summaryrefslogtreecommitdiff
path: root/vendor/github.com/klauspost/compress/flate/deflate.go
blob: 6287951204e8d37b6cd7e183a2309d596cd92f34 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
// Copyright 2009 The Go Authors. All rights reserved.
// Copyright (c) 2015 Klaus Post
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package flate

import (
	"fmt"
	"io"
	"math"
)

const (
	NoCompression      = 0
	BestSpeed          = 1
	BestCompression    = 9
	DefaultCompression = -1

	// HuffmanOnly disables Lempel-Ziv match searching and only performs Huffman
	// entropy encoding. This mode is useful in compressing data that has
	// already been compressed with an LZ style algorithm (e.g. Snappy or LZ4)
	// that lacks an entropy encoder. Compression gains are achieved when
	// certain bytes in the input stream occur more frequently than others.
	//
	// Note that HuffmanOnly produces a compressed output that is
	// RFC 1951 compliant. That is, any valid DEFLATE decompressor will
	// continue to be able to decompress this output.
	HuffmanOnly         = -2
	ConstantCompression = HuffmanOnly // compatibility alias.

	logWindowSize    = 15
	windowSize       = 1 << logWindowSize
	windowMask       = windowSize - 1
	logMaxOffsetSize = 15  // Standard DEFLATE
	minMatchLength   = 4   // The smallest match that the compressor looks for
	maxMatchLength   = 258 // The longest match for the compressor
	minOffsetSize    = 1   // The shortest offset that makes any sense

	// The maximum number of tokens we put into a single flat block, just too
	// stop things from getting too large.
	maxFlateBlockTokens = 1 << 14
	maxStoreBlockSize   = 65535
	hashBits            = 17 // After 17 performance degrades
	hashSize            = 1 << hashBits
	hashMask            = (1 << hashBits) - 1
	hashShift           = (hashBits + minMatchLength - 1) / minMatchLength
	maxHashOffset       = 1 << 24

	skipNever = math.MaxInt32
)

var useSSE42 bool

type compressionLevel struct {
	good, lazy, nice, chain, fastSkipHashing, level int
}

// Compression levels have been rebalanced from zlib deflate defaults
// to give a bigger spread in speed and compression.
// See https://blog.klauspost.com/rebalancing-deflate-compression-levels/
var levels = []compressionLevel{
	{}, // 0
	// Level 1-4 uses specialized algorithm - values not used
	{0, 0, 0, 0, 0, 1},
	{0, 0, 0, 0, 0, 2},
	{0, 0, 0, 0, 0, 3},
	{0, 0, 0, 0, 0, 4},
	// For levels 5-6 we don't bother trying with lazy matches.
	// Lazy matching is at least 30% slower, with 1.5% increase.
	{6, 0, 12, 8, 12, 5},
	{8, 0, 24, 16, 16, 6},
	// Levels 7-9 use increasingly more lazy matching
	// and increasingly stringent conditions for "good enough".
	{8, 8, 24, 16, skipNever, 7},
	{10, 16, 24, 64, skipNever, 8},
	{32, 258, 258, 4096, skipNever, 9},
}

// advancedState contains state for the advanced levels, with bigger hash tables, etc.
type advancedState struct {
	// deflate state
	length         int
	offset         int
	hash           uint32
	maxInsertIndex int
	ii             uint16 // position of last match, intended to overflow to reset.

	// Input hash chains
	// hashHead[hashValue] contains the largest inputIndex with the specified hash value
	// If hashHead[hashValue] is within the current window, then
	// hashPrev[hashHead[hashValue] & windowMask] contains the previous index
	// with the same hash value.
	chainHead  int
	hashHead   [hashSize]uint32
	hashPrev   [windowSize]uint32
	hashOffset int

	// input window: unprocessed data is window[index:windowEnd]
	index      int
	bulkHasher func([]byte, []uint32)
	hashMatch  [maxMatchLength + minMatchLength]uint32
}

type compressor struct {
	compressionLevel

	w *huffmanBitWriter

	// compression algorithm
	fill func(*compressor, []byte) int // copy data to window
	step func(*compressor)             // process window
	sync bool                          // requesting flush

	window        []byte
	windowEnd     int
	blockStart    int  // window index where current tokens start
	byteAvailable bool // if true, still need to process window[index-1].
	err           error

	// queued output tokens
	tokens tokens
	snap   fastEnc
	state  *advancedState
}

func (d *compressor) fillDeflate(b []byte) int {
	s := d.state
	if s.index >= 2*windowSize-(minMatchLength+maxMatchLength) {
		// shift the window by windowSize
		copy(d.window[:], d.window[windowSize:2*windowSize])
		s.index -= windowSize
		d.windowEnd -= windowSize
		if d.blockStart >= windowSize {
			d.blockStart -= windowSize
		} else {
			d.blockStart = math.MaxInt32
		}
		s.hashOffset += windowSize
		if s.hashOffset > maxHashOffset {
			delta := s.hashOffset - 1
			s.hashOffset -= delta
			s.chainHead -= delta
			// Iterate over slices instead of arrays to avoid copying
			// the entire table onto the stack (Issue #18625).
			for i, v := range s.hashPrev[:] {
				if int(v) > delta {
					s.hashPrev[i] = uint32(int(v) - delta)
				} else {
					s.hashPrev[i] = 0
				}
			}
			for i, v := range s.hashHead[:] {
				if int(v) > delta {
					s.hashHead[i] = uint32(int(v) - delta)
				} else {
					s.hashHead[i] = 0
				}
			}
		}
	}
	n := copy(d.window[d.windowEnd:], b)
	d.windowEnd += n
	return n
}

