kairos/luet
1
0
Fork 0
mirror of https://github.com/mudler/luet.git synced 2025-09-04 00:34:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

Switch to go mod

Browse Source
This commit is contained in:
Ettore Di Giacinto
2019-11-10 18:04:06 +01:00
parent f634493dc0
commit 420186b7db
1200 changed files with 139110 additions and 7763 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
vendor/github.com/klauspost/compress/LICENSE generated vendored Normal file
View File

@@ -0,0 +1,28 @@
Copyright (c) 2012 The Go Authors. All rights reserved.
Copyright (c) 2019 Klaus Post. All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above
copyright notice, this list of conditions and the following disclaimer
in the documentation and/or other materials provided with the
distribution.
* Neither the name of Google Inc. nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

42
vendor/github.com/klauspost/compress/flate/crc32_amd64.go generated vendored Normal file
View File

@@ -0,0 +1,42 @@
//+build !noasm
//+build !appengine
//+build !gccgo
// Copyright 2015, Klaus Post, see LICENSE for details.
package flate
import (
"github.com/klauspost/cpuid"
)
// crc32sse returns a hash for the first 4 bytes of the slice
// len(a) must be >= 4.
//go:noescape
func crc32sse(a []byte) uint32
// crc32sseAll calculates hashes for each 4-byte set in a.
// dst must be east len(a) - 4 in size.
// The size is not checked by the assembly.
//go:noescape
func crc32sseAll(a []byte, dst []uint32)
// matchLenSSE4 returns the number of matching bytes in a and b
// up to length 'max'. Both slices must be at least 'max'
// bytes in size.
//
// TODO: drop the "SSE4" name, since it doesn't use any SSE instructions.
//
//go:noescape
func matchLenSSE4(a, b []byte, max int) int
// histogram accumulates a histogram of b in h.
// h must be at least 256 entries in length,
// and must be cleared before calling this function.
//go:noescape
func histogram(b []byte, h []int32)
// Detect SSE 4.2 feature.
func init() {
useSSE42 = cpuid.CPU.SSE42()
}

214
vendor/github.com/klauspost/compress/flate/crc32_amd64.s generated vendored Normal file
View File

@@ -0,0 +1,214 @@
//+build !noasm
//+build !appengine
//+build !gccgo
// Copyright 2015, Klaus Post, see LICENSE for details.
// func crc32sse(a []byte) uint32
TEXT ·crc32sse(SB), 4, $0
MOVQ a+0(FP), R10
XORQ BX, BX
// CRC32 dword (R10), EBX
BYTE $0xF2; BYTE $0x41; BYTE $0x0f
BYTE $0x38; BYTE $0xf1; BYTE $0x1a
MOVL BX, ret+24(FP)
RET
// func crc32sseAll(a []byte, dst []uint32)
TEXT ·crc32sseAll(SB), 4, $0
MOVQ a+0(FP), R8 // R8: src
MOVQ a_len+8(FP), R10 // input length
MOVQ dst+24(FP), R9 // R9: dst
SUBQ $4, R10
JS end
JZ one_crc
MOVQ R10, R13
SHRQ $2, R10 // len/4
ANDQ $3, R13 // len&3
XORQ BX, BX
ADDQ $1, R13
TESTQ R10, R10
JZ rem_loop
crc_loop:
MOVQ (R8), R11
XORQ BX, BX
XORQ DX, DX
XORQ DI, DI
MOVQ R11, R12
SHRQ $8, R11
MOVQ R12, AX
MOVQ R11, CX
SHRQ $16, R12
SHRQ $16, R11
MOVQ R12, SI
// CRC32 EAX, EBX
BYTE $0xF2; BYTE $0x0f
BYTE $0x38; BYTE $0xf1; BYTE $0xd8
// CRC32 ECX, EDX
BYTE $0xF2; BYTE $0x0f
BYTE $0x38; BYTE $0xf1; BYTE $0xd1
// CRC32 ESI, EDI
BYTE $0xF2; BYTE $0x0f
BYTE $0x38; BYTE $0xf1; BYTE $0xfe
MOVL BX, (R9)
MOVL DX, 4(R9)
MOVL DI, 8(R9)
XORQ BX, BX
MOVL R11, AX
// CRC32 EAX, EBX
BYTE $0xF2; BYTE $0x0f
BYTE $0x38; BYTE $0xf1; BYTE $0xd8
MOVL BX, 12(R9)
ADDQ $16, R9
ADDQ $4, R8
XORQ BX, BX
SUBQ $1, R10
JNZ crc_loop
rem_loop:
MOVL (R8), AX
// CRC32 EAX, EBX
BYTE $0xF2; BYTE $0x0f
BYTE $0x38; BYTE $0xf1; BYTE $0xd8
MOVL BX, (R9)
ADDQ $4, R9
ADDQ $1, R8
XORQ BX, BX
SUBQ $1, R13
JNZ rem_loop
end:
RET
one_crc:
MOVQ $1, R13
XORQ BX, BX
JMP rem_loop
// func matchLenSSE4(a, b []byte, max int) int
TEXT ·matchLenSSE4(SB), 4, $0
MOVQ a_base+0(FP), SI
MOVQ b_base+24(FP), DI
MOVQ DI, DX
MOVQ max+48(FP), CX
cmp8:
// As long as we are 8 or more bytes before the end of max, we can load and
// compare 8 bytes at a time. If those 8 bytes are equal, repeat.
CMPQ CX, $8
JLT cmp1
MOVQ (SI), AX
MOVQ (DI), BX
CMPQ AX, BX
JNE bsf
ADDQ $8, SI
ADDQ $8, DI
SUBQ $8, CX
JMP cmp8
bsf:
// If those 8 bytes were not equal, XOR the two 8 byte values, and return
// the index of the first byte that differs. The BSF instruction finds the
// least significant 1 bit, the amd64 architecture is little-endian, and
// the shift by 3 converts a bit index to a byte index.
XORQ AX, BX
BSFQ BX, BX
SHRQ $3, BX
ADDQ BX, DI
// Subtract off &b[0] to convert from &b[ret] to ret, and return.
SUBQ DX, DI
MOVQ DI, ret+56(FP)
RET
cmp1:
// In the slices' tail, compare 1 byte at a time.
CMPQ CX, $0
JEQ matchLenEnd
MOVB (SI), AX
MOVB (DI), BX
CMPB AX, BX
JNE matchLenEnd
ADDQ $1, SI
ADDQ $1, DI
SUBQ $1, CX
JMP cmp1
matchLenEnd:
// Subtract off &b[0] to convert from &b[ret] to ret, and return.
SUBQ DX, DI
MOVQ DI, ret+56(FP)
RET
// func histogram(b []byte, h []int32)
TEXT ·histogram(SB), 4, $0
MOVQ b+0(FP), SI // SI: &b
MOVQ b_len+8(FP), R9 // R9: len(b)
MOVQ h+24(FP), DI // DI: Histogram
MOVQ R9, R8
SHRQ $3, R8
JZ hist1
XORQ R11, R11
loop_hist8:
MOVQ (SI), R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
MOVB R10, R11
INCL (DI)(R11*4)
SHRQ $8, R10
INCL (DI)(R10*4)
ADDQ $8, SI
DECQ R8
JNZ loop_hist8
hist1:
ANDQ $7, R9
JZ end_hist
XORQ R10, R10
loop_hist1:
MOVB (SI), R10
INCL (DI)(R10*4)
INCQ SI
DECQ R9
JNZ loop_hist1
end_hist:
RET

35
vendor/github.com/klauspost/compress/flate/crc32_noasm.go generated vendored Normal file
View File

@@ -0,0 +1,35 @@
//+build !amd64 noasm appengine gccgo
// Copyright 2015, Klaus Post, see LICENSE for details.
package flate
func init() {
useSSE42 = false
}
// crc32sse should never be called.
func crc32sse(a []byte) uint32 {
panic("no assembler")
}
// crc32sseAll should never be called.
func crc32sseAll(a []byte, dst []uint32) {
panic("no assembler")
}
// matchLenSSE4 should never be called.
func matchLenSSE4(a, b []byte, max int) int {
panic("no assembler")
return 0
}
// histogram accumulates a histogram of b in h.
//
// len(h) must be >= 256, and h's elements must be all zeroes.
func histogram(b []byte, h []int32) {
h = h[:256]
for _, t := range b {
h[t]++
}
}

1368
vendor/github.com/klauspost/compress/flate/deflate.go generated vendored Normal file
View File

File diff suppressed because it is too large Load Diff

184
vendor/github.com/klauspost/compress/flate/dict_decoder.go generated vendored Normal file
View File

@@ -0,0 +1,184 @@
// Copyright 2016 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package flate
// dictDecoder implements the LZ77 sliding dictionary as used in decompression.
// LZ77 decompresses data through sequences of two forms of commands:
//
// * Literal insertions: Runs of one or more symbols are inserted into the data
// stream as is. This is accomplished through the writeByte method for a
// single symbol, or combinations of writeSlice/writeMark for multiple symbols.
// Any valid stream must start with a literal insertion if no preset dictionary
// is used.
//
// * Backward copies: Runs of one or more symbols are copied from previously
// emitted data. Backward copies come as the tuple (dist, length) where dist
// determines how far back in the stream to copy from and length determines how
// many bytes to copy. Note that it is valid for the length to be greater than
// the distance. Since LZ77 uses forward copies, that situation is used to
// perform a form of run-length encoding on repeated runs of symbols.
// The writeCopy and tryWriteCopy are used to implement this command.
//
// For performance reasons, this implementation performs little to no sanity
// checks about the arguments. As such, the invariants documented for each
// method call must be respected.
type dictDecoder struct {
hist []byte // Sliding window history
// Invariant: 0 <= rdPos <= wrPos <= len(hist)
wrPos int // Current output position in buffer
rdPos int // Have emitted hist[:rdPos] already
full bool // Has a full window length been written yet?
}
// init initializes dictDecoder to have a sliding window dictionary of the given
// size. If a preset dict is provided, it will initialize the dictionary with
// the contents of dict.
func (dd *dictDecoder) init(size int, dict []byte) {
*dd = dictDecoder{hist: dd.hist}
if cap(dd.hist) < size {
dd.hist = make([]byte, size)
}
dd.hist = dd.hist[:size]
if len(dict) > len(dd.hist) {
dict = dict[len(dict)-len(dd.hist):]
}
dd.wrPos = copy(dd.hist, dict)
if dd.wrPos == len(dd.hist) {
dd.wrPos = 0
dd.full = true
}
dd.rdPos = dd.wrPos
}
// histSize reports the total amount of historical data in the dictionary.
func (dd *dictDecoder) histSize() int {
if dd.full {
return len(dd.hist)
}
return dd.wrPos
}
// availRead reports the number of bytes that can be flushed by readFlush.
func (dd *dictDecoder) availRead() int {
return dd.wrPos - dd.rdPos
}
// availWrite reports the available amount of output buffer space.
func (dd *dictDecoder) availWrite() int {
return len(dd.hist) - dd.wrPos
}
// writeSlice returns a slice of the available buffer to write data to.
//
// This invariant will be kept: len(s) <= availWrite()
func (dd *dictDecoder) writeSlice() []byte {
return dd.hist[dd.wrPos:]
}
// writeMark advances the writer pointer by cnt.
//
// This invariant must be kept: 0 <= cnt <= availWrite()
func (dd *dictDecoder) writeMark(cnt int) {
dd.wrPos += cnt
}
// writeByte writes a single byte to the dictionary.
//
// This invariant must be kept: 0 < availWrite()
func (dd *dictDecoder) writeByte(c byte) {
dd.hist[dd.wrPos] = c
dd.wrPos++
}
// writeCopy copies a string at a given (dist, length) to the output.
// This returns the number of bytes copied and may be less than the requested
// length if the available space in the output buffer is too small.
//
// This invariant must be kept: 0 < dist <= histSize()
func (dd *dictDecoder) writeCopy(dist, length int) int {
dstBase := dd.wrPos
dstPos := dstBase
srcPos := dstPos - dist
endPos := dstPos + length
if endPos > len(dd.hist) {
endPos = len(dd.hist)
}
// Copy non-overlapping section after destination position.
//
// This section is non-overlapping in that the copy length for this section
// is always less than or equal to the backwards distance. This can occur
// if a distance refers to data that wraps-around in the buffer.
// Thus, a backwards copy is performed here; that is, the exact bytes in
// the source prior to the copy is placed in the destination.
if srcPos < 0 {
srcPos += len(dd.hist)
dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:])
srcPos = 0
}
// Copy possibly overlapping section before destination position.
//
// This section can overlap if the copy length for this section is larger
// than the backwards distance. This is allowed by LZ77 so that repeated
// strings can be succinctly represented using (dist, length) pairs.
// Thus, a forwards copy is performed here; that is, the bytes copied is
// possibly dependent on the resulting bytes in the destination as the copy
// progresses along. This is functionally equivalent to the following:
//
// for i := 0; i < endPos-dstPos; i++ {
// dd.hist[dstPos+i] = dd.hist[srcPos+i]
// }
// dstPos = endPos
//
for dstPos < endPos {
dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos])
}
dd.wrPos = dstPos
return dstPos - dstBase
}
// tryWriteCopy tries to copy a string at a given (distance, length) to the
// output. This specialized version is optimized for short distances.
//
// This method is designed to be inlined for performance reasons.
//
// This invariant must be kept: 0 < dist <= histSize()
func (dd *dictDecoder) tryWriteCopy(dist, length int) int {
dstPos := dd.wrPos
endPos := dstPos + length
if dstPos < dist || endPos > len(dd.hist) {
return 0
}
dstBase := dstPos
srcPos := dstPos - dist
// Copy possibly overlapping section before destination position.
loop:
dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos])
if dstPos < endPos {
goto loop // Avoid for-loop so that this function can be inlined
}
dd.wrPos = dstPos
return dstPos - dstBase
}
// readFlush returns a slice of the historical buffer that is ready to be
// emitted to the user. The data returned by readFlush must be fully consumed
// before calling any other dictDecoder methods.
func (dd *dictDecoder) readFlush() []byte {
toRead := dd.hist[dd.rdPos:dd.wrPos]
dd.rdPos = dd.wrPos
if dd.wrPos == len(dd.hist) {
dd.wrPos, dd.rdPos = 0, 0
dd.full = true
}
return toRead
}

265
vendor/github.com/klauspost/compress/flate/gen.go generated vendored Normal file
View File

@@ -0,0 +1,265 @@
// Copyright 2012 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// +build ignore
// This program generates fixedhuff.go
// Invoke as
//
// go run gen.go -output fixedhuff.go
package main
import (
"bytes"
"flag"
"fmt"
"go/format"
"io/ioutil"
"log"
)
var filename = flag.String("output", "fixedhuff.go", "output file name")
const maxCodeLen = 16
// Note: the definition of the huffmanDecoder struct is copied from
// inflate.go, as it is private to the implementation.
// chunk & 15 is number of bits
// chunk >> 4 is value, including table link
const (
huffmanChunkBits = 9
huffmanNumChunks = 1 << huffmanChunkBits
huffmanCountMask = 15
huffmanValueShift = 4
)
type huffmanDecoder struct {
min int // the minimum code length
chunks [huffmanNumChunks]uint32 // chunks as described above
links [][]uint32 // overflow links
linkMask uint32 // mask the width of the link table
}
// Initialize Huffman decoding tables from array of code lengths.
// Following this function, h is guaranteed to be initialized into a complete
// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a
// degenerate case where the tree has only a single symbol with length 1. Empty
// trees are permitted.
func (h *huffmanDecoder) init(bits []int) bool {
// Sanity enables additional runtime tests during Huffman
// table construction. It's intended to be used during
// development to supplement the currently ad-hoc unit tests.
const sanity = false
if h.min != 0 {
*h = huffmanDecoder{}
}
// Count number of codes of each length,
// compute min and max length.
var count [maxCodeLen]int
var min, max int
for _, n := range bits {
if n == 0 {
continue
}
if min == 0 || n < min {
min = n
}
if n > max {
max = n
}
count[n]++
}
// Empty tree. The decompressor.huffSym function will fail later if the tree
// is used. Technically, an empty tree is only valid for the HDIST tree and
// not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree
// is guaranteed to fail since it will attempt to use the tree to decode the
// codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is
// guaranteed to fail later since the compressed data section must be
// composed of at least one symbol (the end-of-block marker).
if max == 0 {
return true
}
code := 0
var nextcode [maxCodeLen]int
for i := min; i <= max; i++ {
code <<= 1
nextcode[i] = code
code += count[i]
}
// Check that the coding is complete (i.e., that we've
// assigned all 2-to-the-max possible bit sequences).
// Exception: To be compatible with zlib, we also need to
// accept degenerate single-code codings. See also
// TestDegenerateHuffmanCoding.
if code != 1<<uint(max) && !(code == 1 && max == 1) {
return false
}
h.min = min
if max > huffmanChunkBits {
numLinks := 1 << (uint(max) - huffmanChunkBits)
h.linkMask = uint32(numLinks - 1)
// create link tables
link := nextcode[huffmanChunkBits+1] >> 1
h.links = make([][]uint32, huffmanNumChunks-link)
for j := uint(link); j < huffmanNumChunks; j++ {
reverse := int(reverseByte[j>>8]) | int(reverseByte[j&0xff])<<8
reverse >>= uint(16 - huffmanChunkBits)
off := j - uint(link)
if sanity && h.chunks[reverse] != 0 {
panic("impossible: overwriting existing chunk")
}
h.chunks[reverse] = uint32(off<<huffmanValueShift | (huffmanChunkBits + 1))
h.links[off] = make([]uint32, numLinks)
}
}
for i, n := range bits {
if n == 0 {
continue
}
code := nextcode[n]
nextcode[n]++
chunk := uint32(i<<huffmanValueShift | n)
reverse := int(reverseByte[code>>8]) | int(reverseByte[code&0xff])<<8
reverse >>= uint(16 - n)
if n <= huffmanChunkBits {
for off := reverse; off < len(h.chunks); off += 1 << uint(n) {
// We should never need to overwrite
// an existing chunk. Also, 0 is
// never a valid chunk, because the
// lower 4 "count" bits should be
// between 1 and 15.
if sanity && h.chunks[off] != 0 {
panic("impossible: overwriting existing chunk")
}
h.chunks[off] = chunk
}
} else {
j := reverse & (huffmanNumChunks - 1)
if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 {
// Longer codes should have been
// associated with a link table above.
panic("impossible: not an indirect chunk")
}
value := h.chunks[j] >> huffmanValueShift
linktab := h.links[value]
reverse >>= huffmanChunkBits
for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) {
if sanity && linktab[off] != 0 {
panic("impossible: overwriting existing chunk")
}
linktab[off] = chunk
}
}
}
if sanity {
// Above we've sanity checked that we never overwrote
// an existing entry. Here we additionally check that
// we filled the tables completely.
for i, chunk := range h.chunks {
if chunk == 0 {
// As an exception, in the degenerate
// single-code case, we allow odd
// chunks to be missing.
if code == 1 && i%2 == 1 {
continue
}
panic("impossible: missing chunk")
}
}
for _, linktab := range h.links {
for _, chunk := range linktab {
if chunk == 0 {
panic("impossible: missing chunk")
}
}
}
}
return true
}
func main() {
flag.Parse()
var h huffmanDecoder
var bits [288]int
initReverseByte()
for i := 0; i < 144; i++ {
bits[i] = 8
}
for i := 144; i < 256; i++ {
bits[i] = 9
}
for i := 256; i < 280; i++ {
bits[i] = 7
}
for i := 280; i < 288; i++ {
bits[i] = 8
}
h.init(bits[:])
if h.links != nil {
log.Fatal("Unexpected links table in fixed Huffman decoder")
}
var buf bytes.Buffer
fmt.Fprintf(&buf, `// Copyright 2013 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.`+"\n\n")
fmt.Fprintln(&buf, "package flate")
fmt.Fprintln(&buf)
fmt.Fprintln(&buf, "// autogenerated by go run gen.go -output fixedhuff.go, DO NOT EDIT")
fmt.Fprintln(&buf)
fmt.Fprintln(&buf, "var fixedHuffmanDecoder = huffmanDecoder{")
fmt.Fprintf(&buf, "\t%d,\n", h.min)
fmt.Fprintln(&buf, "\t[huffmanNumChunks]uint32{")
for i := 0; i < huffmanNumChunks; i++ {
if i&7 == 0 {
fmt.Fprintf(&buf, "\t\t")
} else {
fmt.Fprintf(&buf, " ")
}
fmt.Fprintf(&buf, "0x%04x,", h.chunks[i])
if i&7 == 7 {
fmt.Fprintln(&buf)
}
}
fmt.Fprintln(&buf, "\t},")
fmt.Fprintln(&buf, "\tnil, 0,")
fmt.Fprintln(&buf, "}")
data, err := format.Source(buf.Bytes())
if err != nil {
log.Fatal(err)
}
err = ioutil.WriteFile(*filename, data, 0644)
if err != nil {
log.Fatal(err)
}
}
var reverseByte [256]byte
func initReverseByte() {
for x := 0; x < 256; x++ {
var result byte
for i := uint(0); i < 8; i++ {
result |= byte(((x >> i) & 1) << (7 - i))
}
reverseByte[x] = result
}
}

