aboutsummaryrefslogtreecommitdiffstats
path: root/vendor/github.com/klauspost/compress/s2/encode_all.go
diff options
context:
space:
mode:
Diffstat (limited to 'vendor/github.com/klauspost/compress/s2/encode_all.go')
-rw-r--r--vendor/github.com/klauspost/compress/s2/encode_all.go1048
1 files changed, 1048 insertions, 0 deletions
diff --git a/vendor/github.com/klauspost/compress/s2/encode_all.go b/vendor/github.com/klauspost/compress/s2/encode_all.go
new file mode 100644
index 0000000..5e57995
--- /dev/null
+++ b/vendor/github.com/klauspost/compress/s2/encode_all.go
@@ -0,0 +1,1048 @@
1// Copyright 2016 The Snappy-Go Authors. All rights reserved.
2// Copyright (c) 2019 Klaus Post. All rights reserved.
3// Use of this source code is governed by a BSD-style
4// license that can be found in the LICENSE file.
5
6package s2
7
8import (
9 "bytes"
10 "encoding/binary"
11 "fmt"
12 "math/bits"
13)
14
15func load32(b []byte, i int) uint32 {
16 return binary.LittleEndian.Uint32(b[i:])
17}
18
19func load64(b []byte, i int) uint64 {
20 return binary.LittleEndian.Uint64(b[i:])
21}
22
23// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits.
24// Preferably h should be a constant and should always be <64.
25func hash6(u uint64, h uint8) uint32 {
26 const prime6bytes = 227718039650203
27 return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & 63))
28}
29
30func encodeGo(dst, src []byte) []byte {
31 if n := MaxEncodedLen(len(src)); n < 0 {
32 panic(ErrTooLarge)
33 } else if len(dst) < n {
34 dst = make([]byte, n)
35 }
36
37 // The block starts with the varint-encoded length of the decompressed bytes.
38 d := binary.PutUvarint(dst, uint64(len(src)))
39
40 if len(src) == 0 {
41 return dst[:d]
42 }
43 if len(src) < minNonLiteralBlockSize {
44 d += emitLiteral(dst[d:], src)
45 return dst[:d]
46 }
47 n := encodeBlockGo(dst[d:], src)
48 if n > 0 {
49 d += n
50 return dst[:d]
51 }
52 // Not compressible
53 d += emitLiteral(dst[d:], src)
54 return dst[:d]
55}
56
57// encodeBlockGo encodes a non-empty src to a guaranteed-large-enough dst. It
58// assumes that the varint-encoded length of the decompressed bytes has already
59// been written.
60//
61// It also assumes that:
62//
63// len(dst) >= MaxEncodedLen(len(src)) &&
64// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
65func encodeBlockGo(dst, src []byte) (d int) {
66 // Initialize the hash table.
67 const (
68 tableBits = 14
69 maxTableSize = 1 << tableBits
70
71 debug = false
72 )
73
74 var table [maxTableSize]uint32
75
76 // sLimit is when to stop looking for offset/length copies. The inputMargin
77 // lets us use a fast path for emitLiteral in the main loop, while we are
78 // looking for copies.
79 sLimit := len(src) - inputMargin
80
81 // Bail if we can't compress to at least this.
82 dstLimit := len(src) - len(src)>>5 - 5
83
84 // nextEmit is where in src the next emitLiteral should start from.
85 nextEmit := 0
86
87 // The encoded form must start with a literal, as there are no previous
88 // bytes to copy, so we start looking for hash matches at s == 1.
89 s := 1
90 cv := load64(src, s)
91
92 // We search for a repeat at -1, but don't output repeats when nextEmit == 0
93 repeat := 1
94
95 for {
96 candidate := 0
97 for {
98 // Next src position to check
99 nextS := s + (s-nextEmit)>>6 + 4
100 if nextS > sLimit {
101 goto emitRemainder
102 }
103 hash0 := hash6(cv, tableBits)
104 hash1 := hash6(cv>>8, tableBits)
105 candidate = int(table[hash0])
106 candidate2 := int(table[hash1])
107 table[hash0] = uint32(s)
108 table[hash1] = uint32(s + 1)
109 hash2 := hash6(cv>>16, tableBits)
110
111 // Check repeat at offset checkRep.
