Browse code
vendor: github.com/klauspost/compress v1.15.9
various fixes in zstd compression
- https://github.com/klauspost/compress/releases/tag/v1.15.9
- https://github.com/klauspost/compress/releases/tag/v1.15.8
- https://github.com/klauspost/compress/releases/tag/v1.15.7
- https://github.com/klauspost/compress/releases/tag/v1.15.6
- https://github.com/klauspost/compress/releases/tag/v1.15.5
- https://github.com/klauspost/compress/releases/tag/v1.15.4
- https://github.com/klauspost/compress/releases/tag/v1.15.3
- https://github.com/klauspost/compress/releases/tag/v1.15.2
full diff: https://github.com/klauspost/compress/compare/v1.15.1...v1.15.9
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
Sebastiaan van Stijn authored on 2022/08/26 08:46:28Showing 47 changed files
- vendor.mod
- vendor.sum
- vendor/github.com/klauspost/compress/.gitignore
- vendor/github.com/klauspost/compress/README.md
- vendor/github.com/klauspost/compress/huff0/autogen.go
- vendor/github.com/klauspost/compress/huff0/bitreader.go
- vendor/github.com/klauspost/compress/huff0/bitwriter.go
- vendor/github.com/klauspost/compress/huff0/bytereader.go
- vendor/github.com/klauspost/compress/huff0/compress.go
- vendor/github.com/klauspost/compress/huff0/decompress.go
- vendor/github.com/klauspost/compress/huff0/decompress_8b_amd64.s
- vendor/github.com/klauspost/compress/huff0/decompress_8b_amd64.s.in
- vendor/github.com/klauspost/compress/huff0/decompress_amd64.go
- vendor/github.com/klauspost/compress/huff0/decompress_amd64.s
- vendor/github.com/klauspost/compress/huff0/decompress_amd64.s.in
- vendor/github.com/klauspost/compress/huff0/decompress_generic.go
- vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo.go
- vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.go
- vendor/github.com/klauspost/compress/internal/cpuinfo/cpuinfo_amd64.s
- vendor/github.com/klauspost/compress/zstd/README.md
- vendor/github.com/klauspost/compress/zstd/bitreader.go
- vendor/github.com/klauspost/compress/zstd/bitwriter.go
- vendor/github.com/klauspost/compress/zstd/blockdec.go
- vendor/github.com/klauspost/compress/zstd/bytebuf.go
- vendor/github.com/klauspost/compress/zstd/bytereader.go
- vendor/github.com/klauspost/compress/zstd/decoder.go
- vendor/github.com/klauspost/compress/zstd/decoder_options.go
- vendor/github.com/klauspost/compress/zstd/enc_better.go
- vendor/github.com/klauspost/compress/zstd/enc_dfast.go
- vendor/github.com/klauspost/compress/zstd/encoder.go
- vendor/github.com/klauspost/compress/zstd/encoder_options.go
- vendor/github.com/klauspost/compress/zstd/framedec.go
- vendor/github.com/klauspost/compress/zstd/fse_decoder.go
- vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.go
- vendor/github.com/klauspost/compress/zstd/fse_decoder_amd64.s
- vendor/github.com/klauspost/compress/zstd/fse_decoder_generic.go
- vendor/github.com/klauspost/compress/zstd/fse_encoder.go
- vendor/github.com/klauspost/compress/zstd/fuzz.go
- vendor/github.com/klauspost/compress/zstd/fuzz_none.go
- vendor/github.com/klauspost/compress/zstd/hash.go
- vendor/github.com/klauspost/compress/zstd/seqdec.go
- vendor/github.com/klauspost/compress/zstd/seqdec_amd64.go
- vendor/github.com/klauspost/compress/zstd/seqdec_amd64.s
- vendor/github.com/klauspost/compress/zstd/seqdec_generic.go
- vendor/github.com/klauspost/compress/zstd/zip.go
- vendor/github.com/klauspost/compress/zstd/zstd.go
- vendor/modules.txt
| ... | ... |
@@ -47,7 +47,7 @@ require ( |
| 47 | 47 |
github.com/hashicorp/serf v0.8.5 |
| 48 | 48 |
github.com/imdario/mergo v0.3.12 |
| 49 | 49 |
github.com/ishidawataru/sctp v0.0.0-20210707070123-9a39160e9062 |
| 50 |
- github.com/klauspost/compress v1.15.1 |
|
| 50 |
+ github.com/klauspost/compress v1.15.9 |
|
| 51 | 51 |
github.com/miekg/dns v1.1.27 |
| 52 | 52 |
github.com/mistifyio/go-zfs v2.1.2-0.20190413222219-f784269be439+incompatible |
| 53 | 53 |
github.com/moby/buildkit v0.10.4 |
| ... | ... |
@@ -691,8 +691,9 @@ github.com/kisielk/errcheck v1.5.0/go.mod h1:pFxgyoBC7bSaBwPgfKdkLd5X25qrDl4LWUI |
| 691 | 691 |
github.com/kisielk/gotool v1.0.0/go.mod h1:XhKaO+MFFWcvkIS/tQcRk01m1F5IRFswLeQ+oQHNcck= |
| 692 | 692 |
github.com/klauspost/compress v1.11.3/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs= |
| 693 | 693 |
github.com/klauspost/compress v1.11.13/go.mod h1:aoV0uJVorq1K+umq18yTdKaF57EivdYsUV+/s2qKfXs= |
| 694 |
-github.com/klauspost/compress v1.15.1 h1:y9FcTHGyrebwfP0ZZqFiaxTaiDnUrGkJkI+f583BL1A= |
|
| 695 | 694 |
github.com/klauspost/compress v1.15.1/go.mod h1:/3/Vjq9QcHkK5uEr5lBEmyoZ1iFhe47etQ6QUkpK6sk= |
| 695 |
+github.com/klauspost/compress v1.15.9 h1:wKRjX6JRtDdrE9qwa4b/Cip7ACOshUI4smpCQanqjSY= |
|
| 696 |
+github.com/klauspost/compress v1.15.9/go.mod h1:PhcZ0MbTNciWF3rruxRgKxI5NkcHHrHUDtV4Yw2GlzU= |
|
| 696 | 697 |
github.com/konsorten/go-windows-terminal-sequences v1.0.1/go.mod h1:T0+1ngSBFLxvqU3pZ+m/2kptfBszLMUkC4ZK/EgS/cQ= |
| 697 | 698 |
github.com/konsorten/go-windows-terminal-sequences v1.0.2/go.mod h1:T0+1ngSBFLxvqU3pZ+m/2kptfBszLMUkC4ZK/EgS/cQ= |
| 698 | 699 |
github.com/konsorten/go-windows-terminal-sequences v1.0.3/go.mod h1:T0+1ngSBFLxvqU3pZ+m/2kptfBszLMUkC4ZK/EgS/cQ= |
| ... | ... |
@@ -17,6 +17,72 @@ This package provides various compression algorithms. |
| 17 | 17 |
|
| 18 | 18 |
# changelog |
| 19 | 19 |
|
| 20 |
+* July 13, 2022 (v1.15.8) |
|
| 21 |
+ |
|
| 22 |
+ * gzip: fix stack exhaustion bug in Reader.Read https://github.com/klauspost/compress/pull/641 |
|
| 23 |
+ * s2: Add Index header trim/restore https://github.com/klauspost/compress/pull/638 |
|
| 24 |
+ * zstd: Optimize seqdeq amd64 asm by @greatroar in https://github.com/klauspost/compress/pull/636 |
|
| 25 |
+ * zstd: Improve decoder memcopy https://github.com/klauspost/compress/pull/637 |
|
| 26 |
+ * huff0: Pass a single bitReader pointer to asm by @greatroar in https://github.com/klauspost/compress/pull/634 |
|
| 27 |
+ * zstd: Branchless getBits for amd64 w/o BMI2 by @greatroar in https://github.com/klauspost/compress/pull/640 |
|
| 28 |
+ * gzhttp: Remove header before writing https://github.com/klauspost/compress/pull/639 |
|
| 29 |
+ |
|
| 30 |
+* June 29, 2022 (v1.15.7) |
|
| 31 |
+ |
|
| 32 |
+ * s2: Fix absolute forward seeks https://github.com/klauspost/compress/pull/633 |
|
| 33 |
+ * zip: Merge upstream https://github.com/klauspost/compress/pull/631 |
|
| 34 |
+ * zip: Re-add zip64 fix https://github.com/klauspost/compress/pull/624 |
|
| 35 |
+ * zstd: translate fseDecoder.buildDtable into asm by @WojciechMula in https://github.com/klauspost/compress/pull/598 |
|
| 36 |
+ * flate: Faster histograms https://github.com/klauspost/compress/pull/620 |
|
| 37 |
+ * deflate: Use compound hcode https://github.com/klauspost/compress/pull/622 |
|
| 38 |
+ |
|
| 39 |
+* June 3, 2022 (v1.15.6) |
|
| 40 |
+ * s2: Improve coding for long, close matches https://github.com/klauspost/compress/pull/613 |
|
| 41 |
+ * s2c: Add Snappy/S2 stream recompression https://github.com/klauspost/compress/pull/611 |
|
| 42 |
+ * zstd: Always use configured block size https://github.com/klauspost/compress/pull/605 |
|
| 43 |
+ * zstd: Fix incorrect hash table placement for dict encoding in default https://github.com/klauspost/compress/pull/606 |
|
| 44 |
+ * zstd: Apply default config to ZipDecompressor without options https://github.com/klauspost/compress/pull/608 |
|
| 45 |
+ * gzhttp: Exclude more common archive formats https://github.com/klauspost/compress/pull/612 |
|
| 46 |
+ * s2: Add ReaderIgnoreCRC https://github.com/klauspost/compress/pull/609 |
|
| 47 |
+ * s2: Remove sanity load on index creation https://github.com/klauspost/compress/pull/607 |
|
| 48 |
+ * snappy: Use dedicated function for scoring https://github.com/klauspost/compress/pull/614 |
|
| 49 |
+ * s2c+s2d: Use official snappy framed extension https://github.com/klauspost/compress/pull/610 |
|
| 50 |
+ |
|
| 51 |
+* May 25, 2022 (v1.15.5) |
|
| 52 |
+ * s2: Add concurrent stream decompression https://github.com/klauspost/compress/pull/602 |
|
| 53 |
+ * s2: Fix final emit oob read crash on amd64 https://github.com/klauspost/compress/pull/601 |
|
| 54 |
+ * huff0: asm implementation of Decompress1X by @WojciechMula https://github.com/klauspost/compress/pull/596 |
|
| 55 |
+ * zstd: Use 1 less goroutine for stream decoding https://github.com/klauspost/compress/pull/588 |
|
| 56 |
+ * zstd: Copy literal in 16 byte blocks when possible https://github.com/klauspost/compress/pull/592 |
|
| 57 |
+ * zstd: Speed up when WithDecoderLowmem(false) https://github.com/klauspost/compress/pull/599 |
|
| 58 |
+ * zstd: faster next state update in BMI2 version of decode by @WojciechMula in https://github.com/klauspost/compress/pull/593 |
|
| 59 |
+ * huff0: Do not check max size when reading table. https://github.com/klauspost/compress/pull/586 |
|
| 60 |
+ * flate: Inplace hashing for level 7-9 by @klauspost in https://github.com/klauspost/compress/pull/590 |
|
| 61 |
+ |
|
| 62 |
+ |
|
| 63 |
+* May 11, 2022 (v1.15.4) |
|
| 64 |
+ * huff0: decompress directly into output by @WojciechMula in [#577](https://github.com/klauspost/compress/pull/577) |
|
| 65 |
+ * inflate: Keep dict on stack [#581](https://github.com/klauspost/compress/pull/581) |
|
| 66 |
+ * zstd: Faster decoding memcopy in asm [#583](https://github.com/klauspost/compress/pull/583) |
|
| 67 |
+ * zstd: Fix ignored crc [#580](https://github.com/klauspost/compress/pull/580) |
|
| 68 |
+ |
|
| 69 |
+* May 5, 2022 (v1.15.3) |
|
| 70 |
+ * zstd: Allow to ignore checksum checking by @WojciechMula [#572](https://github.com/klauspost/compress/pull/572) |
|
| 71 |
+ * s2: Fix incorrect seek for io.SeekEnd in [#575](https://github.com/klauspost/compress/pull/575) |
|
| 72 |
+ |
|
| 73 |
+* Apr 26, 2022 (v1.15.2) |
|
| 74 |
+ * zstd: Add x86-64 assembly for decompression on streams and blocks. Contributed by [@WojciechMula](https://github.com/WojciechMula). Typically 2x faster. [#528](https://github.com/klauspost/compress/pull/528) [#531](https://github.com/klauspost/compress/pull/531) [#545](https://github.com/klauspost/compress/pull/545) [#537](https://github.com/klauspost/compress/pull/537) |
|
| 75 |
+ * zstd: Add options to ZipDecompressor and fixes [#539](https://github.com/klauspost/compress/pull/539) |
|
| 76 |
+ * s2: Use sorted search for index [#555](https://github.com/klauspost/compress/pull/555) |
|
| 77 |
+ * Minimum version is Go 1.16, added CI test on 1.18. |
|
| 78 |
+ |
|
| 79 |
+* Mar 11, 2022 (v1.15.1) |
|
| 80 |
+ * huff0: Add x86 assembly of Decode4X by @WojciechMula in [#512](https://github.com/klauspost/compress/pull/512) |
|
| 81 |
+ * zstd: Reuse zip decoders in [#514](https://github.com/klauspost/compress/pull/514) |
|
| 82 |
+ * zstd: Detect extra block data and report as corrupted in [#520](https://github.com/klauspost/compress/pull/520) |
|
| 83 |
+ * zstd: Handle zero sized frame content size stricter in [#521](https://github.com/klauspost/compress/pull/521) |
|
| 84 |
+ * zstd: Add stricter block size checks in [#523](https://github.com/klauspost/compress/pull/523) |
|
| 85 |
+ |
|
| 20 | 86 |
* Mar 3, 2022 (v1.15.0) |
| 21 | 87 |
* zstd: Refactor decoder by @klauspost in [#498](https://github.com/klauspost/compress/pull/498) |
| 22 | 88 |
* zstd: Add stream encoding without goroutines by @klauspost in [#505](https://github.com/klauspost/compress/pull/505) |
| ... | ... |
@@ -60,6 +126,9 @@ While the release has been extensively tested, it is recommended to testing when |
| 60 | 60 |
* zstd: add arm64 xxhash assembly in [#464](https://github.com/klauspost/compress/pull/464) |
| 61 | 61 |
* Add garbled for binaries for s2 in [#445](https://github.com/klauspost/compress/pull/445) |
| 62 | 62 |
|
| 63 |
+<details> |
|
| 64 |
+ <summary>See changes to v1.13.x</summary> |
|
| 65 |
+ |
|
| 63 | 66 |
* Aug 30, 2021 (v1.13.5) |
| 64 | 67 |
* gz/zlib/flate: Alias stdlib errors [#425](https://github.com/klauspost/compress/pull/425) |
| 65 | 68 |
* s2: Add block support to commandline tools [#413](https://github.com/klauspost/compress/pull/413) |
| ... | ... |
@@ -88,6 +157,8 @@ While the release has been extensively tested, it is recommended to testing when |
| 88 | 88 |
* Added [gzhttp](https://github.com/klauspost/compress/tree/master/gzhttp#gzip-handler) which allows wrapping HTTP servers and clients with GZIP compressors. |
| 89 | 89 |
* zstd: Detect short invalid signatures [#382](https://github.com/klauspost/compress/pull/382) |
| 90 | 90 |
* zstd: Spawn decoder goroutine only if needed. [#380](https://github.com/klauspost/compress/pull/380) |
| 91 |
+</details> |
|
| 92 |
+ |
|
| 91 | 93 |
|
| 92 | 94 |
<details> |
| 93 | 95 |
<summary>See changes to v1.12.x</summary> |
| ... | ... |
@@ -165,11 +165,6 @@ func (b *bitReaderShifted) peekBitsFast(n uint8) uint16 {
|
| 165 | 165 |
return uint16(b.value >> ((64 - n) & 63)) |
| 166 | 166 |
} |
| 167 | 167 |
|
| 168 |
-// peekTopBits(n) is equvialent to peekBitFast(64 - n) |
|
| 169 |
-func (b *bitReaderShifted) peekTopBits(n uint8) uint16 {
|
|
| 170 |
- return uint16(b.value >> n) |
|
| 171 |
-} |
|
| 172 |
- |
|
| 173 | 168 |
func (b *bitReaderShifted) advance(n uint8) {
|
| 174 | 169 |
b.bitsRead += n |
| 175 | 170 |
b.value <<= n & 63 |
| ... | ... |
@@ -220,11 +215,6 @@ func (b *bitReaderShifted) fill() {
|
| 220 | 220 |
} |
| 221 | 221 |
} |
| 222 | 222 |
|
| 223 |
-// finished returns true if all bits have been read from the bit stream. |
|
| 224 |
-func (b *bitReaderShifted) finished() bool {
|
|
| 225 |
- return b.off == 0 && b.bitsRead >= 64 |
|
| 226 |
-} |
|
| 227 |
- |
|
| 228 | 223 |
func (b *bitReaderShifted) remaining() uint {
|
| 229 | 224 |
return b.off*8 + uint(64-b.bitsRead) |
| 230 | 225 |
} |
| ... | ... |
@@ -5,8 +5,6 @@ |
| 5 | 5 |
|
| 6 | 6 |
package huff0 |
| 7 | 7 |
|
| 8 |
-import "fmt" |
|
| 9 |
- |
|
| 10 | 8 |
// bitWriter will write bits. |
| 11 | 9 |
// First bit will be LSB of the first byte of output. |
| 12 | 10 |
type bitWriter struct {
|
| ... | ... |
@@ -23,14 +21,6 @@ var bitMask16 = [32]uint16{
|
| 23 | 23 |
0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, |
| 24 | 24 |
0xFFFF, 0xFFFF} /* up to 16 bits */ |
| 25 | 25 |
|
| 26 |
-// addBits16NC will add up to 16 bits. |
|
| 27 |
-// It will not check if there is space for them, |
|
| 28 |
-// so the caller must ensure that it has flushed recently. |
|
| 29 |
-func (b *bitWriter) addBits16NC(value uint16, bits uint8) {
|
|
| 30 |
- b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63) |
|
| 31 |
- b.nBits += bits |
|
| 32 |
-} |
|
| 33 |
- |
|
| 34 | 26 |
// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated. |
| 35 | 27 |
// It will not check if there is space for them, so the caller must ensure that it has flushed recently. |
| 36 | 28 |
func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
|
| ... | ... |
@@ -70,104 +60,6 @@ func (b *bitWriter) encTwoSymbols(ct cTable, av, bv byte) {
|
| 70 | 70 |
b.nBits += encA.nBits + encB.nBits |
| 71 | 71 |
} |
| 72 | 72 |
|
| 73 |
-// addBits16ZeroNC will add up to 16 bits. |
|
| 74 |
-// It will not check if there is space for them, |
|
| 75 |
-// so the caller must ensure that it has flushed recently. |
|
| 76 |
-// This is fastest if bits can be zero. |
|
| 77 |
-func (b *bitWriter) addBits16ZeroNC(value uint16, bits uint8) {
|
|
| 78 |
- if bits == 0 {
|
|
| 79 |
- return |
|
| 80 |
- } |
|
| 81 |
- value <<= (16 - bits) & 15 |
|
| 82 |
- value >>= (16 - bits) & 15 |
|
| 83 |
- b.bitContainer |= uint64(value) << (b.nBits & 63) |
|
| 84 |
- b.nBits += bits |
|
| 85 |
-} |
|
| 86 |
- |
|
| 87 |
-// flush will flush all pending full bytes. |
|
| 88 |
-// There will be at least 56 bits available for writing when this has been called. |
|
| 89 |
-// Using flush32 is faster, but leaves less space for writing. |
|
| 90 |
-func (b *bitWriter) flush() {
|
|
| 91 |
- v := b.nBits >> 3 |
|
| 92 |
- switch v {
|
|
| 93 |
- case 0: |
|
| 94 |
- return |
|
| 95 |
- case 1: |
|
| 96 |
- b.out = append(b.out, |
|
| 97 |
- byte(b.bitContainer), |
|
| 98 |
- ) |
|
| 99 |
- b.bitContainer >>= 1 << 3 |
|
| 100 |
- case 2: |
|
| 101 |
- b.out = append(b.out, |
|
| 102 |
- byte(b.bitContainer), |
|
| 103 |
- byte(b.bitContainer>>8), |
|
| 104 |
- ) |
|
| 105 |
- b.bitContainer >>= 2 << 3 |
|
| 106 |
- case 3: |
|
| 107 |
- b.out = append(b.out, |
|
| 108 |
- byte(b.bitContainer), |
|
| 109 |
- byte(b.bitContainer>>8), |
|
| 110 |
- byte(b.bitContainer>>16), |
|
| 111 |
- ) |
|
| 112 |
- b.bitContainer >>= 3 << 3 |
|
| 113 |
- case 4: |
|
| 114 |
- b.out = append(b.out, |
|
| 115 |
- byte(b.bitContainer), |
|
| 116 |
- byte(b.bitContainer>>8), |
|
| 117 |
- byte(b.bitContainer>>16), |
|
| 118 |
- byte(b.bitContainer>>24), |
|
| 119 |
- ) |
|
| 120 |
- b.bitContainer >>= 4 << 3 |
|
| 121 |
- case 5: |
|
| 122 |
- b.out = append(b.out, |
|
| 123 |
- byte(b.bitContainer), |
|
| 124 |
- byte(b.bitContainer>>8), |
|
| 125 |
- byte(b.bitContainer>>16), |
|
| 126 |
- byte(b.bitContainer>>24), |
|
| 127 |
- byte(b.bitContainer>>32), |
|
| 128 |
- ) |
|
| 129 |
- b.bitContainer >>= 5 << 3 |
|
| 130 |
- case 6: |
|
| 131 |
- b.out = append(b.out, |
|
| 132 |
- byte(b.bitContainer), |
|
| 133 |
- byte(b.bitContainer>>8), |
|
| 134 |
- byte(b.bitContainer>>16), |
|
| 135 |
- byte(b.bitContainer>>24), |
|
| 136 |
- byte(b.bitContainer>>32), |
|
| 137 |
- byte(b.bitContainer>>40), |
|
| 138 |
- ) |
|
| 139 |
- b.bitContainer >>= 6 << 3 |
|
| 140 |
- case 7: |
|
| 141 |
- b.out = append(b.out, |
|
| 142 |
- byte(b.bitContainer), |
|
| 143 |
- byte(b.bitContainer>>8), |
|
| 144 |
- byte(b.bitContainer>>16), |
|
| 145 |
- byte(b.bitContainer>>24), |
|
| 146 |
- byte(b.bitContainer>>32), |
|
| 147 |
- byte(b.bitContainer>>40), |
|
| 148 |
- byte(b.bitContainer>>48), |
|
| 149 |
- ) |
|
| 150 |
- b.bitContainer >>= 7 << 3 |
|
| 151 |
- case 8: |
|
| 152 |
- b.out = append(b.out, |
|
| 153 |
- byte(b.bitContainer), |
|
| 154 |
- byte(b.bitContainer>>8), |
|
| 155 |
- byte(b.bitContainer>>16), |
|
| 156 |
- byte(b.bitContainer>>24), |
|
| 157 |
- byte(b.bitContainer>>32), |
|
| 158 |
- byte(b.bitContainer>>40), |
|
| 159 |
- byte(b.bitContainer>>48), |
|
| 160 |
- byte(b.bitContainer>>56), |
|
| 161 |
- ) |
|
| 162 |
- b.bitContainer = 0 |
|
| 163 |
- b.nBits = 0 |
|
| 164 |
- return |
|
| 165 |
- default: |
|
| 166 |
- panic(fmt.Errorf("bits (%d) > 64", b.nBits))
|
|
| 167 |
- } |
|
| 168 |
- b.nBits &= 7 |
|
| 169 |
-} |
|
| 170 |
- |
|
| 171 | 73 |
// flush32 will flush out, so there are at least 32 bits available for writing. |
| 172 | 74 |
func (b *bitWriter) flush32() {
|
| 173 | 75 |
if b.nBits < 32 {
|
| ... | ... |
@@ -201,10 +93,3 @@ func (b *bitWriter) close() error {
|
| 201 | 201 |
b.flushAlign() |
| 202 | 202 |
return nil |
| 203 | 203 |
} |
| 204 |
- |
|
| 205 |
-// reset and continue writing by appending to out. |
|
| 206 |
-func (b *bitWriter) reset(out []byte) {
|
|
| 207 |
- b.bitContainer = 0 |
|
| 208 |
- b.nBits = 0 |
|
| 209 |
- b.out = out |
|
| 210 |
-} |
| ... | ... |
@@ -20,11 +20,6 @@ func (b *byteReader) init(in []byte) {
|
| 20 | 20 |
b.off = 0 |
| 21 | 21 |
} |
| 22 | 22 |
|
| 23 |
-// advance the stream b n bytes. |
|
| 24 |
-func (b *byteReader) advance(n uint) {
|
|
| 25 |
- b.off += int(n) |
|
| 26 |
-} |
|
| 27 |
- |
|
| 28 | 23 |
// Int32 returns a little endian int32 starting at current offset. |
| 29 | 24 |
func (b byteReader) Int32() int32 {
|
| 30 | 25 |
v3 := int32(b.b[b.off+3]) |
| ... | ... |
@@ -43,11 +38,6 @@ func (b byteReader) Uint32() uint32 {
|
| 43 | 43 |
return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0 |
| 44 | 44 |
} |
| 45 | 45 |
|
| 46 |
-// unread returns the unread portion of the input. |
|
| 47 |
-func (b byteReader) unread() []byte {
|
|
| 48 |
- return b.b[b.off:] |
|
| 49 |
-} |
|
| 50 |
- |
|
| 51 | 46 |
// remain will return the number of bytes remaining. |
| 52 | 47 |
func (b byteReader) remain() int {
|
| 53 | 48 |
return len(b.b) - b.off |
| ... | ... |
@@ -11,7 +11,6 @@ import ( |
| 11 | 11 |
|
| 12 | 12 |
type dTable struct {
|
| 13 | 13 |
single []dEntrySingle |
| 14 |
- double []dEntryDouble |
|
| 15 | 14 |
} |
| 16 | 15 |
|
| 17 | 16 |
// single-symbols decoding |
| ... | ... |
@@ -19,13 +18,6 @@ type dEntrySingle struct {
|
| 19 | 19 |
entry uint16 |
| 20 | 20 |
} |
| 21 | 21 |
|
| 22 |
-// double-symbols decoding |
|
| 23 |
-type dEntryDouble struct {
|
|
| 24 |
- seq [4]byte |
|
| 25 |
- nBits uint8 |
|
| 26 |
- len uint8 |
|
| 27 |
-} |
|
| 28 |
- |
|
| 29 | 22 |
// Uses special code for all tables that are < 8 bits. |
| 30 | 23 |
const use8BitTables = true |
| 31 | 24 |
|
| ... | ... |
@@ -35,7 +27,7 @@ const use8BitTables = true |
| 35 | 35 |
// If no Scratch is provided a new one is allocated. |
| 36 | 36 |
// The returned Scratch can be used for encoding or decoding input using this table. |
| 37 | 37 |
func ReadTable(in []byte, s *Scratch) (s2 *Scratch, remain []byte, err error) {
|
| 38 |
- s, err = s.prepare(in) |
|
| 38 |
+ s, err = s.prepare(nil) |
|
| 39 | 39 |
if err != nil {
|
| 40 | 40 |
return s, nil, err |
| 41 | 41 |
} |
| ... | ... |
@@ -236,108 +228,6 @@ func (d *Decoder) buffer() *[4][256]byte {
|
| 236 | 236 |
return &[4][256]byte{}
|
| 237 | 237 |
} |
| 238 | 238 |
|
| 239 |
-// Decompress1X will decompress a 1X encoded stream. |
|
| 240 |
-// The cap of the output buffer will be the maximum decompressed size. |
|
| 241 |
-// The length of the supplied input must match the end of a block exactly. |
|
| 242 |
-func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
|
|
| 243 |
- if len(d.dt.single) == 0 {
|
|
| 244 |
- return nil, errors.New("no table loaded")
|
|
| 245 |
- } |
|
| 246 |
- if use8BitTables && d.actualTableLog <= 8 {
|
|
| 247 |
- return d.decompress1X8Bit(dst, src) |
|
| 248 |
- } |
|
| 249 |
- var br bitReaderShifted |
|
| 250 |
- err := br.init(src) |
|
| 251 |
- if err != nil {
|
|
| 252 |
- return dst, err |
|
| 253 |
- } |
|
| 254 |
- maxDecodedSize := cap(dst) |
|
| 255 |
- dst = dst[:0] |
|
| 256 |
- |
|
| 257 |
- // Avoid bounds check by always having full sized table. |
|
| 258 |
- const tlSize = 1 << tableLogMax |
|
| 259 |
- const tlMask = tlSize - 1 |
|
| 260 |
- dt := d.dt.single[:tlSize] |
|
| 261 |
- |
|
| 262 |
- // Use temp table to avoid bound checks/append penalty. |
|
| 263 |
- bufs := d.buffer() |
|
| 264 |
- buf := &bufs[0] |
|
| 265 |
- var off uint8 |
|
| 266 |
- |
|
| 267 |
- for br.off >= 8 {
|
|
| 268 |
- br.fillFast() |
|
| 269 |
- v := dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 270 |
- br.advance(uint8(v.entry)) |
|
| 271 |
- buf[off+0] = uint8(v.entry >> 8) |
|
| 272 |
- |
|
| 273 |
- v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 274 |
- br.advance(uint8(v.entry)) |
|
| 275 |
- buf[off+1] = uint8(v.entry >> 8) |
|
| 276 |
- |
|
| 277 |
- // Refill |
|
| 278 |
- br.fillFast() |
|
| 279 |
- |
|
| 280 |
- v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 281 |
- br.advance(uint8(v.entry)) |
|
| 282 |
- buf[off+2] = uint8(v.entry >> 8) |
|
| 283 |
- |
|
| 284 |
- v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 285 |
- br.advance(uint8(v.entry)) |
|
| 286 |
- buf[off+3] = uint8(v.entry >> 8) |
|
| 287 |
- |
|
| 288 |
- off += 4 |
|
| 289 |
- if off == 0 {
|
|
| 290 |
- if len(dst)+256 > maxDecodedSize {
|
|
| 291 |
- br.close() |
|
| 292 |
- d.bufs.Put(bufs) |
|
| 293 |
- return nil, ErrMaxDecodedSizeExceeded |
|
| 294 |
- } |
|
| 295 |
- dst = append(dst, buf[:]...) |
|
| 296 |
- } |
|
| 297 |
- } |
|
| 298 |
- |
|
| 299 |
- if len(dst)+int(off) > maxDecodedSize {
|
|
| 300 |
- d.bufs.Put(bufs) |
|
| 301 |
- br.close() |
|
| 302 |
- return nil, ErrMaxDecodedSizeExceeded |
|
| 303 |
- } |
|
| 304 |
- dst = append(dst, buf[:off]...) |
|
| 305 |
- |
|
| 306 |
- // br < 8, so uint8 is fine |
|
| 307 |
- bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead |
|
| 308 |
- for bitsLeft > 0 {
|
|
| 309 |
- br.fill() |
|
| 310 |
- if false && br.bitsRead >= 32 {
|
|
| 311 |
- if br.off >= 4 {
|
|
| 312 |
- v := br.in[br.off-4:] |
|
| 313 |
- v = v[:4] |
|
| 314 |
- low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 315 |
- br.value = (br.value << 32) | uint64(low) |
|
| 316 |
- br.bitsRead -= 32 |
|
| 317 |
- br.off -= 4 |
|
| 318 |
- } else {
|
|
| 319 |
- for br.off > 0 {
|
|
| 320 |
- br.value = (br.value << 8) | uint64(br.in[br.off-1]) |
|
| 321 |
- br.bitsRead -= 8 |
|
| 322 |
- br.off-- |
|
| 323 |
- } |
|
| 324 |
- } |
|
| 325 |
- } |
|
| 326 |
- if len(dst) >= maxDecodedSize {
|
|
| 327 |
- d.bufs.Put(bufs) |
|
| 328 |
- br.close() |
|
| 329 |
- return nil, ErrMaxDecodedSizeExceeded |
|
| 330 |
- } |
|
| 331 |
- v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 332 |
- nBits := uint8(v.entry) |
|
| 333 |
- br.advance(nBits) |
|
| 334 |
- bitsLeft -= nBits |
|
| 335 |
- dst = append(dst, uint8(v.entry>>8)) |
|
| 336 |
- } |
|
| 337 |
- d.bufs.Put(bufs) |
|
| 338 |
- return dst, br.close() |
|
| 339 |
-} |
|
| 340 |
- |
|
| 341 | 239 |
// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8. |
| 342 | 240 |
// The cap of the output buffer will be the maximum decompressed size. |
| 343 | 241 |
// The length of the supplied input must match the end of a block exactly. |
| ... | ... |
@@ -995,7 +885,6 @@ func (d *Decoder) decompress4X8bitExactly(dst, src []byte) ([]byte, error) {
|
| 995 | 995 |
|
| 996 | 996 |
const shift = 56 |
| 997 | 997 |
const tlSize = 1 << 8 |
| 998 |
- const tlMask = tlSize - 1 |
|
| 999 | 998 |
single := d.dt.single[:tlSize] |
| 1000 | 999 |
|
| 1001 | 1000 |
// Use temp table to avoid bound checks/append penalty. |
| 1002 | 1001 |
deleted file mode 100644 |
| ... | ... |
@@ -1,488 +0,0 @@ |
| 1 |
-// +build !appengine |
|
| 2 |
-// +build gc |
|
| 3 |
-// +build !noasm |
|
| 4 |
- |
|
| 5 |
-#include "textflag.h" |
|
| 6 |
-#include "funcdata.h" |
|
| 7 |
-#include "go_asm.h" |
|
| 8 |
- |
|
| 9 |
-#define bufoff 256 // see decompress.go, we're using [4][256]byte table |
|
| 10 |
- |
|
| 11 |
-// func decompress4x_main_loop_x86(pbr0, pbr1, pbr2, pbr3 *bitReaderShifted, |
|
| 12 |
-// peekBits uint8, buf *byte, tbl *dEntrySingle) (int, bool) |
|
| 13 |
-TEXT ·decompress4x_8b_loop_x86(SB), NOSPLIT, $8 |
|
| 14 |
-#define off R8 |
|
| 15 |
-#define buffer DI |
|
| 16 |
-#define table SI |
|
| 17 |
- |
|
| 18 |
-#define br_bits_read R9 |
|
| 19 |
-#define br_value R10 |
|
| 20 |
-#define br_offset R11 |
|
| 21 |
-#define peek_bits R12 |
|
| 22 |
-#define exhausted DX |
|
| 23 |
- |
|
| 24 |
-#define br0 R13 |
|
| 25 |
-#define br1 R14 |
|
| 26 |
-#define br2 R15 |
|
| 27 |
-#define br3 BP |
|
| 28 |
- |
|
| 29 |
- MOVQ BP, 0(SP) |
|
| 30 |
- |
|
| 31 |
- XORQ exhausted, exhausted // exhausted = false |
|
| 32 |
- XORQ off, off // off = 0 |
|
| 33 |
- |
|
| 34 |
- MOVBQZX peekBits+32(FP), peek_bits |
|
| 35 |
- MOVQ buf+40(FP), buffer |
|
| 36 |
- MOVQ tbl+48(FP), table |
|
| 37 |
- |
|
| 38 |
- MOVQ pbr0+0(FP), br0 |
|
| 39 |
- MOVQ pbr1+8(FP), br1 |
|
| 40 |
- MOVQ pbr2+16(FP), br2 |
|
| 41 |
- MOVQ pbr3+24(FP), br3 |
|
| 42 |
- |
|
| 43 |
-main_loop: |
|
| 44 |
- |
|
| 45 |
- // const stream = 0 |
|
| 46 |
- // br0.fillFast() |
|
| 47 |
- MOVBQZX bitReaderShifted_bitsRead(br0), br_bits_read |
|
| 48 |
- MOVQ bitReaderShifted_value(br0), br_value |
|
| 49 |
- MOVQ bitReaderShifted_off(br0), br_offset |
|
| 50 |
- |
|
| 51 |
- // if b.bitsRead >= 32 {
|
|
| 52 |
- CMPQ br_bits_read, $32 |
|
| 53 |
- JB skip_fill0 |
|
| 54 |
- |
|
| 55 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 56 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 57 |
- |
|
| 58 |
- // v := b.in[b.off-4 : b.off] |
|
| 59 |
- // v = v[:4] |
|
| 60 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 61 |
- MOVQ bitReaderShifted_in(br0), AX |
|
| 62 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 63 |
- |
|
| 64 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 65 |
- MOVQ br_bits_read, CX |
|
| 66 |
- SHLQ CL, AX |
|
| 67 |
- ORQ AX, br_value |
|
| 68 |
- |
|
| 69 |
- // exhausted = exhausted || (br0.off < 4) |
|
| 70 |
- CMPQ br_offset, $4 |
|
| 71 |
- SETLT DL |
|
| 72 |
- ORB DL, DH |
|
| 73 |
- |
|
| 74 |
- // } |
|
| 75 |
-skip_fill0: |
|
| 76 |
- |
|
| 77 |
- // val0 := br0.peekTopBits(peekBits) |
|
| 78 |
- MOVQ br_value, AX |
|
| 79 |
- MOVQ peek_bits, CX |
|
| 80 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 81 |
- |
|
| 82 |
- // v0 := table[val0&mask] |
|
| 83 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 84 |
- |
|
| 85 |
- // br0.advance(uint8(v0.entry)) |
|
| 86 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 87 |
- MOVBQZX AL, CX |
|
| 88 |
- SHLQ CL, br_value // value <<= n |
|
| 89 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 90 |
- |
|
| 91 |
- // val1 := br0.peekTopBits(peekBits) |
|
| 92 |
- MOVQ peek_bits, CX |
|
| 93 |
- MOVQ br_value, AX |
|
| 94 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 95 |
- |
|
| 96 |
- // v1 := table[val1&mask] |
|
| 97 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 98 |
- |
|
| 99 |
- // br0.advance(uint8(v1.entry)) |
|
| 100 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 101 |
- MOVBQZX AL, CX |
|
| 102 |
- SHLQ CX, br_value // value <<= n |
|
| 103 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 104 |
- |
|
| 105 |
- // these two writes get coalesced |
|
| 106 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 107 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 108 |
- MOVW BX, 0(buffer)(off*1) |
|
| 109 |
- |
|
| 110 |
- // SECOND PART: |
|
| 111 |
- // val2 := br0.peekTopBits(peekBits) |
|
| 112 |
- MOVQ br_value, AX |
|
| 113 |
- MOVQ peek_bits, CX |
|
| 114 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 115 |
- |
|
| 116 |
- // v2 := table[val0&mask] |
|
| 117 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 118 |
- |
|
| 119 |
- // br0.advance(uint8(v0.entry)) |
|
| 120 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 121 |
- MOVBQZX AL, CX |
|
| 122 |
- SHLQ CL, br_value // value <<= n |
|
| 123 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 124 |
- |
|
| 125 |
- // val3 := br0.peekTopBits(peekBits) |
|
| 126 |
- MOVQ peek_bits, CX |
|
| 127 |
- MOVQ br_value, AX |
|
| 128 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 129 |
- |
|
| 130 |
- // v3 := table[val1&mask] |
|
| 131 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 132 |
- |
|
| 133 |
- // br0.advance(uint8(v1.entry)) |
|
| 134 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 135 |
- MOVBQZX AL, CX |
|
| 136 |
- SHLQ CX, br_value // value <<= n |
|
| 137 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 138 |
- |
|
| 139 |
- // these two writes get coalesced |
|
| 140 |
- // buf[stream][off+2] = uint8(v2.entry >> 8) |
|
| 141 |
- // buf[stream][off+3] = uint8(v3.entry >> 8) |
|
| 142 |
- MOVW BX, 0+2(buffer)(off*1) |
|
| 143 |
- |
|
| 144 |
- // update the bitrader reader structure |
|
| 145 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br0) |
|
| 146 |
- MOVQ br_value, bitReaderShifted_value(br0) |
|
| 147 |
- MOVQ br_offset, bitReaderShifted_off(br0) |
|
| 148 |
- |
|
| 149 |
- // const stream = 1 |
|
| 150 |
- // br1.fillFast() |
|
| 151 |
- MOVBQZX bitReaderShifted_bitsRead(br1), br_bits_read |
|
| 152 |
- MOVQ bitReaderShifted_value(br1), br_value |
|
| 153 |
- MOVQ bitReaderShifted_off(br1), br_offset |
|
| 154 |
- |
|
| 155 |
- // if b.bitsRead >= 32 {
|
|
| 156 |
- CMPQ br_bits_read, $32 |
|
| 157 |
- JB skip_fill1 |
|
| 158 |
- |
|
| 159 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 160 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 161 |
- |
|
| 162 |
- // v := b.in[b.off-4 : b.off] |
|
| 163 |
- // v = v[:4] |
|
| 164 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 165 |
- MOVQ bitReaderShifted_in(br1), AX |
|
| 166 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 167 |
- |
|
| 168 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 169 |
- MOVQ br_bits_read, CX |
|
| 170 |
- SHLQ CL, AX |
|
| 171 |
- ORQ AX, br_value |
|
| 172 |
- |
|
| 173 |
- // exhausted = exhausted || (br1.off < 4) |
|
| 174 |
- CMPQ br_offset, $4 |
|
| 175 |
- SETLT DL |
|
| 176 |
- ORB DL, DH |
|
| 177 |
- |
|
| 178 |
- // } |
|
| 179 |
-skip_fill1: |
|
| 180 |
- |
|
| 181 |
- // val0 := br1.peekTopBits(peekBits) |
|
| 182 |
- MOVQ br_value, AX |
|
| 183 |
- MOVQ peek_bits, CX |
|
| 184 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 185 |
- |
|
| 186 |
- // v0 := table[val0&mask] |
|
| 187 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 188 |
- |
|
| 189 |
- // br1.advance(uint8(v0.entry)) |
|
| 190 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 191 |
- MOVBQZX AL, CX |
|
| 192 |
- SHLQ CL, br_value // value <<= n |
|
| 193 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 194 |
- |
|
| 195 |
- // val1 := br1.peekTopBits(peekBits) |
|
| 196 |
- MOVQ peek_bits, CX |
|
| 197 |
- MOVQ br_value, AX |
|
| 198 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 199 |
- |
|
| 200 |
- // v1 := table[val1&mask] |
|
| 201 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 202 |
- |
|
| 203 |
- // br1.advance(uint8(v1.entry)) |
|
| 204 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 205 |
- MOVBQZX AL, CX |
|
| 206 |
- SHLQ CX, br_value // value <<= n |
|
| 207 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 208 |
- |
|
| 209 |
- // these two writes get coalesced |
|
| 210 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 211 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 212 |
- MOVW BX, 256(buffer)(off*1) |
|
| 213 |
- |
|
| 214 |
- // SECOND PART: |
|
| 215 |
- // val2 := br1.peekTopBits(peekBits) |
|
| 216 |
- MOVQ br_value, AX |
|
| 217 |
- MOVQ peek_bits, CX |
|
| 218 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 219 |
- |
|
| 220 |
- // v2 := table[val0&mask] |
|
| 221 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 222 |
- |
|
| 223 |
- // br1.advance(uint8(v0.entry)) |
|
| 224 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 225 |
- MOVBQZX AL, CX |
|
| 226 |
- SHLQ CL, br_value // value <<= n |
|
| 227 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 228 |
- |
|
| 229 |
- // val3 := br1.peekTopBits(peekBits) |
|
| 230 |
- MOVQ peek_bits, CX |
|
| 231 |
- MOVQ br_value, AX |
|
| 232 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 233 |
- |
|
| 234 |
- // v3 := table[val1&mask] |
|
| 235 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 236 |
- |
|
| 237 |
- // br1.advance(uint8(v1.entry)) |
|
| 238 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 239 |
- MOVBQZX AL, CX |
|
| 240 |
- SHLQ CX, br_value // value <<= n |
|
| 241 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 242 |
- |
|
| 243 |
- // these two writes get coalesced |
|
| 244 |
- // buf[stream][off+2] = uint8(v2.entry >> 8) |
|
| 245 |
- // buf[stream][off+3] = uint8(v3.entry >> 8) |
|
| 246 |
- MOVW BX, 256+2(buffer)(off*1) |
|
| 247 |
- |
|
| 248 |
- // update the bitrader reader structure |
|
| 249 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br1) |
|
| 250 |
- MOVQ br_value, bitReaderShifted_value(br1) |
|
| 251 |
- MOVQ br_offset, bitReaderShifted_off(br1) |
|
| 252 |
- |
|
| 253 |
- // const stream = 2 |
|
| 254 |
- // br2.fillFast() |
|
| 255 |
- MOVBQZX bitReaderShifted_bitsRead(br2), br_bits_read |
|
| 256 |
- MOVQ bitReaderShifted_value(br2), br_value |
|
| 257 |
- MOVQ bitReaderShifted_off(br2), br_offset |
|
| 258 |
- |
|
| 259 |
- // if b.bitsRead >= 32 {
|
|
| 260 |
- CMPQ br_bits_read, $32 |
|
| 261 |
- JB skip_fill2 |
|
| 262 |
- |
|
| 263 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 264 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 265 |
- |
|
| 266 |
- // v := b.in[b.off-4 : b.off] |
|
| 267 |
- // v = v[:4] |
|
| 268 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 269 |
- MOVQ bitReaderShifted_in(br2), AX |
|
| 270 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 271 |
- |
|
| 272 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 273 |
- MOVQ br_bits_read, CX |
|
| 274 |
- SHLQ CL, AX |
|
| 275 |
- ORQ AX, br_value |
|
| 276 |
- |
|
| 277 |
- // exhausted = exhausted || (br2.off < 4) |
|
| 278 |
- CMPQ br_offset, $4 |
|
| 279 |
- SETLT DL |
|
| 280 |
- ORB DL, DH |
|
| 281 |
- |
|
| 282 |
- // } |
|
| 283 |
-skip_fill2: |
|
| 284 |
- |
|
| 285 |
- // val0 := br2.peekTopBits(peekBits) |
|
| 286 |
- MOVQ br_value, AX |
|
| 287 |
- MOVQ peek_bits, CX |
|
| 288 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 289 |
- |
|
| 290 |
- // v0 := table[val0&mask] |
|
| 291 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 292 |
- |
|
| 293 |
- // br2.advance(uint8(v0.entry)) |
|
| 294 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 295 |
- MOVBQZX AL, CX |
|
| 296 |
- SHLQ CL, br_value // value <<= n |
|
| 297 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 298 |
- |
|
| 299 |
- // val1 := br2.peekTopBits(peekBits) |
|
| 300 |
- MOVQ peek_bits, CX |
|
| 301 |
- MOVQ br_value, AX |
|
| 302 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 303 |
- |
|
| 304 |
- // v1 := table[val1&mask] |
|
| 305 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 306 |
- |
|
| 307 |
- // br2.advance(uint8(v1.entry)) |
|
| 308 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 309 |
- MOVBQZX AL, CX |
|
| 310 |
- SHLQ CX, br_value // value <<= n |
|
| 311 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 312 |
- |
|
| 313 |
- // these two writes get coalesced |
|
| 314 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 315 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 316 |
- MOVW BX, 512(buffer)(off*1) |
|
| 317 |
- |
|
| 318 |
- // SECOND PART: |
|
| 319 |
- // val2 := br2.peekTopBits(peekBits) |
|
| 320 |
- MOVQ br_value, AX |
|
| 321 |
- MOVQ peek_bits, CX |
|
| 322 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 323 |
- |
|
| 324 |
- // v2 := table[val0&mask] |
|
| 325 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 326 |
- |
|
| 327 |
- // br2.advance(uint8(v0.entry)) |
|
| 328 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 329 |
- MOVBQZX AL, CX |
|
| 330 |
- SHLQ CL, br_value // value <<= n |
|
| 331 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 332 |
- |
|
| 333 |
- // val3 := br2.peekTopBits(peekBits) |
|
| 334 |
- MOVQ peek_bits, CX |
|
| 335 |
- MOVQ br_value, AX |
|
| 336 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 337 |
- |
|
| 338 |
- // v3 := table[val1&mask] |
|
| 339 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 340 |
- |
|
| 341 |
- // br2.advance(uint8(v1.entry)) |
|
| 342 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 343 |
- MOVBQZX AL, CX |
|
| 344 |
- SHLQ CX, br_value // value <<= n |
|
| 345 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 346 |
- |
|
| 347 |
- // these two writes get coalesced |
|
| 348 |
- // buf[stream][off+2] = uint8(v2.entry >> 8) |
|
| 349 |
- // buf[stream][off+3] = uint8(v3.entry >> 8) |
|
| 350 |
- MOVW BX, 512+2(buffer)(off*1) |
|
| 351 |
- |
|
| 352 |
- // update the bitrader reader structure |
|
| 353 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br2) |
|
| 354 |
- MOVQ br_value, bitReaderShifted_value(br2) |
|
| 355 |
- MOVQ br_offset, bitReaderShifted_off(br2) |
|
| 356 |
- |
|
| 357 |
- // const stream = 3 |
|
| 358 |
- // br3.fillFast() |
|
| 359 |
- MOVBQZX bitReaderShifted_bitsRead(br3), br_bits_read |
|
| 360 |
- MOVQ bitReaderShifted_value(br3), br_value |
|
| 361 |
- MOVQ bitReaderShifted_off(br3), br_offset |
|
| 362 |
- |
|
| 363 |
- // if b.bitsRead >= 32 {
|
|
| 364 |
- CMPQ br_bits_read, $32 |
|
| 365 |
- JB skip_fill3 |
|
| 366 |
- |
|
| 367 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 368 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 369 |
- |
|
| 370 |
- // v := b.in[b.off-4 : b.off] |
|
| 371 |
- // v = v[:4] |
|
| 372 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 373 |
- MOVQ bitReaderShifted_in(br3), AX |
|
| 374 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 375 |
- |
|
| 376 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 377 |
- MOVQ br_bits_read, CX |
|
| 378 |
- SHLQ CL, AX |
|
| 379 |
- ORQ AX, br_value |
|
| 380 |
- |
|
| 381 |
- // exhausted = exhausted || (br3.off < 4) |
|
| 382 |
- CMPQ br_offset, $4 |
|
| 383 |
- SETLT DL |
|
| 384 |
- ORB DL, DH |
|
| 385 |
- |
|
| 386 |
- // } |
|
| 387 |
-skip_fill3: |
|
| 388 |
- |
|
| 389 |
- // val0 := br3.peekTopBits(peekBits) |
|
| 390 |
- MOVQ br_value, AX |
|
| 391 |
- MOVQ peek_bits, CX |
|
| 392 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 393 |
- |
|
| 394 |
- // v0 := table[val0&mask] |
|
| 395 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 396 |
- |
|
| 397 |
- // br3.advance(uint8(v0.entry)) |
|
| 398 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 399 |
- MOVBQZX AL, CX |
|
| 400 |
- SHLQ CL, br_value // value <<= n |
|
| 401 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 402 |
- |
|
| 403 |
- // val1 := br3.peekTopBits(peekBits) |
|
| 404 |
- MOVQ peek_bits, CX |
|
| 405 |
- MOVQ br_value, AX |
|
| 406 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 407 |
- |
|
| 408 |
- // v1 := table[val1&mask] |
|
| 409 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 410 |
- |
|
| 411 |
- // br3.advance(uint8(v1.entry)) |
|
| 412 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 413 |
- MOVBQZX AL, CX |
|
| 414 |
- SHLQ CX, br_value // value <<= n |
|
| 415 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 416 |
- |
|
| 417 |
- // these two writes get coalesced |
|
| 418 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 419 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 420 |
- MOVW BX, 768(buffer)(off*1) |
|
| 421 |
- |
|
| 422 |
- // SECOND PART: |
|
| 423 |
- // val2 := br3.peekTopBits(peekBits) |
|
| 424 |
- MOVQ br_value, AX |
|
| 425 |
- MOVQ peek_bits, CX |
|
| 426 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 427 |
- |
|
| 428 |
- // v2 := table[val0&mask] |
|
| 429 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 430 |
- |
|
| 431 |
- // br3.advance(uint8(v0.entry)) |
|
| 432 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 433 |
- MOVBQZX AL, CX |
|
| 434 |
- SHLQ CL, br_value // value <<= n |
|
| 435 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 436 |
- |
|
| 437 |
- // val3 := br3.peekTopBits(peekBits) |
|
| 438 |
- MOVQ peek_bits, CX |
|
| 439 |
- MOVQ br_value, AX |
|
| 440 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 441 |
- |
|
| 442 |
- // v3 := table[val1&mask] |
|
| 443 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 444 |
- |
|
| 445 |
- // br3.advance(uint8(v1.entry)) |
|
| 446 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 447 |
- MOVBQZX AL, CX |
|
| 448 |
- SHLQ CX, br_value // value <<= n |
|
| 449 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 450 |
- |
|
| 451 |
- // these two writes get coalesced |
|
| 452 |
- // buf[stream][off+2] = uint8(v2.entry >> 8) |
|
| 453 |
- // buf[stream][off+3] = uint8(v3.entry >> 8) |
|
| 454 |
- MOVW BX, 768+2(buffer)(off*1) |
|
| 455 |
- |
|
| 456 |
- // update the bitrader reader structure |
|
| 457 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br3) |
|
| 458 |
- MOVQ br_value, bitReaderShifted_value(br3) |
|
| 459 |
- MOVQ br_offset, bitReaderShifted_off(br3) |
|
| 460 |
- |
|
| 461 |
- ADDQ $4, off // off += 2 |
|
| 462 |
- |
|
| 463 |
- TESTB DH, DH // any br[i].ofs < 4? |
|
| 464 |
- JNZ end |
|
| 465 |
- |
|
| 466 |
- CMPQ off, $bufoff |
|
| 467 |
- JL main_loop |
|
| 468 |
- |
|
| 469 |
-end: |
|
| 470 |
- MOVQ 0(SP), BP |
|
| 471 |
- |
|
| 472 |
- MOVB off, ret+56(FP) |
|
| 473 |
- RET |
|
| 474 |
- |
|
| 475 |
-#undef off |
|
| 476 |
-#undef buffer |
|
| 477 |
-#undef table |
|
| 478 |
- |
|
| 479 |
-#undef br_bits_read |
|
| 480 |
-#undef br_value |
|
| 481 |
-#undef br_offset |
|
| 482 |
-#undef peek_bits |
|
| 483 |
-#undef exhausted |
|
| 484 |
- |
|
| 485 |
-#undef br0 |
|
| 486 |
-#undef br1 |
|
| 487 |
-#undef br2 |
|
| 488 |
-#undef br3 |
| 489 | 1 |
deleted file mode 100644 |
| ... | ... |
@@ -1,197 +0,0 @@ |
| 1 |
-// +build !appengine |
|
| 2 |
-// +build gc |
|
| 3 |
-// +build !noasm |
|
| 4 |
- |
|
| 5 |
-#include "textflag.h" |
|
| 6 |
-#include "funcdata.h" |
|
| 7 |
-#include "go_asm.h" |
|
| 8 |
- |
|
| 9 |
- |
|
| 10 |
-#define bufoff 256 // see decompress.go, we're using [4][256]byte table |
|
| 11 |
- |
|
| 12 |
-//func decompress4x_main_loop_x86(pbr0, pbr1, pbr2, pbr3 *bitReaderShifted, |
|
| 13 |
-// peekBits uint8, buf *byte, tbl *dEntrySingle) (int, bool) |
|
| 14 |
-TEXT ·decompress4x_8b_loop_x86(SB), NOSPLIT, $8 |
|
| 15 |
-#define off R8 |
|
| 16 |
-#define buffer DI |
|
| 17 |
-#define table SI |
|
| 18 |
- |
|
| 19 |
-#define br_bits_read R9 |
|
| 20 |
-#define br_value R10 |
|
| 21 |
-#define br_offset R11 |
|
| 22 |
-#define peek_bits R12 |
|
| 23 |
-#define exhausted DX |
|
| 24 |
- |
|
| 25 |
-#define br0 R13 |
|
| 26 |
-#define br1 R14 |
|
| 27 |
-#define br2 R15 |
|
| 28 |
-#define br3 BP |
|
| 29 |
- |
|
| 30 |
- MOVQ BP, 0(SP) |
|
| 31 |
- |
|
| 32 |
- XORQ exhausted, exhausted // exhausted = false |
|
| 33 |
- XORQ off, off // off = 0 |
|
| 34 |
- |
|
| 35 |
- MOVBQZX peekBits+32(FP), peek_bits |
|
| 36 |
- MOVQ buf+40(FP), buffer |
|
| 37 |
- MOVQ tbl+48(FP), table |
|
| 38 |
- |
|
| 39 |
- MOVQ pbr0+0(FP), br0 |
|
| 40 |
- MOVQ pbr1+8(FP), br1 |
|
| 41 |
- MOVQ pbr2+16(FP), br2 |
|
| 42 |
- MOVQ pbr3+24(FP), br3 |
|
| 43 |
- |
|
| 44 |
-main_loop: |
|
| 45 |
-{{ define "decode_2_values_x86" }}
|
|
| 46 |
- // const stream = {{ var "id" }}
|
|
| 47 |
- // br{{ var "id"}}.fillFast()
|
|
| 48 |
- MOVBQZX bitReaderShifted_bitsRead(br{{ var "id" }}), br_bits_read
|
|
| 49 |
- MOVQ bitReaderShifted_value(br{{ var "id" }}), br_value
|
|
| 50 |
- MOVQ bitReaderShifted_off(br{{ var "id" }}), br_offset
|
|
| 51 |
- |
|
| 52 |
- // if b.bitsRead >= 32 {
|
|
| 53 |
- CMPQ br_bits_read, $32 |
|
| 54 |
- JB skip_fill{{ var "id" }}
|
|
| 55 |
- |
|
| 56 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 57 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 58 |
- |
|
| 59 |
- // v := b.in[b.off-4 : b.off] |
|
| 60 |
- // v = v[:4] |
|
| 61 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 62 |
- MOVQ bitReaderShifted_in(br{{ var "id" }}), AX
|
|
| 63 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 64 |
- |
|
| 65 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 66 |
- MOVQ br_bits_read, CX |
|
| 67 |
- SHLQ CL, AX |
|
| 68 |
- ORQ AX, br_value |
|
| 69 |
- |
|
| 70 |
- // exhausted = exhausted || (br{{ var "id"}}.off < 4)
|
|
| 71 |
- CMPQ br_offset, $4 |
|
| 72 |
- SETLT DL |
|
| 73 |
- ORB DL, DH |
|
| 74 |
- // } |
|
| 75 |
-skip_fill{{ var "id" }}:
|
|
| 76 |
- |
|
| 77 |
- // val0 := br{{ var "id"}}.peekTopBits(peekBits)
|
|
| 78 |
- MOVQ br_value, AX |
|
| 79 |
- MOVQ peek_bits, CX |
|
| 80 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 81 |
- |
|
| 82 |
- // v0 := table[val0&mask] |
|
| 83 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 84 |
- |
|
| 85 |
- // br{{ var "id"}}.advance(uint8(v0.entry))
|
|
| 86 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 87 |
- MOVBQZX AL, CX |
|
| 88 |
- SHLQ CL, br_value // value <<= n |
|
| 89 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 90 |
- |
|
| 91 |
- // val1 := br{{ var "id"}}.peekTopBits(peekBits)
|
|
| 92 |
- MOVQ peek_bits, CX |
|
| 93 |
- MOVQ br_value, AX |
|
| 94 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 95 |
- |
|
| 96 |
- // v1 := table[val1&mask] |
|
| 97 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 98 |
- |
|
| 99 |
- // br{{ var "id"}}.advance(uint8(v1.entry))
|
|
| 100 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 101 |
- MOVBQZX AL, CX |
|
| 102 |
- SHLQ CX, br_value // value <<= n |
|
| 103 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 104 |
- |
|
| 105 |
- |
|
| 106 |