735
vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go generated vendored Normal file
View File

@@ -0,0 +1,735 @@
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package flate
import (
"io"
)
const (
// The largest offset code.
offsetCodeCount = 30
// The special code used to mark the end of a block.
endBlockMarker = 256
// The first length code.
lengthCodesStart = 257
// The number of codegen codes.
codegenCodeCount = 19
badCode = 255
// bufferFlushSize indicates the buffer size
// after which bytes are flushed to the writer.
// Should preferably be a multiple of 6, since
// we accumulate 6 bytes between writes to the buffer.
bufferFlushSize = 240
// bufferSize is the actual output byte buffer size.
// It must have additional headroom for a flush
// which can contain up to 8 bytes.
bufferSize = bufferFlushSize + 8
)
// The number of extra bits needed by length code X - LENGTH_CODES_START.
var lengthExtraBits = [32]int8{
/* 257 */ 0, 0, 0,
/* 260 */ 0, 0, 0, 0, 0, 1, 1, 1, 1, 2,
/* 270 */ 2, 2, 2, 3, 3, 3, 3, 4, 4, 4,
/* 280 */ 4, 5, 5, 5, 5, 0,
}
// The length indicated by length code X - LENGTH_CODES_START.
var lengthBase = [32]uint8{
0, 1, 2, 3, 4, 5, 6, 7, 8, 10,
12, 14, 16, 20, 24, 28, 32, 40, 48, 56,
64, 80, 96, 112, 128, 160, 192, 224, 255,
}
// offset code word extra bits.
var offsetExtraBits = [64]int8{
0, 0, 0, 0, 1, 1, 2, 2, 3, 3,
4, 4, 5, 5, 6, 6, 7, 7, 8, 8,
9, 9, 10, 10, 11, 11, 12, 12, 13, 13,
/* extended window */
14, 14, 15, 15, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20,
}
var offsetBase = [64]uint32{
/* normal deflate */
0x000000, 0x000001, 0x000002, 0x000003, 0x000004,
0x000006, 0x000008, 0x00000c, 0x000010, 0x000018,
0x000020, 0x000030, 0x000040, 0x000060, 0x000080,
0x0000c0, 0x000100, 0x000180, 0x000200, 0x000300,
0x000400, 0x000600, 0x000800, 0x000c00, 0x001000,
0x001800, 0x002000, 0x003000, 0x004000, 0x006000,
/* extended window */
0x008000, 0x00c000, 0x010000, 0x018000, 0x020000,
0x030000, 0x040000, 0x060000, 0x080000, 0x0c0000,
0x100000, 0x180000, 0x200000, 0x300000,
}
// The odd order in which the codegen code sizes are written.
var codegenOrder = []uint32{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
type huffmanBitWriter struct {
// writer is the underlying writer.
// Do not use it directly; use the write method, which ensures
// that Write errors are sticky.
writer io.Writer
// Data waiting to be written is bytes[0:nbytes]
// and then the low nbits of bits.
bits uint64
nbits uint
bytes [256]byte
codegenFreq [codegenCodeCount]int32
nbytes uint8
literalFreq []int32
offsetFreq []int32
codegen []uint8
literalEncoding *huffmanEncoder
offsetEncoding *huffmanEncoder
codegenEncoding *huffmanEncoder
err error
}
func newHuffmanBitWriter(w io.Writer) *huffmanBitWriter {
return &huffmanBitWriter{
writer: w,
literalFreq: make([]int32, lengthCodesStart+32),
offsetFreq: make([]int32, 32),
codegen: make([]uint8, maxNumLit+offsetCodeCount+1),
literalEncoding: newHuffmanEncoder(maxNumLit),
codegenEncoding: newHuffmanEncoder(codegenCodeCount),
offsetEncoding: newHuffmanEncoder(offsetCodeCount),
}
}
func (w *huffmanBitWriter) reset(writer io.Writer) {
w.writer = writer
w.bits, w.nbits, w.nbytes, w.err = 0, 0, 0, nil
w.bytes = [256]byte{}
}
func (w *huffmanBitWriter) flush() {
if w.err != nil {
w.nbits = 0
return
}
n := w.nbytes
for w.nbits != 0 {
w.bytes[n] = byte(w.bits)
w.bits >>= 8
if w.nbits > 8 { // Avoid underflow
w.nbits -= 8
} else {
w.nbits = 0
}
n++
}
w.bits = 0
w.write(w.bytes[:n])
w.nbytes = 0
}
func (w *huffmanBitWriter) write(b []byte) {
if w.err != nil {
return
}
_, w.err = w.writer.Write(b)
}
func (w *huffmanBitWriter) writeBits(b int32, nb uint) {
w.bits |= uint64(b) << w.nbits
w.nbits += nb
if w.nbits >= 48 {
bits := w.bits
w.bits >>= 48
w.nbits -= 48
n := w.nbytes
w.bytes[n] = byte(bits)
w.bytes[n+1] = byte(bits >> 8)
w.bytes[n+2] = byte(bits >> 16)
w.bytes[n+3] = byte(bits >> 24)
w.bytes[n+4] = byte(bits >> 32)
w.bytes[n+5] = byte(bits >> 40)
n += 6
if n >= bufferFlushSize {
if w.err != nil {
n = 0
return
}
w.write(w.bytes[:n])
n = 0
}
w.nbytes = n
}
}
func (w *huffmanBitWriter) writeBytes(bytes []byte) {
if w.err != nil {
return
}
n := w.nbytes
if w.nbits&7 != 0 {
w.err = InternalError("writeBytes with unfinished bits")
return
}
for w.nbits != 0 {
w.bytes[n] = byte(w.bits)
w.bits >>= 8
w.nbits -= 8
n++
}
if n != 0 {
w.write(w.bytes[:n])
}
w.nbytes = 0
w.write(bytes)
}
// RFC 1951 3.2.7 specifies a special run-length encoding for specifying
// the literal and offset lengths arrays (which are concatenated into a single
// array). This method generates that run-length encoding.
//
// The result is written into the codegen array, and the frequencies
// of each code is written into the codegenFreq array.
// Codes 0-15 are single byte codes. Codes 16-18 are followed by additional
// information. Code badCode is an end marker
//
// numLiterals The number of literals in literalEncoding
// numOffsets The number of offsets in offsetEncoding
// litenc, offenc The literal and offset encoder to use
func (w *huffmanBitWriter) generateCodegen(numLiterals int, numOffsets int, litEnc, offEnc *huffmanEncoder) {
for i := range w.codegenFreq {
w.codegenFreq[i] = 0
}
// Note that we are using codegen both as a temporary variable for holding
// a copy of the frequencies, and as the place where we put the result.
// This is fine because the output is always shorter than the input used
// so far.
codegen := w.codegen // cache
// Copy the concatenated code sizes to codegen. Put a marker at the end.
cgnl := codegen[:numLiterals]
for i := range cgnl {
cgnl[i] = uint8(litEnc.codes[i].len)
}
cgnl = codegen[numLiterals : numLiterals+numOffsets]
for i := range cgnl {
cgnl[i] = uint8(offEnc.codes[i].len)
}
codegen[numLiterals+numOffsets] = badCode
size := codegen[0]
count := 1
outIndex := 0
for inIndex := 1; size != badCode; inIndex++ {
// INVARIANT: We have seen "count" copies of size that have not yet
// had output generated for them.
nextSize := codegen[inIndex]
if nextSize == size {
count++
continue
}
// We need to generate codegen indicating "count" of size.
if size != 0 {
codegen[outIndex] = size
outIndex++
w.codegenFreq[size]++
count--
for count >= 3 {
n := 6
if n > count {
n = count
}
codegen[outIndex] = 16
outIndex++
codegen[outIndex] = uint8(n - 3)
outIndex++
w.codegenFreq[16]++
count -= n
}
} else {
for count >= 11 {
n := 138
if n > count {
n = count
}
codegen[outIndex] = 18
outIndex++
codegen[outIndex] = uint8(n - 11)
outIndex++
w.codegenFreq[18]++
count -= n
}
if count >= 3 {
// count >= 3 && count <= 10
codegen[outIndex] = 17
outIndex++
codegen[outIndex] = uint8(count - 3)
outIndex++
w.codegenFreq[17]++
count = 0
}
}
count--
for ; count >= 0; count-- {
codegen[outIndex] = size
outIndex++
w.codegenFreq[size]++
}
// Set up invariant for next time through the loop.
size = nextSize
count = 1
}
// Marker indicating the end of the codegen.
codegen[outIndex] = badCode
}
// dynamicSize returns the size of dynamically encoded data in bits.
func (w *huffmanBitWriter) dynamicSize(litEnc, offEnc *huffmanEncoder, extraBits int) (size, numCodegens int) {
numCodegens = len(w.codegenFreq)
for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 {
numCodegens--
}
header := 3 + 5 + 5 + 4 + (3 * numCodegens) +
w.codegenEncoding.bitLength(w.codegenFreq[:]) +
int(w.codegenFreq[16])*2 +
int(w.codegenFreq[17])*3 +
int(w.codegenFreq[18])*7
size = header +
litEnc.bitLength(w.literalFreq) +
offEnc.bitLength(w.offsetFreq) +
extraBits
return size, numCodegens
}
// fixedSize returns the size of dynamically encoded data in bits.
func (w *huffmanBitWriter) fixedSize(extraBits int) int {
return 3 +
fixedLiteralEncoding.bitLength(w.literalFreq) +
fixedOffsetEncoding.bitLength(w.offsetFreq) +
extraBits
}
// storedSize calculates the stored size, including header.
// The function returns the size in bits and whether the block
// fits inside a single block.
func (w *huffmanBitWriter) storedSize(in []byte) (int, bool) {
if in == nil {
return 0, false
}
if len(in) <= maxStoreBlockSize {
return (len(in) + 5) * 8, true
}
return 0, false
}
func (w *huffmanBitWriter) writeCode(c hcode) {
w.bits |= uint64(c.code) << w.nbits
w.nbits += uint(c.len)
if w.nbits >= 48 {
bits := w.bits
w.bits >>= 48
w.nbits -= 48
n := w.nbytes
w.bytes[n] = byte(bits)
w.bytes[n+1] = byte(bits >> 8)
w.bytes[n+2] = byte(bits >> 16)
w.bytes[n+3] = byte(bits >> 24)
w.bytes[n+4] = byte(bits >> 32)
w.bytes[n+5] = byte(bits >> 40)
n += 6
if n >= bufferFlushSize {
if w.err != nil {
n = 0
return
}
w.write(w.bytes[:n])
n = 0
}
w.nbytes = n
}
}
// Write the header of a dynamic Huffman block to the output stream.
//
// numLiterals The number of literals specified in codegen
// numOffsets The number of offsets specified in codegen
// numCodegens The number of codegens used in codegen
func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, numCodegens int, isEof bool) {
if w.err != nil {
return
}
var firstBits int32 = 4
if isEof {
firstBits = 5
}
w.writeBits(firstBits, 3)
w.writeBits(int32(numLiterals-257), 5)
w.writeBits(int32(numOffsets-1), 5)
w.writeBits(int32(numCodegens-4), 4)
for i := 0; i < numCodegens; i++ {
value := uint(w.codegenEncoding.codes[codegenOrder[i]].len)
w.writeBits(int32(value), 3)
}
i := 0
for {
var codeWord int = int(w.codegen[i])
i++
if codeWord == badCode {
break
}
w.writeCode(w.codegenEncoding.codes[uint32(codeWord)])
switch codeWord {
case 16:
w.writeBits(int32(w.codegen[i]), 2)
i++
break
case 17:
w.writeBits(int32(w.codegen[i]), 3)
i++
break
case 18:
w.writeBits(int32(w.codegen[i]), 7)
i++
break
}
}
}
func (w *huffmanBitWriter) writeStoredHeader(length int, isEof bool) {
if w.err != nil {
return
}
var flag int32
if isEof {
flag = 1
}
w.writeBits(flag, 3)
w.flush()
w.writeBits(int32(length), 16)
w.writeBits(int32(^uint16(length)), 16)
}
func (w *huffmanBitWriter) writeFixedHeader(isEof bool) {
if w.err != nil {
return
}
// Indicate that we are a fixed Huffman block
var value int32 = 2
if isEof {
value = 3
}
w.writeBits(value, 3)
}
// writeBlock will write a block of tokens with the smallest encoding.
// The original input can be supplied, and if the huffman encoded data
// is larger than the original bytes, the data will be written as a
// stored block.
// If the input is nil, the tokens will always be Huffman encoded.
func (w *huffmanBitWriter) writeBlock(tokens []token, eof bool, input []byte) {
if w.err != nil {
return
}
tokens = append(tokens, endBlockMarker)
numLiterals, numOffsets := w.indexTokens(tokens)
var extraBits int
storedSize, storable := w.storedSize(input)
if storable {
// We only bother calculating the costs of the extra bits required by
// the length of offset fields (which will be the same for both fixed
// and dynamic encoding), if we need to compare those two encodings
// against stored encoding.
for lengthCode := lengthCodesStart + 8; lengthCode < numLiterals; lengthCode++ {
// First eight length codes have extra size = 0.
extraBits += int(w.literalFreq[lengthCode]) * int(lengthExtraBits[lengthCode-lengthCodesStart])
}
for offsetCode := 4; offsetCode < numOffsets; offsetCode++ {
// First four offset codes have extra size = 0.
extraBits += int(w.offsetFreq[offsetCode]) * int(offsetExtraBits[offsetCode&63])
}
}
// Figure out smallest code.
// Fixed Huffman baseline.
var literalEncoding = fixedLiteralEncoding
var offsetEncoding = fixedOffsetEncoding
var size = w.fixedSize(extraBits)
// Dynamic Huffman?
var numCodegens int
// Generate codegen and codegenFrequencies, which indicates how to encode
// the literalEncoding and the offsetEncoding.
w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding)
w.codegenEncoding.generate(w.codegenFreq[:], 7)
dynamicSize, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits)
if dynamicSize < size {
size = dynamicSize
literalEncoding = w.literalEncoding
offsetEncoding = w.offsetEncoding
}
// Stored bytes?
if storable && storedSize < size {
w.writeStoredHeader(len(input), eof)
w.writeBytes(input)
return
}
// Huffman.
if literalEncoding == fixedLiteralEncoding {
w.writeFixedHeader(eof)
} else {
w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
}
// Write the tokens.
w.writeTokens(tokens, literalEncoding.codes, offsetEncoding.codes)
}
// writeBlockDynamic encodes a block using a dynamic Huffman table.
// This should be used if the symbols used have a disproportionate
// histogram distribution.
// If input is supplied and the compression savings are below 1/16th of the
// input size the block is stored.
func (w *huffmanBitWriter) writeBlockDynamic(tokens []token, eof bool, input []byte) {
if w.err != nil {
return
}
tokens = append(tokens, endBlockMarker)
numLiterals, numOffsets := w.indexTokens(tokens)
// Generate codegen and codegenFrequencies, which indicates how to encode
// the literalEncoding and the offsetEncoding.
w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding)
w.codegenEncoding.generate(w.codegenFreq[:], 7)
size, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, 0)
// Store bytes, if we don't get a reasonable improvement.
if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) {
w.writeStoredHeader(len(input), eof)
w.writeBytes(input)
return
}
// Write Huffman table.
w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
// Write the tokens.
w.writeTokens(tokens, w.literalEncoding.codes, w.offsetEncoding.codes)
}
// indexTokens indexes a slice of tokens, and updates
// literalFreq and offsetFreq, and generates literalEncoding
// and offsetEncoding.
// The number of literal and offset tokens is returned.
func (w *huffmanBitWriter) indexTokens(tokens []token) (numLiterals, numOffsets int) {
for i := range w.literalFreq {
w.literalFreq[i] = 0
}
for i := range w.offsetFreq {
w.offsetFreq[i] = 0
}
if len(tokens) == 0 {
return
}
// Only last token should be endBlockMarker.
if tokens[len(tokens)-1] == endBlockMarker {
w.literalFreq[endBlockMarker]++
tokens = tokens[:len(tokens)-1]
}
// Create slices up to the next power of two to avoid bounds checks.
lits := w.literalFreq[:256]
offs := w.offsetFreq[:32]
lengths := w.literalFreq[lengthCodesStart:]
lengths = lengths[:32]
for _, t := range tokens {
if t < endBlockMarker {
lits[t.literal()]++
continue
}
length := t.length()
offset := t.offset()
lengths[lengthCode(length)&31]++
offs[offsetCode(offset)&31]++
}
// get the number of literals
numLiterals = len(w.literalFreq)
for w.literalFreq[numLiterals-1] == 0 {
numLiterals--
}
// get the number of offsets
numOffsets = len(w.offsetFreq)
for numOffsets > 0 && w.offsetFreq[numOffsets-1] == 0 {
numOffsets--
}
if numOffsets == 0 {
// We haven't found a single match. If we want to go with the dynamic encoding,
// we should count at least one offset to be sure that the offset huffman tree could be encoded.
w.offsetFreq[0] = 1
numOffsets = 1
}
w.literalEncoding.generate(w.literalFreq[:maxNumLit], 15)
w.offsetEncoding.generate(w.offsetFreq[:offsetCodeCount], 15)
return
}
// writeTokens writes a slice of tokens to the output.
// codes for literal and offset encoding must be supplied.
func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) {
if w.err != nil {
return
}
if len(tokens) == 0 {
return
}
// Only last token should be endBlockMarker.
var deferEOB bool
if tokens[len(tokens)-1] == endBlockMarker {
tokens = tokens[:len(tokens)-1]
deferEOB = true
}
// Create slices up to the next power of two to avoid bounds checks.
lits := leCodes[:256]
offs := oeCodes[:32]
lengths := leCodes[lengthCodesStart:]
lengths = lengths[:32]
for _, t := range tokens {
if t < matchType {
w.writeCode(lits[t.literal()])
continue
}
// Write the length
length := t.length()
lengthCode := lengthCode(length)
w.writeCode(lengths[lengthCode&31])
extraLengthBits := uint(lengthExtraBits[lengthCode&31])
if extraLengthBits > 0 {
extraLength := int32(length - lengthBase[lengthCode&31])
w.writeBits(extraLength, extraLengthBits)
}
// Write the offset
offset := t.offset()
offsetCode := offsetCode(offset)
w.writeCode(offs[offsetCode&31])
extraOffsetBits := uint(offsetExtraBits[offsetCode&63])
if extraOffsetBits > 0 {
extraOffset := int32(offset - offsetBase[offsetCode&63])
w.writeBits(extraOffset, extraOffsetBits)
}
}
if deferEOB {
w.writeCode(leCodes[endBlockMarker])
}
}
// huffOffset is a static offset encoder used for huffman only encoding.
// It can be reused since we will not be encoding offset values.
var huffOffset *huffmanEncoder
func init() {
w := newHuffmanBitWriter(nil)
w.offsetFreq[0] = 1
huffOffset = newHuffmanEncoder(offsetCodeCount)
huffOffset.generate(w.offsetFreq[:offsetCodeCount], 15)
}
// writeBlockHuff encodes a block of bytes as either
// Huffman encoded literals or uncompressed bytes if the
// results only gains very little from compression.
func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte) {
if w.err != nil {
return
}
// Clear histogram
for i := range w.literalFreq {
w.literalFreq[i] = 0
}
// Add everything as literals
histogram(input, w.literalFreq)
w.literalFreq[endBlockMarker] = 1
const numLiterals = endBlockMarker + 1
const numOffsets = 1
w.literalEncoding.generate(w.literalFreq[:maxNumLit], 15)
// Figure out smallest code.
// Always use dynamic Huffman or Store
var numCodegens int
// Generate codegen and codegenFrequencies, which indicates how to encode
// the literalEncoding and the offsetEncoding.
w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, huffOffset)
w.codegenEncoding.generate(w.codegenFreq[:], 7)
size, numCodegens := w.dynamicSize(w.literalEncoding, huffOffset, 0)
// Store bytes, if we don't get a reasonable improvement.
if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) {
w.writeStoredHeader(len(input), eof)
w.writeBytes(input)
return
}
// Huffman.
w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
encoding := w.literalEncoding.codes[:257]
n := w.nbytes
for _, t := range input {
// Bitwriting inlined, ~30% speedup
c := encoding[t]
w.bits |= uint64(c.code) << w.nbits
w.nbits += uint(c.len)
if w.nbits < 48 {
continue
}
// Store 6 bytes
bits := w.bits
w.bits >>= 48
w.nbits -= 48
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 {
continue
}
w.write(w.bytes[:n])
if w.err != nil {
return // Return early in the event of write failures
}
n = 0
}
w.nbytes = n
w.writeCode(encoding[endBlockMarker])
}

