d.w.err +} + +// hash4 returns a hash representation of the first 4 bytes +// of the supplied slice. +// The caller must ensure that len(b) >= 4. +func hash4(b []byte) uint32 { + b = b[:4] + return hash4u(uint32(b[3])|uint32(b[2])<<8|uint32(b[1])<<16|uint32(b[0])<<24, hashBits) +} + +// bulkHash4 will compute hashes using the same +// algorithm as hash4 +func bulkHash4(b []byte, dst []uint32) { + if len(b) < 4 { + return + } + hb := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24 + dst[0] = hash4u(hb, hashBits) + end := len(b) - 4 + 1 + for i := 1; i < end; i++ { + hb = (hb << 8) | uint32(b[i+3]) + dst[i] = hash4u(hb, hashBits) + } +} + +func (d *compressor) initDeflate() { + d.window = make([]byte, 2*windowSize) + d.byteAvailable = false + d.err = nil + if d.state == nil { + return + } + s := d.state + s.index = 0 + s.hashOffset = 1 + s.length = minMatchLength - 1 + s.offset = 0 + s.hash = 0 + s.chainHead = -1 +} + +// deflateLazy is the same as deflate, but with d.fastSkipHashing == skipNever, +// meaning it always has lazy matching on. +func (d *compressor) deflateLazy() { + s := d.state + // Sanity enables additional runtime tests. + // It's intended to be used during development + // to supplement the currently ad-hoc unit tests. + const sanity = debugDeflate + + if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync { + return + } + + s.maxInsertIndex = d.windowEnd - (minMatchLength - 1) + if s.index < s.maxInsertIndex { + s.hash = hash4(d.window[s.index : s.index+minMatchLength]) + } + + for { + if sanity && s.index > d.windowEnd { + panic("index > windowEnd") + } + lookahead := d.windowEnd - s.index + if lookahead < minMatchLength+maxMatchLength { + if !d.sync { + return + } + if sanity && s.index > d.windowEnd { + panic("index > windowEnd") + } + if lookahead == 0 { + // Flush current output block if any. + if d.byteAvailable { + // There is still one pending token that needs to be flushed + d.tokens.AddLiteral(d.window[s.index-1]) + d.byteAvailable = false + } + if d.tokens.n > 0 { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + return + } + } + if s.index < s.maxInsertIndex { + // Update the hash + s.hash = hash4(d.window[s.index : s.index+minMatchLength]) + ch := s.hashHead[s.hash&hashMask] + s.chainHead = int(ch) + s.hashPrev[s.index&windowMask] = ch + s.hashHead[s.hash&hashMask] = uint32(s.index + s.hashOffset) + } + prevLength := s.length + prevOffset := s.offset + s.length = minMatchLength - 1 + s.offset = 0 + minIndex := s.index - windowSize + if minIndex < 0 { + minIndex = 0 + } + + if s.chainHead-s.hashOffset >= minIndex && lookahead > prevLength && prevLength < d.lazy { + if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok { + s.length = newLength + s.offset = newOffset + } + } + if prevLength >= minMatchLength && s.length <= prevLength { + // There was a match at the previous step, and the current match is + // not better. Output the previous match. + d.tokens.AddMatch(uint32(prevLength-3), uint32(prevOffset-minOffsetSize)) + + // Insert in the hash table all strings up to the end of the match. + // index and index-1 are already inserted. If there is not enough + // lookahead, the last two strings are not inserted into the hash + // table. + var newIndex int + newIndex = s.index + prevLength - 1 + // Calculate missing hashes + end := newIndex + if end > s.maxInsertIndex { + end = s.maxInsertIndex + } + end += minMatchLength - 1 + startindex := s.index + 1 + if startindex > s.maxInsertIndex { + startindex = s.maxInsertIndex + } + tocheck := d.window[startindex:end] + dstSize := len(tocheck) - minMatchLength + 1 + if dstSize > 0 { + dst := s.hashMatch[:dstSize] + bulkHash4(tocheck, dst) + var newH uint32 + for i, val := range dst { + di := i + startindex + newH = val & hashMask + // Get previous value with the same hash. + // Our chain should point to the previous value. + s.hashPrev[di&windowMask] = s.hashHead[newH] + // Set the head of the hash chain to us. + s.hashHead[newH] = uint32(di + s.hashOffset) + } + s.hash = newH + } + + s.index = newIndex + d.byteAvailable = false + s.length = minMatchLength - 1 + if d.tokens.n == maxFlateBlockTokens { + // The block includes the current character + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + } else { + // Reset, if we got a match this run. + if s.length >= minMatchLength { + s.ii = 0 + } + // We have a byte waiting. Emit it. + if d.byteAvailable { + s.ii++ + d.tokens.AddLiteral(d.window[s.index-1]) + if d.tokens.n == maxFlateBlockTokens { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + s.index++ + + // If we have a long run of no matches, skip additional bytes + // Resets when s.ii overflows after 64KB. + if s.ii > 31 { + n := int(s.ii >> 5) + for j := 0; j < n; j++ { + if s.index >= d.windowEnd-1 { + break + } + + d.tokens.AddLiteral(d.window[s.index-1]) + if d.tokens.n == maxFlateBlockTokens { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + s.index++ + } + // Flush last byte + d.tokens.AddLiteral(d.window[s.index-1]) + d.byteAvailable = false + // s.length = minMatchLength - 1 // not needed, since s.ii is reset above, so it should never be > minMatchLength + if d.tokens.n == maxFlateBlockTokens { + if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil { + return + } + d.tokens.Reset() + } + } + } else { + s.index++ + d.byteAvailable = true + } + } + } +} + +func (d *compressor) store() { + if d.windowEnd > 0 && (d.windowEnd == maxStoreBlockSize || d.sync) { + d.err = d.writeStoredBlock(d.window[:d.windowEnd]) + d.windowEnd = 0 + } +} + +// fillWindow will fill the buffer with data for huffman-only compression. +// The number of bytes copied is returned. +func (d *compressor) fillBlock(b []byte) int { + n := copy(d.window[d.windowEnd:], b) + d.windowEnd += n + return n +} + +// storeHuff will compress and store the currently added data, +// if enough has been accumulated or we at the end of the stream. +// Any error that occurred will be in d.err +func (d *compressor) storeHuff() { + if d.windowEnd < len(d.window) && !d.sync || d.windowEnd == 0 { + return + } + d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync) + d.err = d.w.err + d.windowEnd = 0 +} + +// storeFast will compress and store the currently added data, +// if enough has been accumulated or we at the end of the stream. +// Any error that occurred will be in d.err +func (d *compressor) storeFast() { + // We only compress if we have maxStoreBlockSize. + if d.windowEnd < len(d.window) { + if !d.sync { + return + } + // Handle extremely small sizes. + if d.windowEnd < 128 { + if d.windowEnd == 0 { + return + } + if d.windowEnd <= 32 { + d.err = d.writeStoredBlock(d.window[:d.windowEnd]) + } else { + d.w.writeBlockHuff(false, d.window[:d.windowEnd], true) + d.err = d.w.err + } + d.tokens.Reset() + d.windowEnd = 0 + d.fast.Reset() + return + } + } + + d.fast.Encode(&d.tokens, d.window[:d.windowEnd]) + // If we made zero matches, store the block as is. + if d.tokens.n == 0 { + d.err = d.writeStoredBlock(d.window[:d.windowEnd]) + // If we removed less than 1/16th, huffman compress the block. + } else if int(d.tokens.n) > d.windowEnd-(d.windowEnd>>4) { + d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync) + d.err = d.w.err + } else { + d.w.writeBlockDynamic(&d.tokens, false, d.window[:d.windowEnd], d.sync) + d.err = d.w.err + } + d.tokens.Reset() + d.windowEnd = 0 +} + +// write will add input byte to the stream. +// Unless an error occurs all bytes will be consumed. +func (d *compressor) write(b []byte) (n int, err error) { + if d.err != nil { + return 0, d.err + } + n = len(b) + for len(b) > 0 { + d.step(d) + b = b[d.fill(d, b):] + if d.err != nil { + return 0, d.err + } + } + return n, d.err +} + +func (d *compressor) syncFlush() error { + d.sync = true + if d.err != nil { + return d.err + } + d.step(d) + if d.err == nil { + d.w.writeStoredHeader(0, false) + d.w.flush() + d.err = d.w.err + } + d.sync = false + return d.err +} + +func (d *compressor) init(w io.Writer, level int) (err error) { + d.w = newHuffmanBitWriter(w) + + switch { + case level == NoCompression: + d.window = make([]byte, maxStoreBlockSize) + d.fill = (*compressor).fillBlock + d.step = (*compressor).store + case level == ConstantCompression: + d.w.logNewTablePenalty = 4 + d.window = make([]byte, maxStoreBlockSize) + d.fill = (*compressor).fillBlock + d.step = (*compressor).storeHuff + case level == DefaultCompression: + level = 5 + fallthrough + case level >= 1 && level <= 6: + 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.logNewTablePenalty = 10 + d.state = &advancedState{} + d.compressionLevel = levels[level] + d.initDeflate() + d.fill = (*compressor).fillDeflate + d.step = (*compressor).deflateLazy + default: + return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level) + } + d.level = level + return nil +} + +// reset the state of the compressor. +func (d *compressor) reset(w io.Writer) { + d.w.reset(w) + d.sync = false + d.err = nil + // We only need to reset a few things for Snappy. + if d.fast != nil { + d.fast.Reset() + d.windowEnd = 0 + d.tokens.Reset() + return + } + switch d.compressionLevel.chain { + case 0: + // level was NoCompression or ConstantCompresssion. + d.windowEnd = 0 + default: + s := d.state + s.chainHead = -1 + for i := range s.hashHead { + s.hashHead[i] = 0 + } + for i := range s.hashPrev { + s.hashPrev[i] = 0 + } + s.hashOffset = 1 + s.index, d.windowEnd = 0, 0 + d.blockStart, d.byteAvailable = 0, false + d.tokens.Reset() + s.length = minMatchLength - 1 + s.offset = 0 + s.hash = 0 + s.ii = 0 + s.maxInsertIndex = 0 + } +} + +func (d *compressor) close() error { + if d.err != nil { + return d.err + } + d.sync = true + d.step(d) + if d.err != nil { + return d.err + } + if d.w.writeStoredHeader(0, true); d.w.err != nil { + return d.w.err + } + d.w.flush() + d.w.reset(nil) + return d.w.err +} + +// NewWriter returns a new Writer compressing data at the given level. +// Following zlib, levels range from 1 (BestSpeed) to 9 (BestCompression); +// higher levels typically run slower but compress more. +// Level 0 (NoCompression) does not attempt any compression; it only adds the +// necessary DEFLATE framing. +// Level -1 (DefaultCompression) uses the default compression level. +// Level -2 (ConstantCompression) will use Huffman compression only, giving +// a very fast compression for all types of input, but sacrificing considerable +// compression efficiency. +// +// If level is in the range [-2, 9] then the error returned will be nil. +// Otherwise the error returned will be non-nil. +func NewWriter(w io.Writer, level int) (*Writer, error) { + var dw Writer + if err := dw.d.init(w, level); err != nil { + return nil, err + } + return &dw, nil +} + +// NewWriterDict is like NewWriter but initializes the new +// Writer with a preset dictionary. The returned Writer behaves +// as if the dictionary had been written to it without producing +// any compressed output. The compressed data written to w +// can only be decompressed by a Reader initialized with the +// same dictionary. +func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) { + zw, err := NewWriter(w, level) + if err != nil { + return nil, err + } + zw.d.fillWindow(dict) + zw.dict = append(zw.dict, dict...) // duplicate dictionary for Reset method. + return zw, err +} + +// A Writer takes data written to it and writes the compressed +// form of that data to an underlying writer (see NewWriter). +type Writer struct { + d compressor + dict []byte +} + +// Write writes data to w, which will eventually write the +// compressed form of data to its underlying writer. +func (w *Writer) Write(data []byte) (n int, err error) { + return w.d.write(data) +} + +// Flush flushes any pending data to the underlying writer. +// It is useful mainly in compressed network protocols, to ensure that +// a remote reader has enough data to reconstruct a packet. +// Flush does not return until the data has been written. +// Calling Flush when there is no pending data still causes the Writer +// to emit a sync marker of at least 4 bytes. +// If the underlying writer returns an error, Flush returns that error. +// +// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH. +func (w *Writer) Flush() error { + // For more about flushing: + // http://www.bolet.org/~pornin/deflate-flush.html + return w.d.syncFlush() +} + +// Close flushes and closes the writer. +func (w *Writer) Close() error { + return w.d.close() +} + +// Reset discards the writer's state and makes it equivalent to +// the result of NewWriter or NewWriterDict called with dst +// and w's level and dictionary. +func (w *Writer) Reset(dst io.Writer) { + if len(w.dict) > 0 { + // w was created with NewWriterDict + w.d.reset(dst) + if dst != nil { + w.d.fillWindow(w.dict) + } + } else { + // w was created with NewWriter + w.d.reset(dst) + } +} + +// ResetDict discards the writer's state and makes it equivalent to +// the result of NewWriter or NewWriterDict called with dst +// and w's level, but sets a specific dictionary. +func (w *Writer) ResetDict(dst io.Writer, dict []byte) { + w.dict = dict + w.d.reset(dst) + w.d.fillWindow(w.dict) +} diff --git a/github.com/klauspost/compress/flate/deflate_test.go b/github.com/klauspost/compress/flate/deflate_test.go new file mode 100644 index 0000000..079040e --- /dev/null +++ b/github.com/klauspost/compress/flate/deflate_test.go @@ -0,0 +1,658 @@ +// Copyright 2009 The Go Authors. All rights reserved. +// Copyright (c) 2015 Klaus Post +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "bytes" + "fmt" + "io" + "io/ioutil" + "reflect" + "strings" + "sync" + "testing" +) + +type deflateTest struct { + in []byte + level int + out []byte +} + +type deflateInflateTest struct { + in []byte +} + +type reverseBitsTest struct { + in uint16 + bitCount uint8 + out uint16 +} + +var deflateTests = []*deflateTest{ + {[]byte{}, 0, []byte{0x3, 0x0}}, + {[]byte{0x11}, BestCompression, []byte{0x12, 0x4, 0xc, 0x0}}, + {[]byte{0x11}, BestCompression, []byte{0x12, 0x4, 0xc, 0x0}}, + {[]byte{0x11}, BestCompression, []byte{0x12, 0x4, 0xc, 0x0}}, + + {[]byte{0x11}, 0, []byte{0x0, 0x1, 0x0, 0xfe, 0xff, 0x11, 0x3, 0x0}}, + {[]byte{0x11, 0x12}, 0, []byte{0x0, 0x2, 0x0, 0xfd, 0xff, 0x11, 0x12, 0x3, 0x0}}, + {[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}, 0, + []byte{0x0, 0x8, 0x0, 0xf7, 0xff, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x3, 0x0}, + }, + {[]byte{}, 1, []byte{0x3, 0x0}}, + {[]byte{0x11}, BestCompression, []byte{0x12, 0x4, 0xc, 0x0}}, + {[]byte{0x11, 0x12}, BestCompression, []byte{0x12, 0x14, 0x2, 0xc, 0x0}}, + {[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}, BestCompression, []byte{0x12, 0x84, 0x2, 0xc0, 0x0}}, + {[]byte{}, 9, []byte{0x3, 0x0}}, + {[]byte{0x11}, 9, []byte{0x12, 0x4, 0xc, 0x0}}, + {[]byte{0x11, 0x12}, 9, []byte{0x12, 0x14, 0x2, 0xc, 0x0}}, + {[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}, 9, []byte{0x12, 0x84, 0x2, 0xc0, 0x0}}, +} + +var deflateInflateTests = []*deflateInflateTest{ + {[]byte{}}, + {[]byte{0x11}}, + {[]byte{0x11, 0x12}}, + {[]byte{0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11, 0x11}}, + {[]byte{0x11, 0x10, 0x13, 0x41, 0x21, 0x21, 0x41, 0x13, 0x87, 0x78, 0x13}}, + {largeDataChunk()}, +} + +var reverseBitsTests = []*reverseBitsTest{ + {1, 1, 1}, + {1, 2, 2}, + {1, 3, 4}, + {1, 4, 8}, + {1, 5, 16}, + {17, 5, 17}, + {257, 9, 257}, + {29, 5, 23}, +} + +func largeDataChunk() []byte { + result := make([]byte, 100000) + for i := range result { + result[i] = byte(i * i & 0xFF) + } + return result +} + +func TestBulkHash4(t *testing.T) { + for _, x := range deflateTests { + y := x.out + if len(y) >= minMatchLength { + y = append(y, y...) + for j := 4; j < len(y); j++ { + y := y[:j] + dst := make([]uint32, len(y)-minMatchLength+1) + for i := range dst { + dst[i] = uint32(i + 100) + } + bulkHash4(y, dst) + for i, val := range dst { + got := val + expect := hash4(y[i:]) + if got != expect && got == uint32(i)+100 { + t.Errorf("Len:%d Index:%d, expected 0x%08x but not modified", len(y), i, expect) + } else if got != expect { + t.Errorf("Len:%d Index:%d, got 0x%08x expected:0x%08x", len(y), i, got, expect) + } else { + //t.Logf("Len:%d Index:%d OK (0x%08x)", len(y), i, got) + } + } + } + } + } +} + +func TestDeflate(t *testing.T) { + for i, h := range deflateTests { + var buf bytes.Buffer + w, err := NewWriter(&buf, h.level) + if err != nil { + t.Errorf("NewWriter: %v", err) + continue + } + w.Write(h.in) + w.Close() + if !bytes.Equal(buf.Bytes(), h.out) { + t.Errorf("%d: Deflate(%d, %x) = \n%#v, want \n%#v", i, h.level, h.in, buf.Bytes(), h.out) + } + } +} + +// A sparseReader returns a stream consisting of 0s followed by 1<<16 1s. +// This tests missing hash references in a very large input. +type sparseReader struct { + l int64 + cur int64 +} + +func (r *sparseReader) Read(b []byte) (n int, err error) { + if r.cur >= r.l { + return 0, io.EOF + } + n = len(b) + cur := r.cur + int64(n) + if cur > r.l { + n -= int(cur - r.l) + cur = r.l + } + for i := range b[0:n] { + if r.cur+int64(i) >= r.l-1<<16 { + b[i] = 1 + } else { + b[i] = 0 + } + } + r.cur = cur + return +} + +func TestVeryLongSparseChunk(t *testing.T) { + if testing.Short() { + t.Skip("skipping sparse chunk during short test") + } + w, err := NewWriter(ioutil.Discard, 1) + if err != nil { + t.Errorf("NewWriter: %v", err) + return + } + if _, err = io.Copy(w, &sparseReader{l: 23e8}); err != nil { + t.Errorf("Compress failed: %v", err) + return + } +} + +type syncBuffer struct { + buf bytes.Buffer + mu sync.RWMutex + closed bool + ready chan bool +} + +func newSyncBuffer() *syncBuffer { + return &syncBuffer{ready: make(chan bool, 1)} +} + +func (b *syncBuffer) Read(p []byte) (n int, err error) { + for { + b.mu.RLock() + n, err = b.buf.Read(p) + b.mu.RUnlock() + if n > 0 || b.closed { + return + } + <-b.ready + } +} + +func (b *syncBuffer) signal() { + select { + case b.ready <- true: + default: + } +} + +func (b *syncBuffer) Write(p []byte) (n int, err error) { + n, err = b.buf.Write(p) + b.signal() + return +} + +func (b *syncBuffer) WriteMode() { + b.mu.Lock() +} + +func (b *syncBuffer) ReadMode() { + b.mu.Unlock() + b.signal() +} + +func (b *syncBuffer) Close() error { + b.closed = true + b.signal() + return nil +} + +func testSync(t *testing.T, level int, input []byte, name string) { + if len(input) == 0 { + return + } + + t.Logf("--testSync %d, %d, %s", level, len(input), name) + buf := newSyncBuffer() + buf1 := new(bytes.Buffer) + buf.WriteMode() + w, err := NewWriter(io.MultiWriter(buf, buf1), level) + if err != nil { + t.Errorf("NewWriter: %v", err) + return + } + r := NewReader(buf) + + // Write half the input and read back. + for i := 0; i < 2; i++ { + var lo, hi int + if i == 0 { + lo, hi = 0, (len(input)+1)/2 + } else { + lo, hi = (len(input)+1)/2, len(input) + } + t.Logf("#%d: write %d-%d", i, lo, hi) + if _, err := w.Write(input[lo:hi]); err != nil { + t.Errorf("testSync: write: %v", err) + return + } + if i == 0 { + if err := w.Flush(); err != nil { + t.Errorf("testSync: flush: %v", err) + return + } + } else { + if err := w.Close(); err != nil { + t.Errorf("testSync: close: %v", err) + } + } + buf.ReadMode() + out := make([]byte, hi-lo+1) + m, err := io.ReadAtLeast(r, out, hi-lo) + t.Logf("#%d: read %d", i, m) + if m != hi-lo || err != nil { + t.Errorf("testSync/%d (%d, %d, %s): read %d: %d, %v (%d left)", i, level, len(input), name, hi-lo, m, err, buf.buf.Len()) + return + } + if !bytes.Equal(input[lo:hi], out[:hi-lo]) { + t.Errorf("testSync/%d: read wrong bytes: %x vs %x", i, input[lo:hi], out[:hi-lo]) + return + } + // This test originally checked that after reading + // the first half of the input, there was nothing left + // in the read buffer (buf.buf.Len() != 0) but that is + // not necessarily the case: the write Flush may emit + // some extra framing bits that are not necessary + // to process to obtain the first half of the uncompressed + // data. The test ran correctly most of the time, because + // the background goroutine had usually read even + // those extra bits by now, but it's not a useful thing to + // check. + buf.WriteMode() + } + buf.ReadMode() + out := make([]byte, 10) + if n, err := r.Read(out); n > 0 || err != io.EOF { + t.Errorf("testSync (%d, %d, %s): final Read: %d, %v (hex: %x)", level, len(input), name, n, err, out[0:n]) + } + if buf.buf.Len() != 0 { + t.Errorf("testSync (%d, %d, %s): extra data at end", level, len(input), name) + } + r.Close() + + // stream should work for ordinary reader too + r = NewReader(buf1) + out, err = ioutil.ReadAll(r) + if err != nil { + t.Errorf("testSync: read: %s", err) + return + } + r.Close() + if !bytes.Equal(input, out) { + t.Errorf("testSync: decompress(compress(data)) != data: level=%d input=%s", level, name) + } +} + +func testToFromWithLevelAndLimit(t *testing.T, level int, input []byte, name string, limit int) { + var buffer bytes.Buffer + w, err := NewWriter(&buffer, level) + if err != nil { + t.Errorf("NewWriter: %v", err) + return + } + w.Write(input) + w.Close() + if limit > 0 && buffer.Len() > limit { + t.Errorf("level: %d, len(compress(data)) = %d > limit = %d", level, buffer.Len(), limit) + return + } + if limit > 0 { + t.Logf("level: %d - Size:%.2f%%, %d b\n", level, float64(buffer.Len()*100)/float64(limit), buffer.Len()) + } + r := NewReader(&buffer) + out, err := ioutil.ReadAll(r) + if err != nil { + t.Errorf("read: %s", err) + return + } + r.Close() + if !bytes.Equal(input, out) { + t.Errorf("decompress(compress(data)) != data: level=%d input=%s", level, name) + return + } + testSync(t, level, input, name) +} + +func testToFromWithLimit(t *testing.T, input []byte, name string, limit [11]int) { + for i := 0; i < 10; i++ { + testToFromWithLevelAndLimit(t, i, input, name, limit[i]) + } + testToFromWithLevelAndLimit(t, -2, input, name, limit[10]) +} + +func TestDeflateInflate(t *testing.T) { + for i, h := range deflateInflateTests { + testToFromWithLimit(t, h.in, fmt.Sprintf("#%d", i), [11]int{}) + } +} + +func TestReverseBits(t *testing.T) { + for _, h := range reverseBitsTests { + if v := reverseBits(h.in, h.bitCount); v != h.out { + t.Errorf("reverseBits(%v,%v) = %v, want %v", + h.in, h.bitCount, v, h.out) + } + } +} + +type deflateInflateStringTest struct { + filename string + label string + limit [11]int // Number 11 is ConstantCompression +} + +var deflateInflateStringTests = []deflateInflateStringTest{ + { + "../testdata/e.txt", + "2.718281828...", + [...]int{100018, 67900, 50960, 51150, 50930, 50790, 50790, 50790, 50790, 50790, 43683 + 100}, + }, + { + "../testdata/Mark.Twain-Tom.Sawyer.txt", + "Mark.Twain-Tom.Sawyer", + [...]int{387999, 185000, 182361, 179974, 174124, 168819, 162936, 160506, 160295, 160295, 233460 + 100}, + }, +} + +func TestDeflateInflateString(t *testing.T) { + for _, test := range deflateInflateStringTests { + gold, err := ioutil.ReadFile(test.filename) + if err != nil { + t.Error(err) + } + // Remove returns that may be present on Windows + neutral := strings.Map(func(r rune) rune { + if r != '\r' { + return r + } + return -1 + }, string(gold)) + + testToFromWithLimit(t, []byte(neutral), test.label, test.limit) + + if testing.Short() { + break + } + } +} + +func TestReaderDict(t *testing.T) { + const ( + dict = "hello world" + text = "hello again world" + ) + var b bytes.Buffer + w, err := NewWriter(&b, 5) + if err != nil { + t.Fatalf("NewWriter: %v", err) + } + w.Write([]byte(dict)) + w.Flush() + b.Reset() + w.Write([]byte(text)) + w.Close() + + r := NewReaderDict(&b, []byte(dict)) + data, err := ioutil.ReadAll(r) + if err != nil { + t.Fatal(err) + } + if string(data) != "hello again world" { + t.Fatalf("read returned %q want %q", string(data), text) + } +} + +func TestWriterDict(t *testing.T) { + const ( + dict = "hello world Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua." + text = "hello world Lorem ipsum dolor sit amet" + ) + // This test is sensitive to algorithm changes that skip + // data in favour of speed. Higher levels are less prone to this + // so we test level 4-9. + for l := 4; l < 9; l++ { + var b bytes.Buffer + w, err := NewWriter(&b, l) + if err != nil { + t.Fatalf("level %d, NewWriter: %v", l, err) + } + w.Write([]byte(dict)) + w.Flush() + b.Reset() + w.Write([]byte(text)) + w.Close() + + var b1 bytes.Buffer + w, _ = NewWriterDict(&b1, l, []byte(dict)) + w.Write([]byte(text)) + w.Close() + + if !bytes.Equal(b1.Bytes(), b.Bytes()) { + t.Errorf("level %d, writer wrote\n%v\n want\n%v", l, b1.Bytes(), b.Bytes()) + } + } +} + +// See http://code.google.com/p/go/issues/detail?id=2508 +func TestRegression2508(t *testing.T) { + if testing.Short() { + t.Logf("test disabled with -short") + return + } + w, err := NewWriter(ioutil.Discard, 1) + if err != nil { + t.Fatalf("NewWriter: %v", err) + } + buf := make([]byte, 1024) + for i := 0; i < 131072; i++ { + if _, err := w.Write(buf); err != nil { + t.Fatalf("writer failed: %v", err) + } + } + w.Close() +} + +func TestWriterReset(t *testing.T) { + for level := -2; level <= 9; level++ { + if level == -1 { + level++ + } + if testing.Short() && level > 1 { + break + } + w, err := NewWriter(ioutil.Discard, level) + if err != nil { + t.Fatalf("NewWriter: %v", err) + } + buf := []byte("hello world") + for i := 0; i < 1024; i++ { + w.Write(buf) + } + w.Reset(ioutil.Discard) + + wref, err := NewWriter(ioutil.Discard, level) + if err != nil { + t.Fatalf("NewWriter: %v", err) + } + + // DeepEqual doesn't compare functions. + w.d.fill, wref.d.fill = nil, nil + w.d.step, wref.d.step = nil, nil + w.d.state, wref.d.state = nil, nil + w.d.fast, wref.d.fast = nil, nil + + // hashMatch is always overwritten when used. + if w.d.tokens.n != 0 { + t.Errorf("level %d Writer not reset after Reset. %d tokens were present", level, w.d.tokens.n) + } + // As long as the length is 0, we don't care about the content. + w.d.tokens = wref.d.tokens + + // We don't care if there are values in the window, as long as it is at d.index is 0 + w.d.window = wref.d.window + if !reflect.DeepEqual(w, wref) { + t.Errorf("level %d Writer not reset after Reset", level) + } + } + + for i := HuffmanOnly; i <= BestCompression; i++ { + testResetOutput(t, fmt.Sprint("level-", i), func(w io.Writer) (*Writer, error) { return NewWriter(w, i) }) + } + dict := []byte(strings.Repeat("we are the world - how are you?", 3)) + for i := HuffmanOnly; i <= BestCompression; i++ { + testResetOutput(t, fmt.Sprint("dict-level-", i), func(w io.Writer) (*Writer, error) { return NewWriterDict(w, i, dict) }) + } + for i := HuffmanOnly; i <= BestCompression; i++ { + testResetOutput(t, fmt.Sprint("dict-reset-level-", i), func(w io.Writer) (*Writer, error) { + w2, err := NewWriter(nil, i) + if err != nil { + return w2, err + } + w2.ResetDict(w, dict) + return w2, nil + }) + } +} + +func testResetOutput(t *testing.T, name string, newWriter func(w io.Writer) (*Writer, error)) { + t.Run(name, func(t *testing.T) { + buf := new(bytes.Buffer) + w, err := newWriter(buf) + if err != nil { + t.Fatalf("NewWriter: %v", err) + } + b := []byte("hello world - how are you doing?") + for i := 0; i < 1024; i++ { + w.Write(b) + } + w.Close() + out1 := buf.Bytes() + + buf2 := new(bytes.Buffer) + w.Reset(buf2) + for i := 0; i < 1024; i++ { + w.Write(b) + } + w.Close() + out2 := buf2.Bytes() + + if len(out1) != len(out2) { + t.Errorf("got %d, expected %d bytes", len(out2), len(out1)) + } + if bytes.Compare(out1, out2) != 0 { + mm := 0 + for i, b := range out1[:len(out2)] { + if b != out2[i] { + t.Errorf("mismatch index %d: %02x, expected %02x", i, out2[i], b) + } + mm++ + if mm == 10 { + t.Fatal("Stopping") + } + } + } + t.Logf("got %d bytes", len(out1)) + }) +} + +// TestBestSpeed tests that round-tripping through deflate and then inflate +// recovers the original input. The Write sizes are near the thresholds in the +// compressor.encSpeed method (0, 16, 128), as well as near maxStoreBlockSize +// (65535). +func TestBestSpeed(t *testing.T) { + abc := make([]byte, 128) + for i := range abc { + abc[i] = byte(i) + } + abcabc := bytes.Repeat(abc, 131072/len(abc)) + var want []byte + + testCases := [][]int{ + {65536, 0}, + {65536, 1}, + {65536, 1, 256}, + {65536, 1, 65536}, + {65536, 14}, + {65536, 15}, + {65536, 16}, + {65536, 16, 256}, + {65536, 16, 65536}, + {65536, 127}, + {65536, 128}, + {65536, 128, 256}, + {65536, 128, 65536}, + {65536, 129}, + {65536, 65536, 256}, + {65536, 65536, 65536}, + } + + for i, tc := range testCases { + for _, firstN := range []int{1, 65534, 65535, 65536, 65537, 131072} { + tc[0] = firstN + outer: + for _, flush := range []bool{false, true} { + buf := new(bytes.Buffer) + want = want[:0] + + w, err := NewWriter(buf, BestSpeed) + if err != nil { + t.Errorf("i=%d, firstN=%d, flush=%t: NewWriter: %v", i, firstN, flush, err) + continue + } + for _, n := range tc { + want = append(want, abcabc[:n]...) + if _, err := w.Write(abcabc[:n]); err != nil { + t.Errorf("i=%d, firstN=%d, flush=%t: Write: %v", i, firstN, flush, err) + continue outer + } + if !flush { + continue + } + if err := w.Flush(); err != nil { + t.Errorf("i=%d, firstN=%d, flush=%t: Flush: %v", i, firstN, flush, err) + continue outer + } + } + if err := w.Close(); err != nil { + t.Errorf("i=%d, firstN=%d, flush=%t: Close: %v", i, firstN, flush, err) + continue + } + + r := NewReader(buf) + got, err := ioutil.ReadAll(r) + if err != nil { + t.Errorf("i=%d, firstN=%d, flush=%t: ReadAll: %v", i, firstN, flush, err) + continue + } + r.Close() + + if !bytes.Equal(got, want) { + t.Errorf("i=%d, firstN=%d, flush=%t: corruption during deflate-then-inflate", i, firstN, flush) + continue + } + } + } + } +} diff --git a/github.com/klauspost/compress/flate/dict_decoder.go b/github.com/klauspost/compress/flate/dict_decoder.go new file mode 100644 index 0000000..71c75a0 --- /dev/null +++ b/github.com/klauspost/compress/flate/dict_decoder.go @@ -0,0 +1,184 @@ +// 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. + +package flate + +// dictDecoder implements the LZ77 sliding dictionary as used in decompression. +// LZ77 decompresses data through sequences of two forms of commands: +// +// * Literal insertions: Runs of one or more symbols are inserted into the data +// stream as is. This is accomplished through the writeByte method for a +// single symbol, or combinations of writeSlice/writeMark for multiple symbols. +// Any valid stream must start with a literal insertion if no preset dictionary +// is used. +// +// * Backward copies: Runs of one or more symbols are copied from previously +// emitted data. Backward copies come as the tuple (dist, length) where dist +// determines how far back in the stream to copy from and length determines how +// many bytes to copy. Note that it is valid for the length to be greater than +// the distance. Since LZ77 uses forward copies, that situation is used to +// perform a form of run-length encoding on repeated runs of symbols. +// The writeCopy and tryWriteCopy are used to implement this command. +// +// For performance reasons, this implementation performs little to no sanity +// checks about the arguments. As such, the invariants documented for each +// method call must be respected. +type dictDecoder struct { + hist []byte // Sliding window history + + // Invariant: 0 <= rdPos <= wrPos <= len(hist) + wrPos int // Current output position in buffer + rdPos int // Have emitted hist[:rdPos] already + full bool // Has a full window length been written yet? +} + +// init initializes dictDecoder to have a sliding window dictionary of the given +// size. If a preset dict is provided, it will initialize the dictionary with +// the contents of dict. +func (dd *dictDecoder) init(size int, dict []byte) { + *dd = dictDecoder{hist: dd.hist} + + if cap(dd.hist) < size { + dd.hist = make([]byte, size) + } + dd.hist = dd.hist[:size] + + if len(dict) > len(dd.hist) { + dict = dict[len(dict)-len(dd.hist):] + } + dd.wrPos = copy(dd.hist, dict) + if dd.wrPos == len(dd.hist) { + dd.wrPos = 0 + dd.full = true + } + dd.rdPos = dd.wrPos +} + +// histSize reports the total amount of historical data in the dictionary. +func (dd *dictDecoder) histSize() int { + if dd.full { + return len(dd.hist) + } + return dd.wrPos +} + +// availRead reports the number of bytes that can be flushed by readFlush. +func (dd *dictDecoder) availRead() int { + return dd.wrPos - dd.rdPos +} + +// availWrite reports the available amount of output buffer space. +func (dd *dictDecoder) availWrite() int { + return len(dd.hist) - dd.wrPos +} + +// writeSlice returns a slice of the available buffer to write data to. +// +// This invariant will be kept: len(s) <= availWrite() +func (dd *dictDecoder) writeSlice() []byte { + return dd.hist[dd.wrPos:] +} + +// writeMark advances the writer pointer by cnt. +// +// This invariant must be kept: 0 <= cnt <= availWrite() +func (dd *dictDecoder) writeMark(cnt int) { + dd.wrPos += cnt +} + +// writeByte writes a single byte to the dictionary. +// +// This invariant must be kept: 0 < availWrite() +func (dd *dictDecoder) writeByte(c byte) { + dd.hist[dd.wrPos] = c + dd.wrPos++ +} + +// writeCopy copies a string at a given (dist, length) to the output. +// This returns the number of bytes copied and may be less than the requested +// length if the available space in the output buffer is too small. +// +// This invariant must be kept: 0 < dist <= histSize() +func (dd *dictDecoder) writeCopy(dist, length int) int { + dstBase := dd.wrPos + dstPos := dstBase + srcPos := dstPos - dist + endPos := dstPos + length + if endPos > len(dd.hist) { + endPos = len(dd.hist) + } + + // Copy non-overlapping section after destination position. + // + // This section is non-overlapping in that the copy length for this section + // is always less than or equal to the backwards distance. This can occur + // if a distance refers to data that wraps-around in the buffer. + // Thus, a backwards copy is performed here; that is, the exact bytes in + // the source prior to the copy is placed in the destination. + if srcPos < 0 { + srcPos += len(dd.hist) + dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:]) + srcPos = 0 + } + + // Copy possibly overlapping section before destination position. + // + // This section can overlap if the copy length for this section is larger + // than the backwards distance. This is allowed by LZ77 so that repeated + // strings can be succinctly represented using (dist, length) pairs. + // Thus, a forwards copy is performed here; that is, the bytes copied is + // possibly dependent on the resulting bytes in the destination as the copy + // progresses along. This is functionally equivalent to the following: + // + // for i := 0; i < endPos-dstPos; i++ { + // dd.hist[dstPos+i] = dd.hist[srcPos+i] + // } + // dstPos = endPos + // + for dstPos < endPos { + dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos]) + } + + dd.wrPos = dstPos + return dstPos - dstBase +} + +// tryWriteCopy tries to copy a string at a given (distance, length) to the +// output. This specialized version is optimized for short distances. +// +// This method is designed to be inlined for performance reasons. +// +// This invariant must be kept: 0 < dist <= histSize() +func (dd *dictDecoder) tryWriteCopy(dist, length int) int { + dstPos := dd.wrPos + endPos := dstPos + length + if dstPos < dist || endPos > len(dd.hist) { + return 0 + } + dstBase := dstPos + srcPos := dstPos - dist + + // Copy possibly overlapping section before destination position. +loop: + dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos]) + if dstPos < endPos { + goto loop // Avoid for-loop so that this function can be inlined + } + + dd.wrPos = dstPos + return dstPos - dstBase +} + +// readFlush returns a slice of the historical buffer that is ready to be +// emitted to the user. The data returned by readFlush must be fully consumed +// before calling any other dictDecoder methods. +func (dd *dictDecoder) readFlush() []byte { + toRead := dd.hist[dd.rdPos:dd.wrPos] + dd.rdPos = dd.wrPos + if dd.wrPos == len(dd.hist) { + dd.wrPos, dd.rdPos = 0, 0 + dd.full = true + } + return toRead +} diff --git a/github.com/klauspost/compress/flate/dict_decoder_test.go b/github.com/klauspost/compress/flate/dict_decoder_test.go new file mode 100644 index 0000000..9275cff --- /dev/null +++ b/github.com/klauspost/compress/flate/dict_decoder_test.go @@ -0,0 +1,139 @@ +// 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. + +package flate + +import ( + "bytes" + "strings" + "testing" +) + +func TestDictDecoder(t *testing.T) { + const ( + abc = "ABC\n" + fox = "The quick brown fox jumped over the lazy dog!\n" + poem = "The Road Not Taken\nRobert Frost\n" + + "\n" + + "Two roads diverged in a yellow wood,\n" + + "And sorry I could not travel both\n" + + "And be one traveler, long I stood\n" + + "And looked down one as far as I could\n" + + "To where it bent in the undergrowth;\n" + + "\n" + + "Then took the other, as just as fair,\n" + + "And having perhaps the better claim,\n" + + "Because it was grassy and wanted wear;\n" + + "Though as for that the passing there\n" + + "Had worn them really about the same,\n" + + "\n" + + "And both that morning equally lay\n" + + "In leaves no step had trodden black.\n" + + "Oh, I kept the first for another day!\n" + + "Yet knowing how way leads on to way,\n" + + "I doubted if I should ever come back.\n" + + "\n" + + "I shall be telling this with a sigh\n" + + "Somewhere ages and ages hence:\n" + + "Two roads diverged in a wood, and I-\n" + + "I took the one less traveled by,\n" + + "And that has made all the difference.\n" + ) + + var poemRefs = []struct { + dist int // Backward distance (0 if this is an insertion) + length int // Length of copy or insertion + }{ + {0, 38}, {33, 3}, {0, 48}, {79, 3}, {0, 11}, {34, 5}, {0, 6}, {23, 7}, + {0, 8}, {50, 3}, {0, 2}, {69, 3}, {34, 5}, {0, 4}, {97, 3}, {0, 4}, + {43, 5}, {0, 6}, {7, 4}, {88, 7}, {0, 12}, {80, 3}, {0, 2}, {141, 4}, + {0, 1}, {196, 3}, {0, 3}, {157, 3}, {0, 6}, {181, 3}, {0, 2}, {23, 3}, + {77, 3}, {28, 5}, {128, 3}, {110, 4}, {70, 3}, {0, 4}, {85, 6}, {0, 2}, + {182, 6}, {0, 4}, {133, 3}, {0, 7}, {47, 5}, {0, 20}, {112, 5}, {0, 1}, + {58, 3}, {0, 8}, {59, 3}, {0, 4}, {173, 3}, {0, 5}, {114, 3}, {0, 4}, + {92, 5}, {0, 2}, {71, 3}, {0, 2}, {76, 5}, {0, 1}, {46, 3}, {96, 4}, + {130, 4}, {0, 3}, {360, 3}, {0, 3}, {178, 5}, {0, 7}, {75, 3}, {0, 3}, + {45, 6}, {0, 6}, {299, 6}, {180, 3}, {70, 6}, {0, 1}, {48, 3}, {66, 4}, + {0, 3}, {47, 5}, {0, 9}, {325, 3}, {0, 1}, {359, 3}, {318, 3}, {0, 2}, + {199, 3}, {0, 1}, {344, 3}, {0, 3}, {248, 3}, {0, 10}, {310, 3}, {0, 3}, + {93, 6}, {0, 3}, {252, 3}, {157, 4}, {0, 2}, {273, 5}, {0, 14}, {99, 4}, + {0, 1}, {464, 4}, {0, 2}, {92, 4}, {495, 3}, {0, 1}, {322, 4}, {16, 4}, + {0, 3}, {402, 3}, {0, 2}, {237, 4}, {0, 2}, {432, 4}, {0, 1}, {483, 5}, + {0, 2}, {294, 4}, {0, 2}, {306, 3}, {113, 5}, {0, 1}, {26, 4}, {164, 3}, + {488, 4}, {0, 1}, {542, 3}, {248, 6}, {0, 5}, {205, 3}, {0, 8}, {48, 3}, + {449, 6}, {0, 2}, {192, 3}, {328, 4}, {9, 5}, {433, 3}, {0, 3}, {622, 25}, + {615, 5}, {46, 5}, {0, 2}, {104, 3}, {475, 10}, {549, 3}, {0, 4}, {597, 8}, + {314, 3}, {0, 1}, {473, 6}, {317, 5}, {0, 1}, {400, 3}, {0, 3}, {109, 3}, + {151, 3}, {48, 4}, {0, 4}, {125, 3}, {108, 3}, {0, 2}, + } + + var got, want bytes.Buffer + var dd dictDecoder + dd.init(1<<11, nil) + + var writeCopy = func(dist, length int) { + for length > 0 { + cnt := dd.tryWriteCopy(dist, length) + if cnt == 0 { + cnt = dd.writeCopy(dist, length) + } + + length -= cnt + if dd.availWrite() == 0 { + got.Write(dd.readFlush()) + } + } + } + var writeString = func(str string) { + for len(str) > 0 { + cnt := copy(dd.writeSlice(), str) + str = str[cnt:] + dd.writeMark(cnt) + if dd.availWrite() == 0 { + got.Write(dd.readFlush()) + } + } + } + + writeString(".") + want.WriteByte('.') + + str := poem + for _, ref := range poemRefs { + if ref.dist == 0 { + writeString(str[:ref.length]) + } else { + writeCopy(ref.dist, ref.length) + } + str = str[ref.length:] + } + want.WriteString(poem) + + writeCopy(dd.histSize(), 33) + want.Write(want.Bytes()[:33]) + + writeString(abc) + writeCopy(len(abc), 59*len(abc)) + want.WriteString(strings.Repeat(abc, 60)) + + writeString(fox) + writeCopy(len(fox), 9*len(fox)) + want.WriteString(strings.Repeat(fox, 10)) + + writeString(".") + writeCopy(1, 9) + want.WriteString(strings.Repeat(".", 10)) + + writeString(strings.ToUpper(poem)) + writeCopy(len(poem), 7*len(poem)) + want.WriteString(strings.Repeat(strings.ToUpper(poem), 8)) + + writeCopy(dd.histSize(), 10) + want.Write(want.Bytes()[want.Len()-dd.histSize():][:10]) + + got.Write(dd.readFlush()) + if got.String() != want.String() { + t.Errorf("final string mismatch:\ngot %q\nwant %q", got.String(), want.String()) + } +} diff --git a/github.com/klauspost/compress/flate/fast_encoder.go b/github.com/klauspost/compress/flate/fast_encoder.go new file mode 100644 index 0000000..6d4c1e9 --- /dev/null +++ b/github.com/klauspost/compress/flate/fast_encoder.go @@ -0,0 +1,254 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Modified for deflate by Klaus Post (c) 2015. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package flate + +import ( + "fmt" + "math/bits" +) + +type fastEnc interface { + Encode(dst *tokens, src []byte) + Reset() +} + +func newFastEnc(level int) fastEnc { + switch level { + case 1: + return &fastEncL1{fastGen: fastGen{cur: maxStoreBlockSize}} + case 2: + return &fastEncL2{fastGen: fastGen{cur: maxStoreBlockSize}} + case 3: + return &fastEncL3{fastGen: fastGen{cur: maxStoreBlockSize}} + case 4: + return &fastEncL4{fastGen: fastGen{cur: maxStoreBlockSize}} + case 5: + return &fastEncL5{fastGen: fastGen{cur: maxStoreBlockSize}} + case 6: + return &fastEncL6{fastGen: fastGen{cur: maxStoreBlockSize}} + default: + panic("invalid level specified") + } +} + +const ( + 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 = 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 ( + prime3bytes = 506832829 + prime4bytes = 2654435761 + prime5bytes = 889523592379 + prime6bytes = 227718039650203 + prime7bytes = 58295818150454627 + prime8bytes = 0xcf1bbcdcb7a56463 +) + +func load32(b []byte, i int) 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 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 +} + +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 hash(u uint32) uint32 { + return (u * 0x1e35a7bd) >> tableShift +} + +type tableEntry struct { + offset int32 +} + +// fastGen maintains the table for matches, +// and the previous byte block for level 2. +// This is the generic implementation. +type fastGen struct { + hist []byte + cur int32 +} + +func (e *fastGen) addBlock(src []byte) int32 { + // check if we have space already + if len(e.hist)+len(src) > cap(e.hist) { + if cap(e.hist) == 0 { + e.hist = make([]byte, 0, allocHistory) + } else { + if cap(e.hist) < maxMatchOffset*2 { + panic("unexpected buffer size") + } + // Move down + offset := int32(len(e.hist)) - maxMatchOffset + copy(e.hist[0:maxMatchOffset], e.hist[offset:]) + e.cur += offset + e.hist = e.hist[:maxMatchOffset] + } + } + s := int32(len(e.hist)) + e.hist = append(e.hist, src...) + return s +} + +// 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) +} + +type tableEntryPrev struct { + Cur tableEntry + Prev tableEntry +} + +// 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) +} + +// 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)) +} + +// 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)) +} + +// matchlen will return the match length between offsets and t in src. +// The maximum length returned is maxMatchLength - 4. +// It is assumed that s > t, that t >=0 and s < len(src). +func (e *fastGen) matchlen(s, t int32, src []byte) int32 { + if debugDecode { + if t >= s { + panic(fmt.Sprint("t >=s:", t, s)) + } + if int(s) >= len(src) { + panic(fmt.Sprint("s >= len(src):", s, len(src))) + } + if t < 0 { + panic(fmt.Sprint("t < 0:", t)) + } + if s-t > maxMatchOffset { + panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")")) + } + } + s1 := int(s) + maxMatchLength - 4 + if s1 > len(src) { + s1 = len(src) + } + + // Extend the match to be as long as possible. + return int32(matchLen(src[s:s1], src[t:])) +} + +// matchlenLong will return the match length between offsets and t in src. +// It is assumed that s > t, that t >=0 and s < len(src). +func (e *fastGen) matchlenLong(s, t int32, src []byte) int32 { + if debugDecode { + if t >= s { + panic(fmt.Sprint("t >=s:", t, s)) + } + if int(s) >= len(src) { + panic(fmt.Sprint("s >= len(src):", s, len(src))) + } + if t < 0 { + panic(fmt.Sprint("t < 0:", t)) + } + if s-t > maxMatchOffset { + panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")")) + } + } + // Extend the match to be as long as possible. + return int32(matchLen(src[s:], src[t:])) +} + +// Reset the encoding table. +func (e *fastGen) Reset() { + 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)) + } + e.hist = e.hist[:0] +} + +// matchLen returns the maximum length. +// 'a' must be the shortest of the two. +func matchLen(a, b []byte) int { + b = b[:len(a)] + var checked int + if len(a) > 4 { + // Try 4 bytes first + if diff := load32(a, 0) ^ load32(b, 0); diff != 0 { + return bits.TrailingZeros32(diff) >> 3 + } + // Switch to 8 byte matching. + checked = 4 + a = a[4:] + b = b[4:] + for len(a) >= 8 { + b = b[:len(a)] + if diff := load64(a, 0) ^ load64(b, 0); diff != 0 { + return checked + (bits.TrailingZeros64(diff) >> 3) + } + checked += 8 + a = a[8:] + b = b[8:] + } + } + b = b[:len(a)] + for i := range a { + if a[i] != b[i] { + return int(i) + checked + } + } + return len(a) + checked +} diff --git a/github.com/klauspost/compress/flate/flate_test.go b/github.com/klauspost/compress/flate/flate_test.go new file mode 100644 index 0000000..52517d8 --- /dev/null +++ b/github.com/klauspost/compress/flate/flate_test.go @@ -0,0 +1,360 @@ +// 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. + +// This test tests some internals of the flate package. +// The tests in package compress/gzip serve as the +// end-to-end test of the decompressor. + +package flate + +import ( + "archive/zip" + "bytes" + "compress/flate" + "encoding/hex" + "fmt" + "io/ioutil" + "testing" +) + +// The following test should not panic. +func TestIssue5915(t *testing.T) { + bits := []int{4, 0, 0, 6, 4, 3, 2, 3, 3, 4, 4, 5, 0, 0, 0, 0, 5, 5, 6, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 7, 8, 6, 0, 11, 0, 8, 0, 6, 6, 10, 8} + var h huffmanDecoder + if h.init(bits) { + t.Fatalf("Given sequence of bits is bad, and should not succeed.") + } +} + +// The following test should not panic. +func TestIssue5962(t *testing.T) { + bits := []int{4, 0, 0, 6, 4, 3, 2, 3, 3, 4, 4, 5, 0, 0, 0, 0, + 5, 5, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 11} + var h huffmanDecoder + if h.init(bits) { + t.Fatalf("Given sequence of bits is bad, and should not succeed.") + } +} + +// The following test should not panic. +func TestIssue6255(t *testing.T) { + bits1 := []int{1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 11} + bits2 := []int{11, 13} + var h huffmanDecoder + if !h.init(bits1) { + t.Fatalf("Given sequence of bits is good and should succeed.") + } + if h.init(bits2) { + t.Fatalf("Given sequence of bits is bad and should not succeed.") + } +} + +func TestInvalidEncoding(t *testing.T) { + // Initialize Huffman decoder to recognize "0". + var h huffmanDecoder + if !h.init([]int{1}) { + t.Fatal("Failed to initialize Huffman decoder") + } + + // Initialize decompressor with invalid Huffman coding. + var f decompressor + f.r = bytes.NewReader([]byte{0xff}) + + _, err := f.huffSym(&h) + if err == nil { + t.Fatal("Should have rejected invalid bit sequence") + } +} + +func TestRegressions(t *testing.T) { + // Test fuzzer regressions + data, err := ioutil.ReadFile("testdata/regression.zip") + if err != nil { + t.Fatal(err) + } + zr, err := zip.NewReader(bytes.NewReader(data), int64(len(data))) + if err != nil { + t.Fatal(err) + } + for _, tt := range zr.File { + data, err := tt.Open() + if err != nil { + t.Fatal(err) + } + data1, err := ioutil.ReadAll(data) + if err != nil { + t.Fatal(err) + } + for level := 0; level <= 9; level++ { + t.Run(fmt.Sprint(tt.Name+"-level", 1), func(t *testing.T) { + buf := new(bytes.Buffer) + fw, err := NewWriter(buf, level) + if err != nil { + t.Error(err) + } + n, err := fw.Write(data1) + if n != len(data1) { + t.Error("short write") + } + if err != nil { + t.Error(err) + } + err = fw.Close() + if err != nil { + t.Error(err) + } + fr1 := NewReader(buf) + data2, err := ioutil.ReadAll(fr1) + if err != nil { + t.Error(err) + } + if bytes.Compare(data1, data2) != 0 { + t.Error("not equal") + } + // Do it again... + buf.Reset() + fw.Reset(buf) + n, err = fw.Write(data1) + if n != len(data1) { + t.Error("short write") + } + if err != nil { + t.Error(err) + } + err = fw.Close() + if err != nil { + t.Error(err) + } + fr1 = flate.NewReader(buf) + data2, err = ioutil.ReadAll(fr1) + if err != nil { + t.Error(err) + } + if bytes.Compare(data1, data2) != 0 { + t.Error("not equal") + } + }) + } + t.Run(tt.Name+"stateless", func(t *testing.T) { + // Split into two and use history... + buf := new(bytes.Buffer) + err = StatelessDeflate(buf, data1[:len(data1)/2], false, nil) + if err != nil { + t.Error(err) + } + + // Use top half as dictionary... + dict := data1[:len(data1)/2] + err = StatelessDeflate(buf, data1[len(data1)/2:], true, dict) + if err != nil { + t.Error(err) + } + t.Log(buf.Len()) + fr1 := NewReader(buf) + data2, err := ioutil.ReadAll(fr1) + if err != nil { + t.Error(err) + } + if bytes.Compare(data1, data2) != 0 { + fmt.Printf("want:%x\ngot: %x\n", data1, data2) + t.Error("not equal") + } + }) + } +} + +func TestInvalidBits(t *testing.T) { + oversubscribed := []int{1, 2, 3, 4, 4, 5} + incomplete := []int{1, 2, 4, 4} + var h huffmanDecoder + if h.init(oversubscribed) { + t.Fatal("Should reject oversubscribed bit-length set") + } + if h.init(incomplete) { + t.Fatal("Should reject incomplete bit-length set") + } +} + +func TestStreams(t *testing.T) { + // To verify any of these hexstrings as valid or invalid flate streams + // according to the C zlib library, you can use the Python wrapper library: + // >>> hex_string = "010100feff11" + // >>> import zlib + // >>> zlib.decompress(hex_string.decode("hex"), -15) # Negative means raw DEFLATE + // '\x11' + + testCases := []struct { + desc string // Description of the stream + stream string // Hexstring of the input DEFLATE stream + want string // Expected result. Use "fail" to expect failure + }{{ + "degenerate HCLenTree", + "05e0010000000000100000000000000000000000000000000000000000000000" + + "00000000000000000004", + "fail", + }, { + "complete HCLenTree, empty HLitTree, empty HDistTree", + "05e0010400000000000000000000000000000000000000000000000000000000" + + "00000000000000000010", + "fail", + }, { + "empty HCLenTree", + "05e0010000000000000000000000000000000000000000000000000000000000" + + "00000000000000000010", + "fail", + }, { + "complete HCLenTree, complete HLitTree, empty HDistTree, use missing HDist symbol", + "000100feff000de0010400000000100000000000000000000000000000000000" + + "0000000000000000000000000000002c", + "fail", + }, { + "complete HCLenTree, complete HLitTree, degenerate HDistTree, use missing HDist symbol", + "000100feff000de0010000000000000000000000000000000000000000000000" + + "00000000000000000610000000004070", + "fail", + }, { + "complete HCLenTree, empty HLitTree, empty HDistTree", + "05e0010400000000100400000000000000000000000000000000000000000000" + + "0000000000000000000000000008", + "fail", + }, { + "complete HCLenTree, empty HLitTree, degenerate HDistTree", + "05e0010400000000100400000000000000000000000000000000000000000000" + + "0000000000000000000800000008", + "fail", + }, { + "complete HCLenTree, degenerate HLitTree, degenerate HDistTree, use missing HLit symbol", + "05e0010400000000100000000000000000000000000000000000000000000000" + + "0000000000000000001c", + "fail", + }, { + "complete HCLenTree, complete HLitTree, too large HDistTree", + "edff870500000000200400000000000000000000000000000000000000000000" + + "000000000000000000080000000000000004", + "fail", + }, { + "complete HCLenTree, complete HLitTree, empty HDistTree, excessive repeater code", + "edfd870500000000200400000000000000000000000000000000000000000000" + + "000000000000000000e8b100", + "fail", + }, { + "complete HCLenTree, complete HLitTree, empty HDistTree of normal length 30", + "05fd01240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" + + "ffffffffffffffffff07000000fe01", + "", + }, { + "complete HCLenTree, complete HLitTree, empty HDistTree of excessive length 31", + "05fe01240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" + + "ffffffffffffffffff07000000fc03", + "fail", + }, { + "complete HCLenTree, over-subscribed HLitTree, empty HDistTree", + "05e001240000000000fcffffffffffffffffffffffffffffffffffffffffffff" + + "ffffffffffffffffff07f00f", + "fail", + }, { + "complete HCLenTree, under-subscribed HLitTree, empty HDistTree", + "05e001240000000000fcffffffffffffffffffffffffffffffffffffffffffff" + + "fffffffffcffffffff07f00f", + "fail", + }, { + "complete HCLenTree, complete HLitTree with single code, empty HDistTree", + "05e001240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" + + "ffffffffffffffffff07f00f", + "01", + }, { + "complete HCLenTree, complete HLitTree with multiple codes, empty HDistTree", + "05e301240000000000f8ffffffffffffffffffffffffffffffffffffffffffff" + + "ffffffffffffffffff07807f", + "01", + }, { + "complete HCLenTree, complete HLitTree, degenerate HDistTree, use valid HDist symbol", + "000100feff000de0010400000000100000000000000000000000000000000000" + + "0000000000000000000000000000003c", + "00000000", + }, { + "complete HCLenTree, degenerate HLitTree, degenerate HDistTree", + "05e0010400000000100000000000000000000000000000000000000000000000" + + "0000000000000000000c", + "", + }, { + "complete HCLenTree, degenerate HLitTree, empty HDistTree", + "05e0010400000000100000000000000000000000000000000000000000000000" + + "00000000000000000004", + "", + }, { + "complete HCLenTree, complete HLitTree, empty HDistTree, spanning repeater code", + "edfd870500000000200400000000000000000000000000000000000000000000" + + "000000000000000000e8b000", + "", + }, { + "complete HCLenTree with length codes, complete HLitTree, empty HDistTree", + "ede0010400000000100000000000000000000000000000000000000000000000" + + "0000000000000000000400004000", + "", + }, { + "complete HCLenTree, complete HLitTree, degenerate HDistTree, use valid HLit symbol 284 with count 31", + "000100feff00ede0010400000000100000000000000000000000000000000000" + + "000000000000000000000000000000040000407f00", + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "0000000000000000000000000000000000000000000000000000000000000000" + + "000000", + }, { + "complete HCLenTree, complete HLitTree, degenerate HDistTree, use valid HLit and HDist symbols", + "0cc2010d00000082b0ac4aff0eb07d27060000ffff", + "616263616263", + }, { + "fixed block, use reserved symbol 287", + "33180700", + "fail", + }, { + "raw block", + "010100feff11", + "11", + }, { + "issue 10426 - over-subscribed HCLenTree causes a hang", + "344c4a4e494d4b070000ff2e2eff2e2e2e2e2eff", + "fail", + }, { + "issue 11030 - empty HDistTree unexpectedly leads to error", + "05c0070600000080400fff37a0ca", + "", + }, { + "issue 11033 - empty HDistTree unexpectedly leads to error", + "050fb109c020cca5d017dcbca044881ee1034ec149c8980bbc413c2ab35be9dc" + + "b1473449922449922411202306ee97b0383a521b4ffdcf3217f9f7d3adb701", + "3130303634342068652e706870005d05355f7ed957ff084a90925d19e3ebc6d0" + + "c6d7", + }} + + for i, tc := range testCases { + data, err := hex.DecodeString(tc.stream) + if err != nil { + t.Fatal(err) + } + data, err = ioutil.ReadAll(NewReader(bytes.NewReader(data))) + if tc.want == "fail" { + if err == nil { + t.Errorf("#%d (%s): got nil error, want non-nil", i, tc.desc) + } + } else { + if err != nil { + t.Errorf("#%d (%s): %v", i, tc.desc, err) + continue + } + if got := hex.EncodeToString(data); got != tc.want { + t.Errorf("#%d (%s):\ngot %q\nwant %q", i, tc.desc, got, tc.want) + } + + } + } +} diff --git a/github.com/klauspost/compress/flate/gen.go b/github.com/klauspost/compress/flate/gen.go new file mode 100644 index 0000000..154c89a --- /dev/null +++ b/github.com/klauspost/compress/flate/gen.go @@ -0,0 +1,265 @@ +// Copyright 2012 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 ignore + +// This program generates fixedhuff.go +// Invoke as +// +// go run gen.go -output fixedhuff.go + +package main + +import ( + "bytes" + "flag" + "fmt" + "go/format" + "io/ioutil" + "log" +) + +var filename = flag.String("output", "fixedhuff.go", "output file name") + +const maxCodeLen = 16 + +// Note: the definition of the huffmanDecoder struct is copied from +// inflate.go, as it is private to the implementation. + +// chunk & 15 is number of bits +// chunk >> 4 is value, including table link + +const ( + huffmanChunkBits = 9 + huffmanNumChunks = 1 << huffmanChunkBits + huffmanCountMask = 15 + huffmanValueShift = 4 +) + +type huffmanDecoder struct { + min int // the minimum code length + chunks [huffmanNumChunks]uint32 // chunks as described above + links [][]uint32 // overflow links + linkMask uint32 // mask the width of the link table +} + +// Initialize Huffman decoding tables from array of code lengths. +// Following this function, h is guaranteed to be initialized into a complete +// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a +// degenerate case where the tree has only a single symbol with length 1. Empty +// trees are permitted. +func (h *huffmanDecoder) init(bits []int) bool { + // Sanity enables additional runtime tests during Huffman + // table construction. It's intended to be used during + // development to supplement the currently ad-hoc unit tests. + const sanity = false + + if h.min != 0 { + *h = huffmanDecoder{} + } + + // Count number of codes of each length, + // compute min and max length. + var count [maxCodeLen]int + var min, max int + for _, n := range bits { + if n == 0 { + continue + } + if min == 0 || n < min { + min = n + } + if n > max { + max = n + } + count[n]++ + } + + // Empty tree. The decompressor.huffSym function will fail later if the tree + // is used. Technically, an empty tree is only valid for the HDIST tree and + // not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree + // is guaranteed to fail since it will attempt to use the tree to decode the + // codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is + // guaranteed to fail later since the compressed data section must be + // composed of at least one symbol (the end-of-block marker). + if max == 0 { + return true + } + + code := 0 + var nextcode [maxCodeLen]int + for i := min; i <= max; i++ { + code <<= 1 + nextcode[i] = code + code += count[i] + } + + // Check that the coding is complete (i.e., that we've + // assigned all 2-to-the-max possible bit sequences). + // Exception: To be compatible with zlib, we also need to + // accept degenerate single-code codings. See also + // TestDegenerateHuffmanCoding. + if code != 1< huffmanChunkBits { + numLinks := 1 << (uint(max) - huffmanChunkBits) + h.linkMask = uint32(numLinks - 1) + + // create link tables + link := nextcode[huffmanChunkBits+1] >> 1 + h.links = make([][]uint32, huffmanNumChunks-link) + for j := uint(link); j < huffmanNumChunks; j++ { + reverse := int(reverseByte[j>>8]) | int(reverseByte[j&0xff])<<8 + reverse >>= uint(16 - huffmanChunkBits) + off := j - uint(link) + if sanity && h.chunks[reverse] != 0 { + panic("impossible: overwriting existing chunk") + } + h.chunks[reverse] = uint32(off<>8]) | int(reverseByte[code&0xff])<<8 + reverse >>= uint(16 - n) + if n <= huffmanChunkBits { + for off := reverse; off < len(h.chunks); off += 1 << uint(n) { + // We should never need to overwrite + // an existing chunk. Also, 0 is + // never a valid chunk, because the + // lower 4 "count" bits should be + // between 1 and 15. + if sanity && h.chunks[off] != 0 { + panic("impossible: overwriting existing chunk") + } + h.chunks[off] = chunk + } + } else { + j := reverse & (huffmanNumChunks - 1) + if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 { + // Longer codes should have been + // associated with a link table above. + panic("impossible: not an indirect chunk") + } + value := h.chunks[j] >> huffmanValueShift + linktab := h.links[value] + reverse >>= huffmanChunkBits + for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) { + if sanity && linktab[off] != 0 { + panic("impossible: overwriting existing chunk") + } + linktab[off] = chunk + } + } + } + + if sanity { + // Above we've sanity checked that we never overwrote + // an existing entry. Here we additionally check that + // we filled the tables completely. + for i, chunk := range h.chunks { + if chunk == 0 { + // As an exception, in the degenerate + // single-code case, we allow odd + // chunks to be missing. + if code == 1 && i%2 == 1 { + continue + } + panic("impossible: missing chunk") + } + } + for _, linktab := range h.links { + for _, chunk := range linktab { + if chunk == 0 { + panic("impossible: missing chunk") + } + } + } + } + + return true +} + +func main() { + flag.Parse() + + var h huffmanDecoder + var bits [288]int + initReverseByte() + for i := 0; i < 144; i++ { + bits[i] = 8 + } + for i := 144; i < 256; i++ { + bits[i] = 9 + } + for i := 256; i < 280; i++ { + bits[i] = 7 + } + for i := 280; i < 288; i++ { + bits[i] = 8 + } + h.init(bits[:]) + if h.links != nil { + log.Fatal("Unexpected links table in fixed Huffman decoder") + } + + var buf bytes.Buffer + + fmt.Fprintf(&buf, `// Copyright 2013 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.`+"\n\n") + + fmt.Fprintln(&buf, "package flate") + fmt.Fprintln(&buf) + fmt.Fprintln(&buf, "// autogenerated by go run gen.go -output fixedhuff.go, DO NOT EDIT") + fmt.Fprintln(&buf) + fmt.Fprintln(&buf, "var fixedHuffmanDecoder = huffmanDecoder{") + fmt.Fprintf(&buf, "\t%d,\n", h.min) + fmt.Fprintln(&buf, "\t[huffmanNumChunks]uint32{") + for i := 0; i < huffmanNumChunks; i++ { + if i&7 == 0 { + fmt.Fprintf(&buf, "\t\t") + } else { + fmt.Fprintf(&buf, " ") + } + fmt.Fprintf(&buf, "0x%04x,", h.chunks[i]) + if i&7 == 7 { + fmt.Fprintln(&buf) + } + } + fmt.Fprintln(&buf, "\t},") + fmt.Fprintln(&buf, "\tnil, 0,") + fmt.Fprintln(&buf, "}") + + data, err := format.Source(buf.Bytes()) + if err != nil { + log.Fatal(err) + } + err = ioutil.WriteFile(*filename, data, 0644) + if err != nil { + log.Fatal(err) + } +} + +var reverseByte [256]byte + +func initReverseByte() { + for x := 0; x < 256; x++ { + var result byte + for i := uint(0); i < 8; i++ { + result |= byte(((x >> i) & 1) << (7 - i)) + } + reverseByte[x] = result + } +} diff --git a/github.com/klauspost/compress/flate/gen_inflate.go b/github.com/klauspost/compress/flate/gen_inflate.go new file mode 100644 index 0000000..c74a95f --- /dev/null +++ b/github.com/klauspost/compress/flate/gen_inflate.go @@ -0,0 +1,274 @@ +// +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.($TYPE$) + moreBits := func() error { + c, err := fr.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + return nil + } + + 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 & 31) + 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 & 31) + 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).$FUNCNAME$ + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + n = 0 + case v < 269: + length = v*2 - (265*2 - 11) + n = 1 + case v < 273: + length = v*4 - (269*4 - 19) + n = 2 + case v < 277: + length = v*8 - (273*8 - 35) + n = 3 + case v < 281: + length = v*16 - (277*16 - 67) + n = 4 + case v < 285: + length = v*32 - (281*32 - 131) + n = 5 + case v < maxNumLit: + length = 258 + n = 0 + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + if n > 0 { + for f.nb < n { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + } + length += int(f.b & uint32(1<>= n + f.nb -= n + } + + var dist int + if f.hd == nil { + for f.nb < 5 { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + } + dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + if dist, err = f.huffSym(f.hd); err != nil { + if debugDecode { + fmt.Println("huffsym:", err) + } + f.err = err + return + } + } + + 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 + for f.nb < nb { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb + f.nb -= nb + dist = 1<<(nb+1) + 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 > 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, 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/github.com/klauspost/compress/flate/huffman_bit_writer.go b/github.com/klauspost/compress/flate/huffman_bit_writer.go new file mode 100644 index 0000000..53fe1d0 --- /dev/null +++ b/github.com/klauspost/compress/flate/huffman_bit_writer.go @@ -0,0 +1,911 @@ +// 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 + +import ( + "io" +) + +const ( + // The largest offset code. + offsetCodeCount = 30 + + // The special code used to mark the end of a block. + endBlockMarker = 256 + + // The first length code. + lengthCodesStart = 257 + + // The number of codegen codes. + codegenCodeCount = 19 + badCode = 255 + + // bufferFlushSize indicates the buffer size + // after which bytes are flushed to the writer. + // Should preferably be a multiple of 6, since + // we accumulate 6 bytes between writes to the buffer. + bufferFlushSize = 240 + + // bufferSize is the actual output byte buffer size. + // It must have additional headroom for a flush + // which can contain up to 8 bytes. + bufferSize = bufferFlushSize + 8 +) + +// The number of extra bits needed by length code X - LENGTH_CODES_START. +var lengthExtraBits = [32]int8{ + /* 257 */ 0, 0, 0, + /* 260 */ 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, + /* 270 */ 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, + /* 280 */ 4, 5, 5, 5, 5, 0, +} + +// The length indicated by length code X - LENGTH_CODES_START. +var lengthBase = [32]uint8{ + 0, 1, 2, 3, 4, 5, 6, 7, 8, 10, + 12, 14, 16, 20, 24, 28, 32, 40, 48, 56, + 64, 80, 96, 112, 128, 160, 192, 224, 255, +} + +// offset code word extra bits. +var offsetExtraBits = [64]int8{ + 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, + 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, + 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, + /* extended window */ + 14, 14, 15, 15, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20, +} + +var offsetBase = [64]uint32{ + /* normal deflate */ + 0x000000, 0x000001, 0x000002, 0x000003, 0x000004, + 0x000006, 0x000008, 0x00000c, 0x000010, 0x000018, + 0x000020, 0x000030, 0x000040, 0x000060, 0x000080, + 0x0000c0, 0x000100, 0x000180, 0x000200, 0x000300, + 0x000400, 0x000600, 0x000800, 0x000c00, 0x001000, + 0x001800, 0x002000, 0x003000, 0x004000, 0x006000, + + /* extended window */ + 0x008000, 0x00c000, 0x010000, 0x018000, 0x020000, + 0x030000, 0x040000, 0x060000, 0x080000, 0x0c0000, + 0x100000, 0x180000, 0x200000, 0x300000, +} + +// The odd order in which the codegen code sizes are written. +var codegenOrder = []uint32{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15} + +type huffmanBitWriter struct { + // writer is the underlying writer. + // Do not use it directly; use the write method, which ensures + // that Write errors are sticky. + writer io.Writer + + // Data waiting to be written is bytes[0:nbytes] + // and then the low nbits of bits. + bits uint64 + nbits uint16 + nbytes uint8 + literalEncoding *huffmanEncoder + offsetEncoding *huffmanEncoder + codegenEncoding *huffmanEncoder + err error + lastHeader int + // Set between 0 (reused block can be up to 2x the size) + 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 +} + +// Huffman reuse. +// +// The huffmanBitWriter supports reusing huffman tables and thereby combining block sections. +// +// This is controlled by several variables: +// +// If lastHeader is non-zero the Huffman table can be reused. +// This also indicates that a Huffman table has been generated that can output all +// possible symbols. +// It also indicates that an EOB has not yet been emitted, so if a new tabel is generated +// an EOB with the previous table must be written. +// +// 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 '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. + +func newHuffmanBitWriter(w io.Writer) *huffmanBitWriter { + return &huffmanBitWriter{ + writer: w, + literalEncoding: newHuffmanEncoder(literalCount), + codegenEncoding: newHuffmanEncoder(codegenCodeCount), + offsetEncoding: newHuffmanEncoder(offsetCodeCount), + } +} + +func (w *huffmanBitWriter) reset(writer io.Writer) { + w.writer = writer + w.bits, w.nbits, w.nbytes, w.err = 0, 0, 0, nil + w.lastHeader = 0 + w.lastHuffMan = false +} + +func (w *huffmanBitWriter) canReuse(t *tokens) (offsets, lits bool) { + offsets, lits = true, true + a := t.offHist[:offsetCodeCount] + b := w.offsetFreq[:len(a)] + for i := range a { + if b[i] == 0 && a[i] != 0 { + offsets = false + break + } + } + + a = t.extraHist[:literalCount-256] + b = w.literalFreq[256:literalCount] + b = b[:len(a)] + for i := range a { + if b[i] == 0 && a[i] != 0 { + lits = false + break + } + } + if lits { + a = t.litHist[:] + b = w.literalFreq[:len(a)] + for i := range a { + if b[i] == 0 && a[i] != 0 { + lits = false + break + } + } + } + return +} + +func (w *huffmanBitWriter) flush() { + if w.err != nil { + 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) + w.bits >>= 8 + if w.nbits > 8 { // Avoid underflow + w.nbits -= 8 + } else { + w.nbits = 0 + } + n++ + } + w.bits = 0 + w.write(w.bytes[:n]) + w.nbytes = 0 +} + +func (w *huffmanBitWriter) write(b []byte) { + if w.err != nil { + return + } + _, w.err = w.writer.Write(b) +} + +func (w *huffmanBitWriter) writeBits(b int32, nb uint16) { + w.bits |= uint64(b) << (w.nbits & 63) + w.nbits += nb + if w.nbits >= 48 { + w.writeOutBits() + } +} + +func (w *huffmanBitWriter) writeBytes(bytes []byte) { + if w.err != nil { + return + } + n := w.nbytes + if w.nbits&7 != 0 { + w.err = InternalError("writeBytes with unfinished bits") + return + } + for w.nbits != 0 { + w.bytes[n] = byte(w.bits) + w.bits >>= 8 + w.nbits -= 8 + n++ + } + if n != 0 { + w.write(w.bytes[:n]) + } + w.nbytes = 0 + w.write(bytes) +} + +// RFC 1951 3.2.7 specifies a special run-length encoding for specifying +// the literal and offset lengths arrays (which are concatenated into a single +// array). This method generates that run-length encoding. +// +// The result is written into the codegen array, and the frequencies +// of each code is written into the codegenFreq array. +// Codes 0-15 are single byte codes. Codes 16-18 are followed by additional +// information. Code badCode is an end marker +// +// numLiterals The number of literals in literalEncoding +// numOffsets The number of offsets in offsetEncoding +// litenc, offenc The literal and offset encoder to use +func (w *huffmanBitWriter) generateCodegen(numLiterals int, numOffsets int, litEnc, offEnc *huffmanEncoder) { + for i := range w.codegenFreq { + w.codegenFreq[i] = 0 + } + // Note that we are using codegen both as a temporary variable for holding + // a copy of the frequencies, and as the place where we put the result. + // This is fine because the output is always shorter than the input used + // so far. + codegen := w.codegen[:] // cache + // Copy the concatenated code sizes to codegen. Put a marker at the end. + cgnl := codegen[:numLiterals] + for i := range cgnl { + cgnl[i] = uint8(litEnc.codes[i].len) + } + + cgnl = codegen[numLiterals : numLiterals+numOffsets] + for i := range cgnl { + cgnl[i] = uint8(offEnc.codes[i].len) + } + codegen[numLiterals+numOffsets] = badCode + + size := codegen[0] + count := 1 + outIndex := 0 + for inIndex := 1; size != badCode; inIndex++ { + // INVARIANT: We have seen "count" copies of size that have not yet + // had output generated for them. + nextSize := codegen[inIndex] + if nextSize == size { + count++ + continue + } + // We need to generate codegen indicating "count" of size. + if size != 0 { + codegen[outIndex] = size + outIndex++ + w.codegenFreq[size]++ + count-- + for count >= 3 { + n := 6 + if n > count { + n = count + } + codegen[outIndex] = 16 + outIndex++ + codegen[outIndex] = uint8(n - 3) + outIndex++ + w.codegenFreq[16]++ + count -= n + } + } else { + for count >= 11 { + n := 138 + if n > count { + n = count + } + codegen[outIndex] = 18 + outIndex++ + codegen[outIndex] = uint8(n - 11) + outIndex++ + w.codegenFreq[18]++ + count -= n + } + if count >= 3 { + // count >= 3 && count <= 10 + codegen[outIndex] = 17 + outIndex++ + codegen[outIndex] = uint8(count - 3) + outIndex++ + w.codegenFreq[17]++ + count = 0 + } + } + count-- + for ; count >= 0; count-- { + codegen[outIndex] = size + outIndex++ + w.codegenFreq[size]++ + } + // Set up invariant for next time through the loop. + size = nextSize + count = 1 + } + // Marker indicating the end of the codegen. + codegen[outIndex] = badCode +} + +func (w *huffmanBitWriter) codegens() int { + numCodegens := len(w.codegenFreq) + for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 { + numCodegens-- + } + return numCodegens +} + +func (w *huffmanBitWriter) headerSize() (size, numCodegens int) { + numCodegens = len(w.codegenFreq) + for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 { + numCodegens-- + } + return 3 + 5 + 5 + 4 + (3 * numCodegens) + + w.codegenEncoding.bitLength(w.codegenFreq[:]) + + int(w.codegenFreq[16])*2 + + int(w.codegenFreq[17])*3 + + 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() + size = header + + litEnc.bitLength(w.literalFreq[:]) + + offEnc.bitLength(w.offsetFreq[:]) + + extraBits + return size, numCodegens +} + +// extraBitSize will return the number of bits that will be written +// as "extra" bits on matches. +func (w *huffmanBitWriter) extraBitSize() int { + total := 0 + for i, n := range w.literalFreq[257:literalCount] { + total += int(n) * int(lengthExtraBits[i&31]) + } + for i, n := range w.offsetFreq[:offsetCodeCount] { + total += int(n) * int(offsetExtraBits[i&31]) + } + return total +} + +// fixedSize returns the size of dynamically encoded data in bits. +func (w *huffmanBitWriter) fixedSize(extraBits int) int { + return 3 + + fixedLiteralEncoding.bitLength(w.literalFreq[:]) + + fixedOffsetEncoding.bitLength(w.offsetFreq[:]) + + extraBits +} + +// storedSize calculates the stored size, including header. +// The function returns the size in bits and whether the block +// fits inside a single block. +func (w *huffmanBitWriter) storedSize(in []byte) (int, bool) { + if in == nil { + return 0, false + } + if len(in) <= maxStoreBlockSize { + return (len(in) + 5) * 8, true + } + return 0, false +} + +func (w *huffmanBitWriter) writeCode(c hcode) { + // The function does not get inlined if we "& 63" the shift. + w.bits |= uint64(c.code) << w.nbits + w.nbits += c.len + if w.nbits >= 48 { + w.writeOutBits() + } +} + +// writeOutBits will write bits to the buffer. +func (w *huffmanBitWriter) writeOutBits() { + bits := w.bits + w.bits >>= 48 + w.nbits -= 48 + n := w.nbytes + w.bytes[n] = byte(bits) + w.bytes[n+1] = byte(bits >> 8) + w.bytes[n+2] = byte(bits >> 16) + w.bytes[n+3] = byte(bits >> 24) + w.bytes[n+4] = byte(bits >> 32) + w.bytes[n+5] = byte(bits >> 40) + n += 6 + if n >= bufferFlushSize { + if w.err != nil { + n = 0 + return + } + w.write(w.bytes[:n]) + n = 0 + } + w.nbytes = n +} + +// Write the header of a dynamic Huffman block to the output stream. +// +// numLiterals The number of literals specified in codegen +// numOffsets The number of offsets specified in codegen +// numCodegens The number of codegens used in codegen +func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, numCodegens int, isEof bool) { + if w.err != nil { + return + } + var firstBits int32 = 4 + if isEof { + firstBits = 5 + } + w.writeBits(firstBits, 3) + w.writeBits(int32(numLiterals-257), 5) + w.writeBits(int32(numOffsets-1), 5) + w.writeBits(int32(numCodegens-4), 4) + + for i := 0; i < numCodegens; i++ { + value := uint(w.codegenEncoding.codes[codegenOrder[i]].len) + w.writeBits(int32(value), 3) + } + + i := 0 + for { + var codeWord = uint32(w.codegen[i]) + i++ + if codeWord == badCode { + break + } + w.writeCode(w.codegenEncoding.codes[codeWord]) + + switch codeWord { + case 16: + w.writeBits(int32(w.codegen[i]), 2) + i++ + case 17: + w.writeBits(int32(w.codegen[i]), 3) + i++ + case 18: + w.writeBits(int32(w.codegen[i]), 7) + i++ + } + } +} + +// 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 + } + if w.lastHeader > 0 { + // We owe an EOB + 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 + } + w.writeBits(flag, 3) + w.flush() + w.writeBits(int32(length), 16) + w.writeBits(int32(^uint16(length)), 16) +} + +func (w *huffmanBitWriter) writeFixedHeader(isEof bool) { + if w.err != nil { + return + } + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + + // Indicate that we are a fixed Huffman block + var value int32 = 2 + if isEof { + value = 3 + } + w.writeBits(value, 3) +} + +// writeBlock will write a block of tokens with the smallest encoding. +// The original input can be supplied, and if the huffman encoded data +// is larger than the original bytes, the data will be written as a +// stored block. +// If the input is nil, the tokens will always be Huffman encoded. +func (w *huffmanBitWriter) writeBlock(tokens *tokens, eof bool, input []byte) { + if w.err != nil { + return + } + + tokens.AddEOB() + if w.lastHeader > 0 { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + numLiterals, numOffsets := w.indexTokens(tokens, false) + w.generate(tokens) + var extraBits int + storedSize, storable := w.storedSize(input) + if storable { + extraBits = w.extraBitSize() + } + + // Figure out smallest code. + // Fixed Huffman baseline. + var literalEncoding = fixedLiteralEncoding + var offsetEncoding = fixedOffsetEncoding + var size = w.fixedSize(extraBits) + + // Dynamic Huffman? + var numCodegens int + + // Generate codegen and codegenFrequencies, which indicates how to encode + // the literalEncoding and the offsetEncoding. + w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding) + w.codegenEncoding.generate(w.codegenFreq[:], 7) + dynamicSize, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits) + + if dynamicSize < size { + size = dynamicSize + literalEncoding = w.literalEncoding + offsetEncoding = w.offsetEncoding + } + + // Stored bytes? + if storable && storedSize < size { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + + // Huffman. + if literalEncoding == fixedLiteralEncoding { + w.writeFixedHeader(eof) + } else { + w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) + } + + // Write the tokens. + w.writeTokens(tokens.Slice(), literalEncoding.codes, offsetEncoding.codes) +} + +// writeBlockDynamic encodes a block using a dynamic Huffman table. +// This should be used if the symbols used have a disproportionate +// histogram distribution. +// If input is supplied and the compression savings are below 1/16th of the +// input size the block is stored. +func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []byte, sync bool) { + if w.err != nil { + return + } + + sync = sync || eof + if sync { + tokens.AddEOB() + } + + // 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 + w.lastHuffMan = false + } + if !sync { + tokens.Fill() + } + numLiterals, numOffsets := w.indexTokens(tokens, !sync) + + var size int + // Check if we should reuse. + if w.lastHeader > 0 { + // 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. + reuseSize := w.dynamicReuseSize(w.literalEncoding, w.offsetEncoding) + w.extraBitSize() + + // Check if a new table is better. + if newSize < reuseSize { + // Write the EOB we owe. + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + size = newSize + w.lastHeader = 0 + } else { + size = reuseSize + } + // Check if we get a reasonable size decrease. + if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + w.lastHeader = 0 + return + } + } + + // We want a new block/table + if w.lastHeader == 0 { + w.generate(tokens) + // Generate codegen and codegenFrequencies, which indicates how to encode + // the literalEncoding and the offsetEncoding. + w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding) + w.codegenEncoding.generate(w.codegenFreq[:], 7) + var numCodegens int + size, numCodegens = w.dynamicSize(w.literalEncoding, w.offsetEncoding, w.extraBitSize()) + // Store bytes, if we don't get a reasonable improvement. + if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + w.lastHeader = 0 + return + } + + // Write Huffman table. + w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) + w.lastHeader, _ = w.headerSize() + w.lastHuffMan = false + } + + if sync { + w.lastHeader = 0 + } + // Write the tokens. + w.writeTokens(tokens.Slice(), w.literalEncoding.codes, w.offsetEncoding.codes) +} + +// indexTokens indexes a slice of tokens, and updates +// literalFreq and offsetFreq, and generates literalEncoding +// and offsetEncoding. +// The number of literal and offset tokens is returned. +func (w *huffmanBitWriter) indexTokens(t *tokens, filled bool) (numLiterals, numOffsets int) { + copy(w.literalFreq[:], t.litHist[:]) + copy(w.literalFreq[256:], t.extraHist[:]) + copy(w.offsetFreq[:], t.offHist[:offsetCodeCount]) + + if t.n == 0 { + return + } + if filled { + return maxNumLit, maxNumDist + } + // get the number of literals + numLiterals = len(w.literalFreq) + for w.literalFreq[numLiterals-1] == 0 { + numLiterals-- + } + // get the number of offsets + numOffsets = len(w.offsetFreq) + for numOffsets > 0 && w.offsetFreq[numOffsets-1] == 0 { + numOffsets-- + } + if numOffsets == 0 { + // We haven't found a single match. If we want to go with the dynamic encoding, + // we should count at least one offset to be sure that the offset huffman tree could be encoded. + w.offsetFreq[0] = 1 + numOffsets = 1 + } + return +} + +func (w *huffmanBitWriter) generate(t *tokens) { + w.literalEncoding.generate(w.literalFreq[:literalCount], 15) + w.offsetEncoding.generate(w.offsetFreq[:offsetCodeCount], 15) +} + +// writeTokens writes a slice of tokens to the output. +// codes for literal and offset encoding must be supplied. +func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) { + if w.err != nil { + return + } + if len(tokens) == 0 { + return + } + + // Only last token should be endBlockMarker. + var deferEOB bool + if tokens[len(tokens)-1] == endBlockMarker { + tokens = tokens[:len(tokens)-1] + deferEOB = true + } + + // Create slices up to the next power of two to avoid bounds checks. + lits := leCodes[:256] + offs := oeCodes[:32] + lengths := leCodes[lengthCodesStart:] + lengths = lengths[:32] + for _, t := range tokens { + if t < matchType { + w.writeCode(lits[t.literal()]) + continue + } + + // Write the length + length := t.length() + lengthCode := lengthCode(length) + if false { + w.writeCode(lengths[lengthCode&31]) + } else { + // inlined + c := lengths[lengthCode&31] + w.bits |= uint64(c.code) << (w.nbits & 63) + w.nbits += c.len + if w.nbits >= 48 { + w.writeOutBits() + } + } + + extraLengthBits := uint16(lengthExtraBits[lengthCode&31]) + if extraLengthBits > 0 { + extraLength := int32(length - lengthBase[lengthCode&31]) + w.writeBits(extraLength, extraLengthBits) + } + // Write the offset + offset := t.offset() + offsetCode := offsetCode(offset) + if false { + w.writeCode(offs[offsetCode&31]) + } else { + // inlined + c := offs[offsetCode&31] + w.bits |= uint64(c.code) << (w.nbits & 63) + w.nbits += c.len + if w.nbits >= 48 { + w.writeOutBits() + } + } + extraOffsetBits := uint16(offsetExtraBits[offsetCode&63]) + if extraOffsetBits > 0 { + extraOffset := int32(offset - offsetBase[offsetCode&63]) + w.writeBits(extraOffset, extraOffsetBits) + } + } + if deferEOB { + w.writeCode(leCodes[endBlockMarker]) + } +} + +// huffOffset is a static offset encoder used for huffman only encoding. +// It can be reused since we will not be encoding offset values. +var huffOffset *huffmanEncoder + +func init() { + w := newHuffmanBitWriter(nil) + w.offsetFreq[0] = 1 + huffOffset = newHuffmanEncoder(offsetCodeCount) + huffOffset.generate(w.offsetFreq[:offsetCodeCount], 15) +} + +// writeBlockHuff encodes a block of bytes as either +// Huffman encoded literals or uncompressed bytes if the +// results only gains very little from compression. +func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) { + if w.err != nil { + return + } + + // Clear histogram + for i := range w.literalFreq[:] { + w.literalFreq[i] = 0 + } + if !w.lastHuffMan { + for i := range w.offsetFreq[:] { + w.offsetFreq[i] = 0 + } + } + + // Add everything as literals + // 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 { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + + if w.lastHeader > 0 { + reuseSize := w.literalEncoding.bitLength(w.literalFreq[:256]) + estBits += estExtra + + if estBits < reuseSize { + // We owe an EOB + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 + } + } + + const numLiterals = endBlockMarker + 1 + const numOffsets = 1 + if w.lastHeader == 0 { + w.literalFreq[endBlockMarker] = 1 + w.literalEncoding.generate(w.literalFreq[:numLiterals], 15) + + // Generate codegen and codegenFrequencies, which indicates how to encode + // the literalEncoding and the offsetEncoding. + w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, huffOffset) + w.codegenEncoding.generate(w.codegenFreq[:], 7) + numCodegens := w.codegens() + + // Huffman. + w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) + w.lastHuffMan = true + w.lastHeader, _ = w.headerSize() + } + + encoding := w.literalEncoding.codes[:257] + for _, t := range input { + // Bitwriting inlined, ~30% speedup + c := encoding[t] + w.bits |= uint64(c.code) << ((w.nbits) & 63) + w.nbits += c.len + if w.nbits >= 48 { + bits := w.bits + w.bits >>= 48 + w.nbits -= 48 + n := w.nbytes + w.bytes[n] = byte(bits) + w.bytes[n+1] = byte(bits >> 8) + w.bytes[n+2] = byte(bits >> 16) + w.bytes[n+3] = byte(bits >> 24) + w.bytes[n+4] = byte(bits >> 32) + w.bytes[n+5] = byte(bits >> 40) + n += 6 + if n >= bufferFlushSize { + if w.err != nil { + n = 0 + return + } + w.write(w.bytes[:n]) + n = 0 + } + w.nbytes = n + } + } + if eof || sync { + w.writeCode(encoding[endBlockMarker]) + w.lastHeader = 0 + w.lastHuffMan = false + } +} diff --git a/github.com/klauspost/compress/flate/huffman_bit_writer_test.go b/github.com/klauspost/compress/flate/huffman_bit_writer_test.go new file mode 100644 index 0000000..60aef15 --- /dev/null +++ b/github.com/klauspost/compress/flate/huffman_bit_writer_test.go @@ -0,0 +1,382 @@ +// 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. + +package flate + +import ( + "bytes" + "flag" + "fmt" + "io/ioutil" + "os" + "path/filepath" + "strings" + "testing" +) + +var update = flag.Bool("update", false, "update reference files") + +// TestBlockHuff tests huffman encoding against reference files +// to detect possible regressions. +// If encoding/bit allocation changes you can regenerate these files +// by using the -update flag. +func TestBlockHuff(t *testing.T) { + // determine input files + match, err := filepath.Glob("testdata/huffman-*.in") + if err != nil { + t.Fatal(err) + } + + for _, in := range match { + out := in // for files where input and output are identical + if strings.HasSuffix(in, ".in") { + out = in[:len(in)-len(".in")] + ".golden" + } + t.Run(in, func(t *testing.T) { + testBlockHuff(t, in, out) + }) + } +} + +func testBlockHuff(t *testing.T, in, out string) { + all, err := ioutil.ReadFile(in) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + bw := newHuffmanBitWriter(&buf) + bw.logNewTablePenalty = 8 + bw.writeBlockHuff(false, all, false) + bw.flush() + got := buf.Bytes() + + want, err := ioutil.ReadFile(out) + if err != nil && !