xref: /linux/lib/zstd/compress/zstd_lazy.c (revision e61f33273ca755b3e2ebee4520a76097199dc7a8)
1 // SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
2 /*
3  * Copyright (c) Meta Platforms, Inc. and affiliates.
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_compress_internal.h"
13 #include "zstd_lazy.h"
14 #include "../common/bits.h" /* ZSTD_countTrailingZeros64 */
15 
16 #if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
17  || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
18  || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
19  || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
20 
21 #define kLazySkippingStep 8
22 
23 
24 /*-*************************************
25 *  Binary Tree search
26 ***************************************/
27 
28 static
29 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_updateDUBT(ZSTD_MatchState_t * ms,const BYTE * ip,const BYTE * iend,U32 mls)30 void ZSTD_updateDUBT(ZSTD_MatchState_t* ms,
31                 const BYTE* ip, const BYTE* iend,
32                 U32 mls)
33 {
34     const ZSTD_compressionParameters* const cParams = &ms->cParams;
35     U32* const hashTable = ms->hashTable;
36     U32  const hashLog = cParams->hashLog;
37 
38     U32* const bt = ms->chainTable;
39     U32  const btLog  = cParams->chainLog - 1;
40     U32  const btMask = (1 << btLog) - 1;
41 
42     const BYTE* const base = ms->window.base;
43     U32 const target = (U32)(ip - base);
44     U32 idx = ms->nextToUpdate;
45 
46     if (idx != target)
47         DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)",
48                     idx, target, ms->window.dictLimit);
49     assert(ip + 8 <= iend);   /* condition for ZSTD_hashPtr */
50     (void)iend;
51 
52     assert(idx >= ms->window.dictLimit);   /* condition for valid base+idx */
53     for ( ; idx < target ; idx++) {
54         size_t const h  = ZSTD_hashPtr(base + idx, hashLog, mls);   /* assumption : ip + 8 <= iend */
55         U32    const matchIndex = hashTable[h];
56 
57         U32*   const nextCandidatePtr = bt + 2*(idx&btMask);
58         U32*   const sortMarkPtr  = nextCandidatePtr + 1;
59 
60         DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx);
61         hashTable[h] = idx;   /* Update Hash Table */
62         *nextCandidatePtr = matchIndex;   /* update BT like a chain */
63         *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK;
64     }
65     ms->nextToUpdate = target;
66 }
67 
68 
69 /* ZSTD_insertDUBT1() :
70  *  sort one already inserted but unsorted position
71  *  assumption : curr >= btlow == (curr - btmask)
72  *  doesn't fail */
73 static
74 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_insertDUBT1(const ZSTD_MatchState_t * ms,U32 curr,const BYTE * inputEnd,U32 nbCompares,U32 btLow,const ZSTD_dictMode_e dictMode)75 void ZSTD_insertDUBT1(const ZSTD_MatchState_t* ms,
76                  U32 curr, const BYTE* inputEnd,
77                  U32 nbCompares, U32 btLow,
78                  const ZSTD_dictMode_e dictMode)
79 {
80     const ZSTD_compressionParameters* const cParams = &ms->cParams;
81     U32* const bt = ms->chainTable;
82     U32  const btLog  = cParams->chainLog - 1;
83     U32  const btMask = (1 << btLog) - 1;
84     size_t commonLengthSmaller=0, commonLengthLarger=0;
85     const BYTE* const base = ms->window.base;
86     const BYTE* const dictBase = ms->window.dictBase;
87     const U32 dictLimit = ms->window.dictLimit;
88     const BYTE* const ip = (curr>=dictLimit) ? base + curr : dictBase + curr;
89     const BYTE* const iend = (curr>=dictLimit) ? inputEnd : dictBase + dictLimit;
90     const BYTE* const dictEnd = dictBase + dictLimit;
91     const BYTE* const prefixStart = base + dictLimit;
92     const BYTE* match;
93     U32* smallerPtr = bt + 2*(curr&btMask);
94     U32* largerPtr  = smallerPtr + 1;
95     U32 matchIndex = *smallerPtr;   /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */
96     U32 dummy32;   /* to be nullified at the end */
97     U32 const windowValid = ms->window.lowLimit;
98     U32 const maxDistance = 1U << cParams->windowLog;
99     U32 const windowLow = (curr - windowValid > maxDistance) ? curr - maxDistance : windowValid;
100 
101 
102     DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
103                 curr, dictLimit, windowLow);
104     assert(curr >= btLow);
105     assert(ip < iend);   /* condition for ZSTD_count */
106 
107     for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
108         U32* const nextPtr = bt + 2*(matchIndex & btMask);
109         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
110         assert(matchIndex < curr);
111         /* note : all candidates are now supposed sorted,
112          * but it's still possible to have nextPtr[1] == ZSTD_DUBT_UNSORTED_MARK
113          * when a real index has the same value as ZSTD_DUBT_UNSORTED_MARK */
114 
115         if ( (dictMode != ZSTD_extDict)
116           || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
117           || (curr < dictLimit) /* both in extDict */) {
118             const BYTE* const mBase = ( (dictMode != ZSTD_extDict)
119                                      || (matchIndex+matchLength >= dictLimit)) ?
120                                         base : dictBase;
121             assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
122                  || (curr < dictLimit) );
123             match = mBase + matchIndex;
124             matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
125         } else {
126             match = dictBase + matchIndex;
127             matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
128             if (matchIndex+matchLength >= dictLimit)
129                 match = base + matchIndex;   /* preparation for next read of match[matchLength] */
130         }
131 
132         DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
133                     curr, matchIndex, (U32)matchLength);
134 
135         if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
136             break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
137         }
138 
139         if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
140             /* match is smaller than current */
141             *smallerPtr = matchIndex;             /* update smaller idx */
142             commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
143             if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
144             DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u",
145                         matchIndex, btLow, nextPtr[1]);
146             smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
147             matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
148         } else {
149             /* match is larger than current */
150             *largerPtr = matchIndex;
151             commonLengthLarger = matchLength;
152             if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
153             DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u",
154                         matchIndex, btLow, nextPtr[0]);
155             largerPtr = nextPtr;
156             matchIndex = nextPtr[0];
157     }   }
158 
159     *smallerPtr = *largerPtr = 0;
160 }
161 
162 
163 static
164 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_DUBT_findBetterDictMatch(const ZSTD_MatchState_t * ms,const BYTE * const ip,const BYTE * const iend,size_t * offsetPtr,size_t bestLength,U32 nbCompares,U32 const mls,const ZSTD_dictMode_e dictMode)165 size_t ZSTD_DUBT_findBetterDictMatch (
166         const ZSTD_MatchState_t* ms,
167         const BYTE* const ip, const BYTE* const iend,
168         size_t* offsetPtr,
169         size_t bestLength,
170         U32 nbCompares,
171         U32 const mls,
172         const ZSTD_dictMode_e dictMode)
173 {
174     const ZSTD_MatchState_t * const dms = ms->dictMatchState;
175     const ZSTD_compressionParameters* const dmsCParams = &dms->cParams;
176     const U32 * const dictHashTable = dms->hashTable;
177     U32         const hashLog = dmsCParams->hashLog;
178     size_t      const h  = ZSTD_hashPtr(ip, hashLog, mls);
179     U32               dictMatchIndex = dictHashTable[h];
180 
181     const BYTE* const base = ms->window.base;
182     const BYTE* const prefixStart = base + ms->window.dictLimit;
183     U32         const curr = (U32)(ip-base);
184     const BYTE* const dictBase = dms->window.base;
185     const BYTE* const dictEnd = dms->window.nextSrc;
186     U32         const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base);
187     U32         const dictLowLimit = dms->window.lowLimit;
188     U32         const dictIndexDelta = ms->window.lowLimit - dictHighLimit;
189 
190     U32*        const dictBt = dms->chainTable;
191     U32         const btLog  = dmsCParams->chainLog - 1;
192     U32         const btMask = (1 << btLog) - 1;
193     U32         const btLow = (btMask >= dictHighLimit - dictLowLimit) ? dictLowLimit : dictHighLimit - btMask;
194 
195     size_t commonLengthSmaller=0, commonLengthLarger=0;
196 
197     (void)dictMode;
198     assert(dictMode == ZSTD_dictMatchState);
199 
200     for (; nbCompares && (dictMatchIndex > dictLowLimit); --nbCompares) {
201         U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
202         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
203         const BYTE* match = dictBase + dictMatchIndex;
204         matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
205         if (dictMatchIndex+matchLength >= dictHighLimit)
206             match = base + dictMatchIndex + dictIndexDelta;   /* to prepare for next usage of match[matchLength] */
207 
208         if (matchLength > bestLength) {
209             U32 matchIndex = dictMatchIndex + dictIndexDelta;
210             if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
211                 DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
212                     curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, OFFSET_TO_OFFBASE(curr - matchIndex), dictMatchIndex, matchIndex);
213                 bestLength = matchLength, *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
214             }
215             if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
216                 break;   /* drop, to guarantee consistency (miss a little bit of compression) */
