Welcome to mirror list, hosted at ThFree Co, Russian Federation.

zstd_ldm.c « zstd - github.com/wolfpld/tracy.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
blob: 3916da49465d490e6988226c4c9be3cc2beb5d27 (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
/*
 * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
 * 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_ldm.h"

#include "debug.h"
#include "zstd_fast.h"          /* ZSTD_fillHashTable() */
#include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */

#define LDM_BUCKET_SIZE_LOG 3
#define LDM_MIN_MATCH_LENGTH 64
#define LDM_HASH_RLOG 7
#define LDM_HASH_CHAR_OFFSET 10

void ZSTD_ldm_adjustParameters(ldmParams_t* params,
                               ZSTD_compressionParameters const* cParams)
{
    params->windowLog = cParams->windowLog;
    ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
    DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
    if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
    if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
    if (cParams->strategy >= ZSTD_btopt) {
      /* Get out of the way of the optimal parser */
      U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
      assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
      assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
      params->minMatchLength = minMatch;
    }
    if (params->hashLog == 0) {
        params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
        assert(params->hashLog <= ZSTD_HASHLOG_MAX);
    }
    if (params->hashRateLog == 0) {
        params->hashRateLog = params->windowLog < params->hashLog
                                   ? 0
                                   : params->windowLog - params->hashLog;
    }
    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
}

size_t ZSTD_ldm_getTableSize(ldmParams_t params)
{
    size_t const ldmHSize = ((size_t)1) << params.hashLog;
    size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
    size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
    size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
                           + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
    return params.enableLdm ? totalSize : 0;
}

size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
{
    return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
}

/** ZSTD_ldm_getSmallHash() :
 *  numBits should be <= 32
 *  If numBits==0, returns 0.
 *  @return : the most significant numBits of value. */
static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
{
    assert(numBits <= 32);
    return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
}

/** ZSTD_ldm_getChecksum() :
 *  numBitsToDiscard should be <= 32
 *  @return : the next most significant 32 bits after numBitsToDiscard */
static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
{
    assert(numBitsToDiscard <= 32);
    return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
}

/** ZSTD_ldm_getTag() ;
 *  Given the hash, returns the most significant numTagBits bits
 *  after (32 + hbits) bits.
 *
 *  If there are not enough bits remaining, return the last
 *  numTagBits bits. */
static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
{
    assert(numTagBits < 32 && hbits <= 32);
    if (32 - hbits < numTagBits) {
        return hash & (((U32)1 << numTagBits) - 1);
    } else {
        return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
    }
}

/** ZSTD_ldm_getBucket() :
 *  Returns a pointer to the start of the bucket associated with hash. */
static ldmEntry_t* ZSTD_ldm_getBucket(
        ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
{
    return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
}

/** ZSTD_ldm_insertEntry() :
 *  Insert the entry with corresponding hash into the hash table */
static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
                                 size_t const hash, const ldmEntry_t entry,
                                 ldmParams_t const ldmParams)
{
    BYTE* const bucketOffsets = ldmState->bucketOffsets;
    *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
    bucketOffsets[hash]++;
    bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
}

/** ZSTD_ldm_makeEntryAndInsertByTag() :
 *
 *  Gets the small hash, checksum, and tag from the rollingHash.
 *
 *  If the tag matches (1 << ldmParams.hashRateLog)-1, then
 *  creates an ldmEntry from the offset, and inserts it into the hash table.
 *
 *  hBits is the length of the small hash, which is the most significant hBits
 *  of rollingHash. The checksum is the next 32 most significant bits, followed
 *  by ldmParams.hashRateLog bits that make up the tag. */
static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
                                             U64 const rollingHash,
                                             U32 const hBits,
                                             U32 const offset,
                                             ldmParams_t const ldmParams)
{
    U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashRateLog);
    U32 const tagMask = ((U32)1 << ldmParams.hashRateLog) - 1;
    if (tag == tagMask) {
        U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
        U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
        ldmEntry_t entry;
        entry.offset = offset;
        entry.checksum = checksum;
        ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
    }
}

/** ZSTD_ldm_countBackwardsMatch() :
 *  Returns the number of bytes that match backwards before pIn and pMatch.
 *
 *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
static size_t ZSTD_ldm_countBackwardsMatch(
            const BYTE* pIn, const BYTE* pAnchor,
            const BYTE* pMatch, const BYTE* pBase)
{
    size_t matchLength = 0;
    while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
        pIn--;
        pMatch--;
        matchLength++;
    }
    return matchLength;
}