*update { + t.Error(err) + return + } + + t.Logf("Testing %q", in) + if !bytes.Equal(got, want) { + if *update { + if in != out { + t.Logf("Updating %q", out) + if err := ioutil.WriteFile(out, got, 0666); err != nil { + t.Error(err) + } + return + } + // in == out: don't accidentally destroy input + t.Errorf("WARNING: -update did not rewrite input file %s", in) + } + + t.Errorf("%q != %q (see %q)", in, out, in+".got") + if err := ioutil.WriteFile(in+".got", got, 0666); err != nil { + t.Error(err) + } + return + } + t.Log("Output ok") + + // Test if the writer produces the same output after reset. + buf.Reset() + bw.reset(&buf) + bw.writeBlockHuff(false, all, false) + bw.flush() + got = buf.Bytes() + if !bytes.Equal(got, want) { + t.Errorf("after reset %q != %q (see %q)", in, out, in+".reset.got") + if err := ioutil.WriteFile(in+".reset.got", got, 0666); err != nil { + t.Error(err) + } + return + } + t.Log("Reset ok") + testWriterEOF(t, "huff", huffTest{input: in}, true) +} + +type huffTest struct { + tokens []token + input string // File name of input data matching the tokens. + want string // File name of data with the expected output with input available. + wantNoInput string // File name of the expected output when no input is available. +} + +const ml = 0x7fc00000 // Maximum length token. Used to reduce the size of writeBlockTests + +var writeBlockTests = []huffTest{ + { + input: "testdata/huffman-null-max.in", + want: "testdata/huffman-null-max.%s.expect", + wantNoInput: "testdata/huffman-null-max.%s.expect-noinput", + tokens: []token{0x0, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, 0x0, 0x0}, + }, + { + input: "testdata/huffman-pi.in", + want: "testdata/huffman-pi.%s.expect", + wantNoInput: "testdata/huffman-pi.%s.expect-noinput", + tokens: []token{0x33, 0x2e, 0x31, 0x34, 0x31, 0x35, 0x39, 0x32, 0x36, 0x35, 0x33, 0x35, 0x38, 0x39, 0x37, 0x39, 0x33, 0x32, 0x33, 0x38, 0x34, 0x36, 0x32, 0x36, 0x34, 0x33, 0x33, 0x38, 0x33, 0x32, 0x37, 0x39, 0x35, 0x30, 0x32, 0x38, 0x38, 0x34, 0x31, 0x39, 0x37, 0x31, 0x36, 0x39, 0x33, 0x39, 0x39, 0x33, 0x37, 0x35, 0x31, 0x30, 0x35, 0x38, 0x32, 0x30, 0x39, 0x37, 0x34, 0x39, 0x34, 0x34, 0x35, 0x39, 0x32, 0x33, 0x30, 0x37, 0x38, 0x31, 0x36, 0x34, 0x30, 0x36, 0x32, 0x38, 0x36, 0x32, 0x30, 0x38, 0x39, 0x39, 0x38, 0x36, 0x32, 0x38, 0x30, 0x33, 0x34, 0x38, 0x32, 0x35, 0x33, 0x34, 0x32, 0x31, 0x31, 0x37, 0x30, 0x36, 0x37, 0x39, 0x38, 0x32, 0x31, 0x34, 0x38, 0x30, 0x38, 0x36, 0x35, 0x31, 0x33, 0x32, 0x38, 0x32, 0x33, 0x30, 0x36, 0x36, 0x34, 0x37, 0x30, 0x39, 0x33, 0x38, 0x34, 0x34, 0x36, 0x30, 0x39, 0x35, 0x35, 0x30, 0x35, 0x38, 0x32, 0x32, 0x33, 0x31, 0x37, 0x32, 0x35, 0x33, 0x35, 0x39, 0x34, 0x30, 0x38, 0x31, 0x32, 0x38, 0x34, 0x38, 0x31, 0x31, 0x31, 0x37, 0x34, 0x4040007e, 0x34, 0x31, 0x30, 0x32, 0x37, 0x30, 0x31, 0x39, 0x33, 0x38, 0x35, 0x32, 0x31, 0x31, 0x30, 0x35, 0x35, 0x35, 0x39, 0x36, 0x34, 0x34, 0x36, 0x32, 0x32, 0x39, 0x34, 0x38, 0x39, 0x35, 0x34, 0x39, 0x33, 0x30, 0x33, 0x38, 0x31, 0x40400012, 0x32, 0x38, 0x38, 0x31, 0x30, 0x39, 0x37, 0x35, 0x36, 0x36, 0x35, 0x39, 0x33, 0x33, 0x34, 0x34, 0x36, 0x40400047, 0x37, 0x35, 0x36, 0x34, 0x38, 0x32, 0x33, 0x33, 0x37, 0x38, 0x36, 0x37, 0x38, 0x33, 0x31, 0x36, 0x35, 0x32, 0x37, 0x31, 0x32, 0x30, 0x31, 0x39, 0x30, 0x39, 0x31, 0x34, 0x4040001a, 0x35, 0x36, 0x36, 0x39, 0x32, 0x33, 0x34, 0x36, 0x404000b2, 0x36, 0x31, 0x30, 0x34, 0x35, 0x34, 0x33, 0x32, 0x36, 0x40400032, 0x31, 0x33, 0x33, 0x39, 0x33, 0x36, 0x30, 0x37, 0x32, 0x36, 0x30, 0x32, 0x34, 0x39, 0x31, 0x34, 0x31, 0x32, 0x37, 0x33, 0x37, 0x32, 0x34, 0x35, 0x38, 0x37, 0x30, 0x30, 0x36, 0x36, 0x30, 0x36, 0x33, 0x31, 0x35, 0x35, 0x38, 0x38, 0x31, 0x37, 0x34, 0x38, 0x38, 0x31, 0x35, 0x32, 0x30, 0x39, 0x32, 0x30, 0x39, 0x36, 0x32, 0x38, 0x32, 0x39, 0x32, 0x35, 0x34, 0x30, 0x39, 0x31, 0x37, 0x31, 0x35, 0x33, 0x36, 0x34, 0x33, 0x36, 0x37, 0x38, 0x39, 0x32, 0x35, 0x39, 0x30, 0x33, 0x36, 0x30, 0x30, 0x31, 0x31, 0x33, 0x33, 0x30, 0x35, 0x33, 0x30, 0x35, 0x34, 0x38, 0x38, 0x32, 0x30, 0x34, 0x36, 0x36, 0x35, 0x32, 0x31, 0x33, 0x38, 0x34, 0x31, 0x34, 0x36, 0x39, 0x35, 0x31, 0x39, 0x34, 0x31, 0x35, 0x31, 0x31, 0x36, 0x30, 0x39, 0x34, 0x33, 0x33, 0x30, 0x35, 0x37, 0x32, 0x37, 0x30, 0x33, 0x36, 0x35, 0x37, 0x35, 0x39, 0x35, 0x39, 0x31, 0x39, 0x35, 0x33, 0x30, 0x39, 0x32, 0x31, 0x38, 0x36, 0x31, 0x31, 0x37, 0x404000e9, 0x33, 0x32, 0x40400009, 0x39, 0x33, 0x31, 0x30, 0x35, 0x31, 0x31, 0x38, 0x35, 0x34, 0x38, 0x30, 0x37, 0x4040010e, 0x33, 0x37, 0x39, 0x39, 0x36, 0x32, 0x37, 0x34, 0x39, 0x35, 0x36, 0x37, 0x33, 0x35, 0x31, 0x38, 0x38, 0x35, 0x37, 0x35, 0x32, 0x37, 0x32, 0x34, 0x38, 0x39, 0x31, 0x32, 0x32, 0x37, 0x39, 0x33, 0x38, 0x31, 0x38, 0x33, 0x30, 0x31, 0x31, 0x39, 0x34, 0x39, 0x31, 0x32, 0x39, 0x38, 0x33, 0x33, 0x36, 0x37, 0x33, 0x33, 0x36, 0x32, 0x34, 0x34, 0x30, 0x36, 0x35, 0x36, 0x36, 0x34, 0x33, 0x30, 0x38, 0x36, 0x30, 0x32, 0x31, 0x33, 0x39, 0x34, 0x39, 0x34, 0x36, 0x33, 0x39, 0x35, 0x32, 0x32, 0x34, 0x37, 0x33, 0x37, 0x31, 0x39, 0x30, 0x37, 0x30, 0x32, 0x31, 0x37, 0x39, 0x38, 0x40800099, 0x37, 0x30, 0x32, 0x37, 0x37, 0x30, 0x35, 0x33, 0x39, 0x32, 0x31, 0x37, 0x31, 0x37, 0x36, 0x32, 0x39, 0x33, 0x31, 0x37, 0x36, 0x37, 0x35, 0x40800232, 0x37, 0x34, 0x38, 0x31, 0x40400006, 0x36, 0x36, 0x39, 0x34, 0x30, 0x404001e7, 0x30, 0x30, 0x30, 0x35, 0x36, 0x38, 0x31, 0x32, 0x37, 0x31, 0x34, 0x35, 0x32, 0x36, 0x33, 0x35, 0x36, 0x30, 0x38, 0x32, 0x37, 0x37, 0x38, 0x35, 0x37, 0x37, 0x31, 0x33, 0x34, 0x32, 0x37, 0x35, 0x37, 0x37, 0x38, 0x39, 0x36, 0x40400129, 0x33, 0x36, 0x33, 0x37, 0x31, 0x37, 0x38, 0x37, 0x32, 0x31, 0x34, 0x36, 0x38, 0x34, 0x34, 0x30, 0x39, 0x30, 0x31, 0x32, 0x32, 0x34, 0x39, 0x35, 0x33, 0x34, 0x33, 0x30, 0x31, 0x34, 0x36, 0x35, 0x34, 0x39, 0x35, 0x38, 0x35, 0x33, 0x37, 0x31, 0x30, 0x35, 0x30, 0x37, 0x39, 0x404000ca, 0x36, 0x40400153, 0x38, 0x39, 0x32, 0x33, 0x35, 0x34, 0x404001c9, 0x39, 0x35, 0x36, 0x31, 0x31, 0x32, 0x31, 0x32, 0x39, 0x30, 0x32, 0x31, 0x39, 0x36, 0x30, 0x38, 0x36, 0x34, 0x30, 0x33, 0x34, 0x34, 0x31, 0x38, 0x31, 0x35, 0x39, 0x38, 0x31, 0x33, 0x36, 0x32, 0x39, 0x37, 0x37, 0x34, 0x40400074, 0x30, 0x39, 0x39, 0x36, 0x30, 0x35, 0x31, 0x38, 0x37, 0x30, 0x37, 0x32, 0x31, 0x31, 0x33, 0x34, 0x39, 0x40800000, 0x38, 0x33, 0x37, 0x32, 0x39, 0x37, 0x38, 0x30, 0x34, 0x39, 0x39, 0x404002da, 0x39, 0x37, 0x33, 0x31, 0x37, 0x33, 0x32, 0x38, 0x4040018a, 0x36, 0x33, 0x31, 0x38, 0x35, 0x40400301, 0x404002e8, 0x34, 0x35, 0x35, 0x33, 0x34, 0x36, 0x39, 0x30, 0x38, 0x33, 0x30, 0x32, 0x36, 0x34, 0x32, 0x35, 0x32, 0x32, 0x33, 0x30, 0x404002e3, 0x40400267, 0x38, 0x35, 0x30, 0x33, 0x35, 0x32, 0x36, 0x31, 0x39, 0x33, 0x31, 0x31, 0x40400212, 0x31, 0x30, 0x31, 0x30, 0x30, 0x30, 0x33, 0x31, 0x33, 0x37, 0x38, 0x33, 0x38, 0x37, 0x35, 0x32, 0x38, 0x38, 0x36, 0x35, 0x38, 0x37, 0x35, 0x33, 0x33, 0x32, 0x30, 0x38, 0x33, 0x38, 0x31, 0x34, 0x32, 0x30, 0x36, 0x40400140, 0x4040012b, 0x31, 0x34, 0x37, 0x33, 0x30, 0x33, 0x35, 0x39, 0x4080032e, 0x39, 0x30, 0x34, 0x32, 0x38, 0x37, 0x35, 0x35, 0x34, 0x36, 0x38, 0x37, 0x33, 0x31, 0x31, 0x35, 0x39, 0x35, 0x40400355, 0x33, 0x38, 0x38, 0x32, 0x33, 0x35, 0x33, 0x37, 0x38, 0x37, 0x35, 0x4080037f, 0x39, 0x4040013a, 0x31, 0x40400148, 0x38, 0x30, 0x35, 0x33, 0x4040018a, 0x32, 0x32, 0x36, 0x38, 0x30, 0x36, 0x36, 0x31, 0x33, 0x30, 0x30, 0x31, 0x39, 0x32, 0x37, 0x38, 0x37, 0x36, 0x36, 0x31, 0x31, 0x31, 0x39, 0x35, 0x39, 0x40400237, 0x36, 0x40800124, 0x38, 0x39, 0x33, 0x38, 0x30, 0x39, 0x35, 0x32, 0x35, 0x37, 0x32, 0x30, 0x31, 0x30, 0x36, 0x35, 0x34, 0x38, 0x35, 0x38, 0x36, 0x33, 0x32, 0x37, 0x4040009a, 0x39, 0x33, 0x36, 0x31, 0x35, 0x33, 0x40400220, 0x4080015c, 0x32, 0x33, 0x30, 0x33, 0x30, 0x31, 0x39, 0x35, 0x32, 0x30, 0x33, 0x35, 0x33, 0x30, 0x31, 0x38, 0x35, 0x32, 0x40400171, 0x40400075, 0x33, 0x36, 0x32, 0x32, 0x35, 0x39, 0x39, 0x34, 0x31, 0x33, 0x40400254, 0x34, 0x39, 0x37, 0x32, 0x31, 0x37, 0x404000de, 0x33, 0x34, 0x37, 0x39, 0x31, 0x33, 0x31, 0x35, 0x31, 0x35, 0x35, 0x37, 0x34, 0x38, 0x35, 0x37, 0x32, 0x34, 0x32, 0x34, 0x35, 0x34, 0x31, 0x35, 0x30, 0x36, 0x39, 0x4040013f, 0x38, 0x32, 0x39, 0x35, 0x33, 0x33, 0x31, 0x31, 0x36, 0x38, 0x36, 0x31, 0x37, 0x32, 0x37, 0x38, 0x40400337, 0x39, 0x30, 0x37, 0x35, 0x30, 0x39, 0x4040010d, 0x37, 0x35, 0x34, 0x36, 0x33, 0x37, 0x34, 0x36, 0x34, 0x39, 0x33, 0x39, 0x33, 0x31, 0x39, 0x32, 0x35, 0x35, 0x30, 0x36, 0x30, 0x34, 0x30, 0x30, 0x39, 0x4040026b, 0x31, 0x36, 0x37, 0x31, 0x31, 0x33, 0x39, 0x30, 0x30, 0x39, 0x38, 0x40400335, 0x34, 0x30, 0x31, 0x32, 0x38, 0x35, 0x38, 0x33, 0x36, 0x31, 0x36, 0x30, 0x33, 0x35, 0x36, 0x33, 0x37, 0x30, 0x37, 0x36, 0x36, 0x30, 0x31, 0x30, 0x34, 0x40400172, 0x38, 0x31, 0x39, 0x34, 0x32, 0x39, 0x4080041e, 0x404000ef, 0x4040028b, 0x37, 0x38, 0x33, 0x37, 0x34, 0x404004a8, 0x38, 0x32, 0x35, 0x35, 0x33, 0x37, 0x40800209, 0x32, 0x36, 0x38, 0x4040002e, 0x34, 0x30, 0x34, 0x37, 0x404001d1, 0x34, 0x404004b5, 0x4040038d, 0x38, 0x34, 0x404003a8, 0x36, 0x40c0031f, 0x33, 0x33, 0x31, 0x33, 0x36, 0x37, 0x37, 0x30, 0x32, 0x38, 0x39, 0x38, 0x39, 0x31, 0x35, 0x32, 0x40400062, 0x35, 0x32, 0x31, 0x36, 0x32, 0x30, 0x35, 0x36, 0x39, 0x36, 0x40400411, 0x30, 0x35, 0x38, 0x40400477, 0x35, 0x40400498, 0x35, 0x31, 0x31, 0x40400209, 0x38, 0x32, 0x34, 0x33, 0x30, 0x30, 0x33, 0x35, 0x35, 0x38, 0x37, 0x36, 0x34, 0x30, 0x32, 0x34, 0x37, 0x34, 0x39, 0x36, 0x34, 0x37, 0x33, 0x32, 0x36, 0x33, 0x4040043e, 0x39, 0x39, 0x32, 0x4040044b, 0x34, 0x32, 0x36, 0x39, 0x40c002c5, 0x37, 0x404001d6, 0x34, 0x4040053d, 0x4040041d, 0x39, 0x33, 0x34, 0x31, 0x37, 0x404001ad, 0x31, 0x32, 0x4040002a, 0x34, 0x4040019e, 0x31, 0x35, 0x30, 0x33, 0x30, 0x32, 0x38, 0x36, 0x31, 0x38, 0x32, 0x39, 0x37, 0x34, 0x35, 0x35, 0x35, 0x37, 0x30, 0x36, 0x37, 0x34, 0x40400135, 0x35, 0x30, 0x35, 0x34, 0x39, 0x34, 0x35, 0x38, 0x404001c5, 0x39, 0x40400051, 0x35, 0x36, 0x404001ec, 0x37, 0x32, 0x31, 0x30, 0x37, 0x39, 0x40400159, 0x33, 0x30, 0x4040010a, 0x33, 0x32, 0x31, 0x31, 0x36, 0x35, 0x33, 0x34, 0x34, 0x39, 0x38, 0x37, 0x32, 0x30, 0x32, 0x37, 0x4040011b, 0x30, 0x32, 0x33, 0x36, 0x34, 0x4040022e, 0x35, 0x34, 0x39, 0x39, 0x31, 0x31, 0x39, 0x38, 0x40400418, 0x34, 0x4040011b, 0x35, 0x33, 0x35, 0x36, 0x36, 0x33, 0x36, 0x39, 0x40400450, 0x32, 0x36, 0x35, 0x404002e4, 0x37, 0x38, 0x36, 0x32, 0x35, 0x35, 0x31, 0x404003da, 0x31, 0x37, 0x35, 0x37, 0x34, 0x36, 0x37, 0x32, 0x38, 0x39, 0x30, 0x39, 0x37, 0x37, 0x37, 0x37, 0x40800453, 0x30, 0x30, 0x30, 0x404005fd, 0x37, 0x30, 0x404004df, 0x36, 0x404003e9, 0x34, 0x39, 0x31, 0x4040041e, 0x40400297, 0x32, 0x31, 0x34, 0x37, 0x37, 0x32, 0x33, 0x35, 0x30, 0x31, 0x34, 0x31, 0x34, 0x40400643, 0x33, 0x35, 0x36, 0x404004af, 0x31, 0x36, 0x31, 0x33, 0x36, 0x31, 0x31, 0x35, 0x37, 0x33, 0x35, 0x32, 0x35, 0x40400504, 0x33, 0x34, 0x4040005b, 0x31, 0x38, 0x4040047b, 0x38, 0x34, 0x404005e7, 0x33, 0x33, 0x32, 0x33, 0x39, 0x30, 0x37, 0x33, 0x39, 0x34, 0x31, 0x34, 0x33, 0x33, 0x33, 0x34, 0x35, 0x34, 0x37, 0x37, 0x36, 0x32, 0x34, 0x40400242, 0x32, 0x35, 0x31, 0x38, 0x39, 0x38, 0x33, 0x35, 0x36, 0x39, 0x34, 0x38, 0x35, 0x35, 0x36, 0x32, 0x30, 0x39, 0x39, 0x32, 0x31, 0x39, 0x32, 0x32, 0x32, 0x31, 0x38, 0x34, 0x32, 0x37, 0x4040023e, 0x32, 0x404000ba, 0x36, 0x38, 0x38, 0x37, 0x36, 0x37, 0x31, 0x37, 0x39, 0x30, 0x40400055, 0x30, 0x40800106, 0x36, 0x36, 0x404003e7, 0x38, 0x38, 0x36, 0x32, 0x37, 0x32, 0x404006dc, 0x31, 0x37, 0x38, 0x36, 0x30, 0x38, 0x35, 0x37, 0x40400073, 0x33, 0x408002fc, 0x37, 0x39, 0x37, 0x36, 0x36, 0x38, 0x31, 0x404002bd, 0x30, 0x30, 0x39, 0x35, 0x33, 0x38, 0x38, 0x40400638, 0x33, 0x404006a5, 0x30, 0x36, 0x38, 0x30, 0x30, 0x36, 0x34, 0x32, 0x32, 0x35, 0x31, 0x32, 0x35, 0x32, 0x4040057b, 0x37, 0x33, 0x39, 0x32, 0x40400297, 0x40400474, 0x34, 0x408006b3, 0x38, 0x36, 0x32, 0x36, 0x39, 0x34, 0x35, 0x404001e5, 0x34, 0x31, 0x39, 0x36, 0x35, 0x32, 0x38, 0x35, 0x30, 0x40400099, 0x4040039c, 0x31, 0x38, 0x36, 0x33, 0x404001be, 0x34, 0x40800154, 0x32, 0x30, 0x33, 0x39, 0x4040058b, 0x34, 0x35, 0x404002bc, 0x32, 0x33, 0x37, 0x4040042c, 0x36, 0x40400510, 0x35, 0x36, 0x40400638, 0x37, 0x31, 0x39, 0x31, 0x37, 0x32, 0x38, 0x40400171, 0x37, 0x36, 0x34, 0x36, 0x35, 0x37, 0x35, 0x37, 0x33, 0x39, 0x40400101, 0x33, 0x38, 0x39, 0x40400748, 0x38, 0x33, 0x32, 0x36, 0x34, 0x35, 0x39, 0x39, 0x35, 0x38, 0x404006a7, 0x30, 0x34, 0x37, 0x38, 0x404001de, 0x40400328, 0x39, 0x4040002d, 0x36, 0x34, 0x30, 0x37, 0x38, 0x39, 0x35, 0x31, 0x4040008e, 0x36, 0x38, 0x33, 0x4040012f, 0x32, 0x35, 0x39, 0x35, 0x37, 0x30, 0x40400468, 0x38, 0x32, 0x32, 0x404002c8, 0x32, 0x4040061b, 0x34, 0x30, 0x37, 0x37, 0x32, 0x36, 0x37, 0x31, 0x39, 0x34, 0x37, 0x38, 0x40400319, 0x38, 0x32, 0x36, 0x30, 0x31, 0x34, 0x37, 0x36, 0x39, 0x39, 0x30, 0x39, 0x404004e8, 0x30, 0x31, 0x33, 0x36, 0x33, 0x39, 0x34, 0x34, 0x33, 0x4040027f, 0x33, 0x30, 0x40400105, 0x32, 0x30, 0x33, 0x34, 0x39, 0x36, 0x32, 0x35, 0x32, 0x34, 0x35, 0x31, 0x37, 0x404003b5, 0x39, 0x36, 0x35, 0x31, 0x34, 0x33, 0x31, 0x34, 0x32, 0x39, 0x38, 0x30, 0x39, 0x31, 0x39, 0x30, 0x36, 0x35, 0x39, 0x32, 0x40400282, 0x37, 0x32, 0x32, 0x31, 0x36, 0x39, 0x36, 0x34, 0x36, 0x40400419, 0x4040007a, 0x35, 0x4040050e, 0x34, 0x40800565, 0x38, 0x40400559, 0x39, 0x37, 0x4040057b, 0x35, 0x34, 0x4040049d, 0x4040023e, 0x37, 0x4040065a, 0x38, 0x34, 0x36, 0x38, 0x31, 0x33, 0x4040008c, 0x36, 0x38, 0x33, 0x38, 0x36, 0x38, 0x39, 0x34, 0x32, 0x37, 0x37, 0x34, 0x31, 0x35, 0x35, 0x39, 0x39, 0x31, 0x38, 0x35, 0x4040005a, 0x32, 0x34, 0x35, 0x39, 0x35, 0x33, 0x39, 0x35, 0x39, 0x34, 0x33, 0x31, 0x404005b7, 0x37, 0x40400012, 0x36, 0x38, 0x30, 0x38, 0x34, 0x35, 0x404002e7, 0x37, 0x33, 0x4040081e, 0x39, 0x35, 0x38, 0x34, 0x38, 0x36, 0x35, 0x33, 0x38, 0x404006e8, 0x36, 0x32, 0x404000f2, 0x36, 0x30, 0x39, 0x404004b6, 0x36, 0x30, 0x38, 0x30, 0x35, 0x31, 0x32, 0x34, 0x33, 0x38, 0x38, 0x34, 0x4040013a, 0x4040000b, 0x34, 0x31, 0x33, 0x4040030f, 0x37, 0x36, 0x32, 0x37, 0x38, 0x40400341, 0x37, 0x31, 0x35, 0x4040059b, 0x33, 0x35, 0x39, 0x39, 0x37, 0x37, 0x30, 0x30, 0x31, 0x32, 0x39, 0x40400472, 0x38, 0x39, 0x34, 0x34, 0x31, 0x40400277, 0x36, 0x38, 0x35, 0x35, 0x4040005f, 0x34, 0x30, 0x36, 0x33, 0x404008e6, 0x32, 0x30, 0x37, 0x32, 0x32, 0x40400158, 0x40800203, 0x34, 0x38, 0x31, 0x35, 0x38, 0x40400205, 0x404001fe, 0x4040027a, 0x40400298, 0x33, 0x39, 0x34, 0x35, 0x32, 0x32, 0x36, 0x37, 0x40c00496, 0x38, 0x4040058a, 0x32, 0x31, 0x404002ea, 0x32, 0x40400387, 0x35, 0x34, 0x36, 0x36, 0x36, 0x4040051b, 0x32, 0x33, 0x39, 0x38, 0x36, 0x34, 0x35, 0x36, 0x404004c4, 0x31, 0x36, 0x33, 0x35, 0x40800253, 0x40400811, 0x37, 0x404008ad, 0x39, 0x38, 0x4040045e, 0x39, 0x33, 0x36, 0x33, 0x34, 0x4040075b, 0x37, 0x34, 0x33, 0x32, 0x34, 0x4040047b, 0x31, 0x35, 0x30, 0x37, 0x36, 0x404004bb, 0x37, 0x39, 0x34, 0x35, 0x31, 0x30, 0x39, 0x4040003e, 0x30, 0x39, 0x34, 0x30, 0x404006a6, 0x38, 0x38, 0x37, 0x39, 0x37, 0x31, 0x30, 0x38, 0x39, 0x33, 0x404008f0, 0x36, 0x39, 0x31, 0x33, 0x36, 0x38, 0x36, 0x37, 0x32, 0x4040025b, 0x404001fe, 0x35, 0x4040053f, 0x40400468, 0x40400801, 0x31, 0x37, 0x39, 0x32, 0x38, 0x36, 0x38, 0x404008cc, 0x38, 0x37, 0x34, 0x37, 0x4080079e, 0x38, 0x32, 0x34, 0x4040097a, 0x38, 0x4040025b, 0x37, 0x31, 0x34, 0x39, 0x30, 0x39, 0x36, 0x37, 0x35, 0x39, 0x38, 0x404006ef, 0x33, 0x36, 0x35, 0x40400134, 0x38, 0x31, 0x4040005c, 0x40400745, 0x40400936, 0x36, 0x38, 0x32, 0x39, 0x4040057e, 0x38, 0x37, 0x32, 0x32, 0x36, 0x35, 0x38, 0x38, 0x30, 0x40400611, 0x35, 0x40400249, 0x34, 0x32, 0x37, 0x30, 0x34, 0x37, 0x37, 0x35, 0x35, 0x4040081e, 0x33, 0x37, 0x39, 0x36, 0x34, 0x31, 0x34, 0x35, 0x31, 0x35, 0x32, 0x404005fd, 0x32, 0x33, 0x34, 0x33, 0x36, 0x34, 0x35, 0x34, 0x404005de, 0x34, 0x34, 0x34, 0x37, 0x39, 0x35, 0x4040003c, 0x40400523, 0x408008e6, 0x34, 0x31, 0x4040052a, 0x33, 0x40400304, 0x35, 0x32, 0x33, 0x31, 0x40800841, 0x31, 0x36, 0x36, 0x31, 0x404008b2, 0x35, 0x39, 0x36, 0x39, 0x35, 0x33, 0x36, 0x32, 0x33, 0x31, 0x34, 0x404005ff, 0x32, 0x34, 0x38, 0x34, 0x39, 0x33, 0x37, 0x31, 0x38, 0x37, 0x31, 0x31, 0x30, 0x31, 0x34, 0x35, 0x37, 0x36, 0x35, 0x34, 0x40400761, 0x30, 0x32, 0x37, 0x39, 0x39, 0x33, 0x34, 0x34, 0x30, 0x33, 0x37, 0x34, 0x32, 0x30, 0x30, 0x37, 0x4040093f, 0x37, 0x38, 0x35, 0x33, 0x39, 0x30, 0x36, 0x32, 0x31, 0x39, 0x40800299, 0x40400345, 0x38, 0x34, 0x37, 0x408003d2, 0x38, 0x33, 0x33, 0x32, 0x31, 0x34, 0x34, 0x35, 0x37, 0x31, 0x40400284, 0x40400776, 0x34, 0x33, 0x35, 0x30, 0x40400928, 0x40400468, 0x35, 0x33, 0x31, 0x39, 0x31, 0x30, 0x34, 0x38, 0x34, 0x38, 0x31, 0x30, 0x30, 0x35, 0x33, 0x37, 0x30, 0x36, 0x404008bc, 0x4080059d, 0x40800781, 0x31, 0x40400559, 0x37, 0x4040031b, 0x35, 0x404007ec, 0x4040040c, 0x36, 0x33, 0x408007dc, 0x34, 0x40400971, 0x4080034e, 0x408003f5, 0x38, 0x4080052d, 0x40800887, 0x39, 0x40400187, 0x39, 0x31, 0x404008ce, 0x38, 0x31, 0x34, 0x36, 0x37, 0x35, 0x31, 0x4040062b, 0x31, 0x32, 0x33, 0x39, 0x40c001a9, 0x39, 0x30, 0x37, 0x31, 0x38, 0x36, 0x34, 0x39, 0x34, 0x32, 0x33, 0x31, 0x39, 0x36, 0x31, 0x35, 0x36, 0x404001ec, 0x404006bc, 0x39, 0x35, 0x40400926, 0x40400469, 0x4040011b, 0x36, 0x30, 0x33, 0x38, 0x40400a25, 0x4040016f, 0x40400384, 0x36, 0x32, 0x4040045a, 0x35, 0x4040084c, 0x36, 0x33, 0x38, 0x39, 0x33, 0x37, 0x37, 0x38, 0x37, 0x404008c5, 0x404000f8, 0x39, 0x37, 0x39, 0x32, 0x30, 0x37, 0x37, 0x33, 0x404005d7, 0x32, 0x31, 0x38, 0x32, 0x35, 0x36, 0x404007df, 0x36, 0x36, 0x404006d6, 0x34, 0x32, 0x4080067e, 0x36, 0x404006e6, 0x34, 0x34, 0x40400024, 0x35, 0x34, 0x39, 0x32, 0x30, 0x32, 0x36, 0x30, 0x35, 0x40400ab3, 0x408003e4, 0x32, 0x30, 0x31, 0x34, 0x39, 0x404004d2, 0x38, 0x35, 0x30, 0x37, 0x33, 0x40400599, 0x36, 0x36, 0x36, 0x30, 0x40400194, 0x32, 0x34, 0x33, 0x34, 0x30, 0x40400087, 0x30, 0x4040076b, 0x38, 0x36, 0x33, 0x40400956, 0x404007e4, 0x4040042b, 0x40400174, 0x35, 0x37, 0x39, 0x36, 0x32, 0x36, 0x38, 0x35, 0x36, 0x40400140, 0x35, 0x30, 0x38, 0x40400523, 0x35, 0x38, 0x37, 0x39, 0x36, 0x39, 0x39, 0x40400711, 0x35, 0x37, 0x34, 0x40400a18, 0x38, 0x34, 0x30, 0x404008b3, 0x31, 0x34, 0x35, 0x39, 0x31, 0x4040078c, 0x37, 0x30, 0x40400234, 0x30, 0x31, 0x40400be7, 0x31, 0x32, 0x40400c74, 0x30, 0x404003c3, 0x33, 0x39, 0x40400b2a, 0x40400112, 0x37, 0x31, 0x35, 0x404003b0, 0x34, 0x32, 0x30, 0x40800bf2, 0x39, 0x40400bc2, 0x30, 0x37, 0x40400341, 0x40400795, 0x40400aaf, 0x40400c62, 0x32, 0x31, 0x40400960, 0x32, 0x35, 0x31, 0x4040057b, 0x40400944, 0x39, 0x32, 0x404001b2, 0x38, 0x32, 0x36, 0x40400b66, 0x32, 0x40400278, 0x33, 0x32, 0x31, 0x35, 0x37, 0x39, 0x31, 0x39, 0x38, 0x34, 0x31, 0x34, 0x4080087b, 0x39, 0x31, 0x36, 0x34, 0x408006e8, 0x39, 0x40800b58, 0x404008db, 0x37, 0x32, 0x32, 0x40400321, 0x35, 0x404008a4, 0x40400141, 0x39, 0x31, 0x30, 0x404000bc, 0x40400c5b, 0x35, 0x32, 0x38, 0x30, 0x31, 0x37, 0x40400231, 0x37, 0x31, 0x32, 0x40400914, 0x38, 0x33, 0x32, 0x40400373, 0x31, 0x40400589, 0x30, 0x39, 0x33, 0x35, 0x33, 0x39, 0x36, 0x35, 0x37, 0x4040064b, 0x31, 0x30, 0x38, 0x33, 0x40400069, 0x35, 0x31, 0x4040077a, 0x40400d5a, 0x31, 0x34, 0x34, 0x34, 0x32, 0x31, 0x30, 0x30, 0x40400202, 0x30, 0x33, 0x4040019c, 0x31, 0x31, 0x30, 0x33, 0x40400c81, 0x40400009, 0x40400026, 0x40c00602, 0x35, 0x31, 0x36, 0x404005d9, 0x40800883, 0x4040092a, 0x35, 0x40800c42, 0x38, 0x35, 0x31, 0x37, 0x31, 0x34, 0x33, 0x37, 0x40400605, 0x4040006d, 0x31, 0x35, 0x35, 0x36, 0x35, 0x30, 0x38, 0x38, 0x404003b9, 0x39, 0x38, 0x39, 0x38, 0x35, 0x39, 0x39, 0x38, 0x32, 0x33, 0x38, 0x404001cf, 0x404009ba, 0x33, 0x4040016c, 0x4040043e, 0x404009c3, 0x38, 0x40800e05, 0x33, 0x32, 0x40400107, 0x35, 0x40400305, 0x33, 0x404001ca, 0x39, 0x4040041b, 0x39, 0x38, 0x4040087d, 0x34, 0x40400cb8, 0x37, 0x4040064b, 0x30, 0x37, 0x404000e5, 0x34, 0x38, 0x31, 0x34, 0x31, 0x40400539, 0x38, 0x35, 0x39, 0x34, 0x36, 0x31, 0x40400bc9, 0x38, 0x30}, + }, + { + input: "testdata/huffman-rand-1k.in", + want: "testdata/huffman-rand-1k.%s.expect", + wantNoInput: "testdata/huffman-rand-1k.%s.expect-noinput", + tokens: []token{0xf8, 0x8b, 0x96, 0x76, 0x48, 0xd, 0x85, 0x94, 0x25, 0x80, 0xaf, 0xc2, 0xfe, 0x8d, 0xe8, 0x20, 0xeb, 0x17, 0x86, 0xc9, 0xb7, 0xc5, 0xde, 0x6, 0xea, 0x7d, 0x18, 0x8b, 0xe7, 0x3e, 0x7, 0xda, 0xdf, 0xff, 0x6c, 0x73, 0xde, 0xcc, 0xe7, 0x6d, 0x8d, 0x4, 0x19, 0x49, 0x7f, 0x47, 0x1f, 0x48, 0x15, 0xb0, 0xe8, 0x9e, 0xf2, 0x31, 0x59, 0xde, 0x34, 0xb4, 0x5b, 0xe5, 0xe0, 0x9, 0x11, 0x30, 0xc2, 0x88, 0x5b, 0x7c, 0x5d, 0x14, 0x13, 0x6f, 0x23, 0xa9, 0xd, 0xbc, 0x2d, 0x23, 0xbe, 0xd9, 0xed, 0x75, 0x4, 0x6c, 0x99, 0xdf, 0xfd, 0x70, 0x66, 0xe6, 0xee, 0xd9, 0xb1, 0x9e, 0x6e, 0x83, 0x59, 0xd5, 0xd4, 0x80, 0x59, 0x98, 0x77, 0x89, 0x43, 0x38, 0xc9, 0xaf, 0x30, 0x32, 0x9a, 0x20, 0x1b, 0x46, 0x3d, 0x67, 0x6e, 0xd7, 0x72, 0x9e, 0x4e, 0x21, 0x4f, 0xc6, 0xe0, 0xd4, 0x7b, 0x4, 0x8d, 0xa5, 0x3, 0xf6, 0x5, 0x9b, 0x6b, 0xdc, 0x2a, 0x93, 0x77, 0x28, 0xfd, 0xb4, 0x62, 0xda, 0x20, 0xe7, 0x1f, 0xab, 0x6b, 0x51, 0x43, 0x39, 0x2f, 0xa0, 0x92, 0x1, 0x6c, 0x75, 0x3e, 0xf4, 0x35, 0xfd, 0x43, 0x2e, 0xf7, 0xa4, 0x75, 0xda, 0xea, 0x9b, 0xa, 0x64, 0xb, 0xe0, 0x23, 0x29, 0xbd, 0xf7, 0xe7, 0x83, 0x3c, 0xfb, 0xdf, 0xb3, 0xae, 0x4f, 0xa4, 0x47, 0x55, 0x99, 0xde, 0x2f, 0x96, 0x6e, 0x1c, 0x43, 0x4c, 0x87, 0xe2, 0x7c, 0xd9, 0x5f, 0x4c, 0x7c, 0xe8, 0x90, 0x3, 0xdb, 0x30, 0x95, 0xd6, 0x22, 0xc, 0x47, 0xb8, 0x4d, 0x6b, 0xbd, 0x24, 0x11, 0xab, 0x2c, 0xd7, 0xbe, 0x6e, 0x7a, 0xd6, 0x8, 0xa3, 0x98, 0xd8, 0xdd, 0x15, 0x6a, 0xfa, 0x93, 0x30, 0x1, 0x25, 0x1d, 0xa2, 0x74, 0x86, 0x4b, 0x6a, 0x95, 0xe8, 0xe1, 0x4e, 0xe, 0x76, 0xb9, 0x49, 0xa9, 0x5f, 0xa0, 0xa6, 0x63, 0x3c, 0x7e, 0x7e, 0x20, 0x13, 0x4f, 0xbb, 0x66, 0x92, 0xb8, 0x2e, 0xa4, 0xfa, 0x48, 0xcb, 0xae, 0xb9, 0x3c, 0xaf, 0xd3, 0x1f, 0xe1, 0xd5, 0x8d, 0x42, 0x6d, 0xf0, 0xfc, 0x8c, 0xc, 0x0, 0xde, 0x40, 0xab, 0x8b, 0x47, 0x97, 0x4e, 0xa8, 0xcf, 0x8e, 0xdb, 0xa6, 0x8b, 0x20, 0x9, 0x84, 0x7a, 0x66, 0xe5, 0x98, 0x29, 0x2, 0x95, 0xe6, 0x38, 0x32, 0x60, 0x3, 0xe3, 0x9a, 0x1e, 0x54, 0xe8, 0x63, 0x80, 0x48, 0x9c, 0xe7, 0x63, 0x33, 0x6e, 0xa0, 0x65, 0x83, 0xfa, 0xc6, 0xba, 0x7a, 0x43, 0x71, 0x5, 0xf5, 0x68, 0x69, 0x85, 0x9c, 0xba, 0x45, 0xcd, 0x6b, 0xb, 0x19, 0xd1, 0xbb, 0x7f, 0x70, 0x85, 0x92, 0xd1, 0xb4, 0x64, 0x82, 0xb1, 0xe4, 0x62, 0xc5, 0x3c, 0x46, 0x1f, 0x92, 0x31, 0x1c, 0x4e, 0x41, 0x77, 0xf7, 0xe7, 0x87, 0xa2, 0xf, 0x6e, 0xe8, 0x92, 0x3, 0x6b, 0xa, 0xe7, 0xa9, 0x3b, 0x11, 0xda, 0x66, 0x8a, 0x29, 0xda, 0x79, 0xe1, 0x64, 0x8d, 0xe3, 0x54, 0xd4, 0xf5, 0xef, 0x64, 0x87, 0x3b, 0xf4, 0xc2, 0xf4, 0x71, 0x13, 0xa9, 0xe9, 0xe0, 0xa2, 0x6, 0x14, 0xab, 0x5d, 0xa7, 0x96, 0x0, 0xd6, 0xc3, 0xcc, 0x57, 0xed, 0x39, 0x6a, 0x25, 0xcd, 0x76, 0xea, 0xba, 0x3a, 0xf2, 0xa1, 0x95, 0x5d, 0xe5, 0x71, 0xcf, 0x9c, 0x62, 0x9e, 0x6a, 0xfa, 0xd5, 0x31, 0xd1, 0xa8, 0x66, 0x30, 0x33, 0xaa, 0x51, 0x17, 0x13, 0x82, 0x99, 0xc8, 0x14, 0x60, 0x9f, 0x4d, 0x32, 0x6d, 0xda, 0x19, 0x26, 0x21, 0xdc, 0x7e, 0x2e, 0x25, 0x67, 0x72, 0xca, 0xf, 0x92, 0xcd, 0xf6, 0xd6, 0xcb, 0x97, 0x8a, 0x33, 0x58, 0x73, 0x70, 0x91, 0x1d, 0xbf, 0x28, 0x23, 0xa3, 0xc, 0xf1, 0x83, 0xc3, 0xc8, 0x56, 0x77, 0x68, 0xe3, 0x82, 0xba, 0xb9, 0x57, 0x56, 0x57, 0x9c, 0xc3, 0xd6, 0x14, 0x5, 0x3c, 0xb1, 0xaf, 0x93, 0xc8, 0x8a, 0x57, 0x7f, 0x53, 0xfa, 0x2f, 0xaa, 0x6e, 0x66, 0x83, 0xfa, 0x33, 0xd1, 0x21, 0xab, 0x1b, 0x71, 0xb4, 0x7c, 0xda, 0xfd, 0xfb, 0x7f, 0x20, 0xab, 0x5e, 0xd5, 0xca, 0xfd, 0xdd, 0xe0, 0xee, 0xda, 0xba, 0xa8, 0x27, 0x99, 0x97, 0x69, 0xc1, 0x3c, 0x82, 0x8c, 0xa, 0x5c, 0x2d, 0x5b, 0x88, 0x3e, 0x34, 0x35, 0x86, 0x37, 0x46, 0x79, 0xe1, 0xaa, 0x19, 0xfb, 0xaa, 0xde, 0x15, 0x9, 0xd, 0x1a, 0x57, 0xff, 0xb5, 0xf, 0xf3, 0x2b, 0x5a, 0x6a, 0x4d, 0x19, 0x77, 0x71, 0x45, 0xdf, 0x4f, 0xb3, 0xec, 0xf1, 0xeb, 0x18, 0x53, 0x3e, 0x3b, 0x47, 0x8, 0x9a, 0x73, 0xa0, 0x5c, 0x8c, 0x5f, 0xeb, 0xf, 0x3a, 0xc2, 0x43, 0x67, 0xb4, 0x66, 0x67, 0x80, 0x58, 0xe, 0xc1, 0xec, 0x40, 0xd4, 0x22, 0x94, 0xca, 0xf9, 0xe8, 0x92, 0xe4, 0x69, 0x38, 0xbe, 0x67, 0x64, 0xca, 0x50, 0xc7, 0x6, 0x67, 0x42, 0x6e, 0xa3, 0xf0, 0xb7, 0x6c, 0xf2, 0xe8, 0x5f, 0xb1, 0xaf, 0xe7, 0xdb, 0xbb, 0x77, 0xb5, 0xf8, 0xcb, 0x8, 0xc4, 0x75, 0x7e, 0xc0, 0xf9, 0x1c, 0x7f, 0x3c, 0x89, 0x2f, 0xd2, 0x58, 0x3a, 0xe2, 0xf8, 0x91, 0xb6, 0x7b, 0x24, 0x27, 0xe9, 0xae, 0x84, 0x8b, 0xde, 0x74, 0xac, 0xfd, 0xd9, 0xb7, 0x69, 0x2a, 0xec, 0x32, 0x6f, 0xf0, 0x92, 0x84, 0xf1, 0x40, 0xc, 0x8a, 0xbc, 0x39, 0x6e, 0x2e, 0x73, 0xd4, 0x6e, 0x8a, 0x74, 0x2a, 0xdc, 0x60, 0x1f, 0xa3, 0x7, 0xde, 0x75, 0x8b, 0x74, 0xc8, 0xfe, 0x63, 0x75, 0xf6, 0x3d, 0x63, 0xac, 0x33, 0x89, 0xc3, 0xf0, 0xf8, 0x2d, 0x6b, 0xb4, 0x9e, 0x74, 0x8b, 0x5c, 0x33, 0xb4, 0xca, 0xa8, 0xe4, 0x99, 0xb6, 0x90, 0xa1, 0xef, 0xf, 0xd3, 0x61, 0xb2, 0xc6, 0x1a, 0x94, 0x7c, 0x44, 0x55, 0xf4, 0x45, 0xff, 0x9e, 0xa5, 0x5a, 0xc6, 0xa0, 0xe8, 0x2a, 0xc1, 0x8d, 0x6f, 0x34, 0x11, 0xb9, 0xbe, 0x4e, 0xd9, 0x87, 0x97, 0x73, 0xcf, 0x3d, 0x23, 0xae, 0xd5, 0x1a, 0x5e, 0xae, 0x5d, 0x6a, 0x3, 0xf9, 0x22, 0xd, 0x10, 0xd9, 0x47, 0x69, 0x15, 0x3f, 0xee, 0x52, 0xa3, 0x8, 0xd2, 0x3c, 0x51, 0xf4, 0xf8, 0x9d, 0xe4, 0x98, 0x89, 0xc8, 0x67, 0x39, 0xd5, 0x5e, 0x35, 0x78, 0x27, 0xe8, 0x3c, 0x80, 0xae, 0x79, 0x71, 0xd2, 0x93, 0xf4, 0xaa, 0x51, 0x12, 0x1c, 0x4b, 0x1b, 0xe5, 0x6e, 0x15, 0x6f, 0xe4, 0xbb, 0x51, 0x9b, 0x45, 0x9f, 0xf9, 0xc4, 0x8c, 0x2a, 0xfb, 0x1a, 0xdf, 0x55, 0xd3, 0x48, 0x93, 0x27, 0x1, 0x26, 0xc2, 0x6b, 0x55, 0x6d, 0xa2, 0xfb, 0x84, 0x8b, 0xc9, 0x9e, 0x28, 0xc2, 0xef, 0x1a, 0x24, 0xec, 0x9b, 0xae, 0xbd, 0x60, 0xe9, 0x15, 0x35, 0xee, 0x42, 0xa4, 0x33, 0x5b, 0xfa, 0xf, 0xb6, 0xf7, 0x1, 0xa6, 0x2, 0x4c, 0xca, 0x90, 0x58, 0x3a, 0x96, 0x41, 0xe7, 0xcb, 0x9, 0x8c, 0xdb, 0x85, 0x4d, 0xa8, 0x89, 0xf3, 0xb5, 0x8e, 0xfd, 0x75, 0x5b, 0x4f, 0xed, 0xde, 0x3f, 0xeb, 0x38, 0xa3, 0xbe, 0xb0, 0x73, 0xfc, 0xb8, 0x54, 0xf7, 0x4c, 0x30, 0x67, 0x2e, 0x38, 0xa2, 0x54, 0x18, 0xba, 0x8, 0xbf, 0xf2, 0x39, 0xd5, 0xfe, 0xa5, 0x41, 0xc6, 0x66, 0x66, 0xba, 0x81, 0xef, 0x67, 0xe4, 0xe6, 0x3c, 0xc, 0xca, 0xa4, 0xa, 0x79, 0xb3, 0x57, 0x8b, 0x8a, 0x75, 0x98, 0x18, 0x42, 0x2f, 0x29, 0xa3, 0x82, 0xef, 0x9f, 0x86, 0x6, 0x23, 0xe1, 0x75, 0xfa, 0x8, 0xb1, 0xde, 0x17, 0x4a}, + }, + { + input: "testdata/huffman-rand-limit.in", + want: "testdata/huffman-rand-limit.%s.expect", + wantNoInput: "testdata/huffman-rand-limit.%s.expect-noinput", + tokens: []token{0x61, 0x51c00000, 0xa, 0xf8, 0x8b, 0x96, 0x76, 0x48, 0xa, 0x85, 0x94, 0x25, 0x80, 0xaf, 0xc2, 0xfe, 0x8d, 0xe8, 0x20, 0xeb, 0x17, 0x86, 0xc9, 0xb7, 0xc5, 0xde, 0x6, 0xea, 0x7d, 0x18, 0x8b, 0xe7, 0x3e, 0x7, 0xda, 0xdf, 0xff, 0x6c, 0x73, 0xde, 0xcc, 0xe7, 0x6d, 0x8d, 0x4, 0x19, 0x49, 0x7f, 0x47, 0x1f, 0x48, 0x15, 0xb0, 0xe8, 0x9e, 0xf2, 0x31, 0x59, 0xde, 0x34, 0xb4, 0x5b, 0xe5, 0xe0, 0x9, 0x11, 0x30, 0xc2, 0x88, 0x5b, 0x7c, 0x5d, 0x14, 0x13, 0x6f, 0x23, 0xa9, 0xa, 0xbc, 0x2d, 0x23, 0xbe, 0xd9, 0xed, 0x75, 0x4, 0x6c, 0x99, 0xdf, 0xfd, 0x70, 0x66, 0xe6, 0xee, 0xd9, 0xb1, 0x9e, 0x6e, 0x83, 0x59, 0xd5, 0xd4, 0x80, 0x59, 0x98, 0x77, 0x89, 0x43, 0x38, 0xc9, 0xaf, 0x30, 0x32, 0x9a, 0x20, 0x1b, 0x46, 0x3d, 0x67, 0x6e, 0xd7, 0x72, 0x9e, 0x4e, 0x21, 0x4f, 0xc6, 0xe0, 0xd4, 0x7b, 0x4, 0x8d, 0xa5, 0x3, 0xf6, 0x5, 0x9b, 0x6b, 0xdc, 0x2a, 0x93, 0x77, 0x28, 0xfd, 0xb4, 0x62, 0xda, 0x20, 0xe7, 0x1f, 0xab, 0x6b, 0x51, 0x43, 0x39, 0x2f, 0xa0, 0x92, 0x1, 0x6c, 0x75, 0x3e, 0xf4, 0x35, 0xfd, 0x43, 0x2e, 0xf7, 0xa4, 0x75, 0xda, 0xea, 0x9b, 0xa}, + }, + { + input: "testdata/huffman-shifts.in", + want: "testdata/huffman-shifts.%s.expect", + wantNoInput: "testdata/huffman-shifts.%s.expect-noinput", + tokens: []token{0x31, 0x30, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x52400001, 0xd, 0xa, 0x32, 0x33, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7fc00001, 0x7f400001}, + }, + { + input: "testdata/huffman-text-shift.in", + want: "testdata/huffman-text-shift.%s.expect", + wantNoInput: "testdata/huffman-text-shift.%s.expect-noinput", + tokens: []token{0x2f, 0x2f, 0x43, 0x6f, 0x70, 0x79, 0x72, 0x69, 0x67, 0x68, 0x74, 0x32, 0x30, 0x30, 0x39, 0x54, 0x68, 0x47, 0x6f, 0x41, 0x75, 0x74, 0x68, 0x6f, 0x72, 0x2e, 0x41, 0x6c, 0x6c, 0x40800016, 0x72, 0x72, 0x76, 0x64, 0x2e, 0xd, 0xa, 0x2f, 0x2f, 0x55, 0x6f, 0x66, 0x74, 0x68, 0x69, 0x6f, 0x75, 0x72, 0x63, 0x63, 0x6f, 0x64, 0x69, 0x67, 0x6f, 0x76, 0x72, 0x6e, 0x64, 0x62, 0x79, 0x42, 0x53, 0x44, 0x2d, 0x74, 0x79, 0x6c, 0x40400020, 0x6c, 0x69, 0x63, 0x6e, 0x74, 0x68, 0x74, 0x63, 0x6e, 0x62, 0x66, 0x6f, 0x75, 0x6e, 0x64, 0x69, 0x6e, 0x74, 0x68, 0x4c, 0x49, 0x43, 0x45, 0x4e, 0x53, 0x45, 0x66, 0x69, 0x6c, 0x2e, 0xd, 0xa, 0xd, 0xa, 0x70, 0x63, 0x6b, 0x67, 0x6d, 0x69, 0x6e, 0x4040000a, 0x69, 0x6d, 0x70, 0x6f, 0x72, 0x74, 0x22, 0x6f, 0x22, 0x4040000c, 0x66, 0x75, 0x6e, 0x63, 0x6d, 0x69, 0x6e, 0x28, 0x29, 0x7b, 0xd, 0xa, 0x9, 0x76, 0x72, 0x62, 0x3d, 0x6d, 0x6b, 0x28, 0x5b, 0x5d, 0x62, 0x79, 0x74, 0x2c, 0x36, 0x35, 0x35, 0x33, 0x35, 0x29, 0xd, 0xa, 0x9, 0x66, 0x2c, 0x5f, 0x3a, 0x3d, 0x6f, 0x2e, 0x43, 0x72, 0x74, 0x28, 0x22, 0x68, 0x75, 0x66, 0x66, 0x6d, 0x6e, 0x2d, 0x6e, 0x75, 0x6c, 0x6c, 0x2d, 0x6d, 0x78, 0x2e, 0x69, 0x6e, 0x22, 0x40800021, 0x2e, 0x57, 0x72, 0x69, 0x74, 0x28, 0x62, 0x29, 0xd, 0xa, 0x7d, 0xd, 0xa, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x58, 0x78, 0x79, 0x7a, 0x21, 0x22, 0x23, 0xc2, 0xa4, 0x25, 0x26, 0x2f, 0x3f, 0x22}, + }, + { + input: "testdata/huffman-text.in", + want: "testdata/huffman-text.%s.expect", + wantNoInput: "testdata/huffman-text.%s.expect-noinput", + tokens: []token{0x2f, 0x2f, 0x20, 0x43, 0x6f, 0x70, 0x79, 0x72, 0x69, 0x67, 0x68, 0x74, 0x20, 0x32, 0x30, 0x30, 0x39, 0x20, 0x54, 0x68, 0x65, 0x20, 0x47, 0x6f, 0x20, 0x41, 0x75, 0x74, 0x68, 0x6f, 0x72, 0x73, 0x2e, 0x20, 0x41, 0x6c, 0x6c, 0x20, 0x4080001e, 0x73, 0x20, 0x72, 0x65, 0x73, 0x65, 0x72, 0x76, 0x65, 0x64, 0x2e, 0xd, 0xa, 0x2f, 0x2f, 0x20, 0x55, 0x73, 0x65, 0x20, 0x6f, 0x66, 0x20, 0x74, 0x68, 0x69, 0x73, 0x20, 0x73, 0x6f, 0x75, 0x72, 0x63, 0x65, 0x20, 0x63, 0x6f, 0x64, 0x65, 0x20, 0x69, 0x73, 0x20, 0x67, 0x6f, 0x76, 0x65, 0x72, 0x6e, 0x65, 0x64, 0x20, 0x62, 0x79, 0x20, 0x61, 0x20, 0x42, 0x53, 0x44, 0x2d, 0x73, 0x74, 0x79, 0x6c, 0x65, 0x40800036, 0x6c, 0x69, 0x63, 0x65, 0x6e, 0x73, 0x65, 0x20, 0x74, 0x68, 0x61, 0x74, 0x20, 0x63, 0x61, 0x6e, 0x20, 0x62, 0x65, 0x20, 0x66, 0x6f, 0x75, 0x6e, 0x64, 0x20, 0x69, 0x6e, 0x20, 0x74, 0x68, 0x65, 0x20, 0x4c, 0x49, 0x43, 0x45, 0x4e, 0x53, 0x45, 0x20, 0x66, 0x69, 0x6c, 0x65, 0x2e, 0xd, 0xa, 0xd, 0xa, 0x70, 0x61, 0x63, 0x6b, 0x61, 0x67, 0x65, 0x20, 0x6d, 0x61, 0x69, 0x6e, 0x4040000f, 0x69, 0x6d, 0x70, 0x6f, 0x72, 0x74, 0x20, 0x22, 0x6f, 0x73, 0x22, 0x4040000e, 0x66, 0x75, 0x6e, 0x63, 0x4080001b, 0x28, 0x29, 0x20, 0x7b, 0xd, 0xa, 0x9, 0x76, 0x61, 0x72, 0x20, 0x62, 0x20, 0x3d, 0x20, 0x6d, 0x61, 0x6b, 0x65, 0x28, 0x5b, 0x5d, 0x62, 0x79, 0x74, 0x65, 0x2c, 0x20, 0x36, 0x35, 0x35, 0x33, 0x35, 0x29, 0xd, 0xa, 0x9, 0x66, 0x2c, 0x20, 0x5f, 0x20, 0x3a, 0x3d, 0x20, 0x6f, 0x73, 0x2e, 0x43, 0x72, 0x65, 0x61, 0x74, 0x65, 0x28, 0x22, 0x68, 0x75, 0x66, 0x66, 0x6d, 0x61, 0x6e, 0x2d, 0x6e, 0x75, 0x6c, 0x6c, 0x2d, 0x6d, 0x61, 0x78, 0x2e, 0x69, 0x6e, 0x22, 0x4080002a, 0x2e, 0x57, 0x72, 0x69, 0x74, 0x65, 0x28, 0x62, 0x29, 0xd, 0xa, 0x7d, 0xd, 0xa}, + }, + { + input: "testdata/huffman-zero.in", + want: "testdata/huffman-zero.%s.expect", + wantNoInput: "testdata/huffman-zero.%s.expect-noinput", + tokens: []token{0x30, ml, 0x4b800000}, + }, + { + input: "", + want: "", + wantNoInput: "testdata/null-long-match.%s.expect-noinput", + tokens: []token{0x0, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, ml, 0x41400000}, + }, +} + +// TestWriteBlock tests if the writeBlock encoding has changed. +// To update the reference files use the "-update" flag on the test. +func TestWriteBlock(t *testing.T) { + for _, test := range writeBlockTests { + testBlock(t, test, "wb") + } +} + +// TestWriteBlockDynamic tests if the writeBlockDynamic encoding has changed. +// To update the reference files use the "-update" flag on the test. +func TestWriteBlockDynamic(t *testing.T) { + for _, test := range writeBlockTests { + testBlock(t, test, "dyn") + } +} + +// TestWriteBlockDynamic tests if the writeBlockDynamic encoding has changed. +// To update the reference files use the "-update" flag on the test. +func TestWriteBlockDynamicSync(t *testing.T) { + for _, test := range writeBlockTests { + testBlock(t, test, "sync") + } +} + +// testBlock tests a block against its references, +// or regenerate the references, if "-update" flag is set. +func testBlock(t *testing.T, test huffTest, ttype string) { + if test.want != "" { + test.want = fmt.Sprintf(test.want, ttype) + } + const gotSuffix = ".got" + test.wantNoInput = fmt.Sprintf(test.wantNoInput, ttype) + tokens := indexTokens(test.tokens) + if *update { + if test.input != "" { + t.Logf("Updating %q", test.want) + input, err := ioutil.ReadFile(test.input) + if err != nil { + t.Error(err) + return + } + + f, err := os.Create(test.want) + if err != nil { + t.Error(err) + return + } + defer f.Close() + bw := newHuffmanBitWriter(f) + writeToType(t, ttype, bw, tokens, input) + } + + t.Logf("Updating %q", test.wantNoInput) + f, err := os.Create(test.wantNoInput) + if err != nil { + t.Error(err) + return + } + defer f.Close() + bw := newHuffmanBitWriter(f) + writeToType(t, ttype, bw, tokens, nil) + return + } + + if test.input != "" { + t.Logf("Testing %q", test.want) + input, err := ioutil.ReadFile(test.input) + if err != nil { + t.Error(err) + return + } + want, err := ioutil.ReadFile(test.want) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + bw := newHuffmanBitWriter(&buf) + writeToType(t, ttype, bw, tokens, input) + + got := buf.Bytes() + if !bytes.Equal(got, want) { + t.Errorf("writeBlock did not yield expected result for file %q with input. See %q", test.want, test.want+gotSuffix) + if err := ioutil.WriteFile(test.want+gotSuffix, got, 0666); err != nil { + t.Error(err) + } + } + t.Log("Output ok") + + // Test if the writer produces the same output after reset. + buf.Reset() + bw.reset(&buf) + writeToType(t, ttype, bw, tokens, input) + bw.flush() + got = buf.Bytes() + if !bytes.Equal(got, want) { + t.Errorf("reset: writeBlock did not yield expected result for file %q with input. See %q", test.want, test.want+".reset"+gotSuffix) + if err := ioutil.WriteFile(test.want+".reset"+gotSuffix, got, 0666); err != nil { + t.Error(err) + } + return + } + t.Log("Reset ok") + testWriterEOF(t, "wb", test, true) + } + t.Logf("Testing %q", test.wantNoInput) + wantNI, err := ioutil.ReadFile(test.wantNoInput) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + bw := newHuffmanBitWriter(&buf) + writeToType(t, ttype, bw, tokens, nil) + + got := buf.Bytes() + if !bytes.Equal(got, wantNI) { + t.Errorf("writeBlock did not yield expected result for file %q with input. See %q", test.wantNoInput, test.wantNoInput+gotSuffix) + if err := ioutil.WriteFile(test.wantNoInput+gotSuffix, got, 0666); err != nil { + t.Error(err) + } + } else if got[0]&1 == 1 { + t.Error("got unexpected EOF") + return + } + + t.Log("Output ok") + + // Test if the writer produces the same output after reset. + buf.Reset() + bw.reset(&buf) + writeToType(t, ttype, bw, tokens, nil) + bw.flush() + got = buf.Bytes() + if !bytes.Equal(got, wantNI) { + t.Errorf("reset: writeBlock did not yield expected result for file %q without input. See %q", test.wantNoInput, test.wantNoInput+".reset"+gotSuffix) + if err := ioutil.WriteFile(test.wantNoInput+".reset"+gotSuffix, got, 0666); err != nil { + t.Error(err) + } + return + } + t.Log("Reset ok") + testWriterEOF(t, "wb", test, false) +} + +func writeToType(t *testing.T, ttype string, bw *huffmanBitWriter, tok tokens, input []byte) { + switch ttype { + case "wb": + bw.writeBlock(&tok, false, input) + case "dyn": + bw.writeBlockDynamic(&tok, false, input, false) + case "sync": + bw.writeBlockDynamic(&tok, false, input, true) + default: + panic("unknown test type") + } + + if bw.err != nil { + t.Error(bw.err) + return + } + + bw.flush() + if bw.err != nil { + t.Error(bw.err) + return + } +} + +// testWriterEOF tests if the written block contains an EOF marker. +func testWriterEOF(t *testing.T, ttype string, test huffTest, useInput bool) { + if useInput && test.input == "" { + return + } + var input []byte + if useInput { + var err error + input, err = ioutil.ReadFile(test.input) + if err != nil { + t.Error(err) + return + } + } + var buf bytes.Buffer + bw := newHuffmanBitWriter(&buf) + tokens := indexTokens(test.tokens) + switch ttype { + case "wb": + bw.writeBlock(&tokens, true, input) + case "dyn": + bw.writeBlockDynamic(&tokens, true, input, true) + case "huff": + bw.writeBlockHuff(true, input, true) + default: + panic("unknown test type") + } + if bw.err != nil { + t.Error(bw.err) + return + } + + bw.flush() + if bw.err != nil { + t.Error(bw.err) + return + } + b := buf.Bytes() + if len(b) == 0 { + t.Error("no output received") + return + } + if b[0]&1 != 1 { + t.Errorf("block not marked with EOF for input %q", test.input) + return + } + t.Log("EOF ok") +} diff --git a/github.com/klauspost/compress/flate/huffman_code.go b/github.com/klauspost/compress/flate/huffman_code.go new file mode 100644 index 0000000..4c39a30 --- /dev/null +++ b/github.com/klauspost/compress/flate/huffman_code.go @@ -0,0 +1,363 @@ +// 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 + +import ( + "math" + "math/bits" +) + +const ( + maxBitsLimit = 16 + // number of valid literals + literalCount = 286 +) + +// hcode is a huffman code with a bit code and bit length. +type hcode struct { + code, len uint16 +} + +type huffmanEncoder struct { + codes []hcode + freqcache []literalNode + bitCount [17]int32 +} + +type literalNode struct { + literal uint16 + freq uint16 +} + +// A levelInfo describes the state of the constructed tree for a given depth. +type levelInfo struct { + // Our level. for better printing + level int32 + + // The frequency of the last node at this level + lastFreq int32 + + // The frequency of the next character to add to this level + nextCharFreq int32 + + // The frequency of the next pair (from level below) to add to this level. + // Only valid if the "needed" value of the next lower level is 0. + nextPairFreq int32 + + // The number of chains remaining to generate for this level before moving + // up to the next level + needed int32 +} + +// set sets the code and length of an hcode. +func (h *hcode) set(code uint16, length uint16) { + h.len = length + h.code = code +} + +func reverseBits(number uint16, bitLength byte) uint16 { + return bits.Reverse16(number << ((16 - bitLength) & 15)) +} + +func maxNode() literalNode { return literalNode{math.MaxUint16, math.MaxUint16} } + +func newHuffmanEncoder(size int) *huffmanEncoder { + // Make capacity to next power of two. + c := uint(bits.Len32(uint32(size - 1))) + return &huffmanEncoder{codes: make([]hcode, size, 1<= 3 +// The cases of 0, 1, and 2 literals are handled by special case code. +// +// list An array of the literals with non-zero frequencies +// and their associated frequencies. The array is in order of increasing +// frequency, and has as its last element a special element with frequency +// MaxInt32 +// maxBits The maximum number of bits that should be used to encode any literal. +// Must be less than 16. +// return An integer array in which array[i] indicates the number of literals +// that should be encoded in i bits. +func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 { + if maxBits >= maxBitsLimit { + panic("flate: maxBits too large") + } + n := int32(len(list)) + list = list[0 : n+1] + list[n] = maxNode() + + // The tree can't have greater depth than n - 1, no matter what. This + // saves a little bit of work in some small cases + if maxBits > n-1 { + maxBits = n - 1 + } + + // Create information about each of the levels. + // A bogus "Level 0" whose sole purpose is so that + // level1.prev.needed==0. This makes level1.nextPairFreq + // be a legitimate value that never gets chosen. + var levels [maxBitsLimit]levelInfo + // leafCounts[i] counts the number of literals at the left + // of ancestors of the rightmost node at level i. + // leafCounts[i][j] is the number of literals at the left + // of the level j ancestor. + var leafCounts [maxBitsLimit][maxBitsLimit]int32 + + for level := int32(1); level <= maxBits; level++ { + // For every level, the first two items are the first two characters. + // We initialize the levels as if we had already figured this out. + levels[level] = levelInfo{ + level: level, + lastFreq: int32(list[1].freq), + nextCharFreq: int32(list[2].freq), + nextPairFreq: int32(list[0].freq) + int32(list[1].freq), + } + leafCounts[level][level] = 2 + if level == 1 { + levels[level].nextPairFreq = math.MaxInt32 + } + } + + // We need a total of 2*n - 2 items at top level and have already generated 2. + levels[maxBits].needed = 2*n - 4 + + level := maxBits + for { + l := &levels[level] + if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 { + // We've run out of both leafs and pairs. + // End all calculations for this level. + // To make sure we never come back to this level or any lower level, + // set nextPairFreq impossibly large. + l.needed = 0 + levels[level+1].nextPairFreq = math.MaxInt32 + level++ + continue + } + + prevFreq := l.lastFreq + if l.nextCharFreq < l.nextPairFreq { + // The next item on this row is a leaf node. + n := leafCounts[level][level] + 1 + l.lastFreq = l.nextCharFreq + // Lower leafCounts are the same of the previous node. + leafCounts[level][level] = n + e := list[n] + if e.literal < math.MaxUint16 { + l.nextCharFreq = int32(e.freq) + } else { + l.nextCharFreq = math.MaxInt32 + } + } else { + // The next item on this row is a pair from the previous row. + // nextPairFreq isn't valid until we generate two + // more values in the level below + l.lastFreq = l.nextPairFreq + // Take leaf counts from the lower level, except counts[level] remains the same. + copy(leafCounts[level][:level], leafCounts[level-1][:level]) + levels[l.level-1].needed = 2 + } + + if l.needed--; l.needed == 0 { + // We've done everything we need to do for this level. + // Continue calculating one level up. Fill in nextPairFreq + // of that level with the sum of the two nodes we've just calculated on + // this level. + if l.level == maxBits { + // All done! + break + } + levels[l.level+1].nextPairFreq = prevFreq + l.lastFreq + level++ + } else { + // If we stole from below, move down temporarily to replenish it. + for levels[level-1].needed > 0 { + level-- + } + } + } + + // Somethings is wrong if at the end, the top level is null or hasn't used + // all of the leaves. + if leafCounts[maxBits][maxBits] != n { + panic("leafCounts[maxBits][maxBits] != n") + } + + bitCount := h.bitCount[:maxBits+1] + bits := 1 + counts := &leafCounts[maxBits] + for level := maxBits; level > 0; level-- { + // chain.leafCount gives the number of literals requiring at least "bits" + // bits to encode. + bitCount[bits] = counts[level] - counts[level-1] + bits++ + } + return bitCount +} + +// Look at the leaves and assign them a bit count and an encoding as specified +// in RFC 1951 3.2.2 +func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalNode) { + code := uint16(0) + for n, bits := range bitCount { + code <<= 1 + if n == 0 || bits == 0 { + continue + } + // The literals list[len(list)-bits] .. list[len(list)-bits] + // are encoded using "bits" bits, and get the values + // code, code + 1, .... The code values are + // assigned in literal order (not frequency order). + chunk := list[len(list)-int(bits):] + + sortByLiteral(chunk) + for _, node := range chunk { + h.codes[node.literal] = hcode{code: reverseBits(code, uint8(n)), len: uint16(n)} + code++ + } + list = list[0 : len(list)-int(bits)] + } +} + +// Update this Huffman Code object to be the minimum code for the specified frequency count. +// +// freq An array of frequencies, in which frequency[i] gives the frequency of literal i. +// maxBits The maximum number of bits to use for any literal. +func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) { + if h.freqcache == nil { + // Allocate a reusable buffer with the longest possible frequency table. + // Possible lengths are codegenCodeCount, offsetCodeCount and literalCount. + // The largest of these is literalCount, so we allocate for that case. + h.freqcache = make([]literalNode, literalCount+1) + } + list := h.freqcache[:len(freq)+1] + // Number of non-zero literals + count := 0 + // Set list to be the set of all non-zero literals and their frequencies + for i, f := range freq { + if f != 0 { + list[count] = literalNode{uint16(i), f} + count++ + } else { + list[count] = literalNode{} + h.codes[i].len = 0 + } + } + list[len(freq)] = literalNode{} + + list = list[:count] + if count <= 2 { + // Handle the small cases here, because they are awkward for the general case code. With + // two or fewer literals, everything has bit length 1. + for i, node := range list { + // "list" is in order of increasing literal value. + h.codes[node.literal].set(uint16(i), 1) + } + return + } + sortByFreq(list) + + // Get the number of literals for each bit count + bitCount := h.bitCounts(list, maxBits) + // And do the assignment + h.assignEncodingAndSize(bitCount, list) +} + +func atLeastOne(v float32) float32 { + if v < 1 { + return 1 + } + return v +} + +// 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, int) { + h = h[:256] + for _, t := range b { + h[t]++ + } + 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), int(extra + 0.99) +} diff --git a/github.com/klauspost/compress/flate/huffman_sortByFreq.go b/github.com/klauspost/compress/flate/huffman_sortByFreq.go new file mode 100644 index 0000000..2077802 --- /dev/null +++ b/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/github.com/klauspost/compress/flate/huffman_sortByLiteral.go b/github.com/klauspost/compress/flate/huffman_sortByLiteral.go new file mode 100644 index 0000000..93f1aea --- /dev/null +++ b/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/github.com/klauspost/compress/flate/inflate.go b/github.com/klauspost/compress/flate/inflate.go new file mode 100644 index 0000000..3e4259f --- /dev/null +++ b/github.com/klauspost/compress/flate/inflate.go @@ -0,0 +1,1001 @@ +// 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 implements the DEFLATE compressed data format, described in +// RFC 1951. The gzip and zlib packages implement access to DEFLATE-based file +// formats. +package flate + +import ( + "bufio" + "fmt" + "io" + "math/bits" + "strconv" + "sync" +) + +const ( + maxCodeLen = 16 // max length of Huffman code + maxCodeLenMask = 15 // mask for max length of Huffman code + // The next three numbers come from the RFC section 3.2.7, with the + // additional proviso in section 3.2.5 which implies that distance codes + // 30 and 31 should never occur in compressed data. + maxNumLit = 286 + maxNumDist = 30 + numCodes = 19 // number of codes in Huffman meta-code + + debugDecode = false +) + +// Initialize the fixedHuffmanDecoder only once upon first use. +var fixedOnce sync.Once +var fixedHuffmanDecoder huffmanDecoder + +// A CorruptInputError reports the presence of corrupt input at a given offset. +type CorruptInputError int64 + +func (e CorruptInputError) Error() string { + return "flate: corrupt input before offset " + strconv.FormatInt(int64(e), 10) +} + +// An InternalError reports an error in the flate code itself. +type InternalError string + +func (e InternalError) Error() string { return "flate: internal error: " + string(e) } + +// A ReadError reports an error encountered while reading input. +// +// Deprecated: No longer returned. +type ReadError struct { + Offset int64 // byte offset where error occurred + Err error // error returned by underlying Read +} + +func (e *ReadError) Error() string { + return "flate: read error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error() +} + +// A WriteError reports an error encountered while writing output. +// +// Deprecated: No longer returned. +type WriteError struct { + Offset int64 // byte offset where error occurred + Err error // error returned by underlying Write +} + +func (e *WriteError) Error() string { + return "flate: write error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error() +} + +// Resetter resets a ReadCloser returned by NewReader or NewReaderDict to +// to switch to a new underlying Reader. This permits reusing a ReadCloser +// instead of allocating a new one. +type Resetter interface { + // Reset discards any buffered data and resets the Resetter as if it was + // newly initialized with the given reader. + Reset(r io.Reader, dict []byte) error +} + +// The data structure for decoding Huffman tables is based on that of +// zlib. There is a lookup table of a fixed bit width (huffmanChunkBits), +// For codes smaller than the table width, there are multiple entries +// (each combination of trailing bits has the same value). For codes +// larger than the table width, the table contains a link to an overflow +// table. The width of each entry in the link table is the maximum code +// size minus the chunk width. +// +// Note that you can do a lookup in the table even without all bits +// filled. Since the extra bits are zero, and the DEFLATE Huffman codes +// have the property that shorter codes come before longer ones, the +// bit length estimate in the result is a lower bound on the actual +// number of bits. +// +// See the following: +// http://www.gzip.org/algorithm.txt + +// chunk & 15 is number of bits +// chunk >> 4 is value, including table link + +const ( + huffmanChunkBits = 9 + huffmanNumChunks = 1 << huffmanChunkBits + huffmanCountMask = 15 + huffmanValueShift = 4 +) + +type huffmanDecoder struct { + 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 +} + +// Initialize Huffman decoding tables from array of code lengths. +// Following this function, h is guaranteed to be initialized into a complete +// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a +// degenerate case where the tree has only a single symbol with length 1. Empty +// trees are permitted. +func (h *huffmanDecoder) init(lengths []int) bool { + // Sanity enables additional runtime tests during Huffman + // table construction. It's intended to be used during + // development to supplement the currently ad-hoc unit tests. + const sanity = false + + if h.chunks == nil { + h.chunks = &[huffmanNumChunks]uint16{} + } + if h.maxRead != 0 { + *h = huffmanDecoder{chunks: h.chunks, links: h.links} + } + + // Count number of codes of each length, + // compute maxRead and max length. + var count [maxCodeLen]int + var min, max int + for _, n := range lengths { + if n == 0 { + continue + } + if min == 0 || n < min { + min = n + } + if n > max { + max = n + } + count[n&maxCodeLenMask]++ + } + + // Empty tree. The decompressor.huffSym function will fail later if the tree + // is used. Technically, an empty tree is only valid for the HDIST tree and + // not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree + // is guaranteed to fail since it will attempt to use the tree to decode the + // codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is + // guaranteed to fail later since the compressed data section must be + // composed of at least one symbol (the end-of-block marker). + if max == 0 { + return true + } + + code := 0 + var nextcode [maxCodeLen]int + for i := min; i <= max; i++ { + code <<= 1 + nextcode[i&maxCodeLenMask] = code + code += count[i&maxCodeLenMask] + } + + // Check that the coding is complete (i.e., that we've + // assigned all 2-to-the-max possible bit sequences). + // Exception: To be compatible with zlib, we also need to + // accept degenerate single-code codings. See also + // TestDegenerateHuffmanCoding. + if code != 1< huffmanChunkBits { + numLinks := 1 << (uint(max) - huffmanChunkBits) + h.linkMask = uint32(numLinks - 1) + + // create link tables + link := nextcode[huffmanChunkBits+1] >> 1 + if cap(h.links) < huffmanNumChunks-link { + h.links = make([][]uint16, huffmanNumChunks-link) + } else { + h.links = h.links[:huffmanNumChunks-link] + } + for j := uint(link); j < huffmanNumChunks; j++ { + reverse := int(bits.Reverse16(uint16(j))) + reverse >>= uint(16 - huffmanChunkBits) + off := j - uint(link) + if sanity && h.chunks[reverse] != 0 { + panic("impossible: overwriting existing chunk") + } + h.chunks[reverse] = uint16(off<>= uint(16 - n) + if n <= huffmanChunkBits { + for off := reverse; off < len(h.chunks); off += 1 << uint(n) { + // We should never need to overwrite + // an existing chunk. Also, 0 is + // never a valid chunk, because the + // lower 4 "count" bits should be + // between 1 and 15. + if sanity && h.chunks[off] != 0 { + panic("impossible: overwriting existing chunk") + } + h.chunks[off] = chunk + } + } else { + j := reverse & (huffmanNumChunks - 1) + if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 { + // Longer codes should have been + // associated with a link table above. + panic("impossible: not an indirect chunk") + } + value := h.chunks[j] >> huffmanValueShift + linktab := h.links[value] + reverse >>= huffmanChunkBits + for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) { + if sanity && linktab[off] != 0 { + panic("impossible: overwriting existing chunk") + } + linktab[off] = chunk + } + } + } + + if sanity { + // Above we've sanity checked that we never overwrote + // an existing entry. Here we additionally check that + // we filled the tables completely. + for i, chunk := range h.chunks { + if chunk == 0 { + // As an exception, in the degenerate + // single-code case, we allow odd + // chunks to be missing. + if code == 1 && i%2 == 1 { + continue + } + panic("impossible: missing chunk") + } + } + for _, linktab := range h.links { + for _, chunk := range linktab { + if chunk == 0 { + panic("impossible: missing chunk") + } + } + } + } + + return true +} + +// The actual read interface needed by NewReader. +// If the passed in io.Reader does not also have ReadByte, +// the NewReader will introduce its own buffering. +type Reader interface { + io.Reader + io.ByteReader +} + +// Decompress state. +type decompressor struct { + // Input source. + r Reader + roffset int64 + + // Huffman decoders for literal/length, distance. + h1, h2 huffmanDecoder + + // Length arrays used to define Huffman codes. + bits *[maxNumLit + maxNumDist]int + codebits *[numCodes]int + + // Output history, buffer. + dict dictDecoder + + // Next step in the decompression, + // and decompression state. + step func(*decompressor) + stepState int + 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() { + for f.nb < 1+2 { + if f.err = f.moreBits(); f.err != nil { + return + } + } + f.final = f.b&1 == 1 + f.b >>= 1 + typ := f.b & 3 + f.b >>= 2 + f.nb -= 1 + 2 + switch typ { + case 0: + f.dataBlock() + case 1: + // compressed, fixed Huffman tables + f.hl = &fixedHuffmanDecoder + f.hd = nil + f.huffmanBlockDecoder()() + case 2: + // compressed, dynamic Huffman tables + if f.err = f.readHuffman(); f.err != nil { + break + } + f.hl = &f.h1 + f.hd = &f.h2 + f.huffmanBlockDecoder()() + default: + // 3 is reserved. + if debugDecode { + fmt.Println("reserved data block encountered") + } + f.err = CorruptInputError(f.roffset) + } +} + +func (f *decompressor) Read(b []byte) (int, error) { + for { + if len(f.toRead) > 0 { + n := copy(b, f.toRead) + f.toRead = f.toRead[n:] + if len(f.toRead) == 0 { + return n, f.err + } + return n, nil + } + if f.err != nil { + return 0, f.err + } + f.step(f) + if f.err != nil && len(f.toRead) == 0 { + f.toRead = f.dict.readFlush() // Flush what's left in case of error + } + } +} + +// Support the io.WriteTo interface for io.Copy and friends. +func (f *decompressor) WriteTo(w io.Writer) (int64, error) { + total := int64(0) + flushed := false + for { + if len(f.toRead) > 0 { + n, err := w.Write(f.toRead) + total += int64(n) + if err != nil { + f.err = err + return total, err + } + if n != len(f.toRead) { + return total, io.ErrShortWrite + } + f.toRead = f.toRead[:0] + } + if f.err != nil && flushed { + if f.err == io.EOF { + return total, nil + } + return total, f.err + } + if f.err == nil { + f.step(f) + } + if len(f.toRead) == 0 && f.err != nil && !flushed { + f.toRead = f.dict.readFlush() // Flush what's left in case of error + flushed = true + } + } +} + +func (f *decompressor) Close() error { + if f.err == io.EOF { + return nil + } + return f.err +} + +// RFC 1951 section 3.2.7. +// Compression with dynamic Huffman codes + +var codeOrder = [...]int{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15} + +func (f *decompressor) readHuffman() error { + // HLIT[5], HDIST[5], HCLEN[4]. + for f.nb < 5+5+4 { + if err := f.moreBits(); err != nil { + return err + } + } + nlit := int(f.b&0x1F) + 257 + if nlit > maxNumLit { + if debugDecode { + fmt.Println("nlit > maxNumLit", nlit) + } + return CorruptInputError(f.roffset) + } + f.b >>= 5 + ndist := int(f.b&0x1F) + 1 + if ndist > maxNumDist { + if debugDecode { + fmt.Println("ndist > maxNumDist", ndist) + } + return CorruptInputError(f.roffset) + } + f.b >>= 5 + nclen := int(f.b&0xF) + 4 + // numCodes is 19, so nclen is always valid. + f.b >>= 4 + f.nb -= 5 + 5 + 4 + + // (HCLEN+4)*3 bits: code lengths in the magic codeOrder order. + for i := 0; i < nclen; i++ { + for f.nb < 3 { + if err := f.moreBits(); err != nil { + return err + } + } + f.codebits[codeOrder[i]] = int(f.b & 0x7) + f.b >>= 3 + f.nb -= 3 + } + for i := nclen; i < len(codeOrder); i++ { + f.codebits[codeOrder[i]] = 0 + } + if !f.h1.init(f.codebits[0:]) { + if debugDecode { + fmt.Println("init codebits failed") + } + return CorruptInputError(f.roffset) + } + + // HLIT + 257 code lengths, HDIST + 1 code lengths, + // using the code length Huffman code. + for i, n := 0, nlit+ndist; i < n; { + x, err := f.huffSym(&f.h1) + if err != nil { + return err + } + if x < 16 { + // Actual length. + f.bits[i] = x + i++ + continue + } + // Repeat previous length or zero. + var rep int + var nb uint + var b int + switch x { + default: + return InternalError("unexpected length code") + case 16: + rep = 3 + nb = 2 + if i == 0 { + if debugDecode { + fmt.Println("i==0") + } + return CorruptInputError(f.roffset) + } + b = f.bits[i-1] + case 17: + rep = 3 + nb = 3 + b = 0 + case 18: + rep = 11 + nb = 7 + b = 0 + } + for f.nb < nb { + if err := f.moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits:", err) + } + return err + } + } + rep += int(f.b & uint32(1<>= nb + f.nb -= nb + if i+rep > n { + if debugDecode { + fmt.Println("i+rep > n", i, rep, n) + } + return CorruptInputError(f.roffset) + } + for j := 0; j < rep; j++ { + f.bits[i] = b + i++ + } + } + + if !f.h1.init(f.bits[0:nlit]) || !f.h2.init(f.bits[nlit:nlit+ndist]) { + if debugDecode { + fmt.Println("init2 failed") + } + return CorruptInputError(f.roffset) + } + + // 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.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 +} + +// 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) huffmanBlockGeneric() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + + 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 := f.r.ReadByte() + if err != nil { + f.b = b + f.nb = nb + f.err = noEOF(err) + return + } + f.roffset++ + b |= uint32(c) << (nb & 31) + 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 & 31) + 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).huffmanBlockGeneric + f.stepState = stateInit + return + } + goto readLiteral + case v == 256: + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + n = 0 + case v < 269: + length = v*2 - (265*2 - 11) + n = 1 + case v < 273: + length = v*4 - (269*4 - 19) + n = 2 + case v < 277: + length = v*8 - (273*8 - 35) + n = 3 + case v < 281: + length = v*16 - (277*16 - 67) + n = 4 + case v < 285: + length = v*32 - (281*32 - 131) + n = 5 + case v < maxNumLit: + length = 258 + n = 0 + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + if n > 0 { + for f.nb < n { + if err = f.moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + } + length += int(f.b & uint32(1<>= n + f.nb -= n + } + + var dist int + if f.hd == nil { + for f.nb < 5 { + if err = f.moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + } + dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + if dist, err = f.huffSym(f.hd); err != nil { + if debugDecode { + fmt.Println("huffsym:", err) + } + f.err = err + return + } + } + + 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 + for f.nb < nb { + if err = f.moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb + f.nb -= nb + dist = 1<<(nb+1) + 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 > 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, 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).huffmanBlockGeneric // We need to continue this work + f.stepState = stateDict + return + } + goto readLiteral + } +} + +// Copy a single uncompressed data block from input to output. +func (f *decompressor) dataBlock() { + // Uncompressed. + // Discard current half-byte. + 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[offBytes:4]) + f.roffset += int64(nr) + if err != nil { + f.err = noEOF(err) + return + } + 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 { + ncomp := ^n + fmt.Println("uint16(nn) != uint16(^n)", nn, ncomp) + } + f.err = CorruptInputError(f.roffset) + return + } + + if n == 0 { + f.toRead = f.dict.readFlush() + f.finishBlock() + return + } + + f.copyLen = int(n) + f.copyData() +} + +// copyData copies f.copyLen bytes from the underlying reader into f.hist. +// It pauses for reads when f.hist is full. +func (f *decompressor) copyData() { + buf := f.dict.writeSlice() + if len(buf) > f.copyLen { + buf = buf[:f.copyLen] + } + + cnt, err := io.ReadFull(f.r, buf) + f.roffset += int64(cnt) + f.copyLen -= cnt + f.dict.writeMark(cnt) + if err != nil { + f.err = noEOF(err) + return + } + + if f.dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = f.dict.readFlush() + f.step = (*decompressor).copyData + return + } + f.finishBlock() +} + +func (f *decompressor) finishBlock() { + if f.final { + if f.dict.availRead() > 0 { + f.toRead = f.dict.readFlush() + } + f.err = io.EOF + } + f.step = (*decompressor).nextBlock +} + +// noEOF returns err, unless err == io.EOF, in which case it returns io.ErrUnexpectedEOF. +func noEOF(e error) error { + if e == io.EOF { + return io.ErrUnexpectedEOF + } + return e +} + +func (f *decompressor) moreBits() error { + c, err := f.r.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + return nil +} + +// Read the next Huffman-encoded symbol from f according to h. +func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { + // 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(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. + nb, b := f.nb, f.b + for { + for nb < n { + c, err := f.r.ReadByte() + if err != nil { + f.b = b + f.nb = nb + return 0, noEOF(err) + } + f.roffset++ + b |= uint32(c) << (nb & 31) + nb += 8 + } + chunk := h.chunks[b&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = h.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&h.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 0, f.err + } + f.b = b >> (n & 31) + f.nb = nb - n + return int(chunk >> huffmanValueShift), nil + } + } +} + +func makeReader(r io.Reader) Reader { + if rr, ok := r.(Reader); ok { + return rr + } + return bufio.NewReader(r) +} + +func fixedHuffmanDecoderInit() { + fixedOnce.Do(func() { + // These come from the RFC section 3.2.6. + var bits [288]int + for i := 0; i < 144; i++ { + bits[i] = 8 + } + for i := 144; i < 256; i++ { + bits[i] = 9 + } + for i := 256; i < 280; i++ { + bits[i] = 7 + } + for i := 280; i < 288; i++ { + bits[i] = 8 + } + fixedHuffmanDecoder.init(bits[:]) + }) +} + +func (f *decompressor) Reset(r io.Reader, dict []byte) error { + *f = decompressor{ + r: makeReader(r), + bits: f.bits, + codebits: f.codebits, + h1: f.h1, + h2: f.h2, + dict: f.dict, + step: (*decompressor).nextBlock, + } + f.dict.init(maxMatchOffset, dict) + return nil +} + +// NewReader returns a new ReadCloser that can be used +// to read the uncompressed version of r. +// If r does not also implement io.ByteReader, +// the decompressor may read more data than necessary from r. +// It is the caller's responsibility to call Close on the ReadCloser +// when finished reading. +// +// The ReadCloser returned by NewReader also implements Resetter. +func NewReader(r io.Reader) io.ReadCloser { + fixedHuffmanDecoderInit() + + var f decompressor + f.r = makeReader(r) + f.bits = new([maxNumLit + maxNumDist]int) + f.codebits = new([numCodes]int) + f.step = (*decompressor).nextBlock + f.dict.init(maxMatchOffset, nil) + return &f +} + +// NewReaderDict is like NewReader but initializes the reader +// with a preset dictionary. The returned Reader behaves as if +// the uncompressed data stream started with the given dictionary, +// which has already been read. NewReaderDict is typically used +// to read data compressed by NewWriterDict. +// +// The ReadCloser returned by NewReader also implements Resetter. +func NewReaderDict(r io.Reader, dict []byte) io.ReadCloser { + fixedHuffmanDecoderInit() + + var f decompressor + f.r = makeReader(r) + f.bits = new([maxNumLit + maxNumDist]int) + f.codebits = new([numCodes]int) + f.step = (*decompressor).nextBlock + f.dict.init(maxMatchOffset, dict) + return &f +} diff --git a/github.com/klauspost/compress/flate/inflate_gen.go b/github.com/klauspost/compress/flate/inflate_gen.go new file mode 100644 index 0000000..397dc1b --- /dev/null +++ b/github.com/klauspost/compress/flate/inflate_gen.go @@ -0,0 +1,922 @@ +// 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) + moreBits := func() error { + c, err := fr.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + return nil + } + + 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 & 31) + 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 & 31) + 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).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) + n = 0 + case v < 269: + length = v*2 - (265*2 - 11) + n = 1 + case v < 273: + length = v*4 - (269*4 - 19) + n = 2 + case v < 277: + length = v*8 - (273*8 - 35) + n = 3 + case v < 281: + length = v*16 - (277*16 - 67) + n = 4 + case v < 285: + length = v*32 - (281*32 - 131) + n = 5 + case v < maxNumLit: + length = 258 + n = 0 + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + if n > 0 { + for f.nb < n { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + } + length += int(f.b & uint32(1<>= n + f.nb -= n + } + + var dist int + if f.hd == nil { + for f.nb < 5 { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + } + dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + if dist, err = f.huffSym(f.hd); err != nil { + if debugDecode { + fmt.Println("huffsym:", err) + } + f.err = err + return + } + } + + 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 + for f.nb < nb { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb + f.nb -= nb + dist = 1<<(nb+1) + 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 > 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, 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) + moreBits := func() error { + c, err := fr.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + return nil + } + + 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 & 31) + 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 & 31) + 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).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) + n = 0 + case v < 269: + length = v*2 - (265*2 - 11) + n = 1 + case v < 273: + length = v*4 - (269*4 - 19) + n = 2 + case v < 277: + length = v*8 - (273*8 - 35) + n = 3 + case v < 281: + length = v*16 - (277*16 - 67) + n = 4 + case v < 285: + length = v*32 - (281*32 - 131) + n = 5 + case v < maxNumLit: + length = 258 + n = 0 + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + if n > 0 { + for f.nb < n { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + } + length += int(f.b & uint32(1<>= n + f.nb -= n + } + + var dist int + if f.hd == nil { + for f.nb < 5 { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + } + dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + if dist, err = f.huffSym(f.hd); err != nil { + if debugDecode { + fmt.Println("huffsym:", err) + } + f.err = err + return + } + } + + 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 + for f.nb < nb { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb + f.nb -= nb + dist = 1<<(nb+1) + 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 > 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, 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) + moreBits := func() error { + c, err := fr.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + return nil + } + + 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 & 31) + 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 & 31) + 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).