aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/zstd/lib/compress/zstd_fast.c
blob: 5f2c6a2edad5225423fa2699b6b728d9782e2b8d (plain) (blame)
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
/*
 * Copyright (c) Meta Platforms, Inc. and affiliates.
 * All rights reserved.
 *
 * This source code is licensed under both the BSD-style license (found in the
 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
 * in the COPYING file in the root directory of this source tree).
 * You may select, at your option, one of the above-listed licenses.
 */

#include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
#include "zstd_fast.h"

static void ZSTD_fillHashTableForCDict(ZSTD_matchState_t* ms,
                        const void* const end,
                        ZSTD_dictTableLoadMethod_e dtlm)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32  const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
    U32  const mls = cParams->minMatch;
    const BYTE* const base = ms->window.base;
    const BYTE* ip = base + ms->nextToUpdate;
    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
    const U32 fastHashFillStep = 3;

    /* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
     * Feel free to remove this assert if there's a good reason! */
    assert(dtlm == ZSTD_dtlm_full);

    /* Always insert every fastHashFillStep position into the hash table.
     * Insert the other positions if their hash entry is empty.
     */
    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
        U32 const curr = (U32)(ip - base);
        {   size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
            ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr);   }

        if (dtlm == ZSTD_dtlm_fast) continue;
        /* Only load extra positions for ZSTD_dtlm_full */
        {   U32 p;
            for (p = 1; p < fastHashFillStep; ++p) {
                size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
                if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {  /* not yet filled */
                    ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
                }   }   }   }
}

static void ZSTD_fillHashTableForCCtx(ZSTD_matchState_t* ms,
                        const void* const end,
                        ZSTD_dictTableLoadMethod_e dtlm)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32  const hBits = cParams->hashLog;
    U32  const mls = cParams->minMatch;
    const BYTE* const base = ms->window.base;
    const BYTE* ip = base + ms->nextToUpdate;
    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
    const U32 fastHashFillStep = 3;

    /* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
     * Feel free to remove this assert if there's a good reason! */
    assert(dtlm == ZSTD_dtlm_fast);

    /* Always insert every fastHashFillStep position into the hash table.
     * Insert the other positions if their hash entry is empty.
     */
    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
        U32 const curr = (U32)(ip - base);
        size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
        hashTable[hash0] = curr;
        if (dtlm == ZSTD_dtlm_fast) continue;
        /* Only load extra positions for ZSTD_dtlm_full */
        {   U32 p;
            for (p = 1; p < fastHashFillStep; ++p) {
                size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
                if (hashTable[hash] == 0) {  /* not yet filled */
                    hashTable[hash] = curr + p;
    }   }   }   }
}

void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
                        const void* const end,
                        ZSTD_dictTableLoadMethod_e dtlm,
                        ZSTD_tableFillPurpose_e tfp)
{
    if (tfp == ZSTD_tfp_forCDict) {
        ZSTD_fillHashTableForCDict(ms, end, dtlm);
    } else {
        ZSTD_fillHashTableForCCtx(ms, end, dtlm);
    }
}


/**
 * If you squint hard enough (and ignore repcodes), the search operation at any
 * given position is broken into 4 stages:
 *
 * 1. Hash   (map position to hash value via input read)
 * 2. Lookup (map hash val to index via hashtable read)
 * 3. Load   (map index to value at that position via input read)
 * 4. Compare
 *
 * Each of these steps involves a memory read at an address which is computed
 * from the previous step. This means these steps must be sequenced and their
 * latencies are cumulative.
 *
 * Rather than do 1->2->3->4 sequentially for a single position before moving
 * onto the next, this implementation interleaves these operations across the
 * next few positions:
 *
 * R = Repcode Read & Compare
 * H = Hash
 * T = Table Lookup
 * M = Match Read & Compare
 *
 * Pos | Time -->
 * ----+-------------------
 * N   | ... M
 * N+1 | ...   TM
 * N+2 |    R H   T M
 * N+3 |         H    TM
 * N+4 |           R H   T M
 * N+5 |                H   ...
 * N+6 |                  R ...
 *
 * This is very much analogous to the pipelining of execution in a CPU. And just
 * like a CPU, we have to dump the pipeline when we find a match (i.e., take a
 * branch).
 *
 * When this happens, we throw away our current state, and do the following prep
 * to re-enter the loop:
 *
 * Pos | Time -->
 * ----+-------------------
 * N   | H T
 * N+1 |  H
 *
 * This is also the work we do at the beginning to enter the loop initially.
 */
FORCE_INLINE_TEMPLATE size_t
ZSTD_compressBlock_fast_noDict_generic(
        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize,
        U32 const mls, U32 const hasStep)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32 const hlog = cParams->hashLog;
    /* support stepSize of 0 */
    size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2;
    const BYTE* const base = ms->window.base;
    const BYTE* const istart = (const BYTE*)src;
    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
    const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
    const BYTE* const prefixStart = base + prefixStartIndex;
    const BYTE* const iend = istart + srcSize;
    const BYTE* const ilimit = iend - HASH_READ_SIZE;