func (d *compressor) writeBlock(tok tokens, index int, eof bool) error {
	if index > 0 || eof {
		var window []byte
		if d.blockStart <= index {
			window = d.window[d.blockStart:index]
		}
		d.blockStart = index
		d.w.writeBlock(tok.tokens[:tok.n], eof, window)
		return d.w.err
	}
	return nil
}

// writeBlockSkip writes the current block and uses the number of tokens
// to determine if the block should be stored on no matches, or
// only huffman encoded.
func (d *compressor) writeBlockSkip(tok tokens, index int, eof bool) error {
	if index > 0 || eof {
		if d.blockStart <= index {
			window := d.window[d.blockStart:index]
			// If we removed less than a 64th of all literals
			// we huffman compress the block.
			if int(tok.n) > len(window)-int(tok.n>>6) {
				d.w.writeBlockHuff(eof, window)
			} else {
				// Write a dynamic huffman block.
				d.w.writeBlockDynamic(tok.tokens[:tok.n], eof, window)
			}
		} else {
			d.w.writeBlock(tok.tokens[:tok.n], eof, nil)
		}
		d.blockStart = index
		return d.w.err
	}
	return nil
}

// fillWindow will fill the current window with the supplied
// dictionary and calculate all hashes.
// This is much faster than doing a full encode.
// Should only be used after a start/reset.
func (d *compressor) fillWindow(b []byte) {
	// Do not fill window if we are in store-only mode,
	// use constant or Snappy compression.
	switch d.compressionLevel.level {
	case 0, 1, 2:
		return
	}
	s := d.state
	// If we are given too much, cut it.
	if len(b) > windowSize {
		b = b[len(b)-windowSize:]
	}
	// Add all to window.
	n := copy(d.window[d.windowEnd:], b)

	// Calculate 256 hashes at the time (more L1 cache hits)
	loops := (n + 256 - minMatchLength) / 256
	for j := 0; j < loops; j++ {
		startindex := j * 256
		end := startindex + 256 + minMatchLength - 1
		if end > n {
			end = n
		}
		tocheck := d.window[startindex:end]
		dstSize := len(tocheck) - minMatchLength + 1

		if dstSize <= 0 {
			continue
		}

		dst := s.hashMatch[:dstSize]
		s.bulkHasher(tocheck, dst)
		var newH uint32
		for i, val := range dst {
			di := i + startindex
			newH = val & hashMask
			// Get previous value with the same hash.
			// Our chain should point to the previous value.
			s.hashPrev[di&windowMask] = s.hashHead[newH]
			// Set the head of the hash chain to us.
			s.hashHead[newH] = uint32(di + s.hashOffset)
		}
		s.hash = newH
	}
	// Update window information.
	d.windowEnd += n
	s.index = n
}

// Try to find a match starting at index whose length is greater than prevSize.
// We only look at chainCount possibilities before giving up.
// pos = s.index, prevHead = s.chainHead-s.hashOffset, prevLength=minMatchLength-1, lookahead
func (d *compressor) findMatch(pos int, prevHead int, prevLength int, lookahead int) (length, offset int, ok bool) {
	minMatchLook := maxMatchLength
	if lookahead < minMatchLook {
		minMatchLook = lookahead
	}

	win := d.window[0 : pos+minMatchLook]

	// We quit when we get a match that's at least nice long
	nice := len(win) - pos
	if d.nice < nice {
		nice = d.nice
	}

	// If we've got a match that's good enough, only look in 1/4 the chain.
	tries := d.chain
	length = prevLength
	if length >= d.good {
		tries >>= 2
	}

	wEnd := win[pos+length]
	wPos := win[pos:]
	minIndex := pos - windowSize

	for i := prevHead; tries > 0; tries-- {
		if wEnd == win[i+length] {
			n := matchLen(win[i:], wPos, minMatchLook)

			if n > length && (n > minMatchLength || pos-i <= 4096) {
				length = n
				offset = pos - i
				ok = true
				if n >= nice {
					// The match is good enough that we don't try to find a better one.
					break
				}
				wEnd = win[pos+n]
			}
		}
		if i == minIndex {
			// hashPrev[i & windowMask] has already been overwritten, so stop now.
			break
		}
		i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset
		if i < minIndex || i < 0 {
			break
		}
	}
	return
}