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) + n = 0 + case v < 269: + length = v*2 - (265*2 - 11) + n = 1 + case v < 273: + length = v*4 - (269*4 - 19) + n = 2 + case v < 277: + length = v*8 - (273*8 - 35) + n = 3 + case v < 281: + length = v*16 - (277*16 - 67) + n = 4 + case v < 285: + length = v*32 - (281*32 - 131) + n = 5 + case v < maxNumLit: + length = 258 + n = 0 + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + if n > 0 { + for f.nb < n { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + } + length += int(f.b & uint32(1<>= n + f.nb -= n + } + + var dist int + if f.hd == nil { + for f.nb < 5 { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + } + dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + if dist, err = f.huffSym(f.hd); err != nil { + if debugDecode { + fmt.Println("huffsym:", err) + } + f.err = err + return + } + } + + 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 + for f.nb < nb { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb + f.nb -= nb + dist = 1<<(nb+1) + 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 > 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, 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) + moreBits := func() error { + c, err := fr.ReadByte() + if err != nil { + return noEOF(err) + } + f.roffset++ + f.b |= uint32(c) << f.nb + f.nb += 8 + return nil + } + + 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 & 31) + 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 & 31) + 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).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) + n = 0 + case v < 269: + length = v*2 - (265*2 - 11) + n = 1 + case v < 273: + length = v*4 - (269*4 - 19) + n = 2 + case v < 277: + length = v*8 - (273*8 - 35) + n = 3 + case v < 281: + length = v*16 - (277*16 - 67) + n = 4 + case v < 285: + length = v*32 - (281*32 - 131) + n = 5 + case v < maxNumLit: + length = 258 + n = 0 + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + return + } + if n > 0 { + for f.nb < n { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + } + length += int(f.b & uint32(1<>= n + f.nb -= n + } + + var dist int + if f.hd == nil { + for f.nb < 5 { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + } + dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) + f.b >>= 5 + f.nb -= 5 + } else { + if dist, err = f.huffSym(f.hd); err != nil { + if debugDecode { + fmt.Println("huffsym:", err) + } + f.err = err + return + } + } + + 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 + for f.nb < nb { + if err = moreBits(); err != nil { + if debugDecode { + fmt.Println("morebits f.nb>= nb + f.nb -= nb + dist = 1<<(nb+1) + 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 > 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, 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/github.com/klauspost/compress/flate/inflate_test.go b/github.com/klauspost/compress/flate/inflate_test.go new file mode 100644 index 0000000..8402c0c --- /dev/null +++ b/github.com/klauspost/compress/flate/inflate_test.go @@ -0,0 +1,282 @@ +// Copyright 2014 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 + +import ( + "bytes" + "crypto/rand" + "io" + "io/ioutil" + "strconv" + "strings" + "testing" +) + +func TestReset(t *testing.T) { + ss := []string{ + "lorem ipsum izzle fo rizzle", + "the quick brown fox jumped over", + } + + deflated := make([]bytes.Buffer, 2) + for i, s := range ss { + w, _ := NewWriter(&deflated[i], 1) + w.Write([]byte(s)) + w.Close() + } + + inflated := make([]bytes.Buffer, 2) + + f := NewReader(&deflated[0]) + io.Copy(&inflated[0], f) + f.(Resetter).Reset(&deflated[1], nil) + io.Copy(&inflated[1], f) + f.Close() + + for i, s := range ss { + if s != inflated[i].String() { + t.Errorf("inflated[%d]:\ngot %q\nwant %q", i, inflated[i], s) + } + } +} + +func TestReaderTruncated(t *testing.T) { + vectors := []struct{ input, output string }{ + {"\x00", ""}, + {"\x00\f", ""}, + {"\x00\f\x00", ""}, + {"\x00\f\x00\xf3\xff", ""}, + {"\x00\f\x00\xf3\xffhello", "hello"}, + {"\x00\f\x00\xf3\xffhello, world", "hello, world"}, + {"\x02", ""}, + {"\xf2H\xcd", "He"}, + {"\xf2H͙0a\u0084\t", "Hel\x90\x90\x90\x90\x90"}, + {"\xf2H͙0a\u0084\t\x00", "Hel\x90\x90\x90\x90\x90"}, + } + + for i, v := range vectors { + r := strings.NewReader(v.input) + zr := NewReader(r) + b, err := ioutil.ReadAll(zr) + if err != io.ErrUnexpectedEOF { + t.Errorf("test %d, error mismatch: got %v, want io.ErrUnexpectedEOF", i, err) + } + if string(b) != v.output { + t.Errorf("test %d, output mismatch: got %q, want %q", i, b, v.output) + } + } +} + +func TestResetDict(t *testing.T) { + dict := []byte("the lorem fox") + ss := []string{ + "lorem ipsum izzle fo rizzle", + "the quick brown fox jumped over", + } + + deflated := make([]bytes.Buffer, len(ss)) + for i, s := range ss { + w, _ := NewWriterDict(&deflated[i], DefaultCompression, dict) + w.Write([]byte(s)) + w.Close() + } + + inflated := make([]bytes.Buffer, len(ss)) + + f := NewReader(nil) + for i := range inflated { + f.(Resetter).Reset(&deflated[i], dict) + io.Copy(&inflated[i], f) + } + f.Close() + + for i, s := range ss { + if s != inflated[i].String() { + t.Errorf("inflated[%d]:\ngot %q\nwant %q", i, inflated[i], s) + } + } +} + +// Tests ported from zlib/test/infcover.c +type infTest struct { + hex string + id string + n int +} + +var infTests = []infTest{ + {"0 0 0 0 0", "invalid stored block lengths", 1}, + {"3 0", "fixed", 0}, + {"6", "invalid block type", 1}, + {"1 1 0 fe ff 0", "stored", 0}, + {"fc 0 0", "too many length or distance symbols", 1}, + {"4 0 fe ff", "invalid code lengths set", 1}, + {"4 0 24 49 0", "invalid bit length repeat", 1}, + {"4 0 24 e9 ff ff", "invalid bit length repeat", 1}, + {"4 0 24 e9 ff 6d", "invalid code -- missing end-of-block", 1}, + {"4 80 49 92 24 49 92 24 71 ff ff 93 11 0", "invalid literal/lengths set", 1}, + {"4 80 49 92 24 49 92 24 f b4 ff ff c3 84", "invalid distances set", 1}, + {"4 c0 81 8 0 0 0 0 20 7f eb b 0 0", "invalid literal/length code", 1}, + {"2 7e ff ff", "invalid distance code", 1}, + {"c c0 81 0 0 0 0 0 90 ff 6b 4 0", "invalid distance too far back", 1}, + + // also trailer mismatch just in inflate() + {"1f 8b 8 0 0 0 0 0 0 0 3 0 0 0 0 1", "incorrect data check", -1}, + {"1f 8b 8 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 1", "incorrect length check", -1}, + {"5 c0 21 d 0 0 0 80 b0 fe 6d 2f 91 6c", "pull 17", 0}, + {"5 e0 81 91 24 cb b2 2c 49 e2 f 2e 8b 9a 47 56 9f fb fe ec d2 ff 1f", "long code", 0}, + {"ed c0 1 1 0 0 0 40 20 ff 57 1b 42 2c 4f", "length extra", 0}, + {"ed cf c1 b1 2c 47 10 c4 30 fa 6f 35 1d 1 82 59 3d fb be 2e 2a fc f c", "long distance and extra", 0}, + {"ed c0 81 0 0 0 0 80 a0 fd a9 17 a9 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 6", "window end", 0}, +} + +func TestInflate(t *testing.T) { + for _, test := range infTests { + hex := strings.Split(test.hex, " ") + data := make([]byte, len(hex)) + for i, h := range hex { + b, _ := strconv.ParseInt(h, 16, 32) + data[i] = byte(b) + } + buf := bytes.NewReader(data) + r := NewReader(buf) + + _, err := io.Copy(ioutil.Discard, r) + if (test.n == 0 && err == nil) || (test.n != 0 && err != nil) { + t.Logf("%q: OK:", test.id) + t.Logf(" - got %v", err) + continue + } + + if test.n == 0 && err != nil { + t.Errorf("%q: Expected no error, but got %v", test.id, err) + continue + } + + if test.n != 0 && err == nil { + t.Errorf("%q:Expected an error, but got none", test.id) + continue + } + t.Fatal(test.n, err) + } + + for _, test := range infOutTests { + hex := strings.Split(test.hex, " ") + data := make([]byte, len(hex)) + for i, h := range hex { + b, _ := strconv.ParseInt(h, 16, 32) + data[i] = byte(b) + } + buf := bytes.NewReader(data) + r := NewReader(buf) + + _, err := io.Copy(ioutil.Discard, r) + if test.err == (err != nil) { + t.Logf("%q: OK:", test.id) + t.Logf(" - got %v", err) + continue + } + + if test.err == false && err != nil { + t.Errorf("%q: Expected no error, but got %v", test.id, err) + continue + } + + if test.err && err == nil { + t.Errorf("%q: Expected an error, but got none", test.id) + continue + } + t.Fatal(test.err, err) + } + +} + +// Tests ported from zlib/test/infcover.c +// Since zlib inflate is push (writer) instead of pull (reader) +// some of the window size tests have been removed, since they +// are irrelevant. +type infOutTest struct { + hex string + id string + step int + win int + length int + err bool +} + +var infOutTests = []infOutTest{ + {"2 8 20 80 0 3 0", "inflate_fast TYPE return", 0, -15, 258, false}, + {"63 18 5 40 c 0", "window wrap", 3, -8, 300, false}, + {"e5 e0 81 ad 6d cb b2 2c c9 01 1e 59 63 ae 7d ee fb 4d fd b5 35 41 68 ff 7f 0f 0 0 0", "fast length extra bits", 0, -8, 258, true}, + {"25 fd 81 b5 6d 59 b6 6a 49 ea af 35 6 34 eb 8c b9 f6 b9 1e ef 67 49 50 fe ff ff 3f 0 0", "fast distance extra bits", 0, -8, 258, true}, + {"3 7e 0 0 0 0 0", "fast invalid distance code", 0, -8, 258, true}, + {"1b 7 0 0 0 0 0", "fast invalid literal/length code", 0, -8, 258, true}, + {"d c7 1 ae eb 38 c 4 41 a0 87 72 de df fb 1f b8 36 b1 38 5d ff ff 0", "fast 2nd level codes and too far back", 0, -8, 258, true}, + {"63 18 5 8c 10 8 0 0 0 0", "very common case", 0, -8, 259, false}, + {"63 60 60 18 c9 0 8 18 18 18 26 c0 28 0 29 0 0 0", "contiguous and wrap around window", 6, -8, 259, false}, + {"63 0 3 0 0 0 0 0", "copy direct from output", 0, -8, 259, false}, + {"1f 8b 0 0", "bad gzip method", 0, 31, 0, true}, + {"1f 8b 8 80", "bad gzip flags", 0, 31, 0, true}, + {"77 85", "bad zlib method", 0, 15, 0, true}, + {"78 9c", "bad zlib window size", 0, 8, 0, true}, + {"1f 8b 8 1e 0 0 0 0 0 0 1 0 0 0 0 0 0", "bad header crc", 0, 47, 1, true}, + {"1f 8b 8 2 0 0 0 0 0 0 1d 26 3 0 0 0 0 0 0 0 0 0", "check gzip length", 0, 47, 0, true}, + {"78 90", "bad zlib header check", 0, 47, 0, true}, + {"8 b8 0 0 0 1", "need dictionary", 0, 8, 0, true}, + {"63 18 68 30 d0 0 0", "force split window update", 4, -8, 259, false}, + {"3 0", "use fixed blocks", 0, -15, 1, false}, + {"", "bad window size", 0, 1, 0, true}, +} + +func TestWriteTo(t *testing.T) { + input := make([]byte, 100000) + n, err := rand.Read(input) + if err != nil { + t.Fatal(err) + } + if n != len(input) { + t.Fatal("did not fill buffer") + } + compressed := &bytes.Buffer{} + w, err := NewWriter(compressed, -2) + if err != nil { + t.Fatal(err) + } + n, err = w.Write(input) + if err != nil { + t.Fatal(err) + } + if n != len(input) { + t.Fatal("did not fill buffer") + } + w.Close() + buf := compressed.Bytes() + + dec := NewReader(bytes.NewBuffer(buf)) + // ReadAll does not use WriteTo, but we wrap it in a NopCloser to be sure. + readall, err := ioutil.ReadAll(ioutil.NopCloser(dec)) + if err != nil { + t.Fatal(err) + } + if len(readall) != len(input) { + t.Fatal("did not decompress everything") + } + + dec = NewReader(bytes.NewBuffer(buf)) + wtbuf := &bytes.Buffer{} + written, err := dec.(io.WriterTo).WriteTo(wtbuf) + if err != nil { + t.Fatal(err) + } + if written != int64(len(input)) { + t.Error("Returned length did not match, expected", len(input), "got", written) + } + if wtbuf.Len() != len(input) { + t.Error("Actual Length did not match, expected", len(input), "got", wtbuf.Len()) + } + if bytes.Compare(wtbuf.Bytes(), input) != 0 { + t.Fatal("output did not match input") + } +} diff --git a/github.com/klauspost/compress/flate/level1.go b/github.com/klauspost/compress/flate/level1.go new file mode 100644 index 0000000..1e5eea3 --- /dev/null +++ b/github.com/klauspost/compress/flate/level1.go @@ -0,0 +1,179 @@ +package flate + +import "fmt" + +// fastGen maintains the table for matches, +// and the previous byte block for level 2. +// This is the generic implementation. +type fastEncL1 struct { + fastGen + table [tableSize]tableEntry +} + +// EncodeL1 uses a similar algorithm to level 1 +func (e *fastEncL1) Encode(dst *tokens, src []byte) { + const ( + 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 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // 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 := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3232(src, s) + + for { + const skipLog = 5 + const doEvery = 2 + + nextS := s + var candidate tableEntry + for { + nextHash := hash(cv) + candidate = e.table[nextHash] + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + + now := load6432(src, nextS) + e.table[nextHash] = tableEntry{offset: s + e.cur} + nextHash = hash(uint32(now)) + + offset := s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} + break + } + + // Do one right away... + cv = uint32(now) + s = nextS + nextS++ + candidate = e.table[nextHash] + now >>= 8 + e.table[nextHash] = tableEntry{offset: s + e.cur} + + offset = s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} + break + } + cv = uint32(now) + s = nextS + } + + // 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. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + + // Extend the 4-byte match as long as possible. + t := candidate.offset - e.cur + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + + // Save the match found + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + if s >= sLimit { + // 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} + } + goto emitRemainder + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 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 := load6432(src, s-2) + o := e.cur + s - 2 + prevHash := hash(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} + + offset := s - (candidate.offset - e.cur) + if offset > maxMatchOffset || uint32(x) != load3232(src, candidate.offset-e.cur) { + cv = uint32(x >> 8) + s++ + break + } + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/level2.go b/github.com/klauspost/compress/flate/level2.go new file mode 100644 index 0000000..5b986a1 --- /dev/null +++ b/github.com/klauspost/compress/flate/level2.go @@ -0,0 +1,205 @@ +package flate + +import "fmt" + +// fastGen maintains the table for matches, +// and the previous byte block for level 2. +// This is the generic implementation. +type fastEncL2 struct { + fastGen + table [bTableSize]tableEntry +} + +// EncodeL2 uses a similar algorithm to level 1, but is capable +// of matching across blocks giving better compression at a small slowdown. +func (e *fastEncL2) Encode(dst *tokens, src []byte) { + const ( + 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 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // 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 := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3232(src, s) + for { + // When should we start skipping if we haven't found matches in a long while. + const skipLog = 5 + const doEvery = 2 + + nextS := s + var candidate tableEntry + for { + nextHash := hash4u(cv, bTableBits) + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + candidate = e.table[nextHash] + now := load6432(src, nextS) + e.table[nextHash] = tableEntry{offset: s + e.cur} + nextHash = hash4u(uint32(now), bTableBits) + + offset := s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + e.table[nextHash] = tableEntry{offset: nextS + e.cur} + break + } + + // Do one right away... + cv = uint32(now) + s = nextS + nextS++ + candidate = e.table[nextHash] + now >>= 8 + e.table[nextHash] = tableEntry{offset: s + e.cur} + + offset = s - (candidate.offset - e.cur) + if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + break + } + cv = uint32(now) + } + + // 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. + + // 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. + + // Extend the 4-byte match as long as possible. + t := candidate.offset - e.cur + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + // 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} + } + goto emitRemainder + } + + // Store every second hash in-between, but offset by 1. + 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} + // Skip one + x >>= 16 + nextHash = hash4u(uint32(x), bTableBits) + 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} + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 to 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 := load6432(src, s-2) + o := e.cur + s - 2 + prevHash := hash4u(uint32(x), bTableBits) + prevHash2 := hash4u(uint32(x>>8), bTableBits) + 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} + + offset := s - (candidate.offset - e.cur) + if offset > maxMatchOffset || uint32(x>>16) != load3232(src, candidate.offset-e.cur) { + cv = uint32(x >> 24) + s++ + break + } + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/level3.go b/github.com/klauspost/compress/flate/level3.go new file mode 100644 index 0000000..c22b424 --- /dev/null +++ b/github.com/klauspost/compress/flate/level3.go @@ -0,0 +1,229 @@ +package flate + +import "fmt" + +// fastEncL3 +type fastEncL3 struct { + fastGen + table [tableSize]tableEntryPrev +} + +// Encode uses a similar algorithm to level 2, will check up to two candidates. +func (e *fastEncL3) Encode(dst *tokens, src []byte) { + const ( + inputMargin = 8 - 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 { + for i := range e.table[:] { + e.table[i] = tableEntryPrev{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i] + if v.Cur.offset <= minOff { + v.Cur.offset = 0 + } else { + v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset + } + if v.Prev.offset <= minOff { + v.Prev.offset = 0 + } else { + v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset + } + e.table[i] = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // Skip if too small. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // 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 := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3232(src, s) + for { + const skipLog = 6 + nextS := s + var candidate tableEntry + for { + nextHash := hash(cv) + s = nextS + nextS = s + 1 + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + candidates := e.table[nextHash] + now := load3232(src, nextS) + + // 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 + 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 + } + break + } else { + // We only check if value mismatches. + // Offset will always be invalid in other cases. + candidate = candidates.Prev + if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { + break + } + } + cv = now + } + + // 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. + + // Extend the 4-byte match as long as possible. + // + t := candidate.offset - e.cur + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + t += l + // Index first pair after match end. + if int(t+4) < len(src) && t > 0 { + cv := load3232(src, t) + nextHash := hash(cv) + e.table[nextHash] = tableEntryPrev{ + Prev: e.table[nextHash].Cur, + Cur: tableEntry{offset: e.cur + t}, + } + } + goto emitRemainder + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-3 to s. + x := load6432(src, s-3) + prevHash := hash(uint32(x)) + e.table[prevHash] = tableEntryPrev{ + Prev: e.table[prevHash].Cur, + 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}, + } + x >>= 8 + prevHash = hash(uint32(x)) + + e.table[prevHash] = tableEntryPrev{ + Prev: e.table[prevHash].Cur, + Cur: tableEntry{offset: e.cur + s - 1}, + } + x >>= 8 + currHash := hash(uint32(x)) + candidates := e.table[currHash] + cv = uint32(x) + e.table[currHash] = tableEntryPrev{ + Prev: candidates.Cur, + Cur: tableEntry{offset: s + e.cur}, + } + + // Check both candidates + candidate = candidates.Cur + 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 candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { + offset := s - (candidate.offset - e.cur) + if offset <= maxMatchOffset { + continue + } + } + } + cv = uint32(x >> 8) + s++ + break + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/level4.go b/github.com/klauspost/compress/flate/level4.go new file mode 100644 index 0000000..e62f0c0 --- /dev/null +++ b/github.com/klauspost/compress/flate/level4.go @@ -0,0 +1,212 @@ +package flate + +import "fmt" + +type fastEncL4 struct { + fastGen + table [tableSize]tableEntry + bTable [tableSize]tableEntry +} + +func (e *fastEncL4) Encode(dst *tokens, src []byte) { + const ( + 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 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.bTable[:] { + e.bTable[i] = tableEntry{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + for i := range e.bTable[:] { + v := e.bTable[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.bTable[i].offset = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // 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 := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load6432(src, s) + for { + const skipLog = 6 + const doEvery = 1 + + nextS := s + var t int32 + for { + nextHashS := hash4x64(cv, tableBits) + nextHashL := hash7(cv, tableBits) + + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + // Fetch a short+long candidate + sCandidate := e.table[nextHashS] + lCandidate := e.bTable[nextHashL] + next := load6432(src, nextS) + 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) == 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) == 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 && 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 + t = lCandidate.offset - e.cur + } + } + break + } + cv = next + } + + // 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. + + // Extend the 4-byte match as long as possible. + l := e.matchlenLong(s+4, t+4, src) + 4 + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + if debugDeflate { + if t >= s { + panic("s-t") + } + if (s - t) > maxMatchOffset { + panic(fmt.Sprintln("mmo", t)) + } + if l < baseMatchLength { + panic("bml") + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + // 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} + e.bTable[hash7(cv, tableBits)] = tableEntry{offset: s + e.cur} + } + goto emitRemainder + } + + // Store every 3rd hash in-between + if true { + i := nextS + if i < s-1 { + cv := load6432(src, i) + 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(uint32(cv>>8), tableBits)] = t2 + + i += 3 + for ; i < s-1; i += 3 { + cv := load6432(src, i) + 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(uint32(cv>>8), tableBits)] = t2 + } + } + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. + x := load6432(src, s-1) + o := e.cur + s - 1 + prevHashS := hash4x64(x, tableBits) + prevHashL := hash7(x, tableBits) + e.table[prevHashS] = tableEntry{offset: o} + e.bTable[prevHashL] = tableEntry{offset: o} + cv = x >> 8 + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/level5.go b/github.com/klauspost/compress/flate/level5.go new file mode 100644 index 0000000..d513f1f --- /dev/null +++ b/github.com/klauspost/compress/flate/level5.go @@ -0,0 +1,279 @@ +package flate + +import "fmt" + +type fastEncL5 struct { + fastGen + table [tableSize]tableEntry + bTable [tableSize]tableEntryPrev +} + +func (e *fastEncL5) Encode(dst *tokens, src []byte) { + const ( + 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 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.bTable[:] { + e.bTable[i] = tableEntryPrev{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + for i := range e.bTable[:] { + v := e.bTable[i] + if v.Cur.offset <= minOff { + v.Cur.offset = 0 + v.Prev.offset = 0 + } else { + v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset + if v.Prev.offset <= minOff { + v.Prev.offset = 0 + } else { + v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset + } + } + e.bTable[i] = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // 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 := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load6432(src, s) + for { + const skipLog = 6 + const doEvery = 1 + + nextS := s + var l int32 + var t int32 + for { + nextHashS := hash4x64(cv, tableBits) + nextHashL := hash7(cv, tableBits) + + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + // Fetch a short+long candidate + sCandidate := e.table[nextHashS] + lCandidate := e.bTable[nextHashL] + next := load6432(src, nextS) + entry := tableEntry{offset: s + e.cur} + e.table[nextHashS] = entry + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = entry, eLong.Cur + + nextHashS = hash4x64(next, tableBits) + nextHashL = hash7(next, tableBits) + + t = lCandidate.Cur.offset - e.cur + if s-t < maxMatchOffset { + if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) { + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + + t2 := lCandidate.Prev.offset - e.cur + 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 { + t = t2 + l = ml1 + break + } + } + break + } + t = lCandidate.Prev.offset - e.cur + 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} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + break + } + } + + t = sCandidate.offset - e.cur + 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} + eLong := &e.bTable[nextHashL] + 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 load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) { + ml := e.matchlen(nextS+4, t2+4, src) + 4 + if ml > l { + t = t2 + s = nextS + l = ml + break + } + } + // If the previous long is a candidate, use that... + t2 = lCandidate.Prev.offset - e.cur + 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 + s = nextS + l = ml + break + } + } + } + break + } + cv = next + } + + // 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. + + // Extend the 4-byte match as long as possible. + if l == 0 { + l = e.matchlenLong(s+4, t+4, src) + 4 + } else if l == maxMatchLength { + l += e.matchlenLong(s+l, t+l, src) + } + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + if debugDeflate { + if t >= s { + panic(fmt.Sprintln("s-t", s, t)) + } + if (s - t) > maxMatchOffset { + panic(fmt.Sprintln("mmo", s-t)) + } + if l < baseMatchLength { + panic("bml") + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + goto emitRemainder + } + + // Store every 3rd hash in-between. + if true { + const hashEvery = 3 + i := s - l + 1 + if i < s-1 { + cv := load6432(src, i) + 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} + 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} + 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} + t2 := tableEntry{offset: t.offset + 1} + eLong := &e.bTable[hash7(cv, tableBits)] + eLong.Cur, eLong.Prev = t, eLong.Cur + e.table[hash4u(uint32(cv>>8), tableBits)] = t2 + } + } + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. + x := load6432(src, s-1) + o := e.cur + s - 1 + prevHashS := hash4x64(x, tableBits) + prevHashL := hash7(x, tableBits) + e.table[prevHashS] = tableEntry{offset: o} + eLong := &e.bTable[prevHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: o}, eLong.Cur + cv = x >> 8 + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/level6.go b/github.com/klauspost/compress/flate/level6.go new file mode 100644 index 0000000..a52c80e --- /dev/null +++ b/github.com/klauspost/compress/flate/level6.go @@ -0,0 +1,282 @@ +package flate + +import "fmt" + +type fastEncL6 struct { + fastGen + table [tableSize]tableEntry + bTable [tableSize]tableEntryPrev +} + +func (e *fastEncL6) Encode(dst *tokens, src []byte) { + const ( + 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 { + for i := range e.table[:] { + e.table[i] = tableEntry{} + } + for i := range e.bTable[:] { + e.bTable[i] = tableEntryPrev{} + } + e.cur = maxMatchOffset + break + } + // Shift down everything in the table that isn't already too far away. + minOff := e.cur + int32(len(e.hist)) - maxMatchOffset + for i := range e.table[:] { + v := e.table[i].offset + if v <= minOff { + v = 0 + } else { + v = v - e.cur + maxMatchOffset + } + e.table[i].offset = v + } + for i := range e.bTable[:] { + v := e.bTable[i] + if v.Cur.offset <= minOff { + v.Cur.offset = 0 + v.Prev.offset = 0 + } else { + v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset + if v.Prev.offset <= minOff { + v.Prev.offset = 0 + } else { + v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset + } + } + e.bTable[i] = v + } + e.cur = maxMatchOffset + } + + s := e.addBlock(src) + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + dst.n = uint16(len(src)) + return + } + + // Override src + src = e.hist + nextEmit := s + + // 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 := int32(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load6432(src, s) + // Repeat MUST be > 1 and within range + repeat := int32(1) + for { + const skipLog = 7 + const doEvery = 1 + + nextS := s + var l int32 + var t int32 + for { + nextHashS := hash4x64(cv, tableBits) + nextHashL := hash7(cv, tableBits) + s = nextS + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit { + goto emitRemainder + } + // Fetch a short+long candidate + sCandidate := e.table[nextHashS] + lCandidate := e.bTable[nextHashL] + next := load6432(src, nextS) + entry := tableEntry{offset: s + e.cur} + e.table[nextHashS] = entry + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = entry, eLong.Cur + + // Calculate hashes of 'next' + nextHashS = hash4x64(next, tableBits) + nextHashL = hash7(next, tableBits) + + t = lCandidate.Cur.offset - e.cur + if s-t < maxMatchOffset { + 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} + eLong := &e.bTable[nextHashL] + 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) == 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 { + t = t2 + l = ml1 + break + } + } + break + } + // Current value did not match, but check if previous long value does. + t = lCandidate.Prev.offset - e.cur + 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} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + break + } + } + + t = sCandidate.offset - e.cur + 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 + + // Look up next long candidate (at nextS) + lCandidate = e.bTable[nextHashL] + + // Store the next match + e.table[nextHashS] = tableEntry{offset: nextS + e.cur} + eLong := &e.bTable[nextHashL] + eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur + + // Check repeat at s + repOff + const repOff = 1 + t2 := s - repeat + repOff + if load3232(src, t2) == uint32(cv>>(8*repOff)) { + ml := e.matchlen(s+4+repOff, t2+4, src) + 4 + if ml > l { + t = t2 + l = ml + s += repOff + // Not worth checking more. + break + } + } + + // If the next long is a candidate, use that... + t2 = lCandidate.Cur.offset - e.cur + if nextS-t2 < maxMatchOffset { + if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) { + ml := e.matchlen(nextS+4, t2+4, src) + 4 + if ml > l { + t = t2 + s = nextS + l = ml + // This is ok, but check previous as well. + } + } + // If the previous long is a candidate, use that... + t2 = lCandidate.Prev.offset - e.cur + 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 + s = nextS + l = ml + break + } + } + } + break + } + cv = next + } + + // 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. + + // Extend the 4-byte match as long as possible. + if l == 0 { + l = e.matchlenLong(s+4, t+4, src) + 4 + } else if l == maxMatchLength { + l += e.matchlenLong(s+l, t+l, src) + } + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + if false { + if t >= s { + panic(fmt.Sprintln("s-t", s, t)) + } + if (s - t) > maxMatchOffset { + panic(fmt.Sprintln("mmo", s-t)) + } + if l < baseMatchLength { + panic("bml") + } + } + + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + repeat = s - t + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + + if s >= sLimit { + // 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} + eLong := &e.bTable[hash7(cv, tableBits)] + eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur}, eLong.Cur + } + goto emitRemainder + } + + // Store every long hash in-between and every second short. + if true { + for i := nextS + 1; i < s-1; i += 2 { + cv := load6432(src, i) + 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 + eLong.Cur, eLong.Prev = t, eLong.Cur + eLong2.Cur, eLong2.Prev = t2, eLong2.Cur + } + } + + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-1 and at s. + cv = load6432(src, s) + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/reader_test.go b/github.com/klauspost/compress/flate/reader_test.go new file mode 100644 index 0000000..5543964 --- /dev/null +++ b/github.com/klauspost/compress/flate/reader_test.go @@ -0,0 +1,106 @@ +// Copyright 2012 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 + +import ( + "bytes" + "io" + "io/ioutil" + "runtime" + "strings" + "testing" +) + +func TestNlitOutOfRange(t *testing.T) { + // Trying to decode this bogus flate data, which has a Huffman table + // with nlit=288, should not panic. + io.Copy(ioutil.Discard, NewReader(strings.NewReader( + "\xfc\xfe\x36\xe7\x5e\x1c\xef\xb3\x55\x58\x77\xb6\x56\xb5\x43\xf4"+ + "\x6f\xf2\xd2\xe6\x3d\x99\xa0\x85\x8c\x48\xeb\xf8\xda\x83\x04\x2a"+ + "\x75\xc4\xf8\x0f\x12\x11\xb9\xb4\x4b\x09\xa0\xbe\x8b\x91\x4c"))) +} + +const ( + digits = iota + twain + random +) + +var testfiles = []string{ + // Digits is the digits of the irrational number e. Its decimal representation + // does not repeat, but there are only 10 possible digits, so it should be + // reasonably compressible. + digits: "../testdata/e.txt", + // Twain is Project Gutenberg's edition of Mark Twain's classic English novel. + twain: "../testdata/Mark.Twain-Tom.Sawyer.txt", + // Random bytes + random: "../testdata/sharnd.out", +} + +func benchmarkDecode(b *testing.B, testfile, level, n int) { + b.ReportAllocs() + b.StopTimer() + b.SetBytes(int64(n)) + buf0, err := ioutil.ReadFile(testfiles[testfile]) + if err != nil { + b.Fatal(err) + } + if len(buf0) == 0 { + b.Fatalf("test file %q has no data", testfiles[testfile]) + } + compressed := new(bytes.Buffer) + w, err := NewWriter(compressed, level) + if err != nil { + b.Fatal(err) + } + for i := 0; i < n; i += len(buf0) { + if len(buf0) > n-i { + buf0 = buf0[:n-i] + } + io.Copy(w, bytes.NewReader(buf0)) + } + w.Close() + buf1 := compressed.Bytes() + buf0, compressed, w = nil, nil, nil + runtime.GC() + b.StartTimer() + r := NewReader(bytes.NewReader(buf1)) + res := r.(Resetter) + for i := 0; i < b.N; i++ { + res.Reset(bytes.NewReader(buf1), nil) + io.Copy(ioutil.Discard, r) + } +} + +// These short names are so that gofmt doesn't break the BenchmarkXxx function +// bodies below over multiple lines. +const ( + constant = ConstantCompression + speed = BestSpeed + default_ = DefaultCompression + compress = BestCompression +) + +func BenchmarkDecodeDigitsSpeed1e4(b *testing.B) { benchmarkDecode(b, digits, speed, 1e4) } +func BenchmarkDecodeDigitsSpeed1e5(b *testing.B) { benchmarkDecode(b, digits, speed, 1e5) } +func BenchmarkDecodeDigitsSpeed1e6(b *testing.B) { benchmarkDecode(b, digits, speed, 1e6) } +func BenchmarkDecodeDigitsDefault1e4(b *testing.B) { benchmarkDecode(b, digits, default_, 1e4) } +func BenchmarkDecodeDigitsDefault1e5(b *testing.B) { benchmarkDecode(b, digits, default_, 1e5) } +func BenchmarkDecodeDigitsDefault1e6(b *testing.B) { benchmarkDecode(b, digits, default_, 1e6) } +func BenchmarkDecodeDigitsCompress1e4(b *testing.B) { benchmarkDecode(b, digits, compress, 1e4) } +func BenchmarkDecodeDigitsCompress1e5(b *testing.B) { benchmarkDecode(b, digits, compress, 1e5) } +func BenchmarkDecodeDigitsCompress1e6(b *testing.B) { benchmarkDecode(b, digits, compress, 1e6) } +func BenchmarkDecodeTwainSpeed1e4(b *testing.B) { benchmarkDecode(b, twain, speed, 1e4) } +func BenchmarkDecodeTwainSpeed1e5(b *testing.B) { benchmarkDecode(b, twain, speed, 1e5) } +func BenchmarkDecodeTwainSpeed1e6(b *testing.B) { benchmarkDecode(b, twain, speed, 1e6) } +func BenchmarkDecodeTwainDefault1e4(b *testing.B) { benchmarkDecode(b, twain, default_, 1e4) } +func BenchmarkDecodeTwainDefault1e5(b *testing.B) { benchmarkDecode(b, twain, default_, 1e5) } +func BenchmarkDecodeTwainDefault1e6(b *testing.B) { benchmarkDecode(b, twain, default_, 1e6) } +func BenchmarkDecodeTwainCompress1e4(b *testing.B) { benchmarkDecode(b, twain, compress, 1e4) } +func BenchmarkDecodeTwainCompress1e5(b *testing.B) { benchmarkDecode(b, twain, compress, 1e5) } +func BenchmarkDecodeTwainCompress1e6(b *testing.B) { benchmarkDecode(b, twain, compress, 1e6) } +func BenchmarkDecodeRandomSpeed1e4(b *testing.B) { benchmarkDecode(b, random, speed, 1e4) } +func BenchmarkDecodeRandomSpeed1e5(b *testing.B) { benchmarkDecode(b, random, speed, 1e5) } +func BenchmarkDecodeRandomSpeed1e6(b *testing.B) { benchmarkDecode(b, random, speed, 1e6) } diff --git a/github.com/klauspost/compress/flate/stateless.go b/github.com/klauspost/compress/flate/stateless.go new file mode 100644 index 0000000..53e8991 --- /dev/null +++ b/github.com/klauspost/compress/flate/stateless.go @@ -0,0 +1,297 @@ +package flate + +import ( + "io" + "math" + "sync" +) + +const ( + maxStatelessBlock = math.MaxInt16 + // dictionary will be taken from maxStatelessBlock, so limit it. + maxStatelessDict = 8 << 10 + + slTableBits = 13 + slTableSize = 1 << slTableBits + slTableShift = 32 - slTableBits +) + +type statelessWriter struct { + dst io.Writer + closed bool +} + +func (s *statelessWriter) Close() error { + if s.closed { + return nil + } + s.closed = true + // Emit EOF block + return StatelessDeflate(s.dst, nil, true, nil) +} + +func (s *statelessWriter) Write(p []byte) (n int, err error) { + err = StatelessDeflate(s.dst, p, false, nil) + if err != nil { + return 0, err + } + return len(p), nil +} + +func (s *statelessWriter) Reset(w io.Writer) { + s.dst = w + s.closed = false +} + +// NewStatelessWriter will do compression but without maintaining any state +// between Write calls. +// There will be no memory kept between Write calls, +// but compression and speed will be suboptimal. +// Because of this, the size of actual Write calls will affect output size. +func NewStatelessWriter(dst io.Writer) io.WriteCloser { + return &statelessWriter{dst: dst} +} + +// bitWriterPool contains bit writers that can be reused. +var bitWriterPool = sync.Pool{ + New: func() interface{} { + return newHuffmanBitWriter(nil) + }, +} + +// StatelessDeflate allows to compress directly to a Writer without retaining state. +// When returning everything will be flushed. +// 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) + defer func() { + // don't keep a reference to our output + bw.reset(nil) + bitWriterPool.Put(bw) + }() + if eof && len(in) == 0 { + // Just write an EOF block. + // Could be faster... + bw.writeStoredHeader(0, true) + bw.flush() + return bw.err + } + + // Truncate dict + if len(dict) > maxStatelessDict { + dict = dict[len(dict)-maxStatelessDict:] + } + + for len(in) > 0 { + todo := in + 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, int16(len(dict))) + isEof := eof && len(in) == 0 + + if dst.n == 0 { + bw.writeStoredHeader(len(uncompressed), isEof) + if bw.err != nil { + return bw.err + } + 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, uncompressed, len(in) == 0) + } else { + 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 + } + } + if !eof { + // Align, only a stored block can do that. + bw.writeStoredHeader(0, false) + } + bw.flush() + return bw.err +} + +func hashSL(u uint32) uint32 { + return (u * 0x1e35a7bd) >> slTableShift +} + +func load3216(b []byte, i int16) 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 load6416(b []byte, i int16) 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 statelessEnc(dst *tokens, src []byte, startAt int16) { + const ( + inputMargin = 12 - 1 + minNonLiteralBlockSize = 1 + 1 + inputMargin + ) + + type tableEntry struct { + offset int16 + } + + var table [slTableSize]tableEntry + + // This check isn't in the Snappy implementation, but there, the caller + // instead of the callee handles this case. + if len(src)-int(startAt) < minNonLiteralBlockSize { + // We do not fill the token table. + // This will be picked up by caller. + 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 := 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. + sLimit := int16(len(src) - inputMargin) + + // nextEmit is where in src the next emitLiteral should start from. + cv := load3216(src, s) + + for { + const skipLog = 5 + const doEvery = 2 + + nextS := s + var candidate tableEntry + for { + nextHash := hashSL(cv) + candidate = table[nextHash] + nextS = s + doEvery + (s-nextEmit)>>skipLog + if nextS > sLimit || nextS <= 0 { + goto emitRemainder + } + + now := load6416(src, nextS) + table[nextHash] = tableEntry{offset: s} + nextHash = hashSL(uint32(now)) + + if cv == load3216(src, candidate.offset) { + table[nextHash] = tableEntry{offset: nextS} + break + } + + // Do one right away... + cv = uint32(now) + s = nextS + nextS++ + candidate = table[nextHash] + now >>= 8 + table[nextHash] = tableEntry{offset: s} + + if cv == load3216(src, candidate.offset) { + table[nextHash] = tableEntry{offset: nextS} + break + } + cv = uint32(now) + s = nextS + } + + // 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. + for { + // Invariant: we have a 4-byte match at s, and no need to emit any + // literal bytes prior to s. + + // Extend the 4-byte match as long as possible. + t := candidate.offset + l := int16(matchLen(src[s+4:], src[t+4:]) + 4) + + // Extend backwards + for t > 0 && s > nextEmit && src[t-1] == src[s-1] { + s-- + t-- + l++ + } + if nextEmit < s { + emitLiteral(dst, src[nextEmit:s]) + } + + // Save the match found + dst.AddMatchLong(int32(l), uint32(s-t-baseMatchOffset)) + s += l + nextEmit = s + if nextS >= s { + s = nextS + 1 + } + 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-2 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 := load6416(src, s-2) + o := s - 2 + prevHash := hashSL(uint32(x)) + table[prevHash] = tableEntry{offset: o} + x >>= 16 + currHash := hashSL(uint32(x)) + candidate = table[currHash] + table[currHash] = tableEntry{offset: o + 2} + + if uint32(x) != load3216(src, candidate.offset) { + cv = uint32(x >> 8) + s++ + break + } + } + } + +emitRemainder: + if int(nextEmit) < len(src) { + // If nothing was added, don't encode literals. + if dst.n == 0 { + return + } + emitLiteral(dst, src[nextEmit:]) + } +} diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-null-max.dyn.expect new file mode 100644 index 0000000..0a3c71c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-null-max.dyn.expect-noinput new file mode 100644 index 0000000..0a3c71c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.dyn.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.golden b/github.com/klauspost/compress/flate/testdata/huffman-null-max.golden new file mode 100644 index 0000000..fe7b7f4 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.in b/github.com/klauspost/compress/flate/testdata/huffman-null-max.in new file mode 100644 index 0000000..5dfddf0 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.in differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-null-max.sync.expect new file mode 100644 index 0000000..c081651 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-null-max.sync.expect-noinput new file mode 100644 index 0000000..c081651 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-null-max.wb.expect new file mode 100644 index 0000000..c081651 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-null-max.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-null-max.wb.expect-noinput new file mode 100644 index 0000000..c081651 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-null-max.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-pi.dyn.expect new file mode 100644 index 0000000..11756fe Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-pi.dyn.expect-noinput new file mode 100644 index 0000000..11756fe Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.dyn.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.golden b/github.com/klauspost/compress/flate/testdata/huffman-pi.golden new file mode 100644 index 0000000..05fd911 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.in b/github.com/klauspost/compress/flate/testdata/huffman-pi.in new file mode 100644 index 0000000..efaed43 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-pi.in @@ -0,0 +1 @@ +3.141592653589793238462643383279502884197169399375105820974944592307816406286208998628034825342117067982148086513282306647093844609550582231725359408128481117450284102701938521105559644622948954930381964428810975665933446128475648233786783165271201909145648566923460348610454326648213393607260249141273724587006606315588174881520920962829254091715364367892590360011330530548820466521384146951941511609433057270365759591953092186117381932611793105118548074462379962749567351885752724891227938183011949129833673362440656643086021394946395224737190702179860943702770539217176293176752384674818467669405132000568127145263560827785771342757789609173637178721468440901224953430146549585371050792279689258923542019956112129021960864034418159813629774771309960518707211349999998372978049951059731732816096318595024459455346908302642522308253344685035261931188171010003137838752886587533208381420617177669147303598253490428755468731159562863882353787593751957781857780532171226806613001927876611195909216420198938095257201065485863278865936153381827968230301952035301852968995773622599413891249721775283479131515574857242454150695950829533116861727855889075098381754637464939319255060400927701671139009848824012858361603563707660104710181942955596198946767837449448255379774726847104047534646208046684259069491293313677028989152104752162056966024058038150193511253382430035587640247496473263914199272604269922796782354781636009341721641219924586315030286182974555706749838505494588586926995690927210797509302955321165344987202755960236480665499119881834797753566369807426542527862551818417574672890977772793800081647060016145249192173217214772350141441973568548161361157352552133475741849468438523323907394143334547762416862518983569485562099219222184272550254256887671790494601653466804988627232791786085784383827967976681454100953883786360950680064225125205117392984896084128488626945604241965285022210661186306744278622039194945047123713786960956364371917287467764657573962413890865832645995813390478027590099465764078951269468398352595709825822620522489407726719478268482601476990902640136394437455305068203496252451749399651431429809190659250937221696461515709858387410597885959772975498930161753928468138268683868942774155991855925245953959431049972524680845987273644695848653836736222626099124608051243884390451244136549762780797715691435997700129616089441694868555848406353422072225828488648158456028506016842739452267467678895252138522549954666727823986456596116354886230577456498035593634568174324112515076069479451096596094025228879710893145669136867228748940560101503308617928680920874760917824938589009714909675985261365549781893129784821682998948722658804857564014270477555132379641451523746234364542858444795265867821051141354735739523113427166102135969536231442952484937187110145765403590279934403742007310578539062198387447808478489683321445713868751943506430218453191048481005370614680674919278191197939952061419663428754440643745123718192179998391015919561814675142691239748940907186494231961567945208095146550225231603881930142093762137855956638937787083039069792077346722182562599661501421503068038447734549202605414665925201497442850732518666002132434088190710486331734649651453905796268561005508106658796998163574736384052571459102897064140110971206280439039759515677157700420337869936007230558763176359421873125147120532928191826186125867321579198414848829164470609575270695722091756711672291098169091528017350671274858322287183520935396572512108357915136988209144421006751033467110314126711136990865851639831501970165151168517143765761835155650884909989859982387345528331635507647918535893226185489632132933089857064204675259070915481416549859461637180 \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-pi.sync.expect new file mode 100644 index 0000000..e4396ac Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-pi.sync.expect-noinput new file mode 100644 index 0000000..e4396ac Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-pi.wb.expect new file mode 100644 index 0000000..e4396ac Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-pi.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-pi.wb.expect-noinput new file mode 100644 index 0000000..e4396ac Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-pi.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.dyn.expect new file mode 100644 index 0000000..09dc798 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.dyn.expect-noinput new file mode 100644 index 0000000..5162399 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.dyn.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.golden b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.golden new file mode 100644 index 0000000..09dc798 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.in b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.in new file mode 100644 index 0000000..ce038eb Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.in differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.sync.expect new file mode 100644 index 0000000..09dc798 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.sync.expect-noinput new file mode 100644 index 0000000..0c24742 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.wb.expect new file mode 100644 index 0000000..09dc798 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.wb.expect-noinput new file mode 100644 index 0000000..0c24742 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-1k.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.dyn.expect new file mode 100644 index 0000000..57e5932 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.dyn.expect-noinput new file mode 100644 index 0000000..008b9af --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.dyn.expect-noinput @@ -0,0 +1 @@ +athg{5 FkVklkFk]wݱ3:UǺfg۾϶ޅ#@C$,"KBǿgsO 8<}4q˝`X!5i) c|D?;3fm`jɎŃ<񝫑ql]* ʗED'|Ba`$!wlG?DV֏ 0WLȁCKd)+ňx \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.golden b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.golden new file mode 100644 index 0000000..7ef6745 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.in b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.in new file mode 100644 index 0000000..fb5b1be --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.in @@ -0,0 +1,4 @@ +aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa +vH +% ɷ}>lsmIGH1Y4[ 0ˆ[|]o# +-#ulpfٱnYԀYwC8ɯ02 F=gnrN!O{k*w(b kQC9/lu>5C.u diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.sync.expect new file mode 100644 index 0000000..2d65279 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.sync.expect-noinput new file mode 100644 index 0000000..2d65279 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.wb.expect new file mode 100644 index 0000000..881e59c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.wb.expect-noinput new file mode 100644 index 0000000..881e59c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-limit.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-max.golden b/github.com/klauspost/compress/flate/testdata/huffman-rand-max.golden new file mode 100644 index 0000000..47d53c8 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-max.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-rand-max.in b/github.com/klauspost/compress/flate/testdata/huffman-rand-max.in new file mode 100644 index 0000000..8418633 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-rand-max.in differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-shifts.dyn.expect new file mode 100644 index 0000000..2f4fd17 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-shifts.dyn.expect-noinput new file mode 100644 index 0000000..2f4fd17 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.dyn.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.golden b/github.com/klauspost/compress/flate/testdata/huffman-shifts.golden new file mode 100644 index 0000000..89c8add Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.in b/github.com/klauspost/compress/flate/testdata/huffman-shifts.in new file mode 100644 index 0000000..b9f23ee --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-shifts.in @@ -0,0 +1,2 @@ +101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010 +232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323232323 \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-shifts.sync.expect new file mode 100644 index 0000000..7812c1c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-shifts.sync.expect-noinput new file mode 100644 index 0000000..7812c1c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-shifts.wb.expect new file mode 100644 index 0000000..7812c1c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-shifts.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-shifts.wb.expect-noinput new file mode 100644 index 0000000..7812c1c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-shifts.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.dyn.expect new file mode 100644 index 0000000..3a4dcc4 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.dyn.expect-noinput new file mode 100644 index 0000000..29788aa --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.dyn.expect-noinput @@ -0,0 +1,2 @@ +`@5R|@ו1Cᚄ4ϒ|ʂ.zgENL E#2¬EQ<D8.IDHÂD@. E^ @"Ҡ `M +KS4*n%P1nAA`OS^.aJUxx2s4%yWX+&F$I&)I gd<l9 7TCYmE+T"de!eˇ1闍Ș+< \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.golden b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.golden new file mode 100644 index 0000000..80531ad Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.in b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.in new file mode 100644 index 0000000..9d1c422 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.in @@ -0,0 +1,14 @@ +//Copyright2009ThGoAuthor.Allrightrrvd. +//UofthiourccodigovrndbyBSD-tyl +//licnthtcnbfoundinthLICENSEfil. + +pckgmin + +import"o" + +funcmin(){ + vrb=mk([]byt,65535) + f,_:=o.Crt("huffmn-null-mx.in") + f.Writ(b) +} +ABCDEFGHIJKLMNOPQRSTUVXxyz!"#¤%&/?" \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.sync.expect new file mode 100644 index 0000000..71ce3ae Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.sync.expect-noinput new file mode 100644 index 0000000..71ce3ae Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.wb.expect new file mode 100644 index 0000000..71ce3ae Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text-shift.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.wb.expect-noinput new file mode 100644 index 0000000..71ce3ae Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text-shift.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-text.dyn.expect new file mode 100644 index 0000000..1fb84b3 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-text.dyn.expect-noinput new file mode 100644 index 0000000..6ef6dd4 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text.dyn.expect-noinput @@ -0,0 +1,3 @@ +`J|ஏbF=M/MX+Kˊ;޹`.&;$ +A A :F8T h ͍˘P "PI&@ lG p`7TdxDGA^k, OAU!AVJQV2,ށj(,;]X` +*xqF_2>n^AUm Œ2>T gO U+d5ʕd6_i2 \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.golden b/github.com/klauspost/compress/flate/testdata/huffman-text.golden new file mode 100644 index 0000000..b440e84 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-text.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.in b/github.com/klauspost/compress/flate/testdata/huffman-text.in new file mode 100644 index 0000000..e76d564 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text.in @@ -0,0 +1,13 @@ +// 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 main + +import "os" + +func main() { + var b = make([]byte, 65535) + f, _ := os.Create("huffman-null-max.in") + f.Write(b) +} diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-text.sync.expect new file mode 100644 index 0000000..d448727 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text.sync.expect @@ -0,0 +1 @@ +_K0`K0Aasě)^HIɟb߻_>4 a=-^ 1`_ 1 ő:Y-F66!A`aC;ANyr4ߜU!GKС#r:B[G3.L׶bFRuM]^⇳(#Z ivBBH2S]u/ֽWTGnr \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-text.sync.expect-noinput new file mode 100644 index 0000000..d448727 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text.sync.expect-noinput @@ -0,0 +1 @@ +_K0`K0Aasě)^HIɟb߻_>4 a=-^ 1`_ 1 ő:Y-F66!A`aC;ANyr4ߜU!GKС#r:B[G3.L׶bFRuM]^⇳(#Z ivBBH2S]u/ֽWTGnr \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-text.wb.expect new file mode 100644 index 0000000..d448727 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text.wb.expect @@ -0,0 +1 @@ +_K0`K0Aasě)^HIɟb߻_>4 a=-^ 1`_ 1 ő:Y-F66!A`aC;ANyr4ߜU!GKС#r:B[G3.L׶bFRuM]^⇳(#Z ivBBH2S]u/ֽWTGnr \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-text.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-text.wb.expect-noinput new file mode 100644 index 0000000..d448727 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-text.wb.expect-noinput @@ -0,0 +1 @@ +_K0`K0Aasě)^HIɟb߻_>4 a=-^ 1`_ 1 ő:Y-F66!A`aC;ANyr4ߜU!GKС#r:B[G3.L׶bFRuM]^⇳(#Z ivBBH2S]u/ֽWTGnr \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.dyn.expect b/github.com/klauspost/compress/flate/testdata/huffman-zero.dyn.expect new file mode 100644 index 0000000..230433c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-zero.dyn.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-zero.dyn.expect-noinput new file mode 100644 index 0000000..cefc1d3 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-zero.dyn.expect-noinput @@ -0,0 +1 @@ +@hm۶m۶m۶m۶m۶6rk \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.golden b/github.com/klauspost/compress/flate/testdata/huffman-zero.golden new file mode 100644 index 0000000..f0dacf2 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-zero.golden differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.in b/github.com/klauspost/compress/flate/testdata/huffman-zero.in new file mode 100644 index 0000000..349be0e --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/huffman-zero.in @@ -0,0 +1 @@ +00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000 \ No newline at end of file diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.sync.expect b/github.com/klauspost/compress/flate/testdata/huffman-zero.sync.expect new file mode 100644 index 0000000..830348a Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-zero.sync.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-zero.sync.expect-noinput new file mode 100644 index 0000000..830348a Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-zero.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.wb.expect b/github.com/klauspost/compress/flate/testdata/huffman-zero.wb.expect new file mode 100644 index 0000000..dbe401c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-zero.wb.expect differ diff --git a/github.com/klauspost/compress/flate/testdata/huffman-zero.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/huffman-zero.wb.expect-noinput new file mode 100644 index 0000000..dbe401c Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/huffman-zero.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/null-long-match.dyn.expect-noinput b/github.com/klauspost/compress/flate/testdata/null-long-match.dyn.expect-noinput new file mode 100644 index 0000000..14167a3 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/null-long-match.dyn.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/null-long-match.sync.expect-noinput b/github.com/klauspost/compress/flate/testdata/null-long-match.sync.expect-noinput new file mode 100644 index 0000000..8b92d9f Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/null-long-match.sync.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/null-long-match.wb.expect-noinput b/github.com/klauspost/compress/flate/testdata/null-long-match.wb.expect-noinput new file mode 100644 index 0000000..8b92d9f Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/null-long-match.wb.expect-noinput differ diff --git a/github.com/klauspost/compress/flate/testdata/regression.zip b/github.com/klauspost/compress/flate/testdata/regression.zip new file mode 100644 index 0000000..a7be577 Binary files /dev/null and b/github.com/klauspost/compress/flate/testdata/regression.zip differ diff --git a/github.com/klauspost/compress/flate/testdata/tokens.bin b/github.com/klauspost/compress/flate/testdata/tokens.bin new file mode 100644 index 0000000..b93c696 --- /dev/null +++ b/github.com/klauspost/compress/flate/testdata/tokens.bin @@ -0,0 +1,63 @@ + + + name>Wikip끀en./Main_PageMediaWiki 1.6alphaSpecial0" /ɀ1">Talkŀ2">User3 t؀4">܂݀5 6">Image7ڀ89 10">Template 1Ӄ 2">Helpހ3ڀ4">Category5 00">Port101Às҈AaA12005-12-27T18:46:47Z　؀ 614213쁀Ӏ#REDIRECT [[AAA]]adding cur_id=5: {{R from CamelCase}}ҀԂa]]ԀmericanSamoaɂ69ԃˁ4:1to6 ۂ݂ ނppliedEthics858989432-02-25T15:43:11ip>Con script + <Afghaany132002-08-27T03:07:44ZMagnusskewhoops׀<Þ xml:space="rve">#REDIRECT [[҂GeoȀ쁀92-25T15:43:11ip>Con꽁criptmaxnumlit + offHist [32]uint16 // offset codes + litHist [256]uint16 // codes 0->255 + n uint16 // Must be able to contain maxStoreBlockSize + tokens [maxStoreBlockSize + 1]token +} + +func (t *tokens) Reset() { + if t.n == 0 { + return + } + t.n = 0 + t.nLits = 0 + for i := range t.litHist[:] { + t.litHist[i] = 0 + } + for i := range t.extraHist[:] { + t.extraHist[i] = 0 + } + for i := range t.offHist[:] { + t.offHist[i] = 0 + } +} + +func (t *tokens) Fill() { + if t.n == 0 { + return + } + for i, v := range t.litHist[:] { + if v == 0 { + t.litHist[i] = 1 + t.nLits++ + } + } + for i, v := range t.extraHist[:literalCount-256] { + if v == 0 { + t.nLits++ + t.extraHist[i] = 1 + } + } + for i, v := range t.offHist[:offsetCodeCount] { + if v == 0 { + t.offHist[i] = 1 + } + } +} + +func indexTokens(in []token) tokens { + var t tokens + t.indexTokens(in) + return t +} + +func (t *tokens) indexTokens(in []token) { + t.Reset() + for _, tok := range in { + if tok < matchType { + t.AddLiteral(tok.literal()) + continue + } + t.AddMatch(uint32(tok.length()), tok.offset()) + } +} + +// emitLiteral writes a literal chunk and returns the number of bytes written. +func emitLiteral(dst *tokens, lit []byte) { + ol := int(dst.n) + for i, v := range lit { + dst.tokens[(i+ol)&maxStoreBlockSize] = token(v) + dst.litHist[v]++ + } + dst.n += uint16(len(lit)) + dst.nLits += len(lit) +} + +func (t *tokens) AddLiteral(lit byte) { + t.tokens[t.n] = token(lit) + t.litHist[lit]++ + t.n++ + 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 := float32(0) + bits := int(0) + nMatches := 0 + if t.nLits > 0 { + invTotal := 1.0 / float32(t.nLits) + for _, v := range t.litHist[:] { + if v > 0 { + n := float32(v) + shannon += -mFastLog2(n*invTotal) * n + } + } + // Just add 15 for EOB + shannon += 15 + for i, v := range t.extraHist[1 : literalCount-256] { + if v > 0 { + n := float32(v) + shannon += -mFastLog2(n*invTotal) * n + bits += int(lengthExtraBits[i&31]) * int(v) + nMatches += int(v) + } + } + } + if nMatches > 0 { + invTotal := 1.0 / float32(nMatches) + for i, v := range t.offHist[:offsetCodeCount] { + if v > 0 { + 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 debugDeflate { + if xlength >= maxMatchLength+baseMatchLength { + panic(fmt.Errorf("invalid length: %v", xlength)) + } + if xoffset >= maxMatchOffset+baseMatchOffset { + panic(fmt.Errorf("invalid offset: %v", xoffset)) + } + } + t.nLits++ + lengthCode := lengthCodes1[uint8(xlength)] & 31 + t.tokens[t.n] = token(matchType | xlength<= maxMatchOffset+baseMatchOffset { + panic(fmt.Errorf("invalid offset: %v", xoffset)) + } + } + oc := offsetCode(xoffset) & 31 + for xlength > 0 { + xl := xlength + if xl > 258 { + // We need to have at least baseMatchLength left over for next loop. + xl = 258 - baseMatchLength + } + xlength -= xl + xl -= 3 + t.nLits++ + lengthCode := lengthCodes1[uint8(xl)] & 31 + t.tokens[t.n] = token(matchType | uint32(xl)<> lengthShift) } + +// The code is never more than 8 bits, but is returned as uint32 for convenience. +func lengthCode(len uint8) uint32 { return uint32(lengthCodes[len]) } + +// Returns the offset code corresponding to a specific offset +func offsetCode(off uint32) uint32 { + if false { + if off < uint32(len(offsetCodes)) { + return offsetCodes[off&255] + } else if off>>7 < uint32(len(offsetCodes)) { + return offsetCodes[(off>>7)&255] + 14 + } else { + return offsetCodes[(off>>14)&255] + 28 + } + } + if off < uint32(len(offsetCodes)) { + return offsetCodes[uint8(off)] + } + return offsetCodes14[uint8(off>>7)] +} diff --git a/github.com/klauspost/compress/flate/token_test.go b/github.com/klauspost/compress/flate/token_test.go new file mode 100644 index 0000000..a8066c3 --- /dev/null +++ b/github.com/klauspost/compress/flate/token_test.go @@ -0,0 +1,54 @@ +package flate + +import ( + "bytes" + "io/ioutil" + "testing" +) + +type testFatal interface { + Fatal(args ...interface{}) +} + +// loadTestTokens will load test tokens. +// First block from enwik9, varint encoded. +func loadTestTokens(t testFatal) *tokens { + b, err := ioutil.ReadFile("testdata/tokens.bin") + if err != nil { + t.Fatal(err) + } + var tokens tokens + err = tokens.FromVarInt(b) + if err != nil { + t.Fatal(err) + } + return &tokens +} + +func Test_tokens_EstimatedBits(t *testing.T) { + tok := loadTestTokens(t) + // The estimated size, update if method changes. + const expect = 221057 + n := tok.EstimatedBits() + var buf bytes.Buffer + wr := newHuffmanBitWriter(&buf) + wr.writeBlockDynamic(tok, true, nil, true) + if wr.err != nil { + t.Fatal(wr.err) + } + wr.flush() + t.Log("got:", n, "actual:", buf.Len()*8, "(header not part of estimate)") + if n != expect { + t.Error("want:", expect, "bits, got:", n) + } +} + +func Benchmark_tokens_EstimatedBits(b *testing.B) { + tok := loadTestTokens(b) + b.ResetTimer() + // One "byte", one token iteration. + b.SetBytes(1) + for i := 0; i < b.N; i++ { + _ = tok.EstimatedBits() + } +} diff --git a/github.com/klauspost/compress/flate/writer_test.go b/github.com/klauspost/compress/flate/writer_test.go new file mode 100644 index 0000000..46a0644 --- /dev/null +++ b/github.com/klauspost/compress/flate/writer_test.go @@ -0,0 +1,451 @@ +// Copyright 2012 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 + +import ( + "archive/zip" + "bytes" + "fmt" + "io" + "io/ioutil" + "math" + "math/rand" + "runtime" + "strconv" + "strings" + "testing" +) + +func TestWriterRegression(t *testing.T) { + data, err := ioutil.ReadFile("testdata/regression.zip") + if err != nil { + t.Fatal(err) + } + for level := HuffmanOnly; level <= BestCompression; level++ { + t.Run(fmt.Sprint("level_", level), func(t *testing.T) { + zr, err := zip.NewReader(bytes.NewReader(data), int64(len(data))) + if err != nil { + t.Fatal(err) + } + + for _, tt := range zr.File { + if !strings.HasSuffix(t.Name(), "") { + continue + } + + t.Run(tt.Name, func(t *testing.T) { + r, err := tt.Open() + if err != nil { + t.Error(err) + return + } + in, err := ioutil.ReadAll(r) + if err != nil { + t.Error(err) + } + msg := "level " + strconv.Itoa(level) + ":" + buf := new(bytes.Buffer) + fw, err := NewWriter(buf, level) + if err != nil { + t.Fatal(msg + err.Error()) + } + n, err := fw.Write(in) + if n != len(in) { + t.Fatal(msg + "short write") + } + if err != nil { + t.Fatal(msg + err.Error()) + } + err = fw.Close() + if err != nil { + t.Fatal(msg + err.Error()) + } + fr1 := NewReader(buf) + data2, err := ioutil.ReadAll(fr1) + if err != nil { + t.Fatal(msg + err.Error()) + } + if bytes.Compare(in, data2) != 0 { + t.Fatal(msg + "not equal") + } + // Do it again... + msg = "level " + strconv.Itoa(level) + " (reset):" + buf.Reset() + fw.Reset(buf) + n, err = fw.Write(in) + if n != len(in) { + t.Fatal(msg + "short write") + } + if err != nil { + t.Fatal(msg + err.Error()) + } + err = fw.Close() + if err != nil { + t.Fatal(msg + err.Error()) + } + fr1 = NewReader(buf) + data2, err = ioutil.ReadAll(fr1) + if err != nil { + t.Fatal(msg + err.Error()) + } + if bytes.Compare(in, data2) != 0 { + t.Fatal(msg + "not equal") + } + }) + } + }) + } +} + +func benchmarkEncoder(b *testing.B, testfile, level, n int) { + b.SetBytes(int64(n)) + buf0, err := ioutil.ReadFile(testfiles[testfile]) + if err != nil { + b.Fatal(err) + } + if len(buf0) == 0 { + b.Fatalf("test file %q has no data", testfiles[testfile]) + } + buf1 := make([]byte, n) + for i := 0; i < n; i += len(buf0) { + if len(buf0) > n-i { + buf0 = buf0[:n-i] + } + copy(buf1[i:], buf0) + } + buf0 = nil + runtime.GC() + w, err := NewWriter(ioutil.Discard, level) + b.ResetTimer() + b.ReportAllocs() + for i := 0; i < b.N; i++ { + w.Reset(ioutil.Discard) + _, err = w.Write(buf1) + if err != nil { + b.Fatal(err) + } + err = w.Close() + if err != nil { + b.Fatal(err) + } + } +} + +func BenchmarkEncodeDigitsConstant1e4(b *testing.B) { benchmarkEncoder(b, digits, constant, 1e4) } +func BenchmarkEncodeDigitsConstant1e5(b *testing.B) { benchmarkEncoder(b, digits, constant, 1e5) } +func BenchmarkEncodeDigitsConstant1e6(b *testing.B) { benchmarkEncoder(b, digits, constant, 1e6) } +func BenchmarkEncodeDigitsSpeed1e4(b *testing.B) { benchmarkEncoder(b, digits, speed, 1e4) } +func BenchmarkEncodeDigitsSpeed1e5(b *testing.B) { benchmarkEncoder(b, digits, speed, 1e5) } +func BenchmarkEncodeDigitsSpeed1e6(b *testing.B) { benchmarkEncoder(b, digits, speed, 1e6) } +func BenchmarkEncodeDigitsDefault1e4(b *testing.B) { benchmarkEncoder(b, digits, default_, 1e4) } +func BenchmarkEncodeDigitsDefault1e5(b *testing.B) { benchmarkEncoder(b, digits, default_, 1e5) } +func BenchmarkEncodeDigitsDefault1e6(b *testing.B) { benchmarkEncoder(b, digits, default_, 1e6) } +func BenchmarkEncodeDigitsCompress1e4(b *testing.B) { benchmarkEncoder(b, digits, compress, 1e4) } +func BenchmarkEncodeDigitsCompress1e5(b *testing.B) { benchmarkEncoder(b, digits, compress, 1e5) } +func BenchmarkEncodeDigitsCompress1e6(b *testing.B) { benchmarkEncoder(b, digits, compress, 1e6) } +func BenchmarkEncodeDigitsSL1e4(b *testing.B) { benchmarkStatelessEncoder(b, digits, 1e4) } +func BenchmarkEncodeDigitsSL1e5(b *testing.B) { benchmarkStatelessEncoder(b, digits, 1e5) } +func BenchmarkEncodeDigitsSL1e6(b *testing.B) { benchmarkStatelessEncoder(b, digits, 1e6) } +func BenchmarkEncodeTwainConstant1e4(b *testing.B) { benchmarkEncoder(b, twain, constant, 1e4) } +func BenchmarkEncodeTwainConstant1e5(b *testing.B) { benchmarkEncoder(b, twain, constant, 1e5) } +func BenchmarkEncodeTwainConstant1e6(b *testing.B) { benchmarkEncoder(b, twain, constant, 1e6) } +func BenchmarkEncodeTwainSpeed1e4(b *testing.B) { benchmarkEncoder(b, twain, speed, 1e4) } +func BenchmarkEncodeTwainSpeed1e5(b *testing.B) { benchmarkEncoder(b, twain, speed, 1e5) } +func BenchmarkEncodeTwainSpeed1e6(b *testing.B) { benchmarkEncoder(b, twain, speed, 1e6) } +func BenchmarkEncodeTwainDefault1e4(b *testing.B) { benchmarkEncoder(b, twain, default_, 1e4) } +func BenchmarkEncodeTwainDefault1e5(b *testing.B) { benchmarkEncoder(b, twain, default_, 1e5) } +func BenchmarkEncodeTwainDefault1e6(b *testing.B) { benchmarkEncoder(b, twain, default_, 1e6) } +func BenchmarkEncodeTwainCompress1e4(b *testing.B) { benchmarkEncoder(b, twain, compress, 1e4) } +func BenchmarkEncodeTwainCompress1e5(b *testing.B) { benchmarkEncoder(b, twain, compress, 1e5) } +func BenchmarkEncodeTwainCompress1e6(b *testing.B) { benchmarkEncoder(b, twain, compress, 1e6) } +func BenchmarkEncodeTwainSL1e4(b *testing.B) { benchmarkStatelessEncoder(b, twain, 1e4) } +func BenchmarkEncodeTwainSL1e5(b *testing.B) { benchmarkStatelessEncoder(b, twain, 1e5) } +func BenchmarkEncodeTwainSL1e6(b *testing.B) { benchmarkStatelessEncoder(b, twain, 1e6) } + +func benchmarkStatelessEncoder(b *testing.B, testfile, n int) { + b.SetBytes(int64(n)) + buf0, err := ioutil.ReadFile(testfiles[testfile]) + if err != nil { + b.Fatal(err) + } + if len(buf0) == 0 { + b.Fatalf("test file %q has no data", testfiles[testfile]) + } + buf1 := make([]byte, n) + for i := 0; i < n; i += len(buf0) { + if len(buf0) > n-i { + buf0 = buf0[:n-i] + } + copy(buf1[i:], buf0) + } + buf0 = nil + runtime.GC() + b.ResetTimer() + b.ReportAllocs() + for i := 0; i < b.N; i++ { + w := NewStatelessWriter(ioutil.Discard) + _, err = w.Write(buf1) + if err != nil { + b.Fatal(err) + } + err = w.Close() + if err != nil { + b.Fatal(err) + } + } +} + +// A writer that fails after N writes. +type errorWriter struct { + N int +} + +func (e *errorWriter) Write(b []byte) (int, error) { + if e.N <= 0 { + return 0, io.ErrClosedPipe + } + e.N-- + return len(b), nil +} + +// Test if errors from the underlying writer is passed upwards. +func TestWriteError(t *testing.T) { + buf := new(bytes.Buffer) + n := 65536 + if !testing.Short() { + n *= 4 + } + for i := 0; i < n; i++ { + fmt.Fprintf(buf, "asdasfasf%d%dfghfgujyut%dyutyu\n", i, i, i) + } + in := buf.Bytes() + // We create our own buffer to control number of writes. + copyBuf := make([]byte, 128) + for l := 0; l < 10; l++ { + for fail := 1; fail <= 256; fail *= 2 { + // Fail after 'fail' writes + ew := &errorWriter{N: fail} + w, err := NewWriter(ew, l) + if err != nil { + t.Fatalf("NewWriter: level %d: %v", l, err) + } + n, err := copyBuffer(w, bytes.NewBuffer(in), copyBuf) + if err == nil { + t.Fatalf("Level %d: Expected an error, writer was %#v", l, ew) + } + n2, err := w.Write([]byte{1, 2, 2, 3, 4, 5}) + if n2 != 0 { + t.Fatal("Level", l, "Expected 0 length write, got", n) + } + if err == nil { + t.Fatal("Level", l, "Expected an error") + } + err = w.Flush() + if err == nil { + t.Fatal("Level", l, "Expected an error on flush") + } + err = w.Close() + if err == nil { + t.Fatal("Level", l, "Expected an error on close") + } + + w.Reset(ioutil.Discard) + n2, err = w.Write([]byte{1, 2, 3, 4, 5, 6}) + if err != nil { + t.Fatal("Level", l, "Got unexpected error after reset:", err) + } + if n2 == 0 { + t.Fatal("Level", l, "Got 0 length write, expected > 0") + } + if testing.Short() { + return + } + } + } +} + +// Test if errors from the underlying writer is passed upwards. +func TestWriter_Reset(t *testing.T) { + buf := new(bytes.Buffer) + n := 65536 + if !testing.Short() { + n *= 4 + } + for i := 0; i < n; i++ { + fmt.Fprintf(buf, "asdasfasf%d%dfghfgujyut%dyutyu\n", i, i, i) + } + in := buf.Bytes() + for l := 0; l < 10; l++ { + l := l + if testing.Short() && l > 1 { + continue + } + t.Run(fmt.Sprintf("level-%d", l), func(t *testing.T) { + t.Parallel() + offset := 1 + if testing.Short() { + offset = 256 + } + for ; offset <= 256; offset *= 2 { + // Fail after 'fail' writes + w, err := NewWriter(ioutil.Discard, l) + if err != nil { + t.Fatalf("NewWriter: level %d: %v", l, err) + } + if w.d.fast == nil { + t.Skip("Not Fast...") + return + } + for i := 0; i < (bufferReset-len(in)-offset-maxMatchOffset)/maxMatchOffset; i++ { + // skip ahead to where we are close to wrap around... + w.d.fast.Reset() + } + w.d.fast.Reset() + _, err = w.Write(in) + if err != nil { + t.Fatal(err) + } + for i := 0; i < 50; i++ { + // skip ahead again... This should wrap around... + w.d.fast.Reset() + } + w.d.fast.Reset() + + _, err = w.Write(in) + if err != nil { + t.Fatal(err) + } + for i := 0; i < (math.MaxUint32-bufferReset)/maxMatchOffset; i++ { + // skip ahead to where we are close to wrap around... + w.d.fast.Reset() + } + + _, err = w.Write(in) + if err != nil { + t.Fatal(err) + } + err = w.Close() + if err != nil { + t.Fatal(err) + } + } + }) + } +} + +func TestDeterministicL1(t *testing.T) { testDeterministic(1, t) } +func TestDeterministicL2(t *testing.T) { testDeterministic(2, t) } +func TestDeterministicL3(t *testing.T) { testDeterministic(3, t) } +func TestDeterministicL4(t *testing.T) { testDeterministic(4, t) } +func TestDeterministicL5(t *testing.T) { testDeterministic(5, t) } +func TestDeterministicL6(t *testing.T) { testDeterministic(6, t) } +func TestDeterministicL7(t *testing.T) { testDeterministic(7, t) } +func TestDeterministicL8(t *testing.T) { testDeterministic(8, t) } +func TestDeterministicL9(t *testing.T) { testDeterministic(9, t) } +func TestDeterministicL0(t *testing.T) { testDeterministic(0, t) } +func TestDeterministicLM2(t *testing.T) { testDeterministic(-2, t) } + +func testDeterministic(i int, t *testing.T) { + // Test so much we cross a good number of block boundaries. + var length = maxStoreBlockSize*30 + 500 + if testing.Short() { + length /= 10 + } + + // Create a random, but compressible stream. + rng := rand.New(rand.NewSource(1)) + t1 := make([]byte, length) + for i := range t1 { + t1[i] = byte(rng.Int63() & 7) + } + + // Do our first encode. + var b1 bytes.Buffer + br := bytes.NewBuffer(t1) + w, err := NewWriter(&b1, i) + if err != nil { + t.Fatal(err) + } + // Use a very small prime sized buffer. + cbuf := make([]byte, 787) + _, err = copyBuffer(w, br, cbuf) + if err != nil { + t.Fatal(err) + } + w.Close() + + // We choose a different buffer size, + // bigger than a maximum block, and also a prime. + var b2 bytes.Buffer + cbuf = make([]byte, 81761) + br2 := bytes.NewBuffer(t1) + w2, err := NewWriter(&b2, i) + if err != nil { + t.Fatal(err) + } + _, err = copyBuffer(w2, br2, cbuf) + if err != nil { + t.Fatal(err) + } + w2.Close() + + b1b := b1.Bytes() + b2b := b2.Bytes() + + if !bytes.Equal(b1b, b2b) { + t.Errorf("level %d did not produce deterministic result, result mismatch, len(a) = %d, len(b) = %d", i, len(b1b), len(b2b)) + } + + // Test using io.WriterTo interface. + var b3 bytes.Buffer + br = bytes.NewBuffer(t1) + w, err = NewWriter(&b3, i) + if err != nil { + t.Fatal(err) + } + _, err = br.WriteTo(w) + if err != nil { + t.Fatal(err) + } + w.Close() + + b3b := b3.Bytes() + if !bytes.Equal(b1b, b3b) { + t.Errorf("level %d (io.WriterTo) did not produce deterministic result, result mismatch, len(a) = %d, len(b) = %d", i, len(b1b), len(b3b)) + } +} + +// copyBuffer is a copy of io.CopyBuffer, since we want to support older go versions. +// This is modified to never use io.WriterTo or io.ReaderFrom interfaces. +func copyBuffer(dst io.Writer, src io.Reader, buf []byte) (written int64, err error) { + if buf == nil { + buf = make([]byte, 32*1024) + } + for { + nr, er := src.Read(buf) + if nr > 0 { + nw, ew := dst.Write(buf[0:nr]) + if nw > 0 { + written += int64(nw) + } + if ew != nil { + err = ew + break + } + if nr != nw { + err = io.ErrShortWrite + break + } + } + if er == io.EOF { + break + } + if er != nil { + err = er + break + } + } + return written, err +} diff --git a/github.com/klauspost/compress/fse/README.md b/github.com/klauspost/compress/fse/README.md new file mode 100644 index 0000000..27d8ed5 --- /dev/null +++ b/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/github.com/klauspost/compress/fse/bitreader.go b/github.com/klauspost/compress/fse/bitreader.go new file mode 100644 index 0000000..f65eb39 --- /dev/null +++ b/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/github.com/klauspost/compress/fse/bitwriter.go b/github.com/klauspost/compress/fse/bitwriter.go new file mode 100644 index 0000000..43e4636 --- /dev/null +++ b/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/github.com/klauspost/compress/fse/bytereader.go b/github.com/klauspost/compress/fse/bytereader.go new file mode 100644 index 0000000..abade2d --- /dev/null +++ b/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/github.com/klauspost/compress/fse/compress.go