    const BYTE* anchor = istart;
    const BYTE* ip0 = istart;
    const BYTE* ip1;
    const BYTE* ip2;
    const BYTE* ip3;
    U32 current0;

    U32 rep_offset1 = rep[0];
    U32 rep_offset2 = rep[1];
    U32 offsetSaved1 = 0, offsetSaved2 = 0;

    size_t hash0; /* hash for ip0 */
    size_t hash1; /* hash for ip1 */
    U32 idx; /* match idx for ip0 */
    U32 mval; /* src value at match idx */

    U32 offcode;
    const BYTE* match0;
    size_t mLength;

    /* ip0 and ip1 are always adjacent. The targetLength skipping and
     * uncompressibility acceleration is applied to every other position,
     * matching the behavior of #1562. step therefore represents the gap
     * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */
    size_t step;
    const BYTE* nextStep;
    const size_t kStepIncr = (1 << (kSearchStrength - 1));

    DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
    ip0 += (ip0 == prefixStart);
    {   U32 const curr = (U32)(ip0 - base);
        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
        U32 const maxRep = curr - windowLow;
        if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
        if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
    }

    /* start each op */
_start: /* Requires: ip0 */

    step = stepSize;
    nextStep = ip0 + kStepIncr;

    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
    ip1 = ip0 + 1;
    ip2 = ip0 + step;
    ip3 = ip2 + 1;

    if (ip3 >= ilimit) {
        goto _cleanup;
    }

    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
    hash1 = ZSTD_hashPtr(ip1, hlog, mls);

    idx = hashTable[hash0];

    do {
        /* load repcode match for ip[2]*/
        const U32 rval = MEM_read32(ip2 - rep_offset1);

        /* write back hash table entry */
        current0 = (U32)(ip0 - base);
        hashTable[hash0] = current0;

        /* check repcode at ip[2] */
        if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
            ip0 = ip2;
            match0 = ip0 - rep_offset1;
            mLength = ip0[-1] == match0[-1];
            ip0 -= mLength;
            match0 -= mLength;
            offcode = REPCODE1_TO_OFFBASE;
            mLength += 4;

            /* First write next hash table entry; we've already calculated it.
             * This write is known to be safe because the ip1 is before the
             * repcode (ip2). */
            hashTable[hash1] = (U32)(ip1 - base);

            goto _match;
        }

        /* load match for ip[0] */
        if (idx >= prefixStartIndex) {
            mval = MEM_read32(base + idx);
        } else {
            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
        }

        /* check match at ip[0] */
        if (MEM_read32(ip0) == mval) {
            /* found a match! */

            /* First write next hash table entry; we've already calculated it.
             * This write is known to be safe because the ip1 == ip0 + 1, so
             * we know we will resume searching after ip1 */
            hashTable[hash1] = (U32)(ip1 - base);

            goto _offset;
        }

        /* lookup ip[1] */
        idx = hashTable[hash1];

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip3;

        /* write back hash table entry */
        current0 = (U32)(ip0 - base);
        hashTable[hash0] = current0;

        /* load match for ip[0] */
        if (idx >= prefixStartIndex) {
            mval = MEM_read32(base + idx);
        } else {
            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
        }

        /* check match at ip[0] */
        if (MEM_read32(ip0) == mval) {
            /* found a match! */