347
vendor/github.com/klauspost/compress/flate/huffman_code.go generated vendored Normal file
View File

@@ -0,0 +1,347 @@
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package flate
import (
"math"
"math/bits"
"sort"
)
// hcode is a huffman code with a bit code and bit length.
type hcode struct {
code, len uint16
}
type huffmanEncoder struct {
codes []hcode
freqcache []literalNode
bitCount [17]int32
lns byLiteral // stored to avoid repeated allocation in generate
lfs byFreq // stored to avoid repeated allocation in generate
}
type literalNode struct {
literal uint16
freq int32
}
// A levelInfo describes the state of the constructed tree for a given depth.
type levelInfo struct {
// Our level. for better printing
level int32
// The frequency of the last node at this level
lastFreq int32
// The frequency of the next character to add to this level
nextCharFreq int32
// The frequency of the next pair (from level below) to add to this level.
// Only valid if the "needed" value of the next lower level is 0.
nextPairFreq int32
// The number of chains remaining to generate for this level before moving
// up to the next level
needed int32
}
// set sets the code and length of an hcode.
func (h *hcode) set(code uint16, length uint16) {
h.len = length
h.code = code
}
func maxNode() literalNode { return literalNode{math.MaxUint16, math.MaxInt32} }
func newHuffmanEncoder(size int) *huffmanEncoder {
// Make capacity to next power of two.
c := uint(bits.Len32(uint32(size - 1)))
return &huffmanEncoder{codes: make([]hcode, size, 1<<c)}
}
// Generates a HuffmanCode corresponding to the fixed literal table
func generateFixedLiteralEncoding() *huffmanEncoder {
h := newHuffmanEncoder(maxNumLit)
codes := h.codes
var ch uint16
for ch = 0; ch < maxNumLit; ch++ {
var bits uint16
var size uint16
switch {
case ch < 144:
// size 8, 000110000 .. 10111111
bits = ch + 48
size = 8
break
case ch < 256:
// size 9, 110010000 .. 111111111
bits = ch + 400 - 144
size = 9
break
case ch < 280:
// size 7, 0000000 .. 0010111
bits = ch - 256
size = 7
break
default:
// size 8, 11000000 .. 11000111
bits = ch + 192 - 280
size = 8
}
codes[ch] = hcode{code: reverseBits(bits, byte(size)), len: size}
}
return h
}
func generateFixedOffsetEncoding() *huffmanEncoder {
h := newHuffmanEncoder(30)
codes := h.codes
for ch := range codes {
codes[ch] = hcode{code: reverseBits(uint16(ch), 5), len: 5}
}
return h
}
var fixedLiteralEncoding *huffmanEncoder = generateFixedLiteralEncoding()
var fixedOffsetEncoding *huffmanEncoder = generateFixedOffsetEncoding()
func (h *huffmanEncoder) bitLength(freq []int32) int {
var total int
for i, f := range freq {
if f != 0 {
total += int(f) * int(h.codes[i].len)
}
}
return total
}
const maxBitsLimit = 16
// Return the number of literals assigned to each bit size in the Huffman encoding
//
// This method is only called when list.length >= 3
// The cases of 0, 1, and 2 literals are handled by special case code.
//
// list An array of the literals with non-zero frequencies
// and their associated frequencies. The array is in order of increasing
// frequency, and has as its last element a special element with frequency
// MaxInt32
// maxBits The maximum number of bits that should be used to encode any literal.
// Must be less than 16.
// return An integer array in which array[i] indicates the number of literals
// that should be encoded in i bits.
func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 {
if maxBits >= maxBitsLimit {
panic("flate: maxBits too large")
}
n := int32(len(list))
list = list[0 : n+1]
list[n] = maxNode()
// The tree can't have greater depth than n - 1, no matter what. This
// saves a little bit of work in some small cases
if maxBits > n-1 {
maxBits = n - 1
}
// Create information about each of the levels.
// A bogus "Level 0" whose sole purpose is so that
// level1.prev.needed==0. This makes level1.nextPairFreq
// be a legitimate value that never gets chosen.
var levels [maxBitsLimit]levelInfo
// leafCounts[i] counts the number of literals at the left
// of ancestors of the rightmost node at level i.
// leafCounts[i][j] is the number of literals at the left
// of the level j ancestor.
var leafCounts [maxBitsLimit][maxBitsLimit]int32
for level := int32(1); level <= maxBits; level++ {
// For every level, the first two items are the first two characters.
// We initialize the levels as if we had already figured this out.
levels[level] = levelInfo{
level: level,
lastFreq: list[1].freq,
nextCharFreq: list[2].freq,
nextPairFreq: list[0].freq + list[1].freq,
}
leafCounts[level][level] = 2
if level == 1 {
levels[level].nextPairFreq = math.MaxInt32
}
}
// We need a total of 2*n - 2 items at top level and have already generated 2.
levels[maxBits].needed = 2*n - 4
level := maxBits
for {
l := &levels[level]
if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 {
// We've run out of both leafs and pairs.
// End all calculations for this level.
// To make sure we never come back to this level or any lower level,
// set nextPairFreq impossibly large.
l.needed = 0
levels[level+1].nextPairFreq = math.MaxInt32
level++
continue
}
prevFreq := l.lastFreq
if l.nextCharFreq < l.nextPairFreq {
// The next item on this row is a leaf node.
n := leafCounts[level][level] + 1
l.lastFreq = l.nextCharFreq
// Lower leafCounts are the same of the previous node.
leafCounts[level][level] = n
l.nextCharFreq = list[n].freq
} else {
// The next item on this row is a pair from the previous row.
// nextPairFreq isn't valid until we generate two
// more values in the level below
l.lastFreq = l.nextPairFreq
// Take leaf counts from the lower level, except counts[level] remains the same.
copy(leafCounts[level][:level], leafCounts[level-1][:level])
levels[l.level-1].needed = 2
}
if l.needed--; l.needed == 0 {
// We've done everything we need to do for this level.
// Continue calculating one level up. Fill in nextPairFreq
// of that level with the sum of the two nodes we've just calculated on
// this level.
if l.level == maxBits {
// All done!
break
}
levels[l.level+1].nextPairFreq = prevFreq + l.lastFreq
level++
} else {
// If we stole from below, move down temporarily to replenish it.
for levels[level-1].needed > 0 {
level--
}
}
}
// Somethings is wrong if at the end, the top level is null or hasn't used
// all of the leaves.
if leafCounts[maxBits][maxBits] != n {
panic("leafCounts[maxBits][maxBits] != n")
}
bitCount := h.bitCount[:maxBits+1]
bits := 1
counts := &leafCounts[maxBits]
for level := maxBits; level > 0; level-- {
// chain.leafCount gives the number of literals requiring at least "bits"
// bits to encode.
bitCount[bits] = counts[level] - counts[level-1]
bits++
}
return bitCount
}
// Look at the leaves and assign them a bit count and an encoding as specified
// in RFC 1951 3.2.2
func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalNode) {
code := uint16(0)
for n, bits := range bitCount {
code <<= 1
if n == 0 || bits == 0 {
continue
}
// The literals list[len(list)-bits] .. list[len(list)-bits]
// are encoded using "bits" bits, and get the values
// code, code + 1, .... The code values are
// assigned in literal order (not frequency order).
chunk := list[len(list)-int(bits):]
h.lns.sort(chunk)
for _, node := range chunk {
h.codes[node.literal] = hcode{code: reverseBits(code, uint8(n)), len: uint16(n)}
code++
}
list = list[0 : len(list)-int(bits)]
}
}
// Update this Huffman Code object to be the minimum code for the specified frequency count.
//
// freq An array of frequencies, in which frequency[i] gives the frequency of literal i.
// maxBits The maximum number of bits to use for any literal.
func (h *huffmanEncoder) generate(freq []int32, maxBits int32) {
if h.freqcache == nil {
// Allocate a reusable buffer with the longest possible frequency table.
// Possible lengths are codegenCodeCount, offsetCodeCount and maxNumLit.
// The largest of these is maxNumLit, so we allocate for that case.
h.freqcache = make([]literalNode, maxNumLit+1)
}
list := h.freqcache[:len(freq)+1]
// Number of non-zero literals
count := 0
// Set list to be the set of all non-zero literals and their frequencies
for i, f := range freq {
if f != 0 {
list[count] = literalNode{uint16(i), f}
count++
} else {
list[count] = literalNode{}
h.codes[i].len = 0
}
}
list[len(freq)] = literalNode{}
list = list[:count]
if count <= 2 {
// Handle the small cases here, because they are awkward for the general case code. With
// two or fewer literals, everything has bit length 1.
for i, node := range list {
// "list" is in order of increasing literal value.
h.codes[node.literal].set(uint16(i), 1)
}
return
}
h.lfs.sort(list)
// Get the number of literals for each bit count
bitCount := h.bitCounts(list, maxBits)
// And do the assignment
h.assignEncodingAndSize(bitCount, list)
}
type byLiteral []literalNode
func (s *byLiteral) sort(a []literalNode) {
*s = byLiteral(a)
sort.Sort(s)
}
func (s byLiteral) Len() int { return len(s) }
func (s byLiteral) Less(i, j int) bool {
return s[i].literal < s[j].literal
}
func (s byLiteral) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
type byFreq []literalNode
func (s *byFreq) sort(a []literalNode) {
*s = byFreq(a)
sort.Sort(s)
}
func (s byFreq) Len() int { return len(s) }
func (s byFreq) Less(i, j int) bool {
if s[i].freq == s[j].freq {
return s[i].literal < s[j].literal
}
return s[i].freq < s[j].freq
}
func (s byFreq) Swap(i, j int) { s[i], s[j] = s[j], s[i] }

880
vendor/github.com/klauspost/compress/flate/inflate.go generated vendored Normal file
View File