// Try to find a match starting at index whose length is greater than prevSize.
// We only look at chainCount possibilities before giving up.
// pos = s.index, prevHead = s.chainHead-s.hashOffset, prevLength=minMatchLength-1, lookahead
func (d *compressor) findMatchSSE(pos int, prevHead int, prevLength int, lookahead int) (length, offset int, ok bool) {
	minMatchLook := maxMatchLength
	if lookahead < minMatchLook {
		minMatchLook = lookahead
	}

	win := d.window[0 : pos+minMatchLook]

	// We quit when we get a match that's at least nice long
	nice := len(win) - pos
	if d.nice < nice {
		nice = d.nice
	}

	// If we've got a match that's good enough, only look in 1/4 the chain.
	tries := d.chain
	length = prevLength
	if length >= d.good {
		tries >>= 2
	}

	wEnd := win[pos+length]
	wPos := win[pos:]
	minIndex := pos - windowSize

	for i := prevHead; tries > 0; tries-- {
		if wEnd == win[i+length] {
			n := matchLenSSE4(win[i:], wPos, minMatchLook)

			if n > length && (n > minMatchLength || pos-i <= 4096) {
				length = n
				offset = pos - i
				ok = true
				if n >= nice {
					// The match is good enough that we don't try to find a better one.
					break
				}
				wEnd = win[pos+n]
			}
		}
		if i == minIndex {
			// hashPrev[i & windowMask] has already been overwritten, so stop now.
			break
		}
		i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset
		if i < minIndex || i < 0 {
			break
		}
	}
	return
}

func (d *compressor) writeStoredBlock(buf []byte) error {
	if d.w.writeStoredHeader(len(buf), false); d.w.err != nil {
		return d.w.err
	}
	d.w.writeBytes(buf)
	return d.w.err
}

const hashmul = 0x1e35a7bd

// hash4 returns a hash representation of the first 4 bytes
// of the supplied slice.
// The caller must ensure that len(b) >= 4.
func hash4(b []byte) uint32 {
	return ((uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24) * hashmul) >> (32 - hashBits)
}

// bulkHash4 will compute hashes using the same
// algorithm as hash4
func bulkHash4(b []byte, dst []uint32) {
	if len(b) < minMatchLength {
		return
	}
	hb := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24
	dst[0] = (hb * hashmul) >> (32 - hashBits)
	end := len(b) - minMatchLength + 1
	for i := 1; i < end; i++ {
		hb = (hb << 8) | uint32(b[i+3])
		dst[i] = (hb * hashmul) >> (32 - hashBits)
	}
}

// matchLen returns the number of matching bytes in a and b
// up to length 'max'. Both slices must be at least 'max'
// bytes in size.
func matchLen(a, b []byte, max int) int {
	a = a[:max]
	b = b[:len(a)]
	for i, av := range a {
		if b[i] != av {
			return i
		}
	}
	return max
}

func (d *compressor) initDeflate() {
	d.window = make([]byte, 2*windowSize)
	d.byteAvailable = false
	d.err = nil
	if d.state == nil {
		return
	}
	s := d.state
	s.index = 0
	s.hashOffset = 1
	s.length = minMatchLength - 1
	s.offset = 0
	s.hash = 0
	s.chainHead = -1
	s.bulkHasher = bulkHash4
	if useSSE42 {
		s.bulkHasher = crc32sseAll
	}
}

// Assumes that d.fastSkipHashing != skipNever,
// otherwise use deflateLazy
func (d *compressor) deflate() {
	s := d.state
	// Sanity enables additional runtime tests.
	// It's intended to be used during development
	// to supplement the currently ad-hoc unit tests.
	const sanity = false

	if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync {
		return
	}

	s.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
	if s.index < s.maxInsertIndex {
		s.hash = hash4(d.window[s.index : s.index+minMatchLength])
	}

	for {
		if sanity && s.index > d.windowEnd {
			panic("index > windowEnd")
		}
		lookahead := d.windowEnd - s.index
		if lookahead < minMatchLength+maxMatchLength {
			if !d.sync {
				return
			}
			if sanity && s.index > d.windowEnd {
				panic("index > windowEnd")
			}
			if lookahead == 0 {
				if d.tokens.n > 0 {
					if d.err = d.writeBlockSkip(d.tokens, s.index, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
				return
			}
		}
		if s.index < s.maxInsertIndex {
			// Update the hash
			s.hash = hash4(d.window[s.index : s.index+minMatchLength])
			ch := s.hashHead[s.hash&hashMask]
			s.chainHead = int(ch)
			s.hashPrev[s.index&windowMask] = ch
			s.hashHead[s.hash&hashMask] = uint32(s.index + s.hashOffset)
		}
		s.length = minMatchLength - 1
		s.offset = 0
		minIndex := s.index - windowSize
		if minIndex < 0 {
			minIndex = 0
		}