            /* first write next hash table entry; we've already calculated it */
            if (step <= 4) {
                /* We need to avoid writing an index into the hash table >= the
                 * position at which we will pick up our searching after we've
                 * taken this match.
                 *
                 * The minimum possible match has length 4, so the earliest ip0
                 * can be after we take this match will be the current ip0 + 4.
                 * ip1 is ip0 + step - 1. If ip1 is >= ip0 + 4, we can't safely
                 * write this position.
                 */
                hashTable[hash1] = (U32)(ip1 - base);
            }

            goto _offset;
        }

        /* lookup ip[1] */
        idx = hashTable[hash1];

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip0 + step;
        ip3 = ip1 + step;

        /* calculate step */
        if (ip2 >= nextStep) {
            step++;
            PREFETCH_L1(ip1 + 64);
            PREFETCH_L1(ip1 + 128);
            nextStep += kStepIncr;
        }
    } while (ip3 < ilimit);

_cleanup:
    /* Note that there are probably still a couple positions we could search.
     * However, it seems to be a meaningful performance hit to try to search
     * them. So let's not. */

    /* When the repcodes are outside of the prefix, we set them to zero before the loop.
     * When the offsets are still zero, we need to restore them after the block to have a correct
     * repcode history. If only one offset was invalid, it is easy. The tricky case is when both
     * offsets were invalid. We need to figure out which offset to refill with.
     *     - If both offsets are zero they are in the same order.
     *     - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
     *     - If only one is zero, we need to decide which offset to restore.
     *         - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
     *         - It is impossible for rep_offset2 to be non-zero.
     *
     * So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
     * set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
     */
    offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;

    /* save reps for next block */
    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;

    /* Return the last literals size */
    return (size_t)(iend - anchor);

_offset: /* Requires: ip0, idx */

    /* Compute the offset code. */
    match0 = base + idx;
    rep_offset2 = rep_offset1;
    rep_offset1 = (U32)(ip0-match0);
    offcode = OFFSET_TO_OFFBASE(rep_offset1);
    mLength = 4;

    /* Count the backwards match length. */
    while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
        ip0--;
        match0--;
        mLength++;
    }

_match: /* Requires: ip0, match0, offcode */

    /* Count the forward length. */
    mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);

    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);

    ip0 += mLength;
    anchor = ip0;

    /* Fill table and check for immediate repcode. */
    if (ip0 <= ilimit) {
        /* Fill Table */
        assert(base+current0+2 > istart);  /* check base overflow */
        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);

        if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
            while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
                /* store sequence */
                size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
                { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
                ip0 += rLength;
                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
                anchor = ip0;
                continue;   /* faster when present (confirmed on gcc-8) ... (?) */
    }   }   }

    goto _start;
}

#define ZSTD_GEN_FAST_FN(dictMode, mls, step)                                                            \
    static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step(                                      \
            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \
            void const* src, size_t srcSize)                                                       \
    {                                                                                              \
        return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \
    }

ZSTD_GEN_FAST_FN(noDict, 4, 1)
ZSTD_GEN_FAST_FN(noDict, 5, 1)
ZSTD_GEN_FAST_FN(noDict, 6, 1)
ZSTD_GEN_FAST_FN(noDict, 7, 1)

ZSTD_GEN_FAST_FN(noDict, 4, 0)
ZSTD_GEN_FAST_FN(noDict, 5, 0)
ZSTD_GEN_FAST_FN(noDict, 6, 0)
ZSTD_GEN_FAST_FN(noDict, 7, 0)

size_t ZSTD_compressBlock_fast(
        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize)
{
    U32 const mls = ms->cParams.minMatch;
    assert(ms->dictMatchState == NULL);
    if (ms->cParams.targetLength > 1) {
        switch(mls)
        {
        default: /* includes case 3 */
        case 4 :
            return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);
        case 5 :
            return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);
        case 6 :
            return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);
        case 7 :
            return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);
        }
    } else {
        switch(mls)
        {
        default: /* includes case 3 */
        case 4 :
            return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);
        case 5 :
            return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);
        case 6 :
            return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);
        case 7 :
            return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);
        }