217             }
218         }
219 
220         if (match[matchLength] < ip[matchLength]) {
221             if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
222             commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
223             dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
224         } else {
225             /* match is larger than current */
226             if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
227             commonLengthLarger = matchLength;
228             dictMatchIndex = nextPtr[0];
229         }
230     }
231 
232     if (bestLength >= MINMATCH) {
233         U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offsetPtr); (void)mIndex;
234         DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
235                     curr, (U32)bestLength, (U32)*offsetPtr, mIndex);
236     }
237     return bestLength;
238 
239 }
240 
241 
242 static
243 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_DUBT_findBestMatch(ZSTD_MatchState_t * ms,const BYTE * const ip,const BYTE * const iend,size_t * offBasePtr,U32 const mls,const ZSTD_dictMode_e dictMode)244 size_t ZSTD_DUBT_findBestMatch(ZSTD_MatchState_t* ms,
245                         const BYTE* const ip, const BYTE* const iend,
246                         size_t* offBasePtr,
247                         U32 const mls,
248                         const ZSTD_dictMode_e dictMode)
249 {
250     const ZSTD_compressionParameters* const cParams = &ms->cParams;
251     U32*   const hashTable = ms->hashTable;
252     U32    const hashLog = cParams->hashLog;
253     size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
254     U32          matchIndex  = hashTable[h];
255 
256     const BYTE* const base = ms->window.base;
257     U32    const curr = (U32)(ip-base);
258     U32    const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
259 
260     U32*   const bt = ms->chainTable;
261     U32    const btLog  = cParams->chainLog - 1;
262     U32    const btMask = (1 << btLog) - 1;
263     U32    const btLow = (btMask >= curr) ? 0 : curr - btMask;
264     U32    const unsortLimit = MAX(btLow, windowLow);
265 
266     U32*         nextCandidate = bt + 2*(matchIndex&btMask);
267     U32*         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
268     U32          nbCompares = 1U << cParams->searchLog;
269     U32          nbCandidates = nbCompares;
270     U32          previousCandidate = 0;
271 
272     DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", curr);
273     assert(ip <= iend-8);   /* required for h calculation */
274     assert(dictMode != ZSTD_dedicatedDictSearch);
275 
276     /* reach end of unsorted candidates list */
277     while ( (matchIndex > unsortLimit)
278          && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK)
279          && (nbCandidates > 1) ) {
280         DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted",
281                     matchIndex);
282         *unsortedMark = previousCandidate;  /* the unsortedMark becomes a reversed chain, to move up back to original position */
283         previousCandidate = matchIndex;
284         matchIndex = *nextCandidate;
285         nextCandidate = bt + 2*(matchIndex&btMask);
286         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
287         nbCandidates --;
288     }
289 
290     /* nullify last candidate if it's still unsorted
291      * simplification, detrimental to compression ratio, beneficial for speed */
292     if ( (matchIndex > unsortLimit)
293       && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) {
294         DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u",
295                     matchIndex);
296         *nextCandidate = *unsortedMark = 0;
297     }
298 
299     /* batch sort stacked candidates */
300     matchIndex = previousCandidate;
301     while (matchIndex) {  /* will end on matchIndex == 0 */
302         U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
303         U32 const nextCandidateIdx = *nextCandidateIdxPtr;
304         ZSTD_insertDUBT1(ms, matchIndex, iend,
305                          nbCandidates, unsortLimit, dictMode);
306         matchIndex = nextCandidateIdx;
307         nbCandidates++;
308     }
309 
310     /* find longest match */
311     {   size_t commonLengthSmaller = 0, commonLengthLarger = 0;
312         const BYTE* const dictBase = ms->window.dictBase;
313         const U32 dictLimit = ms->window.dictLimit;
314         const BYTE* const dictEnd = dictBase + dictLimit;
315         const BYTE* const prefixStart = base + dictLimit;
316         U32* smallerPtr = bt + 2*(curr&btMask);
317         U32* largerPtr  = bt + 2*(curr&btMask) + 1;
318         U32 matchEndIdx = curr + 8 + 1;
319         U32 dummy32;   /* to be nullified at the end */
320         size_t bestLength = 0;
321 
322         matchIndex  = hashTable[h];
323         hashTable[h] = curr;   /* Update Hash Table */
324 
325         for (; nbCompares && (matchIndex > windowLow); --nbCompares) {
326             U32* const nextPtr = bt + 2*(matchIndex & btMask);
327             size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
328             const BYTE* match;
329 
330             if ((dictMode != ZSTD_extDict) || (matchIndex+matchLength >= dictLimit)) {
331                 match = base + matchIndex;
332                 matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
333             } else {
334                 match = dictBase + matchIndex;
335                 matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
336                 if (matchIndex+matchLength >= dictLimit)
337                     match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
338             }
339 
340             if (matchLength > bestLength) {
341                 if (matchLength > matchEndIdx - matchIndex)
342                     matchEndIdx = matchIndex + (U32)matchLength;
343                 if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr - matchIndex + 1) - ZSTD_highbit32((U32)*offBasePtr)) )
344                     bestLength = matchLength, *offBasePtr = OFFSET_TO_OFFBASE(curr - matchIndex);
345                 if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
346                     if (dictMode == ZSTD_dictMatchState) {
347                         nbCompares = 0; /* in addition to avoiding checking any
348                                          * further in this loop, make sure we
349                                          * skip checking in the dictionary. */
350                     }
351                     break;   /* drop, to guarantee consistency (miss a little bit of compression) */
352                 }
353             }
354 
355             if (match[matchLength] < ip[matchLength]) {
356                 /* match is smaller than current */
357                 *smallerPtr = matchIndex;             /* update smaller idx */
358                 commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
359                 if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
360                 smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
361                 matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
362             } else {
363                 /* match is larger than current */
364                 *largerPtr = matchIndex;
365                 commonLengthLarger = matchLength;
366                 if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
367                 largerPtr = nextPtr;
368                 matchIndex = nextPtr[0];
369         }   }
370 
371         *smallerPtr = *largerPtr = 0;
372 
373         assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
374         if (dictMode == ZSTD_dictMatchState && nbCompares) {
375             bestLength = ZSTD_DUBT_findBetterDictMatch(
376                     ms, ip, iend,
377                     offBasePtr, bestLength, nbCompares,
378                     mls, dictMode);
379         }
380 
381         assert(matchEndIdx > curr+8); /* ensure nextToUpdate is increased */
382         ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
383         if (bestLength >= MINMATCH) {
384             U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offBasePtr); (void)mIndex;
385             DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
386                         curr, (U32)bestLength, (U32)*offBasePtr, mIndex);
387         }
388         return bestLength;
389     }
390 }
391 
392 
393 /* ZSTD_BtFindBestMatch() : Tree updater, providing best match */
394 FORCE_INLINE_TEMPLATE
395 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_BtFindBestMatch(ZSTD_MatchState_t * ms,const BYTE * const ip,const BYTE * const iLimit,size_t * offBasePtr,const U32 mls,const ZSTD_dictMode_e dictMode)396 size_t ZSTD_BtFindBestMatch( ZSTD_MatchState_t* ms,
397                 const BYTE* const ip, const BYTE* const iLimit,
398                       size_t* offBasePtr,
399                 const U32 mls /* template */,
400                 const ZSTD_dictMode_e dictMode)
401 {
402     DEBUGLOG(7, "ZSTD_BtFindBestMatch");
403     if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
404     ZSTD_updateDUBT(ms, ip, iLimit, mls);
405     return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offBasePtr, mls, dictMode);
406 }
407 
408 /* *********************************
409 * Dedicated dict search
410 ***********************************/
411 
ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_MatchState_t * ms,const BYTE * const ip)412 void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_MatchState_t* ms, const BYTE* const ip)
413 {
414     const BYTE* const base = ms->window.base;
415     U32 const target = (U32)(ip - base);
416     U32* const hashTable = ms->hashTable;
417     U32* const chainTable = ms->chainTable;
418     U32 const chainSize = 1 << ms->cParams.chainLog;
419     U32 idx = ms->nextToUpdate;
420     U32 const minChain = chainSize < target - idx ? target - chainSize : idx;
421     U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG;
422     U32 const cacheSize = bucketSize - 1;
423     U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize;
424     U32 const chainLimit = chainAttempts > 255 ? 255 : chainAttempts;
425 
426     /* We know the hashtable is oversized by a factor of `bucketSize`.