b/github.com/klauspost/compress/fse/compress.go new file mode 100644 index 0000000..b69237c --- /dev/null +++ b/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/github.com/klauspost/compress/fse/fse.go b/github.com/klauspost/compress/fse/fse.go new file mode 100644 index 0000000..535cbad --- /dev/null +++ b/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/github.com/klauspost/compress/fse/fse_test.go b/github.com/klauspost/compress/fse/fse_test.go new file mode 100644 index 0000000..9e617f3 --- /dev/null +++ b/github.com/klauspost/compress/fse/fse_test.go @@ -0,0 +1,288 @@ +// 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 ( + "bytes" + "fmt" + "io/ioutil" + "reflect" + "strings" + "testing" +) + +type inputFn func() ([]byte, error) + +var testfiles = []struct { + name string + fn inputFn + err error +}{ + // gettysburg.txt is a small plain text. + {name: "gettysburg", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/gettysburg.txt") }}, + // Digits is the digits of the irrational number e. Its decimal representation + // does not repeat, but there are only 10 possible digits, so it should be + // reasonably compressible. + {name: "digits", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/e.txt") }}, + // Twain is Project Gutenberg's edition of Mark Twain's classic English novel. + {name: "twain", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/Mark.Twain-Tom.Sawyer.txt") }}, + // Random bytes + {name: "random", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/sharnd.out") }, err: ErrIncompressible}, + // Low entropy + {name: "low-ent", fn: func() ([]byte, error) { return []byte(strings.Repeat("1221", 10000)), nil }}, + // Super Low entropy + {name: "superlow-ent", fn: func() ([]byte, error) { return []byte(strings.Repeat("1", 10000) + strings.Repeat("2", 500)), nil }}, + // Zero bytes + {name: "zeroes", fn: func() ([]byte, error) { return make([]byte, 10000), nil }, err: ErrUseRLE}, + {name: "crash1", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash1.bin") }, err: ErrIncompressible}, + {name: "crash2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash2.bin") }, err: ErrIncompressible}, + {name: "crash3", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash3.bin") }, err: ErrIncompressible}, + {name: "endzerobits", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/endzerobits.bin") }, err: nil}, + {name: "endnonzero", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/endnonzero.bin") }, err: ErrIncompressible}, + {name: "case1", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/case1.bin") }, err: ErrIncompressible}, + {name: "case2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/case2.bin") }, err: ErrIncompressible}, + {name: "case3", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/case3.bin") }, err: ErrIncompressible}, + {name: "pngdata.001", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/pngdata.bin") }, err: nil}, + {name: "normcount2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/normcount2.bin") }, err: nil}, +} + +var decTestfiles = []struct { + name string + fn inputFn + err string +}{ + // gettysburg.txt is a small plain text. + {name: "hang1", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/dec-hang1.bin") }, err: "corruption detected (bitCount 252 > 32)"}, + {name: "hang2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/dec-hang2.bin") }, err: "newState (0) == oldState (0) and no bits"}, + {name: "hang3", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/dec-hang3.bin") }, err: "maxSymbolValue too small"}, + {name: "symlen1", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/dec-symlen1.bin") }, err: "symbolLen (257) too big"}, + {name: "crash4", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash4.bin") }, err: "symbolLen (1) too small"}, + {name: "crash5", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash5.bin") }, err: "symbolLen (1) too small"}, + {name: "crash6", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/dec-crash6.bin") }, err: "newState (32768) outside table size (32768)"}, + {name: "something", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/fse-artifact3.bin") }, err: "output size (1048576) > DecompressLimit (1048576)"}, +} + +func TestCompress(t *testing.T) { + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + var s Scratch + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + b, err := Compress(buf0, &s) + if err != test.err { + t.Errorf("want error %v (%T), got %v (%T)", test.err, test.err, err, err) + } + if b == nil { + t.Log(test.name + ": not compressible") + return + } + t.Logf("%s: %d -> %d bytes (%.2f:1)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b))) + }) + } +} + +func ExampleCompress() { + // Read data + data, err := ioutil.ReadFile("../testdata/e.txt") + if err != nil { + panic(err) + } + + // Create re-usable scratch buffer. + var s Scratch + b, err := Compress(data, &s) + if err != nil { + panic(err) + } + fmt.Printf("Compress: %d -> %d bytes (%.2f:1)\n", len(data), len(b), float64(len(data))/float64(len(b))) + // OUTPUT: Compress: 100003 -> 41564 bytes (2.41:1) +} + +func TestDecompress(t *testing.T) { + for _, test := range decTestfiles { + t.Run(test.name, func(t *testing.T) { + var s Scratch + s.DecompressLimit = 1 << 20 + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + b, err := Decompress(buf0, &s) + if fmt.Sprint(err) != test.err { + t.Errorf("want error %q, got %q (%T)", test.err, err, err) + return + } + if err != nil { + return + } + if len(b) == 0 { + t.Error(test.name + ": no output") + return + } + t.Logf("%s: %d -> %d bytes (1:%.2f)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b))) + }) + } +} + +func ExampleDecompress() { + // Read data + data, err := ioutil.ReadFile("../testdata/e.txt") + if err != nil { + panic(err) + } + + // Create re-usable scratch buffer. + var s Scratch + b, err := Compress(data, &s) + if err != nil { + panic(err) + } + + // Since we use the output of compression, it cannot be used as output for decompression. + s.Out = make([]byte, 0, len(data)) + d, err := Decompress(b, &s) + if err != nil { + panic(err) + } + fmt.Printf("Input matches: %t\n", bytes.Equal(d, data)) + // OUTPUT: Input matches: true +} + +func BenchmarkCompress(b *testing.B) { + for _, tt := range testfiles { + test := tt + b.Run(test.name, func(b *testing.B) { + var s Scratch + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + _, err = Compress(buf0, &s) + if err != test.err { + b.Fatal("unexpected error:", err) + } + if err != nil { + b.Skip("skipping benchmark: ", err) + return + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, _ = Compress(buf0, &s) + } + }) + } +} + +func TestReadNCount(t *testing.T) { + for i := range testfiles { + var s Scratch + test := testfiles[i] + t.Run(test.name, func(t *testing.T) { + name := test.name + ": " + buf0, err := testfiles[i].fn() + if err != nil { + t.Fatal(err) + } + b, err := Compress(buf0, &s) + if err != test.err { + t.Error(err) + return + } + if err != nil { + t.Skip(name + err.Error()) + return + } + t.Logf("%s: %d -> %d bytes (%.2f:1)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b))) + //t.Logf("%v", b) + var s2 Scratch + dc, err := Decompress(b, &s2) + if err != nil { + t.Fatal(err) + } + want := s.norm[:s.symbolLen] + got := s2.norm[:s2.symbolLen] + if !reflect.DeepEqual(want, got) { + if s.actualTableLog != s2.actualTableLog { + t.Errorf(name+"norm table, want tablelog: %d, got %d", s.actualTableLog, s2.actualTableLog) + } + if s.symbolLen != s2.symbolLen { + t.Errorf(name+"norm table, want size: %d, got %d", s.symbolLen, s2.symbolLen) + } + t.Errorf(name + "norm table, got delta: \n") + return + } + for i, dec := range s2.decTable { + dd := dec.symbol + ee := s.ct.tableSymbol[i] + if dd != ee { + t.Errorf("table symbol mismatch. idx %d, enc: %v, dec:%v", i, ee, dd) + break + } + } + if dc != nil { + if len(buf0) != len(dc) { + t.Errorf(name+"decompressed, want size: %d, got %d", len(buf0), len(dc)) + if len(buf0) > len(dc) { + buf0 = buf0[:len(dc)] + } else { + dc = dc[:len(buf0)] + } + if !bytes.Equal(buf0, dc) { + t.Errorf(name+"decompressed, got delta: (in) %v != (out) %v\n", buf0, dc) + } + return + } + if !bytes.Equal(buf0, dc) { + t.Errorf(name + "decompressed, got delta.") + } + if !t.Failed() { + t.Log("... roundtrip ok!") + } + } + }) + } +} + +func BenchmarkDecompress(b *testing.B) { + for _, tt := range testfiles { + test := tt + b.Run(test.name, func(b *testing.B) { + var s, s2 Scratch + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + out, err := Compress(buf0, &s) + if err != test.err { + b.Fatal(err) + } + if err != nil { + b.Skip(test.name + ": " + err.Error()) + return + } + got, err := Decompress(out, &s2) + if err != nil { + b.Fatal(err) + } + if !bytes.Equal(buf0, got) { + b.Fatal("output mismatch") + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, err = Decompress(out, &s2) + if err != nil { + b.Fatal(err) + } + } + }) + } +} diff --git a/github.com/klauspost/compress/go.mod b/github.com/klauspost/compress/go.mod new file mode 100644 index 0000000..e25f1c9 --- /dev/null +++ b/github.com/klauspost/compress/go.mod @@ -0,0 +1,3 @@ +module github.com/klauspost/compress + +go 1.13 diff --git a/github.com/klauspost/compress/gzip/example_test.go b/github.com/klauspost/compress/gzip/example_test.go new file mode 100644 index 0000000..ce29e9b --- /dev/null +++ b/github.com/klauspost/compress/gzip/example_test.go @@ -0,0 +1,128 @@ +// 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. + +package gzip_test + +import ( + "bytes" + "compress/gzip" + "fmt" + "io" + "log" + "os" + "time" +) + +func Example_writerReader() { + var buf bytes.Buffer + zw := gzip.NewWriter(&buf) + + // Setting the Header fields is optional. + zw.Name = "a-new-hope.txt" + zw.Comment = "an epic space opera by George Lucas" + zw.ModTime = time.Date(1977, time.May, 25, 0, 0, 0, 0, time.UTC) + + _, err := zw.Write([]byte("A long time ago in a galaxy far, far away...")) + if err != nil { + log.Fatal(err) + } + + if err := zw.Close(); err != nil { + log.Fatal(err) + } + + zr, err := gzip.NewReader(&buf) + if err != nil { + log.Fatal(err) + } + + fmt.Printf("Name: %s\nComment: %s\nModTime: %s\n\n", zr.Name, zr.Comment, zr.ModTime.UTC()) + + if _, err := io.Copy(os.Stdout, zr); err != nil { + log.Fatal(err) + } + + if err := zr.Close(); err != nil { + log.Fatal(err) + } + + // Output: + // Name: a-new-hope.txt + // Comment: an epic space opera by George Lucas + // ModTime: 1977-05-25 00:00:00 +0000 UTC + // + // A long time ago in a galaxy far, far away... +} + +func ExampleReader_Multistream() { + var buf bytes.Buffer + zw := gzip.NewWriter(&buf) + + var files = []struct { + name string + comment string + modTime time.Time + data string + }{ + {"file-1.txt", "file-header-1", time.Date(2006, time.February, 1, 3, 4, 5, 0, time.UTC), "Hello Gophers - 1"}, + {"file-2.txt", "file-header-2", time.Date(2007, time.March, 2, 4, 5, 6, 1, time.UTC), "Hello Gophers - 2"}, + } + + for _, file := range files { + zw.Name = file.name + zw.Comment = file.comment + zw.ModTime = file.modTime + + if _, err := zw.Write([]byte(file.data)); err != nil { + log.Fatal(err) + } + + if err := zw.Close(); err != nil { + log.Fatal(err) + } + + zw.Reset(&buf) + } + + zr, err := gzip.NewReader(&buf) + if err != nil { + log.Fatal(err) + } + + for { + zr.Multistream(false) + fmt.Printf("Name: %s\nComment: %s\nModTime: %s\n\n", zr.Name, zr.Comment, zr.ModTime.UTC()) + + if _, err := io.Copy(os.Stdout, zr); err != nil { + log.Fatal(err) + } + + fmt.Print("\n\n") + + err = zr.Reset(&buf) + if err == io.EOF { + break + } + if err != nil { + log.Fatal(err) + } + } + + if err := zr.Close(); err != nil { + log.Fatal(err) + } + + // Output: + // Name: file-1.txt + // Comment: file-header-1 + // ModTime: 2006-02-01 03:04:05 +0000 UTC + // + // Hello Gophers - 1 + // + // Name: file-2.txt + // Comment: file-header-2 + // ModTime: 2007-03-02 04:05:06 +0000 UTC + // + // Hello Gophers - 2 +} diff --git a/github.com/klauspost/compress/gzip/gunzip.go b/github.com/klauspost/compress/gzip/gunzip.go new file mode 100644 index 0000000..568b5d4 --- /dev/null +++ b/github.com/klauspost/compress/gzip/gunzip.go @@ -0,0 +1,344 @@ +// 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 gzip implements reading and writing of gzip format compressed files, +// as specified in RFC 1952. +package gzip + +import ( + "bufio" + "encoding/binary" + "errors" + "hash/crc32" + "io" + "time" + + "github.com/klauspost/compress/flate" +) + +const ( + gzipID1 = 0x1f + gzipID2 = 0x8b + gzipDeflate = 8 + flagText = 1 << 0 + flagHdrCrc = 1 << 1 + flagExtra = 1 << 2 + flagName = 1 << 3 + flagComment = 1 << 4 +) + +var ( + // ErrChecksum is returned when reading GZIP data that has an invalid checksum. + ErrChecksum = errors.New("gzip: invalid checksum") + // ErrHeader is returned when reading GZIP data that has an invalid header. + ErrHeader = errors.New("gzip: invalid header") +) + +var le = binary.LittleEndian + +// noEOF converts io.EOF to io.ErrUnexpectedEOF. +func noEOF(err error) error { + if err == io.EOF { + return io.ErrUnexpectedEOF + } + return err +} + +// The gzip file stores a header giving metadata about the compressed file. +// That header is exposed as the fields of the Writer and Reader structs. +// +// Strings must be UTF-8 encoded and may only contain Unicode code points +// U+0001 through U+00FF, due to limitations of the GZIP file format. +type Header struct { + Comment string // comment + Extra []byte // "extra data" + ModTime time.Time // modification time + Name string // file name + OS byte // operating system type +} + +// A Reader is an io.Reader that can be read to retrieve +// uncompressed data from a gzip-format compressed file. +// +// In general, a gzip file can be a concatenation of gzip files, +// each with its own header. Reads from the Reader +// return the concatenation of the uncompressed data of each. +// Only the first header is recorded in the Reader fields. +// +// Gzip files store a length and checksum of the uncompressed data. +// The Reader will return a ErrChecksum when Read +// reaches the end of the uncompressed data if it does not +// have the expected length or checksum. Clients should treat data +// returned by Read as tentative until they receive the io.EOF +// marking the end of the data. +type Reader struct { + Header // valid after NewReader or Reader.Reset + r flate.Reader + decompressor io.ReadCloser + digest uint32 // CRC-32, IEEE polynomial (section 8) + size uint32 // Uncompressed size (section 2.3.1) + buf [512]byte + err error + multistream bool +} + +// NewReader creates a new Reader reading the given reader. +// If r does not also implement io.ByteReader, +// the decompressor may read more data than necessary from r. +// +// It is the caller's responsibility to call Close on the Reader when done. +// +// The Reader.Header fields will be valid in the Reader returned. +func NewReader(r io.Reader) (*Reader, error) { + z := new(Reader) + if err := z.Reset(r); err != nil { + return nil, err + } + return z, nil +} + +// Reset discards the Reader z's state and makes it equivalent to the +// result of its original state from NewReader, but reading from r instead. +// This permits reusing a Reader rather than allocating a new one. +func (z *Reader) Reset(r io.Reader) error { + *z = Reader{ + decompressor: z.decompressor, + multistream: true, + } + if rr, ok := r.(flate.Reader); ok { + z.r = rr + } else { + z.r = bufio.NewReader(r) + } + z.Header, z.err = z.readHeader() + return z.err +} + +// Multistream controls whether the reader supports multistream files. +// +// If enabled (the default), the Reader expects the input to be a sequence +// of individually gzipped data streams, each with its own header and +// trailer, ending at EOF. The effect is that the concatenation of a sequence +// of gzipped files is treated as equivalent to the gzip of the concatenation +// of the sequence. This is standard behavior for gzip readers. +// +// Calling Multistream(false) disables this behavior; disabling the behavior +// can be useful when reading file formats that distinguish individual gzip +// data streams or mix gzip data streams with other data streams. +// In this mode, when the Reader reaches the end of the data stream, +// Read returns io.EOF. If the underlying reader implements io.ByteReader, +// it will be left positioned just after the gzip stream. +// To start the next stream, call z.Reset(r) followed by z.Multistream(false). +// If there is no next stream, z.Reset(r) will return io.EOF. +func (z *Reader) Multistream(ok bool) { + z.multistream = ok +} + +// readString reads a NUL-terminated string from z.r. +// It treats the bytes read as being encoded as ISO 8859-1 (Latin-1) and +// will output a string encoded using UTF-8. +// This method always updates z.digest with the data read. +func (z *Reader) readString() (string, error) { + var err error + needConv := false + for i := 0; ; i++ { + if i >= len(z.buf) { + return "", ErrHeader + } + z.buf[i], err = z.r.ReadByte() + if err != nil { + return "", err + } + if z.buf[i] > 0x7f { + needConv = true + } + if z.buf[i] == 0 { + // Digest covers the NUL terminator. + z.digest = crc32.Update(z.digest, crc32.IEEETable, z.buf[:i+1]) + + // Strings are ISO 8859-1, Latin-1 (RFC 1952, section 2.3.1). + if needConv { + s := make([]rune, 0, i) + for _, v := range z.buf[:i] { + s = append(s, rune(v)) + } + return string(s), nil + } + return string(z.buf[:i]), nil + } + } +} + +// readHeader reads the GZIP header according to section 2.3.1. +// This method does not set z.err. +func (z *Reader) readHeader() (hdr Header, err error) { + if _, err = io.ReadFull(z.r, z.buf[:10]); err != nil { + // RFC 1952, section 2.2, says the following: + // A gzip file consists of a series of "members" (compressed data sets). + // + // Other than this, the specification does not clarify whether a + // "series" is defined as "one or more" or "zero or more". To err on the + // side of caution, Go interprets this to mean "zero or more". + // Thus, it is okay to return io.EOF here. + return hdr, err + } + if z.buf[0] != gzipID1 || z.buf[1] != gzipID2 || z.buf[2] != gzipDeflate { + return hdr, ErrHeader + } + flg := z.buf[3] + hdr.ModTime = time.Unix(int64(le.Uint32(z.buf[4:8])), 0) + // z.buf[8] is XFL and is currently ignored. + hdr.OS = z.buf[9] + z.digest = crc32.ChecksumIEEE(z.buf[:10]) + + if flg&flagExtra != 0 { + if _, err = io.ReadFull(z.r, z.buf[:2]); err != nil { + return hdr, noEOF(err) + } + z.digest = crc32.Update(z.digest, crc32.IEEETable, z.buf[:2]) + data := make([]byte, le.Uint16(z.buf[:2])) + if _, err = io.ReadFull(z.r, data); err != nil { + return hdr, noEOF(err) + } + z.digest = crc32.Update(z.digest, crc32.IEEETable, data) + hdr.Extra = data + } + + var s string + if flg&flagName != 0 { + if s, err = z.readString(); err != nil { + return hdr, err + } + hdr.Name = s + } + + if flg&flagComment != 0 { + if s, err = z.readString(); err != nil { + return hdr, err + } + hdr.Comment = s + } + + if flg&flagHdrCrc != 0 { + if _, err = io.ReadFull(z.r, z.buf[:2]); err != nil { + return hdr, noEOF(err) + } + digest := le.Uint16(z.buf[:2]) + if digest != uint16(z.digest) { + return hdr, ErrHeader + } + } + + z.digest = 0 + if z.decompressor == nil { + z.decompressor = flate.NewReader(z.r) + } else { + z.decompressor.(flate.Resetter).Reset(z.r, nil) + } + return hdr, nil +} + +// Read implements io.Reader, reading uncompressed bytes from its underlying Reader. +func (z *Reader) Read(p []byte) (n int, err error) { + if z.err != nil { + return 0, z.err + } + + n, z.err = z.decompressor.Read(p) + z.digest = crc32.Update(z.digest, crc32.IEEETable, p[:n]) + z.size += uint32(n) + if z.err != io.EOF { + // In the normal case we return here. + return n, z.err + } + + // Finished file; check checksum and size. + if _, err := io.ReadFull(z.r, z.buf[:8]); err != nil { + z.err = noEOF(err) + return n, z.err + } + digest := le.Uint32(z.buf[:4]) + size := le.Uint32(z.buf[4:8]) + if digest != z.digest || size != z.size { + z.err = ErrChecksum + return n, z.err + } + z.digest, z.size = 0, 0 + + // File is ok; check if there is another. + if !z.multistream { + return n, io.EOF + } + z.err = nil // Remove io.EOF + + if _, z.err = z.readHeader(); z.err != nil { + return n, z.err + } + + // Read from next file, if necessary. + if n > 0 { + return n, nil + } + return z.Read(p) +} + +// Support the io.WriteTo interface for io.Copy and friends. +func (z *Reader) WriteTo(w io.Writer) (int64, error) { + total := int64(0) + crcWriter := crc32.NewIEEE() + for { + if z.err != nil { + if z.err == io.EOF { + return total, nil + } + return total, z.err + } + + // We write both to output and digest. + mw := io.MultiWriter(w, crcWriter) + n, err := z.decompressor.(io.WriterTo).WriteTo(mw) + total += n + z.size += uint32(n) + if err != nil { + z.err = err + return total, z.err + } + + // Finished file; check checksum + size. + if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil { + if err == io.EOF { + err = io.ErrUnexpectedEOF + } + z.err = err + return total, err + } + z.digest = crcWriter.Sum32() + digest := le.Uint32(z.buf[:4]) + size := le.Uint32(z.buf[4:8]) + if digest != z.digest || size != z.size { + z.err = ErrChecksum + return total, z.err + } + z.digest, z.size = 0, 0 + + // File is ok; check if there is another. + if !z.multistream { + return total, nil + } + crcWriter.Reset() + z.err = nil // Remove io.EOF + + if _, z.err = z.readHeader(); z.err != nil { + if z.err == io.EOF { + return total, nil + } + return total, z.err + } + } +} + +// Close closes the Reader. It does not close the underlying io.Reader. +// In order for the GZIP checksum to be verified, the reader must be +// fully consumed until the io.EOF. +func (z *Reader) Close() error { return z.decompressor.Close() } diff --git a/github.com/klauspost/compress/gzip/gunzip_test.go b/github.com/klauspost/compress/gzip/gunzip_test.go new file mode 100644 index 0000000..bc7ca92 --- /dev/null +++ b/github.com/klauspost/compress/gzip/gunzip_test.go @@ -0,0 +1,716 @@ +// 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 gzip + +import ( + "bytes" + oldgz "compress/gzip" + "crypto/rand" + "io" + "io/ioutil" + "os" + "strings" + "testing" + "time" + + "github.com/klauspost/compress/flate" +) + +type gunzipTest struct { + name string + desc string + raw string + gzip []byte + err error +} + +var gunzipTests = []gunzipTest{ + { // has 1 empty fixed-huffman block + "empty.txt", + "empty.txt", + "", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xf7, 0x5e, 0x14, 0x4a, + 0x00, 0x03, 0x65, 0x6d, 0x70, 0x74, 0x79, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0x03, 0x00, 0x00, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + }, + nil, + }, + { + "", + "empty - with no file name", + "", + []byte{ + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x09, 0x6e, 0x88, + 0x00, 0xff, 0x01, 0x00, 0x00, 0xff, 0xff, 0x00, + 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + }, + nil, + }, + { // has 1 non-empty fixed huffman block + "hello.txt", + "hello.txt", + "hello world\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0x2d, 0x3b, 0x08, 0xaf, 0x0c, 0x00, + 0x00, 0x00, + }, + nil, + }, + { // concatenation + "hello.txt", + "hello.txt x2", + "hello world\n" + + "hello world\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0x2d, 0x3b, 0x08, 0xaf, 0x0c, 0x00, + 0x00, 0x00, + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0x2d, 0x3b, 0x08, 0xaf, 0x0c, 0x00, + 0x00, 0x00, + }, + nil, + }, + { // has a fixed huffman block with some length-distance pairs + "shesells.txt", + "shesells.txt", + "she sells seashells by the seashore\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0x72, 0x66, 0x8b, 0x4a, + 0x00, 0x03, 0x73, 0x68, 0x65, 0x73, 0x65, 0x6c, + 0x6c, 0x73, 0x2e, 0x74, 0x78, 0x74, 0x00, 0x2b, + 0xce, 0x48, 0x55, 0x28, 0x4e, 0xcd, 0xc9, 0x29, + 0x06, 0x92, 0x89, 0xc5, 0x19, 0x60, 0x56, 0x52, + 0xa5, 0x42, 0x09, 0x58, 0x18, 0x28, 0x90, 0x5f, + 0x94, 0xca, 0x05, 0x00, 0x76, 0xb0, 0x3b, 0xeb, + 0x24, 0x00, 0x00, 0x00, + }, + nil, + }, + { // has dynamic huffman blocks + "gettysburg", + "gettysburg", + " Four score and seven years ago our fathers brought forth on\n" + + "this continent, a new nation, conceived in Liberty, and dedicated\n" + + "to the proposition that all men are created equal.\n" + + " Now we are engaged in a great Civil War, testing whether that\n" + + "nation, or any nation so conceived and so dedicated, can long\n" + + "endure.\n" + + " We are met on a great battle-field of that war.\n" + + " We have come to dedicate a portion of that field, as a final\n" + + "resting place for those who here gave their lives that that\n" + + "nation might live. It is altogether fitting and proper that\n" + + "we should do this.\n" + + " But, in a larger sense, we can not dedicate — we can not\n" + + "consecrate — we can not hallow — this ground.\n" + + " The brave men, living and dead, who struggled here, have\n" + + "consecrated it, far above our poor power to add or detract.\n" + + "The world will little note, nor long remember what we say here,\n" + + "but it can never forget what they did here.\n" + + " It is for us the living, rather, to be dedicated here to the\n" + + "unfinished work which they who fought here have thus far so\n" + + "nobly advanced. It is rather for us to be here dedicated to\n" + + "the great task remaining before us — that from these honored\n" + + "dead we take increased devotion to that cause for which they\n" + + "gave the last full measure of devotion —\n" + + " that we here highly resolve that these dead shall not have\n" + + "died in vain — that this nation, under God, shall have a new\n" + + "birth of freedom — and that government of the people, by the\n" + + "people, for the people, shall not perish from this earth.\n" + + "\n" + + "Abraham Lincoln, November 19, 1863, Gettysburg, Pennsylvania\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xd1, 0x12, 0x2b, 0x4a, + 0x00, 0x03, 0x67, 0x65, 0x74, 0x74, 0x79, 0x73, + 0x62, 0x75, 0x72, 0x67, 0x00, 0x65, 0x54, 0xcd, + 0x6e, 0xd4, 0x30, 0x10, 0xbe, 0xfb, 0x29, 0xe6, + 0x01, 0x42, 0xa5, 0x0a, 0x09, 0xc1, 0x11, 0x90, + 0x40, 0x48, 0xa8, 0xe2, 0x80, 0xd4, 0xf3, 0x24, + 0x9e, 0x24, 0x56, 0xbd, 0x9e, 0xc5, 0x76, 0x76, + 0x95, 0x1b, 0x0f, 0xc1, 0x13, 0xf2, 0x24, 0x7c, + 0x63, 0x77, 0x9b, 0x4a, 0x5c, 0xaa, 0x6e, 0x6c, + 0xcf, 0x7c, 0x7f, 0x33, 0x44, 0x5f, 0x74, 0xcb, + 0x54, 0x26, 0xcd, 0x42, 0x9c, 0x3c, 0x15, 0xb9, + 0x48, 0xa2, 0x5d, 0x38, 0x17, 0xe2, 0x45, 0xc9, + 0x4e, 0x67, 0xae, 0xab, 0xe0, 0xf7, 0x98, 0x75, + 0x5b, 0xd6, 0x4a, 0xb3, 0xe6, 0xba, 0x92, 0x26, + 0x57, 0xd7, 0x50, 0x68, 0xd2, 0x54, 0x43, 0x92, + 0x54, 0x07, 0x62, 0x4a, 0x72, 0xa5, 0xc4, 0x35, + 0x68, 0x1a, 0xec, 0x60, 0x92, 0x70, 0x11, 0x4f, + 0x21, 0xd1, 0xf7, 0x30, 0x4a, 0xae, 0xfb, 0xd0, + 0x9a, 0x78, 0xf1, 0x61, 0xe2, 0x2a, 0xde, 0x55, + 0x25, 0xd4, 0xa6, 0x73, 0xd6, 0xb3, 0x96, 0x60, + 0xef, 0xf0, 0x9b, 0x2b, 0x71, 0x8c, 0x74, 0x02, + 0x10, 0x06, 0xac, 0x29, 0x8b, 0xdd, 0x25, 0xf9, + 0xb5, 0x71, 0xbc, 0x73, 0x44, 0x0f, 0x7a, 0xa5, + 0xab, 0xb4, 0x33, 0x49, 0x0b, 0x2f, 0xbd, 0x03, + 0xd3, 0x62, 0x17, 0xe9, 0x73, 0xb8, 0x84, 0x48, + 0x8f, 0x9c, 0x07, 0xaa, 0x52, 0x00, 0x6d, 0xa1, + 0xeb, 0x2a, 0xc6, 0xa0, 0x95, 0x76, 0x37, 0x78, + 0x9a, 0x81, 0x65, 0x7f, 0x46, 0x4b, 0x45, 0x5f, + 0xe1, 0x6d, 0x42, 0xe8, 0x01, 0x13, 0x5c, 0x38, + 0x51, 0xd4, 0xb4, 0x38, 0x49, 0x7e, 0xcb, 0x62, + 0x28, 0x1e, 0x3b, 0x82, 0x93, 0x54, 0x48, 0xf1, + 0xd2, 0x7d, 0xe4, 0x5a, 0xa3, 0xbc, 0x99, 0x83, + 0x44, 0x4f, 0x3a, 0x77, 0x36, 0x57, 0xce, 0xcf, + 0x2f, 0x56, 0xbe, 0x80, 0x90, 0x9e, 0x84, 0xea, + 0x51, 0x1f, 0x8f, 0xcf, 0x90, 0xd4, 0x60, 0xdc, + 0x5e, 0xb4, 0xf7, 0x10, 0x0b, 0x26, 0xe0, 0xff, + 0xc4, 0xd1, 0xe5, 0x67, 0x2e, 0xe7, 0xc8, 0x93, + 0x98, 0x05, 0xb8, 0xa8, 0x45, 0xc0, 0x4d, 0x09, + 0xdc, 0x84, 0x16, 0x2b, 0x0d, 0x9a, 0x21, 0x53, + 0x04, 0x8b, 0xd2, 0x0b, 0xbd, 0xa2, 0x4c, 0xa7, + 0x60, 0xee, 0xd9, 0xe1, 0x1d, 0xd1, 0xb7, 0x4a, + 0x30, 0x8f, 0x63, 0xd5, 0xa5, 0x8b, 0x33, 0x87, + 0xda, 0x1a, 0x18, 0x79, 0xf3, 0xe3, 0xa6, 0x17, + 0x94, 0x2e, 0xab, 0x6e, 0xa0, 0xe3, 0xcd, 0xac, + 0x50, 0x8c, 0xca, 0xa7, 0x0d, 0x76, 0x37, 0xd1, + 0x23, 0xe7, 0x05, 0x57, 0x8b, 0xa4, 0x22, 0x83, + 0xd9, 0x62, 0x52, 0x25, 0xad, 0x07, 0xbb, 0xbf, + 0xbf, 0xff, 0xbc, 0xfa, 0xee, 0x20, 0x73, 0x91, + 0x29, 0xff, 0x7f, 0x02, 0x71, 0x62, 0x84, 0xb5, + 0xf6, 0xb5, 0x25, 0x6b, 0x41, 0xde, 0x92, 0xb7, + 0x76, 0x3f, 0x91, 0x91, 0x31, 0x1b, 0x41, 0x84, + 0x62, 0x30, 0x0a, 0x37, 0xa4, 0x5e, 0x18, 0x3a, + 0x99, 0x08, 0xa5, 0xe6, 0x6d, 0x59, 0x22, 0xec, + 0x33, 0x39, 0x86, 0x26, 0xf5, 0xab, 0x66, 0xc8, + 0x08, 0x20, 0xcf, 0x0c, 0xd7, 0x47, 0x45, 0x21, + 0x0b, 0xf6, 0x59, 0xd5, 0xfe, 0x5c, 0x8d, 0xaa, + 0x12, 0x7b, 0x6f, 0xa1, 0xf0, 0x52, 0x33, 0x4f, + 0xf5, 0xce, 0x59, 0xd3, 0xab, 0x66, 0x10, 0xbf, + 0x06, 0xc4, 0x31, 0x06, 0x73, 0xd6, 0x80, 0xa2, + 0x78, 0xc2, 0x45, 0xcb, 0x03, 0x65, 0x39, 0xc9, + 0x09, 0xd1, 0x06, 0x04, 0x33, 0x1a, 0x5a, 0xf1, + 0xde, 0x01, 0xb8, 0x71, 0x83, 0xc4, 0xb5, 0xb3, + 0xc3, 0x54, 0x65, 0x33, 0x0d, 0x5a, 0xf7, 0x9b, + 0x90, 0x7c, 0x27, 0x1f, 0x3a, 0x58, 0xa3, 0xd8, + 0xfd, 0x30, 0x5f, 0xb7, 0xd2, 0x66, 0xa2, 0x93, + 0x1c, 0x28, 0xb7, 0xe9, 0x1b, 0x0c, 0xe1, 0x28, + 0x47, 0x26, 0xbb, 0xe9, 0x7d, 0x7e, 0xdc, 0x96, + 0x10, 0x92, 0x50, 0x56, 0x7c, 0x06, 0xe2, 0x27, + 0xb4, 0x08, 0xd3, 0xda, 0x7b, 0x98, 0x34, 0x73, + 0x9f, 0xdb, 0xf6, 0x62, 0xed, 0x31, 0x41, 0x13, + 0xd3, 0xa2, 0xa8, 0x4b, 0x3a, 0xc6, 0x1d, 0xe4, + 0x2f, 0x8c, 0xf8, 0xfb, 0x97, 0x64, 0xf4, 0xb6, + 0x2f, 0x80, 0x5a, 0xf3, 0x56, 0xe0, 0x40, 0x50, + 0xd5, 0x19, 0xd0, 0x1e, 0xfc, 0xca, 0xe5, 0xc9, + 0xd4, 0x60, 0x00, 0x81, 0x2e, 0xa3, 0xcc, 0xb6, + 0x52, 0xf0, 0xb4, 0xdb, 0x69, 0x99, 0xce, 0x7a, + 0x32, 0x4c, 0x08, 0xed, 0xaa, 0x10, 0x10, 0xe3, + 0x6f, 0xee, 0x99, 0x68, 0x95, 0x9f, 0x04, 0x71, + 0xb2, 0x49, 0x2f, 0x62, 0xa6, 0x5e, 0xb4, 0xef, + 0x02, 0xed, 0x4f, 0x27, 0xde, 0x4a, 0x0f, 0xfd, + 0xc1, 0xcc, 0xdd, 0x02, 0x8f, 0x08, 0x16, 0x54, + 0xdf, 0xda, 0xca, 0xe0, 0x82, 0xf1, 0xb4, 0x31, + 0x7a, 0xa9, 0x81, 0xfe, 0x90, 0xb7, 0x3e, 0xdb, + 0xd3, 0x35, 0xc0, 0x20, 0x80, 0x33, 0x46, 0x4a, + 0x63, 0xab, 0xd1, 0x0d, 0x29, 0xd2, 0xe2, 0x84, + 0xb8, 0xdb, 0xfa, 0xe9, 0x89, 0x44, 0x86, 0x7c, + 0xe8, 0x0b, 0xe6, 0x02, 0x6a, 0x07, 0x9b, 0x96, + 0xd0, 0xdb, 0x2e, 0x41, 0x4c, 0xa1, 0xd5, 0x57, + 0x45, 0x14, 0xfb, 0xe3, 0xa6, 0x72, 0x5b, 0x87, + 0x6e, 0x0c, 0x6d, 0x5b, 0xce, 0xe0, 0x2f, 0xe2, + 0x21, 0x81, 0x95, 0xb0, 0xe8, 0xb6, 0x32, 0x0b, + 0xb2, 0x98, 0x13, 0x52, 0x5d, 0xfb, 0xec, 0x63, + 0x17, 0x8a, 0x9e, 0x23, 0x22, 0x36, 0xee, 0xcd, + 0xda, 0xdb, 0xcf, 0x3e, 0xf1, 0xc7, 0xf1, 0x01, + 0x12, 0x93, 0x0a, 0xeb, 0x6f, 0xf2, 0x02, 0x15, + 0x96, 0x77, 0x5d, 0xef, 0x9c, 0xfb, 0x88, 0x91, + 0x59, 0xf9, 0x84, 0xdd, 0x9b, 0x26, 0x8d, 0x80, + 0xf9, 0x80, 0x66, 0x2d, 0xac, 0xf7, 0x1f, 0x06, + 0xba, 0x7f, 0xff, 0xee, 0xed, 0x40, 0x5f, 0xa5, + 0xd6, 0xbd, 0x8c, 0x5b, 0x46, 0xd2, 0x7e, 0x48, + 0x4a, 0x65, 0x8f, 0x08, 0x42, 0x60, 0xf7, 0x0f, + 0xb9, 0x16, 0x0b, 0x0c, 0x1a, 0x06, 0x00, 0x00, + }, + nil, + }, + { // has 1 non-empty fixed huffman block then garbage + "hello.txt", + "hello.txt + garbage", + "hello world\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0x2d, 0x3b, 0x08, 0xaf, 0x0c, 0x00, + 0x00, 0x00, 'g', 'a', 'r', 'b', 'a', 'g', 'e', '!', '!', '!', + }, + ErrHeader, + }, + { // has 1 non-empty fixed huffman block not enough header + "hello.txt", + "hello.txt + garbage", + "hello world\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0x2d, 0x3b, 0x08, 0xaf, 0x0c, 0x00, + 0x00, 0x00, gzipID1, + }, + io.ErrUnexpectedEOF, + }, + { // has 1 non-empty fixed huffman block but corrupt checksum + "hello.txt", + "hello.txt + corrupt checksum", + "hello world\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0xff, 0xff, 0xff, 0xff, 0x0c, 0x00, + 0x00, 0x00, + }, + ErrChecksum, + }, + { // has 1 non-empty fixed huffman block but corrupt size + "hello.txt", + "hello.txt + corrupt size", + "hello world\n", + []byte{ + 0x1f, 0x8b, 0x08, 0x08, 0xc8, 0x58, 0x13, 0x4a, + 0x00, 0x03, 0x68, 0x65, 0x6c, 0x6c, 0x6f, 0x2e, + 0x74, 0x78, 0x74, 0x00, 0xcb, 0x48, 0xcd, 0xc9, + 0xc9, 0x57, 0x28, 0xcf, 0x2f, 0xca, 0x49, 0xe1, + 0x02, 0x00, 0x2d, 0x3b, 0x08, 0xaf, 0xff, 0x00, + 0x00, 0x00, + }, + ErrChecksum, + }, + { + "f1l3n4m3.tXt", + "header with all fields used", + "", + []byte{ + 0x1f, 0x8b, 0x08, 0x1e, 0x70, 0xf0, 0xf9, 0x4a, + 0x00, 0xaa, 0x09, 0x00, 0x7a, 0x7a, 0x05, 0x00, + 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x31, 0x6c, + 0x33, 0x6e, 0x34, 0x6d, 0x33, 0x2e, 0x74, 0x58, + 0x74, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, + 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, + 0x0f, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, + 0x17, 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, + 0x1f, 0x20, 0x21, 0x22, 0x23, 0x24, 0x25, 0x26, + 0x27, 0x28, 0x29, 0x2a, 0x2b, 0x2c, 0x2d, 0x2e, + 0x2f, 0x30, 0x31, 0x32, 0x33, 0x34, 0x35, 0x36, + 0x37, 0x38, 0x39, 0x3a, 0x3b, 0x3c, 0x3d, 0x3e, + 0x3f, 0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, + 0x47, 0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, + 0x4f, 0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, + 0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, + 0x5f, 0x60, 0x61, 0x62, 0x63, 0x64, 0x65, 0x66, + 0x67, 0x68, 0x69, 0x6a, 0x6b, 0x6c, 0x6d, 0x6e, + 0x6f, 0x70, 0x71, 0x72, 0x73, 0x74, 0x75, 0x76, + 0x77, 0x78, 0x79, 0x7a, 0x7b, 0x7c, 0x7d, 0x7e, + 0x7f, 0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, + 0x87, 0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, + 0x8f, 0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, + 0x97, 0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, + 0x9f, 0xa0, 0xa1, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, + 0xa7, 0xa8, 0xa9, 0xaa, 0xab, 0xac, 0xad, 0xae, + 0xaf, 0xb0, 0xb1, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, + 0xb7, 0xb8, 0xb9, 0xba, 0xbb, 0xbc, 0xbd, 0xbe, + 0xbf, 0xc0, 0xc1, 0xc2, 0xc3, 0xc4, 0xc5, 0xc6, + 0xc7, 0xc8, 0xc9, 0xca, 0xcb, 0xcc, 0xcd, 0xce, + 0xcf, 0xd0, 0xd1, 0xd2, 0xd3, 0xd4, 0xd5, 0xd6, + 0xd7, 0xd8, 0xd9, 0xda, 0xdb, 0xdc, 0xdd, 0xde, + 0xdf, 0xe0, 0xe1, 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, + 0xe7, 0xe8, 0xe9, 0xea, 0xeb, 0xec, 0xed, 0xee, + 0xef, 0xf0, 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, + 0xf7, 0xf8, 0xf9, 0xfa, 0xfb, 0xfc, 0xfd, 0xfe, + 0xff, 0x00, 0x92, 0xfd, 0x01, 0x00, 0x00, 0xff, + 0xff, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, + 0x00, + }, + nil, + }, + { + "", + "truncated gzip file amid raw-block", + "hello", + []byte{ + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, + 0x00, 0x0c, 0x00, 0xf3, 0xff, 0x68, 0x65, 0x6c, 0x6c, 0x6f, + }, + io.ErrUnexpectedEOF, + }, + { + "", + "truncated gzip file amid fixed-block", + "He", + []byte{ + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, + 0xf2, 0x48, 0xcd, + }, + io.ErrUnexpectedEOF, + }, +} + +func TestDecompressor(t *testing.T) { + b := new(bytes.Buffer) + for _, tt := range gunzipTests { + in := bytes.NewReader(tt.gzip) + gzip, err := NewReader(in) + if err != nil { + t.Errorf("%s: NewReader: %s", tt.name, err) + continue + } + defer gzip.Close() + if tt.name != gzip.Name { + t.Errorf("%s: got name %s", tt.name, gzip.Name) + } + b.Reset() + n, err := io.Copy(b, gzip) + if err != tt.err { + t.Errorf("%s: io.Copy: %v want %v", tt.name, err, tt.err) + } + s := b.String() + if s != tt.raw { + t.Errorf("%s: got %d-byte %q want %d-byte %q", tt.name, n, s, len(tt.raw), tt.raw) + } + + // Test Reader Reset. + in = bytes.NewReader(tt.gzip) + err = gzip.Reset(in) + if err != nil { + t.Errorf("%s: Reset: %s", tt.name, err) + continue + } + if tt.name != gzip.Name { + t.Errorf("%s: got name %s", tt.name, gzip.Name) + } + b.Reset() + n, err = io.Copy(b, gzip) + if err != tt.err { + t.Errorf("%s: io.Copy: %v want %v", tt.name, err, tt.err) + } + s = b.String() + if s != tt.raw { + t.Errorf("%s: got %d-byte %q want %d-byte %q", tt.name, n, s, len(tt.raw), tt.raw) + } + } +} + +func TestIssue6550(t *testing.T) { + f, err := os.Open("testdata/issue6550.gz") + if err != nil { + t.Fatal(err) + } + gzip, err := NewReader(f) + if err != nil { + t.Fatalf("NewReader(testdata/issue6550.gz): %v", err) + } + defer gzip.Close() + done := make(chan bool, 1) + go func() { + _, err := io.Copy(ioutil.Discard, gzip) + if err == nil { + t.Errorf("Copy succeeded") + } else { + t.Logf("Copy failed (correctly): %v", err) + } + done <- true + }() + select { + case <-time.After(1 * time.Second): + t.Errorf("Copy hung") + case <-done: + // ok + } +} + +func TestInitialReset(t *testing.T) { + var r Reader + if err := r.Reset(bytes.NewReader(gunzipTests[1].gzip)); err != nil { + t.Error(err) + } + var buf bytes.Buffer + if _, err := io.Copy(&buf, &r); err != nil { + t.Error(err) + } + if s := buf.String(); s != gunzipTests[1].raw { + t.Errorf("got %q want %q", s, gunzipTests[1].raw) + } +} + +func TestMultistreamFalse(t *testing.T) { + // Find concatenation test. + var tt gunzipTest + for _, tt = range gunzipTests { + if strings.HasSuffix(tt.desc, " x2") { + goto Found + } + } + t.Fatal("cannot find hello.txt x2 in gunzip tests") + +Found: + br := bytes.NewReader(tt.gzip) + var r Reader + if err := r.Reset(br); err != nil { + t.Fatalf("first reset: %v", err) + } + + // Expect two streams with "hello world\n", then real EOF. + const hello = "hello world\n" + + r.Multistream(false) + data, err := ioutil.ReadAll(&r) + if string(data) != hello || err != nil { + t.Fatalf("first stream = %q, %v, want %q, %v", string(data), err, hello, nil) + } + + if err := r.Reset(br); err != nil { + t.Fatalf("second reset: %v", err) + } + r.Multistream(false) + data, err = ioutil.ReadAll(&r) + if string(data) != hello || err != nil { + t.Fatalf("second stream = %q, %v, want %q, %v", string(data), err, hello, nil) + } + + if err := r.Reset(br); err != io.EOF { + t.Fatalf("third reset: err=%v, want io.EOF", err) + } +} + +func TestWriteTo(t *testing.T) { + input := make([]byte, 100000) + n, err := rand.Read(input) + if err != nil { + t.Fatal(err) + } + if n != len(input) { + t.Fatal("did not fill buffer") + } + compressed := &bytes.Buffer{} + // Do it twice to test MultiStream functionality + for i := 0; i < 2; i++ { + w, err := NewWriterLevel(compressed, -2) + if err != nil { + t.Fatal(err) + } + n, err = w.Write(input) + if err != nil { + t.Fatal(err) + } + if n != len(input) { + t.Fatal("did not fill buffer") + } + w.Close() + } + input = append(input, input...) + buf := compressed.Bytes() + + dec, err := NewReader(bytes.NewBuffer(buf)) + if err != nil { + t.Fatal(err) + } + // ReadAll does not use WriteTo, but we wrap it in a NopCloser to be sure. + readall, err := ioutil.ReadAll(ioutil.NopCloser(dec)) + if err != nil { + t.Fatal(err) + } + if len(readall) != len(input) { + t.Errorf("did not decompress everything, want %d, got %d", len(input), len(readall)) + } + if bytes.Compare(readall, input) != 0 { + t.Error("output did not match input") + } + + dec, err = NewReader(bytes.NewBuffer(buf)) + if err != nil { + t.Fatal(err) + } + wtbuf := &bytes.Buffer{} + written, err := dec.WriteTo(wtbuf) + if err != nil { + t.Fatal(err) + } + if written != int64(len(input)) { + t.Error("Returned length did not match, expected", len(input), "got", written) + } + if wtbuf.Len() != len(input) { + t.Error("Actual Length did not match, expected", len(input), "got", wtbuf.Len()) + } + if bytes.Compare(wtbuf.Bytes(), input) != 0 { + t.Fatal("output did not match input") + } +} + +func TestNilStream(t *testing.T) { + // Go liberally interprets RFC 1952 section 2.2 to mean that a gzip file + // consist of zero or more members. Thus, we test that a nil stream is okay. + _, err := NewReader(bytes.NewReader(nil)) + if err != io.EOF { + t.Fatalf("NewReader(nil) on empty stream: got %v, want io.EOF", err) + } +} + +func TestTruncatedStreams(t *testing.T) { + const data = "\x1f\x8b\b\x04\x00\tn\x88\x00\xff\a\x00foo bar\xcbH\xcd\xc9\xc9\xd7Q(\xcf/\xcaI\x01\x04:r\xab\xff\f\x00\x00\x00" + + // Intentionally iterate starting with at least one byte in the stream. + for i := 1; i < len(data)-1; i++ { + r, err := NewReader(strings.NewReader(data[:i])) + if err != nil { + if err != io.ErrUnexpectedEOF { + t.Errorf("NewReader(%d) on truncated stream: got %v, want %v", i, err, io.ErrUnexpectedEOF) + } + continue + } + _, err = io.Copy(ioutil.Discard, r) + if ferr, ok := err.(*flate.ReadError); ok { + err = ferr.Err + } + if err != io.ErrUnexpectedEOF { + t.Errorf("io.Copy(%d) on truncated stream: got %v, want %v", i, err, io.ErrUnexpectedEOF) + } + } +} + +func BenchmarkGunzipCopy(b *testing.B) { + dat, _ := ioutil.ReadFile("testdata/test.json") + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dst := &bytes.Buffer{} + w, _ := NewWriterLevel(dst, 1) + _, err := w.Write(dat) + if err != nil { + b.Fatal(err) + } + w.Close() + input := dst.Bytes() + b.SetBytes(int64(len(dat))) + b.ResetTimer() + for n := 0; n < b.N; n++ { + r, err := NewReader(bytes.NewBuffer(input)) + if err != nil { + b.Fatal(err) + } + _, err = io.Copy(ioutil.Discard, r) + if err != nil { + b.Fatal(err) + } + } +} + +func BenchmarkGunzipNoWriteTo(b *testing.B) { + dat, _ := ioutil.ReadFile("testdata/test.json") + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dst := &bytes.Buffer{} + w, _ := NewWriterLevel(dst, 1) + _, err := w.Write(dat) + if err != nil { + b.Fatal(err) + } + w.Close() + input := dst.Bytes() + r, err := NewReader(bytes.NewBuffer(input)) + if err != nil { + b.Fatal(err) + } + b.SetBytes(int64(len(dat))) + b.ResetTimer() + for n := 0; n < b.N; n++ { + err := r.Reset(bytes.NewBuffer(input)) + if err != nil { + b.Fatal(err) + } + _, err = io.Copy(ioutil.Discard, ioutil.NopCloser(r)) + if err != nil { + b.Fatal(err) + } + } +} + +func BenchmarkGunzipStdlib(b *testing.B) { + dat, _ := ioutil.ReadFile("testdata/test.json") + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dst := &bytes.Buffer{} + w, _ := NewWriterLevel(dst, 1) + _, err := w.Write(dat) + if err != nil { + b.Fatal(err) + } + w.Close() + input := dst.Bytes() + r, err := oldgz.NewReader(bytes.NewBuffer(input)) + if err != nil { + b.Fatal(err) + } + b.SetBytes(int64(len(dat))) + b.ResetTimer() + for n := 0; n < b.N; n++ { + err := r.Reset(bytes.NewBuffer(input)) + if err != nil { + b.Fatal(err) + } + _, err = io.Copy(ioutil.Discard, r) + if err != nil { + b.Fatal(err) + } + } +} +func TestTruncatedGunzip(t *testing.T) { + in := []byte(strings.Repeat("ASDFASDFASDFASDFASDF", 1000)) + var buf bytes.Buffer + enc := NewWriter(&buf) + _, err := enc.Write(in) + if err != nil { + t.Fatal(err) + } + enc.Close() + testdata := buf.Bytes() + for i := 5; i < len(testdata); i += 10 { + timer := time.NewTimer(time.Second) + done := make(chan struct{}) + fail := make(chan struct{}) + go func() { + r, err := NewReader(bytes.NewBuffer(testdata[:i])) + if err == nil { + b, err := ioutil.ReadAll(r) + if err == nil && !bytes.Equal(testdata[:i], b) { + close(fail) + } + } + close(done) + }() + select { + case <-timer.C: + t.Fatal("Timeout decoding") + case <-fail: + t.Fatal("No error, but mismatch") + case <-done: + timer.Stop() + } + } +} diff --git a/github.com/klauspost/compress/gzip/gzip.go b/github.com/klauspost/compress/gzip/gzip.go new file mode 100644 index 0000000..2620385 --- /dev/null +++ b/github.com/klauspost/compress/gzip/gzip.go @@ -0,0 +1,269 @@ +// Copyright 2010 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 gzip + +import ( + "errors" + "fmt" + "hash/crc32" + "io" + + "github.com/klauspost/compress/flate" +) + +// These constants are copied from the flate package, so that code that imports +// "compress/gzip" does not also have to import "compress/flate". +const ( + NoCompression = flate.NoCompression + BestSpeed = flate.BestSpeed + BestCompression = flate.BestCompression + DefaultCompression = flate.DefaultCompression + ConstantCompression = flate.ConstantCompression + HuffmanOnly = flate.HuffmanOnly + + // StatelessCompression will do compression but without maintaining any state + // between Write calls. + // There will be no memory kept between Write calls, + // but compression and speed will be suboptimal. + // Because of this, the size of actual Write calls will affect output size. + StatelessCompression = -3 +) + +// A Writer is an io.WriteCloser. +// Writes to a Writer are compressed and written to w. +type Writer struct { + Header // written at first call to Write, Flush, or Close + w io.Writer + level int + 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 +} + +// NewWriter returns a new Writer. +// Writes to the returned writer are compressed and written to w. +// +// It is the caller's responsibility to call Close on the WriteCloser when done. +// Writes may be buffered and not flushed until Close. +// +// Callers that wish to set the fields in Writer.Header must do so before +// the first call to Write, Flush, or Close. +func NewWriter(w io.Writer) *Writer { + z, _ := NewWriterLevel(w, DefaultCompression) + return z +} + +// NewWriterLevel is like NewWriter but specifies the compression level instead +// of assuming DefaultCompression. +// +// The compression level can be DefaultCompression, NoCompression, or any +// integer value between BestSpeed and BestCompression inclusive. The error +// returned will be nil if the level is valid. +func NewWriterLevel(w io.Writer, level int) (*Writer, error) { + if level < StatelessCompression || level > BestCompression { + return nil, fmt.Errorf("gzip: invalid compression level: %d", level) + } + z := new(Writer) + z.init(w, level) + return z, nil +} + +func (z *Writer) init(w io.Writer, level int) { + compressor := z.compressor + if level != StatelessCompression { + if compressor != nil { + compressor.Reset(w) + } + } + + *z = Writer{ + Header: Header{ + OS: 255, // unknown + }, + w: w, + level: level, + compressor: compressor, + } +} + +// Reset discards the Writer z's state and makes it equivalent to the +// result of its original state from NewWriter or NewWriterLevel, but +// writing to w instead. This permits reusing a Writer rather than +// allocating a new one. +func (z *Writer) Reset(w io.Writer) { + z.init(w, z.level) +} + +// writeBytes writes a length-prefixed byte slice to z.w. +func (z *Writer) writeBytes(b []byte) error { + if len(b) > 0xffff { + return errors.New("gzip.Write: Extra data is too large") + } + le.PutUint16(z.buf[:2], uint16(len(b))) + _, err := z.w.Write(z.buf[:2]) + if err != nil { + return err + } + _, err = z.w.Write(b) + return err +} + +// writeString writes a UTF-8 string s in GZIP's format to z.w. +// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1). +func (z *Writer) writeString(s string) (err error) { + // GZIP stores Latin-1 strings; error if non-Latin-1; convert if non-ASCII. + needconv := false + for _, v := range s { + if v == 0 || v > 0xff { + return errors.New("gzip.Write: non-Latin-1 header string") + } + if v > 0x7f { + needconv = true + } + } + if needconv { + b := make([]byte, 0, len(s)) + for _, v := range s { + b = append(b, byte(v)) + } + _, err = z.w.Write(b) + } else { + _, err = io.WriteString(z.w, s) + } + if err != nil { + return err + } + // GZIP strings are NUL-terminated. + z.buf[0] = 0 + _, err = z.w.Write(z.buf[:1]) + return err +} + +// Write writes a compressed form of p to the underlying io.Writer. The +// compressed bytes are not necessarily flushed until the Writer is closed. +func (z *Writer) Write(p []byte) (int, error) { + if z.err != nil { + return 0, z.err + } + var n int + // Write the GZIP header lazily. + if !z.wroteHeader { + z.wroteHeader = true + z.buf[0] = gzipID1 + z.buf[1] = gzipID2 + z.buf[2] = gzipDeflate + z.buf[3] = 0 + if z.Extra != nil { + z.buf[3] |= 0x04 + } + if z.Name != "" { + z.buf[3] |= 0x08 + } + if z.Comment != "" { + z.buf[3] |= 0x10 + } + le.PutUint32(z.buf[4:8], uint32(z.ModTime.Unix())) + if z.level == BestCompression { + z.buf[8] = 2 + } else if z.level == BestSpeed { + z.buf[8] = 4 + } else { + z.buf[8] = 0 + } + z.buf[9] = z.OS + n, z.err = z.w.Write(z.buf[:10]) + if z.err != nil { + return n, z.err + } + if z.Extra != nil { + z.err = z.writeBytes(z.Extra) + if z.err != nil { + return n, z.err + } + } + if z.Name != "" { + z.err = z.writeString(z.Name) + if z.err != nil { + return n, z.err + } + } + if z.Comment != "" { + z.err = z.writeString(z.Comment) + if z.err != nil { + return n, z.err + } + } + + if z.compressor == nil && z.level != StatelessCompression { + z.compressor, _ = flate.NewWriter(z.w, z.level) + } + } + 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, nil) + } + n, z.err = z.compressor.Write(p) + return n, z.err +} + +// Flush flushes any pending compressed data to the underlying writer. +// +// It is useful mainly in compressed network protocols, to ensure that +// a remote reader has enough data to reconstruct a packet. Flush does +// not return until the data has been written. If the underlying +// writer returns an error, Flush returns that error. +// +// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH. +func (z *Writer) Flush() error { + if z.err != nil { + return z.err + } + if z.closed || z.level == StatelessCompression { + return nil + } + if !z.wroteHeader { + z.Write(nil) + if z.err != nil { + return z.err + } + } + z.err = z.compressor.Flush() + return z.err +} + +// Close closes the Writer, flushing any unwritten data to the underlying +// io.Writer, but does not close the underlying io.Writer. +func (z *Writer) Close() error { + if z.err != nil { + return z.err + } + if z.closed { + return nil + } + z.closed = true + if !z.wroteHeader { + z.Write(nil) + if z.err != nil { + return z.err + } + } + if z.level == StatelessCompression { + z.err = flate.StatelessDeflate(z.w, nil, true, nil) + } else { + z.err = z.compressor.Close() + } + if z.err != nil { + return z.err + } + le.PutUint32(z.buf[:4], z.digest) + le.PutUint32(z.buf[4:8], z.size) + _, z.err = z.w.Write(z.buf[:8]) + return z.err +} diff --git a/github.com/klauspost/compress/gzip/gzip_test.go b/github.com/klauspost/compress/gzip/gzip_test.go new file mode 100644 index 0000000..135637f --- /dev/null +++ b/github.com/klauspost/compress/gzip/gzip_test.go @@ -0,0 +1,521 @@ +// Copyright 2010 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 gzip + +import ( + "bufio" + "bytes" + oldgz "compress/gzip" + "io" + "io/ioutil" + "math/rand" + "testing" + "time" +) + +// TestEmpty tests that an empty payload still forms a valid GZIP stream. +func TestEmpty(t *testing.T) { + buf := new(bytes.Buffer) + + if err := NewWriter(buf).Close(); err != nil { + t.Fatalf("Writer.Close: %v", err) + } + + r, err := NewReader(buf) + if err != nil { + t.Fatalf("NewReader: %v", err) + } + b, err := ioutil.ReadAll(r) + if err != nil { + t.Fatalf("ReadAll: %v", err) + } + if len(b) != 0 { + t.Fatalf("got %d bytes, want 0", len(b)) + } + if err := r.Close(); err != nil { + t.Fatalf("Reader.Close: %v", err) + } +} + +// TestRoundTrip tests that gzipping and then gunzipping is the identity +// function. +func TestRoundTrip(t *testing.T) { + buf := new(bytes.Buffer) + + w := NewWriter(buf) + w.Comment = "comment" + w.Extra = []byte("extra") + w.ModTime = time.Unix(1e8, 0) + w.Name = "name" + if _, err := w.Write([]byte("payload")); err != nil { + t.Fatalf("Write: %v", err) + } + if err := w.Close(); err != nil { + t.Fatalf("Writer.Close: %v", err) + } + + r, err := NewReader(buf) + if err != nil { + t.Fatalf("NewReader: %v", err) + } + b, err := ioutil.ReadAll(r) + if err != nil { + t.Fatalf("ReadAll: %v", err) + } + if string(b) != "payload" { + t.Fatalf("payload is %q, want %q", string(b), "payload") + } + if r.Comment != "comment" { + t.Fatalf("comment is %q, want %q", r.Comment, "comment") + } + if string(r.Extra) != "extra" { + t.Fatalf("extra is %q, want %q", r.Extra, "extra") + } + if r.ModTime.Unix() != 1e8 { + t.Fatalf("mtime is %d, want %d", r.ModTime.Unix(), uint32(1e8)) + } + if r.Name != "name" { + t.Fatalf("name is %q, want %q", r.Name, "name") + } + if err := r.Close(); err != nil { + t.Fatalf("Reader.Close: %v", err) + } +} + +// TestLatin1 tests the internal functions for converting to and from Latin-1. +func TestLatin1(t *testing.T) { + latin1 := []byte{0xc4, 'u', 0xdf, 'e', 'r', 'u', 'n', 'g', 0} + utf8 := "Äußerung" + z := Reader{r: bufio.NewReader(bytes.NewReader(latin1))} + s, err := z.readString() + if err != nil { + t.Fatalf("readString: %v", err) + } + if s != utf8 { + t.Fatalf("read latin-1: got %q, want %q", s, utf8) + } + + buf := bytes.NewBuffer(make([]byte, 0, len(latin1))) + c := Writer{w: buf} + if err = c.writeString(utf8); err != nil { + t.Fatalf("writeString: %v", err) + } + s = buf.String() + if s != string(latin1) { + t.Fatalf("write utf-8: got %q, want %q", s, string(latin1)) + } +} + +// TestLatin1RoundTrip tests that metadata that is representable in Latin-1 +// survives a round trip. +func TestLatin1RoundTrip(t *testing.T) { + testCases := []struct { + name string + ok bool + }{ + {"", true}, + {"ASCII is OK", true}, + {"unless it contains a NUL\x00", false}, + {"no matter where \x00 occurs", false}, + {"\x00\x00\x00", false}, + {"Látin-1 also passes (U+00E1)", true}, + {"but LĀtin Extended-A (U+0100) does not", false}, + {"neither does 日本語", false}, + {"invalid UTF-8 also \xffails", false}, + {"\x00 as does Látin-1 with NUL", false}, + } + for _, tc := range testCases { + buf := new(bytes.Buffer) + + w := NewWriter(buf) + w.Name = tc.name + err := w.Close() + if (err == nil) != tc.ok { + t.Errorf("Writer.Close: name = %q, err = %v", tc.name, err) + continue + } + if !tc.ok { + continue + } + + r, err := NewReader(buf) + if err != nil { + t.Errorf("NewReader: %v", err) + continue + } + _, err = ioutil.ReadAll(r) + if err != nil { + t.Errorf("ReadAll: %v", err) + continue + } + if r.Name != tc.name { + t.Errorf("name is %q, want %q", r.Name, tc.name) + continue + } + if err := r.Close(); err != nil { + t.Errorf("Reader.Close: %v", err) + continue + } + } +} + +func TestWriterFlush(t *testing.T) { + buf := new(bytes.Buffer) + + w := NewWriter(buf) + w.Comment = "comment" + w.Extra = []byte("extra") + w.ModTime = time.Unix(1e8, 0) + w.Name = "name" + + n0 := buf.Len() + if n0 != 0 { + t.Fatalf("buffer size = %d before writes; want 0", n0) + } + + if err := w.Flush(); err != nil { + t.Fatal(err) + } + + n1 := buf.Len() + if n1 == 0 { + t.Fatal("no data after first flush") + } + + w.Write([]byte("x")) + + n2 := buf.Len() + if n1 != n2 { + t.Fatalf("after writing a single byte, size changed from %d to %d; want no change", n1, n2) + } + + if err := w.Flush(); err != nil { + t.Fatal(err) + } + + n3 := buf.Len() + if n2 == n3 { + t.Fatal("Flush didn't flush any data") + } +} + +// Multiple gzip files concatenated form a valid gzip file. +func TestConcat(t *testing.T) { + var buf bytes.Buffer + w := NewWriter(&buf) + w.Write([]byte("hello ")) + w.Close() + w = NewWriter(&buf) + w.Write([]byte("world\n")) + w.Close() + + r, err := NewReader(&buf) + data, err := ioutil.ReadAll(r) + if string(data) != "hello world\n" || err != nil { + t.Fatalf("ReadAll = %q, %v, want %q, nil", data, err, "hello world") + } +} + +func TestWriterReset(t *testing.T) { + buf := new(bytes.Buffer) + buf2 := new(bytes.Buffer) + z := NewWriter(buf) + msg := []byte("hello world") + z.Write(msg) + z.Close() + z.Reset(buf2) + z.Write(msg) + z.Close() + if buf.String() != buf2.String() { + t.Errorf("buf2 %q != original buf of %q", buf2.String(), buf.String()) + } +} + +var testbuf []byte + +func testFile(i, level int, t *testing.T) { + dat, _ := ioutil.ReadFile("testdata/test.json") + dl := len(dat) + if len(testbuf) != i*dl { + // Make results predictable + testbuf = make([]byte, i*dl) + for j := 0; j < i; j++ { + copy(testbuf[j*dl:j*dl+dl], dat) + } + } + + br := bytes.NewBuffer(testbuf) + var buf bytes.Buffer + w, err := NewWriterLevel(&buf, DefaultCompression) + if err != nil { + t.Fatal(err) + } + n, err := io.Copy(w, br) + if err != nil { + t.Fatal(err) + } + if int(n) != len(testbuf) { + t.Fatal("Short write:", n, "!=", testbuf) + } + err = w.Close() + if err != nil { + t.Fatal(err) + } + r, err := NewReader(&buf) + if err != nil { + t.Fatal(err.Error()) + } + decoded, err := ioutil.ReadAll(r) + if err != nil { + t.Fatal(err.Error()) + } + if !bytes.Equal(testbuf, decoded) { + t.Errorf("decoded content does not match.") + } +} + +func TestFile1xM3(t *testing.T) { testFile(1, -3, t) } +func TestFile1xM2(t *testing.T) { testFile(1, -2, t) } +func TestFile1xM1(t *testing.T) { testFile(1, -1, t) } +func TestFile1x0(t *testing.T) { testFile(1, 0, t) } +func TestFile1x1(t *testing.T) { testFile(1, 1, t) } +func TestFile1x2(t *testing.T) { testFile(1, 2, t) } +func TestFile1x3(t *testing.T) { testFile(1, 3, t) } +func TestFile1x4(t *testing.T) { testFile(1, 4, t) } +func TestFile1x5(t *testing.T) { testFile(1, 5, t) } +func TestFile1x6(t *testing.T) { testFile(1, 6, t) } +func TestFile1x7(t *testing.T) { testFile(1, 7, t) } +func TestFile1x8(t *testing.T) { testFile(1, 8, t) } +func TestFile1x9(t *testing.T) { testFile(1, 9, t) } +func TestFile10(t *testing.T) { testFile(10, DefaultCompression, t) } + +func TestFile50(t *testing.T) { + if testing.Short() { + t.Skip("skipping during short test") + } + testFile(50, DefaultCompression, t) +} + +func TestFile200(t *testing.T) { + if testing.Short() { + t.Skip("skipping during short test") + } + testFile(200, BestSpeed, t) +} + +func testBigGzip(i int, t *testing.T) { + if len(testbuf) != i { + // Make results predictable + rand.Seed(1337) + testbuf = make([]byte, i) + for idx := range testbuf { + testbuf[idx] = byte(65 + rand.Intn(20)) + } + } + c := BestCompression + if testing.Short() { + c = BestSpeed + } + + br := bytes.NewBuffer(testbuf) + var buf bytes.Buffer + w, err := NewWriterLevel(&buf, c) + if err != nil { + t.Fatal(err) + } + n, err := io.Copy(w, br) + if err != nil { + t.Fatal(err) + } + if int(n) != len(testbuf) { + t.Fatal("Short write:", n, "!=", len(testbuf)) + } + err = w.Close() + if err != nil { + t.Fatal(err.Error()) + } + + r, err := NewReader(&buf) + if err != nil { + t.Fatal(err.Error()) + } + decoded, err := ioutil.ReadAll(r) + if err != nil { + t.Fatal(err.Error()) + } + if !bytes.Equal(testbuf, decoded) { + t.Errorf("decoded content does not match.") + } +} + +func TestGzip1K(t *testing.T) { testBigGzip(1000, t) } +func TestGzip100K(t *testing.T) { testBigGzip(100000, t) } +func TestGzip1M(t *testing.T) { + if testing.Short() { + t.Skip("skipping during short test") + } + + testBigGzip(1000000, t) +} +func TestGzip10M(t *testing.T) { + if testing.Short() { + t.Skip("skipping during short test") + } + testBigGzip(10000000, t) +} + +// Test if two runs produce identical results. +func TestDeterministicLM2(t *testing.T) { testDeterm(-2, t) } + +// Level 0 is not deterministic since it depends on the size of each write. +// func TestDeterministicL0(t *testing.T) { testDeterm(0, t) } +func TestDeterministicL1(t *testing.T) { testDeterm(1, t) } +func TestDeterministicL2(t *testing.T) { testDeterm(2, t) } +func TestDeterministicL3(t *testing.T) { testDeterm(3, t) } +func TestDeterministicL4(t *testing.T) { testDeterm(4, t) } +func TestDeterministicL5(t *testing.T) { testDeterm(5, t) } +func TestDeterministicL6(t *testing.T) { testDeterm(6, t) } +func TestDeterministicL7(t *testing.T) { testDeterm(7, t) } +func TestDeterministicL8(t *testing.T) { testDeterm(8, t) } +func TestDeterministicL9(t *testing.T) { testDeterm(9, t) } + +func testDeterm(i int, t *testing.T) { + var length = 500000 + if testing.Short() { + length = 100000 + } + rand.Seed(1337) + t1 := make([]byte, length) + for idx := range t1 { + t1[idx] = byte(65 + rand.Intn(8)) + } + + br := bytes.NewBuffer(t1) + var b1 bytes.Buffer + w, err := NewWriterLevel(&b1, i) + if err != nil { + t.Fatal(err) + } + _, err = io.Copy(w, br) + if err != nil { + t.Fatal(err) + } + w.Flush() + w.Close() + + // We recreate the buffer, so we have a goos chance of getting a + // different memory address. + rand.Seed(1337) + t2 := make([]byte, length) + for idx := range t2 { + t2[idx] = byte(65 + rand.Intn(8)) + } + + br2 := bytes.NewBuffer(t2) + var b2 bytes.Buffer + w2, err := NewWriterLevel(&b2, i) + if err != nil { + t.Fatal(err) + } + + // We write the same data, but with a different size than + // the default copy. + for { + _, err = io.CopyN(w2, br2, 1234) + if err == io.EOF { + err = nil + break + } else if err != nil { + break + } + } + if err != nil { + t.Fatal(err) + } + w2.Flush() + w2.Close() + + b1b := b1.Bytes() + b2b := b2.Bytes() + + if bytes.Compare(b1b, b2b) != 0 { + t.Fatalf("Level %d did not produce deterministric result, len(a) = %d, len(b) = %d", i, len(b1b), len(b2b)) + } +} + +func BenchmarkGzipLM3(b *testing.B) { benchmarkGzipN(b, -3) } +func BenchmarkGzipLM2(b *testing.B) { benchmarkGzipN(b, -2) } +func BenchmarkGzipL1(b *testing.B) { benchmarkGzipN(b, 1) } +func BenchmarkGzipL2(b *testing.B) { benchmarkGzipN(b, 2) } +func BenchmarkGzipL3(b *testing.B) { benchmarkGzipN(b, 3) } +func BenchmarkGzipL4(b *testing.B) { benchmarkGzipN(b, 4) } +func BenchmarkGzipL5(b *testing.B) { benchmarkGzipN(b, 5) } +func BenchmarkGzipL6(b *testing.B) { benchmarkGzipN(b, 6) } +func BenchmarkGzipL7(b *testing.B) { benchmarkGzipN(b, 7) } +func BenchmarkGzipL8(b *testing.B) { benchmarkGzipN(b, 8) } +func BenchmarkGzipL9(b *testing.B) { benchmarkGzipN(b, 9) } + +func benchmarkGzipN(b *testing.B, level int) { + dat, _ := ioutil.ReadFile("testdata/test.json") + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + b.SetBytes(int64(len(dat))) + w, _ := NewWriterLevel(ioutil.Discard, level) + b.ResetTimer() + for n := 0; n < b.N; n++ { + w.Reset(ioutil.Discard) + n, err := w.Write(dat) + if n != len(dat) { + panic("short write") + } + if err != nil { + panic(err) + } + err = w.Close() + if err != nil { + panic(err) + } + } +} + +func BenchmarkOldGzipL1(b *testing.B) { benchmarkOldGzipN(b, 1) } +func BenchmarkOldGzipL2(b *testing.B) { benchmarkOldGzipN(b, 2) } +func BenchmarkOldGzipL3(b *testing.B) { benchmarkOldGzipN(b, 3) } +func BenchmarkOldGzipL4(b *testing.B) { benchmarkOldGzipN(b, 4) } +func BenchmarkOldGzipL5(b *testing.B) { benchmarkOldGzipN(b, 5) } +func BenchmarkOldGzipL6(b *testing.B) { benchmarkOldGzipN(b, 6) } +func BenchmarkOldGzipL7(b *testing.B) { benchmarkOldGzipN(b, 7) } +func BenchmarkOldGzipL8(b *testing.B) { benchmarkOldGzipN(b, 8) } +func BenchmarkOldGzipL9(b *testing.B) { benchmarkOldGzipN(b, 9) } + +func benchmarkOldGzipN(b *testing.B, level int) { + dat, _ := ioutil.ReadFile("testdata/test.json") + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + dat = append(dat, dat...) + + b.SetBytes(int64(len(dat))) + w, _ := oldgz.NewWriterLevel(ioutil.Discard, level) + b.ResetTimer() + for n := 0; n < b.N; n++ { + w.Reset(ioutil.Discard) + n, err := w.Write(dat) + if n != len(dat) { + panic("short write") + } + if err != nil { + panic(err) + } + err = w.Close() + if err != nil { + panic(err) + } + } +} diff --git a/github.com/klauspost/compress/gzip/testdata/issue6550.gz b/github.com/klauspost/compress/gzip/testdata/issue6550.gz new file mode 100644 index 0000000..57972b6 Binary files /dev/null and b/github.com/klauspost/compress/gzip/testdata/issue6550.gz differ diff --git a/github.com/klauspost/compress/gzip/testdata/test.json b/github.com/klauspost/compress/gzip/testdata/test.json new file mode 100644 index 0000000..3b7b678 --- /dev/null +++ b/github.com/klauspost/compress/gzip/testdata/test.json @@ -0,0 +1,5902 @@ +[ + { + "_id": "543fa821aeca0fed7f182f01", + "index": 0, + "guid": "3526d142-6d2b-4266-9855-e6ec1589a265", + "isActive": false, + "balance": "$2,156.72", + "picture": "http://placehold.it/32x32", + "age": 29, + "eyeColor": "brown", + "name": { + "first": "Rosella", + "last": "Hale" + }, + "company": "SKINSERVE", + "email": "rosella.hale@skinserve.net", + "phone": "+1 (920) 528-2959", + "address": "324 Imlay Street, Sehili, Guam, 3022", + "about": "Est consectetur ut incididunt commodo elit cillum incididunt consectetur id officia pariatur pariatur cillum. Ipsum non incididunt tempor non. Cillum aliquip aliquip non minim ipsum voluptate incididunt adipisicing aute pariatur laborum minim deserunt laborum. Do do consequat enim adipisicing dolor incididunt reprehenderit sint. Veniam dolor consequat sint ullamco id enim occaecat.\r\n", + "registered": "Wednesday, August 27, 2014 9:12 PM", + "latitude": 43.44586, + "longitude": -65.480986, + "tags": [ + "Lorem", + "ex", + "magna", + "aliqua", + "id", + "sint", + "elit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Etta Stanton" + }, + { + "id": 1, + "name": "Cora Velazquez" + }, + { + "id": 2, + "name": "Deann Guy" + } + ], + "greeting": "Hello, Rosella! You have 6 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8218066e8499ef38bcc", + "index": 1, + "guid": "991a35b5-91db-49e8-8a1e-13688b5ed88d", + "isActive": true, + "balance": "$1,762.71", + "picture": "http://placehold.it/32x32", + "age": 28, + "eyeColor": "green", + "name": { + "first": "Rose", + "last": "Lynn" + }, + "company": "NITRACYR", + "email": "rose.lynn@nitracyr.com", + "phone": "+1 (912) 564-2131", + "address": "485 Pulaski Street, Logan, Mississippi, 7453", + "about": "Minim proident enim eiusmod reprehenderit excepteur laboris. Adipisicing culpa cupidatat eiusmod exercitation reprehenderit anim. Nostrud mollit reprehenderit reprehenderit id magna et id esse cillum et proident. Incididunt eu nisi excepteur est est irure voluptate id nulla. Laboris consectetur aliqua cupidatat ex elit proident officia ex quis. Minim officia eu eiusmod velit. Ullamco dolor non quis aliqua cupidatat amet laborum laborum ad ex proident qui eiusmod ea.\r\n", + "registered": "Sunday, October 5, 2014 10:36 PM", + "latitude": -3.548698, + "longitude": 79.421107, + "tags": [ + "exercitation", + "adipisicing", + "aliqua", + "do", + "id", + "veniam", + "est" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Ada Little" + }, + { + "id": 1, + "name": "Lopez Osborne" + }, + { + "id": 2, + "name": "Tami Leach" + } + ], + "greeting": "Hello, Rose! You have 5 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821255974bb9f89e5ea", + "index": 2, + "guid": "e5727238-63a4-4e1e-88cc-67300826259c", + "isActive": false, + "balance": "$2,131.97", + "picture": "http://placehold.it/32x32", + "age": 21, + "eyeColor": "green", + "name": { + "first": "Gloria", + "last": "Richards" + }, + "company": "SPHERIX", + "email": "gloria.richards@spherix.biz", + "phone": "+1 (884) 536-3434", + "address": "493 Judge Street, Cetronia, Rhode Island, 4439", + "about": "Lorem cupidatat ea et laboris tempor enim non. Sit consequat culpa et qui aute cillum ut ullamco. Nulla duis sit Lorem incididunt mollit nostrud dolor veniam ullamco. Sunt magna id velit in laborum nisi labore. Id deserunt labore dolore dolor aliqua culpa est id duis.\r\n", + "registered": "Saturday, March 29, 2014 8:18 AM", + "latitude": 60.328012, + "longitude": 126.657357, + "tags": [ + "dolore", + "laboris", + "proident", + "cillum", + "in", + "fugiat", + "incididunt" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bowen Cote" + }, + { + "id": 1, + "name": "Olga Gardner" + }, + { + "id": 2, + "name": "Evangeline Howard" + } + ], + "greeting": "Hello, Gloria! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa8212b7e1e8201a38702", + "index": 3, + "guid": "bab757bd-2ebd-4c2c-86b7-0d4d8b059d35", + "isActive": true, + "balance": "$2,509.81", + "picture": "http://placehold.it/32x32", + "age": 39, + "eyeColor": "green", + "name": { + "first": "Casey", + "last": "Hayes" + }, + "company": "SURELOGIC", + "email": "casey.hayes@surelogic.co.uk", + "phone": "+1 (993) 573-3937", + "address": "330 Tapscott Avenue, Eastvale, New Mexico, 928", + "about": "Eu elit sint sunt labore dolor cillum esse ad voluptate commodo. Dolor aliqua do dolore ex tempor sint consequat culpa et consectetur nisi voluptate reprehenderit. Dolor velit eu cillum tempor anim anim. Nostrud laboris eiusmod elit enim duis in consectetur esse anim qui. Et eiusmod culpa nulla anim et officia pariatur reprehenderit eiusmod veniam. Ullamco nisi ea incididunt velit. Ullamco cillum mollit ea aliqua ea eu et enim.\r\n", + "registered": "Sunday, September 14, 2014 8:35 AM", + "latitude": -43.494604, + "longitude": 95.217518, + "tags": [ + "officia", + "sunt", + "dolore", + "qui", + "elit", + "irure", + "cillum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Serrano Wise" + }, + { + "id": 1, + "name": "Lorene Macias" + }, + { + "id": 2, + "name": "Kristen Lott" + } + ], + "greeting": "Hello, Casey! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821bfefa43403d5d054", + "index": 4, + "guid": "675d1598-8c45-4d67-a4df-d38a270de371", + "isActive": false, + "balance": "$3,887.07", + "picture": "http://placehold.it/32x32", + "age": 35, + "eyeColor": "blue", + "name": { + "first": "Price", + "last": "Oconnor" + }, + "company": "ANOCHA", + "email": "price.oconnor@anocha.tv", + "phone": "+1 (855) 410-3197", + "address": "447 Stockholm Street, Templeton, Wisconsin, 2216", + "about": "Cillum veniam esse duis tempor incididunt do dolor officia elit eu. Excepteur velit reprehenderit minim Lorem commodo est. Duis Lorem nisi elit aliquip est deserunt fugiat ut. Nisi tempor ex est pariatur laborum eiusmod anim eu nulla. Nisi enim id aute id ex id nostrud.\r\n", + "registered": "Wednesday, May 14, 2014 5:19 PM", + "latitude": 26.083477, + "longitude": 122.61114, + "tags": [ + "in", + "ad", + "aliqua", + "minim", + "nisi", + "cupidatat", + "id" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Montgomery Mccray" + }, + { + "id": 1, + "name": "Lucia Ferrell" + }, + { + "id": 2, + "name": "Glover Brock" + } + ], + "greeting": "Hello, Price! You have 5 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821a699260d8ed4439a", + "index": 5, + "guid": "5e271270-fef3-48a7-b389-346251b46abc", + "isActive": false, + "balance": "$1,046.50", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "blue", + "name": { + "first": "Rita", + "last": "Huber" + }, + "company": "VURBO", + "email": "rita.huber@vurbo.name", + "phone": "+1 (803) 589-3948", + "address": "838 River Street, Gadsden, American Samoa, 2602", + "about": "Culpa quis qui exercitation velit officia eu id qui consequat qui. Ea fugiat quis fugiat proident velit. Velit et reprehenderit quis irure adipisicing duis dolor id cupidatat ea aliqua elit.\r\n", + "registered": "Friday, April 11, 2014 11:56 AM", + "latitude": 30.717665, + "longitude": -29.687902, + "tags": [ + "veniam", + "ex", + "deserunt", + "cillum", + "sint", + "eu", + "proident" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Oliver Terrell" + }, + { + "id": 1, + "name": "Lora Shepherd" + }, + { + "id": 2, + "name": "Guzman Holman" + } + ], + "greeting": "Hello, Rita! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821b888230e87ce950e", + "index": 6, + "guid": "7e4efd9a-4923-42ae-8924-d6d5fae80ec0", + "isActive": false, + "balance": "$1,205.59", + "picture": "http://placehold.it/32x32", + "age": 30, + "eyeColor": "green", + "name": { + "first": "Peterson", + "last": "Oliver" + }, + "company": "ZENTIX", + "email": "peterson.oliver@zentix.me", + "phone": "+1 (924) 564-2815", + "address": "596 Middleton Street, Walker, Louisiana, 3358", + "about": "Exercitation cillum sit exercitation voluptate duis nostrud incididunt cillum sint minim labore tempor minim ad. Esse ad id pariatur cillum id exercitation ullamco elit. Quis nisi excepteur mollit consectetur id et. Ea voluptate nulla duis minim exercitation aliqua aute nisi enim enim excepteur dolor ad non. Aliquip elit eu enim officia minim enim Lorem tempor. Cillum anim aute sunt cupidatat deserunt consequat.\r\n", + "registered": "Friday, March 28, 2014 2:28 PM", + "latitude": 45.092029, + "longitude": 56.730029, + "tags": [ + "in", + "voluptate", + "sit", + "sit", + "Lorem", + "reprehenderit", + "esse" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Wiley Henry" + }, + { + "id": 1, + "name": "Downs Rowland" + }, + { + "id": 2, + "name": "White Guerra" + } + ], + "greeting": "Hello, Peterson! You have 8 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8210f8589ab846b3d88", + "index": 7, + "guid": "afc25e30-7e70-4a87-bdd3-519e1837969a", + "isActive": false, + "balance": "$3,928.85", + "picture": "http://placehold.it/32x32", + "age": 39, + "eyeColor": "green", + "name": { + "first": "Shauna", + "last": "Morse" + }, + "company": "CENTICE", + "email": "shauna.morse@centice.us", + "phone": "+1 (926) 517-3679", + "address": "752 Dunne Place, Ebro, Kansas, 3215", + "about": "Cupidatat incididunt sit duis tempor labore dolore aute qui magna in. Consequat aute ut veniam laborum aliqua Lorem esse. Cillum in qui sint excepteur eiusmod eiusmod eu anim adipisicing et.\r\n", + "registered": "Saturday, September 6, 2014 2:32 PM", + "latitude": 36.341849, + "longitude": 108.378341, + "tags": [ + "in", + "nulla", + "labore", + "qui", + "id", + "enim", + "fugiat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Lizzie Carson" + }, + { + "id": 1, + "name": "Eliza Hall" + }, + { + "id": 2, + "name": "Baxter Burton" + } + ], + "greeting": "Hello, Shauna! You have 6 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8213c997bd81d4a7fa5", + "index": 8, + "guid": "1337ad27-17e3-459f-90a3-a43b54b88184", + "isActive": true, + "balance": "$2,096.67", + "picture": "http://placehold.it/32x32", + "age": 24, + "eyeColor": "blue", + "name": { + "first": "Glenn", + "last": "Brooks" + }, + "company": "MANGLO", + "email": "glenn.brooks@manglo.ca", + "phone": "+1 (895) 595-2669", + "address": "605 McDonald Avenue, Nicholson, Indiana, 2302", + "about": "Deserunt incididunt ullamco dolore nostrud cupidatat sit consequat adipisicing incididunt sunt. Laboris fugiat et laboris est eu laborum culpa. Labore ad aliquip ut enim aute nulla quis cillum dolor aliqua. Culpa labore occaecat et sunt qui. Velit consequat ad proident non voluptate non mollit eu et cillum tempor. Velit quis deserunt Lorem cupidatat enim ut.\r\n", + "registered": "Friday, March 21, 2014 9:06 AM", + "latitude": -40.51084, + "longitude": -137.771438, + "tags": [ + "enim", + "laboris", + "culpa", + "do", + "nulla", + "anim", + "cillum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Shelly Cardenas" + }, + { + "id": 1, + "name": "Kristine Mendoza" + }, + { + "id": 2, + "name": "Hall Hendrix" + } + ], + "greeting": "Hello, Glenn! You have 10 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821054bcf388259b272", + "index": 9, + "guid": "cd9601fc-7ca7-4d54-830b-145ff5c5c147", + "isActive": true, + "balance": "$1,816.14", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "green", + "name": { + "first": "Maribel", + "last": "Small" + }, + "company": "FUTURITY", + "email": "maribel.small@futurity.org", + "phone": "+1 (825) 532-2134", + "address": "424 Rockaway Parkway, Vale, Alaska, 1834", + "about": "Aliqua irure culpa exercitation nostrud qui exercitation deserunt ullamco culpa aliquip irure. Proident officia in consequat laborum ex adipisicing exercitation proident anim cupidatat excepteur anim. Labore irure pariatur laboris reprehenderit.\r\n", + "registered": "Tuesday, July 29, 2014 9:59 PM", + "latitude": 53.843872, + "longitude": 85.292318, + "tags": [ + "dolore", + "laborum", + "aute", + "aliqua", + "nostrud", + "commodo", + "commodo" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Anthony Bray" + }, + { + "id": 1, + "name": "Vicki Kelly" + }, + { + "id": 2, + "name": "Baird Wagner" + } + ], + "greeting": "Hello, Maribel! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821f14efb6f7f1210b9", + "index": 10, + "guid": "c9400d51-ea8d-4748-9a10-fa0e3a037b23", + "isActive": false, + "balance": "$1,395.15", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "green", + "name": { + "first": "Kendra", + "last": "Knapp" + }, + "company": "UNCORP", + "email": "kendra.knapp@uncorp.biz", + "phone": "+1 (830) 509-3054", + "address": "765 Cameron Court, Ferney, Florida, 1963", + "about": "Duis irure ea qui ut velit nostrud. Lorem laborum excepteur do qui ad sit culpa. Labore mollit mollit deserunt sint aute officia qui laboris dolor aliqua magna in officia.\r\n", + "registered": "Sunday, April 27, 2014 11:55 AM", + "latitude": 84.200593, + "longitude": -155.377179, + "tags": [ + "laborum", + "labore", + "aliquip", + "ex", + "voluptate", + "dolor", + "Lorem" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Marva Cash" + }, + { + "id": 1, + "name": "Aimee Velez" + }, + { + "id": 2, + "name": "Eaton Delgado" + } + ], + "greeting": "Hello, Kendra! You have 7 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa82119105322ba52b407", + "index": 11, + "guid": "b6862230-95da-43d5-97ba-13ed4bcc0744", + "isActive": false, + "balance": "$1,442.51", + "picture": "http://placehold.it/32x32", + "age": 21, + "eyeColor": "brown", + "name": { + "first": "Katrina", + "last": "Ferguson" + }, + "company": "MANTRIX", + "email": "katrina.ferguson@mantrix.io", + "phone": "+1 (938) 541-3037", + "address": "447 Putnam Avenue, Collins, Georgia, 8421", + "about": "Minim aliquip Lorem fugiat et fugiat esse aliqua consectetur non officia esse. Fugiat irure eu ut irure cillum mollit nisi consequat do cillum. Est exercitation deserunt proident ex cupidatat. Elit aliquip pariatur ad minim adipisicing qui. Quis enim laborum incididunt eiusmod deserunt cillum amet enim. Proident et do voluptate esse laboris nisi. Duis cupidatat fugiat adipisicing aute velit et ullamco anim velit velit et excepteur laboris.\r\n", + "registered": "Tuesday, February 4, 2014 5:01 PM", + "latitude": 43.287084, + "longitude": 133.518964, + "tags": [ + "magna", + "officia", + "reprehenderit", + "excepteur", + "cillum", + "veniam", + "officia" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Tameka Mccullough" + }, + { + "id": 1, + "name": "Madden Vincent" + }, + { + "id": 2, + "name": "Jewel Mccarthy" + } + ], + "greeting": "Hello, Katrina! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821ee55902ac207b17a", + "index": 12, + "guid": "50161207-4477-47b9-8aa7-a845c3c2e96f", + "isActive": false, + "balance": "$1,937.64", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "brown", + "name": { + "first": "Gilliam", + "last": "Flowers" + }, + "company": "INTERLOO", + "email": "gilliam.flowers@interloo.net", + "phone": "+1 (930) 564-2474", + "address": "986 Elton Street, Bagtown, Alabama, 9115", + "about": "Velit incididunt ut nulla adipisicing ad qui sint dolor cillum cupidatat in. Commodo aliqua deserunt ea eu irure irure nisi ullamco culpa nostrud. Adipisicing exercitation excepteur et id cupidatat. Ullamco ut incididunt proident est ad deserunt duis id ut. Excepteur cupidatat irure reprehenderit et excepteur minim cillum occaecat adipisicing. Commodo fugiat ad ex consectetur commodo dolore id nisi deserunt commodo aliquip. Veniam amet mollit nulla adipisicing eu minim sit magna incididunt adipisicing.\r\n", + "registered": "Tuesday, April 1, 2014 12:42 AM", + "latitude": -55.19047, + "longitude": 177.975351, + "tags": [ + "sint", + "pariatur", + "incididunt", + "exercitation", + "quis", + "ad", + "sint" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Patsy Hunter" + }, + { + "id": 1, + "name": "Cecilia Green" + }, + { + "id": 2, + "name": "Meyer Jones" + } + ], + "greeting": "Hello, Gilliam! You have 9 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821658f3962ce822678", + "index": 13, + "guid": "7017674f-79b3-43ed-8832-2b787c51f59d", + "isActive": false, + "balance": "$2,291.31", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "blue", + "name": { + "first": "Roberts", + "last": "Floyd" + }, + "company": "OVOLO", + "email": "roberts.floyd@ovolo.com", + "phone": "+1 (935) 401-2916", + "address": "723 Amherst Street, Brady, District Of Columbia, 4241", + "about": "Occaecat incididunt eu do quis est. Est mollit incididunt sint aute sunt. Consectetur incididunt officia eu fugiat quis officia pariatur excepteur sint. In enim nostrud nisi culpa. Ex incididunt exercitation id voluptate.\r\n", + "registered": "Thursday, June 19, 2014 4:16 PM", + "latitude": 72.321258, + "longitude": 28.548926, + "tags": [ + "et", + "ipsum", + "anim", + "dolor", + "commodo", + "do", + "exercitation" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Tracey Vasquez" + }, + { + "id": 1, + "name": "Castro Harrell" + }, + { + "id": 2, + "name": "Sanders Barr" + } + ], + "greeting": "Hello, Roberts! You have 10 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821f95ca5383b9eb53d", + "index": 14, + "guid": "a6532d9a-d291-4a51-92e7-33c50ceecc12", + "isActive": true, + "balance": "$3,310.31", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "green", + "name": { + "first": "Miles", + "last": "Valdez" + }, + "company": "DENTREX", + "email": "miles.valdez@dentrex.biz", + "phone": "+1 (960) 513-3228", + "address": "726 Stillwell Place, Soham, Pennsylvania, 3510", + "about": "Sit labore ex commodo duis tempor labore officia et et est qui ullamco. Aute elit in labore laboris magna duis ipsum excepteur anim laboris ipsum magna magna non. Sint mollit eiusmod in est sint ipsum excepteur do anim cillum cillum.\r\n", + "registered": "Thursday, May 1, 2014 6:08 PM", + "latitude": 88.123309, + "longitude": -121.226418, + "tags": [ + "voluptate", + "sunt", + "anim", + "laboris", + "exercitation", + "deserunt", + "culpa" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bradford Horn" + }, + { + "id": 1, + "name": "Hanson Dillon" + }, + { + "id": 2, + "name": "Whitley Stanley" + } + ], + "greeting": "Hello, Miles! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821c52f80cda8ff36b3", + "index": 15, + "guid": "63820e33-a8c7-410e-afa9-32b7f7017a32", + "isActive": true, + "balance": "$2,616.09", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "brown", + "name": { + "first": "Floyd", + "last": "Barker" + }, + "company": "TALAE", + "email": "floyd.barker@talae.co.uk", + "phone": "+1 (843) 435-2898", + "address": "501 Sullivan Place, Cotopaxi, Nevada, 5498", + "about": "Non deserunt voluptate occaecat est mollit dolor aliqua. Qui elit aute qui aliquip ipsum et labore est aliquip pariatur. Sint deserunt tempor dolore excepteur elit est sint in est ex anim. Nostrud culpa amet eiusmod incididunt. Ea exercitation amet labore cillum culpa duis aute incididunt dolore sunt. Cillum velit laboris quis eiusmod fugiat consectetur sit fugiat irure labore.\r\n", + "registered": "Monday, March 10, 2014 7:48 AM", + "latitude": -86.397923, + "longitude": 171.646534, + "tags": [ + "dolore", + "sit", + "qui", + "id", + "aliquip", + "mollit", + "laborum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Eileen Daniels" + }, + { + "id": 1, + "name": "Genevieve Wood" + }, + { + "id": 2, + "name": "Carver Fields" + } + ], + "greeting": "Hello, Floyd! You have 6 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8219eb621c418384935", + "index": 16, + "guid": "ed67eac2-cfdf-4d28-a734-bc973cba8613", + "isActive": false, + "balance": "$1,917.58", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "green", + "name": { + "first": "Carrillo", + "last": "Cox" + }, + "company": "ARCHITAX", + "email": "carrillo.cox@architax.tv", + "phone": "+1 (818) 444-3875", + "address": "309 Randolph Street, Avoca, Illinois, 770", + "about": "Labore consequat et nostrud officia ad. Sint ipsum ipsum sint laboris adipisicing minim voluptate aliqua proident est commodo nulla. Officia sint ipsum laborum aliquip adipisicing adipisicing ea et reprehenderit dolore. Et deserunt sint incididunt velit dolore voluptate deserunt anim nisi sit est officia fugiat. Velit dolore ea do enim veniam ut do. Duis adipisicing fugiat magna Lorem ullamco quis sint ut cupidatat laborum aute laboris sint aliqua.\r\n", + "registered": "Friday, January 17, 2014 12:19 AM", + "latitude": -31.228015, + "longitude": -82.248255, + "tags": [ + "occaecat", + "nostrud", + "ex", + "dolor", + "magna", + "minim", + "pariatur" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Douglas Mayer" + }, + { + "id": 1, + "name": "Dorothy Riddle" + }, + { + "id": 2, + "name": "Melanie Thompson" + } + ], + "greeting": "Hello, Carrillo! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821e152869356625777", + "index": 17, + "guid": "1dc17af3-a194-42e8-add8-a73853f16da2", + "isActive": true, + "balance": "$1,703.08", + "picture": "http://placehold.it/32x32", + "age": 36, + "eyeColor": "green", + "name": { + "first": "Ana", + "last": "Reese" + }, + "company": "FORTEAN", + "email": "ana.reese@fortean.name", + "phone": "+1 (876) 419-2128", + "address": "451 Brevoort Place, Leola, Tennessee, 3725", + "about": "Consectetur officia irure proident nulla. Anim veniam mollit sit id aliqua. Do reprehenderit culpa magna magna aute est pariatur consequat ut occaecat cillum adipisicing consectetur. Sint qui pariatur id velit deserunt laborum. Minim consequat ut sunt qui. Ex occaecat tempor fugiat sit anim veniam incididunt mollit mollit. Non id anim cillum culpa tempor voluptate aute consequat proident reprehenderit.\r\n", + "registered": "Saturday, June 28, 2014 4:53 AM", + "latitude": 80.18306, + "longitude": 70.818006, + "tags": [ + "mollit", + "voluptate", + "est", + "magna", + "ad", + "duis", + "est" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mclaughlin Johns" + }, + { + "id": 1, + "name": "Leanne Hanson" + }, + { + "id": 2, + "name": "Isabel Leon" + } + ], + "greeting": "Hello, Ana! You have 7 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa82168ed66f19f510aea", + "index": 18, + "guid": "87be73dd-bf5e-48c4-8168-f8f76cda905d", + "isActive": true, + "balance": "$1,307.88", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "green", + "name": { + "first": "Daugherty", + "last": "Ware" + }, + "company": "TYPHONICA", + "email": "daugherty.ware@typhonica.me", + "phone": "+1 (936) 470-3445", + "address": "919 Oriental Boulevard, Westboro, Iowa, 2587", + "about": "Occaecat in nisi et consequat. Laboris minim consequat qui proident id aute occaecat pariatur. Sint esse anim id ex voluptate fugiat culpa anim commodo incididunt.\r\n", + "registered": "Tuesday, March 4, 2014 11:53 PM", + "latitude": -15.007384, + "longitude": -86.496257, + "tags": [ + "consequat", + "nisi", + "duis", + "cupidatat", + "anim", + "eu", + "culpa" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Maddox Wells" + }, + { + "id": 1, + "name": "Maura May" + }, + { + "id": 2, + "name": "Terry Calhoun" + } + ], + "greeting": "Hello, Daugherty! You have 10 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8210abc44da895c5591", + "index": 19, + "guid": "32c4c5d0-b54c-4ea4-999a-f4fa517ac5ce", + "isActive": true, + "balance": "$2,706.98", + "picture": "http://placehold.it/32x32", + "age": 34, + "eyeColor": "green", + "name": { + "first": "Sonja", + "last": "Craft" + }, + "company": "KOZGENE", + "email": "sonja.craft@kozgene.us", + "phone": "+1 (808) 410-3427", + "address": "808 Lawn Court, Blodgett, Massachusetts, 560", + "about": "Eu occaecat reprehenderit ea ad ullamco ea sint cupidatat ex. Deserunt eu est veniam consectetur do anim in. Dolore minim veniam dolore elit sunt labore id eiusmod.\r\n", + "registered": "Sunday, August 31, 2014 12:09 AM", + "latitude": -47.101894, + "longitude": -130.294589, + "tags": [ + "sint", + "cillum", + "magna", + "sit", + "fugiat", + "nisi", + "reprehenderit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Loretta Mcgee" + }, + { + "id": 1, + "name": "Wilson Merritt" + }, + { + "id": 2, + "name": "Susanne Lloyd" + } + ], + "greeting": "Hello, Sonja! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa82134480e13e931193a", + "index": 20, + "guid": "c4a7ded5-aa3e-411f-9956-f4b938ce93dc", + "isActive": false, + "balance": "$3,216.50", + "picture": "http://placehold.it/32x32", + "age": 23, + "eyeColor": "green", + "name": { + "first": "Powers", + "last": "Mathews" + }, + "company": "ANDRYX", + "email": "powers.mathews@andryx.ca", + "phone": "+1 (914) 559-2596", + "address": "545 Nolans Lane, Brenton, Arkansas, 7607", + "about": "In irure et tempor ad commodo culpa reprehenderit excepteur tempor ex. Exercitation eiusmod consequat anim incididunt veniam duis sunt velit sunt aliquip esse adipisicing do. Elit ea incididunt id amet mollit ea in ad ea cupidatat duis minim consectetur incididunt.\r\n", + "registered": "Friday, July 11, 2014 11:27 AM", + "latitude": 64.259332, + "longitude": 111.604942, + "tags": [ + "ut", + "ex", + "cillum", + "commodo", + "pariatur", + "ex", + "reprehenderit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Woodward Witt" + }, + { + "id": 1, + "name": "Hazel Mcfadden" + }, + { + "id": 2, + "name": "Desiree Mclean" + } + ], + "greeting": "Hello, Powers! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821117e0b8bd3f6e566", + "index": 21, + "guid": "fae9ba5c-948b-429c-b1e6-a0a6835f0694", + "isActive": false, + "balance": "$1,046.41", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "brown", + "name": { + "first": "Lola", + "last": "Roy" + }, + "company": "EURON", + "email": "lola.roy@euron.org", + "phone": "+1 (920) 413-2000", + "address": "237 Tabor Court, Berwind, Hawaii, 2276", + "about": "Aute voluptate proident occaecat exercitation aute proident ullamco veniam aute magna velit cupidatat. Occaecat dolor aliquip adipisicing dolor do elit eu elit laboris officia magna dolore. Velit tempor sit ad et occaecat nisi elit excepteur. Non velit sint deserunt culpa magna irure.\r\n", + "registered": "Thursday, February 27, 2014 1:20 AM", + "latitude": 12.90929, + "longitude": 68.693395, + "tags": [ + "ad", + "officia", + "non", + "aute", + "magna", + "minim", + "sint" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Lee Tate" + }, + { + "id": 1, + "name": "Miranda Payne" + }, + { + "id": 2, + "name": "Jocelyn Cantrell" + } + ], + "greeting": "Hello, Lola! You have 9 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8211cf627425d947100", + "index": 22, + "guid": "590c913b-2457-47a5-ad5c-8a5fc3c249f9", + "isActive": true, + "balance": "$2,144.16", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "green", + "name": { + "first": "Marci", + "last": "Mcpherson" + }, + "company": "VIAGREAT", + "email": "marci.mcpherson@viagreat.biz", + "phone": "+1 (916) 417-2166", + "address": "527 Kenilworth Place, Bartonsville, Puerto Rico, 4739", + "about": "Duis velit irure sit sit aliquip sit culpa velit labore velit ipsum amet. Pariatur labore ex et sunt proident ad minim. Aliquip qui adipisicing elit do sunt mollit irure adipisicing in labore cillum. Ut velit dolor cillum irure voluptate ad incididunt consequat cillum esse laborum consequat do.\r\n", + "registered": "Wednesday, August 27, 2014 6:55 AM", + "latitude": 23.135493, + "longitude": -133.213153, + "tags": [ + "ut", + "ex", + "deserunt", + "mollit", + "cillum", + "aliquip", + "excepteur" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Latisha Ortiz" + }, + { + "id": 1, + "name": "Ila Marshall" + }, + { + "id": 2, + "name": "Kathie Strong" + } + ], + "greeting": "Hello, Marci! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa82113ff45aaf64d6b75", + "index": 23, + "guid": "a074b7d2-92de-4728-9032-ac711cc8ca1b", + "isActive": true, + "balance": "$1,927.67", + "picture": "http://placehold.it/32x32", + "age": 26, + "eyeColor": "green", + "name": { + "first": "Lorie", + "last": "Haynes" + }, + "company": "CUBIX", + "email": "lorie.haynes@cubix.io", + "phone": "+1 (914) 479-2574", + "address": "209 Stoddard Place, Grahamtown, New Hampshire, 3422", + "about": "Commodo eu reprehenderit aute veniam occaecat eiusmod ex enim mollit elit. Officia fugiat proident cillum sint sint. In anim occaecat in dolore pariatur occaecat dolore eu duis sint veniam labore tempor id.\r\n", + "registered": "Thursday, June 19, 2014 3:03 AM", + "latitude": -80.694066, + "longitude": 98.315178, + "tags": [ + "proident", + "est", + "nulla", + "minim", + "aute", + "duis", + "ea" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Shana Jensen" + }, + { + "id": 1, + "name": "Audra Hays" + }, + { + "id": 2, + "name": "Shannon Stewart" + } + ], + "greeting": "Hello, Lorie! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8216788a07fe633c5a6", + "index": 24, + "guid": "ff361134-2ce3-4cce-b043-d571e87a041d", + "isActive": false, + "balance": "$1,274.62", + "picture": "http://placehold.it/32x32", + "age": 34, + "eyeColor": "green", + "name": { + "first": "Parker", + "last": "Higgins" + }, + "company": "ISOSTREAM", + "email": "parker.higgins@isostream.net", + "phone": "+1 (965) 467-3975", + "address": "908 Division Place, Homeland, South Carolina, 2577", + "about": "Laborum minim consectetur ipsum incididunt cupidatat ex ad labore eu non est consequat. Tempor eiusmod commodo Lorem enim aliquip ad non sint ipsum culpa amet. Eu sit amet velit est sit cupidatat aliquip magna proident id veniam Lorem dolore. Eiusmod ex amet proident enim ipsum proident mollit adipisicing ut.\r\n", + "registered": "Monday, August 25, 2014 4:51 AM", + "latitude": -28.784274, + "longitude": -151.224185, + "tags": [ + "ea", + "cupidatat", + "do", + "culpa", + "ea", + "ullamco", + "nulla" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bridgette Tyler" + }, + { + "id": 1, + "name": "Harris Pollard" + }, + { + "id": 2, + "name": "Davenport Skinner" + } + ], + "greeting": "Hello, Parker! You have 6 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821ff898d23056717a4", + "index": 25, + "guid": "f5e85a0d-f46e-427e-86a7-657eaaadb169", + "isActive": true, + "balance": "$3,413.58", + "picture": "http://placehold.it/32x32", + "age": 36, + "eyeColor": "brown", + "name": { + "first": "Rios", + "last": "Reilly" + }, + "company": "QUILTIGEN", + "email": "rios.reilly@quiltigen.com", + "phone": "+1 (982) 565-3930", + "address": "789 Beekman Place, Wiscon, Texas, 8745", + "about": "Consectetur qui do sint deserunt voluptate sunt dolor in officia aliquip. Eu irure sit veniam nostrud culpa laboris. Commodo nostrud cillum nulla nostrud.\r\n", + "registered": "Thursday, September 4, 2014 1:54 PM", + "latitude": 6.093115, + "longitude": 145.037939, + "tags": [ + "eu", + "consectetur", + "veniam", + "pariatur", + "laboris", + "ad", + "cupidatat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Sweet Conley" + }, + { + "id": 1, + "name": "Key Grant" + }, + { + "id": 2, + "name": "Guthrie Moss" + } + ], + "greeting": "Hello, Rios! You have 10 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa82125f7c765fcbb1743", + "index": 26, + "guid": "e5f12323-5c7c-4103-b20c-1a633845a28c", + "isActive": true, + "balance": "$1,645.61", + "picture": "http://placehold.it/32x32", + "age": 26, + "eyeColor": "green", + "name": { + "first": "Hurley", + "last": "Cooke" + }, + "company": "QUORDATE", + "email": "hurley.cooke@quordate.biz", + "phone": "+1 (841) 404-3894", + "address": "369 Denton Place, Curtice, South Dakota, 2613", + "about": "Nulla non in aliqua sit mollit pariatur do mollit. Ut pariatur ut velit minim. Fugiat deserunt velit duis consequat labore culpa voluptate sint voluptate consectetur officia voluptate et laborum. Et exercitation ut eu pariatur minim velit elit. Dolore amet officia ipsum voluptate occaecat eiusmod cupidatat do dolore consequat esse consectetur aliquip.\r\n", + "registered": "Sunday, April 13, 2014 11:42 AM", + "latitude": -78.463811, + "longitude": 36.580914, + "tags": [ + "deserunt", + "nisi", + "do", + "enim", + "nisi", + "qui", + "ipsum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Harvey Norman" + }, + { + "id": 1, + "name": "Porter Shannon" + }, + { + "id": 2, + "name": "Reyes Goodman" + } + ], + "greeting": "Hello, Hurley! You have 10 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821f375b9cabd418303", + "index": 27, + "guid": "452fe001-7fff-4f69-9336-00e3e80e9792", + "isActive": true, + "balance": "$2,608.67", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "green", + "name": { + "first": "Jill", + "last": "Blair" + }, + "company": "ORBALIX", + "email": "jill.blair@orbalix.co.uk", + "phone": "+1 (863) 519-2778", + "address": "680 Cleveland Street, Kohatk, Ohio, 1688", + "about": "Do labore sint cupidatat dolor. Mollit nulla voluptate nostrud tempor ad cillum in mollit officia reprehenderit duis commodo veniam ad. Adipisicing enim adipisicing consequat sint minim ut. Cupidatat non ullamco sunt mollit proident. Aliquip dolore dolor excepteur cupidatat. Consectetur duis adipisicing qui enim aute quis veniam deserunt occaecat. Duis elit exercitation ullamco voluptate aliqua.\r\n", + "registered": "Tuesday, September 30, 2014 11:23 PM", + "latitude": -33.279869, + "longitude": 6.221211, + "tags": [ + "nostrud", + "elit", + "adipisicing", + "esse", + "in", + "commodo", + "ea" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Tania Flores" + }, + { + "id": 1, + "name": "Nina Blackburn" + }, + { + "id": 2, + "name": "Mathews Fischer" + } + ], + "greeting": "Hello, Jill! You have 10 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821304372a99199671f", + "index": 28, + "guid": "d160ded3-911c-4ce2-a314-9ed9a8e6fa9b", + "isActive": true, + "balance": "$3,005.54", + "picture": "http://placehold.it/32x32", + "age": 35, + "eyeColor": "brown", + "name": { + "first": "Estela", + "last": "Dalton" + }, + "company": "POWERNET", + "email": "estela.dalton@powernet.tv", + "phone": "+1 (959) 527-2607", + "address": "820 Montauk Avenue, Whitmer, Maine, 3867", + "about": "Commodo est ullamco sit eu irure tempor veniam deserunt in aute cillum tempor. Occaecat velit et deserunt incididunt sint do eu consectetur enim ullamco consectetur esse ipsum pariatur. Tempor exercitation dolore tempor enim. Dolor esse est magna occaecat. Elit culpa sint non ea exercitation. Aliquip nostrud aliquip culpa Lorem cillum incididunt do sit sunt velit id. Proident sit proident est velit consequat cillum officia in et.\r\n", + "registered": "Sunday, April 13, 2014 5:53 AM", + "latitude": 32.713335, + "longitude": 174.505048, + "tags": [ + "aliquip", + "dolore", + "proident", + "pariatur", + "elit", + "cillum", + "id" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Gibson Durham" + }, + { + "id": 1, + "name": "Carolina Cooley" + }, + { + "id": 2, + "name": "Rosa Mcintyre" + } + ], + "greeting": "Hello, Estela! You have 8 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa82187c6143c96c5ce1f", + "index": 29, + "guid": "f62ae8de-8905-4ac0-9b5c-ce12dad96a86", + "isActive": false, + "balance": "$1,859.49", + "picture": "http://placehold.it/32x32", + "age": 34, + "eyeColor": "brown", + "name": { + "first": "Martina", + "last": "Jacobson" + }, + "company": "CUIZINE", + "email": "martina.jacobson@cuizine.name", + "phone": "+1 (927) 493-2997", + "address": "234 Utica Avenue, Hinsdale, Vermont, 459", + "about": "Pariatur nulla ad sint tempor qui in id aliqua ex et ut. Qui occaecat quis veniam mollit officia duis ad ea. Est consectetur sit sint proident sit do. Id ut incididunt tempor id irure. Qui commodo cillum labore anim eiusmod exercitation ea qui nulla qui amet.\r\n", + "registered": "Sunday, February 9, 2014 3:54 AM", + "latitude": -36.70558, + "longitude": -140.397297, + "tags": [ + "voluptate", + "adipisicing", + "do", + "deserunt", + "aliquip", + "est", + "minim" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Griffith Martinez" + }, + { + "id": 1, + "name": "Richard Chavez" + }, + { + "id": 2, + "name": "Mckinney Butler" + } + ], + "greeting": "Hello, Martina! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821392ebf56130b6eaa", + "index": 30, + "guid": "91c5f9c9-d3c7-435a-98cc-06e360c12e1d", + "isActive": true, + "balance": "$3,693.79", + "picture": "http://placehold.it/32x32", + "age": 21, + "eyeColor": "brown", + "name": { + "first": "Althea", + "last": "Valencia" + }, + "company": "BLANET", + "email": "althea.valencia@blanet.me", + "phone": "+1 (887) 501-3212", + "address": "911 Adams Street, Brambleton, Delaware, 5831", + "about": "Quis culpa exercitation dolor anim. Id labore ea aute aliqua. Dolor dolore sit duis anim cillum nostrud officia dolor sit. Laborum tempor dolore id consequat.\r\n", + "registered": "Saturday, January 11, 2014 6:27 PM", + "latitude": -39.101471, + "longitude": -22.991091, + "tags": [ + "eu", + "anim", + "elit", + "pariatur", + "cupidatat", + "cupidatat", + "enim" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Dawson Foreman" + }, + { + "id": 1, + "name": "Sanford Meyer" + }, + { + "id": 2, + "name": "Ruth Barron" + } + ], + "greeting": "Hello, Althea! You have 8 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa82111b129b44454e349", + "index": 31, + "guid": "184ea3dd-af3f-400a-aa64-429b6cac091f", + "isActive": true, + "balance": "$1,351.78", + "picture": "http://placehold.it/32x32", + "age": 20, + "eyeColor": "blue", + "name": { + "first": "Morrison", + "last": "Chan" + }, + "company": "TALKALOT", + "email": "morrison.chan@talkalot.us", + "phone": "+1 (822) 448-3384", + "address": "599 Olive Street, Franklin, Palau, 3303", + "about": "Ex deserunt nulla velit dolore. Sunt sit ea irure incididunt aute sint do veniam. Sit ut ad ipsum est velit ea duis exercitation aliquip consectetur Lorem fugiat eu.\r\n", + "registered": "Sunday, February 16, 2014 2:04 PM", + "latitude": 61.099205, + "longitude": -37.736061, + "tags": [ + "eu", + "deserunt", + "pariatur", + "labore", + "reprehenderit", + "magna", + "consectetur" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Norma Montoya" + }, + { + "id": 1, + "name": "Bailey Gillespie" + }, + { + "id": 2, + "name": "Candace Kent" + } + ], + "greeting": "Hello, Morrison! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821aec56cb0ce50b54b", + "index": 32, + "guid": "b211ab76-8674-4ed0-9a40-2087930468ad", + "isActive": false, + "balance": "$1,492.99", + "picture": "http://placehold.it/32x32", + "age": 34, + "eyeColor": "green", + "name": { + "first": "Walters", + "last": "Potts" + }, + "company": "FLUMBO", + "email": "walters.potts@flumbo.ca", + "phone": "+1 (871) 461-3958", + "address": "438 Highland Avenue, Elwood, Arizona, 9669", + "about": "Lorem do Lorem aliquip ipsum. Elit labore reprehenderit tempor do. Incididunt labore ad eu occaecat enim laborum irure elit nulla Lorem anim sit exercitation velit. Proident ullamco voluptate aute ex et aute mollit nostrud. Adipisicing labore sit irure amet dolore nostrud. Tempor nulla aliqua culpa commodo aliqua ut esse velit mollit ad. Aliqua nulla enim non nisi laboris sint aute duis proident qui officia.\r\n", + "registered": "Saturday, September 6, 2014 11:47 AM", + "latitude": 64.732922, + "longitude": -168.513014, + "tags": [ + "tempor", + "amet", + "dolore", + "proident", + "reprehenderit", + "non", + "tempor" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Freida Bailey" + }, + { + "id": 1, + "name": "Bernice Curry" + }, + { + "id": 2, + "name": "Ochoa Jefferson" + } + ], + "greeting": "Hello, Walters! You have 6 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa82137be18ee6b852a45", + "index": 33, + "guid": "e8c67cca-c977-4668-9aff-46bfde1cd3de", + "isActive": true, + "balance": "$3,747.65", + "picture": "http://placehold.it/32x32", + "age": 24, + "eyeColor": "blue", + "name": { + "first": "Meredith", + "last": "Santana" + }, + "company": "GADTRON", + "email": "meredith.santana@gadtron.org", + "phone": "+1 (836) 438-3637", + "address": "999 Centre Street, Chaparrito, Colorado, 4540", + "about": "Magna nisi laboris sit quis duis anim et ullamco nostrud exercitation. Tempor enim nisi non culpa sit ex elit labore proident veniam dolore anim ex. Nostrud est qui do magna proident et. Nulla ea laboris incididunt elit labore id mollit reprehenderit. Amet in Lorem exercitation tempor voluptate labore anim adipisicing labore in dolor proident labore. Lorem labore duis ex Lorem nulla. Veniam in fugiat ex ullamco officia elit eiusmod enim.\r\n", + "registered": "Sunday, March 2, 2014 1:38 PM", + "latitude": 80.220732, + "longitude": 79.102966, + "tags": [ + "culpa", + "esse", + "velit", + "consectetur", + "incididunt", + "dolore", + "aliqua" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bernadine Manning" + }, + { + "id": 1, + "name": "Daphne Wyatt" + }, + { + "id": 2, + "name": "Keri Harrison" + } + ], + "greeting": "Hello, Meredith! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821ecdd700bb0b77dc2", + "index": 34, + "guid": "b93f3a6f-f05b-4d7e-93b8-4502d9c76cd2", + "isActive": true, + "balance": "$3,059.56", + "picture": "http://placehold.it/32x32", + "age": 25, + "eyeColor": "brown", + "name": { + "first": "Samantha", + "last": "Finley" + }, + "company": "XYQAG", + "email": "samantha.finley@xyqag.biz", + "phone": "+1 (879) 568-2419", + "address": "349 Truxton Street, Haring, Missouri, 8383", + "about": "Consequat do id et quis eiusmod eu irure sunt qui. Mollit minim nulla magna duis nostrud cillum ullamco sunt adipisicing elit ex. Minim fugiat deserunt nostrud esse laboris ullamco sit sit magna. Tempor occaecat Lorem qui ad ut tempor excepteur. Et sunt ullamco officia et Lorem est. Ipsum dolor ut ut elit do nisi in aute consequat. Enim esse ex aliqua anim aliquip cupidatat do Lorem voluptate quis ea culpa incididunt reprehenderit.\r\n", + "registered": "Saturday, March 29, 2014 1:38 PM", + "latitude": 79.209401, + "longitude": -139.211605, + "tags": [ + "irure", + "eiusmod", + "nulla", + "officia", + "eu", + "elit", + "nisi" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mayer Justice" + }, + { + "id": 1, + "name": "Mae Hancock" + }, + { + "id": 2, + "name": "Sherri Bradshaw" + } + ], + "greeting": "Hello, Samantha! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8218ac5365ad300d1bd", + "index": 35, + "guid": "f3b50cb9-da35-47fe-b0fa-fec9fc84cf8e", + "isActive": false, + "balance": "$2,819.28", + "picture": "http://placehold.it/32x32", + "age": 29, + "eyeColor": "blue", + "name": { + "first": "Cohen", + "last": "Finch" + }, + "company": "SYNTAC", + "email": "cohen.finch@syntac.io", + "phone": "+1 (950) 459-2729", + "address": "436 Pineapple Street, Deercroft, Minnesota, 3218", + "about": "Sint velit officia quis esse. Nulla aute laborum veniam dolore tempor adipisicing proident. Duis irure esse nostrud veniam est mollit mollit voluptate eiusmod anim veniam eiusmod. Ullamco sunt sit sint minim ea reprehenderit qui consequat ipsum. Sint id voluptate reprehenderit irure nulla veniam eu Lorem enim nulla. Cupidatat amet pariatur dolor amet ex nostrud dolor ipsum tempor enim nulla aliquip tempor Lorem.\r\n", + "registered": "Wednesday, October 8, 2014 10:04 PM", + "latitude": -84.299718, + "longitude": 52.573184, + "tags": [ + "deserunt", + "eu", + "consectetur", + "ea", + "non", + "officia", + "reprehenderit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Stevenson Mcintosh" + }, + { + "id": 1, + "name": "Chrystal Oneill" + }, + { + "id": 2, + "name": "Janine Rowe" + } + ], + "greeting": "Hello, Cohen! You have 10 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa82161d5c66d4bd0996e", + "index": 36, + "guid": "6dfc2232-f3c6-4818-956a-e8c683fd69fe", + "isActive": true, + "balance": "$3,262.20", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "blue", + "name": { + "first": "Joy", + "last": "Moran" + }, + "company": "UPDAT", + "email": "joy.moran@updat.net", + "phone": "+1 (968) 581-3365", + "address": "279 Kosciusko Street, Smock, Northern Mariana Islands, 5007", + "about": "Quis elit pariatur eu enim magna magna sunt dolore duis commodo. Pariatur sint duis ex aute eu est deserunt culpa fugiat minim non. Tempor consequat consectetur consequat sit incididunt officia id. Incididunt ex eiusmod excepteur aute mollit veniam quis excepteur occaecat excepteur deserunt reprehenderit. Est sit laboris eu dolor. Sunt voluptate quis aliquip nulla ex irure velit in. Aliqua sit id eiusmod amet commodo pariatur deserunt voluptate qui minim ex incididunt voluptate.\r\n", + "registered": "Saturday, February 8, 2014 9:55 PM", + "latitude": 53.735731, + "longitude": 46.00211, + "tags": [ + "exercitation", + "dolor", + "minim", + "incididunt", + "laborum", + "qui", + "sit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Joyce Mckay" + }, + { + "id": 1, + "name": "Turner Murray" + }, + { + "id": 2, + "name": "Jackson Jackson" + } + ], + "greeting": "Hello, Joy! You have 7 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821a9817ca1a9be6517", + "index": 37, + "guid": "84fe8707-cfbc-4434-98bb-faf9cb97471a", + "isActive": true, + "balance": "$3,224.80", + "picture": "http://placehold.it/32x32", + "age": 27, + "eyeColor": "blue", + "name": { + "first": "Brennan", + "last": "Stafford" + }, + "company": "LOVEPAD", + "email": "brennan.stafford@lovepad.com", + "phone": "+1 (873) 405-3600", + "address": "471 Ryder Avenue, Succasunna, Marshall Islands, 2595", + "about": "Et officia quis magna laborum et proident labore elit do Lorem reprehenderit irure. Laborum culpa culpa voluptate commodo consequat non et amet. Mollit cupidatat irure magna sint commodo ipsum proident tempor est. Laboris exercitation aliqua aute deserunt do in aliqua minim ex excepteur. Consequat in minim officia labore laboris laboris occaecat occaecat. Ex qui aliquip sint consectetur elit excepteur incididunt eu non laborum do eu excepteur.\r\n", + "registered": "Saturday, May 31, 2014 3:28 PM", + "latitude": -69.429728, + "longitude": 46.837644, + "tags": [ + "proident", + "ad", + "non", + "duis", + "occaecat", + "proident", + "non" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Jones Kirk" + }, + { + "id": 1, + "name": "Howe Drake" + }, + { + "id": 2, + "name": "Kimberly Jennings" + } + ], + "greeting": "Hello, Brennan! You have 6 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa8219251dc49e1cb846a", + "index": 38, + "guid": "c260bce1-46ac-4e84-9490-13eb8202904e", + "isActive": true, + "balance": "$1,330.69", + "picture": "http://placehold.it/32x32", + "age": 27, + "eyeColor": "brown", + "name": { + "first": "Neal", + "last": "Mooney" + }, + "company": "SOFTMICRO", + "email": "neal.mooney@softmicro.biz", + "phone": "+1 (883) 463-3623", + "address": "818 Lancaster Avenue, Chelsea, New Jersey, 3590", + "about": "Reprehenderit anim nostrud adipisicing non minim ea. Elit deserunt id in mollit nisi. Pariatur in consequat irure aliqua laboris ipsum.\r\n", + "registered": "Wednesday, May 21, 2014 4:33 PM", + "latitude": -57.826881, + "longitude": 154.840249, + "tags": [ + "consequat", + "aliquip", + "pariatur", + "nulla", + "dolore", + "deserunt", + "ipsum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Beach Roman" + }, + { + "id": 1, + "name": "Nash Young" + }, + { + "id": 2, + "name": "Carey Dale" + } + ], + "greeting": "Hello, Neal! You have 10 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821cedc96e4f2209487", + "index": 39, + "guid": "ed7bbe27-811c-44e4-896e-cdf6ef62e048", + "isActive": false, + "balance": "$3,148.21", + "picture": "http://placehold.it/32x32", + "age": 35, + "eyeColor": "green", + "name": { + "first": "Roy", + "last": "Becker" + }, + "company": "KONGENE", + "email": "roy.becker@kongene.co.uk", + "phone": "+1 (895) 426-2172", + "address": "414 Seagate Avenue, Watrous, Virgin Islands, 3905", + "about": "Commodo mollit do minim dolor magna occaecat labore Lorem eiusmod. Occaecat mollit occaecat ex anim est amet irure non minim. Tempor laborum cupidatat tempor ex Lorem cupidatat incididunt ullamco fugiat Lorem consequat labore Lorem non.\r\n", + "registered": "Sunday, August 17, 2014 3:16 PM", + "latitude": -2.609533, + "longitude": -143.844769, + "tags": [ + "do", + "culpa", + "sint", + "ea", + "duis", + "aliqua", + "anim" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Lowery Hull" + }, + { + "id": 1, + "name": "Abbott Oneal" + }, + { + "id": 2, + "name": "Nellie Hammond" + } + ], + "greeting": "Hello, Roy! You have 9 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821668031e7f50e30e9", + "index": 40, + "guid": "14dc3a87-0acf-4edd-93e4-ddcbeeecf96b", + "isActive": false, + "balance": "$2,617.16", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "brown", + "name": { + "first": "Courtney", + "last": "Watson" + }, + "company": "ISOSWITCH", + "email": "courtney.watson@isoswitch.tv", + "phone": "+1 (882) 468-2163", + "address": "385 Douglass Street, Iberia, Oregon, 7802", + "about": "Culpa sunt amet eu magna id quis quis irure velit. Culpa nostrud do enim proident officia. Laboris laborum laborum esse irure proident laborum amet sunt ipsum dolor nulla non ipsum sint. Amet deserunt in esse aliquip laboris proident fugiat nisi cillum ullamco occaecat est. Reprehenderit laborum enim labore ex. Velit do adipisicing irure dolor pariatur duis magna velit. Laborum sint laborum eu anim aliquip adipisicing labore.\r\n", + "registered": "Tuesday, July 22, 2014 6:03 PM", + "latitude": -75.831312, + "longitude": -172.468604, + "tags": [ + "dolor", + "velit", + "et", + "id", + "cupidatat", + "exercitation", + "laborum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mccray Gomez" + }, + { + "id": 1, + "name": "Hoover Rasmussen" + }, + { + "id": 2, + "name": "Hillary Castillo" + } + ], + "greeting": "Hello, Courtney! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8213c480453670f0428", + "index": 41, + "guid": "e7b97ea6-a13f-42ab-a998-e48614000aca", + "isActive": true, + "balance": "$1,499.72", + "picture": "http://placehold.it/32x32", + "age": 35, + "eyeColor": "blue", + "name": { + "first": "Dale", + "last": "Love" + }, + "company": "BULLJUICE", + "email": "dale.love@bulljuice.name", + "phone": "+1 (933) 588-3310", + "address": "603 Myrtle Avenue, Allentown, Utah, 793", + "about": "Ad quis anim commodo nulla et anim minim commodo irure excepteur. Pariatur ut anim aliquip id ex ipsum exercitation irure qui in nisi quis. Cillum deserunt duis dolore quis nostrud incididunt ipsum ea ipsum id fugiat eu voluptate nisi. Exercitation laborum fugiat irure anim. Ex nostrud aliqua deserunt amet.\r\n", + "registered": "Tuesday, July 22, 2014 1:19 PM", + "latitude": 55.335017, + "longitude": 101.730023, + "tags": [ + "ea", + "non", + "fugiat", + "Lorem", + "tempor", + "ut", + "do" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Spears Diaz" + }, + { + "id": 1, + "name": "Nora Dominguez" + }, + { + "id": 2, + "name": "Tamra Paul" + } + ], + "greeting": "Hello, Dale! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821c82bf70ca08206a9", + "index": 42, + "guid": "4367142d-e2b7-475c-b430-ac617295fbfc", + "isActive": false, + "balance": "$2,698.29", + "picture": "http://placehold.it/32x32", + "age": 37, + "eyeColor": "blue", + "name": { + "first": "Gibbs", + "last": "Hunt" + }, + "company": "COWTOWN", + "email": "gibbs.hunt@cowtown.me", + "phone": "+1 (960) 424-3404", + "address": "758 Suydam Place, Adamstown, North Carolina, 2800", + "about": "Veniam qui incididunt officia amet commodo nostrud. Magna consectetur consectetur officia Lorem amet sit officia excepteur minim consectetur pariatur dolore. Mollit fugiat aliqua consectetur qui non elit in aliquip culpa Lorem consectetur velit ad.\r\n", + "registered": "Monday, February 10, 2014 4:10 PM", + "latitude": 62.263652, + "longitude": 64.978136, + "tags": [ + "et", + "dolor", + "aute", + "minim", + "sunt", + "veniam", + "do" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Henry Schmidt" + }, + { + "id": 1, + "name": "Herman Wynn" + }, + { + "id": 2, + "name": "Wilda Grimes" + } + ], + "greeting": "Hello, Gibbs! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8210b1096c8a2f821c4", + "index": 43, + "guid": "5184508c-47f7-48b5-85ac-d15ed747ed07", + "isActive": true, + "balance": "$3,460.45", + "picture": "http://placehold.it/32x32", + "age": 30, + "eyeColor": "green", + "name": { + "first": "Francis", + "last": "Barton" + }, + "company": "COMTRAIL", + "email": "francis.barton@comtrail.us", + "phone": "+1 (881) 419-2936", + "address": "562 Ferris Street, Ola, Maryland, 3838", + "about": "Duis minim laboris in reprehenderit id ut laborum esse consequat. In dolore sunt consequat non fugiat do duis duis. Officia ipsum eiusmod laboris do aliqua aute velit minim nulla nisi. Dolor incididunt enim est eu cupidatat. Dolor commodo sit consectetur irure aliqua ea enim esse reprehenderit ullamco.\r\n", + "registered": "Thursday, April 10, 2014 2:48 PM", + "latitude": -35.457713, + "longitude": 141.805123, + "tags": [ + "tempor", + "officia", + "quis", + "tempor", + "ex", + "mollit", + "amet" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mayra Walters" + }, + { + "id": 1, + "name": "Casey Gross" + }, + { + "id": 2, + "name": "Aisha Santos" + } + ], + "greeting": "Hello, Francis! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821890ad48b2eaabcfb", + "index": 44, + "guid": "07497907-7e6f-471d-af9d-1dcbcf056a56", + "isActive": true, + "balance": "$2,166.79", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "blue", + "name": { + "first": "Dina", + "last": "Travis" + }, + "company": "VENOFLEX", + "email": "dina.travis@venoflex.ca", + "phone": "+1 (921) 485-3865", + "address": "366 Tillary Street, Century, New York, 6335", + "about": "Dolor sunt culpa enim sint officia sint id do ut anim ex in. Dolore est irure aliquip nulla laborum aliqua tempor id ea mollit in ad deserunt. Qui ullamco duis qui elit excepteur. Aute proident duis veniam enim commodo non minim id. Consequat anim eiusmod consectetur ut et labore officia ex ad cillum occaecat. Sit irure officia veniam sint et consequat reprehenderit officia qui. Aute esse nulla ad quis reprehenderit duis.\r\n", + "registered": "Friday, September 12, 2014 5:53 AM", + "latitude": 24.764352, + "longitude": 148.493552, + "tags": [ + "qui", + "officia", + "dolor", + "velit", + "ex", + "veniam", + "ullamco" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Sharpe Foster" + }, + { + "id": 1, + "name": "Kathrine Ayers" + }, + { + "id": 2, + "name": "Cox Acosta" + } + ], + "greeting": "Hello, Dina! You have 8 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa82101996fc944f7f215", + "index": 45, + "guid": "46b331f8-78d3-403f-8c93-cbaac9438998", + "isActive": false, + "balance": "$2,193.84", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "blue", + "name": { + "first": "Haney", + "last": "Garrett" + }, + "company": "EZENTIA", + "email": "haney.garrett@ezentia.org", + "phone": "+1 (965) 596-2629", + "address": "162 Village Road, Southmont, Virginia, 6673", + "about": "Laboris do veniam exercitation officia id eu minim irure Lorem laborum. Sint magna aliquip ad elit eiusmod cillum laborum. Adipisicing aliquip nulla mollit ipsum cupidatat nisi duis irure ullamco. Et elit nulla nisi culpa mollit esse in. Dolore non nulla anim magna enim aliquip quis amet non tempor incididunt dolor.\r\n", + "registered": "Monday, August 18, 2014 3:42 PM", + "latitude": 6.549909, + "longitude": 32.226887, + "tags": [ + "laboris", + "duis", + "culpa", + "qui", + "dolor", + "esse", + "reprehenderit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mariana Sharp" + }, + { + "id": 1, + "name": "Sue Baldwin" + }, + { + "id": 2, + "name": "Ross Arnold" + } + ], + "greeting": "Hello, Haney! You have 10 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821453cc33bd1abd21d", + "index": 46, + "guid": "313f0c45-6eaf-46d9-a28d-ca31e79d2c1c", + "isActive": false, + "balance": "$2,303.36", + "picture": "http://placehold.it/32x32", + "age": 28, + "eyeColor": "blue", + "name": { + "first": "Gale", + "last": "Robles" + }, + "company": "NEWCUBE", + "email": "gale.robles@newcube.biz", + "phone": "+1 (996) 558-2811", + "address": "597 Java Street, Sanborn, North Dakota, 1789", + "about": "Anim fugiat do nisi dolor sunt consequat irure quis laborum. Nisi cupidatat dolore excepteur irure ea minim proident excepteur exercitation ut voluptate deserunt. Ad do amet id voluptate enim commodo ex. Sunt sint quis sint aute do ea aliqua. Enim ullamco dolore proident qui mollit irure consequat. Nostrud sunt adipisicing elit incididunt do laboris ad officia ea amet id reprehenderit nulla.\r\n", + "registered": "Tuesday, July 15, 2014 2:25 PM", + "latitude": -21.549196, + "longitude": -97.373962, + "tags": [ + "ullamco", + "amet", + "sint", + "elit", + "tempor", + "ex", + "pariatur" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bell Gould" + }, + { + "id": 1, + "name": "Denise Kirby" + }, + { + "id": 2, + "name": "Hess Hinton" + } + ], + "greeting": "Hello, Gale! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821ed643a410e2d79dc", + "index": 47, + "guid": "2620cb2b-df00-4303-a5ff-768ffb1697c5", + "isActive": true, + "balance": "$2,890.73", + "picture": "http://placehold.it/32x32", + "age": 39, + "eyeColor": "blue", + "name": { + "first": "Alisha", + "last": "Hamilton" + }, + "company": "FITCORE", + "email": "alisha.hamilton@fitcore.io", + "phone": "+1 (853) 468-3192", + "address": "257 Bayard Street, Coldiron, Oklahoma, 9228", + "about": "Sunt nostrud sunt magna amet excepteur est tempor veniam aliqua. Laboris id aliquip fugiat exercitation dolore veniam et anim duis sit esse ex elit ullamco. Lorem commodo exercitation in sit cillum ipsum do dolor.\r\n", + "registered": "Tuesday, March 11, 2014 12:29 PM", + "latitude": 85.840017, + "longitude": -57.093095, + "tags": [ + "consectetur", + "commodo", + "est", + "ut", + "incididunt", + "elit", + "ipsum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Steele Ellis" + }, + { + "id": 1, + "name": "Serena Emerson" + }, + { + "id": 2, + "name": "Betty Langley" + } + ], + "greeting": "Hello, Alisha! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8215c6db03112e3e13d", + "index": 48, + "guid": "28b59ead-0fd8-480b-8eb8-2d75290934f6", + "isActive": false, + "balance": "$3,497.68", + "picture": "http://placehold.it/32x32", + "age": 25, + "eyeColor": "blue", + "name": { + "first": "Kaufman", + "last": "Williams" + }, + "company": "COMBOGENE", + "email": "kaufman.williams@combogene.net", + "phone": "+1 (820) 465-3213", + "address": "353 Clarendon Road, Wilmington, Michigan, 3811", + "about": "Irure id sint elit mollit occaecat occaecat veniam elit reprehenderit esse officia cillum. Aute aute occaecat ipsum commodo laborum adipisicing fugiat aliquip dolore. Deserunt id excepteur enim eu adipisicing nulla ut non est dolore est. Culpa magna et sit et non ex.\r\n", + "registered": "Sunday, September 7, 2014 5:57 AM", + "latitude": 10.667631, + "longitude": 157.707911, + "tags": [ + "consequat", + "dolor", + "deserunt", + "amet", + "Lorem", + "aliqua", + "minim" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Horn Franco" + }, + { + "id": 1, + "name": "Kathleen Hickman" + }, + { + "id": 2, + "name": "Rosario Scott" + } + ], + "greeting": "Hello, Kaufman! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821045c82593b79538d", + "index": 49, + "guid": "cef964ed-4208-41eb-a8d8-916d95d18f8a", + "isActive": true, + "balance": "$3,808.37", + "picture": "http://placehold.it/32x32", + "age": 28, + "eyeColor": "blue", + "name": { + "first": "Tran", + "last": "Gallegos" + }, + "company": "CONCILITY", + "email": "tran.gallegos@concility.com", + "phone": "+1 (925) 487-2143", + "address": "969 Schermerhorn Street, Watchtower, Idaho, 413", + "about": "Velit ut enim cupidatat adipisicing culpa in non incididunt exercitation dolor pariatur. Deserunt dolor occaecat dolor officia ipsum occaecat tempor nisi. Culpa et culpa aute incididunt et labore sunt cillum nulla. Reprehenderit culpa enim laborum nostrud consectetur velit nulla consequat aliqua non exercitation sunt nulla aliqua. Labore incididunt aliqua aliqua anim duis culpa elit labore. Aliqua ipsum mollit ut sint aliquip in aute do qui amet.\r\n", + "registered": "Friday, May 9, 2014 9:24 AM", + "latitude": -49.148583, + "longitude": 4.911715, + "tags": [ + "labore", + "duis", + "proident", + "adipisicing", + "nisi", + "tempor", + "nisi" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Nicole Holloway" + }, + { + "id": 1, + "name": "Minerva Beasley" + }, + { + "id": 2, + "name": "Bowers Suarez" + } + ], + "greeting": "Hello, Tran! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821fbd8e9836595f96d", + "index": 50, + "guid": "966b2f5e-5686-4381-9ab7-87bc9ac9968b", + "isActive": true, + "balance": "$1,351.67", + "picture": "http://placehold.it/32x32", + "age": 23, + "eyeColor": "green", + "name": { + "first": "Mable", + "last": "Lambert" + }, + "company": "ZENSUS", + "email": "mable.lambert@zensus.biz", + "phone": "+1 (917) 404-3441", + "address": "174 Concord Street, Somerset, Nebraska, 7318", + "about": "Ad minim esse ipsum incididunt incididunt enim Lorem nulla excepteur velit fugiat ullamco amet reprehenderit. Labore velit fugiat proident enim Lorem mollit ex. Tempor voluptate cupidatat officia nostrud qui. Veniam duis voluptate deserunt commodo consectetur eiusmod qui excepteur aliquip. Exercitation laborum Lorem excepteur ipsum aliqua fugiat reprehenderit ea dolore deserunt commodo cillum ipsum eu. Ullamco quis voluptate eiusmod ea aute et pariatur consequat duis occaecat nulla aliquip.\r\n", + "registered": "Sunday, May 25, 2014 4:54 PM", + "latitude": 79.811908, + "longitude": -62.629133, + "tags": [ + "sit", + "non", + "reprehenderit", + "exercitation", + "dolor", + "labore", + "irure" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Rosie Gill" + }, + { + "id": 1, + "name": "Parrish Dean" + }, + { + "id": 2, + "name": "Annmarie Delacruz" + } + ], + "greeting": "Hello, Mable! You have 10 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821c85822cf149b785f", + "index": 51, + "guid": "1afd0e8f-b518-4bca-a4df-29e9b6a961d6", + "isActive": true, + "balance": "$2,628.22", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "brown", + "name": { + "first": "Sadie", + "last": "Clarke" + }, + "company": "UNISURE", + "email": "sadie.clarke@unisure.co.uk", + "phone": "+1 (905) 480-2930", + "address": "326 Gilmore Court, Hamilton, Montana, 9217", + "about": "Dolore do nisi reprehenderit consectetur in. In esse sit proident enim duis veniam quis laboris nulla cillum adipisicing veniam aute. Culpa sunt ex exercitation sit esse exercitation dolor ea enim ad est aute consequat qui. Esse esse nulla eiusmod eiusmod ullamco esse cillum aute ea id ex quis. Pariatur officia Lorem aute officia anim velit velit elit sint voluptate. Aliquip ad velit velit laboris et culpa. Do consectetur aliqua sit sunt eu anim culpa ut incididunt et.\r\n", + "registered": "Sunday, April 27, 2014 10:37 AM", + "latitude": -71.828101, + "longitude": -138.908359, + "tags": [ + "nostrud", + "amet", + "minim", + "occaecat", + "proident", + "sint", + "nostrud" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Christa Estes" + }, + { + "id": 1, + "name": "Alana Schneider" + }, + { + "id": 2, + "name": "Frank Spears" + } + ], + "greeting": "Hello, Sadie! You have 6 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821aa6b44e8d20db81c", + "index": 52, + "guid": "e0220e02-565c-424a-8834-f955ccb72f7d", + "isActive": false, + "balance": "$2,918.63", + "picture": "http://placehold.it/32x32", + "age": 30, + "eyeColor": "brown", + "name": { + "first": "Deana", + "last": "Fletcher" + }, + "company": "VISALIA", + "email": "deana.fletcher@visalia.tv", + "phone": "+1 (815) 430-2641", + "address": "347 Tehama Street, Hollins, Washington, 5953", + "about": "Est consequat id ad Lorem consequat quis ullamco minim pariatur ipsum cillum. Enim exercitation qui duis cillum ea amet ea sint proident officia dolor non. Irure culpa cillum minim officia est culpa sit.\r\n", + "registered": "Monday, May 5, 2014 1:51 AM", + "latitude": 51.516197, + "longitude": 80.400628, + "tags": [ + "ut", + "tempor", + "pariatur", + "ex", + "dolore", + "deserunt", + "culpa" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Hansen Estrada" + }, + { + "id": 1, + "name": "Regina Munoz" + }, + { + "id": 2, + "name": "Bethany Cabrera" + } + ], + "greeting": "Hello, Deana! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa82120c61930bc25e2ff", + "index": 53, + "guid": "99c62c99-ef16-4eb7-a41d-80feafca740a", + "isActive": false, + "balance": "$1,836.96", + "picture": "http://placehold.it/32x32", + "age": 27, + "eyeColor": "brown", + "name": { + "first": "Long", + "last": "Sandoval" + }, + "company": "GINKLE", + "email": "long.sandoval@ginkle.name", + "phone": "+1 (989) 541-2327", + "address": "161 Crown Street, Omar, Kentucky, 8615", + "about": "Et ea eiusmod ex consequat culpa proident. Reprehenderit proident ullamco ullamco aliquip incididunt ullamco sit proident dolore nulla fugiat sit laboris. Adipisicing ullamco laborum nulla exercitation reprehenderit irure ex.\r\n", + "registered": "Tuesday, July 29, 2014 12:14 AM", + "latitude": -20.766582, + "longitude": 74.616145, + "tags": [ + "et", + "consequat", + "duis", + "excepteur", + "sint", + "sunt", + "aliqua" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Aline Rosa" + }, + { + "id": 1, + "name": "Olivia Quinn" + }, + { + "id": 2, + "name": "Fowler Carter" + } + ], + "greeting": "Hello, Long! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821cc7ceb7da88472ed", + "index": 54, + "guid": "3a965f51-48f9-4d10-8a9b-a0b529749c93", + "isActive": true, + "balance": "$2,950.94", + "picture": "http://placehold.it/32x32", + "age": 25, + "eyeColor": "green", + "name": { + "first": "Rush", + "last": "Thornton" + }, + "company": "AQUASSEUR", + "email": "rush.thornton@aquasseur.me", + "phone": "+1 (916) 493-2777", + "address": "344 Sunnyside Avenue, Brethren, California, 9529", + "about": "Duis ut consequat eu laborum voluptate eu Lorem cillum ad in commodo adipisicing. Excepteur aliquip sint dolor voluptate cillum nisi mollit mollit laborum ex culpa adipisicing voluptate. Cupidatat do sit fugiat amet irure. Cupidatat ex ut commodo reprehenderit veniam sit est officia ad pariatur aliquip.\r\n", + "registered": "Thursday, September 25, 2014 11:09 AM", + "latitude": -61.409359, + "longitude": -87.414208, + "tags": [ + "adipisicing", + "consequat", + "magna", + "Lorem", + "consequat", + "voluptate", + "eu" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Hill Good" + }, + { + "id": 1, + "name": "Hartman Rice" + }, + { + "id": 2, + "name": "Petersen Hogan" + } + ], + "greeting": "Hello, Rush! You have 7 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821e221eb5b7ed845c0", + "index": 55, + "guid": "ab0f1517-8c25-46ff-a281-0dcc932e2f9a", + "isActive": false, + "balance": "$2,070.82", + "picture": "http://placehold.it/32x32", + "age": 39, + "eyeColor": "green", + "name": { + "first": "Alyce", + "last": "Perez" + }, + "company": "CAPSCREEN", + "email": "alyce.perez@capscreen.us", + "phone": "+1 (955) 432-2025", + "address": "896 Troutman Street, Williamson, West Virginia, 6491", + "about": "In consequat quis ex sint nisi proident esse excepteur quis nostrud. Enim incididunt ullamco sint quis eiusmod qui tempor ad laboris eiusmod nulla in aliquip sit. Nostrud fugiat fugiat Lorem laboris pariatur eiusmod amet ea do irure et. Excepteur pariatur consequat exercitation amet occaecat do aliqua non deserunt nulla cupidatat tempor id. Mollit ex incididunt et nulla culpa mollit veniam qui amet in excepteur pariatur. Commodo reprehenderit tempor laborum nisi anim minim deserunt eiusmod adipisicing deserunt ut eiusmod excepteur.\r\n", + "registered": "Monday, June 23, 2014 7:37 AM", + "latitude": -4.180393, + "longitude": 21.4789, + "tags": [ + "dolore", + "officia", + "laborum", + "aliquip", + "ex", + "eu", + "sint" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Rosalind Lang" + }, + { + "id": 1, + "name": "Ina Pratt" + }, + { + "id": 2, + "name": "Tamika Mercer" + } + ], + "greeting": "Hello, Alyce! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8215e315c720e186460", + "index": 56, + "guid": "10727900-9be7-46d5-8f2c-7c6834d95a25", + "isActive": false, + "balance": "$1,907.19", + "picture": "http://placehold.it/32x32", + "age": 37, + "eyeColor": "brown", + "name": { + "first": "Snider", + "last": "Johnson" + }, + "company": "IPLAX", + "email": "snider.johnson@iplax.ca", + "phone": "+1 (883) 539-3127", + "address": "424 Hancock Street, Springdale, Wyoming, 7054", + "about": "Incididunt proident amet consectetur cupidatat ex officia labore cupidatat laborum. Tempor proident officia nisi Lorem. Lorem commodo commodo ea voluptate excepteur consequat anim quis excepteur sunt officia.\r\n", + "registered": "Saturday, March 8, 2014 9:23 PM", + "latitude": 16.66716, + "longitude": 17.844641, + "tags": [ + "esse", + "est", + "eu", + "dolor", + "ea", + "voluptate", + "nostrud" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bauer Burt" + }, + { + "id": 1, + "name": "Lowe Boyd" + }, + { + "id": 2, + "name": "Moon Garcia" + } + ], + "greeting": "Hello, Snider! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa82130f82b9ee9265e5f", + "index": 57, + "guid": "ba427ee5-5013-498f-a1a2-97a71b249a6e", + "isActive": true, + "balance": "$2,070.53", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "brown", + "name": { + "first": "Berry", + "last": "Carver" + }, + "company": "EVENTIX", + "email": "berry.carver@eventix.org", + "phone": "+1 (907) 508-2463", + "address": "415 Havens Place, Nile, Connecticut, 6089", + "about": "Quis dolor aute consequat sunt esse dolore Lorem pariatur reprehenderit incididunt aliqua. Officia sunt aute fugiat consectetur id exercitation aliquip velit do fugiat culpa. Et ad amet exercitation veniam ipsum duis qui sunt incididunt. Eiusmod commodo esse aliquip exercitation pariatur consequat nulla nulla quis eiusmod dolor. Ut consectetur qui culpa id veniam dolore pariatur quis est cillum voluptate esse. Sunt eiusmod adipisicing mollit est tempor ipsum dolore tempor. Velit consequat dolore cillum adipisicing id nulla veniam nisi velit in magna id anim.\r\n", + "registered": "Monday, January 13, 2014 10:45 PM", + "latitude": -60.884888, + "longitude": 139.360489, + "tags": [ + "eu", + "deserunt", + "minim", + "quis", + "eiusmod", + "sint", + "dolor" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Tanisha Dudley" + }, + { + "id": 1, + "name": "Dale Mcgowan" + }, + { + "id": 2, + "name": "Torres Pennington" + } + ], + "greeting": "Hello, Berry! You have 7 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821bc354233d50ef914", + "index": 58, + "guid": "6ab626f8-6b71-4f17-ba39-4cc97fdf4855", + "isActive": false, + "balance": "$2,822.25", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "green", + "name": { + "first": "Kramer", + "last": "Berg" + }, + "company": "INTRADISK", + "email": "kramer.berg@intradisk.biz", + "phone": "+1 (901) 534-3326", + "address": "455 Bath Avenue, Hoagland, Guam, 2206", + "about": "Labore ullamco aliquip id incididunt cupidatat pariatur. In magna et aliquip consectetur dolor ullamco aliqua reprehenderit. Ad velit nisi ex culpa consequat. Culpa eiusmod incididunt pariatur esse tempor officia mollit.\r\n", + "registered": "Friday, June 13, 2014 8:45 AM", + "latitude": -43.442578, + "longitude": 69.627031, + "tags": [ + "exercitation", + "dolor", + "quis", + "laboris", + "exercitation", + "sunt", + "ipsum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Love Hutchinson" + }, + { + "id": 1, + "name": "Hayden Marquez" + }, + { + "id": 2, + "name": "Macdonald Hahn" + } + ], + "greeting": "Hello, Kramer! You have 5 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821cf5d51b48b4bd07a", + "index": 59, + "guid": "7b339d1e-d759-4fed-9e58-fd2608cdb0f2", + "isActive": false, + "balance": "$3,836.86", + "picture": "http://placehold.it/32x32", + "age": 20, + "eyeColor": "brown", + "name": { + "first": "Joann", + "last": "Elliott" + }, + "company": "ATOMICA", + "email": "joann.elliott@atomica.io", + "phone": "+1 (992) 429-2667", + "address": "788 Willow Place, Lindisfarne, Mississippi, 3656", + "about": "Ut consequat sunt ipsum minim velit. Lorem eiusmod dolor voluptate est deserunt cupidatat ut ipsum. Et et irure Lorem laborum sint mollit pariatur elit et enim eu eu sunt. Nisi do quis proident enim irure dolore ut Lorem fugiat quis voluptate non reprehenderit dolore.\r\n", + "registered": "Monday, May 12, 2014 6:04 PM", + "latitude": 14.309335, + "longitude": 32.596666, + "tags": [ + "nulla", + "magna", + "dolore", + "incididunt", + "fugiat", + "elit", + "veniam" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mason Hurst" + }, + { + "id": 1, + "name": "Castaneda Davidson" + }, + { + "id": 2, + "name": "Rasmussen Adkins" + } + ], + "greeting": "Hello, Joann! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821876778d3d011eb7b", + "index": 60, + "guid": "a016ab64-b6dd-42bc-8b5f-dbbab7a01d2a", + "isActive": true, + "balance": "$2,795.00", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "green", + "name": { + "first": "Barbara", + "last": "Nolan" + }, + "company": "FURNAFIX", + "email": "barbara.nolan@furnafix.net", + "phone": "+1 (892) 600-2820", + "address": "540 Dennett Place, Mammoth, Rhode Island, 6151", + "about": "In velit officia quis Lorem. Ex quis cillum esse deserunt consectetur et nulla tempor. Lorem reprehenderit cillum excepteur ea veniam commodo et ad ullamco. Ex elit nisi non ipsum aliqua laborum sint aliqua. Reprehenderit consectetur dolore occaecat irure incididunt sunt.\r\n", + "registered": "Monday, April 21, 2014 11:59 AM", + "latitude": 71.103088, + "longitude": -78.48592, + "tags": [ + "eu", + "ullamco", + "cillum", + "est", + "commodo", + "nisi", + "tempor" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Cynthia Aguilar" + }, + { + "id": 1, + "name": "Deanna Graves" + }, + { + "id": 2, + "name": "Bertha Caldwell" + } + ], + "greeting": "Hello, Barbara! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821d6e2aaf18a0b184f", + "index": 61, + "guid": "89ca6bc0-19c2-4d13-a37c-90d02005a6bc", + "isActive": false, + "balance": "$3,134.62", + "picture": "http://placehold.it/32x32", + "age": 39, + "eyeColor": "blue", + "name": { + "first": "Penelope", + "last": "William" + }, + "company": "UTARA", + "email": "penelope.william@utara.com", + "phone": "+1 (968) 575-2395", + "address": "276 Ralph Avenue, Ezel, New Mexico, 2656", + "about": "Pariatur officia anim dolore commodo ipsum labore sint officia. Lorem culpa ea sunt non. Voluptate irure voluptate ut cupidatat nulla nostrud.\r\n", + "registered": "Monday, July 7, 2014 11:29 PM", + "latitude": -83.184502, + "longitude": -91.222471, + "tags": [ + "anim", + "incididunt", + "aliqua", + "id", + "reprehenderit", + "laboris", + "consequat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Maxwell Rocha" + }, + { + "id": 1, + "name": "Roach Bryan" + }, + { + "id": 2, + "name": "Woods Daugherty" + } + ], + "greeting": "Hello, Penelope! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8217abc85acd32fb42a", + "index": 62, + "guid": "cc58e868-7934-40a2-a350-13ad47e86a56", + "isActive": true, + "balance": "$3,734.05", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "green", + "name": { + "first": "Kris", + "last": "Cotton" + }, + "company": "CORPORANA", + "email": "kris.cotton@corporana.biz", + "phone": "+1 (873) 412-2513", + "address": "650 Bushwick Court, Malott, Wisconsin, 9739", + "about": "Aliquip cupidatat exercitation exercitation consectetur. Sit id excepteur ea ut laborum irure ullamco laborum irure reprehenderit nisi aute eu. Ipsum do anim ea veniam do amet pariatur. Lorem consectetur labore deserunt anim deserunt aute.\r\n", + "registered": "Thursday, February 20, 2014 7:18 AM", + "latitude": 41.787092, + "longitude": -44.032192, + "tags": [ + "ex", + "incididunt", + "ut", + "cupidatat", + "commodo", + "commodo", + "occaecat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Angela Middleton" + }, + { + "id": 1, + "name": "Hicks Douglas" + }, + { + "id": 2, + "name": "Shaffer West" + } + ], + "greeting": "Hello, Kris! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821c20dda84da819450", + "index": 63, + "guid": "6ecab19a-0268-4d43-ad43-af2e4941b2e7", + "isActive": false, + "balance": "$2,748.00", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "green", + "name": { + "first": "William", + "last": "Haney" + }, + "company": "PROSURE", + "email": "william.haney@prosure.co.uk", + "phone": "+1 (890) 508-3193", + "address": "930 Hopkins Street, Bluetown, American Samoa, 9860", + "about": "Ad aute aliquip eiusmod tempor ullamco. Et ipsum consequat consequat magna do fugiat sint proident nostrud ad fugiat commodo dolor. Est anim do laboris id esse minim do voluptate occaecat nulla esse. Veniam sit dolore aliqua pariatur quis commodo enim nisi sint excepteur pariatur.\r\n", + "registered": "Friday, January 10, 2014 2:42 PM", + "latitude": 70.057151, + "longitude": -46.509685, + "tags": [ + "pariatur", + "est", + "adipisicing", + "aute", + "in", + "ex", + "eu" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Aida Lindsey" + }, + { + "id": 1, + "name": "Harper Roberson" + }, + { + "id": 2, + "name": "Flora Woods" + } + ], + "greeting": "Hello, William! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8219387ab6d8ebd8c1d", + "index": 64, + "guid": "732dfcb2-f8ab-44bf-a11b-d98f5b589993", + "isActive": true, + "balance": "$1,010.25", + "picture": "http://placehold.it/32x32", + "age": 26, + "eyeColor": "green", + "name": { + "first": "Barrera", + "last": "Sellers" + }, + "company": "KLUGGER", + "email": "barrera.sellers@klugger.tv", + "phone": "+1 (968) 510-3228", + "address": "954 Verona Place, Homeworth, Louisiana, 6862", + "about": "Lorem ex voluptate cupidatat minim officia voluptate enim proident qui mollit dolore ipsum. Non consectetur adipisicing quis consectetur. Non est Lorem ad qui nostrud aute aliqua labore exercitation ea aliquip irure dolor nisi. Excepteur anim ex exercitation velit adipisicing qui excepteur enim culpa consequat sint. Velit consectetur velit culpa eu sint irure culpa consequat anim incididunt ad amet excepteur. Non est anim id sint ipsum id officia dolor commodo dolore labore consectetur.\r\n", + "registered": "Sunday, September 14, 2014 1:44 AM", + "latitude": -88.561319, + "longitude": -44.881241, + "tags": [ + "aliquip", + "eiusmod", + "nisi", + "aliquip", + "minim", + "ullamco", + "commodo" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bond Goff" + }, + { + "id": 1, + "name": "Cathleen Hatfield" + }, + { + "id": 2, + "name": "Pansy Burke" + } + ], + "greeting": "Hello, Barrera! You have 8 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821ab4d9564c853db57", + "index": 65, + "guid": "bba298e6-ebca-4334-afe9-91807ed1b672", + "isActive": true, + "balance": "$2,345.80", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "brown", + "name": { + "first": "Mullen", + "last": "Stephenson" + }, + "company": "POLARIUM", + "email": "mullen.stephenson@polarium.name", + "phone": "+1 (935) 461-2692", + "address": "228 Court Square, Beaulieu, Kansas, 9445", + "about": "Ex magna proident do dolore nostrud aliqua aute dolore enim mollit consectetur sunt pariatur. Ex id duis enim duis laborum do tempor proident exercitation duis. Aute dolor cillum anim incididunt voluptate. Qui proident consectetur sit laboris ex enim excepteur qui.\r\n", + "registered": "Thursday, May 1, 2014 7:57 PM", + "latitude": 34.294733, + "longitude": 138.270754, + "tags": [ + "minim", + "veniam", + "do", + "consequat", + "esse", + "sit", + "do" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bobbi Pate" + }, + { + "id": 1, + "name": "Moran Griffith" + }, + { + "id": 2, + "name": "Marian Hopkins" + } + ], + "greeting": "Hello, Mullen! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8218eb53a92791a3185", + "index": 66, + "guid": "d297bed2-0986-4d22-b120-0e04c253fb34", + "isActive": false, + "balance": "$2,586.70", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "brown", + "name": { + "first": "Leigh", + "last": "Kidd" + }, + "company": "SUNCLIPSE", + "email": "leigh.kidd@sunclipse.me", + "phone": "+1 (870) 427-3520", + "address": "156 Keen Court, Chamizal, Indiana, 4250", + "about": "Ut sunt elit irure eiusmod aliquip consectetur in. Dolore id exercitation irure consectetur. Pariatur occaecat cillum nulla cillum esse deserunt minim consectetur aliqua duis eiusmod. Ea proident aliquip cillum ullamco duis elit Lorem dolore aliqua. Do fugiat culpa reprehenderit ea eu non enim.\r\n", + "registered": "Wednesday, January 15, 2014 1:57 AM", + "latitude": 61.842523, + "longitude": -56.422447, + "tags": [ + "non", + "non", + "incididunt", + "elit", + "aliqua", + "proident", + "nostrud" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Jean Burns" + }, + { + "id": 1, + "name": "Patel Wilkinson" + }, + { + "id": 2, + "name": "Lillie Lane" + } + ], + "greeting": "Hello, Leigh! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8211c12bb8d837e265b", + "index": 67, + "guid": "6560e2f0-6bd2-4e19-bcd1-35297f162890", + "isActive": false, + "balance": "$2,040.59", + "picture": "http://placehold.it/32x32", + "age": 29, + "eyeColor": "green", + "name": { + "first": "Perry", + "last": "Leonard" + }, + "company": "CANDECOR", + "email": "perry.leonard@candecor.us", + "phone": "+1 (945) 556-2907", + "address": "359 Barbey Street, Grandview, Alaska, 9082", + "about": "Dolor fugiat consequat reprehenderit duis duis ex consectetur ea non ut. Irure et elit et mollit consequat dolor exercitation exercitation deserunt culpa mollit nulla. Est sunt deserunt incididunt exercitation eu aliqua qui elit labore id in eu ex.\r\n", + "registered": "Tuesday, August 26, 2014 11:34 PM", + "latitude": -67.974417, + "longitude": 97.189082, + "tags": [ + "sit", + "ex", + "ullamco", + "exercitation", + "adipisicing", + "non", + "laboris" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Alissa Ramsey" + }, + { + "id": 1, + "name": "Adela Bell" + }, + { + "id": 2, + "name": "Pearl Henderson" + } + ], + "greeting": "Hello, Perry! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa8219a0b8e44380bd954", + "index": 68, + "guid": "4aa2bec3-3eaa-464f-9577-27f6c65e64b7", + "isActive": false, + "balance": "$3,911.59", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "blue", + "name": { + "first": "Barbra", + "last": "Mejia" + }, + "company": "ANIXANG", + "email": "barbra.mejia@anixang.ca", + "phone": "+1 (987) 517-2550", + "address": "635 Franklin Street, Allensworth, Florida, 1895", + "about": "Aliqua tempor excepteur velit do exercitation laborum commodo laboris aliqua nostrud. Aute aliquip nisi nulla labore id veniam ad voluptate non eiusmod minim mollit. Minim incididunt nostrud sint ex.\r\n", + "registered": "Thursday, January 16, 2014 11:04 PM", + "latitude": 44.320545, + "longitude": -61.392889, + "tags": [ + "cupidatat", + "excepteur", + "eu", + "consectetur", + "fugiat", + "aliquip", + "deserunt" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Potts Church" + }, + { + "id": 1, + "name": "Dianna Valentine" + }, + { + "id": 2, + "name": "Valeria Whitney" + } + ], + "greeting": "Hello, Barbra! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821b618d050a076972c", + "index": 69, + "guid": "ef47a627-c517-4f5a-931b-d67c8a18614b", + "isActive": false, + "balance": "$1,771.53", + "picture": "http://placehold.it/32x32", + "age": 32, + "eyeColor": "brown", + "name": { + "first": "Gonzales", + "last": "Walker" + }, + "company": "EVIDENDS", + "email": "gonzales.walker@evidends.org", + "phone": "+1 (984) 510-3347", + "address": "336 Pershing Loop, Croom, Georgia, 9128", + "about": "Quis reprehenderit consectetur ad aliqua ad amet incididunt aute irure Lorem veniam. Consequat consequat ut reprehenderit officia cupidatat irure aliqua nostrud veniam velit aliquip magna elit. Ullamco amet nostrud est cupidatat adipisicing fugiat magna anim eu occaecat incididunt. Ut ex ut quis veniam nulla ad ea magna elit incididunt Lorem anim ipsum elit. Aliqua laborum officia magna aliqua. Est quis adipisicing cillum cupidatat sunt velit fugiat mollit exercitation cupidatat sit.\r\n", + "registered": "Saturday, June 21, 2014 9:02 PM", + "latitude": -64.407501, + "longitude": -157.742045, + "tags": [ + "pariatur", + "minim", + "consectetur", + "consequat", + "ad", + "aliqua", + "ut" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Tina Patterson" + }, + { + "id": 1, + "name": "Gabriela Nielsen" + }, + { + "id": 2, + "name": "Amalia Mueller" + } + ], + "greeting": "Hello, Gonzales! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8210426ced41d67a58b", + "index": 70, + "guid": "5d045ae8-c32c-43f6-b404-30a943205f5e", + "isActive": true, + "balance": "$2,054.02", + "picture": "http://placehold.it/32x32", + "age": 36, + "eyeColor": "green", + "name": { + "first": "Clarissa", + "last": "Madden" + }, + "company": "NIQUENT", + "email": "clarissa.madden@niquent.biz", + "phone": "+1 (910) 480-3769", + "address": "637 Scholes Street, Needmore, Alabama, 9344", + "about": "Sunt nulla ad aliquip incididunt ullamco culpa laboris. Consectetur non enim in officia incididunt deserunt. Quis est consequat ipsum ad. Pariatur nostrud voluptate magna occaecat minim irure sint nostrud voluptate ea labore ullamco quis. Mollit veniam consequat commodo sunt.\r\n", + "registered": "Wednesday, January 8, 2014 4:02 AM", + "latitude": 46.115788, + "longitude": 79.731859, + "tags": [ + "reprehenderit", + "nisi", + "id", + "consectetur", + "sunt", + "nostrud", + "laboris" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Mandy Buckner" + }, + { + "id": 1, + "name": "Hickman Brown" + }, + { + "id": 2, + "name": "Kemp Mclaughlin" + } + ], + "greeting": "Hello, Clarissa! You have 8 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8211049b550404f3ce5", + "index": 71, + "guid": "2f9a6201-6728-4509-8d88-c0a614649311", + "isActive": true, + "balance": "$1,324.10", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "blue", + "name": { + "first": "Joyce", + "last": "Callahan" + }, + "company": "OPTYK", + "email": "joyce.callahan@optyk.io", + "phone": "+1 (893) 544-2327", + "address": "567 Crystal Street, Freetown, District Of Columbia, 8319", + "about": "Deserunt in nisi id consequat qui. Sunt velit proident id culpa incididunt velit aute dolore labore. Deserunt qui ea adipisicing cillum irure sit sunt excepteur quis et quis nulla dolore pariatur. Consequat ut et veniam dolor velit nulla veniam fugiat commodo velit fugiat ad veniam ad. Anim consequat labore deserunt eiusmod esse. Laborum labore eu et incididunt commodo dolore eiusmod occaecat. Nisi elit duis mollit cillum id enim.\r\n", + "registered": "Tuesday, February 4, 2014 2:38 AM", + "latitude": -76.437449, + "longitude": -169.66079, + "tags": [ + "ullamco", + "non", + "officia", + "eiusmod", + "duis", + "cupidatat", + "mollit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Hendricks Logan" + }, + { + "id": 1, + "name": "Dolly Baird" + }, + { + "id": 2, + "name": "Wendi Wallace" + } + ], + "greeting": "Hello, Joyce! You have 6 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa82176dab97e42086b90", + "index": 72, + "guid": "9680920d-9303-471b-849e-c30e38e06d45", + "isActive": false, + "balance": "$2,696.40", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "blue", + "name": { + "first": "Felecia", + "last": "Gonzalez" + }, + "company": "CORIANDER", + "email": "felecia.gonzalez@coriander.net", + "phone": "+1 (923) 575-3582", + "address": "315 Borinquen Pl, Rosburg, Pennsylvania, 9619", + "about": "Laboris officia exercitation duis aliqua in sint consectetur. Ad ut labore ipsum ipsum ut culpa labore irure aute. Et exercitation tempor do dolor culpa ad ipsum pariatur adipisicing fugiat. Incididunt amet incididunt minim quis cupidatat pariatur enim fugiat reprehenderit est ipsum labore.\r\n", + "registered": "Monday, January 20, 2014 11:59 PM", + "latitude": -36.201421, + "longitude": 162.994705, + "tags": [ + "cillum", + "reprehenderit", + "non", + "est", + "tempor", + "exercitation", + "fugiat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Gould Waters" + }, + { + "id": 1, + "name": "Ramona Coffey" + }, + { + "id": 2, + "name": "Taylor Byers" + } + ], + "greeting": "Hello, Felecia! You have 10 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821df55849ccd4ca74c", + "index": 73, + "guid": "01b7c49a-6dac-4b65-906b-4483de07a5e8", + "isActive": true, + "balance": "$2,037.32", + "picture": "http://placehold.it/32x32", + "age": 27, + "eyeColor": "blue", + "name": { + "first": "Stanton", + "last": "Rutledge" + }, + "company": "EXOBLUE", + "email": "stanton.rutledge@exoblue.com", + "phone": "+1 (817) 408-2566", + "address": "741 Crooke Avenue, Newry, Nevada, 8555", + "about": "Commodo adipisicing ea ipsum non irure quis excepteur. Qui laborum qui sit cupidatat dolore consectetur tempor esse occaecat cillum qui. Dolore in amet ea ex proident do nulla.\r\n", + "registered": "Thursday, August 28, 2014 8:35 PM", + "latitude": 87.299409, + "longitude": 22.167535, + "tags": [ + "commodo", + "aliqua", + "irure", + "ea", + "dolor", + "aute", + "non" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Best Klein" + }, + { + "id": 1, + "name": "Dickerson Mcknight" + }, + { + "id": 2, + "name": "Gayle Washington" + } + ], + "greeting": "Hello, Stanton! You have 6 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821f3516c63c6fa05e8", + "index": 74, + "guid": "c299ecf4-0528-4280-b6b4-909801b1e9dd", + "isActive": true, + "balance": "$1,695.83", + "picture": "http://placehold.it/32x32", + "age": 27, + "eyeColor": "green", + "name": { + "first": "Thelma", + "last": "Barnett" + }, + "company": "DUOFLEX", + "email": "thelma.barnett@duoflex.biz", + "phone": "+1 (947) 416-2234", + "address": "325 Tampa Court, Zortman, Illinois, 3609", + "about": "Exercitation esse culpa enim anim. Cillum voluptate quis tempor excepteur elit aliquip consequat officia cupidatat laborum ad cupidatat. Mollit exercitation esse fugiat do do id irure tempor et duis. Ut sit reprehenderit velit sit eiusmod in officia nisi commodo magna id in. Lorem sint velit adipisicing aute.\r\n", + "registered": "Tuesday, April 22, 2014 5:11 AM", + "latitude": 20.61329, + "longitude": 48.592063, + "tags": [ + "et", + "excepteur", + "nostrud", + "ullamco", + "quis", + "occaecat", + "in" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Amy Parks" + }, + { + "id": 1, + "name": "Greene Nunez" + }, + { + "id": 2, + "name": "Janna Roth" + } + ], + "greeting": "Hello, Thelma! You have 6 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821637950981c0fa4c8", + "index": 75, + "guid": "bc58930c-58d8-4223-8a98-20295ac61c4e", + "isActive": true, + "balance": "$1,836.33", + "picture": "http://placehold.it/32x32", + "age": 23, + "eyeColor": "green", + "name": { + "first": "Nunez", + "last": "Freeman" + }, + "company": "ZILCH", + "email": "nunez.freeman@zilch.co.uk", + "phone": "+1 (959) 439-2497", + "address": "729 Varick Street, Marne, Tennessee, 1687", + "about": "Sunt nulla ipsum non in nulla. Dolore in fugiat in laborum veniam enim dolore cupidatat. Elit tempor ullamco id in minim excepteur et aute ut mollit aliquip qui consequat. Duis esse magna culpa aliqua ad ipsum deserunt laborum amet sint. Quis voluptate quis quis aute.\r\n", + "registered": "Monday, August 11, 2014 5:27 PM", + "latitude": 85.57657, + "longitude": -145.772132, + "tags": [ + "ullamco", + "tempor", + "et", + "non", + "magna", + "non", + "ex" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Latasha Randolph" + }, + { + "id": 1, + "name": "Neva Porter" + }, + { + "id": 2, + "name": "Drake Nicholson" + } + ], + "greeting": "Hello, Nunez! You have 10 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821ed9d29afc180b718", + "index": 76, + "guid": "45fb73c9-30e9-4e61-b78d-41c8f79a282e", + "isActive": false, + "balance": "$1,298.91", + "picture": "http://placehold.it/32x32", + "age": 25, + "eyeColor": "green", + "name": { + "first": "Bentley", + "last": "Reyes" + }, + "company": "ZYPLE", + "email": "bentley.reyes@zyple.tv", + "phone": "+1 (919) 510-3585", + "address": "227 Oakland Place, Farmers, Iowa, 9827", + "about": "Cillum proident eiusmod id amet anim laboris elit sint ea et non. Aliqua et reprehenderit amet est ea fugiat aute. Minim aute aliquip nulla elit. Duis ad exercitation excepteur laborum anim occaecat nulla sunt. Quis pariatur nulla Lorem consectetur proident sunt amet est et elit eu sunt. Ut irure voluptate consequat amet sint deserunt quis. Incididunt ea culpa commodo fugiat qui veniam quis Lorem incididunt dolor.