diff --git a/fiber.context.go b/fiber.context.go index 4c3b21c4..a2a0572b 100644 --- a/fiber.context.go +++ b/fiber.context.go @@ -8,6 +8,7 @@ import ( "github.com/getevo/evo/lib/log" "github.com/getevo/evo/lib/text" "github.com/gofiber/utils" + "github.com/klauspost/compress/snappy" "github.com/valyala/fasthttp" "io" "mime/multipart" @@ -479,6 +480,9 @@ func (r *Request) Write(body interface{}) { if r.BeforeWrite != nil { data = r.BeforeWrite(r, data) } + if r.Header("x-snappy") == "1" { + data = snappy.Encode(nil, data) + } r.Context.Context().Response.SetBody(data) } diff --git a/go.mod b/go.mod index ee85f2ca..b0588159 100644 --- a/go.mod +++ b/go.mod @@ -33,6 +33,7 @@ require ( github.com/gofiber/websocket v0.5.1 github.com/jackskj/carta v0.2.0 github.com/k0kubun/colorstring v0.0.0-20150214042306-9440f1994b88 // indirect + github.com/klauspost/compress v1.15.11 github.com/leodido/go-urn v1.2.1 github.com/lithammer/shortuuid v3.0.0+incompatible github.com/magefile/mage v1.11.0 diff --git a/go.sum b/go.sum index 4d7f7575..f95227d6 100644 --- a/go.sum +++ b/go.sum @@ -155,6 +155,8 @@ github.com/kisielk/errcheck v1.2.0/go.mod h1:/BMXB+zMLi60iA8Vv6Ksmxu/1UDYcXs4uQL github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck= github.com/klauspost/compress v1.10.7 h1:7rix8v8GpI3ZBb0nSozFRgbtXKv+hOe+qfEpZqybrAg= github.com/klauspost/compress v1.10.7/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs= +github.com/klauspost/compress v1.15.11 h1:Lcadnb3RKGin4FYM/orgq0qde+nc15E5Cbqg4B9Sx9c= +github.com/klauspost/compress v1.15.11/go.mod h1:QPwzmACJjUTFsnSHH934V6woptycfrDDJnH7hvFVbGM= github.com/kr/fs v0.1.0 h1:Jskdu9ieNAYnjxsi0LbQp1ulIKZV1LAFgK1tWhpZgl8= github.com/kr/fs v0.1.0/go.mod h1:FFnZGqtBN9Gxj7eW1uZ42v5BccTP0vu6NEaFoC2HwRg= github.com/kr/pretty v0.1.0 h1:L/CwN0zerZDmRFUapSPitk6f+Q3+0za1rQkzVuMiMFI= diff --git a/vendor/github.com/klauspost/compress/LICENSE b/vendor/github.com/klauspost/compress/LICENSE index 1eb75ef6..87d55747 100644 --- a/vendor/github.com/klauspost/compress/LICENSE +++ b/vendor/github.com/klauspost/compress/LICENSE @@ -26,3 +26,279 @@ DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +------------------ + +Files: gzhttp/* + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright 2016-2017 The New York Times Company + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. + +------------------ + +Files: s2/cmd/internal/readahead/* + +The MIT License (MIT) + +Copyright (c) 2015 Klaus Post + +Permission is hereby granted, free of charge, to any person obtaining a copy +of this software and associated documentation files (the "Software"), to deal +in the Software without restriction, including without limitation the rights +to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +copies of the Software, and to permit persons to whom the Software is +furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all +copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +SOFTWARE. + +--------------------- +Files: snappy/* +Files: internal/snapref/* + +Copyright (c) 2011 The Snappy-Go Authors. All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are +met: + + * Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following disclaimer +in the documentation and/or other materials provided with the +distribution. + * Neither the name of Google Inc. nor the names of its +contributors may be used to endorse or promote products derived from +this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +----------------- + +Files: s2/cmd/internal/filepathx/* + +Copyright 2016 The filepathx Authors + +Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: + +The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. diff --git a/vendor/github.com/klauspost/compress/flate/deflate.go b/vendor/github.com/klauspost/compress/flate/deflate.go index 25dbe3e1..07265dde 100644 --- a/vendor/github.com/klauspost/compress/flate/deflate.go +++ b/vendor/github.com/klauspost/compress/flate/deflate.go @@ -6,6 +6,7 @@ package flate import ( + "encoding/binary" "fmt" "io" "math" @@ -37,15 +38,17 @@ const ( 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 + // The maximum number of tokens we will encode at the time. + // Smaller sizes usually creates less optimal blocks. + // Bigger can make context switching slow. + // We use this for levels 7-9, so we make it big. + maxFlateBlockTokens = 1 << 15 maxStoreBlockSize = 65535 hashBits = 17 // After 17 performance degrades hashSize = 1 << hashBits hashMask = (1 << hashBits) - 1 hashShift = (hashBits + minMatchLength - 1) / minMatchLength - maxHashOffset = 1 << 24 + maxHashOffset = 1 << 28 skipNever = math.MaxInt32 @@ -70,9 +73,9 @@ var levels = []compressionLevel{ {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}, + {8, 12, 16, 24, skipNever, 7}, + {16, 30, 40, 64, skipNever, 8}, + {32, 258, 258, 1024, skipNever, 9}, } // advancedState contains state for the advanced levels, with bigger hash tables, etc. @@ -81,28 +84,29 @@ type advancedState struct { length int offset int maxInsertIndex int + chainHead int + hashOffset int + + ii uint16 // position of last match, intended to overflow to reset. + + // input window: unprocessed data is window[index:windowEnd] + index int + estBitsPerByte int + hashMatch [maxMatchLength + minMatchLength]uint32 // 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. + hashHead [hashSize]uint32 + hashPrev [windowSize]uint32 } type compressor struct { compressionLevel + h *huffmanEncoder w *huffmanBitWriter // compression algorithm @@ -127,7 +131,8 @@ 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]) + //copy(d.window[:], d.window[windowSize:2*windowSize]) + *(*[windowSize]byte)(d.window) = *(*[windowSize]byte)(d.window[windowSize:]) s.index -= windowSize d.windowEnd -= windowSize if d.blockStart >= windowSize { @@ -170,7 +175,8 @@ func (d *compressor) writeBlock(tok *tokens, index int, eof bool) error { window = d.window[d.blockStart:index] } d.blockStart = index - d.w.writeBlock(tok, eof, window) + //d.w.writeBlock(tok, eof, window) + d.w.writeBlockDynamic(tok, eof, window, d.sync) return d.w.err } return nil @@ -253,7 +259,6 @@ func (d *compressor) fillWindow(b []byte) { // 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 @@ -263,7 +268,7 @@ func (d *compressor) fillWindow(b []byte) { // 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) { +func (d *compressor) findMatch(pos int, prevHead int, lookahead int) (length, offset int, ok bool) { minMatchLook := maxMatchLength if lookahead < minMatchLook { minMatchLook = lookahead @@ -279,36 +284,75 @@ func (d *compressor) findMatch(pos int, prevHead int, prevLength int, lookahead // 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 - } + length = minMatchLength - 1 wEnd := win[pos+length] wPos := win[pos:] minIndex := pos - windowSize + if minIndex < 0 { + minIndex = 0 + } + offset = 0 + + cGain := 0 + if d.chain < 100 { + for i := prevHead; tries > 0; tries-- { + if wEnd == win[i+length] { + n := matchLen(win[i:i+minMatchLook], wPos) + if n > length { + 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 { + break + } + } + return + } + // Some like it higher (CSV), some like it lower (JSON) + const baseCost = 6 + // Base is 4 bytes at with an additional cost. + // Matches must be better than this. 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 + if n > length { + // Calculate gain. Estimate + newGain := d.h.bitLengthRaw(wPos[:n]) - int(offsetExtraBits[offsetCode(uint32(pos-i))]) - baseCost - int(lengthExtraBits[lengthCodes[(n-3)&255]]) + + //fmt.Println(n, "gain:", newGain, "prev:", cGain, "raw:", d.h.bitLengthRaw(wPos[:n])) + if newGain > cGain { + length = n + offset = pos - i + cGain = newGain + 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] } - wEnd = win[pos+n] } } - if i == minIndex { + 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 { + if i < minIndex { break } } @@ -327,8 +371,13 @@ func (d *compressor) writeStoredBlock(buf []byte) error { // 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) + return hash4u(binary.LittleEndian.Uint32(b), hashBits) +} + +// 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) } // bulkHash4 will compute hashes using the same @@ -337,11 +386,12 @@ 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 + hb := binary.LittleEndian.Uint32(b) + dst[0] = hash4u(hb, hashBits) end := len(b) - 4 + 1 for i := 1; i < end; i++ { - hb = (hb << 8) | uint32(b[i+3]) + hb = (hb >> 8) | uint32(b[i+3])<<24 dst[i] = hash4u(hb, hashBits) } } @@ -358,7 +408,6 @@ func (d *compressor) initDeflate() { s.hashOffset = 1 s.length = minMatchLength - 1 s.offset = 0 - s.hash = 0 s.chainHead = -1 } @@ -374,11 +423,19 @@ func (d *compressor) deflateLazy() { if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync { return } + if d.windowEnd != s.index && d.chain > 100 { + // Get literal huffman coder. + if d.h == nil { + d.h = newHuffmanEncoder(maxFlateBlockTokens) + } + var tmp [256]uint16 + for _, v := range d.window[s.index:d.windowEnd] { + tmp[v]++ + } + d.h.generate(tmp[:], 15) + } 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 { @@ -410,11 +467,11 @@ func (d *compressor) deflateLazy() { } if s.index < s.maxInsertIndex { // Update the hash - s.hash = hash4(d.window[s.index : s.index+minMatchLength]) - ch := s.hashHead[s.hash&hashMask] + hash := hash4(d.window[s.index:]) + ch := s.hashHead[hash] s.chainHead = int(ch) s.hashPrev[s.index&windowMask] = ch - s.hashHead[s.hash&hashMask] = uint32(s.index + s.hashOffset) + s.hashHead[hash] = uint32(s.index + s.hashOffset) } prevLength := s.length prevOffset := s.offset @@ -426,12 +483,37 @@ func (d *compressor) deflateLazy() { } 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 { + if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, lookahead); ok { s.length = newLength s.offset = newOffset } } + if prevLength >= minMatchLength && s.length <= prevLength { + // Check for better match at end... + // + // checkOff must be >=2 since we otherwise risk checking s.index + // Offset of 2 seems to yield best results. + const checkOff = 2 + prevIndex := s.index - 1 + if prevIndex+prevLength+checkOff < s.maxInsertIndex { + end := lookahead + if lookahead > maxMatchLength { + end = maxMatchLength + } + end += prevIndex + idx := prevIndex + prevLength - (4 - checkOff) + h := hash4(d.window[idx:]) + ch2 := int(s.hashHead[h]) - s.hashOffset - prevLength + (4 - checkOff) + if ch2 > minIndex { + length := matchLen(d.window[prevIndex:end], d.window[ch2:]) + // It seems like a pure length metric is best. + if length > prevLength { + prevLength = length + prevOffset = prevIndex - ch2 + } + } + } // 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)) @@ -440,8 +522,7 @@ func (d *compressor) deflateLazy() { // 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 + newIndex := s.index + prevLength - 1 // Calculate missing hashes end := newIndex if end > s.maxInsertIndex { @@ -467,7 +548,6 @@ func (d *compressor) deflateLazy() { // Set the head of the hash chain to us. s.hashHead[newH] = uint32(di + s.hashOffset) } - s.hash = newH } s.index = newIndex @@ -480,6 +560,7 @@ func (d *compressor) deflateLazy() { } d.tokens.Reset() } + s.ii = 0 } else { // Reset, if we got a match this run. if s.length >= minMatchLength { @@ -499,13 +580,12 @@ func (d *compressor) deflateLazy() { // 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) + if n := int(s.ii) - d.chain; n > 0 { + n = 1 + int(n>>6) 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 { @@ -513,6 +593,14 @@ func (d *compressor) deflateLazy() { } d.tokens.Reset() } + // Index... + if s.index < s.maxInsertIndex { + h := hash4(d.window[s.index:]) + ch := s.hashHead[h] + s.chainHead = int(ch) + s.hashPrev[s.index&windowMask] = ch + s.hashHead[h] = uint32(s.index + s.hashOffset) + } s.index++ } // Flush last byte @@ -612,7 +700,9 @@ func (d *compressor) write(b []byte) (n int, err error) { } n = len(b) for len(b) > 0 { - d.step(d) + if d.windowEnd == len(d.window) || d.sync { + d.step(d) + } b = b[d.fill(d, b):] if d.err != nil { return 0, d.err @@ -645,21 +735,21 @@ func (d *compressor) init(w io.Writer, level int) (err error) { d.fill = (*compressor).fillBlock d.step = (*compressor).store case level == ConstantCompression: - d.w.logNewTablePenalty = 4 - d.window = make([]byte, maxStoreBlockSize) + d.w.logNewTablePenalty = 10 + d.window = make([]byte, 32<<10) d.fill = (*compressor).fillBlock d.step = (*compressor).storeHuff case level == DefaultCompression: level = 5 fallthrough case level >= 1 && level <= 6: - d.w.logNewTablePenalty = 6 + d.w.logNewTablePenalty = 7 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.w.logNewTablePenalty = 8 d.state = &advancedState{} d.compressionLevel = levels[level] d.initDeflate() @@ -703,7 +793,6 @@ func (d *compressor) reset(w io.Writer) { d.tokens.Reset() s.length = minMatchLength - 1 s.offset = 0 - s.hash = 0 s.ii = 0 s.maxInsertIndex = 0 } diff --git a/vendor/github.com/klauspost/compress/flate/dict_decoder.go b/vendor/github.com/klauspost/compress/flate/dict_decoder.go index 71c75a06..bb36351a 100644 --- a/vendor/github.com/klauspost/compress/flate/dict_decoder.go +++ b/vendor/github.com/klauspost/compress/flate/dict_decoder.go @@ -7,19 +7,19 @@ 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. +// - 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. +// - 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 diff --git a/vendor/github.com/klauspost/compress/flate/fast_encoder.go b/vendor/github.com/klauspost/compress/flate/fast_encoder.go index 6d4c1e98..24caf5f7 100644 --- a/vendor/github.com/klauspost/compress/flate/fast_encoder.go +++ b/vendor/github.com/klauspost/compress/flate/fast_encoder.go @@ -6,6 +6,7 @@ package flate import ( + "encoding/binary" "fmt" "math/bits" ) @@ -44,7 +45,7 @@ const ( bTableBits = 17 // Bits used in the big tables bTableSize = 1 << bTableBits // Size of the table - allocHistory = maxStoreBlockSize * 10 // Size to preallocate for history. + allocHistory = maxStoreBlockSize * 5 // Size to preallocate for history. bufferReset = (1 << 31) - allocHistory - maxStoreBlockSize - 1 // Reset the buffer offset when reaching this. ) @@ -57,38 +58,12 @@ const ( 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 + return binary.LittleEndian.Uint32(b[i:]) } 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 + return binary.LittleEndian.Uint64(b[i:]) } type tableEntry struct { @@ -114,7 +89,8 @@ func (e *fastGen) addBlock(src []byte) int32 { } // Move down offset := int32(len(e.hist)) - maxMatchOffset - copy(e.hist[0:maxMatchOffset], e.hist[offset:]) + // copy(e.hist[0:maxMatchOffset], e.hist[offset:]) + *(*[maxMatchOffset]byte)(e.hist) = *(*[maxMatchOffset]byte)(e.hist[offset:]) e.cur += offset e.hist = e.hist[:maxMatchOffset] } @@ -124,39 +100,36 @@ func (e *fastGen) addBlock(src []byte) int32 { 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)) + return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & reg8SizeMask64)) } -// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits. -// Preferably h should be a constant and should always be <64. -func hash6(u uint64, h uint8) uint32 { - return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & 63)) +// hashLen returns a hash of the lowest mls bytes of with length output bits. +// mls must be >=3 and <=8. Any other value will return hash for 4 bytes. +// length should always be < 32. +// Preferably length and mls should be a constant for inlining. +func hashLen(u uint64, length, mls uint8) uint32 { + switch mls { + case 3: + return (uint32(u<<8) * prime3bytes) >> (32 - length) + case 5: + return uint32(((u << (64 - 40)) * prime5bytes) >> (64 - length)) + case 6: + return uint32(((u << (64 - 48)) * prime6bytes) >> (64 - length)) + case 7: + return uint32(((u << (64 - 56)) * prime7bytes) >> (64 - length)) + case 8: + return uint32((u * prime8bytes) >> (64 - length)) + default: + return (uint32(u) * prime4bytes) >> (32 - length) + } } // matchlen will return the match length between offsets and t in src. @@ -189,7 +162,7 @@ func (e *fastGen) matchlen(s, t int32, src []byte) int32 { // 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 debugDeflate { if t >= s { panic(fmt.Sprint("t >=s:", t, s)) } @@ -223,31 +196,20 @@ func (e *fastGen) Reset() { // 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:] + + for len(a) >= 8 { + if diff := binary.LittleEndian.Uint64(a) ^ binary.LittleEndian.Uint64(b); 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 i + checked } } return len(a) + checked diff --git a/vendor/github.com/klauspost/compress/flate/gen_inflate.go b/vendor/github.com/klauspost/compress/flate/gen_inflate.go deleted file mode 100644 index c74a95fe..00000000 --- a/vendor/github.com/klauspost/compress/flate/gen_inflate.go +++ /dev/null @@ -1,274 +0,0 @@ -// +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/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go index 53fe1d06..89a5dd89 100644 --- a/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go +++ b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go @@ -5,7 +5,10 @@ package flate import ( + "encoding/binary" + "fmt" "io" + "math" ) const ( @@ -22,11 +25,15 @@ const ( codegenCodeCount = 19 badCode = 255 + // maxPredefinedTokens is the maximum number of tokens + // where we check if fixed size is smaller. + maxPredefinedTokens = 250 + // 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 + bufferFlushSize = 246 // bufferSize is the actual output byte buffer size. // It must have additional headroom for a flush @@ -34,8 +41,11 @@ const ( bufferSize = bufferFlushSize + 8 ) +// Minimum length code that emits bits. +const lengthExtraBitsMinCode = 8 + // The number of extra bits needed by length code X - LENGTH_CODES_START. -var lengthExtraBits = [32]int8{ +var lengthExtraBits = [32]uint8{ /* 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, @@ -49,28 +59,41 @@ var lengthBase = [32]uint8{ 64, 80, 96, 112, 128, 160, 192, 224, 255, } +// Minimum offset code that emits bits. +const offsetExtraBitsMinCode = 4 + // offset code word extra bits. -var offsetExtraBits = [64]int8{ +var offsetExtraBits = [32]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, + 14, 14, } -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, +var offsetCombined = [32]uint32{} - /* extended window */ - 0x008000, 0x00c000, 0x010000, 0x018000, 0x020000, - 0x030000, 0x040000, 0x060000, 0x080000, 0x0c0000, - 0x100000, 0x180000, 0x200000, 0x300000, +func init() { + var offsetBase = [32]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, + } + + for i := range offsetCombined[:] { + // Don't use extended window values... + if offsetExtraBits[i] == 0 || offsetBase[i] > 0x006000 { + continue + } + offsetCombined[i] = uint32(offsetExtraBits[i]) | (offsetBase[i] << 8) + } } // The odd order in which the codegen code sizes are written. @@ -85,17 +108,18 @@ type huffmanBitWriter struct { // Data waiting to be written is bytes[0:nbytes] // and then the low nbits of bits. bits uint64 - nbits uint16 + nbits uint8 nbytes uint8 + lastHuffMan bool literalEncoding *huffmanEncoder + tmpLitEncoding *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 + bytes [256 + 8]byte literalFreq [lengthCodesStart + 32]uint16 offsetFreq [32]uint16 codegenFreq [codegenCodeCount]uint16 @@ -127,6 +151,7 @@ func newHuffmanBitWriter(w io.Writer) *huffmanBitWriter { return &huffmanBitWriter{ writer: w, literalEncoding: newHuffmanEncoder(literalCount), + tmpLitEncoding: newHuffmanEncoder(literalCount), codegenEncoding: newHuffmanEncoder(codegenCodeCount), offsetEncoding: newHuffmanEncoder(offsetCodeCount), } @@ -139,37 +164,33 @@ func (w *huffmanBitWriter) reset(writer io.Writer) { w.lastHuffMan = false } -func (w *huffmanBitWriter) canReuse(t *tokens) (offsets, lits bool) { - offsets, lits = true, true +func (w *huffmanBitWriter) canReuse(t *tokens) (ok bool) { a := t.offHist[:offsetCodeCount] - b := w.offsetFreq[:len(a)] - for i := range a { - if b[i] == 0 && a[i] != 0 { - offsets = false - break + b := w.offsetEncoding.codes + b = b[:len(a)] + for i, v := range a { + if v != 0 && b[i].zero() { + return false } } a = t.extraHist[:literalCount-256] - b = w.literalFreq[256:literalCount] + b = w.literalEncoding.codes[256:literalCount] b = b[:len(a)] - for i := range a { - if b[i] == 0 && a[i] != 0 { - lits = false - break + for i, v := range a { + if v != 0 && b[i].zero() { + return false } } - if lits { - a = t.litHist[:] - b = w.literalFreq[:len(a)] - for i := range a { - if b[i] == 0 && a[i] != 0 { - lits = false - break - } + + a = t.litHist[:256] + b = w.literalEncoding.codes[:len(a)] + for i, v := range a { + if v != 0 && b[i].zero() { + return false } } - return + return true } func (w *huffmanBitWriter) flush() { @@ -205,7 +226,7 @@ func (w *huffmanBitWriter) write(b []byte) { _, w.err = w.writer.Write(b) } -func (w *huffmanBitWriter) writeBits(b int32, nb uint16) { +func (w *huffmanBitWriter) writeBits(b int32, nb uint8) { w.bits |= uint64(b) << (w.nbits & 63) w.nbits += nb if w.nbits >= 48 { @@ -244,9 +265,9 @@ func (w *huffmanBitWriter) writeBytes(bytes []byte) { // 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 +// 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 @@ -259,12 +280,12 @@ func (w *huffmanBitWriter) generateCodegen(numLiterals int, numOffsets int, litE // 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[i] = litEnc.codes[i].len() } cgnl = codegen[numLiterals : numLiterals+numOffsets] for i := range cgnl { - cgnl[i] = uint8(offEnc.codes[i].len) + cgnl[i] = offEnc.codes[i].len() } codegen[numLiterals+numOffsets] = badCode @@ -407,8 +428,8 @@ func (w *huffmanBitWriter) storedSize(in []byte) (int, bool) { 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 + w.bits |= c.code64() << (w.nbits & 63) + w.nbits += c.len() if w.nbits >= 48 { w.writeOutBits() } @@ -420,13 +441,11 @@ func (w *huffmanBitWriter) writeOutBits() { 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) + + // We over-write, but faster... + binary.LittleEndian.PutUint64(w.bytes[n:], bits) n += 6 + if n >= bufferFlushSize { if w.err != nil { n = 0 @@ -435,14 +454,15 @@ func (w *huffmanBitWriter) writeOutBits() { 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 +// 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 @@ -457,7 +477,7 @@ func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, n w.writeBits(int32(numCodegens-4), 4) for i := 0; i < numCodegens; i++ { - value := uint(w.codegenEncoding.codes[codegenOrder[i]].len) + value := uint(w.codegenEncoding.codes[codegenOrder[i]].len()) w.writeBits(int32(value), 3) } @@ -551,7 +571,7 @@ func (w *huffmanBitWriter) writeBlock(tokens *tokens, eof bool, input []byte) { w.lastHeader = 0 } numLiterals, numOffsets := w.indexTokens(tokens, false) - w.generate(tokens) + w.generate() var extraBits int storedSize, storable := w.storedSize(input) if storable { @@ -562,7 +582,10 @@ func (w *huffmanBitWriter) writeBlock(tokens *tokens, eof bool, input []byte) { // Fixed Huffman baseline. var literalEncoding = fixedLiteralEncoding var offsetEncoding = fixedOffsetEncoding - var size = w.fixedSize(extraBits) + var size = math.MaxInt32 + if tokens.n < maxPredefinedTokens { + size = w.fixedSize(extraBits) + } // Dynamic Huffman? var numCodegens int @@ -580,7 +603,7 @@ func (w *huffmanBitWriter) writeBlock(tokens *tokens, eof bool, input []byte) { } // Stored bytes? - if storable && storedSize < size { + if storable && storedSize <= size { w.writeStoredHeader(len(input), eof) w.writeBytes(input) return @@ -619,22 +642,39 @@ func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []b w.lastHeader = 0 w.lastHuffMan = false } - if !sync { - tokens.Fill() + + // fillReuse enables filling of empty values. + // This will make encodings always reusable without testing. + // However, this does not appear to benefit on most cases. + const fillReuse = false + + // Check if we can reuse... + if !fillReuse && w.lastHeader > 0 && !w.canReuse(tokens) { + w.writeCode(w.literalEncoding.codes[endBlockMarker]) + w.lastHeader = 0 } + numLiterals, numOffsets := w.indexTokens(tokens, !sync) + extraBits := 0 + ssize, storable := w.storedSize(input) + + const usePrefs = true + if storable || w.lastHeader > 0 { + extraBits = w.extraBitSize() + } 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 + newSize += int(w.literalEncoding.codes[endBlockMarker].len()) + 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() + reuseSize := w.dynamicReuseSize(w.literalEncoding, w.offsetEncoding) + extraBits // Check if a new table is better. if newSize < reuseSize { @@ -645,35 +685,83 @@ func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []b } else { size = reuseSize } + + if tokens.n < maxPredefinedTokens { + if preSize := w.fixedSize(extraBits) + 7; usePrefs && preSize < size { + // Check if we get a reasonable size decrease. + if storable && ssize <= size { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + w.writeFixedHeader(eof) + if !sync { + tokens.AddEOB() + } + w.writeTokens(tokens.Slice(), fixedLiteralEncoding.codes, fixedOffsetEncoding.codes) + return + } + } // Check if we get a reasonable size decrease. - if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) { + if storable && ssize <= size { 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) + if fillReuse && !sync { + w.fillTokens() + numLiterals, numOffsets = maxNumLit, maxNumDist + } else { + w.literalFreq[endBlockMarker] = 1 + } + + w.generate() // 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) { + if fillReuse && !sync { + // Reindex for accurate size... + w.indexTokens(tokens, true) + } + size, numCodegens = w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits) + + // Store predefined, if we don't get a reasonable improvement. + if tokens.n < maxPredefinedTokens { + if preSize := w.fixedSize(extraBits); usePrefs && preSize <= size { + // Store bytes, if we don't get an improvement. + if storable && ssize <= preSize { + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + w.writeFixedHeader(eof) + if !sync { + tokens.AddEOB() + } + w.writeTokens(tokens.Slice(), fixedLiteralEncoding.codes, fixedOffsetEncoding.codes) + return + } + } + + if storable && ssize <= size { + // Store bytes, if we don't get an improvement. 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() + if !sync { + w.lastHeader, _ = w.headerSize() + } w.lastHuffMan = false } @@ -684,14 +772,29 @@ func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []b w.writeTokens(tokens.Slice(), w.literalEncoding.codes, w.offsetEncoding.codes) } +func (w *huffmanBitWriter) fillTokens() { + for i, v := range w.literalFreq[:literalCount] { + if v == 0 { + w.literalFreq[i] = 1 + } + } + for i, v := range w.offsetFreq[:offsetCodeCount] { + if v == 0 { + w.offsetFreq[i] = 1 + } + } +} + // 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]) + //copy(w.literalFreq[:], t.litHist[:]) + *(*[256]uint16)(w.literalFreq[:]) = t.litHist + //copy(w.literalFreq[256:], t.extraHist[:]) + *(*[32]uint16)(w.literalFreq[256:]) = t.extraHist + w.offsetFreq = t.offHist if t.n == 0 { return @@ -718,7 +821,7 @@ func (w *huffmanBitWriter) indexTokens(t *tokens, filled bool) (numLiterals, num return } -func (w *huffmanBitWriter) generate(t *tokens) { +func (w *huffmanBitWriter) generate() { w.literalEncoding.generate(w.literalFreq[:literalCount], 15) w.offsetEncoding.generate(w.offsetFreq[:offsetCodeCount], 15) } @@ -745,52 +848,135 @@ func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) offs := oeCodes[:32] lengths := leCodes[lengthCodesStart:] lengths = lengths[:32] + + // Go 1.16 LOVES having these on stack. + bits, nbits, nbytes := w.bits, w.nbits, w.nbytes + for _, t := range tokens { - if t < matchType { - w.writeCode(lits[t.literal()]) + if t < 256 { + //w.writeCode(lits[t.literal()]) + c := lits[t] + bits |= c.code64() << (nbits & 63) + nbits += c.len() + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } continue } // Write the length length := t.length() - lengthCode := lengthCode(length) + lengthCode := lengthCode(length) & 31 if false { - w.writeCode(lengths[lengthCode&31]) + w.writeCode(lengths[lengthCode]) } else { // inlined - c := lengths[lengthCode&31] - w.bits |= uint64(c.code) << (w.nbits & 63) - w.nbits += c.len - if w.nbits >= 48 { - w.writeOutBits() + c := lengths[lengthCode] + bits |= c.code64() << (nbits & 63) + nbits += c.len() + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } } } - extraLengthBits := uint16(lengthExtraBits[lengthCode&31]) - if extraLengthBits > 0 { - extraLength := int32(length - lengthBase[lengthCode&31]) - w.writeBits(extraLength, extraLengthBits) + if lengthCode >= lengthExtraBitsMinCode { + extraLengthBits := lengthExtraBits[lengthCode] + //w.writeBits(extraLength, extraLengthBits) + extraLength := int32(length - lengthBase[lengthCode]) + bits |= uint64(extraLength) << (nbits & 63) + nbits += extraLengthBits + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } } // Write the offset offset := t.offset() - offsetCode := offsetCode(offset) + offsetCode := (offset >> 16) & 31 if false { - w.writeCode(offs[offsetCode&31]) + w.writeCode(offs[offsetCode]) } else { // inlined - c := offs[offsetCode&31] - w.bits |= uint64(c.code) << (w.nbits & 63) - w.nbits += c.len - if w.nbits >= 48 { - w.writeOutBits() + c := offs[offsetCode] + bits |= c.code64() << (nbits & 63) + nbits += c.len() + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } } } - extraOffsetBits := uint16(offsetExtraBits[offsetCode&63]) - if extraOffsetBits > 0 { - extraOffset := int32(offset - offsetBase[offsetCode&63]) - w.writeBits(extraOffset, extraOffsetBits) + + if offsetCode >= offsetExtraBitsMinCode { + offsetComb := offsetCombined[offsetCode] + //w.writeBits(extraOffset, extraOffsetBits) + bits |= uint64((offset-(offsetComb>>8))&matchOffsetOnlyMask) << (nbits & 63) + nbits += uint8(offsetComb) + if nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + } } } + // Restore... + w.bits, w.nbits, w.nbytes = bits, nbits, nbytes + if deferEOB { w.writeCode(leCodes[endBlockMarker]) } @@ -825,43 +1011,78 @@ func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) { } } + const numLiterals = endBlockMarker + 1 + const numOffsets = 1 + // 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 + histogram(input, w.literalFreq[:numLiterals]) + ssize, storable := w.storedSize(input) + if storable && len(input) > 1024 { + // Quick check for incompressible content. + abs := float64(0) + avg := float64(len(input)) / 256 + max := float64(len(input) * 2) + for _, v := range w.literalFreq[:256] { + diff := float64(v) - avg + abs += diff * diff + if abs > max { + break + } + } + if abs < max { + if debugDeflate { + fmt.Println("stored", abs, "<", max) + } + // No chance we can compress this... + w.writeStoredHeader(len(input), eof) + w.writeBytes(input) + return + } + } + w.literalFreq[endBlockMarker] = 1 + w.tmpLitEncoding.generate(w.literalFreq[:numLiterals], 15) + estBits := w.tmpLitEncoding.canReuseBits(w.literalFreq[:numLiterals]) + if estBits < math.MaxInt32 { + estBits += w.lastHeader + if w.lastHeader == 0 { + estBits += guessHeaderSizeBits + } + estBits += estBits >> w.logNewTablePenalty } - estBits += estBits >> w.logNewTablePenalty // Store bytes, if we don't get a reasonable improvement. - ssize, storable := w.storedSize(input) - if storable && ssize < estBits { + if storable && ssize <= estBits { + if debugDeflate { + fmt.Println("stored,", ssize, "<=", estBits) + } w.writeStoredHeader(len(input), eof) w.writeBytes(input) return } if w.lastHeader > 0 { - reuseSize := w.literalEncoding.bitLength(w.literalFreq[:256]) - estBits += estExtra + reuseSize := w.literalEncoding.canReuseBits(w.literalFreq[:256]) if estBits < reuseSize { + if debugDeflate { + fmt.Println("NOT reusing, reuse:", reuseSize/8, "> new:", estBits/8, "header est:", w.lastHeader/8, "bytes") + } // We owe an EOB w.writeCode(w.literalEncoding.codes[endBlockMarker]) w.lastHeader = 0 + } else if debugDeflate { + fmt.Println("reusing, reuse:", reuseSize/8, "> new:", estBits/8, "- header est:", w.lastHeader/8) } } - const numLiterals = endBlockMarker + 1 - const numOffsets = 1 + count := 0 if w.lastHeader == 0 { - w.literalFreq[endBlockMarker] = 1 - w.literalEncoding.generate(w.literalFreq[:numLiterals], 15) - + // Use the temp encoding, so swap. + w.literalEncoding, w.tmpLitEncoding = w.tmpLitEncoding, w.literalEncoding // Generate codegen and codegenFrequencies, which indicates how to encode // the literalEncoding and the offsetEncoding. w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, huffOffset) @@ -872,39 +1093,94 @@ func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) { w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof) w.lastHuffMan = true w.lastHeader, _ = w.headerSize() + if debugDeflate { + count += w.lastHeader + fmt.Println("header:", count/8) + } + } + + encoding := w.literalEncoding.codes[:256] + // Go 1.16 LOVES having these on stack. At least 1.5x the speed. + bits, nbits, nbytes := w.bits, w.nbits, w.nbytes + + if debugDeflate { + count -= int(nbytes)*8 + int(nbits) + } + // Unroll, write 3 codes/loop. + // Fastest number of unrolls. + for len(input) > 3 { + // We must have at least 48 bits free. + if nbits >= 8 { + n := nbits >> 3 + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + bits >>= (n * 8) & 63 + nbits -= n * 8 + nbytes += n + } + if nbytes >= bufferFlushSize { + if w.err != nil { + nbytes = 0 + return + } + if debugDeflate { + count += int(nbytes) * 8 + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 + } + a, b := encoding[input[0]], encoding[input[1]] + bits |= a.code64() << (nbits & 63) + bits |= b.code64() << ((nbits + a.len()) & 63) + c := encoding[input[2]] + nbits += b.len() + a.len() + bits |= c.code64() << (nbits & 63) + nbits += c.len() + input = input[3:] } - encoding := w.literalEncoding.codes[:257] + // Remaining... 