		if s.chainHead-s.hashOffset >= minIndex && lookahead > minMatchLength-1 {
			if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok {
				s.length = newLength
				s.offset = newOffset
			}
		}
		if s.length >= minMatchLength {
			s.ii = 0
			// There was a match at the previous step, and the current match is
			// not better. Output the previous match.
			// "s.length-3" should NOT be "s.length-minMatchLength", since the format always assume 3
			d.tokens.tokens[d.tokens.n] = matchToken(uint32(s.length-3), uint32(s.offset-minOffsetSize))
			d.tokens.n++
			// Insert in the hash table all strings up to the end of the match.
			// index and index-1 are already inserted. If there is not enough
			// lookahead, the last two strings are not inserted into the hash
			// table.
			if s.length <= d.fastSkipHashing {
				var newIndex int
				newIndex = s.index + s.length
				// Calculate missing hashes
				end := newIndex
				if end > s.maxInsertIndex {
					end = s.maxInsertIndex
				}
				end += minMatchLength - 1
				startindex := s.index + 1
				if startindex > s.maxInsertIndex {
					startindex = s.maxInsertIndex
				}
				tocheck := d.window[startindex:end]
				dstSize := len(tocheck) - minMatchLength + 1
				if dstSize > 0 {
					dst := s.hashMatch[:dstSize]
					bulkHash4(tocheck, dst)
					var newH uint32
					for i, val := range dst {
						di := i + startindex
						newH = val & hashMask
						// Get previous value with the same hash.
						// Our chain should point to the previous value.
						s.hashPrev[di&windowMask] = s.hashHead[newH]
						// Set the head of the hash chain to us.
						s.hashHead[newH] = uint32(di + s.hashOffset)
					}
					s.hash = newH
				}
				s.index = newIndex
			} else {
				// For matches this long, we don't bother inserting each individual
				// item into the table.
				s.index += s.length
				if s.index < s.maxInsertIndex {
					s.hash = hash4(d.window[s.index : s.index+minMatchLength])
				}
			}
			if d.tokens.n == maxFlateBlockTokens {
				// The block includes the current character
				if d.err = d.writeBlockSkip(d.tokens, s.index, false); d.err != nil {
					return
				}
				d.tokens.n = 0
			}
		} else {
			s.ii++
			end := s.index + int(s.ii>>uint(d.fastSkipHashing)) + 1
			if end > d.windowEnd {
				end = d.windowEnd
			}
			for i := s.index; i < end; i++ {
				d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[i]))
				d.tokens.n++
				if d.tokens.n == maxFlateBlockTokens {
					if d.err = d.writeBlockSkip(d.tokens, i+1, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
			}
			s.index = end
		}
	}
}

// deflateLazy is the same as deflate, but with d.fastSkipHashing == skipNever,
// meaning it always has lazy matching on.
func (d *compressor) deflateLazy() {
	s := d.state
	// Sanity enables additional runtime tests.
	// It's intended to be used during development
	// to supplement the currently ad-hoc unit tests.
	const sanity = false

	if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync {
		return
	}

	s.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
	if s.index < s.maxInsertIndex {
		s.hash = hash4(d.window[s.index : s.index+minMatchLength])
	}

	for {
		if sanity && s.index > d.windowEnd {
			panic("index > windowEnd")
		}
		lookahead := d.windowEnd - s.index
		if lookahead < minMatchLength+maxMatchLength {
			if !d.sync {
				return
			}
			if sanity && s.index > d.windowEnd {
				panic("index > windowEnd")
			}
			if lookahead == 0 {
				// Flush current output block if any.
				if d.byteAvailable {
					// There is still one pending token that needs to be flushed
					d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
					d.tokens.n++
					d.byteAvailable = false
				}
				if d.tokens.n > 0 {
					if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
				return
			}
		}
		if s.index < s.maxInsertIndex {
			// Update the hash
			s.hash = hash4(d.window[s.index : s.index+minMatchLength])
			ch := s.hashHead[s.hash&hashMask]
			s.chainHead = int(ch)
			s.hashPrev[s.index&windowMask] = ch
			s.hashHead[s.hash&hashMask] = uint32(s.index + s.hashOffset)
		}
		prevLength := s.length
		prevOffset := s.offset
		s.length = minMatchLength - 1
		s.offset = 0
		minIndex := s.index - windowSize
		if minIndex < 0 {
			minIndex = 0
		}

		if s.chainHead-s.hashOffset >= minIndex && lookahead > prevLength && prevLength < d.lazy {
			if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok {
				s.length = newLength
				s.offset = newOffset
			}
		}
		if prevLength >= minMatchLength && s.length <= prevLength {
			// There was a match at the previous step, and the current match is
			// not better. Output the previous match.
			d.tokens.tokens[d.tokens.n] = matchToken(uint32(prevLength-3), uint32(prevOffset-minOffsetSize))
			d.tokens.n++