\r\n", + "registered": "Wednesday, April 9, 2014 1:19 AM", + "latitude": -82.587336, + "longitude": -74.931056, + "tags": [ + "dolore", + "nisi", + "exercitation", + "ullamco", + "excepteur", + "qui", + "ut" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Nadia Pearson" + }, + { + "id": 1, + "name": "Reba Frost" + }, + { + "id": 2, + "name": "Lilia Mcbride" + } + ], + "greeting": "Hello, Bentley! You have 7 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821a936a115315db0b9", + "index": 77, + "guid": "a1b82517-f6a6-437a-9f2d-2c0043591cfa", + "isActive": false, + "balance": "$1,016.00", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "blue", + "name": { + "first": "Morales", + "last": "Shields" + }, + "company": "CORECOM", + "email": "morales.shields@corecom.name", + "phone": "+1 (814) 407-2079", + "address": "472 Coleridge Street, Shasta, Massachusetts, 501", + "about": "Ipsum elit adipisicing nostrud quis ut nisi ullamco consectetur ex laborum reprehenderit anim magna fugiat. In nulla dolor esse exercitation elit exercitation. Laborum sit esse velit magna ea irure est ut velit id nisi sint qui. Laborum voluptate amet cupidatat laborum aute in id duis irure. Enim aliqua enim magna ut consequat. Tempor est commodo elit eu et ut occaecat culpa ex ex. Ullamco cillum sint ipsum tempor anim ullamco sit pariatur excepteur dolor mollit ad quis.\r\n", + "registered": "Saturday, July 5, 2014 5:10 AM", + "latitude": -20.583102, + "longitude": -23.34973, + "tags": [ + "nulla", + "proident", + "enim", + "dolor", + "elit", + "excepteur", + "dolore" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Lacey Perry" + }, + { + "id": 1, + "name": "Stokes Foley" + }, + { + "id": 2, + "name": "Trisha Morales" + } + ], + "greeting": "Hello, Morales! You have 6 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa8214ceb22176ac554f9", + "index": 78, + "guid": "1aa9258a-f508-4e73-adb9-5d2cce0dff79", + "isActive": true, + "balance": "$1,100.42", + "picture": "http://placehold.it/32x32", + "age": 21, + "eyeColor": "blue", + "name": { + "first": "Chandra", + "last": "Patel" + }, + "company": "PROVIDCO", + "email": "chandra.patel@providco.me", + "phone": "+1 (875) 430-3869", + "address": "543 Bainbridge Street, Kidder, Arkansas, 3624", + "about": "Magna ad tempor commodo esse labore mollit sit. Duis laborum excepteur dolor officia exercitation. Elit sunt ex voluptate anim consectetur in ullamco mollit qui non. Cupidatat ipsum et cupidatat cupidatat consequat nulla Lorem culpa minim dolore cupidatat ipsum sint.\r\n", + "registered": "Sunday, June 29, 2014 1:41 AM", + "latitude": 55.150075, + "longitude": -26.185265, + "tags": [ + "amet", + "consectetur", + "occaecat", + "cillum", + "enim", + "ut", + "occaecat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Geneva Snider" + }, + { + "id": 1, + "name": "Rose Michael" + }, + { + "id": 2, + "name": "Kirkland Mason" + } + ], + "greeting": "Hello, Chandra! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821050884ba49ad868d", + "index": 79, + "guid": "88a0d238-e3c9-4b1a-8d19-08c622f9eaae", + "isActive": false, + "balance": "$1,263.98", + "picture": "http://placehold.it/32x32", + "age": 30, + "eyeColor": "green", + "name": { + "first": "Elsa", + "last": "Chambers" + }, + "company": "VICON", + "email": "elsa.chambers@vicon.us", + "phone": "+1 (940) 436-3956", + "address": "190 Albemarle Terrace, Sheatown, Hawaii, 2654", + "about": "Est do esse elit consectetur elit. Aliqua esse duis est sint non. Enim minim laborum ad duis. Proident laboris quis ea amet nulla occaecat ex laboris duis ut velit.\r\n", + "registered": "Saturday, April 19, 2014 3:28 AM", + "latitude": 62.679122, + "longitude": 95.229313, + "tags": [ + "magna", + "ipsum", + "Lorem", + "ut", + "duis", + "elit", + "sit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Laurie Fuentes" + }, + { + "id": 1, + "name": "Juana Blevins" + }, + { + "id": 2, + "name": "Virginia Hester" + } + ], + "greeting": "Hello, Elsa! You have 6 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821eb0dd28cfdc525a9", + "index": 80, + "guid": "e1574006-8d49-4399-8b59-92eaa0ed2be1", + "isActive": true, + "balance": "$2,395.40", + "picture": "http://placehold.it/32x32", + "age": 23, + "eyeColor": "brown", + "name": { + "first": "Morris", + "last": "Trujillo" + }, + "company": "ZOGAK", + "email": "morris.trujillo@zogak.ca", + "phone": "+1 (919) 553-3453", + "address": "127 Clove Road, Keyport, Puerto Rico, 5969", + "about": "Minim do veniam non elit excepteur enim sunt excepteur non amet. Duis eu consequat adipisicing nostrud ad cupidatat tempor occaecat. Ea id consectetur non anim. Irure dolor qui excepteur anim excepteur adipisicing ea sint id aliquip consequat eu id. Ea exercitation officia nostrud ipsum. Voluptate sunt irure aliqua tempor pariatur irure labore ea amet. Quis reprehenderit aliqua pariatur esse id anim laborum ullamco amet.\r\n", + "registered": "Sunday, February 2, 2014 12:52 PM", + "latitude": -86.963921, + "longitude": -157.636932, + "tags": [ + "occaecat", + "amet", + "ex", + "ea", + "exercitation", + "aliqua", + "elit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Lester Watkins" + }, + { + "id": 1, + "name": "Kellie Clayton" + }, + { + "id": 2, + "name": "Valencia Edwards" + } + ], + "greeting": "Hello, Morris! You have 6 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821f5f0ae13a893ce36", + "index": 81, + "guid": "9dd75e1e-8f3c-473e-800e-518254719ca1", + "isActive": true, + "balance": "$1,956.52", + "picture": "http://placehold.it/32x32", + "age": 36, + "eyeColor": "brown", + "name": { + "first": "Gwen", + "last": "Park" + }, + "company": "INTERGEEK", + "email": "gwen.park@intergeek.org", + "phone": "+1 (830) 581-3561", + "address": "514 Beayer Place, Maxville, New Hampshire, 4162", + "about": "Consequat labore commodo nulla veniam aliqua. Tempor ipsum officia exercitation amet elit dolor labore eu voluptate cillum reprehenderit exercitation proident. Id cillum laborum cupidatat reprehenderit anim cillum exercitation culpa aliqua deserunt cupidatat. In proident ea eu nisi est enim. Quis dolor sit aliquip reprehenderit in id ipsum proident duis. Sit eu sint nisi velit. Minim elit nostrud aliquip anim.\r\n", + "registered": "Wednesday, September 17, 2014 9:38 AM", + "latitude": -66.199245, + "longitude": -52.824656, + "tags": [ + "ex", + "magna", + "incididunt", + "veniam", + "mollit", + "eu", + "sunt" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Nieves Potter" + }, + { + "id": 1, + "name": "Welch Reeves" + }, + { + "id": 2, + "name": "Hardy Forbes" + } + ], + "greeting": "Hello, Gwen! You have 8 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821a20d620a7ec793ac", + "index": 82, + "guid": "1ae142ff-8d6a-4e72-a3f1-3ac4349ad0b7", + "isActive": false, + "balance": "$2,560.11", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "green", + "name": { + "first": "Erickson", + "last": "Lancaster" + }, + "company": "SCENTRIC", + "email": "erickson.lancaster@scentric.biz", + "phone": "+1 (980) 465-3465", + "address": "412 Jackson Street, Chumuckla, South Carolina, 3216", + "about": "Id laborum consectetur pariatur non nulla incididunt labore magna minim duis. Ipsum laboris deserunt velit sunt voluptate. Laboris ipsum duis aliquip non aliqua. Commodo veniam mollit voluptate elit nisi nostrud laboris dolor tempor pariatur duis laborum.\r\n", + "registered": "Wednesday, May 7, 2014 8:50 PM", + "latitude": -57.590749, + "longitude": 117.31662, + "tags": [ + "ad", + "elit", + "non", + "fugiat", + "laborum", + "incididunt", + "enim" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Reeves Nguyen" + }, + { + "id": 1, + "name": "Jo Christian" + }, + { + "id": 2, + "name": "Myers Lowe" + } + ], + "greeting": "Hello, Erickson! You have 10 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa8213747fd3443999762", + "index": 83, + "guid": "ec870637-4e65-49c4-abfb-7b99c2d2f094", + "isActive": true, + "balance": "$2,079.21", + "picture": "http://placehold.it/32x32", + "age": 21, + "eyeColor": "blue", + "name": { + "first": "Saundra", + "last": "Kemp" + }, + "company": "LUNCHPOD", + "email": "saundra.kemp@lunchpod.io", + "phone": "+1 (950) 433-2550", + "address": "593 Stuart Street, Edenburg, Texas, 6251", + "about": "Ipsum proident et duis reprehenderit in minim in sint dolore enim aute excepteur cillum eiusmod. Do nulla deserunt quis adipisicing sunt reprehenderit. Dolor pariatur tempor sint ullamco.\r\n", + "registered": "Tuesday, June 24, 2014 6:44 AM", + "latitude": 81.565149, + "longitude": -92.061448, + "tags": [ + "magna", + "commodo", + "esse", + "ad", + "labore", + "amet", + "mollit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Tommie Mays" + }, + { + "id": 1, + "name": "Lou Hubbard" + }, + { + "id": 2, + "name": "Jenna Gentry" + } + ], + "greeting": "Hello, Saundra! You have 9 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821524a5fd2a5f037a2", + "index": 84, + "guid": "e7894fe9-efd4-44ae-afd5-49bfde09b000", + "isActive": true, + "balance": "$2,320.08", + "picture": "http://placehold.it/32x32", + "age": 37, + "eyeColor": "green", + "name": { + "first": "Christensen", + "last": "Wolf" + }, + "company": "GLUKGLUK", + "email": "christensen.wolf@glukgluk.net", + "phone": "+1 (937) 519-3107", + "address": "460 Porter Avenue, Irwin, South Dakota, 2498", + "about": "Ullamco reprehenderit non aliquip amet do deserunt nostrud ea deserunt fugiat. Commodo pariatur est officia dolor dolore in excepteur pariatur laborum ut. Occaecat cillum ullamco nulla eu esse non nisi pariatur ipsum amet do ea ad culpa. Excepteur esse elit laborum deserunt ad consequat. Culpa esse esse nostrud commodo laborum officia sint ea mollit. Dolor culpa pariatur fugiat cillum quis proident non enim esse pariatur duis adipisicing amet. Consequat minim aliqua enim excepteur.\r\n", + "registered": "Saturday, February 8, 2014 10:20 PM", + "latitude": 22.478522, + "longitude": 30.070646, + "tags": [ + "quis", + "ad", + "adipisicing", + "exercitation", + "non", + "eu", + "anim" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Bright Moon" + }, + { + "id": 1, + "name": "Stephenson Sears" + }, + { + "id": 2, + "name": "Fletcher Swanson" + } + ], + "greeting": "Hello, Christensen! You have 9 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa82160fa0826e90b747b", + "index": 85, + "guid": "af08cc83-e8c9-4bcd-84b3-2f3251cf9a02", + "isActive": true, + "balance": "$2,575.60", + "picture": "http://placehold.it/32x32", + "age": 29, + "eyeColor": "brown", + "name": { + "first": "Weaver", + "last": "Parker" + }, + "company": "RETROTEX", + "email": "weaver.parker@retrotex.com", + "phone": "+1 (951) 411-2545", + "address": "560 Madison Place, Sidman, Ohio, 5153", + "about": "Reprehenderit esse dolor tempor consectetur occaecat exercitation consectetur irure commodo. Et eiusmod aute ipsum eu commodo qui id anim proident. Excepteur labore laboris aliqua incididunt ut nisi consequat deserunt dolore officia velit sint consectetur. Laboris sint minim mollit duis. Excepteur nostrud incididunt aute consequat ad magna eu quis. Id dolor eu aliqua deserunt cillum sint ea et nostrud. Lorem amet cillum nulla tempor commodo mollit aliquip do est enim enim.\r\n", + "registered": "Thursday, February 6, 2014 12:55 PM", + "latitude": 87.778566, + "longitude": -122.687026, + "tags": [ + "eu", + "voluptate", + "commodo", + "magna", + "ullamco", + "nulla", + "ea" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Ratliff Mckinney" + }, + { + "id": 1, + "name": "Walker Frye" + }, + { + "id": 2, + "name": "Mcgowan Daniel" + } + ], + "greeting": "Hello, Weaver! You have 5 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8217a543d20bd4b10c1", + "index": 86, + "guid": "5b670ab7-4ee6-4b8e-b379-4e1849b6e329", + "isActive": false, + "balance": "$3,114.39", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "green", + "name": { + "first": "Annabelle", + "last": "Sanders" + }, + "company": "PHORMULA", + "email": "annabelle.sanders@phormula.biz", + "phone": "+1 (982) 489-2678", + "address": "970 Llama Court, Moraida, Maine, 8694", + "about": "Nostrud adipisicing magna nulla magna voluptate duis eu voluptate cupidatat ut dolore excepteur esse dolor. Aliquip exercitation occaecat amet excepteur sit. Velit adipisicing esse labore veniam duis ullamco in ea. Adipisicing eiusmod cillum veniam nostrud sint laboris sit id officia. Esse esse anim sint do ea id. Esse ipsum mollit sit laborum nostrud mollit nulla id.\r\n", + "registered": "Tuesday, January 7, 2014 7:34 AM", + "latitude": 9.515348, + "longitude": -99.138606, + "tags": [ + "ipsum", + "sint", + "dolor", + "laborum", + "est", + "consequat", + "magna" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Juliet Clements" + }, + { + "id": 1, + "name": "Jeannine Pruitt" + }, + { + "id": 2, + "name": "Chambers Warren" + } + ], + "greeting": "Hello, Annabelle! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa82179c1fe89e1ccec4d", + "index": 87, + "guid": "31ecaae4-6443-4c00-a20d-82100c49b488", + "isActive": true, + "balance": "$3,803.83", + "picture": "http://placehold.it/32x32", + "age": 40, + "eyeColor": "blue", + "name": { + "first": "Kelley", + "last": "Miles" + }, + "company": "AQUASURE", + "email": "kelley.miles@aquasure.co.uk", + "phone": "+1 (819) 529-2967", + "address": "680 Monument Walk, Wakulla, Vermont, 8903", + "about": "Ea sint dolor nostrud dolor id commodo esse nisi. Reprehenderit minim dolore nostrud sint incididunt excepteur reprehenderit enim velit velit. Proident officia velit Lorem dolore ullamco occaecat.\r\n", + "registered": "Saturday, May 3, 2014 12:23 AM", + "latitude": 73.767872, + "longitude": -118.631186, + "tags": [ + "consectetur", + "irure", + "nostrud", + "nostrud", + "aliquip", + "quis", + "reprehenderit" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Maynard Townsend" + }, + { + "id": 1, + "name": "Carlene Molina" + }, + { + "id": 2, + "name": "Mai Bentley" + } + ], + "greeting": "Hello, Kelley! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8218de20c1f1e1e93fa", + "index": 88, + "guid": "8f6b0aac-f2ba-45aa-a7c8-76c413bdeb7a", + "isActive": true, + "balance": "$1,898.86", + "picture": "http://placehold.it/32x32", + "age": 38, + "eyeColor": "brown", + "name": { + "first": "Mckay", + "last": "Velasquez" + }, + "company": "NORALEX", + "email": "mckay.velasquez@noralex.tv", + "phone": "+1 (973) 599-3463", + "address": "152 Roebling Street, Mathews, Delaware, 7958", + "about": "Nostrud esse dolor excepteur cillum aliqua. Ea nulla elit minim sint non culpa id. Et ullamco aute laborum incididunt sint quis. Tempor tempor aliqua in sunt. Minim elit dolor quis excepteur exercitation adipisicing. Pariatur incididunt tempor irure proident exercitation deserunt sint.\r\n", + "registered": "Monday, August 4, 2014 5:00 AM", + "latitude": 81.463276, + "longitude": -66.291508, + "tags": [ + "nisi", + "anim", + "qui", + "est", + "qui", + "ipsum", + "ullamco" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Myrna Rollins" + }, + { + "id": 1, + "name": "Tricia Gilliam" + }, + { + "id": 2, + "name": "Collins Obrien" + } + ], + "greeting": "Hello, Mckay! You have 10 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821b3490524365bc954", + "index": 89, + "guid": "43825015-fed6-40c8-bd80-8aaf228067f9", + "isActive": false, + "balance": "$3,035.21", + "picture": "http://placehold.it/32x32", + "age": 40, + "eyeColor": "blue", + "name": { + "first": "Brandy", + "last": "Hayden" + }, + "company": "OMNIGOG", + "email": "brandy.hayden@omnigog.name", + "phone": "+1 (827) 481-2334", + "address": "537 Pioneer Street, Saticoy, Palau, 803", + "about": "Ea velit consectetur ipsum ut elit pariatur id labore sunt eu incididunt est aliqua. Veniam esse officia do non officia cupidatat proident id officia esse tempor non mollit dolore. Elit voluptate nulla exercitation laboris ex ad irure est do enim aute velit aute. Cillum adipisicing nisi dolor velit duis ad fugiat deserunt non commodo Lorem fugiat sint qui. Aute consectetur magna incididunt tempor in esse consectetur magna qui sit. Culpa consequat laborum duis adipisicing dolor in deserunt ut velit ea ex dolore ullamco esse.\r\n", + "registered": "Wednesday, February 12, 2014 7:32 PM", + "latitude": 4.319892, + "longitude": 81.048442, + "tags": [ + "labore", + "ullamco", + "commodo", + "quis", + "aute", + "nulla", + "do" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Riggs Flynn" + }, + { + "id": 1, + "name": "Kidd Guerrero" + }, + { + "id": 2, + "name": "Rosario Wade" + } + ], + "greeting": "Hello, Brandy! You have 10 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8211b737d086264c9a9", + "index": 90, + "guid": "ec5fe190-d16f-4728-bce1-c5140852c583", + "isActive": true, + "balance": "$2,932.98", + "picture": "http://placehold.it/32x32", + "age": 28, + "eyeColor": "green", + "name": { + "first": "Sonia", + "last": "Orr" + }, + "company": "MUSANPOLY", + "email": "sonia.orr@musanpoly.me", + "phone": "+1 (822) 422-2010", + "address": "908 Dekalb Avenue, Elfrida, Arizona, 6925", + "about": "Ut cillum ex irure amet aliquip voluptate Lorem fugiat reprehenderit sunt reprehenderit quis. Nulla laborum sunt elit ad labore ut cupidatat cillum eiusmod est. Ea irure amet excepteur mollit eu ipsum id adipisicing occaecat. Cupidatat sunt do veniam esse enim sint qui voluptate sint. Qui officia ad cupidatat mollit laboris. Duis tempor fugiat ea mollit cupidatat exercitation sunt incididunt.\r\n", + "registered": "Thursday, September 25, 2014 8:21 PM", + "latitude": 71.876999, + "longitude": 79.322401, + "tags": [ + "pariatur", + "sit", + "culpa", + "dolore", + "cupidatat", + "minim", + "cillum" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Morgan Pittman" + }, + { + "id": 1, + "name": "Bullock Cannon" + }, + { + "id": 2, + "name": "Lakeisha Lynch" + } + ], + "greeting": "Hello, Sonia! You have 9 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821b271c92236e5e9b0", + "index": 91, + "guid": "57cff12c-2a18-48c2-b15a-3c393de707b6", + "isActive": false, + "balance": "$1,874.97", + "picture": "http://placehold.it/32x32", + "age": 40, + "eyeColor": "green", + "name": { + "first": "Stephens", + "last": "Whitaker" + }, + "company": "EARTHPLEX", + "email": "stephens.whitaker@earthplex.us", + "phone": "+1 (960) 419-2183", + "address": "368 Division Avenue, Fivepointville, Colorado, 8609", + "about": "Do aliquip laboris irure consectetur esse reprehenderit. Cillum ex deserunt fugiat ut dolore excepteur culpa eiusmod sit ullamco velit consequat consequat aliquip. Nostrud officia est enim velit fugiat laboris.\r\n", + "registered": "Monday, May 26, 2014 8:04 PM", + "latitude": 7.153481, + "longitude": 100.823578, + "tags": [ + "laborum", + "irure", + "dolore", + "enim", + "nisi", + "cillum", + "deserunt" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Perkins Campos" + }, + { + "id": 1, + "name": "Murray Randall" + }, + { + "id": 2, + "name": "Wallace Blackwell" + } + ], + "greeting": "Hello, Stephens! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821c115ee94a728efa4", + "index": 92, + "guid": "7bbe7ace-73f9-4d25-8f20-b50f4b9d4e60", + "isActive": true, + "balance": "$1,168.18", + "picture": "http://placehold.it/32x32", + "age": 33, + "eyeColor": "green", + "name": { + "first": "Shepard", + "last": "Sanchez" + }, + "company": "YOGASM", + "email": "shepard.sanchez@yogasm.ca", + "phone": "+1 (959) 436-3299", + "address": "154 Seeley Street, Witmer, Missouri, 994", + "about": "Adipisicing ut id nulla occaecat enim officia reprehenderit non magna dolor Lorem. Culpa ad proident duis cupidatat nostrud occaecat esse elit pariatur quis Lorem. Velit exercitation anim dolore nisi labore consequat cupidatat magna nostrud sint ut deserunt enim.\r\n", + "registered": "Saturday, February 22, 2014 10:23 PM", + "latitude": -81.429217, + "longitude": -27.374426, + "tags": [ + "aliqua", + "minim", + "consequat", + "aliqua", + "qui", + "proident", + "consequat" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Sara Bruce" + }, + { + "id": 1, + "name": "Kimberley Mcdaniel" + }, + { + "id": 2, + "name": "Tammi England" + } + ], + "greeting": "Hello, Shepard! You have 7 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821a094fdd2592addbd", + "index": 93, + "guid": "69bc8858-bd55-4fa0-95f9-d2cb680a390c", + "isActive": false, + "balance": "$1,746.96", + "picture": "http://placehold.it/32x32", + "age": 35, + "eyeColor": "blue", + "name": { + "first": "Kristy", + "last": "Johnston" + }, + "company": "KONGLE", + "email": "kristy.johnston@kongle.org", + "phone": "+1 (823) 558-3033", + "address": "122 Cadman Plaza, Suitland, Minnesota, 3918", + "about": "Laboris excepteur ea nostrud incididunt est laborum dolor. Consequat ad duis aute proident incididunt commodo adipisicing. Enim voluptate sunt et est excepteur eiusmod commodo. Mollit fugiat reprehenderit ex ullamco magna laboris commodo mollit. Cupidatat tempor tempor minim dolore. Excepteur ipsum esse ipsum nulla.\r\n", + "registered": "Wednesday, January 29, 2014 11:15 PM", + "latitude": 15.335329, + "longitude": -78.001472, + "tags": [ + "veniam", + "consequat", + "ex", + "anim", + "culpa", + "ullamco", + "ex" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Johns Cochran" + }, + { + "id": 1, + "name": "Sheppard Hicks" + }, + { + "id": 2, + "name": "Cervantes Donovan" + } + ], + "greeting": "Hello, Kristy! You have 10 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821311b6242dc75f718", + "index": 94, + "guid": "9de76ab8-2ef7-4b46-a478-5dea8732650a", + "isActive": false, + "balance": "$2,549.05", + "picture": "http://placehold.it/32x32", + "age": 40, + "eyeColor": "blue", + "name": { + "first": "Melendez", + "last": "Golden" + }, + "company": "GRONK", + "email": "melendez.golden@gronk.biz", + "phone": "+1 (986) 553-2124", + "address": "647 Broadway , Reno, Northern Mariana Islands, 9487", + "about": "Aliquip pariatur deserunt culpa eu nostrud eu amet. Adipisicing elit occaecat aliqua ipsum ut. Cillum magna consectetur elit esse sint laboris duis. In ea enim aute eiusmod culpa. Labore quis sunt aliquip excepteur tempor irure amet consequat eu excepteur et occaecat.\r\n", + "registered": "Monday, April 14, 2014 9:41 AM", + "latitude": 55.489443, + "longitude": 3.620039, + "tags": [ + "pariatur", + "magna", + "quis", + "in", + "irure", + "amet", + "esse" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Hilary Dennis" + }, + { + "id": 1, + "name": "Tyler Burgess" + }, + { + "id": 2, + "name": "Eugenia Donaldson" + } + ], + "greeting": "Hello, Melendez! You have 9 unread messages.", + "favoriteFruit": "strawberry" + }, + { + "_id": "543fa821c394a9871527c834", + "index": 95, + "guid": "ec0f31d4-7264-4649-9af9-2716082915d3", + "isActive": true, + "balance": "$2,810.24", + "picture": "http://placehold.it/32x32", + "age": 34, + "eyeColor": "brown", + "name": { + "first": "Eleanor", + "last": "Curtis" + }, + "company": "ZAPPIX", + "email": "eleanor.curtis@zappix.io", + "phone": "+1 (824) 461-3776", + "address": "981 Lyme Avenue, Oneida, Marshall Islands, 6015", + "about": "Commodo labore do pariatur exercitation voluptate ea velit velit qui ex duis. Commodo est excepteur ad labore aute enim eu. Adipisicing irure exercitation sint consectetur exercitation occaecat aute exercitation commodo.\r\n", + "registered": "Monday, May 5, 2014 7:51 AM", + "latitude": -5.604007, + "longitude": -125.532894, + "tags": [ + "aliqua", + "commodo", + "non", + "magna", + "quis", + "velit", + "irure" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Monica Carr" + }, + { + "id": 1, + "name": "Coleman Simpson" + }, + { + "id": 2, + "name": "Wilma Fisher" + } + ], + "greeting": "Hello, Eleanor! You have 7 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa821a5b5e35d24f265b3", + "index": 96, + "guid": "c865c0ff-8505-4d02-af0a-c5a275d03fd5", + "isActive": true, + "balance": "$3,192.24", + "picture": "http://placehold.it/32x32", + "age": 22, + "eyeColor": "blue", + "name": { + "first": "Terrell", + "last": "Greene" + }, + "company": "AUTOGRATE", + "email": "terrell.greene@autograte.net", + "phone": "+1 (933) 557-3825", + "address": "435 Bridgewater Street, Fresno, New Jersey, 8124", + "about": "Elit sit ut anim reprehenderit anim ipsum esse tempor aliqua id ullamco. Exercitation est velit aliquip est elit mollit magna velit. Elit eiusmod esse voluptate consectetur enim id exercitation adipisicing et laborum. Irure fugiat ut sint exercitation dolor nostrud qui mollit eiusmod cupidatat. Commodo ad dolore incididunt ex quis nostrud veniam pariatur aliquip ea reprehenderit reprehenderit.\r\n", + "registered": "Sunday, September 21, 2014 3:31 AM", + "latitude": 76.638623, + "longitude": 147.966829, + "tags": [ + "aliquip", + "qui", + "nisi", + "ut", + "non", + "proident", + "et" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Lucile Jordan" + }, + { + "id": 1, + "name": "Bender Sheppard" + }, + { + "id": 2, + "name": "Milagros Francis" + } + ], + "greeting": "Hello, Terrell! You have 8 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa8216cc444257e522ec0", + "index": 97, + "guid": "90c19881-6521-4012-aba7-f4d143953965", + "isActive": false, + "balance": "$1,509.96", + "picture": "http://placehold.it/32x32", + "age": 31, + "eyeColor": "green", + "name": { + "first": "Holman", + "last": "Hines" + }, + "company": "NORSUP", + "email": "holman.hines@norsup.com", + "phone": "+1 (925) 565-2825", + "address": "902 Monroe Street, Marienthal, Virgin Islands, 2354", + "about": "Eu cupidatat consectetur labore voluptate nulla non amet incididunt labore non magna tempor. Veniam et sint qui reprehenderit reprehenderit sint ut laboris elit. Pariatur in eu dolore culpa nisi laboris officia magna do velit. Cupidatat proident excepteur officia labore irure sunt elit velit dolor commodo. Minim quis sint minim non incididunt Lorem elit cupidatat adipisicing quis esse non sint et. Laborum culpa incididunt incididunt fugiat minim ex deserunt et.\r\n", + "registered": "Sunday, April 27, 2014 8:05 PM", + "latitude": 61.825518, + "longitude": -28.393161, + "tags": [ + "voluptate", + "veniam", + "Lorem", + "ex", + "in", + "est", + "exercitation" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Cecelia Chapman" + }, + { + "id": 1, + "name": "Isabella Castaneda" + }, + { + "id": 2, + "name": "Montoya Chen" + } + ], + "greeting": "Hello, Holman! You have 9 unread messages.", + "favoriteFruit": "apple" + }, + { + "_id": "543fa8210659aa5cdca41f7e", + "index": 98, + "guid": "81adf15d-24b6-454c-a75f-f5be12f6ac75", + "isActive": false, + "balance": "$2,564.95", + "picture": "http://placehold.it/32x32", + "age": 39, + "eyeColor": "green", + "name": { + "first": "Horton", + "last": "Poole" + }, + "company": "GROK", + "email": "horton.poole@grok.biz", + "phone": "+1 (908) 520-3683", + "address": "141 Gatling Place, Hackneyville, Oregon, 2976", + "about": "Consequat nisi reprehenderit incididunt id minim cillum. Lorem reprehenderit fugiat irure dolor excepteur velit. Est nostrud culpa ut reprehenderit in duis id voluptate pariatur voluptate. Exercitation Lorem esse exercitation Lorem esse. Excepteur mollit sit ut voluptate ipsum pariatur anim sint sunt cillum sit consequat.\r\n", + "registered": "Thursday, September 18, 2014 6:11 AM", + "latitude": -3.525694, + "longitude": -103.351985, + "tags": [ + "deserunt", + "voluptate", + "cillum", + "id", + "magna", + "deserunt", + "incididunt" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Hope Lara" + }, + { + "id": 1, + "name": "French Garner" + }, + { + "id": 2, + "name": "Stein Sykes" + } + ], + "greeting": "Hello, Horton! You have 5 unread messages.", + "favoriteFruit": "banana" + }, + { + "_id": "543fa821cf1b15f13d8c3938", + "index": 99, + "guid": "44533089-c11a-4656-b2d1-9ab6be887f30", + "isActive": false, + "balance": "$3,192.31", + "picture": "http://placehold.it/32x32", + "age": 37, + "eyeColor": "green", + "name": { + "first": "Wanda", + "last": "Wiggins" + }, + "company": "ZEPITOPE", + "email": "wanda.wiggins@zepitope.co.uk", + "phone": "+1 (998) 461-3780", + "address": "427 Canton Court, Heil, Utah, 8283", + "about": "Do officia et exercitation dolor esse. Ut nisi eiusmod dolore laborum ad ex mollit minim. Pariatur qui culpa ullamco ex eiusmod.\r\n", + "registered": "Tuesday, August 26, 2014 6:59 PM", + "latitude": 44.770809, + "longitude": 150.936963, + "tags": [ + "consectetur", + "mollit", + "laborum", + "ipsum", + "quis", + "cupidatat", + "in" + ], + "range": [ + 0, + 1, + 2, + 3, + 4, + 5, + 6, + 7, + 8, + 9 + ], + "friends": [ + { + "id": 0, + "name": "Kristie Cain" + }, + { + "id": 1, + "name": "Geraldine Zimmerman" + }, + { + "id": 2, + "name": "Ingrid Harper" + } + ], + "greeting": "Hello, Wanda! You have 5 unread messages.", + "favoriteFruit": "strawberry" + } +] \ No newline at end of file diff --git a/github.com/klauspost/compress/huff0/.gitignore b/github.com/klauspost/compress/huff0/.gitignore new file mode 100644 index 0000000..b3d2629 --- /dev/null +++ b/github.com/klauspost/compress/huff0/.gitignore @@ -0,0 +1 @@ +/huff0-fuzz.zip diff --git a/github.com/klauspost/compress/huff0/README.md b/github.com/klauspost/compress/huff0/README.md new file mode 100644 index 0000000..ec4f909 --- /dev/null +++ b/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/github.com/klauspost/compress/huff0/bitreader.go b/github.com/klauspost/compress/huff0/bitreader.go new file mode 100644 index 0000000..a4979e8 --- /dev/null +++ b/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/github.com/klauspost/compress/huff0/bitwriter.go b/github.com/klauspost/compress/huff0/bitwriter.go new file mode 100644 index 0000000..bda4021 --- /dev/null +++ b/github.com/klauspost/compress/huff0/bitwriter.go @@ -0,0 +1,197 @@ +// 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) + 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 + 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/github.com/klauspost/compress/huff0/bytereader.go b/github.com/klauspost/compress/huff0/bytereader.go new file mode 100644 index 0000000..50bcdf6 --- /dev/null +++ b/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/github.com/klauspost/compress/huff0/compress.go b/github.com/klauspost/compress/huff0/compress.go new file mode 100644 index 0000000..0843cb0 --- /dev/null +++ b/github.com/klauspost/compress/huff0/compress.go @@ -0,0 +1,651 @@ +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 == ReusePolicyPrefer && 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 + } + // 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/github.com/klauspost/compress/huff0/compress_test.go b/github.com/klauspost/compress/huff0/compress_test.go new file mode 100644 index 0000000..ac264a5 --- /dev/null +++ b/github.com/klauspost/compress/huff0/compress_test.go @@ -0,0 +1,718 @@ +package huff0 + +import ( + "bytes" + "fmt" + "io/ioutil" + "math/rand" + "path/filepath" + "strings" + "testing" + + "github.com/klauspost/compress/flate" + "github.com/klauspost/compress/zip" +) + +type inputFn func() ([]byte, error) + +var testfiles = []struct { + name string + fn inputFn + err1X error + err4X error +}{ + // Digits is the digits of the irrational number e. Its decimal representation + // does not repeat, but there are only 10 possible digits, so it should be + // reasonably compressible. + {name: "digits", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/e.txt") }}, + // gettysburg.txt is a small plain text. + {name: "gettysburg", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/gettysburg.txt") }}, + // Twain is Project Gutenberg's edition of Mark Twain's classic English novel. + {name: "twain", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/Mark.Twain-Tom.Sawyer.txt") }}, + // Random bytes + {name: "random", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/sharnd.out") }, err1X: ErrIncompressible, err4X: ErrIncompressible}, + // Low entropy + {name: "low-ent.10k", fn: func() ([]byte, error) { return []byte(strings.Repeat("1221", 10000)), nil }}, + // Super Low entropy + {name: "superlow-ent-10k", fn: func() ([]byte, error) { return []byte(strings.Repeat("1", 10000) + strings.Repeat("2", 500)), nil }}, + // Zero bytes + {name: "zeroes", fn: func() ([]byte, error) { return make([]byte, 10000), nil }, err1X: ErrUseRLE, err4X: ErrUseRLE}, + {name: "crash1", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash1.bin") }, err1X: ErrIncompressible, err4X: ErrIncompressible}, + {name: "crash2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash2.bin") }, err4X: ErrIncompressible}, + {name: "crash3", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/crash3.bin") }, err1X: ErrIncompressible, err4X: ErrIncompressible}, + {name: "endzerobits", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/endzerobits.bin") }, err1X: nil, err4X: ErrIncompressible}, + {name: "endnonzero", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/endnonzero.bin") }, err4X: ErrIncompressible}, + {name: "case1", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/case1.bin") }, err1X: nil}, + {name: "case2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/case2.bin") }, err1X: nil}, + {name: "case3", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/case3.bin") }, err1X: nil}, + {name: "pngdata.001", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/pngdata.bin") }, err1X: nil}, + {name: "normcount2", fn: func() ([]byte, error) { return ioutil.ReadFile("../testdata/normcount2.bin") }, err1X: nil}, +} + +type fuzzInput struct { + name string + fn inputFn +} + +// testfilesExtended is used for regression testing the decoder. +// These files are expected to fail, but not crash +var testfilesExtended []fuzzInput + +func init() { + data, err := ioutil.ReadFile("testdata/regression.zip") + if err != nil { + panic(err) + } + zr, err := zip.NewReader(bytes.NewReader(data), int64(len(data))) + if err != nil { + panic(err) + } + for _, tt := range zr.File { + if tt.UncompressedSize64 == 0 { + continue + } + rc, err := tt.Open() + if err != nil { + panic(err) + } + b, err := ioutil.ReadAll(rc) + if err != nil { + panic(err) + } + testfilesExtended = append(testfilesExtended, fuzzInput{ + name: filepath.Base(tt.Name), + fn: func() ([]byte, error) { + return b, nil + }, + }) + } +} + +func TestCompressRegression(t *testing.T) { + // Match the fuzz function + var testInput = func(t *testing.T, data []byte) int { + var enc Scratch + enc.WantLogLess = 5 + comp, _, err := Compress1X(data, &enc) + if err == ErrIncompressible || err == ErrUseRLE || err == ErrTooBig { + return 0 + } + if err != nil { + panic(err) + } + if len(comp) >= len(data)-len(data)>>enc.WantLogLess { + panic(fmt.Errorf("too large output provided. got %d, but should be < %d", len(comp), len(data)-len(data)>>enc.WantLogLess)) + } + + dec, remain, err := ReadTable(comp, nil) + if err != nil { + panic(err) + } + out, err := dec.Decompress1X(remain) + if err != nil { + t.Error(err) + } + if !bytes.Equal(out, data) { + t.Error("decompression 1x mismatch") + } + // Reuse as 4X + enc.Reuse = ReusePolicyAllow + comp, reUsed, err := Compress4X(data, &enc) + if err == ErrIncompressible || err == ErrUseRLE || err == ErrTooBig { + return 0 + } + if err != nil { + panic(err) + } + if len(comp) >= len(data)-len(data)>>enc.WantLogLess { + panic(fmt.Errorf("too large output provided. got %d, but should be < %d", len(comp), len(data)-len(data)>>enc.WantLogLess)) + } + + remain = comp + if !reUsed { + dec, remain, err = ReadTable(comp, dec) + if err != nil { + panic(err) + } + } + out, err = dec.Decompress4X(remain, len(data)) + if err != nil { + t.Error(err) + } + if !bytes.Equal(out, data) { + t.Error("decompression 4x with reuse mismatch") + } + + enc.Reuse = ReusePolicyNone + comp, reUsed, err = Compress4X(data, &enc) + if err == ErrIncompressible || err == ErrUseRLE || err == ErrTooBig { + return 0 + } + if err != nil { + panic(err) + } + if reUsed { + panic("reused when asked not to") + } + if len(comp) >= len(data)-len(data)>>enc.WantLogLess { + panic(fmt.Errorf("too large output provided. got %d, but should be < %d", len(comp), len(data)-len(data)>>enc.WantLogLess)) + } + + dec, remain, err = ReadTable(comp, dec) + if err != nil { + panic(err) + } + out, err = dec.Decompress4X(remain, len(data)) + if err != nil { + t.Error(err) + } + if !bytes.Equal(out, data) { + t.Error("decompression 4x mismatch") + } + + // Reuse as 1X + dec.Reuse = ReusePolicyAllow + comp, reUsed, err = Compress1X(data, &enc) + if err == ErrIncompressible || err == ErrUseRLE || err == ErrTooBig { + return 0 + } + if err != nil { + panic(err) + } + if len(comp) >= len(data)-len(data)>>enc.WantLogLess { + panic(fmt.Errorf("too large output provided. got %d, but should be < %d", len(comp), len(data)-len(data)>>enc.WantLogLess)) + } + + remain = comp + if !reUsed { + dec, remain, err = ReadTable(comp, dec) + if err != nil { + panic(err) + } + } + out, err = dec.Decompress1X(remain) + if err != nil { + t.Error(err) + } + if !bytes.Equal(out, data) { + t.Error("decompression 1x with reuse mismatch") + } + return 1 + } + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + testInput(t, buf0) + }) + } + for _, test := range testfilesExtended { + t.Run(test.name, func(t *testing.T) { + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + testInput(t, buf0) + }) + } +} + +func TestCompress1X(t *testing.T) { + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + var s Scratch + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress1X(buf0, &s) + if err != test.err1X { + t.Errorf("want error %v (%T), got %v (%T)", test.err1X, test.err1X, err, err) + } + if err != nil { + t.Log(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + min := s.minSize(len(buf0)) + if len(s.OutData) < min { + t.Errorf("output data length (%d) below shannon limit (%d)", len(s.OutData), min) + } + if len(s.OutTable) == 0 { + t.Error("got no table definition") + } + if re { + t.Error("claimed to have re-used.") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + t.Logf("%s: %d -> %d bytes (%.2f:1) re:%t (table: %d bytes)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + s.Out = nil + bRe, _, err := Compress1X(b, &s) + if err == nil { + t.Log("Could re-compress to", len(bRe)) + } + }) + } +} + +func TestCompress4X(t *testing.T) { + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + var s Scratch + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress4X(buf0, &s) + if err != test.err4X { + t.Errorf("want error %v (%T), got %v (%T)", test.err1X, test.err4X, err, err) + } + if err != nil { + t.Log(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutTable) == 0 { + t.Error("got no table definition") + } + if re { + t.Error("claimed to have re-used.") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + }) + } +} + +func TestCompress4XReuse(t *testing.T) { + rng := rand.NewSource(0x1337) + var s Scratch + s.Reuse = ReusePolicyAllow + for i := 0; i < 255; i++ { + if testing.Short() && i > 10 { + break + } + t.Run(fmt.Sprint("test-", i), func(t *testing.T) { + buf0 := make([]byte, BlockSizeMax) + for j := range buf0 { + buf0[j] = byte(int64(i) + (rng.Int63() & 3)) + } + + b, re, err := Compress4X(buf0, &s) + if err != nil { + t.Fatal(err) + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + if re { + t.Error("claimed to have re-used. Unlikely.") + } + + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", t.Name(), len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + }) + } +} + +func TestCompress4XReuseActually(t *testing.T) { + rng := rand.NewSource(0x1337) + var s Scratch + s.Reuse = ReusePolicyAllow + for i := 0; i < 255; i++ { + if testing.Short() && i > 10 { + break + } + t.Run(fmt.Sprint("test-", i), func(t *testing.T) { + buf0 := make([]byte, BlockSizeMax) + for j := range buf0 { + buf0[j] = byte(rng.Int63() & 7) + } + + b, re, err := Compress4X(buf0, &s) + if err != nil { + t.Fatal(err) + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + if re && i == 0 { + t.Error("Claimed to have re-used on first loop.") + } + if !re && i > 0 { + t.Error("Expected table to be reused") + } + + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", t.Name(), len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + }) + } +} +func TestCompress1XReuse(t *testing.T) { + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + var s Scratch + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress1X(buf0, &s) + if err != test.err1X { + t.Errorf("want error %v (%T), got %v (%T)", test.err1X, test.err1X, err, err) + } + if err != nil { + t.Log(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + firstData := len(s.OutData) + s.Reuse = ReusePolicyAllow + b, re, err = Compress1X(buf0, &s) + if err != nil { + t.Errorf("got secondary error %v (%T)", err, err) + return + } + if !re { + t.Error("Didn't re-use even if data was the same") + } + if len(s.OutTable) != 0 { + t.Error("got table definition, don't want any") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + if len(b) != firstData { + t.Errorf("data length did not match first: %d, second:%d", firstData, len(b)) + } + t.Logf("%s: %d -> %d bytes (%.2f:1) %t", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re) + }) + } +} + +func BenchmarkDeflate(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err1X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + dec, err := flate.NewWriter(ioutil.Discard, flate.HuffmanOnly) + if err != nil { + b.Fatal(err) + } + if test.err1X != nil { + b.Skip("skipping") + } + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + dec.Reset(ioutil.Discard) + n, err := dec.Write(buf0) + if err != nil { + b.Fatal(err) + } + if n != len(buf0) { + b.Fatal("mismatch", n, len(buf0)) + } + dec.Close() + } + }) + } +} + +func BenchmarkCompress1XReuseNone(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err1X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + _, re, err := Compress1X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress1X(buf0, &s) + if re { + b.Fatal("reused") + } + } + }) + } +} + +func BenchmarkCompress1XReuseAllow(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err1X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyAllow + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + _, re, err := Compress1X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress1X(buf0, &s) + if !re { + b.Fatal("not reused") + } + } + }) + } +} + +func BenchmarkCompress1XReusePrefer(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err1X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyPrefer + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + _, re, err := Compress1X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress1X(buf0, &s) + if !re { + b.Fatal("not reused") + } + } + }) + } +} + +func BenchmarkCompress4XReuseNone(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err4X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + _, re, err := Compress4X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress4X(buf0, &s) + if re { + b.Fatal("reused") + } + } + }) + } +} + +func BenchmarkCompress4XReuseAllow(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err4X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyAllow + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + _, re, err := Compress4X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress4X(buf0, &s) + if !re { + b.Fatal("not reused") + } + } + }) + } +} + +func BenchmarkCompress4XReusePrefer(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err4X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyPrefer + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + _, re, err := Compress4X(buf0, &s) + if err != test.err4X { + b.Fatal("unexpected error:", err) + } + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress4X(buf0, &s) + if !re { + b.Fatal("not reused") + } + } + }) + } +} + +func BenchmarkCompress1XSizes(b *testing.B) { + test := testfiles[0] + sizes := []int{1e2, 2e2, 5e2, 1e3, 5e3, 1e4, 5e4} + for _, size := range sizes { + b.Run(test.name+"-"+fmt.Sprint(size), func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + buf0 = buf0[:size] + _, re, err := Compress1X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + //b.Log("Size:", len(o)) + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress1X(buf0, &s) + if re { + b.Fatal("reused") + } + } + }) + } +} + +func BenchmarkCompress4XSizes(b *testing.B) { + test := testfiles[0] + sizes := []int{1e2, 2e2, 5e2, 1e3, 5e3, 1e4, 5e4} + for _, size := range sizes { + b.Run(test.name+"-"+fmt.Sprint(size), func(b *testing.B) { + var s Scratch + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + buf0 = buf0[:size] + _, re, err := Compress4X(buf0, &s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + //b.Log("Size:", len(o)) + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, re, _ = Compress4X(buf0, &s) + if re { + b.Fatal("reused") + } + } + }) + } +} diff --git a/github.com/klauspost/compress/huff0/decompress.go b/github.com/klauspost/compress/huff0/decompress.go new file mode 100644 index 0000000..a03b263 --- /dev/null +++ b/github.com/klauspost/compress/huff0/decompress.go @@ -0,0 +1,1146 @@ +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 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, errors.New("input too small for table") + } + // 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) + } + for n, w := range s.huffWeight[:s.symbolLen] { + if w == 0 { + continue + } + length := (uint32(1) << w) >> 1 + d := dEntrySingle{ + entry: uint16(s.actualTableLog+1-w) | (uint16(n) << 8), + } + rank := &rankStats[w] + 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/github.com/klauspost/compress/huff0/decompress_test.go b/github.com/klauspost/compress/huff0/decompress_test.go new file mode 100644 index 0000000..b95d8be --- /dev/null +++ b/github.com/klauspost/compress/huff0/decompress_test.go @@ -0,0 +1,518 @@ +package huff0 + +import ( + "bytes" + "testing" +) + +func TestDecompress1X(t *testing.T) { + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + var s = &Scratch{} + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress1X(buf0, s) + if err != test.err1X { + t.Errorf("want error %v (%T), got %v (%T)", test.err1X, test.err1X, err, err) + } + if err != nil { + t.Log(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutTable) == 0 { + t.Error("got no table definition") + } + if re { + t.Error("claimed to have re-used.") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + + wantRemain := len(s.OutData) + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + + s.Out = nil + var remain []byte + s, remain, err = ReadTable(b, s) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + if s.matches(s.prevTable, &buf); buf.Len() > 0 { + t.Error(buf.String()) + } + if len(remain) != wantRemain { + t.Fatalf("remain mismatch, want %d, got %d bytes", wantRemain, len(remain)) + } + t.Logf("remain: %d bytes, ok", len(remain)) + dc, err := s.Decompress1X(remain) + if err != nil { + t.Error(err) + return + } + if len(buf0) != len(dc) { + t.Errorf(test.name+"decompressed, want size: %d, got %d", len(buf0), len(dc)) + if len(buf0) > len(dc) { + buf0 = buf0[:len(dc)] + } else { + dc = dc[:len(buf0)] + } + if !bytes.Equal(buf0, dc) { + if len(dc) > 1024 { + t.Log(string(dc[:1024])) + t.Errorf(test.name+"decompressed, got delta: \n(in)\t%02x !=\n(out)\t%02x\n", buf0[:1024], dc[:1024]) + } else { + t.Log(string(dc)) + t.Errorf(test.name+"decompressed, got delta: (in) %v != (out) %v\n", buf0, dc) + } + } + return + } + if !bytes.Equal(buf0, dc) { + if len(buf0) > 1024 { + t.Log(string(dc[:1024])) + } else { + t.Log(string(dc)) + } + //t.Errorf(test.name+": decompressed, got delta: \n%s") + t.Errorf(test.name + ": decompressed, got delta") + } + if !t.Failed() { + t.Log("... roundtrip ok!") + } + }) + } +} + +func TestDecompress4X(t *testing.T) { + for _, test := range testfiles { + t.Run(test.name, func(t *testing.T) { + var s = &Scratch{} + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress4X(buf0, s) + if err != test.err4X { + t.Errorf("want error %v (%T), got %v (%T)", test.err1X, test.err1X, err, err) + } + if err != nil { + t.Log(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutTable) == 0 { + t.Error("got no table definition") + } + if re { + t.Error("claimed to have re-used.") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + + wantRemain := len(s.OutData) + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + + s.Out = nil + var remain []byte + s, remain, err = ReadTable(b, s) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + if s.matches(s.prevTable, &buf); buf.Len() > 0 { + t.Error(buf.String()) + } + if len(remain) != wantRemain { + t.Fatalf("remain mismatch, want %d, got %d bytes", wantRemain, len(remain)) + } + t.Logf("remain: %d bytes, ok", len(remain)) + dc, err := s.Decompress4X(remain, len(buf0)) + if err != nil { + t.Error(err) + return + } + if len(buf0) != len(dc) { + t.Errorf(test.name+"decompressed, want size: %d, got %d", len(buf0), len(dc)) + if len(buf0) > len(dc) { + buf0 = buf0[:len(dc)] + } else { + dc = dc[:len(buf0)] + } + if !bytes.Equal(buf0, dc) { + if len(dc) > 1024 { + t.Log(string(dc[:1024])) + t.Errorf(test.name+"decompressed, got delta: \n(in)\t%02x !=\n(out)\t%02x\n", buf0[:1024], dc[:1024]) + } else { + t.Log(string(dc)) + t.Errorf(test.name+"decompressed, got delta: (in) %v != (out) %v\n", buf0, dc) + } + } + return + } + if !bytes.Equal(buf0, dc) { + if len(buf0) > 1024 { + t.Log(string(dc[:1024])) + } else { + t.Log(string(dc)) + } + //t.Errorf(test.name+": decompressed, got delta: \n%s") + t.Errorf(test.name + ": decompressed, got delta") + } + if !t.Failed() { + t.Log("... roundtrip ok!") + } + }) + } +} + +func TestRoundtrip1XFuzz(t *testing.T) { + for _, test := range testfilesExtended { + t.Run(test.name, func(t *testing.T) { + var s = &Scratch{} + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress1X(buf0, s) + if err != nil { + if err == ErrIncompressible || err == ErrUseRLE || err == ErrTooBig { + t.Log(test.name, err.Error()) + return + } + t.Error(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutTable) == 0 { + t.Error("got no table definition") + } + if re { + t.Error("claimed to have re-used.") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + + wantRemain := len(s.OutData) + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + + s.Out = nil + var remain []byte + s, remain, err = ReadTable(b, s) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + if s.matches(s.prevTable, &buf); buf.Len() > 0 { + t.Error(buf.String()) + } + if len(remain) != wantRemain { + t.Fatalf("remain mismatch, want %d, got %d bytes", wantRemain, len(remain)) + } + t.Logf("remain: %d bytes, ok", len(remain)) + dc, err := s.Decompress1X(remain) + if err != nil { + t.Error(err) + return + } + if len(buf0) != len(dc) { + t.Errorf(test.name+"decompressed, want size: %d, got %d", len(buf0), len(dc)) + if len(buf0) > len(dc) { + buf0 = buf0[:len(dc)] + } else { + dc = dc[:len(buf0)] + } + if !bytes.Equal(buf0, dc) { + if len(dc) > 1024 { + t.Log(string(dc[:1024])) + t.Errorf(test.name+"decompressed, got delta: \n(in)\t%02x !=\n(out)\t%02x\n", buf0[:1024], dc[:1024]) + } else { + t.Log(string(dc)) + t.Errorf(test.name+"decompressed, got delta: (in) %v != (out) %v\n", buf0, dc) + } + } + return + } + if !bytes.Equal(buf0, dc) { + if len(buf0) > 1024 { + t.Log(string(dc[:1024])) + } else { + t.Log(string(dc)) + } + //t.Errorf(test.name+": decompressed, got delta: \n%s") + t.Errorf(test.name + ": decompressed, got delta") + } + if !t.Failed() { + t.Log("... roundtrip ok!") + } + }) + } +} + +func TestRoundtrip4XFuzz(t *testing.T) { + for _, test := range testfilesExtended { + t.Run(test.name, func(t *testing.T) { + var s = &Scratch{} + buf0, err := test.fn() + if err != nil { + t.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + b, re, err := Compress4X(buf0, s) + if err != nil { + if err == ErrIncompressible || err == ErrUseRLE || err == ErrTooBig { + t.Log(test.name, err.Error()) + return + } + t.Error(test.name, err.Error()) + return + } + if b == nil { + t.Error("got no output") + return + } + if len(s.OutTable) == 0 { + t.Error("got no table definition") + } + if re { + t.Error("claimed to have re-used.") + } + if len(s.OutData) == 0 { + t.Error("got no data output") + } + + wantRemain := len(s.OutData) + t.Logf("%s: %d -> %d bytes (%.2f:1) %t (table: %d bytes)", test.name, len(buf0), len(b), float64(len(buf0))/float64(len(b)), re, len(s.OutTable)) + + s.Out = nil + var remain []byte + s, remain, err = ReadTable(b, s) + if err != nil { + t.Error(err) + return + } + var buf bytes.Buffer + if s.matches(s.prevTable, &buf); buf.Len() > 0 { + t.Error(buf.String()) + } + if len(remain) != wantRemain { + t.Fatalf("remain mismatch, want %d, got %d bytes", wantRemain, len(remain)) + } + t.Logf("remain: %d bytes, ok", len(remain)) + dc, err := s.Decompress4X(remain, len(buf0)) + if err != nil { + t.Error(err) + return + } + if len(buf0) != len(dc) { + t.Errorf(test.name+"decompressed, want size: %d, got %d", len(buf0), len(dc)) + if len(buf0) > len(dc) { + buf0 = buf0[:len(dc)] + } else { + dc = dc[:len(buf0)] + } + if !bytes.Equal(buf0, dc) { + if len(dc) > 1024 { + t.Log(string(dc[:1024])) + t.Errorf(test.name+"decompressed, got delta: \n(in)\t%02x !=\n(out)\t%02x\n", buf0[:1024], dc[:1024]) + } else { + t.Log(string(dc)) + t.Errorf(test.name+"decompressed, got delta: (in) %v != (out) %v\n", buf0, dc) + } + } + return + } + if !bytes.Equal(buf0, dc) { + if len(buf0) > 1024 { + t.Log(string(dc[:1024])) + } else { + t.Log(string(dc)) + } + //t.Errorf(test.name+": decompressed, got delta: \n%s") + t.Errorf(test.name + ": decompressed, got delta") + } + if !t.Failed() { + t.Log("... roundtrip ok!") + } + }) + } +} + +func BenchmarkDecompress1XTable(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err1X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s = &Scratch{} + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + compressed, _, err := Compress1X(buf0, s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + s.Out = nil + s, remain, _ := ReadTable(compressed, s) + s.Decompress1X(remain) + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + s, remain, err := ReadTable(compressed, s) + if err != nil { + b.Fatal(err) + } + _, err = s.Decompress1X(remain) + if err != nil { + b.Fatal(err) + } + } + }) + } +} + +func BenchmarkDecompress1XNoTable(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err1X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s = &Scratch{} + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + compressed, _, err := Compress1X(buf0, s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + s.Out = nil + s, remain, _ := ReadTable(compressed, s) + s.Decompress1X(remain) + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, err = s.Decompress1X(remain) + if err != nil { + b.Fatal(err) + } + } + }) + } +} + +func BenchmarkDecompress4XNoTable(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err4X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s = &Scratch{} + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + compressed, _, err := Compress4X(buf0, s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + s.Out = nil + s, remain, _ := ReadTable(compressed, s) + s.Decompress4X(remain, len(buf0)) + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + _, err = s.Decompress4X(remain, len(buf0)) + if err != nil { + b.Fatal(err) + } + } + }) + } +} + +func BenchmarkDecompress4XTable(b *testing.B) { + for _, tt := range testfiles { + test := tt + if test.err4X != nil { + continue + } + b.Run(test.name, func(b *testing.B) { + var s = &Scratch{} + s.Reuse = ReusePolicyNone + buf0, err := test.fn() + if err != nil { + b.Fatal(err) + } + if len(buf0) > BlockSizeMax { + buf0 = buf0[:BlockSizeMax] + } + compressed, _, err := Compress4X(buf0, s) + if err != test.err1X { + b.Fatal("unexpected error:", err) + } + s.Out = nil + b.ResetTimer() + b.ReportAllocs() + b.SetBytes(int64(len(buf0))) + for i := 0; i < b.N; i++ { + s, remain, err := ReadTable(compressed, s) + if err != nil { + b.Fatal(err) + } + _, err = s.Decompress4X(remain, len(buf0)) + if err != nil { + b.Fatal(err) + } + } + }) + } +} diff --git a/github.com/klauspost/compress/huff0/huff0.go b/github.com/klauspost/compress/huff0/huff0.go new file mode 100644 index 0000000..177d6c4 --- /dev/null +++ b/github.com/klauspost/compress/huff0/huff0.go @@ -0,0 +1,260 @@ +// 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 +) + +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 +} + +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/github.com/klauspost/compress/huff0/testdata/regression.zip b/github.com/klauspost/compress/huff0/testdata/regression.zip new file mode 100644 index 0000000..9d33c4e Binary files /dev/null and b/github.com/klauspost/compress/huff0/testdata/regression.zip differ diff --git a/github.com/klauspost/compress/s2/.gitignore b/github.com/klauspost/compress/s2/.gitignore new file mode 100644 index 0000000..3a89c6e --- /dev/null +++ b/github.com/klauspost/compress/s2/.gitignore @@ -0,0 +1,15 @@ +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/github.com/klauspost/compress/s2/LICENSE b/github.com/klauspost/compress/s2/LICENSE new file mode 100644 index 0000000..1d2d645 --- /dev/null +++ b/github.com/klauspost/compress/s2/LICENSE @@ -0,0 +1,28 @@ +Copyright (c) 2011 The Snappy-Go Authors. All rights reserved. +Copyright (c) 2019 Klaus Post. 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/github.com/klauspost/compress/s2/README.md b/github.com/klauspost/compress/s2/README.md new file mode 100644 index 0000000..63c5825 --- /dev/null +++ b/github.com/klauspost/compress/s2/README.md @@ -0,0 +1,493 @@ +# S2 Compression + +S2 is an extension of [Snappy](https://github.com/google/snappy). + +S2 is aimed for high throughput, which is why it features concurrent compression for bigger payloads. + +Decoding is compatible with Snappy compressed content, but content compressed with S2 cannot be decompressed by Snappy. +This means that S2 can seamlessly replace Snappy without converting compressed content. + +S2 is designed to have high throughput on content that cannot be compressed. +This is important so you don't have to worry about spending CPU cycles on already compressed data. + +## Benefits over Snappy + +* Better compression +* Concurrent stream compression +* Faster decompression +* Ability to quickly skip forward in compressed stream +* Compatible with reading Snappy compressed content +* Offers alternative, more efficient, but slightly slower compression mode. +* Smaller block size overhead on incompressible blocks. +* Block concatenation +* Automatic stream size padding. +* Snappy compatible block compression. + +## Drawbacks over Snappy + +* Not optimized for 32 bit systems. +* Uses slightly more memory (4MB per core) due to larger blocks and concurrency (configurable). + +# Usage + +Installation: `go get -u github.com/klauspost/compress/s2` + +Full package documentation: + +[![godoc][1]][2] + +[1]: https://godoc.org/github.com/klauspost/compress?status.svg +[2]: https://godoc.org/github.com/klauspost/compress/s2 + +## Compression + +```Go +func EncodeStream(src io.Reader, dst io.Writer) error { + enc := s2.NewWriter(dst) + _, err := io.Copy(enc, src) + if err != nil { + enc.Close() + return err + } + // Blocks until compression is done. + return enc.Close() +} +``` + +You should always call `enc.Close()`, otherwise you will leak resources and your encode will be incomplete. + +For the best throughput, you should attempt to reuse the `Writer` using the `Reset()` method. + +The Writer in S2 is always buffered, therefore `NewBufferedWriter` in Snappy can be replaced with `NewWriter` in S2. +It is possible to flush any buffered data using the `Flush()` method. +This will block until all data sent to the encoder has been written to the output. + +S2 also supports the `io.ReaderFrom` interface, which will consume all input from a reader. + +As a final method to compress data, if you have a single block of data you would like to have encoded as a stream, +a slightly more efficient method is to use the `EncodeBuffer` method. +This will take ownership of the buffer until the stream is closed. + +```Go +func EncodeStream(src []byte, dst io.Writer) error { + enc := s2.NewWriter(dst) + // The encoder owns the buffer until Flush or Close is called. + err := enc.EncodeBuffer(buf) + if err != nil { + enc.Close() + return err + } + // Blocks until compression is done. + return enc.Close() +} +``` + +Each call to `EncodeBuffer` will result in discrete blocks being created without buffering, +so it should only be used a single time per stream. +If you need to write several blocks, you should use the regular io.Writer interface. + + +## Decompression + +```Go +func DecodeStream(src io.Reader, dst io.Writer) error + dec := s2.NewReader(src) + _, err := io.Copy(dst, dec) + return err +} +``` + +Similar to the Writer, a Reader can be reused using the `Reset` method. + +For the best possible throughput, there is a `EncodeBuffer(buf []byte)` function available. +However, it requires that the provided buffer isn't used after it is handed over to S2 and until the stream is flushed or closed. + +For smaller data blocks, there is also a non-streaming interface: `Encode()`, `EncodeBetter()` and `Decode()`. +Do however note that these functions (similar to Snappy) does not provide validation of data, +so data corruption may be undetected. Stream encoding provides CRC checks of data. + +It is possible to efficiently skip forward in a compressed stream using the `Skip()` method. +For big skips the decompressor is able to skip blocks without decompressing them. + +## Single Blocks + +Similar to Snappy S2 offers single block compression. +Blocks do not offer the same flexibility and safety as streams, but may be preferable for very small payloads, less than 100K. + +Using a simple `dst := s2.Encode(nil, src)` will compress `src` and return the compressed result. It is possible to provide a destination buffer. If the buffer has a capacity of `s2.MaxEncodedLen(len(src))` it will be used. If not a new will be allocated. Alternatively `EncodeBetter` can also be used for better, but slightly slower compression. + +Similarly to decompress a block you can use `dst, err := s2.Decode(nil, src)`. Again an optional destination buffer can be supplied. +The `s2.DecodedLen(src)` can be used to get the minimum capacity needed. If that is not satisfied a new buffer will be allocated. + +Block function always operate on a single goroutine since it should only be used for small payloads. + +# Commandline tools + +Some very simply commandline tools are provided; `s2c` for compression and `s2d` for decompression. + +Binaries can be downloaded on the [Releases Page](https://github.com/klauspost/compress/releases). + +Installing then requires Go to be installed. To install them, use: + +`go install github.com/klauspost/compress/s2/cmd/s2c && go install github.com/klauspost/compress/s2/cmd/s2d` + +To build binaries to the current folder use: + +`go build github.com/klauspost/compress/s2/cmd/s2c && go build github.com/klauspost/compress/s2/cmd/s2d` + + +## s2c + +``` +Usage: s2c [options] file1 file2 + +Compresses all files supplied as input separately. +Output files are written as 'filename.ext.s2'. +By default output files will be overwritten. +Use - as the only file name to read from stdin and write to stdout. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +Options: + -bench int + Run benchmark n times. No output will be written + -blocksize string + Max block size. Examples: 64K, 256K, 1M, 4M. Must be power of two and <= 4MB (default "4M") + -c Write all output to stdout. Multiple input files will be concatenated + -cpu int + Compress using this amount of threads (default CPU_THREADS]) + -faster + Compress faster, but with a minor compression loss + -help + Display help + -pad string + Pad size to a multiple of this value, Examples: 500, 64K, 256K, 1M, 4M, etc (default "1") + -q Don't write any output to terminal, except errors + -rm + Delete source file(s) after successful compression + -safe + Do not overwrite output files +``` + +## s2d + +``` +Usage: s2d [options] file1 file2 + +Decompresses all files supplied as input. Input files must end with '.s2' or '.snappy'. +Output file names have the extension removed. By default output files will be overwritten. +Use - as the only file name to read from stdin and write to stdout. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +Options: + -bench int + Run benchmark n times. No output will be written + -c Write all output to stdout. Multiple input files will be concatenated + -help + Display help + -q Don't write any output to terminal, except errors + -rm + Delete source file(s) after successful decompression + -safe + Do not overwrite output files + +``` + +# Performance + +This section will focus on comparisons to Snappy. +This package is solely aimed at replacing Snappy as a high speed compression package. +If you are mainly looking for better compression [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) +gives better compression, but typically at speeds slightly below "better" mode in this package. + +Compression is increased compared to Snappy, mostly around 5-20% and the throughput is typically 25-40% increased (single threaded) compared to the Snappy Go implementation. + +Streams are concurrently compressed. The stream will be distributed among all available CPU cores for the best possible throughput. + +A "better" compression mode is also available. This allows to trade a bit of speed for a minor compression gain. +The content compressed in this mode is fully compatible with the standard decoder. + +Snappy vs S2 **compression** speed on 16 core (32 thread) computer, using all threads and a single thread (1 CPU): + +| File | S2 speed | S2 Throughput | S2 % smaller | S2 "better" | "better" throughput | "better" % smaller | +|-----------------------------------------------------------------------------------------------------|----------|---------------|--------------|-------------|---------------------|--------------------| +| [rawstudio-mint14.tar](https://files.klauspost.com/compress/rawstudio-mint14.7z) | 12.70x | 10556 MB/s | 7.35% | 4.15x | 3455 MB/s | 12.79% | +| (1 CPU) | 1.14x | 948 MB/s | - | 0.42x | 349 MB/s | - | +| [github-june-2days-2019.json](https://files.klauspost.com/compress/github-june-2days-2019.json.zst) | 17.13x | 14484 MB/s | 31.60% | 10.09x | 8533 MB/s | 37.71% | +| (1 CPU) | 1.33x | 1127 MB/s | - | 0.70x | 589 MB/s | - | +| [github-ranks-backup.bin](https://files.klauspost.com/compress/github-ranks-backup.bin.zst) | 15.14x | 12000 MB/s | -5.79% | 6.59x | 5223 MB/s | 5.80% | +| (1 CPU) | 1.11x | 877 MB/s | - | 0.47x | 370 MB/s | - | +| [consensus.db.10gb](https://files.klauspost.com/compress/consensus.db.10gb.zst) | 14.62x | 12116 MB/s | 15.90% | 5.35x | 4430 MB/s | 16.08% | +| (1 CPU) | 1.38x | 1146 MB/s | - | 0.38x | 312 MB/s | - | +| [adresser.json](https://files.klauspost.com/compress/adresser.json.zst) | 8.83x | 17579 MB/s | 43.86% | 6.54x | 13011 MB/s | 47.23% | +| (1 CPU) | 1.14x | 2259 MB/s | - | 0.74x | 1475 MB/s | - | +| [gob-stream](https://files.klauspost.com/compress/gob-stream.7z) | 16.72x | 14019 MB/s | 24.02% | 10.11x | 8477 MB/s | 30.48% | +| (1 CPU) | 1.24x | 1043 MB/s | - | 0.70x | 586 MB/s | - | +| [10gb.tar](http://mattmahoney.net/dc/10gb.html) | 13.33x | 9254 MB/s | 1.84% | 6.75x | 4686 MB/s | 6.72% | +| (1 CPU) | 0.97x | 672 MB/s | - | 0.53x | 366 MB/s | - | +| sharnd.out.2gb | 2.11x | 12639 MB/s | 0.01% | 1.98x | 11833 MB/s | 0.01% | +| (1 CPU) | 0.93x | 5594 MB/s | - | 1.34x | 8030 MB/s | - | +| [enwik9](http://mattmahoney.net/dc/textdata.html) | 19.34x | 8220 MB/s | 3.98% | 7.87x | 3345 MB/s | 15.82% | +| (1 CPU) | 1.06x | 452 MB/s | - | 0.50x | 213 MB/s | - | +| [silesia.tar](http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip) | 10.48x | 6124 MB/s | 5.67% | 3.76x | 2197 MB/s | 12.60% | +| (1 CPU) | 0.97x | 568 MB/s | - | 0.46x | 271 MB/s | - | +| [enwik10](https://encode.su/threads/3315-enwik10-benchmark-results) | 21.07x | 9020 MB/s | 6.36% | 6.91x | 2959 MB/s | 16.95% | +| (1 CPU) | 1.07x | 460 MB/s | - | 0.51x | 220 MB/s | - | + +### Legend + +* `S2 speed`: Speed of S2 compared to Snappy, using 16 cores and 1 core. +* `S2 throughput`: Throughput of S2 in MB/s. +* `S2 % smaller`: How many percent of the Snappy output size is S2 better. +* `S2 "better"`: Speed when enabling "better" compression mode in S2 compared to Snappy. +* `"better" throughput`: Speed when enabling "better" compression mode in S2 compared to Snappy. +* `"better" % smaller`: How many percent of the Snappy output size is S2 better when using "better" compression. + +There is a good speedup across the board when using a single thread and a significant speedup when using multiple threads. + +Machine generated data gets by far the biggest compression boost, with size being being reduced by up to 45% of Snappy size. + +The "better" compression mode sees a good improvement in all cases, but usually at a performance cost. + +Incompressible content (`sharnd.out.2gb`, 2GB random data) sees the smallest speedup. +This is likely dominated by synchronization overhead, which is confirmed by the fact that single threaded performance is higher (see above). + +## Decompression + +S2 attempts to create content that is also fast to decompress, except in "better" mode where the smallest representation is used. + +S2 vs Snappy **decompression** speed. Both operating on single core: + +| File | S2 Throughput | vs. Snappy | Better Throughput | vs. Snappy | +|-----------------------------------------------------------------------------------------------------|---------------|------------|-------------------|------------| +| [rawstudio-mint14.tar](https://files.klauspost.com/compress/rawstudio-mint14.7z) | 2117 MB/s | 1.14x | 1738 MB/s | 0.94x | +| [github-june-2days-2019.json](https://files.klauspost.com/compress/github-june-2days-2019.json.zst) | 2401 MB/s | 1.25x | 2307 MB/s | 1.20x | +| [github-ranks-backup.bin](https://files.klauspost.com/compress/github-ranks-backup.bin.zst) | 2075 MB/s | 0.98x | 1764 MB/s | 0.83x | +| [consensus.db.10gb](https://files.klauspost.com/compress/consensus.db.10gb.zst) | 2967 MB/s | 1.05x | 2885 MB/s | 1.02x | +| [adresser.json](https://files.klauspost.com/compress/adresser.json.zst) | 4141 MB/s | 1.07x | 4184 MB/s | 1.08x | +| [gob-stream](https://files.klauspost.com/compress/gob-stream.7z) | 2264 MB/s | 1.12x | 2185 MB/s | 1.08x | +| [10gb.tar](http://mattmahoney.net/dc/10gb.html) | 1525 MB/s | 1.03x | 1347 MB/s | 0.91x | +| sharnd.out.2gb | 3813 MB/s | 0.79x | 3900 MB/s | 0.81x | +| [enwik9](http://mattmahoney.net/dc/textdata.html) | 1246 MB/s | 1.29x | 967 MB/s | 1.00x | +| [silesia.tar](http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip) | 1433 MB/s | 1.12x | 1203 MB/s | 0.94x | +| [enwik10](https://encode.su/threads/3315-enwik10-benchmark-results) | 1284 MB/s | 1.32x | 1010 MB/s | 1.04x | + +### Legend + +* `S2 Throughput`: Decompression speed of S2 encoded content. +* `Better Throughput`: Decompression speed of S2 "better" encoded content. +* `vs Snappy`: Decompression speed of S2 "better" mode compared to Snappy and absolute speed. + + +While the decompression code hasn't changed, there is a significant speedup in decompression speed. +S2 prefers longer matches and will typically only find matches that are 6 bytes or longer. +While this reduces compression a bit, it improves decompression speed. + +The "better" compression mode will actively look for shorter matches, which is why it has a decompression speed quite similar to Snappy. + +Without assembly decompression is also very fast; single goroutine decompression speed. No assembly: + +| File | S2 Throughput | S2 throughput | +|--------------------------------|--------------|---------------| +| consensus.db.10gb.s2 | 1.84x | 2289.8 MB/s | +| 10gb.tar.s2 | 1.30x | 867.07 MB/s | +| rawstudio-mint14.tar.s2 | 1.66x | 1329.65 MB/s | +| github-june-2days-2019.json.s2 | 2.36x | 1831.59 MB/s | +| github-ranks-backup.bin.s2 | 1.73x | 1390.7 MB/s | +| enwik9.s2 | 1.67x | 681.53 MB/s | +| adresser.json.s2 | 3.41x | 4230.53 MB/s | +| silesia.tar.s2 | 1.52x | 811.58 | + +Even though S2 typically compresses better than Snappy, decompression speed is always better. + +## Block compression + + +When compressing blocks no concurrent compression is performed just as Snappy. +This is because blocks are for smaller payloads and generally will not benefit from concurrent compression. + +An important change is that incompressible blocks will not be more than at most 10 bytes bigger than the input. +In rare, worst case scenario Snappy blocks could be significantly bigger than the input. + +### Mixed content blocks + +The most reliable is a wide dataset. +For this we use `webdevdata.org-2015-01-07-subset`, 53927 files, total input size: 4,014,526,923 bytes. +Single goroutine used. + +| * | Input | Output | Reduction | MB/s | +|-------------------|------------|------------|-----------|--------| +| S2 | 4014526923 | 1062282489 | 73.54% | **861.44** | +| S2 Better | 4014526923 | 981221284 | **75.56%** | 399.54 | +| Snappy | 4014526923 | 1128667736 | 71.89% | 741.29 | +| S2, Snappy Output | 4014526923 | 1093784815 | 72.75% | 843.66 | + +S2 delivers both the best single threaded throuhput with regular mode and the best compression rate with "better" mode. + +When outputting Snappy compatible output it still delivers better throughput (100MB/s more) and better compression. + +As can be seen from the other benchmarks decompression should also be easier on the S2 generated output. + +### Standard block compression + +Benchmarking single block performance is subject to a lot more variation since it only tests a limited number of file patterns. +So individual benchmarks should only be seen as a guideline and the overall picture is more important. + +These micro-benchmarks are with data in cache and trained branch predictors. For a more realistic benchmark see the mixed content above. + +Block compression. Parallel benchmark running on 16 cores, 16 goroutines. + +AMD64 assembly is use for both S2 and Snappy. + +| Absolute Perf | Snappy size | S2 Size | Snappy Speed | S2 Speed | Snappy dec | S2 dec | +|-----------------------|-------------|---------|--------------|-------------|-------------|-------------| +| html | 22843 | 21111 | 16246 MB/s | 17438 MB/s | 40972 MB/s | 49263 MB/s | +| urls.10K | 335492 | 287326 | 7943 MB/s | 9693 MB/s | 22523 MB/s | 26484 MB/s | +| fireworks.jpeg | 123034 | 123100 | 349544 MB/s | 273889 MB/s | 718321 MB/s | 827552 MB/s | +| fireworks.jpeg (200B) | 146 | 155 | 8869 MB/s | 17773 MB/s | 33691 MB/s | 52421 MB/s | +| paper-100k.pdf | 85304 | 84459 | 167546 MB/s | 101263 MB/s | 326905 MB/s | 291944 MB/s | +| html_x_4 | 92234 | 21113 | 15194 MB/s | 50670 MB/s | 30843 MB/s | 32217 MB/s | +| alice29.txt | 88034 | 85975 | 5936 MB/s | 6139 MB/s | 12882 MB/s | 20044 MB/s | +| asyoulik.txt | 77503 | 79650 | 5517 MB/s | 6366 MB/s | 12735 MB/s | 22806 MB/s | +| lcet10.txt | 234661 | 220670 | 6235 MB/s | 6067 MB/s | 14519 MB/s | 18697 MB/s | +| plrabn12.txt | 319267 | 317985 | 5159 MB/s | 5726 MB/s | 11923 MB/s | 19901 MB/s | +| geo.protodata | 23335 | 18690 | 21220 MB/s | 26529 MB/s | 56271 MB/s | 62540 MB/s | +| kppkn.gtb | 69526 | 65312 | 9732 MB/s | 8559 MB/s | 18491 MB/s | 18969 MB/s | +| alice29.txt (128B) | 80 | 82 | 6691 MB/s | 15489 MB/s | 31883 MB/s | 38874 MB/s | +| alice29.txt (1000B) | 774 | 774 | 12204 MB/s | 13000 MB/s | 48056 MB/s | 52341 MB/s | +| alice29.txt (10000B) | 6648 | 6933 | 10044 MB/s | 12806 MB/s | 32378 MB/s | 46322 MB/s | +| alice29.txt (20000B) | 12686 | 13574 | 7733 MB/s | 11210 MB/s | 30566 MB/s | 58969 MB/s | + + +| Relative Perf | Snappy size | S2 size improved | S2 Speed | S2 Dec Speed | +|-----------------------|-------------|------------------|----------|--------------| +| html | 22.31% | 7.58% | 1.07x | 1.20x | +| urls.10K | 47.78% | 14.36% | 1.22x | 1.18x | +| fireworks.jpeg | 99.95% | -0.05% | 0.78x | 1.15x | +| fireworks.jpeg (200B) | 73.00% | -6.16% | 2.00x | 1.56x | +| paper-100k.pdf | 83.30% | 0.99% | 0.60x | 0.89x | +| html_x_4 | 22.52% | 77.11% | 3.33x | 1.04x | +| alice29.txt | 57.88% | 2.34% | 1.03x | 1.56x | +| asyoulik.txt | 61.91% | -2.77% | 1.15x | 1.79x | +| lcet10.txt | 54.99% | 5.96% | 0.97x | 1.29x | +| plrabn12.txt | 66.26% | 0.40% | 1.11x | 1.67x | +| geo.protodata | 19.68% | 19.91% | 1.25x | 1.11x | +| kppkn.gtb | 37.72% | 6.06% | 0.88x | 1.03x | +| alice29.txt (128B) | 62.50% | -2.50% | 2.31x | 1.22x | +| alice29.txt (1000B) | 77.40% | 0.00% | 1.07x | 1.09x | +| alice29.txt (10000B) | 66.48% | -4.29% | 1.27x | 1.43x | +| alice29.txt (20000B) | 63.43% | -7.00% | 1.45x | 1.93x | + +Speed is generally at or above Snappy. Small blocks gets a significant speedup, although at the expense of size. + +Decompression speed is better than Snappy, except in one case. + +Since payloads are very small the variance in terms of size is rather big, so they should only be seen as a general guideline. + +Size is on average around Snappy, but varies on content type. +In cases where compression is worse, it usually is compensated by a speed boost. + + +### Better compression + +Benchmarking single block performance is subject to a lot more variation since it only tests a limited number of file patterns. +So individual benchmarks should only be seen as a guideline and the overall picture is more important. + +| Absolute Perf | Snappy size | Better Size | Snappy Speed | Better Speed | Snappy dec | Better dec | +|-----------------------|-------------|-------------|--------------|--------------|-------------|-------------| +| html | 22843 | 19833 | 16246 MB/s | 7731 MB/s | 40972 MB/s | 40292 MB/s | +| urls.10K | 335492 | 253529 | 7943 MB/s | 3980 MB/s | 22523 MB/s | 20981 MB/s | +| fireworks.jpeg | 123034 | 123100 | 349544 MB/s | 9760 MB/s | 718321 MB/s | 823698 MB/s | +| fireworks.jpeg (200B) | 146 | 142 | 8869 MB/s | 594 MB/s | 33691 MB/s | 30101 MB/s | +| paper-100k.pdf | 85304 | 82915 | 167546 MB/s | 7470 MB/s | 326905 MB/s | 198869 MB/s | +| html_x_4 | 92234 | 19841 | 15194 MB/s | 23403 MB/s | 30843 MB/s | 30937 MB/s | +| alice29.txt | 88034 | 73218 | 5936 MB/s | 2945 MB/s | 12882 MB/s | 16611 MB/s | +| asyoulik.txt | 77503 | 66844 | 5517 MB/s | 2739 MB/s | 12735 MB/s | 14975 MB/s | +| lcet10.txt | 234661 | 190589 | 6235 MB/s | 3099 MB/s | 14519 MB/s | 16634 MB/s | +| plrabn12.txt | 319267 | 270828 | 5159 MB/s | 2600 MB/s | 11923 MB/s | 13382 MB/s | +| geo.protodata | 23335 | 18278 | 21220 MB/s | 11208 MB/s | 56271 MB/s | 57961 MB/s | +| kppkn.gtb | 69526 | 