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 nbits >= 48 { + binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits) + //*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits + bits >>= 48 + nbits -= 48 + nbytes += 6 + if nbytes >= bufferFlushSize { if w.err != nil { - n = 0 + nbytes = 0 return } - w.write(w.bytes[:n]) - n = 0 + if debugDeflate { + count += int(nbytes) * 8 + } + _, w.err = w.writer.Write(w.bytes[:nbytes]) + nbytes = 0 } - w.nbytes = n + } + // Bitwriting inlined, ~30% speedup + c := encoding[t] + bits |= c.code64() << (nbits & 63) + + nbits += c.len() + if debugDeflate { + count += int(c.len()) } } + // Restore... + w.bits, w.nbits, w.nbytes = bits, nbits, nbytes + + if debugDeflate { + nb := count + int(nbytes)*8 + int(nbits) + fmt.Println("wrote", nb, "bits,", nb/8, "bytes.") + } + // Flush if needed to have space. + if w.nbits >= 48 { + w.writeOutBits() + } + if eof || sync { - w.writeCode(encoding[endBlockMarker]) + w.writeCode(w.literalEncoding.codes[endBlockMarker]) w.lastHeader = 0 w.lastHuffMan = false } diff --git a/vendor/github.com/klauspost/compress/flate/huffman_code.go b/vendor/github.com/klauspost/compress/flate/huffman_code.go index 4c39a301..be7b58b4 100644 --- a/vendor/github.com/klauspost/compress/flate/huffman_code.go +++ b/vendor/github.com/klauspost/compress/flate/huffman_code.go @@ -16,14 +16,28 @@ const ( ) // hcode is a huffman code with a bit code and bit length. -type hcode struct { - code, len uint16 +type hcode uint32 + +func (h hcode) len() uint8 { + return uint8(h) +} + +func (h hcode) code64() uint64 { + return uint64(h >> 8) +} + +func (h hcode) zero() bool { + return h == 0 } type huffmanEncoder struct { - codes []hcode - freqcache []literalNode - bitCount [17]int32 + codes []hcode + bitCount [17]int32 + + // 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. + freqcache [literalCount + 1]literalNode } type literalNode struct { @@ -52,9 +66,12 @@ type levelInfo struct { } // set sets the code and length of an hcode. -func (h *hcode) set(code uint16, length uint16) { - h.len = length - h.code = code +func (h *hcode) set(code uint16, length uint8) { + *h = hcode(length) | (hcode(code) << 8) +} + +func newhcode(code uint16, length uint8) hcode { + return hcode(length) | (hcode(code) << 8) } func reverseBits(number uint16, bitLength byte) uint16 { @@ -76,7 +93,7 @@ func generateFixedLiteralEncoding() *huffmanEncoder { var ch uint16 for ch = 0; ch < literalCount; ch++ { var bits uint16 - var size uint16 + var size uint8 switch { case ch < 144: // size 8, 000110000 .. 10111111 @@ -95,7 +112,7 @@ func generateFixedLiteralEncoding() *huffmanEncoder { bits = ch + 192 - 280 size = 8 } - codes[ch] = hcode{code: reverseBits(bits, byte(size)), len: size} + codes[ch] = newhcode(reverseBits(bits, size), size) } return h } @@ -104,7 +121,7 @@ func generateFixedOffsetEncoding() *huffmanEncoder { h := newHuffmanEncoder(30) codes := h.codes for ch := range codes { - codes[ch] = hcode{code: reverseBits(uint16(ch), 5), len: 5} + codes[ch] = newhcode(reverseBits(uint16(ch), 5), 5) } return h } @@ -116,7 +133,30 @@ func (h *huffmanEncoder) bitLength(freq []uint16) int { var total int for i, f := range freq { if f != 0 { - total += int(f) * int(h.codes[i].len) + total += int(f) * int(h.codes[i].len()) + } + } + return total +} + +func (h *huffmanEncoder) bitLengthRaw(b []byte) int { + var total int + for _, f := range b { + total += int(h.codes[f].len()) + } + return total +} + +// canReuseBits returns the number of bits or math.MaxInt32 if the encoder cannot be reused. +func (h *huffmanEncoder) canReuseBits(freq []uint16) int { + var total int + for i, f := range freq { + if f != 0 { + code := h.codes[i] + if code.zero() { + return math.MaxInt32 + } + total += int(f) * int(code.len()) } } return total @@ -128,13 +168,18 @@ func (h *huffmanEncoder) bitLength(freq []uint16) int { // 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 +// +// 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. +// +// 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. +// +// that should be encoded in i bits. func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 { if maxBits >= maxBitsLimit { panic("flate: maxBits too large") @@ -160,14 +205,19 @@ func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 { // of the level j ancestor. var leafCounts [maxBitsLimit][maxBitsLimit]int32 + // Descending to only have 1 bounds check. + l2f := int32(list[2].freq) + l1f := int32(list[1].freq) + l0f := int32(list[0].freq) + int32(list[1].freq) + 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), + lastFreq: l1f, + nextCharFreq: l2f, + nextPairFreq: l0f, } leafCounts[level][level] = 2 if level == 1 { @@ -178,8 +228,8 @@ func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 { // 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 { + level := uint32(maxBits) + for level < 16 { l := &levels[level] if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 { // We've run out of both leafs and pairs. @@ -211,7 +261,13 @@ func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 { // 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]) + if true { + save := leafCounts[level][level] + leafCounts[level] = leafCounts[level-1] + leafCounts[level][level] = save + } else { + copy(leafCounts[level][:level], leafCounts[level-1][:level]) + } levels[l.level-1].needed = 2 } @@ -269,7 +325,7 @@ func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalN sortByLiteral(chunk) for _, node := range chunk { - h.codes[node.literal] = hcode{code: reverseBits(code, uint8(n)), len: uint16(n)} + h.codes[node.literal] = newhcode(reverseBits(code, uint8(n)), uint8(n)) code++ } list = list[0 : len(list)-int(bits)] @@ -281,13 +337,8 @@ func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalN // 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] + codes := h.codes[:len(freq)] // Number of non-zero literals count := 0 // Set list to be the set of all non-zero literals and their frequencies @@ -296,11 +347,10 @@ func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) { list[count] = literalNode{uint16(i), f} count++ } else { - list[count] = literalNode{} - h.codes[i].len = 0 + codes[i] = 0 } } - list[len(freq)] = literalNode{} + list[count] = literalNode{} list = list[:count] if count <= 2 { @@ -320,44 +370,48 @@ func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) { h.assignEncodingAndSize(bitCount, list) } +// atLeastOne clamps the result between 1 and 15. func atLeastOne(v float32) float32 { if v < 1 { return 1 } + if v > 15 { + return 15 + } 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 - } - } +func histogram(b []byte, h []uint16) { + if true && len(b) >= 8<<10 { + // Split for bigger inputs + histogramSplit(b, h) } else { - for _, v := range h[:] { - if v > 0 { - n := float32(v) - shannon += atLeastOne(-mFastLog2(n*invTotal)) * n - } + h = h[:256] + for _, t := range b { + h[t]++ } } +} - return int(shannon + 0.99), int(extra + 0.99) +func histogramSplit(b []byte, h []uint16) { + // Tested, and slightly faster than 2-way. + // Writing to separate arrays and combining is also slightly slower. + h = h[:256] + for len(b)&3 != 0 { + h[b[0]]++ + b = b[1:] + } + n := len(b) / 4 + x, y, z, w := b[:n], b[n:], b[n+n:], b[n+n+n:] + y, z, w = y[:len(x)], z[:len(x)], w[:len(x)] + for i, t := range x { + v0 := &h[t] + v1 := &h[y[i]] + v3 := &h[w[i]] + v2 := &h[z[i]] + *v0++ + *v1++ + *v2++ + *v3++ + } } diff --git a/vendor/github.com/klauspost/compress/flate/inflate.go b/vendor/github.com/klauspost/compress/flate/inflate.go index 3e4259f1..414c0bea 100644 --- a/vendor/github.com/klauspost/compress/flate/inflate.go +++ b/vendor/github.com/klauspost/compress/flate/inflate.go @@ -9,10 +9,10 @@ package flate import ( "bufio" + "compress/flate" "fmt" "io" "math/bits" - "strconv" "sync" ) @@ -29,16 +29,26 @@ const ( debugDecode = false ) +// Value of length - 3 and extra bits. +type lengthExtra struct { + length, extra uint8 +} + +var decCodeToLen = [32]lengthExtra{{length: 0x0, extra: 0x0}, {length: 0x1, extra: 0x0}, {length: 0x2, extra: 0x0}, {length: 0x3, extra: 0x0}, {length: 0x4, extra: 0x0}, {length: 0x5, extra: 0x0}, {length: 0x6, extra: 0x0}, {length: 0x7, extra: 0x0}, {length: 0x8, extra: 0x1}, {length: 0xa, extra: 0x1}, {length: 0xc, extra: 0x1}, {length: 0xe, extra: 0x1}, {length: 0x10, extra: 0x2}, {length: 0x14, extra: 0x2}, {length: 0x18, extra: 0x2}, {length: 0x1c, extra: 0x2}, {length: 0x20, extra: 0x3}, {length: 0x28, extra: 0x3}, {length: 0x30, extra: 0x3}, {length: 0x38, extra: 0x3}, {length: 0x40, extra: 0x4}, {length: 0x50, extra: 0x4}, {length: 0x60, extra: 0x4}, {length: 0x70, extra: 0x4}, {length: 0x80, extra: 0x5}, {length: 0xa0, extra: 0x5}, {length: 0xc0, extra: 0x5}, {length: 0xe0, extra: 0x5}, {length: 0xff, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}} + +var bitMask32 = [32]uint32{ + 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF, + 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, + 0x1ffff, 0x3ffff, 0x7FFFF, 0xfFFFF, 0x1fFFFF, 0x3fFFFF, 0x7fFFFF, 0xffFFFF, + 0x1ffFFFF, 0x3ffFFFF, 0x7ffFFFF, 0xfffFFFF, 0x1fffFFFF, 0x3fffFFFF, 0x7fffFFFF, +} // up to 32 bits + // 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) -} +type CorruptInputError = flate.CorruptInputError // An InternalError reports an error in the flate code itself. type InternalError string @@ -48,26 +58,12 @@ func (e InternalError) Error() string { return "flate: internal error: " + strin // 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() -} +type ReadError = flate.ReadError // 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() -} +type WriteError = flate.WriteError // Resetter resets a ReadCloser returned by NewReader or NewReaderDict to // to switch to a new underlying Reader. This permits reusing a ReadCloser @@ -339,11 +335,17 @@ func (f *decompressor) nextBlock() { switch typ { case 0: f.dataBlock() + if debugDecode { + fmt.Println("stored block") + } case 1: // compressed, fixed Huffman tables f.hl = &fixedHuffmanDecoder f.hd = nil f.huffmanBlockDecoder()() + if debugDecode { + fmt.Println("predefinied huffman block") + } case 2: // compressed, dynamic Huffman tables if f.err = f.readHuffman(); f.err != nil { @@ -352,6 +354,9 @@ func (f *decompressor) nextBlock() { f.hl = &f.h1 f.hd = &f.h2 f.huffmanBlockDecoder()() + if debugDecode { + fmt.Println("dynamic huffman block") + } default: // 3 is reserved. if debugDecode { @@ -522,8 +527,8 @@ func (f *decompressor) readHuffman() error { return err } } - rep += int(f.b & uint32(1<>= nb + rep += int(f.b & uint32(1<<(nb®SizeMaskUint32)-1)) + f.b >>= nb & regSizeMaskUint32 f.nb -= nb if i+rep > n { if debugDecode { @@ -561,219 +566,6 @@ func (f *decompressor) readHuffman() error { 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. @@ -869,7 +661,7 @@ func (f *decompressor) moreBits() error { return noEOF(err) } f.roffset++ - f.b |= uint32(c) << f.nb + f.b |= uint32(c) << (f.nb & regSizeMaskUint32) f.nb += 8 return nil } @@ -894,7 +686,7 @@ func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { return 0, noEOF(err) } f.roffset++ - b |= uint32(c) << (nb & 31) + b |= uint32(c) << (nb & regSizeMaskUint32) nb += 8 } chunk := h.chunks[b&(huffmanNumChunks-1)] @@ -913,7 +705,7 @@ func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) { f.err = CorruptInputError(f.roffset) return 0, f.err } - f.b = b >> (n & 31) + f.b = b >> (n & regSizeMaskUint32) f.nb = nb - n return int(chunk >> huffmanValueShift), nil } diff --git a/vendor/github.com/klauspost/compress/flate/inflate_gen.go b/vendor/github.com/klauspost/compress/flate/inflate_gen.go index 397dc1b1..61342b6b 100644 --- a/vendor/github.com/klauspost/compress/flate/inflate_gen.go +++ b/vendor/github.com/klauspost/compress/flate/inflate_gen.go @@ -20,16 +20,11 @@ func (f *decompressor) huffmanBytesBuffer() { 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 - } + + // 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. + fnb, fb, dict := f.nb, f.b, &f.dict switch f.stepState { case stateInit: @@ -49,128 +44,150 @@ readLiteral: // 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 { + for fnb < n { c, err := fr.ReadByte() if err != nil { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb f.err = noEOF(err) return } f.roffset++ - b |= uint32(c) << (nb & 31) - nb += 8 + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] n = uint(chunk & huffmanCountMask) if n > huffmanChunkBits { - chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] n = uint(chunk & huffmanCountMask) } - if n <= nb { + if n <= fnb { if n == 0 { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("huffsym: n==0") } f.err = CorruptInputError(f.roffset) return } - f.b = b >> (n & 31) - f.nb = nb - n + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - 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() + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanBytesBuffer f.stepState = stateInit + f.b, f.nb = fb, fnb return } goto readLiteral case v == 256: + f.b, f.nb = fb, fnb 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 { + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits n>0:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") } - length += int(f.b & uint32(1<>= n - f.nb -= n + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return } - var dist int + var dist uint32 if f.hd == nil { - for f.nb < 5 { - if err = moreBits(); err != nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb<5:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) - f.b >>= 5 - f.nb -= 5 + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 } else { - if dist, err = f.huffSym(f.hd); err != nil { - if debugDecode { - fmt.Println("huffsym:", err) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break } - f.err = err - return } } @@ -180,21 +197,28 @@ readLiteral: 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 { + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb>= nb - f.nb -= nb - dist = 1<<(nb+1) + 1 + extra + extra |= fb & bitMask32[nb] + fb >>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra default: + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("dist too big:", dist, maxNumDist) } @@ -203,35 +227,38 @@ readLiteral: } // No check on length; encoding can be prescient. - if dist > f.dict.histSize() { + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb if debugDecode { - fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) } f.err = CorruptInputError(f.roffset) return } - f.copyLen, f.copyDist = length, dist + f.copyLen, f.copyDist = length, int(dist) goto copyHistory } copyHistory: // Perform a backwards copy according to RFC section 3.2.3. { - cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) if cnt == 0 { - cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + cnt = dict.writeCopy(f.copyDist, f.copyLen) } f.copyLen -= cnt - if f.dict.availWrite() == 0 || f.copyLen > 0 { - f.toRead = f.dict.readFlush() + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanBytesBuffer // We need to continue this work f.stepState = stateDict + f.b, f.nb = fb, fnb return } goto readLiteral } + // Not reached } // Decode a single Huffman block from f. @@ -244,16 +271,11 @@ func (f *decompressor) huffmanBytesReader() { 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 - } + + // 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. + fnb, fb, dict := f.nb, f.b, &f.dict switch f.stepState { case stateInit: @@ -273,128 +295,150 @@ readLiteral: // 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 { + for fnb < n { c, err := fr.ReadByte() if err != nil { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb f.err = noEOF(err) return } f.roffset++ - b |= uint32(c) << (nb & 31) - nb += 8 + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] n = uint(chunk & huffmanCountMask) if n > huffmanChunkBits { - chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] n = uint(chunk & huffmanCountMask) } - if n <= nb { + if n <= fnb { if n == 0 { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("huffsym: n==0") } f.err = CorruptInputError(f.roffset) return } - f.b = b >> (n & 31) - f.nb = nb - n + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - 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() + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanBytesReader f.stepState = stateInit + f.b, f.nb = fb, fnb return } goto readLiteral case v == 256: + f.b, f.nb = fb, fnb 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 { + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits n>0:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") } - length += int(f.b & uint32(1<>= n - f.nb -= n + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return } - var dist int + var dist uint32 if f.hd == nil { - for f.nb < 5 { - if err = moreBits(); err != nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb<5:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) - f.b >>= 5 - f.nb -= 5 + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 } else { - if dist, err = f.huffSym(f.hd); err != nil { - if debugDecode { - fmt.Println("huffsym:", err) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break } - f.err = err - return } } @@ -404,21 +448,28 @@ readLiteral: 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 { + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb>= nb - f.nb -= nb - dist = 1<<(nb+1) + 1 + extra + extra |= fb & bitMask32[nb] + fb >>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra default: + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("dist too big:", dist, maxNumDist) } @@ -427,35 +478,38 @@ readLiteral: } // No check on length; encoding can be prescient. - if dist > f.dict.histSize() { + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb if debugDecode { - fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) } f.err = CorruptInputError(f.roffset) return } - f.copyLen, f.copyDist = length, dist + f.copyLen, f.copyDist = length, int(dist) goto copyHistory } copyHistory: // Perform a backwards copy according to RFC section 3.2.3. { - cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) if cnt == 0 { - cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + cnt = dict.writeCopy(f.copyDist, f.copyLen) } f.copyLen -= cnt - if f.dict.availWrite() == 0 || f.copyLen > 0 { - f.toRead = f.dict.readFlush() + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanBytesReader // We need to continue this work f.stepState = stateDict + f.b, f.nb = fb, fnb return } goto readLiteral } + // Not reached } // Decode a single Huffman block from f. @@ -468,16 +522,11 @@ func (f *decompressor) huffmanBufioReader() { 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 - } + + // 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. + fnb, fb, dict := f.nb, f.b, &f.dict switch f.stepState { case stateInit: @@ -497,128 +546,150 @@ readLiteral: // 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 { + for fnb < n { c, err := fr.ReadByte() if err != nil { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb f.err = noEOF(err) return } f.roffset++ - b |= uint32(c) << (nb & 31) - nb += 8 + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] n = uint(chunk & huffmanCountMask) if n > huffmanChunkBits { - chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] n = uint(chunk & huffmanCountMask) } - if n <= nb { + if n <= fnb { if n == 0 { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("huffsym: n==0") } f.err = CorruptInputError(f.roffset) return } - f.b = b >> (n & 31) - f.nb = nb - n + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - 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() + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanBufioReader f.stepState = stateInit + f.b, f.nb = fb, fnb return } goto readLiteral case v == 256: + f.b, f.nb = fb, fnb 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 { + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits n>0:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - length += int(f.b & uint32(1<>= n - f.nb -= n + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") + } + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return } - var dist int + var dist uint32 if f.hd == nil { - for f.nb < 5 { - if err = moreBits(); err != nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb<5:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) - f.b >>= 5 - f.nb -= 5 + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 } else { - if dist, err = f.huffSym(f.hd); err != nil { - if debugDecode { - fmt.Println("huffsym:", err) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break } - f.err = err - return } } @@ -628,21 +699,28 @@ readLiteral: 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 { + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb>= nb - f.nb -= nb - dist = 1<<(nb+1) + 1 + extra + extra |= fb & bitMask32[nb] + fb >>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra default: + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("dist too big:", dist, maxNumDist) } @@ -651,35 +729,38 @@ readLiteral: } // No check on length; encoding can be prescient. - if dist > f.dict.histSize() { + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb if debugDecode { - fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) } f.err = CorruptInputError(f.roffset) return } - f.copyLen, f.copyDist = length, dist + f.copyLen, f.copyDist = length, int(dist) goto copyHistory } copyHistory: // Perform a backwards copy according to RFC section 3.2.3. { - cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) if cnt == 0 { - cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + cnt = dict.writeCopy(f.copyDist, f.copyLen) } f.copyLen -= cnt - if f.dict.availWrite() == 0 || f.copyLen > 0 { - f.toRead = f.dict.readFlush() + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanBufioReader // We need to continue this work f.stepState = stateDict + f.b, f.nb = fb, fnb return } goto readLiteral } + // Not reached } // Decode a single Huffman block from f. @@ -692,16 +773,11 @@ func (f *decompressor) huffmanStringsReader() { 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 - } + + // 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. + fnb, fb, dict := f.nb, f.b, &f.dict switch f.stepState { case stateInit: @@ -721,128 +797,401 @@ readLiteral: // 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 { + for fnb < n { c, err := fr.ReadByte() if err != nil { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb f.err = noEOF(err) return } f.roffset++ - b |= uint32(c) << (nb & 31) - nb += 8 + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - chunk := f.hl.chunks[b&(huffmanNumChunks-1)] + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] n = uint(chunk & huffmanCountMask) if n > huffmanChunkBits { - chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask] + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] n = uint(chunk & huffmanCountMask) } - if n <= nb { + if n <= fnb { if n == 0 { - f.b = b - f.nb = nb + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("huffsym: n==0") } f.err = CorruptInputError(f.roffset) return } - f.b = b >> (n & 31) - f.nb = nb - n + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - 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() + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() f.step = (*decompressor).huffmanStringsReader f.stepState = stateInit + f.b, f.nb = fb, fnb return } goto readLiteral case v == 256: + f.b, f.nb = fb, fnb 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 + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits n>0:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n default: if debugDecode { fmt.Println(v, ">= maxNumLit") } f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb return } - if n > 0 { - for f.nb < n { - if err = moreBits(); err != nil { + + var dist uint32 + if f.hd == nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb<5:", err) + } + f.err = err + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 + } else { + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break + } + } + } + + switch { + case dist < 4: + dist++ + case dist < maxNumDist: + nb := uint(dist-2) >> 1 + // have 1 bit in bottom of dist, need nb more. + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("morebits f.nb>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra + default: + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist too big:", dist, maxNumDist) + } + f.err = CorruptInputError(f.roffset) + return + } + + // No check on length; encoding can be prescient. + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) + } + f.err = CorruptInputError(f.roffset) + return + } + + f.copyLen, f.copyDist = length, int(dist) + goto copyHistory + } + +copyHistory: + // Perform a backwards copy according to RFC section 3.2.3. + { + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) + if cnt == 0 { + cnt = dict.writeCopy(f.copyDist, f.copyLen) + } + f.copyLen -= cnt + + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanStringsReader // We need to continue this work + f.stepState = stateDict + f.b, f.nb = fb, fnb + return + } + goto readLiteral + } + // Not reached +} + +// 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) huffmanGenericReader() { + const ( + stateInit = iota // Zero value must be stateInit + stateDict + ) + fr := f.r.