112 const checkRep = 1
113 if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
114 base := s + checkRep
115 // Extend back
116 for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
117 i--
118 base--
119 }
120 d += emitLiteral(dst[d:], src[nextEmit:base])
121
122 // Extend forward
123 candidate := s - repeat + 4 + checkRep
124 s += 4 + checkRep
125 for s <= sLimit {
126 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
127 s += bits.TrailingZeros64(diff) >> 3
128 break
129 }
130 s += 8
131 candidate += 8
132 }
133 if debug {
134 // Validate match.
135 if s <= candidate {
136 panic("s <= candidate")
137 }
138 a := src[base:s]
139 b := src[base-repeat : base-repeat+(s-base)]
140 if !bytes.Equal(a, b) {
141 panic("mismatch")
142 }
143 }
144 if nextEmit > 0 {
145 // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
146 d += emitRepeat(dst[d:], repeat, s-base)
147 } else {
148 // First match, cannot be repeat.
149 d += emitCopy(dst[d:], repeat, s-base)
150 }
151 nextEmit = s
152 if s >= sLimit {
153 goto emitRemainder
154 }
155
156 cv = load64(src, s)
157 continue
158 }
159
160 if uint32(cv) == load32(src, candidate) {
161 break
162 }
163 candidate = int(table[hash2])
164 if uint32(cv>>8) == load32(src, candidate2) {
165 table[hash2] = uint32(s + 2)
166 candidate = candidate2
167 s++
168 break
169 }
170 table[hash2] = uint32(s + 2)
171 if uint32(cv>>16) == load32(src, candidate) {
172 s += 2
173 break
174 }
175
176 cv = load64(src, nextS)
177 s = nextS
178 }
179
180 // Extend backwards.
181 // The top bytes will be rechecked to get the full match.
182 for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
183 candidate--
184 s--
185 }
186
187 // Bail if we exceed the maximum size.
188 if d+(s-nextEmit) > dstLimit {
189 return 0
190 }
191
192 // A 4-byte match has been found. We'll later see if more than 4 bytes
193 // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
194 // them as literal bytes.
195
196 d += emitLiteral(dst[d:], src[nextEmit:s])
197
198 // Call emitCopy, and then see if another emitCopy could be our next
199 // move. Repeat until we find no match for the input immediately after
200 // what was consumed by the last emitCopy call.
201 //
202 // If we exit this loop normally then we need to call emitLiteral next,
203 // though we don't yet know how big the literal will be. We handle that
204 // by proceeding to the next iteration of the main loop. We also can
205 // exit this loop via goto if we get close to exhausting the input.
206 for {
207 // Invariant: we have a 4-byte match at s, and no need to emit any
208 // literal bytes prior to s.
209 base := s
210 repeat = base - candidate
211
212 // Extend the 4-byte match as long as possible.
213 s += 4
214 candidate += 4
215 for s <= len(src)-8 {
216 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
217 s += bits.TrailingZeros64(diff) >> 3
218 break
219 }
220 s += 8
221 candidate += 8
222 }
223
224 d += emitCopy(dst[d:], repeat, s-base)
225 if debug {
226 // Validate match.
227 if s <= candidate {
228 panic("s <= candidate")
229 }
230 a := src[base:s]
231 b := src[base-repeat : base-repeat+(s-base)]
232 if !bytes.Equal(a, b) {
233 panic("mismatch")
234 }
235 }
236
237 nextEmit = s
238 if s >= sLimit {
239 goto emitRemainder
240 }
241
242 if d > dstLimit {
243 // Do we have space for more, if not bail.
244 return 0
245 }
246 // Check for an immediate match, otherwise start search at s+1
247 x := load64(src, s-2)
248 m2Hash := hash6(x, tableBits)
249 currHash := hash6(x>>16, tableBits)
250 candidate = int(table[currHash])
251 table[m2Hash] = uint32(s - 2)
252 table[currHash] = uint32(s)
253 if debug && s == candidate {
254 panic("s == candidate")
255 }
256 if uint32(x>>16) != load32(src, candidate) {
257 cv = load64(src, s+1)
258 s++
259 break
260 }
261 }
262 }
263
264emitRemainder:
265 if nextEmit < len(src) {
266 // Bail if we exceed the maximum size.