427      * We are going to temporarily pretend `bucketSize == 1`, keeping only a
428      * single entry. We will use the rest of the space to construct a temporary
429      * chaintable.
430      */
431     U32 const hashLog = ms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
432     U32* const tmpHashTable = hashTable;
433     U32* const tmpChainTable = hashTable + ((size_t)1 << hashLog);
434     U32 const tmpChainSize = (U32)((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog;
435     U32 const tmpMinChain = tmpChainSize < target ? target - tmpChainSize : idx;
436     U32 hashIdx;
437 
438     assert(ms->cParams.chainLog <= 24);
439     assert(ms->cParams.hashLog > ms->cParams.chainLog);
440     assert(idx != 0);
441     assert(tmpMinChain <= minChain);
442 
443     /* fill conventional hash table and conventional chain table */
444     for ( ; idx < target; idx++) {
445         U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch);
446         if (idx >= tmpMinChain) {
447             tmpChainTable[idx - tmpMinChain] = hashTable[h];
448         }
449         tmpHashTable[h] = idx;
450     }
451 
452     /* sort chains into ddss chain table */
453     {
454         U32 chainPos = 0;
455         for (hashIdx = 0; hashIdx < (1U << hashLog); hashIdx++) {
456             U32 count;
457             U32 countBeyondMinChain = 0;
458             U32 i = tmpHashTable[hashIdx];
459             for (count = 0; i >= tmpMinChain && count < cacheSize; count++) {
460                 /* skip through the chain to the first position that won't be
461                  * in the hash cache bucket */
462                 if (i < minChain) {
463                     countBeyondMinChain++;
464                 }
465                 i = tmpChainTable[i - tmpMinChain];
466             }
467             if (count == cacheSize) {
468                 for (count = 0; count < chainLimit;) {
469                     if (i < minChain) {
470                         if (!i || ++countBeyondMinChain > cacheSize) {
471                             /* only allow pulling `cacheSize` number of entries
472                              * into the cache or chainTable beyond `minChain`,
473                              * to replace the entries pulled out of the
474                              * chainTable into the cache. This lets us reach
475                              * back further without increasing the total number
476                              * of entries in the chainTable, guaranteeing the
477                              * DDSS chain table will fit into the space
478                              * allocated for the regular one. */
479                             break;
480                         }
481                     }
482                     chainTable[chainPos++] = i;
483                     count++;
484                     if (i < tmpMinChain) {
485                         break;
486                     }
487                     i = tmpChainTable[i - tmpMinChain];
488                 }
489             } else {
490                 count = 0;
491             }
492             if (count) {
493                 tmpHashTable[hashIdx] = ((chainPos - count) << 8) + count;
494             } else {
495                 tmpHashTable[hashIdx] = 0;
496             }
497         }
498         assert(chainPos <= chainSize); /* I believe this is guaranteed... */
499     }
500 
501     /* move chain pointers into the last entry of each hash bucket */
502     for (hashIdx = (1 << hashLog); hashIdx; ) {
503         U32 const bucketIdx = --hashIdx << ZSTD_LAZY_DDSS_BUCKET_LOG;
504         U32 const chainPackedPointer = tmpHashTable[hashIdx];
505         U32 i;
506         for (i = 0; i < cacheSize; i++) {
507             hashTable[bucketIdx + i] = 0;
508         }
509         hashTable[bucketIdx + bucketSize - 1] = chainPackedPointer;
510     }
511 
512     /* fill the buckets of the hash table */
513     for (idx = ms->nextToUpdate; idx < target; idx++) {
514         U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch)
515                    << ZSTD_LAZY_DDSS_BUCKET_LOG;
516         U32 i;
517         /* Shift hash cache down 1. */
518         for (i = cacheSize - 1; i; i--)
519             hashTable[h + i] = hashTable[h + i - 1];
520         hashTable[h] = idx;
521     }
522 
523     ms->nextToUpdate = target;
524 }
525 
526 /* Returns the longest match length found in the dedicated dict search structure.
527  * If none are longer than the argument ml, then ml will be returned.
528  */
529 FORCE_INLINE_TEMPLATE
ZSTD_dedicatedDictSearch_lazy_search(size_t * offsetPtr,size_t ml,U32 nbAttempts,const ZSTD_MatchState_t * const dms,const BYTE * const ip,const BYTE * const iLimit,const BYTE * const prefixStart,const U32 curr,const U32 dictLimit,const size_t ddsIdx)530 size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nbAttempts,
531                                             const ZSTD_MatchState_t* const dms,
532                                             const BYTE* const ip, const BYTE* const iLimit,
533                                             const BYTE* const prefixStart, const U32 curr,
534                                             const U32 dictLimit, const size_t ddsIdx) {
535     const U32 ddsLowestIndex  = dms->window.dictLimit;
536     const BYTE* const ddsBase = dms->window.base;
537     const BYTE* const ddsEnd  = dms->window.nextSrc;
538     const U32 ddsSize         = (U32)(ddsEnd - ddsBase);
539     const U32 ddsIndexDelta   = dictLimit - ddsSize;
540     const U32 bucketSize      = (1 << ZSTD_LAZY_DDSS_BUCKET_LOG);
541     const U32 bucketLimit     = nbAttempts < bucketSize - 1 ? nbAttempts : bucketSize - 1;
542     U32 ddsAttempt;
543     U32 matchIndex;
544 
545     for (ddsAttempt = 0; ddsAttempt < bucketSize - 1; ddsAttempt++) {
546         PREFETCH_L1(ddsBase + dms->hashTable[ddsIdx + ddsAttempt]);
547     }
548 
549     {
550         U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
551         U32 const chainIndex = chainPackedPointer >> 8;
552 
553         PREFETCH_L1(&dms->chainTable[chainIndex]);
554     }
555 
556     for (ddsAttempt = 0; ddsAttempt < bucketLimit; ddsAttempt++) {
557         size_t currentMl=0;
558         const BYTE* match;
559         matchIndex = dms->hashTable[ddsIdx + ddsAttempt];
560         match = ddsBase + matchIndex;
561 
562         if (!matchIndex) {
563             return ml;
564         }
565 
566         /* guaranteed by table construction */
567         (void)ddsLowestIndex;
568         assert(matchIndex >= ddsLowestIndex);
569         assert(match+4 <= ddsEnd);
570         if (MEM_read32(match) == MEM_read32(ip)) {
571             /* assumption : matchIndex <= dictLimit-4 (by table construction) */
572             currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
573         }
574 
575         /* save best solution */
576         if (currentMl > ml) {
577             ml = currentMl;
578             *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
579             if (ip+currentMl == iLimit) {
580                 /* best possible, avoids read overflow on next attempt */
581                 return ml;
582             }
583         }
584     }
585 
586     {
587         U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1];
588         U32 chainIndex = chainPackedPointer >> 8;
589         U32 const chainLength = chainPackedPointer & 0xFF;
590         U32 const chainAttempts = nbAttempts - ddsAttempt;
591         U32 const chainLimit = chainAttempts > chainLength ? chainLength : chainAttempts;
592         U32 chainAttempt;
593 
594         for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++) {
595             PREFETCH_L1(ddsBase + dms->chainTable[chainIndex + chainAttempt]);
596         }
597 
598         for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++, chainIndex++) {
599             size_t currentMl=0;
600             const BYTE* match;
601             matchIndex = dms->chainTable[chainIndex];
602             match = ddsBase + matchIndex;
603 
604             /* guaranteed by table construction */
605             assert(matchIndex >= ddsLowestIndex);
606             assert(match+4 <= ddsEnd);
607             if (MEM_read32(match) == MEM_read32(ip)) {
608                 /* assumption : matchIndex <= dictLimit-4 (by table construction) */
609                 currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4;
610             }
611 
612             /* save best solution */
613             if (currentMl > ml) {
614                 ml = currentMl;
615                 *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta));
616                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
617             }
618         }
619     }
620     return ml;
621 }
622 
623 
624 /* *********************************
625 *  Hash Chain
626 ***********************************/
627 #define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & (mask)]
628 
629 /* Update chains up to ip (excluded)
630    Assumption : always within prefix (i.e. not within extDict) */
631 FORCE_INLINE_TEMPLATE
632 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_insertAndFindFirstIndex_internal(ZSTD_MatchState_t * ms,const ZSTD_compressionParameters * const cParams,const BYTE * ip,U32 const mls,U32 const lazySkipping)633 U32 ZSTD_insertAndFindFirstIndex_internal(
634                         ZSTD_MatchState_t* ms,
635                         const ZSTD_compressionParameters* const cParams,
