xref: /freebsd/sys/contrib/openzfs/module/zstd/lib/compress/zstd_ldm.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only
2 /*
3  * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
4  * All rights reserved.
5  *
6  * This source code is licensed under both the BSD-style license (found in the
7  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
8  * in the COPYING file in the root directory of this source tree).
9  * You may select, at your option, one of the above-listed licenses.
10  */
11 
12 #include "zstd_ldm.h"
13 
14 #include "../common/debug.h"
15 #include "zstd_fast.h"          /* ZSTD_fillHashTable() */
16 #include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
17 
18 #define LDM_BUCKET_SIZE_LOG 3
19 #define LDM_MIN_MATCH_LENGTH 64
20 #define LDM_HASH_RLOG 7
21 #define LDM_HASH_CHAR_OFFSET 10
22 
ZSTD_ldm_adjustParameters(ldmParams_t * params,ZSTD_compressionParameters const * cParams)23 void ZSTD_ldm_adjustParameters(ldmParams_t* params,
24                                ZSTD_compressionParameters const* cParams)
25 {
26     params->windowLog = cParams->windowLog;
27     ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
28     DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
29     if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
30     if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
31     if (cParams->strategy >= ZSTD_btopt) {
32       /* Get out of the way of the optimal parser */
33       U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
34       assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
35       assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
36       params->minMatchLength = minMatch;
37     }
38     if (params->hashLog == 0) {
39         params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
40         assert(params->hashLog <= ZSTD_HASHLOG_MAX);
41     }
42     if (params->hashRateLog == 0) {
43         params->hashRateLog = params->windowLog < params->hashLog
44                                    ? 0
45                                    : params->windowLog - params->hashLog;
46     }
47     params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
48 }
49 
ZSTD_ldm_getTableSize(ldmParams_t params)50 size_t ZSTD_ldm_getTableSize(ldmParams_t params)
51 {
52     size_t const ldmHSize = ((size_t)1) << params.hashLog;
53     size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
54     size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
55     size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
56                            + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
57     return params.enableLdm ? totalSize : 0;
58 }
59 
ZSTD_ldm_getMaxNbSeq(ldmParams_t params,size_t maxChunkSize)60 size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
61 {
62     return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
63 }
64 
65 /** ZSTD_ldm_getSmallHash() :
66  *  numBits should be <= 32
67  *  If numBits==0, returns 0.
68  *  @return : the most significant numBits of value. */
ZSTD_ldm_getSmallHash(U64 value,U32 numBits)69 static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
70 {
71     assert(numBits <= 32);
72     return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
73 }
74 
75 /** ZSTD_ldm_getChecksum() :
76  *  numBitsToDiscard should be <= 32
77  *  @return : the next most significant 32 bits after numBitsToDiscard */
ZSTD_ldm_getChecksum(U64 hash,U32 numBitsToDiscard)78 static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
79 {
80     assert(numBitsToDiscard <= 32);
81     return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
82 }
83 
84 /** ZSTD_ldm_getTag() ;
85  *  Given the hash, returns the most significant numTagBits bits
86  *  after (32 + hbits) bits.
87  *
88  *  If there are not enough bits remaining, return the last
89  *  numTagBits bits. */
ZSTD_ldm_getTag(U64 hash,U32 hbits,U32 numTagBits)90 static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
91 {
92     assert(numTagBits < 32 && hbits <= 32);
93     if (32 - hbits < numTagBits) {
94         return hash & (((U32)1 << numTagBits) - 1);
95     } else {
96         return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
97     }
98 }
99 
100 /** ZSTD_ldm_getBucket() :
101  *  Returns a pointer to the start of the bucket associated with hash. */
ZSTD_ldm_getBucket(ldmState_t * ldmState,size_t hash,ldmParams_t const ldmParams)102 static ldmEntry_t* ZSTD_ldm_getBucket(
103         ldmState_t* ldmState, size_t hash, ldmParams_t const ldmParams)
104 {
105     return ldmState->hashTable + (hash << ldmParams.bucketSizeLog);
106 }
107 
108 /** ZSTD_ldm_insertEntry() :
109  *  Insert the entry with corresponding hash into the hash table */
ZSTD_ldm_insertEntry(ldmState_t * ldmState,size_t const hash,const ldmEntry_t entry,ldmParams_t const ldmParams)110 static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
111                                  size_t const hash, const ldmEntry_t entry,
112                                  ldmParams_t const ldmParams)
113 {
114     BYTE* const bucketOffsets = ldmState->bucketOffsets;
115     *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
116     bucketOffsets[hash]++;
117     bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
118 }
119 
120 /** ZSTD_ldm_makeEntryAndInsertByTag() :
121  *
122  *  Gets the small hash, checksum, and tag from the rollingHash.