267 if d+len(src)-nextEmit > dstLimit {
268 return 0
269 }
270 d += emitLiteral(dst[d:], src[nextEmit:])
271 }
272 return d
273}
274
275func encodeBlockSnappyGo(dst, src []byte) (d int) {
276 // Initialize the hash table.
277 const (
278 tableBits = 14
279 maxTableSize = 1 << tableBits
280 )
281
282 var table [maxTableSize]uint32
283
284 // sLimit is when to stop looking for offset/length copies. The inputMargin
285 // lets us use a fast path for emitLiteral in the main loop, while we are
286 // looking for copies.
287 sLimit := len(src) - inputMargin
288
289 // Bail if we can't compress to at least this.
290 dstLimit := len(src) - len(src)>>5 - 5
291
292 // nextEmit is where in src the next emitLiteral should start from.
293 nextEmit := 0
294
295 // The encoded form must start with a literal, as there are no previous
296 // bytes to copy, so we start looking for hash matches at s == 1.
297 s := 1
298 cv := load64(src, s)
299
300 // We search for a repeat at -1, but don't output repeats when nextEmit == 0
301 repeat := 1
302
303 for {
304 candidate := 0
305 for {
306 // Next src position to check
307 nextS := s + (s-nextEmit)>>6 + 4
308 if nextS > sLimit {
309 goto emitRemainder
310 }
311 hash0 := hash6(cv, tableBits)
312 hash1 := hash6(cv>>8, tableBits)
313 candidate = int(table[hash0])
314 candidate2 := int(table[hash1])
315 table[hash0] = uint32(s)
316 table[hash1] = uint32(s + 1)
317 hash2 := hash6(cv>>16, tableBits)
318
319 // Check repeat at offset checkRep.
320 const checkRep = 1
321 if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
322 base := s + checkRep
323 // Extend back
324 for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
325 i--
326 base--
327 }
328 d += emitLiteral(dst[d:], src[nextEmit:base])
329
330 // Extend forward
331 candidate := s - repeat + 4 + checkRep
332 s += 4 + checkRep
333 for s <= sLimit {
334 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
335 s += bits.TrailingZeros64(diff) >> 3
336 break
337 }
338 s += 8
339 candidate += 8
340 }
341
342 d += emitCopyNoRepeat(dst[d:], repeat, s-base)
343 nextEmit = s
344 if s >= sLimit {
345 goto emitRemainder
346 }
347
348 cv = load64(src, s)
349 continue
350 }
351
352 if uint32(cv) == load32(src, candidate) {
353 break
354 }
355 candidate = int(table[hash2])
356 if uint32(cv>>8) == load32(src, candidate2) {
357 table[hash2] = uint32(s + 2)
358 candidate = candidate2
359 s++
360 break
361 }
362 table[hash2] = uint32(s + 2)
363 if uint32(cv>>16) == load32(src, candidate) {
364 s += 2
365 break
366 }
367
368 cv = load64(src, nextS)
369 s = nextS
370 }
371
372 // Extend backwards
373 for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
374 candidate--
375 s--
376 }
377
378 // Bail if we exceed the maximum size.
379 if d+(s-nextEmit) > dstLimit {
380 return 0
381 }
382
383 // A 4-byte match has been found. We'll later see if more than 4 bytes
384 // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
385 // them as literal bytes.
386
387 d += emitLiteral(dst[d:], src[nextEmit:s])
388
389 // Call emitCopy, and then see if another emitCopy could be our next
390 // move. Repeat until we find no match for the input immediately after
391 // what was consumed by the last emitCopy call.
392 //
393 // If we exit this loop normally then we need to call emitLiteral next,
394 // though we don't yet know how big the literal will be. We handle that
395 // by proceeding to the next iteration of the main loop. We also can
396 // exit this loop via goto if we get close to exhausting the input.