61851 | 9732 MB/s | 4556 MB/s | 18491 MB/s | 16524 MB/s | +| alice29.txt (128B) | 80 | 81 | 6691 MB/s | 529 MB/s | 31883 MB/s | 34225 MB/s | +| alice29.txt (1000B) | 774 | 748 | 12204 MB/s | 1943 MB/s | 48056 MB/s | 42068 MB/s | +| alice29.txt (10000B) | 6648 | 6234 | 10044 MB/s | 2949 MB/s | 32378 MB/s | 28813 MB/s | +| alice29.txt (20000B) | 12686 | 11584 | 7733 MB/s | 2822 MB/s | 30566 MB/s | 27315 MB/s | + + +| Relative Perf | Snappy size | Better size | Better Speed | Better dec | +|-----------------------|-------------|-------------|--------------|------------| +| html | 22.31% | 13.18% | 0.48x | 0.98x | +| urls.10K | 47.78% | 24.43% | 0.50x | 0.93x | +| fireworks.jpeg | 99.95% | -0.05% | 0.03x | 1.15x | +| fireworks.jpeg (200B) | 73.00% | 2.74% | 0.07x | 0.89x | +| paper-100k.pdf | 83.30% | 2.80% | 0.07x | 0.61x | +| html_x_4 | 22.52% | 78.49% | 0.04x | 1.00x | +| alice29.txt | 57.88% | 16.83% | 1.54x | 1.29x | +| asyoulik.txt | 61.91% | 13.75% | 0.50x | 1.18x | +| lcet10.txt | 54.99% | 18.78% | 0.50x | 1.15x | +| plrabn12.txt | 66.26% | 15.17% | 0.50x | 1.12x | +| geo.protodata | 19.68% | 21.67% | 0.50x | 1.03x | +| kppkn.gtb | 37.72% | 11.04% | 0.53x | 0.89x | +| alice29.txt (128B) | 62.50% | -1.25% | 0.47x | 1.07x | +| alice29.txt (1000B) | 77.40% | 3.36% | 0.08x | 0.88x | +| alice29.txt (10000B) | 66.48% | 6.23% | 0.16x | 0.89x | +| alice29.txt (20000B) | 63.43% | 8.69% | 0.29x | 0.89x | + +Except for the mostly incompressible JPEG image compression is better and usually in the +double digits in terms of percentage reduction over Snappy. + +The PDF sample shows a significant slowdown compared to Snappy, as this mode tries harder +to compress the data. Very small blocks are also not favorable for better compression, so throughput is way down. + +This mode aims to provide better compression at the expense of performance and achieves that +without a huge performance pentalty, except on very small blocks. + +Decompression speed suffers a little compared to the regular S2 mode, +but still manages to be close to Snappy in spite of increased compression. + +# Concatenating blocks and streams. + +Concatenating streams will concatenate the output of both without recompressing them. +While this is inefficient in terms of compression it might be usable in certain scenarios. +The 10 byte 'stream identifier' of the second stream can optionally be stripped, but it is not a requirement. + +Blocks can be concatenated using the `ConcatBlocks` function. + +Snappy blocks/streams can safely be concatenated with S2 blocks and streams. + +# Format Extensions + +* Frame [Stream identifier](https://github.com/google/snappy/blob/master/framing_format.txt#L68) changed from `sNaPpY` to `S2sTwO`. +* [Framed compressed blocks](https://github.com/google/snappy/blob/master/format_description.txt) can be up to 4MB (up from 64KB). +* Compressed blocks can have an offset of `0`, which indicates to repeat the last seen offset. + +Repeat offsets must be encoded as a [2.2.1. Copy with 1-byte offset (01)](https://github.com/google/snappy/blob/master/format_description.txt#L89), where the offset is 0. + +The length is specified by reading the 3-bit length specified in the tag and decode using this table: + +| Length | Actual Length | +|--------|----------------------| +| 0 | 4 | +| 1 | 5 | +| 2 | 6 | +| 3 | 7 | +| 4 | 8 | +| 5 | 8 + read 1 byte | +| 6 | 260 + read 2 bytes | +| 7 | 65540 + read 3 bytes | + +This allows any repeat offset + length to be represented by 2 to 5 bytes. + +Lengths are stored as little endian values. + +The first copy of a block cannot be a repeat offset and the offset is not carried across blocks in streams. + +Default streaming block size is 1MB. + +# LICENSE + +This code is based on the [Snappy-Go](https://github.com/golang/snappy) implementation. + +Use of this source code is governed by a BSD-style license that can be found in the LICENSE file. diff --git a/github.com/klauspost/compress/s2/_generate/gen.go b/github.com/klauspost/compress/s2/_generate/gen.go new file mode 100644 index 0000000..e242150 --- /dev/null +++ b/github.com/klauspost/compress/s2/_generate/gen.go @@ -0,0 +1,1817 @@ +package main + +//go:generate go run gen.go -out ../encodeblock_amd64.s -stubs ../encodeblock_amd64.go -pkg=s2 + +import ( + "fmt" + "math" + "runtime" + + . "github.com/mmcloughlin/avo/build" + "github.com/mmcloughlin/avo/buildtags" + . "github.com/mmcloughlin/avo/operand" + "github.com/mmcloughlin/avo/reg" +) + +// insert extra checks here and there. +const debug = false + +func main() { + Constraint(buildtags.Not("appengine").ToConstraint()) + Constraint(buildtags.Not("noasm").ToConstraint()) + Constraint(buildtags.Term("gc").ToConstraint()) + + o := options{ + snappy: false, + } + o.genEncodeBlockAsm("encodeBlockAsm", 14, 6, 6) + o.genEncodeBlockAsm("encodeBlockAsm12B", 12, 5, 5) + o.genEncodeBlockAsm("encodeBlockAsm10B", 10, 5, 4) + o.genEncodeBlockAsm("encodeBlockAsm8B", 8, 4, 4) + + // Snappy compatible + o.snappy = true + o.genEncodeBlockAsm("encodeSnappyBlockAsm", 14, 6, 6) + o.genEncodeBlockAsm("encodeSnappyBlockAsm12B", 12, 5, 5) + o.genEncodeBlockAsm("encodeSnappyBlockAsm10B", 10, 5, 4) + o.genEncodeBlockAsm("encodeSnappyBlockAsm8B", 8, 4, 4) + + o.snappy = false + o.genEmitLiteral() + o.genEmitRepeat() + o.genEmitCopy() + o.snappy = true + o.genEmitCopyNoRepeat() + o.snappy = false + o.genMatchLen() + Generate() +} + +func debugval(v Op) { + value := reg.R15 + MOVQ(v, value) + INT(Imm(3)) +} + +func debugval32(v Op) { + value := reg.R15L + MOVL(v, value) + INT(Imm(3)) +} + +var assertCounter int + +// assert will insert code if debug is enabled. +// The code should jump to 'ok' is assertion is success. +func assert(fn func(ok LabelRef)) { + if debug { + caller := [100]uintptr{0} + runtime.Callers(2, caller[:]) + frame, _ := runtime.CallersFrames(caller[:]).Next() + + ok := fmt.Sprintf("assert_check_%d_ok_srcline_%d", assertCounter, frame.Line) + fn(LabelRef(ok)) + // Emit several since delve is imprecise. + INT(Imm(3)) + INT(Imm(3)) + Label(ok) + assertCounter++ + } +} + +type options struct { + snappy bool + vmbi2 bool +} + +func (o options) genEncodeBlockAsm(name string, tableBits, skipLog, hashBytes int) { + TEXT(name, 0, "func(dst, src []byte) int") + Doc(name+" 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.", "") + Pragma("noescape") + + const literalMaxOverhead = 4 + + var tableSize = 4 * (1 << tableBits) + // Memzero needs at least 128 bytes. + if tableSize < 128 { + panic("tableSize must be at least 128 bytes") + } + + lenSrcBasic, err := Param("src").Len().Resolve() + if err != nil { + panic(err) + } + lenSrcQ := lenSrcBasic.Addr + + lenDstBasic, err := Param("dst").Len().Resolve() + if err != nil { + panic(err) + } + lenDstQ := lenDstBasic.Addr + + // Bail if we can't compress to at least this. + dstLimitPtrQ := AllocLocal(8) + + // sLimitL is when to stop looking for offset/length copies. + sLimitL := AllocLocal(4) + + // nextEmitL keeps track of the point we have emitted to. + nextEmitL := AllocLocal(4) + + // Repeat stores the last match offset. + repeatL := AllocLocal(4) + + // nextSTempL keeps nextS while other functions are being called. + nextSTempL := AllocLocal(4) + + // Alloc table last + table := AllocLocal(tableSize) + + dst := GP64() + { + dstBaseBasic, err := Param("dst").Base().Resolve() + if err != nil { + panic(err) + } + dstBaseQ := dstBaseBasic.Addr + MOVQ(dstBaseQ, dst) + } + + srcBaseBasic, err := Param("src").Base().Resolve() + if err != nil { + panic(err) + } + srcBaseQ := srcBaseBasic.Addr + + // Zero table + { + iReg := GP64() + MOVQ(U32(tableSize/8/16), iReg) + tablePtr := GP64() + LEAQ(table, tablePtr) + zeroXmm := XMM() + PXOR(zeroXmm, zeroXmm) + + Label("zero_loop_" + name) + for i := 0; i < 8; i++ { + MOVOU(zeroXmm, Mem{Base: tablePtr, Disp: i * 16}) + } + ADDQ(U8(16*8), tablePtr) + DECQ(iReg) + JNZ(LabelRef("zero_loop_" + name)) + } + + { + // nextEmit is offset n src where the next emitLiteral should start from. + MOVL(U32(0), nextEmitL) + + const inputMargin = 8 + tmp, tmp2, tmp3 := GP64(), GP64(), GP64() + MOVQ(lenSrcQ, tmp) + LEAQ(Mem{Base: tmp, Disp: -5}, tmp2) + // sLimitL := len(src) - inputMargin + LEAQ(Mem{Base: tmp, Disp: -inputMargin}, tmp3) + + assert(func(ok LabelRef) { + CMPQ(tmp3, lenSrcQ) + JL(ok) + }) + + MOVL(tmp3.As32(), sLimitL) + + // dstLimit := (len(src) - 5 ) - len(src)>>5 + SHRQ(U8(5), tmp) + SUBL(tmp.As32(), tmp2.As32()) // tmp2 = tmp2 - tmp + + assert(func(ok LabelRef) { + // if len(src) > len(src) - len(src)>>5 - 5: ok + CMPQ(lenSrcQ, tmp2) + JGE(ok) + }) + + LEAQ(Mem{Base: dst, Index: tmp2, Scale: 1}, tmp2) + MOVQ(tmp2, dstLimitPtrQ) + } + + // s = 1 + s := GP32() + MOVL(U32(1), s) + // repeatL = 1 + MOVL(s, repeatL) + + src := GP64() + Load(Param("src").Base(), src) + + // Load cv + Label("search_loop_" + name) + candidate := GP32() + { + assert(func(ok LabelRef) { + // Check if somebody changed src + tmp := GP64() + MOVQ(srcBaseQ, tmp) + CMPQ(tmp, src) + JEQ(ok) + }) + + cv := GP64() + MOVQ(Mem{Base: src, Index: s, Scale: 1}, cv) + nextS := GP32() + // nextS := s + (s-nextEmit)>>6 + 4 + { + tmp := GP64() + MOVL(s, tmp.As32()) // tmp = s + SUBL(nextEmitL, tmp.As32()) // tmp = s - nextEmit + SHRL(U8(skipLog), tmp.As32()) // tmp = (s - nextEmit) >> skipLog + LEAL(Mem{Base: s, Disp: 4, Index: tmp, Scale: 1}, nextS) + } + // if nextS > sLimit {goto emitRemainder} + { + CMPL(nextS.As32(), sLimitL) + JGE(LabelRef("emit_remainder_" + name)) + } + assert(func(ok LabelRef) { + // Check if s is valid (we should have jumped above if not) + tmp := GP64() + MOVQ(lenSrcQ, tmp) + CMPQ(tmp, s.As64()) + JG(ok) + }) + // move nextS to stack. + MOVL(nextS.As32(), nextSTempL) + + candidate2 := GP32() + hasher := hashN(hashBytes, tableBits) + { + hash0, hash1 := GP64(), GP64() + MOVQ(cv, hash0) + MOVQ(cv, hash1) + SHRQ(U8(8), hash1) + hasher.hash(hash0) + hasher.hash(hash1) + MOVL(table.Idx(hash0, 4), candidate) + MOVL(table.Idx(hash1, 4), candidate2) + assert(func(ok LabelRef) { + CMPQ(hash0, U32(tableSize)) + JL(ok) + }) + assert(func(ok LabelRef) { + CMPQ(hash1, U32(tableSize)) + JL(ok) + }) + + MOVL(s, table.Idx(hash0, 4)) + tmp := GP32() + LEAL(Mem{Base: s, Disp: 1}, tmp) + MOVL(tmp, table.Idx(hash1, 4)) + } + + // Can be moved up if registers are available. + hash2 := GP64() + { + // hash2 := hash6(cv>>16, tableBits) + // hasher = hash6(tableBits) + MOVQ(cv, hash2) + SHRQ(U8(16), hash2) + hasher.hash(hash2) + assert(func(ok LabelRef) { + CMPQ(hash2, U32(tableSize)) + JL(ok) + }) + } + + // En/disable repeat matching. + if true { + // Check repeat at offset checkRep + const checkRep = 1 + { + // rep = s - repeat + rep := GP32() + MOVL(s, rep) + SUBL(repeatL, rep) // rep = s - repeat + + // if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) { + left, right := GP64(), GP64() + MOVL(Mem{Base: src, Index: rep, Disp: checkRep, Scale: 1}, right.As32()) + MOVQ(cv, left) + SHRQ(U8(checkRep*8), left) + CMPL(left.As32(), right.As32()) + // BAIL, no repeat. + JNE(LabelRef("no_repeat_found_" + name)) + } + // base = s + checkRep + base := GP32() + LEAL(Mem{Base: s, Disp: checkRep}, base) + + // nextEmit before repeat. + nextEmit := GP32() + MOVL(nextEmitL, nextEmit) + + // Extend back + if true { + i := GP32() + MOVL(base, i) + SUBL(repeatL, i) + JZ(LabelRef("repeat_extend_back_end_" + name)) + + Label("repeat_extend_back_loop_" + name) + // if base <= nextemit {exit} + CMPL(base.As32(), nextEmit) + JLE(LabelRef("repeat_extend_back_end_" + name)) + // if src[i-1] == src[base-1] + tmp, tmp2 := GP64(), GP64() + MOVB(Mem{Base: src, Index: i, Scale: 1, Disp: -1}, tmp.As8()) + MOVB(Mem{Base: src, Index: base, Scale: 1, Disp: -1}, tmp2.As8()) + CMPB(tmp.As8(), tmp2.As8()) + JNE(LabelRef("repeat_extend_back_end_" + name)) + LEAL(Mem{Base: base, Disp: -1}, base) + DECL(i) + JNZ(LabelRef("repeat_extend_back_loop_" + name)) + } + Label("repeat_extend_back_end_" + name) + + // Base is now at start. Emit until base. + // d += emitLiteral(dst[d:], src[nextEmit:base]) + if true { + o.emitLiteralsDstP(nextEmitL, base, src, dst, "repeat_emit_"+name) + } + + // Extend forward + { + // s += 4 + checkRep + ADDL(U8(4+checkRep), s) + + if true { + // candidate := s - repeat + 4 + checkRep + MOVL(s, candidate) + SUBL(repeatL, candidate) // candidate = s - repeat + + // srcLeft = len(src) - s + srcLeft := GP64() + MOVQ(lenSrcQ, srcLeft) + SUBL(s, srcLeft.As32()) + assert(func(ok LabelRef) { + // if srcleft < maxint32: ok + CMPQ(srcLeft, U32(0x7fffffff)) + JL(ok) + }) + // Forward address + forwardStart := GP64() + LEAQ(Mem{Base: src, Index: s, Scale: 1}, forwardStart) + // End address + backStart := GP64() + LEAQ(Mem{Base: src, Index: candidate, Scale: 1}, backStart) + + length := o.matchLen("repeat_extend_"+name, forwardStart, backStart, srcLeft, LabelRef("repeat_extend_forward_end_"+name)) + forwardStart, backStart, srcLeft = nil, nil, nil + Label("repeat_extend_forward_end_" + name) + // s+= length + ADDL(length.As32(), s) + } + } + // Emit + if true { + // length = s-base + length := GP32() + MOVL(s, length) + SUBL(base.As32(), length) // length = s - base + + offsetVal := GP32() + MOVL(repeatL, offsetVal) + + if !o.snappy { + // if nextEmit == 0 {do copy instead...} + TESTL(nextEmit, nextEmit) + JZ(LabelRef("repeat_as_copy_" + name)) + + // Emit as repeat... + o.emitRepeat("match_repeat_"+name, length, offsetVal, nil, dst, LabelRef("repeat_end_emit_"+name)) + + // Emit as copy instead... + Label("repeat_as_copy_" + name) + } + o.emitCopy("repeat_as_copy_"+name, length, offsetVal, nil, dst, LabelRef("repeat_end_emit_"+name)) + + Label("repeat_end_emit_" + name) + // Store new dst and nextEmit + MOVL(s, nextEmitL) + } + // if s >= sLimit is picked up on next loop. + if false { + CMPL(s.As32(), sLimitL) + JGE(LabelRef("emit_remainder_" + name)) + } + JMP(LabelRef("search_loop_" + name)) + } + Label("no_repeat_found_" + name) + { + // Check candidates are ok. All must be < s and < len(src) + assert(func(ok LabelRef) { + tmp := GP64() + MOVQ(lenSrcQ, tmp) + CMPL(tmp.As32(), candidate) + JG(ok) + }) + assert(func(ok LabelRef) { + CMPL(s, candidate) + JG(ok) + }) + assert(func(ok LabelRef) { + tmp := GP64() + MOVQ(lenSrcQ, tmp) + CMPL(tmp.As32(), candidate2) + JG(ok) + }) + assert(func(ok LabelRef) { + CMPL(s, candidate2) + JG(ok) + }) + + CMPL(Mem{Base: src, Index: candidate, Scale: 1}, cv.As32()) + JEQ(LabelRef("candidate_match_" + name)) + + tmp := GP32() + // cv >>= 8 + SHRQ(U8(8), cv) + + // candidate = int(table[hash2]) - load early. + MOVL(table.Idx(hash2, 4), candidate) + assert(func(ok LabelRef) { + tmp := GP64() + MOVQ(lenSrcQ, tmp) + CMPL(tmp.As32(), candidate) + JG(ok) + }) + assert(func(ok LabelRef) { + // We may get s and s+1 + tmp := GP32() + LEAL(Mem{Base: s, Disp: 2}, tmp) + CMPL(tmp, candidate) + JG(ok) + }) + + LEAL(Mem{Base: s, Disp: 2}, tmp) + + //if uint32(cv>>8) == load32(src, candidate2) + CMPL(Mem{Base: src, Index: candidate2, Scale: 1}, cv.As32()) + JEQ(LabelRef("candidate2_match_" + name)) + + // table[hash2] = uint32(s + 2) + MOVL(tmp, table.Idx(hash2, 4)) + + // cv >>= 8 (>> 16 total) + SHRQ(U8(8), cv) + + // if uint32(cv>>16) == load32(src, candidate) + CMPL(Mem{Base: src, Index: candidate, Scale: 1}, cv.As32()) + JEQ(LabelRef("candidate3_match_" + name)) + + // No match found, next loop + // s = nextS + MOVL(nextSTempL, s) + JMP(LabelRef("search_loop_" + name)) + + // Matches candidate at s + 2 (3rd check) + Label("candidate3_match_" + name) + ADDL(U8(2), s) + JMP(LabelRef("candidate_match_" + name)) + + // Match at s + 1 (we calculated the hash, lets store it) + Label("candidate2_match_" + name) + // table[hash2] = uint32(s + 2) + MOVL(tmp, table.Idx(hash2, 4)) + // s++ + INCL(s) + MOVL(candidate2, candidate) + } + } + + Label("candidate_match_" + name) + // We have a match at 's' with src offset in "candidate" that matches at least 4 bytes. + // Extend backwards + if true { + ne := GP32() + MOVL(nextEmitL, ne) + TESTL(candidate, candidate) + JZ(LabelRef("match_extend_back_end_" + name)) + + // candidate is tested when decremented, so we loop back here. + Label("match_extend_back_loop_" + name) + // if s <= nextEmit {exit} + CMPL(s, ne) + JLE(LabelRef("match_extend_back_end_" + name)) + // if src[candidate-1] == src[s-1] + tmp, tmp2 := GP64(), GP64() + MOVB(Mem{Base: src, Index: candidate, Scale: 1, Disp: -1}, tmp.As8()) + MOVB(Mem{Base: src, Index: s, Scale: 1, Disp: -1}, tmp2.As8()) + CMPB(tmp.As8(), tmp2.As8()) + JNE(LabelRef("match_extend_back_end_" + name)) + LEAL(Mem{Base: s, Disp: -1}, s) + DECL(candidate) + JZ(LabelRef("match_extend_back_end_" + name)) + JMP(LabelRef("match_extend_back_loop_" + name)) + } + Label("match_extend_back_end_" + name) + + // Bail if we exceed the maximum size. + if true { + // tmp = s-nextEmit + tmp := GP64() + MOVL(s, tmp.As32()) + SUBL(nextEmitL, tmp.As32()) + // tmp = &dst + s-nextEmit + LEAQ(Mem{Base: dst, Index: tmp, Scale: 1, Disp: literalMaxOverhead}, tmp) + CMPQ(tmp, dstLimitPtrQ) + JL(LabelRef("match_dst_size_check_" + name)) + ri, err := ReturnIndex(0).Resolve() + if err != nil { + panic(err) + } + MOVQ(U32(0), ri.Addr) + RET() + } + Label("match_dst_size_check_" + name) + { + base := GP32() + MOVL(s, base.As32()) + o.emitLiteralsDstP(nextEmitL, base, src, dst, "match_emit_"+name) + } + cv := GP64() + Label("match_nolit_loop_" + name) + { + // Update repeat + { + // repeat = base - candidate + repeatVal := GP64().As32() + MOVL(s, repeatVal) + SUBL(candidate, repeatVal) + MOVL(repeatVal, repeatL) + } + // s+=4, candidate+=4 + ADDL(U8(4), s) + ADDL(U8(4), candidate) + // Extend the 4-byte match as long as possible and emit copy. + { + assert(func(ok LabelRef) { + // s must be > candidate cannot be equal. + CMPL(s, candidate) + JG(ok) + }) + // srcLeft = len(src) - s + srcLeft := GP64() + MOVQ(lenSrcQ, srcLeft) + SUBL(s, srcLeft.As32()) + assert(func(ok LabelRef) { + // if srcleft < maxint32: ok + CMPQ(srcLeft, U32(0x7fffffff)) + JL(ok) + }) + + a, b := GP64(), GP64() + LEAQ(Mem{Base: src, Index: s, Scale: 1}, a) + LEAQ(Mem{Base: src, Index: candidate, Scale: 1}, b) + length := o.matchLen("match_nolit_"+name, + a, b, + srcLeft, + LabelRef("match_nolit_end_"+name), + ) + Label("match_nolit_end_" + name) + assert(func(ok LabelRef) { + CMPL(length.As32(), U32(math.MaxInt32)) + JL(ok) + }) + a, b, srcLeft = nil, nil, nil + + // s += length (length is destroyed, use it now) + ADDL(length.As32(), s) + + // Load offset from repeat value. + offset := GP64() + MOVL(repeatL, offset.As32()) + + // length += 4 + ADDL(U8(4), length.As32()) + MOVL(s, nextEmitL) // nextEmit = s + o.emitCopy("match_nolit_"+name, length, offset, nil, dst, LabelRef("match_nolit_emitcopy_end_"+name)) + Label("match_nolit_emitcopy_end_" + name) + + // if s >= sLimit { end } + { + CMPL(s.As32(), sLimitL) + JGE(LabelRef("emit_remainder_" + name)) + } + // Start load s-2 as early as possible... + MOVQ(Mem{Base: src, Index: s, Scale: 1, Disp: -2}, cv) + // Bail if we exceed the maximum size. + { + CMPQ(dst, dstLimitPtrQ) + JL(LabelRef("match_nolit_dst_ok_" + name)) + ri, err := ReturnIndex(0).Resolve() + if err != nil { + panic(err) + } + MOVQ(U32(0), ri.Addr) + RET() + Label("match_nolit_dst_ok_" + name) + } + } + // cv must be set to value at s-2 before arriving here + { + // Check for an immediate match, otherwise start search at s+1 + // Index s-2 + hasher := hashN(hashBytes, tableBits) + hash0, hash1 := GP64(), GP64() + MOVQ(cv, hash0) // src[s-2] + SHRQ(U8(16), cv) + MOVQ(cv, hash1) // src[s] + hasher.hash(hash0) + hasher.hash(hash1) + sm2 := GP32() // s - 2 + LEAL(Mem{Base: s, Disp: -2}, sm2) + assert(func(ok LabelRef) { + CMPQ(hash0, U32(tableSize)) + JL(ok) + }) + assert(func(ok LabelRef) { + CMPQ(hash1, U32(tableSize)) + JL(ok) + }) + addr := GP64() + LEAQ(table.Idx(hash1, 4), addr) + MOVL(Mem{Base: addr}, candidate) + MOVL(sm2, table.Idx(hash0, 4)) + MOVL(s, Mem{Base: addr}) + CMPL(Mem{Base: src, Index: candidate, Scale: 1}, cv.As32()) + JEQ(LabelRef("match_nolit_loop_" + name)) + INCL(s) + } + JMP(LabelRef("search_loop_" + name)) + } + + Label("emit_remainder_" + name) + // Bail if we exceed the maximum size. + // if d+len(src)-nextEmitL > dstLimitPtrQ { return 0 + { + // remain = len(src) - nextEmit + remain := GP64() + MOVQ(lenSrcQ, remain) + SUBL(nextEmitL, remain.As32()) + + dstExpect := GP64() + // dst := dst + (len(src)-nextEmitL) + + LEAQ(Mem{Base: dst, Index: remain, Scale: 1, Disp: literalMaxOverhead}, dstExpect) + CMPQ(dstExpect, dstLimitPtrQ) + JL(LabelRef("emit_remainder_ok_" + name)) + ri, err := ReturnIndex(0).Resolve() + if err != nil { + panic(err) + } + MOVQ(U32(0), ri.Addr) + RET() + Label("emit_remainder_ok_" + name) + } + // emitLiteral(dst[d:], src[nextEmitL:]) + emitEnd := GP64() + MOVQ(lenSrcQ, emitEnd) + + // Emit final literals. + o.emitLiteralsDstP(nextEmitL, emitEnd, src, dst, "emit_remainder_"+name) + + // Assert size is < limit + assert(func(ok LabelRef) { + // if dstBaseQ < dstLimitPtrQ: ok + CMPQ(dst, dstLimitPtrQ) + JL(ok) + }) + + // length := start - base (ptr arithmetic) + length := GP64() + base := Load(Param("dst").Base(), GP64()) + MOVQ(dst, length) + SUBQ(base, length) + + // Assert size is < len(src) + assert(func(ok LabelRef) { + // if len(src) >= length: ok + CMPQ(lenSrcQ, length) + JGE(ok) + }) + // Assert size is < len(dst) + assert(func(ok LabelRef) { + // if len(dst) >= length: ok + CMPQ(lenDstQ, length) + JGE(ok) + }) + Store(length, ReturnIndex(0)) + RET() +} + +// emitLiterals emits literals from nextEmit to base, updates nextEmit, dstBase. +// Checks if base == nextemit. +// src & base are untouched. +func (o options) emitLiterals(nextEmitL Mem, base reg.GPVirtual, src reg.GPVirtual, dstBase Mem, name string) { + nextEmit, litLen, dstBaseTmp, litBase := GP32(), GP32(), GP64(), GP64() + MOVL(nextEmitL, nextEmit) + CMPL(nextEmit, base.As32()) + JEQ(LabelRef("emit_literal_skip_" + name)) + MOVL(base.As32(), litLen.As32()) + + // Base is now next emit. + MOVL(base.As32(), nextEmitL) + + // litBase = src[nextEmitL:] + LEAQ(Mem{Base: src, Index: nextEmit, Scale: 1}, litBase) + SUBL(nextEmit, litLen.As32()) // litlen = base - nextEmit + + // Load (and store when we return) + MOVQ(dstBase, dstBaseTmp) + o.emitLiteral(name, litLen, nil, dstBaseTmp, litBase, LabelRef("emit_literal_done_"+name), true) + Label("emit_literal_done_" + name) + + // Emitted length must be > litlen. + // We have already checked for len(0) above. + assert(func(ok LabelRef) { + tmp := GP64() + MOVQ(dstBaseTmp, tmp) + SUBQ(dstBase, tmp) // tmp = dstBaseTmp - dstBase + // if tmp > litLen: ok + CMPQ(tmp, litLen.As64()) + JG(ok) + }) + // Store updated dstBase + MOVQ(dstBaseTmp, dstBase) + Label("emit_literal_skip_" + name) +} + +// emitLiterals emits literals from nextEmit to base, updates nextEmit, dstBase. +// Checks if base == nextemit. +// src & base are untouched. +func (o options) emitLiteralsDstP(nextEmitL Mem, base reg.GPVirtual, src, dst reg.GPVirtual, name string) { + nextEmit, litLen, litBase := GP32(), GP32(), GP64() + MOVL(nextEmitL, nextEmit) + CMPL(nextEmit, base.As32()) + JEQ(LabelRef("emit_literal_done_" + name)) + MOVL(base.As32(), litLen.As32()) + + // Base is now next emit. + MOVL(base.As32(), nextEmitL) + + // litBase = src[nextEmitL:] + LEAQ(Mem{Base: src, Index: nextEmit, Scale: 1}, litBase) + SUBL(nextEmit, litLen.As32()) // litlen = base - nextEmit + + // Load (and store when we return) + o.emitLiteral(name, litLen, nil, dst, litBase, LabelRef("emit_literal_done_"+name), true) + Label("emit_literal_done_" + name) +} + +type hashGen struct { + bytes int + tablebits int + mulreg reg.GPVirtual +} + +// hashN uses multiply to get a 'output' hash on the hash of the lowest 'bytes' bytes in value. +func hashN(hashBytes, tablebits int) hashGen { + h := hashGen{ + bytes: hashBytes, + tablebits: tablebits, + mulreg: GP64(), + } + primebytes := uint64(0) + switch hashBytes { + case 3: + primebytes = 506832829 + case 4: + primebytes = 2654435761 + case 5: + primebytes = 889523592379 + case 6: + primebytes = 227718039650203 + case 7: + primebytes = 58295818150454627 + case 8: + primebytes = 0xcf1bbcdcb7a56463 + default: + panic("invalid hash length") + } + MOVQ(Imm(primebytes), h.mulreg) + return h +} + +// hash uses multiply to get hash of the value. +func (h hashGen) hash(val reg.GPVirtual) { + // Move value to top of register. + SHLQ(U8(64-8*h.bytes), val) + IMULQ(h.mulreg, val) + // Move value to bottom + SHRQ(U8(64-h.tablebits), val) +} + +func (o options) genEmitLiteral() { + TEXT("emitLiteral", NOSPLIT, "func(dst, lit []byte) int") + Doc("emitLiteral writes a literal chunk and returns the number of bytes written.", "", + "It assumes that:", + " dst is long enough to hold the encoded bytes", + " 0 <= len(lit) && len(lit) <= math.MaxUint32", "") + Pragma("noescape") + + dstBase, litBase, litLen, retval := GP64(), GP64(), GP64(), GP64() + Load(Param("lit").Len(), litLen) + Load(Param("dst").Base(), dstBase) + Load(Param("lit").Base(), litBase) + TESTQ(litLen, litLen) + JZ(LabelRef("emit_literal_end_standalone_skip")) + o.emitLiteral("standalone", litLen, retval, dstBase, litBase, "emit_literal_end_standalone", false) + + Label("emit_literal_end_standalone_skip") + XORQ(retval, retval) + + Label("emit_literal_end_standalone") + Store(retval, ReturnIndex(0)) + RET() + +} + +// emitLiteral can be used for inlining an emitLiteral call. +// litLen must be > 0. +// stack must have at least 32 bytes. +// retval will contain emitted bytes, but can be nil if this is not interesting. +// dstBase and litBase are updated. +// Uses 2 GP registers. With AVX 4 registers. +// If updateDst is true dstBase will have the updated end pointer and an additional register will be used. +func (o options) emitLiteral(name string, litLen, retval, dstBase, litBase reg.GPVirtual, end LabelRef, updateDst bool) { + n := GP32() + n16 := GP32() + + // litLen must be > 0 + assert(func(ok LabelRef) { + TESTL(litLen.As32(), litLen.As32()) + JNZ(ok) + }) + + // We always add litLen bytes + if retval != nil { + MOVL(litLen.As32(), retval.As32()) + } + // n = litlen - 1 + LEAL(Mem{Base: litLen.As32(), Disp: -1}, n) + + // Find number of bytes to emit for tag. + CMPL(n.As32(), U8(60)) + JLT(LabelRef("one_byte_" + name)) + CMPL(n.As32(), U32(1<<8)) + JLT(LabelRef("two_bytes_" + name)) + CMPL(n.As32(), U32(1<<16)) + JLT(LabelRef("three_bytes_" + name)) + CMPL(n.As32(), U32(1<<24)) + JLT(LabelRef("four_bytes_" + name)) + + Label("five_bytes_" + name) + MOVB(U8(252), Mem{Base: dstBase}) + MOVL(n.As32(), Mem{Base: dstBase, Disp: 1}) + if retval != nil { + ADDQ(U8(5), retval) + } + ADDQ(U8(5), dstBase) + JMP(LabelRef("memmove_long_" + name)) + + Label("four_bytes_" + name) + MOVL(n, n16) + SHRL(U8(16), n16.As32()) + MOVB(U8(248), Mem{Base: dstBase}) + MOVW(n.As16(), Mem{Base: dstBase, Disp: 1}) + MOVB(n16.As8(), Mem{Base: dstBase, Disp: 3}) + if retval != nil { + ADDQ(U8(4), retval) + } + ADDQ(U8(4), dstBase) + JMP(LabelRef("memmove_long_" + name)) + + Label("three_bytes_" + name) + MOVB(U8(0xf4), Mem{Base: dstBase}) + MOVW(n.As16(), Mem{Base: dstBase, Disp: 1}) + if retval != nil { + ADDQ(U8(3), retval) + } + ADDQ(U8(3), dstBase) + JMP(LabelRef("memmove_long_" + name)) + + Label("two_bytes_" + name) + MOVB(U8(0xf0), Mem{Base: dstBase}) + MOVB(n.As8(), Mem{Base: dstBase, Disp: 1}) + if retval != nil { + ADDQ(U8(2), retval) + } + ADDQ(U8(2), dstBase) + CMPL(n.As32(), U8(64)) + JL(LabelRef("memmove_" + name)) + JMP(LabelRef("memmove_long_" + name)) + + Label("one_byte_" + name) + SHLB(U8(2), n.As8()) + MOVB(n.As8(), Mem{Base: dstBase}) + if retval != nil { + ADDQ(U8(1), retval) + } + ADDQ(U8(1), dstBase) + // Fallthrough + + Label("memmove_" + name) + + // copy(dst[i:], lit) + dstEnd := GP64() + copyEnd := end + if updateDst { + copyEnd = LabelRef("memmove_end_copy_" + name) + LEAQ(Mem{Base: dstBase, Index: litLen, Scale: 1}, dstEnd) + } + length := GP64() + MOVL(litLen.As32(), length.As32()) + + // updates litBase. + o.genMemMoveShort("emit_lit_memmove_"+name, dstBase, litBase, length, copyEnd) + + if updateDst { + Label("memmove_end_copy_" + name) + MOVQ(dstEnd, dstBase) + } + JMP(end) + + // > 32 bytes + Label("memmove_long_" + name) + + // copy(dst[i:], lit) + dstEnd = GP64() + copyEnd = end + if updateDst { + copyEnd = LabelRef("memmove_end_copy_long_" + name) + LEAQ(Mem{Base: dstBase, Index: litLen, Scale: 1}, dstEnd) + } + length = GP64() + MOVL(litLen.As32(), length.As32()) + + // updates litBase. + o.genMemMoveLong("emit_lit_memmove_long_"+name, dstBase, litBase, length, copyEnd) + + if updateDst { + Label("memmove_end_copy_long_" + name) + MOVQ(dstEnd, dstBase) + } + JMP(end) + // Should be unreachable + if debug { + INT(Imm(3)) + } + return +} + +// genEmitRepeat generates a standlone emitRepeat. +func (o options) genEmitRepeat() { + TEXT("emitRepeat", NOSPLIT, "func(dst []byte, offset, length int) int") + Doc("emitRepeat writes a repeat chunk and returns the number of bytes written.", + "Length must be at least 4 and < 1<<32", "") + Pragma("noescape") + + dstBase, offset, length, retval := GP64(), GP64(), GP64(), GP64() + + // retval = 0 + XORQ(retval, retval) + + Load(Param("dst").Base(), dstBase) + Load(Param("offset"), offset) + Load(Param("length"), length) + o.emitRepeat("standalone", length, offset, retval, dstBase, LabelRef("gen_emit_repeat_end")) + Label("gen_emit_repeat_end") + Store(retval, ReturnIndex(0)) + RET() +} + +// emitRepeat can be used for inlining an emitRepeat call. +// length >= 4 and < 1<<32 +// length is modified. dstBase is updated. retval is added to input. +// retval can be nil. +// Will jump to end label when finished. +// Uses 1 GP register. +func (o options) emitRepeat(name string, length, offset, retval, dstBase reg.GPVirtual, end LabelRef) { + Label("emit_repeat_again_" + name) + tmp := GP32() + MOVL(length.As32(), tmp) // Copy length + // length -= 4 + LEAL(Mem{Base: length, Disp: -4}, length.As32()) + + // if length <= 4 (use copied value) + CMPL(tmp.As32(), U8(8)) + JLE(LabelRef("repeat_two_" + name)) + + // length < 8 && offset < 2048 + CMPL(tmp.As32(), U8(12)) + JGE(LabelRef("cant_repeat_two_offset_" + name)) + CMPL(offset.As32(), U32(2048)) + JLT(LabelRef("repeat_two_offset_" + name)) + + const maxRepeat = ((1 << 24) - 1) + 65536 + Label("cant_repeat_two_offset_" + name) + CMPL(length.As32(), U32((1<<8)+4)) + JLT(LabelRef("repeat_three_" + name)) // if length < (1<<8)+4 + CMPL(length.As32(), U32((1<<16)+(1<<8))) + JLT(LabelRef("repeat_four_" + name)) // if length < (1 << 16) + (1 << 8) + CMPL(length.As32(), U32(maxRepeat)) + JLT(LabelRef("repeat_five_" + name)) // If less than 24 bits to represent. + + // We have have more than 24 bits + // Emit so we have at least 4 bytes left. + LEAL(Mem{Base: length, Disp: -(maxRepeat - 4)}, length.As32()) // length -= (maxRepeat - 4) + MOVW(U16(7<<2|tagCopy1), Mem{Base: dstBase}) // dst[0] = 7<<2 | tagCopy1, dst[1] = 0 + MOVW(U16(65531), Mem{Base: dstBase, Disp: 2}) // 0xfffb + MOVB(U8(255), Mem{Base: dstBase, Disp: 4}) + ADDQ(U8(5), dstBase) + if retval != nil { + ADDQ(U8(5), retval) + } + JMP(LabelRef("emit_repeat_again_" + name)) + + // Must be able to be within 5 bytes. + Label("repeat_five_" + name) + LEAL(Mem{Base: length, Disp: -65536}, length.As32()) // length -= 65536 + MOVL(length.As32(), offset.As32()) + MOVW(U16(7<<2|tagCopy1), Mem{Base: dstBase}) // dst[0] = 7<<2 | tagCopy1, dst[1] = 0 + MOVW(length.As16(), Mem{Base: dstBase, Disp: 2}) // dst[2] = uint8(length), dst[3] = uint8(length >> 8) + SARL(U8(16), offset.As32()) // offset = length >> 16 + MOVB(offset.As8(), Mem{Base: dstBase, Disp: 4}) // dst[4] = length >> 16 + if retval != nil { + ADDQ(U8(5), retval) // i += 5 + } + ADDQ(U8(5), dstBase) // dst += 5 + JMP(end) + + Label("repeat_four_" + name) + LEAL(Mem{Base: length, Disp: -256}, length.As32()) // length -= 256 + MOVW(U16(6<<2|tagCopy1), Mem{Base: dstBase}) // dst[0] = 6<<2 | tagCopy1, dst[1] = 0 + MOVW(length.As16(), Mem{Base: dstBase, Disp: 2}) // dst[2] = uint8(length), dst[3] = uint8(length >> 8) + if retval != nil { + ADDQ(U8(4), retval) // i += 4 + } + ADDQ(U8(4), dstBase) // dst += 4 + JMP(end) + + Label("repeat_three_" + name) + LEAL(Mem{Base: length, Disp: -4}, length.As32()) // length -= 4 + MOVW(U16(5<<2|tagCopy1), Mem{Base: dstBase}) // dst[0] = 5<<2 | tagCopy1, dst[1] = 0 + MOVB(length.As8(), Mem{Base: dstBase, Disp: 2}) // dst[2] = uint8(length) + if retval != nil { + ADDQ(U8(3), retval) // i += 3 + } + ADDQ(U8(3), dstBase) // dst += 3 + JMP(end) + + Label("repeat_two_" + name) + // dst[0] = uint8(length)<<2 | tagCopy1, dst[1] = 0 + SHLL(U8(2), length.As32()) + ORL(U8(tagCopy1), length.As32()) + MOVW(length.As16(), Mem{Base: dstBase}) // dst[0] = 7<<2 | tagCopy1, dst[1] = 0 + if retval != nil { + ADDQ(U8(2), retval) // i += 2 + } + ADDQ(U8(2), dstBase) // dst += 2 + JMP(end) + + Label("repeat_two_offset_" + name) + // Emit the remaining copy, encoded as 2 bytes. + // dst[1] = uint8(offset) + // dst[0] = uint8(offset>>8)<<5 | uint8(length)<<2 | tagCopy1 + tmp = GP64() + XORQ(tmp, tmp) + // Use scale and displacement to shift and subtract values from length. + LEAL(Mem{Base: tmp, Index: length, Scale: 4, Disp: tagCopy1}, length.As32()) + MOVB(offset.As8(), Mem{Base: dstBase, Disp: 1}) // Store offset lower byte + SARL(U8(8), offset.As32()) // Remove lower + SHLL(U8(5), offset.As32()) // Shift back up + ORL(offset.As32(), length.As32()) // OR result + MOVB(length.As8(), Mem{Base: dstBase, Disp: 0}) + if retval != nil { + ADDQ(U8(2), retval) // i += 2 + } + ADDQ(U8(2), dstBase) // dst += 2 + + JMP(end) +} + +// 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 <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 + +// genEmitCopy generates a standlone emitCopy +func (o options) genEmitCopy() { + TEXT("emitCopy", NOSPLIT, "func(dst []byte, offset, length int) int") + Doc("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 <= math.MaxUint32", + " 4 <= length && length <= 1 << 24", "") + Pragma("noescape") + + dstBase, offset, length, retval := GP64(), GP64(), GP64(), GP64() + + // i := 0 + XORQ(retval, retval) + + Load(Param("dst").Base(), dstBase) + Load(Param("offset"), offset) + Load(Param("length"), length) + o.emitCopy("standalone", length, offset, retval, dstBase, LabelRef("gen_emit_copy_end")) + Label("gen_emit_copy_end") + Store(retval, ReturnIndex(0)) + RET() +} + +// 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 <= math.MaxUint32 +// 4 <= length && length <= 1 << 24 + +// genEmitCopy generates a standlone emitCopy +func (o options) genEmitCopyNoRepeat() { + TEXT("emitCopyNoRepeat", NOSPLIT, "func(dst []byte, offset, length int) int") + Doc("emitCopyNoRepeat 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 <= math.MaxUint32", + " 4 <= length && length <= 1 << 24", "") + Pragma("noescape") + + dstBase, offset, length, retval := GP64(), GP64(), GP64(), GP64() + + // i := 0 + XORQ(retval, retval) + + Load(Param("dst").Base(), dstBase) + Load(Param("offset"), offset) + Load(Param("length"), length) + o.emitCopy("standalone_snappy", length, offset, retval, dstBase, "gen_emit_copy_end_snappy") + Label("gen_emit_copy_end_snappy") + Store(retval, ReturnIndex(0)) + RET() +} + +const ( + tagLiteral = 0x00 + tagCopy1 = 0x01 + tagCopy2 = 0x02 + tagCopy4 = 0x03 +) + +// emitCopy can be used for inlining an emitCopy call. +// length is modified (and junk). dstBase is updated. retval is added to input. +// retval can be nil. +// Will jump to end label when finished. +// Uses 2 GP registers. +func (o options) emitCopy(name string, length, offset, retval, dstBase reg.GPVirtual, end LabelRef) { + //if offset >= 65536 { + CMPL(offset.As32(), U32(65536)) + JL(LabelRef("two_byte_offset_" + name)) + + // offset is >= 65536 + // if length <= 64 goto four_bytes_remain_ + Label("four_bytes_loop_back_" + name) + CMPL(length.As32(), U8(64)) + JLE(LabelRef("four_bytes_remain_" + name)) + + // Emit a length 64 copy, encoded as 5 bytes. + // dst[0] = 63<<2 | tagCopy4 + MOVB(U8(63<<2|tagCopy4), Mem{Base: dstBase}) + // dst[4] = uint8(offset >> 24) + // dst[3] = uint8(offset >> 16) + // dst[2] = uint8(offset >> 8) + // dst[1] = uint8(offset) + MOVL(offset.As32(), Mem{Base: dstBase, Disp: 1}) + // length -= 64 + LEAL(Mem{Base: length, Disp: -64}, length.As32()) + if retval != nil { + ADDQ(U8(5), retval) // i+=5 + } + ADDQ(U8(5), dstBase) // dst+=5 + + // if length >= 4 { + CMPL(length.As32(), U8(4)) + JL(LabelRef("four_bytes_remain_" + name)) + + // Emit remaining as repeats + // return 5 + emitRepeat(dst[5:], offset, length) + // Inline call to emitRepeat. Will jump to end + if !o.snappy { + o.emitRepeat(name+"_emit_copy", length, offset, retval, dstBase, end) + } + JMP(LabelRef("four_bytes_loop_back_" + name)) + + Label("four_bytes_remain_" + name) + // if length == 0 { + // return i + // } + TESTL(length.As32(), length.As32()) + JZ(end) + + // Emit a copy, offset encoded as 4 bytes. + // dst[i+0] = uint8(length-1)<<2 | tagCopy4 + // dst[i+1] = uint8(offset) + // dst[i+2] = uint8(offset >> 8) + // dst[i+3] = uint8(offset >> 16) + // dst[i+4] = uint8(offset >> 24) + tmp := GP64() + MOVB(U8(tagCopy4), tmp.As8()) + // Use displacement to subtract 1 from upshifted length. + LEAL(Mem{Base: tmp, Disp: -(1 << 2), Index: length, Scale: 4}, length.As32()) + MOVB(length.As8(), Mem{Base: dstBase}) + MOVL(offset.As32(), Mem{Base: dstBase, Disp: 1}) + // return i + 5 + if retval != nil { + ADDQ(U8(5), retval) + } + ADDQ(U8(5), dstBase) + JMP(end) + + Label("two_byte_offset_" + name) + // Offset no more than 2 bytes. + + //if length > 64 { + CMPL(length.As32(), U8(64)) + JLE(LabelRef("two_byte_offset_short_" + name)) + // Emit a length 60 copy, encoded as 3 bytes. + // Emit remaining as repeat value (minimum 4 bytes). + // dst[2] = uint8(offset >> 8) + // dst[1] = uint8(offset) + // dst[0] = 59<<2 | tagCopy2 + MOVB(U8(59<<2|tagCopy2), Mem{Base: dstBase}) + MOVW(offset.As16(), Mem{Base: dstBase, Disp: 1}) + // length -= 60 + LEAL(Mem{Base: length, Disp: -60}, length.As32()) + + // Emit remaining as repeats, at least 4 bytes remain. + // return 3 + emitRepeat(dst[3:], offset, length) + //} + ADDQ(U8(3), dstBase) + if retval != nil { + ADDQ(U8(3), retval) + } + // Inline call to emitRepeat. Will jump to end + if !o.snappy { + o.emitRepeat(name+"_emit_copy_short", length, offset, retval, dstBase, end) + } + JMP(LabelRef("two_byte_offset_" + name)) + + Label("two_byte_offset_short_" + name) + //if length >= 12 || offset >= 2048 { + CMPL(length.As32(), U8(12)) + JGE(LabelRef("emit_copy_three_" + name)) + CMPL(offset.As32(), U32(2048)) + JGE(LabelRef("emit_copy_three_" + name)) + + // Emit the remaining copy, encoded as 2 bytes. + // dst[1] = uint8(offset) + // dst[0] = uint8(offset>>8)<<5 | uint8(length-4)<<2 | tagCopy1 + tmp = GP64() + MOVB(U8(tagCopy1), tmp.As8()) + // Use scale and displacement to shift and subtract values from length. + LEAL(Mem{Base: tmp, Index: length, Scale: 4, Disp: -(4 << 2)}, length.As32()) + MOVB(offset.As8(), Mem{Base: dstBase, Disp: 1}) // Store offset lower byte + SHRL(U8(8), offset.As32()) // Remove lower + SHLL(U8(5), offset.As32()) // Shift back up + ORL(offset.As32(), length.As32()) // OR result + MOVB(length.As8(), Mem{Base: dstBase, Disp: 0}) + if retval != nil { + ADDQ(U8(2), retval) // i += 2 + } + ADDQ(U8(2), dstBase) // dst += 2 + // return 2 + JMP(end) + + Label("emit_copy_three_" + name) + // // Emit the remaining copy, encoded as 3 bytes. + // dst[2] = uint8(offset >> 8) + // dst[1] = uint8(offset) + // dst[0] = uint8(length-1)<<2 | tagCopy2 + tmp = GP64() + MOVB(U8(tagCopy2), tmp.As8()) + LEAL(Mem{Base: tmp, Disp: -(1 << 2), Index: length, Scale: 4}, length.As32()) + MOVB(length.As8(), Mem{Base: dstBase}) + MOVW(offset.As16(), Mem{Base: dstBase, Disp: 1}) + // return 3 + if retval != nil { + ADDQ(U8(3), retval) // i += 3 + } + ADDQ(U8(3), dstBase) // dst += 3 + JMP(end) +} + +// func memmove(to, from unsafe.Pointer, n uintptr) +// src and dst may not overlap. +// Non AVX uses 2 GP register, 16 SSE2 registers. +// AVX uses 4 GP registers 16 AVX/SSE registers. +// All passed registers may be updated. +// Length must be 1 -> 64 bytes +func (o options) genMemMoveShort(name string, dst, src, length reg.GPVirtual, end LabelRef) { + AX, CX := GP64(), GP64() + name += "_memmove_" + + // Only enable if length can be 0. + if false { + TESTQ(length, length) + JEQ(end) + } + assert(func(ok LabelRef) { + CMPQ(length, U8(64)) + JBE(ok) + }) + assert(func(ok LabelRef) { + TESTQ(length, length) + JNZ(ok) + }) + Label(name + "tail") + CMPQ(length, U8(3)) + JB(LabelRef(name + "move_1or2")) + JE(LabelRef(name + "move_3")) + CMPQ(length, U8(8)) + JB(LabelRef(name + "move_4through7")) + CMPQ(length, U8(16)) + JBE(LabelRef(name + "move_8through16")) + CMPQ(length, U8(32)) + JBE(LabelRef(name + "move_17through32")) + if debug { + CMPQ(length, U8(64)) + JBE(LabelRef(name + "move_33through64")) + INT(U8(3)) + } + JMP(LabelRef(name + "move_33through64")) + + //genMemMoveLong(name, dst, src, length, end) + + Label(name + "move_1or2") + MOVB(Mem{Base: src}, AX.As8()) + MOVB(Mem{Base: src, Disp: -1, Index: length, Scale: 1}, CX.As8()) + MOVB(AX.As8(), Mem{Base: dst}) + MOVB(CX.As8(), Mem{Base: dst, Disp: -1, Index: length, Scale: 1}) + JMP(end) + + Label(name + "move_3") + MOVW(Mem{Base: src}, AX.As16()) + MOVB(Mem{Base: src, Disp: 2}, CX.As8()) + MOVW(AX.As16(), Mem{Base: dst}) + MOVB(CX.As8(), Mem{Base: dst, Disp: 2}) + JMP(end) + + Label(name + "move_4through7") + MOVL(Mem{Base: src}, AX.As32()) + MOVL(Mem{Base: src, Disp: -4, Index: length, Scale: 1}, CX.As32()) + MOVL(AX.As32(), Mem{Base: dst}) + MOVL(CX.As32(), Mem{Base: dst, Disp: -4, Index: length, Scale: 1}) + JMP(end) + + Label(name + "move_8through16") + MOVQ(Mem{Base: src}, AX) + MOVQ(Mem{Base: src, Disp: -8, Index: length, Scale: 1}, CX) + MOVQ(AX, Mem{Base: dst}) + MOVQ(CX, Mem{Base: dst, Disp: -8, Index: length, Scale: 1}) + JMP(end) + + Label(name + "move_17through32") + X0, X1, X2, X3 := XMM(), XMM(), XMM(), XMM() + + MOVOU(Mem{Base: src}, X0) + MOVOU(Mem{Base: src, Disp: -16, Index: length, Scale: 1}, X1) + MOVOU(X0, Mem{Base: dst}) + MOVOU(X1, Mem{Base: dst, Disp: -16, Index: length, Scale: 1}) + JMP(end) + + Label(name + "move_33through64") + MOVOU(Mem{Base: src}, X0) + MOVOU(Mem{Base: src, Disp: 16}, X1) + MOVOU(Mem{Base: src, Disp: -32, Index: length, Scale: 1}, X2) + MOVOU(Mem{Base: src, Disp: -16, Index: length, Scale: 1}, X3) + MOVOU(X0, Mem{Base: dst}) + MOVOU(X1, Mem{Base: dst, Disp: 16}) + MOVOU(X2, Mem{Base: dst, Disp: -32, Index: length, Scale: 1}) + MOVOU(X3, Mem{Base: dst, Disp: -16, Index: length, Scale: 1}) + JMP(end) +} + +// func genMemMoveLong(to, from unsafe.Pointer, n uintptr) +// src and dst may not overlap. +// length must be >= 64 bytes. +// Non AVX uses 2 GP register, 16 SSE2 registers. +// AVX uses 4 GP registers 16 AVX/SSE registers. +// All passed registers may be updated. +func (o options) genMemMoveLong(name string, dst, src, length reg.GPVirtual, end LabelRef) { + name += "large_" + + assert(func(ok LabelRef) { + CMPQ(length, U8(64)) + JAE(ok) + }) + + // These are disabled. + // AVX is ever so slightly faster, but it is disabled for simplicity. + const branchLoops = false + const avx = false && branchLoops + if branchLoops { + CMPQ(length, U8(128)) + JBE(LabelRef(name + "move_65through128")) + CMPQ(length, U32(256)) + JBE(LabelRef(name + "move_129through256")) + if avx { + JMP(LabelRef(name + "avxUnaligned")) + } else { + JMP(LabelRef(name + "forward_sse")) + } + + X0, X1, X2, X3, X4, X5, X6, X7 := XMM(), XMM(), XMM(), XMM(), XMM(), XMM(), XMM(), XMM() + X8, X9, X10, X11, X12, X13, X14, X15 := XMM(), XMM(), XMM(), XMM(), XMM(), XMM(), XMM(), XMM() + Label(name + "move_65through128") + MOVOU(Mem{Base: src}, X0) + MOVOU(Mem{Base: src, Disp: 16}, X1) + MOVOU(Mem{Base: src, Disp: 32}, X2) + MOVOU(Mem{Base: src, Disp: 48}, X3) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -64}, X12) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -48}, X13) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -32}, X14) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -16}, X15) + MOVOU(X0, Mem{Base: dst}) + MOVOU(X1, Mem{Base: dst, Disp: 16}) + MOVOU(X2, Mem{Base: dst, Disp: 32}) + MOVOU(X3, Mem{Base: dst, Disp: 48}) + MOVOU(X12, Mem{Base: dst, Index: length, Scale: 1, Disp: -64}) + MOVOU(X13, Mem{Base: dst, Index: length, Scale: 1, Disp: -48}) + MOVOU(X14, Mem{Base: dst, Index: length, Scale: 1, Disp: -32}) + MOVOU(X15, Mem{Base: dst, Index: length, Scale: 1, Disp: -16}) + JMP(end) + + Label(name + "move_129through256") + MOVOU(Mem{Base: src}, X0) + MOVOU(Mem{Base: src, Disp: 16}, X1) + MOVOU(Mem{Base: src, Disp: 32}, X2) + MOVOU(Mem{Base: src, Disp: 48}, X3) + MOVOU(Mem{Base: src, Disp: 64}, X4) + MOVOU(Mem{Base: src, Disp: 80}, X5) + MOVOU(Mem{Base: src, Disp: 96}, X6) + MOVOU(Mem{Base: src, Disp: 112}, X7) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -128}, X8) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -112}, X9) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -96}, X10) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -80}, X11) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -64}, X12) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -48}, X13) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -32}, X14) + MOVOU(Mem{Base: src, Index: length, Scale: 1, Disp: -16}, X15) + MOVOU(X0, Mem{Base: dst}) + MOVOU(X1, Mem{Base: dst, Disp: 16}) + MOVOU(X2, Mem{Base: dst, Disp: 32}) + MOVOU(X3, Mem{Base: dst, Disp: 48}) + MOVOU(X4, Mem{Base: dst, Disp: 64}) + MOVOU(X5, Mem{Base: dst, Disp: 80}) + MOVOU(X6, Mem{Base: dst, Disp: 96}) + MOVOU(X7, Mem{Base: dst, Disp: 112}) + MOVOU(X8, Mem{Base: dst, Index: length, Scale: 1, Disp: -128}) + MOVOU(X9, Mem{Base: dst, Index: length, Scale: 1, Disp: -112}) + MOVOU(X10, Mem{Base: dst, Index: length, Scale: 1, Disp: -96}) + MOVOU(X11, Mem{Base: dst, Index: length, Scale: 1, Disp: -80}) + MOVOU(X12, Mem{Base: dst, Index: length, Scale: 1, Disp: -64}) + MOVOU(X13, Mem{Base: dst, Index: length, Scale: 1, Disp: -48}) + MOVOU(X14, Mem{Base: dst, Index: length, Scale: 1, Disp: -32}) + MOVOU(X15, Mem{Base: dst, Index: length, Scale: 1, Disp: -16}) + JMP(end) + if avx { + Label(name + "avxUnaligned") + AX, CX, R8, R10 := GP64(), GP64(), GP64(), GP64() + // Memory layout on the source side + // src CX + // |<---------length before correction--------->| + // | |<--length corrected-->| | + // | | |<--- AX --->| + // |<-R11->| |<-128 bytes->| + // +----------------------------------------+ + // | Head | Body | Tail | + // +-------+------------------+-------------+ + // ^ ^ ^ + // | | | + // Save head into Y4 Save tail into X5..X12 + // | + // src+R11, where R11 = ((dst & -32) + 32) - dst + // Algorithm: + // 1. Unaligned save of the tail's 128 bytes + // 2. Unaligned save of the head's 32 bytes + // 3. Destination-aligned copying of body (128 bytes per iteration) + // 4. Put head on the new place + // 5. Put the tail on the new place + // It can be important to satisfy processor's pipeline requirements for + // small sizes as the cost of unaligned memory region copying is + // comparable with the cost of main loop. So code is slightly messed there. + // There is more clean implementation of that algorithm for bigger sizes + // where the cost of unaligned part copying is negligible. + // You can see it after gobble_big_data_fwd label. + Y0, Y1, Y2, Y3, Y4 := YMM(), YMM(), YMM(), YMM(), YMM() + + LEAQ(Mem{Base: src, Index: length, Scale: 1}, CX) + MOVQ(dst, R10) + // CX points to the end of buffer so we need go back slightly. We will use negative offsets there. + MOVOU(Mem{Base: CX, Disp: -0x80}, X5) + MOVOU(Mem{Base: CX, Disp: -0x70}, X6) + MOVQ(U32(0x80), AX) + + // Align destination address + ANDQ(U32(0xffffffe0), dst) + ADDQ(U8(32), dst) + // Continue tail saving. + MOVOU(Mem{Base: CX, Disp: -0x60}, X7) + MOVOU(Mem{Base: CX, Disp: -0x50}, X8) + // Make R8 delta between aligned and unaligned destination addresses. + MOVQ(dst, R8) + SUBQ(R10, R8) + // Continue tail saving. + MOVOU(Mem{Base: CX, Disp: -0x40}, X9) + MOVOU(Mem{Base: CX, Disp: -0x30}, X10) + // Let's make bytes-to-copy value adjusted as we've prepared unaligned part for copying. + SUBQ(R8, length) + // Continue tail saving. + MOVOU(Mem{Base: CX, Disp: -0x20}, X11) + MOVOU(Mem{Base: CX, Disp: -0x10}, X12) + // The tail will be put on its place after main body copying. + // It's time for the unaligned heading part. + VMOVDQU(Mem{Base: src}, Y4) + // Adjust source address to point past head. + ADDQ(R8, src) + SUBQ(AX, length) + + // Aligned memory copying there + Label(name + "gobble_128_loop") + VMOVDQU(Mem{Base: src}, Y0) + VMOVDQU(Mem{Base: src, Disp: 0x20}, Y1) + VMOVDQU(Mem{Base: src, Disp: 0x40}, Y2) + VMOVDQU(Mem{Base: src, Disp: 0x60}, Y3) + ADDQ(AX, src) + VMOVDQA(Y0, Mem{Base: dst}) + VMOVDQA(Y1, Mem{Base: dst, Disp: 0x20}) + VMOVDQA(Y2, Mem{Base: dst, Disp: 0x40}) + VMOVDQA(Y3, Mem{Base: dst, Disp: 0x60}) + ADDQ(AX, dst) + SUBQ(AX, length) + JA(LabelRef(name + "gobble_128_loop")) + // Now we can store unaligned parts. + ADDQ(AX, length) + ADDQ(dst, length) + VMOVDQU(Y4, Mem{Base: R10}) + VZEROUPPER() + MOVOU(X5, Mem{Base: length, Disp: -0x80}) + MOVOU(X6, Mem{Base: length, Disp: -0x70}) + MOVOU(X7, Mem{Base: length, Disp: -0x60}) + MOVOU(X8, Mem{Base: length, Disp: -0x50}) + MOVOU(X9, Mem{Base: length, Disp: -0x40}) + MOVOU(X10, Mem{Base: length, Disp: -0x30}) + MOVOU(X11, Mem{Base: length, Disp: -0x20}) + MOVOU(X12, Mem{Base: length, Disp: -0x10}) + JMP(end) + + return + } + } + + // Store start and end for sse_tail + Label(name + "forward_sse") + X0, X1, X2, X3, X4, X5, X6, X7 := XMM(), XMM(), XMM(), XMM(), XMM(), XMM(), XMM(), XMM() + X8, X9, X10, X11 := XMM(), XMM(), XMM(), XMM() + + MOVOU(Mem{Base: src}, X0) + MOVOU(Mem{Base: src, Disp: 16}, X1) + MOVOU(Mem{Base: src, Disp: -32, Index: length, Scale: 1}, X2) + MOVOU(Mem{Base: src, Disp: -16, Index: length, Scale: 1}, X3) + + // forward (only) + dstAlign := GP64() + bigLoops := GP64() + MOVQ(length, bigLoops) + SHRQ(U8(7), bigLoops) // bigLoops = length / 128 + + MOVQ(dst, dstAlign) + ANDL(U32(31), dstAlign.As32()) + srcOff := GP64() + MOVQ(U32(64), srcOff) + SUBQ(dstAlign, srcOff) + + // Move 128 bytes/loop + DECQ(bigLoops) + JA(LabelRef(name + "forward_sse_loop_32")) + + // Can be moved inside loop for less regs. + srcPos := GP64() + LEAQ(Mem{Disp: -32, Base: src, Scale: 1, Index: srcOff}, srcPos) + dstPos := GP64() + LEAQ(Mem{Disp: -32, Base: dst, Scale: 1, Index: srcOff}, dstPos) + + Label(name + "big_loop_back") + + MOVOU(Mem{Disp: 0, Base: srcPos}, X4) + MOVOU(Mem{Disp: 16, Base: srcPos}, X5) + MOVOU(Mem{Disp: 32, Base: srcPos}, X6) + MOVOU(Mem{Disp: 48, Base: srcPos}, X7) + MOVOU(Mem{Disp: 64, Base: srcPos}, X8) + MOVOU(Mem{Disp: 80, Base: srcPos}, X9) + MOVOU(Mem{Disp: 96, Base: srcPos}, X10) + MOVOU(Mem{Disp: 112, Base: srcPos}, X11) + + MOVOA(X4, Mem{Disp: 0, Base: dstPos}) + MOVOA(X5, Mem{Disp: 16, Base: dstPos}) + MOVOA(X6, Mem{Disp: 32, Base: dstPos}) + MOVOA(X7, Mem{Disp: 48, Base: dstPos}) + MOVOA(X8, Mem{Disp: 64, Base: dstPos}) + MOVOA(X9, Mem{Disp: 80, Base: dstPos}) + MOVOA(X10, Mem{Disp: 96, Base: dstPos}) + MOVOA(X11, Mem{Disp: 112, Base: dstPos}) + ADDQ(U8(128), dstPos) + ADDQ(U8(128), srcPos) + ADDQ(U8(128), srcOff) // This could be outside the loop, but we lose a reg if we do. + DECQ(bigLoops) + JNA(LabelRef(name + "big_loop_back")) + + Label(name + "forward_sse_loop_32") + MOVOU(Mem{Disp: -32, Base: src, Scale: 1, Index: srcOff}, X4) + MOVOU(Mem{Disp: -16, Base: src, Scale: 1, Index: srcOff}, X5) + MOVOA(X4, Mem{Disp: -32, Base: dst, Scale: 1, Index: srcOff}) + MOVOA(X5, Mem{Disp: -16, Base: dst, Scale: 1, Index: srcOff}) + ADDQ(U8(32), srcOff) + CMPQ(length, srcOff) + JAE(LabelRef(name + "forward_sse_loop_32")) + + // sse_tail patches up the beginning and end of the transfer. + MOVOU(X0, Mem{Base: dst, Disp: 0}) + MOVOU(X1, Mem{Base: dst, Disp: 16}) + MOVOU(X2, Mem{Base: dst, Disp: -32, Index: length, Scale: 1}) + MOVOU(X3, Mem{Base: dst, Disp: -16, Index: length, Scale: 1}) + + JMP(end) + return +} + +// genMatchLen generates standalone matchLen. +func (o options) genMatchLen() { + TEXT("matchLen", NOSPLIT, "func(a, b []byte) int") + Doc("matchLen returns how many bytes match in a and b", "", + "It assumes that:", + " len(a) <= len(b)", "") + Pragma("noescape") + + aBase, bBase, length := GP64(), GP64(), GP64() + + Load(Param("a").Base(), aBase) + Load(Param("b").Base(), bBase) + Load(Param("a").Len(), length) + l := o.matchLen("standalone", aBase, bBase, length, LabelRef("gen_match_len_end")) + Label("gen_match_len_end") + Store(l.As64(), ReturnIndex(0)) + RET() +} + +// matchLen returns the number of matching bytes of a and b. +// len is the maximum number of bytes to match. +// Will jump to end when done and returns the length. +// Uses 2 GP registers. +func (o options) matchLen(name string, a, b, len reg.GPVirtual, end LabelRef) reg.GPVirtual { + if false { + return o.matchLenAlt(name, a, b, len, end) + } + tmp, matched := GP64(), GP32() + XORL(matched, matched) + + CMPL(len.As32(), U8(8)) + JL(LabelRef("matchlen_single_" + name)) + + Label("matchlen_loopback_" + name) + MOVQ(Mem{Base: a, Index: matched, Scale: 1}, tmp) + XORQ(Mem{Base: b, Index: matched, Scale: 1}, tmp) + TESTQ(tmp, tmp) + JZ(LabelRef("matchlen_loop_" + name)) + // Not all match. + BSFQ(tmp, tmp) + SARQ(U8(3), tmp) + LEAL(Mem{Base: matched, Index: tmp, Scale: 1}, matched) + JMP(end) + + // All 8 byte matched, update and loop. + Label("matchlen_loop_" + name) + LEAL(Mem{Base: len, Disp: -8}, len.As32()) + LEAL(Mem{Base: matched, Disp: 8}, matched) + CMPL(len.As32(), U8(8)) + JGE(LabelRef("matchlen_loopback_" + name)) + + // Less than 8 bytes left. + Label("matchlen_single_" + name) + TESTL(len.As32(), len.As32()) + JZ(end) + Label("matchlen_single_loopback_" + name) + MOVB(Mem{Base: a, Index: matched, Scale: 1}, tmp.As8()) + CMPB(Mem{Base: b, Index: matched, Scale: 1}, tmp.As8()) + JNE(end) + LEAL(Mem{Base: matched, Disp: 1}, matched) + DECL(len.As32()) + JNZ(LabelRef("matchlen_single_loopback_" + name)) + JMP(end) + return matched +} + +// matchLen returns the number of matching bytes of a and b. +// len is the maximum number of bytes to match. +// Will jump to end when done and returns the length. +// Uses 3 GP registers. +// It is better on longer matches. +func (o options) matchLenAlt(name string, a, b, len reg.GPVirtual, end LabelRef) reg.GPVirtual { + tmp, tmp2, matched := GP64(), GP64(), GP32() + XORL(matched, matched) + + CMPL(len.As32(), U8(16)) + JB(LabelRef("matchlen_short_" + name)) + + Label("matchlen_loopback_" + name) + MOVQ(Mem{Base: a}, tmp) + MOVQ(Mem{Base: a, Disp: 8}, tmp2) + XORQ(Mem{Base: b, Disp: 0}, tmp) + XORQ(Mem{Base: b, Disp: 8}, tmp2) + endTest := func(xored reg.GPVirtual, disp int, ok LabelRef) { + TESTQ(xored, xored) + JZ(ok) + // Not all match. + BSFQ(xored, xored) + SARQ(U8(3), xored) + LEAL(Mem{Base: matched, Index: xored, Scale: 1, Disp: disp}, matched) + JMP(end) + } + endTest(tmp, 0, LabelRef("matchlen_loop_tmp2_"+name)) + Label("matchlen_loop_tmp2_" + name) + endTest(tmp2, 8, LabelRef("matchlen_loop_"+name)) + + // All 16 byte matched, update and loop. + Label("matchlen_loop_" + name) + SUBL(U8(16), len.As32()) + ADDL(U8(16), matched) + ADDQ(U8(16), a) + ADDQ(U8(16), b) + CMPL(len.As32(), U8(16)) + JAE(LabelRef("matchlen_loopback_" + name)) + + // Test 4 bytes at the time... + Label("matchlen_short_" + name) + lenoff := 0 + if true { + lenoff = 4 + SUBL(U8(4), len.As32()) + JC(LabelRef("matchlen_single_resume_" + name)) + + Label("matchlen_four_loopback_" + name) + assert(func(ok LabelRef) { + CMPL(len.As32(), U32(math.MaxInt32)) + JL(ok) + }) + + MOVL(Mem{Base: a}, tmp.As32()) + XORL(Mem{Base: b}, tmp.As32()) + { + JZ(LabelRef("matchlen_four_loopback_next" + name)) + BSFL(tmp.As32(), tmp.As32()) + SARQ(U8(3), tmp) + LEAL(Mem{Base: matched, Index: tmp, Scale: 1}, matched) + JMP(end) + } + Label("matchlen_four_loopback_next" + name) + ADDL(U8(4), matched) + ADDQ(U8(4), a) + ADDQ(U8(4), b) + SUBL(U8(4), len.As32()) + JNC(LabelRef("matchlen_four_loopback_" + name)) + } + + // Test one at the time + Label("matchlen_single_resume_" + name) + if true { + // Less than 16 bytes left. + if lenoff > 0 { + ADDL(U8(lenoff), len.As32()) + } + TESTL(len.As32(), len.As32()) + JZ(end) + + Label("matchlen_single_loopback_" + name) + MOVB(Mem{Base: a}, tmp.As8()) + CMPB(Mem{Base: b}, tmp.As8()) + JNE(end) + INCL(matched) + INCQ(a) + INCQ(b) + DECL(len.As32()) + JNZ(LabelRef("matchlen_single_loopback_" + name)) + } + JMP(end) + return matched +} diff --git a/github.com/klauspost/compress/s2/_generate/go.mod b/github.com/klauspost/compress/s2/_generate/go.mod new file mode 100644 index 0000000..f7f3c44 --- /dev/null +++ b/github.com/klauspost/compress/s2/_generate/go.mod @@ -0,0 +1,5 @@ +module github.com/klauspost/compress + +go 1.13 + +require github.com/mmcloughlin/avo v0.0.0-20200523190732-4439b6b2c061 diff --git a/github.com/klauspost/compress/s2/_generate/go.sum b/github.com/klauspost/compress/s2/_generate/go.sum new file mode 100644 index 0000000..6494ef5 --- /dev/null +++ b/github.com/klauspost/compress/s2/_generate/go.sum @@ -0,0 +1,23 @@ +github.com/mmcloughlin/avo v0.0.0-20200523190732-4439b6b2c061 h1:UCU8+cLbbvyxi0sQ9fSeoEhZgvrrD9HKMtX6Gmc1vk8= +github.com/mmcloughlin/avo v0.0.0-20200523190732-4439b6b2c061/go.mod h1:wqKykBG2QzQDJEzvRkcS8x6MiSJkF52hXZsXcjaB3ls= +github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74= +golang.org/x/arch v0.0.0-20190909030613-46d78d1859ac/go.mod h1:flIaEI6LNU6xOCD5PaJvn9wGP0agmIOqjrtsKGRguv4= +golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w= +golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI= +golang.org/x/mod v0.2.0/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA= +golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg= +golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s= +golang.org/x/net v0.0.0-20200226121028-0de0cce0169b/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s= +golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM= +golang.org/x/sync v0.0.0-20190911185100-cd5d95a43a6e/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM= +golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY= +golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs= +golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ= +golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo= +golang.org/x/tools v0.0.0-20200425043458-8463f397d07c h1:iHhCR0b26amDCiiO+kBguKZom9aMF+NrFxh9zeKR/XU= +golang.org/x/tools v0.0.0-20200425043458-8463f397d07c/go.mod h1:EkVYQZoAsY45+roYkvgYkIh4xh/qjgUK9TdY2XT94GE= +golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0= +golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0= +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= +rsc.io/pdf v0.1.1/go.mod h1:n8OzWcQ6Sp37PL01nO98y4iUCRdTGarVfzxY20ICaU4= diff --git a/github.com/klauspost/compress/s2/cmd/internal/readahead/LICENSE b/github.com/klauspost/compress/s2/cmd/internal/readahead/LICENSE new file mode 100644 index 0000000..5cec7ee --- /dev/null +++ b/github.com/klauspost/compress/s2/cmd/internal/readahead/LICENSE @@ -0,0 +1,22 @@ +The MIT License (MIT) + +Copyright (c) 2015 Klaus Post + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. + diff --git a/github.com/klauspost/compress/s2/cmd/internal/readahead/README.md b/github.com/klauspost/compress/s2/cmd/internal/readahead/README.md new file mode 100644 index 0000000..e6d0e92 --- /dev/null +++ b/github.com/klauspost/compress/s2/cmd/internal/readahead/README.md @@ -0,0 +1,57 @@ +# readahead +Asynchronous read-ahead for Go readers + +This package will allow you to add readhead to any reader. This means a separate goroutine will perform reads from your upstream reader, so you can request from this reader without delay. + +This is helpful for splitting an input stream into concurrent processing, and also helps smooth out **bursts** of input or output. + +This should be fully transparent, except that once an error has been returned from the Reader, it will not recover. A panic will be caught and returned as an error. + +The readahead object also fulfills the [`io.WriterTo`](https://golang.org/pkg/io/#WriterTo) interface, which is likely to speed up `io.Copy` and other code that use the interface. + +See an introduction: [An Async Read-ahead Package for Go](https://blog.klauspost.com/an-async-read-ahead-package-for-go/) + +[![GoDoc][1]][2] [![Build Status][3]][4] + +[1]: https://godoc.org/github.com/klauspost/readahead?status.svg +[2]: https://godoc.org/github.com/klauspost/readahead +[3]: https://travis-ci.org/klauspost/readahead.svg +[4]: https://travis-ci.org/klauspost/readahead + +# usage + +To get the package use `go get -u github.com/klauspost/readahead`. + +Here is a simple example that does file copy. Error handling has been omitted for brevity. +```Go +input, _ := os.Open("input.txt") +output, _ := os.Create("output.txt") +defer input.Close() +defer output.Close() + +// Create a read-ahead Reader with default settings +ra := readahead.NewReader(input) +defer ra.Close() + +// Copy the content to our output +_, _ = io.Copy(output, ra) +``` + +# settings + +You can finetune the read-ahead for your specific use case, and adjust the number of buffers and the size of each buffer. + +The default the size of each buffer is 1MB, and there are 4 buffers. Do not make your buffers too small since there is a small overhead for passing buffers between goroutines. Other than that you are free to experiment with buffer sizes. + +# contributions + +On this project contributions in terms of new features is limited to: + +* Features that are widely usable and +* Features that have extensive tests + +This package is meant to be simple and stable, so therefore these strict requirements. + +# license + +This package is released under the MIT license. See the supplied LICENSE file for more info. diff --git a/github.com/klauspost/compress/s2/cmd/internal/readahead/reader.go b/github.com/klauspost/compress/s2/cmd/internal/readahead/reader.go new file mode 100644 index 0000000..84a2694 --- /dev/null +++ b/github.com/klauspost/compress/s2/cmd/internal/readahead/reader.go @@ -0,0 +1,275 @@ +// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file. + +// The readahead package will do asynchronous read-ahead from an input io.Reader +// and make the data available as an io.Reader. +// +// This should be fully transparent, except that once an error +// has been returned from the Reader, it will not recover. +// +// The readahead object also fulfills the io.WriterTo interface, which +// is likely to speed up copies. +// +// Package home: https://github.com/klauspost/readahead +// +package readahead + +import ( + "errors" + "fmt" + "io" +) + +type seekable struct { + *reader +} + +type ReadSeekCloser interface { + io.ReadCloser + io.Seeker +} + +type reader struct { + in io.Reader // Input reader + closer io.Closer // Optional closer + ready chan *buffer // Buffers ready to be handed to the reader + reuse chan *buffer // Buffers to reuse for input reading + exit chan struct{} // Closes when finished + buffers int // Number of buffers + size int // Size of each buffer + err error // If an error has occurred it is here + cur *buffer // Current buffer being served + exited chan struct{} // Channel is closed been the async reader shuts down +} + +// NewReaderSize returns a reader with a custom number of buffers and size. +// buffers is the number of queued buffers and size is the size of each +// buffer in bytes. +func NewReaderSize(rd io.Reader, buffers, size int) (res io.ReadCloser, err error) { + if size <= 0 { + return nil, fmt.Errorf("buffer size too small") + } + if buffers <= 0 { + return nil, fmt.Errorf("number of buffers too small") + } + if rd == nil { + return nil, fmt.Errorf("nil input reader supplied") + } + a := &reader{} + if _, ok := rd.(io.Seeker); ok { + res = &seekable{a} + } else { + res = a + } + a.init(rd, buffers, size) + return +} + +// initialize the reader +func (a *reader) init(rd io.Reader, buffers, size int) { + a.in = rd + a.ready = make(chan *buffer, buffers) + a.reuse = make(chan *buffer, buffers) + a.exit = make(chan struct{}, 0) + a.exited = make(chan struct{}, 0) + a.buffers = buffers + a.size = size + a.cur = nil + a.err = nil + + // Create buffers + for i := 0; i < buffers; i++ { + a.reuse <- newBuffer(size) + } + + // Start async reader + go func() { + // Ensure that when we exit this is signalled. + defer close(a.exited) + defer close(a.ready) + for { + select { + case b := <-a.reuse: + err := b.read(a.in) + a.ready <- b + if err != nil { + return + } + case <-a.exit: + return + } + } + }() +} + +// fill will check if the current buffer is empty and fill it if it is. +// If an error was returned at the end of the current buffer it is returned. +func (a *reader) fill() (err error) { + if a.cur.isEmpty() { + if a.cur != nil { + a.reuse <- a.cur + a.cur = nil + } + b, ok := <-a.ready + if !ok { + if a.err == nil { + a.err = errors.New("readahead: read after Close") + } + return a.err + } + a.cur = b + } + return nil +} + +// Read will return the next available data. +func (a *reader) Read(p []byte) (n int, err error) { + if a.err != nil { + return 0, a.err + } + // Swap buffer and maybe return error + err = a.fill() + if err != nil { + return 0, err + } + + // Copy what we can + n = copy(p, a.cur.buffer()) + a.cur.inc(n) + + // If at end of buffer, return any error, if present + if a.cur.isEmpty() { + a.err = a.cur.err + return n, a.err + } + return n, nil +} + +func (a *seekable) Seek(offset int64, whence int) (res int64, err error) { + //Not checking the result as seekable receiver guarantees it to be assertable + seeker, _ := a.in.(io.Seeker) + //Make sure the async routine is closed + select { + case <-a.exited: + case a.exit <- struct{}{}: + <-a.exited + } + if whence == io.SeekCurrent { + //If need to seek based on current position, take into consideration the bytes we read but the consumer + //doesn't know about + err = nil + for a.cur != nil { + if err = a.fill(); err == nil && a.cur != nil { + offset -= int64(len(a.cur.buffer())) + a.cur.offset = len(a.cur.buf) + } + } + } + //Seek the actual Seeker + if res, err = seeker.Seek(offset, whence); err == nil { + //If the seek was successful, reinitialize ourselves (with the new position). + a.init(a.in, a.buffers, a.size) + } + 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 (a *reader) WriteTo(w io.Writer) (n int64, err error) { + if a.err != nil { + return 0, a.err + } + n = 0 + for { + err = a.fill() + if err != nil { + return n, err + } + n2, err := w.Write(a.cur.buffer()) + a.cur.inc(n2) + n += int64(n2) + if err != nil { + return n, err + } + if a.cur.err != nil { + // io.Writer should return nil if we are at EOF. + if a.cur.err == io.EOF { + a.err = a.cur.err + return n, nil + } + a.err = a.cur.err + return n, a.cur.err + } + } +} + +// Close will ensure that the underlying async reader is shut down. +// It will also close the input supplied on newAsyncReader. +func (a *reader) Close() (err error) { + select { + case <-a.exited: + case a.exit <- struct{}{}: + <-a.exited + } + if a.closer != nil { + // Only call once + c := a.closer + a.closer = nil + return c.Close() + } + a.err = errors.New("readahead: read after Close") + return nil +} + +// Internal buffer representing a single read. +// If an error is present, it must be returned +// once all buffer content has been served. +type buffer struct { + buf []byte + err error + offset int + size int +} + +func newBuffer(size int) *buffer { + return &buffer{buf: make([]byte, size), err: nil, size: size} +} + +// isEmpty returns true is offset is at end of +// buffer, or if the buffer is nil +func (b *buffer) isEmpty() bool { + if b == nil { + return true + } + if len(b.buf)-b.offset <= 0 { + return true + } + return false +} + +// read into start of the buffer from the supplied reader, +// resets the offset and updates the size of the buffer. +// Any error encountered during the read is returned. +func (b *buffer) read(rd io.Reader) (err error) { + defer func() { + if r := recover(); r != nil { + err = fmt.Errorf("panic reading: %v", r) + b.err = err + } + }() + var n int + n, b.err = rd.Read(b.buf[0:b.size]) + b.buf = b.buf[0:n] + b.offset = 0 + return b.err +} + +// Return the buffer at current offset +func (b *buffer) buffer() []byte { + return b.buf[b.offset:] +} + +// inc will increment the read offset +func (b *buffer) inc(n int) { + b.offset += n +} diff --git a/github.com/klauspost/compress/s2/cmd/s2c/main.go b/github.com/klauspost/compress/s2/cmd/s2c/main.go new file mode 100644 index 0000000..a630de6 --- /dev/null +++ b/github.com/klauspost/compress/s2/cmd/s2c/main.go @@ -0,0 +1,253 @@ +package main + +import ( + "bufio" + "errors" + "flag" + "fmt" + "io" + "io/ioutil" + "log" + "os" + "os/signal" + "path/filepath" + "runtime" + "runtime/pprof" + "runtime/trace" + "strconv" + "strings" + "sync" + "time" + "unicode" + + "github.com/klauspost/compress/s2" + "github.com/klauspost/compress/s2/cmd/internal/readahead" +) + +var ( + faster = flag.Bool("faster", false, "Compress faster, but with a minor compression loss") + cpu = flag.Int("cpu", runtime.GOMAXPROCS(0), "Compress using this amount of threads") + blockSize = flag.String("blocksize", "4M", "Max block size. Examples: 64K, 256K, 1M, 4M. Must be power of two and <= 4MB") + safe = flag.Bool("safe", false, "Do not overwrite output files") + padding = flag.String("pad", "1", "Pad size to a multiple of this value, Examples: 500, 64K, 256K, 1M, 4M, etc") + stdout = flag.Bool("c", false, "Write all output to stdout. Multiple input files will be concatenated") + remove = flag.Bool("rm", false, "Delete source file(s) after successful compression") + quiet = flag.Bool("q", false, "Don't write any output to terminal, except errors") + bench = flag.Int("bench", 0, "Run benchmark n times. No output will be written") + help = flag.Bool("help", false, "Display help") + + cpuprofile, memprofile, traceprofile string + + version = "(dev)" + date = "(unknown)" +) + +func main() { + if false { + flag.StringVar(&cpuprofile, "cpuprofile", "", "write cpu profile to file") + flag.StringVar(&memprofile, "memprofile", "", "write mem profile to file") + flag.StringVar(&traceprofile, "traceprofile", "", "write trace profile to file") + } + flag.Parse() + sz, err := toSize(*blockSize) + exitErr(err) + pad, err := toSize(*padding) + exitErr(err) + + args := flag.Args() + if len(args) == 0 || *help { + _, _ = fmt.Fprintf(os.Stderr, "s2 compress v%v, built at %v.\n\n", version, date) + _, _ = fmt.Fprintf(os.Stderr, "Copyright (c) 2011 The Snappy-Go Authors. All rights reserved.\n"+ + "Copyright (c) 2019 Klaus Post. All rights reserved.\n\n") + _, _ = fmt.Fprintln(os.Stderr, `Usage: s2c [options] file1 file2 + +Compresses all files supplied as input separately. +Output files are written as 'filename.ext.s2'. +By default output files will be overwritten. +Use - as the only file name to read from stdin and write to stdout. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +Options:`) + flag.PrintDefaults() + } + opts := []s2.WriterOption{s2.WriterBlockSize(int(sz)), s2.WriterConcurrency(*cpu), s2.WriterPadding(int(pad))} + if !*faster { + opts = append(opts, s2.WriterBetterCompression()) + } + wr := s2.NewWriter(nil, opts...) + + // No args, use stdin/stdout + if len(args) == 1 && args[0] == "-" { + // Catch interrupt, so we don't exit at once. + // os.Stdin will return EOF, so we should be able to get everything. + signal.Notify(make(chan os.Signal), os.Interrupt) + wr.Reset(os.Stdout) + _, err = wr.ReadFrom(os.Stdin) + printErr(err) + printErr(wr.Close()) + return + } + var files []string + + for _, pattern := range args { + found, err := filepath.Glob(pattern) + exitErr(err) + if len(found) == 0 { + exitErr(fmt.Errorf("unable to find file %v", pattern)) + } + files = append(files, found...) + } + if cpuprofile != "" { + f, err := os.Create(cpuprofile) + if err != nil { + log.Fatal(err) + } + pprof.StartCPUProfile(f) + defer pprof.StopCPUProfile() + } + if memprofile != "" { + f, err := os.Create(memprofile) + if err != nil { + log.Fatal(err) + } + defer f.Close() + defer pprof.WriteHeapProfile(f) + } + if traceprofile != "" { + f, err := os.Create(traceprofile) + if err != nil { + log.Fatal(err) + } + defer f.Close() + err = trace.Start(f) + if err != nil { + log.Fatal(err) + } + defer trace.Stop() + } + + *quiet = *quiet || *stdout + allFiles := files + for i := 0; i < *bench; i++ { + files = append(files, allFiles...) + } + for _, filename := range files { + func() { + var closeOnce sync.Once + dstFilename := fmt.Sprintf("%s%s", filename, ".s2") + if *bench > 0 { + dstFilename = "(discarded)" + } + if !*quiet { + fmt.Print("Compressing ", filename, " -> ", dstFilename) + } + // Input file. + file, err := os.Open(filename) + exitErr(err) + defer closeOnce.Do(func() { file.Close() }) + src, err := readahead.NewReaderSize(file, *cpu+1, 1<<20) + exitErr(err) + defer src.Close() + finfo, err := file.Stat() + exitErr(err) + var out io.Writer + switch { + case *bench > 0: + out = ioutil.Discard + case *stdout: + out = os.Stdout + default: + mode := finfo.Mode() // use the same mode for the output file + if *safe { + _, err := os.Stat(dstFilename) + if !os.IsNotExist(err) { + exitErr(errors.New("destination file exists")) + } + } + dstFile, err := os.OpenFile(dstFilename, os.O_CREATE|os.O_WRONLY, mode) + exitErr(err) + defer dstFile.Close() + bw := bufio.NewWriterSize(dstFile, int(sz)*2) + defer bw.Flush() + out = bw + } + wc := wCounter{out: out} + wr.Reset(&wc) + defer wr.Close() + start := time.Now() + input, err := wr.ReadFrom(src) + exitErr(err) + err = wr.Close() + exitErr(err) + if !*quiet { + elapsed := time.Since(start) + mbpersec := (float64(input) / (1024 * 1024)) / (float64(elapsed) / (float64(time.Second))) + pct := float64(wc.n) * 100 / float64(input) + fmt.Printf(" %d -> %d [%.02f%%]; %.01fMB/s\n", input, wc.n, pct, mbpersec) + } + if *remove { + closeOnce.Do(func() { + file.Close() + if !*quiet { + fmt.Println("Removing", filename) + } + err := os.Remove(filename) + exitErr(err) + }) + } + }() + } +} + +func printErr(err error) { + if err != nil { + fmt.Fprintln(os.Stderr, "\nERROR:", err.Error()) + } +} + +func exitErr(err error) { + if err != nil { + fmt.Fprintln(os.Stderr, "\nERROR:", err.Error()) + os.Exit(2) + } +} + +// toSize converts a size indication to bytes. +func toSize(size string) (uint64, error) { + size = strings.ToUpper(strings.TrimSpace(size)) + firstLetter := strings.IndexFunc(size, unicode.IsLetter) + if firstLetter == -1 { + firstLetter = len(size) + } + + bytesString, multiple := size[:firstLetter], size[firstLetter:] + bytes, err := strconv.ParseUint(bytesString, 10, 64) + if err != nil { + return 0, fmt.Errorf("unable to parse size: %v", err) + } + + switch multiple { + case "M", "MB", "MIB": + return bytes * 1 << 20, nil + case "K", "KB", "KIB": + return bytes * 1 << 10, nil + case "B", "": + return bytes, nil + default: + return 0, fmt.Errorf("unknown size suffix: %v", multiple) + } +} + +type wCounter struct { + n int + out io.Writer +} + +func (w *wCounter) Write(p []byte) (n int, err error) { + n, err = w.out.Write(p) + w.n += n + return n, err + +} diff --git a/github.com/klauspost/compress/s2/cmd/s2d/main.go b/github.com/klauspost/compress/s2/cmd/s2d/main.go new file mode 100644 index 0000000..2cc1437 --- /dev/null +++ b/github.com/klauspost/compress/s2/cmd/s2d/main.go @@ -0,0 +1,178 @@ +package main + +import ( + "bufio" + "errors" + "flag" + "fmt" + "io" + "io/ioutil" + "os" + "path/filepath" + "strings" + "sync" + "time" + + "github.com/klauspost/compress/s2" + "github.com/klauspost/compress/s2/cmd/internal/readahead" +) + +var ( + safe = flag.Bool("safe", false, "Do not overwrite output files") + verify = flag.Bool("verify", false, "Verify files, but do not write output") + stdout = flag.Bool("c", false, "Write all output to stdout. Multiple input files will be concatenated") + remove = flag.Bool("rm", false, "Delete source file(s) after successful decompression") + quiet = flag.Bool("q", false, "Don't write any output to terminal, except errors") + bench = flag.Int("bench", 0, "Run benchmark n times. No output will be written") + help = flag.Bool("help", false, "Display help") + + version = "(dev)" + date = "(unknown)" +) + +func main() { + flag.Parse() + r := s2.NewReader(nil) + + // No args, use stdin/stdout + args := flag.Args() + if len(args) == 0 || *help { + _, _ = fmt.Fprintf(os.Stderr, "s2 decompress v%v, built at %v.\n\n", version, date) + _, _ = fmt.Fprintf(os.Stderr, "Copyright (c) 2011 The Snappy-Go Authors. All rights reserved.\n"+ + "Copyright (c) 2019 Klaus Post. All rights reserved.\n\n") + _, _ = fmt.Fprintln(os.Stderr, `Usage: s2d [options] file1 file2 + +Decompresses all files supplied as input. Input files must end with '.s2' or '.snappy'. +Output file names have the extension removed. By default output files will be overwritten. +Use - as the only file name to read from stdin and write to stdout. + +Wildcards are accepted: testdir/*.txt will compress all files in testdir ending with .txt +Directories can be wildcards as well. testdir/*/*.txt will match testdir/subdir/b.txt + +Options:`) + flag.PrintDefaults() + } + if len(args) == 1 && args[0] == "-" { + r.Reset(os.Stdin) + if !*verify { + _, err := io.Copy(os.Stdout, r) + exitErr(err) + } else { + _, err := io.Copy(ioutil.Discard, r) + exitErr(err) + } + return + } + var files []string + + for _, pattern := range args { + found, err := filepath.Glob(pattern) + exitErr(err) + if len(found) == 0 { + exitErr(fmt.Errorf("unable to find file %v", pattern)) + } + files = append(files, found...) + } + + *quiet = *quiet || *stdout + allFiles := files + for i := 0; i < *bench; i++ { + files = append(files, allFiles...) + } + + for _, filename := range files { + dstFilename := filename + switch { + case strings.HasSuffix(filename, ".s2"): + dstFilename = strings.TrimSuffix(filename, ".s2") + case strings.HasSuffix(filename, ".snappy"): + dstFilename = strings.TrimSuffix(filename, ".snappy") + default: + fmt.Println("Skipping", filename) + continue + } + if *bench > 0 { + dstFilename = "(discarded)" + } + if *verify { + dstFilename = "(verify)" + } + + func() { + var closeOnce sync.Once + if !