			// Insert in the hash table all strings up to the end of the match.
			// index and index-1 are already inserted. If there is not enough
			// lookahead, the last two strings are not inserted into the hash
			// table.
			var newIndex int
			newIndex = s.index + prevLength - 1
			// Calculate missing hashes
			end := newIndex
			if end > s.maxInsertIndex {
				end = s.maxInsertIndex
			}
			end += minMatchLength - 1
			startindex := s.index + 1
			if startindex > s.maxInsertIndex {
				startindex = s.maxInsertIndex
			}
			tocheck := d.window[startindex:end]
			dstSize := len(tocheck) - minMatchLength + 1
			if dstSize > 0 {
				dst := s.hashMatch[:dstSize]
				bulkHash4(tocheck, dst)
				var newH uint32
				for i, val := range dst {
					di := i + startindex
					newH = val & hashMask
					// Get previous value with the same hash.
					// Our chain should point to the previous value.
					s.hashPrev[di&windowMask] = s.hashHead[newH]
					// Set the head of the hash chain to us.
					s.hashHead[newH] = uint32(di + s.hashOffset)
				}
				s.hash = newH
			}

			s.index = newIndex
			d.byteAvailable = false
			s.length = minMatchLength - 1
			if d.tokens.n == maxFlateBlockTokens {
				// The block includes the current character
				if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
					return
				}
				d.tokens.n = 0
			}
		} else {
			// Reset, if we got a match this run.
			if s.length >= minMatchLength {
				s.ii = 0
			}
			// We have a byte waiting. Emit it.
			if d.byteAvailable {
				s.ii++
				d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
				d.tokens.n++
				if d.tokens.n == maxFlateBlockTokens {
					if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
				s.index++

				// If we have a long run of no matches, skip additional bytes
				// Resets when s.ii overflows after 64KB.
				if s.ii > 31 {
					n := int(s.ii >> 5)
					for j := 0; j < n; j++ {
						if s.index >= d.windowEnd-1 {
							break
						}

						d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
						d.tokens.n++
						if d.tokens.n == maxFlateBlockTokens {
							if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
								return
							}
							d.tokens.n = 0
						}
						s.index++
					}
					// Flush last byte
					d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
					d.tokens.n++
					d.byteAvailable = false
					// s.length = minMatchLength - 1 // not needed, since s.ii is reset above, so it should never be > minMatchLength
					if d.tokens.n == maxFlateBlockTokens {
						if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
							return
						}
						d.tokens.n = 0
					}
				}
			} else {
				s.index++
				d.byteAvailable = true
			}
		}
	}
}

// Assumes that d.fastSkipHashing != skipNever,
// otherwise use deflateLazySSE
func (d *compressor) deflateSSE() {
	s := d.state
	// Sanity enables additional runtime tests.
	// It's intended to be used during development
	// to supplement the currently ad-hoc unit tests.
	const sanity = false

	if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync {
		return
	}

	s.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
	if s.index < s.maxInsertIndex {
		s.hash = crc32sse(d.window[s.index:s.index+minMatchLength]) & hashMask
	}

	for {
		if sanity && s.index > d.windowEnd {
			panic("index > windowEnd")
		}
		lookahead := d.windowEnd - s.index
		if lookahead < minMatchLength+maxMatchLength {
			if !d.sync {
				return
			}
			if sanity && s.index > d.windowEnd {
				panic("index > windowEnd")
			}
			if lookahead == 0 {
				if d.tokens.n > 0 {
					if d.err = d.writeBlockSkip(d.tokens, s.index, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
				return
			}
		}
		if s.index < s.maxInsertIndex {
			// Update the hash
			s.hash = crc32sse(d.window[s.index:s.index+minMatchLength]) & hashMask
			ch := s.hashHead[s.hash]
			s.chainHead = int(ch)
			s.hashPrev[s.index&windowMask] = ch
			s.hashHead[s.hash] = uint32(s.index + s.hashOffset)
		}
		s.length = minMatchLength - 1
		s.offset = 0
		minIndex := s.index - windowSize
		if minIndex < 0 {
			minIndex = 0
		}

		if s.chainHead-s.hashOffset >= minIndex && lookahead > minMatchLength-1 {
			if newLength, newOffset, ok := d.findMatchSSE(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok {
				s.length = newLength
				s.offset = newOffset
			}
		}
		if s.length >= minMatchLength {
			s.ii = 0
			// There was a match at the previous step, and the current match is
			// not better. Output the previous match.
			// "s.length-3" should NOT be "s.length-minMatchLength", since the format always assume 3
			d.tokens.tokens[d.tokens.n] = matchToken(uint32(s.length-3), uint32(s.offset-minOffsetSize))
			d.tokens.n++
			// Insert in the hash table all strings up to the end of the match.
			// index and index-1 are already inserted. If there is not enough
			// lookahead, the last two strings are not inserted into the hash
			// table.
			if s.length <= d.fastSkipHashing {
				var newIndex int
				newIndex = s.index + s.length
				// Calculate missing hashes
				end := newIndex
				if end > s.maxInsertIndex {
					end = s.maxInsertIndex
				}
				end += minMatchLength - 1
				startindex := s.index + 1
				if startindex > s.maxInsertIndex {
					startindex = s.maxInsertIndex
				}
				tocheck := d.window[startindex:end]
				dstSize := len(tocheck) - minMatchLength + 1
				if dstSize > 0 {
					dst := s.hashMatch[:dstSize]