(Reader) + + // 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. + fnb, fb, dict := f.nb, f.b, &f.dict + + 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) + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hl.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hl.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hl.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + v = int(chunk >> huffmanValueShift) + break + } + } + } + + var length int + switch { + case v < 256: + dict.writeByte(byte(v)) + if dict.availWrite() == 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanGenericReader + f.stepState = stateInit + f.b, f.nb = fb, fnb + return + } + goto readLiteral + case v == 256: + f.b, f.nb = fb, fnb + f.finishBlock() + return + // otherwise, reference to older data + case v < 265: + length = v - (257 - 3) + case v < maxNumLit: + val := decCodeToLen[(v - 257)] + length = int(val.length) + 3 + n := uint(val.extra) + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits n>0:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + length += int(fb & bitMask32[n]) + fb >>= n & regSizeMaskUint32 + fnb -= n + default: + if debugDecode { + fmt.Println(v, ">= maxNumLit") } - length += int(f.b & uint32(1<>= n - f.nb -= n + f.err = CorruptInputError(f.roffset) + f.b, f.nb = fb, fnb + return } - var dist int + var dist uint32 if f.hd == nil { - for f.nb < 5 { - if err = moreBits(); err != nil { + for fnb < 5 { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb<5:", err) } f.err = err return } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 } - dist = int(bits.Reverse8(uint8(f.b & 0x1F << 3))) - f.b >>= 5 - f.nb -= 5 + dist = uint32(bits.Reverse8(uint8(fb & 0x1F << 3))) + fb >>= 5 + fnb -= 5 } else { - if dist, err = f.huffSym(f.hd); err != nil { - if debugDecode { - fmt.Println("huffsym:", err) + // Since a huffmanDecoder can be empty or be composed of a degenerate tree + // with single element, huffSym must error on these two edge cases. In both + // cases, the chunks slice will be 0 for the invalid sequence, leading it + // satisfy the n == 0 check below. + n := uint(f.hd.maxRead) + // Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers, + // but is smart enough to keep local variables in registers, so use nb and b, + // inline call to moreBits and reassign b,nb back to f on return. + for { + for fnb < n { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb + f.err = noEOF(err) + return + } + f.roffset++ + fb |= uint32(c) << (fnb & regSizeMaskUint32) + fnb += 8 + } + chunk := f.hd.chunks[fb&(huffmanNumChunks-1)] + n = uint(chunk & huffmanCountMask) + if n > huffmanChunkBits { + chunk = f.hd.links[chunk>>huffmanValueShift][(fb>>huffmanChunkBits)&f.hd.linkMask] + n = uint(chunk & huffmanCountMask) + } + if n <= fnb { + if n == 0 { + f.b, f.nb = fb, fnb + if debugDecode { + fmt.Println("huffsym: n==0") + } + f.err = CorruptInputError(f.roffset) + return + } + fb = fb >> (n & regSizeMaskUint32) + fnb = fnb - n + dist = uint32(chunk >> huffmanValueShift) + break } - f.err = err - return } } @@ -852,21 +1201,28 @@ readLiteral: 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 { + extra := (dist & 1) << (nb & regSizeMaskUint32) + for fnb < nb { + c, err := fr.ReadByte() + if err != nil { + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("morebits f.nb>= nb - f.nb -= nb - dist = 1<<(nb+1) + 1 + extra + extra |= fb & bitMask32[nb] + fb >>= nb & regSizeMaskUint32 + fnb -= nb + dist = 1<<((nb+1)®SizeMaskUint32) + 1 + extra + // slower: dist = bitMask32[nb+1] + 2 + extra default: + f.b, f.nb = fb, fnb if debugDecode { fmt.Println("dist too big:", dist, maxNumDist) } @@ -875,35 +1231,38 @@ readLiteral: } // No check on length; encoding can be prescient. - if dist > f.dict.histSize() { + if dist > uint32(dict.histSize()) { + f.b, f.nb = fb, fnb if debugDecode { - fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize()) + fmt.Println("dist > dict.histSize():", dist, dict.histSize()) } f.err = CorruptInputError(f.roffset) return } - f.copyLen, f.copyDist = length, dist + f.copyLen, f.copyDist = length, int(dist) goto copyHistory } copyHistory: // Perform a backwards copy according to RFC section 3.2.3. { - cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen) + cnt := dict.tryWriteCopy(f.copyDist, f.copyLen) if cnt == 0 { - cnt = f.dict.writeCopy(f.copyDist, f.copyLen) + cnt = 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 + if dict.availWrite() == 0 || f.copyLen > 0 { + f.toRead = dict.readFlush() + f.step = (*decompressor).huffmanGenericReader // We need to continue this work f.stepState = stateDict + f.b, f.nb = fb, fnb return } goto readLiteral } + // Not reached } func (f *decompressor) huffmanBlockDecoder() func() { @@ -916,7 +1275,9 @@ func (f *decompressor) huffmanBlockDecoder() func() { return f.huffmanBufioReader case *strings.Reader: return f.huffmanStringsReader + case Reader: + return f.huffmanGenericReader default: - return f.huffmanBlockGeneric + return f.huffmanGenericReader } } diff --git a/vendor/github.com/klauspost/compress/flate/level1.go b/vendor/github.com/klauspost/compress/flate/level1.go index 1e5eea39..703b9a89 100644 --- a/vendor/github.com/klauspost/compress/flate/level1.go +++ b/vendor/github.com/klauspost/compress/flate/level1.go @@ -1,6 +1,10 @@ package flate -import "fmt" +import ( + "encoding/binary" + "fmt" + "math/bits" +) // fastGen maintains the table for matches, // and the previous byte block for level 2. @@ -15,6 +19,7 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { const ( inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin + hashBytes = 5 ) if debugDeflate && e.cur < 0 { panic(fmt.Sprint("e.cur < 0: ", e.cur)) @@ -64,7 +69,7 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { sLimit := int32(len(src) - inputMargin) // nextEmit is where in src the next emitLiteral should start from. - cv := load3232(src, s) + cv := load6432(src, s) for { const skipLog = 5 @@ -73,7 +78,7 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { nextS := s var candidate tableEntry for { - nextHash := hash(cv) + nextHash := hashLen(cv, tableBits, hashBytes) candidate = e.table[nextHash] nextS = s + doEvery + (s-nextEmit)>>skipLog if nextS > sLimit { @@ -82,16 +87,16 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { now := load6432(src, nextS) e.table[nextHash] = tableEntry{offset: s + e.cur} - nextHash = hash(uint32(now)) + nextHash = hashLen(now, tableBits, hashBytes) offset := s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + if offset < maxMatchOffset && uint32(cv) == load3232(src, candidate.offset-e.cur) { e.table[nextHash] = tableEntry{offset: nextS + e.cur} break } // Do one right away... - cv = uint32(now) + cv = now s = nextS nextS++ candidate = e.table[nextHash] @@ -99,11 +104,11 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { e.table[nextHash] = tableEntry{offset: s + e.cur} offset = s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + if offset < maxMatchOffset && uint32(cv) == load3232(src, candidate.offset-e.cur) { e.table[nextHash] = tableEntry{offset: nextS + e.cur} break } - cv = uint32(now) + cv = now s = nextS } @@ -116,7 +121,32 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { // Extend the 4-byte match as long as possible. t := candidate.offset - e.cur - l := e.matchlenLong(s+4, t+4, src) + 4 + var l = int32(4) + if false { + l = e.matchlenLong(s+4, t+4, src) + 4 + } else { + // inlined: + a := src[s+4:] + b := src[t+4:] + for len(a) >= 8 { + if diff := binary.LittleEndian.Uint64(a) ^ binary.LittleEndian.Uint64(b); diff != 0 { + l += int32(bits.TrailingZeros64(diff) >> 3) + break + } + l += 8 + a = a[8:] + b = b[8:] + } + if len(a) < 8 { + b = b[:len(a)] + for i := range a { + if a[i] != b[i] { + break + } + l++ + } + } + } // Extend backwards for t > 0 && s > nextEmit && src[t-1] == src[s-1] { @@ -125,11 +155,43 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } // Save the match found - dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + if false { + dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) + } else { + // Inlined... + xoffset := uint32(s - t - baseMatchOffset) + xlength := l + oc := offsetCode(xoffset) + xoffset |= oc << 16 + for xlength > 0 { + xl := xlength + if xl > 258 { + if xl > 258+baseMatchLength { + xl = 258 + } else { + xl = 258 - baseMatchLength + } + } + xlength -= xl + xl -= baseMatchLength + dst.extraHist[lengthCodes1[uint8(xl)]]++ + dst.offHist[oc]++ + dst.tokens[dst.n] = token(matchType | uint32(xl)<= s { @@ -137,9 +199,9 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { } 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} + if int(s+l+8) < len(src) { + cv := load6432(src, s) + e.table[hashLen(cv, tableBits, hashBytes)] = tableEntry{offset: s + e.cur} } goto emitRemainder } @@ -152,16 +214,16 @@ func (e *fastEncL1) Encode(dst *tokens, src []byte) { // three load32 calls. x := load6432(src, s-2) o := e.cur + s - 2 - prevHash := hash(uint32(x)) + prevHash := hashLen(x, tableBits, hashBytes) e.table[prevHash] = tableEntry{offset: o} x >>= 16 - currHash := hash(uint32(x)) + currHash := hashLen(x, tableBits, hashBytes) 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) + cv = x >> 8 s++ break } diff --git a/vendor/github.com/klauspost/compress/flate/level2.go b/vendor/github.com/klauspost/compress/flate/level2.go index 5b986a19..876dfbe3 100644 --- a/vendor/github.com/klauspost/compress/flate/level2.go +++ b/vendor/github.com/klauspost/compress/flate/level2.go @@ -16,6 +16,7 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { const ( inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin + hashBytes = 5 ) if debugDeflate && e.cur < 0 { @@ -66,7 +67,7 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { sLimit := int32(len(src) - inputMargin) // nextEmit is where in src the next emitLiteral should start from. - cv := load3232(src, s) + cv := load6432(src, s) for { // When should we start skipping if we haven't found matches