@@ -0,0 +1,880 @@
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Package flate implements the DEFLATE compressed data format, described in
// RFC 1951. The gzip and zlib packages implement access to DEFLATE-based file
// formats.
package flate
import (
"bufio"
"io"
"math/bits"
"strconv"
"sync"
)
const (
maxCodeLen = 16 // max length of Huffman code
maxCodeLenMask = 15 // mask for max length of Huffman code
// The next three numbers come from the RFC section 3.2.7, with the
// additional proviso in section 3.2.5 which implies that distance codes
// 30 and 31 should never occur in compressed data.
maxNumLit = 286
maxNumDist = 30
numCodes = 19 // number of codes in Huffman meta-code
)
// Initialize the fixedHuffmanDecoder only once upon first use.
var fixedOnce sync.Once
var fixedHuffmanDecoder huffmanDecoder
// A CorruptInputError reports the presence of corrupt input at a given offset.
type CorruptInputError int64
func (e CorruptInputError) Error() string {
return "flate: corrupt input before offset " + strconv.FormatInt(int64(e), 10)
}
// An InternalError reports an error in the flate code itself.
type InternalError string
func (e InternalError) Error() string { return "flate: internal error: " + string(e) }
// A ReadError reports an error encountered while reading input.
//
// Deprecated: No longer returned.
type ReadError struct {
Offset int64 // byte offset where error occurred
Err error // error returned by underlying Read
}
func (e *ReadError) Error() string {
return "flate: read error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error()
}
// A WriteError reports an error encountered while writing output.
//
// Deprecated: No longer returned.
type WriteError struct {
Offset int64 // byte offset where error occurred
Err error // error returned by underlying Write
}
func (e *WriteError) Error() string {
return "flate: write error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error()
}
// Resetter resets a ReadCloser returned by NewReader or NewReaderDict to
// to switch to a new underlying Reader. This permits reusing a ReadCloser
// instead of allocating a new one.
type Resetter interface {
// Reset discards any buffered data and resets the Resetter as if it was
// newly initialized with the given reader.
Reset(r io.Reader, dict []byte) error
}
// The data structure for decoding Huffman tables is based on that of
// zlib. There is a lookup table of a fixed bit width (huffmanChunkBits),
// For codes smaller than the table width, there are multiple entries
// (each combination of trailing bits has the same value). For codes
// larger than the table width, the table contains a link to an overflow
// table. The width of each entry in the link table is the maximum code
// size minus the chunk width.
//
// Note that you can do a lookup in the table even without all bits
// filled. Since the extra bits are zero, and the DEFLATE Huffman codes
// have the property that shorter codes come before longer ones, the
// bit length estimate in the result is a lower bound on the actual
// number of bits.
//
// See the following:
// http://www.gzip.org/algorithm.txt
// chunk & 15 is number of bits
// chunk >> 4 is value, including table link
const (
huffmanChunkBits = 9
huffmanNumChunks = 1 << huffmanChunkBits
huffmanCountMask = 15
huffmanValueShift = 4
)
type huffmanDecoder struct {
min int // the minimum code length
chunks *[huffmanNumChunks]uint32 // chunks as described above
links [][]uint32 // overflow links
linkMask uint32 // mask the width of the link table
}
// Initialize Huffman decoding tables from array of code lengths.
// Following this function, h is guaranteed to be initialized into a complete
// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a
// degenerate case where the tree has only a single symbol with length 1. Empty
// trees are permitted.
func (h *huffmanDecoder) init(lengths []int) bool {
// Sanity enables additional runtime tests during Huffman
// table construction. It's intended to be used during
// development to supplement the currently ad-hoc unit tests.
const sanity = false
if h.chunks == nil {
h.chunks = &[huffmanNumChunks]uint32{}
}
if h.min != 0 {
*h = huffmanDecoder{chunks: h.chunks, links: h.links}
}
// Count number of codes of each length,
// compute min and max length.
var count [maxCodeLen]int
var min, max int
for _, n := range lengths {
if n == 0 {
continue
}
if min == 0 || n < min {
min = n
}
if n > max {
max = n
}
count[n&maxCodeLenMask]++
}
// Empty tree. The decompressor.huffSym function will fail later if the tree
// is used. Technically, an empty tree is only valid for the HDIST tree and
// not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree
// is guaranteed to fail since it will attempt to use the tree to decode the
// codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is
// guaranteed to fail later since the compressed data section must be
// composed of at least one symbol (the end-of-block marker).
if max == 0 {
return true
}
code := 0
var nextcode [maxCodeLen]int
for i := min; i <= max; i++ {
code <<= 1
nextcode[i&maxCodeLenMask] = code
code += count[i&maxCodeLenMask]
}
// Check that the coding is complete (i.e., that we've
// assigned all 2-to-the-max possible bit sequences).
// Exception: To be compatible with zlib, we also need to
// accept degenerate single-code codings. See also
// TestDegenerateHuffmanCoding.
if code != 1<<uint(max) && !(code == 1 && max == 1) {
return false
}
h.min = min
chunks := h.chunks[:]
for i := range chunks {
chunks[i] = 0
}
if max > huffmanChunkBits {
numLinks := 1 << (uint(max) - huffmanChunkBits)
h.linkMask = uint32(numLinks - 1)
// create link tables
link := nextcode[huffmanChunkBits+1] >> 1
if cap(h.links) < huffmanNumChunks-link {
h.links = make([][]uint32, huffmanNumChunks-link)
} else {
h.links = h.links[:huffmanNumChunks-link]
}
for j := uint(link); j < huffmanNumChunks; j++ {
reverse := int(bits.Reverse16(uint16(j)))
reverse >>= uint(16 - huffmanChunkBits)
off := j - uint(link)
if sanity && h.chunks[reverse] != 0 {
panic("impossible: overwriting existing chunk")
}
h.chunks[reverse] = uint32(off<<huffmanValueShift | (huffmanChunkBits + 1))
if cap(h.links[off]) < numLinks {
h.links[off] = make([]uint32, numLinks)
} else {
links := h.links[off][:0]
h.links[off] = links[:numLinks]
}
}
} else {
h.links = h.links[:0]
}
for i, n := range lengths {
if n == 0 {
continue
}
code := nextcode[n]
nextcode[n]++
chunk := uint32(i<<huffmanValueShift | n)
reverse := int(bits.Reverse16(uint16(code)))
reverse >>= uint(16 - n)
if n <= huffmanChunkBits {
for off := reverse; off < len(h.chunks); off += 1 << uint(n) {
// We should never need to overwrite
// an existing chunk. Also, 0 is
// never a valid chunk, because the
// lower 4 "count" bits should be
// between 1 and 15.
if sanity && h.chunks[off] != 0 {
panic("impossible: overwriting existing chunk")
}
h.chunks[off] = chunk
}
} else {
j := reverse & (huffmanNumChunks - 1)
if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 {
// Longer codes should have been
// associated with a link table above.
panic("impossible: not an indirect chunk")
}
value := h.chunks[j] >> huffmanValueShift
linktab := h.links[value]
reverse >>= huffmanChunkBits
for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) {
if sanity && linktab[off] != 0 {
panic("impossible: overwriting existing chunk")
}
linktab[off] = chunk
}
}
}
if sanity {
// Above we've sanity checked that we never overwrote
// an existing entry. Here we additionally check that
// we filled the tables completely.
for i, chunk := range h.chunks {
if chunk == 0 {
// As an exception, in the degenerate
// single-code case, we allow odd
// chunks to be missing.
if code == 1 && i%2 == 1 {
continue
}
panic("impossible: missing chunk")
}
}
for _, linktab := range h.links {
for _, chunk := range linktab {
if chunk == 0 {
panic("impossible: missing chunk")
}
}
}
}
return true
}
// The actual read interface needed by NewReader.
// If the passed in io.Reader does not also have ReadByte,
// the NewReader will introduce its own buffering.
type Reader interface {
io.Reader
io.ByteReader
}
// Decompress state.
type decompressor struct {
// Input source.
r Reader
roffset int64
// Input bits, in top of b.
b uint32
nb uint
// Huffman decoders for literal/length, distance.
h1, h2 huffmanDecoder
// Length arrays used to define Huffman codes.
bits *[maxNumLit + maxNumDist]int
codebits *[numCodes]int
// Output history, buffer.
dict dictDecoder
// Temporary buffer (avoids repeated allocation).
buf [4]byte
// Next step in the decompression,
// and decompression state.
step func(*decompressor)
stepState int
final bool
err error
toRead []byte
hl, hd *huffmanDecoder
copyLen int
copyDist int
}
func (f *decompressor) nextBlock() {
for f.nb < 1+2 {
if f.err = f.moreBits(); f.err != nil {
return
}
}
f.final = f.b&1 == 1
f.b >>= 1
typ := f.b & 3
f.b >>= 2
f.nb -= 1 + 2
switch typ {
case 0:
f.dataBlock()
case 1:
// compressed, fixed Huffman tables
f.hl = &fixedHuffmanDecoder
f.hd = nil
f.huffmanBlock()
case 2:
// compressed, dynamic Huffman tables
if f.err = f.readHuffman(); f.err != nil {
break
}
f.hl = &f.h1
f.hd = &f.h2
f.huffmanBlock()
default:
// 3 is reserved.
f.err = CorruptInputError(f.roffset)
}
}
func (f *decompressor) Read(b []byte) (int, error) {
for {
if len(f.toRead) > 0 {
n := copy(b, f.toRead)
f.toRead = f.toRead[n:]
if len(f.toRead) == 0 {
return n, f.err
}
return n, nil
}
if f.err != nil {
return 0, f.err
}
f.step(f)
if f.err != nil && len(f.toRead) == 0 {
f.toRead = f.dict.readFlush() // Flush what's left in case of error
}
}
}
// Support the io.WriteTo interface for io.Copy and friends.
func (f *decompressor) WriteTo(w io.Writer) (int64, error) {
total := int64(0)
flushed := false
for {
if len(f.toRead) > 0 {
n, err := w.Write(f.toRead)
total += int64(n)
if err != nil {
f.err = err
return total, err
}
if n != len(f.toRead) {
return total, io.ErrShortWrite
}
f.toRead = f.toRead[:0]
}
if f.err != nil && flushed {
if f.err == io.EOF {
return total, nil
}
return total, f.err
}
if f.err == nil {
f.step(f)
}
if len(f.toRead) == 0 && f.err != nil && !flushed {
f.toRead = f.dict.readFlush() // Flush what's left in case of error
flushed = true
}
}
}
func (f *decompressor) Close() error {
if f.err == io.EOF {
return nil
}
return f.err
}
// RFC 1951 section 3.2.7.
// Compression with dynamic Huffman codes
var codeOrder = [...]int{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
func (f *decompressor) readHuffman() error {
// HLIT[5], HDIST[5], HCLEN[4].
for f.nb < 5+5+4 {
if err := f.moreBits(); err != nil {
return err
}
}
nlit := int(f.b&0x1F) + 257
if nlit > maxNumLit {
return CorruptInputError(f.roffset)
}
f.b >>= 5
ndist := int(f.b&0x1F) + 1
if ndist > maxNumDist {
return CorruptInputError(f.roffset)
}
f.b >>= 5
nclen := int(f.b&0xF) + 4
// numCodes is 19, so nclen is always valid.
f.b >>= 4
f.nb -= 5 + 5 + 4
// (HCLEN+4)*3 bits: code lengths in the magic codeOrder order.
for i := 0; i < nclen; i++ {
for f.nb < 3 {
if err := f.moreBits(); err != nil {
return err
}
}
f.codebits[codeOrder[i]] = int(f.b & 0x7)
f.b >>= 3
f.nb -= 3
}
for i := nclen; i < len(codeOrder); i++ {
f.codebits[codeOrder[i]] = 0
}
if !f.h1.init(f.codebits[0:]) {
return CorruptInputError(f.roffset)
}
// HLIT + 257 code lengths, HDIST + 1 code lengths,
// using the code length Huffman code.
for i, n := 0, nlit+ndist; i < n; {
x, err := f.huffSym(&f.h1)
if err != nil {
return err
}
if x < 16 {
// Actual length.
f.bits[i] = x
i++
continue
}
// Repeat previous length or zero.
var rep int
var nb uint
var b int
switch x {
default:
return InternalError("unexpected length code")
case 16:
rep = 3
nb = 2
if i == 0 {
return CorruptInputError(f.roffset)
}
b = f.bits[i-1]
case 17:
rep = 3
nb = 3
b = 0
case 18:
rep = 11
nb = 7
b = 0
}
for f.nb < nb {
if err := f.moreBits(); err != nil {
return err
}
}
rep += int(f.b & uint32(1<<nb-1))
f.b >>= nb
f.nb -= nb
if i+rep > n {
return CorruptInputError(f.roffset)
}
for j := 0; j < rep; j++ {
f.bits[i] = b
i++
}
}
if !f.h1.init(f.bits[0:nlit]) || !f.h2.init(f.bits[nlit:nlit+ndist]) {
return CorruptInputError(f.roffset)
}
// As an optimization, we can initialize the min bits to read at a time
// for the HLIT tree to the length of the EOB marker since we know that
// every block must terminate with one. This preserves the property that
// we never read any extra bytes after the end of the DEFLATE stream.
if f.h1.min < f.bits[endBlockMarker] {
f.h1.min = f.bits[endBlockMarker]
}
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) huffmanBlock() {
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.
{
v, err := f.huffSym(f.hl)
if err != nil {
f.err = err
return
}
var n uint // number of bits extra
var length int
switch {
case v < 256:
f.dict.writeByte(byte(v))
if f.dict.availWrite() == 0 {
f.toRead = f.dict.readFlush()
f.step = (*decompressor).huffmanBlock
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:
f.err = CorruptInputError(f.roffset)
return
}
if n > 0 {
for f.nb < n {
if err = f.moreBits(); err != nil {
f.err = err
return
}
}
length += int(f.b & uint32(1<<n-1))
f.b >>= n
f.nb -= n
}
var dist int
if f.hd == nil {
for f.nb < 5 {
if err = f.moreBits(); err != nil {
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 {
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 {
f.err = err
return
}
}
extra |= int(f.b & uint32(1<<nb-1))
f.b >>= nb
f.nb -= nb
dist = 1<<(nb+1) + 1 + extra
default:
f.err = CorruptInputError(f.roffset)
return
}
// No check on length; encoding can be prescient.
if 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).huffmanBlock // We need to continue this work
f.stepState = stateDict
return
}
goto readLiteral
}
}
// Copy a single uncompressed data block from input to output.
func (f *decompressor) dataBlock() {
// Uncompressed.
// Discard current half-byte.
f.nb = 0
f.b = 0
// Length then ones-complement of length.
nr, err := io.ReadFull(f.r, f.buf[0:4])
f.roffset += int64(nr)
if err != nil {
f.err = noEOF(err)
return
}
n := int(f.buf[0]) | int(f.buf[1])<<8
nn := int(f.buf[2]) | int(f.buf[3])<<8
if uint16(nn) != uint16(^n) {
f.err = CorruptInputError(f.roffset)
return
}
if n == 0 {
f.toRead = f.dict.readFlush()
f.finishBlock()
return
}
f.copyLen = n
f.copyData()
}
// copyData copies f.copyLen bytes from the underlying reader into f.hist.
// It pauses for reads when f.hist is full.
func (f *decompressor) copyData() {
buf := f.dict.writeSlice()
if len(buf) > f.copyLen {
buf = buf[:f.copyLen]
}
cnt, err := io.ReadFull(f.r, buf)
f.roffset += int64(cnt)
f.copyLen -= cnt
f.dict.writeMark(cnt)
if err != nil {
f.err = noEOF(err)
return
}
if f.dict.availWrite() == 0 || f.copyLen > 0 {
f.toRead = f.dict.readFlush()
f.step = (*decompressor).copyData
return
}
f.finishBlock()
}
func (f *decompressor) finishBlock() {
if f.final {
if f.dict.availRead() > 0 {
f.toRead = f.dict.readFlush()
}
f.err = io.EOF
}
f.step = (*decompressor).nextBlock
}
// noEOF returns err, unless err == io.EOF, in which case it returns io.ErrUnexpectedEOF.
func noEOF(e error) error {
if e == io.EOF {
return io.ErrUnexpectedEOF
}
return e
}
func (f *decompressor) moreBits() error {
c, err := f.r.ReadByte()
if err != nil {
return noEOF(err)
}
f.roffset++
f.b |= uint32(c) << f.nb
f.nb += 8
return nil
}
// Read the next Huffman-encoded symbol from f according to h.
func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) {
// Since a huffmanDecoder can be empty or be composed of a degenerate tree
// with single element, huffSym must error on these two edge cases. In both
// cases, the chunks slice will be 0 for the invalid sequence, leading it
// satisfy the n == 0 check below.
n := uint(h.min)
// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
// but is smart enough to keep local variables in registers, so use nb and b,
// inline call to moreBits and reassign b,nb back to f on return.
nb, b := f.nb, f.b
for {
for nb < n {
c, err := f.r.ReadByte()
if err != nil {
f.b = b
f.nb = nb
return 0, noEOF(err)
}
f.roffset++
b |= uint32(c) << (nb & 31)
nb += 8
}
chunk := h.chunks[b&(huffmanNumChunks-1)]
n = uint(chunk & huffmanCountMask)
if n > huffmanChunkBits {
chunk = h.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&h.linkMask]
n = uint(chunk & huffmanCountMask)
}
if n <= nb {
if n == 0 {
f.b = b
f.nb = nb
f.err = CorruptInputError(f.roffset)
return 0, f.err
}
f.b = b >> (n & 31)
f.nb = nb - n
return int(chunk >> huffmanValueShift), nil
}
}
}
func makeReader(r io.Reader) Reader {
if rr, ok := r.(Reader); ok {
return rr
}
return bufio.NewReader(r)
}
func fixedHuffmanDecoderInit() {
fixedOnce.Do(func() {
// These come from the RFC section 3.2.6.
var bits [288]int
for i := 0; i < 144; i++ {
bits[i] = 8
}
for i := 144; i < 256; i++ {
bits[i] = 9
}
for i := 256; i < 280; i++ {
bits[i] = 7
}
for i := 280; i < 288; i++ {
bits[i] = 8
}
fixedHuffmanDecoder.init(bits[:])
})
}
func (f *decompressor) Reset(r io.Reader, dict []byte) error {
*f = decompressor{
r: makeReader(r),
bits: f.bits,
codebits: f.codebits,
h1: f.h1,
h2: f.h2,
dict: f.dict,
step: (*decompressor).nextBlock,
}
f.dict.init(maxMatchOffset, dict)
return nil
}
// NewReader returns a new ReadCloser that can be used
// to read the uncompressed version of r.
// If r does not also implement io.ByteReader,
// the decompressor may read more data than necessary from r.
// It is the caller's responsibility to call Close on the ReadCloser
// when finished reading.
//
// The ReadCloser returned by NewReader also implements Resetter.
func NewReader(r io.Reader) io.ReadCloser {
fixedHuffmanDecoderInit()
var f decompressor
f.r = makeReader(r)
f.bits = new([maxNumLit + maxNumDist]int)
f.codebits = new([numCodes]int)
f.step = (*decompressor).nextBlock
f.dict.init(maxMatchOffset, nil)
return &f
}
// NewReaderDict is like NewReader but initializes the reader
// with a preset dictionary. The returned Reader behaves as if
// the uncompressed data stream started with the given dictionary,
// which has already been read. NewReaderDict is typically used
// to read data compressed by NewWriterDict.
//
// The ReadCloser returned by NewReader also implements Resetter.
func NewReaderDict(r io.Reader, dict []byte) io.ReadCloser {
fixedHuffmanDecoderInit()
var f decompressor
f.r = makeReader(r)
f.bits = new([maxNumLit + maxNumDist]int)
f.codebits = new([numCodes]int)
f.step = (*decompressor).nextBlock
f.dict.init(maxMatchOffset, dict)
return &f
}

48
vendor/github.com/klauspost/compress/flate/reverse_bits.go generated vendored Normal file
View File

@@ -0,0 +1,48 @@
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package flate
var reverseByte = [256]byte{
0x00, 0x80, 0x40, 0xc0, 0x20, 0xa0, 0x60, 0xe0,
0x10, 0x90, 0x50, 0xd0, 0x30, 0xb0, 0x70, 0xf0,
0x08, 0x88, 0x48, 0xc8, 0x28, 0xa8, 0x68, 0xe8,
0x18, 0x98, 0x58, 0xd8, 0x38, 0xb8, 0x78, 0xf8,
0x04, 0x84, 0x44, 0xc4, 0x24, 0xa4, 0x64, 0xe4,
0x14, 0x94, 0x54, 0xd4, 0x34, 0xb4, 0x74, 0xf4,
0x0c, 0x8c, 0x4c, 0xcc, 0x2c, 0xac, 0x6c, 0xec,
0x1c, 0x9c, 0x5c, 0xdc, 0x3c, 0xbc, 0x7c, 0xfc,
0x02, 0x82, 0x42, 0xc2, 0x22, 0xa2, 0x62, 0xe2,
0x12, 0x92, 0x52, 0xd2, 0x32, 0xb2, 0x72, 0xf2,
0x0a, 0x8a, 0x4a, 0xca, 0x2a, 0xaa, 0x6a, 0xea,
0x1a, 0x9a, 0x5a, 0xda, 0x3a, 0xba, 0x7a, 0xfa,
0x06, 0x86, 0x46, 0xc6, 0x26, 0xa6, 0x66, 0xe6,
0x16, 0x96, 0x56, 0xd6, 0x36, 0xb6, 0x76, 0xf6,
0x0e, 0x8e, 0x4e, 0xce, 0x2e, 0xae, 0x6e, 0xee,
0x1e, 0x9e, 0x5e, 0xde, 0x3e, 0xbe, 0x7e, 0xfe,
0x01, 0x81, 0x41, 0xc1, 0x21, 0xa1, 0x61, 0xe1,
0x11, 0x91, 0x51, 0xd1, 0x31, 0xb1, 0x71, 0xf1,
0x09, 0x89, 0x49, 0xc9, 0x29, 0xa9, 0x69, 0xe9,
0x19, 0x99, 0x59, 0xd9, 0x39, 0xb9, 0x79, 0xf9,
0x05, 0x85, 0x45, 0xc5, 0x25, 0xa5, 0x65, 0xe5,
0x15, 0x95, 0x55, 0xd5, 0x35, 0xb5, 0x75, 0xf5,
0x0d, 0x8d, 0x4d, 0xcd, 0x2d, 0xad, 0x6d, 0xed,
0x1d, 0x9d, 0x5d, 0xdd, 0x3d, 0xbd, 0x7d, 0xfd,
0x03, 0x83, 0x43, 0xc3, 0x23, 0xa3, 0x63, 0xe3,
0x13, 0x93, 0x53, 0xd3, 0x33, 0xb3, 0x73, 0xf3,
0x0b, 0x8b, 0x4b, 0xcb, 0x2b, 0xab, 0x6b, 0xeb,
0x1b, 0x9b, 0x5b, 0xdb, 0x3b, 0xbb, 0x7b, 0xfb,
0x07, 0x87, 0x47, 0xc7, 0x27, 0xa7, 0x67, 0xe7,
0x17, 0x97, 0x57, 0xd7, 0x37, 0xb7, 0x77, 0xf7,
0x0f, 0x8f, 0x4f, 0xcf, 0x2f, 0xaf, 0x6f, 0xef,
0x1f, 0x9f, 0x5f, 0xdf, 0x3f, 0xbf, 0x7f, 0xff,
}
func reverseUint16(v uint16) uint16 {
return uint16(reverseByte[v>>8]) | uint16(reverseByte[v&0xFF])<<8
}
func reverseBits(number uint16, bitLength byte) uint16 {
return reverseUint16(number << uint8(16-bitLength))
}

900
vendor/github.com/klauspost/compress/flate/snappy.go generated vendored Normal file
View File

