xref: /freebsd/sys/contrib/zstd/lib/compress/zstd_ldm.c (revision 85732ac8)
1 /*
2  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
3  * All rights reserved.
4  *
5  * This source code is licensed under both the BSD-style license (found in the
6  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7  * in the COPYING file in the root directory of this source tree).
8  */
9 
10 #include "zstd_ldm.h"
11 
12 #include "debug.h"
13 #include "zstd_fast.h"          /* ZSTD_fillHashTable() */
14 #include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
15 
16 #define LDM_BUCKET_SIZE_LOG 3
17 #define LDM_MIN_MATCH_LENGTH 64
18 #define LDM_HASH_RLOG 7
19 #define LDM_HASH_CHAR_OFFSET 10
20 
21 void ZSTD_ldm_adjustParameters(ldmParams_t* params,
22                                ZSTD_compressionParameters const* cParams)
23 {
24     params->windowLog = cParams->windowLog;
25     ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
26     DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
27     if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
28     if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
29     if (cParams->strategy >= ZSTD_btopt) {
30       /* Get out of the way of the optimal parser */
31       U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
32       assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
33       assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
34       params->minMatchLength = minMatch;
35     }
36     if (params->hashLog == 0) {
37         params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
38         assert(params->hashLog <= ZSTD_HASHLOG_MAX);
39     }
40     if (params->hashEveryLog == 0) {
41         params->hashEveryLog = params->windowLog < params->hashLog
42                                    ? 0
43                                    : params->windowLog - params->hashLog;
44     }
45     params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
46 }
47 
48 size_t ZSTD_ldm_getTableSize(ldmParams_t params)
49 {
50     size_t const ldmHSize = ((size_t)1) << params.hashLog;
51     size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
52     size_t const ldmBucketSize =
53         ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
54     size_t const totalSize = ldmBucketSize + ldmHSize * sizeof(ldmEntry_t);
55     return params.enableLdm ? totalSize : 0;
56 }
57 
58 size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
59 {
60     return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
61 }
62 
63 /** ZSTD_ldm_getSmallHash() :
64  *  numBits should be <= 32
65  *  If numBits==0, returns 0.
66  *  @return : the most significant numBits of value. */
67 static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
68 {
69     assert(numBits <= 32);
70     return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
71 }
72 
73 /** ZSTD_ldm_getChecksum() :
74  *  numBitsToDiscard should be <= 32
75  *  @return : the next most significant 32 bits after numBitsToDiscard */
76 static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
77 {
78     assert(numBitsToDiscard <= 32);
79     return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
80 }
81 
82 /** ZSTD_ldm_getTag() ;
83  *  Given the hash, returns the most significant numTagBits bits
84  *  after (32 + hbits) bits.
85  *
86  *  If there are not enough bits remaining, return the last
87  *  numTagBits bits. */
88 static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
89 {
90     assert(numTagBits < 32 && hbits <= 32);
91     if (32 - hbits < numTagBits) {
92         return hash & (((U32)1 << numTagBits) - 1);
93     } else {
94         return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
95     }
96 }
97 
98 /** ZSTD_ldm_getBucket() :
99  *  Returns a pointer to the start of the bucket associated with hash. */
100 static ldmEntry_t* ZSTD_ldm_getBucket(
101         ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
102 {
103     return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
104 }
105 
106 /** ZSTD_ldm_insertEntry() :
107  *  Insert the entry with corresponding hash into the hash table */
108 static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
109                                  size_t const hash, const ldmEntry_t entry,
110                                  ldmParams_t const ldmParams)
111 {
112     BYTE* const bucketOffsets = ldmState->bucketOffsets;
113     *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
114     bucketOffsets[hash]++;
115     bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
116 }
117 
118 /** ZSTD_ldm_makeEntryAndInsertByTag() :
119  *
120  *  Gets the small hash, checksum, and tag from the rollingHash.
121  *
122  *  If the tag matches (1 << ldmParams.hashEveryLog)-1, then
123  *  creates an ldmEntry from the offset, and inserts it into the hash table.
124  *
125  *  hBits is the length of the small hash, which is the most significant hBits
126  *  of rollingHash. The checksum is the next 32 most significant bits, followed
127  *  by ldmParams.hashEveryLog bits that make up the tag. */
128 static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
129                                              U64 const rollingHash,
130                                              U32 const hBits,
131                                              U32 const offset,
132                                              ldmParams_t const ldmParams)
133 {
134     U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog);
135     U32 const tagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
136     if (tag == tagMask) {
137         U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
138         U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
139         ldmEntry_t entry;
140         entry.offset = offset;
141         entry.checksum = checksum;
142         ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
143     }
144 }
145 
146 /** ZSTD_ldm_getRollingHash() :
147  *  Get a 64-bit hash using the first len bytes from buf.