123  *
124  *  If the tag matches (1 << ldmParams.hashRateLog)-1, then
125  *  creates an ldmEntry from the offset, and inserts it into the hash table.
126  *
127  *  hBits is the length of the small hash, which is the most significant hBits
128  *  of rollingHash. The checksum is the next 32 most significant bits, followed
129  *  by ldmParams.hashRateLog bits that make up the tag. */
ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t * ldmState,U64 const rollingHash,U32 const hBits,U32 const offset,ldmParams_t const ldmParams)130 static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
131                                              U64 const rollingHash,
132                                              U32 const hBits,
133                                              U32 const offset,
134                                              ldmParams_t const ldmParams)
135 {
136     U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashRateLog);
137     U32 const tagMask = ((U32)1 << ldmParams.hashRateLog) - 1;
138     if (tag == tagMask) {
139         U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
140         U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
141         ldmEntry_t entry;
142         entry.offset = offset;
143         entry.checksum = checksum;
144         ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
145     }
146 }
147 
148 /** ZSTD_ldm_countBackwardsMatch() :
149  *  Returns the number of bytes that match backwards before pIn and pMatch.
150  *
151  *  We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
ZSTD_ldm_countBackwardsMatch(const BYTE * pIn,const BYTE * pAnchor,const BYTE * pMatch,const BYTE * pBase)152 static size_t ZSTD_ldm_countBackwardsMatch(
153             const BYTE* pIn, const BYTE* pAnchor,
154             const BYTE* pMatch, const BYTE* pBase)
155 {
156     size_t matchLength = 0;
157     while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
158         pIn--;
159         pMatch--;
160         matchLength++;
161     }
162     return matchLength;
163 }
164 
165 /** ZSTD_ldm_fillFastTables() :
166  *
167  *  Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
168  *  This is similar to ZSTD_loadDictionaryContent.
169  *
170  *  The tables for the other strategies are filled within their
171  *  block compressors. */
ZSTD_ldm_fillFastTables(ZSTD_matchState_t * ms,void const * end)172 static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
173                                       void const* end)
174 {
175     const BYTE* const iend = (const BYTE*)end;
176 
177     switch(ms->cParams.strategy)
178     {
179     case ZSTD_fast:
180         ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
181         break;
182 
183     case ZSTD_dfast:
184         ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
185         break;
186 
187     case ZSTD_greedy:
188     case ZSTD_lazy:
189     case ZSTD_lazy2:
190     case ZSTD_btlazy2:
191     case ZSTD_btopt:
192     case ZSTD_btultra:
193     case ZSTD_btultra2:
194         break;
195     default:
196         assert(0);  /* not possible : not a valid strategy id */
197     }
198 
199     return 0;
200 }
201 
202 /** ZSTD_ldm_fillLdmHashTable() :
203  *
204  *  Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
205  *  lastHash is the rolling hash that corresponds to lastHashed.