@@ -0,0 +1,900 @@
// Copyright 2011 The Snappy-Go Authors. All rights reserved.
// Modified for deflate by Klaus Post (c) 2015.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package flate
// emitLiteral writes a literal chunk and returns the number of bytes written.
func emitLiteral(dst *tokens, lit []byte) {
ol := int(dst.n)
for i, v := range lit {
dst.tokens[(i+ol)&maxStoreBlockSize] = token(v)
}
dst.n += uint16(len(lit))
}
// emitCopy writes a copy chunk and returns the number of bytes written.
func emitCopy(dst *tokens, offset, length int) {
dst.tokens[dst.n] = matchToken(uint32(length-3), uint32(offset-minOffsetSize))
dst.n++
}
type fastEnc interface {
Encode(dst *tokens, src []byte)
Reset()
}
func newFastEnc(level int) fastEnc {
switch level {
case 1:
return &snappyL1{}
case 2:
return &snappyL2{snappyGen: snappyGen{cur: maxStoreBlockSize, prev: make([]byte, 0, maxStoreBlockSize)}}
case 3:
return &snappyL3{snappyGen: snappyGen{cur: maxStoreBlockSize, prev: make([]byte, 0, maxStoreBlockSize)}}
case 4:
return &snappyL4{snappyL3{snappyGen: snappyGen{cur: maxStoreBlockSize, prev: make([]byte, 0, maxStoreBlockSize)}}}
default:
panic("invalid level specified")
}
}
const (
tableBits = 14 // Bits used in the table
tableSize = 1 << tableBits // Size of the table
tableMask = tableSize - 1 // Mask for table indices. Redundant, but can eliminate bounds checks.
tableShift = 32 - tableBits // Right-shift to get the tableBits most significant bits of a uint32.
baseMatchOffset = 1 // The smallest match offset
baseMatchLength = 3 // The smallest match length per the RFC section 3.2.5
maxMatchOffset = 1 << 15 // The largest match offset
)
func load32(b []byte, i int) uint32 {
b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line.
return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
}
func load64(b []byte, i int) uint64 {
b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line.
return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
}
func hash(u uint32) uint32 {
return (u * 0x1e35a7bd) >> tableShift
}
// snappyL1 encapsulates level 1 compression
type snappyL1 struct{}
func (e *snappyL1) Reset() {}
func (e *snappyL1) Encode(dst *tokens, src []byte) {
const (
inputMargin = 16 - 1
minNonLiteralBlockSize = 1 + 1 + inputMargin
)
// This check isn't in the Snappy implementation, but there, the caller
// instead of the callee handles this case.
if len(src) < minNonLiteralBlockSize {
// We do not fill the token table.
// This will be picked up by caller.
dst.n = uint16(len(src))
return
}
// Initialize the hash table.
//
// The table element type is uint16, as s < sLimit and sLimit < len(src)
// and len(src) <= maxStoreBlockSize and maxStoreBlockSize == 65535.
var table [tableSize]uint16
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := len(src) - inputMargin
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := 0
// The encoded form must start with a literal, as there are no previous
// bytes to copy, so we start looking for hash matches at s == 1.
s := 1
nextHash := hash(load32(src, s))
for {
// Copied from the C++ snappy implementation:
//
// Heuristic match skipping: If 32 bytes are scanned with no matches
// found, start looking only at every other byte. If 32 more bytes are
// scanned (or skipped), look at every third byte, etc.. When a match
// is found, immediately go back to looking at every byte. This is a
// small loss (~5% performance, ~0.1% density) for compressible data
// due to more bookkeeping, but for non-compressible data (such as
// JPEG) it's a huge win since the compressor quickly "realizes" the
// data is incompressible and doesn't bother looking for matches
// everywhere.
//
// The "skip" variable keeps track of how many bytes there are since
// the last match; dividing it by 32 (ie. right-shifting by five) gives
// the number of bytes to move ahead for each iteration.
skip := 32
nextS := s
candidate := 0
for {
s = nextS
bytesBetweenHashLookups := skip >> 5
nextS = s + bytesBetweenHashLookups
skip += bytesBetweenHashLookups
if nextS > sLimit {
goto emitRemainder
}
candidate = int(table[nextHash&tableMask])
table[nextHash&tableMask] = uint16(s)
nextHash = hash(load32(src, nextS))
if s-candidate <= maxMatchOffset && load32(src, s) == load32(src, candidate) {
break
}
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
emitLiteral(dst, src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
base := s
// Extend the 4-byte match as long as possible.
//
// This is an inlined version of Snappy's:
// s = extendMatch(src, candidate+4, s+4)
s += 4
s1 := base + maxMatchLength
if s1 > len(src) {
s1 = len(src)
}
a := src[s:s1]
b := src[candidate+4:]
b = b[:len(a)]
l := len(a)
for i := range a {
if a[i] != b[i] {
l = i
break
}
}
s += l
// matchToken is flate's equivalent of Snappy's emitCopy.
dst.tokens[dst.n] = matchToken(uint32(s-base-baseMatchLength), uint32(base-candidate-baseMatchOffset))
dst.n++
nextEmit = s
if s >= sLimit {
goto emitRemainder
}
// We could immediately start working at s now, but to improve
// compression we first update the hash table at s-1 and at s. If
// another emitCopy is not our next move, also calculate nextHash
// at s+1. At least on GOARCH=amd64, these three hash calculations
// are faster as one load64 call (with some shifts) instead of
// three load32 calls.
x := load64(src, s-1)
prevHash := hash(uint32(x >> 0))
table[prevHash&tableMask] = uint16(s - 1)
currHash := hash(uint32(x >> 8))
candidate = int(table[currHash&tableMask])
table[currHash&tableMask] = uint16(s)
if s-candidate > maxMatchOffset || uint32(x>>8) != load32(src, candidate) {
nextHash = hash(uint32(x >> 16))
s++
break
}
}
}
emitRemainder:
if nextEmit < len(src) {
emitLiteral(dst, src[nextEmit:])
}
}
type tableEntry struct {
val uint32
offset int32
}
func load3232(b []byte, i int32) uint32 {
b = b[i : i+4 : len(b)] // Help the compiler eliminate bounds checks on the next line.
return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
}
func load6432(b []byte, i int32) uint64 {
b = b[i : i+8 : len(b)] // Help the compiler eliminate bounds checks on the next line.
return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
}
// snappyGen maintains the table for matches,
// and the previous byte block for level 2.
// This is the generic implementation.
type snappyGen struct {
prev []byte
cur int32
}
// snappyGen maintains the table for matches,
// and the previous byte block for level 2.
// This is the generic implementation.
type snappyL2 struct {
snappyGen
table [tableSize]tableEntry
}
// EncodeL2 uses a similar algorithm to level 1, but is capable
// of matching across blocks giving better compression at a small slowdown.
func (e *snappyL2) Encode(dst *tokens, src []byte) {
const (
inputMargin = 8 - 1
minNonLiteralBlockSize = 1 + 1 + inputMargin
)
// Protect against e.cur wraparound.
if e.cur > 1<<30 {
for i := range e.table[:] {
e.table[i] = tableEntry{}
}
e.cur = maxStoreBlockSize
}
// This check isn't in the Snappy implementation, but there, the caller
// instead of the callee handles this case.
if len(src) < minNonLiteralBlockSize {
// We do not fill the token table.
// This will be picked up by caller.
dst.n = uint16(len(src))
e.cur += maxStoreBlockSize
e.prev = e.prev[:0]
return
}
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := int32(len(src) - inputMargin)
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := int32(0)
s := int32(0)
cv := load3232(src, s)
nextHash := hash(cv)
for {
// Copied from the C++ snappy implementation:
//
// Heuristic match skipping: If 32 bytes are scanned with no matches
// found, start looking only at every other byte. If 32 more bytes are
// scanned (or skipped), look at every third byte, etc.. When a match
// is found, immediately go back to looking at every byte. This is a
// small loss (~5% performance, ~0.1% density) for compressible data
// due to more bookkeeping, but for non-compressible data (such as
// JPEG) it's a huge win since the compressor quickly "realizes" the
// data is incompressible and doesn't bother looking for matches
// everywhere.
//
// The "skip" variable keeps track of how many bytes there are since
// the last match; dividing it by 32 (ie. right-shifting by five) gives
// the number of bytes to move ahead for each iteration.
skip := int32(32)
nextS := s
var candidate tableEntry
for {
s = nextS
bytesBetweenHashLookups := skip >> 5
nextS = s + bytesBetweenHashLookups
skip += bytesBetweenHashLookups
if nextS > sLimit {
goto emitRemainder
}
candidate = e.table[nextHash&tableMask]
now := load3232(src, nextS)
e.table[nextHash&tableMask] = tableEntry{offset: s + e.cur, val: cv}
nextHash = hash(now)
offset := s - (candidate.offset - e.cur)
if offset > maxMatchOffset || cv != candidate.val {
// Out of range or not matched.
cv = now
continue
}
break
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
emitLiteral(dst, src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
// Extend the 4-byte match as long as possible.
//
s += 4
t := candidate.offset - e.cur + 4
l := e.matchlen(s, t, src)
// matchToken is flate's equivalent of Snappy's emitCopy. (length,offset)
dst.tokens[dst.n] = matchToken(uint32(l+4-baseMatchLength), uint32(s-t-baseMatchOffset))
dst.n++
s += l
nextEmit = s
if s >= sLimit {
t += l
// Index first pair after match end.
if int(t+4) < len(src) && t > 0 {
cv := load3232(src, t)
e.table[hash(cv)&tableMask] = tableEntry{offset: t + e.cur, val: cv}
}
goto emitRemainder
}
// We could immediately start working at s now, but to improve
// compression we first update the hash table at s-1 and at s. If
// another emitCopy is not our next move, also calculate nextHash
// at s+1. At least on GOARCH=amd64, these three hash calculations
// are faster as one load64 call (with some shifts) instead of
// three load32 calls.
x := load6432(src, s-1)
prevHash := hash(uint32(x))
e.table[prevHash&tableMask] = tableEntry{offset: e.cur + s - 1, val: uint32(x)}
x >>= 8
currHash := hash(uint32(x))
candidate = e.table[currHash&tableMask]
e.table[currHash&tableMask] = tableEntry{offset: e.cur + s, val: uint32(x)}
offset := s - (candidate.offset - e.cur)
if offset > maxMatchOffset || uint32(x) != candidate.val {
cv = uint32(x >> 8)
nextHash = hash(cv)
s++
break
}
}
}
emitRemainder:
if int(nextEmit) < len(src) {
emitLiteral(dst, src[nextEmit:])
}
e.cur += int32(len(src))
e.prev = e.prev[:len(src)]
copy(e.prev, src)
}
type tableEntryPrev struct {
Cur tableEntry
Prev tableEntry
}
// snappyL3
type snappyL3 struct {
snappyGen
table [tableSize]tableEntryPrev
}
// Encode uses a similar algorithm to level 2, will check up to two candidates.
func (e *snappyL3) Encode(dst *tokens, src []byte) {
const (
inputMargin = 8 - 1
minNonLiteralBlockSize = 1 + 1 + inputMargin
)
// Protect against e.cur wraparound.
if e.cur > 1<<30 {
for i := range e.table[:] {
e.table[i] = tableEntryPrev{}
}
e.snappyGen = snappyGen{cur: maxStoreBlockSize, prev: e.prev[:0]}
}
// This check isn't in the Snappy implementation, but there, the caller
// instead of the callee handles this case.
if len(src) < minNonLiteralBlockSize {
// We do not fill the token table.
// This will be picked up by caller.
dst.n = uint16(len(src))
e.cur += maxStoreBlockSize
e.prev = e.prev[:0]
return
}
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := int32(len(src) - inputMargin)
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := int32(0)
s := int32(0)
cv := load3232(src, s)
nextHash := hash(cv)
for {
// Copied from the C++ snappy implementation:
//
// Heuristic match skipping: If 32 bytes are scanned with no matches
// found, start looking only at every other byte. If 32 more bytes are
// scanned (or skipped), look at every third byte, etc.. When a match
// is found, immediately go back to looking at every byte. This is a
// small loss (~5% performance, ~0.1% density) for compressible data
// due to more bookkeeping, but for non-compressible data (such as
// JPEG) it's a huge win since the compressor quickly "realizes" the
// data is incompressible and doesn't bother looking for matches
// everywhere.
//
// The "skip" variable keeps track of how many bytes there are since
// the last match; dividing it by 32 (ie. right-shifting by five) gives
// the number of bytes to move ahead for each iteration.
skip := int32(32)
nextS := s
var candidate tableEntry
for {
s = nextS
bytesBetweenHashLookups := skip >> 5
nextS = s + bytesBetweenHashLookups
skip += bytesBetweenHashLookups
if nextS > sLimit {
goto emitRemainder
}
candidates := e.table[nextHash&tableMask]
now := load3232(src, nextS)
e.table[nextHash&tableMask] = tableEntryPrev{Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur, val: cv}}
nextHash = hash(now)
// Check both candidates
candidate = candidates.Cur
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset <= maxMatchOffset {
break
}
} else {
// We only check if value mismatches.
// Offset will always be invalid in other cases.
candidate = candidates.Prev
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset <= maxMatchOffset {
break
}
}
}
cv = now
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
emitLiteral(dst, src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
// Extend the 4-byte match as long as possible.
//
s += 4
t := candidate.offset - e.cur + 4
l := e.matchlen(s, t, src)
// matchToken is flate's equivalent of Snappy's emitCopy. (length,offset)
dst.tokens[dst.n] = matchToken(uint32(l+4-baseMatchLength), uint32(s-t-baseMatchOffset))
dst.n++
s += l
nextEmit = s
if s >= sLimit {
t += l
// Index first pair after match end.
if int(t+4) < len(src) && t > 0 {
cv := load3232(src, t)
nextHash = hash(cv)
e.table[nextHash&tableMask] = tableEntryPrev{
Prev: e.table[nextHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + t, val: cv},
}
}
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. If
// another emitCopy is not our next move, also calculate nextHash
// at s+1. At least on GOARCH=amd64, these three hash calculations
// are faster as one load64 call (with some shifts) instead of
// three load32 calls.
x := load6432(src, s-3)
prevHash := hash(uint32(x))
e.table[prevHash&tableMask] = tableEntryPrev{
Prev: e.table[prevHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + s - 3, val: uint32(x)},
}
x >>= 8
prevHash = hash(uint32(x))
e.table[prevHash&tableMask] = tableEntryPrev{
Prev: e.table[prevHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + s - 2, val: uint32(x)},
}
x >>= 8
prevHash = hash(uint32(x))
e.table[prevHash&tableMask] = tableEntryPrev{
Prev: e.table[prevHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + s - 1, val: uint32(x)},
}
x >>= 8
currHash := hash(uint32(x))
candidates := e.table[currHash&tableMask]
cv = uint32(x)
e.table[currHash&tableMask] = tableEntryPrev{
Prev: candidates.Cur,
Cur: tableEntry{offset: s + e.cur, val: cv},
}
// Check both candidates
candidate = candidates.Cur
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset <= maxMatchOffset {
continue
}
} else {
// We only check if value mismatches.
// Offset will always be invalid in other cases.
candidate = candidates.Prev
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset <= maxMatchOffset {
continue
}
}
}
cv = uint32(x >> 8)
nextHash = hash(cv)
s++
break
}
}
emitRemainder:
if int(nextEmit) < len(src) {
emitLiteral(dst, src[nextEmit:])
}
e.cur += int32(len(src))
e.prev = e.prev[:len(src)]
copy(e.prev, src)
}
// snappyL4
type snappyL4 struct {
snappyL3
}
// Encode uses a similar algorithm to level 3,
// but will check up to two candidates if first isn't long enough.
func (e *snappyL4) Encode(dst *tokens, src []byte) {
const (
inputMargin = 8 - 3
minNonLiteralBlockSize = 1 + 1 + inputMargin
matchLenGood = 12
)
// Protect against e.cur wraparound.
if e.cur > 1<<30 {
for i := range e.table[:] {
e.table[i] = tableEntryPrev{}
}
e.snappyGen = snappyGen{cur: maxStoreBlockSize, prev: e.prev[:0]}
}
// This check isn't in the Snappy implementation, but there, the caller
// instead of the callee handles this case.
if len(src) < minNonLiteralBlockSize {
// We do not fill the token table.
// This will be picked up by caller.
dst.n = uint16(len(src))
e.cur += maxStoreBlockSize
e.prev = e.prev[:0]
return
}
// sLimit is when to stop looking for offset/length copies. The inputMargin
// lets us use a fast path for emitLiteral in the main loop, while we are
// looking for copies.
sLimit := int32(len(src) - inputMargin)
// nextEmit is where in src the next emitLiteral should start from.
nextEmit := int32(0)
s := int32(0)
cv := load3232(src, s)
nextHash := hash(cv)
for {
// Copied from the C++ snappy implementation:
//
// Heuristic match skipping: If 32 bytes are scanned with no matches
// found, start looking only at every other byte. If 32 more bytes are
// scanned (or skipped), look at every third byte, etc.. When a match
// is found, immediately go back to looking at every byte. This is a
// small loss (~5% performance, ~0.1% density) for compressible data
// due to more bookkeeping, but for non-compressible data (such as
// JPEG) it's a huge win since the compressor quickly "realizes" the
// data is incompressible and doesn't bother looking for matches
// everywhere.
//
// The "skip" variable keeps track of how many bytes there are since
// the last match; dividing it by 32 (ie. right-shifting by five) gives
// the number of bytes to move ahead for each iteration.
skip := int32(32)
nextS := s
var candidate tableEntry
var candidateAlt tableEntry
for {
s = nextS
bytesBetweenHashLookups := skip >> 5
nextS = s + bytesBetweenHashLookups
skip += bytesBetweenHashLookups
if nextS > sLimit {
goto emitRemainder
}
candidates := e.table[nextHash&tableMask]
now := load3232(src, nextS)
e.table[nextHash&tableMask] = tableEntryPrev{Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur, val: cv}}
nextHash = hash(now)
// Check both candidates
candidate = candidates.Cur
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset < maxMatchOffset {
offset = s - (candidates.Prev.offset - e.cur)
if cv == candidates.Prev.val && offset < maxMatchOffset {
candidateAlt = candidates.Prev
}
break
}
} else {
// We only check if value mismatches.
// Offset will always be invalid in other cases.
candidate = candidates.Prev
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset < maxMatchOffset {
break
}
}
}
cv = now
}
// A 4-byte match has been found. We'll later see if more than 4 bytes
// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
// them as literal bytes.
emitLiteral(dst, src[nextEmit:s])
// Call emitCopy, and then see if another emitCopy could be our next
// move. Repeat until we find no match for the input immediately after
// what was consumed by the last emitCopy call.
//
// If we exit this loop normally then we need to call emitLiteral next,
// though we don't yet know how big the literal will be. We handle that
// by proceeding to the next iteration of the main loop. We also can
// exit this loop via goto if we get close to exhausting the input.
for {
// Invariant: we have a 4-byte match at s, and no need to emit any
// literal bytes prior to s.
// Extend the 4-byte match as long as possible.
//
s += 4
t := candidate.offset - e.cur + 4
l := e.matchlen(s, t, src)
// Try alternative candidate if match length < matchLenGood.
if l < matchLenGood-4 && candidateAlt.offset != 0 {
t2 := candidateAlt.offset - e.cur + 4
l2 := e.matchlen(s, t2, src)
if l2 > l {
l = l2
t = t2
}
}
// matchToken is flate's equivalent of Snappy's emitCopy. (length,offset)
dst.tokens[dst.n] = matchToken(uint32(l+4-baseMatchLength), uint32(s-t-baseMatchOffset))
dst.n++
s += l
nextEmit = s
if s >= sLimit {
t += l
// Index first pair after match end.
if int(t+4) < len(src) && t > 0 {
cv := load3232(src, t)
nextHash = hash(cv)
e.table[nextHash&tableMask] = tableEntryPrev{
Prev: e.table[nextHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + t, val: cv},
}
}
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. If
// another emitCopy is not our next move, also calculate nextHash
// at s+1. At least on GOARCH=amd64, these three hash calculations
// are faster as one load64 call (with some shifts) instead of
// three load32 calls.
x := load6432(src, s-3)
prevHash := hash(uint32(x))
e.table[prevHash&tableMask] = tableEntryPrev{
Prev: e.table[prevHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + s - 3, val: uint32(x)},
}
x >>= 8
prevHash = hash(uint32(x))
e.table[prevHash&tableMask] = tableEntryPrev{
Prev: e.table[prevHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + s - 2, val: uint32(x)},
}
x >>= 8
prevHash = hash(uint32(x))
e.table[prevHash&tableMask] = tableEntryPrev{
Prev: e.table[prevHash&tableMask].Cur,
Cur: tableEntry{offset: e.cur + s - 1, val: uint32(x)},
}
x >>= 8
currHash := hash(uint32(x))
candidates := e.table[currHash&tableMask]
cv = uint32(x)
e.table[currHash&tableMask] = tableEntryPrev{
Prev: candidates.Cur,
Cur: tableEntry{offset: s + e.cur, val: cv},
}
// Check both candidates
candidate = candidates.Cur
candidateAlt = tableEntry{}
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset <= maxMatchOffset {
offset = s - (candidates.Prev.offset - e.cur)
if cv == candidates.Prev.val && offset <= maxMatchOffset {
candidateAlt = candidates.Prev
}
continue
}
} else {
// We only check if value mismatches.
// Offset will always be invalid in other cases.
candidate = candidates.Prev
if cv == candidate.val {
offset := s - (candidate.offset - e.cur)
if offset <= maxMatchOffset {
continue
}
}
}
cv = uint32(x >> 8)
nextHash = hash(cv)
s++
break
}
}
emitRemainder:
if int(nextEmit) < len(src) {
emitLiteral(dst, src[nextEmit:])
}
e.cur += int32(len(src))
e.prev = e.prev[:len(src)]
copy(e.prev, src)
}
func (e *snappyGen) matchlen(s, t int32, src []byte) int32 {
s1 := int(s) + maxMatchLength - 4
if s1 > len(src) {
s1 = len(src)
}
// If we are inside the current block
if t >= 0 {
b := src[t:]
a := src[s:s1]
b = b[:len(a)]
// Extend the match to be as long as possible.
for i := range a {
if a[i] != b[i] {
return int32(i)
}
}
return int32(len(a))
}
// We found a match in the previous block.
tp := int32(len(e.prev)) + t
if tp < 0 {
return 0
}
// Extend the match to be as long as possible.
a := src[s:s1]
b := e.prev[tp:]
if len(b) > len(a) {
b = b[:len(a)]
}
a = a[:len(b)]
for i := range b {
if a[i] != b[i] {
return int32(i)
}
}
// If we reached our limit, we matched everything we are
// allowed to in the previous block and we return.
n := int32(len(b))
if int(s+n) == s1 {
return n
}
// Continue looking for more matches in the current block.
a = src[s+n : s1]
b = src[:len(a)]
for i := range a {
if a[i] != b[i] {
return int32(i) + n
}
}
return int32(len(a)) + n
}
// Reset the encoding table.
func (e *snappyGen) Reset() {
e.prev = e.prev[:0]
e.cur += maxMatchOffset
}

106
vendor/github.com/klauspost/compress/flate/token.go generated vendored Normal file
View File

@@ -0,0 +1,106 @@
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package flate
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<<lengthShift - 1
typeMask = 3 << 30
literalType = 0 << 30
matchType = 1 << 30
)
// The length code for length X (MIN_MATCH_LENGTH <= X <= MAX_MATCH_LENGTH)
// is lengthCodes[length - MIN_MATCH_LENGTH]
var lengthCodes = [256]uint8{
0, 1, 2, 3, 4, 5, 6, 7, 8, 8,
9, 9, 10, 10, 11, 11, 12, 12, 12, 12,
13, 13, 13, 13, 14, 14, 14, 14, 15, 15,
15, 15, 16, 16, 16, 16, 16, 16, 16, 16,
17, 17, 17, 17, 17, 17, 17, 17, 18, 18,
18, 18, 18, 18, 18, 18, 19, 19, 19, 19,
19, 19, 19, 19, 20, 20, 20, 20, 20, 20,
20, 20, 20, 20, 20, 20, 20, 20, 20, 20,
21, 21, 21, 21, 21, 21, 21, 21, 21, 21,
21, 21, 21, 21, 21, 21, 22, 22, 22, 22,
22, 22, 22, 22, 22, 22, 22, 22, 22, 22,
22, 22, 23, 23, 23, 23, 23, 23, 23, 23,
23, 23, 23, 23, 23, 23, 23, 23, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
25, 25, 26, 26, 26, 26, 26, 26, 26, 26,
26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
26, 26, 26, 26, 27, 27, 27, 27, 27, 27,
27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
27, 27, 27, 27, 27, 28,
}
var offsetCodes = [256]uint32{
0, 1, 2, 3, 4, 4, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7,
8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,
10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
}
type token uint32
type tokens struct {
tokens [maxStoreBlockSize + 1]token
n uint16 // Must be able to contain maxStoreBlockSize
}
// Convert a literal into a literal token.
func literalToken(literal uint32) token { return token(literalType + literal) }
// Convert a < xlength, xoffset > pair into a match token.
func matchToken(xlength uint32, xoffset uint32) token {
return token(matchType + xlength<<lengthShift + xoffset)
}
// Returns the type of a token
func (t token) typ() uint32 { return uint32(t) & typeMask }
// Returns the literal of a literal token
func (t token) literal() uint8 { return uint8(t) }
// Returns the extra offset of a match token
func (t token) offset() uint32 { return uint32(t) & offsetMask }
func (t token) length() uint8 { return uint8(t >> lengthShift) }
// The code is never more than 8 bits, but is returned as uint32 for convenience.
func lengthCode(len uint8) uint32 { return uint32(lengthCodes[len]) }
// Returns the offset code corresponding to a specific offset
func offsetCode(off uint32) uint32 {
if off < uint32(len(offsetCodes)) {
return offsetCodes[off&255]
} else if off>>7 < uint32(len(offsetCodes)) {
return offsetCodes[(off>>7)&255] + 14
} else {
return offsetCodes[(off>>14)&255] + 28
}
}

24
vendor/github.com/klauspost/cpuid/.gitignore generated vendored Normal file
View File

@@ -0,0 +1,24 @@
# Compiled Object files, Static and Dynamic libs (Shared Objects)
*.o
*.a
*.so
# Folders
_obj
_test
# Architecture specific extensions/prefixes
*.[568vq]
[568vq].out
*.cgo1.go
*.cgo2.c
_cgo_defun.c
_cgo_gotypes.go
_cgo_export.*
_testmain.go
*.exe
*.test
*.prof

23
vendor/github.com/klauspost/cpuid/.travis.yml generated vendored Normal file
View File

@@ -0,0 +1,23 @@
language: go
sudo: false
os:
- linux
- osx
go:
- 1.8.x
- 1.9.x
- 1.10.x
- master
script:
- go vet ./...
- go test -v ./...
- go test -race ./...
- diff <(gofmt -d .) <("")
matrix:
allow_failures:
- go: 'master'
fast_finish: true

35
vendor/github.com/klauspost/cpuid/CONTRIBUTING.txt generated vendored Normal file
View File

@@ -0,0 +1,35 @@
Developer Certificate of Origin
Version 1.1
Copyright (C) 2015- Klaus Post & Contributors.
Email: klauspost@gmail.com
Everyone is permitted to copy and distribute verbatim copies of this
license document, but changing it is not allowed.
Developer's Certificate of Origin 1.1
By making a contribution to this project, I certify that:
(a) The contribution was created in whole or in part by me and I
have the right to submit it under the open source license
indicated in the file; or
(b) The contribution is based upon previous work that, to the best
of my knowledge, is covered under an appropriate open source
license and I have the right under that license to submit that
work with modifications, whether created in whole or in part
by me, under the same open source license (unless I am
permitted to submit under a different license), as indicated
in the file; or
(c) The contribution was provided directly to me by some other
person who certified (a), (b) or (c) and I have not modified
it.
(d) I understand and agree that this project and the contribution
are public and that a record of the contribution (including all
personal information I submit with it, including my sign-off) is
maintained indefinitely and may be redistributed consistent with
this project or the open source license(s) involved.