- // these two writes get coalesced |
|
| 107 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 108 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 109 |
- MOVW BX, {{ var "bufofs" }}(buffer)(off*1)
|
|
| 110 |
- |
|
| 111 |
- // SECOND PART: |
|
| 112 |
- // val2 := br{{ var "id"}}.peekTopBits(peekBits)
|
|
| 113 |
- MOVQ br_value, AX |
|
| 114 |
- MOVQ peek_bits, CX |
|
| 115 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 116 |
- |
|
| 117 |
- // v2 := table[val0&mask] |
|
| 118 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 119 |
- |
|
| 120 |
- // br{{ var "id"}}.advance(uint8(v0.entry))
|
|
| 121 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 122 |
- MOVBQZX AL, CX |
|
| 123 |
- SHLQ CL, br_value // value <<= n |
|
| 124 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 125 |
- |
|
| 126 |
- // val3 := br{{ var "id"}}.peekTopBits(peekBits)
|
|
| 127 |
- MOVQ peek_bits, CX |
|
| 128 |
- MOVQ br_value, AX |
|
| 129 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 130 |
- |
|
| 131 |
- // v3 := table[val1&mask] |
|
| 132 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 133 |
- |
|
| 134 |
- // br{{ var "id"}}.advance(uint8(v1.entry))
|
|
| 135 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 136 |
- MOVBQZX AL, CX |
|
| 137 |
- SHLQ CX, br_value // value <<= n |
|
| 138 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 139 |
- |
|
| 140 |
- |
|
| 141 |
- // these two writes get coalesced |
|
| 142 |
- // buf[stream][off+2] = uint8(v2.entry >> 8) |
|
| 143 |
- // buf[stream][off+3] = uint8(v3.entry >> 8) |
|
| 144 |
- MOVW BX, {{ var "bufofs" }}+2(buffer)(off*1)
|
|
| 145 |
- |
|
| 146 |
- // update the bitrader reader structure |
|
| 147 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br{{ var "id" }})
|
|
| 148 |
- MOVQ br_value, bitReaderShifted_value(br{{ var "id" }})
|
|
| 149 |
- MOVQ br_offset, bitReaderShifted_off(br{{ var "id" }})
|
|
| 150 |
-{{ end }}
|
|
| 151 |
- |
|
| 152 |
- {{ set "id" "0" }}
|
|
| 153 |
- {{ set "ofs" "0" }}
|
|
| 154 |
- {{ set "bufofs" "0" }} {{/* id * bufoff */}}
|
|
| 155 |
- {{ template "decode_2_values_x86" . }}
|
|
| 156 |
- |
|
| 157 |
- {{ set "id" "1" }}
|
|
| 158 |
- {{ set "ofs" "8" }}
|
|
| 159 |
- {{ set "bufofs" "256" }}
|
|
| 160 |
- {{ template "decode_2_values_x86" . }}
|
|
| 161 |
- |
|
| 162 |
- {{ set "id" "2" }}
|
|
| 163 |
- {{ set "ofs" "16" }}
|
|
| 164 |
- {{ set "bufofs" "512" }}
|
|
| 165 |
- {{ template "decode_2_values_x86" . }}
|
|
| 166 |
- |
|
| 167 |
- {{ set "id" "3" }}
|
|
| 168 |
- {{ set "ofs" "24" }}
|
|
| 169 |
- {{ set "bufofs" "768" }}
|
|
| 170 |
- {{ template "decode_2_values_x86" . }}
|
|
| 171 |
- |
|
| 172 |
- ADDQ $4, off // off += 2 |
|
| 173 |
- |
|
| 174 |
- TESTB DH, DH // any br[i].ofs < 4? |
|
| 175 |
- JNZ end |
|
| 176 |
- |
|
| 177 |
- CMPQ off, $bufoff |
|
| 178 |
- JL main_loop |
|
| 179 |
-end: |
|
| 180 |
- MOVQ 0(SP), BP |
|
| 181 |
- |
|
| 182 |
- MOVB off, ret+56(FP) |
|
| 183 |
- RET |
|
| 184 |
-#undef off |
|
| 185 |
-#undef buffer |
|
| 186 |
-#undef table |
|
| 187 |
- |
|
| 188 |
-#undef br_bits_read |
|
| 189 |
-#undef br_value |
|
| 190 |
-#undef br_offset |
|
| 191 |
-#undef peek_bits |
|
| 192 |
-#undef exhausted |
|
| 193 |
- |
|
| 194 |
-#undef br0 |
|
| 195 |
-#undef br1 |
|
| 196 |
-#undef br2 |
|
| 197 |
-#undef br3 |
| ... | ... |
@@ -2,30 +2,40 @@ |
| 2 | 2 |
// +build amd64,!appengine,!noasm,gc |
| 3 | 3 |
|
| 4 | 4 |
// This file contains the specialisation of Decoder.Decompress4X |
| 5 |
-// that uses an asm implementation of its main loop. |
|
| 5 |
+// and Decoder.Decompress1X that use an asm implementation of thir main loops. |
|
| 6 | 6 |
package huff0 |
| 7 | 7 |
|
| 8 | 8 |
import ( |
| 9 | 9 |
"errors" |
| 10 | 10 |
"fmt" |
| 11 |
+ |
|
| 12 |
+ "github.com/klauspost/compress/internal/cpuinfo" |
|
| 11 | 13 |
) |
| 12 | 14 |
|
| 13 | 15 |
// decompress4x_main_loop_x86 is an x86 assembler implementation |
| 14 | 16 |
// of Decompress4X when tablelog > 8. |
| 15 |
-// go:noescape |
|
| 16 |
-func decompress4x_main_loop_x86(pbr0, pbr1, pbr2, pbr3 *bitReaderShifted, |
|
| 17 |
- peekBits uint8, buf *byte, tbl *dEntrySingle) uint8 |
|
| 17 |
+//go:noescape |
|
| 18 |
+func decompress4x_main_loop_amd64(ctx *decompress4xContext) |
|
| 18 | 19 |
|
| 19 | 20 |
// decompress4x_8b_loop_x86 is an x86 assembler implementation |
| 20 | 21 |
// of Decompress4X when tablelog <= 8 which decodes 4 entries |
| 21 | 22 |
// per loop. |
| 22 |
-// go:noescape |
|
| 23 |
-func decompress4x_8b_loop_x86(pbr0, pbr1, pbr2, pbr3 *bitReaderShifted, |
|
| 24 |
- peekBits uint8, buf *byte, tbl *dEntrySingle) uint8 |
|
| 23 |
+//go:noescape |
|
| 24 |
+func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext) |
|
| 25 | 25 |
|
| 26 | 26 |
// fallback8BitSize is the size where using Go version is faster. |
| 27 | 27 |
const fallback8BitSize = 800 |
| 28 | 28 |
|
| 29 |
+type decompress4xContext struct {
|
|
| 30 |
+ pbr *[4]bitReaderShifted |
|
| 31 |
+ peekBits uint8 |
|
| 32 |
+ out *byte |
|
| 33 |
+ dstEvery int |
|
| 34 |
+ tbl *dEntrySingle |
|
| 35 |
+ decoded int |
|
| 36 |
+ limit *byte |
|
| 37 |
+} |
|
| 38 |
+ |
|
| 29 | 39 |
// Decompress4X will decompress a 4X encoded stream. |
| 30 | 40 |
// The length of the supplied input must match the end of a block exactly. |
| 31 | 41 |
// The *capacity* of the dst slice must match the destination size of |
| ... | ... |
@@ -42,6 +52,7 @@ func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
|
| 42 | 42 |
if cap(dst) < fallback8BitSize && use8BitTables {
|
| 43 | 43 |
return d.decompress4X8bit(dst, src) |
| 44 | 44 |
} |
| 45 |
+ |
|
| 45 | 46 |
var br [4]bitReaderShifted |
| 46 | 47 |
// Decode "jump table" |
| 47 | 48 |
start := 6 |
| ... | ... |
@@ -71,70 +82,25 @@ func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
|
| 71 | 71 |
const tlMask = tlSize - 1 |
| 72 | 72 |
single := d.dt.single[:tlSize] |
| 73 | 73 |
|
| 74 |
- // Use temp table to avoid bound checks/append penalty. |
|
| 75 |
- buf := d.buffer() |
|
| 76 |
- var off uint8 |
|
| 77 | 74 |
var decoded int |
| 78 | 75 |
|
| 79 |
- const debug = false |
|
| 80 |
- |
|
| 81 |
- // see: bitReaderShifted.peekBitsFast() |
|
| 82 |
- peekBits := uint8((64 - d.actualTableLog) & 63) |
|
| 83 |
- |
|
| 84 |
- // Decode 2 values from each decoder/loop. |
|
| 85 |
- const bufoff = 256 |
|
| 86 |
- for {
|
|
| 87 |
- if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
|
|
| 88 |
- break |
|
| 76 |
+ if len(out) > 4*4 && !(br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4) {
|
|
| 77 |
+ ctx := decompress4xContext{
|
|
| 78 |
+ pbr: &br, |
|
| 79 |
+ peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast() |
|
| 80 |
+ out: &out[0], |
|
| 81 |
+ dstEvery: dstEvery, |
|
| 82 |
+ tbl: &single[0], |
|
| 83 |
+ limit: &out[dstEvery-4], // Always stop decoding when first buffer gets here to avoid writing OOB on last. |
|
| 89 | 84 |
} |
| 90 |
- |
|
| 91 | 85 |
if use8BitTables {
|
| 92 |
- off = decompress4x_8b_loop_x86(&br[0], &br[1], &br[2], &br[3], peekBits, &buf[0][0], &single[0]) |
|
| 86 |
+ decompress4x_8b_main_loop_amd64(&ctx) |
|
| 93 | 87 |
} else {
|
| 94 |
- off = decompress4x_main_loop_x86(&br[0], &br[1], &br[2], &br[3], peekBits, &buf[0][0], &single[0]) |
|
| 95 |
- } |
|
| 96 |
- if debug {
|
|
| 97 |
- fmt.Print("DEBUG: ")
|
|
| 98 |
- fmt.Printf("off=%d,", off)
|
|
| 99 |
- for i := 0; i < 4; i++ {
|
|
| 100 |
- fmt.Printf(" br[%d]={bitsRead=%d, value=%x, off=%d}",
|
|
| 101 |
- i, br[i].bitsRead, br[i].value, br[i].off) |
|
| 102 |
- } |
|
| 103 |
- fmt.Println("")
|
|
| 104 |
- } |
|
| 105 |
- |
|
| 106 |
- if off != 0 {
|
|
| 107 |
- break |
|
| 88 |
+ decompress4x_main_loop_amd64(&ctx) |
|
| 108 | 89 |
} |
| 109 | 90 |
|
| 110 |
- if bufoff > dstEvery {
|
|
| 111 |
- d.bufs.Put(buf) |
|
| 112 |
- return nil, errors.New("corruption detected: stream overrun 1")
|
|
| 113 |
- } |
|
| 114 |
- copy(out, buf[0][:]) |
|
| 115 |
- copy(out[dstEvery:], buf[1][:]) |
|
| 116 |
- copy(out[dstEvery*2:], buf[2][:]) |
|
| 117 |
- copy(out[dstEvery*3:], buf[3][:]) |
|
| 118 |
- out = out[bufoff:] |
|
| 119 |
- decoded += bufoff * 4 |
|
| 120 |
- // There must at least be 3 buffers left. |
|
| 121 |
- if len(out) < dstEvery*3 {
|
|
| 122 |
- d.bufs.Put(buf) |
|
| 123 |
- return nil, errors.New("corruption detected: stream overrun 2")
|
|
| 124 |
- } |
|
| 125 |
- } |
|
| 126 |
- if off > 0 {
|
|
| 127 |
- ioff := int(off) |
|
| 128 |
- if len(out) < dstEvery*3+ioff {
|
|
| 129 |
- d.bufs.Put(buf) |
|
| 130 |
- return nil, errors.New("corruption detected: stream overrun 3")
|
|
| 131 |
- } |
|
| 132 |
- copy(out, buf[0][:off]) |
|
| 133 |
- copy(out[dstEvery:], buf[1][:off]) |
|
| 134 |
- copy(out[dstEvery*2:], buf[2][:off]) |
|
| 135 |
- copy(out[dstEvery*3:], buf[3][:off]) |
|
| 136 |
- decoded += int(off) * 4 |
|
| 137 |
- out = out[off:] |
|
| 91 |
+ decoded = ctx.decoded |
|
| 92 |
+ out = out[decoded/4:] |
|
| 138 | 93 |
} |
| 139 | 94 |
|
| 140 | 95 |
// Decode remaining. |
| ... | ... |
@@ -150,7 +116,6 @@ func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
|
| 150 | 150 |
for bitsLeft > 0 {
|
| 151 | 151 |
br.fill() |
| 152 | 152 |
if offset >= endsAt {
|
| 153 |
- d.bufs.Put(buf) |
|
| 154 | 153 |
return nil, errors.New("corruption detected: stream overrun 4")
|
| 155 | 154 |
} |
| 156 | 155 |
|
| ... | ... |
@@ -164,7 +129,6 @@ func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
|
| 164 | 164 |
offset++ |
| 165 | 165 |
} |
| 166 | 166 |
if offset != endsAt {
|
| 167 |
- d.bufs.Put(buf) |
|
| 168 | 167 |
return nil, fmt.Errorf("corruption detected: short output block %d, end %d != %d", i, offset, endsAt)
|
| 169 | 168 |
} |
| 170 | 169 |
decoded += offset - dstEvery*i |
| ... | ... |
@@ -173,9 +137,86 @@ func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
|
| 173 | 173 |
return nil, err |
| 174 | 174 |
} |
| 175 | 175 |
} |
| 176 |
- d.bufs.Put(buf) |
|
| 177 | 176 |
if dstSize != decoded {
|
| 178 | 177 |
return nil, errors.New("corruption detected: short output block")
|
| 179 | 178 |
} |
| 180 | 179 |
return dst, nil |
| 181 | 180 |
} |
| 181 |
+ |
|
| 182 |
+// decompress4x_main_loop_x86 is an x86 assembler implementation |
|
| 183 |
+// of Decompress1X when tablelog > 8. |
|
| 184 |
+//go:noescape |
|
| 185 |
+func decompress1x_main_loop_amd64(ctx *decompress1xContext) |
|
| 186 |
+ |
|
| 187 |
+// decompress4x_main_loop_x86 is an x86 with BMI2 assembler implementation |
|
| 188 |
+// of Decompress1X when tablelog > 8. |
|
| 189 |
+//go:noescape |
|
| 190 |
+func decompress1x_main_loop_bmi2(ctx *decompress1xContext) |
|
| 191 |
+ |
|
| 192 |
+type decompress1xContext struct {
|
|
| 193 |
+ pbr *bitReaderShifted |
|
| 194 |
+ peekBits uint8 |
|
| 195 |
+ out *byte |
|
| 196 |
+ outCap int |
|
| 197 |
+ tbl *dEntrySingle |
|
| 198 |
+ decoded int |
|
| 199 |
+} |
|
| 200 |
+ |
|
| 201 |
+// Error reported by asm implementations |
|
| 202 |
+const error_max_decoded_size_exeeded = -1 |
|
| 203 |
+ |
|
| 204 |
+// Decompress1X will decompress a 1X encoded stream. |
|
| 205 |
+// The cap of the output buffer will be the maximum decompressed size. |
|
| 206 |
+// The length of the supplied input must match the end of a block exactly. |
|
| 207 |
+func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
|
|
| 208 |
+ if len(d.dt.single) == 0 {
|
|
| 209 |
+ return nil, errors.New("no table loaded")
|
|
| 210 |
+ } |
|
| 211 |
+ var br bitReaderShifted |
|
| 212 |
+ err := br.init(src) |
|
| 213 |
+ if err != nil {
|
|
| 214 |
+ return dst, err |
|
| 215 |
+ } |
|
| 216 |
+ maxDecodedSize := cap(dst) |
|
| 217 |
+ dst = dst[:maxDecodedSize] |
|
| 218 |
+ |
|
| 219 |
+ const tlSize = 1 << tableLogMax |
|
| 220 |
+ const tlMask = tlSize - 1 |
|
| 221 |
+ |
|
| 222 |
+ if maxDecodedSize >= 4 {
|
|
| 223 |
+ ctx := decompress1xContext{
|
|
| 224 |
+ pbr: &br, |
|
| 225 |
+ out: &dst[0], |
|
| 226 |
+ outCap: maxDecodedSize, |
|
| 227 |
+ peekBits: uint8((64 - d.actualTableLog) & 63), // see: bitReaderShifted.peekBitsFast() |
|
| 228 |
+ tbl: &d.dt.single[0], |
|
| 229 |
+ } |
|
| 230 |
+ |
|
| 231 |
+ if cpuinfo.HasBMI2() {
|
|
| 232 |
+ decompress1x_main_loop_bmi2(&ctx) |
|
| 233 |
+ } else {
|
|
| 234 |
+ decompress1x_main_loop_amd64(&ctx) |
|
| 235 |
+ } |
|
| 236 |
+ if ctx.decoded == error_max_decoded_size_exeeded {
|
|
| 237 |
+ return nil, ErrMaxDecodedSizeExceeded |
|
| 238 |
+ } |
|
| 239 |
+ |
|
| 240 |
+ dst = dst[:ctx.decoded] |
|
| 241 |
+ } |
|
| 242 |
+ |
|
| 243 |
+ // br < 8, so uint8 is fine |
|
| 244 |
+ bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead |
|
| 245 |
+ for bitsLeft > 0 {
|
|
| 246 |
+ br.fill() |
|
| 247 |
+ if len(dst) >= maxDecodedSize {
|
|
| 248 |
+ br.close() |
|
| 249 |
+ return nil, ErrMaxDecodedSizeExceeded |
|
| 250 |
+ } |
|
| 251 |
+ v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 252 |
+ nBits := uint8(v.entry) |
|
| 253 |
+ br.advance(nBits) |
|
| 254 |
+ bitsLeft -= nBits |
|
| 255 |
+ dst = append(dst, uint8(v.entry>>8)) |
|
| 256 |
+ } |
|
| 257 |
+ return dst, br.close() |
|
| 258 |
+} |
| ... | ... |
@@ -1,506 +1,847 @@ |
| 1 |
-// +build !appengine |
|
| 2 |
-// +build gc |
|
| 3 |
-// +build !noasm |
|
| 4 |
- |
|
| 5 |
-#include "textflag.h" |
|
| 6 |
-#include "funcdata.h" |
|
| 7 |
-#include "go_asm.h" |
|
| 8 |
- |
|
| 9 |
-#ifdef GOAMD64_v4 |
|
| 10 |
-#ifndef GOAMD64_v3 |
|
| 11 |
-#define GOAMD64_v3 |
|
| 12 |
-#endif |
|
| 13 |
-#endif |
|
| 14 |
- |
|
| 15 |
-#define bufoff 256 // see decompress.go, we're using [4][256]byte table |
|
| 16 |
- |
|
| 17 |
-// func decompress4x_main_loop_x86(pbr0, pbr1, pbr2, pbr3 *bitReaderShifted, |
|
| 18 |
-// peekBits uint8, buf *byte, tbl *dEntrySingle) (int, bool) |
|
| 19 |
-TEXT ·decompress4x_main_loop_x86(SB), NOSPLIT, $8 |
|
| 20 |
-#define off R8 |
|
| 21 |
-#define buffer DI |
|
| 22 |
-#define table SI |
|
| 23 |
- |
|
| 24 |
-#define br_bits_read R9 |
|
| 25 |
-#define br_value R10 |
|
| 26 |
-#define br_offset R11 |
|
| 27 |
-#define peek_bits R12 |
|
| 28 |
-#define exhausted DX |
|
| 29 |
- |
|
| 30 |
-#define br0 R13 |
|
| 31 |
-#define br1 R14 |
|
| 32 |
-#define br2 R15 |
|
| 33 |
-#define br3 BP |
|
| 34 |
- |
|
| 35 |
- MOVQ BP, 0(SP) |
|
| 36 |
- |
|
| 37 |
- XORQ exhausted, exhausted // exhausted = false |
|
| 38 |
- XORQ off, off // off = 0 |
|
| 39 |
- |
|
| 40 |
- MOVBQZX peekBits+32(FP), peek_bits |
|
| 41 |
- MOVQ buf+40(FP), buffer |
|
| 42 |
- MOVQ tbl+48(FP), table |
|
| 43 |
- |
|
| 44 |
- MOVQ pbr0+0(FP), br0 |
|
| 45 |
- MOVQ pbr1+8(FP), br1 |
|
| 46 |
- MOVQ pbr2+16(FP), br2 |
|
| 47 |
- MOVQ pbr3+24(FP), br3 |
|
| 1 |
+// Code generated by command: go run gen.go -out ../decompress_amd64.s -pkg=huff0. DO NOT EDIT. |
|
| 48 | 2 |
|
| 49 |
-main_loop: |
|
| 50 |
- |
|
| 51 |
- // const stream = 0 |
|
| 52 |
- // br0.fillFast() |
|
| 53 |
- MOVBQZX bitReaderShifted_bitsRead(br0), br_bits_read |
|
| 54 |
- MOVQ bitReaderShifted_value(br0), br_value |
|
| 55 |
- MOVQ bitReaderShifted_off(br0), br_offset |
|
| 3 |
+//go:build amd64 && !appengine && !noasm && gc |
|
| 4 |
+// +build amd64,!appengine,!noasm,gc |
|
| 56 | 5 |
|
| 57 |
- // We must have at least 2 * max tablelog left |
|
| 58 |
- CMPQ br_bits_read, $64-22 |
|
| 59 |
- JBE skip_fill0 |
|
| 6 |
+// func decompress4x_main_loop_amd64(ctx *decompress4xContext) |
|
| 7 |
+TEXT ·decompress4x_main_loop_amd64(SB), $0-8 |
|
| 8 |
+ XORQ DX, DX |
|
| 60 | 9 |
|
| 61 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 62 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 10 |
+ // Preload values |
|
| 11 |
+ MOVQ ctx+0(FP), AX |
|
| 12 |
+ MOVBQZX 8(AX), DI |
|
| 13 |
+ MOVQ 16(AX), SI |
|
| 14 |
+ MOVQ 48(AX), BX |
|
| 15 |
+ MOVQ 24(AX), R9 |
|
| 16 |
+ MOVQ 32(AX), R10 |
|
| 17 |
+ MOVQ (AX), R11 |
|
| 63 | 18 |
|
| 64 |
- // v := b.in[b.off-4 : b.off] |
|
| 65 |
- // v = v[:4] |
|
| 66 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 67 |
- MOVQ bitReaderShifted_in(br0), AX |
|
| 19 |
+ // Main loop |
|
| 20 |
+main_loop: |
|
| 21 |
+ MOVQ SI, R8 |
|
| 22 |
+ CMPQ R8, BX |
|
| 23 |
+ SETGE DL |
|
| 24 |
+ |
|
| 25 |
+ // br0.fillFast32() |
|
| 26 |
+ MOVQ 32(R11), R12 |
|
| 27 |
+ MOVBQZX 40(R11), R13 |
|
| 28 |
+ CMPQ R13, $0x20 |
|
| 29 |
+ JBE skip_fill0 |
|
| 30 |
+ MOVQ 24(R11), AX |
|
| 31 |
+ SUBQ $0x20, R13 |
|
| 32 |
+ SUBQ $0x04, AX |
|
| 33 |
+ MOVQ (R11), R14 |
|
| 68 | 34 |
|
| 69 | 35 |
// b.value |= uint64(low) << (b.bitsRead & 63) |
| 70 |
-#ifdef GOAMD64_v3 |
|
| 71 |
- SHLXQ br_bits_read, 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) << (b.bitsRead & 63) |
|
| 72 |
- |
|
| 73 |
-#else |
|
| 74 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 75 |
- MOVQ br_bits_read, CX |
|
| 76 |
- SHLQ CL, AX |
|
| 77 |
- |
|
| 78 |
-#endif |
|
| 79 |
- |
|
| 80 |
- ORQ AX, br_value |
|
| 36 |
+ MOVL (AX)(R14*1), R14 |
|
| 37 |
+ MOVQ R13, CX |
|
| 38 |
+ SHLQ CL, R14 |
|
| 39 |
+ MOVQ AX, 24(R11) |
|
| 40 |
+ ORQ R14, R12 |
|
| 81 | 41 |
|
| 82 | 42 |
// exhausted = exhausted || (br0.off < 4) |
| 83 |
- CMPQ br_offset, $4 |
|
| 84 |
- SETLT DL |
|
| 85 |
- ORB DL, DH |
|
| 43 |
+ CMPQ AX, $0x04 |
|
| 44 |
+ SETLT AL |
|
| 45 |
+ ORB AL, DL |
|
| 86 | 46 |
|
| 87 |
- // } |
|
| 88 | 47 |
skip_fill0: |
| 89 |
- |
|
| 90 | 48 |
// val0 := br0.peekTopBits(peekBits) |
| 91 |
-#ifdef GOAMD64_v3 |
|
| 92 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 93 |
- |
|
| 94 |
-#else |
|
| 95 |
- MOVQ br_value, AX |
|
| 96 |
- MOVQ peek_bits, CX |
|
| 97 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 98 |
- |
|
| 99 |
-#endif |
|
| 49 |
+ MOVQ R12, R14 |
|
| 50 |
+ MOVQ DI, CX |
|
| 51 |
+ SHRQ CL, R14 |
|
| 100 | 52 |
|
| 101 | 53 |
// v0 := table[val0&mask] |
| 102 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 103 |
- |
|
| 104 |
- // br0.advance(uint8(v0.entry)) |
|
| 105 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 106 |
- |
|
| 107 |
-#ifdef GOAMD64_v3 |
|
| 108 |
- MOVBQZX AL, CX |
|
| 109 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 110 |
- |
|
| 111 |
-#else |
|
| 112 |
- MOVBQZX AL, CX |
|
| 113 |
- SHLQ CL, br_value // value <<= n |
|
| 114 |
- |
|
| 115 |
-#endif |
|
| 116 |
- |
|
| 117 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 54 |
+ MOVW (R10)(R14*2), CX |
|
| 118 | 55 |
|
| 119 |
-#ifdef GOAMD64_v3 |
|
| 120 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 56 |
+ // br0.advance(uint8(v0.entry) |
|
| 57 |
+ MOVB CH, AL |
|
| 58 |
+ SHLQ CL, R12 |
|
| 59 |
+ ADDB CL, R13 |
|
| 121 | 60 |
|
| 122 |
-#else |
|
| 123 | 61 |
// val1 := br0.peekTopBits(peekBits) |
| 124 |
- MOVQ peek_bits, CX |
|
| 125 |
- MOVQ br_value, AX |
|
| 126 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 127 |
- |
|
| 128 |
-#endif |
|
| 62 |
+ MOVQ DI, CX |
|
| 63 |
+ MOVQ R12, R14 |
|
| 64 |
+ SHRQ CL, R14 |
|
| 129 | 65 |
|
| 130 | 66 |
// v1 := table[val1&mask] |
| 131 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 67 |
+ MOVW (R10)(R14*2), CX |
|
| 132 | 68 |
|
| 133 | 69 |
// br0.advance(uint8(v1.entry)) |
| 134 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 135 |
- |
|
| 136 |
-#ifdef GOAMD64_v3 |
|
| 137 |
- MOVBQZX AL, CX |
|
| 138 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 139 |
- |
|
| 140 |
-#else |
|
| 141 |
- MOVBQZX AL, CX |
|
| 142 |
- SHLQ CL, br_value // value <<= n |
|
| 143 |
- |
|
| 144 |
-#endif |
|
| 145 |
- |
|
| 146 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 70 |
+ MOVB CH, AH |
|
| 71 |
+ SHLQ CL, R12 |
|
| 72 |
+ ADDB CL, R13 |
|
| 147 | 73 |
|
| 148 | 74 |
// these two writes get coalesced |
| 149 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 150 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 151 |
- MOVW BX, 0(buffer)(off*1) |
|
| 152 |
- |
|
| 153 |
- // update the bitrader reader structure |
|
| 154 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br0) |
|
| 155 |
- MOVQ br_value, bitReaderShifted_value(br0) |
|
| 156 |
- MOVQ br_offset, bitReaderShifted_off(br0) |
|
| 157 |
- |
|
| 158 |
- // const stream = 1 |
|
| 159 |
- // br1.fillFast() |
|
| 160 |
- MOVBQZX bitReaderShifted_bitsRead(br1), br_bits_read |
|
| 161 |
- MOVQ bitReaderShifted_value(br1), br_value |
|
| 162 |
- MOVQ bitReaderShifted_off(br1), br_offset |
|
| 163 |
- |
|
| 164 |
- // We must have at least 2 * max tablelog left |
|
| 165 |
- CMPQ br_bits_read, $64-22 |
|
| 166 |
- JBE skip_fill1 |
|
| 167 |
- |
|
| 168 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 169 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 170 |
- |
|
| 171 |
- // v := b.in[b.off-4 : b.off] |
|
| 172 |
- // v = v[:4] |
|
| 173 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 174 |
- MOVQ bitReaderShifted_in(br1), AX |
|
| 75 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 76 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 77 |
+ MOVW AX, (R8) |
|
| 78 |
+ |
|
| 79 |
+ // update the bitreader structure |
|
| 80 |
+ MOVQ R12, 32(R11) |
|
| 81 |
+ MOVB R13, 40(R11) |
|
| 82 |
+ ADDQ R9, R8 |
|
| 83 |
+ |
|
| 84 |
+ // br1.fillFast32() |
|
| 85 |
+ MOVQ 80(R11), R12 |
|
| 86 |
+ MOVBQZX 88(R11), R13 |
|
| 87 |
+ CMPQ R13, $0x20 |
|
| 88 |
+ JBE skip_fill1 |
|
| 89 |
+ MOVQ 72(R11), AX |
|
| 90 |
+ SUBQ $0x20, R13 |
|
| 91 |
+ SUBQ $0x04, AX |
|
| 92 |
+ MOVQ 48(R11), R14 |
|
| 175 | 93 |
|
| 176 | 94 |
// b.value |= uint64(low) << (b.bitsRead & 63) |
| 177 |
-#ifdef GOAMD64_v3 |
|
| 178 |
- SHLXQ br_bits_read, 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) << (b.bitsRead & 63) |
|
| 179 |
- |
|
| 180 |
-#else |
|
| 181 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 182 |
- MOVQ br_bits_read, CX |
|
| 183 |
- SHLQ CL, AX |
|
| 184 |
- |
|
| 185 |
-#endif |
|
| 186 |
- |
|
| 187 |
- ORQ AX, br_value |
|
| 95 |
+ MOVL (AX)(R14*1), R14 |
|
| 96 |
+ MOVQ R13, CX |
|
| 97 |
+ SHLQ CL, R14 |
|
| 98 |
+ MOVQ AX, 72(R11) |
|
| 99 |
+ ORQ R14, R12 |
|
| 188 | 100 |
|
| 189 | 101 |
// exhausted = exhausted || (br1.off < 4) |
| 190 |
- CMPQ br_offset, $4 |
|
| 191 |
- SETLT DL |
|
| 192 |
- ORB DL, DH |
|
| 102 |
+ CMPQ AX, $0x04 |
|
| 103 |
+ SETLT AL |
|
| 104 |
+ ORB AL, DL |
|
| 193 | 105 |
|
| 194 |
- // } |
|
| 195 | 106 |
skip_fill1: |
| 196 |
- |
|
| 197 | 107 |
// val0 := br1.peekTopBits(peekBits) |
| 198 |
-#ifdef GOAMD64_v3 |
|
| 199 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 200 |
- |
|
| 201 |
-#else |
|
| 202 |
- MOVQ br_value, AX |
|
| 203 |
- MOVQ peek_bits, CX |
|
| 204 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 205 |
- |
|
| 206 |
-#endif |
|
| 108 |
+ MOVQ R12, R14 |
|
| 109 |
+ MOVQ DI, CX |
|
| 110 |
+ SHRQ CL, R14 |
|
| 207 | 111 |
|
| 208 | 112 |
// v0 := table[val0&mask] |
| 209 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 210 |
- |
|
| 211 |
- // br1.advance(uint8(v0.entry)) |
|
| 212 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 213 |
- |
|
| 214 |
-#ifdef GOAMD64_v3 |
|
| 215 |
- MOVBQZX AL, CX |
|
| 216 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 217 |
- |
|
| 218 |
-#else |
|
| 219 |
- MOVBQZX AL, CX |
|
| 220 |
- SHLQ CL, br_value // value <<= n |
|
| 221 |
- |
|
| 222 |
-#endif |
|
| 223 |
- |
|
| 224 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 113 |
+ MOVW (R10)(R14*2), CX |
|
| 225 | 114 |
|
| 226 |
-#ifdef GOAMD64_v3 |
|
| 227 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 115 |
+ // br1.advance(uint8(v0.entry) |
|
| 116 |
+ MOVB CH, AL |
|
| 117 |
+ SHLQ CL, R12 |
|
| 118 |
+ ADDB CL, R13 |
|
| 228 | 119 |
|
| 229 |
-#else |
|
| 230 | 120 |
// val1 := br1.peekTopBits(peekBits) |
| 231 |
- MOVQ peek_bits, CX |
|
| 232 |
- MOVQ br_value, AX |
|
| 233 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 234 |
- |
|
| 235 |
-#endif |
|
| 121 |
+ MOVQ DI, CX |
|
| 122 |
+ MOVQ R12, R14 |
|
| 123 |
+ SHRQ CL, R14 |
|
| 236 | 124 |
|
| 237 | 125 |
// v1 := table[val1&mask] |
| 238 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 126 |
+ MOVW (R10)(R14*2), CX |
|
| 239 | 127 |
|
| 240 | 128 |
// br1.advance(uint8(v1.entry)) |
| 241 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 242 |
- |
|
| 243 |
-#ifdef GOAMD64_v3 |
|
| 244 |
- MOVBQZX AL, CX |
|
| 245 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 246 |
- |
|
| 247 |
-#else |
|
| 248 |
- MOVBQZX AL, CX |
|
| 249 |
- SHLQ CL, br_value // value <<= n |
|
| 250 |
- |
|
| 251 |
-#endif |
|
| 252 |
- |
|
| 253 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 129 |
+ MOVB CH, AH |
|
| 130 |
+ SHLQ CL, R12 |
|
| 131 |
+ ADDB CL, R13 |
|
| 254 | 132 |
|
| 255 | 133 |
// these two writes get coalesced |
| 256 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 257 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 258 |
- MOVW BX, 256(buffer)(off*1) |
|
| 259 |
- |
|
| 260 |
- // update the bitrader reader structure |
|
| 261 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br1) |
|
| 262 |
- MOVQ br_value, bitReaderShifted_value(br1) |
|
| 263 |
- MOVQ br_offset, bitReaderShifted_off(br1) |
|
| 264 |
- |
|
| 265 |
- // const stream = 2 |
|
| 266 |
- // br2.fillFast() |
|
| 267 |
- MOVBQZX bitReaderShifted_bitsRead(br2), br_bits_read |
|
| 268 |
- MOVQ bitReaderShifted_value(br2), br_value |
|
| 269 |
- MOVQ bitReaderShifted_off(br2), br_offset |
|
| 270 |
- |
|
| 271 |
- // We must have at least 2 * max tablelog left |
|
| 272 |
- CMPQ br_bits_read, $64-22 |
|
| 273 |
- JBE skip_fill2 |
|
| 274 |
- |
|
| 275 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 276 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 277 |
- |
|
| 278 |
- // v := b.in[b.off-4 : b.off] |
|
| 279 |
- // v = v[:4] |
|
| 280 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 281 |
- MOVQ bitReaderShifted_in(br2), AX |
|
| 134 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 135 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 136 |
+ MOVW AX, (R8) |
|
| 137 |
+ |
|
| 138 |
+ // update the bitreader structure |
|
| 139 |
+ MOVQ R12, 80(R11) |
|
| 140 |
+ MOVB R13, 88(R11) |
|
| 141 |
+ ADDQ R9, R8 |
|
| 142 |
+ |
|
| 143 |
+ // br2.fillFast32() |
|
| 144 |
+ MOVQ 128(R11), R12 |
|
| 145 |
+ MOVBQZX 136(R11), R13 |
|
| 146 |
+ CMPQ R13, $0x20 |
|
| 147 |
+ JBE skip_fill2 |
|
| 148 |
+ MOVQ 120(R11), AX |
|
| 149 |
+ SUBQ $0x20, R13 |
|
| 150 |
+ SUBQ $0x04, AX |
|
| 151 |
+ MOVQ 96(R11), R14 |
|
| 282 | 152 |
|
| 283 | 153 |
// b.value |= uint64(low) << (b.bitsRead & 63) |
| 284 |
-#ifdef GOAMD64_v3 |
|
| 285 |
- SHLXQ br_bits_read, 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) << (b.bitsRead & 63) |
|
| 286 |
- |
|
| 287 |
-#else |
|
| 288 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 289 |
- MOVQ br_bits_read, CX |
|
| 290 |
- SHLQ CL, AX |
|
| 291 |
- |
|
| 292 |
-#endif |
|
| 293 |
- |
|
| 294 |
- ORQ AX, br_value |
|
| 154 |
+ MOVL (AX)(R14*1), R14 |
|
| 155 |
+ MOVQ R13, CX |
|
| 156 |
+ SHLQ CL, R14 |
|
| 157 |
+ MOVQ AX, 120(R11) |
|
| 158 |
+ ORQ R14, R12 |
|
| 295 | 159 |
|
| 296 | 160 |
// exhausted = exhausted || (br2.off < 4) |
| 297 |
- CMPQ br_offset, $4 |
|
| 298 |
- SETLT DL |
|
| 299 |
- ORB DL, DH |
|
| 161 |
+ CMPQ AX, $0x04 |
|
| 162 |
+ SETLT AL |
|
| 163 |
+ ORB AL, DL |
|
| 300 | 164 |
|
| 301 |
- // } |
|
| 302 | 165 |
skip_fill2: |
| 303 |
- |
|
| 304 | 166 |
// val0 := br2.peekTopBits(peekBits) |
| 305 |
-#ifdef GOAMD64_v3 |
|
| 306 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 307 |
- |
|
| 308 |
-#else |
|
| 309 |
- MOVQ br_value, AX |
|
| 310 |
- MOVQ peek_bits, CX |
|
| 311 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 312 |
- |
|
| 313 |
-#endif |
|
| 167 |
+ MOVQ R12, R14 |
|
| 168 |
+ MOVQ DI, CX |
|
| 169 |
+ SHRQ CL, R14 |
|
| 314 | 170 |
|
| 315 | 171 |
// v0 := table[val0&mask] |
| 316 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 317 |
- |
|
| 318 |
- // br2.advance(uint8(v0.entry)) |
|
| 319 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 320 |
- |
|
| 321 |
-#ifdef GOAMD64_v3 |
|
| 322 |
- MOVBQZX AL, CX |
|
| 323 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 324 |
- |
|
| 325 |
-#else |
|
| 326 |
- MOVBQZX AL, CX |
|
| 327 |
- SHLQ CL, br_value // value <<= n |
|
| 328 |
- |
|
| 329 |
-#endif |
|
| 172 |
+ MOVW (R10)(R14*2), CX |
|
| 330 | 173 |
|
| 331 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 174 |
+ // br2.advance(uint8(v0.entry) |
|
| 175 |
+ MOVB CH, AL |
|
| 176 |
+ SHLQ CL, R12 |
|
| 177 |
+ ADDB CL, R13 |
|
| 332 | 178 |
|
| 333 |
-#ifdef GOAMD64_v3 |
|
| 334 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 335 |
- |
|
| 336 |
-#else |
|
| 337 | 179 |
// val1 := br2.peekTopBits(peekBits) |
| 338 |
- MOVQ peek_bits, CX |
|
| 339 |
- MOVQ br_value, AX |
|
| 340 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 341 |
- |
|
| 342 |
-#endif |
|
| 180 |
+ MOVQ DI, CX |
|
| 181 |
+ MOVQ R12, R14 |
|
| 182 |
+ SHRQ CL, R14 |
|
| 343 | 183 |
|
| 344 | 184 |
// v1 := table[val1&mask] |
| 345 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 185 |
+ MOVW (R10)(R14*2), CX |
|
| 346 | 186 |
|
| 347 | 187 |
// br2.advance(uint8(v1.entry)) |
| 348 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 188 |
+ MOVB CH, AH |
|
| 189 |
+ SHLQ CL, R12 |
|
| 190 |
+ ADDB CL, R13 |
|
| 349 | 191 |
|
| 350 |
-#ifdef GOAMD64_v3 |
|
| 351 |
- MOVBQZX AL, CX |
|
| 352 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 353 |
- |
|
| 354 |
-#else |
|
| 355 |
- MOVBQZX AL, CX |
|
| 356 |
- SHLQ CL, br_value // value <<= n |
|
| 192 |
+ // these two writes get coalesced |
|
| 193 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 194 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 195 |
+ MOVW AX, (R8) |
|
| 196 |
+ |
|
| 197 |
+ // update the bitreader structure |
|
| 198 |
+ MOVQ R12, 128(R11) |
|
| 199 |
+ MOVB R13, 136(R11) |
|
| 200 |
+ ADDQ R9, R8 |
|
| 201 |
+ |
|
| 202 |
+ // br3.fillFast32() |
|
| 203 |
+ MOVQ 176(R11), R12 |
|
| 204 |
+ MOVBQZX 184(R11), R13 |
|
| 205 |
+ CMPQ R13, $0x20 |
|
| 206 |
+ JBE skip_fill3 |
|
| 207 |
+ MOVQ 168(R11), AX |
|
| 208 |
+ SUBQ $0x20, R13 |
|
| 209 |
+ SUBQ $0x04, AX |
|
| 210 |
+ MOVQ 144(R11), R14 |
|
| 357 | 211 |
|
| 358 |
-#endif |
|
| 212 |
+ // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 213 |
+ MOVL (AX)(R14*1), R14 |
|
| 214 |
+ MOVQ R13, CX |
|
| 215 |
+ SHLQ CL, R14 |
|
| 216 |
+ MOVQ AX, 168(R11) |
|
| 217 |
+ ORQ R14, R12 |
|
| 359 | 218 |
|
| 360 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 219 |
+ // exhausted = exhausted || (br3.off < 4) |
|
| 220 |
+ CMPQ AX, $0x04 |
|
| 221 |
+ SETLT AL |
|
| 222 |
+ ORB AL, DL |
|
| 361 | 223 |
|
| 362 |
- // these two writes get coalesced |
|
| 363 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 364 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 365 |
- MOVW BX, 512(buffer)(off*1) |
|
| 224 |
+skip_fill3: |
|
| 225 |
+ // val0 := br3.peekTopBits(peekBits) |
|
| 226 |
+ MOVQ R12, R14 |
|
| 227 |
+ MOVQ DI, CX |
|
| 228 |
+ SHRQ CL, R14 |
|
| 366 | 229 |
|
| 367 |
- // update the bitrader reader structure |
|
| 368 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br2) |
|
| 369 |
- MOVQ br_value, bitReaderShifted_value(br2) |
|
| 370 |
- MOVQ br_offset, bitReaderShifted_off(br2) |
|
| 230 |
+ // v0 := table[val0&mask] |
|
| 231 |
+ MOVW (R10)(R14*2), CX |
|
| 371 | 232 |
|
| 372 |
- // const stream = 3 |
|
| 373 |
- // br3.fillFast() |
|
| 374 |
- MOVBQZX bitReaderShifted_bitsRead(br3), br_bits_read |
|
| 375 |
- MOVQ bitReaderShifted_value(br3), br_value |
|
| 376 |
- MOVQ bitReaderShifted_off(br3), br_offset |
|
| 233 |
+ // br3.advance(uint8(v0.entry) |
|
| 234 |
+ MOVB CH, AL |
|
| 235 |
+ SHLQ CL, R12 |
|
| 236 |
+ ADDB CL, R13 |
|
| 377 | 237 |
|
| 378 |
- // We must have at least 2 * max tablelog left |
|
| 379 |
- CMPQ br_bits_read, $64-22 |
|
| 380 |
- JBE skip_fill3 |
|
| 238 |
+ // val1 := br3.peekTopBits(peekBits) |
|
| 239 |
+ MOVQ DI, CX |
|
| 240 |
+ MOVQ R12, R14 |
|
| 241 |
+ SHRQ CL, R14 |
|
| 381 | 242 |
|
| 382 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 383 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 243 |
+ // v1 := table[val1&mask] |
|
| 244 |
+ MOVW (R10)(R14*2), CX |
|
| 384 | 245 |
|
| 385 |
- // v := b.in[b.off-4 : b.off] |
|
| 386 |
- // v = v[:4] |
|
| 387 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 388 |
- MOVQ bitReaderShifted_in(br3), AX |
|
| 246 |
+ // br3.advance(uint8(v1.entry)) |
|
| 247 |
+ MOVB CH, AH |
|
| 248 |
+ SHLQ CL, R12 |
|
| 249 |
+ ADDB CL, R13 |
|
| 389 | 250 |
|
| 390 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 391 |
-#ifdef GOAMD64_v3 |
|
| 392 |
- SHLXQ br_bits_read, 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) << (b.bitsRead & 63) |
|
| 251 |
+ // these two writes get coalesced |
|
| 252 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 253 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 254 |
+ MOVW AX, (R8) |
|
| 255 |
+ |
|
| 256 |
+ // update the bitreader structure |
|
| 257 |
+ MOVQ R12, 176(R11) |
|
| 258 |
+ MOVB R13, 184(R11) |
|
| 259 |
+ ADDQ $0x02, SI |
|
| 260 |
+ TESTB DL, DL |
|
| 261 |
+ JZ main_loop |
|
| 262 |
+ MOVQ ctx+0(FP), AX |
|
| 263 |
+ SUBQ 16(AX), SI |
|
| 264 |
+ SHLQ $0x02, SI |
|
| 265 |
+ MOVQ SI, 40(AX) |
|
| 266 |
+ RET |
|
| 393 | 267 |
|
| 394 |
-#else |
|
| 395 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 396 |
- MOVQ br_bits_read, CX |
|
| 397 |
- SHLQ CL, AX |
|
| 268 |
+// func decompress4x_8b_main_loop_amd64(ctx *decompress4xContext) |
|
| 269 |
+TEXT ·decompress4x_8b_main_loop_amd64(SB), $0-8 |
|
| 270 |
+ XORQ DX, DX |
|
| 398 | 271 |
|
| 399 |
-#endif |
|
| 272 |
+ // Preload values |
|
| 273 |
+ MOVQ ctx+0(FP), CX |
|
| 274 |
+ MOVBQZX 8(CX), DI |
|
| 275 |
+ MOVQ 16(CX), BX |
|
| 276 |
+ MOVQ 48(CX), SI |
|
| 277 |
+ MOVQ 24(CX), R9 |
|
| 278 |
+ MOVQ 32(CX), R10 |
|
| 279 |
+ MOVQ (CX), R11 |
|
| 400 | 280 |
|
| 401 |
- ORQ AX, br_value |
|
| 281 |
+ // Main loop |
|
| 282 |
+main_loop: |
|
| 283 |
+ MOVQ BX, R8 |
|
| 284 |
+ CMPQ R8, SI |
|
| 285 |
+ SETGE DL |
|
| 286 |
+ |
|
| 287 |
+ // br0.fillFast32() |
|
| 288 |
+ MOVQ 32(R11), R12 |
|
| 289 |
+ MOVBQZX 40(R11), R13 |
|
| 290 |
+ CMPQ R13, $0x20 |
|
| 291 |
+ JBE skip_fill0 |
|
| 292 |
+ MOVQ 24(R11), R14 |
|
| 293 |
+ SUBQ $0x20, R13 |
|
| 294 |
+ SUBQ $0x04, R14 |
|
| 295 |
+ MOVQ (R11), R15 |
|
| 402 | 296 |
|
| 403 |
- // exhausted = exhausted || (br3.off < 4) |
|
| 404 |
- CMPQ br_offset, $4 |
|
| 405 |
- SETLT DL |
|
| 406 |
- ORB DL, DH |
|
| 297 |
+ // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 298 |
+ MOVL (R14)(R15*1), R15 |
|
| 299 |
+ MOVQ R13, CX |
|
| 300 |
+ SHLQ CL, R15 |
|
| 301 |
+ MOVQ R14, 24(R11) |
|
| 302 |
+ ORQ R15, R12 |
|
| 407 | 303 |
|
| 408 |
- // } |
|
| 409 |
-skip_fill3: |
|
| 304 |
+ // exhausted = exhausted || (br0.off < 4) |
|
| 305 |
+ CMPQ R14, $0x04 |
|
| 306 |
+ SETLT AL |
|
| 307 |
+ ORB AL, DL |
|
| 410 | 308 |
|
| 411 |
- // val0 := br3.peekTopBits(peekBits) |
|
| 412 |
-#ifdef GOAMD64_v3 |
|
| 413 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 309 |
+skip_fill0: |
|
| 310 |
+ // val0 := br0.peekTopBits(peekBits) |
|
| 311 |
+ MOVQ R12, R14 |
|
| 312 |
+ MOVQ DI, CX |
|
| 313 |
+ SHRQ CL, R14 |
|
| 414 | 314 |
|
| 415 |
-#else |
|
| 416 |
- MOVQ br_value, AX |
|
| 417 |
- MOVQ peek_bits, CX |
|
| 418 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 315 |
+ // v0 := table[val0&mask] |
|
| 316 |
+ MOVW (R10)(R14*2), CX |
|
| 419 | 317 |
|
| 420 |
-#endif |
|
| 318 |
+ // br0.advance(uint8(v0.entry) |
|
| 319 |
+ MOVB CH, AL |
|
| 320 |
+ SHLQ CL, R12 |
|
| 321 |
+ ADDB CL, R13 |
|
| 421 | 322 |
|
| 422 |
- // v0 := table[val0&mask] |
|
| 423 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 323 |
+ // val1 := br0.peekTopBits(peekBits) |
|
| 324 |
+ MOVQ R12, R14 |
|
| 325 |
+ MOVQ DI, CX |
|
| 326 |
+ SHRQ CL, R14 |
|
| 327 |
+ |
|
| 328 |
+ // v1 := table[val0&mask] |
|
| 329 |
+ MOVW (R10)(R14*2), CX |
|
| 330 |
+ |
|
| 331 |
+ // br0.advance(uint8(v1.entry) |
|
| 332 |
+ MOVB CH, AH |
|
| 333 |
+ SHLQ CL, R12 |
|
| 334 |
+ ADDB CL, R13 |
|
| 335 |
+ BSWAPL AX |
|
| 336 |
+ |
|
| 337 |
+ // val2 := br0.peekTopBits(peekBits) |
|
| 338 |
+ MOVQ R12, R14 |
|
| 339 |
+ MOVQ DI, CX |
|
| 340 |
+ SHRQ CL, R14 |
|
| 341 |
+ |
|
| 342 |
+ // v2 := table[val0&mask] |
|
| 343 |
+ MOVW (R10)(R14*2), CX |
|
| 344 |
+ |
|
| 345 |
+ // br0.advance(uint8(v2.entry) |
|
| 346 |
+ MOVB CH, AH |
|
| 347 |
+ SHLQ CL, R12 |
|
| 348 |
+ ADDB CL, R13 |
|
| 349 |
+ |
|
| 350 |
+ // val3 := br0.peekTopBits(peekBits) |
|
| 351 |
+ MOVQ R12, R14 |
|
| 352 |
+ MOVQ DI, CX |
|
| 353 |
+ SHRQ CL, R14 |
|
| 354 |
+ |
|
| 355 |
+ // v3 := table[val0&mask] |
|
| 356 |
+ MOVW (R10)(R14*2), CX |
|
| 357 |
+ |
|
| 358 |
+ // br0.advance(uint8(v3.entry) |
|
| 359 |
+ MOVB CH, AL |
|
| 360 |
+ SHLQ CL, R12 |
|
| 361 |
+ ADDB CL, R13 |
|
| 362 |
+ BSWAPL AX |
|
| 363 |
+ |
|
| 364 |
+ // these four writes get coalesced |
|
| 365 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 366 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 367 |
+ // out[id * dstEvery + 3] = uint8(v2.entry >> 8) |
|
| 368 |
+ // out[id * dstEvery + 4] = uint8(v3.entry >> 8) |
|
| 369 |
+ MOVL AX, (R8) |
|
| 370 |
+ |
|
| 371 |
+ // update the bitreader structure |
|
| 372 |
+ MOVQ R12, 32(R11) |
|
| 373 |
+ MOVB R13, 40(R11) |
|
| 374 |
+ ADDQ R9, R8 |
|
| 375 |
+ |
|
| 376 |
+ // br1.fillFast32() |
|
| 377 |
+ MOVQ 80(R11), R12 |
|
| 378 |
+ MOVBQZX 88(R11), R13 |
|
| 379 |
+ CMPQ R13, $0x20 |
|
| 380 |
+ JBE skip_fill1 |
|
| 381 |
+ MOVQ 72(R11), R14 |
|
| 382 |
+ SUBQ $0x20, R13 |
|
| 383 |
+ SUBQ $0x04, R14 |
|
| 384 |
+ MOVQ 48(R11), R15 |
|
| 424 | 385 |
|
| 425 |
- // br3.advance(uint8(v0.entry)) |
|
| 426 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 386 |
+ // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 387 |
+ MOVL (R14)(R15*1), R15 |
|
| 388 |
+ MOVQ R13, CX |
|
| 389 |
+ SHLQ CL, R15 |
|
| 390 |
+ MOVQ R14, 72(R11) |
|
| 391 |
+ ORQ R15, R12 |
|
| 427 | 392 |
|
| 428 |
-#ifdef GOAMD64_v3 |
|
| 429 |
- MOVBQZX AL, CX |
|
| 430 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 393 |
+ // exhausted = exhausted || (br1.off < 4) |
|
| 394 |
+ CMPQ R14, $0x04 |
|
| 395 |
+ SETLT AL |
|
| 396 |
+ ORB AL, DL |
|
| 431 | 397 |
|
| 432 |
-#else |
|
| 433 |
- MOVBQZX AL, CX |
|
| 434 |
- SHLQ CL, br_value // value <<= n |
|
| 398 |
+skip_fill1: |
|
| 399 |
+ // val0 := br1.peekTopBits(peekBits) |
|
| 400 |
+ MOVQ R12, R14 |
|
| 401 |
+ MOVQ DI, CX |
|
| 402 |
+ SHRQ CL, R14 |
|
| 435 | 403 |
|
| 436 |
-#endif |
|
| 404 |
+ // v0 := table[val0&mask] |
|
| 405 |
+ MOVW (R10)(R14*2), CX |
|
| 437 | 406 |
|
| 438 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 407 |
+ // br1.advance(uint8(v0.entry) |
|
| 408 |
+ MOVB CH, AL |
|
| 409 |
+ SHLQ CL, R12 |
|
| 410 |
+ ADDB CL, R13 |
|
| 439 | 411 |
|
| 440 |
-#ifdef GOAMD64_v3 |
|
| 441 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 412 |
+ // val1 := br1.peekTopBits(peekBits) |
|
| 413 |
+ MOVQ R12, R14 |
|
| 414 |
+ MOVQ DI, CX |
|
| 415 |
+ SHRQ CL, R14 |
|
| 416 |
+ |
|
| 417 |
+ // v1 := table[val0&mask] |
|
| 418 |
+ MOVW (R10)(R14*2), CX |
|
| 419 |
+ |
|
| 420 |
+ // br1.advance(uint8(v1.entry) |
|
| 421 |
+ MOVB CH, AH |
|
| 422 |
+ SHLQ CL, R12 |
|
| 423 |
+ ADDB CL, R13 |
|
| 424 |
+ BSWAPL AX |
|
| 425 |
+ |
|
| 426 |
+ // val2 := br1.peekTopBits(peekBits) |
|
| 427 |
+ MOVQ R12, R14 |
|
| 428 |
+ MOVQ DI, CX |
|
| 429 |
+ SHRQ CL, R14 |
|
| 430 |
+ |
|
| 431 |
+ // v2 := table[val0&mask] |
|
| 432 |
+ MOVW (R10)(R14*2), CX |
|
| 433 |
+ |
|
| 434 |
+ // br1.advance(uint8(v2.entry) |
|
| 435 |
+ MOVB CH, AH |
|
| 436 |
+ SHLQ CL, R12 |
|
| 437 |
+ ADDB CL, R13 |
|
| 438 |
+ |
|
| 439 |
+ // val3 := br1.peekTopBits(peekBits) |
|
| 440 |
+ MOVQ R12, R14 |
|
| 441 |
+ MOVQ DI, CX |
|
| 442 |
+ SHRQ CL, R14 |
|
| 443 |
+ |
|
| 444 |
+ // v3 := table[val0&mask] |
|
| 445 |
+ MOVW (R10)(R14*2), CX |
|
| 446 |
+ |
|
| 447 |
+ // br1.advance(uint8(v3.entry) |
|
| 448 |
+ MOVB CH, AL |
|
| 449 |
+ SHLQ CL, R12 |
|
| 450 |
+ ADDB CL, R13 |
|
| 451 |
+ BSWAPL AX |
|
| 452 |
+ |
|
| 453 |
+ // these four writes get coalesced |
|
| 454 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 455 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 456 |
+ // out[id * dstEvery + 3] = uint8(v2.entry >> 8) |
|
| 457 |
+ // out[id * dstEvery + 4] = uint8(v3.entry >> 8) |
|
| 458 |
+ MOVL AX, (R8) |
|
| 459 |
+ |
|
| 460 |
+ // update the bitreader structure |
|
| 461 |
+ MOVQ R12, 80(R11) |
|
| 462 |
+ MOVB R13, 88(R11) |
|
| 463 |
+ ADDQ R9, R8 |
|
| 464 |
+ |
|
| 465 |
+ // br2.fillFast32() |
|
| 466 |
+ MOVQ 128(R11), R12 |
|
| 467 |
+ MOVBQZX 136(R11), R13 |
|
| 468 |
+ CMPQ R13, $0x20 |
|
| 469 |
+ JBE skip_fill2 |
|
| 470 |
+ MOVQ 120(R11), R14 |
|
| 471 |
+ SUBQ $0x20, R13 |
|
| 472 |
+ SUBQ $0x04, R14 |
|
| 473 |
+ MOVQ 96(R11), R15 |
|
| 442 | 474 |
|
| 443 |
-#else |
|
| 444 |
- // val1 := br3.peekTopBits(peekBits) |
|
| 445 |
- MOVQ peek_bits, CX |
|
| 446 |
- MOVQ br_value, AX |
|
| 447 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 475 |
+ // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 476 |
+ MOVL (R14)(R15*1), R15 |
|
| 477 |
+ MOVQ R13, CX |
|
| 478 |
+ SHLQ CL, R15 |
|
| 479 |
+ MOVQ R14, 120(R11) |
|
| 480 |
+ ORQ R15, R12 |
|
| 448 | 481 |
|
| 449 |
-#endif |
|
| 482 |
+ // exhausted = exhausted || (br2.off < 4) |
|
| 483 |
+ CMPQ R14, $0x04 |
|
| 484 |
+ SETLT AL |
|
| 485 |
+ ORB AL, DL |
|
| 450 | 486 |
|
| 451 |
- // v1 := table[val1&mask] |
|
| 452 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 487 |
+skip_fill2: |
|
| 488 |
+ // val0 := br2.peekTopBits(peekBits) |
|
| 489 |
+ MOVQ R12, R14 |
|
| 490 |
+ MOVQ DI, CX |
|
| 491 |
+ SHRQ CL, R14 |
|
| 453 | 492 |
|
| 454 |
- // br3.advance(uint8(v1.entry)) |
|
| 455 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 493 |
+ // v0 := table[val0&mask] |
|
| 494 |
+ MOVW (R10)(R14*2), CX |
|
| 456 | 495 |
|
| 457 |
-#ifdef GOAMD64_v3 |
|
| 458 |
- MOVBQZX AL, CX |
|
| 459 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 496 |
+ // br2.advance(uint8(v0.entry) |
|
| 497 |
+ MOVB CH, AL |
|
| 498 |
+ SHLQ CL, R12 |
|
| 499 |
+ ADDB CL, R13 |
|
| 460 | 500 |
|
| 461 |
-#else |
|
| 462 |
- MOVBQZX AL, CX |
|
| 463 |
- SHLQ CL, br_value // value <<= n |
|
| 501 |
+ // val1 := br2.peekTopBits(peekBits) |
|
| 502 |
+ MOVQ R12, R14 |
|
| 503 |
+ MOVQ DI, CX |
|
| 504 |
+ SHRQ CL, R14 |
|
| 505 |
+ |
|
| 506 |
+ // v1 := table[val0&mask] |
|
| 507 |
+ MOVW (R10)(R14*2), CX |
|
| 508 |
+ |
|
| 509 |
+ // br2.advance(uint8(v1.entry) |
|
| 510 |
+ MOVB CH, AH |
|
| 511 |
+ SHLQ CL, R12 |
|
| 512 |
+ ADDB CL, R13 |
|
| 513 |
+ BSWAPL AX |
|
| 514 |
+ |
|
| 515 |
+ // val2 := br2.peekTopBits(peekBits) |
|
| 516 |
+ MOVQ R12, R14 |
|
| 517 |
+ MOVQ DI, CX |
|
| 518 |
+ SHRQ CL, R14 |
|
| 519 |
+ |
|
| 520 |
+ // v2 := table[val0&mask] |
|
| 521 |
+ MOVW (R10)(R14*2), CX |
|
| 522 |
+ |
|
| 523 |
+ // br2.advance(uint8(v2.entry) |
|
| 524 |
+ MOVB CH, AH |
|
| 525 |
+ SHLQ CL, R12 |
|
| 526 |
+ ADDB CL, R13 |
|
| 527 |
+ |
|
| 528 |
+ // val3 := br2.peekTopBits(peekBits) |
|
| 529 |
+ MOVQ R12, R14 |
|
| 530 |
+ MOVQ DI, CX |
|
| 531 |
+ SHRQ CL, R14 |
|
| 532 |
+ |
|
| 533 |
+ // v3 := table[val0&mask] |
|
| 534 |
+ MOVW (R10)(R14*2), CX |
|
| 535 |
+ |
|
| 536 |
+ // br2.advance(uint8(v3.entry) |
|
| 537 |
+ MOVB CH, AL |
|
| 538 |
+ SHLQ CL, R12 |
|
| 539 |
+ ADDB CL, R13 |
|
| 540 |
+ BSWAPL AX |
|
| 541 |
+ |
|
| 542 |
+ // these four writes get coalesced |
|
| 543 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 544 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 545 |
+ // out[id * dstEvery + 3] = uint8(v2.entry >> 8) |
|
| 546 |
+ // out[id * dstEvery + 4] = uint8(v3.entry >> 8) |
|
| 547 |
+ MOVL AX, (R8) |
|
| 548 |
+ |
|
| 549 |
+ // update the bitreader structure |
|
| 550 |
+ MOVQ R12, 128(R11) |
|
| 551 |
+ MOVB R13, 136(R11) |
|
| 552 |
+ ADDQ R9, R8 |
|
| 553 |
+ |
|
| 554 |
+ // br3.fillFast32() |
|
| 555 |
+ MOVQ 176(R11), R12 |
|
| 556 |
+ MOVBQZX 184(R11), R13 |
|
| 557 |
+ CMPQ R13, $0x20 |
|
| 558 |
+ JBE skip_fill3 |
|
| 559 |
+ MOVQ 168(R11), R14 |
|
| 560 |
+ SUBQ $0x20, R13 |
|
| 561 |
+ SUBQ $0x04, R14 |
|
| 562 |
+ MOVQ 144(R11), R15 |
|
| 464 | 563 |
|
| 465 |
-#endif |
|
| 564 |
+ // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 565 |
+ MOVL (R14)(R15*1), R15 |
|
| 566 |
+ MOVQ R13, CX |
|
| 567 |
+ SHLQ CL, R15 |
|
| 568 |
+ MOVQ R14, 168(R11) |
|
| 569 |
+ ORQ R15, R12 |
|
| 466 | 570 |
|
| 467 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 571 |
+ // exhausted = exhausted || (br3.off < 4) |
|
| 572 |
+ CMPQ R14, $0x04 |
|
| 573 |
+ SETLT AL |
|
| 574 |
+ ORB AL, DL |
|
| 468 | 575 |
|
| 469 |
- // these two writes get coalesced |
|
| 470 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 471 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 472 |
- MOVW BX, 768(buffer)(off*1) |
|
| 576 |
+skip_fill3: |
|
| 577 |
+ // val0 := br3.peekTopBits(peekBits) |
|
| 578 |
+ MOVQ R12, R14 |
|
| 579 |
+ MOVQ DI, CX |
|
| 580 |
+ SHRQ CL, R14 |
|
| 473 | 581 |
|
| 474 |
- // update the bitrader reader structure |
|
| 475 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br3) |
|
| 476 |
- MOVQ br_value, bitReaderShifted_value(br3) |
|
| 477 |
- MOVQ br_offset, bitReaderShifted_off(br3) |
|
| 582 |
+ // v0 := table[val0&mask] |
|
| 583 |
+ MOVW (R10)(R14*2), CX |
|
| 478 | 584 |
|
| 479 |
- ADDQ $2, off // off += 2 |
|
| 585 |
+ // br3.advance(uint8(v0.entry) |
|
| 586 |
+ MOVB CH, AL |
|
| 587 |
+ SHLQ CL, R12 |
|
| 588 |
+ ADDB CL, R13 |
|
| 480 | 589 |
|
| 481 |
- TESTB DH, DH // any br[i].ofs < 4? |
|
| 482 |
- JNZ end |
|
| 590 |
+ // val1 := br3.peekTopBits(peekBits) |
|
| 591 |
+ MOVQ R12, R14 |
|
| 592 |
+ MOVQ DI, CX |
|
| 593 |
+ SHRQ CL, R14 |
|
| 594 |
+ |
|
| 595 |
+ // v1 := table[val0&mask] |
|
| 596 |
+ MOVW (R10)(R14*2), CX |
|
| 597 |
+ |
|
| 598 |
+ // br3.advance(uint8(v1.entry) |
|
| 599 |
+ MOVB CH, AH |
|
| 600 |
+ SHLQ CL, R12 |
|
| 601 |
+ ADDB CL, R13 |
|
| 602 |
+ BSWAPL AX |
|
| 603 |
+ |
|
| 604 |
+ // val2 := br3.peekTopBits(peekBits) |
|
| 605 |
+ MOVQ R12, R14 |
|
| 606 |
+ MOVQ DI, CX |
|
| 607 |
+ SHRQ CL, R14 |
|
| 608 |
+ |
|
| 609 |
+ // v2 := table[val0&mask] |
|
| 610 |
+ MOVW (R10)(R14*2), CX |
|
| 611 |
+ |
|
| 612 |
+ // br3.advance(uint8(v2.entry) |
|
| 613 |
+ MOVB CH, AH |
|
| 614 |
+ SHLQ CL, R12 |
|
| 615 |
+ ADDB CL, R13 |
|
| 616 |
+ |
|
| 617 |
+ // val3 := br3.peekTopBits(peekBits) |
|
| 618 |
+ MOVQ R12, R14 |
|
| 619 |
+ MOVQ DI, CX |
|
| 620 |
+ SHRQ CL, R14 |
|
| 621 |
+ |
|
| 622 |
+ // v3 := table[val0&mask] |
|
| 623 |
+ MOVW (R10)(R14*2), CX |
|
| 624 |
+ |
|
| 625 |
+ // br3.advance(uint8(v3.entry) |
|
| 626 |
+ MOVB CH, AL |
|
| 627 |
+ SHLQ CL, R12 |
|
| 628 |
+ ADDB CL, R13 |
|
| 629 |
+ BSWAPL AX |
|
| 630 |
+ |
|
| 631 |
+ // these four writes get coalesced |
|
| 632 |
+ // out[id * dstEvery + 0] = uint8(v0.entry >> 8) |
|
| 633 |
+ // out[id * dstEvery + 1] = uint8(v1.entry >> 8) |
|
| 634 |
+ // out[id * dstEvery + 3] = uint8(v2.entry >> 8) |
|
| 635 |
+ // out[id * dstEvery + 4] = uint8(v3.entry >> 8) |
|
| 636 |
+ MOVL AX, (R8) |
|
| 637 |
+ |
|
| 638 |
+ // update the bitreader structure |
|
| 639 |
+ MOVQ R12, 176(R11) |
|
| 640 |
+ MOVB R13, 184(R11) |
|
| 641 |
+ ADDQ $0x04, BX |
|
| 642 |
+ TESTB DL, DL |
|
| 643 |
+ JZ main_loop |
|
| 644 |
+ MOVQ ctx+0(FP), AX |
|
| 645 |
+ SUBQ 16(AX), BX |
|
| 646 |
+ SHLQ $0x02, BX |
|
| 647 |
+ MOVQ BX, 40(AX) |
|
| 648 |
+ RET |
|
| 483 | 649 |
|
| 484 |
- CMPQ off, $bufoff |
|
| 485 |
- JL main_loop |
|
| 650 |
+// func decompress1x_main_loop_amd64(ctx *decompress1xContext) |
|
| 651 |
+TEXT ·decompress1x_main_loop_amd64(SB), $0-8 |
|
| 652 |
+ MOVQ ctx+0(FP), CX |
|
| 653 |
+ MOVQ 16(CX), DX |
|
| 654 |
+ MOVQ 24(CX), BX |
|
| 655 |
+ CMPQ BX, $0x04 |
|
| 656 |
+ JB error_max_decoded_size_exeeded |
|
| 657 |
+ LEAQ (DX)(BX*1), BX |