/** ZSTD_ldm_fillFastTables() :
 *
 *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
 *  This is similar to ZSTD_loadDictionaryContent.
 *
 *  The tables for the other strategies are filled within their
 *  block compressors. */
static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
                                      void const* end)
{
    const BYTE* const iend = (const BYTE*)end;

    switch(ms->cParams.strategy)
    {
    case ZSTD_fast:
        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
        break;

    case ZSTD_dfast:
        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
        break;

    case ZSTD_greedy:
    case ZSTD_lazy:
    case ZSTD_lazy2:
    case ZSTD_btlazy2:
    case ZSTD_btopt:
    case ZSTD_btultra:
    case ZSTD_btultra2:
        break;
    default:
        assert(0);  /* not possible : not a valid strategy id */
    }

    return 0;
}

/** ZSTD_ldm_fillLdmHashTable() :
 *
 *  Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
 *  lastHash is the rolling hash that corresponds to lastHashed.
 *
 *  Returns the rolling hash corresponding to position iend-1. */
static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
                                     U64 lastHash, const BYTE* lastHashed,
                                     const BYTE* iend, const BYTE* base,
                                     U32 hBits, ldmParams_t const ldmParams)
{
    U64 rollingHash = lastHash;
    const BYTE* cur = lastHashed + 1;

    while (cur < iend) {
        rollingHash = ZSTD_rollingHash_rotate(rollingHash, cur[-1],
                                              cur[ldmParams.minMatchLength-1],
                                              state->hashPower);
        ZSTD_ldm_makeEntryAndInsertByTag(state,
                                         rollingHash, hBits,
                                         (U32)(cur - base), ldmParams);
        ++cur;
    }
    return rollingHash;
}

void ZSTD_ldm_fillHashTable(
            ldmState_t* state, const BYTE* ip,
            const BYTE* iend, ldmParams_t const* params)
{
    DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
    if ((size_t)(iend - ip) >= params->minMatchLength) {
        U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
        ZSTD_ldm_fillLdmHashTable(
            state, startingHash, ip, iend - params->minMatchLength, state->window.base,
            params->hashLog - params->bucketSizeLog,
            *params);
    }
}


/** ZSTD_ldm_limitTableUpdate() :
 *
 *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
 *  if it is far way
 *  (after a long match, only update tables a limited amount). */
static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
{
    U32 const current = (U32)(anchor - ms->window.base);
    if (current > ms->nextToUpdate + 1024) {
        ms->nextToUpdate =
            current - MIN(512, current - ms->nextToUpdate - 1024);
    }
}

static size_t ZSTD_ldm_generateSequences_internal(
        ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
        ldmParams_t const* params, void const* src, size_t srcSize)
{
    /* LDM parameters */
    int const extDict = ZSTD_window_hasExtDict(ldmState->window);
    U32 const minMatchLength = params->minMatchLength;
    U64 const hashPower = ldmState->hashPower;
    U32 const hBits = params->hashLog - params->bucketSizeLog;
    U32 const ldmBucketSize = 1U << params->bucketSizeLog;
    U32 const hashRateLog = params->hashRateLog;
    U32 const ldmTagMask = (1U << params->hashRateLog) - 1;
    /* Prefix and extDict parameters */
    U32 const dictLimit = ldmState->window.dictLimit;
    U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
    BYTE const* const base = ldmState->window.base;
    BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
    BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
    BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
    BYTE const* const lowPrefixPtr = base + dictLimit;
    /* Input bounds */
    BYTE const* const istart = (BYTE const*)src;
    BYTE const* const iend = istart + srcSize;
    BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
    /* Input positions */
    BYTE const* anchor = istart;
    BYTE const* ip = istart;
    /* Rolling hash */
    BYTE const* lastHashed = NULL;
    U64 rollingHash = 0;

    while (ip <= ilimit) {
        size_t mLength;
        U32 const current = (U32)(ip - base);
        size_t forwardMatchLength = 0, backwardMatchLength = 0;
        ldmEntry_t* bestEntry = NULL;
        if (ip != istart) {
            rollingHash = ZSTD_rollingHash_rotate(rollingHash, lastHashed[0],
                                                  lastHashed[minMatchLength],
                                                  hashPower);
        } else {
            rollingHash = ZSTD_rollingHash_compute(ip, minMatchLength);
        }
        lastHashed = ip;

        /* Do not insert and do not look for a match */
        if (ZSTD_ldm_getTag(rollingHash, hBits, hashRateLog) != ldmTagMask) {
           ip++;
           continue;
        }