148  *
149  *  Giving bytes s = s_1, s_2, ... s_k, the hash is defined to be
150  *  H(s) = s_1*(a^(k-1)) + s_2*(a^(k-2)) + ... + s_k*(a^0)
151  *
152  *  where the constant a is defined to be prime8bytes.
153  *
154  *  The implementation adds an offset to each byte, so
155  *  H(s) = (s_1 + HASH_CHAR_OFFSET)*(a^(k-1)) + ... */
156 static U64 ZSTD_ldm_getRollingHash(const BYTE* buf, U32 len)
157 {
158     U64 ret = 0;
159     U32 i;
160     for (i = 0; i < len; i++) {
161         ret *= prime8bytes;
162         ret += buf[i] + LDM_HASH_CHAR_OFFSET;
163     }
164     return ret;
165 }
166 
167 /** ZSTD_ldm_ipow() :
168  *  Return base^exp. */
169 static U64 ZSTD_ldm_ipow(U64 base, U64 exp)
170 {
171     U64 ret = 1;
172     while (exp) {
173         if (exp & 1) { ret *= base; }
174         exp >>= 1;
175         base *= base;
176     }
177     return ret;
178 }
179 
180 U64 ZSTD_ldm_getHashPower(U32 minMatchLength) {
181     DEBUGLOG(4, "ZSTD_ldm_getHashPower: mml=%u", minMatchLength);
182     assert(minMatchLength >= ZSTD_LDM_MINMATCH_MIN);
183     return ZSTD_ldm_ipow(prime8bytes, minMatchLength - 1);
184 }
185 
186 /** ZSTD_ldm_updateHash() :
187  *  Updates hash by removing toRemove and adding toAdd. */
188 static U64 ZSTD_ldm_updateHash(U64 hash, BYTE toRemove, BYTE toAdd, U64 hashPower)
189 {
190     hash -= ((toRemove + LDM_HASH_CHAR_OFFSET) * hashPower);
191     hash *= prime8bytes;
192     hash += toAdd + LDM_HASH_CHAR_OFFSET;
193     return hash;
194 }
195 
196 /** ZSTD_ldm_countBackwardsMatch() :
197  *  Returns the number of bytes that match backwards before pIn and pMatch.
198  *
199  *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
200 static size_t ZSTD_ldm_countBackwardsMatch(
201             const BYTE* pIn, const BYTE* pAnchor,
202             const BYTE* pMatch, const BYTE* pBase)
203 {
204     size_t matchLength = 0;
205     while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
206         pIn--;
207         pMatch--;
208         matchLength++;
209     }
210     return matchLength;
211 }
212 
213 /** ZSTD_ldm_fillFastTables() :
214  *
215  *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
216  *  This is similar to ZSTD_loadDictionaryContent.
217  *
218  *  The tables for the other strategies are filled within their
219  *  block compressors. */
220 static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
221                                       void const* end)
222 {
223     const BYTE* const iend = (const BYTE*)end;
224 
225     switch(ms->cParams.strategy)
226     {
227     case ZSTD_fast:
228         ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
229         break;
230 
231     case ZSTD_dfast:
232         ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
233         break;
234 
235     case ZSTD_greedy:
236     case ZSTD_lazy:
237     case ZSTD_lazy2:
238     case ZSTD_btlazy2:
239     case ZSTD_btopt:
240     case ZSTD_btultra:
241         break;
242     default:
243         assert(0);  /* not possible : not a valid strategy id */
244     }
245 
246     return 0;
247 }
248 
249 /** ZSTD_ldm_fillLdmHashTable() :
250  *
251  *  Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
252  *  lastHash is the rolling hash that corresponds to lastHashed.