    }
}

FORCE_INLINE_TEMPLATE
size_t ZSTD_compressBlock_fast_dictMatchState_generic(
        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32 const hlog = cParams->hashLog;
    /* support stepSize of 0 */
    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
    const BYTE* const base = ms->window.base;
    const BYTE* const istart = (const BYTE*)src;
    const BYTE* ip0 = istart;
    const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
    const BYTE* anchor = istart;
    const U32   prefixStartIndex = ms->window.dictLimit;
    const BYTE* const prefixStart = base + prefixStartIndex;
    const BYTE* const iend = istart + srcSize;
    const BYTE* const ilimit = iend - HASH_READ_SIZE;
    U32 offset_1=rep[0], offset_2=rep[1];

    const ZSTD_matchState_t* const dms = ms->dictMatchState;
    const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
    const U32* const dictHashTable = dms->hashTable;
    const U32 dictStartIndex       = dms->window.dictLimit;
    const BYTE* const dictBase     = dms->window.base;
    const BYTE* const dictStart    = dictBase + dictStartIndex;
    const BYTE* const dictEnd      = dms->window.nextSrc;
    const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
    const U32 dictAndPrefixLength  = (U32)(istart - prefixStart + dictEnd - dictStart);
    const U32 dictHBits            = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;

    /* if a dictionary is still attached, it necessarily means that
     * it is within window size. So we just check it. */
    const U32 maxDistance = 1U << cParams->windowLog;
    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
    assert(endIndex - prefixStartIndex <= maxDistance);
    (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */

    (void)hasStep; /* not currently specialized on whether it's accelerated */

    /* ensure there will be no underflow
     * when translating a dict index into a local index */
    assert(prefixStartIndex >= (U32)(dictEnd - dictBase));

    if (ms->prefetchCDictTables) {
        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
        PREFETCH_AREA(dictHashTable, hashTableBytes)
    }

    /* init */
    DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
    ip0 += (dictAndPrefixLength == 0);
    /* dictMatchState repCode checks don't currently handle repCode == 0
     * disabling. */
    assert(offset_1 <= dictAndPrefixLength);
    assert(offset_2 <= dictAndPrefixLength);

    /* Outer search loop */
    assert(stepSize >= 1);
    while (ip1 <= ilimit) {   /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
        size_t mLength;
        size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);

        size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
        U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
        int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);

        U32 matchIndex = hashTable[hash0];
        U32 curr = (U32)(ip0 - base);
        size_t step = stepSize;
        const size_t kStepIncr = 1 << kSearchStrength;
        const BYTE* nextStep = ip0 + kStepIncr;

        /* Inner search loop */
        while (1) {
            const BYTE* match = base + matchIndex;
            const U32 repIndex = curr + 1 - offset_1;
            const BYTE* repMatch = (repIndex < prefixStartIndex) ?
                                   dictBase + (repIndex - dictIndexDelta) :
                                   base + repIndex;
            const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
            size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
            hashTable[hash0] = curr;   /* update hash table */

            if (((U32) ((prefixStartIndex - 1) - repIndex) >=
                 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
                && (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
                const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
                mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
                ip0++;
                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
                break;
            }

            if (dictTagsMatch) {
                /* Found a possible dict match */
                const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
                const BYTE* dictMatch = dictBase + dictMatchIndex;
                if (dictMatchIndex > dictStartIndex &&
                    MEM_read32(dictMatch) == MEM_read32(ip0)) {
                    /* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
                    if (matchIndex <= prefixStartIndex) {
                        U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
                        mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
                        while (((ip0 > anchor) & (dictMatch > dictStart))
                            && (ip0[-1] == dictMatch[-1])) {
                            ip0--;
                            dictMatch--;
                            mLength++;
                        } /* catch up */
                        offset_2 = offset_1;
                        offset_1 = offset;
                        ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
                        break;
                    }
                }
            }

            if (matchIndex > prefixStartIndex && MEM_read32(match) == MEM_read32(ip0)) {
                /* found a regular match */
                U32 const offset = (U32) (ip0 - match);
                mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
                while (((ip0 > anchor) & (match > prefixStart))
                       && (ip0[-1] == match[-1])) {
                    ip0--;
                    match--;
                    mLength++;
                } /* catch up */
                offset_2 = offset_1;
                offset_1 = offset;
                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
                break;
            }

            /* Prepare for next iteration */
            dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
            dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
            matchIndex = hashTable[hash1];

            if (ip1 >= nextStep) {
                step++;
                nextStep += kStepIncr;
            }
            ip0 = ip1;
            ip1 = ip1 + step;
            if (ip1 > ilimit) goto _cleanup;

            curr = (U32)(ip0 - base);
            hash0 = hash1;
        }   /* end inner search loop */