206  *
207  *  Returns the rolling hash corresponding to position iend-1. */
ZSTD_ldm_fillLdmHashTable(ldmState_t * state,U64 lastHash,const BYTE * lastHashed,const BYTE * iend,const BYTE * base,U32 hBits,ldmParams_t const ldmParams)208 static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
209                                      U64 lastHash, const BYTE* lastHashed,
210                                      const BYTE* iend, const BYTE* base,
211                                      U32 hBits, ldmParams_t const ldmParams)
212 {
213     U64 rollingHash = lastHash;
214     const BYTE* cur = lastHashed + 1;
215 
216     while (cur < iend) {
217         rollingHash = ZSTD_rollingHash_rotate(rollingHash, cur[-1],
218                                               cur[ldmParams.minMatchLength-1],
219                                               state->hashPower);
220         ZSTD_ldm_makeEntryAndInsertByTag(state,
221                                          rollingHash, hBits,
222                                          (U32)(cur - base), ldmParams);
223         ++cur;
224     }
225     return rollingHash;
226 }
227 
ZSTD_ldm_fillHashTable(ldmState_t * state,const BYTE * ip,const BYTE * iend,ldmParams_t const * params)228 void ZSTD_ldm_fillHashTable(
229             ldmState_t* state, const BYTE* ip,
230             const BYTE* iend, ldmParams_t const* params)
231 {
232     DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
233     if ((size_t)(iend - ip) >= params->minMatchLength) {
234         U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
235         ZSTD_ldm_fillLdmHashTable(
236             state, startingHash, ip, iend - params->minMatchLength, state->window.base,
237             params->hashLog - params->bucketSizeLog,
238             *params);
239     }
240 }
241 
242 
243 /** ZSTD_ldm_limitTableUpdate() :
244  *
245  *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
246  *  if it is far way
247  *  (after a long match, only update tables a limited amount). */
ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t * ms,const BYTE * anchor)248 static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
249 {
250     U32 const current = (U32)(anchor - ms->window.base);
251     if (current > ms->nextToUpdate + 1024) {
252         ms->nextToUpdate =
253             current - MIN(512, current - ms->nextToUpdate - 1024);
254     }
255 }
256 
ZSTD_ldm_generateSequences_internal(ldmState_t * ldmState,rawSeqStore_t * rawSeqStore,ldmParams_t const * params,void const * src,size_t srcSize)257 static size_t ZSTD_ldm_generateSequences_internal(
258         ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
259         ldmParams_t const* params, void const* src, size_t srcSize)
260 {
261     /* LDM parameters */
262     int const extDict = ZSTD_window_hasExtDict(ldmState->window);
263     U32 const minMatchLength = params->minMatchLength;
264     U64 const hashPower = ldmState->hashPower;
265     U32 const hBits = params->hashLog - params->bucketSizeLog;
266     U32 const ldmBucketSize = 1U << params->bucketSizeLog;
267     U32 const hashRateLog = params->hashRateLog;
268     U32 const ldmTagMask = (1U << params->hashRateLog) - 1;
269     /* Prefix and extDict parameters */
270     U32 const dictLimit = ldmState->window.dictLimit;
271     U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
272     BYTE const* const base = ldmState->window.base;
273     BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
274     BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
275     BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
276     BYTE const* const lowPrefixPtr = base + dictLimit;
277     /* Input bounds */
278     BYTE const* const istart = (BYTE const*)src;
279     BYTE const* const iend = istart + srcSize;
280     BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
281     /* Input positions */
282     BYTE const* anchor = istart;
283     BYTE const* ip = istart;
284     /* Rolling hash */
285     BYTE const* lastHashed = NULL;
286     U64 rollingHash = 0;
287 
288     while (ip <= ilimit) {
289         size_t mLength;
290         U32 const current = (U32)(ip - base);
291         size_t forwardMatchLength = 0, backwardMatchLength = 0;
292         ldmEntry_t* bestEntry = NULL;
293         if (ip != istart) {
294             rollingHash = ZSTD_rollingHash_rotate(rollingHash, lastHashed[0],
295                                                   lastHashed[minMatchLength],
296                                                   hashPower);
297         } else {
298             rollingHash = ZSTD_rollingHash_compute(ip, minMatchLength);
299         }
300         lastHashed = ip;
301 
302         /* Do not insert and do not look for a match */
303         if (ZSTD_ldm_getTag(rollingHash, hBits, hashRateLog) != ldmTagMask) {
304            ip++;
305            continue;
306         }
307 
308         /* Get the best entry and compute the match lengths */
309         {
310             ldmEntry_t* const bucket =
311                 ZSTD_ldm_getBucket(ldmState,
312                                    ZSTD_ldm_getSmallHash(rollingHash, hBits),
313                                    *params);
314             ldmEntry_t* cur;
315             size_t bestMatchLength = 0;