253  *
254  *  Returns the rolling hash corresponding to position iend-1. */
255 static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
256                                      U64 lastHash, const BYTE* lastHashed,
257                                      const BYTE* iend, const BYTE* base,
258                                      U32 hBits, ldmParams_t const ldmParams)
259 {
260     U64 rollingHash = lastHash;
261     const BYTE* cur = lastHashed + 1;
262 
263     while (cur < iend) {
264         rollingHash = ZSTD_ldm_updateHash(rollingHash, cur[-1],
265                                           cur[ldmParams.minMatchLength-1],
266                                           state->hashPower);
267         ZSTD_ldm_makeEntryAndInsertByTag(state,
268                                          rollingHash, hBits,
269                                          (U32)(cur - base), ldmParams);
270         ++cur;
271     }
272     return rollingHash;
273 }
274 
275 
276 /** ZSTD_ldm_limitTableUpdate() :
277  *
278  *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
279  *  if it is far way
280  *  (after a long match, only update tables a limited amount). */
281 static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
282 {
283     U32 const current = (U32)(anchor - ms->window.base);
284     if (current > ms->nextToUpdate + 1024) {
285         ms->nextToUpdate =
286             current - MIN(512, current - ms->nextToUpdate - 1024);
287     }
288 }
289 
290 static size_t ZSTD_ldm_generateSequences_internal(
291         ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
292         ldmParams_t const* params, void const* src, size_t srcSize)
293 {
294     /* LDM parameters */
295     int const extDict = ZSTD_window_hasExtDict(ldmState->window);
296     U32 const minMatchLength = params->minMatchLength;
297     U64 const hashPower = ldmState->hashPower;
298     U32 const hBits = params->hashLog - params->bucketSizeLog;
299     U32 const ldmBucketSize = 1U << params->bucketSizeLog;
300     U32 const hashEveryLog = params->hashEveryLog;
301     U32 const ldmTagMask = (1U << params->hashEveryLog) - 1;
302     /* Prefix and extDict parameters */
303     U32 const dictLimit = ldmState->window.dictLimit;
304     U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
305     BYTE const* const base = ldmState->window.base;
306     BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
307     BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
308     BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
309     BYTE const* const lowPrefixPtr = base + dictLimit;
310     /* Input bounds */
311     BYTE const* const istart = (BYTE const*)src;
312     BYTE const* const iend = istart + srcSize;
313     BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
314     /* Input positions */
315     BYTE const* anchor = istart;
316     BYTE const* ip = istart;
317     /* Rolling hash */
318     BYTE const* lastHashed = NULL;
319     U64 rollingHash = 0;
320 
321     while (ip <= ilimit) {
322         size_t mLength;
323         U32 const current = (U32)(ip - base);
324         size_t forwardMatchLength = 0, backwardMatchLength = 0;
325         ldmEntry_t* bestEntry = NULL;
326         if (ip != istart) {
327             rollingHash = ZSTD_ldm_updateHash(rollingHash, lastHashed[0],
328                                               lastHashed[minMatchLength],
329                                               hashPower);
330         } else {
331             rollingHash = ZSTD_ldm_getRollingHash(ip, minMatchLength);
332         }
333         lastHashed = ip;
334 
335         /* Do not insert and do not look for a match */
336         if (ZSTD_ldm_getTag(rollingHash, hBits, hashEveryLog) != ldmTagMask) {
337            ip++;
338            continue;
339         }
340 
341         /* Get the best entry and compute the match lengths */
342         {
343             ldmEntry_t* const bucket =
344                 ZSTD_ldm_getBucket(ldmState,
345                                    ZSTD_ldm_getSmallHash(rollingHash, hBits),
346                                    *params);
347             ldmEntry_t* cur;
348             size_t bestMatchLength = 0;
349             U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
350 
351             for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
352                 size_t curForwardMatchLength, curBackwardMatchLength,
353                        curTotalMatchLength;
354                 if (cur->checksum != checksum || cur->offset <= lowestIndex) {
355                     continue;
356                 }
357                 if (extDict) {
358                     BYTE const* const curMatchBase =
359                         cur->offset < dictLimit ? dictBase : base;
360                     BYTE const* const pMatch = curMatchBase + cur->offset;
361                     BYTE const* const matchEnd =
362                         cur->offset < dictLimit ? dictEnd : iend;
363                     BYTE const* const lowMatchPtr =
364                         cur->offset < dictLimit ? dictStart : lowPrefixPtr;
365 
366                     curForwardMatchLength = ZSTD_count_2segments(
367                                                 ip, pMatch, iend,
368                                                 matchEnd, lowPrefixPtr);
369                     if (curForwardMatchLength < minMatchLength) {
370                         continue;
371                     }
372                     curBackwardMatchLength =
373                         ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
374                                                      lowMatchPtr);
375                     curTotalMatchLength = curForwardMatchLength +
376                                           curBackwardMatchLength;
377                 } else { /* !extDict */
378                     BYTE const* const pMatch = base + cur->offset;
379                     curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
380                     if (curForwardMatchLength < minMatchLength) {
381                         continue;
382                     }
383                     curBackwardMatchLength =