        /* match found */
        assert(mLength);
        ip0 += mLength;
        anchor = ip0;

        if (ip0 <= ilimit) {
            /* Fill Table */
            assert(base+curr+2 > istart);  /* check base overflow */
            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);

            /* check immediate repcode */
            while (ip0 <= ilimit) {
                U32 const current2 = (U32)(ip0-base);
                U32 const repIndex2 = current2 - offset_2;
                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
                        dictBase - dictIndexDelta + repIndex2 :
                        base + repIndex2;
                if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
                   && (MEM_read32(repMatch2) == MEM_read32(ip0))) {
                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
                    size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
                    ip0 += repLength2;
                    anchor = ip0;
                    continue;
                }
                break;
            }
        }

        /* Prepare for next iteration */
        assert(ip0 == anchor);
        ip1 = ip0 + stepSize;
    }

_cleanup:
    /* save reps for next block */
    rep[0] = offset_1;
    rep[1] = offset_2;

    /* Return the last literals size */
    return (size_t)(iend - anchor);
}


ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)

size_t ZSTD_compressBlock_fast_dictMatchState(
        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize)
{
    U32 const mls = ms->cParams.minMatch;
    assert(ms->dictMatchState != NULL);
    switch(mls)
    {
    default: /* includes case 3 */
    case 4 :
        return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);
    case 5 :
        return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);
    case 6 :
        return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);
    case 7 :
        return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);
    }
}


static size_t ZSTD_compressBlock_fast_extDict_generic(
        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32 const hlog = cParams->hashLog;
    /* support stepSize of 0 */
    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
    const BYTE* const base = ms->window.base;
    const BYTE* const dictBase = ms->window.dictBase;
    const BYTE* const istart = (const BYTE*)src;
    const BYTE* anchor = istart;
    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
    const U32   dictStartIndex = lowLimit;
    const BYTE* const dictStart = dictBase + dictStartIndex;
    const U32   dictLimit = ms->window.dictLimit;
    const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
    const BYTE* const prefixStart = base + prefixStartIndex;
    const BYTE* const dictEnd = dictBase + prefixStartIndex;
    const BYTE* const iend = istart + srcSize;
    const BYTE* const ilimit = iend - 8;
    U32 offset_1=rep[0], offset_2=rep[1];
    U32 offsetSaved1 = 0, offsetSaved2 = 0;

    const BYTE* ip0 = istart;
    const BYTE* ip1;
    const BYTE* ip2;
    const BYTE* ip3;
    U32 current0;


    size_t hash0; /* hash for ip0 */
    size_t hash1; /* hash for ip1 */
    U32 idx; /* match idx for ip0 */
    const BYTE* idxBase; /* base pointer for idx */

    U32 offcode;
    const BYTE* match0;
    size_t mLength;
    const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */

    size_t step;
    const BYTE* nextStep;
    const size_t kStepIncr = (1 << (kSearchStrength - 1));

    (void)hasStep; /* not currently specialized on whether it's accelerated */

    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);

    /* switch to "regular" variant if extDict is invalidated due to maxDistance */
    if (prefixStartIndex == dictStartIndex)
        return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);

    {   U32 const curr = (U32)(ip0 - base);
        U32 const maxRep = curr - dictStartIndex;
        if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
        if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
    }

    /* start each op */
_start: /* Requires: ip0 */

    step = stepSize;
    nextStep = ip0 + kStepIncr;

    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
    ip1 = ip0 + 1;
    ip2 = ip0 + step;
    ip3 = ip2 + 1;

    if (ip3 >= ilimit) {
        goto _cleanup;
    }

    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
    hash1 = ZSTD_hashPtr(ip1, hlog, mls);

    idx = hashTable[hash0];
    idxBase = idx < prefixStartIndex ? dictBase : base;

    do {
        {   /* load repcode match for ip[2] */
            U32 const current2 = (U32)(ip2 - base);
            U32 const repIndex = current2 - offset_1;
            const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
            U32 rval;
            if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
                 & (offset_1 > 0) ) {
                rval = MEM_read32(repBase + repIndex);
            } else {
                rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
            }