in a long while. const skipLog = 5 @@ -75,7 +76,7 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { nextS := s var candidate tableEntry for { - nextHash := hash4u(cv, bTableBits) + nextHash := hashLen(cv, bTableBits, hashBytes) s = nextS nextS = s + doEvery + (s-nextEmit)>>skipLog if nextS > sLimit { @@ -84,16 +85,16 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { candidate = e.table[nextHash] now := load6432(src, nextS) e.table[nextHash] = tableEntry{offset: s + e.cur} - nextHash = hash4u(uint32(now), bTableBits) + nextHash = hashLen(now, bTableBits, hashBytes) offset := s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + if offset < maxMatchOffset && uint32(cv) == load3232(src, candidate.offset-e.cur) { e.table[nextHash] = tableEntry{offset: nextS + e.cur} break } // Do one right away... - cv = uint32(now) + cv = now s = nextS nextS++ candidate = e.table[nextHash] @@ -101,10 +102,10 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { e.table[nextHash] = tableEntry{offset: s + e.cur} offset = s - (candidate.offset - e.cur) - if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) { + if offset < maxMatchOffset && uint32(cv) == load3232(src, candidate.offset-e.cur) { break } - cv = uint32(now) + cv = now } // A 4-byte match has been found. We'll later see if more than 4 bytes @@ -134,7 +135,15 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) @@ -146,25 +155,25 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { 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} + if int(s+l+8) < len(src) { + cv := load6432(src, s) + e.table[hashLen(cv, bTableBits, hashBytes)] = 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) + x := load6432(src, i) + nextHash := hashLen(x, bTableBits, hashBytes) e.table[nextHash] = tableEntry{offset: e.cur + i} // Skip one x >>= 16 - nextHash = hash4u(uint32(x), bTableBits) + nextHash = hashLen(x, bTableBits, hashBytes) e.table[nextHash] = tableEntry{offset: e.cur + i + 2} // Skip one x >>= 16 - nextHash = hash4u(uint32(x), bTableBits) + nextHash = hashLen(x, bTableBits, hashBytes) e.table[nextHash] = tableEntry{offset: e.cur + i + 4} } @@ -176,17 +185,17 @@ func (e *fastEncL2) Encode(dst *tokens, src []byte) { // three load32 calls. x := load6432(src, s-2) o := e.cur + s - 2 - prevHash := hash4u(uint32(x), bTableBits) - prevHash2 := hash4u(uint32(x>>8), bTableBits) + prevHash := hashLen(x, bTableBits, hashBytes) + prevHash2 := hashLen(x>>8, bTableBits, hashBytes) e.table[prevHash] = tableEntry{offset: o} e.table[prevHash2] = tableEntry{offset: o + 1} - currHash := hash4u(uint32(x>>16), bTableBits) + currHash := hashLen(x>>16, bTableBits, hashBytes) 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) + cv = x >> 24 s++ break } diff --git a/vendor/github.com/klauspost/compress/flate/level3.go b/vendor/github.com/klauspost/compress/flate/level3.go index c22b4244..7aa2b72a 100644 --- a/vendor/github.com/klauspost/compress/flate/level3.go +++ b/vendor/github.com/klauspost/compress/flate/level3.go @@ -5,14 +5,17 @@ import "fmt" // fastEncL3 type fastEncL3 struct { fastGen - table [tableSize]tableEntryPrev + table [1 << 16]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 + inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin + tableBits = 16 + tableSize = 1 << tableBits + hashBytes = 5 ) if debugDeflate && e.cur < 0 { @@ -67,20 +70,20 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { sLimit := int32(len(src) - inputMargin) // nextEmit is where in src the next emitLiteral should start from. - cv := load3232(src, s) + cv := load6432(src, s) for { - const skipLog = 6 + const skipLog = 7 nextS := s var candidate tableEntry for { - nextHash := hash(cv) + nextHash := hashLen(cv, tableBits, hashBytes) s = nextS nextS = s + 1 + (s-nextEmit)>>skipLog if nextS > sLimit { goto emitRemainder } candidates := e.table[nextHash] - now := load3232(src, nextS) + now := load6432(src, nextS) // Safe offset distance until s + 4... minOffset := e.cur + s - (maxMatchOffset - 4) @@ -94,8 +97,8 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { continue } - if cv == load3232(src, candidate.offset-e.cur) { - if candidates.Prev.offset < minOffset || cv != load3232(src, candidates.Prev.offset-e.cur) { + if uint32(cv) == load3232(src, candidate.offset-e.cur) { + if candidates.Prev.offset < minOffset || uint32(cv) != load3232(src, candidates.Prev.offset-e.cur) { break } // Both match and are valid, pick longest. @@ -110,7 +113,7 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { // We only check if value mismatches. // Offset will always be invalid in other cases. candidate = candidates.Prev - if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { + if candidate.offset > minOffset && uint32(cv) == load3232(src, candidate.offset-e.cur) { break } } @@ -141,7 +144,15 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } dst.AddMatchLong(l, uint32(s-t-baseMatchOffset)) @@ -154,9 +165,9 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { 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) + if int(t+8) < len(src) && t > 0 { + cv = load6432(src, t) + nextHash := hashLen(cv, tableBits, hashBytes) e.table[nextHash] = tableEntryPrev{ Prev: e.table[nextHash].Cur, Cur: tableEntry{offset: e.cur + t}, @@ -165,32 +176,33 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { 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}, + // Store every 5th hash in-between. + for i := s - l + 2; i < s-5; i += 6 { + nextHash := hashLen(load6432(src, i), tableBits, hashBytes) + e.table[nextHash] = tableEntryPrev{ + Prev: e.table[nextHash].Cur, + Cur: tableEntry{offset: e.cur + i}} } - x >>= 8 - prevHash = hash(uint32(x)) + // We could immediately start working at s now, but to improve + // compression we first update the hash table at s-2 to s. + x := load6432(src, s-2) + prevHash := hashLen(x, tableBits, hashBytes) e.table[prevHash] = tableEntryPrev{ Prev: e.table[prevHash].Cur, Cur: tableEntry{offset: e.cur + s - 2}, } x >>= 8 - prevHash = hash(uint32(x)) + prevHash = hashLen(x, tableBits, hashBytes) e.table[prevHash] = tableEntryPrev{ Prev: e.table[prevHash].Cur, Cur: tableEntry{offset: e.cur + s - 1}, } x >>= 8 - currHash := hash(uint32(x)) + currHash := hashLen(x, tableBits, hashBytes) candidates := e.table[currHash] - cv = uint32(x) + cv = x e.table[currHash] = tableEntryPrev{ Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur}, @@ -200,18 +212,18 @@ func (e *fastEncL3) Encode(dst *tokens, src []byte) { 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. + if candidate.offset > minOffset { + if uint32(cv) == load3232(src, candidate.offset-e.cur) { + // Found a match... + continue + } candidate = candidates.Prev - if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) { - offset := s - (candidate.offset - e.cur) - if offset <= maxMatchOffset { - continue - } + if candidate.offset > minOffset && uint32(cv) == load3232(src, candidate.offset-e.cur) { + // Match at prev... + continue } } - cv = uint32(x >> 8) + cv = x >> 8 s++ break } diff --git a/vendor/github.com/klauspost/compress/flate/level4.go b/vendor/github.com/klauspost/compress/flate/level4.go index e62f0c02..23c08b32 100644 --- a/vendor/github.com/klauspost/compress/flate/level4.go +++ b/vendor/github.com/klauspost/compress/flate/level4.go @@ -12,6 +12,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { const ( inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin + hashShortBytes = 4 ) if debugDeflate && e.cur < 0 { panic(fmt.Sprint("e.cur < 0: ", e.cur)) @@ -80,7 +81,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { nextS := s var t int32 for { - nextHashS := hash4x64(cv, tableBits) + nextHashS := hashLen(cv, tableBits, hashShortBytes) nextHashL := hash7(cv, tableBits) s = nextS @@ -135,7 +136,15 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } if debugDeflate { if t >= s { @@ -160,7 +169,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { // Index first pair after match end. if int(s+8) < len(src) { cv := load6432(src, s) - e.table[hash4x64(cv, tableBits)] = tableEntry{offset: s + e.cur} + e.table[hashLen(cv, tableBits, hashShortBytes)] = tableEntry{offset: s + e.cur} e.bTable[hash7(cv, tableBits)] = tableEntry{offset: s + e.cur} } goto emitRemainder @@ -175,7 +184,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { 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 + e.table[hashLen(cv>>8, tableBits, hashShortBytes)] = t2 i += 3 for ; i < s-1; i += 3 { @@ -184,7 +193,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { 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 + e.table[hashLen(cv>>8, tableBits, hashShortBytes)] = t2 } } } @@ -193,7 +202,7 @@ func (e *fastEncL4) Encode(dst *tokens, src []byte) { // 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) + prevHashS := hashLen(x, tableBits, hashShortBytes) prevHashL := hash7(x, tableBits) e.table[prevHashS] = tableEntry{offset: o} e.bTable[prevHashL] = tableEntry{offset: o} diff --git a/vendor/github.com/klauspost/compress/flate/level5.go b/vendor/github.com/klauspost/compress/flate/level5.go index d513f1ff..83ef50ba 100644 --- a/vendor/github.com/klauspost/compress/flate/level5.go +++ b/vendor/github.com/klauspost/compress/flate/level5.go @@ -12,6 +12,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { const ( inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin + hashShortBytes = 4 ) if debugDeflate && e.cur < 0 { panic(fmt.Sprint("e.cur < 0: ", e.cur)) @@ -88,7 +89,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { var l int32 var t int32 for { - nextHashS := hash4x64(cv, tableBits) + nextHashS := hashLen(cv, tableBits, hashShortBytes) nextHashL := hash7(cv, tableBits) s = nextS @@ -105,7 +106,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { eLong := &e.bTable[nextHashL] eLong.Cur, eLong.Prev = entry, eLong.Cur - nextHashS = hash4x64(next, tableBits) + nextHashS = hashLen(next, tableBits, hashShortBytes) nextHashL = hash7(next, tableBits) t = lCandidate.Cur.offset - e.cur @@ -182,12 +183,34 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { // 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 { + // Extend the 4-byte match as long as possible. l = e.matchlenLong(s+4, t+4, src) + 4 } else if l == maxMatchLength { l += e.matchlenLong(s+l, t+l, src) } + + // Try to locate a better match by checking the end of best match... + if sAt := s + l; l < 30 && sAt < sLimit { + // Allow some bytes at the beginning to mismatch. + // Sweet spot is 2/3 bytes depending on input. + // 3 is only a little better when it is but sometimes a lot worse. + // The skipped bytes are tested in Extend backwards, + // and still picked up as part of the match if they do. + const skipBeginning = 2 + eLong := e.bTable[hash7(load6432(src, sAt), tableBits)].Cur.offset + t2 := eLong - e.cur - l + skipBeginning + s2 := s + skipBeginning + off := s2 - t2 + if t2 >= 0 && off < maxMatchOffset && off > 0 { + if l2 := e.matchlenLong(s2, t2, src); l2 > l { + t = t2 + l = l2 + s = s2 + } + } + } + // Extend backwards for t > 0 && s > nextEmit && src[t-1] == src[s-1] { s-- @@ -195,7 +218,15 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } if debugDeflate { if t >= s { @@ -227,7 +258,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { if i < s-1 { cv := load6432(src, i) t := tableEntry{offset: i + e.cur} - e.table[hash4x64(cv, tableBits)] = t + e.table[hashLen(cv, tableBits, hashShortBytes)] = t eLong := &e.bTable[hash7(cv, tableBits)] eLong.Cur, eLong.Prev = t, eLong.Cur @@ -240,7 +271,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { // 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 + e.table[hashLen(cv, tableBits, hashShortBytes)] = t // Skip one - otherwise we risk hitting 's' i += 4 @@ -250,7 +281,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { 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 + e.table[hashLen(cv>>8, tableBits, hashShortBytes)] = t2 } } } @@ -259,7 +290,7 @@ func (e *fastEncL5) Encode(dst *tokens, src []byte) { // 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) + prevHashS := hashLen(x, tableBits, hashShortBytes) prevHashL := hash7(x, tableBits) e.table[prevHashS] = tableEntry{offset: o} eLong := &e.bTable[prevHashL] diff --git a/vendor/github.com/klauspost/compress/flate/level6.go b/vendor/github.com/klauspost/compress/flate/level6.go index a52c80ea..f1e9d98f 100644 --- a/vendor/github.com/klauspost/compress/flate/level6.go +++ b/vendor/github.com/klauspost/compress/flate/level6.go @@ -12,6 +12,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { const ( inputMargin = 12 - 1 minNonLiteralBlockSize = 1 + 1 + inputMargin + hashShortBytes = 4 ) if debugDeflate && e.cur < 0 { panic(fmt.Sprint("e.cur < 0: ", e.cur)) @@ -90,7 +91,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { var l int32 var t int32 for { - nextHashS := hash4x64(cv, tableBits) + nextHashS := hashLen(cv, tableBits, hashShortBytes) nextHashL := hash7(cv, tableBits) s = nextS nextS = s + doEvery + (s-nextEmit)>>skipLog @@ -107,7 +108,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { eLong.Cur, eLong.Prev = entry, eLong.Cur // Calculate hashes of 'next' - nextHashS = hash4x64(next, tableBits) + nextHashS = hashLen(next, tableBits, hashShortBytes) nextHashL = hash7(next, tableBits) t = lCandidate.Cur.offset - e.cur @@ -211,6 +212,40 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { l += e.matchlenLong(s+l, t+l, src) } + // Try to locate a better match by checking the end-of-match... + if sAt := s + l; sAt < sLimit { + // Allow some bytes at the beginning to mismatch. + // Sweet spot is 2/3 bytes depending on input. + // 3 is only a little better when it is but sometimes a lot worse. + // The skipped bytes are tested in Extend backwards, + // and still picked up as part of the match if they do. + const skipBeginning = 2 + eLong := &e.bTable[hash7(load6432(src, sAt), tableBits)] + // Test current + t2 := eLong.Cur.offset - e.cur - l + skipBeginning + s2 := s + skipBeginning + off := s2 - t2 + if off < maxMatchOffset { + if off > 0 && t2 >= 0 { + if l2 := e.matchlenLong(s2, t2, src); l2 > l { + t = t2 + l = l2 + s = s2 + } + } + // Test next: + t2 = eLong.Prev.offset - e.cur - l + skipBeginning + off := s2 - t2 + if off > 0 && off < maxMatchOffset && t2 >= 0 { + if l2 := e.matchlenLong(s2, t2, src); l2 > l { + t = t2 + l = l2 + s = s2 + } + } + } + } + // Extend backwards for t > 0 && s > nextEmit && src[t-1] == src[s-1] { s-- @@ -218,7 +253,15 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } if false { if t >= s { @@ -244,7 +287,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { // Index after match end. for i := nextS + 1; i < int32(len(src))-8; i += 2 { cv := load6432(src, i) - e.table[hash4x64(cv, tableBits)] = tableEntry{offset: i + e.cur} + e.table[hashLen(cv, tableBits, hashShortBytes)] = tableEntry{offset: i + e.cur} eLong := &e.bTable[hash7(cv, tableBits)] eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur}, eLong.Cur } @@ -259,7 +302,7 @@ func (e *fastEncL6) Encode(dst *tokens, src []byte) { 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 + e.table[hashLen(cv, tableBits, hashShortBytes)] = t eLong.Cur, eLong.Prev = t, eLong.Cur eLong2.Cur, eLong2.Prev = t2, eLong2.Cur } diff --git a/vendor/github.com/klauspost/compress/flate/stateless.go b/vendor/github.com/klauspost/compress/flate/stateless.go index 53e89912..93a1d150 100644 --- a/vendor/github.com/klauspost/compress/flate/stateless.go +++ b/vendor/github.com/klauspost/compress/flate/stateless.go @@ -59,9 +59,9 @@ var bitWriterPool = sync.Pool{ }, } -// StatelessDeflate allows to compress directly to a Writer without retaining state. +// StatelessDeflate allows compressing 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. +// Up to 8KB of an optional dictionary can be given which is 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 { @@ -249,7 +249,15 @@ func statelessEnc(dst *tokens, src []byte, startAt int16) { l++ } if nextEmit < s { - emitLiteral(dst, src[nextEmit:s]) + if false { + emitLiteral(dst, src[nextEmit:s]) + } else { + for _, v := range src[nextEmit:s] { + dst.tokens[dst.n] = token(v) + dst.litHist[v]++ + dst.n++ + } + } } // Save the match found diff --git a/vendor/github.com/klauspost/compress/flate/token.go b/vendor/github.com/klauspost/compress/flate/token.go index f9abf606..d818790c 100644 --- a/vendor/github.com/klauspost/compress/flate/token.go +++ b/vendor/github.com/klauspost/compress/flate/token.go @@ -13,14 +13,16 @@ import ( ) const ( - // 2 bits: type 0 = literal 1=EOF 2=Match 3=Unused - // 8 bits: xlength = length - MIN_MATCH_LENGTH - // 22 bits xoffset = offset - MIN_OFFSET_SIZE, or literal - lengthShift = 22 - offsetMask = 1<maxnumlit offHist [32]uint16 // offset codes litHist [256]uint16 // codes 0->255 - n uint16 // Must be able to contain maxStoreBlockSize + nFilled int + n uint16 // Must be able to contain maxStoreBlockSize tokens [maxStoreBlockSize + 1]token } @@ -139,7 +141,7 @@ func (t *tokens) Reset() { return } t.n = 0 - t.nLits = 0 + t.nFilled = 0 for i := range t.litHist[:] { t.litHist[i] = 0 } @@ -158,12 +160,12 @@ func (t *tokens) Fill() { for i, v := range t.litHist[:] { if v == 0 { t.litHist[i] = 1 - t.nLits++ + t.nFilled++ } } for i, v := range t.extraHist[:literalCount-256] { if v == 0 { - t.nLits++ + t.nFilled++ t.extraHist[i] = 1 } } @@ -187,26 +189,23 @@ func (t *tokens) indexTokens(in []token) { t.AddLiteral(tok.literal()) continue } - t.AddMatch(uint32(tok.length()), tok.offset()) + t.AddMatch(uint32(tok.length()), tok.offset()&matchOffsetOnlyMask) } } // 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) + for _, v := range lit { + dst.tokens[dst.n] = token(v) dst.litHist[v]++ + dst.n++ } - 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 @@ -227,12 +226,13 @@ func (t *tokens) EstimatedBits() int { shannon := float32(0) bits := int(0) nMatches := 0 - if t.nLits > 0 { - invTotal := 1.0 / float32(t.nLits) + total := int(t.n) + t.nFilled + if total > 0 { + invTotal := 1.0 / float32(total) for _, v := range t.litHist[:] { if v > 0 { n := float32(v) - shannon += -mFastLog2(n*invTotal) * n + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n } } // Just add 15 for EOB @@ -240,7 +240,7 @@ func (t *tokens) EstimatedBits() int { for i, v := range t.extraHist[1 : literalCount-256] { if v > 0 { n := float32(v) - shannon += -mFastLog2(n*invTotal) * n + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n bits += int(lengthExtraBits[i&31]) * int(v) nMatches += int(v) } @@ -251,7 +251,7 @@ func (t *tokens) EstimatedBits() int { for i, v := range t.offHist[:offsetCodeCount] { if v > 0 { n := float32(v) - shannon += -mFastLog2(n*invTotal) * n + shannon += atLeastOne(-mFastLog2(n*invTotal)) * n bits += int(offsetExtraBits[i&31]) * int(v) } } @@ -270,11 +270,12 @@ func (t *tokens) AddMatch(xlength uint32, xoffset uint32) { panic(fmt.Errorf("invalid offset: %v", xoffset)) } } - t.nLits++ - lengthCode := lengthCodes1[uint8(xlength)] & 31 + oCode := offsetCode(xoffset) + xoffset |= oCode << 16 + + t.extraHist[lengthCodes1[uint8(xlength)]]++ + t.offHist[oCode&31]++ t.tokens[t.n] = token(matchType | xlength< 0 { xl := xlength if xl > 258 { // We need to have at least baseMatchLength left over for next loop. - xl = 258 - baseMatchLength + if xl > 258+baseMatchLength { + xl = 258 + } else { + xl = 258 - baseMatchLength + } } xlength -= xl - xl -= 3 - t.nLits++ - lengthCode := lengthCodes1[uint8(xl)] & 31 + xl -= baseMatchLength + t.extraHist[lengthCodes1[uint8(xl)]]++ + t.offHist[oc&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]) } +// Convert length to code. +func lengthCode(len uint8) uint8 { return lengthCodes[len] } // Returns the offset code corresponding to a specific offset func offsetCode(off uint32) uint32 { diff --git a/vendor/github.com/klauspost/compress/gzip/gunzip.go b/vendor/github.com/klauspost/compress/gzip/gunzip.go index 568b5d4f..66fe5ddf 100644 --- a/vendor/github.com/klauspost/compress/gzip/gunzip.go +++ b/vendor/github.com/klauspost/compress/gzip/gunzip.go @@ -8,8 +8,8 @@ package gzip import ( "bufio" + "compress/gzip" "encoding/binary" - "errors" "hash/crc32" "io" "time" @@ -30,9 +30,9 @@ const ( var ( // ErrChecksum is returned when reading GZIP data that has an invalid checksum. - ErrChecksum = errors.New("gzip: invalid checksum") + ErrChecksum = gzip.ErrChecksum // ErrHeader is returned when reading GZIP data that has an invalid header. - ErrHeader = errors.New("gzip: invalid header") + ErrHeader = gzip.ErrHeader ) var le = binary.LittleEndian @@ -75,6 +75,7 @@ type Header struct { type Reader struct { Header // valid after NewReader or Reader.Reset r flate.Reader + br *bufio.Reader decompressor io.ReadCloser digest uint32 // CRC-32, IEEE polynomial (section 8) size uint32 // Uncompressed size (section 2.3.1) @@ -109,7 +110,13 @@ func (z *Reader) Reset(r io.Reader) error { if rr, ok := r.(flate.Reader); ok { z.r = rr } else { - z.r = bufio.NewReader(r) + // Reuse if we can. + if z.br != nil { + z.br.Reset(r) + } else { + z.br = bufio.NewReader(r) + } + z.r = z.br } z.Header, z.err = z.readHeader() return z.err @@ -245,42 +252,40 @@ func (z *Reader) Read(p []byte) (n int, err error) { 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 - } + for n == 0 { + 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 + // 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 + // 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 + 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) + return n, nil } // Support the io.WriteTo interface for io.Copy and friends. diff --git a/vendor/github.com/klauspost/compress/zlib/reader.go b/vendor/github.com/klauspost/compress/zlib/reader.go index d9091e83..f127d477 100644 --- a/vendor/github.com/klauspost/compress/zlib/reader.go +++ b/vendor/github.com/klauspost/compress/zlib/reader.go @@ -25,7 +25,7 @@ package zlib import ( "bufio" - "errors" + "compress/zlib" "hash" "hash/adler32" "io" @@ -37,11 +37,11 @@ const zlibDeflate = 8 var ( // ErrChecksum is returned when reading ZLIB data that has an invalid checksum. - ErrChecksum = errors.New("zlib: invalid checksum") + ErrChecksum = zlib.ErrChecksum // ErrDictionary is returned when reading ZLIB data that has an invalid dictionary. - ErrDictionary = errors.New("zlib: invalid dictionary") + ErrDictionary = zlib.ErrDictionary // ErrHeader is returned when reading ZLIB data that has an invalid header. - ErrHeader = errors.New("zlib: invalid header") + ErrHeader = zlib.ErrHeader ) type reader struct { diff --git a/vendor/modules.txt b/vendor/modules.txt index 58122aa3..752cf15b 100644 --- a/vendor/modules.txt +++ b/vendor/modules.txt @@ -139,9 +139,12 @@ github.com/jinzhu/inflection github.com/jinzhu/now # github.com/k0kubun/colorstring v0.0.0-20150214042306-9440f1994b88 ## explicit -# github.com/klauspost/compress v1.10.7 +# github.com/klauspost/compress v1.15.11 +## explicit github.com/klauspost/compress/flate github.com/klauspost/compress/gzip +github.com/klauspost/compress/s2 +github.com/klauspost/compress/snappy github.com/klauspost/compress/zlib # github.com/kr/fs v0.1.0 github.com/kr/fs