397 for {
398 // Invariant: we have a 4-byte match at s, and no need to emit any
399 // literal bytes prior to s.
400 base := s
401 repeat = base - candidate
402
403 // Extend the 4-byte match as long as possible.
404 s += 4
405 candidate += 4
406 for s <= len(src)-8 {
407 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
408 s += bits.TrailingZeros64(diff) >> 3
409 break
410 }
411 s += 8
412 candidate += 8
413 }
414
415 d += emitCopyNoRepeat(dst[d:], repeat, s-base)
416 if false {
417 // Validate match.
418 a := src[base:s]
419 b := src[base-repeat : base-repeat+(s-base)]
420 if !bytes.Equal(a, b) {
421 panic("mismatch")
422 }
423 }
424
425 nextEmit = s
426 if s >= sLimit {
427 goto emitRemainder
428 }
429
430 if d > dstLimit {
431 // Do we have space for more, if not bail.
432 return 0
433 }
434 // Check for an immediate match, otherwise start search at s+1
435 x := load64(src, s-2)
436 m2Hash := hash6(x, tableBits)
437 currHash := hash6(x>>16, tableBits)
438 candidate = int(table[currHash])
439 table[m2Hash] = uint32(s - 2)
440 table[currHash] = uint32(s)
441 if uint32(x>>16) != load32(src, candidate) {
442 cv = load64(src, s+1)
443 s++
444 break
445 }
446 }
447 }
448
449emitRemainder:
450 if nextEmit < len(src) {
451 // Bail if we exceed the maximum size.
452 if d+len(src)-nextEmit > dstLimit {
453 return 0
454 }
455 d += emitLiteral(dst[d:], src[nextEmit:])
456 }
457 return d
458}
459
460// encodeBlockGo encodes a non-empty src to a guaranteed-large-enough dst. It
461// assumes that the varint-encoded length of the decompressed bytes has already
462// been written.
463//
464// It also assumes that:
465//
466// len(dst) >= MaxEncodedLen(len(src)) &&
467// minNonLiteralBlockSize <= len(src) && len(src) <= maxBlockSize
468func encodeBlockDictGo(dst, src []byte, dict *Dict) (d int) {
469 // Initialize the hash table.
470 const (
471 tableBits = 14
472 maxTableSize = 1 << tableBits
473 maxAhead = 8 // maximum bytes ahead without checking sLimit
474
475 debug = false
476 )
477 dict.initFast()
478
479 var table [maxTableSize]uint32
480
481 // sLimit is when to stop looking for offset/length copies. The inputMargin
482 // lets us use a fast path for emitLiteral in the main loop, while we are
483 // looking for copies.
484 sLimit := len(src) - inputMargin
485 if sLimit > MaxDictSrcOffset-maxAhead {
486 sLimit = MaxDictSrcOffset - maxAhead
487 }
488
489 // Bail if we can't compress to at least this.
490 dstLimit := len(src) - len(src)>>5 - 5
491
492 // nextEmit is where in src the next emitLiteral should start from.
493 nextEmit := 0
494
495 // The encoded form can start with a dict entry (copy or repeat).
496 s := 0
497
498 // Convert dict repeat to offset
499 repeat := len(dict.dict) - dict.repeat
500 cv := load64(src, 0)
501
502 // While in dict
503searchDict:
504 for {
505 // Next src position to check
506 nextS := s + (s-nextEmit)>>6 + 4
507 hash0 := hash6(cv, tableBits)
508 hash1 := hash6(cv>>8, tableBits)
509 if nextS > sLimit {
510 if debug {
511 fmt.Println("slimit reached", s, nextS)
512 }
513 break searchDict
514 }
515 candidateDict := int(dict.fastTable[hash0])
516 candidateDict2 := int(dict.fastTable[hash1])
517 candidate2 := int(table[hash1])
518 candidate := int(table[hash0])
519 table[hash0] = uint32(s)
520 table[hash1] = uint32(s + 1)
521 hash2 := hash6(cv>>16, tableBits)
522
523 // Check repeat at offset checkRep.