316             U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
317 
318             for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
319                 size_t curForwardMatchLength, curBackwardMatchLength,
320                        curTotalMatchLength;
321                 if (cur->checksum != checksum || cur->offset <= lowestIndex) {
322                     continue;
323                 }
324                 if (extDict) {
325                     BYTE const* const curMatchBase =
326                         cur->offset < dictLimit ? dictBase : base;
327                     BYTE const* const pMatch = curMatchBase + cur->offset;
328                     BYTE const* const matchEnd =
329                         cur->offset < dictLimit ? dictEnd : iend;
330                     BYTE const* const lowMatchPtr =
331                         cur->offset < dictLimit ? dictStart : lowPrefixPtr;
332 
333                     curForwardMatchLength = ZSTD_count_2segments(
334                                                 ip, pMatch, iend,
335                                                 matchEnd, lowPrefixPtr);
336                     if (curForwardMatchLength < minMatchLength) {
337                         continue;
338                     }
339                     curBackwardMatchLength =
340                         ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
341                                                      lowMatchPtr);
342                     curTotalMatchLength = curForwardMatchLength +
343                                           curBackwardMatchLength;
344                 } else { /* !extDict */
345                     BYTE const* const pMatch = base + cur->offset;
346                     curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
347                     if (curForwardMatchLength < minMatchLength) {
348                         continue;
349                     }
350                     curBackwardMatchLength =
351                         ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
352                                                      lowPrefixPtr);
353                     curTotalMatchLength = curForwardMatchLength +
354                                           curBackwardMatchLength;
355                 }
356 
357                 if (curTotalMatchLength > bestMatchLength) {
358                     bestMatchLength = curTotalMatchLength;
359                     forwardMatchLength = curForwardMatchLength;
360                     backwardMatchLength = curBackwardMatchLength;
361                     bestEntry = cur;
362                 }
363             }
364         }
365 
366         /* No match found -- continue searching */
367         if (bestEntry == NULL) {
368             ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
369                                              hBits, current,
370                                              *params);
371             ip++;
372             continue;
373         }
374 
375         /* Match found */
376         mLength = forwardMatchLength + backwardMatchLength;
377         ip -= backwardMatchLength;
378 
379         {
380             /* Store the sequence:
381              * ip = current - backwardMatchLength
382              * The match is at (bestEntry->offset - backwardMatchLength)
383              */
384             U32 const matchIndex = bestEntry->offset;
385             U32 const offset = current - matchIndex;
386             rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
387 
388             /* Out of sequence storage */
389             if (rawSeqStore->size == rawSeqStore->capacity)
390                 return ERROR(dstSize_tooSmall);
391             seq->litLength = (U32)(ip - anchor);
392             seq->matchLength = (U32)mLength;
393             seq->offset = offset;
394             rawSeqStore->size++;
395         }
396 
397         /* Insert the current entry into the hash table */
398         ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
399                                          (U32)(lastHashed - base),
400                                          *params);
401 
402         assert(ip + backwardMatchLength == lastHashed);
403 
404         /* Fill the hash table from lastHashed+1 to ip+mLength*/
405         /* Heuristic: don't need to fill the entire table at end of block */
406         if (ip + mLength <= ilimit) {
407             rollingHash = ZSTD_ldm_fillLdmHashTable(
408                               ldmState, rollingHash, lastHashed,
409                               ip + mLength, base, hBits, *params);
410             lastHashed = ip + mLength - 1;
411         }
412         ip += mLength;
413         anchor = ip;
414     }
415     return iend - anchor;
416 }
417 
418 /*! ZSTD_ldm_reduceTable() :
419  *  reduce table indexes by `reducerValue` */
ZSTD_ldm_reduceTable(ldmEntry_t * const table,U32 const size,U32 const reducerValue)420 static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
421                                  U32 const reducerValue)
422 {
423     U32 u;
424     for (u = 0; u < size; u++) {
425         if (table[u].offset < reducerValue) table[u].offset = 0;
426         else table[u].offset -= reducerValue;
427     }
428 }
429 