            /* write back hash table entry */
            current0 = (U32)(ip0 - base);
            hashTable[hash0] = current0;

            /* check repcode at ip[2] */
            if (MEM_read32(ip2) == rval) {
                ip0 = ip2;
                match0 = repBase + repIndex;
                matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
                assert((match0 != prefixStart) & (match0 != dictStart));
                mLength = ip0[-1] == match0[-1];
                ip0 -= mLength;
                match0 -= mLength;
                offcode = REPCODE1_TO_OFFBASE;
                mLength += 4;
                goto _match;
        }   }

        {   /* load match for ip[0] */
            U32 const mval = idx >= dictStartIndex ?
                    MEM_read32(idxBase + idx) :
                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */

            /* check match at ip[0] */
            if (MEM_read32(ip0) == mval) {
                /* found a match! */
                goto _offset;
        }   }

        /* lookup ip[1] */
        idx = hashTable[hash1];
        idxBase = idx < prefixStartIndex ? dictBase : base;

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip3;

        /* write back hash table entry */
        current0 = (U32)(ip0 - base);
        hashTable[hash0] = current0;

        {   /* load match for ip[0] */
            U32 const mval = idx >= dictStartIndex ?
                    MEM_read32(idxBase + idx) :
                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */

            /* check match at ip[0] */
            if (MEM_read32(ip0) == mval) {
                /* found a match! */
                goto _offset;
        }   }

        /* lookup ip[1] */
        idx = hashTable[hash1];
        idxBase = idx < prefixStartIndex ? dictBase : base;

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip0 + step;
        ip3 = ip1 + step;

        /* calculate step */
        if (ip2 >= nextStep) {
            step++;
            PREFETCH_L1(ip1 + 64);
            PREFETCH_L1(ip1 + 128);
            nextStep += kStepIncr;
        }
    } while (ip3 < ilimit);

_cleanup:
    /* Note that there are probably still a couple positions we could search.
     * However, it seems to be a meaningful performance hit to try to search
     * them. So let's not. */

    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;

    /* save reps for next block */
    rep[0] = offset_1 ? offset_1 : offsetSaved1;
    rep[1] = offset_2 ? offset_2 : offsetSaved2;

    /* Return the last literals size */
    return (size_t)(iend - anchor);

_offset: /* Requires: ip0, idx, idxBase */

    /* Compute the offset code. */
    {   U32 const offset = current0 - idx;
        const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
        matchEnd = idx < prefixStartIndex ? dictEnd : iend;
        match0 = idxBase + idx;
        offset_2 = offset_1;
        offset_1 = offset;
        offcode = OFFSET_TO_OFFBASE(offset);
        mLength = 4;

        /* Count the backwards match length. */
        while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
            ip0--;
            match0--;
            mLength++;
    }   }

_match: /* Requires: ip0, match0, offcode, matchEnd */

    /* Count the forward length. */
    assert(matchEnd != 0);
    mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);

    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);

    ip0 += mLength;
    anchor = ip0;

    /* write next hash table entry */
    if (ip1 < ip0) {
        hashTable[hash1] = (U32)(ip1 - base);
    }

    /* Fill table and check for immediate repcode. */
    if (ip0 <= ilimit) {
        /* Fill Table */
        assert(base+current0+2 > istart);  /* check base overflow */
        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);

        while (ip0 <= ilimit) {
            U32 const repIndex2 = (U32)(ip0-base) - offset_2;
            const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
            if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 > 0))  /* intentional underflow */
                 && (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
                const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
                size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
                ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
                ip0 += repLength2;
                anchor = ip0;
                continue;
            }
            break;
    }   }

    goto _start;
}

ZSTD_GEN_FAST_FN(extDict, 4, 0)
ZSTD_GEN_FAST_FN(extDict, 5, 0)
ZSTD_GEN_FAST_FN(extDict, 6, 0)
ZSTD_GEN_FAST_FN(extDict, 7, 0)

size_t ZSTD_compressBlock_fast_extDict(
        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize)
{
    U32 const mls = ms->cParams.minMatch;
    assert(ms->dictMatchState == NULL);
    switch(mls)
    {
    default: /* includes case 3 */
    case 4 :
        return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);
    case 5 :
        return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);
    case 6 :
        return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);
    case 7 :
        return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);
    }
}