636                         const BYTE* ip, U32 const mls, U32 const lazySkipping)
637 {
638     U32* const hashTable  = ms->hashTable;
639     const U32 hashLog = cParams->hashLog;
640     U32* const chainTable = ms->chainTable;
641     const U32 chainMask = (1 << cParams->chainLog) - 1;
642     const BYTE* const base = ms->window.base;
643     const U32 target = (U32)(ip - base);
644     U32 idx = ms->nextToUpdate;
645 
646     while(idx < target) { /* catch up */
647         size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
648         NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
649         hashTable[h] = idx;
650         idx++;
651         /* Stop inserting every position when in the lazy skipping mode. */
652         if (lazySkipping)
653             break;
654     }
655 
656     ms->nextToUpdate = target;
657     return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
658 }
659 
ZSTD_insertAndFindFirstIndex(ZSTD_MatchState_t * ms,const BYTE * ip)660 U32 ZSTD_insertAndFindFirstIndex(ZSTD_MatchState_t* ms, const BYTE* ip) {
661     const ZSTD_compressionParameters* const cParams = &ms->cParams;
662     return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch, /* lazySkipping*/ 0);
663 }
664 
665 /* inlining is important to hardwire a hot branch (template emulation) */
666 FORCE_INLINE_TEMPLATE
667 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_HcFindBestMatch(ZSTD_MatchState_t * ms,const BYTE * const ip,const BYTE * const iLimit,size_t * offsetPtr,const U32 mls,const ZSTD_dictMode_e dictMode)668 size_t ZSTD_HcFindBestMatch(
669                         ZSTD_MatchState_t* ms,
670                         const BYTE* const ip, const BYTE* const iLimit,
671                         size_t* offsetPtr,
672                         const U32 mls, const ZSTD_dictMode_e dictMode)
673 {
674     const ZSTD_compressionParameters* const cParams = &ms->cParams;
675     U32* const chainTable = ms->chainTable;
676     const U32 chainSize = (1 << cParams->chainLog);
677     const U32 chainMask = chainSize-1;
678     const BYTE* const base = ms->window.base;
679     const BYTE* const dictBase = ms->window.dictBase;
680     const U32 dictLimit = ms->window.dictLimit;
681     const BYTE* const prefixStart = base + dictLimit;
682     const BYTE* const dictEnd = dictBase + dictLimit;
683     const U32 curr = (U32)(ip-base);
684     const U32 maxDistance = 1U << cParams->windowLog;
685     const U32 lowestValid = ms->window.lowLimit;
686     const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
687     const U32 isDictionary = (ms->loadedDictEnd != 0);
688     const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
689     const U32 minChain = curr > chainSize ? curr - chainSize : 0;
690     U32 nbAttempts = 1U << cParams->searchLog;
691     size_t ml=4-1;
692 
693     const ZSTD_MatchState_t* const dms = ms->dictMatchState;
694     const U32 ddsHashLog = dictMode == ZSTD_dedicatedDictSearch
695                          ? dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
696     const size_t ddsIdx = dictMode == ZSTD_dedicatedDictSearch
697                         ? ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG : 0;
698 
699     U32 matchIndex;
700 
701     if (dictMode == ZSTD_dedicatedDictSearch) {
702         const U32* entry = &dms->hashTable[ddsIdx];
703         PREFETCH_L1(entry);
704     }
705 
706     /* HC4 match finder */
707     matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls, ms->lazySkipping);
708 
709     for ( ; (matchIndex>=lowLimit) & (nbAttempts>0) ; nbAttempts--) {
710         size_t currentMl=0;
711         if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
712             const BYTE* const match = base + matchIndex;
713             assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
714             /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
715             if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
716                 currentMl = ZSTD_count(ip, match, iLimit);
717         } else {
718             const BYTE* const match = dictBase + matchIndex;
719             assert(match+4 <= dictEnd);
720             if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
721                 currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
722         }
723 
724         /* save best solution */
725         if (currentMl > ml) {
726             ml = currentMl;
727             *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
728             if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
729         }
730 
731         if (matchIndex <= minChain) break;
732         matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
733     }
734 
735     assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
736     if (dictMode == ZSTD_dedicatedDictSearch) {
737         ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts, dms,
738                                                   ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
739     } else if (dictMode == ZSTD_dictMatchState) {
740         const U32* const dmsChainTable = dms->chainTable;
741         const U32 dmsChainSize         = (1 << dms->cParams.chainLog);
742         const U32 dmsChainMask         = dmsChainSize - 1;
743         const U32 dmsLowestIndex       = dms->window.dictLimit;
744         const BYTE* const dmsBase      = dms->window.base;
745         const BYTE* const dmsEnd       = dms->window.nextSrc;
746         const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
747         const U32 dmsIndexDelta        = dictLimit - dmsSize;
748         const U32 dmsMinChain = dmsSize > dmsChainSize ? dmsSize - dmsChainSize : 0;
749 
750         matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)];
751 
752         for ( ; (matchIndex>=dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
753             size_t currentMl=0;
754             const BYTE* const match = dmsBase + matchIndex;
755             assert(match+4 <= dmsEnd);
756             if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
757                 currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
758 
759             /* save best solution */
760             if (currentMl > ml) {
761                 ml = currentMl;
762                 assert(curr > matchIndex + dmsIndexDelta);
763                 *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
764                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
765             }
766 
767             if (matchIndex <= dmsMinChain) break;
768 
769             matchIndex = dmsChainTable[matchIndex & dmsChainMask];
770         }
771     }
772 
773     return ml;
774 }
775 
776 /* *********************************
777 * (SIMD) Row-based matchfinder
778 ***********************************/
779 /* Constants for row-based hash */
780 #define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1)
781 #define ZSTD_ROW_HASH_MAX_ENTRIES 64    /* absolute maximum number of entries per row, for all configurations */
782 
783 #define ZSTD_ROW_HASH_CACHE_MASK (ZSTD_ROW_HASH_CACHE_SIZE - 1)
784 
785 typedef U64 ZSTD_VecMask;   /* Clarifies when we are interacting with a U64 representing a mask of matches */
786 
787 /* ZSTD_VecMask_next():
788  * Starting from the LSB, returns the idx of the next non-zero bit.
789  * Basically counting the nb of trailing zeroes.
790  */
ZSTD_VecMask_next(ZSTD_VecMask val)791 MEM_STATIC U32 ZSTD_VecMask_next(ZSTD_VecMask val) {
792     return ZSTD_countTrailingZeros64(val);
793 }
794 
795 /* ZSTD_row_nextIndex():
796  * Returns the next index to insert at within a tagTable row, and updates the "head"
797  * value to reflect the update. Essentially cycles backwards from [1, {entries per row})
798  */
ZSTD_row_nextIndex(BYTE * const tagRow,U32 const rowMask)799 FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextIndex(BYTE* const tagRow, U32 const rowMask) {
800     U32 next = (*tagRow-1) & rowMask;
801     next += (next == 0) ? rowMask : 0; /* skip first position */
802     *tagRow = (BYTE)next;
803     return next;
804 }
805 
806 /* ZSTD_isAligned():
807  * Checks that a pointer is aligned to "align" bytes which must be a power of 2.
808  */
ZSTD_isAligned(void const * ptr,size_t align)809 MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) {
810     assert((align & (align - 1)) == 0);
811     return (((size_t)ptr) & (align - 1)) == 0;
812 }
813 
814 /* ZSTD_row_prefetch():
815  * Performs prefetching for the hashTable and tagTable at a given row.
816  */
ZSTD_row_prefetch(U32 const * hashTable,BYTE const * tagTable,U32 const relRow,U32 const rowLog)817 FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, BYTE const* tagTable, U32 const relRow, U32 const rowLog) {
818     PREFETCH_L1(hashTable + relRow);
819     if (rowLog >= 5) {
820         PREFETCH_L1(hashTable + relRow + 16);
821         /* Note: prefetching more of the hash table does not appear to be beneficial for 128-entry rows */
822     }
823     PREFETCH_L1(tagTable + relRow);
824     if (rowLog == 6) {
825         PREFETCH_L1(tagTable + relRow + 32);
826     }
827     assert(rowLog == 4 || rowLog == 5 || rowLog == 6);
828     assert(ZSTD_isAligned(hashTable + relRow, 64));                 /* prefetched hash row always 64-byte aligned */
829     assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on correct multiple of bytes (32,64,128) */
830 }
831 
832 /* ZSTD_row_fillHashCache():
833  * Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries,
834  * but not beyond iLimit.