ZSTD_ldm_generateSequences(ldmState_t * ldmState,rawSeqStore_t * sequences,ldmParams_t const * params,void const * src,size_t srcSize)430 size_t ZSTD_ldm_generateSequences(
431         ldmState_t* ldmState, rawSeqStore_t* sequences,
432         ldmParams_t const* params, void const* src, size_t srcSize)
433 {
434     U32 const maxDist = 1U << params->windowLog;
435     BYTE const* const istart = (BYTE const*)src;
436     BYTE const* const iend = istart + srcSize;
437     size_t const kMaxChunkSize = 1 << 20;
438     size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
439     size_t chunk;
440     size_t leftoverSize = 0;
441 
442     assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
443     /* Check that ZSTD_window_update() has been called for this chunk prior
444      * to passing it to this function.
445      */
446     assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
447     /* The input could be very large (in zstdmt), so it must be broken up into
448      * chunks to enforce the maximum distance and handle overflow correction.
449      */
450     assert(sequences->pos <= sequences->size);
451     assert(sequences->size <= sequences->capacity);
452     for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
453         BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
454         size_t const remaining = (size_t)(iend - chunkStart);
455         BYTE const *const chunkEnd =
456             (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
457         size_t const chunkSize = chunkEnd - chunkStart;
458         size_t newLeftoverSize;
459         size_t const prevSize = sequences->size;
460 
461         assert(chunkStart < iend);
462         /* 1. Perform overflow correction if necessary. */
463         if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
464             U32 const ldmHSize = 1U << params->hashLog;
465             U32 const correction = ZSTD_window_correctOverflow(
466                 &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
467             ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
468             /* invalidate dictionaries on overflow correction */
469             ldmState->loadedDictEnd = 0;
470         }
471         /* 2. We enforce the maximum offset allowed.
472          *
473          * kMaxChunkSize should be small enough that we don't lose too much of
474          * the window through early invalidation.
475          * TODO: * Test the chunk size.
476          *       * Try invalidation after the sequence generation and test the
477          *         the offset against maxDist directly.
478          *
479          * NOTE: Because of dictionaries + sequence splitting we MUST make sure
480          * that any offset used is valid at the END of the sequence, since it may
481          * be split into two sequences. This condition holds when using
482          * ZSTD_window_enforceMaxDist(), but if we move to checking offsets
483          * against maxDist directly, we'll have to carefully handle that case.
484          */
485         ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
486         /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
487         newLeftoverSize = ZSTD_ldm_generateSequences_internal(
488             ldmState, sequences, params, chunkStart, chunkSize);
489         if (ZSTD_isError(newLeftoverSize))
490             return newLeftoverSize;
491         /* 4. We add the leftover literals from previous iterations to the first
492          *    newly generated sequence, or add the `newLeftoverSize` if none are
493          *    generated.
494          */
495         /* Prepend the leftover literals from the last call */
496         if (prevSize < sequences->size) {
497             sequences->seq[prevSize].litLength += (U32)leftoverSize;
498             leftoverSize = newLeftoverSize;
499         } else {
500             assert(newLeftoverSize == chunkSize);
501             leftoverSize += chunkSize;
502         }
503     }
504     return 0;
505 }
506 
ZSTD_ldm_skipSequences(rawSeqStore_t * rawSeqStore,size_t srcSize,U32 const minMatch)507 void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
508     while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
509         rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
510         if (srcSize <= seq->litLength) {
511             /* Skip past srcSize literals */
512             seq->litLength -= (U32)srcSize;
513             return;
514         }
515         srcSize -= seq->litLength;
516         seq->litLength = 0;
517         if (srcSize < seq->matchLength) {
518             /* Skip past the first srcSize of the match */
519             seq->matchLength -= (U32)srcSize;
520             if (seq->matchLength < minMatch) {
521                 /* The match is too short, omit it */
522                 if (rawSeqStore->pos + 1 < rawSeqStore->size) {
523                     seq[1].litLength += seq[0].matchLength;
524                 }
525                 rawSeqStore->pos++;
526             }
527             return;
528         }
529         srcSize -= seq->matchLength;
530         seq->matchLength = 0;
531         rawSeqStore->pos++;
532     }
533 }
534 
535 /**
536  * If the sequence length is longer than remaining then the sequence is split
537  * between this block and the next.