22
vendor/github.com/klauspost/cpuid/LICENSE generated vendored Normal file
View File

@@ -0,0 +1,22 @@
The MIT License (MIT)
Copyright (c) 2015 Klaus Post
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

147
vendor/github.com/klauspost/cpuid/README.md generated vendored Normal file
View File

@@ -0,0 +1,147 @@
# cpuid
Package cpuid provides information about the CPU running the current program.
CPU features are detected on startup, and kept for fast access through the life of the application.
Currently x86 / x64 (AMD64) is supported, and no external C (cgo) code is used, which should make the library very easy to use.
You can access the CPU information by accessing the shared CPU variable of the cpuid library.
Package home: https://github.com/klauspost/cpuid
[![GoDoc][1]][2] [![Build Status][3]][4]
[1]: https://godoc.org/github.com/klauspost/cpuid?status.svg
[2]: https://godoc.org/github.com/klauspost/cpuid
[3]: https://travis-ci.org/klauspost/cpuid.svg
[4]: https://travis-ci.org/klauspost/cpuid
# features
## CPU Instructions
* **CMOV** (i686 CMOV)
* **NX** (NX (No-Execute) bit)
* **AMD3DNOW** (AMD 3DNOW)
* **AMD3DNOWEXT** (AMD 3DNowExt)
* **MMX** (standard MMX)
* **MMXEXT** (SSE integer functions or AMD MMX ext)
* **SSE** (SSE functions)
* **SSE2** (P4 SSE functions)
* **SSE3** (Prescott SSE3 functions)
* **SSSE3** (Conroe SSSE3 functions)
* **SSE4** (Penryn SSE4.1 functions)
* **SSE4A** (AMD Barcelona microarchitecture SSE4a instructions)
* **SSE42** (Nehalem SSE4.2 functions)
* **AVX** (AVX functions)
* **AVX2** (AVX2 functions)
* **FMA3** (Intel FMA 3)
* **FMA4** (Bulldozer FMA4 functions)
* **XOP** (Bulldozer XOP functions)
* **F16C** (Half-precision floating-point conversion)
* **BMI1** (Bit Manipulation Instruction Set 1)
* **BMI2** (Bit Manipulation Instruction Set 2)
* **TBM** (AMD Trailing Bit Manipulation)
* **LZCNT** (LZCNT instruction)
* **POPCNT** (POPCNT instruction)
* **AESNI** (Advanced Encryption Standard New Instructions)
* **CLMUL** (Carry-less Multiplication)
* **HTT** (Hyperthreading (enabled))
* **HLE** (Hardware Lock Elision)
* **RTM** (Restricted Transactional Memory)
* **RDRAND** (RDRAND instruction is available)
* **RDSEED** (RDSEED instruction is available)
* **ADX** (Intel ADX (Multi-Precision Add-Carry Instruction Extensions))
* **SHA** (Intel SHA Extensions)
* **AVX512F** (AVX-512 Foundation)
* **AVX512DQ** (AVX-512 Doubleword and Quadword Instructions)
* **AVX512IFMA** (AVX-512 Integer Fused Multiply-Add Instructions)
* **AVX512PF** (AVX-512 Prefetch Instructions)
* **AVX512ER** (AVX-512 Exponential and Reciprocal Instructions)
* **AVX512CD** (AVX-512 Conflict Detection Instructions)
* **AVX512BW** (AVX-512 Byte and Word Instructions)
* **AVX512VL** (AVX-512 Vector Length Extensions)
* **AVX512VBMI** (AVX-512 Vector Bit Manipulation Instructions)
* **MPX** (Intel MPX (Memory Protection Extensions))
* **ERMS** (Enhanced REP MOVSB/STOSB)
* **RDTSCP** (RDTSCP Instruction)
* **CX16** (CMPXCHG16B Instruction)
* **SGX** (Software Guard Extensions, with activation details)
## Performance
* **RDTSCP()** Returns current cycle count. Can be used for benchmarking.
* **SSE2SLOW** (SSE2 is supported, but usually not faster)
* **SSE3SLOW** (SSE3 is supported, but usually not faster)
* **ATOM** (Atom processor, some SSSE3 instructions are slower)
* **Cache line** (Probable size of a cache line).
* **L1, L2, L3 Cache size** on newer Intel/AMD CPUs.
## Cpu Vendor/VM
* **Intel**
* **AMD**
* **VIA**
* **Transmeta**
* **NSC**
* **KVM** (Kernel-based Virtual Machine)
* **MSVM** (Microsoft Hyper-V or Windows Virtual PC)
* **VMware**
* **XenHVM**
* **Bhyve**
* **Hygon**
# installing
```go get github.com/klauspost/cpuid```
# example
```Go
package main
import (
"fmt"
"github.com/klauspost/cpuid"
)
func main() {
// Print basic CPU information:
fmt.Println("Name:", cpuid.CPU.BrandName)
fmt.Println("PhysicalCores:", cpuid.CPU.PhysicalCores)
fmt.Println("ThreadsPerCore:", cpuid.CPU.ThreadsPerCore)
fmt.Println("LogicalCores:", cpuid.CPU.LogicalCores)
fmt.Println("Family", cpuid.CPU.Family, "Model:", cpuid.CPU.Model)
fmt.Println("Features:", cpuid.CPU.Features)
fmt.Println("Cacheline bytes:", cpuid.CPU.CacheLine)
fmt.Println("L1 Data Cache:", cpuid.CPU.Cache.L1D, "bytes")
fmt.Println("L1 Instruction Cache:", cpuid.CPU.Cache.L1D, "bytes")
fmt.Println("L2 Cache:", cpuid.CPU.Cache.L2, "bytes")
fmt.Println("L3 Cache:", cpuid.CPU.Cache.L3, "bytes")
// Test if we have a specific feature:
if cpuid.CPU.SSE() {
fmt.Println("We have Streaming SIMD Extensions")
}
}
```
Sample output:
```
>go run main.go
Name: Intel(R) Core(TM) i5-2540M CPU @ 2.60GHz
PhysicalCores: 2
ThreadsPerCore: 2
LogicalCores: 4
Family 6 Model: 42
Features: CMOV,MMX,MMXEXT,SSE,SSE2,SSE3,SSSE3,SSE4.1,SSE4.2,AVX,AESNI,CLMUL
Cacheline bytes: 64
We have Streaming SIMD Extensions
```
# private package
In the "private" folder you can find an autogenerated version of the library you can include in your own packages.
For this purpose all exports are removed, and functions and constants are lowercased.
This is not a recommended way of using the library, but provided for convenience, if it is difficult for you to use external packages.
# license
This code is published under an MIT license. See LICENSE file for more information.

1049
vendor/github.com/klauspost/cpuid/cpuid.go generated vendored Normal file
View File

File diff suppressed because it is too large Load Diff

42
vendor/github.com/klauspost/cpuid/cpuid_386.s generated vendored Normal file
View File

@@ -0,0 +1,42 @@
// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
// +build 386,!gccgo
// func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32)
TEXT ·asmCpuid(SB), 7, $0
XORL CX, CX
MOVL op+0(FP), AX
CPUID
MOVL AX, eax+4(FP)
MOVL BX, ebx+8(FP)
MOVL CX, ecx+12(FP)
MOVL DX, edx+16(FP)
RET
// func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32)
TEXT ·asmCpuidex(SB), 7, $0
MOVL op+0(FP), AX
MOVL op2+4(FP), CX
CPUID
MOVL AX, eax+8(FP)
MOVL BX, ebx+12(FP)
MOVL CX, ecx+16(FP)
MOVL DX, edx+20(FP)
RET
// func xgetbv(index uint32) (eax, edx uint32)
TEXT ·asmXgetbv(SB), 7, $0
MOVL index+0(FP), CX
BYTE $0x0f; BYTE $0x01; BYTE $0xd0 // XGETBV
MOVL AX, eax+4(FP)
MOVL DX, edx+8(FP)
RET
// func asmRdtscpAsm() (eax, ebx, ecx, edx uint32)
TEXT ·asmRdtscpAsm(SB), 7, $0
BYTE $0x0F; BYTE $0x01; BYTE $0xF9 // RDTSCP
MOVL AX, eax+0(FP)
MOVL BX, ebx+4(FP)
MOVL CX, ecx+8(FP)
MOVL DX, edx+12(FP)
RET

42
vendor/github.com/klauspost/cpuid/cpuid_amd64.s generated vendored Normal file
View File

@@ -0,0 +1,42 @@
// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
//+build amd64,!gccgo
// func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32)
TEXT ·asmCpuid(SB), 7, $0
XORQ CX, CX
MOVL op+0(FP), AX
CPUID
MOVL AX, eax+8(FP)
MOVL BX, ebx+12(FP)
MOVL CX, ecx+16(FP)
MOVL DX, edx+20(FP)
RET
// func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32)
TEXT ·asmCpuidex(SB), 7, $0
MOVL op+0(FP), AX
MOVL op2+4(FP), CX
CPUID
MOVL AX, eax+8(FP)
MOVL BX, ebx+12(FP)
MOVL CX, ecx+16(FP)
MOVL DX, edx+20(FP)
RET
// func asmXgetbv(index uint32) (eax, edx uint32)
TEXT ·asmXgetbv(SB), 7, $0
MOVL index+0(FP), CX
BYTE $0x0f; BYTE $0x01; BYTE $0xd0 // XGETBV
MOVL AX, eax+8(FP)
MOVL DX, edx+12(FP)
RET
// func asmRdtscpAsm() (eax, ebx, ecx, edx uint32)
TEXT ·asmRdtscpAsm(SB), 7, $0
BYTE $0x0F; BYTE $0x01; BYTE $0xF9 // RDTSCP
MOVL AX, eax+0(FP)
MOVL BX, ebx+4(FP)
MOVL CX, ecx+8(FP)
MOVL DX, edx+12(FP)
RET

17
vendor/github.com/klauspost/cpuid/detect_intel.go generated vendored Normal file
View File

@@ -0,0 +1,17 @@
// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
// +build 386,!gccgo amd64,!gccgo
package cpuid
func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32)
func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32)
func asmXgetbv(index uint32) (eax, edx uint32)
func asmRdtscpAsm() (eax, ebx, ecx, edx uint32)
func initCPU() {
cpuid = asmCpuid
cpuidex = asmCpuidex
xgetbv = asmXgetbv
rdtscpAsm = asmRdtscpAsm
}

23
vendor/github.com/klauspost/cpuid/detect_ref.go generated vendored Normal file
View File

@@ -0,0 +1,23 @@
// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
// +build !amd64,!386 gccgo
package cpuid
func initCPU() {
cpuid = func(op uint32) (eax, ebx, ecx, edx uint32) {
return 0, 0, 0, 0
}
cpuidex = func(op, op2 uint32) (eax, ebx, ecx, edx uint32) {
return 0, 0, 0, 0
}
xgetbv = func(index uint32) (eax, edx uint32) {
return 0, 0
}
rdtscpAsm = func() (eax, ebx, ecx, edx uint32) {
return 0, 0, 0, 0
}
}

4
vendor/github.com/klauspost/cpuid/generate.go generated vendored Normal file
View File

@@ -0,0 +1,4 @@
package cpuid
//go:generate go run private-gen.go
//go:generate gofmt -w ./private

476
vendor/github.com/klauspost/cpuid/private-gen.go generated vendored Normal file
View File

@@ -0,0 +1,476 @@
// +build ignore
package main
import (
"bytes"
"fmt"
"go/ast"
"go/parser"
"go/printer"
"go/token"
"io"
"io/ioutil"
"log"
"os"
"reflect"
"strings"
"unicode"
"unicode/utf8"
)
var inFiles = []string{"cpuid.go", "cpuid_test.go"}
var copyFiles = []string{"cpuid_amd64.s", "cpuid_386.s", "detect_ref.go", "detect_intel.go"}
var fileSet = token.NewFileSet()
var reWrites = []rewrite{
initRewrite("CPUInfo -> cpuInfo"),
initRewrite("Vendor -> vendor"),
initRewrite("Flags -> flags"),
initRewrite("Detect -> detect"),
initRewrite("CPU -> cpu"),
}
var excludeNames = map[string]bool{"string": true, "join": true, "trim": true,
// cpuid_test.go
"t": true, "println": true, "logf": true, "log": true, "fatalf": true, "fatal": true,
}
var excludePrefixes = []string{"test", "benchmark"}
func main() {
Package := "private"
parserMode := parser.ParseComments
exported := make(map[string]rewrite)
for _, file := range inFiles {
in, err := os.Open(file)
if err != nil {
log.Fatalf("opening input", err)
}
src, err := ioutil.ReadAll(in)
if err != nil {
log.Fatalf("reading input", err)
}
astfile, err := parser.ParseFile(fileSet, file, src, parserMode)
if err != nil {
log.Fatalf("parsing input", err)
}
for _, rw := range reWrites {
astfile = rw(astfile)
}
// Inspect the AST and print all identifiers and literals.
var startDecl token.Pos
var endDecl token.Pos
ast.Inspect(astfile, func(n ast.Node) bool {
var s string
switch x := n.(type) {
case *ast.Ident:
if x.IsExported() {
t := strings.ToLower(x.Name)
for _, pre := range excludePrefixes {
if strings.HasPrefix(t, pre) {
return true
}
}
if excludeNames[t] != true {
//if x.Pos() > startDecl && x.Pos() < endDecl {
exported[x.Name] = initRewrite(x.Name + " -> " + t)
}
}
case *ast.GenDecl:
if x.Tok == token.CONST && x.Lparen > 0 {
startDecl = x.Lparen
endDecl = x.Rparen
// fmt.Printf("Decl:%s -> %s\n", fileSet.Position(startDecl), fileSet.Position(endDecl))
}
}
if s != "" {
fmt.Printf("%s:\t%s\n", fileSet.Position(n.Pos()), s)
}
return true
})
for _, rw := range exported {
astfile = rw(astfile)
}
var buf bytes.Buffer
printer.Fprint(&buf, fileSet, astfile)
// Remove package documentation and insert information
s := buf.String()
ind := strings.Index(buf.String(), "\npackage cpuid")
s = s[ind:]
s = "// Generated, DO NOT EDIT,\n" +
"// but copy it to your own project and rename the package.\n" +
"// See more at http://github.com/klauspost/cpuid\n" +
s
outputName := Package + string(os.PathSeparator) + file
err = ioutil.WriteFile(outputName, []byte(s), 0644)
if err != nil {
log.Fatalf("writing output: %s", err)
}
log.Println("Generated", outputName)
}
for _, file := range copyFiles {
dst := ""
if strings.HasPrefix(file, "cpuid") {
dst = Package + string(os.PathSeparator) + file
} else {
dst = Package + string(os.PathSeparator) + "cpuid_" + file
}
err := copyFile(file, dst)
if err != nil {
log.Fatalf("copying file: %s", err)
}
log.Println("Copied", dst)
}
}
// CopyFile copies a file from src to dst. If src and dst files exist, and are
// the same, then return success. Copy the file contents from src to dst.
func copyFile(src, dst string) (err error) {
sfi, err := os.Stat(src)
if err != nil {
return
}
if !sfi.Mode().IsRegular() {
// cannot copy non-regular files (e.g., directories,
// symlinks, devices, etc.)
return fmt.Errorf("CopyFile: non-regular source file %s (%q)", sfi.Name(), sfi.Mode().String())
}
dfi, err := os.Stat(dst)
if err != nil {
if !os.IsNotExist(err) {
return
}
} else {
if !(dfi.Mode().IsRegular()) {
return fmt.Errorf("CopyFile: non-regular destination file %s (%q)", dfi.Name(), dfi.Mode().String())
}
if os.SameFile(sfi, dfi) {
return
}
}
err = copyFileContents(src, dst)
return
}
// copyFileContents copies the contents of the file named src to the file named
// by dst. The file will be created if it does not already exist. If the
// destination file exists, all it's contents will be replaced by the contents
// of the source file.
func copyFileContents(src, dst string) (err error) {
in, err := os.Open(src)
if err != nil {
return
}
defer in.Close()
out, err := os.Create(dst)
if err != nil {
return
}
defer func() {
cerr := out.Close()
if err == nil {
err = cerr
}
}()
if _, err = io.Copy(out, in); err != nil {
return
}
err = out.Sync()
return
}
type rewrite func(*ast.File) *ast.File
// Mostly copied from gofmt
func initRewrite(rewriteRule string) rewrite {
f := strings.Split(rewriteRule, "->")
if len(f) != 2 {
fmt.Fprintf(os.Stderr, "rewrite rule must be of the form 'pattern -> replacement'\n")
os.Exit(2)
}
pattern := parseExpr(f[0], "pattern")
replace := parseExpr(f[1], "replacement")
return func(p *ast.File) *ast.File { return rewriteFile(pattern, replace, p) }
}
// parseExpr parses s as an expression.
// It might make sense to expand this to allow statement patterns,
// but there are problems with preserving formatting and also
// with what a wildcard for a statement looks like.
func parseExpr(s, what string) ast.Expr {
x, err := parser.ParseExpr(s)
if err != nil {
fmt.Fprintf(os.Stderr, "parsing %s %s at %s\n", what, s, err)
os.Exit(2)
}
return x
}
// Keep this function for debugging.
/*
func dump(msg string, val reflect.Value) {
fmt.Printf("%s:\n", msg)
ast.Print(fileSet, val.Interface())
fmt.Println()
}
*/
// rewriteFile applies the rewrite rule 'pattern -> replace' to an entire file.
func rewriteFile(pattern, replace ast.Expr, p *ast.File) *ast.File {
cmap := ast.NewCommentMap(fileSet, p, p.Comments)
m := make(map[string]reflect.Value)
pat := reflect.ValueOf(pattern)
repl := reflect.ValueOf(replace)
var rewriteVal func(val reflect.Value) reflect.Value