					crc32sseAll(tocheck, dst)
					var newH uint32
					for i, val := range dst {
						di := i + startindex
						newH = val & hashMask
						// Get previous value with the same hash.
						// Our chain should point to the previous value.
						s.hashPrev[di&windowMask] = s.hashHead[newH]
						// Set the head of the hash chain to us.
						s.hashHead[newH] = uint32(di + s.hashOffset)
					}
					s.hash = newH
				}
				s.index = newIndex
			} else {
				// For matches this long, we don't bother inserting each individual
				// item into the table.
				s.index += s.length
				if s.index < s.maxInsertIndex {
					s.hash = crc32sse(d.window[s.index:s.index+minMatchLength]) & hashMask
				}
			}
			if d.tokens.n == maxFlateBlockTokens {
				// The block includes the current character
				if d.err = d.writeBlockSkip(d.tokens, s.index, false); d.err != nil {
					return
				}
				d.tokens.n = 0
			}
		} else {
			s.ii++
			end := s.index + int(s.ii>>5) + 1
			if end > d.windowEnd {
				end = d.windowEnd
			}
			for i := s.index; i < end; i++ {
				d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[i]))
				d.tokens.n++
				if d.tokens.n == maxFlateBlockTokens {
					if d.err = d.writeBlockSkip(d.tokens, i+1, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
			}
			s.index = end
		}
	}
}

// deflateLazy is the same as deflate, but with d.fastSkipHashing == skipNever,
// meaning it always has lazy matching on.
func (d *compressor) deflateLazySSE() {
	s := d.state
	// Sanity enables additional runtime tests.
	// It's intended to be used during development
	// to supplement the currently ad-hoc unit tests.
	const sanity = false

	if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync {
		return
	}

	s.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
	if s.index < s.maxInsertIndex {
		s.hash = crc32sse(d.window[s.index:s.index+minMatchLength]) & hashMask
	}

	for {
		if sanity && s.index > d.windowEnd {
			panic("index > windowEnd")
		}
		lookahead := d.windowEnd - s.index
		if lookahead < minMatchLength+maxMatchLength {
			if !d.sync {
				return
			}
			if sanity && s.index > d.windowEnd {
				panic("index > windowEnd")
			}
			if lookahead == 0 {
				// Flush current output block if any.
				if d.byteAvailable {
					// There is still one pending token that needs to be flushed
					d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
					d.tokens.n++
					d.byteAvailable = false
				}
				if d.tokens.n > 0 {
					if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
				return
			}
		}
		if s.index < s.maxInsertIndex {
			// Update the hash
			s.hash = crc32sse(d.window[s.index:s.index+minMatchLength]) & hashMask
			ch := s.hashHead[s.hash]
			s.chainHead = int(ch)
			s.hashPrev[s.index&windowMask] = ch
			s.hashHead[s.hash] = uint32(s.index + s.hashOffset)
		}
		prevLength := s.length
		prevOffset := s.offset
		s.length = minMatchLength - 1
		s.offset = 0
		minIndex := s.index - windowSize
		if minIndex < 0 {
			minIndex = 0
		}

		if s.chainHead-s.hashOffset >= minIndex && lookahead > prevLength && prevLength < d.lazy {
			if newLength, newOffset, ok := d.findMatchSSE(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok {
				s.length = newLength
				s.offset = newOffset
			}
		}
		if prevLength >= minMatchLength && s.length <= prevLength {
			// There was a match at the previous step, and the current match is
			// not better. Output the previous match.
			d.tokens.tokens[d.tokens.n] = matchToken(uint32(prevLength-3), uint32(prevOffset-minOffsetSize))
			d.tokens.n++

			// Insert in the hash table all strings up to the end of the match.
			// index and index-1 are already inserted. If there is not enough
			// lookahead, the last two strings are not inserted into the hash
			// table.
			var newIndex int
			newIndex = s.index + prevLength - 1
			// Calculate missing hashes
			end := newIndex
			if end > s.maxInsertIndex {
				end = s.maxInsertIndex
			}
			end += minMatchLength - 1
			startindex := s.index + 1
			if startindex > s.maxInsertIndex {
				startindex = s.maxInsertIndex
			}
			tocheck := d.window[startindex:end]
			dstSize := len(tocheck) - minMatchLength + 1
			if dstSize > 0 {
				dst := s.hashMatch[:dstSize]
				crc32sseAll(tocheck, dst)
				var newH uint32
				for i, val := range dst {
					di := i + startindex
					newH = val & hashMask
					// Get previous value with the same hash.
					// Our chain should point to the previous value.
					s.hashPrev[di&windowMask] = s.hashHead[newH]
					// Set the head of the hash chain to us.
					s.hashHead[newH] = uint32(di + s.hashOffset)
				}
				s.hash = newH
			}

			s.index = newIndex
			d.byteAvailable = false
			s.length = minMatchLength - 1
			if d.tokens.n == maxFlateBlockTokens {
				// The block includes the current character
				if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
					return
				}
				d.tokens.n = 0
			}
		} else {
			// Reset, if we got a match this run.
			if s.length >= minMatchLength {
				s.ii = 0
			}
			// We have a byte waiting. Emit it.
			if d.byteAvailable {
				s.ii++
				d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
				d.tokens.n++
				if d.tokens.n == maxFlateBlockTokens {
					if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
						return
					}
					d.tokens.n = 0
				}
				s.index++