538  *
539  * Returns the current sequence to handle, or if the rest of the block should
540  * be literals, it returns a sequence with offset == 0.
541  */
maybeSplitSequence(rawSeqStore_t * rawSeqStore,U32 const remaining,U32 const minMatch)542 static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
543                                  U32 const remaining, U32 const minMatch)
544 {
545     rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
546     assert(sequence.offset > 0);
547     /* Likely: No partial sequence */
548     if (remaining >= sequence.litLength + sequence.matchLength) {
549         rawSeqStore->pos++;
550         return sequence;
551     }
552     /* Cut the sequence short (offset == 0 ==> rest is literals). */
553     if (remaining <= sequence.litLength) {
554         sequence.offset = 0;
555     } else if (remaining < sequence.litLength + sequence.matchLength) {
556         sequence.matchLength = remaining - sequence.litLength;
557         if (sequence.matchLength < minMatch) {
558             sequence.offset = 0;
559         }
560     }
561     /* Skip past `remaining` bytes for the future sequences. */
562     ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
563     return sequence;
564 }
565 
ZSTD_ldm_blockCompress(rawSeqStore_t * rawSeqStore,ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)566 size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
567     ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
568     void const* src, size_t srcSize)
569 {
570     const ZSTD_compressionParameters* const cParams = &ms->cParams;
571     unsigned const minMatch = cParams->minMatch;
572     ZSTD_blockCompressor const blockCompressor =
573         ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
574     /* Input bounds */
575     BYTE const* const istart = (BYTE const*)src;
576     BYTE const* const iend = istart + srcSize;
577     /* Input positions */
578     BYTE const* ip = istart;
579 
580     DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
581     assert(rawSeqStore->pos <= rawSeqStore->size);
582     assert(rawSeqStore->size <= rawSeqStore->capacity);
583     /* Loop through each sequence and apply the block compressor to the lits */
584     while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
585         /* maybeSplitSequence updates rawSeqStore->pos */
586         rawSeq const sequence = maybeSplitSequence(rawSeqStore,
587                                                    (U32)(iend - ip), minMatch);
588         int i;
589         /* End signal */
590         if (sequence.offset == 0)
591             break;
592 
593         assert(ip + sequence.litLength + sequence.matchLength <= iend);
594 
595         /* Fill tables for block compressor */
596         ZSTD_ldm_limitTableUpdate(ms, ip);
597         ZSTD_ldm_fillFastTables(ms, ip);
598         /* Run the block compressor */
599         DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
600         {
601             size_t const newLitLength =
602                 blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
603             ip += sequence.litLength;
604             /* Update the repcodes */
605             for (i = ZSTD_REP_NUM - 1; i > 0; i--)
606                 rep[i] = rep[i-1];
607             rep[0] = sequence.offset;
608             /* Store the sequence */
609             ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
610                           sequence.offset + ZSTD_REP_MOVE,
611                           sequence.matchLength - MINMATCH);
612             ip += sequence.matchLength;
613         }
614     }
615     /* Fill the tables for the block compressor */
616     ZSTD_ldm_limitTableUpdate(ms, ip);
617     ZSTD_ldm_fillFastTables(ms, ip);
618     /* Compress the last literals */
619     return blockCompressor(ms, seqStore, rep, ip, iend - ip);
620 }
621