|
| 658 |
+ MOVQ (CX), SI |
|
| 659 |
+ MOVQ (SI), R8 |
|
| 660 |
+ MOVQ 24(SI), R9 |
|
| 661 |
+ MOVQ 32(SI), R10 |
|
| 662 |
+ MOVBQZX 40(SI), R11 |
|
| 663 |
+ MOVQ 32(CX), SI |
|
| 664 |
+ MOVBQZX 8(CX), DI |
|
| 665 |
+ JMP loop_condition |
|
| 486 | 666 |
|
| 487 |
-end: |
|
| 488 |
- MOVQ 0(SP), BP |
|
| 667 |
+main_loop: |
|
| 668 |
+ // Check if we have room for 4 bytes in the output buffer |
|
| 669 |
+ LEAQ 4(DX), CX |
|
| 670 |
+ CMPQ CX, BX |
|
| 671 |
+ JGE error_max_decoded_size_exeeded |
|
| 672 |
+ |
|
| 673 |
+ // Decode 4 values |
|
| 674 |
+ CMPQ R11, $0x20 |
|
| 675 |
+ JL bitReader_fillFast_1_end |
|
| 676 |
+ SUBQ $0x20, R11 |
|
| 677 |
+ SUBQ $0x04, R9 |
|
| 678 |
+ MOVL (R8)(R9*1), R12 |
|
| 679 |
+ MOVQ R11, CX |
|
| 680 |
+ SHLQ CL, R12 |
|
| 681 |
+ ORQ R12, R10 |
|
| 682 |
+ |
|
| 683 |
+bitReader_fillFast_1_end: |
|
| 684 |
+ MOVQ DI, CX |
|
| 685 |
+ MOVQ R10, R12 |
|
| 686 |
+ SHRQ CL, R12 |
|
| 687 |
+ MOVW (SI)(R12*2), CX |
|
| 688 |
+ MOVB CH, AL |
|
| 689 |
+ MOVBQZX CL, CX |
|
| 690 |
+ ADDQ CX, R11 |
|
| 691 |
+ SHLQ CL, R10 |
|
| 692 |
+ MOVQ DI, CX |
|
| 693 |
+ MOVQ R10, R12 |
|
| 694 |
+ SHRQ CL, R12 |
|
| 695 |
+ MOVW (SI)(R12*2), CX |
|
| 696 |
+ MOVB CH, AH |
|
| 697 |
+ MOVBQZX CL, CX |
|
| 698 |
+ ADDQ CX, R11 |
|
| 699 |
+ SHLQ CL, R10 |
|
| 700 |
+ BSWAPL AX |
|
| 701 |
+ CMPQ R11, $0x20 |
|
| 702 |
+ JL bitReader_fillFast_2_end |
|
| 703 |
+ SUBQ $0x20, R11 |
|
| 704 |
+ SUBQ $0x04, R9 |
|
| 705 |
+ MOVL (R8)(R9*1), R12 |
|
| 706 |
+ MOVQ R11, CX |
|
| 707 |
+ SHLQ CL, R12 |
|
| 708 |
+ ORQ R12, R10 |
|
| 709 |
+ |
|
| 710 |
+bitReader_fillFast_2_end: |
|
| 711 |
+ MOVQ DI, CX |
|
| 712 |
+ MOVQ R10, R12 |
|
| 713 |
+ SHRQ CL, R12 |
|
| 714 |
+ MOVW (SI)(R12*2), CX |
|
| 715 |
+ MOVB CH, AH |
|
| 716 |
+ MOVBQZX CL, CX |
|
| 717 |
+ ADDQ CX, R11 |
|
| 718 |
+ SHLQ CL, R10 |
|
| 719 |
+ MOVQ DI, CX |
|
| 720 |
+ MOVQ R10, R12 |
|
| 721 |
+ SHRQ CL, R12 |
|
| 722 |
+ MOVW (SI)(R12*2), CX |
|
| 723 |
+ MOVB CH, AL |
|
| 724 |
+ MOVBQZX CL, CX |
|
| 725 |
+ ADDQ CX, R11 |
|
| 726 |
+ SHLQ CL, R10 |
|
| 727 |
+ BSWAPL AX |
|
| 728 |
+ |
|
| 729 |
+ // Store the decoded values |
|
| 730 |
+ MOVL AX, (DX) |
|
| 731 |
+ ADDQ $0x04, DX |
|
| 732 |
+ |
|
| 733 |
+loop_condition: |
|
| 734 |
+ CMPQ R9, $0x08 |
|
| 735 |
+ JGE main_loop |
|
| 736 |
+ |
|
| 737 |
+ // Update ctx structure |
|
| 738 |
+ MOVQ ctx+0(FP), AX |
|
| 739 |
+ SUBQ 16(AX), DX |
|
| 740 |
+ MOVQ DX, 40(AX) |
|
| 741 |
+ MOVQ (AX), AX |
|
| 742 |
+ MOVQ R9, 24(AX) |
|
| 743 |
+ MOVQ R10, 32(AX) |
|
| 744 |
+ MOVB R11, 40(AX) |
|
| 745 |
+ RET |
|
| 489 | 746 |
|
| 490 |
- MOVB off, ret+56(FP) |
|
| 747 |
+ // Report error |
|
| 748 |
+error_max_decoded_size_exeeded: |
|
| 749 |
+ MOVQ ctx+0(FP), AX |
|
| 750 |
+ MOVQ $-1, CX |
|
| 751 |
+ MOVQ CX, 40(AX) |
|
| 491 | 752 |
RET |
| 492 | 753 |
|
| 493 |
-#undef off |
|
| 494 |
-#undef buffer |
|
| 495 |
-#undef table |
|
| 754 |
+// func decompress1x_main_loop_bmi2(ctx *decompress1xContext) |
|
| 755 |
+// Requires: BMI2 |
|
| 756 |
+TEXT ·decompress1x_main_loop_bmi2(SB), $0-8 |
|
| 757 |
+ MOVQ ctx+0(FP), CX |
|
| 758 |
+ MOVQ 16(CX), DX |
|
| 759 |
+ MOVQ 24(CX), BX |
|
| 760 |
+ CMPQ BX, $0x04 |
|
| 761 |
+ JB error_max_decoded_size_exeeded |
|
| 762 |
+ LEAQ (DX)(BX*1), BX |
|
| 763 |
+ MOVQ (CX), SI |
|
| 764 |
+ MOVQ (SI), R8 |
|
| 765 |
+ MOVQ 24(SI), R9 |
|
| 766 |
+ MOVQ 32(SI), R10 |
|
| 767 |
+ MOVBQZX 40(SI), R11 |
|
| 768 |
+ MOVQ 32(CX), SI |
|
| 769 |
+ MOVBQZX 8(CX), DI |
|
| 770 |
+ JMP loop_condition |
|
| 496 | 771 |
|
| 497 |
-#undef br_bits_read |
|
| 498 |
-#undef br_value |
|
| 499 |
-#undef br_offset |
|
| 500 |
-#undef peek_bits |
|
| 501 |
-#undef exhausted |
|
| 772 |
+main_loop: |
|
| 773 |
+ // Check if we have room for 4 bytes in the output buffer |
|
| 774 |
+ LEAQ 4(DX), CX |
|
| 775 |
+ CMPQ CX, BX |
|
| 776 |
+ JGE error_max_decoded_size_exeeded |
|
| 777 |
+ |
|
| 778 |
+ // Decode 4 values |
|
| 779 |
+ CMPQ R11, $0x20 |
|
| 780 |
+ JL bitReader_fillFast_1_end |
|
| 781 |
+ SUBQ $0x20, R11 |
|
| 782 |
+ SUBQ $0x04, R9 |
|
| 783 |
+ MOVL (R8)(R9*1), CX |
|
| 784 |
+ SHLXQ R11, CX, CX |
|
| 785 |
+ ORQ CX, R10 |
|
| 786 |
+ |
|
| 787 |
+bitReader_fillFast_1_end: |
|
| 788 |
+ SHRXQ DI, R10, CX |
|
| 789 |
+ MOVW (SI)(CX*2), CX |
|
| 790 |
+ MOVB CH, AL |
|
| 791 |
+ MOVBQZX CL, CX |
|
| 792 |
+ ADDQ CX, R11 |
|
| 793 |
+ SHLXQ CX, R10, R10 |
|
| 794 |
+ SHRXQ DI, R10, CX |
|
| 795 |
+ MOVW (SI)(CX*2), CX |
|
| 796 |
+ MOVB CH, AH |
|
| 797 |
+ MOVBQZX CL, CX |
|
| 798 |
+ ADDQ CX, R11 |
|
| 799 |
+ SHLXQ CX, R10, R10 |
|
| 800 |
+ BSWAPL AX |
|
| 801 |
+ CMPQ R11, $0x20 |
|
| 802 |
+ JL bitReader_fillFast_2_end |
|
| 803 |
+ SUBQ $0x20, R11 |
|
| 804 |
+ SUBQ $0x04, R9 |
|
| 805 |
+ MOVL (R8)(R9*1), CX |
|
| 806 |
+ SHLXQ R11, CX, CX |
|
| 807 |
+ ORQ CX, R10 |
|
| 808 |
+ |
|
| 809 |
+bitReader_fillFast_2_end: |
|
| 810 |
+ SHRXQ DI, R10, CX |
|
| 811 |
+ MOVW (SI)(CX*2), CX |
|
| 812 |
+ MOVB CH, AH |
|
| 813 |
+ MOVBQZX CL, CX |
|
| 814 |
+ ADDQ CX, R11 |
|
| 815 |
+ SHLXQ CX, R10, R10 |
|
| 816 |
+ SHRXQ DI, R10, CX |
|
| 817 |
+ MOVW (SI)(CX*2), CX |
|
| 818 |
+ MOVB CH, AL |
|
| 819 |
+ MOVBQZX CL, CX |
|
| 820 |
+ ADDQ CX, R11 |
|
| 821 |
+ SHLXQ CX, R10, R10 |
|
| 822 |
+ BSWAPL AX |
|
| 823 |
+ |
|
| 824 |
+ // Store the decoded values |
|
| 825 |
+ MOVL AX, (DX) |
|
| 826 |
+ ADDQ $0x04, DX |
|
| 827 |
+ |
|
| 828 |
+loop_condition: |
|
| 829 |
+ CMPQ R9, $0x08 |
|
| 830 |
+ JGE main_loop |
|
| 831 |
+ |
|
| 832 |
+ // Update ctx structure |
|
| 833 |
+ MOVQ ctx+0(FP), AX |
|
| 834 |
+ SUBQ 16(AX), DX |
|
| 835 |
+ MOVQ DX, 40(AX) |
|
| 836 |
+ MOVQ (AX), AX |
|
| 837 |
+ MOVQ R9, 24(AX) |
|
| 838 |
+ MOVQ R10, 32(AX) |
|
| 839 |
+ MOVB R11, 40(AX) |
|
| 840 |
+ RET |
|
| 502 | 841 |
|
| 503 |
-#undef br0 |
|
| 504 |
-#undef br1 |
|
| 505 |
-#undef br2 |
|
| 506 |
-#undef br3 |
|
| 842 |
+ // Report error |
|
| 843 |
+error_max_decoded_size_exeeded: |
|
| 844 |
+ MOVQ ctx+0(FP), AX |
|
| 845 |
+ MOVQ $-1, CX |
|
| 846 |
+ MOVQ CX, 40(AX) |
|
| 847 |
+ RET |
| 507 | 848 |
deleted file mode 100644 |
| ... | ... |
@@ -1,195 +0,0 @@ |
| 1 |
-// +build !appengine |
|
| 2 |
-// +build gc |
|
| 3 |
-// +build !noasm |
|
| 4 |
- |
|
| 5 |
-#include "textflag.h" |
|
| 6 |
-#include "funcdata.h" |
|
| 7 |
-#include "go_asm.h" |
|
| 8 |
- |
|
| 9 |
-#ifdef GOAMD64_v4 |
|
| 10 |
-#ifndef GOAMD64_v3 |
|
| 11 |
-#define GOAMD64_v3 |
|
| 12 |
-#endif |
|
| 13 |
-#endif |
|
| 14 |
- |
|
| 15 |
-#define bufoff 256 // see decompress.go, we're using [4][256]byte table |
|
| 16 |
- |
|
| 17 |
-//func decompress4x_main_loop_x86(pbr0, pbr1, pbr2, pbr3 *bitReaderShifted, |
|
| 18 |
-// peekBits uint8, buf *byte, tbl *dEntrySingle) (int, bool) |
|
| 19 |
-TEXT ·decompress4x_main_loop_x86(SB), NOSPLIT, $8 |
|
| 20 |
-#define off R8 |
|
| 21 |
-#define buffer DI |
|
| 22 |
-#define table SI |
|
| 23 |
- |
|
| 24 |
-#define br_bits_read R9 |
|
| 25 |
-#define br_value R10 |
|
| 26 |
-#define br_offset R11 |
|
| 27 |
-#define peek_bits R12 |
|
| 28 |
-#define exhausted DX |
|
| 29 |
- |
|
| 30 |
-#define br0 R13 |
|
| 31 |
-#define br1 R14 |
|
| 32 |
-#define br2 R15 |
|
| 33 |
-#define br3 BP |
|
| 34 |
- |
|
| 35 |
- MOVQ BP, 0(SP) |
|
| 36 |
- |
|
| 37 |
- XORQ exhausted, exhausted // exhausted = false |
|
| 38 |
- XORQ off, off // off = 0 |
|
| 39 |
- |
|
| 40 |
- MOVBQZX peekBits+32(FP), peek_bits |
|
| 41 |
- MOVQ buf+40(FP), buffer |
|
| 42 |
- MOVQ tbl+48(FP), table |
|
| 43 |
- |
|
| 44 |
- MOVQ pbr0+0(FP), br0 |
|
| 45 |
- MOVQ pbr1+8(FP), br1 |
|
| 46 |
- MOVQ pbr2+16(FP), br2 |
|
| 47 |
- MOVQ pbr3+24(FP), br3 |
|
| 48 |
- |
|
| 49 |
-main_loop: |
|
| 50 |
-{{ define "decode_2_values_x86" }}
|
|
| 51 |
- // const stream = {{ var "id" }}
|
|
| 52 |
- // br{{ var "id"}}.fillFast()
|
|
| 53 |
- MOVBQZX bitReaderShifted_bitsRead(br{{ var "id" }}), br_bits_read
|
|
| 54 |
- MOVQ bitReaderShifted_value(br{{ var "id" }}), br_value
|
|
| 55 |
- MOVQ bitReaderShifted_off(br{{ var "id" }}), br_offset
|
|
| 56 |
- |
|
| 57 |
- // We must have at least 2 * max tablelog left |
|
| 58 |
- CMPQ br_bits_read, $64-22 |
|
| 59 |
- JBE skip_fill{{ var "id" }}
|
|
| 60 |
- |
|
| 61 |
- SUBQ $32, br_bits_read // b.bitsRead -= 32 |
|
| 62 |
- SUBQ $4, br_offset // b.off -= 4 |
|
| 63 |
- |
|
| 64 |
- // v := b.in[b.off-4 : b.off] |
|
| 65 |
- // v = v[:4] |
|
| 66 |
- // low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 67 |
- MOVQ bitReaderShifted_in(br{{ var "id" }}), AX
|
|
| 68 |
- |
|
| 69 |
- // b.value |= uint64(low) << (b.bitsRead & 63) |
|
| 70 |
-#ifdef GOAMD64_v3 |
|
| 71 |
- SHLXQ br_bits_read, 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) << (b.bitsRead & 63) |
|
| 72 |
-#else |
|
| 73 |
- MOVL 0(br_offset)(AX*1), AX // AX = uint32(b.in[b.off:b.off+4]) |
|
| 74 |
- MOVQ br_bits_read, CX |
|
| 75 |
- SHLQ CL, AX |
|
| 76 |
-#endif |
|
| 77 |
- |
|
| 78 |
- ORQ AX, br_value |
|
| 79 |
- |
|
| 80 |
- // exhausted = exhausted || (br{{ var "id"}}.off < 4)
|
|
| 81 |
- CMPQ br_offset, $4 |
|
| 82 |
- SETLT DL |
|
| 83 |
- ORB DL, DH |
|
| 84 |
- // } |
|
| 85 |
-skip_fill{{ var "id" }}:
|
|
| 86 |
- |
|
| 87 |
- // val0 := br{{ var "id"}}.peekTopBits(peekBits)
|
|
| 88 |
-#ifdef GOAMD64_v3 |
|
| 89 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 90 |
-#else |
|
| 91 |
- MOVQ br_value, AX |
|
| 92 |
- MOVQ peek_bits, CX |
|
| 93 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 94 |
-#endif |
|
| 95 |
- |
|
| 96 |
- // v0 := table[val0&mask] |
|
| 97 |
- MOVW 0(table)(AX*2), AX // AX - v0 |
|
| 98 |
- |
|
| 99 |
- // br{{ var "id"}}.advance(uint8(v0.entry))
|
|
| 100 |
- MOVB AH, BL // BL = uint8(v0.entry >> 8) |
|
| 101 |
- |
|
| 102 |
-#ifdef GOAMD64_v3 |
|
| 103 |
- MOVBQZX AL, CX |
|
| 104 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 105 |
-#else |
|
| 106 |
- MOVBQZX AL, CX |
|
| 107 |
- SHLQ CL, br_value // value <<= n |
|
| 108 |
-#endif |
|
| 109 |
- |
|
| 110 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 111 |
- |
|
| 112 |
- |
|
| 113 |
-#ifdef GOAMD64_v3 |
|
| 114 |
- SHRXQ peek_bits, br_value, AX // AX = (value >> peek_bits) & mask |
|
| 115 |
-#else |
|
| 116 |
- // val1 := br{{ var "id"}}.peekTopBits(peekBits)
|
|
| 117 |
- MOVQ peek_bits, CX |
|
| 118 |
- MOVQ br_value, AX |
|
| 119 |
- SHRQ CL, AX // AX = (value >> peek_bits) & mask |
|
| 120 |
-#endif |
|
| 121 |
- |
|
| 122 |
- // v1 := table[val1&mask] |
|
| 123 |
- MOVW 0(table)(AX*2), AX // AX - v1 |
|
| 124 |
- |
|
| 125 |
- // br{{ var "id"}}.advance(uint8(v1.entry))
|
|
| 126 |
- MOVB AH, BH // BH = uint8(v1.entry >> 8) |
|
| 127 |
- |
|
| 128 |
-#ifdef GOAMD64_v3 |
|
| 129 |
- MOVBQZX AL, CX |
|
| 130 |
- SHLXQ AX, br_value, br_value // value <<= n |
|
| 131 |
-#else |
|
| 132 |
- MOVBQZX AL, CX |
|
| 133 |
- SHLQ CL, br_value // value <<= n |
|
| 134 |
-#endif |
|
| 135 |
- |
|
| 136 |
- ADDQ CX, br_bits_read // bits_read += n |
|
| 137 |
- |
|
| 138 |
- |
|
| 139 |
- // these two writes get coalesced |
|
| 140 |
- // buf[stream][off] = uint8(v0.entry >> 8) |
|
| 141 |
- // buf[stream][off+1] = uint8(v1.entry >> 8) |
|
| 142 |
- MOVW BX, {{ var "bufofs" }}(buffer)(off*1)
|
|
| 143 |
- |
|
| 144 |
- // update the bitrader reader structure |
|
| 145 |
- MOVB br_bits_read, bitReaderShifted_bitsRead(br{{ var "id" }})
|
|
| 146 |
- MOVQ br_value, bitReaderShifted_value(br{{ var "id" }})
|
|
| 147 |
- MOVQ br_offset, bitReaderShifted_off(br{{ var "id" }})
|
|
| 148 |
-{{ end }}
|
|
| 149 |
- |
|
| 150 |
- {{ set "id" "0" }}
|
|
| 151 |
- {{ set "ofs" "0" }}
|
|
| 152 |
- {{ set "bufofs" "0" }} {{/* id * bufoff */}}
|
|
| 153 |
- {{ template "decode_2_values_x86" . }}
|
|
| 154 |
- |
|
| 155 |
- {{ set "id" "1" }}
|
|
| 156 |
- {{ set "ofs" "8" }}
|
|
| 157 |
- {{ set "bufofs" "256" }}
|
|
| 158 |
- {{ template "decode_2_values_x86" . }}
|
|
| 159 |
- |
|
| 160 |
- {{ set "id" "2" }}
|
|
| 161 |
- {{ set "ofs" "16" }}
|
|
| 162 |
- {{ set "bufofs" "512" }}
|
|
| 163 |
- {{ template "decode_2_values_x86" . }}
|
|
| 164 |
- |
|
| 165 |
- {{ set "id" "3" }}
|
|
| 166 |
- {{ set "ofs" "24" }}
|
|
| 167 |
- {{ set "bufofs" "768" }}
|
|
| 168 |
- {{ template "decode_2_values_x86" . }}
|
|
| 169 |
- |
|
| 170 |
- ADDQ $2, off // off += 2 |
|
| 171 |
- |
|
| 172 |
- TESTB DH, DH // any br[i].ofs < 4? |
|
| 173 |
- JNZ end |
|
| 174 |
- |
|
| 175 |
- CMPQ off, $bufoff |
|
| 176 |
- JL main_loop |
|
| 177 |
-end: |
|
| 178 |
- MOVQ 0(SP), BP |
|
| 179 |
- |
|
| 180 |
- MOVB off, ret+56(FP) |
|
| 181 |
- RET |
|
| 182 |
-#undef off |
|
| 183 |
-#undef buffer |
|
| 184 |
-#undef table |
|
| 185 |
- |
|
| 186 |
-#undef br_bits_read |
|
| 187 |
-#undef br_value |
|
| 188 |
-#undef br_offset |
|
| 189 |
-#undef peek_bits |
|
| 190 |
-#undef exhausted |
|
| 191 |
- |
|
| 192 |
-#undef br0 |
|
| 193 |
-#undef br1 |
|
| 194 |
-#undef br2 |
|
| 195 |
-#undef br3 |
| ... | ... |
@@ -191,3 +191,105 @@ func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
|
| 191 | 191 |
} |
| 192 | 192 |
return dst, nil |
| 193 | 193 |
} |
| 194 |
+ |
|
| 195 |
+// Decompress1X will decompress a 1X encoded stream. |
|
| 196 |
+// The cap of the output buffer will be the maximum decompressed size. |
|
| 197 |
+// The length of the supplied input must match the end of a block exactly. |
|
| 198 |
+func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
|
|
| 199 |
+ if len(d.dt.single) == 0 {
|
|
| 200 |
+ return nil, errors.New("no table loaded")
|
|
| 201 |
+ } |
|
| 202 |
+ if use8BitTables && d.actualTableLog <= 8 {
|
|
| 203 |
+ return d.decompress1X8Bit(dst, src) |
|
| 204 |
+ } |
|
| 205 |
+ var br bitReaderShifted |
|
| 206 |
+ err := br.init(src) |
|
| 207 |
+ if err != nil {
|
|
| 208 |
+ return dst, err |
|
| 209 |
+ } |
|
| 210 |
+ maxDecodedSize := cap(dst) |
|
| 211 |
+ dst = dst[:0] |
|
| 212 |
+ |
|
| 213 |
+ // Avoid bounds check by always having full sized table. |
|
| 214 |
+ const tlSize = 1 << tableLogMax |
|
| 215 |
+ const tlMask = tlSize - 1 |
|
| 216 |
+ dt := d.dt.single[:tlSize] |
|
| 217 |
+ |
|
| 218 |
+ // Use temp table to avoid bound checks/append penalty. |
|
| 219 |
+ bufs := d.buffer() |
|
| 220 |
+ buf := &bufs[0] |
|
| 221 |
+ var off uint8 |
|
| 222 |
+ |
|
| 223 |
+ for br.off >= 8 {
|
|
| 224 |
+ br.fillFast() |
|
| 225 |
+ v := dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 226 |
+ br.advance(uint8(v.entry)) |
|
| 227 |
+ buf[off+0] = uint8(v.entry >> 8) |
|
| 228 |
+ |
|
| 229 |
+ v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 230 |
+ br.advance(uint8(v.entry)) |
|
| 231 |
+ buf[off+1] = uint8(v.entry >> 8) |
|
| 232 |
+ |
|
| 233 |
+ // Refill |
|
| 234 |
+ br.fillFast() |
|
| 235 |
+ |
|
| 236 |
+ v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 237 |
+ br.advance(uint8(v.entry)) |
|
| 238 |
+ buf[off+2] = uint8(v.entry >> 8) |
|
| 239 |
+ |
|
| 240 |
+ v = dt[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 241 |
+ br.advance(uint8(v.entry)) |
|
| 242 |
+ buf[off+3] = uint8(v.entry >> 8) |
|
| 243 |
+ |
|
| 244 |
+ off += 4 |
|
| 245 |
+ if off == 0 {
|
|
| 246 |
+ if len(dst)+256 > maxDecodedSize {
|
|
| 247 |
+ br.close() |
|
| 248 |
+ d.bufs.Put(bufs) |
|
| 249 |
+ return nil, ErrMaxDecodedSizeExceeded |
|
| 250 |
+ } |
|
| 251 |
+ dst = append(dst, buf[:]...) |
|
| 252 |
+ } |
|
| 253 |
+ } |
|
| 254 |
+ |
|
| 255 |
+ if len(dst)+int(off) > maxDecodedSize {
|
|
| 256 |
+ d.bufs.Put(bufs) |
|
| 257 |
+ br.close() |
|
| 258 |
+ return nil, ErrMaxDecodedSizeExceeded |
|
| 259 |
+ } |
|
| 260 |
+ dst = append(dst, buf[:off]...) |
|
| 261 |
+ |
|
| 262 |
+ // br < 8, so uint8 is fine |
|
| 263 |
+ bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead |
|
| 264 |
+ for bitsLeft > 0 {
|
|
| 265 |
+ br.fill() |
|
| 266 |
+ if false && br.bitsRead >= 32 {
|
|
| 267 |
+ if br.off >= 4 {
|
|
| 268 |
+ v := br.in[br.off-4:] |
|
| 269 |
+ v = v[:4] |
|
| 270 |
+ low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24) |
|
| 271 |
+ br.value = (br.value << 32) | uint64(low) |
|
| 272 |
+ br.bitsRead -= 32 |
|
| 273 |
+ br.off -= 4 |
|
| 274 |
+ } else {
|
|
| 275 |
+ for br.off > 0 {
|
|
| 276 |
+ br.value = (br.value << 8) | uint64(br.in[br.off-1]) |
|
| 277 |
+ br.bitsRead -= 8 |
|
| 278 |
+ br.off-- |
|
| 279 |
+ } |
|
| 280 |
+ } |
|
| 281 |
+ } |
|
| 282 |
+ if len(dst) >= maxDecodedSize {
|
|
| 283 |
+ d.bufs.Put(bufs) |
|
| 284 |
+ br.close() |
|
| 285 |
+ return nil, ErrMaxDecodedSizeExceeded |
|
| 286 |
+ } |
|
| 287 |
+ v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask] |
|
| 288 |
+ nBits := uint8(v.entry) |
|
| 289 |
+ br.advance(nBits) |
|
| 290 |
+ bitsLeft -= nBits |
|
| 291 |
+ dst = append(dst, uint8(v.entry>>8)) |
|
| 292 |
+ } |
|
| 293 |
+ d.bufs.Put(bufs) |
|
| 294 |
+ return dst, br.close() |
|
| 295 |
+} |
| 194 | 296 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,34 @@ |
| 0 |
+// Package cpuinfo gives runtime info about the current CPU. |
|
| 1 |
+// |
|
| 2 |
+// This is a very limited module meant for use internally |
|
| 3 |
+// in this project. For more versatile solution check |
|
| 4 |
+// https://github.com/klauspost/cpuid. |
|
| 5 |
+package cpuinfo |
|
| 6 |
+ |
|
| 7 |
+// HasBMI1 checks whether an x86 CPU supports the BMI1 extension. |
|
| 8 |
+func HasBMI1() bool {
|
|
| 9 |
+ return hasBMI1 |
|
| 10 |
+} |
|
| 11 |
+ |
|
| 12 |
+// HasBMI2 checks whether an x86 CPU supports the BMI2 extension. |
|
| 13 |
+func HasBMI2() bool {
|
|
| 14 |
+ return hasBMI2 |
|
| 15 |
+} |
|
| 16 |
+ |
|
| 17 |
+// DisableBMI2 will disable BMI2, for testing purposes. |
|
| 18 |
+// Call returned function to restore previous state. |
|
| 19 |
+func DisableBMI2() func() {
|
|
| 20 |
+ old := hasBMI2 |
|
| 21 |
+ hasBMI2 = false |
|
| 22 |
+ return func() {
|
|
| 23 |
+ hasBMI2 = old |
|
| 24 |
+ } |
|
| 25 |
+} |
|
| 26 |
+ |
|
| 27 |
+// HasBMI checks whether an x86 CPU supports both BMI1 and BMI2 extensions. |
|
| 28 |
+func HasBMI() bool {
|
|
| 29 |
+ return HasBMI1() && HasBMI2() |
|
| 30 |
+} |
|
| 31 |
+ |
|
| 32 |
+var hasBMI1 bool |
|
| 33 |
+var hasBMI2 bool |
| 0 | 11 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,36 @@ |
| 0 |
+// +build !appengine |
|
| 1 |
+// +build gc |
|
| 2 |
+// +build !noasm |
|
| 3 |
+ |
|
| 4 |
+#include "textflag.h" |
|
| 5 |
+#include "funcdata.h" |
|
| 6 |
+#include "go_asm.h" |
|
| 7 |
+ |
|
| 8 |
+TEXT ·x86extensions(SB), NOSPLIT, $0 |
|
| 9 |
+ // 1. determine max EAX value |
|
| 10 |
+ XORQ AX, AX |
|
| 11 |
+ CPUID |
|
| 12 |
+ |
|
| 13 |
+ CMPQ AX, $7 |
|
| 14 |
+ JB unsupported |
|
| 15 |
+ |
|
| 16 |
+ // 2. EAX = 7, ECX = 0 --- see Table 3-8 "Information Returned by CPUID Instruction" |
|
| 17 |
+ MOVQ $7, AX |
|
| 18 |
+ MOVQ $0, CX |
|
| 19 |
+ CPUID |
|
| 20 |
+ |
|
| 21 |
+ BTQ $3, BX // bit 3 = BMI1 |
|
| 22 |
+ SETCS AL |
|
| 23 |
+ |
|
| 24 |
+ BTQ $8, BX // bit 8 = BMI2 |
|
| 25 |
+ SETCS AH |
|
| 26 |
+ |
|
| 27 |
+ MOVB AL, bmi1+0(FP) |
|
| 28 |
+ MOVB AH, bmi2+1(FP) |
|
| 29 |
+ RET |
|
| 30 |
+ |
|
| 31 |
+unsupported: |
|
| 32 |
+ XORQ AX, AX |
|
| 33 |
+ MOVB AL, bmi1+0(FP) |
|
| 34 |
+ MOVB AL, bmi2+1(FP) |
|
| 35 |
+ RET |
| ... | ... |
@@ -386,47 +386,31 @@ In practice this means that concurrency is often limited to utilizing about 3 co |
| 386 | 386 |
|
| 387 | 387 |
### Benchmarks |
| 388 | 388 |
|
| 389 |
-These are some examples of performance compared to [datadog cgo library](https://github.com/DataDog/zstd). |
|
| 390 |
- |
|
| 391 | 389 |
The first two are streaming decodes and the last are smaller inputs. |
| 392 |
- |
|
| 390 |
+ |
|
| 391 |
+Running on AMD Ryzen 9 3950X 16-Core Processor. AMD64 assembly used. |
|
| 392 |
+ |
|
| 393 | 393 |
``` |
| 394 |
-BenchmarkDecoderSilesia-8 3 385000067 ns/op 550.51 MB/s 5498 B/op 8 allocs/op |
|
| 395 |
-BenchmarkDecoderSilesiaCgo-8 6 197666567 ns/op 1072.25 MB/s 270672 B/op 8 allocs/op |
|
| 396 |
- |
|
| 397 |
-BenchmarkDecoderEnwik9-8 1 2027001600 ns/op 493.34 MB/s 10496 B/op 18 allocs/op |
|
| 398 |
-BenchmarkDecoderEnwik9Cgo-8 2 979499200 ns/op 1020.93 MB/s 270672 B/op 8 allocs/op |
|
| 399 |
- |
|
| 400 |
-Concurrent performance: |
|
| 401 |
- |
|
| 402 |
-BenchmarkDecoder_DecodeAllParallel/kppkn.gtb.zst-16 28915 42469 ns/op 4340.07 MB/s 114 B/op 0 allocs/op |
|
| 403 |
-BenchmarkDecoder_DecodeAllParallel/geo.protodata.zst-16 116505 9965 ns/op 11900.16 MB/s 16 B/op 0 allocs/op |
|
| 404 |
-BenchmarkDecoder_DecodeAllParallel/plrabn12.txt.zst-16 8952 134272 ns/op 3588.70 MB/s 915 B/op 0 allocs/op |
|
| 405 |
-BenchmarkDecoder_DecodeAllParallel/lcet10.txt.zst-16 11820 102538 ns/op 4161.90 MB/s 594 B/op 0 allocs/op |
|
| 406 |
-BenchmarkDecoder_DecodeAllParallel/asyoulik.txt.zst-16 34782 34184 ns/op 3661.88 MB/s 60 B/op 0 allocs/op |
|
| 407 |
-BenchmarkDecoder_DecodeAllParallel/alice29.txt.zst-16 27712 43447 ns/op 3500.58 MB/s 99 B/op 0 allocs/op |
|
| 408 |
-BenchmarkDecoder_DecodeAllParallel/html_x_4.zst-16 62826 18750 ns/op 21845.10 MB/s 104 B/op 0 allocs/op |
|
| 409 |
-BenchmarkDecoder_DecodeAllParallel/paper-100k.pdf.zst-16 631545 1794 ns/op 57078.74 MB/s 2 B/op 0 allocs/op |
|
| 410 |
-BenchmarkDecoder_DecodeAllParallel/fireworks.jpeg.zst-16 1690140 712 ns/op 172938.13 MB/s 1 B/op 0 allocs/op |
|
| 411 |
-BenchmarkDecoder_DecodeAllParallel/urls.10K.zst-16 10432 113593 ns/op 6180.73 MB/s 1143 B/op 0 allocs/op |
|
| 412 |
-BenchmarkDecoder_DecodeAllParallel/html.zst-16 113206 10671 ns/op 9596.27 MB/s 15 B/op 0 allocs/op |
|
| 413 |
-BenchmarkDecoder_DecodeAllParallel/comp-data.bin.zst-16 1530615 779 ns/op 5229.49 MB/s 0 B/op 0 allocs/op |
|
| 414 |
- |
|
| 415 |
-BenchmarkDecoder_DecodeAllParallelCgo/kppkn.gtb.zst-16 65217 16192 ns/op 11383.34 MB/s 46 B/op 0 allocs/op |
|
| 416 |
-BenchmarkDecoder_DecodeAllParallelCgo/geo.protodata.zst-16 292671 4039 ns/op 29363.19 MB/s 6 B/op 0 allocs/op |
|
| 417 |
-BenchmarkDecoder_DecodeAllParallelCgo/plrabn12.txt.zst-16 26314 46021 ns/op 10470.43 MB/s 293 B/op 0 allocs/op |
|
| 418 |
-BenchmarkDecoder_DecodeAllParallelCgo/lcet10.txt.zst-16 33897 34900 ns/op 12227.96 MB/s 205 B/op 0 allocs/op |
|
| 419 |
-BenchmarkDecoder_DecodeAllParallelCgo/asyoulik.txt.zst-16 104348 11433 ns/op 10949.01 MB/s 20 B/op 0 allocs/op |
|
| 420 |
-BenchmarkDecoder_DecodeAllParallelCgo/alice29.txt.zst-16 75949 15510 ns/op 9805.60 MB/s 32 B/op 0 allocs/op |
|
| 421 |
-BenchmarkDecoder_DecodeAllParallelCgo/html_x_4.zst-16 173910 6756 ns/op 60624.29 MB/s 37 B/op 0 allocs/op |
|
| 422 |
-BenchmarkDecoder_DecodeAllParallelCgo/paper-100k.pdf.zst-16 923076 1339 ns/op 76474.87 MB/s 1 B/op 0 allocs/op |
|
| 423 |
-BenchmarkDecoder_DecodeAllParallelCgo/fireworks.jpeg.zst-16 922920 1351 ns/op 91102.57 MB/s 2 B/op 0 allocs/op |
|
| 424 |
-BenchmarkDecoder_DecodeAllParallelCgo/urls.10K.zst-16 27649 43618 ns/op 16096.19 MB/s 407 B/op 0 allocs/op |
|
| 425 |
-BenchmarkDecoder_DecodeAllParallelCgo/html.zst-16 279073 4160 ns/op 24614.18 MB/s 6 B/op 0 allocs/op |
|
| 426 |
-BenchmarkDecoder_DecodeAllParallelCgo/comp-data.bin.zst-16 749938 1579 ns/op 2581.71 MB/s 0 B/op 0 allocs/op |
|
| 394 |
+BenchmarkDecoderSilesia-32 5 206878840 ns/op 1024.50 MB/s 49808 B/op 43 allocs/op |
|
| 395 |
+BenchmarkDecoderEnwik9-32 1 1271809000 ns/op 786.28 MB/s 72048 B/op 52 allocs/op |
|
| 396 |
+ |
|
| 397 |
+Concurrent blocks, performance: |
|
| 398 |
+ |
|
| 399 |
+BenchmarkDecoder_DecodeAllParallel/kppkn.gtb.zst-32 67356 17857 ns/op 10321.96 MB/s 22.48 pct 102 B/op 0 allocs/op |
|
| 400 |
+BenchmarkDecoder_DecodeAllParallel/geo.protodata.zst-32 266656 4421 ns/op 26823.21 MB/s 11.89 pct 19 B/op 0 allocs/op |
|
| 401 |
+BenchmarkDecoder_DecodeAllParallel/plrabn12.txt.zst-32 20992 56842 ns/op 8477.17 MB/s 39.90 pct 754 B/op 0 allocs/op |
|
| 402 |
+BenchmarkDecoder_DecodeAllParallel/lcet10.txt.zst-32 27456 43932 ns/op 9714.01 MB/s 33.27 pct 524 B/op 0 allocs/op |
|
| 403 |
+BenchmarkDecoder_DecodeAllParallel/asyoulik.txt.zst-32 78432 15047 ns/op 8319.15 MB/s 40.34 pct 66 B/op 0 allocs/op |
|
| 404 |
+BenchmarkDecoder_DecodeAllParallel/alice29.txt.zst-32 65800 18436 ns/op 8249.63 MB/s 37.75 pct 88 B/op 0 allocs/op |
|
| 405 |
+BenchmarkDecoder_DecodeAllParallel/html_x_4.zst-32 102993 11523 ns/op 35546.09 MB/s 3.637 pct 143 B/op 0 allocs/op |
|
| 406 |
+BenchmarkDecoder_DecodeAllParallel/paper-100k.pdf.zst-32 1000000 1070 ns/op 95720.98 MB/s 80.53 pct 3 B/op 0 allocs/op |
|
| 407 |
+BenchmarkDecoder_DecodeAllParallel/fireworks.jpeg.zst-32 749802 1752 ns/op 70272.35 MB/s 100.0 pct 5 B/op 0 allocs/op |
|
| 408 |
+BenchmarkDecoder_DecodeAllParallel/urls.10K.zst-32 22640 52934 ns/op 13263.37 MB/s 26.25 pct 1014 B/op 0 allocs/op |
|
| 409 |
+BenchmarkDecoder_DecodeAllParallel/html.zst-32 226412 5232 ns/op 19572.27 MB/s 14.49 pct 20 B/op 0 allocs/op |
|
| 410 |
+BenchmarkDecoder_DecodeAllParallel/comp-data.bin.zst-32 923041 1276 ns/op 3194.71 MB/s 31.26 pct 0 B/op 0 allocs/op |
|
| 427 | 411 |
``` |
| 428 | 412 |
|
| 429 |
-This reflects the performance around May 2020, but this may be out of date. |
|
| 413 |
+This reflects the performance around May 2022, but this may be out of date. |
|
| 430 | 414 |
|
| 431 | 415 |
## Zstd inside ZIP files |
| 432 | 416 |
|
| ... | ... |
@@ -63,13 +63,6 @@ func (b *bitReader) get32BitsFast(n uint8) uint32 {
|
| 63 | 63 |
return v |
| 64 | 64 |
} |
| 65 | 65 |
|
| 66 |
-func (b *bitReader) get16BitsFast(n uint8) uint16 {
|
|
| 67 |
- const regMask = 64 - 1 |
|
| 68 |
- v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask)) |
|
| 69 |
- b.bitsRead += n |
|
| 70 |
- return v |
|
| 71 |
-} |
|
| 72 |
- |
|
| 73 | 66 |
// fillFast() will make sure at least 32 bits are available. |
| 74 | 67 |
// There must be at least 4 bytes available. |
| 75 | 68 |
func (b *bitReader) fillFast() {
|
| ... | ... |
@@ -5,8 +5,6 @@ |
| 5 | 5 |
|
| 6 | 6 |
package zstd |
| 7 | 7 |
|
| 8 |
-import "fmt" |
|
| 9 |
- |
|
| 10 | 8 |
// bitWriter will write bits. |
| 11 | 9 |
// First bit will be LSB of the first byte of output. |
| 12 | 10 |
type bitWriter struct {
|
| ... | ... |
@@ -73,80 +71,6 @@ func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
|
| 73 | 73 |
b.nBits += bits |
| 74 | 74 |
} |
| 75 | 75 |
|
| 76 |
-// flush will flush all pending full bytes. |
|
| 77 |
-// There will be at least 56 bits available for writing when this has been called. |
|
| 78 |
-// Using flush32 is faster, but leaves less space for writing. |
|
| 79 |
-func (b *bitWriter) flush() {
|
|
| 80 |
- v := b.nBits >> 3 |
|
| 81 |
- switch v {
|
|
| 82 |
- case 0: |
|
| 83 |
- case 1: |
|
| 84 |
- b.out = append(b.out, |
|
| 85 |
- byte(b.bitContainer), |
|
| 86 |
- ) |
|
| 87 |
- case 2: |
|
| 88 |
- b.out = append(b.out, |
|
| 89 |
- byte(b.bitContainer), |
|
| 90 |
- byte(b.bitContainer>>8), |
|
| 91 |
- ) |
|
| 92 |
- case 3: |
|
| 93 |
- b.out = append(b.out, |
|
| 94 |
- byte(b.bitContainer), |
|
| 95 |
- byte(b.bitContainer>>8), |
|
| 96 |
- byte(b.bitContainer>>16), |
|
| 97 |
- ) |
|
| 98 |
- case 4: |
|
| 99 |
- b.out = append(b.out, |
|
| 100 |
- byte(b.bitContainer), |
|
| 101 |
- byte(b.bitContainer>>8), |
|
| 102 |
- byte(b.bitContainer>>16), |
|
| 103 |
- byte(b.bitContainer>>24), |
|
| 104 |
- ) |
|
| 105 |
- case 5: |
|
| 106 |
- b.out = append(b.out, |
|
| 107 |
- byte(b.bitContainer), |
|
| 108 |
- byte(b.bitContainer>>8), |
|
| 109 |
- byte(b.bitContainer>>16), |
|
| 110 |
- byte(b.bitContainer>>24), |
|
| 111 |
- byte(b.bitContainer>>32), |
|
| 112 |
- ) |
|
| 113 |
- case 6: |
|
| 114 |
- b.out = append(b.out, |
|
| 115 |
- byte(b.bitContainer), |
|
| 116 |
- byte(b.bitContainer>>8), |
|
| 117 |
- byte(b.bitContainer>>16), |
|
| 118 |
- byte(b.bitContainer>>24), |
|
| 119 |
- byte(b.bitContainer>>32), |
|
| 120 |
- byte(b.bitContainer>>40), |
|
| 121 |
- ) |
|
| 122 |
- case 7: |
|
| 123 |
- b.out = append(b.out, |
|
| 124 |
- byte(b.bitContainer), |
|
| 125 |
- byte(b.bitContainer>>8), |
|
| 126 |
- byte(b.bitContainer>>16), |
|
| 127 |
- byte(b.bitContainer>>24), |
|
| 128 |
- byte(b.bitContainer>>32), |
|
| 129 |
- byte(b.bitContainer>>40), |
|
| 130 |
- byte(b.bitContainer>>48), |
|
| 131 |
- ) |
|
| 132 |
- case 8: |
|
| 133 |
- b.out = append(b.out, |
|
| 134 |
- byte(b.bitContainer), |
|
| 135 |
- byte(b.bitContainer>>8), |
|
| 136 |
- byte(b.bitContainer>>16), |
|
| 137 |
- byte(b.bitContainer>>24), |
|
| 138 |
- byte(b.bitContainer>>32), |
|
| 139 |
- byte(b.bitContainer>>40), |
|
| 140 |
- byte(b.bitContainer>>48), |
|
| 141 |
- byte(b.bitContainer>>56), |
|
| 142 |
- ) |
|
| 143 |
- default: |
|
| 144 |
- panic(fmt.Errorf("bits (%d) > 64", b.nBits))
|
|
| 145 |
- } |
|
| 146 |
- b.bitContainer >>= v << 3 |
|
| 147 |
- b.nBits &= 7 |
|
| 148 |
-} |
|
| 149 |
- |
|
| 150 | 76 |
// flush32 will flush out, so there are at least 32 bits available for writing. |
| 151 | 77 |
func (b *bitWriter) flush32() {
|
| 152 | 78 |
if b.nBits < 32 {
|
| ... | ... |
@@ -5,9 +5,14 @@ |
| 5 | 5 |
package zstd |
| 6 | 6 |
|
| 7 | 7 |
import ( |
| 8 |
+ "bytes" |
|
| 9 |
+ "encoding/binary" |
|
| 8 | 10 |
"errors" |
| 9 | 11 |
"fmt" |
| 10 | 12 |
"io" |
| 13 |
+ "io/ioutil" |
|
| 14 |
+ "os" |
|
| 15 |
+ "path/filepath" |
|
| 11 | 16 |
"sync" |
| 12 | 17 |
|
| 13 | 18 |
"github.com/klauspost/compress/huff0" |
| ... | ... |
@@ -38,14 +43,14 @@ const ( |
| 38 | 38 |
// maxCompressedBlockSize is the biggest allowed compressed block size (128KB) |
| 39 | 39 |
maxCompressedBlockSize = 128 << 10 |
| 40 | 40 |
|
| 41 |
+ compressedBlockOverAlloc = 16 |
|
| 42 |
+ maxCompressedBlockSizeAlloc = 128<<10 + compressedBlockOverAlloc |
|
| 43 |
+ |
|
| 41 | 44 |
// Maximum possible block size (all Raw+Uncompressed). |
| 42 | 45 |
maxBlockSize = (1 << 21) - 1 |
| 43 | 46 |
|
| 44 |
- // https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#literals_section_header |
|
| 45 |
- maxCompressedLiteralSize = 1 << 18 |
|
| 46 |
- maxRLELiteralSize = 1 << 20 |
|
| 47 |
- maxMatchLen = 131074 |
|
| 48 |
- maxSequences = 0x7f00 + 0xffff |
|
| 47 |
+ maxMatchLen = 131074 |
|
| 48 |
+ maxSequences = 0x7f00 + 0xffff |
|
| 49 | 49 |
|
| 50 | 50 |
// We support slightly less than the reference decoder to be able to |
| 51 | 51 |
// use ints on 32 bit archs. |
| ... | ... |
@@ -97,7 +102,6 @@ type blockDec struct {
|
| 97 | 97 |
|
| 98 | 98 |
// Block is RLE, this is the size. |
| 99 | 99 |
RLESize uint32 |
| 100 |
- tmp [4]byte |
|
| 101 | 100 |
|
| 102 | 101 |
Type blockType |
| 103 | 102 |
|
| ... | ... |
@@ -136,7 +140,7 @@ func (b *blockDec) reset(br byteBuffer, windowSize uint64) error {
|
| 136 | 136 |
b.Type = blockType((bh >> 1) & 3) |
| 137 | 137 |
// find size. |
| 138 | 138 |
cSize := int(bh >> 3) |
| 139 |
- maxSize := maxBlockSize |
|
| 139 |
+ maxSize := maxCompressedBlockSizeAlloc |
|
| 140 | 140 |
switch b.Type {
|
| 141 | 141 |
case blockTypeReserved: |
| 142 | 142 |
return ErrReservedBlockType |
| ... | ... |
@@ -157,9 +161,9 @@ func (b *blockDec) reset(br byteBuffer, windowSize uint64) error {
|
| 157 | 157 |
println("Data size on stream:", cSize)
|
| 158 | 158 |
} |
| 159 | 159 |
b.RLESize = 0 |
| 160 |
- maxSize = maxCompressedBlockSize |
|
| 160 |
+ maxSize = maxCompressedBlockSizeAlloc |
|
| 161 | 161 |
if windowSize < maxCompressedBlockSize && b.lowMem {
|
| 162 |
- maxSize = int(windowSize) |
|
| 162 |
+ maxSize = int(windowSize) + compressedBlockOverAlloc |
|
| 163 | 163 |
} |
| 164 | 164 |
if cSize > maxCompressedBlockSize || uint64(cSize) > b.WindowSize {
|
| 165 | 165 |
if debugDecoder {
|
| ... | ... |
@@ -190,9 +194,9 @@ func (b *blockDec) reset(br byteBuffer, windowSize uint64) error {
|
| 190 | 190 |
// Read block data. |
| 191 | 191 |
if cap(b.dataStorage) < cSize {
|
| 192 | 192 |
if b.lowMem || cSize > maxCompressedBlockSize {
|
| 193 |
- b.dataStorage = make([]byte, 0, cSize) |
|
| 193 |
+ b.dataStorage = make([]byte, 0, cSize+compressedBlockOverAlloc) |
|
| 194 | 194 |
} else {
|
| 195 |
- b.dataStorage = make([]byte, 0, maxCompressedBlockSize) |
|
| 195 |
+ b.dataStorage = make([]byte, 0, maxCompressedBlockSizeAlloc) |
|
| 196 | 196 |
} |
| 197 | 197 |
} |
| 198 | 198 |
if cap(b.dst) <= maxSize {
|
| ... | ... |
@@ -360,14 +364,9 @@ func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err |
| 360 | 360 |
} |
| 361 | 361 |
if cap(b.literalBuf) < litRegenSize {
|
| 362 | 362 |
if b.lowMem {
|
| 363 |
- b.literalBuf = make([]byte, litRegenSize) |
|
| 363 |
+ b.literalBuf = make([]byte, litRegenSize, litRegenSize+compressedBlockOverAlloc) |
|
| 364 | 364 |
} else {
|
| 365 |
- if litRegenSize > maxCompressedLiteralSize {
|
|
| 366 |
- // Exceptional |
|
| 367 |
- b.literalBuf = make([]byte, litRegenSize) |
|
| 368 |
- } else {
|
|
| 369 |
- b.literalBuf = make([]byte, litRegenSize, maxCompressedLiteralSize) |
|
| 370 |
- } |
|
| 365 |
+ b.literalBuf = make([]byte, litRegenSize, maxCompressedBlockSize+compressedBlockOverAlloc) |
|
| 371 | 366 |
} |
| 372 | 367 |
} |
| 373 | 368 |
literals = b.literalBuf[:litRegenSize] |
| ... | ... |
@@ -397,14 +396,14 @@ func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err |
| 397 | 397 |
// Ensure we have space to store it. |
| 398 | 398 |
if cap(b.literalBuf) < litRegenSize {
|
| 399 | 399 |
if b.lowMem {
|
| 400 |
- b.literalBuf = make([]byte, 0, litRegenSize) |
|
| 400 |
+ b.literalBuf = make([]byte, 0, litRegenSize+compressedBlockOverAlloc) |
|
| 401 | 401 |
} else {
|
| 402 |
- b.literalBuf = make([]byte, 0, maxCompressedLiteralSize) |
|
| 402 |
+ b.literalBuf = make([]byte, 0, maxCompressedBlockSize+compressedBlockOverAlloc) |
|
| 403 | 403 |
} |
| 404 | 404 |
} |
| 405 | 405 |
var err error |
| 406 | 406 |
// Use our out buffer. |
| 407 |
- huff.MaxDecodedSize = maxCompressedBlockSize |
|
| 407 |
+ huff.MaxDecodedSize = litRegenSize |
|
| 408 | 408 |
if fourStreams {
|
| 409 | 409 |
literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals) |
| 410 | 410 |
} else {
|
| ... | ... |
@@ -429,9 +428,9 @@ func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err |
| 429 | 429 |
// Ensure we have space to store it. |
| 430 | 430 |
if cap(b.literalBuf) < litRegenSize {
|
| 431 | 431 |
if b.lowMem {
|
| 432 |
- b.literalBuf = make([]byte, 0, litRegenSize) |
|
| 432 |
+ b.literalBuf = make([]byte, 0, litRegenSize+compressedBlockOverAlloc) |
|
| 433 | 433 |
} else {
|
| 434 |
- b.literalBuf = make([]byte, 0, maxCompressedBlockSize) |
|
| 434 |
+ b.literalBuf = make([]byte, 0, maxCompressedBlockSize+compressedBlockOverAlloc) |
|
| 435 | 435 |
} |
| 436 | 436 |
} |
| 437 | 437 |
huff := hist.huffTree |
| ... | ... |
@@ -448,7 +447,7 @@ func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err |
| 448 | 448 |
return in, err |
| 449 | 449 |
} |
| 450 | 450 |
hist.huffTree = huff |
| 451 |
- huff.MaxDecodedSize = maxCompressedBlockSize |
|
| 451 |
+ huff.MaxDecodedSize = litRegenSize |
|
| 452 | 452 |
// Use our out buffer. |
| 453 | 453 |
if fourStreams {
|
| 454 | 454 |
literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals) |
| ... | ... |
@@ -463,6 +462,8 @@ func (b *blockDec) decodeLiterals(in []byte, hist *history) (remain []byte, err |
| 463 | 463 |
if len(literals) != litRegenSize {
|
| 464 | 464 |
return in, fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
|
| 465 | 465 |
} |
| 466 |
+ // Re-cap to get extra size. |
|
| 467 |
+ literals = b.literalBuf[:len(literals)] |
|
| 466 | 468 |
if debugDecoder {
|
| 467 | 469 |
printf("Decompressed %d literals into %d bytes\n", litCompSize, litRegenSize)
|
| 468 | 470 |
} |
| ... | ... |
@@ -486,10 +487,15 @@ func (b *blockDec) decodeCompressed(hist *history) error {
|
| 486 | 486 |
b.dst = append(b.dst, hist.decoders.literals...) |
| 487 | 487 |
return nil |
| 488 | 488 |
} |
| 489 |
- err = hist.decoders.decodeSync(hist) |
|
| 489 |
+ before := len(hist.decoders.out) |
|
| 490 |
+ err = hist.decoders.decodeSync(hist.b[hist.ignoreBuffer:]) |
|
| 490 | 491 |
if err != nil {
|
| 491 | 492 |
return err |
| 492 | 493 |
} |
| 494 |
+ if hist.decoders.maxSyncLen > 0 {
|
|
| 495 |
+ hist.decoders.maxSyncLen += uint64(before) |
|
| 496 |
+ hist.decoders.maxSyncLen -= uint64(len(hist.decoders.out)) |
|
| 497 |
+ } |
|
| 493 | 498 |
b.dst = hist.decoders.out |
| 494 | 499 |
hist.recentOffsets = hist.decoders.prevOffset |
| 495 | 500 |
return nil |
| ... | ... |
@@ -632,6 +638,22 @@ func (b *blockDec) prepareSequences(in []byte, hist *history) (err error) {
|
| 632 | 632 |
println("initializing sequences:", err)
|
| 633 | 633 |
return err |
| 634 | 634 |
} |
| 635 |
+ // Extract blocks... |
|
| 636 |
+ if false && hist.dict == nil {
|
|
| 637 |
+ fatalErr := func(err error) {
|
|
| 638 |
+ if err != nil {
|
|
| 639 |
+ panic(err) |
|
| 640 |
+ } |
|
| 641 |
+ } |
|
| 642 |
+ fn := fmt.Sprintf("n-%d-lits-%d-prev-%d-%d-%d-win-%d.blk", hist.decoders.nSeqs, len(hist.decoders.literals), hist.recentOffsets[0], hist.recentOffsets[1], hist.recentOffsets[2], hist.windowSize)
|
|
| 643 |
+ var buf bytes.Buffer |
|
| 644 |
+ fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.litLengths.fse)) |
|
| 645 |
+ fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.matchLengths.fse)) |
|
| 646 |
+ fatalErr(binary.Write(&buf, binary.LittleEndian, hist.decoders.offsets.fse)) |
|
| 647 |
+ buf.Write(in) |
|
| 648 |
+ ioutil.WriteFile(filepath.Join("testdata", "seqs", fn), buf.Bytes(), os.ModePerm)
|
|
| 649 |
+ } |
|
| 650 |
+ |
|
| 635 | 651 |
return nil |
| 636 | 652 |
} |
| 637 | 653 |
|
| ... | ... |
@@ -650,6 +672,7 @@ func (b *blockDec) decodeSequences(hist *history) error {
|
| 650 | 650 |
} |
| 651 | 651 |
hist.decoders.windowSize = hist.windowSize |
| 652 | 652 |
hist.decoders.prevOffset = hist.recentOffsets |
| 653 |
+ |
|
| 653 | 654 |
err := hist.decoders.decode(b.sequence) |
| 654 | 655 |
hist.recentOffsets = hist.decoders.prevOffset |
| 655 | 656 |
return err |
| ... | ... |
@@ -23,7 +23,7 @@ type byteBuffer interface {
|
| 23 | 23 |
readByte() (byte, error) |
| 24 | 24 |
|
| 25 | 25 |
// Skip n bytes. |
| 26 |
- skipN(n int) error |
|
| 26 |
+ skipN(n int64) error |
|
| 27 | 27 |
} |
| 28 | 28 |
|
| 29 | 29 |
// in-memory buffer |
| ... | ... |
@@ -52,10 +52,6 @@ func (b *byteBuf) readBig(n int, dst []byte) ([]byte, error) {
|
| 52 | 52 |
return r, nil |
| 53 | 53 |
} |
| 54 | 54 |
|
| 55 |
-func (b *byteBuf) remain() []byte {
|
|
| 56 |
- return *b |
|
| 57 |
-} |
|
| 58 |
- |
|
| 59 | 55 |
func (b *byteBuf) readByte() (byte, error) {
|
| 60 | 56 |
bb := *b |
| 61 | 57 |
if len(bb) < 1 {
|
| ... | ... |
@@ -66,9 +62,12 @@ func (b *byteBuf) readByte() (byte, error) {
|
| 66 | 66 |
return r, nil |
| 67 | 67 |
} |
| 68 | 68 |
|
| 69 |
-func (b *byteBuf) skipN(n int) error {
|
|
| 69 |
+func (b *byteBuf) skipN(n int64) error {
|
|
| 70 | 70 |
bb := *b |
| 71 |
- if len(bb) < n {
|
|
| 71 |
+ if n < 0 {
|
|
| 72 |
+ return fmt.Errorf("negative skip (%d) requested", n)
|
|
| 73 |
+ } |
|
| 74 |
+ if int64(len(bb)) < n {
|
|
| 72 | 75 |
return io.ErrUnexpectedEOF |
| 73 | 76 |
} |
| 74 | 77 |
*b = bb[n:] |
| ... | ... |
@@ -124,9 +123,9 @@ func (r *readerWrapper) readByte() (byte, error) {
|
| 124 | 124 |
return r.tmp[0], nil |
| 125 | 125 |
} |
| 126 | 126 |
|
| 127 |
-func (r *readerWrapper) skipN(n int) error {
|
|
| 128 |
- n2, err := io.CopyN(ioutil.Discard, r.r, int64(n)) |
|
| 129 |
- if n2 != int64(n) {
|
|
| 127 |
+func (r *readerWrapper) skipN(n int64) error {
|
|
| 128 |
+ n2, err := io.CopyN(ioutil.Discard, r.r, n) |
|
| 129 |
+ if n2 != n {
|
|
| 130 | 130 |
err = io.ErrUnexpectedEOF |
| 131 | 131 |
} |
| 132 | 132 |
return err |
| ... | ... |
@@ -13,12 +13,6 @@ type byteReader struct {
|
| 13 | 13 |
off int |
| 14 | 14 |
} |
| 15 | 15 |
|
| 16 |
-// init will initialize the reader and set the input. |
|
| 17 |
-func (b *byteReader) init(in []byte) {
|
|
| 18 |
- b.b = in |
|
| 19 |
- b.off = 0 |
|
| 20 |
-} |
|
| 21 |
- |
|
| 22 | 16 |
// advance the stream b n bytes. |
| 23 | 17 |
func (b *byteReader) advance(n uint) {
|
| 24 | 18 |
b.off += int(n) |
| ... | ... |
@@ -347,18 +347,23 @@ func (d *Decoder) DecodeAll(input, dst []byte) ([]byte, error) {
|
| 347 | 347 |
} |
| 348 | 348 |
frame.history.setDict(&dict) |
| 349 | 349 |
} |
| 350 |
- |
|
| 351 |
- if frame.FrameContentSize != fcsUnknown && frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) {
|
|
| 352 |
- return dst, ErrDecoderSizeExceeded |
|
| 350 |
+ if frame.WindowSize > d.o.maxWindowSize {
|
|
| 351 |
+ if debugDecoder {
|
|
| 352 |
+ println("window size exceeded:", frame.WindowSize, ">", d.o.maxWindowSize)
|
|
| 353 |
+ } |
|
| 354 |
+ return dst, ErrWindowSizeExceeded |
|
| 353 | 355 |
} |
| 354 |
- if frame.FrameContentSize < 1<<30 {
|
|
| 355 |
- // Never preallocate more than 1 GB up front. |
|
| 356 |
+ if frame.FrameContentSize != fcsUnknown {
|
|
| 357 |
+ if frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) {
|
|
| 358 |
+ return dst, ErrDecoderSizeExceeded |
|
| 359 |
+ } |
|
| 356 | 360 |
if cap(dst)-len(dst) < int(frame.FrameContentSize) {
|
| 357 |
- dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize)) |
|
| 361 |
+ dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize)+compressedBlockOverAlloc) |
|
| 358 | 362 |
copy(dst2, dst) |
| 359 | 363 |
dst = dst2 |
| 360 | 364 |
} |
| 361 | 365 |
} |
| 366 |
+ |
|
| 362 | 367 |
if cap(dst) == 0 {
|
| 363 | 368 |
// Allocate len(input) * 2 by default if nothing is provided |
| 364 | 369 |
// and we didn't get frame content size. |
| ... | ... |
@@ -437,7 +442,7 @@ func (d *Decoder) nextBlock(blocking bool) (ok bool) {
|
| 437 | 437 |
println("got", len(d.current.b), "bytes, error:", d.current.err, "data crc:", tmp)
|
| 438 | 438 |
} |
| 439 | 439 |
|
| 440 |
- if len(next.b) > 0 {
|
|
| 440 |
+ if !d.o.ignoreChecksum && len(next.b) > 0 {
|
|
| 441 | 441 |
n, err := d.current.crc.Write(next.b) |
| 442 | 442 |
if err == nil {
|
| 443 | 443 |
if n != len(next.b) {
|
| ... | ... |
@@ -449,7 +454,7 @@ func (d *Decoder) nextBlock(blocking bool) (ok bool) {
|
| 449 | 449 |
got := d.current.crc.Sum64() |
| 450 | 450 |
var tmp [4]byte |
| 451 | 451 |
binary.LittleEndian.PutUint32(tmp[:], uint32(got)) |
| 452 |
- if !bytes.Equal(tmp[:], next.d.checkCRC) && !ignoreCRC {
|
|
| 452 |
+ if !d.o.ignoreChecksum && !bytes.Equal(tmp[:], next.d.checkCRC) {
|
|
| 453 | 453 |
if debugDecoder {
|
| 454 | 454 |
println("CRC Check Failed:", tmp[:], " (got) !=", next.d.checkCRC, "(on stream)")
|
| 455 | 455 |
} |
| ... | ... |
@@ -533,9 +538,15 @@ func (d *Decoder) nextBlockSync() (ok bool) {
|
| 533 | 533 |
|
| 534 | 534 |
// Update/Check CRC |
| 535 | 535 |
if d.frame.HasCheckSum {
|
| 536 |
- d.frame.crc.Write(d.current.b) |
|
| 536 |
+ if !d.o.ignoreChecksum {
|
|
| 537 |
+ d.frame.crc.Write(d.current.b) |
|
| 538 |
+ } |
|
| 537 | 539 |
if d.current.d.Last {
|
| 538 |
- d.current.err = d.frame.checkCRC() |
|
| 540 |
+ if !d.o.ignoreChecksum {
|
|
| 541 |
+ d.current.err = d.frame.checkCRC() |
|
| 542 |
+ } else {
|
|
| 543 |
+ d.current.err = d.frame.consumeCRC() |
|
| 544 |
+ } |
|
| 539 | 545 |
if d.current.err != nil {
|
| 540 | 546 |
println("CRC error:", d.current.err)
|
| 541 | 547 |
return false |
| ... | ... |
@@ -629,60 +640,18 @@ func (d *Decoder) startSyncDecoder(r io.Reader) error {
|
| 629 | 629 |
|
| 630 | 630 |
// Create Decoder: |
| 631 | 631 |
// ASYNC: |
| 632 |
-// Spawn 4 go routines. |
|
| 633 |
-// 0: Read frames and decode blocks. |
|
| 634 |
-// 1: Decode block and literals. Receives hufftree and seqdecs, returns seqdecs and huff tree. |
|
| 635 |
-// 2: Wait for recentOffsets if needed. Decode sequences, send recentOffsets. |
|
| 636 |
-// 3: Wait for stream history, execute sequences, send stream history. |
|
| 632 |
+// Spawn 3 go routines. |
|
| 633 |
+// 0: Read frames and decode block literals. |
|
| 634 |
+// 1: Decode sequences. |
|
| 635 |
+// 2: Execute sequences, send to output. |
|
| 637 | 636 |
func (d *Decoder) startStreamDecoder(ctx context.Context, r io.Reader, output chan decodeOutput) {
|
| 638 | 637 |
defer d.streamWg.Done() |
| 639 | 638 |
br := readerWrapper{r: r}
|
| 640 | 639 |
|
| 641 |
- var seqPrepare = make(chan *blockDec, d.o.concurrent) |
|
| 642 | 640 |
var seqDecode = make(chan *blockDec, d.o.concurrent) |
| 643 | 641 |
var seqExecute = make(chan *blockDec, d.o.concurrent) |
| 644 | 642 |
|
| 645 |
- // Async 1: Prepare blocks... |
|
| 646 |
- go func() {
|
|
| 647 |
- var hist history |
|
| 648 |
- var hasErr bool |
|
| 649 |
- for block := range seqPrepare {
|
|
| 650 |
- if hasErr {
|
|
| 651 |
- if block != nil {
|
|
| 652 |
- seqDecode <- block |
|
| 653 |
- } |
|
| 654 |
- continue |
|
| 655 |
- } |
|
| 656 |
- if block.async.newHist != nil {
|
|
| 657 |
- if debugDecoder {
|
|
| 658 |
- println("Async 1: new history")
|
|
| 659 |
- } |
|
| 660 |
- hist.reset() |
|
| 661 |
- if block.async.newHist.dict != nil {
|
|
| 662 |
- hist.setDict(block.async.newHist.dict) |
|
| 663 |
- } |
|
| 664 |
- } |
|
| 665 |
- if block.err != nil || block.Type != blockTypeCompressed {
|
|
| 666 |
- hasErr = block.err != nil |
|
| 667 |
- seqDecode <- block |
|
| 668 |
- continue |
|
| 669 |
- } |
|
| 670 |
- |
|
| 671 |
- remain, err := block.decodeLiterals(block.data, &hist) |
|
| 672 |
- block.err = err |
|
| 673 |
- hasErr = block.err != nil |
|
| 674 |
- if err == nil {
|
|
| 675 |
- block.async.literals = hist.decoders.literals |
|
| 676 |
- block.async.seqData = remain |
|
| 677 |
- } else if debugDecoder {
|
|
| 678 |
- println("decodeLiterals error:", err)
|
|
| 679 |
- } |
|
| 680 |
- seqDecode <- block |
|
| 681 |
- } |
|
| 682 |
- close(seqDecode) |
|
| 683 |
- }() |
|
| 684 |
- |
|
| 685 |
- // Async 2: Decode sequences... |
|
| 643 |
+ // Async 1: Decode sequences... |
|
| 686 | 644 |
go func() {
|
| 687 | 645 |
var hist history |
| 688 | 646 |
var hasErr bool |
| ... | ... |
@@ -696,7 +665,7 @@ func (d *Decoder) startStreamDecoder(ctx context.Context, r io.Reader, output ch |
| 696 | 696 |
} |
| 697 | 697 |
if block.async.newHist != nil {
|
| 698 | 698 |
if debugDecoder {
|
| 699 |
- println("Async 2: new history, recent:", block.async.newHist.recentOffsets)
|
|
| 699 |
+ println("Async 1: new history, recent:", block.async.newHist.recentOffsets)
|
|
| 700 | 700 |
} |
| 701 | 701 |
hist.decoders = block.async.newHist.decoders |
| 702 | 702 |
hist.recentOffsets = block.async.newHist.recentOffsets |
| ... | ... |
@@ -750,7 +719,7 @@ func (d *Decoder) startStreamDecoder(ctx context.Context, r io.Reader, output ch |
| 750 | 750 |
} |
| 751 | 751 |
if block.async.newHist != nil {
|
| 752 | 752 |
if debugDecoder {
|
| 753 |
- println("Async 3: new history")
|
|
| 753 |
+ println("Async 2: new history")
|
|
| 754 | 754 |
} |
| 755 | 755 |
hist.windowSize = block.async.newHist.windowSize |
| 756 | 756 |
hist.allocFrameBuffer = block.async.newHist.allocFrameBuffer |
| ... | ... |
@@ -837,6 +806,33 @@ func (d *Decoder) startStreamDecoder(ctx context.Context, r io.Reader, output ch |
| 837 | 837 |
|
| 838 | 838 |
decodeStream: |
| 839 | 839 |
for {
|
| 840 |
+ var hist history |
|
| 841 |
+ var hasErr bool |
|
| 842 |
+ |
|
| 843 |
+ decodeBlock := func(block *blockDec) {
|
|
| 844 |
+ if hasErr {
|
|
| 845 |
+ if block != nil {
|
|
| 846 |
+ seqDecode <- block |
|
| 847 |
+ } |
|
| 848 |
+ return |
|
| 849 |
+ } |
|
| 850 |
+ if block.err != nil || block.Type != blockTypeCompressed {
|
|
| 851 |
+ hasErr = block.err != nil |
|
| 852 |
+ seqDecode <- block |
|
| 853 |
+ return |
|
| 854 |
+ } |
|
| 855 |
+ |
|
| 856 |
+ remain, err := block.decodeLiterals(block.data, &hist) |
|
| 857 |
+ block.err = err |
|
| 858 |
+ hasErr = block.err != nil |
|
| 859 |
+ if err == nil {
|
|
| 860 |
+ block.async.literals = hist.decoders.literals |
|
| 861 |
+ block.async.seqData = remain |
|
| 862 |
+ } else if debugDecoder {
|
|
| 863 |
+ println("decodeLiterals error:", err)
|
|
| 864 |
+ } |
|
| 865 |
+ seqDecode <- block |
|
| 866 |
+ } |
|
| 840 | 867 |
frame := d.frame |
| 841 | 868 |
if debugDecoder {
|
| 842 | 869 |
println("New frame...")