524 const checkRep = 1
525
526 if repeat > s {
527 candidate := len(dict.dict) - repeat + s
528 if repeat-s >= 4 && uint32(cv) == load32(dict.dict, candidate) {
529 // Extend back
530 base := s
531 for i := candidate; base > nextEmit && i > 0 && dict.dict[i-1] == src[base-1]; {
532 i--
533 base--
534 }
535 d += emitLiteral(dst[d:], src[nextEmit:base])
536 if debug && nextEmit != base {
537 fmt.Println("emitted ", base-nextEmit, "literals")
538 }
539 s += 4
540 candidate += 4
541 for candidate < len(dict.dict)-8 && s <= len(src)-8 {
542 if diff := load64(src, s) ^ load64(dict.dict, candidate); diff != 0 {
543 s += bits.TrailingZeros64(diff) >> 3
544 break
545 }
546 s += 8
547 candidate += 8
548 }
549 d += emitRepeat(dst[d:], repeat, s-base)
550 if debug {
551 fmt.Println("emitted dict repeat length", s-base, "offset:", repeat, "s:", s)
552 }
553 nextEmit = s
554 if s >= sLimit {
555 break searchDict
556 }
557 cv = load64(src, s)
558 continue
559 }
560 } else if uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
561 base := s + checkRep
562 // Extend back
563 for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
564 i--
565 base--
566 }
567 d += emitLiteral(dst[d:], src[nextEmit:base])
568 if debug && nextEmit != base {
569 fmt.Println("emitted ", base-nextEmit, "literals")
570 }
571
572 // Extend forward
573 candidate := s - repeat + 4 + checkRep
574 s += 4 + checkRep
575 for s <= sLimit {
576 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
577 s += bits.TrailingZeros64(diff) >> 3
578 break
579 }
580 s += 8
581 candidate += 8
582 }
583 if debug {
584 // Validate match.
585 if s <= candidate {
586 panic("s <= candidate")
587 }
588 a := src[base:s]
589 b := src[base-repeat : base-repeat+(s-base)]
590 if !bytes.Equal(a, b) {
591 panic("mismatch")
592 }
593 }
594
595 if nextEmit > 0 {
596 // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
597 d += emitRepeat(dst[d:], repeat, s-base)
598 } else {
599 // First match, cannot be repeat.
600 d += emitCopy(dst[d:], repeat, s-base)
601 }
602
603 nextEmit = s
604 if s >= sLimit {
605 break searchDict
606 }
607 if debug {
608 fmt.Println("emitted reg repeat", s-base, "s:", s)
609 }
610 cv = load64(src, s)
611 continue searchDict
612 }
613 if s == 0 {
614 cv = load64(src, nextS)
615 s = nextS
616 continue searchDict
617 }
618 // Start with table. These matches will always be closer.
619 if uint32(cv) == load32(src, candidate) {
620 goto emitMatch
621 }
622 candidate = int(table[hash2])
623 if uint32(cv>>8) == load32(src, candidate2) {
624 table[hash2] = uint32(s + 2)
625 candidate = candidate2
626 s++
627 goto emitMatch
628 }
629
630 // Check dict. Dicts have longer offsets, so we want longer matches.
631 if cv == load64(dict.dict, candidateDict) {
632 table[hash2] = uint32(s + 2)
633 goto emitDict
634 }
635
636 candidateDict = int(dict.fastTable[hash2])
637 // Check if upper 7 bytes match
638 if candidateDict2 >= 1 {
639 if cv^load64(dict.dict, candidateDict2-1) < (1 << 8) {
640 table[hash2] = uint32(s + 2)
641 candidateDict = candidateDict2
642 s++
643 goto emitDict
644 }
645 }
646
647 table[hash2] = uint32(s + 2)
648 if uint32(cv>>16) == load32(src, candidate) {
649 s += 2
650 goto emitMatch
651 }
652 if candidateDict >= 2 {
653 // Check if upper 6 bytes match
654 if cv^load64(dict.dict, candidateDict-2) < (1 << 16) {
655 s += 2
656 goto emitDict
657 }
658 }
659
660 cv = load64(src, nextS)
661 s = nextS
662 continue searchDict
663
664 emitDict:
665 {
666 if debug {
667 if load32(dict.dict, candidateDict) != load32(src, s) {
668 panic("dict emit mismatch")
669 }
670 }
671 // Extend backwards.