rewriteVal = func(val reflect.Value) reflect.Value {
// don't bother if val is invalid to start with
if !val.IsValid() {
return reflect.Value{}
}
for k := range m {
delete(m, k)
}
val = apply(rewriteVal, val)
if match(m, pat, val) {
val = subst(m, repl, reflect.ValueOf(val.Interface().(ast.Node).Pos()))
}
return val
}
r := apply(rewriteVal, reflect.ValueOf(p)).Interface().(*ast.File)
r.Comments = cmap.Filter(r).Comments() // recreate comments list
return r
}
// set is a wrapper for x.Set(y); it protects the caller from panics if x cannot be changed to y.
func set(x, y reflect.Value) {
// don't bother if x cannot be set or y is invalid
if !x.CanSet() || !y.IsValid() {
return
}
defer func() {
if x := recover(); x != nil {
if s, ok := x.(string); ok &&
(strings.Contains(s, "type mismatch") || strings.Contains(s, "not assignable")) {
// x cannot be set to y - ignore this rewrite
return
}
panic(x)
}
}()
x.Set(y)
}
// Values/types for special cases.
var (
objectPtrNil = reflect.ValueOf((*ast.Object)(nil))
scopePtrNil = reflect.ValueOf((*ast.Scope)(nil))
identType = reflect.TypeOf((*ast.Ident)(nil))
objectPtrType = reflect.TypeOf((*ast.Object)(nil))
positionType = reflect.TypeOf(token.NoPos)
callExprType = reflect.TypeOf((*ast.CallExpr)(nil))
scopePtrType = reflect.TypeOf((*ast.Scope)(nil))
)
// apply replaces each AST field x in val with f(x), returning val.
// To avoid extra conversions, f operates on the reflect.Value form.
func apply(f func(reflect.Value) reflect.Value, val reflect.Value) reflect.Value {
if !val.IsValid() {
return reflect.Value{}
}
// *ast.Objects introduce cycles and are likely incorrect after
// rewrite; don't follow them but replace with nil instead
if val.Type() == objectPtrType {
return objectPtrNil
}
// similarly for scopes: they are likely incorrect after a rewrite;
// replace them with nil
if val.Type() == scopePtrType {
return scopePtrNil
}
switch v := reflect.Indirect(val); v.Kind() {
case reflect.Slice:
for i := 0; i < v.Len(); i++ {
e := v.Index(i)
set(e, f(e))
}
case reflect.Struct:
for i := 0; i < v.NumField(); i++ {
e := v.Field(i)
set(e, f(e))
}
case reflect.Interface:
e := v.Elem()
set(v, f(e))
}
return val
}
func isWildcard(s string) bool {
rune, size := utf8.DecodeRuneInString(s)
return size == len(s) && unicode.IsLower(rune)
}
// match returns true if pattern matches val,
// recording wildcard submatches in m.
// If m == nil, match checks whether pattern == val.
func match(m map[string]reflect.Value, pattern, val reflect.Value) bool {
// Wildcard matches any expression. If it appears multiple
// times in the pattern, it must match the same expression
// each time.
if m != nil && pattern.IsValid() && pattern.Type() == identType {
name := pattern.Interface().(*ast.Ident).Name
if isWildcard(name) && val.IsValid() {
// wildcards only match valid (non-nil) expressions.
if _, ok := val.Interface().(ast.Expr); ok && !val.IsNil() {
if old, ok := m[name]; ok {
return match(nil, old, val)
}
m[name] = val
return true
}
}
}
// Otherwise, pattern and val must match recursively.
if !pattern.IsValid() || !val.IsValid() {
return !pattern.IsValid() && !val.IsValid()
}
if pattern.Type() != val.Type() {
return false
}
// Special cases.
switch pattern.Type() {
case identType:
// For identifiers, only the names need to match
// (and none of the other *ast.Object information).
// This is a common case, handle it all here instead
// of recursing down any further via reflection.
p := pattern.Interface().(*ast.Ident)
v := val.Interface().(*ast.Ident)
return p == nil && v == nil || p != nil && v != nil && p.Name == v.Name
case objectPtrType, positionType:
// object pointers and token positions always match
return true
case callExprType:
// For calls, the Ellipsis fields (token.Position) must
// match since that is how f(x) and f(x...) are different.
// Check them here but fall through for the remaining fields.
p := pattern.Interface().(*ast.CallExpr)
v := val.Interface().(*ast.CallExpr)
if p.Ellipsis.IsValid() != v.Ellipsis.IsValid() {
return false
}
}
p := reflect.Indirect(pattern)
v := reflect.Indirect(val)
if !p.IsValid() || !v.IsValid() {
return !p.IsValid() && !v.IsValid()
}
switch p.Kind() {
case reflect.Slice:
if p.Len() != v.Len() {
return false
}
for i := 0; i < p.Len(); i++ {
if !match(m, p.Index(i), v.Index(i)) {
return false
}
}
return true
case reflect.Struct:
for i := 0; i < p.NumField(); i++ {
if !match(m, p.Field(i), v.Field(i)) {
return false
}
}
return true
case reflect.Interface:
return match(m, p.Elem(), v.Elem())
}
// Handle token integers, etc.
return p.Interface() == v.Interface()
}
// subst returns a copy of pattern with values from m substituted in place
// of wildcards and pos used as the position of tokens from the pattern.
// if m == nil, subst returns a copy of pattern and doesn't change the line
// number information.
func subst(m map[string]reflect.Value, pattern reflect.Value, pos reflect.Value) reflect.Value {
if !pattern.IsValid() {
return reflect.Value{}
}
// Wildcard gets replaced with map value.
if m != nil && pattern.Type() == identType {
name := pattern.Interface().(*ast.Ident).Name
if isWildcard(name) {
if old, ok := m[name]; ok {
return subst(nil, old, reflect.Value{})
}
}
}
if pos.IsValid() && pattern.Type() == positionType {
// use new position only if old position was valid in the first place
if old := pattern.Interface().(token.Pos); !old.IsValid() {
return pattern
}
return pos
}
// Otherwise copy.
switch p := pattern; p.Kind() {
case reflect.Slice:
v := reflect.MakeSlice(p.Type(), p.Len(), p.Len())
for i := 0; i < p.Len(); i++ {
v.Index(i).Set(subst(m, p.Index(i), pos))
}
return v
case reflect.Struct:
v := reflect.New(p.Type()).Elem()
for i := 0; i < p.NumField(); i++ {
v.Field(i).Set(subst(m, p.Field(i), pos))
}
return v
case reflect.Ptr:
v := reflect.New(p.Type()).Elem()
if elem := p.Elem(); elem.IsValid() {
v.Set(subst(m, elem, pos).Addr())
}
return v
case reflect.Interface:
v := reflect.New(p.Type()).Elem()
if elem := p.Elem(); elem.IsValid() {
v.Set(subst(m, elem, pos))
}
return v
}
return pattern
}

24
vendor/github.com/klauspost/pgzip/.gitignore generated vendored Normal file
View File

@@ -0,0 +1,24 @@
# Compiled Object files, Static and Dynamic libs (Shared Objects)
*.o
*.a
*.so
# Folders
_obj
_test
# Architecture specific extensions/prefixes
*.[568vq]
[568vq].out
*.cgo1.go
*.cgo2.c
_cgo_defun.c
_cgo_gotypes.go
_cgo_export.*
_testmain.go
*.exe
*.test
*.prof

21
vendor/github.com/klauspost/pgzip/.travis.yml generated vendored Normal file
View File

@@ -0,0 +1,21 @@
language: go
sudo: false
os:
- linux
- osx
go:
- 1.9.x
- 1.10.x
- master
script:
- go test -v -cpu=1,2,4 .
- go test -v -cpu=2 -race -short .
matrix:
allow_failures:
- go: 'master'
fast_finish: true

27
vendor/github.com/klauspost/pgzip/GO_LICENSE generated vendored Normal file
View File

@@ -0,0 +1,27 @@
Copyright (c) 2012 The 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.

22
vendor/github.com/klauspost/pgzip/LICENSE generated vendored Normal file
View File

@@ -0,0 +1,22 @@
The MIT License (MIT)
Copyright (c) 2014 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.

136
vendor/github.com/klauspost/pgzip/README.md generated vendored Normal file
View File

@@ -0,0 +1,136 @@
pgzip
=====
Go parallel gzip compression/decompression. This is a fully gzip compatible drop in replacement for "compress/gzip".
This will split compression into blocks that are compressed in parallel.
This can be useful for compressing big amounts of data. The output is a standard gzip file.
The gzip decompression is modified so it decompresses ahead of the current reader.
This means that reads will be non-blocking if the decompressor can keep ahead of your code reading from it.
CRC calculation also takes place in a separate goroutine.
You should only use this if you are (de)compressing big amounts of data,
say **more than 1MB** at the time, otherwise you will not see any benefit,
and it will likely be faster to use the internal gzip library
or [this package](https://github.com/klauspost/compress).
It is important to note that this library creates and reads *standard gzip files*.
You do not have to match the compressor/decompressor to get the described speedups,
and the gzip files are fully compatible with other gzip readers/writers.
A golang variant of this is [bgzf](https://godoc.org/github.com/biogo/hts/bgzf),
which has the same feature, as well as seeking in the resulting file.
The only drawback is a slightly bigger overhead compared to this and pure gzip.
See a comparison below.
[![GoDoc][1]][2] [![Build Status][3]][4]
[1]: https://godoc.org/github.com/klauspost/pgzip?status.svg
[2]: https://godoc.org/github.com/klauspost/pgzip
[3]: https://travis-ci.org/klauspost/pgzip.svg
[4]: https://travis-ci.org/klauspost/pgzip
Installation
====
```go get github.com/klauspost/pgzip/...```
You might need to get/update the dependencies:
```
go get -u github.com/klauspost/compress
go get -u github.com/klauspost/crc32
```
Usage
====
[Godoc Doumentation](https://godoc.org/github.com/klauspost/pgzip)
To use as a replacement for gzip, exchange
```import "compress/gzip"```
with
```import gzip "github.com/klauspost/pgzip"```.
# Changes
* Oct 6, 2016: Fixed an issue if the destination writer returned an error.
* Oct 6, 2016: Better buffer reuse, should now generate less garbage.
* Oct 6, 2016: Output does not change based on write sizes.
* Dec 8, 2015: Decoder now supports the io.WriterTo interface, giving a speedup and less GC pressure.
* Oct 9, 2015: Reduced allocations by ~35 by using sync.Pool. ~15% overall speedup.
Changes in [github.com/klauspost/compress](https://github.com/klauspost/compress#changelog) are also carried over, so see that for more changes.
## Compression
The simplest way to use this is to simply do the same as you would when using [compress/gzip](http://golang.org/pkg/compress/gzip).
To change the block size, use the added (*pgzip.Writer).SetConcurrency(blockSize, blocks int) function. With this you can control the approximate size of your blocks, as well as how many you want to be processing in parallel. Default values for this is SetConcurrency(250000, 16), meaning blocks are split at 250000 bytes and up to 16 blocks can be processing at once before the writer blocks.
Example:
```
var b bytes.Buffer
w := gzip.NewWriter(&b)
w.SetConcurrency(100000, 10)
w.Write([]byte("hello, world\n"))
w.Close()
```
To get any performance gains, you should at least be compressing more than 1 megabyte of data at the time.
You should at least have a block size of 100k and at least a number of blocks that match the number of cores your would like to utilize, but about twice the number of blocks would be the best.
Another side effect of this is, that it is likely to speed up your other code, since writes to the compressor only blocks if the compressor is already compressing the number of blocks you have specified. This also means you don't have worry about buffering input to the compressor.
## Decompression
Decompression works similar to compression. That means that you simply call pgzip the same way as you would call [compress/gzip](http://golang.org/pkg/compress/gzip).
The only difference is that if you want to specify your own readahead, you have to use `pgzip.NewReaderN(r io.Reader, blockSize, blocks int)` to get a reader with your custom blocksizes. The `blockSize` is the size of each block decoded, and `blocks` is the maximum number of blocks that is decoded ahead.
See [Example on playground](http://play.golang.org/p/uHv1B5NbDh)
Performance
====
## Compression
See my blog post in [Benchmarks of Golang Gzip](https://blog.klauspost.com/go-gzipdeflate-benchmarks/).
Compression cost is usually about 0.2% with default settings with a block size of 250k.
Example with GOMAXPROC set to 8 (quad core with 8 hyperthreads)
Content is [Matt Mahoneys 10GB corpus](http://mattmahoney.net/dc/10gb.html). Compression level 6.
Compressor | MB/sec | speedup | size | size overhead (lower=better)
------------|----------|---------|------|---------
[gzip](http://golang.org/pkg/compress/gzip) (golang) | 7.21MB/s | 1.0x | 4786608902 | 0%
[gzip](http://github.com/klauspost/compress/gzip) (klauspost) | 10.98MB/s | 1.52x | 4781331645 | -0.11%
[pgzip](https://github.com/klauspost/pgzip) (klauspost) | 50.76MB/s|7.04x | 4784121440 | -0.052%
[bgzf](https://godoc.org/github.com/biogo/hts/bgzf) (biogo) | 38.65MB/s | 5.36x | 4924899484 | 2.889%
[pargzip](https://godoc.org/github.com/golang/build/pargzip) (builder) | 32.00MB/s | 4.44x | 4791226567 | 0.096%
pgzip also contains a [linear time compression](https://github.com/klauspost/compress#linear-time-compression) mode, that will allow compression at ~150MB per core per second, independent of the content.
See the [complete sheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing) for different content types and compression settings.
## Decompression
The decompression speedup is there because it allows you to do other work while the decompression is taking place.
In the example above, the numbers are as follows on a 4 CPU machine:
Decompressor | Time | Speedup
-------------|------|--------
[gzip](http://golang.org/pkg/compress/gzip) (golang) | 1m28.85s | 0%
[pgzip](https://github.com/klauspost/pgzip) (golang) | 43.48s | 104%
But wait, since gzip decompression is inherently singlethreaded (aside from CRC calculation) how can it be more than 100% faster? Because pgzip due to its design also acts as a buffer. When using unbuffered gzip, you are also waiting for io when you are decompressing. If the gzip decoder can keep up, it will always have data ready for your reader, and you will not be waiting for input to the gzip decompressor to complete.
This is pretty much an optimal situation for pgzip, but it reflects most common usecases for CPU intensive gzip usage.
I haven't included [bgzf](https://godoc.org/github.com/biogo/hts/bgzf) in this comparison, since it only can decompress files created by a compatible encoder, and therefore cannot be considered a generic gzip decompressor. But if you are able to compress your files with a bgzf compatible program, you can expect it to scale beyond 100%.
# License
This contains large portions of code from the go repository - see GO_LICENSE for more information. The changes are released under MIT License. See LICENSE for more information.