				// If we have a long run of no matches, skip additional bytes
				// Resets when s.ii overflows after 64KB.
				if s.ii > 31 {
					n := int(s.ii >> 6)
					for j := 0; j < n; j++ {
						if s.index >= d.windowEnd-1 {
							break
						}

						d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
						d.tokens.n++
						if d.tokens.n == maxFlateBlockTokens {
							if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
								return
							}
							d.tokens.n = 0
						}
						s.index++
					}
					// Flush last byte
					d.tokens.tokens[d.tokens.n] = literalToken(uint32(d.window[s.index-1]))
					d.tokens.n++
					d.byteAvailable = false
					// s.length = minMatchLength - 1 // not needed, since s.ii is reset above, so it should never be > minMatchLength
					if d.tokens.n == maxFlateBlockTokens {
						if d.err = d.writeBlock(d.tokens, s.index, false); d.err != nil {
							return
						}
						d.tokens.n = 0
					}
				}
			} else {
				s.index++
				d.byteAvailable = true
			}
		}
	}
}

func (d *compressor) store() {
	if d.windowEnd > 0 && (d.windowEnd == maxStoreBlockSize || d.sync) {
		d.err = d.writeStoredBlock(d.window[:d.windowEnd])
		d.windowEnd = 0
	}
}

// fillWindow will fill the buffer with data for huffman-only compression.
// The number of bytes copied is returned.
func (d *compressor) fillBlock(b []byte) int {
	n := copy(d.window[d.windowEnd:], b)
	d.windowEnd += n
	return n
}

// storeHuff will compress and store the currently added data,
// if enough has been accumulated or we at the end of the stream.
// Any error that occurred will be in d.err
func (d *compressor) storeHuff() {
	if d.windowEnd < len(d.window) && !d.sync || d.windowEnd == 0 {
		return
	}
	d.w.writeBlockHuff(false, d.window[:d.windowEnd])
	d.err = d.w.err
	d.windowEnd = 0
}

// storeHuff will compress and store the currently added data,
// if enough has been accumulated or we at the end of the stream.
// Any error that occurred will be in d.err
func (d *compressor) storeSnappy() {
	// We only compress if we have maxStoreBlockSize.
	if d.windowEnd < maxStoreBlockSize {
		if !d.sync {
			return
		}
		// Handle extremely small sizes.
		if d.windowEnd < 128 {
			if d.windowEnd == 0 {
				return
			}
			if d.windowEnd <= 32 {
				d.err = d.writeStoredBlock(d.window[:d.windowEnd])
				d.tokens.n = 0
				d.windowEnd = 0
			} else {
				d.w.writeBlockHuff(false, d.window[:d.windowEnd])
				d.err = d.w.err
			}
			d.tokens.n = 0
			d.windowEnd = 0
			d.snap.Reset()
			return
		}
	}

	d.snap.Encode(&d.tokens, d.window[:d.windowEnd])
	// If we made zero matches, store the block as is.
	if int(d.tokens.n) == d.windowEnd {
		d.err = d.writeStoredBlock(d.window[:d.windowEnd])
		// If we removed less than 1/16th, huffman compress the block.
	} else if int(d.tokens.n) > d.windowEnd-(d.windowEnd>>4) {
		d.w.writeBlockHuff(false, d.window[:d.windowEnd])
		d.err = d.w.err
	} else {
		d.w.writeBlockDynamic(d.tokens.tokens[:d.tokens.n], false, d.window[:d.windowEnd])
		d.err = d.w.err
	}
	d.tokens.n = 0
	d.windowEnd = 0
}

// write will add input byte to the stream.
// Unless an error occurs all bytes will be consumed.
func (d *compressor) write(b []byte) (n int, err error) {
	if d.err != nil {
		return 0, d.err
	}
	n = len(b)
	for len(b) > 0 {
		d.step(d)
		b = b[d.fill(d, b):]
		if d.err != nil {
			return 0, d.err
		}
	}
	return n, d.err
}

func (d *compressor) syncFlush() error {
	d.sync = true
	if d.err != nil {
		return d.err
	}
	d.step(d)
	if d.err == nil {
		d.w.writeStoredHeader(0, false)
		d.w.flush()
		d.err = d.w.err
	}
	d.sync = false
	return d.err
}

func (d *compressor) init(w io.Writer, level int) (err error) {
	d.w = newHuffmanBitWriter(w)

	switch {
	case level == NoCompression:
		d.window = make([]byte, maxStoreBlockSize)
		d.fill = (*compressor).fillBlock
		d.step = (*compressor).store
	case level == ConstantCompression:
		d.window = make([]byte, maxStoreBlockSize)
		d.fill = (*compressor).fillBlock
		d.step = (*compressor).storeHuff
	case level >= 1 && level <= 4:
		d.snap = newFastEnc(level)
		d.window = make([]byte, maxStoreBlockSize)
		d.fill = (*compressor).fillBlock
		d.step = (*compressor).storeSnappy
	case level == DefaultCompression:
		level = 5
		fallthrough
	case 5 <= level && level <= 9:
		d.state = &advancedState{}
		d.compressionLevel = levels[level]
		d.initDeflate()
		d.fill = (*compressor).fillDeflate
		if d.fastSkipHashing == skipNever {
			if useSSE42 {
				d.step = (*compressor).deflateLazySSE
			} else {
				d.step = (*compressor).deflateLazy
			}
		} else {
			if useSSE42 {
				d.step = (*compressor).deflateSSE
			} else {
				d.step = (*compressor).deflate