672 // The top bytes will be rechecked to get the full match.
673 for candidateDict > 0 && s > nextEmit && dict.dict[candidateDict-1] == src[s-1] {
674 candidateDict--
675 s--
676 }
677
678 // Bail if we exceed the maximum size.
679 if d+(s-nextEmit) > dstLimit {
680 return 0
681 }
682
683 // A 4-byte match has been found. We'll later see if more than 4 bytes
684 // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
685 // them as literal bytes.
686
687 d += emitLiteral(dst[d:], src[nextEmit:s])
688 if debug && nextEmit != s {
689 fmt.Println("emitted ", s-nextEmit, "literals")
690 }
691 {
692 // Invariant: we have a 4-byte match at s, and no need to emit any
693 // literal bytes prior to s.
694 base := s
695 repeat = s + (len(dict.dict)) - candidateDict
696
697 // Extend the 4-byte match as long as possible.
698 s += 4
699 candidateDict += 4
700 for s <= len(src)-8 && len(dict.dict)-candidateDict >= 8 {
701 if diff := load64(src, s) ^ load64(dict.dict, candidateDict); diff != 0 {
702 s += bits.TrailingZeros64(diff) >> 3
703 break
704 }
705 s += 8
706 candidateDict += 8
707 }
708
709 // Matches longer than 64 are split.
710 if s <= sLimit || s-base < 8 {
711 d += emitCopy(dst[d:], repeat, s-base)
712 } else {
713 // Split to ensure we don't start a copy within next block
714 d += emitCopy(dst[d:], repeat, 4)
715 d += emitRepeat(dst[d:], repeat, s-base-4)
716 }
717 if false {
718 // Validate match.
719 if s <= candidate {
720 panic("s <= candidate")
721 }
722 a := src[base:s]
723 b := dict.dict[base-repeat : base-repeat+(s-base)]
724 if !bytes.Equal(a, b) {
725 panic("mismatch")
726 }
727 }
728 if debug {
729 fmt.Println("emitted dict copy, length", s-base, "offset:", repeat, "s:", s)
730 }
731 nextEmit = s
732 if s >= sLimit {
733 break searchDict
734 }
735
736 if d > dstLimit {
737 // Do we have space for more, if not bail.
738 return 0
739 }
740
741 // Index and continue loop to try new candidate.
742 x := load64(src, s-2)
743 m2Hash := hash6(x, tableBits)
744 currHash := hash6(x>>8, tableBits)
745 table[m2Hash] = uint32(s - 2)
746 table[currHash] = uint32(s - 1)
747 cv = load64(src, s)
748 }
749 continue
750 }
751 emitMatch:
752
753 // Extend backwards.
754 // The top bytes will be rechecked to get the full match.
755 for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
756 candidate--
757 s--
758 }
759
760 // Bail if we exceed the maximum size.
761 if d+(s-nextEmit) > dstLimit {
762 return 0
763 }
764
765 // A 4-byte match has been found. We'll later see if more than 4 bytes
766 // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
767 // them as literal bytes.
768
769 d += emitLiteral(dst[d:], src[nextEmit:s])
770 if debug && nextEmit != s {
771 fmt.Println("emitted ", s-nextEmit, "literals")
772 }
773 // Call emitCopy, and then see if another emitCopy could be our next
774 // move. Repeat until we find no match for the input immediately after
775 // what was consumed by the last emitCopy call.
776 //
777 // If we exit this loop normally then we need to call emitLiteral next,
778 // though we don't yet know how big the literal will be. We handle that
779 // by proceeding to the next iteration of the main loop. We also can
780 // exit this loop via goto if we get close to exhausting the input.
781 for {
782 // Invariant: we have a 4-byte match at s, and no need to emit any
783 // literal bytes prior to s.
784 base := s
785 repeat = base - candidate
786
787 // Extend the 4-byte match as long as possible.