384                         ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
385                                                      lowPrefixPtr);
386                     curTotalMatchLength = curForwardMatchLength +
387                                           curBackwardMatchLength;
388                 }
389 
390                 if (curTotalMatchLength > bestMatchLength) {
391                     bestMatchLength = curTotalMatchLength;
392                     forwardMatchLength = curForwardMatchLength;
393                     backwardMatchLength = curBackwardMatchLength;
394                     bestEntry = cur;
395                 }
396             }
397         }
398 
399         /* No match found -- continue searching */
400         if (bestEntry == NULL) {
401             ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
402                                              hBits, current,
403                                              *params);
404             ip++;
405             continue;
406         }
407 
408         /* Match found */
409         mLength = forwardMatchLength + backwardMatchLength;
410         ip -= backwardMatchLength;
411 
412         {
413             /* Store the sequence:
414              * ip = current - backwardMatchLength
415              * The match is at (bestEntry->offset - backwardMatchLength)
416              */
417             U32 const matchIndex = bestEntry->offset;
418             U32 const offset = current - matchIndex;
419             rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
420 
421             /* Out of sequence storage */
422             if (rawSeqStore->size == rawSeqStore->capacity)
423                 return ERROR(dstSize_tooSmall);
424             seq->litLength = (U32)(ip - anchor);
425             seq->matchLength = (U32)mLength;
426             seq->offset = offset;
427             rawSeqStore->size++;
428         }
429 
430         /* Insert the current entry into the hash table */
431         ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
432                                          (U32)(lastHashed - base),
433                                          *params);
434 
435         assert(ip + backwardMatchLength == lastHashed);
436 
437         /* Fill the hash table from lastHashed+1 to ip+mLength*/
438         /* Heuristic: don't need to fill the entire table at end of block */
439         if (ip + mLength <= ilimit) {
440             rollingHash = ZSTD_ldm_fillLdmHashTable(
441                               ldmState, rollingHash, lastHashed,
442                               ip + mLength, base, hBits, *params);
443             lastHashed = ip + mLength - 1;
444         }
445         ip += mLength;
446         anchor = ip;
447     }
448     return iend - anchor;
449 }
450 
451 /*! ZSTD_ldm_reduceTable() :
452  *  reduce table indexes by `reducerValue` */
453 static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
454                                  U32 const reducerValue)
455 {
456     U32 u;
457     for (u = 0; u < size; u++) {
458         if (table[u].offset < reducerValue) table[u].offset = 0;
459         else table[u].offset -= reducerValue;
460     }
461 }
462 
463 size_t ZSTD_ldm_generateSequences(
464         ldmState_t* ldmState, rawSeqStore_t* sequences,
465         ldmParams_t const* params, void const* src, size_t srcSize)
466 {
467     U32 const maxDist = 1U << params->windowLog;
468     BYTE const* const istart = (BYTE const*)src;
469     BYTE const* const iend = istart + srcSize;
470     size_t const kMaxChunkSize = 1 << 20;
471     size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
472     size_t chunk;
473     size_t leftoverSize = 0;
474 
475     assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
476     /* Check that ZSTD_window_update() has been called for this chunk prior
477      * to passing it to this function.
478      */
479     assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
480     /* The input could be very large (in zstdmt), so it must be broken up into
481      * chunks to enforce the maximmum distance and handle overflow correction.
482      */
483     assert(sequences->pos <= sequences->size);
484     assert(sequences->size <= sequences->capacity);
485     for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
486         BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
487         size_t const remaining = (size_t)(iend - chunkStart);
488         BYTE const *const chunkEnd =
489             (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
490         size_t const chunkSize = chunkEnd - chunkStart;
491         size_t newLeftoverSize;
492         size_t const prevSize = sequences->size;
493 
494         assert(chunkStart < iend);
495         /* 1. Perform overflow correction if necessary. */
496         if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
497             U32 const ldmHSize = 1U << params->hashLog;
498             U32 const correction = ZSTD_window_correctOverflow(
499                 &ldmState->window, /* cycleLog */ 0, maxDist, src);
500             ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
501         }
502         /* 2. We enforce the maximum offset allowed.
503          *
504          * kMaxChunkSize should be small enough that we don't lose too much of
505          * the window through early invalidation.
506          * TODO: * Test the chunk size.
507          *       * Try invalidation after the sequence generation and test the
508          *         the offset against maxDist directly.
509          */
510         ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, NULL, NULL);
511         /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
512         newLeftoverSize = ZSTD_ldm_generateSequences_internal(
513             ldmState, sequences, params, chunkStart, chunkSize);
514         if (ZSTD_isError(newLeftoverSize))
515             return newLeftoverSize;
516         /* 4. We add the leftover literals from previous iterations to the first
517          *    newly generated sequence, or add the `newLeftoverSize` if none are
518          *    generated.