|
| ... | ... |
@@ -863,7 +859,7 @@ decodeStream: |
| 863 | 863 |
case <-ctx.Done(): |
| 864 | 864 |
case dec := <-d.decoders: |
| 865 | 865 |
dec.sendErr(err) |
| 866 |
- seqPrepare <- dec |
|
| 866 |
+ decodeBlock(dec) |
|
| 867 | 867 |
} |
| 868 | 868 |
break decodeStream |
| 869 | 869 |
} |
| ... | ... |
@@ -883,6 +879,10 @@ decodeStream: |
| 883 | 883 |
if debugDecoder {
|
| 884 | 884 |
println("Alloc History:", h.allocFrameBuffer)
|
| 885 | 885 |
} |
| 886 |
+ hist.reset() |
|
| 887 |
+ if h.dict != nil {
|
|
| 888 |
+ hist.setDict(h.dict) |
|
| 889 |
+ } |
|
| 886 | 890 |
dec.async.newHist = &h |
| 887 | 891 |
dec.async.fcs = frame.FrameContentSize |
| 888 | 892 |
historySent = true |
| ... | ... |
@@ -909,7 +909,7 @@ decodeStream: |
| 909 | 909 |
} |
| 910 | 910 |
err = dec.err |
| 911 | 911 |
last := dec.Last |
| 912 |
- seqPrepare <- dec |
|
| 912 |
+ decodeBlock(dec) |
|
| 913 | 913 |
if err != nil {
|
| 914 | 914 |
break decodeStream |
| 915 | 915 |
} |
| ... | ... |
@@ -918,7 +918,7 @@ decodeStream: |
| 918 | 918 |
} |
| 919 | 919 |
} |
| 920 | 920 |
} |
| 921 |
- close(seqPrepare) |
|
| 921 |
+ close(seqDecode) |
|
| 922 | 922 |
wg.Wait() |
| 923 | 923 |
d.frame.history.b = frameHistCache |
| 924 | 924 |
} |
| ... | ... |
@@ -19,6 +19,7 @@ type decoderOptions struct {
|
| 19 | 19 |
maxDecodedSize uint64 |
| 20 | 20 |
maxWindowSize uint64 |
| 21 | 21 |
dicts []dict |
| 22 |
+ ignoreChecksum bool |
|
| 22 | 23 |
} |
| 23 | 24 |
|
| 24 | 25 |
func (o *decoderOptions) setDefault() {
|
| ... | ... |
@@ -31,7 +32,7 @@ func (o *decoderOptions) setDefault() {
|
| 31 | 31 |
if o.concurrent > 4 {
|
| 32 | 32 |
o.concurrent = 4 |
| 33 | 33 |
} |
| 34 |
- o.maxDecodedSize = 1 << 63 |
|
| 34 |
+ o.maxDecodedSize = 64 << 30 |
|
| 35 | 35 |
} |
| 36 | 36 |
|
| 37 | 37 |
// WithDecoderLowmem will set whether to use a lower amount of memory, |
| ... | ... |
@@ -66,7 +67,7 @@ func WithDecoderConcurrency(n int) DOption {
|
| 66 | 66 |
// WithDecoderMaxMemory allows to set a maximum decoded size for in-memory |
| 67 | 67 |
// non-streaming operations or maximum window size for streaming operations. |
| 68 | 68 |
// This can be used to control memory usage of potentially hostile content. |
| 69 |
-// Maximum and default is 1 << 63 bytes. |
|
| 69 |
+// Maximum is 1 << 63 bytes. Default is 64GiB. |
|
| 70 | 70 |
func WithDecoderMaxMemory(n uint64) DOption {
|
| 71 | 71 |
return func(o *decoderOptions) error {
|
| 72 | 72 |
if n == 0 {
|
| ... | ... |
@@ -112,3 +113,11 @@ func WithDecoderMaxWindow(size uint64) DOption {
|
| 112 | 112 |
return nil |
| 113 | 113 |
} |
| 114 | 114 |
} |
| 115 |
+ |
|
| 116 |
+// IgnoreChecksum allows to forcibly ignore checksum checking. |
|
| 117 |
+func IgnoreChecksum(b bool) DOption {
|
|
| 118 |
+ return func(o *decoderOptions) error {
|
|
| 119 |
+ o.ignoreChecksum = b |
|
| 120 |
+ return nil |
|
| 121 |
+ } |
|
| 122 |
+} |
| ... | ... |
@@ -156,8 +156,8 @@ encodeLoop: |
| 156 | 156 |
panic("offset0 was 0")
|
| 157 | 157 |
} |
| 158 | 158 |
|
| 159 |
- nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 160 | 159 |
nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) |
| 160 |
+ nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 161 | 161 |
candidateL := e.longTable[nextHashL] |
| 162 | 162 |
candidateS := e.table[nextHashS] |
| 163 | 163 |
|
| ... | ... |
@@ -518,8 +518,8 @@ encodeLoop: |
| 518 | 518 |
} |
| 519 | 519 |
|
| 520 | 520 |
// Store this, since we have it. |
| 521 |
- nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 522 | 521 |
nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) |
| 522 |
+ nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 523 | 523 |
|
| 524 | 524 |
// We have at least 4 byte match. |
| 525 | 525 |
// No need to check backwards. We come straight from a match |
| ... | ... |
@@ -674,8 +674,8 @@ encodeLoop: |
| 674 | 674 |
panic("offset0 was 0")
|
| 675 | 675 |
} |
| 676 | 676 |
|
| 677 |
- nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 678 | 677 |
nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) |
| 678 |
+ nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 679 | 679 |
candidateL := e.longTable[nextHashL] |
| 680 | 680 |
candidateS := e.table[nextHashS] |
| 681 | 681 |
|
| ... | ... |
@@ -1047,8 +1047,8 @@ encodeLoop: |
| 1047 | 1047 |
} |
| 1048 | 1048 |
|
| 1049 | 1049 |
// Store this, since we have it. |
| 1050 |
- nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 1051 | 1050 |
nextHashL := hashLen(cv, betterLongTableBits, betterLongLen) |
| 1051 |
+ nextHashS := hashLen(cv, betterShortTableBits, betterShortLen) |
|
| 1052 | 1052 |
|
| 1053 | 1053 |
// We have at least 4 byte match. |
| 1054 | 1054 |
// No need to check backwards. We come straight from a match |
| ... | ... |
@@ -127,8 +127,8 @@ encodeLoop: |
| 127 | 127 |
panic("offset0 was 0")
|
| 128 | 128 |
} |
| 129 | 129 |
|
| 130 |
- nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 131 | 130 |
nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) |
| 131 |
+ nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 132 | 132 |
candidateL := e.longTable[nextHashL] |
| 133 | 133 |
candidateS := e.table[nextHashS] |
| 134 | 134 |
|
| ... | ... |
@@ -439,8 +439,8 @@ encodeLoop: |
| 439 | 439 |
var t int32 |
| 440 | 440 |
for {
|
| 441 | 441 |
|
| 442 |
- nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 443 | 442 |
nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) |
| 443 |
+ nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 444 | 444 |
candidateL := e.longTable[nextHashL] |
| 445 | 445 |
candidateS := e.table[nextHashS] |
| 446 | 446 |
|
| ... | ... |
@@ -785,8 +785,8 @@ encodeLoop: |
| 785 | 785 |
panic("offset0 was 0")
|
| 786 | 786 |
} |
| 787 | 787 |
|
| 788 |
- nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 789 | 788 |
nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) |
| 789 |
+ nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 790 | 790 |
candidateL := e.longTable[nextHashL] |
| 791 | 791 |
candidateS := e.table[nextHashS] |
| 792 | 792 |
|
| ... | ... |
@@ -969,7 +969,7 @@ encodeLoop: |
| 969 | 969 |
te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
|
| 970 | 970 |
te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
|
| 971 | 971 |
longHash1 := hashLen(cv0, dFastLongTableBits, dFastLongLen) |
| 972 |
- longHash2 := hashLen(cv0, dFastLongTableBits, dFastLongLen) |
|
| 972 |
+ longHash2 := hashLen(cv1, dFastLongTableBits, dFastLongLen) |
|
| 973 | 973 |
e.longTable[longHash1] = te0 |
| 974 | 974 |
e.longTable[longHash2] = te1 |
| 975 | 975 |
e.markLongShardDirty(longHash1) |
| ... | ... |
@@ -1002,8 +1002,8 @@ encodeLoop: |
| 1002 | 1002 |
} |
| 1003 | 1003 |
|
| 1004 | 1004 |
// Store this, since we have it. |
| 1005 |
- nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 1006 | 1005 |
nextHashL := hashLen(cv, dFastLongTableBits, dFastLongLen) |
| 1006 |
+ nextHashS := hashLen(cv, dFastShortTableBits, dFastShortLen) |
|
| 1007 | 1007 |
|
| 1008 | 1008 |
// We have at least 4 byte match. |
| 1009 | 1009 |
// No need to check backwards. We come straight from a match |
| ... | ... |
@@ -528,8 +528,8 @@ func (e *Encoder) EncodeAll(src, dst []byte) []byte {
|
| 528 | 528 |
// If a non-single block is needed the encoder will reset again. |
| 529 | 529 |
e.encoders <- enc |
| 530 | 530 |
}() |
| 531 |
- // Use single segments when above minimum window and below 1MB. |
|
| 532 |
- single := len(src) < 1<<20 && len(src) > MinWindowSize |
|
| 531 |
+ // Use single segments when above minimum window and below window size. |
|
| 532 |
+ single := len(src) <= e.o.windowSize && len(src) > MinWindowSize |
|
| 533 | 533 |
if e.o.single != nil {
|
| 534 | 534 |
single = *e.o.single |
| 535 | 535 |
} |
| ... | ... |
@@ -551,7 +551,7 @@ func (e *Encoder) EncodeAll(src, dst []byte) []byte {
|
| 551 | 551 |
} |
| 552 | 552 |
|
| 553 | 553 |
// If we can do everything in one block, prefer that. |
| 554 |
- if len(src) <= maxCompressedBlockSize {
|
|
| 554 |
+ if len(src) <= e.o.blockSize {
|
|
| 555 | 555 |
enc.Reset(e.o.dict, true) |
| 556 | 556 |
// Slightly faster with no history and everything in one block. |
| 557 | 557 |
if e.o.crc {
|
| ... | ... |
@@ -283,7 +283,7 @@ func WithNoEntropyCompression(b bool) EOption {
|
| 283 | 283 |
// a decoder is allowed to reject a compressed frame which requests a memory size beyond decoder's authorized range. |
| 284 | 284 |
// For broader compatibility, decoders are recommended to support memory sizes of at least 8 MB. |
| 285 | 285 |
// This is only a recommendation, each decoder is free to support higher or lower limits, depending on local limitations. |
| 286 |
-// If this is not specified, block encodes will automatically choose this based on the input size. |
|
| 286 |
+// If this is not specified, block encodes will automatically choose this based on the input size and the window size. |
|
| 287 | 287 |
// This setting has no effect on streamed encodes. |
| 288 | 288 |
func WithSingleSegment(b bool) EOption {
|
| 289 | 289 |
return func(o *encoderOptions) error {
|
| ... | ... |
@@ -106,7 +106,7 @@ func (d *frameDec) reset(br byteBuffer) error {
|
| 106 | 106 |
} |
| 107 | 107 |
n := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24) |
| 108 | 108 |
println("Skipping frame with", n, "bytes.")
|
| 109 |
- err = br.skipN(int(n)) |
|
| 109 |
+ err = br.skipN(int64(n)) |
|
| 110 | 110 |
if err != nil {
|
| 111 | 111 |
if debugDecoder {
|
| 112 | 112 |
println("Reading discarded frame", err)
|
| ... | ... |
@@ -231,20 +231,27 @@ func (d *frameDec) reset(br byteBuffer) error {
|
| 231 | 231 |
d.crc.Reset() |
| 232 | 232 |
} |
| 233 | 233 |
|
| 234 |
+ if d.WindowSize > d.o.maxWindowSize {
|
|
| 235 |
+ if debugDecoder {
|
|
| 236 |
+ printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize)
|
|
| 237 |
+ } |
|
| 238 |
+ return ErrWindowSizeExceeded |
|
| 239 |
+ } |
|
| 240 |
+ |
|
| 234 | 241 |
if d.WindowSize == 0 && d.SingleSegment {
|
| 235 | 242 |
// We may not need window in this case. |
| 236 | 243 |
d.WindowSize = d.FrameContentSize |
| 237 | 244 |
if d.WindowSize < MinWindowSize {
|
| 238 | 245 |
d.WindowSize = MinWindowSize |
| 239 | 246 |
} |
| 240 |
- } |
|
| 241 |
- |
|
| 242 |
- if d.WindowSize > uint64(d.o.maxWindowSize) {
|
|
| 243 |
- if debugDecoder {
|
|
| 244 |
- printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize)
|
|
| 247 |
+ if d.WindowSize > d.o.maxDecodedSize {
|
|
| 248 |
+ if debugDecoder {
|
|
| 249 |
+ printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize)
|
|
| 250 |
+ } |
|
| 251 |
+ return ErrDecoderSizeExceeded |
|
| 245 | 252 |
} |
| 246 |
- return ErrWindowSizeExceeded |
|
| 247 | 253 |
} |
| 254 |
+ |
|
| 248 | 255 |
// The minimum Window_Size is 1 KB. |
| 249 | 256 |
if d.WindowSize < MinWindowSize {
|
| 250 | 257 |
if debugDecoder {
|
| ... | ... |
@@ -253,10 +260,11 @@ func (d *frameDec) reset(br byteBuffer) error {
|
| 253 | 253 |
return ErrWindowSizeTooSmall |
| 254 | 254 |
} |
| 255 | 255 |
d.history.windowSize = int(d.WindowSize) |
| 256 |
- if d.o.lowMem && d.history.windowSize < maxBlockSize {
|
|
| 256 |
+ if !d.o.lowMem || d.history.windowSize < maxBlockSize {
|
|
| 257 |
+ // Alloc 2x window size if not low-mem, or very small window size. |
|
| 257 | 258 |
d.history.allocFrameBuffer = d.history.windowSize * 2 |
| 258 |
- // TODO: Maybe use FrameContent size |
|
| 259 | 259 |
} else {
|
| 260 |
+ // Alloc with one additional block |
|
| 260 | 261 |
d.history.allocFrameBuffer = d.history.windowSize + maxBlockSize |
| 261 | 262 |
} |
| 262 | 263 |
|
| ... | ... |
@@ -290,13 +298,6 @@ func (d *frameDec) checkCRC() error {
|
| 290 | 290 |
if !d.HasCheckSum {
|
| 291 | 291 |
return nil |
| 292 | 292 |
} |
| 293 |
- var tmp [4]byte |
|
| 294 |
- got := d.crc.Sum64() |
|
| 295 |
- // Flip to match file order. |
|
| 296 |
- tmp[0] = byte(got >> 0) |
|
| 297 |
- tmp[1] = byte(got >> 8) |
|
| 298 |
- tmp[2] = byte(got >> 16) |
|
| 299 |
- tmp[3] = byte(got >> 24) |
|
| 300 | 293 |
|
| 301 | 294 |
// We can overwrite upper tmp now |
| 302 | 295 |
want, err := d.rawInput.readSmall(4) |
| ... | ... |
@@ -305,7 +306,19 @@ func (d *frameDec) checkCRC() error {
|
| 305 | 305 |
return err |
| 306 | 306 |
} |
| 307 | 307 |
|
| 308 |
- if !bytes.Equal(tmp[:], want) && !ignoreCRC {
|
|
| 308 |
+ if d.o.ignoreChecksum {
|
|
| 309 |
+ return nil |
|
| 310 |
+ } |
|
| 311 |
+ |
|
| 312 |
+ var tmp [4]byte |
|
| 313 |
+ got := d.crc.Sum64() |
|
| 314 |
+ // Flip to match file order. |
|
| 315 |
+ tmp[0] = byte(got >> 0) |
|
| 316 |
+ tmp[1] = byte(got >> 8) |
|
| 317 |
+ tmp[2] = byte(got >> 16) |
|
| 318 |
+ tmp[3] = byte(got >> 24) |
|
| 319 |
+ |
|
| 320 |
+ if !bytes.Equal(tmp[:], want) {
|
|
| 309 | 321 |
if debugDecoder {
|
| 310 | 322 |
println("CRC Check Failed:", tmp[:], "!=", want)
|
| 311 | 323 |
} |
| ... | ... |
@@ -317,6 +330,19 @@ func (d *frameDec) checkCRC() error {
|
| 317 | 317 |
return nil |
| 318 | 318 |
} |
| 319 | 319 |
|
| 320 |
+// consumeCRC reads the checksum data if the frame has one. |
|
| 321 |
+func (d *frameDec) consumeCRC() error {
|
|
| 322 |
+ if d.HasCheckSum {
|
|
| 323 |
+ _, err := d.rawInput.readSmall(4) |
|
| 324 |
+ if err != nil {
|
|
| 325 |
+ println("CRC missing?", err)
|
|
| 326 |
+ return err |
|
| 327 |
+ } |
|
| 328 |
+ } |
|
| 329 |
+ |
|
| 330 |
+ return nil |
|
| 331 |
+} |
|
| 332 |
+ |
|
| 320 | 333 |
// runDecoder will create a sync decoder that will decode a block of data. |
| 321 | 334 |
func (d *frameDec) runDecoder(dst []byte, dec *blockDec) ([]byte, error) {
|
| 322 | 335 |
saved := d.history.b |
| ... | ... |
@@ -326,6 +352,19 @@ func (d *frameDec) runDecoder(dst []byte, dec *blockDec) ([]byte, error) {
|
| 326 | 326 |
d.history.ignoreBuffer = len(dst) |
| 327 | 327 |
// Store input length, so we only check new data. |
| 328 | 328 |
crcStart := len(dst) |
| 329 |
+ d.history.decoders.maxSyncLen = 0 |
|
| 330 |
+ if d.FrameContentSize != fcsUnknown {
|
|
| 331 |
+ d.history.decoders.maxSyncLen = d.FrameContentSize + uint64(len(dst)) |
|
| 332 |
+ if d.history.decoders.maxSyncLen > d.o.maxDecodedSize {
|
|
| 333 |
+ return dst, ErrDecoderSizeExceeded |
|
| 334 |
+ } |
|
| 335 |
+ if uint64(cap(dst)) < d.history.decoders.maxSyncLen {
|
|
| 336 |
+ // Alloc for output |
|
| 337 |
+ dst2 := make([]byte, len(dst), d.history.decoders.maxSyncLen+compressedBlockOverAlloc) |
|
| 338 |
+ copy(dst2, dst) |
|
| 339 |
+ dst = dst2 |
|
| 340 |
+ } |
|
| 341 |
+ } |
|
| 329 | 342 |
var err error |
| 330 | 343 |
for {
|
| 331 | 344 |
err = dec.reset(d.rawInput, d.WindowSize) |
| ... | ... |
@@ -360,13 +399,17 @@ func (d *frameDec) runDecoder(dst []byte, dec *blockDec) ([]byte, error) {
|
| 360 | 360 |
if d.FrameContentSize != fcsUnknown && uint64(len(d.history.b)-crcStart) != d.FrameContentSize {
|
| 361 | 361 |
err = ErrFrameSizeMismatch |
| 362 | 362 |
} else if d.HasCheckSum {
|
| 363 |
- var n int |
|
| 364 |
- n, err = d.crc.Write(dst[crcStart:]) |
|
| 365 |
- if err == nil {
|
|
| 366 |
- if n != len(dst)-crcStart {
|
|
| 367 |
- err = io.ErrShortWrite |
|
| 368 |
- } else {
|
|
| 369 |
- err = d.checkCRC() |
|
| 363 |
+ if d.o.ignoreChecksum {
|
|
| 364 |
+ err = d.consumeCRC() |
|
| 365 |
+ } else {
|
|
| 366 |
+ var n int |
|
| 367 |
+ n, err = d.crc.Write(dst[crcStart:]) |
|
| 368 |
+ if err == nil {
|
|
| 369 |
+ if n != len(dst)-crcStart {
|
|
| 370 |
+ err = io.ErrShortWrite |
|
| 371 |
+ } else {
|
|
| 372 |
+ err = d.checkCRC() |
|
| 373 |
+ } |
|
| 370 | 374 |
} |
| 371 | 375 |
} |
| 372 | 376 |
} |
| ... | ... |
@@ -5,8 +5,10 @@ |
| 5 | 5 |
package zstd |
| 6 | 6 |
|
| 7 | 7 |
import ( |
| 8 |
+ "encoding/binary" |
|
| 8 | 9 |
"errors" |
| 9 | 10 |
"fmt" |
| 11 |
+ "io" |
|
| 10 | 12 |
) |
| 11 | 13 |
|
| 12 | 14 |
const ( |
| ... | ... |
@@ -178,10 +180,32 @@ func (s *fseDecoder) readNCount(b *byteReader, maxSymbol uint16) error {
|
| 178 | 178 |
return fmt.Errorf("corruption detected (total %d != %d)", gotTotal, 1<<s.actualTableLog)
|
| 179 | 179 |
} |
| 180 | 180 |
b.advance((bitCount + 7) >> 3) |
| 181 |
- // println(s.norm[:s.symbolLen], s.symbolLen) |
|
| 182 | 181 |
return s.buildDtable() |
| 183 | 182 |
} |
| 184 | 183 |
|
| 184 |
+func (s *fseDecoder) mustReadFrom(r io.Reader) {
|
|
| 185 |
+ fatalErr := func(err error) {
|
|
| 186 |
+ if err != nil {
|
|
| 187 |
+ panic(err) |
|
| 188 |
+ } |
|
| 189 |
+ } |
|
| 190 |
+ // dt [maxTablesize]decSymbol // Decompression table. |
|
| 191 |
+ // symbolLen uint16 // Length of active part of the symbol table. |
|
| 192 |
+ // actualTableLog uint8 // Selected tablelog. |
|
| 193 |
+ // maxBits uint8 // Maximum number of additional bits |
|
| 194 |
+ // // used for table creation to avoid allocations. |
|
| 195 |
+ // stateTable [256]uint16 |
|
| 196 |
+ // norm [maxSymbolValue + 1]int16 |
|
| 197 |
+ // preDefined bool |
|
| 198 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.dt)) |
|
| 199 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.symbolLen)) |
|
| 200 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.actualTableLog)) |
|
| 201 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.maxBits)) |
|
| 202 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.stateTable)) |
|
| 203 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.norm)) |
|
| 204 |
+ fatalErr(binary.Read(r, binary.LittleEndian, &s.preDefined)) |
|
| 205 |
+} |
|
| 206 |
+ |
|
| 185 | 207 |
// decSymbol contains information about a state entry, |
| 186 | 208 |
// Including the state offset base, the output symbol and |
| 187 | 209 |
// the number of bits to read for the low part of the destination state. |
| ... | ... |
@@ -204,18 +228,10 @@ func (d decSymbol) newState() uint16 {
|
| 204 | 204 |
return uint16(d >> 16) |
| 205 | 205 |
} |
| 206 | 206 |
|
| 207 |
-func (d decSymbol) baseline() uint32 {
|
|
| 208 |
- return uint32(d >> 32) |
|
| 209 |
-} |
|
| 210 |
- |
|
| 211 | 207 |
func (d decSymbol) baselineInt() int {
|
| 212 | 208 |
return int(d >> 32) |
| 213 | 209 |
} |
| 214 | 210 |
|
| 215 |
-func (d *decSymbol) set(nbits, addBits uint8, newState uint16, baseline uint32) {
|
|
| 216 |
- *d = decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32) |
|
| 217 |
-} |
|
| 218 |
- |
|
| 219 | 211 |
func (d *decSymbol) setNBits(nBits uint8) {
|
| 220 | 212 |
const mask = 0xffffffffffffff00 |
| 221 | 213 |
*d = (*d & mask) | decSymbol(nBits) |
| ... | ... |
@@ -231,11 +247,6 @@ func (d *decSymbol) setNewState(state uint16) {
|
| 231 | 231 |
*d = (*d & mask) | decSymbol(state)<<16 |
| 232 | 232 |
} |
| 233 | 233 |
|
| 234 |
-func (d *decSymbol) setBaseline(baseline uint32) {
|
|
| 235 |
- const mask = 0xffffffff |
|
| 236 |
- *d = (*d & mask) | decSymbol(baseline)<<32 |
|
| 237 |
-} |
|
| 238 |
- |
|
| 239 | 234 |
func (d *decSymbol) setExt(addBits uint8, baseline uint32) {
|
| 240 | 235 |
const mask = 0xffff00ff |
| 241 | 236 |
*d = (*d & mask) | (decSymbol(addBits) << 8) | (decSymbol(baseline) << 32) |
| ... | ... |
@@ -257,68 +268,6 @@ func (s *fseDecoder) setRLE(symbol decSymbol) {
|
| 257 | 257 |
s.dt[0] = symbol |
| 258 | 258 |
} |
| 259 | 259 |
|
| 260 |
-// buildDtable will build the decoding table. |
|
| 261 |
-func (s *fseDecoder) buildDtable() error {
|
|
| 262 |
- tableSize := uint32(1 << s.actualTableLog) |
|
| 263 |
- highThreshold := tableSize - 1 |
|
| 264 |
- symbolNext := s.stateTable[:256] |
|
| 265 |
- |
|
| 266 |
- // Init, lay down lowprob symbols |
|
| 267 |
- {
|
|
| 268 |
- for i, v := range s.norm[:s.symbolLen] {
|
|
| 269 |
- if v == -1 {
|
|
| 270 |
- s.dt[highThreshold].setAddBits(uint8(i)) |
|
| 271 |
- highThreshold-- |
|
| 272 |
- symbolNext[i] = 1 |
|
| 273 |
- } else {
|
|
| 274 |
- symbolNext[i] = uint16(v) |
|
| 275 |
- } |
|
| 276 |
- } |
|
| 277 |
- } |
|
| 278 |
- // Spread symbols |
|
| 279 |
- {
|
|
| 280 |
- tableMask := tableSize - 1 |
|
| 281 |
- step := tableStep(tableSize) |
|
| 282 |
- position := uint32(0) |
|
| 283 |
- for ss, v := range s.norm[:s.symbolLen] {
|
|
| 284 |
- for i := 0; i < int(v); i++ {
|
|
| 285 |
- s.dt[position].setAddBits(uint8(ss)) |
|
| 286 |
- position = (position + step) & tableMask |
|
| 287 |
- for position > highThreshold {
|
|
| 288 |
- // lowprob area |
|
| 289 |
- position = (position + step) & tableMask |
|
| 290 |
- } |
|
| 291 |
- } |
|
| 292 |
- } |
|
| 293 |
- if position != 0 {
|
|
| 294 |
- // position must reach all cells once, otherwise normalizedCounter is incorrect |
|
| 295 |
- return errors.New("corrupted input (position != 0)")
|
|
| 296 |
- } |
|
| 297 |
- } |
|
| 298 |
- |
|
| 299 |
- // Build Decoding table |
|
| 300 |
- {
|
|
| 301 |
- tableSize := uint16(1 << s.actualTableLog) |
|
| 302 |
- for u, v := range s.dt[:tableSize] {
|
|
| 303 |
- symbol := v.addBits() |
|
| 304 |
- nextState := symbolNext[symbol] |
|
| 305 |
- symbolNext[symbol] = nextState + 1 |
|
| 306 |
- nBits := s.actualTableLog - byte(highBits(uint32(nextState))) |
|
| 307 |
- s.dt[u&maxTableMask].setNBits(nBits) |
|
| 308 |
- newState := (nextState << nBits) - tableSize |
|
| 309 |
- if newState > tableSize {
|
|
| 310 |
- return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
|
|
| 311 |
- } |
|
| 312 |
- if newState == uint16(u) && nBits == 0 {
|
|
| 313 |
- // Seems weird that this is possible with nbits > 0. |
|
| 314 |
- return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
|
|
| 315 |
- } |
|
| 316 |
- s.dt[u&maxTableMask].setNewState(newState) |
|
| 317 |
- } |
|
| 318 |
- } |
|
| 319 |
- return nil |
|
| 320 |
-} |
|
| 321 |
- |
|
| 322 | 260 |
// transform will transform the decoder table into a table usable for |
| 323 | 261 |
// decoding without having to apply the transformation while decoding. |
| 324 | 262 |
// The state will contain the base value and the number of bits to read. |
| ... | ... |
@@ -352,34 +301,7 @@ func (s *fseState) init(br *bitReader, tableLog uint8, dt []decSymbol) {
|
| 352 | 352 |
s.state = dt[br.getBits(tableLog)] |
| 353 | 353 |
} |
| 354 | 354 |
|
| 355 |
-// next returns the current symbol and sets the next state. |
|
| 356 |
-// At least tablelog bits must be available in the bit reader. |
|
| 357 |
-func (s *fseState) next(br *bitReader) {
|
|
| 358 |
- lowBits := uint16(br.getBits(s.state.nbBits())) |
|
| 359 |
- s.state = s.dt[s.state.newState()+lowBits] |
|
| 360 |
-} |
|
| 361 |
- |
|
| 362 |
-// finished returns true if all bits have been read from the bitstream |
|
| 363 |
-// and the next state would require reading bits from the input. |
|
| 364 |
-func (s *fseState) finished(br *bitReader) bool {
|
|
| 365 |
- return br.finished() && s.state.nbBits() > 0 |
|
| 366 |
-} |
|
| 367 |
- |
|
| 368 |
-// final returns the current state symbol without decoding the next. |
|
| 369 |
-func (s *fseState) final() (int, uint8) {
|
|
| 370 |
- return s.state.baselineInt(), s.state.addBits() |
|
| 371 |
-} |
|
| 372 |
- |
|
| 373 | 355 |
// final returns the current state symbol without decoding the next. |
| 374 | 356 |
func (s decSymbol) final() (int, uint8) {
|
| 375 | 357 |
return s.baselineInt(), s.addBits() |
| 376 | 358 |
} |
| 377 |
- |
|
| 378 |
-// nextFast returns the next symbol and sets the next state. |
|
| 379 |
-// This can only be used if no symbols are 0 bits. |
|
| 380 |
-// At least tablelog bits must be available in the bit reader. |
|
| 381 |
-func (s *fseState) nextFast(br *bitReader) (uint32, uint8) {
|
|
| 382 |
- lowBits := br.get16BitsFast(s.state.nbBits()) |
|
| 383 |
- s.state = s.dt[s.state.newState()+lowBits] |
|
| 384 |
- return s.state.baseline(), s.state.addBits() |
|
| 385 |
-} |
| 386 | 359 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,64 @@ |
| 0 |
+//go:build amd64 && !appengine && !noasm && gc |
|
| 1 |
+// +build amd64,!appengine,!noasm,gc |
|
| 2 |
+ |
|
| 3 |
+package zstd |
|
| 4 |
+ |
|
| 5 |
+import ( |
|
| 6 |
+ "fmt" |
|
| 7 |
+) |
|
| 8 |
+ |
|
| 9 |
+type buildDtableAsmContext struct {
|
|
| 10 |
+ // inputs |
|
| 11 |
+ stateTable *uint16 |
|
| 12 |
+ norm *int16 |
|
| 13 |
+ dt *uint64 |
|
| 14 |
+ |
|
| 15 |
+ // outputs --- set by the procedure in the case of error; |
|
| 16 |
+ // for interpretation please see the error handling part below |
|
| 17 |
+ errParam1 uint64 |
|
| 18 |
+ errParam2 uint64 |
|
| 19 |
+} |
|
| 20 |
+ |
|
| 21 |
+// buildDtable_asm is an x86 assembly implementation of fseDecoder.buildDtable. |
|
| 22 |
+// Function returns non-zero exit code on error. |
|
| 23 |
+// go:noescape |
|
| 24 |
+func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int |
|
| 25 |
+ |
|
| 26 |
+// please keep in sync with _generate/gen_fse.go |
|
| 27 |
+const ( |
|
| 28 |
+ errorCorruptedNormalizedCounter = 1 |
|
| 29 |
+ errorNewStateTooBig = 2 |
|
| 30 |
+ errorNewStateNoBits = 3 |
|
| 31 |
+) |
|
| 32 |
+ |
|
| 33 |
+// buildDtable will build the decoding table. |
|
| 34 |
+func (s *fseDecoder) buildDtable() error {
|
|
| 35 |
+ ctx := buildDtableAsmContext{
|
|
| 36 |
+ stateTable: &s.stateTable[0], |
|
| 37 |
+ norm: &s.norm[0], |
|
| 38 |
+ dt: (*uint64)(&s.dt[0]), |
|
| 39 |
+ } |
|
| 40 |
+ code := buildDtable_asm(s, &ctx) |
|
| 41 |
+ |
|
| 42 |
+ if code != 0 {
|
|
| 43 |
+ switch code {
|
|
| 44 |
+ case errorCorruptedNormalizedCounter: |
|
| 45 |
+ position := ctx.errParam1 |
|
| 46 |
+ return fmt.Errorf("corrupted input (position=%d, expected 0)", position)
|
|
| 47 |
+ |
|
| 48 |
+ case errorNewStateTooBig: |
|
| 49 |
+ newState := decSymbol(ctx.errParam1) |
|
| 50 |
+ size := ctx.errParam2 |
|
| 51 |
+ return fmt.Errorf("newState (%d) outside table size (%d)", newState, size)
|
|
| 52 |
+ |
|
| 53 |
+ case errorNewStateNoBits: |
|
| 54 |
+ newState := decSymbol(ctx.errParam1) |
|
| 55 |
+ oldState := decSymbol(ctx.errParam2) |
|
| 56 |
+ return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, oldState)
|
|
| 57 |
+ |
|
| 58 |
+ default: |
|
| 59 |
+ return fmt.Errorf("buildDtable_asm returned unhandled nonzero code = %d", code)
|
|
| 60 |
+ } |
|
| 61 |
+ } |
|
| 62 |
+ return nil |
|
| 63 |
+} |
| 0 | 64 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,127 @@ |
| 0 |
+// Code generated by command: go run gen_fse.go -out ../fse_decoder_amd64.s -pkg=zstd. DO NOT EDIT. |
|
| 1 |
+ |
|
| 2 |
+//go:build !appengine && !noasm && gc && !noasm |
|
| 3 |
+// +build !appengine,!noasm,gc,!noasm |
|
| 4 |
+ |
|
| 5 |
+// func buildDtable_asm(s *fseDecoder, ctx *buildDtableAsmContext) int |
|
| 6 |
+TEXT ·buildDtable_asm(SB), $0-24 |
|
| 7 |
+ MOVQ ctx+8(FP), CX |
|
| 8 |
+ MOVQ s+0(FP), DI |
|
| 9 |
+ |
|
| 10 |
+ // Load values |
|
| 11 |
+ MOVBQZX 4098(DI), DX |
|
| 12 |
+ XORQ AX, AX |
|
| 13 |
+ BTSQ DX, AX |
|
| 14 |
+ MOVQ (CX), BX |
|
| 15 |
+ MOVQ 16(CX), SI |
|
| 16 |
+ LEAQ -1(AX), R8 |
|
| 17 |
+ MOVQ 8(CX), CX |
|
| 18 |
+ MOVWQZX 4096(DI), DI |
|
| 19 |
+ |
|
| 20 |
+ // End load values |
|
| 21 |
+ // Init, lay down lowprob symbols |
|
| 22 |
+ XORQ R9, R9 |
|
| 23 |
+ JMP init_main_loop_condition |
|
| 24 |
+ |
|
| 25 |
+init_main_loop: |
|
| 26 |
+ MOVWQSX (CX)(R9*2), R10 |
|
| 27 |
+ CMPW R10, $-1 |
|
| 28 |
+ JNE do_not_update_high_threshold |
|
| 29 |
+ MOVB R9, 1(SI)(R8*8) |
|
| 30 |
+ DECQ R8 |
|
| 31 |
+ MOVQ $0x0000000000000001, R10 |
|
| 32 |
+ |
|
| 33 |
+do_not_update_high_threshold: |
|
| 34 |
+ MOVW R10, (BX)(R9*2) |
|
| 35 |
+ INCQ R9 |
|
| 36 |
+ |
|
| 37 |
+init_main_loop_condition: |
|
| 38 |
+ CMPQ R9, DI |
|
| 39 |
+ JL init_main_loop |
|
| 40 |
+ |
|
| 41 |
+ // Spread symbols |
|
| 42 |
+ // Calculate table step |
|
| 43 |
+ MOVQ AX, R9 |
|
| 44 |
+ SHRQ $0x01, R9 |
|
| 45 |
+ MOVQ AX, R10 |
|
| 46 |
+ SHRQ $0x03, R10 |
|
| 47 |
+ LEAQ 3(R9)(R10*1), R9 |
|
| 48 |
+ |
|
| 49 |
+ // Fill add bits values |
|
| 50 |
+ LEAQ -1(AX), R10 |
|
| 51 |
+ XORQ R11, R11 |
|
| 52 |
+ XORQ R12, R12 |
|
| 53 |
+ JMP spread_main_loop_condition |
|
| 54 |
+ |
|
| 55 |
+spread_main_loop: |
|
| 56 |
+ XORQ R13, R13 |
|
| 57 |
+ MOVWQSX (CX)(R12*2), R14 |
|
| 58 |
+ JMP spread_inner_loop_condition |
|
| 59 |
+ |
|
| 60 |
+spread_inner_loop: |
|
| 61 |
+ MOVB R12, 1(SI)(R11*8) |
|
| 62 |
+ |
|
| 63 |
+adjust_position: |
|
| 64 |
+ ADDQ R9, R11 |
|
| 65 |
+ ANDQ R10, R11 |
|
| 66 |
+ CMPQ R11, R8 |
|
| 67 |
+ JG adjust_position |
|
| 68 |
+ INCQ R13 |
|
| 69 |
+ |
|
| 70 |
+spread_inner_loop_condition: |
|
| 71 |
+ CMPQ R13, R14 |
|
| 72 |
+ JL spread_inner_loop |
|
| 73 |
+ INCQ R12 |
|
| 74 |
+ |
|
| 75 |
+spread_main_loop_condition: |
|
| 76 |
+ CMPQ R12, DI |
|
| 77 |
+ JL spread_main_loop |
|
| 78 |
+ TESTQ R11, R11 |
|
| 79 |
+ JZ spread_check_ok |
|
| 80 |
+ MOVQ ctx+8(FP), AX |
|
| 81 |
+ MOVQ R11, 24(AX) |
|
| 82 |
+ MOVQ $+1, ret+16(FP) |
|
| 83 |
+ RET |
|
| 84 |
+ |
|
| 85 |
+spread_check_ok: |
|
| 86 |
+ // Build Decoding table |
|
| 87 |
+ XORQ DI, DI |
|
| 88 |
+ |
|
| 89 |
+build_table_main_table: |
|
| 90 |
+ MOVBQZX 1(SI)(DI*8), CX |
|
| 91 |
+ MOVWQZX (BX)(CX*2), R8 |
|
| 92 |
+ LEAQ 1(R8), R9 |
|
| 93 |
+ MOVW R9, (BX)(CX*2) |
|
| 94 |
+ MOVQ R8, R9 |
|
| 95 |
+ BSRQ R9, R9 |
|
| 96 |
+ MOVQ DX, CX |
|
| 97 |
+ SUBQ R9, CX |
|
| 98 |
+ SHLQ CL, R8 |
|
| 99 |
+ SUBQ AX, R8 |
|
| 100 |
+ MOVB CL, (SI)(DI*8) |
|
| 101 |
+ MOVW R8, 2(SI)(DI*8) |
|
| 102 |
+ CMPQ R8, AX |
|
| 103 |
+ JLE build_table_check1_ok |
|
| 104 |
+ MOVQ ctx+8(FP), CX |
|
| 105 |
+ MOVQ R8, 24(CX) |
|
| 106 |
+ MOVQ AX, 32(CX) |
|
| 107 |
+ MOVQ $+2, ret+16(FP) |
|
| 108 |
+ RET |
|
| 109 |
+ |
|
| 110 |
+build_table_check1_ok: |
|
| 111 |
+ TESTB CL, CL |
|
| 112 |
+ JNZ build_table_check2_ok |
|
| 113 |
+ CMPW R8, DI |
|
| 114 |
+ JNE build_table_check2_ok |
|
| 115 |
+ MOVQ ctx+8(FP), AX |
|
| 116 |
+ MOVQ R8, 24(AX) |
|
| 117 |
+ MOVQ DI, 32(AX) |
|
| 118 |
+ MOVQ $+3, ret+16(FP) |
|
| 119 |
+ RET |
|
| 120 |
+ |
|
| 121 |
+build_table_check2_ok: |
|
| 122 |
+ INCQ DI |
|
| 123 |
+ CMPQ DI, AX |
|
| 124 |
+ JL build_table_main_table |
|
| 125 |
+ MOVQ $+0, ret+16(FP) |
|
| 126 |
+ RET |
| 0 | 127 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,72 @@ |
| 0 |
+//go:build !amd64 || appengine || !gc || noasm |
|
| 1 |
+// +build !amd64 appengine !gc noasm |
|
| 2 |
+ |
|
| 3 |
+package zstd |
|
| 4 |
+ |
|
| 5 |
+import ( |
|
| 6 |
+ "errors" |
|
| 7 |
+ "fmt" |
|
| 8 |
+) |
|
| 9 |
+ |
|
| 10 |
+// buildDtable will build the decoding table. |
|
| 11 |
+func (s *fseDecoder) buildDtable() error {
|
|
| 12 |
+ tableSize := uint32(1 << s.actualTableLog) |
|
| 13 |
+ highThreshold := tableSize - 1 |
|
| 14 |
+ symbolNext := s.stateTable[:256] |
|
| 15 |
+ |
|
| 16 |
+ // Init, lay down lowprob symbols |
|
| 17 |
+ {
|
|
| 18 |
+ for i, v := range s.norm[:s.symbolLen] {
|
|
| 19 |
+ if v == -1 {
|
|
| 20 |
+ s.dt[highThreshold].setAddBits(uint8(i)) |
|
| 21 |
+ highThreshold-- |
|
| 22 |
+ symbolNext[i] = 1 |
|
| 23 |
+ } else {
|
|
| 24 |
+ symbolNext[i] = uint16(v) |
|
| 25 |
+ } |
|
| 26 |
+ } |
|
| 27 |
+ } |
|
| 28 |
+ |
|
| 29 |
+ // Spread symbols |
|
| 30 |
+ {
|
|
| 31 |
+ tableMask := tableSize - 1 |
|
| 32 |
+ step := tableStep(tableSize) |
|
| 33 |
+ position := uint32(0) |
|
| 34 |
+ for ss, v := range s.norm[:s.symbolLen] {
|
|
| 35 |
+ for i := 0; i < int(v); i++ {
|
|
| 36 |
+ s.dt[position].setAddBits(uint8(ss)) |
|
| 37 |
+ position = (position + step) & tableMask |
|
| 38 |
+ for position > highThreshold {
|
|
| 39 |
+ // lowprob area |
|
| 40 |
+ position = (position + step) & tableMask |
|
| 41 |
+ } |
|
| 42 |
+ } |
|
| 43 |
+ } |
|
| 44 |
+ if position != 0 {
|
|
| 45 |
+ // position must reach all cells once, otherwise normalizedCounter is incorrect |
|
| 46 |
+ return errors.New("corrupted input (position != 0)")
|
|
| 47 |
+ } |
|
| 48 |
+ } |
|
| 49 |
+ |
|
| 50 |
+ // Build Decoding table |
|
| 51 |
+ {
|
|
| 52 |
+ tableSize := uint16(1 << s.actualTableLog) |
|
| 53 |
+ for u, v := range s.dt[:tableSize] {
|
|
| 54 |
+ symbol := v.addBits() |
|
| 55 |
+ nextState := symbolNext[symbol] |
|
| 56 |
+ symbolNext[symbol] = nextState + 1 |
|
| 57 |
+ nBits := s.actualTableLog - byte(highBits(uint32(nextState))) |
|
| 58 |
+ s.dt[u&maxTableMask].setNBits(nBits) |
|
| 59 |
+ newState := (nextState << nBits) - tableSize |
|
| 60 |
+ if newState > tableSize {
|
|
| 61 |
+ return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
|
|
| 62 |
+ } |
|
| 63 |
+ if newState == uint16(u) && nBits == 0 {
|
|
| 64 |
+ // Seems weird that this is possible with nbits > 0. |
|
| 65 |
+ return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
|
|
| 66 |
+ } |
|
| 67 |
+ s.dt[u&maxTableMask].setNewState(newState) |
|
| 68 |
+ } |
|
| 69 |
+ } |
|
| 70 |
+ return nil |
|
| 71 |
+} |
| ... | ... |
@@ -76,21 +76,6 @@ func (s *fseEncoder) HistogramFinished(maxSymbol uint8, maxCount int) {
|
| 76 | 76 |
s.clearCount = maxCount != 0 |
| 77 | 77 |
} |
| 78 | 78 |
|
| 79 |
-// prepare will prepare and allocate scratch tables used for both compression and decompression. |
|
| 80 |
-func (s *fseEncoder) prepare() (*fseEncoder, error) {
|
|
| 81 |
- if s == nil {
|
|
| 82 |
- s = &fseEncoder{}
|
|
| 83 |
- } |
|
| 84 |
- s.useRLE = false |
|
| 85 |
- if s.clearCount && s.maxCount == 0 {
|
|
| 86 |
- for i := range s.count {
|
|
| 87 |
- s.count[i] = 0 |
|
| 88 |
- } |
|
| 89 |
- s.clearCount = false |
|
| 90 |
- } |
|
| 91 |
- return s, nil |
|
| 92 |
-} |
|
| 93 |
- |
|
| 94 | 79 |
// allocCtable will allocate tables needed for compression. |
| 95 | 80 |
// If existing tables a re big enough, they are simply re-used. |
| 96 | 81 |
func (s *fseEncoder) allocCtable() {
|
| ... | ... |
@@ -709,14 +694,6 @@ func (c *cState) init(bw *bitWriter, ct *cTable, first symbolTransform) {
|
| 709 | 709 |
c.state = c.stateTable[lu] |
| 710 | 710 |
} |
| 711 | 711 |
|
| 712 |
-// encode the output symbol provided and write it to the bitstream. |
|
| 713 |
-func (c *cState) encode(symbolTT symbolTransform) {
|
|
| 714 |
- nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16 |
|
| 715 |
- dstState := int32(c.state>>(nbBitsOut&15)) + int32(symbolTT.deltaFindState) |
|
| 716 |
- c.bw.addBits16NC(c.state, uint8(nbBitsOut)) |
|
| 717 |
- c.state = c.stateTable[dstState] |
|
| 718 |
-} |
|
| 719 |
- |
|
| 720 | 712 |
// flush will write the tablelog to the output and flush the remaining full bytes. |
| 721 | 713 |
func (c *cState) flush(tableLog uint8) {
|
| 722 | 714 |
c.bw.flush32() |
| 723 | 715 |
deleted file mode 100644 |
| ... | ... |
@@ -1,11 +0,0 @@ |
| 1 |
-//go:build ignorecrc |
|
| 2 |
-// +build ignorecrc |
|
| 3 |
- |
|
| 4 |
-// Copyright 2019+ Klaus Post. All rights reserved. |
|
| 5 |
-// License information can be found in the LICENSE file. |
|
| 6 |
-// Based on work by Yann Collet, released under BSD License. |
|
| 7 |
- |
|
| 8 |
-package zstd |
|
| 9 |
- |
|
| 10 |
-// ignoreCRC can be used for fuzz testing to ignore CRC values... |
|
| 11 |
-const ignoreCRC = true |
| 12 | 1 |
deleted file mode 100644 |
| ... | ... |
@@ -1,11 +0,0 @@ |
| 1 |
-//go:build !ignorecrc |
|
| 2 |
-// +build !ignorecrc |
|
| 3 |
- |
|
| 4 |
-// Copyright 2019+ Klaus Post. All rights reserved. |
|
| 5 |
-// License information can be found in the LICENSE file. |
|
| 6 |
-// Based on work by Yann Collet, released under BSD License. |
|
| 7 |
- |
|
| 8 |
-package zstd |
|
| 9 |
- |
|
| 10 |
-// ignoreCRC can be used for fuzz testing to ignore CRC values... |
|
| 11 |
-const ignoreCRC = false |
| ... | ... |
@@ -33,9 +33,3 @@ func hashLen(u uint64, length, mls uint8) uint32 {
|
| 33 | 33 |
return (uint32(u) * prime4bytes) >> (32 - length) |
| 34 | 34 |
} |
| 35 | 35 |
} |
| 36 |
- |
|
| 37 |
-// hash3 returns the hash of the lower 3 bytes of u to fit in a hash table with h bits. |
|
| 38 |
-// Preferably h should be a constant and should always be <32. |
|
| 39 |
-func hash3(u uint32, h uint8) uint32 {
|
|
| 40 |
- return ((u << (32 - 24)) * prime3bytes) >> ((32 - h) & 31) |
|
| 41 |
-} |
| ... | ... |
@@ -73,6 +73,7 @@ type sequenceDecs struct {
|
| 73 | 73 |
seqSize int |
| 74 | 74 |
windowSize int |
| 75 | 75 |
maxBits uint8 |
| 76 |
+ maxSyncLen uint64 |
|
| 76 | 77 |
} |
| 77 | 78 |
|
| 78 | 79 |
// initialize all 3 decoders from the stream input. |
| ... | ... |
@@ -98,153 +99,13 @@ func (s *sequenceDecs) initialize(br *bitReader, hist *history, out []byte) erro |
| 98 | 98 |
return nil |
| 99 | 99 |
} |
| 100 | 100 |
|
| 101 |
-// decode sequences from the stream with the provided history. |
|
| 102 |
-func (s *sequenceDecs) decode(seqs []seqVals) error {
|
|
| 103 |
- br := s.br |
|
| 104 |
- |
|
| 105 |
- // Grab full sizes tables, to avoid bounds checks. |
|
| 106 |
- llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize] |
|
| 107 |
- llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state |
|
| 108 |
- s.seqSize = 0 |
|
| 109 |
- litRemain := len(s.literals) |
|
| 110 |
- maxBlockSize := maxCompressedBlockSize |
|
| 111 |
- if s.windowSize < maxBlockSize {
|
|
| 112 |
- maxBlockSize = s.windowSize |
|
| 113 |
- } |
|
| 114 |
- for i := range seqs {
|
|
| 115 |
- var ll, mo, ml int |
|
| 116 |
- if br.off > 4+((maxOffsetBits+16+16)>>3) {
|
|
| 117 |
- // inlined function: |
|
| 118 |
- // ll, mo, ml = s.nextFast(br, llState, mlState, ofState) |
|
| 119 |
- |
|
| 120 |
- // Final will not read from stream. |
|
| 121 |
- var llB, mlB, moB uint8 |
|
| 122 |
- ll, llB = llState.final() |
|
| 123 |
- ml, mlB = mlState.final() |
|
| 124 |
- mo, moB = ofState.final() |
|
| 125 |
- |
|
| 126 |
- // extra bits are stored in reverse order. |
|
| 127 |
- br.fillFast() |
|
| 128 |
- mo += br.getBits(moB) |
|
| 129 |
- if s.maxBits > 32 {
|
|
| 130 |
- br.fillFast() |
|
| 131 |
- } |
|
| 132 |
- ml += br.getBits(mlB) |
|
| 133 |
- ll += br.getBits(llB) |
|
| 134 |
- |
|
| 135 |
- if moB > 1 {
|
|
| 136 |
- s.prevOffset[2] = s.prevOffset[1] |
|
| 137 |
- s.prevOffset[1] = s.prevOffset[0] |
|
| 138 |
- s.prevOffset[0] = mo |
|
| 139 |
- } else {
|
|
| 140 |
- // mo = s.adjustOffset(mo, ll, moB) |
|
| 141 |
- // Inlined for rather big speedup |
|
| 142 |
- if ll == 0 {
|
|
| 143 |
- // There is an exception though, when current sequence's literals_length = 0. |
|
| 144 |
- // In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2, |
|
| 145 |
- // an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte. |
|
| 146 |
- mo++ |
|
| 147 |
- } |
|
| 148 |
- |
|
| 149 |
- if mo == 0 {
|
|
| 150 |
- mo = s.prevOffset[0] |
|
| 151 |
- } else {
|
|
| 152 |
- var temp int |
|
| 153 |
- if mo == 3 {
|
|
| 154 |
- temp = s.prevOffset[0] - 1 |
|
| 155 |
- } else {
|
|
| 156 |
- temp = s.prevOffset[mo] |
|
| 157 |
- } |
|
| 158 |
- |
|
| 159 |
- if temp == 0 {
|
|
| 160 |
- // 0 is not valid; input is corrupted; force offset to 1 |
|
| 161 |
- println("WARNING: temp was 0")
|
|
| 162 |
- temp = 1 |
|
| 163 |
- } |
|
| 164 |
- |
|
| 165 |
- if mo != 1 {
|
|
| 166 |
- s.prevOffset[2] = s.prevOffset[1] |
|
| 167 |
- } |
|
| 168 |
- s.prevOffset[1] = s.prevOffset[0] |
|
| 169 |
- s.prevOffset[0] = temp |
|
| 170 |
- mo = temp |
|
| 171 |
- } |
|
| 172 |
- } |
|
| 173 |
- br.fillFast() |
|
| 174 |
- } else {
|
|
| 175 |
- if br.overread() {
|
|
| 176 |
- if debugDecoder {
|
|
| 177 |
- printf("reading sequence %d, exceeded available data\n", i)
|
|
| 178 |
- } |
|
| 179 |
- return io.ErrUnexpectedEOF |
|
| 180 |
- } |
|
| 181 |
- ll, mo, ml = s.next(br, llState, mlState, ofState) |
|
| 182 |
- br.fill() |
|
| 183 |
- } |
|
| 184 |
- |
|
| 185 |
- if debugSequences {
|
|
| 186 |
- println("Seq", i, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml)
|
|
| 187 |
- } |
|
| 188 |
- // Evaluate. |
|
| 189 |
- // We might be doing this async, so do it early. |
|
| 190 |
- if mo == 0 && ml > 0 {
|
|
| 191 |
- return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
|
|
| 192 |
- } |
|
| 193 |
- if ml > maxMatchLen {
|
|
| 194 |
- return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
|
|
| 195 |
- } |
|
| 196 |
- s.seqSize += ll + ml |
|
| 197 |
- if s.seqSize > maxBlockSize {
|
|
| 198 |
- return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
|
|
| 199 |
- } |
|
| 200 |
- litRemain -= ll |
|
| 201 |
- if litRemain < 0 {
|
|
| 202 |
- return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, litRemain+ll)
|
|
| 203 |
- } |
|
| 204 |
- seqs[i] = seqVals{
|
|
| 205 |
- ll: ll, |
|
| 206 |
- ml: ml, |
|
| 207 |
- mo: mo, |
|
| 208 |
- } |
|
| 209 |
- if i == len(seqs)-1 {
|
|
| 210 |
- // This is the last sequence, so we shouldn't update state. |
|
| 211 |
- break |
|
| 212 |
- } |
|
| 213 |
- |
|
| 214 |
- // Manually inlined, ~ 5-20% faster |
|
| 215 |
- // Update all 3 states at once. Approx 20% faster. |
|
| 216 |
- nBits := llState.nbBits() + mlState.nbBits() + ofState.nbBits() |
|
| 217 |
- if nBits == 0 {
|
|
| 218 |
- llState = llTable[llState.newState()&maxTableMask] |
|
| 219 |
- mlState = mlTable[mlState.newState()&maxTableMask] |
|
| 220 |
- ofState = ofTable[ofState.newState()&maxTableMask] |
|
| 221 |
- } else {
|
|
| 222 |
- bits := br.get32BitsFast(nBits) |
|
| 223 |
- lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31)) |
|
| 224 |
- llState = llTable[(llState.newState()+lowBits)&maxTableMask] |
|
| 225 |
- |
|
| 226 |
- lowBits = uint16(bits >> (ofState.nbBits() & 31)) |
|
| 227 |
- lowBits &= bitMask[mlState.nbBits()&15] |
|
| 228 |
- mlState = mlTable[(mlState.newState()+lowBits)&maxTableMask] |
|
| 229 |
- |
|
| 230 |
- lowBits = uint16(bits) & bitMask[ofState.nbBits()&15] |
|
| 231 |
- ofState = ofTable[(ofState.newState()+lowBits)&maxTableMask] |
|
| 232 |
- } |
|
| 233 |
- } |
|
| 234 |
- s.seqSize += litRemain |
|
| 235 |
- if s.seqSize > maxBlockSize {
|
|
| 236 |
- return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
|
|
| 237 |
- } |
|
| 238 |
- err := br.close() |
|
| 239 |
- if err != nil {
|
|
| 240 |
- printf("Closing sequences: %v, %+v\n", err, *br)
|
|
| 241 |
- } |
|
| 242 |
- return err |
|
| 243 |
-} |
|
| 244 |
- |
|
| 245 | 101 |
// execute will execute the decoded sequence with the provided history. |
| 246 | 102 |
// The sequence must be evaluated before being sent. |
| 247 | 103 |
func (s *sequenceDecs) execute(seqs []seqVals, hist []byte) error {
|
| 104 |
+ if len(s.dict) == 0 {
|
|
| 105 |
+ return s.executeSimple(seqs, hist) |
|
| 106 |
+ } |
|
| 107 |
+ |
|
| 248 | 108 |
// Ensure we have enough output size... |
| 249 | 109 |
if len(s.out)+s.seqSize > cap(s.out) {
|
| 250 | 110 |
addBytes := s.seqSize + len(s.out) |
| ... | ... |
@@ -327,6 +188,7 @@ func (s *sequenceDecs) execute(seqs []seqVals, hist []byte) error {
|
| 327 | 327 |
} |
| 328 | 328 |
} |
| 329 | 329 |
} |
| 330 |
+ |
|
| 330 | 331 |
// Add final literals |
| 331 | 332 |
copy(out[t:], s.literals) |
| 332 | 333 |
if debugDecoder {
|
| ... | ... |
@@ -341,14 +203,18 @@ func (s *sequenceDecs) execute(seqs []seqVals, hist []byte) error {
|
| 341 | 341 |
} |
| 342 | 342 |
|
| 343 | 343 |
// decode sequences from the stream with the provided history. |
| 344 |
-func (s *sequenceDecs) decodeSync(history *history) error {
|
|
| 344 |
+func (s *sequenceDecs) decodeSync(hist []byte) error {
|
|
| 345 |
+ supported, err := s.decodeSyncSimple(hist) |
|
| 346 |
+ if supported {
|
|
| 347 |
+ return err |
|
| 348 |
+ } |
|
| 349 |
+ |
|
| 345 | 350 |