788 s += 4
789 candidate += 4
790 for s <= len(src)-8 {
791 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
792 s += bits.TrailingZeros64(diff) >> 3
793 break
794 }
795 s += 8
796 candidate += 8
797 }
798
799 d += emitCopy(dst[d:], repeat, s-base)
800 if debug {
801 // Validate match.
802 if s <= candidate {
803 panic("s <= candidate")
804 }
805 a := src[base:s]
806 b := src[base-repeat : base-repeat+(s-base)]
807 if !bytes.Equal(a, b) {
808 panic("mismatch")
809 }
810 }
811 if debug {
812 fmt.Println("emitted src copy, length", s-base, "offset:", repeat, "s:", s)
813 }
814 nextEmit = s
815 if s >= sLimit {
816 break searchDict
817 }
818
819 if d > dstLimit {
820 // Do we have space for more, if not bail.
821 return 0
822 }
823 // Check for an immediate match, otherwise start search at s+1
824 x := load64(src, s-2)
825 m2Hash := hash6(x, tableBits)
826 currHash := hash6(x>>16, tableBits)
827 candidate = int(table[currHash])
828 table[m2Hash] = uint32(s - 2)
829 table[currHash] = uint32(s)
830 if debug && s == candidate {
831 panic("s == candidate")
832 }
833 if uint32(x>>16) != load32(src, candidate) {
834 cv = load64(src, s+1)
835 s++
836 break
837 }
838 }
839 }
840
841 // Search without dict:
842 if repeat > s {
843 repeat = 0
844 }
845
846 // No more dict
847 sLimit = len(src) - inputMargin
848 if s >= sLimit {
849 goto emitRemainder
850 }
851 if debug {
852 fmt.Println("non-dict matching at", s, "repeat:", repeat)
853 }
854 cv = load64(src, s)
855 if debug {
856 fmt.Println("now", s, "->", sLimit, "out:", d, "left:", len(src)-s, "nextemit:", nextEmit, "dstLimit:", dstLimit, "s:", s)
857 }
858 for {
859 candidate := 0
860 for {
861 // Next src position to check
862 nextS := s + (s-nextEmit)>>6 + 4
863 if nextS > sLimit {
864 goto emitRemainder
865 }
866 hash0 := hash6(cv, tableBits)
867 hash1 := hash6(cv>>8, tableBits)
868 candidate = int(table[hash0])
869 candidate2 := int(table[hash1])
870 table[hash0] = uint32(s)
871 table[hash1] = uint32(s + 1)
872 hash2 := hash6(cv>>16, tableBits)
873
874 // Check repeat at offset checkRep.
875 const checkRep = 1
876 if repeat > 0 && uint32(cv>>(checkRep*8)) == load32(src, s-repeat+checkRep) {
877 base := s + checkRep
878 // Extend back
879 for i := base - repeat; base > nextEmit && i > 0 && src[i-1] == src[base-1]; {
880 i--
881 base--
882 }
883 d += emitLiteral(dst[d:], src[nextEmit:base])
884 if debug && nextEmit != base {
885 fmt.Println("emitted ", base-nextEmit, "literals")
886 }
887 // Extend forward
888 candidate := s - repeat + 4 + checkRep
889 s += 4 + checkRep
890 for s <= sLimit {
891 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
892 s += bits.TrailingZeros64(diff) >> 3
893 break
894 }
895 s += 8
896 candidate += 8
897 }
898 if debug {
899 // Validate match.
900 if s <= candidate {
901 panic("s <= candidate")
902 }
903 a := src[base:s]
904 b := src[base-repeat : base-repeat+(s-base)]
905 if !bytes.Equal(a, b) {
906 panic("mismatch")
907 }
908 }
909 if nextEmit > 0 {
910 // same as `add := emitCopy(dst[d:], repeat, s-base)` but skips storing offset.
911 d += emitRepeat(dst[d:], repeat, s-base)
912 } else {
913 // First match, cannot be repeat.