573
vendor/github.com/klauspost/pgzip/gunzip.go generated vendored Normal file
View File

@@ -0,0 +1,573 @@
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
// Package pgzip implements reading and writing of gzip format compressed files,
// as specified in RFC 1952.
//
// This is a drop in replacement for "compress/gzip".
// This will split compression into blocks that are compressed in parallel.
// This can be useful for compressing big amounts of data.
// The gzip decompression has not been modified, but remains in the package,
// so you can use it as a complete replacement for "compress/gzip".
//
// See more at https://github.com/klauspost/pgzip
package pgzip
import (
"bufio"
"errors"
"hash"
"hash/crc32"
"io"
"sync"
"time"
"github.com/klauspost/compress/flate"
)
const (
gzipID1 = 0x1f
gzipID2 = 0x8b
gzipDeflate = 8
flagText = 1 << 0
flagHdrCrc = 1 << 1
flagExtra = 1 << 2
flagName = 1 << 3
flagComment = 1 << 4
)
func makeReader(r io.Reader) flate.Reader {
if rr, ok := r.(flate.Reader); ok {
return rr
}
return bufio.NewReader(r)
}
var (
// ErrChecksum is returned when reading GZIP data that has an invalid checksum.
ErrChecksum = errors.New("gzip: invalid checksum")
// ErrHeader is returned when reading GZIP data that has an invalid header.
ErrHeader = errors.New("gzip: invalid header")
)
// The gzip file stores a header giving metadata about the compressed file.
// That header is exposed as the fields of the Writer and Reader structs.
type Header struct {
Comment string // comment
Extra []byte // "extra data"
ModTime time.Time // modification time
Name string // file name
OS byte // operating system type
}
// A Reader is an io.Reader that can be read to retrieve
// uncompressed data from a gzip-format compressed file.
//
// In general, a gzip file can be a concatenation of gzip files,
// each with its own header. Reads from the Reader
// return the concatenation of the uncompressed data of each.
// Only the first header is recorded in the Reader fields.
//
// Gzip files store a length and checksum of the uncompressed data.
// The Reader will return a ErrChecksum when Read
// reaches the end of the uncompressed data if it does not
// have the expected length or checksum. Clients should treat data
// returned by Read as tentative until they receive the io.EOF
// marking the end of the data.
type Reader struct {
Header
r flate.Reader
decompressor io.ReadCloser
digest hash.Hash32
size uint32
flg byte
buf [512]byte
err error
closeErr chan error
multistream bool
readAhead chan read
roff int // read offset
current []byte
closeReader chan struct{}
lastBlock bool
blockSize int
blocks int
activeRA bool // Indication if readahead is active
mu sync.Mutex // Lock for above
blockPool chan []byte
}
type read struct {
b []byte
err error
}
// NewReader creates a new Reader reading the given reader.
// The implementation buffers input and may read more data than necessary from r.
// It is the caller's responsibility to call Close on the Reader when done.
func NewReader(r io.Reader) (*Reader, error) {
z := new(Reader)
z.blocks = defaultBlocks
z.blockSize = defaultBlockSize
z.r = makeReader(r)
z.digest = crc32.NewIEEE()
z.multistream = true
z.blockPool = make(chan []byte, z.blocks)
for i := 0; i < z.blocks; i++ {
z.blockPool <- make([]byte, z.blockSize)
}
if err := z.readHeader(true); err != nil {
return nil, err
}
return z, nil
}
// NewReaderN creates a new Reader reading the given reader.
// The implementation buffers input and may read more data than necessary from r.
// It is the caller's responsibility to call Close on the Reader when done.
//
// With this you can control the approximate size of your blocks,
// as well as how many blocks you want to have prefetched.
//
// Default values for this is blockSize = 250000, blocks = 16,
// meaning up to 16 blocks of maximum 250000 bytes will be
// prefetched.
func NewReaderN(r io.Reader, blockSize, blocks int) (*Reader, error) {
z := new(Reader)
z.blocks = blocks
z.blockSize = blockSize
z.r = makeReader(r)
z.digest = crc32.NewIEEE()
z.multistream = true
// Account for too small values
if z.blocks <= 0 {
z.blocks = defaultBlocks
}
if z.blockSize <= 512 {
z.blockSize = defaultBlockSize
}
z.blockPool = make(chan []byte, z.blocks)
for i := 0; i < z.blocks; i++ {
z.blockPool <- make([]byte, z.blockSize)
}
if err := z.readHeader(true); err != nil {
return nil, err
}
return z, nil
}
// Reset discards the Reader z's state and makes it equivalent to the
// result of its original state from NewReader, but reading from r instead.
// This permits reusing a Reader rather than allocating a new one.
func (z *Reader) Reset(r io.Reader) error {
z.killReadAhead()
z.r = makeReader(r)
z.digest = crc32.NewIEEE()
z.size = 0
z.err = nil
z.multistream = true
// Account for uninitialized values
if z.blocks <= 0 {
z.blocks = defaultBlocks
}
if z.blockSize <= 512 {
z.blockSize = defaultBlockSize
}
if z.blockPool == nil {
z.blockPool = make(chan []byte, z.blocks)
for i := 0; i < z.blocks; i++ {
z.blockPool <- make([]byte, z.blockSize)
}
}
return z.readHeader(true)
}
// Multistream controls whether the reader supports multistream files.
//
// If enabled (the default), the Reader expects the input to be a sequence
// of individually gzipped data streams, each with its own header and
// trailer, ending at EOF. The effect is that the concatenation of a sequence
// of gzipped files is treated as equivalent to the gzip of the concatenation
// of the sequence. This is standard behavior for gzip readers.
//
// Calling Multistream(false) disables this behavior; disabling the behavior
// can be useful when reading file formats that distinguish individual gzip
// data streams or mix gzip data streams with other data streams.
// In this mode, when the Reader reaches the end of the data stream,
// Read returns io.EOF. If the underlying reader implements io.ByteReader,
// it will be left positioned just after the gzip stream.
// To start the next stream, call z.Reset(r) followed by z.Multistream(false).
// If there is no next stream, z.Reset(r) will return io.EOF.
func (z *Reader) Multistream(ok bool) {
z.multistream = ok
}
// GZIP (RFC 1952) is little-endian, unlike ZLIB (RFC 1950).
func get4(p []byte) uint32 {
return uint32(p[0]) | uint32(p[1])<<8 | uint32(p[2])<<16 | uint32(p[3])<<24
}
func (z *Reader) readString() (string, error) {
var err error
needconv := false
for i := 0; ; i++ {
if i >= len(z.buf) {
return "", ErrHeader
}
z.buf[i], err = z.r.ReadByte()
if err != nil {
return "", err
}
if z.buf[i] > 0x7f {
needconv = true
}
if z.buf[i] == 0 {
// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1).
if needconv {
s := make([]rune, 0, i)
for _, v := range z.buf[0:i] {
s = append(s, rune(v))
}
return string(s), nil
}
return string(z.buf[0:i]), nil
}
}
}
func (z *Reader) read2() (uint32, error) {
_, err := io.ReadFull(z.r, z.buf[0:2])
if err != nil {
return 0, err
}
return uint32(z.buf[0]) | uint32(z.buf[1])<<8, nil
}
func (z *Reader) readHeader(save bool) error {
z.killReadAhead()
_, err := io.ReadFull(z.r, z.buf[0:10])
if err != nil {
return err
}
if z.buf[0] != gzipID1 || z.buf[1] != gzipID2 || z.buf[2] != gzipDeflate {
return ErrHeader
}
z.flg = z.buf[3]
if save {
z.ModTime = time.Unix(int64(get4(z.buf[4:8])), 0)
// z.buf[8] is xfl, ignored
z.OS = z.buf[9]
}
z.digest.Reset()
z.digest.Write(z.buf[0:10])
if z.flg&flagExtra != 0 {
n, err := z.read2()
if err != nil {
return err
}
data := make([]byte, n)
if _, err = io.ReadFull(z.r, data); err != nil {
return err
}
if save {
z.Extra = data
}
}
var s string
if z.flg&flagName != 0 {
if s, err = z.readString(); err != nil {
return err
}
if save {
z.Name = s
}
}
if z.flg&flagComment != 0 {
if s, err = z.readString(); err != nil {
return err
}
if save {
z.Comment = s
}
}
if z.flg&flagHdrCrc != 0 {
n, err := z.read2()
if err != nil {
return err
}
sum := z.digest.Sum32() & 0xFFFF
if n != sum {
return ErrHeader
}
}
z.digest.Reset()
z.decompressor = flate.NewReader(z.r)
z.doReadAhead()
return nil
}
func (z *Reader) killReadAhead() error {
z.mu.Lock()
defer z.mu.Unlock()
if z.activeRA {
if z.closeReader != nil {
close(z.closeReader)
}
// Wait for decompressor to be closed and return error, if any.
e, ok := <-z.closeErr
z.activeRA = false
if !ok {
// Channel is closed, so if there was any error it has already been returned.
return nil
}
return e
}
return nil
}
// Starts readahead.
// Will return on error (including io.EOF)
// or when z.closeReader is closed.
func (z *Reader) doReadAhead() {
z.mu.Lock()
defer z.mu.Unlock()
z.activeRA = true
if z.blocks <= 0 {
z.blocks = defaultBlocks
}
if z.blockSize <= 512 {
z.blockSize = defaultBlockSize
}
ra := make(chan read, z.blocks)
z.readAhead = ra
closeReader := make(chan struct{}, 0)
z.closeReader = closeReader
z.lastBlock = false
closeErr := make(chan error, 1)
z.closeErr = closeErr
z.size = 0
z.roff = 0
z.current = nil
decomp := z.decompressor
go func() {
defer func() {
closeErr <- decomp.Close()
close(closeErr)
close(ra)
}()
// We hold a local reference to digest, since
// it way be changed by reset.
digest := z.digest
var wg sync.WaitGroup
for {
var buf []byte
select {
case buf = <-z.blockPool:
case <-closeReader:
return
}
buf = buf[0:z.blockSize]
// Try to fill the buffer
n, err := io.ReadFull(decomp, buf)
if err == io.ErrUnexpectedEOF {
if n > 0 {
err = nil
} else {
// If we got zero bytes, we need to establish if
// we reached end of stream or truncated stream.
_, err = decomp.Read([]byte{})
if err == io.EOF {
err = nil
}
}
}
if n < len(buf) {
buf = buf[0:n]
}
wg.Wait()
wg.Add(1)
go func() {
digest.Write(buf)
wg.Done()
}()
z.size += uint32(n)
// If we return any error, out digest must be ready
if err != nil {
wg.Wait()
}
select {
case z.readAhead <- read{b: buf, err: err}:
case <-closeReader:
// Sent on close, we don't care about the next results
return
}
if err != nil {
return
}
}
}()
}
func (z *Reader) Read(p []byte) (n int, err error) {
if z.err != nil {
return 0, z.err
}
if len(p) == 0 {
return 0, nil
}
for {
if len(z.current) == 0 && !z.lastBlock {
read := <-z.readAhead
if read.err != nil {
// If not nil, the reader will have exited
z.closeReader = nil
if read.err != io.EOF {
z.err = read.err
return
}
if read.err == io.EOF {
z.lastBlock = true
err = nil
}
}
z.current = read.b
z.roff = 0
}
avail := z.current[z.roff:]
if len(p) >= len(avail) {
// If len(p) >= len(current), return all content of current
n = copy(p, avail)
z.blockPool <- z.current
z.current = nil
if z.lastBlock {
err = io.EOF
break
}
} else {
// We copy as much as there is space for
n = copy(p, avail)
z.roff += n
}
return
}
// Finished file; check checksum + size.
if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil {
z.err = err
return 0, err
}
crc32, isize := get4(z.buf[0:4]), get4(z.buf[4:8])
sum := z.digest.Sum32()
if sum != crc32 || isize != z.size {
z.err = ErrChecksum
return 0, z.err
}
// File is ok; should we attempt reading one more?
if !z.multistream {
return 0, io.EOF
}
// Is there another?
if err = z.readHeader(false); err != nil {
z.err = err
return
}
// Yes. Reset and read from it.
return z.Read(p)
}
func (z *Reader) WriteTo(w io.Writer) (n int64, err error) {
total := int64(0)
for {
if z.err != nil {
return total, z.err
}
// We write both to output and digest.
for {
// Read from input
read := <-z.readAhead
if read.err != nil {
// If not nil, the reader will have exited
z.closeReader = nil
if read.err != io.EOF {
z.err = read.err
return total, z.err
}
if read.err == io.EOF {
z.lastBlock = true
err = nil
}
}
// Write what we got
n, err := w.Write(read.b)
if n != len(read.b) {
return total, io.ErrShortWrite
}
total += int64(n)
if err != nil {
return total, err
}
// Put block back
z.blockPool <- read.b
if z.lastBlock {
break
}
}
// Finished file; check checksum + size.
if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil {
z.err = err
return total, err
}
crc32, isize := get4(z.buf[0:4]), get4(z.buf[4:8])
sum := z.digest.Sum32()
if sum != crc32 || isize != z.size {
z.err = ErrChecksum
return total, z.err
}
// File is ok; should we attempt reading one more?
if !z.multistream {
return total, nil
}
// Is there another?
err = z.readHeader(false)
if err == io.EOF {
return total, nil
}
if err != nil {
z.err = err
return total, err
}
}
}
// Close closes the Reader. It does not close the underlying io.Reader.
func (z *Reader) Close() error {
return z.killReadAhead()
}

501
vendor/github.com/klauspost/pgzip/gzip.go generated vendored Normal file
View File

@@ -0,0 +1,501 @@
// Copyright 2010 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
package pgzip
import (
"bytes"
"errors"
"fmt"
"hash"
"hash/crc32"
"io"
"sync"
"time"
"github.com/klauspost/compress/flate"
)
const (
defaultBlockSize = 256 << 10
tailSize = 16384
defaultBlocks = 16
)
// These constants are copied from the flate package, so that code that imports
// "compress/gzip" does not also have to import "compress/flate".
const (
NoCompression = flate.NoCompression
BestSpeed = flate.BestSpeed
BestCompression = flate.BestCompression
DefaultCompression = flate.DefaultCompression
ConstantCompression = flate.ConstantCompression
HuffmanOnly = flate.HuffmanOnly
)
// A Writer is an io.WriteCloser.
// Writes to a Writer are compressed and written to w.
type Writer struct {
Header
w io.Writer
level int
wroteHeader bool
blockSize int
blocks int
currentBuffer []byte
prevTail []byte
digest hash.Hash32
size int
closed bool
buf [10]byte
errMu sync.RWMutex
err error
pushedErr chan struct{}
results chan result
dictFlatePool sync.Pool
dstPool sync.Pool
wg sync.WaitGroup
}
type result struct {
result chan []byte
notifyWritten chan struct{}
}
// Use SetConcurrency to finetune the concurrency level if needed.
//
// With this you can control the approximate size of your blocks,
// as well as how many you want to be processing in parallel.
//
// Default values for this is SetConcurrency(250000, 16),
// meaning blocks are split at 250000 bytes and up to 16 blocks
// can be processing at once before the writer blocks.
func (z *Writer) SetConcurrency(blockSize, blocks int) error {
if blockSize <= tailSize {
return fmt.Errorf("gzip: block size cannot be less than or equal to %d", tailSize)
}
if blocks <= 0 {
return errors.New("gzip: blocks cannot be zero or less")
}
if blockSize == z.blockSize && blocks == z.blocks {
return nil
}
z.blockSize = blockSize
z.results = make(chan result, blocks)
z.blocks = blocks
z.dstPool = sync.Pool{New: func() interface{} { return make([]byte, 0, blockSize+(blockSize)>>4) }}
return nil
}
// NewWriter returns a new Writer.
// Writes to the returned writer are compressed and written to w.
//
// It is the caller's responsibility to call Close on the WriteCloser when done.
// Writes may be buffered and not flushed until Close.
//
// Callers that wish to set the fields in Writer.Header must do so before
// the first call to Write or Close. The Comment and Name header fields are
// UTF-8 strings in Go, but the underlying format requires NUL-terminated ISO
// 8859-1 (Latin-1). NUL or non-Latin-1 runes in those strings will lead to an
// error on Write.
func NewWriter(w io.Writer) *Writer {
z, _ := NewWriterLevel(w, DefaultCompression)
return z
}
// NewWriterLevel is like NewWriter but specifies the compression level instead
// of assuming DefaultCompression.
//
// The compression level can be DefaultCompression, NoCompression, or any
// integer value between BestSpeed and BestCompression inclusive. The error
// returned will be nil if the level is valid.
func NewWriterLevel(w io.Writer, level int) (*Writer, error) {
if level < ConstantCompression || level > BestCompression {
return nil, fmt.Errorf("gzip: invalid compression level: %d", level)
}
z := new(Writer)
z.SetConcurrency(defaultBlockSize, defaultBlocks)
z.init(w, level)
return z, nil
}
// This function must be used by goroutines to set an
// error condition, since z.err access is restricted
// to the callers goruotine.
func (z *Writer) pushError(err error) {
z.errMu.Lock()
if z.err != nil {
z.errMu.Unlock()
return
}
z.err = err
close(z.pushedErr)
z.errMu.Unlock()
}
func (z *Writer) init(w io.Writer, level int) {
z.wg.Wait()
digest := z.digest
if digest != nil {
digest.Reset()
} else {
digest = crc32.NewIEEE()
}
z.Header = Header{OS: 255}
z.w = w
z.level = level
z.digest = digest
z.pushedErr = make(chan struct{}, 0)
z.results = make(chan result, z.blocks)
z.err = nil
z.closed = false
z.Comment = ""
z.Extra = nil
z.ModTime = time.Time{}
z.wroteHeader = false
z.currentBuffer = nil
z.buf = [10]byte{}
z.prevTail = nil
z.size = 0
if z.dictFlatePool.New == nil {
z.dictFlatePool.New = func() interface{} {
f, _ := flate.NewWriterDict(w, level, nil)
return f
}
}
}
// Reset discards the Writer z's state and makes it equivalent to the
// result of its original state from NewWriter or NewWriterLevel, but
// writing to w instead. This permits reusing a Writer rather than
// allocating a new one.
func (z *Writer) Reset(w io.Writer) {
if z.results != nil && !z.closed {
close(z.results)
}
z.SetConcurrency(defaultBlockSize, defaultBlocks)
z.init(w, z.level)
}
// GZIP (RFC 1952) is little-endian, unlike ZLIB (RFC 1950).
func put2(p []byte, v uint16) {
p[0] = uint8(v >> 0)
p[1] = uint8(v >> 8)
}
func put4(p []byte, v uint32) {
p[0] = uint8(v >> 0)
p[1] = uint8(v >> 8)
p[2] = uint8(v >> 16)
p[3] = uint8(v >> 24)
}
// writeBytes writes a length-prefixed byte slice to z.w.
func (z *Writer) writeBytes(b []byte) error {
if len(b) > 0xffff {
return errors.New("gzip.Write: Extra data is too large")
}
put2(z.buf[0:2], uint16(len(b)))
_, err := z.w.Write(z.buf[0:2])
if err != nil {
return err
}
_, err = z.w.Write(b)
return err
}
// writeString writes a UTF-8 string s in GZIP's format to z.w.
// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1).
func (z *Writer) writeString(s string) (err error) {
// GZIP stores Latin-1 strings; error if non-Latin-1; convert if non-ASCII.
needconv := false
for _, v := range s {
if v == 0 || v > 0xff {
return errors.New("gzip.Write: non-Latin-1 header string")
}
if v > 0x7f {
needconv = true
}
}
if needconv {
b := make([]byte, 0, len(s))
for _, v := range s {
b = append(b, byte(v))
}
_, err = z.w.Write(b)
} else {
_, err = io.WriteString(z.w, s)
}
if err != nil {
return err
}
// GZIP strings are NUL-terminated.
z.buf[0] = 0
_, err = z.w.Write(z.buf[0:1])
return err
}
// compressCurrent will compress the data currently buffered
// This should only be called from the main writer/flush/closer
func (z *Writer) compressCurrent(flush bool) {
r := result{}
r.result = make(chan []byte, 1)
r.notifyWritten = make(chan struct{}, 0)
select {
case z.results <- r:
case <-z.pushedErr:
return
}
// If block given is more than twice the block size, split it.
c := z.currentBuffer
if len(c) > z.blockSize*2 {
c = c[:z.blockSize]
z.wg.Add(1)
go z.compressBlock(c, z.prevTail, r, false)
z.prevTail = c[len(c)-tailSize:]
z.currentBuffer = z.currentBuffer[z.blockSize:]
z.compressCurrent(flush)
// Last one flushes if needed
return
}
z.wg.Add(1)
go z.compressBlock(c, z.prevTail, r, z.closed)
if len(c) > tailSize {
z.prevTail = c[len(c)-tailSize:]
} else {
z.prevTail = nil
}
z.currentBuffer = z.dstPool.Get().([]byte)
z.currentBuffer = z.currentBuffer[:0]
// Wait if flushing
if flush {
<-r.notifyWritten
}
}
// Returns an error if it has been set.
// Cannot be used by functions that are from internal goroutines.
func (z *Writer) checkError() error {
z.errMu.RLock()
err := z.err
z.errMu.RUnlock()
return err
}
// Write writes a compressed form of p to the underlying io.Writer. The
// compressed bytes are not necessarily flushed to output until
// the Writer is closed or Flush() is called.
//
// The function will return quickly, if there are unused buffers.
// The sent slice (p) is copied, and the caller is free to re-use the buffer
// when the function returns.
//
// Errors that occur during compression will be reported later, and a nil error
// does not signify that the compression succeeded (since it is most likely still running)
// That means that the call that returns an error may not be the call that caused it.
// Only Flush and Close functions are guaranteed to return any errors up to that point.
func (z *Writer) Write(p []byte) (int, error) {
if err := z.checkError(); err != nil {
return 0, err
}
// Write the GZIP header lazily.
if !z.wroteHeader {
z.wroteHeader = true
z.buf[0] = gzipID1
z.buf[1] = gzipID2
z.buf[2] = gzipDeflate
z.buf[3] = 0
if z.Extra != nil {
z.buf[3] |= 0x04
}
if z.Name != "" {
z.buf[3] |= 0x08
}
if z.Comment != "" {
z.buf[3] |= 0x10
}
put4(z.buf[4:8], uint32(z.ModTime.Unix()))
if z.level == BestCompression {
z.buf[8] = 2
} else if z.level == BestSpeed {
z.buf[8] = 4
} else {
z.buf[8] = 0
}
z.buf[9] = z.OS
var n int
var err error
n, err = z.w.Write(z.buf[0:10])
if err != nil {
z.pushError(err)
return n, err
}
if z.Extra != nil {
err = z.writeBytes(z.Extra)
if err != nil {
z.pushError(err)
return n, err
}
}
if z.Name != "" {
err = z.writeString(z.Name)
if err != nil {
z.pushError(err)
return n, err
}
}
if z.Comment != "" {
err = z.writeString(z.Comment)
if err != nil {
z.pushError(err)
return n, err
}
}
// Start receiving data from compressors
go func() {
listen := z.results
for {
r, ok := <-listen
// If closed, we are finished.
if !ok {
return
}
buf := <-r.result
n, err := z.w.Write(buf)
if err != nil {
z.pushError(err)
close(r.notifyWritten)
return
}
if n != len(buf) {
z.pushError(fmt.Errorf("gzip: short write %d should be %d", n, len(buf)))
close(r.notifyWritten)
return
}
z.dstPool.Put(buf)
close(r.notifyWritten)
}
}()
z.currentBuffer = make([]byte, 0, z.blockSize)
}
q := p
for len(q) > 0 {
length := len(q)
if length+len(z.currentBuffer) > z.blockSize {
length = z.blockSize - len(z.currentBuffer)
}
z.digest.Write(q[:length])
z.currentBuffer = append(z.currentBuffer, q[:length]...)
if len(z.currentBuffer) >= z.blockSize {
z.compressCurrent(false)
if err := z.checkError(); err != nil {
return len(p) - len(q) - length, err
}
}
z.size += length
q = q[length:]
}
return len(p), z.checkError()
}
// Step 1: compresses buffer to buffer
// Step 2: send writer to channel
// Step 3: Close result channel to indicate we are done
func (z *Writer) compressBlock(p, prevTail []byte, r result, closed bool) {
defer func() {
close(r.result)
z.wg.Done()
}()
buf := z.dstPool.Get().([]byte)
dest := bytes.NewBuffer(buf[:0])
compressor := z.dictFlatePool.Get().(*flate.Writer)
compressor.ResetDict(dest, prevTail)
compressor.Write(p)
err := compressor.Flush()
if err != nil {
z.pushError(err)
return
}
if closed {
err = compressor.Close()
if err != nil {
z.pushError(err)
return
}
}
z.dictFlatePool.Put(compressor)
// Read back buffer
buf = dest.Bytes()
r.result <- buf
}
// Flush flushes any pending compressed data to the underlying writer.
//
// It is useful mainly in compressed network protocols, to ensure that
// a remote reader has enough data to reconstruct a packet. Flush does
// not return until the data has been written. If the underlying
// writer returns an error, Flush returns that error.
//
// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH.
func (z *Writer) Flush() error {
if err := z.checkError(); err != nil {
return err
}
if z.closed {
return nil
}
if !z.wroteHeader {
_, err := z.Write(nil)
if err != nil {
return err
}
}
// We send current block to compression
z.compressCurrent(true)
return z.checkError()
}
// UncompressedSize will return the number of bytes written.
// pgzip only, not a function in the official gzip package.
func (z *Writer) UncompressedSize() int {
return z.size
}
// Close closes the Writer, flushing any unwritten data to the underlying
// io.Writer, but does not close the underlying io.Writer.
func (z *Writer) Close() error {
if err := z.checkError(); err != nil {
return err
}
if z.closed {
return nil
}
z.closed = true
if !z.wroteHeader {
z.Write(nil)
if err := z.checkError(); err != nil {
return err
}
}
z.compressCurrent(true)
if err := z.checkError(); err != nil {
return err
}
close(z.results)
put4(z.buf[0:4], z.digest.Sum32())
put4(z.buf[4:8], uint32(z.size))
_, err := z.w.Write(z.buf[0:8])
if err != nil {
z.pushError(err)
return err
}
return nil
}