			}
		}
	default:
		return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level)
	}
	return nil
}

// reset the state of the compressor.
func (d *compressor) reset(w io.Writer) {
	d.w.reset(w)
	d.sync = false
	d.err = nil
	// We only need to reset a few things for Snappy.
	if d.snap != nil {
		d.snap.Reset()
		d.windowEnd = 0
		d.tokens.n = 0
		return
	}
	switch d.compressionLevel.chain {
	case 0:
		// level was NoCompression or ConstantCompresssion.
		d.windowEnd = 0
	default:
		s := d.state
		s.chainHead = -1
		for i := range s.hashHead {
			s.hashHead[i] = 0
		}
		for i := range s.hashPrev {
			s.hashPrev[i] = 0
		}
		s.hashOffset = 1
		s.index, d.windowEnd = 0, 0
		d.blockStart, d.byteAvailable = 0, false
		d.tokens.n = 0
		s.length = minMatchLength - 1
		s.offset = 0
		s.hash = 0
		s.ii = 0
		s.maxInsertIndex = 0
	}
}

func (d *compressor) close() error {
	if d.err != nil {
		return d.err
	}
	d.sync = true
	d.step(d)
	if d.err != nil {
		return d.err
	}
	if d.w.writeStoredHeader(0, true); d.w.err != nil {
		return d.w.err
	}
	d.w.flush()
	return d.w.err
}

// NewWriter returns a new Writer compressing data at the given level.
// Following zlib, levels range from 1 (BestSpeed) to 9 (BestCompression);
// higher levels typically run slower but compress more.
// Level 0 (NoCompression) does not attempt any compression; it only adds the
// necessary DEFLATE framing.
// Level -1 (DefaultCompression) uses the default compression level.
// Level -2 (ConstantCompression) will use Huffman compression only, giving
// a very fast compression for all types of input, but sacrificing considerable
// compression efficiency.
//
// If level is in the range [-2, 9] then the error returned will be nil.
// Otherwise the error returned will be non-nil.
func NewWriter(w io.Writer, level int) (*Writer, error) {
	var dw Writer
	if err := dw.d.init(w, level); err != nil {
		return nil, err
	}
	return &dw, nil
}

// NewWriterDict is like NewWriter but initializes the new
// Writer with a preset dictionary.  The returned Writer behaves
// as if the dictionary had been written to it without producing
// any compressed output.  The compressed data written to w
// can only be decompressed by a Reader initialized with the
// same dictionary.
func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) {
	dw := &dictWriter{w}
	zw, err := NewWriter(dw, level)
	if err != nil {
		return nil, err
	}
	zw.d.fillWindow(dict)
	zw.dict = append(zw.dict, dict...) // duplicate dictionary for Reset method.
	return zw, err
}

type dictWriter struct {
	w io.Writer
}

func (w *dictWriter) Write(b []byte) (n int, err error) {
	return w.w.Write(b)
}

// A Writer takes data written to it and writes the compressed
// form of that data to an underlying writer (see NewWriter).
type Writer struct {
	d    compressor
	dict []byte
}

// Write writes data to w, which will eventually write the
// compressed form of data to its underlying writer.
func (w *Writer) Write(data []byte) (n int, err error) {
	return w.d.write(data)
}

// Flush flushes any pending data to the underlying writer.
// It is useful mainly in compressed network protocols, to ensure that
// a remote reader has enough data to reconstruct a packet.
// Flush does not return until the data has been written.
// Calling Flush when there is no pending data still causes the Writer
// to emit a sync marker of at least 4 bytes.
// If the underlying writer returns an error, Flush returns that error.
//
// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH.
func (w *Writer) Flush() error {
	// For more about flushing:
	// http://www.bolet.org/~pornin/deflate-flush.html
	return w.d.syncFlush()
}

// Close flushes and closes the writer.
func (w *Writer) Close() error {
	return w.d.close()
}

// Reset discards the writer's state and makes it equivalent to
// the result of NewWriter or NewWriterDict called with dst
// and w's level and dictionary.
func (w *Writer) Reset(dst io.Writer) {
	if dw, ok := w.d.w.writer.(*dictWriter); ok {
		// w was created with NewWriterDict
		dw.w = dst
		w.d.reset(dw)
		w.d.fillWindow(w.dict)
	} else {
		// w was created with NewWriter
		w.d.reset(dst)
	}
}

// ResetDict discards the writer's state and makes it equivalent to
// the result of NewWriter or NewWriterDict called with dst
// and w's level, but sets a specific dictionary.
func (w *Writer) ResetDict(dst io.Writer, dict []byte) {
	w.dict = dict
	w.d.reset(dst)
	w.d.fillWindow(w.dict)
}