        /* Get the best entry and compute the match lengths */
        {
            ldmEntry_t* const bucket =
                ZSTD_ldm_getBucket(ldmState,
                                   ZSTD_ldm_getSmallHash(rollingHash, hBits),
                                   *params);
            ldmEntry_t* cur;
            size_t bestMatchLength = 0;
            U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);

            for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
                size_t curForwardMatchLength, curBackwardMatchLength,
                       curTotalMatchLength;
                if (cur->checksum != checksum || cur->offset <= lowestIndex) {
                    continue;
                }
                if (extDict) {
                    BYTE const* const curMatchBase =
                        cur->offset < dictLimit ? dictBase : base;
                    BYTE const* const pMatch = curMatchBase + cur->offset;
                    BYTE const* const matchEnd =
                        cur->offset < dictLimit ? dictEnd : iend;
                    BYTE const* const lowMatchPtr =
                        cur->offset < dictLimit ? dictStart : lowPrefixPtr;

                    curForwardMatchLength = ZSTD_count_2segments(
                                                ip, pMatch, iend,
                                                matchEnd, lowPrefixPtr);
                    if (curForwardMatchLength < minMatchLength) {
                        continue;
                    }
                    curBackwardMatchLength =
                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
                                                     lowMatchPtr);
                    curTotalMatchLength = curForwardMatchLength +
                                          curBackwardMatchLength;
                } else { /* !extDict */
                    BYTE const* const pMatch = base + cur->offset;
                    curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
                    if (curForwardMatchLength < minMatchLength) {
                        continue;
                    }
                    curBackwardMatchLength =
                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
                                                     lowPrefixPtr);
                    curTotalMatchLength = curForwardMatchLength +
                                          curBackwardMatchLength;
                }

                if (curTotalMatchLength > bestMatchLength) {
                    bestMatchLength = curTotalMatchLength;
                    forwardMatchLength = curForwardMatchLength;
                    backwardMatchLength = curBackwardMatchLength;
                    bestEntry = cur;
                }
            }
        }

        /* No match found -- continue searching */
        if (bestEntry == NULL) {
            ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
                                             hBits, current,
                                             *params);
            ip++;
            continue;
        }

        /* Match found */
        mLength = forwardMatchLength + backwardMatchLength;
        ip -= backwardMatchLength;

        {
            /* Store the sequence:
             * ip = current - backwardMatchLength
             * The match is at (bestEntry->offset - backwardMatchLength)
             */
            U32 const matchIndex = bestEntry->offset;
            U32 const offset = current - matchIndex;
            rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;

            /* Out of sequence storage */
            if (rawSeqStore->size == rawSeqStore->capacity)
                return ERROR(dstSize_tooSmall);
            seq->litLength = (U32)(ip - anchor);
            seq->matchLength = (U32)mLength;
            seq->offset = offset;
            rawSeqStore->size++;
        }

        /* Insert the current entry into the hash table */
        ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
                                         (U32)(lastHashed - base),
                                         *params);

        assert(ip + backwardMatchLength == lastHashed);

        /* Fill the hash table from lastHashed+1 to ip+mLength*/
        /* Heuristic: don't need to fill the entire table at end of block */
        if (ip + mLength <= ilimit) {
            rollingHash = ZSTD_ldm_fillLdmHashTable(
                              ldmState, rollingHash, lastHashed,
                              ip + mLength, base, hBits, *params);
            lastHashed = ip + mLength - 1;
        }
        ip += mLength;
        anchor = ip;
    }
    return iend - anchor;
}

/*! ZSTD_ldm_reduceTable() :
 *  reduce table indexes by `reducerValue` */
static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
                                 U32 const reducerValue)
{
    U32 u;
    for (u = 0; u < size; u++) {
        if (table[u].offset < reducerValue) table[u].offset = 0;
        else table[u].offset -= reducerValue;
    }
}

size_t ZSTD_ldm_generateSequences(
        ldmState_t* ldmState, rawSeqStore_t* sequences,
        ldmParams_t const* params, void const* src, size_t srcSize)
{
    U32 const maxDist = 1U << params->windowLog;
    BYTE const* const istart = (BYTE const*)src;
    BYTE const* const iend = istart + srcSize;
    size_t const kMaxChunkSize = 1 << 20;
    size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
    size_t chunk;
    size_t leftoverSize = 0;

    assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
    /* Check that ZSTD_window_update() has been called for this chunk prior
     * to passing it to this function.
     */
    assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
    /* The input could be very large (in zstdmt), so it must be broken up into
     * chunks to enforce the maximum distance and handle overflow correction.
     */
    assert(sequences->pos <= sequences->size);
    assert(sequences->size <= sequences->capacity);
    for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
        BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
        size_t const remaining = (size_t)(iend - chunkStart);
        BYTE const *const chunkEnd =
            (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
        size_t const chunkSize = chunkEnd - chunkStart;
        size_t newLeftoverSize;
        size_t const prevSize = sequences->size;