835  */
836 FORCE_INLINE_TEMPLATE
837 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_row_fillHashCache(ZSTD_MatchState_t * ms,const BYTE * base,U32 const rowLog,U32 const mls,U32 idx,const BYTE * const iLimit)838 void ZSTD_row_fillHashCache(ZSTD_MatchState_t* ms, const BYTE* base,
839                                    U32 const rowLog, U32 const mls,
840                                    U32 idx, const BYTE* const iLimit)
841 {
842     U32 const* const hashTable = ms->hashTable;
843     BYTE const* const tagTable = ms->tagTable;
844     U32 const hashLog = ms->rowHashLog;
845     U32 const maxElemsToPrefetch = (base + idx) > iLimit ? 0 : (U32)(iLimit - (base + idx) + 1);
846     U32 const lim = idx + MIN(ZSTD_ROW_HASH_CACHE_SIZE, maxElemsToPrefetch);
847 
848     for (; idx < lim; ++idx) {
849         U32 const hash = (U32)ZSTD_hashPtrSalted(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
850         U32 const row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
851         ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
852         ms->hashCache[idx & ZSTD_ROW_HASH_CACHE_MASK] = hash;
853     }
854 
855     DEBUGLOG(6, "ZSTD_row_fillHashCache(): [%u %u %u %u %u %u %u %u]", ms->hashCache[0], ms->hashCache[1],
856                                                      ms->hashCache[2], ms->hashCache[3], ms->hashCache[4],
857                                                      ms->hashCache[5], ms->hashCache[6], ms->hashCache[7]);
858 }
859 
860 /* ZSTD_row_nextCachedHash():
861  * Returns the hash of base + idx, and replaces the hash in the hash cache with the byte at
862  * base + idx + ZSTD_ROW_HASH_CACHE_SIZE. Also prefetches the appropriate rows from hashTable and tagTable.
863  */
864 FORCE_INLINE_TEMPLATE
865 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_row_nextCachedHash(U32 * cache,U32 const * hashTable,BYTE const * tagTable,BYTE const * base,U32 idx,U32 const hashLog,U32 const rowLog,U32 const mls,U64 const hashSalt)866 U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTable,
867                                                   BYTE const* tagTable, BYTE const* base,
868                                                   U32 idx, U32 const hashLog,
869                                                   U32 const rowLog, U32 const mls,
870                                                   U64 const hashSalt)
871 {
872     U32 const newHash = (U32)ZSTD_hashPtrSalted(base+idx+ZSTD_ROW_HASH_CACHE_SIZE, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
873     U32 const row = (newHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
874     ZSTD_row_prefetch(hashTable, tagTable, row, rowLog);
875     {   U32 const hash = cache[idx & ZSTD_ROW_HASH_CACHE_MASK];
876         cache[idx & ZSTD_ROW_HASH_CACHE_MASK] = newHash;
877         return hash;
878     }
879 }
880 
881 /* ZSTD_row_update_internalImpl():
882  * Updates the hash table with positions starting from updateStartIdx until updateEndIdx.
883  */
884 FORCE_INLINE_TEMPLATE
885 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_row_update_internalImpl(ZSTD_MatchState_t * ms,U32 updateStartIdx,U32 const updateEndIdx,U32 const mls,U32 const rowLog,U32 const rowMask,U32 const useCache)886 void ZSTD_row_update_internalImpl(ZSTD_MatchState_t* ms,
887                                   U32 updateStartIdx, U32 const updateEndIdx,
888                                   U32 const mls, U32 const rowLog,
889                                   U32 const rowMask, U32 const useCache)
890 {
891     U32* const hashTable = ms->hashTable;
892     BYTE* const tagTable = ms->tagTable;
893     U32 const hashLog = ms->rowHashLog;
894     const BYTE* const base = ms->window.base;
895 
896     DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx);
897     for (; updateStartIdx < updateEndIdx; ++updateStartIdx) {
898         U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls, ms->hashSalt)
899                                   : (U32)ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt);
900         U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
901         U32* const row = hashTable + relRow;
902         BYTE* tagRow = tagTable + relRow;
903         U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
904 
905         assert(hash == ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt));
906         tagRow[pos] = hash & ZSTD_ROW_HASH_TAG_MASK;
907         row[pos] = updateStartIdx;
908     }
909 }
910 
911 /* ZSTD_row_update_internal():
912  * Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate.
913  * Skips sections of long matches as is necessary.
914  */
915 FORCE_INLINE_TEMPLATE
916 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_row_update_internal(ZSTD_MatchState_t * ms,const BYTE * ip,U32 const mls,U32 const rowLog,U32 const rowMask,U32 const useCache)917 void ZSTD_row_update_internal(ZSTD_MatchState_t* ms, const BYTE* ip,
918                               U32 const mls, U32 const rowLog,
919                               U32 const rowMask, U32 const useCache)
920 {
921     U32 idx = ms->nextToUpdate;
922     const BYTE* const base = ms->window.base;
923     const U32 target = (U32)(ip - base);
924     const U32 kSkipThreshold = 384;
925     const U32 kMaxMatchStartPositionsToUpdate = 96;
926     const U32 kMaxMatchEndPositionsToUpdate = 32;
927 
928     if (useCache) {
929         /* Only skip positions when using hash cache, i.e.
930          * if we are loading a dict, don't skip anything.
931          * If we decide to skip, then we only update a set number
932          * of positions at the beginning and end of the match.
933          */
934         if (UNLIKELY(target - idx > kSkipThreshold)) {
935             U32 const bound = idx + kMaxMatchStartPositionsToUpdate;
936             ZSTD_row_update_internalImpl(ms, idx, bound, mls, rowLog, rowMask, useCache);
937             idx = target - kMaxMatchEndPositionsToUpdate;
938             ZSTD_row_fillHashCache(ms, base, rowLog, mls, idx, ip+1);
939         }
940     }
941     assert(target >= idx);
942     ZSTD_row_update_internalImpl(ms, idx, target, mls, rowLog, rowMask, useCache);
943     ms->nextToUpdate = target;
944 }
945 
946 /* ZSTD_row_update():
947  * External wrapper for ZSTD_row_update_internal(). Used for filling the hashtable during dictionary
948  * processing.
949  */
ZSTD_row_update(ZSTD_MatchState_t * const ms,const BYTE * ip)950 void ZSTD_row_update(ZSTD_MatchState_t* const ms, const BYTE* ip) {
951     const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
952     const U32 rowMask = (1u << rowLog) - 1;
953     const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */);
954 
955     DEBUGLOG(5, "ZSTD_row_update(), rowLog=%u", rowLog);
956     ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* don't use cache */);
957 }
958 
959 /* Returns the mask width of bits group of which will be set to 1. Given not all
960  * architectures have easy movemask instruction, this helps to iterate over
961  * groups of bits easier and faster.
962  */
963 FORCE_INLINE_TEMPLATE U32
ZSTD_row_matchMaskGroupWidth(const U32 rowEntries)964 ZSTD_row_matchMaskGroupWidth(const U32 rowEntries)
965 {
966     assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
967     assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
968     (void)rowEntries;
969 #if defined(ZSTD_ARCH_ARM_NEON)
970     /* NEON path only works for little endian */
971     if (!MEM_isLittleEndian()) {
972         return 1;
973     }
974     if (rowEntries == 16) {
975         return 4;
976     }
977     if (rowEntries == 32) {
978         return 2;
979     }
980     if (rowEntries == 64) {
981         return 1;
982     }
983 #endif
984     return 1;
985 }
986 
987 #if defined(ZSTD_ARCH_X86_SSE2)
988 FORCE_INLINE_TEMPLATE ZSTD_VecMask
ZSTD_row_getSSEMask(int nbChunks,const BYTE * const src,const BYTE tag,const U32 head)989 ZSTD_row_getSSEMask(int nbChunks, const BYTE* const src, const BYTE tag, const U32 head)
990 {
991     const __m128i comparisonMask = _mm_set1_epi8((char)tag);
992     int matches[4] = {0};
993     int i;
994     assert(nbChunks == 1 || nbChunks == 2 || nbChunks == 4);
995     for (i=0; i<nbChunks; i++) {
996         const __m128i chunk = _mm_loadu_si128((const __m128i*)(const void*)(src + 16*i));
997         const __m128i equalMask = _mm_cmpeq_epi8(chunk, comparisonMask);
998         matches[i] = _mm_movemask_epi8(equalMask);
999     }
1000     if (nbChunks == 1) return ZSTD_rotateRight_U16((U16)matches[0], head);
1001     if (nbChunks == 2) return ZSTD_rotateRight_U32((U32)matches[1] << 16 | (U32)matches[0], head);
1002     assert(nbChunks == 4);
1003     return ZSTD_rotateRight_U64((U64)matches[3] << 48 | (U64)matches[2] << 32 | (U64)matches[1] << 16 | (U64)matches[0], head);
1004 }
1005 #endif
1006 
1007 #if defined(ZSTD_ARCH_ARM_NEON)
1008 FORCE_INLINE_TEMPLATE ZSTD_VecMask
ZSTD_row_getNEONMask(const U32 rowEntries,const BYTE * const src,const BYTE tag,const U32 headGrouped)1009 ZSTD_row_getNEONMask(const U32 rowEntries, const BYTE* const src, const BYTE tag, const U32 headGrouped)
1010 {
1011     assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
1012     if (rowEntries == 16) {
1013         /* vshrn_n_u16 shifts by 4 every u16 and narrows to 8 lower bits.