br := s.br |
| 346 | 351 |
seqs := s.nSeqs |
| 347 | 352 |
startSize := len(s.out) |
| 348 | 353 |
// Grab full sizes tables, to avoid bounds checks. |
| 349 | 354 |
llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize] |
| 350 | 355 |
llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state |
| 351 |
- hist := history.b[history.ignoreBuffer:] |
|
| 352 | 356 |
out := s.out |
| 353 | 357 |
maxBlockSize := maxCompressedBlockSize |
| 354 | 358 |
if s.windowSize < maxBlockSize {
|
| ... | ... |
@@ -433,7 +299,7 @@ func (s *sequenceDecs) decodeSync(history *history) error {
|
| 433 | 433 |
} |
| 434 | 434 |
size := ll + ml + len(out) |
| 435 | 435 |
if size-startSize > maxBlockSize {
|
| 436 |
- return fmt.Errorf("output (%d) bigger than max block size (%d)", size, maxBlockSize)
|
|
| 436 |
+ return fmt.Errorf("output (%d) bigger than max block size (%d)", size-startSize, maxBlockSize)
|
|
| 437 | 437 |
} |
| 438 | 438 |
if size > cap(out) {
|
| 439 | 439 |
// Not enough size, which can happen under high volume block streaming conditions |
| ... | ... |
@@ -463,13 +329,13 @@ func (s *sequenceDecs) decodeSync(history *history) error {
|
| 463 | 463 |
|
| 464 | 464 |
if mo > len(out)+len(hist) || mo > s.windowSize {
|
| 465 | 465 |
if len(s.dict) == 0 {
|
| 466 |
- return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist))
|
|
| 466 |
+ return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist)-startSize)
|
|
| 467 | 467 |
} |
| 468 | 468 |
|
| 469 | 469 |
// we may be in dictionary. |
| 470 | 470 |
dictO := len(s.dict) - (mo - (len(out) + len(hist))) |
| 471 | 471 |
if dictO < 0 || dictO >= len(s.dict) {
|
| 472 |
- return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist))
|
|
| 472 |
+ return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(out)+len(hist)-startSize)
|
|
| 473 | 473 |
} |
| 474 | 474 |
end := dictO + ml |
| 475 | 475 |
if end > len(s.dict) {
|
| ... | ... |
@@ -530,6 +396,7 @@ func (s *sequenceDecs) decodeSync(history *history) error {
|
| 530 | 530 |
ofState = ofTable[ofState.newState()&maxTableMask] |
| 531 | 531 |
} else {
|
| 532 | 532 |
bits := br.get32BitsFast(nBits) |
| 533 |
+ |
|
| 533 | 534 |
lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31)) |
| 534 | 535 |
llState = llTable[(llState.newState()+lowBits)&maxTableMask] |
| 535 | 536 |
|
| ... | ... |
@@ -543,8 +410,8 @@ func (s *sequenceDecs) decodeSync(history *history) error {
|
| 543 | 543 |
} |
| 544 | 544 |
|
| 545 | 545 |
// Check if space for literals |
| 546 |
- if len(s.literals)+len(s.out)-startSize > maxBlockSize {
|
|
| 547 |
- return fmt.Errorf("output (%d) bigger than max block size (%d)", len(s.out), maxBlockSize)
|
|
| 546 |
+ if size := len(s.literals) + len(s.out) - startSize; size > maxBlockSize {
|
|
| 547 |
+ return fmt.Errorf("output (%d) bigger than max block size (%d)", size, maxBlockSize)
|
|
| 548 | 548 |
} |
| 549 | 549 |
|
| 550 | 550 |
// Add final literals |
| ... | ... |
@@ -552,16 +419,6 @@ func (s *sequenceDecs) decodeSync(history *history) error {
|
| 552 | 552 |
return br.close() |
| 553 | 553 |
} |
| 554 | 554 |
|
| 555 |
-// update states, at least 27 bits must be available. |
|
| 556 |
-func (s *sequenceDecs) update(br *bitReader) {
|
|
| 557 |
- // Max 8 bits |
|
| 558 |
- s.litLengths.state.next(br) |
|
| 559 |
- // Max 9 bits |
|
| 560 |
- s.matchLengths.state.next(br) |
|
| 561 |
- // Max 8 bits |
|
| 562 |
- s.offsets.state.next(br) |
|
| 563 |
-} |
|
| 564 |
- |
|
| 565 | 555 |
var bitMask [16]uint16 |
| 566 | 556 |
|
| 567 | 557 |
func init() {
|
| ... | ... |
@@ -570,87 +427,6 @@ func init() {
|
| 570 | 570 |
} |
| 571 | 571 |
} |
| 572 | 572 |
|
| 573 |
-// update states, at least 27 bits must be available. |
|
| 574 |
-func (s *sequenceDecs) updateAlt(br *bitReader) {
|
|
| 575 |
- // Update all 3 states at once. Approx 20% faster. |
|
| 576 |
- a, b, c := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state |
|
| 577 |
- |
|
| 578 |
- nBits := a.nbBits() + b.nbBits() + c.nbBits() |
|
| 579 |
- if nBits == 0 {
|
|
| 580 |
- s.litLengths.state.state = s.litLengths.state.dt[a.newState()] |
|
| 581 |
- s.matchLengths.state.state = s.matchLengths.state.dt[b.newState()] |
|
| 582 |
- s.offsets.state.state = s.offsets.state.dt[c.newState()] |
|
| 583 |
- return |
|
| 584 |
- } |
|
| 585 |
- bits := br.get32BitsFast(nBits) |
|
| 586 |
- lowBits := uint16(bits >> ((c.nbBits() + b.nbBits()) & 31)) |
|
| 587 |
- s.litLengths.state.state = s.litLengths.state.dt[a.newState()+lowBits] |
|
| 588 |
- |
|
| 589 |
- lowBits = uint16(bits >> (c.nbBits() & 31)) |
|
| 590 |
- lowBits &= bitMask[b.nbBits()&15] |
|
| 591 |
- s.matchLengths.state.state = s.matchLengths.state.dt[b.newState()+lowBits] |
|
| 592 |
- |
|
| 593 |
- lowBits = uint16(bits) & bitMask[c.nbBits()&15] |
|
| 594 |
- s.offsets.state.state = s.offsets.state.dt[c.newState()+lowBits] |
|
| 595 |
-} |
|
| 596 |
- |
|
| 597 |
-// nextFast will return new states when there are at least 4 unused bytes left on the stream when done. |
|
| 598 |
-func (s *sequenceDecs) nextFast(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) {
|
|
| 599 |
- // Final will not read from stream. |
|
| 600 |
- ll, llB := llState.final() |
|
| 601 |
- ml, mlB := mlState.final() |
|
| 602 |
- mo, moB := ofState.final() |
|
| 603 |
- |
|
| 604 |
- // extra bits are stored in reverse order. |
|
| 605 |
- br.fillFast() |
|
| 606 |
- mo += br.getBits(moB) |
|
| 607 |
- if s.maxBits > 32 {
|
|
| 608 |
- br.fillFast() |
|
| 609 |
- } |
|
| 610 |
- ml += br.getBits(mlB) |
|
| 611 |
- ll += br.getBits(llB) |
|
| 612 |
- |
|
| 613 |
- if moB > 1 {
|
|
| 614 |
- s.prevOffset[2] = s.prevOffset[1] |
|
| 615 |
- s.prevOffset[1] = s.prevOffset[0] |
|
| 616 |
- s.prevOffset[0] = mo |
|
| 617 |
- return |
|
| 618 |
- } |
|
| 619 |
- // mo = s.adjustOffset(mo, ll, moB) |
|
| 620 |
- // Inlined for rather big speedup |
|
| 621 |
- if ll == 0 {
|
|
| 622 |
- // There is an exception though, when current sequence's literals_length = 0. |
|
| 623 |
- // In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2, |
|
| 624 |
- // an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte. |
|
| 625 |
- mo++ |
|
| 626 |
- } |
|
| 627 |
- |
|
| 628 |
- if mo == 0 {
|
|
| 629 |
- mo = s.prevOffset[0] |
|
| 630 |
- return |
|
| 631 |
- } |
|
| 632 |
- var temp int |
|
| 633 |
- if mo == 3 {
|
|
| 634 |
- temp = s.prevOffset[0] - 1 |
|
| 635 |
- } else {
|
|
| 636 |
- temp = s.prevOffset[mo] |
|
| 637 |
- } |
|
| 638 |
- |
|
| 639 |
- if temp == 0 {
|
|
| 640 |
- // 0 is not valid; input is corrupted; force offset to 1 |
|
| 641 |
- println("temp was 0")
|
|
| 642 |
- temp = 1 |
|
| 643 |
- } |
|
| 644 |
- |
|
| 645 |
- if mo != 1 {
|
|
| 646 |
- s.prevOffset[2] = s.prevOffset[1] |
|
| 647 |
- } |
|
| 648 |
- s.prevOffset[1] = s.prevOffset[0] |
|
| 649 |
- s.prevOffset[0] = temp |
|
| 650 |
- mo = temp |
|
| 651 |
- return |
|
| 652 |
-} |
|
| 653 |
- |
|
| 654 | 573 |
func (s *sequenceDecs) next(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) {
|
| 655 | 574 |
// Final will not read from stream. |
| 656 | 575 |
ll, llB := llState.final() |
| 657 | 576 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,368 @@ |
| 0 |
+//go:build amd64 && !appengine && !noasm && gc |
|
| 1 |
+// +build amd64,!appengine,!noasm,gc |
|
| 2 |
+ |
|
| 3 |
+package zstd |
|
| 4 |
+ |
|
| 5 |
+import ( |
|
| 6 |
+ "fmt" |
|
| 7 |
+ |
|
| 8 |
+ "github.com/klauspost/compress/internal/cpuinfo" |
|
| 9 |
+) |
|
| 10 |
+ |
|
| 11 |
+type decodeSyncAsmContext struct {
|
|
| 12 |
+ llTable []decSymbol |
|
| 13 |
+ mlTable []decSymbol |
|
| 14 |
+ ofTable []decSymbol |
|
| 15 |
+ llState uint64 |
|
| 16 |
+ mlState uint64 |
|
| 17 |
+ ofState uint64 |
|
| 18 |
+ iteration int |
|
| 19 |
+ litRemain int |
|
| 20 |
+ out []byte |
|
| 21 |
+ outPosition int |
|
| 22 |
+ literals []byte |
|
| 23 |
+ litPosition int |
|
| 24 |
+ history []byte |
|
| 25 |
+ windowSize int |
|
| 26 |
+ ll int // set on error (not for all errors, please refer to _generate/gen.go) |
|
| 27 |
+ ml int // set on error (not for all errors, please refer to _generate/gen.go) |
|
| 28 |
+ mo int // set on error (not for all errors, please refer to _generate/gen.go) |
|
| 29 |
+} |
|
| 30 |
+ |
|
| 31 |
+// sequenceDecs_decodeSync_amd64 implements the main loop of sequenceDecs.decodeSync in x86 asm. |
|
| 32 |
+// |
|
| 33 |
+// Please refer to seqdec_generic.go for the reference implementation. |
|
| 34 |
+//go:noescape |
|
| 35 |
+func sequenceDecs_decodeSync_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 36 |
+ |
|
| 37 |
+// sequenceDecs_decodeSync_bmi2 implements the main loop of sequenceDecs.decodeSync in x86 asm with BMI2 extensions. |
|
| 38 |
+//go:noescape |
|
| 39 |
+func sequenceDecs_decodeSync_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 40 |
+ |
|
| 41 |
+// sequenceDecs_decodeSync_safe_amd64 does the same as above, but does not write more than output buffer. |
|
| 42 |
+//go:noescape |
|
| 43 |
+func sequenceDecs_decodeSync_safe_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 44 |
+ |
|
| 45 |
+// sequenceDecs_decodeSync_safe_bmi2 does the same as above, but does not write more than output buffer. |
|
| 46 |
+//go:noescape |
|
| 47 |
+func sequenceDecs_decodeSync_safe_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 48 |
+ |
|
| 49 |
+// decode sequences from the stream with the provided history but without a dictionary. |
|
| 50 |
+func (s *sequenceDecs) decodeSyncSimple(hist []byte) (bool, error) {
|
|
| 51 |
+ if len(s.dict) > 0 {
|
|
| 52 |
+ return false, nil |
|
| 53 |
+ } |
|
| 54 |
+ if s.maxSyncLen == 0 && cap(s.out)-len(s.out) < maxCompressedBlockSize {
|
|
| 55 |
+ return false, nil |
|
| 56 |
+ } |
|
| 57 |
+ |
|
| 58 |
+ // FIXME: Using unsafe memory copies leads to rare, random crashes |
|
| 59 |
+ // with fuzz testing. It is therefore disabled for now. |
|
| 60 |
+ const useSafe = true |
|
| 61 |
+ /* |
|
| 62 |
+ useSafe := false |
|
| 63 |
+ if s.maxSyncLen == 0 && cap(s.out)-len(s.out) < maxCompressedBlockSizeAlloc {
|
|
| 64 |
+ useSafe = true |
|
| 65 |
+ } |
|
| 66 |
+ if s.maxSyncLen > 0 && cap(s.out)-len(s.out)-compressedBlockOverAlloc < int(s.maxSyncLen) {
|
|
| 67 |
+ useSafe = true |
|
| 68 |
+ } |
|
| 69 |
+ if cap(s.literals) < len(s.literals)+compressedBlockOverAlloc {
|
|
| 70 |
+ useSafe = true |
|
| 71 |
+ } |
|
| 72 |
+ */ |
|
| 73 |
+ |
|
| 74 |
+ br := s.br |
|
| 75 |
+ |
|
| 76 |
+ maxBlockSize := maxCompressedBlockSize |
|
| 77 |
+ if s.windowSize < maxBlockSize {
|
|
| 78 |
+ maxBlockSize = s.windowSize |
|
| 79 |
+ } |
|
| 80 |
+ |
|
| 81 |
+ ctx := decodeSyncAsmContext{
|
|
| 82 |
+ llTable: s.litLengths.fse.dt[:maxTablesize], |
|
| 83 |
+ mlTable: s.matchLengths.fse.dt[:maxTablesize], |
|
| 84 |
+ ofTable: s.offsets.fse.dt[:maxTablesize], |
|
| 85 |
+ llState: uint64(s.litLengths.state.state), |
|
| 86 |
+ mlState: uint64(s.matchLengths.state.state), |
|
| 87 |
+ ofState: uint64(s.offsets.state.state), |
|
| 88 |
+ iteration: s.nSeqs - 1, |
|
| 89 |
+ litRemain: len(s.literals), |
|
| 90 |
+ out: s.out, |
|
| 91 |
+ outPosition: len(s.out), |
|
| 92 |
+ literals: s.literals, |
|
| 93 |
+ windowSize: s.windowSize, |
|
| 94 |
+ history: hist, |
|
| 95 |
+ } |
|
| 96 |
+ |
|
| 97 |
+ s.seqSize = 0 |
|
| 98 |
+ startSize := len(s.out) |
|
| 99 |
+ |
|
| 100 |
+ var errCode int |
|
| 101 |
+ if cpuinfo.HasBMI2() {
|
|
| 102 |
+ if useSafe {
|
|
| 103 |
+ errCode = sequenceDecs_decodeSync_safe_bmi2(s, br, &ctx) |
|
| 104 |
+ } else {
|
|
| 105 |
+ errCode = sequenceDecs_decodeSync_bmi2(s, br, &ctx) |
|
| 106 |
+ } |
|
| 107 |
+ } else {
|
|
| 108 |
+ if useSafe {
|
|
| 109 |
+ errCode = sequenceDecs_decodeSync_safe_amd64(s, br, &ctx) |
|
| 110 |
+ } else {
|
|
| 111 |
+ errCode = sequenceDecs_decodeSync_amd64(s, br, &ctx) |
|
| 112 |
+ } |
|
| 113 |
+ } |
|
| 114 |
+ switch errCode {
|
|
| 115 |
+ case noError: |
|
| 116 |
+ break |
|
| 117 |
+ |
|
| 118 |
+ case errorMatchLenOfsMismatch: |
|
| 119 |
+ return true, fmt.Errorf("zero matchoff and matchlen (%d) > 0", ctx.ml)
|
|
| 120 |
+ |
|
| 121 |
+ case errorMatchLenTooBig: |
|
| 122 |
+ return true, fmt.Errorf("match len (%d) bigger than max allowed length", ctx.ml)
|
|
| 123 |
+ |
|
| 124 |
+ case errorMatchOffTooBig: |
|
| 125 |
+ return true, fmt.Errorf("match offset (%d) bigger than current history (%d)",
|
|
| 126 |
+ ctx.mo, ctx.outPosition+len(hist)-startSize) |
|
| 127 |
+ |
|
| 128 |
+ case errorNotEnoughLiterals: |
|
| 129 |
+ return true, fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available",
|
|
| 130 |
+ ctx.ll, ctx.litRemain+ctx.ll) |
|
| 131 |
+ |
|
| 132 |
+ case errorNotEnoughSpace: |
|
| 133 |
+ size := ctx.outPosition + ctx.ll + ctx.ml |
|
| 134 |
+ if debugDecoder {
|
|
| 135 |
+ println("msl:", s.maxSyncLen, "cap", cap(s.out), "bef:", startSize, "sz:", size-startSize, "mbs:", maxBlockSize, "outsz:", cap(s.out)-startSize)
|
|
| 136 |
+ } |
|
| 137 |
+ return true, fmt.Errorf("output (%d) bigger than max block size (%d)", size-startSize, maxBlockSize)
|
|
| 138 |
+ |
|
| 139 |
+ default: |
|
| 140 |
+ return true, fmt.Errorf("sequenceDecs_decode returned erronous code %d", errCode)
|
|
| 141 |
+ } |
|
| 142 |
+ |
|
| 143 |
+ s.seqSize += ctx.litRemain |
|
| 144 |
+ if s.seqSize > maxBlockSize {
|
|
| 145 |
+ return true, fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
|
|
| 146 |
+ } |
|
| 147 |
+ err := br.close() |
|
| 148 |
+ if err != nil {
|
|
| 149 |
+ printf("Closing sequences: %v, %+v\n", err, *br)
|
|
| 150 |
+ return true, err |
|
| 151 |
+ } |
|
| 152 |
+ |
|
| 153 |
+ s.literals = s.literals[ctx.litPosition:] |
|
| 154 |
+ t := ctx.outPosition |
|
| 155 |
+ s.out = s.out[:t] |
|
| 156 |
+ |
|
| 157 |
+ // Add final literals |
|
| 158 |
+ s.out = append(s.out, s.literals...) |
|
| 159 |
+ if debugDecoder {
|
|
| 160 |
+ t += len(s.literals) |
|
| 161 |
+ if t != len(s.out) {
|
|
| 162 |
+ panic(fmt.Errorf("length mismatch, want %d, got %d", len(s.out), t))
|
|
| 163 |
+ } |
|
| 164 |
+ } |
|
| 165 |
+ |
|
| 166 |
+ return true, nil |
|
| 167 |
+} |
|
| 168 |
+ |
|
| 169 |
+// -------------------------------------------------------------------------------- |
|
| 170 |
+ |
|
| 171 |
+type decodeAsmContext struct {
|
|
| 172 |
+ llTable []decSymbol |
|
| 173 |
+ mlTable []decSymbol |
|
| 174 |
+ ofTable []decSymbol |
|
| 175 |
+ llState uint64 |
|
| 176 |
+ mlState uint64 |
|
| 177 |
+ ofState uint64 |
|
| 178 |
+ iteration int |
|
| 179 |
+ seqs []seqVals |
|
| 180 |
+ litRemain int |
|
| 181 |
+} |
|
| 182 |
+ |
|
| 183 |
+const noError = 0 |
|
| 184 |
+ |
|
| 185 |
+// error reported when mo == 0 && ml > 0 |
|
| 186 |
+const errorMatchLenOfsMismatch = 1 |
|
| 187 |
+ |
|
| 188 |
+// error reported when ml > maxMatchLen |
|
| 189 |
+const errorMatchLenTooBig = 2 |
|
| 190 |
+ |
|
| 191 |
+// error reported when mo > available history or mo > s.windowSize |
|
| 192 |
+const errorMatchOffTooBig = 3 |
|
| 193 |
+ |
|
| 194 |
+// error reported when the sum of literal lengths exeeceds the literal buffer size |
|
| 195 |
+const errorNotEnoughLiterals = 4 |
|
| 196 |
+ |
|
| 197 |
+// error reported when capacity of `out` is too small |
|
| 198 |
+const errorNotEnoughSpace = 5 |
|
| 199 |
+ |
|
| 200 |
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm. |
|
| 201 |
+// |
|
| 202 |
+// Please refer to seqdec_generic.go for the reference implementation. |
|
| 203 |
+//go:noescape |
|
| 204 |
+func sequenceDecs_decode_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 205 |
+ |
|
| 206 |
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm. |
|
| 207 |
+// |
|
| 208 |
+// Please refer to seqdec_generic.go for the reference implementation. |
|
| 209 |
+//go:noescape |
|
| 210 |
+func sequenceDecs_decode_56_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 211 |
+ |
|
| 212 |
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm with BMI2 extensions. |
|
| 213 |
+//go:noescape |
|
| 214 |
+func sequenceDecs_decode_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 215 |
+ |
|
| 216 |
+// sequenceDecs_decode implements the main loop of sequenceDecs in x86 asm with BMI2 extensions. |
|
| 217 |
+//go:noescape |
|
| 218 |
+func sequenceDecs_decode_56_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 219 |
+ |
|
| 220 |
+// decode sequences from the stream without the provided history. |
|
| 221 |
+func (s *sequenceDecs) decode(seqs []seqVals) error {
|
|
| 222 |
+ br := s.br |
|
| 223 |
+ |
|
| 224 |
+ maxBlockSize := maxCompressedBlockSize |
|
| 225 |
+ if s.windowSize < maxBlockSize {
|
|
| 226 |
+ maxBlockSize = s.windowSize |
|
| 227 |
+ } |
|
| 228 |
+ |
|
| 229 |
+ ctx := decodeAsmContext{
|
|
| 230 |
+ llTable: s.litLengths.fse.dt[:maxTablesize], |
|
| 231 |
+ mlTable: s.matchLengths.fse.dt[:maxTablesize], |
|
| 232 |
+ ofTable: s.offsets.fse.dt[:maxTablesize], |
|
| 233 |
+ llState: uint64(s.litLengths.state.state), |
|
| 234 |
+ mlState: uint64(s.matchLengths.state.state), |
|
| 235 |
+ ofState: uint64(s.offsets.state.state), |
|
| 236 |
+ seqs: seqs, |
|
| 237 |
+ iteration: len(seqs) - 1, |
|
| 238 |
+ litRemain: len(s.literals), |
|
| 239 |
+ } |
|
| 240 |
+ |
|
| 241 |
+ s.seqSize = 0 |
|
| 242 |
+ lte56bits := s.maxBits+s.offsets.fse.actualTableLog+s.matchLengths.fse.actualTableLog+s.litLengths.fse.actualTableLog <= 56 |
|
| 243 |
+ var errCode int |
|
| 244 |
+ if cpuinfo.HasBMI2() {
|
|
| 245 |
+ if lte56bits {
|
|
| 246 |
+ errCode = sequenceDecs_decode_56_bmi2(s, br, &ctx) |
|
| 247 |
+ } else {
|
|
| 248 |
+ errCode = sequenceDecs_decode_bmi2(s, br, &ctx) |
|
| 249 |
+ } |
|
| 250 |
+ } else {
|
|
| 251 |
+ if lte56bits {
|
|
| 252 |
+ errCode = sequenceDecs_decode_56_amd64(s, br, &ctx) |
|
| 253 |
+ } else {
|
|
| 254 |
+ errCode = sequenceDecs_decode_amd64(s, br, &ctx) |
|
| 255 |
+ } |
|
| 256 |
+ } |
|
| 257 |
+ if errCode != 0 {
|
|
| 258 |
+ i := len(seqs) - ctx.iteration - 1 |
|
| 259 |
+ switch errCode {
|
|
| 260 |
+ case errorMatchLenOfsMismatch: |
|
| 261 |
+ ml := ctx.seqs[i].ml |
|
| 262 |
+ return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
|
|
| 263 |
+ |
|
| 264 |
+ case errorMatchLenTooBig: |
|
| 265 |
+ ml := ctx.seqs[i].ml |
|
| 266 |
+ return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
|
|
| 267 |
+ |
|
| 268 |
+ case errorNotEnoughLiterals: |
|
| 269 |
+ ll := ctx.seqs[i].ll |
|
| 270 |
+ return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, ctx.litRemain+ll)
|
|
| 271 |
+ } |
|
| 272 |
+ |
|
| 273 |
+ return fmt.Errorf("sequenceDecs_decode_amd64 returned erronous code %d", errCode)
|
|
| 274 |
+ } |
|
| 275 |
+ |
|
| 276 |
+ if ctx.litRemain < 0 {
|
|
| 277 |
+ return fmt.Errorf("literal count is too big: total available %d, total requested %d",
|
|
| 278 |
+ len(s.literals), len(s.literals)-ctx.litRemain) |
|
| 279 |
+ } |
|
| 280 |
+ |
|
| 281 |
+ s.seqSize += ctx.litRemain |
|
| 282 |
+ if s.seqSize > maxBlockSize {
|
|
| 283 |
+ return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
|
|
| 284 |
+ } |
|
| 285 |
+ err := br.close() |
|
| 286 |
+ if err != nil {
|
|
| 287 |
+ printf("Closing sequences: %v, %+v\n", err, *br)
|
|
| 288 |
+ } |
|
| 289 |
+ return err |
|
| 290 |
+} |
|
| 291 |
+ |
|
| 292 |
+// -------------------------------------------------------------------------------- |
|
| 293 |
+ |
|
| 294 |
+type executeAsmContext struct {
|
|
| 295 |
+ seqs []seqVals |
|
| 296 |
+ seqIndex int |
|
| 297 |
+ out []byte |
|
| 298 |
+ history []byte |
|
| 299 |
+ literals []byte |
|
| 300 |
+ outPosition int |
|
| 301 |
+ litPosition int |
|
| 302 |
+ windowSize int |
|
| 303 |
+} |
|
| 304 |
+ |
|
| 305 |
+// sequenceDecs_executeSimple_amd64 implements the main loop of sequenceDecs.executeSimple in x86 asm. |
|
| 306 |
+// |
|
| 307 |
+// Returns false if a match offset is too big. |
|
| 308 |
+// |
|
| 309 |
+// Please refer to seqdec_generic.go for the reference implementation. |
|
| 310 |
+//go:noescape |
|
| 311 |
+func sequenceDecs_executeSimple_amd64(ctx *executeAsmContext) bool |
|
| 312 |
+ |
|
| 313 |
+// Same as above, but with safe memcopies |
|
| 314 |
+//go:noescape |
|
| 315 |
+func sequenceDecs_executeSimple_safe_amd64(ctx *executeAsmContext) bool |
|
| 316 |
+ |
|
| 317 |
+// executeSimple handles cases when dictionary is not used. |
|
| 318 |
+func (s *sequenceDecs) executeSimple(seqs []seqVals, hist []byte) error {
|
|
| 319 |
+ // Ensure we have enough output size... |
|
| 320 |
+ if len(s.out)+s.seqSize+compressedBlockOverAlloc > cap(s.out) {
|
|
| 321 |
+ addBytes := s.seqSize + len(s.out) + compressedBlockOverAlloc |
|
| 322 |
+ s.out = append(s.out, make([]byte, addBytes)...) |
|
| 323 |
+ s.out = s.out[:len(s.out)-addBytes] |
|
| 324 |
+ } |
|
| 325 |
+ |
|
| 326 |
+ if debugDecoder {
|
|
| 327 |
+ printf("Execute %d seqs with literals: %d into %d bytes\n", len(seqs), len(s.literals), s.seqSize)
|
|
| 328 |
+ } |
|
| 329 |
+ |
|
| 330 |
+ var t = len(s.out) |
|
| 331 |
+ out := s.out[:t+s.seqSize] |
|
| 332 |
+ |
|
| 333 |
+ ctx := executeAsmContext{
|
|
| 334 |
+ seqs: seqs, |
|
| 335 |
+ seqIndex: 0, |
|
| 336 |
+ out: out, |
|
| 337 |
+ history: hist, |
|
| 338 |
+ outPosition: t, |
|
| 339 |
+ litPosition: 0, |
|
| 340 |
+ literals: s.literals, |
|
| 341 |
+ windowSize: s.windowSize, |
|
| 342 |
+ } |
|
| 343 |
+ var ok bool |
|
| 344 |
+ if cap(s.literals) < len(s.literals)+compressedBlockOverAlloc {
|
|
| 345 |
+ ok = sequenceDecs_executeSimple_safe_amd64(&ctx) |
|
| 346 |
+ } else {
|
|
| 347 |
+ ok = sequenceDecs_executeSimple_amd64(&ctx) |
|
| 348 |
+ } |
|
| 349 |
+ if !ok {
|
|
| 350 |
+ return fmt.Errorf("match offset (%d) bigger than current history (%d)",
|
|
| 351 |
+ seqs[ctx.seqIndex].mo, ctx.outPosition+len(hist)) |
|
| 352 |
+ } |
|
| 353 |
+ s.literals = s.literals[ctx.litPosition:] |
|
| 354 |
+ t = ctx.outPosition |
|
| 355 |
+ |
|
| 356 |
+ // Add final literals |
|
| 357 |
+ copy(out[t:], s.literals) |
|
| 358 |
+ if debugDecoder {
|
|
| 359 |
+ t += len(s.literals) |
|
| 360 |
+ if t != len(out) {
|
|
| 361 |
+ panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize))
|
|
| 362 |
+ } |
|
| 363 |
+ } |
|
| 364 |
+ s.out = out |
|
| 365 |
+ |
|
| 366 |
+ return nil |
|
| 367 |
+} |
| 0 | 368 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,4100 @@ |
| 0 |
+// Code generated by command: go run gen.go -out ../seqdec_amd64.s -pkg=zstd. DO NOT EDIT. |
|
| 1 |
+ |
|
| 2 |
+//go:build !appengine && !noasm && gc && !noasm |
|
| 3 |
+// +build !appengine,!noasm,gc,!noasm |
|
| 4 |
+ |
|
| 5 |
+// func sequenceDecs_decode_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 6 |
+// Requires: CMOV |
|
| 7 |
+TEXT ·sequenceDecs_decode_amd64(SB), $8-32 |
|
| 8 |
+ MOVQ br+8(FP), AX |
|
| 9 |
+ MOVQ 32(AX), DX |
|
| 10 |
+ MOVBQZX 40(AX), BX |
|
| 11 |
+ MOVQ 24(AX), SI |
|
| 12 |
+ MOVQ (AX), AX |
|
| 13 |
+ ADDQ SI, AX |
|
| 14 |
+ MOVQ AX, (SP) |
|
| 15 |
+ MOVQ ctx+16(FP), AX |
|
| 16 |
+ MOVQ 72(AX), DI |
|
| 17 |
+ MOVQ 80(AX), R8 |
|
| 18 |
+ MOVQ 88(AX), R9 |
|
| 19 |
+ MOVQ 104(AX), R10 |
|
| 20 |
+ MOVQ s+0(FP), AX |
|
| 21 |
+ MOVQ 144(AX), R11 |
|
| 22 |
+ MOVQ 152(AX), R12 |
|
| 23 |
+ MOVQ 160(AX), R13 |
|
| 24 |
+ |
|
| 25 |
+sequenceDecs_decode_amd64_main_loop: |
|
| 26 |
+ MOVQ (SP), R14 |
|
| 27 |
+ |
|
| 28 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 29 |
+ CMPQ SI, $0x08 |
|
| 30 |
+ JL sequenceDecs_decode_amd64_fill_byte_by_byte |
|
| 31 |
+ MOVQ BX, AX |
|
| 32 |
+ SHRQ $0x03, AX |
|
| 33 |
+ SUBQ AX, R14 |
|
| 34 |
+ MOVQ (R14), DX |
|
| 35 |
+ SUBQ AX, SI |
|
| 36 |
+ ANDQ $0x07, BX |
|
| 37 |
+ JMP sequenceDecs_decode_amd64_fill_end |
|
| 38 |
+ |
|
| 39 |
+sequenceDecs_decode_amd64_fill_byte_by_byte: |
|
| 40 |
+ CMPQ SI, $0x00 |
|
| 41 |
+ JLE sequenceDecs_decode_amd64_fill_end |
|
| 42 |
+ CMPQ BX, $0x07 |
|
| 43 |
+ JLE sequenceDecs_decode_amd64_fill_end |
|
| 44 |
+ SHLQ $0x08, DX |
|
| 45 |
+ SUBQ $0x01, R14 |
|
| 46 |
+ SUBQ $0x01, SI |
|
| 47 |
+ SUBQ $0x08, BX |
|
| 48 |
+ MOVBQZX (R14), AX |
|
| 49 |
+ ORQ AX, DX |
|
| 50 |
+ JMP sequenceDecs_decode_amd64_fill_byte_by_byte |
|
| 51 |
+ |
|
| 52 |
+sequenceDecs_decode_amd64_fill_end: |
|
| 53 |
+ // Update offset |
|
| 54 |
+ MOVQ R9, AX |
|
| 55 |
+ MOVQ BX, CX |
|
| 56 |
+ MOVQ DX, R15 |
|
| 57 |
+ SHLQ CL, R15 |
|
| 58 |
+ MOVB AH, CL |
|
| 59 |
+ SHRQ $0x20, AX |
|
| 60 |
+ TESTQ CX, CX |
|
| 61 |
+ JZ sequenceDecs_decode_amd64_of_update_zero |
|
| 62 |
+ ADDQ CX, BX |
|
| 63 |
+ CMPQ BX, $0x40 |
|
| 64 |
+ JA sequenceDecs_decode_amd64_of_update_zero |
|
| 65 |
+ CMPQ CX, $0x40 |
|
| 66 |
+ JAE sequenceDecs_decode_amd64_of_update_zero |
|
| 67 |
+ NEGQ CX |
|
| 68 |
+ SHRQ CL, R15 |
|
| 69 |
+ ADDQ R15, AX |
|
| 70 |
+ |
|
| 71 |
+sequenceDecs_decode_amd64_of_update_zero: |
|
| 72 |
+ MOVQ AX, 16(R10) |
|
| 73 |
+ |
|
| 74 |
+ // Update match length |
|
| 75 |
+ MOVQ R8, AX |
|
| 76 |
+ MOVQ BX, CX |
|
| 77 |
+ MOVQ DX, R15 |
|
| 78 |
+ SHLQ CL, R15 |
|
| 79 |
+ MOVB AH, CL |
|
| 80 |
+ SHRQ $0x20, AX |
|
| 81 |
+ TESTQ CX, CX |
|
| 82 |
+ JZ sequenceDecs_decode_amd64_ml_update_zero |
|
| 83 |
+ ADDQ CX, BX |
|
| 84 |
+ CMPQ BX, $0x40 |
|
| 85 |
+ JA sequenceDecs_decode_amd64_ml_update_zero |
|
| 86 |
+ CMPQ CX, $0x40 |
|
| 87 |
+ JAE sequenceDecs_decode_amd64_ml_update_zero |
|
| 88 |
+ NEGQ CX |
|
| 89 |
+ SHRQ CL, R15 |
|
| 90 |
+ ADDQ R15, AX |
|
| 91 |
+ |
|
| 92 |
+sequenceDecs_decode_amd64_ml_update_zero: |
|
| 93 |
+ MOVQ AX, 8(R10) |
|
| 94 |
+ |
|
| 95 |
+ // Fill bitreader to have enough for the remaining |
|
| 96 |
+ CMPQ SI, $0x08 |
|
| 97 |
+ JL sequenceDecs_decode_amd64_fill_2_byte_by_byte |
|
| 98 |
+ MOVQ BX, AX |
|
| 99 |
+ SHRQ $0x03, AX |
|
| 100 |
+ SUBQ AX, R14 |
|
| 101 |
+ MOVQ (R14), DX |
|
| 102 |
+ SUBQ AX, SI |
|
| 103 |
+ ANDQ $0x07, BX |
|
| 104 |
+ JMP sequenceDecs_decode_amd64_fill_2_end |
|
| 105 |
+ |
|
| 106 |
+sequenceDecs_decode_amd64_fill_2_byte_by_byte: |
|
| 107 |
+ CMPQ SI, $0x00 |
|
| 108 |
+ JLE sequenceDecs_decode_amd64_fill_2_end |
|
| 109 |
+ CMPQ BX, $0x07 |
|
| 110 |
+ JLE sequenceDecs_decode_amd64_fill_2_end |
|
| 111 |
+ SHLQ $0x08, DX |
|
| 112 |
+ SUBQ $0x01, R14 |
|
| 113 |
+ SUBQ $0x01, SI |
|
| 114 |
+ SUBQ $0x08, BX |
|
| 115 |
+ MOVBQZX (R14), AX |
|
| 116 |
+ ORQ AX, DX |
|
| 117 |
+ JMP sequenceDecs_decode_amd64_fill_2_byte_by_byte |
|
| 118 |
+ |
|
| 119 |
+sequenceDecs_decode_amd64_fill_2_end: |
|
| 120 |
+ // Update literal length |
|
| 121 |
+ MOVQ DI, AX |
|
| 122 |
+ MOVQ BX, CX |
|
| 123 |
+ MOVQ DX, R15 |
|
| 124 |
+ SHLQ CL, R15 |
|
| 125 |
+ MOVB AH, CL |
|
| 126 |
+ SHRQ $0x20, AX |
|
| 127 |
+ TESTQ CX, CX |
|
| 128 |
+ JZ sequenceDecs_decode_amd64_ll_update_zero |
|
| 129 |
+ ADDQ CX, BX |
|
| 130 |
+ CMPQ BX, $0x40 |
|
| 131 |
+ JA sequenceDecs_decode_amd64_ll_update_zero |
|
| 132 |
+ CMPQ CX, $0x40 |
|
| 133 |
+ JAE sequenceDecs_decode_amd64_ll_update_zero |
|
| 134 |
+ NEGQ CX |
|
| 135 |
+ SHRQ CL, R15 |
|
| 136 |
+ ADDQ R15, AX |
|
| 137 |
+ |
|
| 138 |
+sequenceDecs_decode_amd64_ll_update_zero: |
|
| 139 |
+ MOVQ AX, (R10) |
|
| 140 |
+ |
|
| 141 |
+ // Fill bitreader for state updates |
|
| 142 |
+ MOVQ R14, (SP) |
|
| 143 |
+ MOVQ R9, AX |
|
| 144 |
+ SHRQ $0x08, AX |
|
| 145 |
+ MOVBQZX AL, AX |
|
| 146 |
+ MOVQ ctx+16(FP), CX |
|
| 147 |
+ CMPQ 96(CX), $0x00 |
|
| 148 |
+ JZ sequenceDecs_decode_amd64_skip_update |
|
| 149 |
+ |
|
| 150 |
+ // Update Literal Length State |
|
| 151 |
+ MOVBQZX DI, R14 |
|
| 152 |
+ SHRQ $0x10, DI |
|
| 153 |
+ MOVWQZX DI, DI |
|
| 154 |
+ LEAQ (BX)(R14*1), CX |
|
| 155 |
+ MOVQ DX, R15 |
|
| 156 |
+ MOVQ CX, BX |
|
| 157 |
+ ROLQ CL, R15 |
|
| 158 |
+ MOVL $0x00000001, BP |
|
| 159 |
+ MOVB R14, CL |
|
| 160 |
+ SHLL CL, BP |
|
| 161 |
+ DECL BP |
|
| 162 |
+ ANDQ BP, R15 |
|
| 163 |
+ ADDQ R15, DI |
|
| 164 |
+ |
|
| 165 |
+ // Load ctx.llTable |
|
| 166 |
+ MOVQ ctx+16(FP), CX |
|
| 167 |
+ MOVQ (CX), CX |
|
| 168 |
+ MOVQ (CX)(DI*8), DI |
|
| 169 |
+ |
|
| 170 |
+ // Update Match Length State |
|
| 171 |
+ MOVBQZX R8, R14 |
|
| 172 |
+ SHRQ $0x10, R8 |
|
| 173 |
+ MOVWQZX R8, R8 |
|
| 174 |
+ LEAQ (BX)(R14*1), CX |
|
| 175 |
+ MOVQ DX, R15 |
|
| 176 |
+ MOVQ CX, BX |
|
| 177 |
+ ROLQ CL, R15 |
|
| 178 |
+ MOVL $0x00000001, BP |
|
| 179 |
+ MOVB R14, CL |
|
| 180 |
+ SHLL CL, BP |
|
| 181 |
+ DECL BP |
|
| 182 |
+ ANDQ BP, R15 |
|
| 183 |
+ ADDQ R15, R8 |
|
| 184 |
+ |
|
| 185 |
+ // Load ctx.mlTable |
|
| 186 |
+ MOVQ ctx+16(FP), CX |
|
| 187 |
+ MOVQ 24(CX), CX |
|
| 188 |
+ MOVQ (CX)(R8*8), R8 |
|
| 189 |
+ |
|
| 190 |
+ // Update Offset State |
|
| 191 |
+ MOVBQZX R9, R14 |
|
| 192 |
+ SHRQ $0x10, R9 |
|
| 193 |
+ MOVWQZX R9, R9 |
|
| 194 |
+ LEAQ (BX)(R14*1), CX |
|
| 195 |
+ MOVQ DX, R15 |
|
| 196 |
+ MOVQ CX, BX |
|
| 197 |
+ ROLQ CL, R15 |
|
| 198 |
+ MOVL $0x00000001, BP |
|
| 199 |
+ MOVB R14, CL |
|
| 200 |
+ SHLL CL, BP |
|
| 201 |
+ DECL BP |
|
| 202 |
+ ANDQ BP, R15 |
|
| 203 |
+ ADDQ R15, R9 |
|
| 204 |
+ |
|
| 205 |
+ // Load ctx.ofTable |
|
| 206 |
+ MOVQ ctx+16(FP), CX |
|
| 207 |
+ MOVQ 48(CX), CX |
|
| 208 |
+ MOVQ (CX)(R9*8), R9 |
|
| 209 |
+ |
|
| 210 |
+sequenceDecs_decode_amd64_skip_update: |
|
| 211 |
+ // Adjust offset |
|
| 212 |
+ MOVQ 16(R10), CX |
|
| 213 |
+ CMPQ AX, $0x01 |
|
| 214 |
+ JBE sequenceDecs_decode_amd64_adjust_offsetB_1_or_0 |
|
| 215 |
+ MOVQ R12, R13 |
|
| 216 |
+ MOVQ R11, R12 |
|
| 217 |
+ MOVQ CX, R11 |
|
| 218 |
+ JMP sequenceDecs_decode_amd64_after_adjust |
|
| 219 |
+ |
|
| 220 |
+sequenceDecs_decode_amd64_adjust_offsetB_1_or_0: |
|
| 221 |
+ CMPQ (R10), $0x00000000 |
|
| 222 |
+ JNE sequenceDecs_decode_amd64_adjust_offset_maybezero |
|
| 223 |
+ INCQ CX |
|
| 224 |
+ JMP sequenceDecs_decode_amd64_adjust_offset_nonzero |
|
| 225 |
+ |
|
| 226 |
+sequenceDecs_decode_amd64_adjust_offset_maybezero: |
|
| 227 |
+ TESTQ CX, CX |
|
| 228 |
+ JNZ sequenceDecs_decode_amd64_adjust_offset_nonzero |
|
| 229 |
+ MOVQ R11, CX |
|
| 230 |
+ JMP sequenceDecs_decode_amd64_after_adjust |
|
| 231 |
+ |
|
| 232 |
+sequenceDecs_decode_amd64_adjust_offset_nonzero: |
|
| 233 |
+ CMPQ CX, $0x01 |
|
| 234 |
+ JB sequenceDecs_decode_amd64_adjust_zero |
|
| 235 |
+ JEQ sequenceDecs_decode_amd64_adjust_one |
|
| 236 |
+ CMPQ CX, $0x02 |
|
| 237 |
+ JA sequenceDecs_decode_amd64_adjust_three |
|
| 238 |
+ JMP sequenceDecs_decode_amd64_adjust_two |
|
| 239 |
+ |
|
| 240 |
+sequenceDecs_decode_amd64_adjust_zero: |
|
| 241 |
+ MOVQ R11, AX |
|
| 242 |
+ JMP sequenceDecs_decode_amd64_adjust_test_temp_valid |
|
| 243 |
+ |
|
| 244 |
+sequenceDecs_decode_amd64_adjust_one: |
|
| 245 |
+ MOVQ R12, AX |
|
| 246 |
+ JMP sequenceDecs_decode_amd64_adjust_test_temp_valid |
|
| 247 |
+ |
|
| 248 |
+sequenceDecs_decode_amd64_adjust_two: |
|
| 249 |
+ MOVQ R13, AX |
|
| 250 |
+ JMP sequenceDecs_decode_amd64_adjust_test_temp_valid |
|
| 251 |
+ |
|
| 252 |
+sequenceDecs_decode_amd64_adjust_three: |
|
| 253 |
+ LEAQ -1(R11), AX |
|
| 254 |
+ |
|
| 255 |
+sequenceDecs_decode_amd64_adjust_test_temp_valid: |
|
| 256 |
+ TESTQ AX, AX |
|
| 257 |
+ JNZ sequenceDecs_decode_amd64_adjust_temp_valid |
|
| 258 |
+ MOVQ $0x00000001, AX |
|
| 259 |
+ |
|
| 260 |
+sequenceDecs_decode_amd64_adjust_temp_valid: |
|
| 261 |
+ CMPQ CX, $0x01 |
|
| 262 |
+ CMOVQNE R12, R13 |
|
| 263 |
+ MOVQ R11, R12 |
|
| 264 |
+ MOVQ AX, R11 |
|
| 265 |
+ MOVQ AX, CX |
|
| 266 |
+ |
|
| 267 |
+sequenceDecs_decode_amd64_after_adjust: |
|
| 268 |
+ MOVQ CX, 16(R10) |
|
| 269 |
+ |
|
| 270 |
+ // Check values |
|
| 271 |
+ MOVQ 8(R10), AX |
|
| 272 |
+ MOVQ (R10), R14 |
|
| 273 |
+ LEAQ (AX)(R14*1), R15 |
|
| 274 |
+ MOVQ s+0(FP), BP |
|
| 275 |
+ ADDQ R15, 256(BP) |
|
| 276 |
+ MOVQ ctx+16(FP), R15 |
|
| 277 |
+ SUBQ R14, 128(R15) |
|
| 278 |
+ JS error_not_enough_literals |
|
| 279 |
+ CMPQ AX, $0x00020002 |
|
| 280 |
+ JA sequenceDecs_decode_amd64_error_match_len_too_big |
|
| 281 |
+ TESTQ CX, CX |
|
| 282 |
+ JNZ sequenceDecs_decode_amd64_match_len_ofs_ok |
|
| 283 |
+ TESTQ AX, AX |
|
| 284 |
+ JNZ sequenceDecs_decode_amd64_error_match_len_ofs_mismatch |
|
| 285 |
+ |
|
| 286 |
+sequenceDecs_decode_amd64_match_len_ofs_ok: |
|
| 287 |
+ ADDQ $0x18, R10 |
|
| 288 |
+ MOVQ ctx+16(FP), AX |
|
| 289 |
+ DECQ 96(AX) |
|
| 290 |
+ JNS sequenceDecs_decode_amd64_main_loop |
|
| 291 |
+ MOVQ s+0(FP), AX |
|
| 292 |
+ MOVQ R11, 144(AX) |
|
| 293 |
+ MOVQ R12, 152(AX) |
|
| 294 |
+ MOVQ R13, 160(AX) |
|
| 295 |
+ MOVQ br+8(FP), AX |
|
| 296 |
+ MOVQ DX, 32(AX) |
|
| 297 |
+ MOVB BL, 40(AX) |
|
| 298 |
+ MOVQ SI, 24(AX) |
|
| 299 |
+ |
|
| 300 |
+ // Return success |
|
| 301 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 302 |
+ RET |
|
| 303 |
+ |
|
| 304 |
+ // Return with match length error |
|
| 305 |
+sequenceDecs_decode_amd64_error_match_len_ofs_mismatch: |
|
| 306 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 307 |
+ RET |
|
| 308 |
+ |
|
| 309 |
+ // Return with match too long error |
|
| 310 |
+sequenceDecs_decode_amd64_error_match_len_too_big: |
|
| 311 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 312 |
+ RET |
|
| 313 |
+ |
|
| 314 |
+ // Return with match offset too long error |
|
| 315 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 316 |
+ RET |
|
| 317 |
+ |
|
| 318 |
+ // Return with not enough literals error |
|
| 319 |
+error_not_enough_literals: |
|
| 320 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 321 |
+ RET |
|
| 322 |
+ |
|
| 323 |
+ // Return with not enough output space error |
|
| 324 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 325 |
+ RET |
|
| 326 |
+ |
|
| 327 |
+// func sequenceDecs_decode_56_amd64(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 328 |
+// Requires: CMOV |
|
| 329 |
+TEXT ·sequenceDecs_decode_56_amd64(SB), $8-32 |
|
| 330 |
+ MOVQ br+8(FP), AX |
|
| 331 |
+ MOVQ 32(AX), DX |
|
| 332 |
+ MOVBQZX 40(AX), BX |
|
| 333 |
+ MOVQ 24(AX), SI |
|
| 334 |
+ MOVQ (AX), AX |
|
| 335 |
+ ADDQ SI, AX |
|
| 336 |
+ MOVQ AX, (SP) |
|
| 337 |
+ MOVQ ctx+16(FP), AX |
|
| 338 |
+ MOVQ 72(AX), DI |
|
| 339 |
+ MOVQ 80(AX), R8 |
|
| 340 |
+ MOVQ 88(AX), R9 |
|
| 341 |
+ MOVQ 104(AX), R10 |
|
| 342 |
+ MOVQ s+0(FP), AX |
|
| 343 |
+ MOVQ 144(AX), R11 |
|
| 344 |
+ MOVQ 152(AX), R12 |
|
| 345 |
+ MOVQ 160(AX), R13 |
|
| 346 |
+ |
|
| 347 |
+sequenceDecs_decode_56_amd64_main_loop: |
|
| 348 |
+ MOVQ (SP), R14 |
|
| 349 |
+ |
|
| 350 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 351 |
+ CMPQ SI, $0x08 |
|
| 352 |
+ JL sequenceDecs_decode_56_amd64_fill_byte_by_byte |
|
| 353 |
+ MOVQ BX, AX |
|
| 354 |
+ SHRQ $0x03, AX |
|
| 355 |
+ SUBQ AX, R14 |
|
| 356 |
+ MOVQ (R14), DX |
|
| 357 |
+ SUBQ AX, SI |
|
| 358 |
+ ANDQ $0x07, BX |
|
| 359 |
+ JMP sequenceDecs_decode_56_amd64_fill_end |
|
| 360 |
+ |
|
| 361 |
+sequenceDecs_decode_56_amd64_fill_byte_by_byte: |
|
| 362 |
+ CMPQ SI, $0x00 |
|
| 363 |
+ JLE sequenceDecs_decode_56_amd64_fill_end |
|
| 364 |
+ CMPQ BX, $0x07 |
|
| 365 |
+ JLE sequenceDecs_decode_56_amd64_fill_end |
|
| 366 |
+ SHLQ $0x08, DX |
|
| 367 |
+ SUBQ $0x01, R14 |
|
| 368 |
+ SUBQ $0x01, SI |
|
| 369 |
+ SUBQ $0x08, BX |
|
| 370 |
+ MOVBQZX (R14), AX |
|
| 371 |
+ ORQ AX, DX |
|
| 372 |
+ JMP sequenceDecs_decode_56_amd64_fill_byte_by_byte |
|
| 373 |
+ |
|
| 374 |
+sequenceDecs_decode_56_amd64_fill_end: |
|
| 375 |
+ // Update offset |
|
| 376 |
+ MOVQ R9, AX |
|
| 377 |
+ MOVQ BX, CX |
|
| 378 |
+ MOVQ DX, R15 |
|
| 379 |
+ SHLQ CL, R15 |
|
| 380 |
+ MOVB AH, CL |
|
| 381 |
+ SHRQ $0x20, AX |
|
| 382 |
+ TESTQ CX, CX |
|
| 383 |
+ JZ sequenceDecs_decode_56_amd64_of_update_zero |
|
| 384 |
+ ADDQ CX, BX |
|
| 385 |
+ CMPQ BX, $0x40 |
|
| 386 |
+ JA sequenceDecs_decode_56_amd64_of_update_zero |
|
| 387 |
+ CMPQ CX, $0x40 |
|
| 388 |
+ JAE sequenceDecs_decode_56_amd64_of_update_zero |
|
| 389 |
+ NEGQ CX |
|
| 390 |
+ SHRQ CL, R15 |
|
| 391 |
+ ADDQ R15, AX |
|
| 392 |
+ |
|
| 393 |
+sequenceDecs_decode_56_amd64_of_update_zero: |
|
| 394 |
+ MOVQ AX, 16(R10) |
|
| 395 |
+ |
|
| 396 |
+ // Update match length |
|
| 397 |
+ MOVQ R8, AX |
|
| 398 |
+ MOVQ BX, CX |
|
| 399 |
+ MOVQ DX, R15 |
|
| 400 |
+ SHLQ CL, R15 |
|
| 401 |
+ MOVB AH, CL |
|
| 402 |
+ SHRQ $0x20, AX |
|
| 403 |
+ TESTQ CX, CX |
|
| 404 |
+ JZ sequenceDecs_decode_56_amd64_ml_update_zero |
|
| 405 |
+ ADDQ CX, BX |
|
| 406 |
+ CMPQ BX, $0x40 |
|
| 407 |
+ JA sequenceDecs_decode_56_amd64_ml_update_zero |
|
| 408 |
+ CMPQ CX, $0x40 |
|
| 409 |
+ JAE sequenceDecs_decode_56_amd64_ml_update_zero |
|
| 410 |
+ NEGQ CX |
|
| 411 |
+ SHRQ CL, R15 |
|
| 412 |
+ ADDQ R15, AX |
|
| 413 |
+ |
|
| 414 |
+sequenceDecs_decode_56_amd64_ml_update_zero: |
|
| 415 |
+ MOVQ AX, 8(R10) |
|
| 416 |
+ |
|
| 417 |
+ // Update literal length |
|
| 418 |
+ MOVQ DI, AX |
|
| 419 |
+ MOVQ BX, CX |
|
| 420 |
+ MOVQ DX, R15 |
|
| 421 |
+ SHLQ CL, R15 |
|
| 422 |
+ MOVB AH, CL |
|
| 423 |
+ SHRQ $0x20, AX |
|
| 424 |
+ TESTQ CX, CX |
|
| 425 |
+ JZ sequenceDecs_decode_56_amd64_ll_update_zero |
|
| 426 |
+ ADDQ CX, BX |
|
| 427 |
+ CMPQ BX, $0x40 |
|
| 428 |
+ JA sequenceDecs_decode_56_amd64_ll_update_zero |
|
| 429 |
+ CMPQ CX, $0x40 |
|
| 430 |
+ JAE sequenceDecs_decode_56_amd64_ll_update_zero |
|
| 431 |
+ NEGQ CX |
|
| 432 |
+ SHRQ CL, R15 |
|
| 433 |
+ ADDQ R15, AX |
|
| 434 |
+ |
|
| 435 |
+sequenceDecs_decode_56_amd64_ll_update_zero: |
|
| 436 |
+ MOVQ AX, (R10) |
|
| 437 |
+ |
|
| 438 |
+ // Fill bitreader for state updates |
|
| 439 |
+ MOVQ R14, (SP) |
|
| 440 |
+ MOVQ R9, AX |
|
| 441 |
+ SHRQ $0x08, AX |
|
| 442 |
+ MOVBQZX AL, AX |
|
| 443 |
+ MOVQ ctx+16(FP), CX |
|
| 444 |
+ CMPQ 96(CX), $0x00 |
|
| 445 |
+ JZ sequenceDecs_decode_56_amd64_skip_update |
|
| 446 |
+ |
|
| 447 |
+ // Update Literal Length State |
|
| 448 |
+ MOVBQZX DI, R14 |
|
| 449 |
+ SHRQ $0x10, DI |
|
| 450 |
+ MOVWQZX DI, DI |
|
| 451 |
+ LEAQ (BX)(R14*1), CX |
|
| 452 |
+ MOVQ DX, R15 |
|
| 453 |
+ MOVQ CX, BX |
|
| 454 |
+ ROLQ CL, R15 |
|
| 455 |
+ MOVL $0x00000001, BP |
|
| 456 |
+ MOVB R14, CL |
|
| 457 |
+ SHLL CL, BP |
|
| 458 |
+ DECL BP |
|
| 459 |
+ ANDQ BP, R15 |
|
| 460 |
+ ADDQ R15, DI |
|
| 461 |
+ |
|
| 462 |
+ // Load ctx.llTable |
|
| 463 |
+ MOVQ ctx+16(FP), CX |
|
| 464 |
+ MOVQ (CX), CX |
|
| 465 |
+ MOVQ (CX)(DI*8), DI |
|
| 466 |
+ |
|
| 467 |
+ // Update Match Length State |
|
| 468 |
+ MOVBQZX R8, R14 |
|
| 469 |
+ SHRQ $0x10, R8 |
|
| 470 |
+ MOVWQZX R8, R8 |
|
| 471 |
+ LEAQ (BX)(R14*1), CX |
|
| 472 |
+ MOVQ DX, R15 |
|
| 473 |
+ MOVQ CX, BX |
|
| 474 |
+ ROLQ CL, R15 |
|
| 475 |
+ MOVL $0x00000001, BP |
|
| 476 |
+ MOVB R14, CL |
|
| 477 |
+ SHLL CL, BP |
|
| 478 |
+ DECL BP |
|
| 479 |
+ ANDQ BP, R15 |
|
| 480 |
+ ADDQ R15, R8 |
|
| 481 |
+ |
|
| 482 |
+ // Load ctx.mlTable |
|
| 483 |
+ MOVQ ctx+16(FP), CX |
|
| 484 |
+ MOVQ 24(CX), CX |
|
| 485 |
+ MOVQ (CX)(R8*8), R8 |
|
| 486 |
+ |
|
| 487 |
+ // Update Offset State |
|
| 488 |
+ MOVBQZX R9, R14 |
|
| 489 |
+ SHRQ $0x10, R9 |
|
| 490 |
+ MOVWQZX R9, R9 |
|
| 491 |
+ LEAQ (BX)(R14*1), CX |
|
| 492 |
+ MOVQ DX, R15 |
|
| 493 |
+ MOVQ CX, BX |
|
| 494 |
+ ROLQ CL, R15 |
|
| 495 |
+ MOVL $0x00000001, BP |
|
| 496 |
+ MOVB R14, CL |
|
| 497 |
+ SHLL CL, BP |
|
| 498 |
+ DECL BP |
|
| 499 |
+ ANDQ BP, R15 |
|
| 500 |
+ ADDQ R15, R9 |
|
| 501 |
+ |
|
| 502 |
+ // Load ctx.ofTable |
|
| 503 |
+ MOVQ ctx+16(FP), CX |
|
| 504 |
+ MOVQ 48(CX), CX |
|
| 505 |
+ MOVQ (CX)(R9*8), R9 |
|
| 506 |
+ |
|
| 507 |
+sequenceDecs_decode_56_amd64_skip_update: |
|
| 508 |
+ // Adjust offset |
|
| 509 |
+ MOVQ 16(R10), CX |
|
| 510 |
+ CMPQ AX, $0x01 |
|
| 511 |
+ JBE sequenceDecs_decode_56_amd64_adjust_offsetB_1_or_0 |
|
| 512 |
+ MOVQ R12, R13 |
|
| 513 |
+ MOVQ R11, R12 |
|
| 514 |
+ MOVQ CX, R11 |
|
| 515 |
+ JMP sequenceDecs_decode_56_amd64_after_adjust |
|
| 516 |
+ |
|
| 517 |
+sequenceDecs_decode_56_amd64_adjust_offsetB_1_or_0: |
|
| 518 |
+ CMPQ (R10), $0x00000000 |
|
| 519 |
+ JNE sequenceDecs_decode_56_amd64_adjust_offset_maybezero |
|
| 520 |
+ INCQ CX |
|
| 521 |
+ JMP sequenceDecs_decode_56_amd64_adjust_offset_nonzero |
|
| 522 |
+ |
|
| 523 |
+sequenceDecs_decode_56_amd64_adjust_offset_maybezero: |
|
| 524 |
+ TESTQ CX, CX |
|
| 525 |
+ JNZ sequenceDecs_decode_56_amd64_adjust_offset_nonzero |
|
| 526 |
+ MOVQ R11, CX |
|
| 527 |
+ JMP sequenceDecs_decode_56_amd64_after_adjust |
|
| 528 |
+ |
|
| 529 |
+sequenceDecs_decode_56_amd64_adjust_offset_nonzero: |
|
| 530 |
+ CMPQ CX, $0x01 |
|
| 531 |
+ JB sequenceDecs_decode_56_amd64_adjust_zero |
|
| 532 |
+ JEQ sequenceDecs_decode_56_amd64_adjust_one |
|
| 533 |
+ CMPQ CX, $0x02 |
|
| 534 |
+ JA sequenceDecs_decode_56_amd64_adjust_three |
|
| 535 |
+ JMP sequenceDecs_decode_56_amd64_adjust_two |
|
| 536 |
+ |
|
| 537 |
+sequenceDecs_decode_56_amd64_adjust_zero: |
|
| 538 |
+ MOVQ R11, AX |
|
| 539 |
+ JMP sequenceDecs_decode_56_amd64_adjust_test_temp_valid |
|
| 540 |
+ |
|
| 541 |
+sequenceDecs_decode_56_amd64_adjust_one: |
|
| 542 |
+ MOVQ R12, AX |
|
| 543 |
+ JMP sequenceDecs_decode_56_amd64_adjust_test_temp_valid |
|
| 544 |
+ |
|
| 545 |
+sequenceDecs_decode_56_amd64_adjust_two: |
|
| 546 |
+ MOVQ R13, AX |
|
| 547 |
+ JMP sequenceDecs_decode_56_amd64_adjust_test_temp_valid |
|
| 548 |
+ |
|
| 549 |
+sequenceDecs_decode_56_amd64_adjust_three: |
|
| 550 |
+ LEAQ -1(R11), AX |
|
| 551 |
+ |
|
| 552 |
+sequenceDecs_decode_56_amd64_adjust_test_temp_valid: |
|
| 553 |
+ TESTQ AX, AX |
|
| 554 |
+ JNZ sequenceDecs_decode_56_amd64_adjust_temp_valid |
|
| 555 |
+ MOVQ $0x00000001, AX |
|
| 556 |
+ |
|
| 557 |
+sequenceDecs_decode_56_amd64_adjust_temp_valid: |
|
| 558 |
+ CMPQ CX, $0x01 |
|
| 559 |
+ CMOVQNE R12, R13 |
|
| 560 |
+ MOVQ R11, R12 |
|
| 561 |
+ MOVQ AX, R11 |
|
| 562 |
+ MOVQ AX, CX |
|
| 563 |
+ |
|
| 564 |
+sequenceDecs_decode_56_amd64_after_adjust: |
|
| 565 |
+ MOVQ CX, 16(R10) |
|
| 566 |
+ |
|
| 567 |
+ // Check values |
|
| 568 |
+ MOVQ 8(R10), AX |
|
| 569 |
+ MOVQ (R10), R14 |
|
| 570 |
+ LEAQ (AX)(R14*1), R15 |
|
| 571 |
+ MOVQ s+0(FP), BP |
|
| 572 |
+ ADDQ R15, 256(BP) |
|
| 573 |
+ MOVQ ctx+16(FP), R15 |
|
| 574 |
+ SUBQ R14, 128(R15) |
|
| 575 |
+ JS error_not_enough_literals |
|
| 576 |
+ CMPQ AX, $0x00020002 |
|
| 577 |
+ JA sequenceDecs_decode_56_amd64_error_match_len_too_big |
|
| 578 |
+ TESTQ CX, CX |
|
| 579 |
+ JNZ sequenceDecs_decode_56_amd64_match_len_ofs_ok |
|
| 580 |
+ TESTQ AX, AX |
|
| 581 |
+ JNZ sequenceDecs_decode_56_amd64_error_match_len_ofs_mismatch |
|
| 582 |
+ |
|
| 583 |
+sequenceDecs_decode_56_amd64_match_len_ofs_ok: |
|
| 584 |
+ ADDQ $0x18, R10 |
|
| 585 |
+ MOVQ ctx+16(FP), AX |
|
| 586 |
+ DECQ 96(AX) |
|
| 587 |
+ JNS sequenceDecs_decode_56_amd64_main_loop |
|
| 588 |
+ MOVQ s+0(FP), AX |
|
| 589 |
+ MOVQ R11, 144(AX) |
|
| 590 |
+ MOVQ R12, 152(AX) |
|
| 591 |
+ MOVQ R13, 160(AX) |
|
| 592 |
+ MOVQ br+8(FP), AX |
|
| 593 |
+ MOVQ DX, 32(AX) |
|
| 594 |
+ MOVB BL, 40(AX) |
|
| 595 |
+ MOVQ SI, 24(AX) |
|
| 596 |
+ |
|
| 597 |
+ // Return success |
|
| 598 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 599 |
+ RET |
|
| 600 |
+ |
|
| 601 |
+ // Return with match length error |
|
| 602 |
+sequenceDecs_decode_56_amd64_error_match_len_ofs_mismatch: |
|
| 603 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 604 |
+ RET |
|
| 605 |
+ |
|
| 606 |