519          */
520         /* Prepend the leftover literals from the last call */
521         if (prevSize < sequences->size) {
522             sequences->seq[prevSize].litLength += (U32)leftoverSize;
523             leftoverSize = newLeftoverSize;
524         } else {
525             assert(newLeftoverSize == chunkSize);
526             leftoverSize += chunkSize;
527         }
528     }
529     return 0;
530 }
531 
532 void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
533     while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
534         rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
535         if (srcSize <= seq->litLength) {
536             /* Skip past srcSize literals */
537             seq->litLength -= (U32)srcSize;
538             return;
539         }
540         srcSize -= seq->litLength;
541         seq->litLength = 0;
542         if (srcSize < seq->matchLength) {
543             /* Skip past the first srcSize of the match */
544             seq->matchLength -= (U32)srcSize;
545             if (seq->matchLength < minMatch) {
546                 /* The match is too short, omit it */
547                 if (rawSeqStore->pos + 1 < rawSeqStore->size) {
548                     seq[1].litLength += seq[0].matchLength;
549                 }
550                 rawSeqStore->pos++;
551             }
552             return;
553         }
554         srcSize -= seq->matchLength;
555         seq->matchLength = 0;
556         rawSeqStore->pos++;
557     }
558 }
559 
560 /**
561  * If the sequence length is longer than remaining then the sequence is split
562  * between this block and the next.
563  *
564  * Returns the current sequence to handle, or if the rest of the block should
565  * be literals, it returns a sequence with offset == 0.
566  */
567 static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
568                                  U32 const remaining, U32 const minMatch)
569 {
570     rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
571     assert(sequence.offset > 0);
572     /* Likely: No partial sequence */
573     if (remaining >= sequence.litLength + sequence.matchLength) {
574         rawSeqStore->pos++;
575         return sequence;
576     }
577     /* Cut the sequence short (offset == 0 ==> rest is literals). */
578     if (remaining <= sequence.litLength) {
579         sequence.offset = 0;
580     } else if (remaining < sequence.litLength + sequence.matchLength) {
581         sequence.matchLength = remaining - sequence.litLength;
582         if (sequence.matchLength < minMatch) {
583             sequence.offset = 0;
584         }
585     }
586     /* Skip past `remaining` bytes for the future sequences. */
587     ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
588     return sequence;
589 }
590 
591 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
592     ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
593     void const* src, size_t srcSize)
594 {
595     const ZSTD_compressionParameters* const cParams = &ms->cParams;
596     unsigned const minMatch = cParams->searchLength;
597     ZSTD_blockCompressor const blockCompressor =
598         ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
599     /* Input bounds */
600     BYTE const* const istart = (BYTE const*)src;
601     BYTE const* const iend = istart + srcSize;
602     /* Input positions */
603     BYTE const* ip = istart;
604 
605     DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
606     assert(rawSeqStore->pos <= rawSeqStore->size);
607     assert(rawSeqStore->size <= rawSeqStore->capacity);
608     /* Loop through each sequence and apply the block compressor to the lits */
609     while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
610         /* maybeSplitSequence updates rawSeqStore->pos */
611         rawSeq const sequence = maybeSplitSequence(rawSeqStore,
612                                                    (U32)(iend - ip), minMatch);
613         int i;
614         /* End signal */
615         if (sequence.offset == 0)
616             break;
617 
618         assert(sequence.offset <= (1U << cParams->windowLog));
619         assert(ip + sequence.litLength + sequence.matchLength <= iend);
620 
621         /* Fill tables for block compressor */
622         ZSTD_ldm_limitTableUpdate(ms, ip);
623         ZSTD_ldm_fillFastTables(ms, ip);
624         /* Run the block compressor */
625         DEBUGLOG(5, "calling block compressor on segment of size %u", sequence.litLength);
626         {
627             size_t const newLitLength =
628                 blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
629             ip += sequence.litLength;
630             /* Update the repcodes */
631             for (i = ZSTD_REP_NUM - 1; i > 0; i--)
632                 rep[i] = rep[i-1];
633             rep[0] = sequence.offset;
634             /* Store the sequence */
635             ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength,
636                           sequence.offset + ZSTD_REP_MOVE,
637                           sequence.matchLength - MINMATCH);
638             ip += sequence.matchLength;
639         }
640     }
641     /* Fill the tables for the block compressor */
642     ZSTD_ldm_limitTableUpdate(ms, ip);
643     ZSTD_ldm_fillFastTables(ms, ip);
644     /* Compress the last literals */
645     return blockCompressor(ms, seqStore, rep, ip, iend - ip);
646 }
647