1014          * After that groups of 4 bits represent the equalMask. We lower
1015          * all bits except the highest in these groups by doing AND with
1016          * 0x88 = 0b10001000.
1017          */
1018         const uint8x16_t chunk = vld1q_u8(src);
1019         const uint16x8_t equalMask = vreinterpretq_u16_u8(vceqq_u8(chunk, vdupq_n_u8(tag)));
1020         const uint8x8_t res = vshrn_n_u16(equalMask, 4);
1021         const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0);
1022         return ZSTD_rotateRight_U64(matches, headGrouped) & 0x8888888888888888ull;
1023     } else if (rowEntries == 32) {
1024         /* Same idea as with rowEntries == 16 but doing AND with
1025          * 0x55 = 0b01010101.
1026          */
1027         const uint16x8x2_t chunk = vld2q_u16((const uint16_t*)(const void*)src);
1028         const uint8x16_t chunk0 = vreinterpretq_u8_u16(chunk.val[0]);
1029         const uint8x16_t chunk1 = vreinterpretq_u8_u16(chunk.val[1]);
1030         const uint8x16_t dup = vdupq_n_u8(tag);
1031         const uint8x8_t t0 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk0, dup)), 6);
1032         const uint8x8_t t1 = vshrn_n_u16(vreinterpretq_u16_u8(vceqq_u8(chunk1, dup)), 6);
1033         const uint8x8_t res = vsli_n_u8(t0, t1, 4);
1034         const U64 matches = vget_lane_u64(vreinterpret_u64_u8(res), 0) ;
1035         return ZSTD_rotateRight_U64(matches, headGrouped) & 0x5555555555555555ull;
1036     } else { /* rowEntries == 64 */
1037         const uint8x16x4_t chunk = vld4q_u8(src);
1038         const uint8x16_t dup = vdupq_n_u8(tag);
1039         const uint8x16_t cmp0 = vceqq_u8(chunk.val[0], dup);
1040         const uint8x16_t cmp1 = vceqq_u8(chunk.val[1], dup);
1041         const uint8x16_t cmp2 = vceqq_u8(chunk.val[2], dup);
1042         const uint8x16_t cmp3 = vceqq_u8(chunk.val[3], dup);
1043 
1044         const uint8x16_t t0 = vsriq_n_u8(cmp1, cmp0, 1);
1045         const uint8x16_t t1 = vsriq_n_u8(cmp3, cmp2, 1);
1046         const uint8x16_t t2 = vsriq_n_u8(t1, t0, 2);
1047         const uint8x16_t t3 = vsriq_n_u8(t2, t2, 4);
1048         const uint8x8_t t4 = vshrn_n_u16(vreinterpretq_u16_u8(t3), 4);
1049         const U64 matches = vget_lane_u64(vreinterpret_u64_u8(t4), 0);
1050         return ZSTD_rotateRight_U64(matches, headGrouped);
1051     }
1052 }
1053 #endif
1054 
1055 /* Returns a ZSTD_VecMask (U64) that has the nth group (determined by
1056  * ZSTD_row_matchMaskGroupWidth) of bits set to 1 if the newly-computed "tag"
1057  * matches the hash at the nth position in a row of the tagTable.
1058  * Each row is a circular buffer beginning at the value of "headGrouped". So we
1059  * must rotate the "matches" bitfield to match up with the actual layout of the
1060  * entries within the hashTable */
1061 FORCE_INLINE_TEMPLATE ZSTD_VecMask
ZSTD_row_getMatchMask(const BYTE * const tagRow,const BYTE tag,const U32 headGrouped,const U32 rowEntries)1062 ZSTD_row_getMatchMask(const BYTE* const tagRow, const BYTE tag, const U32 headGrouped, const U32 rowEntries)
1063 {
1064     const BYTE* const src = tagRow;
1065     assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64);
1066     assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES);
1067     assert(ZSTD_row_matchMaskGroupWidth(rowEntries) * rowEntries <= sizeof(ZSTD_VecMask) * 8);
1068 
1069 #if defined(ZSTD_ARCH_X86_SSE2)
1070 
1071     return ZSTD_row_getSSEMask(rowEntries / 16, src, tag, headGrouped);
1072 
1073 #else /* SW or NEON-LE */
1074 
1075 # if defined(ZSTD_ARCH_ARM_NEON)
1076   /* This NEON path only works for little endian - otherwise use SWAR below */
1077     if (MEM_isLittleEndian()) {
1078         return ZSTD_row_getNEONMask(rowEntries, src, tag, headGrouped);
1079     }
1080 # endif /* ZSTD_ARCH_ARM_NEON */
1081     /* SWAR */
1082     {   const int chunkSize = sizeof(size_t);
1083         const size_t shiftAmount = ((chunkSize * 8) - chunkSize);
1084         const size_t xFF = ~((size_t)0);
1085         const size_t x01 = xFF / 0xFF;
1086         const size_t x80 = x01 << 7;
1087         const size_t splatChar = tag * x01;
1088         ZSTD_VecMask matches = 0;
1089         int i = rowEntries - chunkSize;
1090         assert((sizeof(size_t) == 4) || (sizeof(size_t) == 8));
1091         if (MEM_isLittleEndian()) { /* runtime check so have two loops */
1092             const size_t extractMagic = (xFF / 0x7F) >> chunkSize;
1093             do {
1094                 size_t chunk = MEM_readST(&src[i]);
1095                 chunk ^= splatChar;
1096                 chunk = (((chunk | x80) - x01) | chunk) & x80;
1097                 matches <<= chunkSize;
1098                 matches |= (chunk * extractMagic) >> shiftAmount;
1099                 i -= chunkSize;
1100             } while (i >= 0);
1101         } else { /* big endian: reverse bits during extraction */
1102             const size_t msb = xFF ^ (xFF >> 1);
1103             const size_t extractMagic = (msb / 0x1FF) | msb;
1104             do {
1105                 size_t chunk = MEM_readST(&src[i]);
1106                 chunk ^= splatChar;
1107                 chunk = (((chunk | x80) - x01) | chunk) & x80;
1108                 matches <<= chunkSize;
1109                 matches |= ((chunk >> 7) * extractMagic) >> shiftAmount;
1110                 i -= chunkSize;
1111             } while (i >= 0);
1112         }
1113         matches = ~matches;
1114         if (rowEntries == 16) {
1115             return ZSTD_rotateRight_U16((U16)matches, headGrouped);
1116         } else if (rowEntries == 32) {
1117             return ZSTD_rotateRight_U32((U32)matches, headGrouped);
1118         } else {
1119             return ZSTD_rotateRight_U64((U64)matches, headGrouped);
1120         }
1121     }
1122 #endif
1123 }
1124 
1125 /* The high-level approach of the SIMD row based match finder is as follows:
1126  * - Figure out where to insert the new entry:
1127  *      - Generate a hash for current input position and split it into a one byte of tag and `rowHashLog` bits of index.
1128  *           - The hash is salted by a value that changes on every context reset, so when the same table is used
1129  *             we will avoid collisions that would otherwise slow us down by introducing phantom matches.
1130  *      - The hashTable is effectively split into groups or "rows" of 15 or 31 entries of U32, and the index determines
1131  *        which row to insert into.
1132  *      - Determine the correct position within the row to insert the entry into. Each row of 15 or 31 can
1133  *        be considered as a circular buffer with a "head" index that resides in the tagTable (overall 16 or 32 bytes
1134  *        per row).
1135  * - Use SIMD to efficiently compare the tags in the tagTable to the 1-byte tag calculated for the position and
1136  *   generate a bitfield that we can cycle through to check the collisions in the hash table.
1137  * - Pick the longest match.
1138  * - Insert the tag into the equivalent row and position in the tagTable.