+ // Return with match too long error |
|
| 607 |
+sequenceDecs_decode_56_amd64_error_match_len_too_big: |
|
| 608 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 609 |
+ RET |
|
| 610 |
+ |
|
| 611 |
+ // Return with match offset too long error |
|
| 612 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 613 |
+ RET |
|
| 614 |
+ |
|
| 615 |
+ // Return with not enough literals error |
|
| 616 |
+error_not_enough_literals: |
|
| 617 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 618 |
+ RET |
|
| 619 |
+ |
|
| 620 |
+ // Return with not enough output space error |
|
| 621 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 622 |
+ RET |
|
| 623 |
+ |
|
| 624 |
+// func sequenceDecs_decode_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 625 |
+// Requires: BMI, BMI2, CMOV |
|
| 626 |
+TEXT ·sequenceDecs_decode_bmi2(SB), $8-32 |
|
| 627 |
+ MOVQ br+8(FP), CX |
|
| 628 |
+ MOVQ 32(CX), AX |
|
| 629 |
+ MOVBQZX 40(CX), DX |
|
| 630 |
+ MOVQ 24(CX), BX |
|
| 631 |
+ MOVQ (CX), CX |
|
| 632 |
+ ADDQ BX, CX |
|
| 633 |
+ MOVQ CX, (SP) |
|
| 634 |
+ MOVQ ctx+16(FP), CX |
|
| 635 |
+ MOVQ 72(CX), SI |
|
| 636 |
+ MOVQ 80(CX), DI |
|
| 637 |
+ MOVQ 88(CX), R8 |
|
| 638 |
+ MOVQ 104(CX), R9 |
|
| 639 |
+ MOVQ s+0(FP), CX |
|
| 640 |
+ MOVQ 144(CX), R10 |
|
| 641 |
+ MOVQ 152(CX), R11 |
|
| 642 |
+ MOVQ 160(CX), R12 |
|
| 643 |
+ |
|
| 644 |
+sequenceDecs_decode_bmi2_main_loop: |
|
| 645 |
+ MOVQ (SP), R13 |
|
| 646 |
+ |
|
| 647 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 648 |
+ CMPQ BX, $0x08 |
|
| 649 |
+ JL sequenceDecs_decode_bmi2_fill_byte_by_byte |
|
| 650 |
+ MOVQ DX, CX |
|
| 651 |
+ SHRQ $0x03, CX |
|
| 652 |
+ SUBQ CX, R13 |
|
| 653 |
+ MOVQ (R13), AX |
|
| 654 |
+ SUBQ CX, BX |
|
| 655 |
+ ANDQ $0x07, DX |
|
| 656 |
+ JMP sequenceDecs_decode_bmi2_fill_end |
|
| 657 |
+ |
|
| 658 |
+sequenceDecs_decode_bmi2_fill_byte_by_byte: |
|
| 659 |
+ CMPQ BX, $0x00 |
|
| 660 |
+ JLE sequenceDecs_decode_bmi2_fill_end |
|
| 661 |
+ CMPQ DX, $0x07 |
|
| 662 |
+ JLE sequenceDecs_decode_bmi2_fill_end |
|
| 663 |
+ SHLQ $0x08, AX |
|
| 664 |
+ SUBQ $0x01, R13 |
|
| 665 |
+ SUBQ $0x01, BX |
|
| 666 |
+ SUBQ $0x08, DX |
|
| 667 |
+ MOVBQZX (R13), CX |
|
| 668 |
+ ORQ CX, AX |
|
| 669 |
+ JMP sequenceDecs_decode_bmi2_fill_byte_by_byte |
|
| 670 |
+ |
|
| 671 |
+sequenceDecs_decode_bmi2_fill_end: |
|
| 672 |
+ // Update offset |
|
| 673 |
+ MOVQ $0x00000808, CX |
|
| 674 |
+ BEXTRQ CX, R8, R14 |
|
| 675 |
+ MOVQ AX, R15 |
|
| 676 |
+ LEAQ (DX)(R14*1), CX |
|
| 677 |
+ ROLQ CL, R15 |
|
| 678 |
+ BZHIQ R14, R15, R15 |
|
| 679 |
+ MOVQ CX, DX |
|
| 680 |
+ MOVQ R8, CX |
|
| 681 |
+ SHRQ $0x20, CX |
|
| 682 |
+ ADDQ R15, CX |
|
| 683 |
+ MOVQ CX, 16(R9) |
|
| 684 |
+ |
|
| 685 |
+ // Update match length |
|
| 686 |
+ MOVQ $0x00000808, CX |
|
| 687 |
+ BEXTRQ CX, DI, R14 |
|
| 688 |
+ MOVQ AX, R15 |
|
| 689 |
+ LEAQ (DX)(R14*1), CX |
|
| 690 |
+ ROLQ CL, R15 |
|
| 691 |
+ BZHIQ R14, R15, R15 |
|
| 692 |
+ MOVQ CX, DX |
|
| 693 |
+ MOVQ DI, CX |
|
| 694 |
+ SHRQ $0x20, CX |
|
| 695 |
+ ADDQ R15, CX |
|
| 696 |
+ MOVQ CX, 8(R9) |
|
| 697 |
+ |
|
| 698 |
+ // Fill bitreader to have enough for the remaining |
|
| 699 |
+ CMPQ BX, $0x08 |
|
| 700 |
+ JL sequenceDecs_decode_bmi2_fill_2_byte_by_byte |
|
| 701 |
+ MOVQ DX, CX |
|
| 702 |
+ SHRQ $0x03, CX |
|
| 703 |
+ SUBQ CX, R13 |
|
| 704 |
+ MOVQ (R13), AX |
|
| 705 |
+ SUBQ CX, BX |
|
| 706 |
+ ANDQ $0x07, DX |
|
| 707 |
+ JMP sequenceDecs_decode_bmi2_fill_2_end |
|
| 708 |
+ |
|
| 709 |
+sequenceDecs_decode_bmi2_fill_2_byte_by_byte: |
|
| 710 |
+ CMPQ BX, $0x00 |
|
| 711 |
+ JLE sequenceDecs_decode_bmi2_fill_2_end |
|
| 712 |
+ CMPQ DX, $0x07 |
|
| 713 |
+ JLE sequenceDecs_decode_bmi2_fill_2_end |
|
| 714 |
+ SHLQ $0x08, AX |
|
| 715 |
+ SUBQ $0x01, R13 |
|
| 716 |
+ SUBQ $0x01, BX |
|
| 717 |
+ SUBQ $0x08, DX |
|
| 718 |
+ MOVBQZX (R13), CX |
|
| 719 |
+ ORQ CX, AX |
|
| 720 |
+ JMP sequenceDecs_decode_bmi2_fill_2_byte_by_byte |
|
| 721 |
+ |
|
| 722 |
+sequenceDecs_decode_bmi2_fill_2_end: |
|
| 723 |
+ // Update literal length |
|
| 724 |
+ MOVQ $0x00000808, CX |
|
| 725 |
+ BEXTRQ CX, SI, R14 |
|
| 726 |
+ MOVQ AX, R15 |
|
| 727 |
+ LEAQ (DX)(R14*1), CX |
|
| 728 |
+ ROLQ CL, R15 |
|
| 729 |
+ BZHIQ R14, R15, R15 |
|
| 730 |
+ MOVQ CX, DX |
|
| 731 |
+ MOVQ SI, CX |
|
| 732 |
+ SHRQ $0x20, CX |
|
| 733 |
+ ADDQ R15, CX |
|
| 734 |
+ MOVQ CX, (R9) |
|
| 735 |
+ |
|
| 736 |
+ // Fill bitreader for state updates |
|
| 737 |
+ MOVQ R13, (SP) |
|
| 738 |
+ MOVQ $0x00000808, CX |
|
| 739 |
+ BEXTRQ CX, R8, R13 |
|
| 740 |
+ MOVQ ctx+16(FP), CX |
|
| 741 |
+ CMPQ 96(CX), $0x00 |
|
| 742 |
+ JZ sequenceDecs_decode_bmi2_skip_update |
|
| 743 |
+ LEAQ (SI)(DI*1), R14 |
|
| 744 |
+ ADDQ R8, R14 |
|
| 745 |
+ MOVBQZX R14, R14 |
|
| 746 |
+ LEAQ (DX)(R14*1), CX |
|
| 747 |
+ MOVQ AX, R15 |
|
| 748 |
+ MOVQ CX, DX |
|
| 749 |
+ ROLQ CL, R15 |
|
| 750 |
+ BZHIQ R14, R15, R15 |
|
| 751 |
+ |
|
| 752 |
+ // Update Offset State |
|
| 753 |
+ BZHIQ R8, R15, CX |
|
| 754 |
+ SHRXQ R8, R15, R15 |
|
| 755 |
+ MOVQ $0x00001010, R14 |
|
| 756 |
+ BEXTRQ R14, R8, R8 |
|
| 757 |
+ ADDQ CX, R8 |
|
| 758 |
+ |
|
| 759 |
+ // Load ctx.ofTable |
|
| 760 |
+ MOVQ ctx+16(FP), CX |
|
| 761 |
+ MOVQ 48(CX), CX |
|
| 762 |
+ MOVQ (CX)(R8*8), R8 |
|
| 763 |
+ |
|
| 764 |
+ // Update Match Length State |
|
| 765 |
+ BZHIQ DI, R15, CX |
|
| 766 |
+ SHRXQ DI, R15, R15 |
|
| 767 |
+ MOVQ $0x00001010, R14 |
|
| 768 |
+ BEXTRQ R14, DI, DI |
|
| 769 |
+ ADDQ CX, DI |
|
| 770 |
+ |
|
| 771 |
+ // Load ctx.mlTable |
|
| 772 |
+ MOVQ ctx+16(FP), CX |
|
| 773 |
+ MOVQ 24(CX), CX |
|
| 774 |
+ MOVQ (CX)(DI*8), DI |
|
| 775 |
+ |
|
| 776 |
+ // Update Literal Length State |
|
| 777 |
+ BZHIQ SI, R15, CX |
|
| 778 |
+ MOVQ $0x00001010, R14 |
|
| 779 |
+ BEXTRQ R14, SI, SI |
|
| 780 |
+ ADDQ CX, SI |
|
| 781 |
+ |
|
| 782 |
+ // Load ctx.llTable |
|
| 783 |
+ MOVQ ctx+16(FP), CX |
|
| 784 |
+ MOVQ (CX), CX |
|
| 785 |
+ MOVQ (CX)(SI*8), SI |
|
| 786 |
+ |
|
| 787 |
+sequenceDecs_decode_bmi2_skip_update: |
|
| 788 |
+ // Adjust offset |
|
| 789 |
+ MOVQ 16(R9), CX |
|
| 790 |
+ CMPQ R13, $0x01 |
|
| 791 |
+ JBE sequenceDecs_decode_bmi2_adjust_offsetB_1_or_0 |
|
| 792 |
+ MOVQ R11, R12 |
|
| 793 |
+ MOVQ R10, R11 |
|
| 794 |
+ MOVQ CX, R10 |
|
| 795 |
+ JMP sequenceDecs_decode_bmi2_after_adjust |
|
| 796 |
+ |
|
| 797 |
+sequenceDecs_decode_bmi2_adjust_offsetB_1_or_0: |
|
| 798 |
+ CMPQ (R9), $0x00000000 |
|
| 799 |
+ JNE sequenceDecs_decode_bmi2_adjust_offset_maybezero |
|
| 800 |
+ INCQ CX |
|
| 801 |
+ JMP sequenceDecs_decode_bmi2_adjust_offset_nonzero |
|
| 802 |
+ |
|
| 803 |
+sequenceDecs_decode_bmi2_adjust_offset_maybezero: |
|
| 804 |
+ TESTQ CX, CX |
|
| 805 |
+ JNZ sequenceDecs_decode_bmi2_adjust_offset_nonzero |
|
| 806 |
+ MOVQ R10, CX |
|
| 807 |
+ JMP sequenceDecs_decode_bmi2_after_adjust |
|
| 808 |
+ |
|
| 809 |
+sequenceDecs_decode_bmi2_adjust_offset_nonzero: |
|
| 810 |
+ CMPQ CX, $0x01 |
|
| 811 |
+ JB sequenceDecs_decode_bmi2_adjust_zero |
|
| 812 |
+ JEQ sequenceDecs_decode_bmi2_adjust_one |
|
| 813 |
+ CMPQ CX, $0x02 |
|
| 814 |
+ JA sequenceDecs_decode_bmi2_adjust_three |
|
| 815 |
+ JMP sequenceDecs_decode_bmi2_adjust_two |
|
| 816 |
+ |
|
| 817 |
+sequenceDecs_decode_bmi2_adjust_zero: |
|
| 818 |
+ MOVQ R10, R13 |
|
| 819 |
+ JMP sequenceDecs_decode_bmi2_adjust_test_temp_valid |
|
| 820 |
+ |
|
| 821 |
+sequenceDecs_decode_bmi2_adjust_one: |
|
| 822 |
+ MOVQ R11, R13 |
|
| 823 |
+ JMP sequenceDecs_decode_bmi2_adjust_test_temp_valid |
|
| 824 |
+ |
|
| 825 |
+sequenceDecs_decode_bmi2_adjust_two: |
|
| 826 |
+ MOVQ R12, R13 |
|
| 827 |
+ JMP sequenceDecs_decode_bmi2_adjust_test_temp_valid |
|
| 828 |
+ |
|
| 829 |
+sequenceDecs_decode_bmi2_adjust_three: |
|
| 830 |
+ LEAQ -1(R10), R13 |
|
| 831 |
+ |
|
| 832 |
+sequenceDecs_decode_bmi2_adjust_test_temp_valid: |
|
| 833 |
+ TESTQ R13, R13 |
|
| 834 |
+ JNZ sequenceDecs_decode_bmi2_adjust_temp_valid |
|
| 835 |
+ MOVQ $0x00000001, R13 |
|
| 836 |
+ |
|
| 837 |
+sequenceDecs_decode_bmi2_adjust_temp_valid: |
|
| 838 |
+ CMPQ CX, $0x01 |
|
| 839 |
+ CMOVQNE R11, R12 |
|
| 840 |
+ MOVQ R10, R11 |
|
| 841 |
+ MOVQ R13, R10 |
|
| 842 |
+ MOVQ R13, CX |
|
| 843 |
+ |
|
| 844 |
+sequenceDecs_decode_bmi2_after_adjust: |
|
| 845 |
+ MOVQ CX, 16(R9) |
|
| 846 |
+ |
|
| 847 |
+ // Check values |
|
| 848 |
+ MOVQ 8(R9), R13 |
|
| 849 |
+ MOVQ (R9), R14 |
|
| 850 |
+ LEAQ (R13)(R14*1), R15 |
|
| 851 |
+ MOVQ s+0(FP), BP |
|
| 852 |
+ ADDQ R15, 256(BP) |
|
| 853 |
+ MOVQ ctx+16(FP), R15 |
|
| 854 |
+ SUBQ R14, 128(R15) |
|
| 855 |
+ JS error_not_enough_literals |
|
| 856 |
+ CMPQ R13, $0x00020002 |
|
| 857 |
+ JA sequenceDecs_decode_bmi2_error_match_len_too_big |
|
| 858 |
+ TESTQ CX, CX |
|
| 859 |
+ JNZ sequenceDecs_decode_bmi2_match_len_ofs_ok |
|
| 860 |
+ TESTQ R13, R13 |
|
| 861 |
+ JNZ sequenceDecs_decode_bmi2_error_match_len_ofs_mismatch |
|
| 862 |
+ |
|
| 863 |
+sequenceDecs_decode_bmi2_match_len_ofs_ok: |
|
| 864 |
+ ADDQ $0x18, R9 |
|
| 865 |
+ MOVQ ctx+16(FP), CX |
|
| 866 |
+ DECQ 96(CX) |
|
| 867 |
+ JNS sequenceDecs_decode_bmi2_main_loop |
|
| 868 |
+ MOVQ s+0(FP), CX |
|
| 869 |
+ MOVQ R10, 144(CX) |
|
| 870 |
+ MOVQ R11, 152(CX) |
|
| 871 |
+ MOVQ R12, 160(CX) |
|
| 872 |
+ MOVQ br+8(FP), CX |
|
| 873 |
+ MOVQ AX, 32(CX) |
|
| 874 |
+ MOVB DL, 40(CX) |
|
| 875 |
+ MOVQ BX, 24(CX) |
|
| 876 |
+ |
|
| 877 |
+ // Return success |
|
| 878 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 879 |
+ RET |
|
| 880 |
+ |
|
| 881 |
+ // Return with match length error |
|
| 882 |
+sequenceDecs_decode_bmi2_error_match_len_ofs_mismatch: |
|
| 883 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 884 |
+ RET |
|
| 885 |
+ |
|
| 886 |
+ // Return with match too long error |
|
| 887 |
+sequenceDecs_decode_bmi2_error_match_len_too_big: |
|
| 888 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 889 |
+ RET |
|
| 890 |
+ |
|
| 891 |
+ // Return with match offset too long error |
|
| 892 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 893 |
+ RET |
|
| 894 |
+ |
|
| 895 |
+ // Return with not enough literals error |
|
| 896 |
+error_not_enough_literals: |
|
| 897 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 898 |
+ RET |
|
| 899 |
+ |
|
| 900 |
+ // Return with not enough output space error |
|
| 901 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 902 |
+ RET |
|
| 903 |
+ |
|
| 904 |
+// func sequenceDecs_decode_56_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeAsmContext) int |
|
| 905 |
+// Requires: BMI, BMI2, CMOV |
|
| 906 |
+TEXT ·sequenceDecs_decode_56_bmi2(SB), $8-32 |
|
| 907 |
+ MOVQ br+8(FP), CX |
|
| 908 |
+ MOVQ 32(CX), AX |
|
| 909 |
+ MOVBQZX 40(CX), DX |
|
| 910 |
+ MOVQ 24(CX), BX |
|
| 911 |
+ MOVQ (CX), CX |
|
| 912 |
+ ADDQ BX, CX |
|
| 913 |
+ MOVQ CX, (SP) |
|
| 914 |
+ MOVQ ctx+16(FP), CX |
|
| 915 |
+ MOVQ 72(CX), SI |
|
| 916 |
+ MOVQ 80(CX), DI |
|
| 917 |
+ MOVQ 88(CX), R8 |
|
| 918 |
+ MOVQ 104(CX), R9 |
|
| 919 |
+ MOVQ s+0(FP), CX |
|
| 920 |
+ MOVQ 144(CX), R10 |
|
| 921 |
+ MOVQ 152(CX), R11 |
|
| 922 |
+ MOVQ 160(CX), R12 |
|
| 923 |
+ |
|
| 924 |
+sequenceDecs_decode_56_bmi2_main_loop: |
|
| 925 |
+ MOVQ (SP), R13 |
|
| 926 |
+ |
|
| 927 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 928 |
+ CMPQ BX, $0x08 |
|
| 929 |
+ JL sequenceDecs_decode_56_bmi2_fill_byte_by_byte |
|
| 930 |
+ MOVQ DX, CX |
|
| 931 |
+ SHRQ $0x03, CX |
|
| 932 |
+ SUBQ CX, R13 |
|
| 933 |
+ MOVQ (R13), AX |
|
| 934 |
+ SUBQ CX, BX |
|
| 935 |
+ ANDQ $0x07, DX |
|
| 936 |
+ JMP sequenceDecs_decode_56_bmi2_fill_end |
|
| 937 |
+ |
|
| 938 |
+sequenceDecs_decode_56_bmi2_fill_byte_by_byte: |
|
| 939 |
+ CMPQ BX, $0x00 |
|
| 940 |
+ JLE sequenceDecs_decode_56_bmi2_fill_end |
|
| 941 |
+ CMPQ DX, $0x07 |
|
| 942 |
+ JLE sequenceDecs_decode_56_bmi2_fill_end |
|
| 943 |
+ SHLQ $0x08, AX |
|
| 944 |
+ SUBQ $0x01, R13 |
|
| 945 |
+ SUBQ $0x01, BX |
|
| 946 |
+ SUBQ $0x08, DX |
|
| 947 |
+ MOVBQZX (R13), CX |
|
| 948 |
+ ORQ CX, AX |
|
| 949 |
+ JMP sequenceDecs_decode_56_bmi2_fill_byte_by_byte |
|
| 950 |
+ |
|
| 951 |
+sequenceDecs_decode_56_bmi2_fill_end: |
|
| 952 |
+ // Update offset |
|
| 953 |
+ MOVQ $0x00000808, CX |
|
| 954 |
+ BEXTRQ CX, R8, R14 |
|
| 955 |
+ MOVQ AX, R15 |
|
| 956 |
+ LEAQ (DX)(R14*1), CX |
|
| 957 |
+ ROLQ CL, R15 |
|
| 958 |
+ BZHIQ R14, R15, R15 |
|
| 959 |
+ MOVQ CX, DX |
|
| 960 |
+ MOVQ R8, CX |
|
| 961 |
+ SHRQ $0x20, CX |
|
| 962 |
+ ADDQ R15, CX |
|
| 963 |
+ MOVQ CX, 16(R9) |
|
| 964 |
+ |
|
| 965 |
+ // Update match length |
|
| 966 |
+ MOVQ $0x00000808, CX |
|
| 967 |
+ BEXTRQ CX, DI, R14 |
|
| 968 |
+ MOVQ AX, R15 |
|
| 969 |
+ LEAQ (DX)(R14*1), CX |
|
| 970 |
+ ROLQ CL, R15 |
|
| 971 |
+ BZHIQ R14, R15, R15 |
|
| 972 |
+ MOVQ CX, DX |
|
| 973 |
+ MOVQ DI, CX |
|
| 974 |
+ SHRQ $0x20, CX |
|
| 975 |
+ ADDQ R15, CX |
|
| 976 |
+ MOVQ CX, 8(R9) |
|
| 977 |
+ |
|
| 978 |
+ // Update literal length |
|
| 979 |
+ MOVQ $0x00000808, CX |
|
| 980 |
+ BEXTRQ CX, SI, R14 |
|
| 981 |
+ MOVQ AX, R15 |
|
| 982 |
+ LEAQ (DX)(R14*1), CX |
|
| 983 |
+ ROLQ CL, R15 |
|
| 984 |
+ BZHIQ R14, R15, R15 |
|
| 985 |
+ MOVQ CX, DX |
|
| 986 |
+ MOVQ SI, CX |
|
| 987 |
+ SHRQ $0x20, CX |
|
| 988 |
+ ADDQ R15, CX |
|
| 989 |
+ MOVQ CX, (R9) |
|
| 990 |
+ |
|
| 991 |
+ // Fill bitreader for state updates |
|
| 992 |
+ MOVQ R13, (SP) |
|
| 993 |
+ MOVQ $0x00000808, CX |
|
| 994 |
+ BEXTRQ CX, R8, R13 |
|
| 995 |
+ MOVQ ctx+16(FP), CX |
|
| 996 |
+ CMPQ 96(CX), $0x00 |
|
| 997 |
+ JZ sequenceDecs_decode_56_bmi2_skip_update |
|
| 998 |
+ LEAQ (SI)(DI*1), R14 |
|
| 999 |
+ ADDQ R8, R14 |
|
| 1000 |
+ MOVBQZX R14, R14 |
|
| 1001 |
+ LEAQ (DX)(R14*1), CX |
|
| 1002 |
+ MOVQ AX, R15 |
|
| 1003 |
+ MOVQ CX, DX |
|
| 1004 |
+ ROLQ CL, R15 |
|
| 1005 |
+ BZHIQ R14, R15, R15 |
|
| 1006 |
+ |
|
| 1007 |
+ // Update Offset State |
|
| 1008 |
+ BZHIQ R8, R15, CX |
|
| 1009 |
+ SHRXQ R8, R15, R15 |
|
| 1010 |
+ MOVQ $0x00001010, R14 |
|
| 1011 |
+ BEXTRQ R14, R8, R8 |
|
| 1012 |
+ ADDQ CX, R8 |
|
| 1013 |
+ |
|
| 1014 |
+ // Load ctx.ofTable |
|
| 1015 |
+ MOVQ ctx+16(FP), CX |
|
| 1016 |
+ MOVQ 48(CX), CX |
|
| 1017 |
+ MOVQ (CX)(R8*8), R8 |
|
| 1018 |
+ |
|
| 1019 |
+ // Update Match Length State |
|
| 1020 |
+ BZHIQ DI, R15, CX |
|
| 1021 |
+ SHRXQ DI, R15, R15 |
|
| 1022 |
+ MOVQ $0x00001010, R14 |
|
| 1023 |
+ BEXTRQ R14, DI, DI |
|
| 1024 |
+ ADDQ CX, DI |
|
| 1025 |
+ |
|
| 1026 |
+ // Load ctx.mlTable |
|
| 1027 |
+ MOVQ ctx+16(FP), CX |
|
| 1028 |
+ MOVQ 24(CX), CX |
|
| 1029 |
+ MOVQ (CX)(DI*8), DI |
|
| 1030 |
+ |
|
| 1031 |
+ // Update Literal Length State |
|
| 1032 |
+ BZHIQ SI, R15, CX |
|
| 1033 |
+ MOVQ $0x00001010, R14 |
|
| 1034 |
+ BEXTRQ R14, SI, SI |
|
| 1035 |
+ ADDQ CX, SI |
|
| 1036 |
+ |
|
| 1037 |
+ // Load ctx.llTable |
|
| 1038 |
+ MOVQ ctx+16(FP), CX |
|
| 1039 |
+ MOVQ (CX), CX |
|
| 1040 |
+ MOVQ (CX)(SI*8), SI |
|
| 1041 |
+ |
|
| 1042 |
+sequenceDecs_decode_56_bmi2_skip_update: |
|
| 1043 |
+ // Adjust offset |
|
| 1044 |
+ MOVQ 16(R9), CX |
|
| 1045 |
+ CMPQ R13, $0x01 |
|
| 1046 |
+ JBE sequenceDecs_decode_56_bmi2_adjust_offsetB_1_or_0 |
|
| 1047 |
+ MOVQ R11, R12 |
|
| 1048 |
+ MOVQ R10, R11 |
|
| 1049 |
+ MOVQ CX, R10 |
|
| 1050 |
+ JMP sequenceDecs_decode_56_bmi2_after_adjust |
|
| 1051 |
+ |
|
| 1052 |
+sequenceDecs_decode_56_bmi2_adjust_offsetB_1_or_0: |
|
| 1053 |
+ CMPQ (R9), $0x00000000 |
|
| 1054 |
+ JNE sequenceDecs_decode_56_bmi2_adjust_offset_maybezero |
|
| 1055 |
+ INCQ CX |
|
| 1056 |
+ JMP sequenceDecs_decode_56_bmi2_adjust_offset_nonzero |
|
| 1057 |
+ |
|
| 1058 |
+sequenceDecs_decode_56_bmi2_adjust_offset_maybezero: |
|
| 1059 |
+ TESTQ CX, CX |
|
| 1060 |
+ JNZ sequenceDecs_decode_56_bmi2_adjust_offset_nonzero |
|
| 1061 |
+ MOVQ R10, CX |
|
| 1062 |
+ JMP sequenceDecs_decode_56_bmi2_after_adjust |
|
| 1063 |
+ |
|
| 1064 |
+sequenceDecs_decode_56_bmi2_adjust_offset_nonzero: |
|
| 1065 |
+ CMPQ CX, $0x01 |
|
| 1066 |
+ JB sequenceDecs_decode_56_bmi2_adjust_zero |
|
| 1067 |
+ JEQ sequenceDecs_decode_56_bmi2_adjust_one |
|
| 1068 |
+ CMPQ CX, $0x02 |
|
| 1069 |
+ JA sequenceDecs_decode_56_bmi2_adjust_three |
|
| 1070 |
+ JMP sequenceDecs_decode_56_bmi2_adjust_two |
|
| 1071 |
+ |
|
| 1072 |
+sequenceDecs_decode_56_bmi2_adjust_zero: |
|
| 1073 |
+ MOVQ R10, R13 |
|
| 1074 |
+ JMP sequenceDecs_decode_56_bmi2_adjust_test_temp_valid |
|
| 1075 |
+ |
|
| 1076 |
+sequenceDecs_decode_56_bmi2_adjust_one: |
|
| 1077 |
+ MOVQ R11, R13 |
|
| 1078 |
+ JMP sequenceDecs_decode_56_bmi2_adjust_test_temp_valid |
|
| 1079 |
+ |
|
| 1080 |
+sequenceDecs_decode_56_bmi2_adjust_two: |
|
| 1081 |
+ MOVQ R12, R13 |
|
| 1082 |
+ JMP sequenceDecs_decode_56_bmi2_adjust_test_temp_valid |
|
| 1083 |
+ |
|
| 1084 |
+sequenceDecs_decode_56_bmi2_adjust_three: |
|
| 1085 |
+ LEAQ -1(R10), R13 |
|
| 1086 |
+ |
|
| 1087 |
+sequenceDecs_decode_56_bmi2_adjust_test_temp_valid: |
|
| 1088 |
+ TESTQ R13, R13 |
|
| 1089 |
+ JNZ sequenceDecs_decode_56_bmi2_adjust_temp_valid |
|
| 1090 |
+ MOVQ $0x00000001, R13 |
|
| 1091 |
+ |
|
| 1092 |
+sequenceDecs_decode_56_bmi2_adjust_temp_valid: |
|
| 1093 |
+ CMPQ CX, $0x01 |
|
| 1094 |
+ CMOVQNE R11, R12 |
|
| 1095 |
+ MOVQ R10, R11 |
|
| 1096 |
+ MOVQ R13, R10 |
|
| 1097 |
+ MOVQ R13, CX |
|
| 1098 |
+ |
|
| 1099 |
+sequenceDecs_decode_56_bmi2_after_adjust: |
|
| 1100 |
+ MOVQ CX, 16(R9) |
|
| 1101 |
+ |
|
| 1102 |
+ // Check values |
|
| 1103 |
+ MOVQ 8(R9), R13 |
|
| 1104 |
+ MOVQ (R9), R14 |
|
| 1105 |
+ LEAQ (R13)(R14*1), R15 |
|
| 1106 |
+ MOVQ s+0(FP), BP |
|
| 1107 |
+ ADDQ R15, 256(BP) |
|
| 1108 |
+ MOVQ ctx+16(FP), R15 |
|
| 1109 |
+ SUBQ R14, 128(R15) |
|
| 1110 |
+ JS error_not_enough_literals |
|
| 1111 |
+ CMPQ R13, $0x00020002 |
|
| 1112 |
+ JA sequenceDecs_decode_56_bmi2_error_match_len_too_big |
|
| 1113 |
+ TESTQ CX, CX |
|
| 1114 |
+ JNZ sequenceDecs_decode_56_bmi2_match_len_ofs_ok |
|
| 1115 |
+ TESTQ R13, R13 |
|
| 1116 |
+ JNZ sequenceDecs_decode_56_bmi2_error_match_len_ofs_mismatch |
|
| 1117 |
+ |
|
| 1118 |
+sequenceDecs_decode_56_bmi2_match_len_ofs_ok: |
|
| 1119 |
+ ADDQ $0x18, R9 |
|
| 1120 |
+ MOVQ ctx+16(FP), CX |
|
| 1121 |
+ DECQ 96(CX) |
|
| 1122 |
+ JNS sequenceDecs_decode_56_bmi2_main_loop |
|
| 1123 |
+ MOVQ s+0(FP), CX |
|
| 1124 |
+ MOVQ R10, 144(CX) |
|
| 1125 |
+ MOVQ R11, 152(CX) |
|
| 1126 |
+ MOVQ R12, 160(CX) |
|
| 1127 |
+ MOVQ br+8(FP), CX |
|
| 1128 |
+ MOVQ AX, 32(CX) |
|
| 1129 |
+ MOVB DL, 40(CX) |
|
| 1130 |
+ MOVQ BX, 24(CX) |
|
| 1131 |
+ |
|
| 1132 |
+ // Return success |
|
| 1133 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 1134 |
+ RET |
|
| 1135 |
+ |
|
| 1136 |
+ // Return with match length error |
|
| 1137 |
+sequenceDecs_decode_56_bmi2_error_match_len_ofs_mismatch: |
|
| 1138 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 1139 |
+ RET |
|
| 1140 |
+ |
|
| 1141 |
+ // Return with match too long error |
|
| 1142 |
+sequenceDecs_decode_56_bmi2_error_match_len_too_big: |
|
| 1143 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 1144 |
+ RET |
|
| 1145 |
+ |
|
| 1146 |
+ // Return with match offset too long error |
|
| 1147 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 1148 |
+ RET |
|
| 1149 |
+ |
|
| 1150 |
+ // Return with not enough literals error |
|
| 1151 |
+error_not_enough_literals: |
|
| 1152 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 1153 |
+ RET |
|
| 1154 |
+ |
|
| 1155 |
+ // Return with not enough output space error |
|
| 1156 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 1157 |
+ RET |
|
| 1158 |
+ |
|
| 1159 |
+// func sequenceDecs_executeSimple_amd64(ctx *executeAsmContext) bool |
|
| 1160 |
+// Requires: SSE |
|
| 1161 |
+TEXT ·sequenceDecs_executeSimple_amd64(SB), $8-9 |
|
| 1162 |
+ MOVQ ctx+0(FP), R10 |
|
| 1163 |
+ MOVQ 8(R10), CX |
|
| 1164 |
+ TESTQ CX, CX |
|
| 1165 |
+ JZ empty_seqs |
|
| 1166 |
+ MOVQ (R10), AX |
|
| 1167 |
+ MOVQ 24(R10), DX |
|
| 1168 |
+ MOVQ 32(R10), BX |
|
| 1169 |
+ MOVQ 80(R10), SI |
|
| 1170 |
+ MOVQ 104(R10), DI |
|
| 1171 |
+ MOVQ 120(R10), R8 |
|
| 1172 |
+ MOVQ 56(R10), R9 |
|
| 1173 |
+ MOVQ 64(R10), R10 |
|
| 1174 |
+ ADDQ R10, R9 |
|
| 1175 |
+ |
|
| 1176 |
+ // seqsBase += 24 * seqIndex |
|
| 1177 |
+ LEAQ (DX)(DX*2), R11 |
|
| 1178 |
+ SHLQ $0x03, R11 |
|
| 1179 |
+ ADDQ R11, AX |
|
| 1180 |
+ |
|
| 1181 |
+ // outBase += outPosition |
|
| 1182 |
+ ADDQ DI, BX |
|
| 1183 |
+ |
|
| 1184 |
+main_loop: |
|
| 1185 |
+ MOVQ (AX), R11 |
|
| 1186 |
+ MOVQ 16(AX), R12 |
|
| 1187 |
+ MOVQ 8(AX), R13 |
|
| 1188 |
+ |
|
| 1189 |
+ // Copy literals |
|
| 1190 |
+ TESTQ R11, R11 |
|
| 1191 |
+ JZ check_offset |
|
| 1192 |
+ XORQ R14, R14 |
|
| 1193 |
+ |
|
| 1194 |
+copy_1: |
|
| 1195 |
+ MOVUPS (SI)(R14*1), X0 |
|
| 1196 |
+ MOVUPS X0, (BX)(R14*1) |
|
| 1197 |
+ ADDQ $0x10, R14 |
|
| 1198 |
+ CMPQ R14, R11 |
|
| 1199 |
+ JB copy_1 |
|
| 1200 |
+ ADDQ R11, SI |
|
| 1201 |
+ ADDQ R11, BX |
|
| 1202 |
+ ADDQ R11, DI |
|
| 1203 |
+ |
|
| 1204 |
+ // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) |
|
| 1205 |
+check_offset: |
|
| 1206 |
+ LEAQ (DI)(R10*1), R11 |
|
| 1207 |
+ CMPQ R12, R11 |
|
| 1208 |
+ JG error_match_off_too_big |
|
| 1209 |
+ CMPQ R12, R8 |
|
| 1210 |
+ JG error_match_off_too_big |
|
| 1211 |
+ |
|
| 1212 |
+ // Copy match from history |
|
| 1213 |
+ MOVQ R12, R11 |
|
| 1214 |
+ SUBQ DI, R11 |
|
| 1215 |
+ JLS copy_match |
|
| 1216 |
+ MOVQ R9, R14 |
|
| 1217 |
+ SUBQ R11, R14 |
|
| 1218 |
+ CMPQ R13, R11 |
|
| 1219 |
+ JG copy_all_from_history |
|
| 1220 |
+ MOVQ R13, R11 |
|
| 1221 |
+ SUBQ $0x10, R11 |
|
| 1222 |
+ JB copy_4_small |
|
| 1223 |
+ |
|
| 1224 |
+copy_4_loop: |
|
| 1225 |
+ MOVUPS (R14), X0 |
|
| 1226 |
+ MOVUPS X0, (BX) |
|
| 1227 |
+ ADDQ $0x10, R14 |
|
| 1228 |
+ ADDQ $0x10, BX |
|
| 1229 |
+ SUBQ $0x10, R11 |
|
| 1230 |
+ JAE copy_4_loop |
|
| 1231 |
+ LEAQ 16(R14)(R11*1), R14 |
|
| 1232 |
+ LEAQ 16(BX)(R11*1), BX |
|
| 1233 |
+ MOVUPS -16(R14), X0 |
|
| 1234 |
+ MOVUPS X0, -16(BX) |
|
| 1235 |
+ JMP copy_4_end |
|
| 1236 |
+ |
|
| 1237 |
+copy_4_small: |
|
| 1238 |
+ CMPQ R13, $0x03 |
|
| 1239 |
+ JE copy_4_move_3 |
|
| 1240 |
+ CMPQ R13, $0x08 |
|
| 1241 |
+ JB copy_4_move_4through7 |
|
| 1242 |
+ JMP copy_4_move_8through16 |
|
| 1243 |
+ |
|
| 1244 |
+copy_4_move_3: |
|
| 1245 |
+ MOVW (R14), R11 |
|
| 1246 |
+ MOVB 2(R14), R12 |
|
| 1247 |
+ MOVW R11, (BX) |
|
| 1248 |
+ MOVB R12, 2(BX) |
|
| 1249 |
+ ADDQ R13, R14 |
|
| 1250 |
+ ADDQ R13, BX |
|
| 1251 |
+ JMP copy_4_end |
|
| 1252 |
+ |
|
| 1253 |
+copy_4_move_4through7: |
|
| 1254 |
+ MOVL (R14), R11 |
|
| 1255 |
+ MOVL -4(R14)(R13*1), R12 |
|
| 1256 |
+ MOVL R11, (BX) |
|
| 1257 |
+ MOVL R12, -4(BX)(R13*1) |
|
| 1258 |
+ ADDQ R13, R14 |
|
| 1259 |
+ ADDQ R13, BX |
|
| 1260 |
+ JMP copy_4_end |
|
| 1261 |
+ |
|
| 1262 |
+copy_4_move_8through16: |
|
| 1263 |
+ MOVQ (R14), R11 |
|
| 1264 |
+ MOVQ -8(R14)(R13*1), R12 |
|
| 1265 |
+ MOVQ R11, (BX) |
|
| 1266 |
+ MOVQ R12, -8(BX)(R13*1) |
|
| 1267 |
+ ADDQ R13, R14 |
|
| 1268 |
+ ADDQ R13, BX |
|
| 1269 |
+ |
|
| 1270 |
+copy_4_end: |
|
| 1271 |
+ ADDQ R13, DI |
|
| 1272 |
+ ADDQ $0x18, AX |
|
| 1273 |
+ INCQ DX |
|
| 1274 |
+ CMPQ DX, CX |
|
| 1275 |
+ JB main_loop |
|
| 1276 |
+ JMP loop_finished |
|
| 1277 |
+ |
|
| 1278 |
+copy_all_from_history: |
|
| 1279 |
+ MOVQ R11, R15 |
|
| 1280 |
+ SUBQ $0x10, R15 |
|
| 1281 |
+ JB copy_5_small |
|
| 1282 |
+ |
|
| 1283 |
+copy_5_loop: |
|
| 1284 |
+ MOVUPS (R14), X0 |
|
| 1285 |
+ MOVUPS X0, (BX) |
|
| 1286 |
+ ADDQ $0x10, R14 |
|
| 1287 |
+ ADDQ $0x10, BX |
|
| 1288 |
+ SUBQ $0x10, R15 |
|
| 1289 |
+ JAE copy_5_loop |
|
| 1290 |
+ LEAQ 16(R14)(R15*1), R14 |
|
| 1291 |
+ LEAQ 16(BX)(R15*1), BX |
|
| 1292 |
+ MOVUPS -16(R14), X0 |
|
| 1293 |
+ MOVUPS X0, -16(BX) |
|
| 1294 |
+ JMP copy_5_end |
|
| 1295 |
+ |
|
| 1296 |
+copy_5_small: |
|
| 1297 |
+ CMPQ R11, $0x03 |
|
| 1298 |
+ JE copy_5_move_3 |
|
| 1299 |
+ JB copy_5_move_1or2 |
|
| 1300 |
+ CMPQ R11, $0x08 |
|
| 1301 |
+ JB copy_5_move_4through7 |
|
| 1302 |
+ JMP copy_5_move_8through16 |
|
| 1303 |
+ |
|
| 1304 |
+copy_5_move_1or2: |
|
| 1305 |
+ MOVB (R14), R15 |
|
| 1306 |
+ MOVB -1(R14)(R11*1), BP |
|
| 1307 |
+ MOVB R15, (BX) |
|
| 1308 |
+ MOVB BP, -1(BX)(R11*1) |
|
| 1309 |
+ ADDQ R11, R14 |
|
| 1310 |
+ ADDQ R11, BX |
|
| 1311 |
+ JMP copy_5_end |
|
| 1312 |
+ |
|
| 1313 |
+copy_5_move_3: |
|
| 1314 |
+ MOVW (R14), R15 |
|
| 1315 |
+ MOVB 2(R14), BP |
|
| 1316 |
+ MOVW R15, (BX) |
|
| 1317 |
+ MOVB BP, 2(BX) |
|
| 1318 |
+ ADDQ R11, R14 |
|
| 1319 |
+ ADDQ R11, BX |
|
| 1320 |
+ JMP copy_5_end |
|
| 1321 |
+ |
|
| 1322 |
+copy_5_move_4through7: |
|
| 1323 |
+ MOVL (R14), R15 |
|
| 1324 |
+ MOVL -4(R14)(R11*1), BP |
|
| 1325 |
+ MOVL R15, (BX) |
|
| 1326 |
+ MOVL BP, -4(BX)(R11*1) |
|
| 1327 |
+ ADDQ R11, R14 |
|
| 1328 |
+ ADDQ R11, BX |
|
| 1329 |
+ JMP copy_5_end |
|
| 1330 |
+ |
|
| 1331 |
+copy_5_move_8through16: |
|
| 1332 |
+ MOVQ (R14), R15 |
|
| 1333 |
+ MOVQ -8(R14)(R11*1), BP |
|
| 1334 |
+ MOVQ R15, (BX) |
|
| 1335 |
+ MOVQ BP, -8(BX)(R11*1) |
|
| 1336 |
+ ADDQ R11, R14 |
|
| 1337 |
+ ADDQ R11, BX |
|
| 1338 |
+ |
|
| 1339 |
+copy_5_end: |
|
| 1340 |
+ ADDQ R11, DI |
|
| 1341 |
+ SUBQ R11, R13 |
|
| 1342 |
+ |
|
| 1343 |
+ // Copy match from the current buffer |
|
| 1344 |
+copy_match: |
|
| 1345 |
+ MOVQ BX, R11 |
|
| 1346 |
+ SUBQ R12, R11 |
|
| 1347 |
+ |
|
| 1348 |
+ // ml <= mo |
|
| 1349 |
+ CMPQ R13, R12 |
|
| 1350 |
+ JA copy_overlapping_match |
|
| 1351 |
+ |
|
| 1352 |
+ // Copy non-overlapping match |
|
| 1353 |
+ ADDQ R13, DI |
|
| 1354 |
+ MOVQ BX, R12 |
|
| 1355 |
+ ADDQ R13, BX |
|
| 1356 |
+ |
|
| 1357 |
+copy_2: |
|
| 1358 |
+ MOVUPS (R11), X0 |
|
| 1359 |
+ MOVUPS X0, (R12) |
|
| 1360 |
+ ADDQ $0x10, R11 |
|
| 1361 |
+ ADDQ $0x10, R12 |
|
| 1362 |
+ SUBQ $0x10, R13 |
|
| 1363 |
+ JHI copy_2 |
|
| 1364 |
+ JMP handle_loop |
|
| 1365 |
+ |
|
| 1366 |
+ // Copy overlapping match |
|
| 1367 |
+copy_overlapping_match: |
|
| 1368 |
+ ADDQ R13, DI |
|
| 1369 |
+ |
|
| 1370 |
+copy_slow_3: |
|
| 1371 |
+ MOVB (R11), R12 |
|
| 1372 |
+ MOVB R12, (BX) |
|
| 1373 |
+ INCQ R11 |
|
| 1374 |
+ INCQ BX |
|
| 1375 |
+ DECQ R13 |
|
| 1376 |
+ JNZ copy_slow_3 |
|
| 1377 |
+ |
|
| 1378 |
+handle_loop: |
|
| 1379 |
+ ADDQ $0x18, AX |
|
| 1380 |
+ INCQ DX |
|
| 1381 |
+ CMPQ DX, CX |
|
| 1382 |
+ JB main_loop |
|
| 1383 |
+ |
|
| 1384 |
+loop_finished: |
|
| 1385 |
+ // Return value |
|
| 1386 |
+ MOVB $0x01, ret+8(FP) |
|
| 1387 |
+ |
|
| 1388 |
+ // Update the context |
|
| 1389 |
+ MOVQ ctx+0(FP), AX |
|
| 1390 |
+ MOVQ DX, 24(AX) |
|
| 1391 |
+ MOVQ DI, 104(AX) |
|
| 1392 |
+ MOVQ 80(AX), CX |
|
| 1393 |
+ SUBQ CX, SI |
|
| 1394 |
+ MOVQ SI, 112(AX) |
|
| 1395 |
+ RET |
|
| 1396 |
+ |
|
| 1397 |
+error_match_off_too_big: |
|
| 1398 |
+ // Return value |
|
| 1399 |
+ MOVB $0x00, ret+8(FP) |
|
| 1400 |
+ |
|
| 1401 |
+ // Update the context |
|
| 1402 |
+ MOVQ ctx+0(FP), AX |
|
| 1403 |
+ MOVQ DX, 24(AX) |
|
| 1404 |
+ MOVQ DI, 104(AX) |
|
| 1405 |
+ MOVQ 80(AX), CX |
|
| 1406 |
+ SUBQ CX, SI |
|
| 1407 |
+ MOVQ SI, 112(AX) |
|
| 1408 |
+ RET |
|
| 1409 |
+ |
|
| 1410 |
+empty_seqs: |
|
| 1411 |
+ // Return value |
|
| 1412 |
+ MOVB $0x01, ret+8(FP) |
|
| 1413 |
+ RET |
|
| 1414 |
+ |
|
| 1415 |
+// func sequenceDecs_executeSimple_safe_amd64(ctx *executeAsmContext) bool |
|
| 1416 |
+// Requires: SSE |
|
| 1417 |
+TEXT ·sequenceDecs_executeSimple_safe_amd64(SB), $8-9 |
|
| 1418 |
+ MOVQ ctx+0(FP), R10 |
|
| 1419 |
+ MOVQ 8(R10), CX |
|
| 1420 |
+ TESTQ CX, CX |
|
| 1421 |
+ JZ empty_seqs |
|
| 1422 |
+ MOVQ (R10), AX |
|
| 1423 |
+ MOVQ 24(R10), DX |
|
| 1424 |
+ MOVQ 32(R10), BX |
|
| 1425 |
+ MOVQ 80(R10), SI |
|
| 1426 |
+ MOVQ 104(R10), DI |
|
| 1427 |
+ MOVQ 120(R10), R8 |
|
| 1428 |
+ MOVQ 56(R10), R9 |
|
| 1429 |
+ MOVQ 64(R10), R10 |
|
| 1430 |
+ ADDQ R10, R9 |
|
| 1431 |
+ |
|
| 1432 |
+ // seqsBase += 24 * seqIndex |
|
| 1433 |
+ LEAQ (DX)(DX*2), R11 |
|
| 1434 |
+ SHLQ $0x03, R11 |
|
| 1435 |
+ ADDQ R11, AX |
|
| 1436 |
+ |
|
| 1437 |
+ // outBase += outPosition |
|
| 1438 |
+ ADDQ DI, BX |
|
| 1439 |
+ |
|
| 1440 |
+main_loop: |
|
| 1441 |
+ MOVQ (AX), R11 |
|
| 1442 |
+ MOVQ 16(AX), R12 |
|
| 1443 |
+ MOVQ 8(AX), R13 |
|
| 1444 |
+ |
|
| 1445 |
+ // Copy literals |
|
| 1446 |
+ TESTQ R11, R11 |
|
| 1447 |
+ JZ check_offset |
|
| 1448 |
+ MOVQ R11, R14 |
|
| 1449 |
+ SUBQ $0x10, R14 |
|
| 1450 |
+ JB copy_1_small |
|
| 1451 |
+ |
|
| 1452 |
+copy_1_loop: |
|
| 1453 |
+ MOVUPS (SI), X0 |
|
| 1454 |
+ MOVUPS X0, (BX) |
|
| 1455 |
+ ADDQ $0x10, SI |
|
| 1456 |
+ ADDQ $0x10, BX |
|
| 1457 |
+ SUBQ $0x10, R14 |
|
| 1458 |
+ JAE copy_1_loop |
|
| 1459 |
+ LEAQ 16(SI)(R14*1), SI |
|
| 1460 |
+ LEAQ 16(BX)(R14*1), BX |
|
| 1461 |
+ MOVUPS -16(SI), X0 |
|
| 1462 |
+ MOVUPS X0, -16(BX) |
|
| 1463 |
+ JMP copy_1_end |
|
| 1464 |
+ |
|
| 1465 |
+copy_1_small: |
|
| 1466 |
+ CMPQ R11, $0x03 |
|
| 1467 |
+ JE copy_1_move_3 |
|
| 1468 |
+ JB copy_1_move_1or2 |
|
| 1469 |
+ CMPQ R11, $0x08 |
|
| 1470 |
+ JB copy_1_move_4through7 |
|
| 1471 |
+ JMP copy_1_move_8through16 |
|
| 1472 |
+ |
|
| 1473 |
+copy_1_move_1or2: |
|
| 1474 |
+ MOVB (SI), R14 |
|
| 1475 |
+ MOVB -1(SI)(R11*1), R15 |
|
| 1476 |
+ MOVB R14, (BX) |
|
| 1477 |
+ MOVB R15, -1(BX)(R11*1) |
|
| 1478 |
+ ADDQ R11, SI |
|
| 1479 |
+ ADDQ R11, BX |
|
| 1480 |
+ JMP copy_1_end |
|
| 1481 |
+ |
|
| 1482 |
+copy_1_move_3: |
|
| 1483 |
+ MOVW (SI), R14 |
|
| 1484 |
+ MOVB 2(SI), R15 |
|
| 1485 |
+ MOVW R14, (BX) |
|
| 1486 |
+ MOVB R15, 2(BX) |
|
| 1487 |
+ ADDQ R11, SI |
|
| 1488 |
+ ADDQ R11, BX |
|
| 1489 |
+ JMP copy_1_end |
|
| 1490 |
+ |
|
| 1491 |
+copy_1_move_4through7: |
|
| 1492 |
+ MOVL (SI), R14 |
|
| 1493 |
+ MOVL -4(SI)(R11*1), R15 |
|
| 1494 |
+ MOVL R14, (BX) |
|
| 1495 |
+ MOVL R15, -4(BX)(R11*1) |
|
| 1496 |
+ ADDQ R11, SI |
|
| 1497 |
+ ADDQ R11, BX |
|
| 1498 |
+ JMP copy_1_end |
|
| 1499 |
+ |
|
| 1500 |
+copy_1_move_8through16: |
|
| 1501 |
+ MOVQ (SI), R14 |
|
| 1502 |
+ MOVQ -8(SI)(R11*1), R15 |
|
| 1503 |
+ MOVQ R14, (BX) |
|
| 1504 |
+ MOVQ R15, -8(BX)(R11*1) |
|
| 1505 |
+ ADDQ R11, SI |
|
| 1506 |
+ ADDQ R11, BX |
|
| 1507 |
+ |
|
| 1508 |
+copy_1_end: |
|
| 1509 |
+ ADDQ R11, DI |
|
| 1510 |
+ |
|
| 1511 |
+ // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) |
|
| 1512 |
+check_offset: |
|
| 1513 |
+ LEAQ (DI)(R10*1), R11 |
|
| 1514 |
+ CMPQ R12, R11 |
|
| 1515 |
+ JG error_match_off_too_big |
|
| 1516 |
+ CMPQ R12, R8 |
|
| 1517 |
+ JG error_match_off_too_big |
|
| 1518 |
+ |
|
| 1519 |
+ // Copy match from history |
|
| 1520 |
+ MOVQ R12, R11 |
|
| 1521 |
+ SUBQ DI, R11 |
|
| 1522 |
+ JLS copy_match |
|
| 1523 |
+ MOVQ R9, R14 |
|
| 1524 |
+ SUBQ R11, R14 |
|
| 1525 |
+ CMPQ R13, R11 |
|
| 1526 |
+ JG copy_all_from_history |
|
| 1527 |
+ MOVQ R13, R11 |
|
| 1528 |
+ SUBQ $0x10, R11 |
|
| 1529 |
+ JB copy_4_small |
|
| 1530 |
+ |
|
| 1531 |
+copy_4_loop: |
|
| 1532 |
+ MOVUPS (R14), X0 |
|
| 1533 |
+ MOVUPS X0, (BX) |
|
| 1534 |
+ ADDQ $0x10, R14 |
|
| 1535 |
+ ADDQ $0x10, BX |
|
| 1536 |
+ SUBQ $0x10, R11 |
|
| 1537 |
+ JAE copy_4_loop |
|
| 1538 |
+ LEAQ 16(R14)(R11*1), R14 |
|
| 1539 |
+ LEAQ 16(BX)(R11*1), BX |
|
| 1540 |
+ MOVUPS -16(R14), X0 |
|
| 1541 |
+ MOVUPS X0, -16(BX) |
|
| 1542 |
+ JMP copy_4_end |
|
| 1543 |
+ |
|
| 1544 |
+copy_4_small: |
|
| 1545 |
+ CMPQ R13, $0x03 |
|
| 1546 |
+ JE copy_4_move_3 |
|
| 1547 |
+ CMPQ R13, $0x08 |
|
| 1548 |
+ JB copy_4_move_4through7 |
|
| 1549 |
+ JMP copy_4_move_8through16 |
|
| 1550 |
+ |
|
| 1551 |
+copy_4_move_3: |
|
| 1552 |
+ MOVW (R14), R11 |
|
| 1553 |
+ MOVB 2(R14), R12 |
|
| 1554 |
+ MOVW R11, (BX) |
|
| 1555 |
+ MOVB R12, 2(BX) |
|
| 1556 |
+ ADDQ R13, R14 |
|
| 1557 |
+ ADDQ R13, BX |
|
| 1558 |
+ JMP copy_4_end |
|
| 1559 |
+ |
|
| 1560 |
+copy_4_move_4through7: |
|
| 1561 |
+ MOVL (R14), R11 |
|
| 1562 |
+ MOVL -4(R14)(R13*1), R12 |
|
| 1563 |
+ MOVL R11, (BX) |
|
| 1564 |
+ MOVL R12, -4(BX)(R13*1) |
|
| 1565 |
+ ADDQ R13, R14 |
|
| 1566 |
+ ADDQ R13, BX |
|
| 1567 |
+ JMP copy_4_end |
|
| 1568 |
+ |
|
| 1569 |
+copy_4_move_8through16: |
|
| 1570 |
+ MOVQ (R14), R11 |
|
| 1571 |
+ MOVQ -8(R14)(R13*1), R12 |
|
| 1572 |
+ MOVQ R11, (BX) |
|
| 1573 |
+ MOVQ R12, -8(BX)(R13*1) |
|
| 1574 |
+ ADDQ R13, R14 |
|
| 1575 |
+ ADDQ R13, BX |
|
| 1576 |
+ |
|
| 1577 |
+copy_4_end: |
|
| 1578 |
+ ADDQ R13, DI |
|
| 1579 |
+ ADDQ $0x18, AX |
|
| 1580 |
+ INCQ DX |
|
| 1581 |
+ CMPQ DX, CX |
|
| 1582 |
+ JB main_loop |
|
| 1583 |
+ JMP loop_finished |
|
| 1584 |
+ |
|
| 1585 |
+copy_all_from_history: |
|
| 1586 |
+ MOVQ R11, R15 |
|
| 1587 |
+ SUBQ $0x10, R15 |
|
| 1588 |
+ JB copy_5_small |
|
| 1589 |
+ |
|
| 1590 |
+copy_5_loop: |
|
| 1591 |
+ MOVUPS (R14), X0 |
|
| 1592 |
+ MOVUPS X0, (BX) |
|
| 1593 |
+ ADDQ $0x10, R14 |
|
| 1594 |
+ ADDQ $0x10, BX |
|
| 1595 |
+ SUBQ $0x10, R15 |
|
| 1596 |
+ JAE copy_5_loop |
|
| 1597 |
+ LEAQ 16(R14)(R15*1), R14 |
|
| 1598 |
+ LEAQ 16(BX)(R15*1), BX |
|
| 1599 |
+ MOVUPS -16(R14), X0 |
|
| 1600 |
+ MOVUPS X0, -16(BX) |
|
| 1601 |
+ JMP copy_5_end |
|
| 1602 |
+ |
|
| 1603 |
+copy_5_small: |
|
| 1604 |
+ CMPQ R11, $0x03 |
|
| 1605 |
+ JE copy_5_move_3 |
|
| 1606 |
+ JB copy_5_move_1or2 |
|
| 1607 |
+ CMPQ R11, $0x08 |
|
| 1608 |
+ JB copy_5_move_4through7 |
|
| 1609 |
+ JMP copy_5_move_8through16 |
|
| 1610 |
+ |
|
| 1611 |
+copy_5_move_1or2: |
|
| 1612 |
+ MOVB (R14), R15 |
|
| 1613 |
+ MOVB -1(R14)(R11*1), BP |
|
| 1614 |
+ MOVB R15, (BX) |
|
| 1615 |
+ MOVB BP, -1(BX)(R11*1) |
|
| 1616 |
+ ADDQ R11, R14 |
|
| 1617 |
+ ADDQ R11, BX |
|
| 1618 |
+ JMP copy_5_end |
|
| 1619 |
+ |
|
| 1620 |
+copy_5_move_3: |
|
| 1621 |
+ MOVW (R14), R15 |
|
| 1622 |
+ MOVB 2(R14), BP |
|
| 1623 |
+ MOVW R15, (BX) |
|
| 1624 |
+ MOVB BP, 2(BX) |
|
| 1625 |
+ ADDQ R11, R14 |
|
| 1626 |
+ ADDQ R11, BX |
|
| 1627 |
+ JMP copy_5_end |
|
| 1628 |
+ |
|
| 1629 |
+copy_5_move_4through7: |
|
| 1630 |
+ MOVL (R14), R15 |
|
| 1631 |
+ MOVL -4(R14)(R11*1), BP |
|
| 1632 |
+ MOVL R15, (BX) |
|
| 1633 |
+ MOVL BP, -4(BX)(R11*1) |
|
| 1634 |
+ ADDQ R11, R14 |
|
| 1635 |
+ ADDQ R11, BX |
|
| 1636 |
+ JMP copy_5_end |
|
| 1637 |
+ |
|
| 1638 |
+copy_5_move_8through16: |
|
| 1639 |
+ MOVQ (R14), R15 |
|
| 1640 |
+ MOVQ -8(R14)(R11*1), BP |
|
| 1641 |
+ MOVQ R15, (BX) |
|
| 1642 |
+ MOVQ BP, -8(BX)(R11*1) |
|
| 1643 |
+ ADDQ R11, R14 |
|
| 1644 |
+ ADDQ R11, BX |
|
| 1645 |
+ |
|
| 1646 |
+copy_5_end: |
|
| 1647 |
+ ADDQ R11, DI |
|
| 1648 |
+ SUBQ R11, R13 |
|
| 1649 |
+ |
|
| 1650 |
+ // Copy match from the current buffer |
|
| 1651 |
+copy_match: |
|
| 1652 |
+ MOVQ BX, R11 |
|
| 1653 |
+ SUBQ R12, R11 |
|
| 1654 |
+ |
|
| 1655 |
+ // ml <= mo |
|
| 1656 |
+ CMPQ R13, R12 |
|
| 1657 |
+ JA copy_overlapping_match |
|
| 1658 |
+ |
|
| 1659 |
+ // Copy non-overlapping match |
|
| 1660 |
+ ADDQ R13, DI |
|
| 1661 |
+ MOVQ R13, R12 |
|
| 1662 |
+ SUBQ $0x10, R12 |
|
| 1663 |
+ JB copy_2_small |
|
| 1664 |
+ |
|
| 1665 |
+copy_2_loop: |
|
| 1666 |
+ MOVUPS (R11), X0 |
|
| 1667 |
+ MOVUPS X0, (BX) |
|
| 1668 |
+ ADDQ $0x10, R11 |
|
| 1669 |
+ ADDQ $0x10, BX |
|
| 1670 |
+ SUBQ $0x10, R12 |
|
| 1671 |
+ JAE copy_2_loop |
|
| 1672 |
+ LEAQ 16(R11)(R12*1), R11 |
|
| 1673 |
+ LEAQ 16(BX)(R12*1), BX |
|
| 1674 |
+ MOVUPS -16(R11), X0 |
|
| 1675 |
+ MOVUPS X0, -16(BX) |
|
| 1676 |
+ JMP copy_2_end |
|
| 1677 |
+ |
|
| 1678 |
+copy_2_small: |
|
| 1679 |
+ CMPQ R13, $0x03 |
|
| 1680 |
+ JE copy_2_move_3 |
|
| 1681 |
+ JB copy_2_move_1or2 |
|
| 1682 |
+ CMPQ R13, $0x08 |
|
| 1683 |
+ JB copy_2_move_4through7 |
|
| 1684 |
+ JMP copy_2_move_8through16 |
|
| 1685 |
+ |
|
| 1686 |
+copy_2_move_1or2: |
|
| 1687 |
+ MOVB (R11), R12 |
|
| 1688 |
+ MOVB -1(R11)(R13*1), R14 |
|
| 1689 |
+ MOVB R12, (BX) |
|
| 1690 |
+ MOVB R14, -1(BX)(R13*1) |
|
| 1691 |
+ ADDQ R13, R11 |
|
| 1692 |
+ ADDQ R13, BX |
|
| 1693 |
+ JMP copy_2_end |
|
| 1694 |
+ |
|
| 1695 |
+copy_2_move_3: |
|
| 1696 |
+ MOVW (R11), R12 |
|
| 1697 |
+ MOVB 2(R11), R14 |
|
| 1698 |
+ MOVW R12, (BX) |
|
| 1699 |
+ MOVB R14, 2(BX) |
|
| 1700 |
+ ADDQ R13, R11 |
|
| 1701 |
+ ADDQ R13, BX |
|
| 1702 |
+ JMP copy_2_end |
|
| 1703 |
+ |
|
| 1704 |
+copy_2_move_4through7: |
|
| 1705 |
+ MOVL (R11), R12 |
|
| 1706 |
+ MOVL -4(R11)(R13*1), R14 |
|
| 1707 |
+ MOVL R12, (BX) |
|
| 1708 |
+ MOVL R14, -4(BX)(R13*1) |
|
| 1709 |
+ ADDQ R13, R11 |
|
| 1710 |
+ ADDQ R13, BX |
|
| 1711 |
+ JMP copy_2_end |
|
| 1712 |
+ |
|
| 1713 |
+copy_2_move_8through16: |
|
| 1714 |
+ MOVQ (R11), R12 |
|
| 1715 |
+ MOVQ -8(R11)(R13*1), R14 |
|
| 1716 |
+ MOVQ R12, (BX) |
|
| 1717 |
+ MOVQ R14, -8(BX)(R13*1) |
|
| 1718 |
+ ADDQ R13, R11 |
|
| 1719 |
+ ADDQ R13, BX |
|
| 1720 |
+ |
|
| 1721 |
+copy_2_end: |
|
| 1722 |
+ JMP handle_loop |
|
| 1723 |
+ |
|
| 1724 |
+ // Copy overlapping match |
|
| 1725 |
+copy_overlapping_match: |
|
| 1726 |
+ ADDQ R13, DI |
|
| 1727 |
+ |
|
| 1728 |
+copy_slow_3: |
|
| 1729 |
+ MOVB (R11), R12 |
|
| 1730 |
+ MOVB R12, (BX) |
|
| 1731 |
+ INCQ R11 |
|
| 1732 |
+ INCQ BX |
|
| 1733 |
+ DECQ R13 |
|
| 1734 |
+ JNZ copy_slow_3 |
|
| 1735 |
+ |
|
| 1736 |
+handle_loop: |
|
| 1737 |
+ ADDQ $0x18, AX |
|
| 1738 |
+ INCQ DX |
|
| 1739 |
+ CMPQ DX, CX |
|
| 1740 |
+ JB main_loop |
|
| 1741 |
+ |
|
| 1742 |
+loop_finished: |
|
| 1743 |
+ // Return value |
|
| 1744 |
+ MOVB $0x01, ret+8(FP) |
|
| 1745 |
+ |
|
| 1746 |
+ // Update the context |
|
| 1747 |
+ MOVQ ctx+0(FP), AX |
|
| 1748 |
+ MOVQ DX, 24(AX) |
|
| 1749 |
+ MOVQ DI, 104(AX) |
|
| 1750 |
+ MOVQ 80(AX), CX |
|
| 1751 |
+ SUBQ CX, SI |
|
| 1752 |
+ MOVQ SI, 112(AX) |
|
| 1753 |
+ RET |
|
| 1754 |
+ |
|
| 1755 |
+error_match_off_too_big: |
|
| 1756 |
+ // Return value |
|
| 1757 |
+ MOVB $0x00, ret+8(FP) |
|
| 1758 |
+ |
|
| 1759 |
+ // Update the context |
|
| 1760 |
+ MOVQ ctx+0(FP), AX |
|
| 1761 |
+ MOVQ DX, 24(AX) |
|
| 1762 |
+ MOVQ DI, 104(AX) |
|
| 1763 |
+ MOVQ 80(AX), CX |
|
| 1764 |
+ SUBQ CX, SI |
|
| 1765 |
+ MOVQ SI, 112(AX) |
|
| 1766 |
+ RET |
|
| 1767 |
+ |
|
| 1768 |
+empty_seqs: |
|
| 1769 |
+ // Return value |
|
| 1770 |
+ MOVB $0x01, ret+8(FP) |
|
| 1771 |
+ RET |
|
| 1772 |
+ |
|
| 1773 |
+// func sequenceDecs_decodeSync_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 1774 |
+// Requires: CMOV, SSE |
|
| 1775 |
+TEXT ·sequenceDecs_decodeSync_amd64(SB), $64-32 |
|
| 1776 |
+ MOVQ br+8(FP), AX |
|
| 1777 |
+ MOVQ 32(AX), DX |
|
| 1778 |
+ MOVBQZX 40(AX), BX |
|
| 1779 |
+ MOVQ 24(AX), SI |
|
| 1780 |
+ MOVQ (AX), AX |
|
| 1781 |
+ ADDQ SI, AX |
|
| 1782 |
+ MOVQ AX, (SP) |
|
| 1783 |
+ MOVQ ctx+16(FP), AX |
|
| 1784 |
+ MOVQ 72(AX), DI |
|
| 1785 |
+ MOVQ 80(AX), R8 |
|
| 1786 |
+ MOVQ 88(AX), R9 |
|
| 1787 |
+ XORQ CX, CX |
|
| 1788 |
+ MOVQ CX, 8(SP) |
|
| 1789 |
+ MOVQ CX, 16(SP) |
|
| 1790 |
+ MOVQ CX, 24(SP) |
|
| 1791 |
+ MOVQ 112(AX), R10 |
|
| 1792 |
+ MOVQ 128(AX), CX |
|
| 1793 |
+ MOVQ CX, 32(SP) |
|
| 1794 |
+ MOVQ 144(AX), R11 |
|
| 1795 |
+ MOVQ 136(AX), R12 |
|
| 1796 |
+ MOVQ 200(AX), CX |
|
| 1797 |
+ MOVQ CX, 56(SP) |
|
| 1798 |
+ MOVQ 176(AX), CX |
|
| 1799 |
+ MOVQ CX, 48(SP) |
|
| 1800 |
+ MOVQ 184(AX), AX |
|
| 1801 |
+ MOVQ AX, 40(SP) |
|
| 1802 |
+ MOVQ 40(SP), AX |
|
| 1803 |
+ ADDQ AX, 48(SP) |
|
| 1804 |
+ |
|
| 1805 |
+ // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) |
|
| 1806 |
+ ADDQ R10, 32(SP) |
|
| 1807 |
+ |
|
| 1808 |
+ // outBase += outPosition |
|
| 1809 |
+ ADDQ R12, R10 |
|
| 1810 |
+ |
|
| 1811 |
+sequenceDecs_decodeSync_amd64_main_loop: |
|
| 1812 |
+ MOVQ (SP), R13 |
|
| 1813 |
+ |
|
| 1814 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 1815 |
+ CMPQ SI, $0x08 |
|
| 1816 |
+ JL sequenceDecs_decodeSync_amd64_fill_byte_by_byte |
|
| 1817 |
+ MOVQ BX, AX |
|
| 1818 |
+ SHRQ $0x03, AX |
|
| 1819 |
+ SUBQ AX, R13 |
|
| 1820 |
+ MOVQ (R13), DX |
|
| 1821 |
+ SUBQ AX, SI |
|
| 1822 |
+ ANDQ $0x07, BX |
|
| 1823 |
+ JMP sequenceDecs_decodeSync_amd64_fill_end |
|
| 1824 |
+ |
|
| 1825 |
+sequenceDecs_decodeSync_amd64_fill_byte_by_byte: |
|
| 1826 |
+ CMPQ SI, $0x00 |
|
| 1827 |
+ JLE sequenceDecs_decodeSync_amd64_fill_end |
|
| 1828 |
+ CMPQ BX, $0x07 |
|
| 1829 |
+ JLE sequenceDecs_decodeSync_amd64_fill_end |
|
| 1830 |
+ SHLQ $0x08, DX |
|
| 1831 |
+ SUBQ $0x01, R13 |
|
| 1832 |
+ SUBQ $0x01, SI |
|
| 1833 |
+ SUBQ $0x08, BX |
|
| 1834 |
+ MOVBQZX (R13), AX |
|
| 1835 |
+ ORQ AX, DX |
|
| 1836 |
+ JMP sequenceDecs_decodeSync_amd64_fill_byte_by_byte |
|
| 1837 |
+ |
|
| 1838 |
+sequenceDecs_decodeSync_amd64_fill_end: |
|
| 1839 |
+ // Update offset |
|
| 1840 |
+ MOVQ R9, AX |
|
| 1841 |
+ MOVQ BX, CX |
|
| 1842 |
+ MOVQ DX, R14 |
|
| 1843 |
+ SHLQ CL, R14 |
|
| 1844 |
+ MOVB AH, CL |
|
| 1845 |
+ SHRQ $0x20, AX |
|
| 1846 |
+ TESTQ CX, CX |
|
| 1847 |
+ JZ sequenceDecs_decodeSync_amd64_of_update_zero |
|
| 1848 |
+ ADDQ CX, BX |
|
| 1849 |
+ CMPQ BX, $0x40 |
|
| 1850 |
+ JA sequenceDecs_decodeSync_amd64_of_update_zero |
|
| 1851 |
+ CMPQ CX, $0x40 |
|
| 1852 |
+ JAE sequenceDecs_decodeSync_amd64_of_update_zero |
|
| 1853 |
+ NEGQ CX |
|
| 1854 |
+ SHRQ CL, R14 |
|
| 1855 |
+ ADDQ R14, AX |
|
| 1856 |
+ |
|
| 1857 |
+sequenceDecs_decodeSync_amd64_of_update_zero: |
|
| 1858 |
+ MOVQ AX, 8(SP) |
|
| 1859 |
+ |
|
| 1860 |
+ // Update match length |
|
| 1861 |
+ MOVQ R8, AX |
|
| 1862 |
+ MOVQ BX, CX |
|
| 1863 |
+ MOVQ DX, R14 |
|
| 1864 |
+ SHLQ CL, R14 |
|
| 1865 |
+ MOVB AH, CL |
|
| 1866 |
+ SHRQ $0x20, AX |
|
| 1867 |
+ TESTQ CX, CX |
|
| 1868 |
+ JZ sequenceDecs_decodeSync_amd64_ml_update_zero |
|
| 1869 |
+ ADDQ CX, BX |
|
| 1870 |
+ CMPQ BX, $0x40 |
|
| 1871 |
+ JA sequenceDecs_decodeSync_amd64_ml_update_zero |
|
| 1872 |
+ CMPQ CX, $0x40 |
|
| 1873 |
+ JAE sequenceDecs_decodeSync_amd64_ml_update_zero |
|
| 1874 |
+ NEGQ CX |
|
| 1875 |
+ SHRQ CL, R14 |
|
| 1876 |
+ ADDQ R14, AX |
|
| 1877 |
+ |
|
| 1878 |
+sequenceDecs_decodeSync_amd64_ml_update_zero: |
|
| 1879 |
+ MOVQ AX, 16(SP) |
|
| 1880 |
+ |
|
| 1881 |
+ // Fill bitreader to have enough for the remaining |
|
| 1882 |
+ CMPQ SI, $0x08 |
|
| 1883 |
+ JL sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte |
|
| 1884 |
+ MOVQ BX, AX |
|
| 1885 |
+ SHRQ $0x03, AX |
|
| 1886 |
+ SUBQ AX, R13 |
|
| 1887 |
+ MOVQ (R13), DX |
|
| 1888 |
+ SUBQ AX, SI |
|
| 1889 |
+ ANDQ $0x07, BX |
|
| 1890 |
+ JMP sequenceDecs_decodeSync_amd64_fill_2_end |
|
| 1891 |
+ |
|
| 1892 |
+sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte: |
|
| 1893 |
+ CMPQ SI, $0x00 |
|
| 1894 |
+ JLE sequenceDecs_decodeSync_amd64_fill_2_end |
|
| 1895 |
+ CMPQ BX, $0x07 |
|
| 1896 |
+ JLE sequenceDecs_decodeSync_amd64_fill_2_end |
|
| 1897 |
+ SHLQ $0x08, DX |
|
| 1898 |
+ SUBQ $0x01, R13 |
|
| 1899 |
+ SUBQ $0x01, SI |
|
| 1900 |
+ SUBQ $0x08, BX |
|
| 1901 |
+ MOVBQZX (R13), AX |
|
| 1902 |
+ ORQ AX, DX |
|
| 1903 |
+ JMP sequenceDecs_decodeSync_amd64_fill_2_byte_by_byte |
|
| 1904 |
+ |
|
| 1905 |
+sequenceDecs_decodeSync_amd64_fill_2_end: |
|
| 1906 |
+ // Update literal length |
|
| 1907 |
+ MOVQ DI, AX |
|
| 1908 |
+ MOVQ BX, CX |
|
| 1909 |
+ MOVQ DX, R14 |
|
| 1910 |
+ SHLQ CL, R14 |
|
| 1911 |
+ MOVB AH, CL |
|
| 1912 |
+ SHRQ $0x20, AX |
|
| 1913 |
+ TESTQ CX, CX |
|
| 1914 |
+ JZ sequenceDecs_decodeSync_amd64_ll_update_zero |
|
| 1915 |
+ ADDQ CX, BX |
|
| 1916 |
+ CMPQ BX, $0x40 |
|
| 1917 |
+ JA sequenceDecs_decodeSync_amd64_ll_update_zero |
|
| 1918 |
+ CMPQ CX, $0x40 |
|
| 1919 |
+ JAE sequenceDecs_decodeSync_amd64_ll_update_zero |