        assert(chunkStart < iend);
        /* 1. Perform overflow correction if necessary. */
        if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
            U32 const ldmHSize = 1U << params->hashLog;
            U32 const correction = ZSTD_window_correctOverflow(
                &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
            ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
            /* invalidate dictionaries on overflow correction */
            ldmState->loadedDictEnd = 0;
        }
        /* 2. We enforce the maximum offset allowed.
         *
         * kMaxChunkSize should be small enough that we don't lose too much of
         * the window through early invalidation.
         * TODO: * Test the chunk size.
         *       * Try invalidation after the sequence generation and test the
         *         the offset against maxDist directly.
         *
         * NOTE: Because of dictionaries + sequence splitting we MUST make sure
         * that any offset used is valid at the END of the sequence, since it may
         * be split into two sequences. This condition holds when using
         * ZSTD_window_enforceMaxDist(), but if we move to checking offsets
         * against maxDist directly, we'll have to carefully handle that case.
         */
        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
        /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
        newLeftoverSize = ZSTD_ldm_generateSequences_internal(
            ldmState, sequences, params, chunkStart, chunkSize);
        if (ZSTD_isError(newLeftoverSize))
            return newLeftoverSize;
        /* 4. We add the leftover literals from previous iterations to the first
         *    newly generated sequence, or add the `newLeftoverSize` if none are
         *    generated.
         */
        /* Prepend the leftover literals from the last call */
        if (prevSize < sequences->size) {
            sequences->seq[prevSize].litLength += (U32)leftoverSize;
            leftoverSize = newLeftoverSize;
        } else {
            assert(newLeftoverSize == chunkSize);
            leftoverSize += chunkSize;
        }
    }
    return 0;
}

void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
        if (srcSize <= seq->litLength) {
            /* Skip past srcSize literals */
            seq->litLength -= (U32)srcSize;
            return;
        }
        srcSize -= seq->litLength;
        seq->litLength = 0;
        if (srcSize < seq->matchLength) {
            /* Skip past the first srcSize of the match */
            seq->matchLength -= (U32)srcSize;
            if (seq->matchLength < minMatch) {
                /* The match is too short, omit it */
                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
                    seq[1].litLength += seq[0].matchLength;
                }
                rawSeqStore->pos++;
            }
            return;
        }
        srcSize -= seq->matchLength;
        seq->matchLength = 0;
        rawSeqStore->pos++;
    }
}

/**
 * If the sequence length is longer than remaining then the sequence is split
 * between this block and the next.
 *
 * Returns the current sequence to handle, or if the rest of the block should
 * be literals, it returns a sequence with offset == 0.
 */
static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
                                 U32 const remaining, U32 const minMatch)
{
    rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
    assert(sequence.offset > 0);
    /* Likely: No partial sequence */
    if (remaining >= sequence.litLength + sequence.matchLength) {
        rawSeqStore->pos++;
        return sequence;
    }
    /* Cut the sequence short (offset == 0 ==> rest is literals). */
    if (remaining <= sequence.litLength) {
        sequence.offset = 0;
    } else if (remaining < sequence.litLength + sequence.matchLength) {
        sequence.matchLength = remaining - sequence.litLength;
        if (sequence.matchLength < minMatch) {
            sequence.offset = 0;
        }
    }
    /* Skip past `remaining` bytes for the future sequences. */
    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
    return sequence;
}

size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
    ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
    void const* src, size_t srcSize)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    unsigned const minMatch = cParams->minMatch;
    ZSTD_blockCompressor const blockCompressor =
        ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
    /* Input bounds */
    BYTE const* const istart = (BYTE const*)src;
    BYTE const* const iend = istart + srcSize;
    /* Input positions */
    BYTE const* ip = istart;

    DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
    assert(rawSeqStore->pos <= rawSeqStore->size);
    assert(rawSeqStore->size <= rawSeqStore->capacity);
    /* Loop through each sequence and apply the block compressor to the lits */
    while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
        /* maybeSplitSequence updates rawSeqStore->pos */
        rawSeq const sequence = maybeSplitSequence(rawSeqStore,
                                                   (U32)(iend - ip), minMatch);
        int i;
        /* End signal */
        if (sequence.offset == 0)
            break;

        assert(ip + sequence.litLength + sequence.matchLength <= iend);

        /* Fill tables for block compressor */
        ZSTD_ldm_limitTableUpdate(ms, ip);
        ZSTD_ldm_fillFastTables(ms, ip);
        /* Run the block compressor */
        DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
        {
            size_t const newLitLength =
                blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
            ip += sequence.litLength;
            /* Update the repcodes */
            for (i = ZSTD_REP_NUM - 1; i > 0; i--)
                rep[i] = rep[i-1];
            rep[0] = sequence.offset;
            /* Store the sequence */
            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
                          sequence.offset + ZSTD_REP_MOVE,
                          sequence.matchLength - MINMATCH);
            ip += sequence.matchLength;
        }
    }
    /* Fill the tables for the block compressor */
    ZSTD_ldm_limitTableUpdate(ms, ip);
    ZSTD_ldm_fillFastTables(ms, ip);
    /* Compress the last literals */
    return blockCompressor(ms, seqStore, rep, ip, iend - ip);
}