1139  */
1140 FORCE_INLINE_TEMPLATE
1141 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_RowFindBestMatch(ZSTD_MatchState_t * ms,const BYTE * const ip,const BYTE * const iLimit,size_t * offsetPtr,const U32 mls,const ZSTD_dictMode_e dictMode,const U32 rowLog)1142 size_t ZSTD_RowFindBestMatch(
1143                         ZSTD_MatchState_t* ms,
1144                         const BYTE* const ip, const BYTE* const iLimit,
1145                         size_t* offsetPtr,
1146                         const U32 mls, const ZSTD_dictMode_e dictMode,
1147                         const U32 rowLog)
1148 {
1149     U32* const hashTable = ms->hashTable;
1150     BYTE* const tagTable = ms->tagTable;
1151     U32* const hashCache = ms->hashCache;
1152     const U32 hashLog = ms->rowHashLog;
1153     const ZSTD_compressionParameters* const cParams = &ms->cParams;
1154     const BYTE* const base = ms->window.base;
1155     const BYTE* const dictBase = ms->window.dictBase;
1156     const U32 dictLimit = ms->window.dictLimit;
1157     const BYTE* const prefixStart = base + dictLimit;
1158     const BYTE* const dictEnd = dictBase + dictLimit;
1159     const U32 curr = (U32)(ip-base);
1160     const U32 maxDistance = 1U << cParams->windowLog;
1161     const U32 lowestValid = ms->window.lowLimit;
1162     const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
1163     const U32 isDictionary = (ms->loadedDictEnd != 0);
1164     const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
1165     const U32 rowEntries = (1U << rowLog);
1166     const U32 rowMask = rowEntries - 1;
1167     const U32 cappedSearchLog = MIN(cParams->searchLog, rowLog); /* nb of searches is capped at nb entries per row */
1168     const U32 groupWidth = ZSTD_row_matchMaskGroupWidth(rowEntries);
1169     const U64 hashSalt = ms->hashSalt;
1170     U32 nbAttempts = 1U << cappedSearchLog;
1171     size_t ml=4-1;
1172     U32 hash;
1173 
1174     /* DMS/DDS variables that may be referenced laster */
1175     const ZSTD_MatchState_t* const dms = ms->dictMatchState;
1176 
1177     /* Initialize the following variables to satisfy static analyzer */
1178     size_t ddsIdx = 0;
1179     U32 ddsExtraAttempts = 0; /* cctx hash tables are limited in searches, but allow extra searches into DDS */
1180     U32 dmsTag = 0;
1181     U32* dmsRow = NULL;
1182     BYTE* dmsTagRow = NULL;
1183 
1184     if (dictMode == ZSTD_dedicatedDictSearch) {
1185         const U32 ddsHashLog = dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
1186         {   /* Prefetch DDS hashtable entry */
1187             ddsIdx = ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG;
1188             PREFETCH_L1(&dms->hashTable[ddsIdx]);
1189         }
1190         ddsExtraAttempts = cParams->searchLog > rowLog ? 1U << (cParams->searchLog - rowLog) : 0;
1191     }
1192 
1193     if (dictMode == ZSTD_dictMatchState) {
1194         /* Prefetch DMS rows */
1195         U32* const dmsHashTable = dms->hashTable;
1196         BYTE* const dmsTagTable = dms->tagTable;
1197         U32 const dmsHash = (U32)ZSTD_hashPtr(ip, dms->rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls);
1198         U32 const dmsRelRow = (dmsHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
1199         dmsTag = dmsHash & ZSTD_ROW_HASH_TAG_MASK;
1200         dmsTagRow = (BYTE*)(dmsTagTable + dmsRelRow);
1201         dmsRow = dmsHashTable + dmsRelRow;
1202         ZSTD_row_prefetch(dmsHashTable, dmsTagTable, dmsRelRow, rowLog);
1203     }
1204 
1205     /* Update the hashTable and tagTable up to (but not including) ip */
1206     if (!ms->lazySkipping) {
1207         ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 1 /* useCache */);
1208         hash = ZSTD_row_nextCachedHash(hashCache, hashTable, tagTable, base, curr, hashLog, rowLog, mls, hashSalt);
1209     } else {
1210         /* Stop inserting every position when in the lazy skipping mode.
1211          * The hash cache is also not kept up to date in this mode.
1212          */
1213         hash = (U32)ZSTD_hashPtrSalted(ip, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt);
1214         ms->nextToUpdate = curr;
1215     }
1216     ms->hashSaltEntropy += hash; /* collect salt entropy */
1217 
1218     {   /* Get the hash for ip, compute the appropriate row */
1219         U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog;
1220         U32 const tag = hash & ZSTD_ROW_HASH_TAG_MASK;
1221         U32* const row = hashTable + relRow;
1222         BYTE* tagRow = (BYTE*)(tagTable + relRow);
1223         U32 const headGrouped = (*tagRow & rowMask) * groupWidth;
1224         U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
1225         size_t numMatches = 0;
1226         size_t currMatch = 0;
1227         ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, headGrouped, rowEntries);
1228 
1229         /* Cycle through the matches and prefetch */
1230         for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
1231             U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
1232             U32 const matchIndex = row[matchPos];
1233             if(matchPos == 0) continue;
1234             assert(numMatches < rowEntries);
1235             if (matchIndex < lowLimit)
1236                 break;
1237             if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
1238                 PREFETCH_L1(base + matchIndex);
1239             } else {
1240                 PREFETCH_L1(dictBase + matchIndex);
1241             }
1242             matchBuffer[numMatches++] = matchIndex;
1243             --nbAttempts;
1244         }
1245 
1246         /* Speed opt: insert current byte into hashtable too. This allows us to avoid one iteration of the loop
1247            in ZSTD_row_update_internal() at the next search. */
1248         {
1249             U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask);
1250             tagRow[pos] = (BYTE)tag;
1251             row[pos] = ms->nextToUpdate++;
1252         }
1253 
1254         /* Return the longest match */
1255         for (; currMatch < numMatches; ++currMatch) {
1256             U32 const matchIndex = matchBuffer[currMatch];
1257             size_t currentMl=0;
1258             assert(matchIndex < curr);
1259             assert(matchIndex >= lowLimit);
1260 
1261             if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
1262                 const BYTE* const match = base + matchIndex;
1263                 assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
1264                 /* read 4B starting from (match + ml + 1 - sizeof(U32)) */
1265                 if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3))   /* potentially better */
1266                     currentMl = ZSTD_count(ip, match, iLimit);
1267             } else {
1268                 const BYTE* const match = dictBase + matchIndex;
1269                 assert(match+4 <= dictEnd);
1270                 if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
1271                     currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
1272             }
1273 
1274             /* Save best solution */
1275             if (currentMl > ml) {
1276                 ml = currentMl;
1277                 *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex);
1278                 if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
1279             }
1280         }
1281     }
1282 
1283     assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */
1284     if (dictMode == ZSTD_dedicatedDictSearch) {
1285         ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts + ddsExtraAttempts, dms,
1286                                                   ip, iLimit, prefixStart, curr, dictLimit, ddsIdx);
1287     } else if (dictMode == ZSTD_dictMatchState) {
1288         /* TODO: Measure and potentially add prefetching to DMS */
1289         const U32 dmsLowestIndex       = dms->window.dictLimit;
1290         const BYTE* const dmsBase      = dms->window.base;
1291         const BYTE* const dmsEnd       = dms->window.nextSrc;
1292         const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
1293         const U32 dmsIndexDelta        = dictLimit - dmsSize;
1294 
1295         {   U32 const headGrouped = (*dmsTagRow & rowMask) * groupWidth;
1296             U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES];
1297             size_t numMatches = 0;
1298             size_t currMatch = 0;
1299             ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, headGrouped, rowEntries);
1300 
1301             for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) {
1302                 U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask;
1303                 U32 const matchIndex = dmsRow[matchPos];
1304                 if(matchPos == 0) continue;
1305                 if (matchIndex < dmsLowestIndex)
1306                     break;
1307                 PREFETCH_L1(dmsBase + matchIndex);
1308                 matchBuffer[numMatches++] = matchIndex;
1309                 --nbAttempts;
1310             }
1311 
1312             /* Return the longest match */
1313             for (; currMatch < numMatches; ++currMatch) {
1314                 U32 const matchIndex = matchBuffer[currMatch];
1315                 size_t currentMl=0;
1316                 assert(matchIndex >= dmsLowestIndex);
1317                 assert(matchIndex < curr);
1318 
1319                 {   const BYTE* const match = dmsBase + matchIndex;
1320                     assert(match+4 <= dmsEnd);
1321                     if (MEM_read32(match) == MEM_read32(ip))
1322                         currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
1323                 }
1324 
1325                 if (currentMl > ml) {
1326                     ml = currentMl;
1327                     assert(curr > matchIndex + dmsIndexDelta);
1328                     *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta));
1329                     if (ip+currentMl == iLimit) break;
1330                 }
1331             }
1332         }
1333     }
1334     return ml;
1335 }
1336 
1337 
1338 /*
1339  * Generate search functions templated on (dictMode, mls, rowLog).
1340  * These functions are outlined for code size & compilation time.
1341  * ZSTD_searchMax() dispatches to the correct implementation function.
1342  *
1343  * TODO: The start of the search function involves loading and calculating a
1344  * bunch of constants from the ZSTD_MatchState_t. These computations could be
1345  * done in an initialization function, and saved somewhere in the match state.
1346  * Then we could pass a pointer to the saved state instead of the match state,
1347  * and avoid duplicate computations.