|
| 1920 |
+ NEGQ CX |
|
| 1921 |
+ SHRQ CL, R14 |
|
| 1922 |
+ ADDQ R14, AX |
|
| 1923 |
+ |
|
| 1924 |
+sequenceDecs_decodeSync_amd64_ll_update_zero: |
|
| 1925 |
+ MOVQ AX, 24(SP) |
|
| 1926 |
+ |
|
| 1927 |
+ // Fill bitreader for state updates |
|
| 1928 |
+ MOVQ R13, (SP) |
|
| 1929 |
+ MOVQ R9, AX |
|
| 1930 |
+ SHRQ $0x08, AX |
|
| 1931 |
+ MOVBQZX AL, AX |
|
| 1932 |
+ MOVQ ctx+16(FP), CX |
|
| 1933 |
+ CMPQ 96(CX), $0x00 |
|
| 1934 |
+ JZ sequenceDecs_decodeSync_amd64_skip_update |
|
| 1935 |
+ |
|
| 1936 |
+ // Update Literal Length State |
|
| 1937 |
+ MOVBQZX DI, R13 |
|
| 1938 |
+ SHRQ $0x10, DI |
|
| 1939 |
+ MOVWQZX DI, DI |
|
| 1940 |
+ LEAQ (BX)(R13*1), CX |
|
| 1941 |
+ MOVQ DX, R14 |
|
| 1942 |
+ MOVQ CX, BX |
|
| 1943 |
+ ROLQ CL, R14 |
|
| 1944 |
+ MOVL $0x00000001, R15 |
|
| 1945 |
+ MOVB R13, CL |
|
| 1946 |
+ SHLL CL, R15 |
|
| 1947 |
+ DECL R15 |
|
| 1948 |
+ ANDQ R15, R14 |
|
| 1949 |
+ ADDQ R14, DI |
|
| 1950 |
+ |
|
| 1951 |
+ // Load ctx.llTable |
|
| 1952 |
+ MOVQ ctx+16(FP), CX |
|
| 1953 |
+ MOVQ (CX), CX |
|
| 1954 |
+ MOVQ (CX)(DI*8), DI |
|
| 1955 |
+ |
|
| 1956 |
+ // Update Match Length State |
|
| 1957 |
+ MOVBQZX R8, R13 |
|
| 1958 |
+ SHRQ $0x10, R8 |
|
| 1959 |
+ MOVWQZX R8, R8 |
|
| 1960 |
+ LEAQ (BX)(R13*1), CX |
|
| 1961 |
+ MOVQ DX, R14 |
|
| 1962 |
+ MOVQ CX, BX |
|
| 1963 |
+ ROLQ CL, R14 |
|
| 1964 |
+ MOVL $0x00000001, R15 |
|
| 1965 |
+ MOVB R13, CL |
|
| 1966 |
+ SHLL CL, R15 |
|
| 1967 |
+ DECL R15 |
|
| 1968 |
+ ANDQ R15, R14 |
|
| 1969 |
+ ADDQ R14, R8 |
|
| 1970 |
+ |
|
| 1971 |
+ // Load ctx.mlTable |
|
| 1972 |
+ MOVQ ctx+16(FP), CX |
|
| 1973 |
+ MOVQ 24(CX), CX |
|
| 1974 |
+ MOVQ (CX)(R8*8), R8 |
|
| 1975 |
+ |
|
| 1976 |
+ // Update Offset State |
|
| 1977 |
+ MOVBQZX R9, R13 |
|
| 1978 |
+ SHRQ $0x10, R9 |
|
| 1979 |
+ MOVWQZX R9, R9 |
|
| 1980 |
+ LEAQ (BX)(R13*1), CX |
|
| 1981 |
+ MOVQ DX, R14 |
|
| 1982 |
+ MOVQ CX, BX |
|
| 1983 |
+ ROLQ CL, R14 |
|
| 1984 |
+ MOVL $0x00000001, R15 |
|
| 1985 |
+ MOVB R13, CL |
|
| 1986 |
+ SHLL CL, R15 |
|
| 1987 |
+ DECL R15 |
|
| 1988 |
+ ANDQ R15, R14 |
|
| 1989 |
+ ADDQ R14, R9 |
|
| 1990 |
+ |
|
| 1991 |
+ // Load ctx.ofTable |
|
| 1992 |
+ MOVQ ctx+16(FP), CX |
|
| 1993 |
+ MOVQ 48(CX), CX |
|
| 1994 |
+ MOVQ (CX)(R9*8), R9 |
|
| 1995 |
+ |
|
| 1996 |
+sequenceDecs_decodeSync_amd64_skip_update: |
|
| 1997 |
+ // Adjust offset |
|
| 1998 |
+ MOVQ s+0(FP), CX |
|
| 1999 |
+ MOVQ 8(SP), R13 |
|
| 2000 |
+ CMPQ AX, $0x01 |
|
| 2001 |
+ JBE sequenceDecs_decodeSync_amd64_adjust_offsetB_1_or_0 |
|
| 2002 |
+ MOVUPS 144(CX), X0 |
|
| 2003 |
+ MOVQ R13, 144(CX) |
|
| 2004 |
+ MOVUPS X0, 152(CX) |
|
| 2005 |
+ JMP sequenceDecs_decodeSync_amd64_after_adjust |
|
| 2006 |
+ |
|
| 2007 |
+sequenceDecs_decodeSync_amd64_adjust_offsetB_1_or_0: |
|
| 2008 |
+ CMPQ 24(SP), $0x00000000 |
|
| 2009 |
+ JNE sequenceDecs_decodeSync_amd64_adjust_offset_maybezero |
|
| 2010 |
+ INCQ R13 |
|
| 2011 |
+ JMP sequenceDecs_decodeSync_amd64_adjust_offset_nonzero |
|
| 2012 |
+ |
|
| 2013 |
+sequenceDecs_decodeSync_amd64_adjust_offset_maybezero: |
|
| 2014 |
+ TESTQ R13, R13 |
|
| 2015 |
+ JNZ sequenceDecs_decodeSync_amd64_adjust_offset_nonzero |
|
| 2016 |
+ MOVQ 144(CX), R13 |
|
| 2017 |
+ JMP sequenceDecs_decodeSync_amd64_after_adjust |
|
| 2018 |
+ |
|
| 2019 |
+sequenceDecs_decodeSync_amd64_adjust_offset_nonzero: |
|
| 2020 |
+ MOVQ R13, AX |
|
| 2021 |
+ XORQ R14, R14 |
|
| 2022 |
+ MOVQ $-1, R15 |
|
| 2023 |
+ CMPQ R13, $0x03 |
|
| 2024 |
+ CMOVQEQ R14, AX |
|
| 2025 |
+ CMOVQEQ R15, R14 |
|
| 2026 |
+ ADDQ 144(CX)(AX*8), R14 |
|
| 2027 |
+ JNZ sequenceDecs_decodeSync_amd64_adjust_temp_valid |
|
| 2028 |
+ MOVQ $0x00000001, R14 |
|
| 2029 |
+ |
|
| 2030 |
+sequenceDecs_decodeSync_amd64_adjust_temp_valid: |
|
| 2031 |
+ CMPQ R13, $0x01 |
|
| 2032 |
+ JZ sequenceDecs_decodeSync_amd64_adjust_skip |
|
| 2033 |
+ MOVQ 152(CX), AX |
|
| 2034 |
+ MOVQ AX, 160(CX) |
|
| 2035 |
+ |
|
| 2036 |
+sequenceDecs_decodeSync_amd64_adjust_skip: |
|
| 2037 |
+ MOVQ 144(CX), AX |
|
| 2038 |
+ MOVQ AX, 152(CX) |
|
| 2039 |
+ MOVQ R14, 144(CX) |
|
| 2040 |
+ MOVQ R14, R13 |
|
| 2041 |
+ |
|
| 2042 |
+sequenceDecs_decodeSync_amd64_after_adjust: |
|
| 2043 |
+ MOVQ R13, 8(SP) |
|
| 2044 |
+ |
|
| 2045 |
+ // Check values |
|
| 2046 |
+ MOVQ 16(SP), AX |
|
| 2047 |
+ MOVQ 24(SP), CX |
|
| 2048 |
+ LEAQ (AX)(CX*1), R14 |
|
| 2049 |
+ MOVQ s+0(FP), R15 |
|
| 2050 |
+ ADDQ R14, 256(R15) |
|
| 2051 |
+ MOVQ ctx+16(FP), R14 |
|
| 2052 |
+ SUBQ CX, 104(R14) |
|
| 2053 |
+ JS error_not_enough_literals |
|
| 2054 |
+ CMPQ AX, $0x00020002 |
|
| 2055 |
+ JA sequenceDecs_decodeSync_amd64_error_match_len_too_big |
|
| 2056 |
+ TESTQ R13, R13 |
|
| 2057 |
+ JNZ sequenceDecs_decodeSync_amd64_match_len_ofs_ok |
|
| 2058 |
+ TESTQ AX, AX |
|
| 2059 |
+ JNZ sequenceDecs_decodeSync_amd64_error_match_len_ofs_mismatch |
|
| 2060 |
+ |
|
| 2061 |
+sequenceDecs_decodeSync_amd64_match_len_ofs_ok: |
|
| 2062 |
+ MOVQ 24(SP), AX |
|
| 2063 |
+ MOVQ 8(SP), CX |
|
| 2064 |
+ MOVQ 16(SP), R13 |
|
| 2065 |
+ |
|
| 2066 |
+ // Check if we have enough space in s.out |
|
| 2067 |
+ LEAQ (AX)(R13*1), R14 |
|
| 2068 |
+ ADDQ R10, R14 |
|
| 2069 |
+ CMPQ R14, 32(SP) |
|
| 2070 |
+ JA error_not_enough_space |
|
| 2071 |
+ |
|
| 2072 |
+ // Copy literals |
|
| 2073 |
+ TESTQ AX, AX |
|
| 2074 |
+ JZ check_offset |
|
| 2075 |
+ XORQ R14, R14 |
|
| 2076 |
+ |
|
| 2077 |
+copy_1: |
|
| 2078 |
+ MOVUPS (R11)(R14*1), X0 |
|
| 2079 |
+ MOVUPS X0, (R10)(R14*1) |
|
| 2080 |
+ ADDQ $0x10, R14 |
|
| 2081 |
+ CMPQ R14, AX |
|
| 2082 |
+ JB copy_1 |
|
| 2083 |
+ ADDQ AX, R11 |
|
| 2084 |
+ ADDQ AX, R10 |
|
| 2085 |
+ ADDQ AX, R12 |
|
| 2086 |
+ |
|
| 2087 |
+ // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) |
|
| 2088 |
+check_offset: |
|
| 2089 |
+ MOVQ R12, AX |
|
| 2090 |
+ ADDQ 40(SP), AX |
|
| 2091 |
+ CMPQ CX, AX |
|
| 2092 |
+ JG error_match_off_too_big |
|
| 2093 |
+ CMPQ CX, 56(SP) |
|
| 2094 |
+ JG error_match_off_too_big |
|
| 2095 |
+ |
|
| 2096 |
+ // Copy match from history |
|
| 2097 |
+ MOVQ CX, AX |
|
| 2098 |
+ SUBQ R12, AX |
|
| 2099 |
+ JLS copy_match |
|
| 2100 |
+ MOVQ 48(SP), R14 |
|
| 2101 |
+ SUBQ AX, R14 |
|
| 2102 |
+ CMPQ R13, AX |
|
| 2103 |
+ JG copy_all_from_history |
|
| 2104 |
+ MOVQ R13, AX |
|
| 2105 |
+ SUBQ $0x10, AX |
|
| 2106 |
+ JB copy_4_small |
|
| 2107 |
+ |
|
| 2108 |
+copy_4_loop: |
|
| 2109 |
+ MOVUPS (R14), X0 |
|
| 2110 |
+ MOVUPS X0, (R10) |
|
| 2111 |
+ ADDQ $0x10, R14 |
|
| 2112 |
+ ADDQ $0x10, R10 |
|
| 2113 |
+ SUBQ $0x10, AX |
|
| 2114 |
+ JAE copy_4_loop |
|
| 2115 |
+ LEAQ 16(R14)(AX*1), R14 |
|
| 2116 |
+ LEAQ 16(R10)(AX*1), R10 |
|
| 2117 |
+ MOVUPS -16(R14), X0 |
|
| 2118 |
+ MOVUPS X0, -16(R10) |
|
| 2119 |
+ JMP copy_4_end |
|
| 2120 |
+ |
|
| 2121 |
+copy_4_small: |
|
| 2122 |
+ CMPQ R13, $0x03 |
|
| 2123 |
+ JE copy_4_move_3 |
|
| 2124 |
+ CMPQ R13, $0x08 |
|
| 2125 |
+ JB copy_4_move_4through7 |
|
| 2126 |
+ JMP copy_4_move_8through16 |
|
| 2127 |
+ |
|
| 2128 |
+copy_4_move_3: |
|
| 2129 |
+ MOVW (R14), AX |
|
| 2130 |
+ MOVB 2(R14), CL |
|
| 2131 |
+ MOVW AX, (R10) |
|
| 2132 |
+ MOVB CL, 2(R10) |
|
| 2133 |
+ ADDQ R13, R14 |
|
| 2134 |
+ ADDQ R13, R10 |
|
| 2135 |
+ JMP copy_4_end |
|
| 2136 |
+ |
|
| 2137 |
+copy_4_move_4through7: |
|
| 2138 |
+ MOVL (R14), AX |
|
| 2139 |
+ MOVL -4(R14)(R13*1), CX |
|
| 2140 |
+ MOVL AX, (R10) |
|
| 2141 |
+ MOVL CX, -4(R10)(R13*1) |
|
| 2142 |
+ ADDQ R13, R14 |
|
| 2143 |
+ ADDQ R13, R10 |
|
| 2144 |
+ JMP copy_4_end |
|
| 2145 |
+ |
|
| 2146 |
+copy_4_move_8through16: |
|
| 2147 |
+ MOVQ (R14), AX |
|
| 2148 |
+ MOVQ -8(R14)(R13*1), CX |
|
| 2149 |
+ MOVQ AX, (R10) |
|
| 2150 |
+ MOVQ CX, -8(R10)(R13*1) |
|
| 2151 |
+ ADDQ R13, R14 |
|
| 2152 |
+ ADDQ R13, R10 |
|
| 2153 |
+ |
|
| 2154 |
+copy_4_end: |
|
| 2155 |
+ ADDQ R13, R12 |
|
| 2156 |
+ JMP handle_loop |
|
| 2157 |
+ JMP loop_finished |
|
| 2158 |
+ |
|
| 2159 |
+copy_all_from_history: |
|
| 2160 |
+ MOVQ AX, R15 |
|
| 2161 |
+ SUBQ $0x10, R15 |
|
| 2162 |
+ JB copy_5_small |
|
| 2163 |
+ |
|
| 2164 |
+copy_5_loop: |
|
| 2165 |
+ MOVUPS (R14), X0 |
|
| 2166 |
+ MOVUPS X0, (R10) |
|
| 2167 |
+ ADDQ $0x10, R14 |
|
| 2168 |
+ ADDQ $0x10, R10 |
|
| 2169 |
+ SUBQ $0x10, R15 |
|
| 2170 |
+ JAE copy_5_loop |
|
| 2171 |
+ LEAQ 16(R14)(R15*1), R14 |
|
| 2172 |
+ LEAQ 16(R10)(R15*1), R10 |
|
| 2173 |
+ MOVUPS -16(R14), X0 |
|
| 2174 |
+ MOVUPS X0, -16(R10) |
|
| 2175 |
+ JMP copy_5_end |
|
| 2176 |
+ |
|
| 2177 |
+copy_5_small: |
|
| 2178 |
+ CMPQ AX, $0x03 |
|
| 2179 |
+ JE copy_5_move_3 |
|
| 2180 |
+ JB copy_5_move_1or2 |
|
| 2181 |
+ CMPQ AX, $0x08 |
|
| 2182 |
+ JB copy_5_move_4through7 |
|
| 2183 |
+ JMP copy_5_move_8through16 |
|
| 2184 |
+ |
|
| 2185 |
+copy_5_move_1or2: |
|
| 2186 |
+ MOVB (R14), R15 |
|
| 2187 |
+ MOVB -1(R14)(AX*1), BP |
|
| 2188 |
+ MOVB R15, (R10) |
|
| 2189 |
+ MOVB BP, -1(R10)(AX*1) |
|
| 2190 |
+ ADDQ AX, R14 |
|
| 2191 |
+ ADDQ AX, R10 |
|
| 2192 |
+ JMP copy_5_end |
|
| 2193 |
+ |
|
| 2194 |
+copy_5_move_3: |
|
| 2195 |
+ MOVW (R14), R15 |
|
| 2196 |
+ MOVB 2(R14), BP |
|
| 2197 |
+ MOVW R15, (R10) |
|
| 2198 |
+ MOVB BP, 2(R10) |
|
| 2199 |
+ ADDQ AX, R14 |
|
| 2200 |
+ ADDQ AX, R10 |
|
| 2201 |
+ JMP copy_5_end |
|
| 2202 |
+ |
|
| 2203 |
+copy_5_move_4through7: |
|
| 2204 |
+ MOVL (R14), R15 |
|
| 2205 |
+ MOVL -4(R14)(AX*1), BP |
|
| 2206 |
+ MOVL R15, (R10) |
|
| 2207 |
+ MOVL BP, -4(R10)(AX*1) |
|
| 2208 |
+ ADDQ AX, R14 |
|
| 2209 |
+ ADDQ AX, R10 |
|
| 2210 |
+ JMP copy_5_end |
|
| 2211 |
+ |
|
| 2212 |
+copy_5_move_8through16: |
|
| 2213 |
+ MOVQ (R14), R15 |
|
| 2214 |
+ MOVQ -8(R14)(AX*1), BP |
|
| 2215 |
+ MOVQ R15, (R10) |
|
| 2216 |
+ MOVQ BP, -8(R10)(AX*1) |
|
| 2217 |
+ ADDQ AX, R14 |
|
| 2218 |
+ ADDQ AX, R10 |
|
| 2219 |
+ |
|
| 2220 |
+copy_5_end: |
|
| 2221 |
+ ADDQ AX, R12 |
|
| 2222 |
+ SUBQ AX, R13 |
|
| 2223 |
+ |
|
| 2224 |
+ // Copy match from the current buffer |
|
| 2225 |
+copy_match: |
|
| 2226 |
+ MOVQ R10, AX |
|
| 2227 |
+ SUBQ CX, AX |
|
| 2228 |
+ |
|
| 2229 |
+ // ml <= mo |
|
| 2230 |
+ CMPQ R13, CX |
|
| 2231 |
+ JA copy_overlapping_match |
|
| 2232 |
+ |
|
| 2233 |
+ // Copy non-overlapping match |
|
| 2234 |
+ ADDQ R13, R12 |
|
| 2235 |
+ MOVQ R10, CX |
|
| 2236 |
+ ADDQ R13, R10 |
|
| 2237 |
+ |
|
| 2238 |
+copy_2: |
|
| 2239 |
+ MOVUPS (AX), X0 |
|
| 2240 |
+ MOVUPS X0, (CX) |
|
| 2241 |
+ ADDQ $0x10, AX |
|
| 2242 |
+ ADDQ $0x10, CX |
|
| 2243 |
+ SUBQ $0x10, R13 |
|
| 2244 |
+ JHI copy_2 |
|
| 2245 |
+ JMP handle_loop |
|
| 2246 |
+ |
|
| 2247 |
+ // Copy overlapping match |
|
| 2248 |
+copy_overlapping_match: |
|
| 2249 |
+ ADDQ R13, R12 |
|
| 2250 |
+ |
|
| 2251 |
+copy_slow_3: |
|
| 2252 |
+ MOVB (AX), CL |
|
| 2253 |
+ MOVB CL, (R10) |
|
| 2254 |
+ INCQ AX |
|
| 2255 |
+ INCQ R10 |
|
| 2256 |
+ DECQ R13 |
|
| 2257 |
+ JNZ copy_slow_3 |
|
| 2258 |
+ |
|
| 2259 |
+handle_loop: |
|
| 2260 |
+ MOVQ ctx+16(FP), AX |
|
| 2261 |
+ DECQ 96(AX) |
|
| 2262 |
+ JNS sequenceDecs_decodeSync_amd64_main_loop |
|
| 2263 |
+ |
|
| 2264 |
+loop_finished: |
|
| 2265 |
+ MOVQ br+8(FP), AX |
|
| 2266 |
+ MOVQ DX, 32(AX) |
|
| 2267 |
+ MOVB BL, 40(AX) |
|
| 2268 |
+ MOVQ SI, 24(AX) |
|
| 2269 |
+ |
|
| 2270 |
+ // Update the context |
|
| 2271 |
+ MOVQ ctx+16(FP), AX |
|
| 2272 |
+ MOVQ R12, 136(AX) |
|
| 2273 |
+ MOVQ 144(AX), CX |
|
| 2274 |
+ SUBQ CX, R11 |
|
| 2275 |
+ MOVQ R11, 168(AX) |
|
| 2276 |
+ |
|
| 2277 |
+ // Return success |
|
| 2278 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 2279 |
+ RET |
|
| 2280 |
+ |
|
| 2281 |
+ // Return with match length error |
|
| 2282 |
+sequenceDecs_decodeSync_amd64_error_match_len_ofs_mismatch: |
|
| 2283 |
+ MOVQ 16(SP), AX |
|
| 2284 |
+ MOVQ ctx+16(FP), CX |
|
| 2285 |
+ MOVQ AX, 216(CX) |
|
| 2286 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 2287 |
+ RET |
|
| 2288 |
+ |
|
| 2289 |
+ // Return with match too long error |
|
| 2290 |
+sequenceDecs_decodeSync_amd64_error_match_len_too_big: |
|
| 2291 |
+ MOVQ ctx+16(FP), AX |
|
| 2292 |
+ MOVQ 16(SP), CX |
|
| 2293 |
+ MOVQ CX, 216(AX) |
|
| 2294 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 2295 |
+ RET |
|
| 2296 |
+ |
|
| 2297 |
+ // Return with match offset too long error |
|
| 2298 |
+error_match_off_too_big: |
|
| 2299 |
+ MOVQ ctx+16(FP), AX |
|
| 2300 |
+ MOVQ 8(SP), CX |
|
| 2301 |
+ MOVQ CX, 224(AX) |
|
| 2302 |
+ MOVQ R12, 136(AX) |
|
| 2303 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 2304 |
+ RET |
|
| 2305 |
+ |
|
| 2306 |
+ // Return with not enough literals error |
|
| 2307 |
+error_not_enough_literals: |
|
| 2308 |
+ MOVQ ctx+16(FP), AX |
|
| 2309 |
+ MOVQ 24(SP), CX |
|
| 2310 |
+ MOVQ CX, 208(AX) |
|
| 2311 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 2312 |
+ RET |
|
| 2313 |
+ |
|
| 2314 |
+ // Return with not enough output space error |
|
| 2315 |
+error_not_enough_space: |
|
| 2316 |
+ MOVQ ctx+16(FP), AX |
|
| 2317 |
+ MOVQ 24(SP), CX |
|
| 2318 |
+ MOVQ CX, 208(AX) |
|
| 2319 |
+ MOVQ 16(SP), CX |
|
| 2320 |
+ MOVQ CX, 216(AX) |
|
| 2321 |
+ MOVQ R12, 136(AX) |
|
| 2322 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 2323 |
+ RET |
|
| 2324 |
+ |
|
| 2325 |
+// func sequenceDecs_decodeSync_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 2326 |
+// Requires: BMI, BMI2, CMOV, SSE |
|
| 2327 |
+TEXT ·sequenceDecs_decodeSync_bmi2(SB), $64-32 |
|
| 2328 |
+ MOVQ br+8(FP), CX |
|
| 2329 |
+ MOVQ 32(CX), AX |
|
| 2330 |
+ MOVBQZX 40(CX), DX |
|
| 2331 |
+ MOVQ 24(CX), BX |
|
| 2332 |
+ MOVQ (CX), CX |
|
| 2333 |
+ ADDQ BX, CX |
|
| 2334 |
+ MOVQ CX, (SP) |
|
| 2335 |
+ MOVQ ctx+16(FP), CX |
|
| 2336 |
+ MOVQ 72(CX), SI |
|
| 2337 |
+ MOVQ 80(CX), DI |
|
| 2338 |
+ MOVQ 88(CX), R8 |
|
| 2339 |
+ XORQ R9, R9 |
|
| 2340 |
+ MOVQ R9, 8(SP) |
|
| 2341 |
+ MOVQ R9, 16(SP) |
|
| 2342 |
+ MOVQ R9, 24(SP) |
|
| 2343 |
+ MOVQ 112(CX), R9 |
|
| 2344 |
+ MOVQ 128(CX), R10 |
|
| 2345 |
+ MOVQ R10, 32(SP) |
|
| 2346 |
+ MOVQ 144(CX), R10 |
|
| 2347 |
+ MOVQ 136(CX), R11 |
|
| 2348 |
+ MOVQ 200(CX), R12 |
|
| 2349 |
+ MOVQ R12, 56(SP) |
|
| 2350 |
+ MOVQ 176(CX), R12 |
|
| 2351 |
+ MOVQ R12, 48(SP) |
|
| 2352 |
+ MOVQ 184(CX), CX |
|
| 2353 |
+ MOVQ CX, 40(SP) |
|
| 2354 |
+ MOVQ 40(SP), CX |
|
| 2355 |
+ ADDQ CX, 48(SP) |
|
| 2356 |
+ |
|
| 2357 |
+ // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) |
|
| 2358 |
+ ADDQ R9, 32(SP) |
|
| 2359 |
+ |
|
| 2360 |
+ // outBase += outPosition |
|
| 2361 |
+ ADDQ R11, R9 |
|
| 2362 |
+ |
|
| 2363 |
+sequenceDecs_decodeSync_bmi2_main_loop: |
|
| 2364 |
+ MOVQ (SP), R12 |
|
| 2365 |
+ |
|
| 2366 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 2367 |
+ CMPQ BX, $0x08 |
|
| 2368 |
+ JL sequenceDecs_decodeSync_bmi2_fill_byte_by_byte |
|
| 2369 |
+ MOVQ DX, CX |
|
| 2370 |
+ SHRQ $0x03, CX |
|
| 2371 |
+ SUBQ CX, R12 |
|
| 2372 |
+ MOVQ (R12), AX |
|
| 2373 |
+ SUBQ CX, BX |
|
| 2374 |
+ ANDQ $0x07, DX |
|
| 2375 |
+ JMP sequenceDecs_decodeSync_bmi2_fill_end |
|
| 2376 |
+ |
|
| 2377 |
+sequenceDecs_decodeSync_bmi2_fill_byte_by_byte: |
|
| 2378 |
+ CMPQ BX, $0x00 |
|
| 2379 |
+ JLE sequenceDecs_decodeSync_bmi2_fill_end |
|
| 2380 |
+ CMPQ DX, $0x07 |
|
| 2381 |
+ JLE sequenceDecs_decodeSync_bmi2_fill_end |
|
| 2382 |
+ SHLQ $0x08, AX |
|
| 2383 |
+ SUBQ $0x01, R12 |
|
| 2384 |
+ SUBQ $0x01, BX |
|
| 2385 |
+ SUBQ $0x08, DX |
|
| 2386 |
+ MOVBQZX (R12), CX |
|
| 2387 |
+ ORQ CX, AX |
|
| 2388 |
+ JMP sequenceDecs_decodeSync_bmi2_fill_byte_by_byte |
|
| 2389 |
+ |
|
| 2390 |
+sequenceDecs_decodeSync_bmi2_fill_end: |
|
| 2391 |
+ // Update offset |
|
| 2392 |
+ MOVQ $0x00000808, CX |
|
| 2393 |
+ BEXTRQ CX, R8, R13 |
|
| 2394 |
+ MOVQ AX, R14 |
|
| 2395 |
+ LEAQ (DX)(R13*1), CX |
|
| 2396 |
+ ROLQ CL, R14 |
|
| 2397 |
+ BZHIQ R13, R14, R14 |
|
| 2398 |
+ MOVQ CX, DX |
|
| 2399 |
+ MOVQ R8, CX |
|
| 2400 |
+ SHRQ $0x20, CX |
|
| 2401 |
+ ADDQ R14, CX |
|
| 2402 |
+ MOVQ CX, 8(SP) |
|
| 2403 |
+ |
|
| 2404 |
+ // Update match length |
|
| 2405 |
+ MOVQ $0x00000808, CX |
|
| 2406 |
+ BEXTRQ CX, DI, R13 |
|
| 2407 |
+ MOVQ AX, R14 |
|
| 2408 |
+ LEAQ (DX)(R13*1), CX |
|
| 2409 |
+ ROLQ CL, R14 |
|
| 2410 |
+ BZHIQ R13, R14, R14 |
|
| 2411 |
+ MOVQ CX, DX |
|
| 2412 |
+ MOVQ DI, CX |
|
| 2413 |
+ SHRQ $0x20, CX |
|
| 2414 |
+ ADDQ R14, CX |
|
| 2415 |
+ MOVQ CX, 16(SP) |
|
| 2416 |
+ |
|
| 2417 |
+ // Fill bitreader to have enough for the remaining |
|
| 2418 |
+ CMPQ BX, $0x08 |
|
| 2419 |
+ JL sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte |
|
| 2420 |
+ MOVQ DX, CX |
|
| 2421 |
+ SHRQ $0x03, CX |
|
| 2422 |
+ SUBQ CX, R12 |
|
| 2423 |
+ MOVQ (R12), AX |
|
| 2424 |
+ SUBQ CX, BX |
|
| 2425 |
+ ANDQ $0x07, DX |
|
| 2426 |
+ JMP sequenceDecs_decodeSync_bmi2_fill_2_end |
|
| 2427 |
+ |
|
| 2428 |
+sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte: |
|
| 2429 |
+ CMPQ BX, $0x00 |
|
| 2430 |
+ JLE sequenceDecs_decodeSync_bmi2_fill_2_end |
|
| 2431 |
+ CMPQ DX, $0x07 |
|
| 2432 |
+ JLE sequenceDecs_decodeSync_bmi2_fill_2_end |
|
| 2433 |
+ SHLQ $0x08, AX |
|
| 2434 |
+ SUBQ $0x01, R12 |
|
| 2435 |
+ SUBQ $0x01, BX |
|
| 2436 |
+ SUBQ $0x08, DX |
|
| 2437 |
+ MOVBQZX (R12), CX |
|
| 2438 |
+ ORQ CX, AX |
|
| 2439 |
+ JMP sequenceDecs_decodeSync_bmi2_fill_2_byte_by_byte |
|
| 2440 |
+ |
|
| 2441 |
+sequenceDecs_decodeSync_bmi2_fill_2_end: |
|
| 2442 |
+ // Update literal length |
|
| 2443 |
+ MOVQ $0x00000808, CX |
|
| 2444 |
+ BEXTRQ CX, SI, R13 |
|
| 2445 |
+ MOVQ AX, R14 |
|
| 2446 |
+ LEAQ (DX)(R13*1), CX |
|
| 2447 |
+ ROLQ CL, R14 |
|
| 2448 |
+ BZHIQ R13, R14, R14 |
|
| 2449 |
+ MOVQ CX, DX |
|
| 2450 |
+ MOVQ SI, CX |
|
| 2451 |
+ SHRQ $0x20, CX |
|
| 2452 |
+ ADDQ R14, CX |
|
| 2453 |
+ MOVQ CX, 24(SP) |
|
| 2454 |
+ |
|
| 2455 |
+ // Fill bitreader for state updates |
|
| 2456 |
+ MOVQ R12, (SP) |
|
| 2457 |
+ MOVQ $0x00000808, CX |
|
| 2458 |
+ BEXTRQ CX, R8, R12 |
|
| 2459 |
+ MOVQ ctx+16(FP), CX |
|
| 2460 |
+ CMPQ 96(CX), $0x00 |
|
| 2461 |
+ JZ sequenceDecs_decodeSync_bmi2_skip_update |
|
| 2462 |
+ LEAQ (SI)(DI*1), R13 |
|
| 2463 |
+ ADDQ R8, R13 |
|
| 2464 |
+ MOVBQZX R13, R13 |
|
| 2465 |
+ LEAQ (DX)(R13*1), CX |
|
| 2466 |
+ MOVQ AX, R14 |
|
| 2467 |
+ MOVQ CX, DX |
|
| 2468 |
+ ROLQ CL, R14 |
|
| 2469 |
+ BZHIQ R13, R14, R14 |
|
| 2470 |
+ |
|
| 2471 |
+ // Update Offset State |
|
| 2472 |
+ BZHIQ R8, R14, CX |
|
| 2473 |
+ SHRXQ R8, R14, R14 |
|
| 2474 |
+ MOVQ $0x00001010, R13 |
|
| 2475 |
+ BEXTRQ R13, R8, R8 |
|
| 2476 |
+ ADDQ CX, R8 |
|
| 2477 |
+ |
|
| 2478 |
+ // Load ctx.ofTable |
|
| 2479 |
+ MOVQ ctx+16(FP), CX |
|
| 2480 |
+ MOVQ 48(CX), CX |
|
| 2481 |
+ MOVQ (CX)(R8*8), R8 |
|
| 2482 |
+ |
|
| 2483 |
+ // Update Match Length State |
|
| 2484 |
+ BZHIQ DI, R14, CX |
|
| 2485 |
+ SHRXQ DI, R14, R14 |
|
| 2486 |
+ MOVQ $0x00001010, R13 |
|
| 2487 |
+ BEXTRQ R13, DI, DI |
|
| 2488 |
+ ADDQ CX, DI |
|
| 2489 |
+ |
|
| 2490 |
+ // Load ctx.mlTable |
|
| 2491 |
+ MOVQ ctx+16(FP), CX |
|
| 2492 |
+ MOVQ 24(CX), CX |
|
| 2493 |
+ MOVQ (CX)(DI*8), DI |
|
| 2494 |
+ |
|
| 2495 |
+ // Update Literal Length State |
|
| 2496 |
+ BZHIQ SI, R14, CX |
|
| 2497 |
+ MOVQ $0x00001010, R13 |
|
| 2498 |
+ BEXTRQ R13, SI, SI |
|
| 2499 |
+ ADDQ CX, SI |
|
| 2500 |
+ |
|
| 2501 |
+ // Load ctx.llTable |
|
| 2502 |
+ MOVQ ctx+16(FP), CX |
|
| 2503 |
+ MOVQ (CX), CX |
|
| 2504 |
+ MOVQ (CX)(SI*8), SI |
|
| 2505 |
+ |
|
| 2506 |
+sequenceDecs_decodeSync_bmi2_skip_update: |
|
| 2507 |
+ // Adjust offset |
|
| 2508 |
+ MOVQ s+0(FP), CX |
|
| 2509 |
+ MOVQ 8(SP), R13 |
|
| 2510 |
+ CMPQ R12, $0x01 |
|
| 2511 |
+ JBE sequenceDecs_decodeSync_bmi2_adjust_offsetB_1_or_0 |
|
| 2512 |
+ MOVUPS 144(CX), X0 |
|
| 2513 |
+ MOVQ R13, 144(CX) |
|
| 2514 |
+ MOVUPS X0, 152(CX) |
|
| 2515 |
+ JMP sequenceDecs_decodeSync_bmi2_after_adjust |
|
| 2516 |
+ |
|
| 2517 |
+sequenceDecs_decodeSync_bmi2_adjust_offsetB_1_or_0: |
|
| 2518 |
+ CMPQ 24(SP), $0x00000000 |
|
| 2519 |
+ JNE sequenceDecs_decodeSync_bmi2_adjust_offset_maybezero |
|
| 2520 |
+ INCQ R13 |
|
| 2521 |
+ JMP sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero |
|
| 2522 |
+ |
|
| 2523 |
+sequenceDecs_decodeSync_bmi2_adjust_offset_maybezero: |
|
| 2524 |
+ TESTQ R13, R13 |
|
| 2525 |
+ JNZ sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero |
|
| 2526 |
+ MOVQ 144(CX), R13 |
|
| 2527 |
+ JMP sequenceDecs_decodeSync_bmi2_after_adjust |
|
| 2528 |
+ |
|
| 2529 |
+sequenceDecs_decodeSync_bmi2_adjust_offset_nonzero: |
|
| 2530 |
+ MOVQ R13, R12 |
|
| 2531 |
+ XORQ R14, R14 |
|
| 2532 |
+ MOVQ $-1, R15 |
|
| 2533 |
+ CMPQ R13, $0x03 |
|
| 2534 |
+ CMOVQEQ R14, R12 |
|
| 2535 |
+ CMOVQEQ R15, R14 |
|
| 2536 |
+ ADDQ 144(CX)(R12*8), R14 |
|
| 2537 |
+ JNZ sequenceDecs_decodeSync_bmi2_adjust_temp_valid |
|
| 2538 |
+ MOVQ $0x00000001, R14 |
|
| 2539 |
+ |
|
| 2540 |
+sequenceDecs_decodeSync_bmi2_adjust_temp_valid: |
|
| 2541 |
+ CMPQ R13, $0x01 |
|
| 2542 |
+ JZ sequenceDecs_decodeSync_bmi2_adjust_skip |
|
| 2543 |
+ MOVQ 152(CX), R12 |
|
| 2544 |
+ MOVQ R12, 160(CX) |
|
| 2545 |
+ |
|
| 2546 |
+sequenceDecs_decodeSync_bmi2_adjust_skip: |
|
| 2547 |
+ MOVQ 144(CX), R12 |
|
| 2548 |
+ MOVQ R12, 152(CX) |
|
| 2549 |
+ MOVQ R14, 144(CX) |
|
| 2550 |
+ MOVQ R14, R13 |
|
| 2551 |
+ |
|
| 2552 |
+sequenceDecs_decodeSync_bmi2_after_adjust: |
|
| 2553 |
+ MOVQ R13, 8(SP) |
|
| 2554 |
+ |
|
| 2555 |
+ // Check values |
|
| 2556 |
+ MOVQ 16(SP), CX |
|
| 2557 |
+ MOVQ 24(SP), R12 |
|
| 2558 |
+ LEAQ (CX)(R12*1), R14 |
|
| 2559 |
+ MOVQ s+0(FP), R15 |
|
| 2560 |
+ ADDQ R14, 256(R15) |
|
| 2561 |
+ MOVQ ctx+16(FP), R14 |
|
| 2562 |
+ SUBQ R12, 104(R14) |
|
| 2563 |
+ JS error_not_enough_literals |
|
| 2564 |
+ CMPQ CX, $0x00020002 |
|
| 2565 |
+ JA sequenceDecs_decodeSync_bmi2_error_match_len_too_big |
|
| 2566 |
+ TESTQ R13, R13 |
|
| 2567 |
+ JNZ sequenceDecs_decodeSync_bmi2_match_len_ofs_ok |
|
| 2568 |
+ TESTQ CX, CX |
|
| 2569 |
+ JNZ sequenceDecs_decodeSync_bmi2_error_match_len_ofs_mismatch |
|
| 2570 |
+ |
|
| 2571 |
+sequenceDecs_decodeSync_bmi2_match_len_ofs_ok: |
|
| 2572 |
+ MOVQ 24(SP), CX |
|
| 2573 |
+ MOVQ 8(SP), R12 |
|
| 2574 |
+ MOVQ 16(SP), R13 |
|
| 2575 |
+ |
|
| 2576 |
+ // Check if we have enough space in s.out |
|
| 2577 |
+ LEAQ (CX)(R13*1), R14 |
|
| 2578 |
+ ADDQ R9, R14 |
|
| 2579 |
+ CMPQ R14, 32(SP) |
|
| 2580 |
+ JA error_not_enough_space |
|
| 2581 |
+ |
|
| 2582 |
+ // Copy literals |
|
| 2583 |
+ TESTQ CX, CX |
|
| 2584 |
+ JZ check_offset |
|
| 2585 |
+ XORQ R14, R14 |
|
| 2586 |
+ |
|
| 2587 |
+copy_1: |
|
| 2588 |
+ MOVUPS (R10)(R14*1), X0 |
|
| 2589 |
+ MOVUPS X0, (R9)(R14*1) |
|
| 2590 |
+ ADDQ $0x10, R14 |
|
| 2591 |
+ CMPQ R14, CX |
|
| 2592 |
+ JB copy_1 |
|
| 2593 |
+ ADDQ CX, R10 |
|
| 2594 |
+ ADDQ CX, R9 |
|
| 2595 |
+ ADDQ CX, R11 |
|
| 2596 |
+ |
|
| 2597 |
+ // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) |
|
| 2598 |
+check_offset: |
|
| 2599 |
+ MOVQ R11, CX |
|
| 2600 |
+ ADDQ 40(SP), CX |
|
| 2601 |
+ CMPQ R12, CX |
|
| 2602 |
+ JG error_match_off_too_big |
|
| 2603 |
+ CMPQ R12, 56(SP) |
|
| 2604 |
+ JG error_match_off_too_big |
|
| 2605 |
+ |
|
| 2606 |
+ // Copy match from history |
|
| 2607 |
+ MOVQ R12, CX |
|
| 2608 |
+ SUBQ R11, CX |
|
| 2609 |
+ JLS copy_match |
|
| 2610 |
+ MOVQ 48(SP), R14 |
|
| 2611 |
+ SUBQ CX, R14 |
|
| 2612 |
+ CMPQ R13, CX |
|
| 2613 |
+ JG copy_all_from_history |
|
| 2614 |
+ MOVQ R13, CX |
|
| 2615 |
+ SUBQ $0x10, CX |
|
| 2616 |
+ JB copy_4_small |
|
| 2617 |
+ |
|
| 2618 |
+copy_4_loop: |
|
| 2619 |
+ MOVUPS (R14), X0 |
|
| 2620 |
+ MOVUPS X0, (R9) |
|
| 2621 |
+ ADDQ $0x10, R14 |
|
| 2622 |
+ ADDQ $0x10, R9 |
|
| 2623 |
+ SUBQ $0x10, CX |
|
| 2624 |
+ JAE copy_4_loop |
|
| 2625 |
+ LEAQ 16(R14)(CX*1), R14 |
|
| 2626 |
+ LEAQ 16(R9)(CX*1), R9 |
|
| 2627 |
+ MOVUPS -16(R14), X0 |
|
| 2628 |
+ MOVUPS X0, -16(R9) |
|
| 2629 |
+ JMP copy_4_end |
|
| 2630 |
+ |
|
| 2631 |
+copy_4_small: |
|
| 2632 |
+ CMPQ R13, $0x03 |
|
| 2633 |
+ JE copy_4_move_3 |
|
| 2634 |
+ CMPQ R13, $0x08 |
|
| 2635 |
+ JB copy_4_move_4through7 |
|
| 2636 |
+ JMP copy_4_move_8through16 |
|
| 2637 |
+ |
|
| 2638 |
+copy_4_move_3: |
|
| 2639 |
+ MOVW (R14), CX |
|
| 2640 |
+ MOVB 2(R14), R12 |
|
| 2641 |
+ MOVW CX, (R9) |
|
| 2642 |
+ MOVB R12, 2(R9) |
|
| 2643 |
+ ADDQ R13, R14 |
|
| 2644 |
+ ADDQ R13, R9 |
|
| 2645 |
+ JMP copy_4_end |
|
| 2646 |
+ |
|
| 2647 |
+copy_4_move_4through7: |
|
| 2648 |
+ MOVL (R14), CX |
|
| 2649 |
+ MOVL -4(R14)(R13*1), R12 |
|
| 2650 |
+ MOVL CX, (R9) |
|
| 2651 |
+ MOVL R12, -4(R9)(R13*1) |
|
| 2652 |
+ ADDQ R13, R14 |
|
| 2653 |
+ ADDQ R13, R9 |
|
| 2654 |
+ JMP copy_4_end |
|
| 2655 |
+ |
|
| 2656 |
+copy_4_move_8through16: |
|
| 2657 |
+ MOVQ (R14), CX |
|
| 2658 |
+ MOVQ -8(R14)(R13*1), R12 |
|
| 2659 |
+ MOVQ CX, (R9) |
|
| 2660 |
+ MOVQ R12, -8(R9)(R13*1) |
|
| 2661 |
+ ADDQ R13, R14 |
|
| 2662 |
+ ADDQ R13, R9 |
|
| 2663 |
+ |
|
| 2664 |
+copy_4_end: |
|
| 2665 |
+ ADDQ R13, R11 |
|
| 2666 |
+ JMP handle_loop |
|
| 2667 |
+ JMP loop_finished |
|
| 2668 |
+ |
|
| 2669 |
+copy_all_from_history: |
|
| 2670 |
+ MOVQ CX, R15 |
|
| 2671 |
+ SUBQ $0x10, R15 |
|
| 2672 |
+ JB copy_5_small |
|
| 2673 |
+ |
|
| 2674 |
+copy_5_loop: |
|
| 2675 |
+ MOVUPS (R14), X0 |
|
| 2676 |
+ MOVUPS X0, (R9) |
|
| 2677 |
+ ADDQ $0x10, R14 |
|
| 2678 |
+ ADDQ $0x10, R9 |
|
| 2679 |
+ SUBQ $0x10, R15 |
|
| 2680 |
+ JAE copy_5_loop |
|
| 2681 |
+ LEAQ 16(R14)(R15*1), R14 |
|
| 2682 |
+ LEAQ 16(R9)(R15*1), R9 |
|
| 2683 |
+ MOVUPS -16(R14), X0 |
|
| 2684 |
+ MOVUPS X0, -16(R9) |
|
| 2685 |
+ JMP copy_5_end |
|
| 2686 |
+ |
|
| 2687 |
+copy_5_small: |
|
| 2688 |
+ CMPQ CX, $0x03 |
|
| 2689 |
+ JE copy_5_move_3 |
|
| 2690 |
+ JB copy_5_move_1or2 |
|
| 2691 |
+ CMPQ CX, $0x08 |
|
| 2692 |
+ JB copy_5_move_4through7 |
|
| 2693 |
+ JMP copy_5_move_8through16 |
|
| 2694 |
+ |
|
| 2695 |
+copy_5_move_1or2: |
|
| 2696 |
+ MOVB (R14), R15 |
|
| 2697 |
+ MOVB -1(R14)(CX*1), BP |
|
| 2698 |
+ MOVB R15, (R9) |
|
| 2699 |
+ MOVB BP, -1(R9)(CX*1) |
|
| 2700 |
+ ADDQ CX, R14 |
|
| 2701 |
+ ADDQ CX, R9 |
|
| 2702 |
+ JMP copy_5_end |
|
| 2703 |
+ |
|
| 2704 |
+copy_5_move_3: |
|
| 2705 |
+ MOVW (R14), R15 |
|
| 2706 |
+ MOVB 2(R14), BP |
|
| 2707 |
+ MOVW R15, (R9) |
|
| 2708 |
+ MOVB BP, 2(R9) |
|
| 2709 |
+ ADDQ CX, R14 |
|
| 2710 |
+ ADDQ CX, R9 |
|
| 2711 |
+ JMP copy_5_end |
|
| 2712 |
+ |
|
| 2713 |
+copy_5_move_4through7: |
|
| 2714 |
+ MOVL (R14), R15 |
|
| 2715 |
+ MOVL -4(R14)(CX*1), BP |
|
| 2716 |
+ MOVL R15, (R9) |
|
| 2717 |
+ MOVL BP, -4(R9)(CX*1) |
|
| 2718 |
+ ADDQ CX, R14 |
|
| 2719 |
+ ADDQ CX, R9 |
|
| 2720 |
+ JMP copy_5_end |
|
| 2721 |
+ |
|
| 2722 |
+copy_5_move_8through16: |
|
| 2723 |
+ MOVQ (R14), R15 |
|
| 2724 |
+ MOVQ -8(R14)(CX*1), BP |
|
| 2725 |
+ MOVQ R15, (R9) |
|
| 2726 |
+ MOVQ BP, -8(R9)(CX*1) |
|
| 2727 |
+ ADDQ CX, R14 |
|
| 2728 |
+ ADDQ CX, R9 |
|
| 2729 |
+ |
|
| 2730 |
+copy_5_end: |
|
| 2731 |
+ ADDQ CX, R11 |
|
| 2732 |
+ SUBQ CX, R13 |
|
| 2733 |
+ |
|
| 2734 |
+ // Copy match from the current buffer |
|
| 2735 |
+copy_match: |
|
| 2736 |
+ MOVQ R9, CX |
|
| 2737 |
+ SUBQ R12, CX |
|
| 2738 |
+ |
|
| 2739 |
+ // ml <= mo |
|
| 2740 |
+ CMPQ R13, R12 |
|
| 2741 |
+ JA copy_overlapping_match |
|
| 2742 |
+ |
|
| 2743 |
+ // Copy non-overlapping match |
|
| 2744 |
+ ADDQ R13, R11 |
|
| 2745 |
+ MOVQ R9, R12 |
|
| 2746 |
+ ADDQ R13, R9 |
|
| 2747 |
+ |
|
| 2748 |
+copy_2: |
|
| 2749 |
+ MOVUPS (CX), X0 |
|
| 2750 |
+ MOVUPS X0, (R12) |
|
| 2751 |
+ ADDQ $0x10, CX |
|
| 2752 |
+ ADDQ $0x10, R12 |
|
| 2753 |
+ SUBQ $0x10, R13 |
|
| 2754 |
+ JHI copy_2 |
|
| 2755 |
+ JMP handle_loop |
|
| 2756 |
+ |
|
| 2757 |
+ // Copy overlapping match |
|
| 2758 |
+copy_overlapping_match: |
|
| 2759 |
+ ADDQ R13, R11 |
|
| 2760 |
+ |
|
| 2761 |
+copy_slow_3: |
|
| 2762 |
+ MOVB (CX), R12 |
|
| 2763 |
+ MOVB R12, (R9) |
|
| 2764 |
+ INCQ CX |
|
| 2765 |
+ INCQ R9 |
|
| 2766 |
+ DECQ R13 |
|
| 2767 |
+ JNZ copy_slow_3 |
|
| 2768 |
+ |
|
| 2769 |
+handle_loop: |
|
| 2770 |
+ MOVQ ctx+16(FP), CX |
|
| 2771 |
+ DECQ 96(CX) |
|
| 2772 |
+ JNS sequenceDecs_decodeSync_bmi2_main_loop |
|
| 2773 |
+ |
|
| 2774 |
+loop_finished: |
|
| 2775 |
+ MOVQ br+8(FP), CX |
|
| 2776 |
+ MOVQ AX, 32(CX) |
|
| 2777 |
+ MOVB DL, 40(CX) |
|
| 2778 |
+ MOVQ BX, 24(CX) |
|
| 2779 |
+ |
|
| 2780 |
+ // Update the context |
|
| 2781 |
+ MOVQ ctx+16(FP), AX |
|
| 2782 |
+ MOVQ R11, 136(AX) |
|
| 2783 |
+ MOVQ 144(AX), CX |
|
| 2784 |
+ SUBQ CX, R10 |
|
| 2785 |
+ MOVQ R10, 168(AX) |
|
| 2786 |
+ |
|
| 2787 |
+ // Return success |
|
| 2788 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 2789 |
+ RET |
|
| 2790 |
+ |
|
| 2791 |
+ // Return with match length error |
|
| 2792 |
+sequenceDecs_decodeSync_bmi2_error_match_len_ofs_mismatch: |
|
| 2793 |
+ MOVQ 16(SP), AX |
|
| 2794 |
+ MOVQ ctx+16(FP), CX |
|
| 2795 |
+ MOVQ AX, 216(CX) |
|
| 2796 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 2797 |
+ RET |
|
| 2798 |
+ |
|
| 2799 |
+ // Return with match too long error |
|
| 2800 |
+sequenceDecs_decodeSync_bmi2_error_match_len_too_big: |
|
| 2801 |
+ MOVQ ctx+16(FP), AX |
|
| 2802 |
+ MOVQ 16(SP), CX |
|
| 2803 |
+ MOVQ CX, 216(AX) |
|
| 2804 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 2805 |
+ RET |
|
| 2806 |
+ |
|
| 2807 |
+ // Return with match offset too long error |
|
| 2808 |
+error_match_off_too_big: |
|
| 2809 |
+ MOVQ ctx+16(FP), AX |
|
| 2810 |
+ MOVQ 8(SP), CX |
|
| 2811 |
+ MOVQ CX, 224(AX) |
|
| 2812 |
+ MOVQ R11, 136(AX) |
|
| 2813 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 2814 |
+ RET |
|
| 2815 |
+ |
|
| 2816 |
+ // Return with not enough literals error |
|
| 2817 |
+error_not_enough_literals: |
|
| 2818 |
+ MOVQ ctx+16(FP), AX |
|
| 2819 |
+ MOVQ 24(SP), CX |
|
| 2820 |
+ MOVQ CX, 208(AX) |
|
| 2821 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 2822 |
+ RET |
|
| 2823 |
+ |
|
| 2824 |
+ // Return with not enough output space error |
|
| 2825 |
+error_not_enough_space: |
|
| 2826 |
+ MOVQ ctx+16(FP), AX |
|
| 2827 |
+ MOVQ 24(SP), CX |
|
| 2828 |
+ MOVQ CX, 208(AX) |
|
| 2829 |
+ MOVQ 16(SP), CX |
|
| 2830 |
+ MOVQ CX, 216(AX) |
|
| 2831 |
+ MOVQ R11, 136(AX) |
|
| 2832 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 2833 |
+ RET |
|
| 2834 |
+ |
|
| 2835 |
+// func sequenceDecs_decodeSync_safe_amd64(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 2836 |
+// Requires: CMOV, SSE |
|
| 2837 |
+TEXT ·sequenceDecs_decodeSync_safe_amd64(SB), $64-32 |
|
| 2838 |
+ MOVQ br+8(FP), AX |
|
| 2839 |
+ MOVQ 32(AX), DX |
|
| 2840 |
+ MOVBQZX 40(AX), BX |
|
| 2841 |
+ MOVQ 24(AX), SI |
|
| 2842 |
+ MOVQ (AX), AX |
|
| 2843 |
+ ADDQ SI, AX |
|
| 2844 |
+ MOVQ AX, (SP) |
|
| 2845 |
+ MOVQ ctx+16(FP), AX |
|
| 2846 |
+ MOVQ 72(AX), DI |
|
| 2847 |
+ MOVQ 80(AX), R8 |
|
| 2848 |
+ MOVQ 88(AX), R9 |
|
| 2849 |
+ XORQ CX, CX |
|
| 2850 |
+ MOVQ CX, 8(SP) |
|
| 2851 |
+ MOVQ CX, 16(SP) |
|
| 2852 |
+ MOVQ CX, 24(SP) |
|
| 2853 |
+ MOVQ 112(AX), R10 |
|
| 2854 |
+ MOVQ 128(AX), CX |
|
| 2855 |
+ MOVQ CX, 32(SP) |
|
| 2856 |
+ MOVQ 144(AX), R11 |
|
| 2857 |
+ MOVQ 136(AX), R12 |
|
| 2858 |
+ MOVQ 200(AX), CX |
|
| 2859 |
+ MOVQ CX, 56(SP) |
|
| 2860 |
+ MOVQ 176(AX), CX |
|
| 2861 |
+ MOVQ CX, 48(SP) |
|
| 2862 |
+ MOVQ 184(AX), AX |
|
| 2863 |
+ MOVQ AX, 40(SP) |
|
| 2864 |
+ MOVQ 40(SP), AX |
|
| 2865 |
+ ADDQ AX, 48(SP) |
|
| 2866 |
+ |
|
| 2867 |
+ // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) |
|
| 2868 |
+ ADDQ R10, 32(SP) |
|
| 2869 |
+ |
|
| 2870 |
+ // outBase += outPosition |
|
| 2871 |
+ ADDQ R12, R10 |
|
| 2872 |
+ |
|
| 2873 |
+sequenceDecs_decodeSync_safe_amd64_main_loop: |
|
| 2874 |
+ MOVQ (SP), R13 |
|
| 2875 |
+ |
|
| 2876 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 2877 |
+ CMPQ SI, $0x08 |
|
| 2878 |
+ JL sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte |
|
| 2879 |
+ MOVQ BX, AX |
|
| 2880 |
+ SHRQ $0x03, AX |
|
| 2881 |
+ SUBQ AX, R13 |
|
| 2882 |
+ MOVQ (R13), DX |
|
| 2883 |
+ SUBQ AX, SI |
|
| 2884 |
+ ANDQ $0x07, BX |
|
| 2885 |
+ JMP sequenceDecs_decodeSync_safe_amd64_fill_end |
|
| 2886 |
+ |
|
| 2887 |
+sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte: |
|
| 2888 |
+ CMPQ SI, $0x00 |
|
| 2889 |
+ JLE sequenceDecs_decodeSync_safe_amd64_fill_end |
|
| 2890 |
+ CMPQ BX, $0x07 |
|
| 2891 |
+ JLE sequenceDecs_decodeSync_safe_amd64_fill_end |
|
| 2892 |
+ SHLQ $0x08, DX |
|
| 2893 |
+ SUBQ $0x01, R13 |
|
| 2894 |
+ SUBQ $0x01, SI |
|
| 2895 |
+ SUBQ $0x08, BX |
|
| 2896 |
+ MOVBQZX (R13), AX |
|
| 2897 |
+ ORQ AX, DX |
|
| 2898 |
+ JMP sequenceDecs_decodeSync_safe_amd64_fill_byte_by_byte |
|
| 2899 |
+ |
|
| 2900 |
+sequenceDecs_decodeSync_safe_amd64_fill_end: |
|
| 2901 |
+ // Update offset |
|
| 2902 |
+ MOVQ R9, AX |
|
| 2903 |
+ MOVQ BX, CX |
|
| 2904 |
+ MOVQ DX, R14 |
|
| 2905 |
+ SHLQ CL, R14 |
|
| 2906 |
+ MOVB AH, CL |
|
| 2907 |
+ SHRQ $0x20, AX |
|
| 2908 |
+ TESTQ CX, CX |
|
| 2909 |
+ JZ sequenceDecs_decodeSync_safe_amd64_of_update_zero |
|
| 2910 |
+ ADDQ CX, BX |
|
| 2911 |
+ CMPQ BX, $0x40 |
|
| 2912 |
+ JA sequenceDecs_decodeSync_safe_amd64_of_update_zero |
|
| 2913 |
+ CMPQ CX, $0x40 |
|
| 2914 |
+ JAE sequenceDecs_decodeSync_safe_amd64_of_update_zero |
|
| 2915 |
+ NEGQ CX |
|
| 2916 |
+ SHRQ CL, R14 |
|
| 2917 |
+ ADDQ R14, AX |
|
| 2918 |
+ |
|
| 2919 |
+sequenceDecs_decodeSync_safe_amd64_of_update_zero: |
|
| 2920 |
+ MOVQ AX, 8(SP) |
|
| 2921 |
+ |
|
| 2922 |
+ // Update match length |
|
| 2923 |
+ MOVQ R8, AX |
|
| 2924 |
+ MOVQ BX, CX |
|
| 2925 |
+ MOVQ DX, R14 |
|
| 2926 |
+ SHLQ CL, R14 |
|
| 2927 |
+ MOVB AH, CL |
|
| 2928 |
+ SHRQ $0x20, AX |
|
| 2929 |
+ TESTQ CX, CX |
|
| 2930 |
+ JZ sequenceDecs_decodeSync_safe_amd64_ml_update_zero |
|
| 2931 |
+ ADDQ CX, BX |
|
| 2932 |
+ CMPQ BX, $0x40 |
|
| 2933 |
+ JA sequenceDecs_decodeSync_safe_amd64_ml_update_zero |
|
| 2934 |
+ CMPQ CX, $0x40 |
|
| 2935 |
+ JAE sequenceDecs_decodeSync_safe_amd64_ml_update_zero |
|
| 2936 |
+ NEGQ CX |
|
| 2937 |
+ SHRQ CL, R14 |
|
| 2938 |
+ ADDQ R14, AX |
|
| 2939 |
+ |
|
| 2940 |
+sequenceDecs_decodeSync_safe_amd64_ml_update_zero: |
|
| 2941 |
+ MOVQ AX, 16(SP) |
|
| 2942 |
+ |
|
| 2943 |
+ // Fill bitreader to have enough for the remaining |
|
| 2944 |
+ CMPQ SI, $0x08 |
|
| 2945 |
+ JL sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte |
|
| 2946 |
+ MOVQ BX, AX |
|
| 2947 |
+ SHRQ $0x03, AX |
|
| 2948 |
+ SUBQ AX, R13 |
|
| 2949 |
+ MOVQ (R13), DX |
|
| 2950 |
+ SUBQ AX, SI |
|
| 2951 |
+ ANDQ $0x07, BX |
|
| 2952 |
+ JMP sequenceDecs_decodeSync_safe_amd64_fill_2_end |
|
| 2953 |
+ |
|
| 2954 |
+sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte: |
|
| 2955 |
+ CMPQ SI, $0x00 |
|
| 2956 |
+ JLE sequenceDecs_decodeSync_safe_amd64_fill_2_end |
|
| 2957 |
+ CMPQ BX, $0x07 |
|
| 2958 |
+ JLE sequenceDecs_decodeSync_safe_amd64_fill_2_end |
|
| 2959 |
+ SHLQ $0x08, DX |
|
| 2960 |
+ SUBQ $0x01, R13 |
|
| 2961 |
+ SUBQ $0x01, SI |
|
| 2962 |
+ SUBQ $0x08, BX |
|
| 2963 |
+ MOVBQZX (R13), AX |
|
| 2964 |
+ ORQ AX, DX |
|
| 2965 |
+ JMP sequenceDecs_decodeSync_safe_amd64_fill_2_byte_by_byte |
|
| 2966 |
+ |
|
| 2967 |
+sequenceDecs_decodeSync_safe_amd64_fill_2_end: |
|
| 2968 |
+ // Update literal length |
|
| 2969 |
+ MOVQ DI, AX |
|
| 2970 |
+ MOVQ BX, CX |
|
| 2971 |
+ MOVQ DX, R14 |
|
| 2972 |
+ SHLQ CL, R14 |
|
| 2973 |
+ MOVB AH, CL |
|
| 2974 |
+ SHRQ $0x20, AX |
|
| 2975 |
+ TESTQ CX, CX |
|
| 2976 |
+ JZ sequenceDecs_decodeSync_safe_amd64_ll_update_zero |
|
| 2977 |
+ ADDQ CX, BX |
|
| 2978 |
+ CMPQ BX, $0x40 |
|
| 2979 |
+ JA sequenceDecs_decodeSync_safe_amd64_ll_update_zero |
|
| 2980 |
+ CMPQ CX, $0x40 |
|
| 2981 |
+ JAE sequenceDecs_decodeSync_safe_amd64_ll_update_zero |
|
| 2982 |
+ NEGQ CX |
|
| 2983 |
+ SHRQ CL, R14 |
|
| 2984 |
+ ADDQ R14, AX |
|
| 2985 |
+ |
|
| 2986 |
+sequenceDecs_decodeSync_safe_amd64_ll_update_zero: |
|
| 2987 |
+ MOVQ AX, 24(SP) |
|
| 2988 |
+ |
|
| 2989 |
+ // Fill bitreader for state updates |
|
| 2990 |
+ MOVQ R13, (SP) |
|
| 2991 |
+ MOVQ R9, AX |
|
| 2992 |
+ SHRQ $0x08, AX |
|
| 2993 |
+ MOVBQZX AL, AX |
|
| 2994 |
+ MOVQ ctx+16(FP), CX |
|
| 2995 |
+ CMPQ 96(CX), $0x00 |
|
| 2996 |
+ JZ sequenceDecs_decodeSync_safe_amd64_skip_update |
|
| 2997 |
+ |
|
| 2998 |
+ // Update Literal Length State |
|
| 2999 |
+ MOVBQZX DI, R13 |
|
| 3000 |
+ SHRQ $0x10, DI |
|
| 3001 |
+ MOVWQZX DI, DI |
|
| 3002 |
+ LEAQ (BX)(R13*1), CX |
|
| 3003 |
+ MOVQ DX, R14 |
|
| 3004 |
+ MOVQ CX, BX |
|
| 3005 |
+ ROLQ CL, R14 |
|
| 3006 |
+ MOVL $0x00000001, R15 |
|
| 3007 |
+ MOVB R13, CL |
|
| 3008 |
+ SHLL CL, R15 |
|
| 3009 |
+ DECL R15 |
|
| 3010 |
+ ANDQ R15, R14 |
|
| 3011 |
+ ADDQ R14, DI |
|
| 3012 |
+ |
|
| 3013 |
+ // Load ctx.llTable |
|
| 3014 |
+ MOVQ ctx+16(FP), CX |
|
| 3015 |
+ MOVQ (CX), CX |
|
| 3016 |
+ MOVQ (CX)(DI*8), DI |
|
| 3017 |
+ |
|
| 3018 |
+ // Update Match Length State |
|
| 3019 |
+ MOVBQZX R8, R13 |
|
| 3020 |
+ SHRQ $0x10, R8 |
|
| 3021 |
+ MOVWQZX R8, R8 |
|
| 3022 |
+ LEAQ (BX)(R13*1), CX |
|
| 3023 |
+ MOVQ DX, R14 |
|
| 3024 |
+ MOVQ CX, BX |
|
| 3025 |
+ ROLQ CL, R14 |
|
| 3026 |
+ MOVL $0x00000001, R15 |
|
| 3027 |
+ MOVB R13, CL |
|
| 3028 |
+ SHLL CL, R15 |
|
| 3029 |
+ DECL R15 |
|
| 3030 |
+ ANDQ R15, R14 |
|
| 3031 |
+ ADDQ R14, R8 |
|
| 3032 |
+ |
|
| 3033 |
+ // Load ctx.mlTable |
|
| 3034 |
+ MOVQ ctx+16(FP), CX |
|
| 3035 |
+ MOVQ 24(CX), CX |
|
| 3036 |
+ MOVQ (CX)(R8*8), R8 |
|
| 3037 |
+ |
|
| 3038 |
+ // Update Offset State |
|
| 3039 |
+ MOVBQZX R9, R13 |
|
| 3040 |
+ SHRQ $0x10, R9 |
|
| 3041 |
+ MOVWQZX R9, R9 |
|
| 3042 |
+ LEAQ (BX)(R13*1), CX |
|
| 3043 |
+ MOVQ DX, R14 |
|
| 3044 |
+ MOVQ CX, BX |
|
| 3045 |
+ ROLQ CL, R14 |
|
| 3046 |
+ MOVL $0x00000001, R15 |
|
| 3047 |
+ MOVB R13, CL |
|
| 3048 |
+ SHLL CL, R15 |
|
| 3049 |
+ DECL R15 |
|
| 3050 |
+ ANDQ R15, R14 |
|
| 3051 |
+ ADDQ R14, R9 |
|
| 3052 |
+ |
|
| 3053 |
+ // Load ctx.ofTable |
|
| 3054 |
+ MOVQ ctx+16(FP), CX |
|
| 3055 |
+ MOVQ 48(CX), CX |
|
| 3056 |
+ MOVQ (CX)(R9*8), R9 |
|
| 3057 |
+ |
|
| 3058 |
+sequenceDecs_decodeSync_safe_amd64_skip_update: |
|
| 3059 |
+ // Adjust offset |
|
| 3060 |
+ MOVQ s+0(FP), CX |
|
| 3061 |
+ MOVQ 8(SP), R13 |
|
| 3062 |
+ CMPQ AX, $0x01 |
|
| 3063 |
+ JBE sequenceDecs_decodeSync_safe_amd64_adjust_offsetB_1_or_0 |
|
| 3064 |
+ MOVUPS 144(CX), X0 |
|
| 3065 |
+ MOVQ R13, 144(CX) |
|
| 3066 |
+ MOVUPS X0, 152(CX) |
|
| 3067 |
+ JMP sequenceDecs_decodeSync_safe_amd64_after_adjust |
|
| 3068 |
+ |
|
| 3069 |
+sequenceDecs_decodeSync_safe_amd64_adjust_offsetB_1_or_0: |
|
| 3070 |
+ CMPQ 24(SP), $0x00000000 |
|
| 3071 |
+ JNE sequenceDecs_decodeSync_safe_amd64_adjust_offset_maybezero |
|
| 3072 |
+ INCQ R13 |
|
| 3073 |
+ JMP sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero |
|
| 3074 |
+ |
|
| 3075 |
+sequenceDecs_decodeSync_safe_amd64_adjust_offset_maybezero: |
|
| 3076 |
+ TESTQ R13, R13 |
|
| 3077 |
+ JNZ sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero |
|
| 3078 |
+ MOVQ 144(CX), R13 |
|
| 3079 |
+ JMP sequenceDecs_decodeSync_safe_amd64_after_adjust |
|
| 3080 |
+ |
|
| 3081 |
+sequenceDecs_decodeSync_safe_amd64_adjust_offset_nonzero: |
|
| 3082 |
+ MOVQ R13, AX |
|
| 3083 |
+ XORQ R14, R14 |
|
| 3084 |
+ MOVQ $-1, R15 |
|
| 3085 |
+ CMPQ R13, $0x03 |
|
| 3086 |
+ CMOVQEQ R14, AX |
|
| 3087 |
+ CMOVQEQ R15, R14 |
|
| 3088 |
+ ADDQ 144(CX)(AX*8), R14 |
|
| 3089 |
+ JNZ sequenceDecs_decodeSync_safe_amd64_adjust_temp_valid |
|
| 3090 |
+ MOVQ $0x00000001, R14 |
|
| 3091 |
+ |
|
| 3092 |
+sequenceDecs_decodeSync_safe_amd64_adjust_temp_valid: |
|
| 3093 |
+ CMPQ R13, $0x01 |
|
| 3094 |
+ JZ sequenceDecs_decodeSync_safe_amd64_adjust_skip |
|
| 3095 |
+ MOVQ 152(CX), AX |
|
| 3096 |
+ MOVQ AX, 160(CX) |
|
| 3097 |
+ |
|
| 3098 |
+sequenceDecs_decodeSync_safe_amd64_adjust_skip: |
|
| 3099 |
+ MOVQ 144(CX), AX |
|
| 3100 |
+ MOVQ AX, 152(CX) |
|
| 3101 |
+ MOVQ R14, 144(CX) |
|
| 3102 |
+ MOVQ R14, R13 |
|
| 3103 |
+ |
|
| 3104 |
+sequenceDecs_decodeSync_safe_amd64_after_adjust: |
|
| 3105 |
+ MOVQ R13, 8(SP) |
|
| 3106 |
+ |
|
| 3107 |
+ // Check values |
|
| 3108 |
+ MOVQ 16(SP), AX |
|
| 3109 |
+ MOVQ 24(SP), CX |
|
| 3110 |
+ LEAQ (AX)(CX*1), R14 |
|
| 3111 |
+ MOVQ s+0(FP), R15 |
|
| 3112 |
+ ADDQ R14, 256(R15) |
|
| 3113 |
+ MOVQ ctx+16(FP), R14 |
|
| 3114 |
+ SUBQ CX, 104(R14) |
|
| 3115 |
+ JS error_not_enough_literals |
|
| 3116 |
+ CMPQ AX, $0x00020002 |
|
| 3117 |
+ JA sequenceDecs_decodeSync_safe_amd64_error_match_len_too_big |
|
| 3118 |
+ TESTQ R13, R13 |
|
| 3119 |
+ JNZ sequenceDecs_decodeSync_safe_amd64_match_len_ofs_ok |
|
| 3120 |
+ TESTQ AX, AX |
|
| 3121 |
+ JNZ sequenceDecs_decodeSync_safe_amd64_error_match_len_ofs_mismatch |
|
| 3122 |
+ |
|
| 3123 |
+sequenceDecs_decodeSync_safe_amd64_match_len_ofs_ok: |
|
| 3124 |
+ MOVQ 24(SP), AX |
|
| 3125 |
+ MOVQ 8(SP), CX |
|
| 3126 |
+ MOVQ 16(SP), R13 |
|
| 3127 |
+ |
|
| 3128 |
+ // Check if we have enough space in s.out |
|
| 3129 |
+ LEAQ (AX)(R13*1), R14 |
|
| 3130 |
+ ADDQ R10, R14 |
|
| 3131 |
+ CMPQ R14, 32(SP) |
|
| 3132 |
+ JA error_not_enough_space |
|
| 3133 |
+ |
|
| 3134 |
+ // Copy literals |
|
| 3135 |
+ TESTQ AX, AX |
|
| 3136 |
+ JZ check_offset |
|
| 3137 |
+ MOVQ AX, R14 |
|
| 3138 |
+ SUBQ $0x10, R14 |
|
| 3139 |
+ JB copy_1_small |
|
| 3140 |
+ |
|
| 3141 |
+copy_1_loop: |
|
| 3142 |
+ MOVUPS (R11), X0 |
|
| 3143 |
+ MOVUPS X0, (R10) |
|
| 3144 |
+ ADDQ $0x10, R11 |
|
| 3145 |
+ ADDQ $0x10, R10 |
|
| 3146 |
+ SUBQ $0x10, R14 |
|
| 3147 |
+ JAE copy_1_loop |
|
| 3148 |
+ LEAQ 16(R11)(R14*1), R11 |
|
| 3149 |
+ LEAQ 16(R10)(R14*1), R10 |
|
| 3150 |
+ MOVUPS -16(R11), X0 |
|
| 3151 |
+ MOVUPS X0, -16(R10) |
|
| 3152 |
+ JMP copy_1_end |
|
| 3153 |
+ |
|
| 3154 |
+copy_1_small: |
|
| 3155 |
+ CMPQ AX, $0x03 |
|
| 3156 |
+ JE copy_1_move_3 |
|
| 3157 |
+ JB copy_1_move_1or2 |
|
| 3158 |
+ CMPQ AX, $0x08 |
|
| 3159 |
+ JB copy_1_move_4through7 |
|
| 3160 |
+ JMP copy_1_move_8through16 |
|
| 3161 |
+ |
|
| 3162 |
+copy_1_move_1or2: |
|
| 3163 |
+ MOVB (R11), R14 |
|
| 3164 |
+ MOVB -1(R11)(AX*1), R15 |
|
| 3165 |
+ MOVB R14, (R10) |
|
| 3166 |
+ MOVB R15, -1(R10)(AX*1) |
|
| 3167 |
+ ADDQ AX, R11 |
|
| 3168 |
+ ADDQ AX, R10 |
|
| 3169 |
+ JMP copy_1_end |
|
| 3170 |
+ |
|
| 3171 |
+copy_1_move_3: |
|
| 3172 |
+ MOVW (R11), R14 |
|
| 3173 |
+ MOVB 2(R11), R15 |
|
| 3174 |
+ MOVW R14, (R10) |
|
| 3175 |
+ MOVB R15, 2(R10) |
|
| 3176 |
+ ADDQ AX, R11 |
|
| 3177 |
+ ADDQ AX, R10 |
|
| 3178 |
+ JMP copy_1_end |
|
| 3179 |
+ |
|
| 3180 |
+copy_1_move_4through7: |
|
| 3181 |
+ MOVL (R11), R14 |
|
| 3182 |
+ MOVL -4(R11)(AX*1), R15 |
|
| 3183 |
+ MOVL R14, (R10) |
|
| 3184 |
+ MOVL R15, -4(R10)(AX*1) |
|
| 3185 |
+ ADDQ AX, R11 |
|
| 3186 |
+ ADDQ AX, R10 |
|
| 3187 |
+ JMP copy_1_end |
|
| 3188 |
+ |
|
| 3189 |
+copy_1_move_8through16: |
|
| 3190 |
+ MOVQ (R11), R14 |
|
| 3191 |
+ MOVQ -8(R11)(AX*1), R15 |
|
| 3192 |
+ MOVQ R14, (R10) |
|
| 3193 |
+ MOVQ R15, -8(R10)(AX*1) |
|
| 3194 |
+ ADDQ AX, R11 |
|
| 3195 |
+ ADDQ AX, R10 |
|
| 3196 |
+ |
|
| 3197 |
+copy_1_end: |
|
| 3198 |
+ ADDQ AX, R12 |
|
| 3199 |
+ |
|
| 3200 |
+ // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) |
|
| 3201 |
+check_offset: |
|
| 3202 |
+ MOVQ R12, AX |
|
| 3203 |
+ ADDQ 40(SP), AX |
|
| 3204 |
+ CMPQ CX, AX |
|
| 3205 |
+ JG error_match_off_too_big |
|
| 3206 |
+ CMPQ CX, 56(SP) |
|
| 3207 |
+ JG error_match_off_too_big |