914 d += emitCopy(dst[d:], repeat, s-base)
915 }
916 if debug {
917 fmt.Println("emitted src repeat length", s-base, "offset:", repeat, "s:", s)
918 }
919 nextEmit = s
920 if s >= sLimit {
921 goto emitRemainder
922 }
923
924 cv = load64(src, s)
925 continue
926 }
927
928 if uint32(cv) == load32(src, candidate) {
929 break
930 }
931 candidate = int(table[hash2])
932 if uint32(cv>>8) == load32(src, candidate2) {
933 table[hash2] = uint32(s + 2)
934 candidate = candidate2
935 s++
936 break
937 }
938 table[hash2] = uint32(s + 2)
939 if uint32(cv>>16) == load32(src, candidate) {
940 s += 2
941 break
942 }
943
944 cv = load64(src, nextS)
945 s = nextS
946 }
947
948 // Extend backwards.
949 // The top bytes will be rechecked to get the full match.
950 for candidate > 0 && s > nextEmit && src[candidate-1] == src[s-1] {
951 candidate--
952 s--
953 }
954
955 // Bail if we exceed the maximum size.
956 if d+(s-nextEmit) > dstLimit {
957 return 0
958 }
959
960 // A 4-byte match has been found. We'll later see if more than 4 bytes
961 // match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
962 // them as literal bytes.
963
964 d += emitLiteral(dst[d:], src[nextEmit:s])
965 if debug && nextEmit != s {
966 fmt.Println("emitted ", s-nextEmit, "literals")
967 }
968 // Call emitCopy, and then see if another emitCopy could be our next
969 // move. Repeat until we find no match for the input immediately after
970 // what was consumed by the last emitCopy call.
971 //
972 // If we exit this loop normally then we need to call emitLiteral next,
973 // though we don't yet know how big the literal will be. We handle that
974 // by proceeding to the next iteration of the main loop. We also can
975 // exit this loop via goto if we get close to exhausting the input.
976 for {
977 // Invariant: we have a 4-byte match at s, and no need to emit any
978 // literal bytes prior to s.
979 base := s
980 repeat = base - candidate
981
982 // Extend the 4-byte match as long as possible.
983 s += 4
984 candidate += 4
985 for s <= len(src)-8 {
986 if diff := load64(src, s) ^ load64(src, candidate); diff != 0 {
987 s += bits.TrailingZeros64(diff) >> 3
988 break
989 }
990 s += 8
991 candidate += 8
992 }
993
994 d += emitCopy(dst[d:], repeat, s-base)
995 if debug {
996 // Validate match.
997 if s <= candidate {
998 panic("s <= candidate")
999 }
1000 a := src[base:s]
1001 b := src[base-repeat : base-repeat+(s-base)]
1002 if !bytes.Equal(a, b) {
1003 panic("mismatch")
1004 }
1005 }
1006 if debug {
1007 fmt.Println("emitted src copy, length", s-base, "offset:", repeat, "s:", s)
1008 }
1009 nextEmit = s
1010 if s >= sLimit {
1011 goto emitRemainder
1012 }
1013
1014 if d > dstLimit {
1015 // Do we have space for more, if not bail.
1016 return 0
1017 }
1018 // Check for an immediate match, otherwise start search at s+1
1019 x := load64(src, s-2)
1020 m2Hash := hash6(x, tableBits)
1021 currHash := hash6(x>>16, tableBits)
1022 candidate = int(table[currHash])
1023 table[m2Hash] = uint32(s - 2)
1024 table[currHash] = uint32(s)
1025 if debug && s == candidate {
1026 panic("s == candidate")
1027 }
1028 if uint32(x>>16) != load32(src, candidate) {
1029 cv = load64(src, s+1)
1030 s++
1031 break
1032 }
1033 }
1034 }
1035
1036emitRemainder:
1037 if nextEmit < len(src) {
1038 // Bail if we exceed the maximum size.
1039 if d+len(src)-nextEmit > dstLimit {
1040 return 0
1041 }
1042 d += emitLiteral(dst[d:], src[nextEmit:])
1043 if debug && nextEmit != s {
1044 fmt.Println("emitted ", len(src)-nextEmit, "literals")
1045 }
1046 }
1047 return d
1048}
generated by cgit v1.2.3 (git 2.49.0) at 2025-11-05 11:15:44 +0000