*quiet { + fmt.Print("Decompressing ", filename, " -> ", dstFilename) + } + // Input file. + file, err := os.Open(filename) + exitErr(err) + defer closeOnce.Do(func() { file.Close() }) + rc := rCounter{in: file} + src, err := readahead.NewReaderSize(&rc, 2, 4<<20) + exitErr(err) + defer src.Close() + finfo, err := file.Stat() + exitErr(err) + mode := finfo.Mode() // use the same mode for the output file + if *safe { + _, err := os.Stat(dstFilename) + if !os.IsNotExist(err) { + exitErr(errors.New("destination files exists")) + } + } + var out io.Writer + switch { + case *bench > 0 || *verify: + out = ioutil.Discard + case *stdout: + out = os.Stdout + default: + dstFile, err := os.OpenFile(dstFilename, os.O_CREATE|os.O_WRONLY, mode) + exitErr(err) + defer dstFile.Close() + bw := bufio.NewWriterSize(dstFile, 4<<20) + defer bw.Flush() + out = bw + } + r.Reset(src) + start := time.Now() + output, err := io.Copy(out, r) + exitErr(err) + if !*quiet { + elapsed := time.Since(start) + mbPerSec := (float64(output) / (1024 * 1024)) / (float64(elapsed) / (float64(time.Second))) + pct := float64(output) * 100 / float64(rc.n) + fmt.Printf(" %d -> %d [%.02f%%]; %.01fMB/s\n", rc.n, output, pct, mbPerSec) + } + if *remove && !*verify { + closeOnce.Do(func() { + file.Close() + if !*quiet { + fmt.Println("Removing", filename) + } + err := os.Remove(filename) + exitErr(err) + }) + } + }() + } +} + +func exitErr(err error) { + if err != nil { + fmt.Fprintln(os.Stderr, "\nERROR:", err.Error()) + os.Exit(2) + } +} + +type rCounter struct { + n int + in io.Reader +} + +func (w *rCounter) Read(p []byte) (n int, err error) { + n, err = w.in.Read(p) + w.n += n + return n, err + +} diff --git a/github.com/klauspost/compress/s2/cpuid_amd64.go b/github.com/klauspost/compress/s2/cpuid_amd64.go new file mode 100644 index 0000000..114ccd2 --- /dev/null +++ b/github.com/klauspost/compress/s2/cpuid_amd64.go @@ -0,0 +1,1194 @@ +// Generated, DO NOT EDIT, +// but copy it to your own project and rename the package. +// See more at http://github.com/klauspost/cpuid + +// +build !appengine +// +build gc +// +build !noasm + +package s2 + +import "strings" + +func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32) +func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32) +func asmXgetbv(index uint32) (eax, edx uint32) +func asmRdtscpAsm() (eax, ebx, ecx, edx uint32) + +func initCPU() { + cpuid = asmCpuid + cpuidex = asmCpuidex + xgetbv = asmXgetbv + rdtscpAsm = asmRdtscpAsm +} + +// Vendor is a representation of a CPU vendor. +type vendor int + +const ( + other vendor = iota + intel + amd + via + transmeta + nsc + kvm // Kernel-based Virtual Machine + msvm // Microsoft Hyper-V or Windows Virtual PC + vmware + xenhvm + bhyve + hygon +) + +const ( + cmov = 1 << iota // i686 CMOV + nx // NX (No-Execute) bit + amd3dnow // AMD 3DNOW + amd3dnowext // AMD 3DNowExt + mmx // standard MMX + mmxext // SSE integer functions or AMD MMX ext + sse // SSE functions + sse2 // P4 SSE functions + sse3 // Prescott SSE3 functions + ssse3 // Conroe SSSE3 functions + sse4 // Penryn SSE4.1 functions + sse4a // AMD Barcelona microarchitecture SSE4a instructions + sse42 // Nehalem SSE4.2 functions + avx // AVX functions + avx2 // AVX2 functions + fma3 // Intel FMA 3 + fma4 // Bulldozer FMA4 functions + xop // Bulldozer XOP functions + f16c // Half-precision floating-point conversion + bmi1 // Bit Manipulation Instruction Set 1 + bmi2 // Bit Manipulation Instruction Set 2 + tbm // AMD Trailing Bit Manipulation + lzcnt // LZCNT instruction + popcnt // POPCNT instruction + aesni // Advanced Encryption Standard New Instructions + clmul // Carry-less Multiplication + htt // Hyperthreading (enabled) + hle // Hardware Lock Elision + rtm // Restricted Transactional Memory + rdrand // RDRAND instruction is available + rdseed // RDSEED instruction is available + adx // Intel ADX (Multi-Precision Add-Carry Instruction Extensions) + sha // Intel SHA Extensions + avx512f // AVX-512 Foundation + avx512dq // AVX-512 Doubleword and Quadword Instructions + avx512ifma // AVX-512 Integer Fused Multiply-Add Instructions + avx512pf // AVX-512 Prefetch Instructions + avx512er // AVX-512 Exponential and Reciprocal Instructions + avx512cd // AVX-512 Conflict Detection Instructions + avx512bw // AVX-512 Byte and Word Instructions + avx512vl // AVX-512 Vector Length Extensions + avx512vbmi // AVX-512 Vector Bit Manipulation Instructions + avx512vbmi2 // AVX-512 Vector Bit Manipulation Instructions, Version 2 + avx512vnni // AVX-512 Vector Neural Network Instructions + avx512vpopcntdq // AVX-512 Vector Population Count Doubleword and Quadword + gfni // Galois Field New Instructions + vaes // Vector AES + avx512bitalg // AVX-512 Bit Algorithms + vpclmulqdq // Carry-Less Multiplication Quadword + avx512bf16 // AVX-512 BFLOAT16 Instructions + avx512vp2intersect // AVX-512 Intersect for D/Q + mpx // Intel MPX (Memory Protection Extensions) + erms // Enhanced REP MOVSB/STOSB + rdtscp // RDTSCP Instruction + cx16 // CMPXCHG16B Instruction + sgx // Software Guard Extensions + sgxlc // Software Guard Extensions Launch Control + ibpb // Indirect Branch Restricted Speculation (IBRS) and Indirect Branch Predictor Barrier (IBPB) + stibp // Single Thread Indirect Branch Predictors + vmx // Virtual Machine Extensions + + // Performance indicators + sse2slow // SSE2 is supported, but usually not faster + sse3slow // SSE3 is supported, but usually not faster + atom // Atom processor, some SSSE3 instructions are slower +) + +var flagNames = map[flags]string{ + cmov: "CMOV", // i686 CMOV + nx: "NX", // NX (No-Execute) bit + amd3dnow: "AMD3DNOW", // AMD 3DNOW + amd3dnowext: "AMD3DNOWEXT", // AMD 3DNowExt + mmx: "MMX", // Standard MMX + mmxext: "MMXEXT", // SSE integer functions or AMD MMX ext + sse: "SSE", // SSE functions + sse2: "SSE2", // P4 SSE2 functions + sse3: "SSE3", // Prescott SSE3 functions + ssse3: "SSSE3", // Conroe SSSE3 functions + sse4: "SSE4.1", // Penryn SSE4.1 functions + sse4a: "SSE4A", // AMD Barcelona microarchitecture SSE4a instructions + sse42: "SSE4.2", // Nehalem SSE4.2 functions + avx: "AVX", // AVX functions + avx2: "AVX2", // AVX functions + fma3: "FMA3", // Intel FMA 3 + fma4: "FMA4", // Bulldozer FMA4 functions + xop: "XOP", // Bulldozer XOP functions + f16c: "F16C", // Half-precision floating-point conversion + bmi1: "BMI1", // Bit Manipulation Instruction Set 1 + bmi2: "BMI2", // Bit Manipulation Instruction Set 2 + tbm: "TBM", // AMD Trailing Bit Manipulation + lzcnt: "LZCNT", // LZCNT instruction + popcnt: "POPCNT", // POPCNT instruction + aesni: "AESNI", // Advanced Encryption Standard New Instructions + clmul: "CLMUL", // Carry-less Multiplication + htt: "HTT", // Hyperthreading (enabled) + hle: "HLE", // Hardware Lock Elision + rtm: "RTM", // Restricted Transactional Memory + rdrand: "RDRAND", // RDRAND instruction is available + rdseed: "RDSEED", // RDSEED instruction is available + adx: "ADX", // Intel ADX (Multi-Precision Add-Carry Instruction Extensions) + sha: "SHA", // Intel SHA Extensions + avx512f: "AVX512F", // AVX-512 Foundation + avx512dq: "AVX512DQ", // AVX-512 Doubleword and Quadword Instructions + avx512ifma: "AVX512IFMA", // AVX-512 Integer Fused Multiply-Add Instructions + avx512pf: "AVX512PF", // AVX-512 Prefetch Instructions + avx512er: "AVX512ER", // AVX-512 Exponential and Reciprocal Instructions + avx512cd: "AVX512CD", // AVX-512 Conflict Detection Instructions + avx512bw: "AVX512BW", // AVX-512 Byte and Word Instructions + avx512vl: "AVX512VL", // AVX-512 Vector Length Extensions + avx512vbmi: "AVX512VBMI", // AVX-512 Vector Bit Manipulation Instructions + avx512vbmi2: "AVX512VBMI2", // AVX-512 Vector Bit Manipulation Instructions, Version 2 + avx512vnni: "AVX512VNNI", // AVX-512 Vector Neural Network Instructions + avx512vpopcntdq: "AVX512VPOPCNTDQ", // AVX-512 Vector Population Count Doubleword and Quadword + gfni: "GFNI", // Galois Field New Instructions + vaes: "VAES", // Vector AES + avx512bitalg: "AVX512BITALG", // AVX-512 Bit Algorithms + vpclmulqdq: "VPCLMULQDQ", // Carry-Less Multiplication Quadword + avx512bf16: "AVX512BF16", // AVX-512 BFLOAT16 Instruction + avx512vp2intersect: "AVX512VP2INTERSECT", // AVX-512 Intersect for D/Q + mpx: "MPX", // Intel MPX (Memory Protection Extensions) + erms: "ERMS", // Enhanced REP MOVSB/STOSB + rdtscp: "RDTSCP", // RDTSCP Instruction + cx16: "CX16", // CMPXCHG16B Instruction + sgx: "SGX", // Software Guard Extensions + sgxlc: "SGXLC", // Software Guard Extensions Launch Control + ibpb: "IBPB", // Indirect Branch Restricted Speculation and Indirect Branch Predictor Barrier + stibp: "STIBP", // Single Thread Indirect Branch Predictors + vmx: "VMX", // Virtual Machine Extensions + + // Performance indicators + sse2slow: "SSE2SLOW", // SSE2 supported, but usually not faster + sse3slow: "SSE3SLOW", // SSE3 supported, but usually not faster + atom: "ATOM", // Atom processor, some SSSE3 instructions are slower + +} + +// CPUInfo contains information about the detected system CPU. +type cpuInfo struct { + brandname string // Brand name reported by the CPU + vendorid vendor // Comparable CPU vendor ID + features flags // Features of the CPU + physicalcores int // Number of physical processor cores in your CPU. Will be 0 if undetectable. + threadspercore int // Number of threads per physical core. Will be 1 if undetectable. + logicalcores int // Number of physical cores times threads that can run on each core through the use of hyperthreading. Will be 0 if undetectable. + family int // CPU family number + model int // CPU model number + cacheline int // Cache line size in bytes. Will be 0 if undetectable. + cache struct { + l1i int // L1 Instruction Cache (per core or shared). Will be -1 if undetected + l1d int // L1 Data Cache (per core or shared). Will be -1 if undetected + l2 int // L2 Cache (per core or shared). Will be -1 if undetected + l3 int // L3 Instruction Cache (per core or shared). Will be -1 if undetected + } + sgx sgxsupport + maxFunc uint32 + maxExFunc uint32 +} + +var cpuid func(op uint32) (eax, ebx, ecx, edx uint32) +var cpuidex func(op, op2 uint32) (eax, ebx, ecx, edx uint32) +var xgetbv func(index uint32) (eax, edx uint32) +var rdtscpAsm func() (eax, ebx, ecx, edx uint32) + +// CPU contains information about the CPU as detected on startup, +// or when Detect last was called. +// +// Use this as the primary entry point to you data, +// this way queries are +var cpu cpuInfo + +func init() { + initCPU() + detect() +} + +// Detect will re-detect current CPU info. +// This will replace the content of the exported CPU variable. +// +// Unless you expect the CPU to change while you are running your program +// you should not need to call this function. +// If you call this, you must ensure that no other goroutine is accessing the +// exported CPU variable. +func detect() { + cpu.maxFunc = maxFunctionID() + cpu.maxExFunc = maxExtendedFunction() + cpu.brandname = brandName() + cpu.cacheline = cacheLine() + cpu.family, cpu.model = familyModel() + cpu.features = support() + cpu.sgx = hasSGX(cpu.features&sgx != 0, cpu.features&sgxlc != 0) + cpu.threadspercore = threadsPerCore() + cpu.logicalcores = logicalCores() + cpu.physicalcores = physicalCores() + cpu.vendorid = vendorID() + cpu.cacheSize() +} + +// Generated here: http://play.golang.org/p/BxFH2Gdc0G + +// Cmov indicates support of CMOV instructions +func (c cpuInfo) cmov() bool { + return c.features&cmov != 0 +} + +// Amd3dnow indicates support of AMD 3DNOW! instructions +func (c cpuInfo) amd3dnow() bool { + return c.features&amd3dnow != 0 +} + +// Amd3dnowExt indicates support of AMD 3DNOW! Extended instructions +func (c cpuInfo) amd3dnowext() bool { + return c.features&amd3dnowext != 0 +} + +// VMX indicates support of VMX +func (c cpuInfo) vmx() bool { + return c.features&vmx != 0 +} + +// MMX indicates support of MMX instructions +func (c cpuInfo) mmx() bool { + return c.features&mmx != 0 +} + +// MMXExt indicates support of MMXEXT instructions +// (SSE integer functions or AMD MMX ext) +func (c cpuInfo) mmxext() bool { + return c.features&mmxext != 0 +} + +// SSE indicates support of SSE instructions +func (c cpuInfo) sse() bool { + return c.features&sse != 0 +} + +// SSE2 indicates support of SSE 2 instructions +func (c cpuInfo) sse2() bool { + return c.features&sse2 != 0 +} + +// SSE3 indicates support of SSE 3 instructions +func (c cpuInfo) sse3() bool { + return c.features&sse3 != 0 +} + +// SSSE3 indicates support of SSSE 3 instructions +func (c cpuInfo) ssse3() bool { + return c.features&ssse3 != 0 +} + +// SSE4 indicates support of SSE 4 (also called SSE 4.1) instructions +func (c cpuInfo) sse4() bool { + return c.features&sse4 != 0 +} + +// SSE42 indicates support of SSE4.2 instructions +func (c cpuInfo) sse42() bool { + return c.features&sse42 != 0 +} + +// AVX indicates support of AVX instructions +// and operating system support of AVX instructions +func (c cpuInfo) avx() bool { + return c.features&avx != 0 +} + +// AVX2 indicates support of AVX2 instructions +func (c cpuInfo) avx2() bool { + return c.features&avx2 != 0 +} + +// FMA3 indicates support of FMA3 instructions +func (c cpuInfo) fma3() bool { + return c.features&fma3 != 0 +} + +// FMA4 indicates support of FMA4 instructions +func (c cpuInfo) fma4() bool { + return c.features&fma4 != 0 +} + +// XOP indicates support of XOP instructions +func (c cpuInfo) xop() bool { + return c.features&xop != 0 +} + +// F16C indicates support of F16C instructions +func (c cpuInfo) f16c() bool { + return c.features&f16c != 0 +} + +// BMI1 indicates support of BMI1 instructions +func (c cpuInfo) bmi1() bool { + return c.features&bmi1 != 0 +} + +// BMI2 indicates support of BMI2 instructions +func (c cpuInfo) bmi2() bool { + return c.features&bmi2 != 0 +} + +// TBM indicates support of TBM instructions +// (AMD Trailing Bit Manipulation) +func (c cpuInfo) tbm() bool { + return c.features&tbm != 0 +} + +// Lzcnt indicates support of LZCNT instruction +func (c cpuInfo) lzcnt() bool { + return c.features&lzcnt != 0 +} + +// Popcnt indicates support of POPCNT instruction +func (c cpuInfo) popcnt() bool { + return c.features&popcnt != 0 +} + +// HTT indicates the processor has Hyperthreading enabled +func (c cpuInfo) htt() bool { + return c.features&htt != 0 +} + +// SSE2Slow indicates that SSE2 may be slow on this processor +func (c cpuInfo) sse2slow() bool { + return c.features&sse2slow != 0 +} + +// SSE3Slow indicates that SSE3 may be slow on this processor +func (c cpuInfo) sse3slow() bool { + return c.features&sse3slow != 0 +} + +// AesNi indicates support of AES-NI instructions +// (Advanced Encryption Standard New Instructions) +func (c cpuInfo) aesni() bool { + return c.features&aesni != 0 +} + +// Clmul indicates support of CLMUL instructions +// (Carry-less Multiplication) +func (c cpuInfo) clmul() bool { + return c.features&clmul != 0 +} + +// NX indicates support of NX (No-Execute) bit +func (c cpuInfo) nx() bool { + return c.features&nx != 0 +} + +// SSE4A indicates support of AMD Barcelona microarchitecture SSE4a instructions +func (c cpuInfo) sse4a() bool { + return c.features&sse4a != 0 +} + +// HLE indicates support of Hardware Lock Elision +func (c cpuInfo) hle() bool { + return c.features&hle != 0 +} + +// RTM indicates support of Restricted Transactional Memory +func (c cpuInfo) rtm() bool { + return c.features&rtm != 0 +} + +// Rdrand indicates support of RDRAND instruction is available +func (c cpuInfo) rdrand() bool { + return c.features&rdrand != 0 +} + +// Rdseed indicates support of RDSEED instruction is available +func (c cpuInfo) rdseed() bool { + return c.features&rdseed != 0 +} + +// ADX indicates support of Intel ADX (Multi-Precision Add-Carry Instruction Extensions) +func (c cpuInfo) adx() bool { + return c.features&adx != 0 +} + +// SHA indicates support of Intel SHA Extensions +func (c cpuInfo) sha() bool { + return c.features&sha != 0 +} + +// AVX512F indicates support of AVX-512 Foundation +func (c cpuInfo) avx512f() bool { + return c.features&avx512f != 0 +} + +// AVX512DQ indicates support of AVX-512 Doubleword and Quadword Instructions +func (c cpuInfo) avx512dq() bool { + return c.features&avx512dq != 0 +} + +// AVX512IFMA indicates support of AVX-512 Integer Fused Multiply-Add Instructions +func (c cpuInfo) avx512ifma() bool { + return c.features&avx512ifma != 0 +} + +// AVX512PF indicates support of AVX-512 Prefetch Instructions +func (c cpuInfo) avx512pf() bool { + return c.features&avx512pf != 0 +} + +// AVX512ER indicates support of AVX-512 Exponential and Reciprocal Instructions +func (c cpuInfo) avx512er() bool { + return c.features&avx512er != 0 +} + +// AVX512CD indicates support of AVX-512 Conflict Detection Instructions +func (c cpuInfo) avx512cd() bool { + return c.features&avx512cd != 0 +} + +// AVX512BW indicates support of AVX-512 Byte and Word Instructions +func (c cpuInfo) avx512bw() bool { + return c.features&avx512bw != 0 +} + +// AVX512VL indicates support of AVX-512 Vector Length Extensions +func (c cpuInfo) avx512vl() bool { + return c.features&avx512vl != 0 +} + +// AVX512VBMI indicates support of AVX-512 Vector Bit Manipulation Instructions +func (c cpuInfo) avx512vbmi() bool { + return c.features&avx512vbmi != 0 +} + +// AVX512VBMI2 indicates support of AVX-512 Vector Bit Manipulation Instructions, Version 2 +func (c cpuInfo) avx512vbmi2() bool { + return c.features&avx512vbmi2 != 0 +} + +// AVX512VNNI indicates support of AVX-512 Vector Neural Network Instructions +func (c cpuInfo) avx512vnni() bool { + return c.features&avx512vnni != 0 +} + +// AVX512VPOPCNTDQ indicates support of AVX-512 Vector Population Count Doubleword and Quadword +func (c cpuInfo) avx512vpopcntdq() bool { + return c.features&avx512vpopcntdq != 0 +} + +// GFNI indicates support of Galois Field New Instructions +func (c cpuInfo) gfni() bool { + return c.features&gfni != 0 +} + +// VAES indicates support of Vector AES +func (c cpuInfo) vaes() bool { + return c.features&vaes != 0 +} + +// AVX512BITALG indicates support of AVX-512 Bit Algorithms +func (c cpuInfo) avx512bitalg() bool { + return c.features&avx512bitalg != 0 +} + +// VPCLMULQDQ indicates support of Carry-Less Multiplication Quadword +func (c cpuInfo) vpclmulqdq() bool { + return c.features&vpclmulqdq != 0 +} + +// AVX512BF16 indicates support of +func (c cpuInfo) avx512bf16() bool { + return c.features&avx512bf16 != 0 +} + +// AVX512VP2INTERSECT indicates support of +func (c cpuInfo) avx512vp2intersect() bool { + return c.features&avx512vp2intersect != 0 +} + +// MPX indicates support of Intel MPX (Memory Protection Extensions) +func (c cpuInfo) mpx() bool { + return c.features&mpx != 0 +} + +// ERMS indicates support of Enhanced REP MOVSB/STOSB +func (c cpuInfo) erms() bool { + return c.features&erms != 0 +} + +// RDTSCP Instruction is available. +func (c cpuInfo) rdtscp() bool { + return c.features&rdtscp != 0 +} + +// CX16 indicates if CMPXCHG16B instruction is available. +func (c cpuInfo) cx16() bool { + return c.features&cx16 != 0 +} + +// TSX is split into HLE (Hardware Lock Elision) and RTM (Restricted Transactional Memory) detection. +// So TSX simply checks that. +func (c cpuInfo) tsx() bool { + return c.features&(hle|rtm) == hle|rtm +} + +// Atom indicates an Atom processor +func (c cpuInfo) atom() bool { + return c.features&atom != 0 +} + +// Intel returns true if vendor is recognized as Intel +func (c cpuInfo) intel() bool { + return c.vendorid == intel +} + +// AMD returns true if vendor is recognized as AMD +func (c cpuInfo) amd() bool { + return c.vendorid == amd +} + +// Hygon returns true if vendor is recognized as Hygon +func (c cpuInfo) hygon() bool { + return c.vendorid == hygon +} + +// Transmeta returns true if vendor is recognized as Transmeta +func (c cpuInfo) transmeta() bool { + return c.vendorid == transmeta +} + +// NSC returns true if vendor is recognized as National Semiconductor +func (c cpuInfo) nsc() bool { + return c.vendorid == nsc +} + +// VIA returns true if vendor is recognized as VIA +func (c cpuInfo) via() bool { + return c.vendorid == via +} + +// RTCounter returns the 64-bit time-stamp counter +// Uses the RDTSCP instruction. The value 0 is returned +// if the CPU does not support the instruction. +func (c cpuInfo) rtcounter() uint64 { + if !c.rdtscp() { + return 0 + } + a, _, _, d := rdtscpAsm() + return uint64(a) | (uint64(d) << 32) +} + +// Ia32TscAux returns the IA32_TSC_AUX part of the RDTSCP. +// This variable is OS dependent, but on Linux contains information +// about the current cpu/core the code is running on. +// If the RDTSCP instruction isn't supported on the CPU, the value 0 is returned. +func (c cpuInfo) ia32tscaux() uint32 { + if !c.rdtscp() { + return 0 + } + _, _, ecx, _ := rdtscpAsm() + return ecx +} + +// LogicalCPU will return the Logical CPU the code is currently executing on. +// This is likely to change when the OS re-schedules the running thread +// to another CPU. +// If the current core cannot be detected, -1 will be returned. +func (c cpuInfo) logicalcpu() int { + if c.maxFunc < 1 { + return -1 + } + _, ebx, _, _ := cpuid(1) + return int(ebx >> 24) +} + +// VM Will return true if the cpu id indicates we are in +// a virtual machine. This is only a hint, and will very likely +// have many false negatives. +func (c cpuInfo) vm() bool { + switch c.vendorid { + case msvm, kvm, vmware, xenhvm, bhyve: + return true + } + return false +} + +// Flags contains detected cpu features and caracteristics +type flags uint64 + +// String returns a string representation of the detected +// CPU features. +func (f flags) String() string { + return strings.Join(f.strings(), ",") +} + +// Strings returns and array of the detected features. +func (f flags) strings() []string { + s := support() + r := make([]string, 0, 20) + for i := uint(0); i < 64; i++ { + key := flags(1 << i) + val := flagNames[key] + if s&key != 0 { + r = append(r, val) + } + } + return r +} + +func maxExtendedFunction() uint32 { + eax, _, _, _ := cpuid(0x80000000) + return eax +} + +func maxFunctionID() uint32 { + a, _, _, _ := cpuid(0) + return a +} + +func brandName() string { + if maxExtendedFunction() >= 0x80000004 { + v := make([]uint32, 0, 48) + for i := uint32(0); i < 3; i++ { + a, b, c, d := cpuid(0x80000002 + i) + v = append(v, a, b, c, d) + } + return strings.Trim(string(valAsString(v...)), " ") + } + return "unknown" +} + +func threadsPerCore() int { + mfi := maxFunctionID() + if mfi < 0x4 || vendorID() != intel { + return 1 + } + + if mfi < 0xb { + _, b, _, d := cpuid(1) + if (d & (1 << 28)) != 0 { + // v will contain logical core count + v := (b >> 16) & 255 + if v > 1 { + a4, _, _, _ := cpuid(4) + // physical cores + v2 := (a4 >> 26) + 1 + if v2 > 0 { + return int(v) / int(v2) + } + } + } + return 1 + } + _, b, _, _ := cpuidex(0xb, 0) + if b&0xffff == 0 { + return 1 + } + return int(b & 0xffff) +} + +func logicalCores() int { + mfi := maxFunctionID() + switch vendorID() { + case intel: + // Use this on old Intel processors + if mfi < 0xb { + if mfi < 1 { + return 0 + } + // CPUID.1:EBX[23:16] represents the maximum number of addressable IDs (initial APIC ID) + // that can be assigned to logical processors in a physical package. + // The value may not be the same as the number of logical processors that are present in the hardware of a physical package. + _, ebx, _, _ := cpuid(1) + logical := (ebx >> 16) & 0xff + return int(logical) + } + _, b, _, _ := cpuidex(0xb, 1) + return int(b & 0xffff) + case amd, hygon: + _, b, _, _ := cpuid(1) + return int((b >> 16) & 0xff) + default: + return 0 + } +} + +func familyModel() (int, int) { + if maxFunctionID() < 0x1 { + return 0, 0 + } + eax, _, _, _ := cpuid(1) + family := ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff) + model := ((eax >> 4) & 0xf) + ((eax >> 12) & 0xf0) + return int(family), int(model) +} + +func physicalCores() int { + switch vendorID() { + case intel: + return logicalCores() / threadsPerCore() + case amd, hygon: + if maxExtendedFunction() >= 0x80000008 { + _, _, c, _ := cpuid(0x80000008) + return int(c&0xff) + 1 + } + } + return 0 +} + +// Except from http://en.wikipedia.org/wiki/CPUID#EAX.3D0:_Get_vendor_ID +var vendorMapping = map[string]vendor{ + "AMDisbetter!": amd, + "AuthenticAMD": amd, + "CentaurHauls": via, + "GenuineIntel": intel, + "TransmetaCPU": transmeta, + "GenuineTMx86": transmeta, + "Geode by NSC": nsc, + "VIA VIA VIA ": via, + "KVMKVMKVMKVM": kvm, + "Microsoft Hv": msvm, + "VMwareVMware": vmware, + "XenVMMXenVMM": xenhvm, + "bhyve bhyve ": bhyve, + "HygonGenuine": hygon, +} + +func vendorID() vendor { + _, b, c, d := cpuid(0) + v := valAsString(b, d, c) + vend, ok := vendorMapping[string(v)] + if !ok { + return other + } + return vend +} + +func cacheLine() int { + if maxFunctionID() < 0x1 { + return 0 + } + + _, ebx, _, _ := cpuid(1) + cache := (ebx & 0xff00) >> 5 // cflush size + if cache == 0 && maxExtendedFunction() >= 0x80000006 { + _, _, ecx, _ := cpuid(0x80000006) + cache = ecx & 0xff // cacheline size + } + // TODO: Read from Cache and TLB Information + return int(cache) +} + +func (c *cpuInfo) cacheSize() { + c.cache.l1d = -1 + c.cache.l1i = -1 + c.cache.l2 = -1 + c.cache.l3 = -1 + vendor := vendorID() + switch vendor { + case intel: + if maxFunctionID() < 4 { + return + } + for i := uint32(0); ; i++ { + eax, ebx, ecx, _ := cpuidex(4, i) + cacheType := eax & 15 + if cacheType == 0 { + break + } + cacheLevel := (eax >> 5) & 7 + coherency := int(ebx&0xfff) + 1 + partitions := int((ebx>>12)&0x3ff) + 1 + associativity := int((ebx>>22)&0x3ff) + 1 + sets := int(ecx) + 1 + size := associativity * partitions * coherency * sets + switch cacheLevel { + case 1: + if cacheType == 1 { + // 1 = Data Cache + c.cache.l1d = size + } else if cacheType == 2 { + // 2 = Instruction Cache + c.cache.l1i = size + } else { + if c.cache.l1d < 0 { + c.cache.l1i = size + } + if c.cache.l1i < 0 { + c.cache.l1i = size + } + } + case 2: + c.cache.l2 = size + case 3: + c.cache.l3 = size + } + } + case amd, hygon: + // Untested. + if maxExtendedFunction() < 0x80000005 { + return + } + _, _, ecx, edx := cpuid(0x80000005) + c.cache.l1d = int(((ecx >> 24) & 0xFF) * 1024) + c.cache.l1i = int(((edx >> 24) & 0xFF) * 1024) + + if maxExtendedFunction() < 0x80000006 { + return + } + _, _, ecx, _ = cpuid(0x80000006) + c.cache.l2 = int(((ecx >> 16) & 0xFFFF) * 1024) + } + + return +} + +type sgxepcsection struct { + baseaddress uint64 + epcsize uint64 +} + +type sgxsupport struct { + available bool + launchcontrol bool + sgx1supported bool + sgx2supported bool + maxenclavesizenot64 int64 + maxenclavesize64 int64 + epcsections []sgxepcsection +} + +func hasSGX(available, lc bool) (rval sgxsupport) { + rval.available = available + + if !available { + return + } + + rval.launchcontrol = lc + + a, _, _, d := cpuidex(0x12, 0) + rval.sgx1supported = a&0x01 != 0 + rval.sgx2supported = a&0x02 != 0 + rval.maxenclavesizenot64 = 1 << (d & 0xFF) // pow 2 + rval.maxenclavesize64 = 1 << ((d >> 8) & 0xFF) // pow 2 + rval.epcsections = make([]sgxepcsection, 0) + + for subleaf := uint32(2); subleaf < 2+8; subleaf++ { + eax, ebx, ecx, edx := cpuidex(0x12, subleaf) + leafType := eax & 0xf + + if leafType == 0 { + // Invalid subleaf, stop iterating + break + } else if leafType == 1 { + // EPC Section subleaf + baseAddress := uint64(eax&0xfffff000) + (uint64(ebx&0x000fffff) << 32) + size := uint64(ecx&0xfffff000) + (uint64(edx&0x000fffff) << 32) + + section := sgxepcsection{baseaddress: baseAddress, epcsize: size} + rval.epcsections = append(rval.epcsections, section) + } + } + + return +} + +func support() flags { + mfi := maxFunctionID() + vend := vendorID() + if mfi < 0x1 { + return 0 + } + rval := uint64(0) + _, _, c, d := cpuid(1) + if (d & (1 << 15)) != 0 { + rval |= cmov + } + if (d & (1 << 23)) != 0 { + rval |= mmx + } + if (d & (1 << 25)) != 0 { + rval |= mmxext + } + if (d & (1 << 25)) != 0 { + rval |= sse + } + if (d & (1 << 26)) != 0 { + rval |= sse2 + } + if (c & 1) != 0 { + rval |= sse3 + } + if (c & (1 << 5)) != 0 { + rval |= vmx + } + if (c & 0x00000200) != 0 { + rval |= ssse3 + } + if (c & 0x00080000) != 0 { + rval |= sse4 + } + if (c & 0x00100000) != 0 { + rval |= sse42 + } + if (c & (1 << 25)) != 0 { + rval |= aesni + } + if (c & (1 << 1)) != 0 { + rval |= clmul + } + if c&(1<<23) != 0 { + rval |= popcnt + } + if c&(1<<30) != 0 { + rval |= rdrand + } + if c&(1<<29) != 0 { + rval |= f16c + } + if c&(1<<13) != 0 { + rval |= cx16 + } + if vend == intel && (d&(1<<28)) != 0 && mfi >= 4 { + if threadsPerCore() > 1 { + rval |= htt + } + } + + // Check XGETBV, OXSAVE and AVX bits + if c&(1<<26) != 0 && c&(1<<27) != 0 && c&(1<<28) != 0 { + // Check for OS support + eax, _ := xgetbv(0) + if (eax & 0x6) == 0x6 { + rval |= avx + if (c & 0x00001000) != 0 { + rval |= fma3 + } + } + } + + // Check AVX2, AVX2 requires OS support, but BMI1/2 don't. + if mfi >= 7 { + _, ebx, ecx, edx := cpuidex(7, 0) + eax1, _, _, _ := cpuidex(7, 1) + if (rval&avx) != 0 && (ebx&0x00000020) != 0 { + rval |= avx2 + } + if (ebx & 0x00000008) != 0 { + rval |= bmi1 + if (ebx & 0x00000100) != 0 { + rval |= bmi2 + } + } + if ebx&(1<<2) != 0 { + rval |= sgx + } + if ebx&(1<<4) != 0 { + rval |= hle + } + if ebx&(1<<9) != 0 { + rval |= erms + } + if ebx&(1<<11) != 0 { + rval |= rtm + } + if ebx&(1<<14) != 0 { + rval |= mpx + } + if ebx&(1<<18) != 0 { + rval |= rdseed + } + if ebx&(1<<19) != 0 { + rval |= adx + } + if ebx&(1<<29) != 0 { + rval |= sha + } + if edx&(1<<26) != 0 { + rval |= ibpb + } + if ecx&(1<<30) != 0 { + rval |= sgxlc + } + if edx&(1<<27) != 0 { + rval |= stibp + } + + // Only detect AVX-512 features if XGETBV is supported + if c&((1<<26)|(1<<27)) == (1<<26)|(1<<27) { + // Check for OS support + eax, _ := xgetbv(0) + + // Verify that XCR0[7:5] = ‘111b’ (OPMASK state, upper 256-bit of ZMM0-ZMM15 and + // ZMM16-ZMM31 state are enabled by OS) + /// and that XCR0[2:1] = ‘11b’ (XMM state and YMM state are enabled by OS). + if (eax>>5)&7 == 7 && (eax>>1)&3 == 3 { + if ebx&(1<<16) != 0 { + rval |= avx512f + } + if ebx&(1<<17) != 0 { + rval |= avx512dq + } + if ebx&(1<<21) != 0 { + rval |= avx512ifma + } + if ebx&(1<<26) != 0 { + rval |= avx512pf + } + if ebx&(1<<27) != 0 { + rval |= avx512er + } + if ebx&(1<<28) != 0 { + rval |= avx512cd + } + if ebx&(1<<30) != 0 { + rval |= avx512bw + } + if ebx&(1<<31) != 0 { + rval |= avx512vl + } + // ecx + if ecx&(1<<1) != 0 { + rval |= avx512vbmi + } + if ecx&(1<<6) != 0 { + rval |= avx512vbmi2 + } + if ecx&(1<<8) != 0 { + rval |= gfni + } + if ecx&(1<<9) != 0 { + rval |= vaes + } + if ecx&(1<<10) != 0 { + rval |= vpclmulqdq + } + if ecx&(1<<11) != 0 { + rval |= avx512vnni + } + if ecx&(1<<12) != 0 { + rval |= avx512bitalg + } + if ecx&(1<<14) != 0 { + rval |= avx512vpopcntdq + } + // edx + if edx&(1<<8) != 0 { + rval |= avx512vp2intersect + } + // cpuid eax 07h,ecx=1 + if eax1&(1<<5) != 0 { + rval |= avx512bf16 + } + } + } + } + + if maxExtendedFunction() >= 0x80000001 { + _, _, c, d := cpuid(0x80000001) + if (c & (1 << 5)) != 0 { + rval |= lzcnt + rval |= popcnt + } + if (d & (1 << 31)) != 0 { + rval |= amd3dnow + } + if (d & (1 << 30)) != 0 { + rval |= amd3dnowext + } + if (d & (1 << 23)) != 0 { + rval |= mmx + } + if (d & (1 << 22)) != 0 { + rval |= mmxext + } + if (c & (1 << 6)) != 0 { + rval |= sse4a + } + if d&(1<<20) != 0 { + rval |= nx + } + if d&(1<<27) != 0 { + rval |= rdtscp + } + + /* Allow for selectively disabling SSE2 functions on AMD processors + with SSE2 support but not SSE4a. This includes Athlon64, some + Opteron, and some Sempron processors. MMX, SSE, or 3DNow! are faster + than SSE2 often enough to utilize this special-case flag. + AV_CPU_FLAG_SSE2 and AV_CPU_FLAG_SSE2SLOW are both set in this case + so that SSE2 is used unless explicitly disabled by checking + AV_CPU_FLAG_SSE2SLOW. */ + if vendorID() != intel && + rval&sse2 != 0 && (c&0x00000040) == 0 { + rval |= sse2slow + } + + /* XOP and FMA4 use the AVX instruction coding scheme, so they can't be + * used unless the OS has AVX support. */ + if (rval & avx) != 0 { + if (c & 0x00000800) != 0 { + rval |= xop + } + if (c & 0x00010000) != 0 { + rval |= fma4 + } + } + + if vendorID() == intel { + family, model := familyModel() + if family == 6 && (model == 9 || model == 13 || model == 14) { + /* 6/9 (pentium-m "banias"), 6/13 (pentium-m "dothan"), and + * 6/14 (core1 "yonah") theoretically support sse2, but it's + * usually slower than mmx. */ + if (rval & sse2) != 0 { + rval |= sse2slow + } + if (rval & sse3) != 0 { + rval |= sse3slow + } + } + /* The Atom processor has SSSE3 support, which is useful in many cases, + * but sometimes the SSSE3 version is slower than the SSE2 equivalent + * on the Atom, but is generally faster on other processors supporting + * SSSE3. This flag allows for selectively disabling certain SSSE3 + * functions on the Atom. */ + if family == 6 && model == 28 { + rval |= atom + } + } + } + return flags(rval) +} + +func valAsString(values ...uint32) []byte { + r := make([]byte, 4*len(values)) + for i, v := range values { + dst := r[i*4:] + dst[0] = byte(v & 0xff) + dst[1] = byte((v >> 8) & 0xff) + dst[2] = byte((v >> 16) & 0xff) + dst[3] = byte((v >> 24) & 0xff) + switch { + case dst[0] == 0: + return r[:i*4] + case dst[1] == 0: + return r[:i*4+1] + case dst[2] == 0: + return r[:i*4+2] + case dst[3] == 0: + return r[:i*4+3] + } + } + return r +} diff --git a/github.com/klauspost/compress/s2/cpuid_amd64.s b/github.com/klauspost/compress/s2/cpuid_amd64.s new file mode 100644 index 0000000..4cc19a4 --- /dev/null +++ b/github.com/klauspost/compress/s2/cpuid_amd64.s @@ -0,0 +1,44 @@ +// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file. + +// +build !appengine +// +build gc +// +build !noasm + +// func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32) +TEXT ·asmCpuid(SB), 7, $0 + XORQ CX, CX + MOVL op+0(FP), AX + CPUID + MOVL AX, eax+8(FP) + MOVL BX, ebx+12(FP) + MOVL CX, ecx+16(FP) + MOVL DX, edx+20(FP) + RET + +// func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32) +TEXT ·asmCpuidex(SB), 7, $0 + MOVL op+0(FP), AX + MOVL op2+4(FP), CX + CPUID + MOVL AX, eax+8(FP) + MOVL BX, ebx+12(FP) + MOVL CX, ecx+16(FP) + MOVL DX, edx+20(FP) + RET + +// func asmXgetbv(index uint32) (eax, edx uint32) +TEXT ·asmXgetbv(SB), 7, $0 + MOVL index+0(FP), CX + BYTE $0x0f; BYTE $0x01; BYTE $0xd0 // XGETBV + MOVL AX, eax+8(FP) + MOVL DX, edx+12(FP) + RET + +// func asmRdtscpAsm() (eax, ebx, ecx, edx uint32) +TEXT ·asmRdtscpAsm(SB), 7, $0 + BYTE $0x0F; BYTE $0x01; BYTE $0xF9 // RDTSCP + MOVL AX, eax+0(FP) + MOVL BX, ebx+4(FP) + MOVL CX, ecx+8(FP) + MOVL DX, edx+12(FP) + RET diff --git a/github.com/klauspost/compress/s2/decode.go b/github.com/klauspost/compress/s2/decode.go new file mode 100644 index 0000000..039ddb9 --- /dev/null +++ b/github.com/klauspost/compress/s2/decode.go @@ -0,0 +1,403 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 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. + +package s2 + +import ( + "encoding/binary" + "errors" + "io" +) + +var ( + // ErrCorrupt reports that the input is invalid. + ErrCorrupt = errors.New("s2: corrupt input") + // ErrCRC reports that the input failed CRC validation (streams only) + ErrCRC = errors.New("s2: corrupt input, crc mismatch") + // ErrTooLarge reports that the uncompressed length is too large. + ErrTooLarge = errors.New("s2: decoded block is too large") + // ErrUnsupported reports that the input isn't supported. + ErrUnsupported = errors.New("s2: unsupported input") + + errUnsupportedLiteralLength = errors.New("s2: 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 <= cap(dst) { + dst = dst[:dLen] + } else { + dst = make([]byte, dLen) + } + switch s2Decode(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 with S2 changes. +func NewReader(r io.Reader) *Reader { + return &Reader{ + r: r, + buf: make([]byte, MaxEncodedLen(maxBlockSize)+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) { + if n > maxBlockSize { + r.err = ErrCorrupt + return 0, r.err + } + r.decoded = make([]byte, n) + } + if _, err := Decode(r.decoded, buf); err != nil { + r.err = err + return 0, r.err + } + if crc(r.decoded[:n]) != checksum { + r.err = ErrCRC + 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) { + if n > maxBlockSize { + r.err = ErrCorrupt + return 0, r.err + } + r.decoded = make([]byte, n) + } + if !r.readFull(r.decoded[:n], false) { + return 0, r.err + } + if crc(r.decoded[:n]) != checksum { + r.err = ErrCRC + 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 + } + if string(r.buf[:len(magicBody)]) != magicBody { + if string(r.buf[:len(magicBody)]) != magicBodySnappy { + 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 + } + } +} + +// Skip will skip n bytes forward in the decompressed output. +// For larger skips this consumes less CPU and is faster than reading output and discarding it. +// CRC is not checked on skipped blocks. +// io.ErrUnexpectedEOF is returned if the stream ends before all bytes have been skipped. +// If a decoding error is encountered subsequent calls to Read will also fail. +func (r *Reader) Skip(n int64) error { + if n < 0 { + return errors.New("attempted negative skip") + } + if r.err != nil { + return r.err + } + + for n > 0 { + if r.i < r.j { + // Skip in buffer. + // decoded[i:j] contains decoded bytes that have not yet been passed on. + left := int64(r.j - r.i) + if left >= n { + r.i += int(n) + return nil + } + n -= int64(r.j - r.i) + r.i, r.j = 0, 0 + } + + // Buffer empty; read blocks until we have content. + if !r.readFull(r.buf[:4], true) { + if r.err == io.EOF { + r.err = io.ErrUnexpectedEOF + } + return r.err + } + chunkType := r.buf[0] + if !r.readHeader { + if chunkType != chunkTypeStreamIdentifier { + r.err = ErrCorrupt + return r.err + } + r.readHeader = true + } + chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16 + if chunkLen > len(r.buf) { + r.err = ErrUnsupported + return r.err + } + + // The chunk types are specified at + // https://github.com/google/snappy/blob/master/framing_format.txt + switch chunkType { + case chunkTypeCompressedData: + // Section 4.2. Compressed data (chunk type 0x00). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return r.err + } + buf := r.buf[:chunkLen] + if !r.readFull(buf, false) { + return r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + buf = buf[checksumSize:] + + dLen, err := DecodedLen(buf) + if err != nil { + r.err = err + return r.err + } + if dLen > maxBlockSize { + r.err = ErrCorrupt + return r.err + } + // Check if destination is within this block + if int64(dLen) > n { + if len(r.decoded) < dLen { + r.decoded = make([]byte, dLen) + } + if _, err := Decode(r.decoded, buf); err != nil { + r.err = err + return r.err + } + if crc(r.decoded[:dLen]) != checksum { + r.err = ErrCorrupt + return r.err + } + } else { + // Skip block completely + n -= int64(dLen) + dLen = 0 + } + r.i, r.j = 0, dLen + continue + case chunkTypeUncompressedData: + // Section 4.3. Uncompressed data (chunk type 0x01). + if chunkLen < checksumSize { + r.err = ErrCorrupt + return r.err + } + buf := r.buf[:checksumSize] + if !r.readFull(buf, false) { + return r.err + } + checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24 + // Read directly into r.decoded instead of via r.buf. + n2 := chunkLen - checksumSize + if n2 > len(r.decoded) { + if n2 > maxBlockSize { + r.err = ErrCorrupt + return r.err + } + r.decoded = make([]byte, n2) + } + if !r.readFull(r.decoded[:n2], false) { + return r.err + } + if int64(n2) < n { + if crc(r.decoded[:n2]) != checksum { + r.err = ErrCorrupt + return r.err + } + } + r.i, r.j = 0, n2 + continue + case chunkTypeStreamIdentifier: + // Section 4.1. Stream identifier (chunk type 0xff). + if chunkLen != len(magicBody) { + r.err = ErrCorrupt + return r.err + } + if !r.readFull(r.buf[:len(magicBody)], false) { + return r.err + } + if string(r.buf[:len(magicBody)]) != magicBody { + if string(r.buf[:len(magicBody)]) != magicBodySnappy { + r.err = ErrCorrupt + return r.err + } + } + + continue + } + + if chunkType <= 0x7f { + // Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f). + r.err = ErrUnsupported + return r.err + } + // Section 4.4 Padding (chunk type 0xfe). + // Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd). + if !r.readFull(r.buf[:chunkLen], false) { + return r.err + } + + return io.ErrUnexpectedEOF + } + return nil +} diff --git a/github.com/klauspost/compress/s2/decode_amd64.go b/github.com/klauspost/compress/s2/decode_amd64.go new file mode 100644 index 0000000..8a89efe --- /dev/null +++ b/github.com/klauspost/compress/s2/decode_amd64.go @@ -0,0 +1,15 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 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. + +// +build !appengine +// +build gc +// +build !noasm + +package s2 + +// decode has the same semantics as in decode_other.go. +// +//go:noescape +func s2Decode(dst, src []byte) int diff --git a/github.com/klauspost/compress/s2/decode_amd64.s b/github.com/klauspost/compress/s2/decode_amd64.s new file mode 100644 index 0000000..2faee6e --- /dev/null +++ b/github.com/klauspost/compress/s2/decode_amd64.s @@ -0,0 +1,570 @@ +// Copyright 2016 The Go Authors. All rights reserved. +// Copyright (c) 2019 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. + +// +build !appengine +// +build gc +// +build !noasm + +#include "textflag.h" + +#define R_TMP0 AX +#define R_TMP1 BX +#define R_LEN CX +#define R_OFF DX +#define R_SRC SI +#define R_DST DI +#define R_DBASE R8 +#define R_DLEN R9 +#define R_DEND R10 +#define R_SBASE R11 +#define R_SLEN R12 +#define R_SEND R13 +#define R_TMP2 R14 +#define R_TMP3 R15 + +// 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: +// - R_TMP0 scratch +// - R_TMP1 scratch +// - R_LEN length or x (shared) +// - R_X length or x (shared) +// - R_OFF offset +// - R_SRC &src[s] +// - R_DST &dst[d] +// + R_DBASE dst_base +// + R_DLEN dst_len +// + R_DEND dst_base + dst_len +// + R_SBASE src_base +// + R_SLEN src_len +// + R_SEND src_base + src_len +// - R_TMP2 used by doCopy +// - R_TMP3 used by doCopy +// +// The registers R_DBASE-R_SEND (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 R_DST - R_DBASE, and len(dst)-d is R_DEND - R_DST. +// The s variable is implicitly R_SRC - R_SBASE, and len(src)-s is R_SEND - R_SRC. +TEXT ·s2Decode(SB), NOSPLIT, $48-56 + // Initialize R_SRC, R_DST and R_DBASE-R_SEND. + MOVQ dst_base+0(FP), R_DBASE + MOVQ dst_len+8(FP), R_DLEN + MOVQ R_DBASE, R_DST + MOVQ R_DBASE, R_DEND + ADDQ R_DLEN, R_DEND + MOVQ src_base+24(FP), R_SBASE + MOVQ src_len+32(FP), R_SLEN + MOVQ R_SBASE, R_SRC + MOVQ R_SBASE, R_SEND + ADDQ R_SLEN, R_SEND + XORQ R_OFF, R_OFF + +loop: + // for s < len(src) + CMPQ R_SRC, R_SEND + JEQ end + + // R_LEN = uint32(src[s]) + // + // switch src[s] & 0x03 + MOVBLZX (R_SRC), R_LEN + MOVL R_LEN, R_TMP1 + ANDL $3, R_TMP1 + CMPL R_TMP1, $1 + JAE tagCopy + + // ---------------------------------------- + // The code below handles literal tags. + + // case tagLiteral: + // x := uint32(src[s] >> 2) + // switch + SHRL $2, R_LEN + CMPL R_LEN, $60 + JAE tagLit60Plus + + // case x < 60: + // s++ + INCQ R_SRC + +doLit: + // This is the end of the inner "switch", when we have a literal tag. + // + // We assume that R_LEN == 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 + // R_LEN can hold 64 bits, so the increment cannot overflow. + INCQ R_LEN + + // Prepare to check if copying length bytes will run past the end of dst or + // src. + // + // R_TMP0 = len(dst) - d + // R_TMP1 = len(src) - s + MOVQ R_DEND, R_TMP0 + SUBQ R_DST, R_TMP0 + MOVQ R_SEND, R_TMP1 + SUBQ R_SRC, R_TMP1 + + // !!! 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 R_LEN, $16 + JGT callMemmove + CMPQ R_TMP0, $16 + JLT callMemmove + CMPQ R_TMP1, $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(R_SRC), X0 + MOVOU X0, 0(R_DST) + + // d += length + // s += length + ADDQ R_LEN, R_DST + ADDQ R_LEN, R_SRC + JMP loop + +callMemmove: + // if length > len(dst)-d || length > len(src)-s { etc } + CMPQ R_LEN, R_TMP0 + JGT errCorrupt + CMPQ R_LEN, R_TMP1 + JGT errCorrupt + + // copy(dst[d:], src[s:s+length]) + // + // This means calling runtime·memmove(&dst[d], &src[s], length), so we push + // R_DST, R_SRC and R_LEN as arguments. Coincidentally, we also need to spill those + // three registers to the stack, to save local variables across the CALL. + MOVQ R_DST, 0(SP) + MOVQ R_SRC, 8(SP) + MOVQ R_LEN, 16(SP) + MOVQ R_DST, 24(SP) + MOVQ R_SRC, 32(SP) + MOVQ R_LEN, 40(SP) + CALL runtime·memmove(SB) + + // Restore local variables: unspill registers from the stack and + // re-calculate R_DBASE-R_SEND. + MOVQ 24(SP), R_DST + MOVQ 32(SP), R_SRC + MOVQ 40(SP), R_LEN + MOVQ dst_base+0(FP), R_DBASE + MOVQ dst_len+8(FP), R_DLEN + MOVQ R_DBASE, R_DEND + ADDQ R_DLEN, R_DEND + MOVQ src_base+24(FP), R_SBASE + MOVQ src_len+32(FP), R_SLEN + MOVQ R_SBASE, R_SEND + ADDQ R_SLEN, R_SEND + + // d += length + // s += length + ADDQ R_LEN, R_DST + ADDQ R_LEN, R_SRC + 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 R_LEN, R_SRC + SUBQ $58, R_SRC + CMPQ R_SRC, R_SEND + JA errCorrupt + + // case x == 60: + CMPL R_LEN, $61 + JEQ tagLit61 + JA tagLit62Plus + + // x = uint32(src[s-1]) + MOVBLZX -1(R_SRC), R_LEN + JMP doLit + +tagLit61: + // case x == 61: + // x = uint32(src[s-2]) | uint32(src[s-1])<<8 + MOVWLZX -2(R_SRC), R_LEN + JMP doLit + +tagLit62Plus: + CMPL R_LEN, $62 + JA tagLit63 + + // case x == 62: + // x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16 + MOVWLZX -3(R_SRC), R_LEN + MOVBLZX -1(R_SRC), R_TMP1 + SHLL $16, R_TMP1 + ORL R_TMP1, R_LEN + 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(R_SRC), R_LEN + JMP doLit + +// The code above handles literal tags. +// ---------------------------------------- +// The code below handles copy tags. + +tagCopy4: + // case tagCopy4: + // s += 5 + ADDQ $5, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = 1 + int(src[s-5])>>2 + SHRQ $2, R_LEN + INCQ R_LEN + + // offset = int(uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24) + MOVLQZX -4(R_SRC), R_OFF + JMP doCopy + +tagCopy2: + // case tagCopy2: + // s += 3 + ADDQ $3, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = 1 + int(src[s-3])>>2 + SHRQ $2, R_LEN + INCQ R_LEN + + // offset = int(uint32(src[s-2]) | uint32(src[s-1])<<8) + MOVWQZX -2(R_SRC), R_OFF + JMP doCopy + +tagCopy: + // We have a copy tag. We assume that: + // - R_TMP1 == src[s] & 0x03 + // - R_LEN == src[s] + CMPQ R_TMP1, $2 + JEQ tagCopy2 + JA tagCopy4 + + // case tagCopy1: + // s += 2 + ADDQ $2, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // offset = int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + MOVQ R_LEN, R_TMP0 + ANDQ $0xe0, R_TMP0 + SHLQ $3, R_TMP0 + MOVBQZX -1(R_SRC), R_TMP1 + ORQ R_TMP1, R_TMP0 + + // length = 4 + int(src[s-2])>>2&0x7 + SHRQ $2, R_LEN + ANDQ $7, R_LEN + ADDQ $4, R_LEN + + // check if repeat code + CMPQ R_TMP0, $0 + JE repeatCode + + // This is a regular copy, transfer our temporary value to R_OFF (length) + MOVQ R_TMP0, R_OFF + JMP doCopy + +// This is a repeat code. +repeatCode: + // If length < 9, reuse last offset, with the length already calculated. + CMPQ R_LEN, $9 + JL doCopyRepeat + + // Read additional bytes for length. + JE repeatLen1 + + // Rare, so the extra branch shouldn't hurt too much. + CMPQ R_LEN, $10 + JE repeatLen2 + JMP repeatLen3 + +// Read repeat lengths. +repeatLen1: + // s ++ + ADDQ $1, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = src[s-1] + 8 + MOVBQZX -1(R_SRC), R_LEN + ADDL $8, R_LEN + JMP doCopyRepeat + +repeatLen2: + // s +=2 + ADDQ $2, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = uint32(src[s-2]) | (uint32(src[s-1])<<8) + (1 << 8) + MOVWQZX -2(R_SRC), R_LEN + ADDL $260, R_LEN + JMP doCopyRepeat + +repeatLen3: + // s +=3 + ADDQ $3, R_SRC + + // if uint(s) > uint(len(src)) { etc } + CMPQ R_SRC, R_SEND + JA errCorrupt + + // length = uint32(src[s-3]) | (uint32(src[s-2])<<8) | (uint32(src[s-1])<<16) + (1 << 16) + MOVBLZX -1(R_SRC), R_TMP0 + MOVWLZX -3(R_SRC), R_LEN + SHLL $16, R_TMP0 + ORL R_TMP0, R_LEN + ADDL $65540, R_LEN + JMP doCopyRepeat + +doCopy: + // This is the end of the outer "switch", when we have a copy tag. + // + // We assume that: + // - R_LEN == length && R_LEN > 0 + // - R_OFF == offset + + // if d < offset { etc } + MOVQ R_DST, R_TMP1 + SUBQ R_DBASE, R_TMP1 + CMPQ R_TMP1, R_OFF + JLT errCorrupt + + // Repeat values can skip the test above, since any offset > 0 will be in dst. +doCopyRepeat: + // if offset <= 0 { etc } + CMPQ R_OFF, $0 + JLE errCorrupt + + // if length > len(dst)-d { etc } + MOVQ R_DEND, R_TMP1 + SUBQ R_DST, R_TMP1 + CMPQ R_LEN, R_TMP1 + JGT errCorrupt + + // forwardCopy(dst[d:d+length], dst[d-offset:]); d += length + // + // Set: + // - R_TMP2 = len(dst)-d + // - R_TMP3 = &dst[d-offset] + MOVQ R_DEND, R_TMP2 + SUBQ R_DST, R_TMP2 + MOVQ R_DST, R_TMP3 + SUBQ R_OFF, R_TMP3 + + // !!! 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 R_LEN, $16 + JGT slowForwardCopy + CMPQ R_OFF, $8 + JLT slowForwardCopy + CMPQ R_TMP2, $16 + JLT slowForwardCopy + MOVQ 0(R_TMP3), R_TMP0 + MOVQ R_TMP0, 0(R_DST) + MOVQ 8(R_TMP3), R_TMP1 + MOVQ R_TMP1, 8(R_DST) + ADDQ R_LEN, R_DST + 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, R_TMP2 + CMPQ R_LEN, R_TMP2 + JGT verySlowForwardCopy + + // We want to keep the offset, so we use R_TMP2 from here. + MOVQ R_OFF, R_TMP2 + +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 + // // R_TMP3, is unchanged. + // } + CMPQ R_TMP2, $8 + JGE fixUpSlowForwardCopy + MOVQ (R_TMP3), R_TMP1 + MOVQ R_TMP1, (R_DST) + SUBQ R_TMP2, R_LEN + ADDQ R_TMP2, R_DST + ADDQ R_TMP2, R_TMP2 + JMP makeOffsetAtLeast8 + +fixUpSlowForwardCopy: + // !!! Add length (which might be negative now) to d (implied by R_DST 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 R_DST to R_TMP0 so that, if + // length is positive, copying the remaining length bytes will write to the + // right place. + MOVQ R_DST, R_TMP0 + ADDQ R_LEN, R_DST + +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 R_LEN, $0 + JLE loop + MOVQ (R_TMP3), R_TMP1 + MOVQ R_TMP1, (R_TMP0) + ADDQ $8, R_TMP3 + ADDQ $8, R_TMP0 + SUBQ $8, R_LEN + 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 (R_TMP3), R_TMP1 + MOVB R_TMP1, (R_DST) + INCQ R_TMP3 + INCQ R_DST + DECQ R_LEN + 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 R_DST, R_DEND + JNE errCorrupt + + // return 0 + MOVQ $0, ret+48(FP) + RET + +errCorrupt: + // return decodeErrCodeCorrupt + MOVQ $1, ret+48(FP) + RET diff --git a/github.com/klauspost/compress/s2/decode_other.go b/github.com/klauspost/compress/s2/decode_other.go new file mode 100644 index 0000000..4cb6100 --- /dev/null +++ b/github.com/klauspost/compress/s2/decode_other.go @@ -0,0 +1,161 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 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. + +// +build !amd64 appengine !gc noasm + +package s2 + +import "fmt" + +// 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 s2Decode(dst, src []byte) int { + const debug = false + if debug { + fmt.Println("Starting decode, dst len:", len(dst)) + } + var d, s, length int + offset := 0 + 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 + } + if debug { + fmt.Println("literals, length:", length, "d-after:", d+length) + } + + 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 = int(src[s-2]) >> 2 & 0x7 + toffset := int(uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])) + if toffset == 0 { + if debug { + fmt.Print("(repeat) ") + } + // keep last offset + switch length { + case 5: + s += 1 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(uint32(src[s-1])) + 4 + case 6: + s += 2 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(uint32(src[s-2])|(uint32(src[s-1])<<8)) + (1 << 8) + case 7: + s += 3 + if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line. + return decodeErrCodeCorrupt + } + length = int(uint32(src[s-3])|(uint32(src[s-2])<<8)|(uint32(src[s-1])<<16)) + (1 << 16) + default: // 0-> 4 + } + } else { + offset = toffset + } + length += 4 + 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 + } + + if debug { + fmt.Println("copy, length:", length, "offset:", offset, "d-after:", d+length) + } + + // 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/github.com/klauspost/compress/s2/decode_test.go b/github.com/klauspost/compress/s2/decode_test.go new file mode 100644 index 0000000..0a6a812 --- /dev/null +++ b/github.com/klauspost/compress/s2/decode_test.go @@ -0,0 +1,43 @@ +// Copyright (c) 2019 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. + +package s2 + +import ( + "bytes" + "io/ioutil" + "strings" + "testing" + + "github.com/klauspost/compress/zip" +) + +func TestDecodeRegression(t *testing.T) { + data, err := ioutil.ReadFile("testdata/dec-block-regressions.zip") + if err != nil { + t.Fatal(err) + } + zr, err := zip.NewReader(bytes.NewReader(data), int64(len(data))) + if err != nil { + t.Fatal(err) + } + for _, tt := range zr.File { + if !strings.HasSuffix(t.Name(), "") { + continue + } + t.Run(tt.Name, func(t *testing.T) { + r, err := tt.Open() + if err != nil { + t.Error(err) + return + } + in, err := ioutil.ReadAll(r) + if err != nil { + t.Error(err) + } + got, err := Decode(nil, in) + t.Log("Received:", len(got), err) + }) + } +} diff --git a/github.com/klauspost/compress/s2/encode.go b/github.com/klauspost/compress/s2/encode.go new file mode 100644 index 0000000..53431d9 --- /dev/null +++ b/github.com/klauspost/compress/s2/encode.go @@ -0,0 +1,924 @@ +// Copyright 2011 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 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. + +package s2 + +import ( + "crypto/rand" + "encoding/binary" + "errors" + "fmt" + "io" + "math" + "math/bits" + "runtime" + "sync" +) + +// 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. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func Encode(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if cap(dst) < n { + dst = make([]byte, n) + } else { + dst = dst[:n] + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlock(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeBetter 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. +// +// EncodeBetter compresses better than Encode but typically with a +// 10-40% speed decrease on both compression and decompression. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeBetter(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))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlockBetter(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// EncodeSnappy 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 output is Snappy compatible and will likely decompress faster. +// +// The dst and src must not overlap. It is valid to pass a nil dst. +// +// The blocks will require the same amount of memory to decode as encoding, +// and does not make for concurrent decoding. +// Also note that blocks do not contain CRC information, so corruption may be undetected. +// +// If you need to encode larger amounts of data, consider using +// the streaming interface which gives all of these features. +func EncodeSnappy(dst, src []byte) []byte { + if n := MaxEncodedLen(len(src)); n < 0 { + panic(ErrTooLarge) + } else if cap(dst) < n { + dst = make([]byte, n) + } else { + dst = dst[:n] + } + + // The block starts with the varint-encoded length of the decompressed bytes. + d := binary.PutUvarint(dst, uint64(len(src))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + + n := encodeBlockSnappy(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// ConcatBlocks will concatenate the supplied blocks and append them to the supplied destination. +// If the destination is nil or too small, a new will be allocated. +// The blocks are not validated, so garbage in = garbage out. +// dst may not overlap block data. +// Any data in dst is preserved as is, so it will not be considered a block. +func ConcatBlocks(dst []byte, blocks ...[]byte) ([]byte, error) { + totalSize := uint64(0) + compSize := 0 + for _, b := range blocks { + l, hdr, err := decodedLen(b) + if err != nil { + return nil, err + } + totalSize += uint64(l) + compSize += len(b) - hdr + } + if totalSize == 0 { + dst = append(dst, 0) + return dst, nil + } + if totalSize > math.MaxUint32 { + return nil, ErrTooLarge + } + var tmp [binary.MaxVarintLen32]byte + hdrSize := binary.PutUvarint(tmp[:], totalSize) + wantSize := hdrSize + compSize + + if cap(dst)-len(dst) < wantSize { + dst = append(make([]byte, 0, wantSize+len(dst)), dst...) + } + dst = append(dst, tmp[:hdrSize]...) + for _, b := range blocks { + _, hdr, err := decodedLen(b) + if err != nil { + return nil, err + } + dst = append(dst, b[hdr:]...) + } + return dst, nil +} + +// 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 = 8 + +// minNonLiteralBlockSize is the minimum size of the input to encodeBlock that +// will be accepted by the encoder. +const minNonLiteralBlockSize = 32 + +// MaxBlockSize is the maximum value where MaxEncodedLen will return a valid block size. +// Blocks this big are highly discouraged, though. +const MaxBlockSize = math.MaxUint32 - binary.MaxVarintLen32 - 5 + +// 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. +// 32 bit platforms will have lower thresholds for rejecting big content. +func MaxEncodedLen(srcLen int) int { + n := uint64(srcLen) + if n > 0xffffffff { + // Also includes negative. + return -1 + } + // Size of the varint encoded block size. + n = n + uint64((bits.Len64(n)+7)/7) + + // Add maximum size of encoding block as literals. + n += uint64(literalExtraSize(int64(srcLen))) + if n > 0xffffffff { + return -1 + } + return int(n) +} + +var errClosed = errors.New("s2: Writer is closed") + +// NewWriter returns a new Writer that compresses to w, using the +// framing format described at +// https://github.com/google/snappy/blob/master/framing_format.txt +// +// Users must call Close to guarantee all data has been forwarded to +// the underlying io.Writer and that resources are released. +// They may also call Flush zero or more times before calling Close. +func NewWriter(w io.Writer, opts ...WriterOption) *Writer { + w2 := Writer{ + blockSize: defaultBlockSize, + concurrency: runtime.GOMAXPROCS(0), + } + for _, opt := range opts { + if err := opt(&w2); err != nil { + w2.errState = err + return &w2 + } + } + w2.obufLen = obufHeaderLen + MaxEncodedLen(w2.blockSize) + w2.paramsOK = true + w2.ibuf = make([]byte, 0, w2.blockSize) + w2.buffers.New = func() interface{} { + return make([]byte, w2.obufLen) + } + w2.Reset(w) + return &w2 +} + +// Writer is an io.Writer that can write Snappy-compressed bytes. +type Writer struct { + errMu sync.Mutex + errState error + + // ibuf is a buffer for the incoming (uncompressed) bytes. + ibuf []byte + + blockSize int + obufLen int + concurrency int + written int64 + output chan chan result + buffers sync.Pool + pad int + + writer io.Writer + writerWg sync.WaitGroup + + // wroteStreamHeader is whether we have written the stream header. + wroteStreamHeader bool + paramsOK bool + better bool +} + +type result []byte + +// err returns the previously set error. +// If no error has been set it is set to err if not nil. +func (w *Writer) err(err error) error { + w.errMu.Lock() + errSet := w.errState + if errSet == nil && err != nil { + w.errState = err + errSet = err + } + w.errMu.Unlock() + return errSet +} + +// 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) { + if !w.paramsOK { + return + } + // Close previous writer, if any. + if w.output != nil { + close(w.output) + w.writerWg.Wait() + w.output = nil + } + w.errState = nil + w.ibuf = w.ibuf[:0] + w.wroteStreamHeader = false + w.written = 0 + w.writer = writer + // If we didn't get a writer, stop here. + if writer == nil { + return + } + // If no concurrency requested, don't spin up writer goroutine. + if w.concurrency == 1 { + return + } + + toWrite := make(chan chan result, w.concurrency) + w.output = toWrite + w.writerWg.Add(1) + + // Start a writer goroutine that will write all output in order. + go func() { + defer w.writerWg.Done() + + // Get a queued write. + for write := range toWrite { + // Wait for the data to be available. + in := <-write + if len(in) > 0 { + if w.err(nil) == nil { + // Don't expose data from previous buffers. + toWrite := in[:len(in):len(in)] + // Write to output. + n, err := writer.Write(toWrite) + if err == nil && n != len(toWrite) { + err = io.ErrShortBuffer + } + _ = w.err(err) + w.written += int64(n) + } + } + if cap(in) >= w.obufLen { + w.buffers.Put([]byte(in)) + } + // close the incoming write request. + // This can be used for synchronizing flushes. + close(write) + } + }() +} + +// Write satisfies the io.Writer interface. +func (w *Writer) Write(p []byte) (nRet int, errRet error) { + // If we exceed the input buffer size, start writing + for len(p) > (cap(w.ibuf)-len(w.ibuf)) && w.err(nil) == 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.write(w.ibuf) + w.ibuf = w.ibuf[:0] + } + nRet += n + p = p[n:] + } + if err := w.err(nil); err != nil { + return nRet, err + } + // p should always be able to fit into w.ibuf now. + n := copy(w.ibuf[len(w.ibuf):cap(w.ibuf)], p) + w.ibuf = w.ibuf[:len(w.ibuf)+n] + nRet += n + return nRet, nil +} + +// ReadFrom implements the io.ReaderFrom interface. +// Using this is typically more efficient since it avoids a memory copy. +// 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. +func (w *Writer) ReadFrom(r io.Reader) (n int64, err error) { + if len(w.ibuf) > 0 { + err := w.Flush() + if err != nil { + return 0, err + } + } + for { + inbuf := w.buffers.Get().([]byte)[:w.blockSize+obufHeaderLen] + n2, err := io.ReadFull(r, inbuf[obufHeaderLen:]) + if err != nil { + if err == io.ErrUnexpectedEOF { + err = io.EOF + } + if err != io.EOF { + return n, w.err(err) + } + } + if n2 == 0 { + break + } + n += int64(n2) + err2 := w.writeFull(inbuf[:n2+obufHeaderLen]) + if w.err(err2) != nil { + break + } + + if err != nil { + // We got EOF and wrote everything + break + } + } + + return n, w.err(nil) +} + +// EncodeBuffer will add a buffer to the stream. +// This is the fastest way to encode a stream, +// but the input buffer cannot be written to by the caller +// until this function, Flush or Close has been called. +// +// Note that input is not buffered. +// This means that each write will result in discrete blocks being created. +// For buffered writes, use the regular Write function. +func (w *Writer) EncodeBuffer(buf []byte) (err error) { + if err := w.err(nil); err != nil { + return err + } + + // Flush queued data first. + if len(w.ibuf) > 0 { + err := w.Flush() + if err != nil { + return err + } + } + if w.concurrency == 1 { + _, err := w.writeSync(buf) + return err + } + + // Spawn goroutine and write block to output channel. + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + hWriter <- []byte(magicChunk) + } + + for len(buf) > 0 { + // Cut input. + uncompressed := buf + if len(uncompressed) > w.blockSize { + uncompressed = uncompressed[:w.blockSize] + } + buf = buf[len(uncompressed):] + // Get an output buffer. + obuf := w.buffers.Get().([]byte)[:len(uncompressed)+obufHeaderLen] + output := make(chan result) + // Queue output now, so we keep order. + w.output <- output + go func() { + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + var n2 int + if w.better { + n2 = encodeBlockBetter(obuf[obufHeaderLen+n:], uncompressed) + } else { + n2 = encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + } + + // Check if we should use this, or store as uncompressed instead. + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + // copy uncompressed + copy(obuf[obufHeaderLen:], uncompressed) + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + // Queue final output. + output <- obuf + }() + } + return nil +} + +func (w *Writer) write(p []byte) (nRet int, errRet error) { + if err := w.err(nil); err != nil { + return 0, err + } + if w.concurrency == 1 { + return w.writeSync(p) + } + + // Spawn goroutine and write block to output channel. + for len(p) > 0 { + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + hWriter <- []byte(magicChunk) + } + + var uncompressed []byte + if len(p) > w.blockSize { + uncompressed, p = p[:w.blockSize], p[w.blockSize:] + } else { + uncompressed, p = p, nil + } + + // Copy input. + // If the block is incompressible, this is used for the result. + inbuf := w.buffers.Get().([]byte)[:len(uncompressed)+obufHeaderLen] + obuf := w.buffers.Get().([]byte)[:w.obufLen] + copy(inbuf[obufHeaderLen:], uncompressed) + uncompressed = inbuf[obufHeaderLen:] + + output := make(chan result) + // Queue output now, so we keep order. + w.output <- output + go func() { + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + var n2 int + if w.better { + n2 = encodeBlockBetter(obuf[obufHeaderLen+n:], uncompressed) + } else { + n2 = encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + } + + // Check if we should use this, or store as uncompressed instead. + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + // Use input as output. + obuf, inbuf = inbuf, obuf + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + // Queue final output. + output <- obuf + + // Put unused buffer back in pool. + w.buffers.Put(inbuf) + }() + nRet += len(uncompressed) + } + return nRet, nil +} + +// writeFull is a special version of write that will always write the full buffer. +// Data to be compressed should start at offset obufHeaderLen and fill the remainder of the buffer. +// The data will be written as a single block. +// The caller is not allowed to use inbuf after this function has been called. +func (w *Writer) writeFull(inbuf []byte) (errRet error) { + if err := w.err(nil); err != nil { + return err + } + + if w.concurrency == 1 { + _, err := w.writeSync(inbuf[obufHeaderLen:]) + return err + } + + // Spawn goroutine and write block to output channel. + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + hWriter := make(chan result) + w.output <- hWriter + hWriter <- []byte(magicChunk) + } + + // Get an output buffer. + obuf := w.buffers.Get().([]byte)[:w.obufLen] + uncompressed := inbuf[obufHeaderLen:] + + output := make(chan result) + // Queue output now, so we keep order. + w.output <- output + go func() { + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + var n2 int + if w.better { + n2 = encodeBlockBetter(obuf[obufHeaderLen+n:], uncompressed) + } else { + n2 = encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + } + + // Check if we should use this, or store as uncompressed instead. + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + // Use input as output. + obuf, inbuf = inbuf, obuf + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + // Queue final output. + output <- obuf + + // Put unused buffer back in pool. + w.buffers.Put(inbuf) + }() + return nil +} + +func (w *Writer) writeSync(p []byte) (nRet int, errRet error) { + if err := w.err(nil); err != nil { + return 0, err + } + if !w.wroteStreamHeader { + w.wroteStreamHeader = true + n, err := w.writer.Write([]byte(magicChunk)) + if err != nil { + return 0, w.err(err) + } + if n != len(magicChunk) { + return 0, w.err(io.ErrShortWrite) + } + w.written += int64(n) + } + + for len(p) > 0 { + var uncompressed []byte + if len(p) > w.blockSize { + uncompressed, p = p[:w.blockSize], p[w.blockSize:] + } else { + uncompressed, p = p, nil + } + + obuf := w.buffers.Get().([]byte)[:w.obufLen] + checksum := crc(uncompressed) + + // Set to uncompressed. + chunkType := uint8(chunkTypeUncompressedData) + chunkLen := 4 + len(uncompressed) + + // Attempt compressing. + n := binary.PutUvarint(obuf[obufHeaderLen:], uint64(len(uncompressed))) + var n2 int + if w.better { + n2 = encodeBlockBetter(obuf[obufHeaderLen+n:], uncompressed) + } else { + n2 = encodeBlock(obuf[obufHeaderLen+n:], uncompressed) + } + + if n2 > 0 { + chunkType = uint8(chunkTypeCompressedData) + chunkLen = 4 + n + n2 + obuf = obuf[:obufHeaderLen+n+n2] + } else { + obuf = obuf[:8] + } + + // Fill in the per-chunk header that comes before the body. + obuf[0] = chunkType + obuf[1] = uint8(chunkLen >> 0) + obuf[2] = uint8(chunkLen >> 8) + obuf[3] = uint8(chunkLen >> 16) + obuf[4] = uint8(checksum >> 0) + obuf[5] = uint8(checksum >> 8) + obuf[6] = uint8(checksum >> 16) + obuf[7] = uint8(checksum >> 24) + + n, err := w.writer.Write(obuf) + if err != nil { + return 0, w.err(err) + } + if n != len(obuf) { + return 0, w.err(io.ErrShortWrite) + } + w.written += int64(n) + if chunkType == chunkTypeUncompressedData { + // Write uncompressed data. + n, err := w.writer.Write(uncompressed) + if err != nil { + return 0, w.err(err) + } + if n != len(uncompressed) { + return 0, w.err(io.ErrShortWrite) + } + w.written += int64(n) + } + w.buffers.Put(obuf) + // Queue final output. + nRet += len(uncompressed) + } + return nRet, nil +} + +// Flush flushes the Writer to its underlying io.Writer. +// This does not apply padding. +func (w *Writer) Flush() error { + if err := w.err(nil); err != nil { + return err + } + + // Queue any data still in input buffer. + if len(w.ibuf) != 0 { + _, err := w.write(w.ibuf) + w.ibuf = w.ibuf[:0] + err = w.err(err) + if err != nil { + return err + } + } + if w.output == nil { + return w.err(nil) + } + + // Send empty buffer + res := make(chan result) + w.output <- res + // Block until this has been picked up. + res <- nil + // When it is closed, we have flushed. + <-res + return w.err(nil) +} + +// Close calls Flush and then closes the Writer. +// Calling Close multiple times is ok. +func (w *Writer) Close() error { + err := w.Flush() + if w.output != nil { + close(w.output) + w.writerWg.Wait() + w.output = nil + } + if w.err(nil) == nil && w.writer != nil && w.pad > 0 { + add := calcSkippableFrame(w.written, int64(w.pad)) + frame, err := skippableFrame(w.ibuf[:0], add, rand.Reader) + if err = w.err(err); err != nil { + return err + } + _, err2 := w.writer.Write(frame) + _ = w.err(err2) + } + _ = w.err(errClosed) + if err == errClosed { + return nil + } + return err +} + +const skippableFrameHeader = 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 < maxBlockSize + skippableFrameHeader +func skippableFrame(dst []byte, total int, r io.Reader) ([]byte, error) { + if total == 0 { + return dst, nil + } + if total < skippableFrameHeader { + return dst, fmt.Errorf("s2: requested skippable frame (%d) < 4", total) + } + if int64(total) >= maxBlockSize+skippableFrameHeader { + return dst, fmt.Errorf("s2: requested skippable frame (%d) >= max 1<<24", total) + } + // Chunk type 0xfe "Section 4.4 Padding (chunk type 0xfe)" + dst = append(dst, chunkTypePadding) + f := uint32(total - skippableFrameHeader) + // Add chunk length. + dst = append(dst, uint8(f), uint8(f>>8), uint8(f>>16)) + // Add data + start := len(dst) + dst = append(dst, make([]byte, f)...) + _, err := io.ReadFull(r, dst[start:]) + return dst, err +} + +// WriterOption is an option for creating a encoder. +type WriterOption func(*Writer) error + +// WriterConcurrency 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 WriterConcurrency(n int) WriterOption { + return func(w *Writer) error { + if n <= 0 { + return errors.New("concurrency must be at least 1") + } + w.concurrency = n + return nil + } +} + +// WriterBetterCompression will enable better compression. +// EncodeBetter compresses better than Encode but typically with a +// 10-40% speed decrease on both compression and decompression. +func WriterBetterCompression() WriterOption { + return func(w *Writer) error { + w.better = true + return nil + } +} + +// WriterBlockSize allows to override the default block size. +// Blocks will be this size or smaller. +// Minimum size is 4KB and and maximum size is 4MB. +// +// Bigger blocks may give bigger throughput on systems with many cores, +// and will increase compression slightly, but it will limit the possible +// concurrency for smaller payloads for both encoding and decoding. +// Default block size is 1MB. +func WriterBlockSize(n int) WriterOption { + return func(w *Writer) error { + if w.blockSize > maxBlockSize || w.blockSize < minBlockSize { + return errors.New("s2: block size too large. Must be <= 4MB and >=4KB") + } + w.blockSize = n + return nil + } +} + +// WriterPadding 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 <= 4MB. +// The padded area will be filled with data from crypto/rand.Reader. +// The padding will be applied whenever Close is called on the writer. +func WriterPadding(n int) WriterOption { + return func(w *Writer) error { + if n <= 0 { + return fmt.Errorf("s2: padding must be at least 1") + } + // No need to waste our time. + if n == 1 { + w.pad = 0 + } + if n > maxBlockSize { + return fmt.Errorf("s2: padding must less than 4MB") + } + w.pad = n + return nil + } +} diff --git a/github.com/klauspost/compress/s2/encode_all.go b/github.com/klauspost/compress/s2/encode_all.go new file mode 100644 index 0000000..f1d27b6 --- /dev/null +++ b/github.com/klauspost/compress/s2/encode_all.go @@ -0,0 +1,276 @@ +// Copyright 2016 The Snappy-Go Authors. All rights reserved. +// Copyright (c) 2019 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. + +package s2 + +import ( + "bytes" + "encoding/binary" + "math/bits" +) + +func load32(b []byte, i int) uint32 { + b = b[i:] + b = b[:4] + 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:] + 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 +} + +// 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 { + const prime6bytes = 227718039650203 + return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & 63)) +} + +func encodeGo(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))) + + if len(src) == 0 { + return dst[:d] + } + if len(src) < minNonLiteralBlockSize { + d += emitLiteral(dst[d:], src) + return dst[:d] + } + n := encodeBlockGo(dst[d:], src) + if n > 0 { + d += n + return dst[:d] + } + // Not compressible + d += emitLiteral(dst[d:], src) + return dst[:d] +} + +// encodeBlockGo 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 encodeBlockGo(dst, src []byte) (d int) { + // Initialize the hash table. + const ( + tableBits = 14 + maxTableSize = 1 << tableBits + + debug = false + ) + + var table [maxTableSize]uint32 + + // 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 + + // Bail if we can't compress to at least this. + dstLimit := len(src) - len(src)>>5 - 5 + + // 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 + cv := load64(src, s) + + // We search for a repeat at -1, but don't output repeats when nextEmit == 0 + repeat := 1 + + for { + candidate := 0 + for { + // Next src position to check + nextS := s + (s-nextEmit)>>6 + 4 + if nextS > sLimit { + goto emitRemainder + } + hash0 := hash6(cv, tableBits) + hash1 := hash6(cv>>8, tableBits) + candidate = int(table[hash0]) + candidate2 := int(table[hash1]) + table[hash0] = uint32(s) + table[hash1] = uint32(s + 1) + hash2 := hash6(cv>>16, tableBits) + + // Check repeat at offset checkRep. + const checkRep = 1 + if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) { + base := s + checkRep + // Extend back + for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; { + i-- + base-- + } + d += emitLiteral(dst[d:], src[nextEmit:base]) + + // Extend forward + candidate := s - repeat + 4 + checkRep + s += 4 + checkRep + for s <= sLimit { + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + if debug { + // Validate match. + if s <= candidate { + panic("s <= candidate") + } + a := src[base:s] + b := src[base-repeat : base-repeat+(s-base)] + if !bytes.Equal(a, b) { + panic("mismatch") + } + } + if nextEmit > 0 { + // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset. + d += emitRepeat(dst[d:], repeat, s-base) + } else { + // First match, cannot be repeat. + d += emitCopy(dst[d:], repeat, s-base) + } + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + cv = load64(src, s) + continue + } + + if uint32(cv) == load32(src, candidate) { + break + } + candidate = int(table[hash2]) + if uint32(cv>>8) == load32(src, candidate2) { + table[hash2] = uint32(s + 2) + candidate = candidate2 + s++ + break + } + table[hash2] = uint32(s + 2) + if uint32(cv>>16) == load32(src, candidate) { + s += 2 + break + } + + cv = load64(src, nextS) + s = nextS + } + + // Extend backwards. + // The top bytes will be rechecked to get the full match. + for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] { + candidate-- + s-- + } + + // Bail if we exceed the maximum size. + if d+(s-nextEmit) > dstLimit { + return 0 + } + + // 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 + repeat = base - candidate + + // Extend the 4-byte match as long as possible. + s += 4 + candidate += 4 + for s <= len(src)-8 { + if diff := load64(src, s) ^ load64(src, candidate); diff != 0 { + s += bits.TrailingZeros64(diff) >> 3 + break + } + s += 8 + candidate += 8 + } + + d += emitCopy(dst[d:], repeat, s-base) + if debug { + // Validate match. + if s <= candidate { + panic("s <= candidate") + } + a := src[base:s] + b := src[base-repeat : base-repeat+(s-base)] + if !bytes.Equal(a, b) { + panic("mismatch") + } + } + + nextEmit = s + if s >= sLimit { + goto emitRemainder + } + + if d > dstLimit { + // Do we have space for more, if not bail. + return 0 + } + // Check for an immediate match, otherwise start search at s+1 + x := load64(src, s-2) + m2Hash := hash6(x, tableBits) + currHash := hash6(x>>16, tableBits) + candidate = int(table[currHash]) + table[m2Hash] = uint32(s - 2) + table[currHash] = uint32(s) + if debug && s == candidate { + panic("s == candidate") + } + if uint32(x>>16) != load32(src, candidate) { + cv = load64(src, s+1) + s++ + break + } + } + } + +emitRemainder: + if nextEmit < len(src) { + // Bail if we exceed the maximum size. + if d+len(src)-nextEmit > dstLimit { + return 0 + } + d += emitLiteral(dst[d:], src[nextEmit:]) + } + return d +} diff --git a/github.com/klauspost/compress/s2/encode_amd64.go b/github.com/klauspost/compress/s2/encode_amd64.go new file mode 100644 index 0000000..253f84f --- /dev/null +++ b/github.com/klauspost/compress/s2/encode_amd64.go @@ -0,0 +1,68 @@ +// +build !appengine +// +build !noasm +// +build gc + +package s2 + +// 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)) &&