|
| 3208 |
+ |
|
| 3209 |
+ // Copy match from history |
|
| 3210 |
+ MOVQ CX, AX |
|
| 3211 |
+ SUBQ R12, AX |
|
| 3212 |
+ JLS copy_match |
|
| 3213 |
+ MOVQ 48(SP), R14 |
|
| 3214 |
+ SUBQ AX, R14 |
|
| 3215 |
+ CMPQ R13, AX |
|
| 3216 |
+ JG copy_all_from_history |
|
| 3217 |
+ MOVQ R13, AX |
|
| 3218 |
+ SUBQ $0x10, AX |
|
| 3219 |
+ JB copy_4_small |
|
| 3220 |
+ |
|
| 3221 |
+copy_4_loop: |
|
| 3222 |
+ MOVUPS (R14), X0 |
|
| 3223 |
+ MOVUPS X0, (R10) |
|
| 3224 |
+ ADDQ $0x10, R14 |
|
| 3225 |
+ ADDQ $0x10, R10 |
|
| 3226 |
+ SUBQ $0x10, AX |
|
| 3227 |
+ JAE copy_4_loop |
|
| 3228 |
+ LEAQ 16(R14)(AX*1), R14 |
|
| 3229 |
+ LEAQ 16(R10)(AX*1), R10 |
|
| 3230 |
+ MOVUPS -16(R14), X0 |
|
| 3231 |
+ MOVUPS X0, -16(R10) |
|
| 3232 |
+ JMP copy_4_end |
|
| 3233 |
+ |
|
| 3234 |
+copy_4_small: |
|
| 3235 |
+ CMPQ R13, $0x03 |
|
| 3236 |
+ JE copy_4_move_3 |
|
| 3237 |
+ CMPQ R13, $0x08 |
|
| 3238 |
+ JB copy_4_move_4through7 |
|
| 3239 |
+ JMP copy_4_move_8through16 |
|
| 3240 |
+ |
|
| 3241 |
+copy_4_move_3: |
|
| 3242 |
+ MOVW (R14), AX |
|
| 3243 |
+ MOVB 2(R14), CL |
|
| 3244 |
+ MOVW AX, (R10) |
|
| 3245 |
+ MOVB CL, 2(R10) |
|
| 3246 |
+ ADDQ R13, R14 |
|
| 3247 |
+ ADDQ R13, R10 |
|
| 3248 |
+ JMP copy_4_end |
|
| 3249 |
+ |
|
| 3250 |
+copy_4_move_4through7: |
|
| 3251 |
+ MOVL (R14), AX |
|
| 3252 |
+ MOVL -4(R14)(R13*1), CX |
|
| 3253 |
+ MOVL AX, (R10) |
|
| 3254 |
+ MOVL CX, -4(R10)(R13*1) |
|
| 3255 |
+ ADDQ R13, R14 |
|
| 3256 |
+ ADDQ R13, R10 |
|
| 3257 |
+ JMP copy_4_end |
|
| 3258 |
+ |
|
| 3259 |
+copy_4_move_8through16: |
|
| 3260 |
+ MOVQ (R14), AX |
|
| 3261 |
+ MOVQ -8(R14)(R13*1), CX |
|
| 3262 |
+ MOVQ AX, (R10) |
|
| 3263 |
+ MOVQ CX, -8(R10)(R13*1) |
|
| 3264 |
+ ADDQ R13, R14 |
|
| 3265 |
+ ADDQ R13, R10 |
|
| 3266 |
+ |
|
| 3267 |
+copy_4_end: |
|
| 3268 |
+ ADDQ R13, R12 |
|
| 3269 |
+ JMP handle_loop |
|
| 3270 |
+ JMP loop_finished |
|
| 3271 |
+ |
|
| 3272 |
+copy_all_from_history: |
|
| 3273 |
+ MOVQ AX, R15 |
|
| 3274 |
+ SUBQ $0x10, R15 |
|
| 3275 |
+ JB copy_5_small |
|
| 3276 |
+ |
|
| 3277 |
+copy_5_loop: |
|
| 3278 |
+ MOVUPS (R14), X0 |
|
| 3279 |
+ MOVUPS X0, (R10) |
|
| 3280 |
+ ADDQ $0x10, R14 |
|
| 3281 |
+ ADDQ $0x10, R10 |
|
| 3282 |
+ SUBQ $0x10, R15 |
|
| 3283 |
+ JAE copy_5_loop |
|
| 3284 |
+ LEAQ 16(R14)(R15*1), R14 |
|
| 3285 |
+ LEAQ 16(R10)(R15*1), R10 |
|
| 3286 |
+ MOVUPS -16(R14), X0 |
|
| 3287 |
+ MOVUPS X0, -16(R10) |
|
| 3288 |
+ JMP copy_5_end |
|
| 3289 |
+ |
|
| 3290 |
+copy_5_small: |
|
| 3291 |
+ CMPQ AX, $0x03 |
|
| 3292 |
+ JE copy_5_move_3 |
|
| 3293 |
+ JB copy_5_move_1or2 |
|
| 3294 |
+ CMPQ AX, $0x08 |
|
| 3295 |
+ JB copy_5_move_4through7 |
|
| 3296 |
+ JMP copy_5_move_8through16 |
|
| 3297 |
+ |
|
| 3298 |
+copy_5_move_1or2: |
|
| 3299 |
+ MOVB (R14), R15 |
|
| 3300 |
+ MOVB -1(R14)(AX*1), BP |
|
| 3301 |
+ MOVB R15, (R10) |
|
| 3302 |
+ MOVB BP, -1(R10)(AX*1) |
|
| 3303 |
+ ADDQ AX, R14 |
|
| 3304 |
+ ADDQ AX, R10 |
|
| 3305 |
+ JMP copy_5_end |
|
| 3306 |
+ |
|
| 3307 |
+copy_5_move_3: |
|
| 3308 |
+ MOVW (R14), R15 |
|
| 3309 |
+ MOVB 2(R14), BP |
|
| 3310 |
+ MOVW R15, (R10) |
|
| 3311 |
+ MOVB BP, 2(R10) |
|
| 3312 |
+ ADDQ AX, R14 |
|
| 3313 |
+ ADDQ AX, R10 |
|
| 3314 |
+ JMP copy_5_end |
|
| 3315 |
+ |
|
| 3316 |
+copy_5_move_4through7: |
|
| 3317 |
+ MOVL (R14), R15 |
|
| 3318 |
+ MOVL -4(R14)(AX*1), BP |
|
| 3319 |
+ MOVL R15, (R10) |
|
| 3320 |
+ MOVL BP, -4(R10)(AX*1) |
|
| 3321 |
+ ADDQ AX, R14 |
|
| 3322 |
+ ADDQ AX, R10 |
|
| 3323 |
+ JMP copy_5_end |
|
| 3324 |
+ |
|
| 3325 |
+copy_5_move_8through16: |
|
| 3326 |
+ MOVQ (R14), R15 |
|
| 3327 |
+ MOVQ -8(R14)(AX*1), BP |
|
| 3328 |
+ MOVQ R15, (R10) |
|
| 3329 |
+ MOVQ BP, -8(R10)(AX*1) |
|
| 3330 |
+ ADDQ AX, R14 |
|
| 3331 |
+ ADDQ AX, R10 |
|
| 3332 |
+ |
|
| 3333 |
+copy_5_end: |
|
| 3334 |
+ ADDQ AX, R12 |
|
| 3335 |
+ SUBQ AX, R13 |
|
| 3336 |
+ |
|
| 3337 |
+ // Copy match from the current buffer |
|
| 3338 |
+copy_match: |
|
| 3339 |
+ MOVQ R10, AX |
|
| 3340 |
+ SUBQ CX, AX |
|
| 3341 |
+ |
|
| 3342 |
+ // ml <= mo |
|
| 3343 |
+ CMPQ R13, CX |
|
| 3344 |
+ JA copy_overlapping_match |
|
| 3345 |
+ |
|
| 3346 |
+ // Copy non-overlapping match |
|
| 3347 |
+ ADDQ R13, R12 |
|
| 3348 |
+ MOVQ R13, CX |
|
| 3349 |
+ SUBQ $0x10, CX |
|
| 3350 |
+ JB copy_2_small |
|
| 3351 |
+ |
|
| 3352 |
+copy_2_loop: |
|
| 3353 |
+ MOVUPS (AX), X0 |
|
| 3354 |
+ MOVUPS X0, (R10) |
|
| 3355 |
+ ADDQ $0x10, AX |
|
| 3356 |
+ ADDQ $0x10, R10 |
|
| 3357 |
+ SUBQ $0x10, CX |
|
| 3358 |
+ JAE copy_2_loop |
|
| 3359 |
+ LEAQ 16(AX)(CX*1), AX |
|
| 3360 |
+ LEAQ 16(R10)(CX*1), R10 |
|
| 3361 |
+ MOVUPS -16(AX), X0 |
|
| 3362 |
+ MOVUPS X0, -16(R10) |
|
| 3363 |
+ JMP copy_2_end |
|
| 3364 |
+ |
|
| 3365 |
+copy_2_small: |
|
| 3366 |
+ CMPQ R13, $0x03 |
|
| 3367 |
+ JE copy_2_move_3 |
|
| 3368 |
+ JB copy_2_move_1or2 |
|
| 3369 |
+ CMPQ R13, $0x08 |
|
| 3370 |
+ JB copy_2_move_4through7 |
|
| 3371 |
+ JMP copy_2_move_8through16 |
|
| 3372 |
+ |
|
| 3373 |
+copy_2_move_1or2: |
|
| 3374 |
+ MOVB (AX), CL |
|
| 3375 |
+ MOVB -1(AX)(R13*1), R14 |
|
| 3376 |
+ MOVB CL, (R10) |
|
| 3377 |
+ MOVB R14, -1(R10)(R13*1) |
|
| 3378 |
+ ADDQ R13, AX |
|
| 3379 |
+ ADDQ R13, R10 |
|
| 3380 |
+ JMP copy_2_end |
|
| 3381 |
+ |
|
| 3382 |
+copy_2_move_3: |
|
| 3383 |
+ MOVW (AX), CX |
|
| 3384 |
+ MOVB 2(AX), R14 |
|
| 3385 |
+ MOVW CX, (R10) |
|
| 3386 |
+ MOVB R14, 2(R10) |
|
| 3387 |
+ ADDQ R13, AX |
|
| 3388 |
+ ADDQ R13, R10 |
|
| 3389 |
+ JMP copy_2_end |
|
| 3390 |
+ |
|
| 3391 |
+copy_2_move_4through7: |
|
| 3392 |
+ MOVL (AX), CX |
|
| 3393 |
+ MOVL -4(AX)(R13*1), R14 |
|
| 3394 |
+ MOVL CX, (R10) |
|
| 3395 |
+ MOVL R14, -4(R10)(R13*1) |
|
| 3396 |
+ ADDQ R13, AX |
|
| 3397 |
+ ADDQ R13, R10 |
|
| 3398 |
+ JMP copy_2_end |
|
| 3399 |
+ |
|
| 3400 |
+copy_2_move_8through16: |
|
| 3401 |
+ MOVQ (AX), CX |
|
| 3402 |
+ MOVQ -8(AX)(R13*1), R14 |
|
| 3403 |
+ MOVQ CX, (R10) |
|
| 3404 |
+ MOVQ R14, -8(R10)(R13*1) |
|
| 3405 |
+ ADDQ R13, AX |
|
| 3406 |
+ ADDQ R13, R10 |
|
| 3407 |
+ |
|
| 3408 |
+copy_2_end: |
|
| 3409 |
+ JMP handle_loop |
|
| 3410 |
+ |
|
| 3411 |
+ // Copy overlapping match |
|
| 3412 |
+copy_overlapping_match: |
|
| 3413 |
+ ADDQ R13, R12 |
|
| 3414 |
+ |
|
| 3415 |
+copy_slow_3: |
|
| 3416 |
+ MOVB (AX), CL |
|
| 3417 |
+ MOVB CL, (R10) |
|
| 3418 |
+ INCQ AX |
|
| 3419 |
+ INCQ R10 |
|
| 3420 |
+ DECQ R13 |
|
| 3421 |
+ JNZ copy_slow_3 |
|
| 3422 |
+ |
|
| 3423 |
+handle_loop: |
|
| 3424 |
+ MOVQ ctx+16(FP), AX |
|
| 3425 |
+ DECQ 96(AX) |
|
| 3426 |
+ JNS sequenceDecs_decodeSync_safe_amd64_main_loop |
|
| 3427 |
+ |
|
| 3428 |
+loop_finished: |
|
| 3429 |
+ MOVQ br+8(FP), AX |
|
| 3430 |
+ MOVQ DX, 32(AX) |
|
| 3431 |
+ MOVB BL, 40(AX) |
|
| 3432 |
+ MOVQ SI, 24(AX) |
|
| 3433 |
+ |
|
| 3434 |
+ // Update the context |
|
| 3435 |
+ MOVQ ctx+16(FP), AX |
|
| 3436 |
+ MOVQ R12, 136(AX) |
|
| 3437 |
+ MOVQ 144(AX), CX |
|
| 3438 |
+ SUBQ CX, R11 |
|
| 3439 |
+ MOVQ R11, 168(AX) |
|
| 3440 |
+ |
|
| 3441 |
+ // Return success |
|
| 3442 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 3443 |
+ RET |
|
| 3444 |
+ |
|
| 3445 |
+ // Return with match length error |
|
| 3446 |
+sequenceDecs_decodeSync_safe_amd64_error_match_len_ofs_mismatch: |
|
| 3447 |
+ MOVQ 16(SP), AX |
|
| 3448 |
+ MOVQ ctx+16(FP), CX |
|
| 3449 |
+ MOVQ AX, 216(CX) |
|
| 3450 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 3451 |
+ RET |
|
| 3452 |
+ |
|
| 3453 |
+ // Return with match too long error |
|
| 3454 |
+sequenceDecs_decodeSync_safe_amd64_error_match_len_too_big: |
|
| 3455 |
+ MOVQ ctx+16(FP), AX |
|
| 3456 |
+ MOVQ 16(SP), CX |
|
| 3457 |
+ MOVQ CX, 216(AX) |
|
| 3458 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 3459 |
+ RET |
|
| 3460 |
+ |
|
| 3461 |
+ // Return with match offset too long error |
|
| 3462 |
+error_match_off_too_big: |
|
| 3463 |
+ MOVQ ctx+16(FP), AX |
|
| 3464 |
+ MOVQ 8(SP), CX |
|
| 3465 |
+ MOVQ CX, 224(AX) |
|
| 3466 |
+ MOVQ R12, 136(AX) |
|
| 3467 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 3468 |
+ RET |
|
| 3469 |
+ |
|
| 3470 |
+ // Return with not enough literals error |
|
| 3471 |
+error_not_enough_literals: |
|
| 3472 |
+ MOVQ ctx+16(FP), AX |
|
| 3473 |
+ MOVQ 24(SP), CX |
|
| 3474 |
+ MOVQ CX, 208(AX) |
|
| 3475 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 3476 |
+ RET |
|
| 3477 |
+ |
|
| 3478 |
+ // Return with not enough output space error |
|
| 3479 |
+error_not_enough_space: |
|
| 3480 |
+ MOVQ ctx+16(FP), AX |
|
| 3481 |
+ MOVQ 24(SP), CX |
|
| 3482 |
+ MOVQ CX, 208(AX) |
|
| 3483 |
+ MOVQ 16(SP), CX |
|
| 3484 |
+ MOVQ CX, 216(AX) |
|
| 3485 |
+ MOVQ R12, 136(AX) |
|
| 3486 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 3487 |
+ RET |
|
| 3488 |
+ |
|
| 3489 |
+// func sequenceDecs_decodeSync_safe_bmi2(s *sequenceDecs, br *bitReader, ctx *decodeSyncAsmContext) int |
|
| 3490 |
+// Requires: BMI, BMI2, CMOV, SSE |
|
| 3491 |
+TEXT ·sequenceDecs_decodeSync_safe_bmi2(SB), $64-32 |
|
| 3492 |
+ MOVQ br+8(FP), CX |
|
| 3493 |
+ MOVQ 32(CX), AX |
|
| 3494 |
+ MOVBQZX 40(CX), DX |
|
| 3495 |
+ MOVQ 24(CX), BX |
|
| 3496 |
+ MOVQ (CX), CX |
|
| 3497 |
+ ADDQ BX, CX |
|
| 3498 |
+ MOVQ CX, (SP) |
|
| 3499 |
+ MOVQ ctx+16(FP), CX |
|
| 3500 |
+ MOVQ 72(CX), SI |
|
| 3501 |
+ MOVQ 80(CX), DI |
|
| 3502 |
+ MOVQ 88(CX), R8 |
|
| 3503 |
+ XORQ R9, R9 |
|
| 3504 |
+ MOVQ R9, 8(SP) |
|
| 3505 |
+ MOVQ R9, 16(SP) |
|
| 3506 |
+ MOVQ R9, 24(SP) |
|
| 3507 |
+ MOVQ 112(CX), R9 |
|
| 3508 |
+ MOVQ 128(CX), R10 |
|
| 3509 |
+ MOVQ R10, 32(SP) |
|
| 3510 |
+ MOVQ 144(CX), R10 |
|
| 3511 |
+ MOVQ 136(CX), R11 |
|
| 3512 |
+ MOVQ 200(CX), R12 |
|
| 3513 |
+ MOVQ R12, 56(SP) |
|
| 3514 |
+ MOVQ 176(CX), R12 |
|
| 3515 |
+ MOVQ R12, 48(SP) |
|
| 3516 |
+ MOVQ 184(CX), CX |
|
| 3517 |
+ MOVQ CX, 40(SP) |
|
| 3518 |
+ MOVQ 40(SP), CX |
|
| 3519 |
+ ADDQ CX, 48(SP) |
|
| 3520 |
+ |
|
| 3521 |
+ // Calculate poiter to s.out[cap(s.out)] (a past-end pointer) |
|
| 3522 |
+ ADDQ R9, 32(SP) |
|
| 3523 |
+ |
|
| 3524 |
+ // outBase += outPosition |
|
| 3525 |
+ ADDQ R11, R9 |
|
| 3526 |
+ |
|
| 3527 |
+sequenceDecs_decodeSync_safe_bmi2_main_loop: |
|
| 3528 |
+ MOVQ (SP), R12 |
|
| 3529 |
+ |
|
| 3530 |
+ // Fill bitreader to have enough for the offset and match length. |
|
| 3531 |
+ CMPQ BX, $0x08 |
|
| 3532 |
+ JL sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte |
|
| 3533 |
+ MOVQ DX, CX |
|
| 3534 |
+ SHRQ $0x03, CX |
|
| 3535 |
+ SUBQ CX, R12 |
|
| 3536 |
+ MOVQ (R12), AX |
|
| 3537 |
+ SUBQ CX, BX |
|
| 3538 |
+ ANDQ $0x07, DX |
|
| 3539 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_fill_end |
|
| 3540 |
+ |
|
| 3541 |
+sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte: |
|
| 3542 |
+ CMPQ BX, $0x00 |
|
| 3543 |
+ JLE sequenceDecs_decodeSync_safe_bmi2_fill_end |
|
| 3544 |
+ CMPQ DX, $0x07 |
|
| 3545 |
+ JLE sequenceDecs_decodeSync_safe_bmi2_fill_end |
|
| 3546 |
+ SHLQ $0x08, AX |
|
| 3547 |
+ SUBQ $0x01, R12 |
|
| 3548 |
+ SUBQ $0x01, BX |
|
| 3549 |
+ SUBQ $0x08, DX |
|
| 3550 |
+ MOVBQZX (R12), CX |
|
| 3551 |
+ ORQ CX, AX |
|
| 3552 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_fill_byte_by_byte |
|
| 3553 |
+ |
|
| 3554 |
+sequenceDecs_decodeSync_safe_bmi2_fill_end: |
|
| 3555 |
+ // Update offset |
|
| 3556 |
+ MOVQ $0x00000808, CX |
|
| 3557 |
+ BEXTRQ CX, R8, R13 |
|
| 3558 |
+ MOVQ AX, R14 |
|
| 3559 |
+ LEAQ (DX)(R13*1), CX |
|
| 3560 |
+ ROLQ CL, R14 |
|
| 3561 |
+ BZHIQ R13, R14, R14 |
|
| 3562 |
+ MOVQ CX, DX |
|
| 3563 |
+ MOVQ R8, CX |
|
| 3564 |
+ SHRQ $0x20, CX |
|
| 3565 |
+ ADDQ R14, CX |
|
| 3566 |
+ MOVQ CX, 8(SP) |
|
| 3567 |
+ |
|
| 3568 |
+ // Update match length |
|
| 3569 |
+ MOVQ $0x00000808, CX |
|
| 3570 |
+ BEXTRQ CX, DI, R13 |
|
| 3571 |
+ MOVQ AX, R14 |
|
| 3572 |
+ LEAQ (DX)(R13*1), CX |
|
| 3573 |
+ ROLQ CL, R14 |
|
| 3574 |
+ BZHIQ R13, R14, R14 |
|
| 3575 |
+ MOVQ CX, DX |
|
| 3576 |
+ MOVQ DI, CX |
|
| 3577 |
+ SHRQ $0x20, CX |
|
| 3578 |
+ ADDQ R14, CX |
|
| 3579 |
+ MOVQ CX, 16(SP) |
|
| 3580 |
+ |
|
| 3581 |
+ // Fill bitreader to have enough for the remaining |
|
| 3582 |
+ CMPQ BX, $0x08 |
|
| 3583 |
+ JL sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte |
|
| 3584 |
+ MOVQ DX, CX |
|
| 3585 |
+ SHRQ $0x03, CX |
|
| 3586 |
+ SUBQ CX, R12 |
|
| 3587 |
+ MOVQ (R12), AX |
|
| 3588 |
+ SUBQ CX, BX |
|
| 3589 |
+ ANDQ $0x07, DX |
|
| 3590 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_fill_2_end |
|
| 3591 |
+ |
|
| 3592 |
+sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte: |
|
| 3593 |
+ CMPQ BX, $0x00 |
|
| 3594 |
+ JLE sequenceDecs_decodeSync_safe_bmi2_fill_2_end |
|
| 3595 |
+ CMPQ DX, $0x07 |
|
| 3596 |
+ JLE sequenceDecs_decodeSync_safe_bmi2_fill_2_end |
|
| 3597 |
+ SHLQ $0x08, AX |
|
| 3598 |
+ SUBQ $0x01, R12 |
|
| 3599 |
+ SUBQ $0x01, BX |
|
| 3600 |
+ SUBQ $0x08, DX |
|
| 3601 |
+ MOVBQZX (R12), CX |
|
| 3602 |
+ ORQ CX, AX |
|
| 3603 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_fill_2_byte_by_byte |
|
| 3604 |
+ |
|
| 3605 |
+sequenceDecs_decodeSync_safe_bmi2_fill_2_end: |
|
| 3606 |
+ // Update literal length |
|
| 3607 |
+ MOVQ $0x00000808, CX |
|
| 3608 |
+ BEXTRQ CX, SI, R13 |
|
| 3609 |
+ MOVQ AX, R14 |
|
| 3610 |
+ LEAQ (DX)(R13*1), CX |
|
| 3611 |
+ ROLQ CL, R14 |
|
| 3612 |
+ BZHIQ R13, R14, R14 |
|
| 3613 |
+ MOVQ CX, DX |
|
| 3614 |
+ MOVQ SI, CX |
|
| 3615 |
+ SHRQ $0x20, CX |
|
| 3616 |
+ ADDQ R14, CX |
|
| 3617 |
+ MOVQ CX, 24(SP) |
|
| 3618 |
+ |
|
| 3619 |
+ // Fill bitreader for state updates |
|
| 3620 |
+ MOVQ R12, (SP) |
|
| 3621 |
+ MOVQ $0x00000808, CX |
|
| 3622 |
+ BEXTRQ CX, R8, R12 |
|
| 3623 |
+ MOVQ ctx+16(FP), CX |
|
| 3624 |
+ CMPQ 96(CX), $0x00 |
|
| 3625 |
+ JZ sequenceDecs_decodeSync_safe_bmi2_skip_update |
|
| 3626 |
+ LEAQ (SI)(DI*1), R13 |
|
| 3627 |
+ ADDQ R8, R13 |
|
| 3628 |
+ MOVBQZX R13, R13 |
|
| 3629 |
+ LEAQ (DX)(R13*1), CX |
|
| 3630 |
+ MOVQ AX, R14 |
|
| 3631 |
+ MOVQ CX, DX |
|
| 3632 |
+ ROLQ CL, R14 |
|
| 3633 |
+ BZHIQ R13, R14, R14 |
|
| 3634 |
+ |
|
| 3635 |
+ // Update Offset State |
|
| 3636 |
+ BZHIQ R8, R14, CX |
|
| 3637 |
+ SHRXQ R8, R14, R14 |
|
| 3638 |
+ MOVQ $0x00001010, R13 |
|
| 3639 |
+ BEXTRQ R13, R8, R8 |
|
| 3640 |
+ ADDQ CX, R8 |
|
| 3641 |
+ |
|
| 3642 |
+ // Load ctx.ofTable |
|
| 3643 |
+ MOVQ ctx+16(FP), CX |
|
| 3644 |
+ MOVQ 48(CX), CX |
|
| 3645 |
+ MOVQ (CX)(R8*8), R8 |
|
| 3646 |
+ |
|
| 3647 |
+ // Update Match Length State |
|
| 3648 |
+ BZHIQ DI, R14, CX |
|
| 3649 |
+ SHRXQ DI, R14, R14 |
|
| 3650 |
+ MOVQ $0x00001010, R13 |
|
| 3651 |
+ BEXTRQ R13, DI, DI |
|
| 3652 |
+ ADDQ CX, DI |
|
| 3653 |
+ |
|
| 3654 |
+ // Load ctx.mlTable |
|
| 3655 |
+ MOVQ ctx+16(FP), CX |
|
| 3656 |
+ MOVQ 24(CX), CX |
|
| 3657 |
+ MOVQ (CX)(DI*8), DI |
|
| 3658 |
+ |
|
| 3659 |
+ // Update Literal Length State |
|
| 3660 |
+ BZHIQ SI, R14, CX |
|
| 3661 |
+ MOVQ $0x00001010, R13 |
|
| 3662 |
+ BEXTRQ R13, SI, SI |
|
| 3663 |
+ ADDQ CX, SI |
|
| 3664 |
+ |
|
| 3665 |
+ // Load ctx.llTable |
|
| 3666 |
+ MOVQ ctx+16(FP), CX |
|
| 3667 |
+ MOVQ (CX), CX |
|
| 3668 |
+ MOVQ (CX)(SI*8), SI |
|
| 3669 |
+ |
|
| 3670 |
+sequenceDecs_decodeSync_safe_bmi2_skip_update: |
|
| 3671 |
+ // Adjust offset |
|
| 3672 |
+ MOVQ s+0(FP), CX |
|
| 3673 |
+ MOVQ 8(SP), R13 |
|
| 3674 |
+ CMPQ R12, $0x01 |
|
| 3675 |
+ JBE sequenceDecs_decodeSync_safe_bmi2_adjust_offsetB_1_or_0 |
|
| 3676 |
+ MOVUPS 144(CX), X0 |
|
| 3677 |
+ MOVQ R13, 144(CX) |
|
| 3678 |
+ MOVUPS X0, 152(CX) |
|
| 3679 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_after_adjust |
|
| 3680 |
+ |
|
| 3681 |
+sequenceDecs_decodeSync_safe_bmi2_adjust_offsetB_1_or_0: |
|
| 3682 |
+ CMPQ 24(SP), $0x00000000 |
|
| 3683 |
+ JNE sequenceDecs_decodeSync_safe_bmi2_adjust_offset_maybezero |
|
| 3684 |
+ INCQ R13 |
|
| 3685 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero |
|
| 3686 |
+ |
|
| 3687 |
+sequenceDecs_decodeSync_safe_bmi2_adjust_offset_maybezero: |
|
| 3688 |
+ TESTQ R13, R13 |
|
| 3689 |
+ JNZ sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero |
|
| 3690 |
+ MOVQ 144(CX), R13 |
|
| 3691 |
+ JMP sequenceDecs_decodeSync_safe_bmi2_after_adjust |
|
| 3692 |
+ |
|
| 3693 |
+sequenceDecs_decodeSync_safe_bmi2_adjust_offset_nonzero: |
|
| 3694 |
+ MOVQ R13, R12 |
|
| 3695 |
+ XORQ R14, R14 |
|
| 3696 |
+ MOVQ $-1, R15 |
|
| 3697 |
+ CMPQ R13, $0x03 |
|
| 3698 |
+ CMOVQEQ R14, R12 |
|
| 3699 |
+ CMOVQEQ R15, R14 |
|
| 3700 |
+ ADDQ 144(CX)(R12*8), R14 |
|
| 3701 |
+ JNZ sequenceDecs_decodeSync_safe_bmi2_adjust_temp_valid |
|
| 3702 |
+ MOVQ $0x00000001, R14 |
|
| 3703 |
+ |
|
| 3704 |
+sequenceDecs_decodeSync_safe_bmi2_adjust_temp_valid: |
|
| 3705 |
+ CMPQ R13, $0x01 |
|
| 3706 |
+ JZ sequenceDecs_decodeSync_safe_bmi2_adjust_skip |
|
| 3707 |
+ MOVQ 152(CX), R12 |
|
| 3708 |
+ MOVQ R12, 160(CX) |
|
| 3709 |
+ |
|
| 3710 |
+sequenceDecs_decodeSync_safe_bmi2_adjust_skip: |
|
| 3711 |
+ MOVQ 144(CX), R12 |
|
| 3712 |
+ MOVQ R12, 152(CX) |
|
| 3713 |
+ MOVQ R14, 144(CX) |
|
| 3714 |
+ MOVQ R14, R13 |
|
| 3715 |
+ |
|
| 3716 |
+sequenceDecs_decodeSync_safe_bmi2_after_adjust: |
|
| 3717 |
+ MOVQ R13, 8(SP) |
|
| 3718 |
+ |
|
| 3719 |
+ // Check values |
|
| 3720 |
+ MOVQ 16(SP), CX |
|
| 3721 |
+ MOVQ 24(SP), R12 |
|
| 3722 |
+ LEAQ (CX)(R12*1), R14 |
|
| 3723 |
+ MOVQ s+0(FP), R15 |
|
| 3724 |
+ ADDQ R14, 256(R15) |
|
| 3725 |
+ MOVQ ctx+16(FP), R14 |
|
| 3726 |
+ SUBQ R12, 104(R14) |
|
| 3727 |
+ JS error_not_enough_literals |
|
| 3728 |
+ CMPQ CX, $0x00020002 |
|
| 3729 |
+ JA sequenceDecs_decodeSync_safe_bmi2_error_match_len_too_big |
|
| 3730 |
+ TESTQ R13, R13 |
|
| 3731 |
+ JNZ sequenceDecs_decodeSync_safe_bmi2_match_len_ofs_ok |
|
| 3732 |
+ TESTQ CX, CX |
|
| 3733 |
+ JNZ sequenceDecs_decodeSync_safe_bmi2_error_match_len_ofs_mismatch |
|
| 3734 |
+ |
|
| 3735 |
+sequenceDecs_decodeSync_safe_bmi2_match_len_ofs_ok: |
|
| 3736 |
+ MOVQ 24(SP), CX |
|
| 3737 |
+ MOVQ 8(SP), R12 |
|
| 3738 |
+ MOVQ 16(SP), R13 |
|
| 3739 |
+ |
|
| 3740 |
+ // Check if we have enough space in s.out |
|
| 3741 |
+ LEAQ (CX)(R13*1), R14 |
|
| 3742 |
+ ADDQ R9, R14 |
|
| 3743 |
+ CMPQ R14, 32(SP) |
|
| 3744 |
+ JA error_not_enough_space |
|
| 3745 |
+ |
|
| 3746 |
+ // Copy literals |
|
| 3747 |
+ TESTQ CX, CX |
|
| 3748 |
+ JZ check_offset |
|
| 3749 |
+ MOVQ CX, R14 |
|
| 3750 |
+ SUBQ $0x10, R14 |
|
| 3751 |
+ JB copy_1_small |
|
| 3752 |
+ |
|
| 3753 |
+copy_1_loop: |
|
| 3754 |
+ MOVUPS (R10), X0 |
|
| 3755 |
+ MOVUPS X0, (R9) |
|
| 3756 |
+ ADDQ $0x10, R10 |
|
| 3757 |
+ ADDQ $0x10, R9 |
|
| 3758 |
+ SUBQ $0x10, R14 |
|
| 3759 |
+ JAE copy_1_loop |
|
| 3760 |
+ LEAQ 16(R10)(R14*1), R10 |
|
| 3761 |
+ LEAQ 16(R9)(R14*1), R9 |
|
| 3762 |
+ MOVUPS -16(R10), X0 |
|
| 3763 |
+ MOVUPS X0, -16(R9) |
|
| 3764 |
+ JMP copy_1_end |
|
| 3765 |
+ |
|
| 3766 |
+copy_1_small: |
|
| 3767 |
+ CMPQ CX, $0x03 |
|
| 3768 |
+ JE copy_1_move_3 |
|
| 3769 |
+ JB copy_1_move_1or2 |
|
| 3770 |
+ CMPQ CX, $0x08 |
|
| 3771 |
+ JB copy_1_move_4through7 |
|
| 3772 |
+ JMP copy_1_move_8through16 |
|
| 3773 |
+ |
|
| 3774 |
+copy_1_move_1or2: |
|
| 3775 |
+ MOVB (R10), R14 |
|
| 3776 |
+ MOVB -1(R10)(CX*1), R15 |
|
| 3777 |
+ MOVB R14, (R9) |
|
| 3778 |
+ MOVB R15, -1(R9)(CX*1) |
|
| 3779 |
+ ADDQ CX, R10 |
|
| 3780 |
+ ADDQ CX, R9 |
|
| 3781 |
+ JMP copy_1_end |
|
| 3782 |
+ |
|
| 3783 |
+copy_1_move_3: |
|
| 3784 |
+ MOVW (R10), R14 |
|
| 3785 |
+ MOVB 2(R10), R15 |
|
| 3786 |
+ MOVW R14, (R9) |
|
| 3787 |
+ MOVB R15, 2(R9) |
|
| 3788 |
+ ADDQ CX, R10 |
|
| 3789 |
+ ADDQ CX, R9 |
|
| 3790 |
+ JMP copy_1_end |
|
| 3791 |
+ |
|
| 3792 |
+copy_1_move_4through7: |
|
| 3793 |
+ MOVL (R10), R14 |
|
| 3794 |
+ MOVL -4(R10)(CX*1), R15 |
|
| 3795 |
+ MOVL R14, (R9) |
|
| 3796 |
+ MOVL R15, -4(R9)(CX*1) |
|
| 3797 |
+ ADDQ CX, R10 |
|
| 3798 |
+ ADDQ CX, R9 |
|
| 3799 |
+ JMP copy_1_end |
|
| 3800 |
+ |
|
| 3801 |
+copy_1_move_8through16: |
|
| 3802 |
+ MOVQ (R10), R14 |
|
| 3803 |
+ MOVQ -8(R10)(CX*1), R15 |
|
| 3804 |
+ MOVQ R14, (R9) |
|
| 3805 |
+ MOVQ R15, -8(R9)(CX*1) |
|
| 3806 |
+ ADDQ CX, R10 |
|
| 3807 |
+ ADDQ CX, R9 |
|
| 3808 |
+ |
|
| 3809 |
+copy_1_end: |
|
| 3810 |
+ ADDQ CX, R11 |
|
| 3811 |
+ |
|
| 3812 |
+ // Malformed input if seq.mo > t+len(hist) || seq.mo > s.windowSize) |
|
| 3813 |
+check_offset: |
|
| 3814 |
+ MOVQ R11, CX |
|
| 3815 |
+ ADDQ 40(SP), CX |
|
| 3816 |
+ CMPQ R12, CX |
|
| 3817 |
+ JG error_match_off_too_big |
|
| 3818 |
+ CMPQ R12, 56(SP) |
|
| 3819 |
+ JG error_match_off_too_big |
|
| 3820 |
+ |
|
| 3821 |
+ // Copy match from history |
|
| 3822 |
+ MOVQ R12, CX |
|
| 3823 |
+ SUBQ R11, CX |
|
| 3824 |
+ JLS copy_match |
|
| 3825 |
+ MOVQ 48(SP), R14 |
|
| 3826 |
+ SUBQ CX, R14 |
|
| 3827 |
+ CMPQ R13, CX |
|
| 3828 |
+ JG copy_all_from_history |
|
| 3829 |
+ MOVQ R13, CX |
|
| 3830 |
+ SUBQ $0x10, CX |
|
| 3831 |
+ JB copy_4_small |
|
| 3832 |
+ |
|
| 3833 |
+copy_4_loop: |
|
| 3834 |
+ MOVUPS (R14), X0 |
|
| 3835 |
+ MOVUPS X0, (R9) |
|
| 3836 |
+ ADDQ $0x10, R14 |
|
| 3837 |
+ ADDQ $0x10, R9 |
|
| 3838 |
+ SUBQ $0x10, CX |
|
| 3839 |
+ JAE copy_4_loop |
|
| 3840 |
+ LEAQ 16(R14)(CX*1), R14 |
|
| 3841 |
+ LEAQ 16(R9)(CX*1), R9 |
|
| 3842 |
+ MOVUPS -16(R14), X0 |
|
| 3843 |
+ MOVUPS X0, -16(R9) |
|
| 3844 |
+ JMP copy_4_end |
|
| 3845 |
+ |
|
| 3846 |
+copy_4_small: |
|
| 3847 |
+ CMPQ R13, $0x03 |
|
| 3848 |
+ JE copy_4_move_3 |
|
| 3849 |
+ CMPQ R13, $0x08 |
|
| 3850 |
+ JB copy_4_move_4through7 |
|
| 3851 |
+ JMP copy_4_move_8through16 |
|
| 3852 |
+ |
|
| 3853 |
+copy_4_move_3: |
|
| 3854 |
+ MOVW (R14), CX |
|
| 3855 |
+ MOVB 2(R14), R12 |
|
| 3856 |
+ MOVW CX, (R9) |
|
| 3857 |
+ MOVB R12, 2(R9) |
|
| 3858 |
+ ADDQ R13, R14 |
|
| 3859 |
+ ADDQ R13, R9 |
|
| 3860 |
+ JMP copy_4_end |
|
| 3861 |
+ |
|
| 3862 |
+copy_4_move_4through7: |
|
| 3863 |
+ MOVL (R14), CX |
|
| 3864 |
+ MOVL -4(R14)(R13*1), R12 |
|
| 3865 |
+ MOVL CX, (R9) |
|
| 3866 |
+ MOVL R12, -4(R9)(R13*1) |
|
| 3867 |
+ ADDQ R13, R14 |
|
| 3868 |
+ ADDQ R13, R9 |
|
| 3869 |
+ JMP copy_4_end |
|
| 3870 |
+ |
|
| 3871 |
+copy_4_move_8through16: |
|
| 3872 |
+ MOVQ (R14), CX |
|
| 3873 |
+ MOVQ -8(R14)(R13*1), R12 |
|
| 3874 |
+ MOVQ CX, (R9) |
|
| 3875 |
+ MOVQ R12, -8(R9)(R13*1) |
|
| 3876 |
+ ADDQ R13, R14 |
|
| 3877 |
+ ADDQ R13, R9 |
|
| 3878 |
+ |
|
| 3879 |
+copy_4_end: |
|
| 3880 |
+ ADDQ R13, R11 |
|
| 3881 |
+ JMP handle_loop |
|
| 3882 |
+ JMP loop_finished |
|
| 3883 |
+ |
|
| 3884 |
+copy_all_from_history: |
|
| 3885 |
+ MOVQ CX, R15 |
|
| 3886 |
+ SUBQ $0x10, R15 |
|
| 3887 |
+ JB copy_5_small |
|
| 3888 |
+ |
|
| 3889 |
+copy_5_loop: |
|
| 3890 |
+ MOVUPS (R14), X0 |
|
| 3891 |
+ MOVUPS X0, (R9) |
|
| 3892 |
+ ADDQ $0x10, R14 |
|
| 3893 |
+ ADDQ $0x10, R9 |
|
| 3894 |
+ SUBQ $0x10, R15 |
|
| 3895 |
+ JAE copy_5_loop |
|
| 3896 |
+ LEAQ 16(R14)(R15*1), R14 |
|
| 3897 |
+ LEAQ 16(R9)(R15*1), R9 |
|
| 3898 |
+ MOVUPS -16(R14), X0 |
|
| 3899 |
+ MOVUPS X0, -16(R9) |
|
| 3900 |
+ JMP copy_5_end |
|
| 3901 |
+ |
|
| 3902 |
+copy_5_small: |
|
| 3903 |
+ CMPQ CX, $0x03 |
|
| 3904 |
+ JE copy_5_move_3 |
|
| 3905 |
+ JB copy_5_move_1or2 |
|
| 3906 |
+ CMPQ CX, $0x08 |
|
| 3907 |
+ JB copy_5_move_4through7 |
|
| 3908 |
+ JMP copy_5_move_8through16 |
|
| 3909 |
+ |
|
| 3910 |
+copy_5_move_1or2: |
|
| 3911 |
+ MOVB (R14), R15 |
|
| 3912 |
+ MOVB -1(R14)(CX*1), BP |
|
| 3913 |
+ MOVB R15, (R9) |
|
| 3914 |
+ MOVB BP, -1(R9)(CX*1) |
|
| 3915 |
+ ADDQ CX, R14 |
|
| 3916 |
+ ADDQ CX, R9 |
|
| 3917 |
+ JMP copy_5_end |
|
| 3918 |
+ |
|
| 3919 |
+copy_5_move_3: |
|
| 3920 |
+ MOVW (R14), R15 |
|
| 3921 |
+ MOVB 2(R14), BP |
|
| 3922 |
+ MOVW R15, (R9) |
|
| 3923 |
+ MOVB BP, 2(R9) |
|
| 3924 |
+ ADDQ CX, R14 |
|
| 3925 |
+ ADDQ CX, R9 |
|
| 3926 |
+ JMP copy_5_end |
|
| 3927 |
+ |
|
| 3928 |
+copy_5_move_4through7: |
|
| 3929 |
+ MOVL (R14), R15 |
|
| 3930 |
+ MOVL -4(R14)(CX*1), BP |
|
| 3931 |
+ MOVL R15, (R9) |
|
| 3932 |
+ MOVL BP, -4(R9)(CX*1) |
|
| 3933 |
+ ADDQ CX, R14 |
|
| 3934 |
+ ADDQ CX, R9 |
|
| 3935 |
+ JMP copy_5_end |
|
| 3936 |
+ |
|
| 3937 |
+copy_5_move_8through16: |
|
| 3938 |
+ MOVQ (R14), R15 |
|
| 3939 |
+ MOVQ -8(R14)(CX*1), BP |
|
| 3940 |
+ MOVQ R15, (R9) |
|
| 3941 |
+ MOVQ BP, -8(R9)(CX*1) |
|
| 3942 |
+ ADDQ CX, R14 |
|
| 3943 |
+ ADDQ CX, R9 |
|
| 3944 |
+ |
|
| 3945 |
+copy_5_end: |
|
| 3946 |
+ ADDQ CX, R11 |
|
| 3947 |
+ SUBQ CX, R13 |
|
| 3948 |
+ |
|
| 3949 |
+ // Copy match from the current buffer |
|
| 3950 |
+copy_match: |
|
| 3951 |
+ MOVQ R9, CX |
|
| 3952 |
+ SUBQ R12, CX |
|
| 3953 |
+ |
|
| 3954 |
+ // ml <= mo |
|
| 3955 |
+ CMPQ R13, R12 |
|
| 3956 |
+ JA copy_overlapping_match |
|
| 3957 |
+ |
|
| 3958 |
+ // Copy non-overlapping match |
|
| 3959 |
+ ADDQ R13, R11 |
|
| 3960 |
+ MOVQ R13, R12 |
|
| 3961 |
+ SUBQ $0x10, R12 |
|
| 3962 |
+ JB copy_2_small |
|
| 3963 |
+ |
|
| 3964 |
+copy_2_loop: |
|
| 3965 |
+ MOVUPS (CX), X0 |
|
| 3966 |
+ MOVUPS X0, (R9) |
|
| 3967 |
+ ADDQ $0x10, CX |
|
| 3968 |
+ ADDQ $0x10, R9 |
|
| 3969 |
+ SUBQ $0x10, R12 |
|
| 3970 |
+ JAE copy_2_loop |
|
| 3971 |
+ LEAQ 16(CX)(R12*1), CX |
|
| 3972 |
+ LEAQ 16(R9)(R12*1), R9 |
|
| 3973 |
+ MOVUPS -16(CX), X0 |
|
| 3974 |
+ MOVUPS X0, -16(R9) |
|
| 3975 |
+ JMP copy_2_end |
|
| 3976 |
+ |
|
| 3977 |
+copy_2_small: |
|
| 3978 |
+ CMPQ R13, $0x03 |
|
| 3979 |
+ JE copy_2_move_3 |
|
| 3980 |
+ JB copy_2_move_1or2 |
|
| 3981 |
+ CMPQ R13, $0x08 |
|
| 3982 |
+ JB copy_2_move_4through7 |
|
| 3983 |
+ JMP copy_2_move_8through16 |
|
| 3984 |
+ |
|
| 3985 |
+copy_2_move_1or2: |
|
| 3986 |
+ MOVB (CX), R12 |
|
| 3987 |
+ MOVB -1(CX)(R13*1), R14 |
|
| 3988 |
+ MOVB R12, (R9) |
|
| 3989 |
+ MOVB R14, -1(R9)(R13*1) |
|
| 3990 |
+ ADDQ R13, CX |
|
| 3991 |
+ ADDQ R13, R9 |
|
| 3992 |
+ JMP copy_2_end |
|
| 3993 |
+ |
|
| 3994 |
+copy_2_move_3: |
|
| 3995 |
+ MOVW (CX), R12 |
|
| 3996 |
+ MOVB 2(CX), R14 |
|
| 3997 |
+ MOVW R12, (R9) |
|
| 3998 |
+ MOVB R14, 2(R9) |
|
| 3999 |
+ ADDQ R13, CX |
|
| 4000 |
+ ADDQ R13, R9 |
|
| 4001 |
+ JMP copy_2_end |
|
| 4002 |
+ |
|
| 4003 |
+copy_2_move_4through7: |
|
| 4004 |
+ MOVL (CX), R12 |
|
| 4005 |
+ MOVL -4(CX)(R13*1), R14 |
|
| 4006 |
+ MOVL R12, (R9) |
|
| 4007 |
+ MOVL R14, -4(R9)(R13*1) |
|
| 4008 |
+ ADDQ R13, CX |
|
| 4009 |
+ ADDQ R13, R9 |
|
| 4010 |
+ JMP copy_2_end |
|
| 4011 |
+ |
|
| 4012 |
+copy_2_move_8through16: |
|
| 4013 |
+ MOVQ (CX), R12 |
|
| 4014 |
+ MOVQ -8(CX)(R13*1), R14 |
|
| 4015 |
+ MOVQ R12, (R9) |
|
| 4016 |
+ MOVQ R14, -8(R9)(R13*1) |
|
| 4017 |
+ ADDQ R13, CX |
|
| 4018 |
+ ADDQ R13, R9 |
|
| 4019 |
+ |
|
| 4020 |
+copy_2_end: |
|
| 4021 |
+ JMP handle_loop |
|
| 4022 |
+ |
|
| 4023 |
+ // Copy overlapping match |
|
| 4024 |
+copy_overlapping_match: |
|
| 4025 |
+ ADDQ R13, R11 |
|
| 4026 |
+ |
|
| 4027 |
+copy_slow_3: |
|
| 4028 |
+ MOVB (CX), R12 |
|
| 4029 |
+ MOVB R12, (R9) |
|
| 4030 |
+ INCQ CX |
|
| 4031 |
+ INCQ R9 |
|
| 4032 |
+ DECQ R13 |
|
| 4033 |
+ JNZ copy_slow_3 |
|
| 4034 |
+ |
|
| 4035 |
+handle_loop: |
|
| 4036 |
+ MOVQ ctx+16(FP), CX |
|
| 4037 |
+ DECQ 96(CX) |
|
| 4038 |
+ JNS sequenceDecs_decodeSync_safe_bmi2_main_loop |
|
| 4039 |
+ |
|
| 4040 |
+loop_finished: |
|
| 4041 |
+ MOVQ br+8(FP), CX |
|
| 4042 |
+ MOVQ AX, 32(CX) |
|
| 4043 |
+ MOVB DL, 40(CX) |
|
| 4044 |
+ MOVQ BX, 24(CX) |
|
| 4045 |
+ |
|
| 4046 |
+ // Update the context |
|
| 4047 |
+ MOVQ ctx+16(FP), AX |
|
| 4048 |
+ MOVQ R11, 136(AX) |
|
| 4049 |
+ MOVQ 144(AX), CX |
|
| 4050 |
+ SUBQ CX, R10 |
|
| 4051 |
+ MOVQ R10, 168(AX) |
|
| 4052 |
+ |
|
| 4053 |
+ // Return success |
|
| 4054 |
+ MOVQ $0x00000000, ret+24(FP) |
|
| 4055 |
+ RET |
|
| 4056 |
+ |
|
| 4057 |
+ // Return with match length error |
|
| 4058 |
+sequenceDecs_decodeSync_safe_bmi2_error_match_len_ofs_mismatch: |
|
| 4059 |
+ MOVQ 16(SP), AX |
|
| 4060 |
+ MOVQ ctx+16(FP), CX |
|
| 4061 |
+ MOVQ AX, 216(CX) |
|
| 4062 |
+ MOVQ $0x00000001, ret+24(FP) |
|
| 4063 |
+ RET |
|
| 4064 |
+ |
|
| 4065 |
+ // Return with match too long error |
|
| 4066 |
+sequenceDecs_decodeSync_safe_bmi2_error_match_len_too_big: |
|
| 4067 |
+ MOVQ ctx+16(FP), AX |
|
| 4068 |
+ MOVQ 16(SP), CX |
|
| 4069 |
+ MOVQ CX, 216(AX) |
|
| 4070 |
+ MOVQ $0x00000002, ret+24(FP) |
|
| 4071 |
+ RET |
|
| 4072 |
+ |
|
| 4073 |
+ // Return with match offset too long error |
|
| 4074 |
+error_match_off_too_big: |
|
| 4075 |
+ MOVQ ctx+16(FP), AX |
|
| 4076 |
+ MOVQ 8(SP), CX |
|
| 4077 |
+ MOVQ CX, 224(AX) |
|
| 4078 |
+ MOVQ R11, 136(AX) |
|
| 4079 |
+ MOVQ $0x00000003, ret+24(FP) |
|
| 4080 |
+ RET |
|
| 4081 |
+ |
|
| 4082 |
+ // Return with not enough literals error |
|
| 4083 |
+error_not_enough_literals: |
|
| 4084 |
+ MOVQ ctx+16(FP), AX |
|
| 4085 |
+ MOVQ 24(SP), CX |
|
| 4086 |
+ MOVQ CX, 208(AX) |
|
| 4087 |
+ MOVQ $0x00000004, ret+24(FP) |
|
| 4088 |
+ RET |
|
| 4089 |
+ |
|
| 4090 |
+ // Return with not enough output space error |
|
| 4091 |
+error_not_enough_space: |
|
| 4092 |
+ MOVQ ctx+16(FP), AX |
|
| 4093 |
+ MOVQ 24(SP), CX |
|
| 4094 |
+ MOVQ CX, 208(AX) |
|
| 4095 |
+ MOVQ 16(SP), CX |
|
| 4096 |
+ MOVQ CX, 216(AX) |
|
| 4097 |
+ MOVQ R11, 136(AX) |
|
| 4098 |
+ MOVQ $0x00000005, ret+24(FP) |
|
| 4099 |
+ RET |
| 0 | 4100 |
new file mode 100644 |
| ... | ... |
@@ -0,0 +1,237 @@ |
| 0 |
+//go:build !amd64 || appengine || !gc || noasm |
|
| 1 |
+// +build !amd64 appengine !gc noasm |
|
| 2 |
+ |
|
| 3 |
+package zstd |
|
| 4 |
+ |
|
| 5 |
+import ( |
|
| 6 |
+ "fmt" |
|
| 7 |
+ "io" |
|
| 8 |
+) |
|
| 9 |
+ |
|
| 10 |
+// decode sequences from the stream with the provided history but without dictionary. |
|
| 11 |
+func (s *sequenceDecs) decodeSyncSimple(hist []byte) (bool, error) {
|
|
| 12 |
+ return false, nil |
|
| 13 |
+} |
|
| 14 |
+ |
|
| 15 |
+// decode sequences from the stream without the provided history. |
|
| 16 |
+func (s *sequenceDecs) decode(seqs []seqVals) error {
|
|
| 17 |
+ br := s.br |
|
| 18 |
+ |
|
| 19 |
+ // Grab full sizes tables, to avoid bounds checks. |
|
| 20 |
+ llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize] |
|
| 21 |
+ llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state |
|
| 22 |
+ s.seqSize = 0 |
|
| 23 |
+ litRemain := len(s.literals) |
|
| 24 |
+ |
|
| 25 |
+ maxBlockSize := maxCompressedBlockSize |
|
| 26 |
+ if s.windowSize < maxBlockSize {
|
|
| 27 |
+ maxBlockSize = s.windowSize |
|
| 28 |
+ } |
|
| 29 |
+ for i := range seqs {
|
|
| 30 |
+ var ll, mo, ml int |
|
| 31 |
+ if br.off > 4+((maxOffsetBits+16+16)>>3) {
|
|
| 32 |
+ // inlined function: |
|
| 33 |
+ // ll, mo, ml = s.nextFast(br, llState, mlState, ofState) |
|
| 34 |
+ |
|
| 35 |
+ // Final will not read from stream. |
|
| 36 |
+ var llB, mlB, moB uint8 |
|
| 37 |
+ ll, llB = llState.final() |
|
| 38 |
+ ml, mlB = mlState.final() |
|
| 39 |
+ mo, moB = ofState.final() |
|
| 40 |
+ |
|
| 41 |
+ // extra bits are stored in reverse order. |
|
| 42 |
+ br.fillFast() |
|
| 43 |
+ mo += br.getBits(moB) |
|
| 44 |
+ if s.maxBits > 32 {
|
|
| 45 |
+ br.fillFast() |
|
| 46 |
+ } |
|
| 47 |
+ ml += br.getBits(mlB) |
|
| 48 |
+ ll += br.getBits(llB) |
|
| 49 |
+ |
|
| 50 |
+ if moB > 1 {
|
|
| 51 |
+ s.prevOffset[2] = s.prevOffset[1] |
|
| 52 |
+ s.prevOffset[1] = s.prevOffset[0] |
|
| 53 |
+ s.prevOffset[0] = mo |
|
| 54 |
+ } else {
|
|
| 55 |
+ // mo = s.adjustOffset(mo, ll, moB) |
|
| 56 |
+ // Inlined for rather big speedup |
|
| 57 |
+ if ll == 0 {
|
|
| 58 |
+ // There is an exception though, when current sequence's literals_length = 0. |
|
| 59 |
+ // In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2, |
|
| 60 |
+ // an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte. |
|
| 61 |
+ mo++ |
|
| 62 |
+ } |
|
| 63 |
+ |
|
| 64 |
+ if mo == 0 {
|
|
| 65 |
+ mo = s.prevOffset[0] |
|
| 66 |
+ } else {
|
|
| 67 |
+ var temp int |
|
| 68 |
+ if mo == 3 {
|
|
| 69 |
+ temp = s.prevOffset[0] - 1 |
|
| 70 |
+ } else {
|
|
| 71 |
+ temp = s.prevOffset[mo] |
|
| 72 |
+ } |
|
| 73 |
+ |
|
| 74 |
+ if temp == 0 {
|
|
| 75 |
+ // 0 is not valid; input is corrupted; force offset to 1 |
|
| 76 |
+ println("WARNING: temp was 0")
|
|
| 77 |
+ temp = 1 |
|
| 78 |
+ } |
|
| 79 |
+ |
|
| 80 |
+ if mo != 1 {
|
|
| 81 |
+ s.prevOffset[2] = s.prevOffset[1] |
|
| 82 |
+ } |
|
| 83 |
+ s.prevOffset[1] = s.prevOffset[0] |
|
| 84 |
+ s.prevOffset[0] = temp |
|
| 85 |
+ mo = temp |
|
| 86 |
+ } |
|
| 87 |
+ } |
|
| 88 |
+ br.fillFast() |
|
| 89 |
+ } else {
|
|
| 90 |
+ if br.overread() {
|
|
| 91 |
+ if debugDecoder {
|
|
| 92 |
+ printf("reading sequence %d, exceeded available data\n", i)
|
|
| 93 |
+ } |
|
| 94 |
+ return io.ErrUnexpectedEOF |
|
| 95 |
+ } |
|
| 96 |
+ ll, mo, ml = s.next(br, llState, mlState, ofState) |
|
| 97 |
+ br.fill() |
|
| 98 |
+ } |
|
| 99 |
+ |
|
| 100 |
+ if debugSequences {
|
|
| 101 |
+ println("Seq", i, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml)
|
|
| 102 |
+ } |
|
| 103 |
+ // Evaluate. |
|
| 104 |
+ // We might be doing this async, so do it early. |
|
| 105 |
+ if mo == 0 && ml > 0 {
|
|
| 106 |
+ return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
|
|
| 107 |
+ } |
|
| 108 |
+ if ml > maxMatchLen {
|
|
| 109 |
+ return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
|
|
| 110 |
+ } |
|
| 111 |
+ s.seqSize += ll + ml |
|
| 112 |
+ if s.seqSize > maxBlockSize {
|
|
| 113 |
+ return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
|
|
| 114 |
+ } |
|
| 115 |
+ litRemain -= ll |
|
| 116 |
+ if litRemain < 0 {
|
|
| 117 |
+ return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, litRemain+ll)
|
|
| 118 |
+ } |
|
| 119 |
+ seqs[i] = seqVals{
|
|
| 120 |
+ ll: ll, |
|
| 121 |
+ ml: ml, |
|
| 122 |
+ mo: mo, |
|
| 123 |
+ } |
|
| 124 |
+ if i == len(seqs)-1 {
|
|
| 125 |
+ // This is the last sequence, so we shouldn't update state. |
|
| 126 |
+ break |
|
| 127 |
+ } |
|
| 128 |
+ |
|
| 129 |
+ // Manually inlined, ~ 5-20% faster |
|
| 130 |
+ // Update all 3 states at once. Approx 20% faster. |
|
| 131 |
+ nBits := llState.nbBits() + mlState.nbBits() + ofState.nbBits() |
|
| 132 |
+ if nBits == 0 {
|
|
| 133 |
+ llState = llTable[llState.newState()&maxTableMask] |
|
| 134 |
+ mlState = mlTable[mlState.newState()&maxTableMask] |
|
| 135 |
+ ofState = ofTable[ofState.newState()&maxTableMask] |
|
| 136 |
+ } else {
|
|
| 137 |
+ bits := br.get32BitsFast(nBits) |
|
| 138 |
+ lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31)) |
|
| 139 |
+ llState = llTable[(llState.newState()+lowBits)&maxTableMask] |
|
| 140 |
+ |
|
| 141 |
+ lowBits = uint16(bits >> (ofState.nbBits() & 31)) |
|
| 142 |
+ lowBits &= bitMask[mlState.nbBits()&15] |
|
| 143 |
+ mlState = mlTable[(mlState.newState()+lowBits)&maxTableMask] |
|
| 144 |
+ |
|
| 145 |
+ lowBits = uint16(bits) & bitMask[ofState.nbBits()&15] |
|
| 146 |
+ ofState = ofTable[(ofState.newState()+lowBits)&maxTableMask] |
|
| 147 |
+ } |
|
| 148 |
+ } |
|
| 149 |
+ s.seqSize += litRemain |
|
| 150 |
+ if s.seqSize > maxBlockSize {
|
|
| 151 |
+ return fmt.Errorf("output (%d) bigger than max block size (%d)", s.seqSize, maxBlockSize)
|
|
| 152 |
+ } |
|
| 153 |
+ err := br.close() |
|
| 154 |
+ if err != nil {
|
|
| 155 |
+ printf("Closing sequences: %v, %+v\n", err, *br)
|
|
| 156 |
+ } |
|
| 157 |
+ return err |
|
| 158 |
+} |
|
| 159 |
+ |
|
| 160 |
+// executeSimple handles cases when a dictionary is not used. |
|
| 161 |
+func (s *sequenceDecs) executeSimple(seqs []seqVals, hist []byte) error {
|
|
| 162 |
+ // Ensure we have enough output size... |
|
| 163 |
+ if len(s.out)+s.seqSize > cap(s.out) {
|
|
| 164 |
+ addBytes := s.seqSize + len(s.out) |
|
| 165 |
+ s.out = append(s.out, make([]byte, addBytes)...) |
|
| 166 |
+ s.out = s.out[:len(s.out)-addBytes] |
|
| 167 |
+ } |
|
| 168 |
+ |
|
| 169 |
+ if debugDecoder {
|
|
| 170 |
+ printf("Execute %d seqs with literals: %d into %d bytes\n", len(seqs), len(s.literals), s.seqSize)
|
|
| 171 |
+ } |
|
| 172 |
+ |
|
| 173 |
+ var t = len(s.out) |
|
| 174 |
+ out := s.out[:t+s.seqSize] |
|
| 175 |
+ |
|
| 176 |
+ for _, seq := range seqs {
|
|
| 177 |
+ // Add literals |
|
| 178 |
+ copy(out[t:], s.literals[:seq.ll]) |
|
| 179 |
+ t += seq.ll |
|
| 180 |
+ s.literals = s.literals[seq.ll:] |
|
| 181 |
+ |
|
| 182 |
+ // Malformed input |
|
| 183 |
+ if seq.mo > t+len(hist) || seq.mo > s.windowSize {
|
|
| 184 |
+ return fmt.Errorf("match offset (%d) bigger than current history (%d)", seq.mo, t+len(hist))
|
|
| 185 |
+ } |
|
| 186 |
+ |
|
| 187 |
+ // Copy from history. |
|
| 188 |
+ if v := seq.mo - t; v > 0 {
|
|
| 189 |
+ // v is the start position in history from end. |
|
| 190 |
+ start := len(hist) - v |
|
| 191 |
+ if seq.ml > v {
|
|
| 192 |
+ // Some goes into the current block. |
|
| 193 |
+ // Copy remainder of history |
|
| 194 |
+ copy(out[t:], hist[start:]) |
|
| 195 |
+ t += v |
|
| 196 |
+ seq.ml -= v |
|
| 197 |
+ } else {
|
|
| 198 |
+ copy(out[t:], hist[start:start+seq.ml]) |
|
| 199 |
+ t += seq.ml |
|
| 200 |
+ continue |
|
| 201 |
+ } |
|
| 202 |
+ } |
|
| 203 |
+ |
|
| 204 |
+ // We must be in the current buffer now |
|
| 205 |
+ if seq.ml > 0 {
|
|
| 206 |
+ start := t - seq.mo |
|
| 207 |
+ if seq.ml <= t-start {
|
|
| 208 |
+ // No overlap |
|
| 209 |
+ copy(out[t:], out[start:start+seq.ml]) |
|
| 210 |
+ t += seq.ml |
|
| 211 |
+ } else {
|
|
| 212 |
+ // Overlapping copy |
|
| 213 |
+ // Extend destination slice and copy one byte at the time. |
|
| 214 |
+ src := out[start : start+seq.ml] |
|
| 215 |
+ dst := out[t:] |
|
| 216 |
+ dst = dst[:len(src)] |
|
| 217 |
+ t += len(src) |
|
| 218 |
+ // Destination is the space we just added. |
|
| 219 |
+ for i := range src {
|
|
| 220 |
+ dst[i] = src[i] |
|
| 221 |
+ } |
|
| 222 |
+ } |
|
| 223 |
+ } |
|
| 224 |
+ } |
|
| 225 |
+ // Add final literals |
|
| 226 |
+ copy(out[t:], s.literals) |
|
| 227 |
+ if debugDecoder {
|
|
| 228 |
+ t += len(s.literals) |
|
| 229 |
+ if t != len(out) {
|
|
| 230 |
+ panic(fmt.Errorf("length mismatch, want %d, got %d, ss: %d", len(out), t, s.seqSize))
|
|
| 231 |
+ } |
|
| 232 |
+ } |
|
| 233 |
+ s.out = out |
|
| 234 |
+ |
|
| 235 |
+ return nil |
|
| 236 |
+} |
| ... | ... |
@@ -18,26 +18,44 @@ const ZipMethodWinZip = 93 |
| 18 | 18 |
// See https://pkware.cachefly.net/webdocs/APPNOTE/APPNOTE-6.3.9.TXT |
| 19 | 19 |
const ZipMethodPKWare = 20 |
| 20 | 20 |
|
| 21 |
-var zipReaderPool sync.Pool |
|
| 21 |
+// zipReaderPool is the default reader pool. |
|
| 22 |
+var zipReaderPool = sync.Pool{New: func() interface{} {
|
|
| 23 |
+ z, err := NewReader(nil, WithDecoderLowmem(true), WithDecoderMaxWindow(128<<20), WithDecoderConcurrency(1)) |
|
| 24 |
+ if err != nil {
|
|
| 25 |
+ panic(err) |
|
| 26 |
+ } |
|
| 27 |
+ return z |
|
| 28 |
+}} |
|
| 22 | 29 |
|
| 23 | 30 |
// newZipReader creates a pooled zip decompressor. |
| 24 |
-func newZipReader(r io.Reader) io.ReadCloser {
|
|
| 25 |
- dec, ok := zipReaderPool.Get().(*Decoder) |
|
| 26 |
- if ok {
|
|
| 27 |
- dec.Reset(r) |
|
| 28 |
- } else {
|
|
| 29 |
- d, err := NewReader(r, WithDecoderConcurrency(1), WithDecoderLowmem(true)) |
|
| 30 |
- if err != nil {
|
|
| 31 |
- panic(err) |
|
| 31 |
+func newZipReader(opts ...DOption) func(r io.Reader) io.ReadCloser {
|
|
| 32 |
+ pool := &zipReaderPool |
|
| 33 |
+ if len(opts) > 0 {
|
|
| 34 |
+ opts = append([]DOption{WithDecoderLowmem(true), WithDecoderMaxWindow(128 << 20)}, opts...)
|
|
| 35 |
+ // Force concurrency 1 |
|
| 36 |
+ opts = append(opts, WithDecoderConcurrency(1)) |
|
| 37 |
+ // Create our own pool |
|
| 38 |
+ pool = &sync.Pool{}
|
|
| 39 |
+ } |
|
| 40 |
+ return func(r io.Reader) io.ReadCloser {
|
|
| 41 |
+ dec, ok := pool.Get().(*Decoder) |
|
| 42 |
+ if ok {
|
|
| 43 |
+ dec.Reset(r) |
|
| 44 |
+ } else {
|
|
| 45 |
+ d, err := NewReader(r, opts...) |
|
| 46 |
+ if err != nil {
|
|
| 47 |
+ panic(err) |
|
| 48 |
+ } |
|
| 49 |
+ dec = d |
|
| 32 | 50 |
} |
| 33 |
- dec = d |
|
| 51 |
+ return &pooledZipReader{dec: dec, pool: pool}
|
|
| 34 | 52 |
} |
| 35 |
- return &pooledZipReader{dec: dec}
|
|
| 36 | 53 |
} |
| 37 | 54 |
|
| 38 | 55 |
type pooledZipReader struct {
|
| 39 |
- mu sync.Mutex // guards Close and Read |
|
| 40 |
- dec *Decoder |
|
| 56 |
+ mu sync.Mutex // guards Close and Read |
|
| 57 |
+ pool *sync.Pool |
|
| 58 |
+ dec *Decoder |
|
| 41 | 59 |
} |
| 42 | 60 |
|
| 43 | 61 |
func (r *pooledZipReader) Read(p []byte) (n int, err error) {
|
| ... | ... |
@@ -48,8 +66,8 @@ func (r *pooledZipReader) Read(p []byte) (n int, err error) {
|
| 48 | 48 |
} |
| 49 | 49 |
dec, err := r.dec.Read(p) |
| 50 | 50 |
if err == io.EOF {
|
| 51 |
- err = r.dec.Reset(nil) |
|
| 52 |
- zipReaderPool.Put(r.dec) |
|
| 51 |
+ r.dec.Reset(nil) |
|
| 52 |
+ r.pool.Put(r.dec) |
|
| 53 | 53 |
r.dec = nil |
| 54 | 54 |
} |
| 55 | 55 |
return dec, err |
| ... | ... |
@@ -61,7 +79,7 @@ func (r *pooledZipReader) Close() error {
|
| 61 | 61 |
var err error |
| 62 | 62 |
if r.dec != nil {
|
| 63 | 63 |
err = r.dec.Reset(nil) |
| 64 |
- zipReaderPool.Put(r.dec) |
|
| 64 |
+ r.pool.Put(r.dec) |
|
| 65 | 65 |
r.dec = nil |
| 66 | 66 |
} |
| 67 | 67 |
return err |
| ... | ... |
@@ -115,6 +133,9 @@ func ZipCompressor(opts ...EOption) func(w io.Writer) (io.WriteCloser, error) {
|
| 115 | 115 |
|
| 116 | 116 |
// ZipDecompressor returns a decompressor that can be registered with zip libraries. |
| 117 | 117 |
// See ZipCompressor for example. |
| 118 |
-func ZipDecompressor() func(r io.Reader) io.ReadCloser {
|
|
| 119 |
- return newZipReader |
|
| 118 |
+// Options can be specified. WithDecoderConcurrency(1) is forced, |
|
| 119 |
+// and by default a 128MB maximum decompression window is specified. |
|
| 120 |
+// The window size can be overridden if required. |
|
| 121 |
+func ZipDecompressor(opts ...DOption) func(r io.Reader) io.ReadCloser {
|
|
| 122 |
+ return newZipReader(opts...) |
|
| 120 | 123 |
} |
| ... | ... |
@@ -110,17 +110,6 @@ func printf(format string, a ...interface{}) {
|
| 110 | 110 |
} |
| 111 | 111 |
} |
| 112 | 112 |
|
| 113 |
-// matchLenFast does matching, but will not match the last up to 7 bytes. |
|
| 114 |
-func matchLenFast(a, b []byte) int {
|
|
| 115 |
- endI := len(a) & (math.MaxInt32 - 7) |
|
| 116 |
- for i := 0; i < endI; i += 8 {
|
|
| 117 |
- if diff := load64(a, i) ^ load64(b, i); diff != 0 {
|
|
| 118 |
- return i + bits.TrailingZeros64(diff)>>3 |
|
| 119 |
- } |
|
| 120 |
- } |
|
| 121 |
- return endI |
|
| 122 |
-} |
|
| 123 |
- |
|
| 124 | 113 |
// matchLen returns the maximum length. |
| 125 | 114 |
// a must be the shortest of the two. |
| 126 | 115 |
// The function also returns whether all bytes matched. |
| ... | ... |
@@ -454,11 +454,12 @@ github.com/ishidawataru/sctp |
| 454 | 454 |
# github.com/jmespath/go-jmespath v0.3.0 |
| 455 | 455 |
## explicit; go 1.14 |
| 456 | 456 |
github.com/jmespath/go-jmespath |
| 457 |
-# github.com/klauspost/compress v1.15.1 |
|
| 458 |
-## explicit; go 1.15 |
|
| 457 |
+# github.com/klauspost/compress v1.15.9 |
|
| 458 |
+## explicit; go 1.16 |
|
| 459 | 459 |
github.com/klauspost/compress |
| 460 | 460 |
github.com/klauspost/compress/fse |
| 461 | 461 |
github.com/klauspost/compress/huff0 |
| 462 |
+github.com/klauspost/compress/internal/cpuinfo |
|
| 462 | 463 |
github.com/klauspost/compress/internal/snapref |
| 463 | 464 |
github.com/klauspost/compress/zstd |
| 464 | 465 |
github.com/klauspost/compress/zstd/internal/xxhash |