1348  *
1349  * TODO: Move the match re-winding into searchMax. This improves compression
1350  * ratio, and unlocks further simplifications with the next TODO.
1351  *
1352  * TODO: Try moving the repcode search into searchMax. After the re-winding
1353  * and repcode search are in searchMax, there is no more logic in the match
1354  * finder loop that requires knowledge about the dictMode. So we should be
1355  * able to avoid force inlining it, and we can join the extDict loop with
1356  * the single segment loop. It should go in searchMax instead of its own
1357  * function to avoid having multiple virtual function calls per search.
1358  */
1359 
1360 #define ZSTD_BT_SEARCH_FN(dictMode, mls) ZSTD_BtFindBestMatch_##dictMode##_##mls
1361 #define ZSTD_HC_SEARCH_FN(dictMode, mls) ZSTD_HcFindBestMatch_##dictMode##_##mls
1362 #define ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog) ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog
1363 
1364 #define ZSTD_SEARCH_FN_ATTRS FORCE_NOINLINE
1365 
1366 #define GEN_ZSTD_BT_SEARCH_FN(dictMode, mls)                                           \
1367     ZSTD_SEARCH_FN_ATTRS size_t ZSTD_BT_SEARCH_FN(dictMode, mls)(                      \
1368             ZSTD_MatchState_t* ms,                                                     \
1369             const BYTE* ip, const BYTE* const iLimit,                                  \
1370             size_t* offBasePtr)                                                        \
1371     {                                                                                  \
1372         assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                           \
1373         return ZSTD_BtFindBestMatch(ms, ip, iLimit, offBasePtr, mls, ZSTD_##dictMode); \
1374     }                                                                                  \
1375 
1376 #define GEN_ZSTD_HC_SEARCH_FN(dictMode, mls)                                          \
1377     ZSTD_SEARCH_FN_ATTRS size_t ZSTD_HC_SEARCH_FN(dictMode, mls)(                     \
1378             ZSTD_MatchState_t* ms,                                                    \
1379             const BYTE* ip, const BYTE* const iLimit,                                 \
1380             size_t* offsetPtr)                                                        \
1381     {                                                                                 \
1382         assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                          \
1383         return ZSTD_HcFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \
1384     }                                                                                 \
1385 
1386 #define GEN_ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)                                          \
1387     ZSTD_SEARCH_FN_ATTRS size_t ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(                     \
1388             ZSTD_MatchState_t* ms,                                                             \
1389             const BYTE* ip, const BYTE* const iLimit,                                          \
1390             size_t* offsetPtr)                                                                 \
1391     {                                                                                          \
1392         assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls);                                   \
1393         assert(MAX(4, MIN(6, ms->cParams.searchLog)) == rowLog);                               \
1394         return ZSTD_RowFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode, rowLog); \
1395     }                                                                                          \
1396 
1397 #define ZSTD_FOR_EACH_ROWLOG(X, dictMode, mls) \
1398     X(dictMode, mls, 4)                        \
1399     X(dictMode, mls, 5)                        \
1400     X(dictMode, mls, 6)
1401 
1402 #define ZSTD_FOR_EACH_MLS_ROWLOG(X, dictMode) \
1403     ZSTD_FOR_EACH_ROWLOG(X, dictMode, 4)      \
1404     ZSTD_FOR_EACH_ROWLOG(X, dictMode, 5)      \
1405     ZSTD_FOR_EACH_ROWLOG(X, dictMode, 6)
1406 
1407 #define ZSTD_FOR_EACH_MLS(X, dictMode) \
1408     X(dictMode, 4)                     \
1409     X(dictMode, 5)                     \
1410     X(dictMode, 6)
1411 
1412 #define ZSTD_FOR_EACH_DICT_MODE(X, ...) \
1413     X(__VA_ARGS__, noDict)              \
1414     X(__VA_ARGS__, extDict)             \
1415     X(__VA_ARGS__, dictMatchState)      \
1416     X(__VA_ARGS__, dedicatedDictSearch)
1417 
1418 /* Generate row search fns for each combination of (dictMode, mls, rowLog) */
1419 ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS_ROWLOG, GEN_ZSTD_ROW_SEARCH_FN)
1420 /* Generate binary Tree search fns for each combination of (dictMode, mls) */
1421 ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_BT_SEARCH_FN)
1422 /* Generate hash chain search fns for each combination of (dictMode, mls) */
1423 ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_HC_SEARCH_FN)
1424 
1425 typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e;
1426 
1427 #define GEN_ZSTD_CALL_BT_SEARCH_FN(dictMode, mls)                         \
1428     case mls:                                                             \
1429         return ZSTD_BT_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
1430 #define GEN_ZSTD_CALL_HC_SEARCH_FN(dictMode, mls)                         \
1431     case mls:                                                             \
1432         return ZSTD_HC_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr);
1433 #define GEN_ZSTD_CALL_ROW_SEARCH_FN(dictMode, mls, rowLog)                         \
1434     case rowLog:                                                                   \
1435         return ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(ms, ip, iend, offsetPtr);
1436 
1437 #define ZSTD_SWITCH_MLS(X, dictMode)   \
1438     switch (mls) {                     \
1439         ZSTD_FOR_EACH_MLS(X, dictMode) \
1440     }
1441 
1442 #define ZSTD_SWITCH_ROWLOG(dictMode, mls)                                    \
1443     case mls:                                                                \
1444         switch (rowLog) {                                                    \
1445             ZSTD_FOR_EACH_ROWLOG(GEN_ZSTD_CALL_ROW_SEARCH_FN, dictMode, mls) \
1446         }                                                                    \
1447         ZSTD_UNREACHABLE;                                                    \
1448         break;
1449 
1450 #define ZSTD_SWITCH_SEARCH_METHOD(dictMode)                       \
1451     switch (searchMethod) {                                       \
1452         case search_hashChain:                                    \
1453             ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_HC_SEARCH_FN, dictMode) \
1454             break;                                                \
1455         case search_binaryTree:                                   \
1456             ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_BT_SEARCH_FN, dictMode) \
1457             break;                                                \
1458         case search_rowHash:                                      \
1459             ZSTD_SWITCH_MLS(ZSTD_SWITCH_ROWLOG, dictMode)         \
1460             break;                                                \
1461     }                                                             \
1462     ZSTD_UNREACHABLE;
1463 
1464 /*
1465  * Searches for the longest match at @p ip.
1466  * Dispatches to the correct implementation function based on the
1467  * (searchMethod, dictMode, mls, rowLog). We use switch statements
1468  * here instead of using an indirect function call through a function
1469  * pointer because after Spectre and Meltdown mitigations, indirect
1470  * function calls can be very costly, especially in the kernel.
1471  *
1472  * NOTE: dictMode and searchMethod should be templated, so those switch
1473  * statements should be optimized out. Only the mls & rowLog switches
1474  * should be left.
1475  *
1476  * @param ms The match state.
1477  * @param ip The position to search at.
1478  * @param iend The end of the input data.
1479  * @param[out] offsetPtr Stores the match offset into this pointer.
1480  * @param mls The minimum search length, in the range [4, 6].
1481  * @param rowLog The row log (if applicable), in the range [4, 6].
1482  * @param searchMethod The search method to use (templated).
1483  * @param dictMode The dictMode (templated).
1484  *
1485  * @returns The length of the longest match found, or < mls if no match is found.
1486  * If a match is found its offset is stored in @p offsetPtr.
1487  */
ZSTD_searchMax(ZSTD_MatchState_t * ms,const BYTE * ip,const BYTE * iend,size_t * offsetPtr,U32 const mls,U32 const rowLog,searchMethod_e const searchMethod,ZSTD_dictMode_e const dictMode)1488 FORCE_INLINE_TEMPLATE size_t ZSTD_searchMax(
1489     ZSTD_MatchState_t* ms,
1490     const BYTE* ip,
1491     const BYTE* iend,
1492     size_t* offsetPtr,
1493     U32 const mls,
1494     U32 const rowLog,
1495     searchMethod_e const searchMethod,
1496     ZSTD_dictMode_e const dictMode)
1497 {
1498     if (dictMode == ZSTD_noDict) {
1499         ZSTD_SWITCH_SEARCH_METHOD(noDict)
1500     } else if (dictMode == ZSTD_extDict) {
1501         ZSTD_SWITCH_SEARCH_METHOD(extDict)
1502     } else if (dictMode == ZSTD_dictMatchState) {
1503         ZSTD_SWITCH_SEARCH_METHOD(dictMatchState)
1504     } else if (dictMode == ZSTD_dedicatedDictSearch) {
1505         ZSTD_SWITCH_SEARCH_METHOD(dedicatedDictSearch)
1506     }
1507     ZSTD_UNREACHABLE;
1508     return 0;
1509 }
1510 
1511 /* *******************************
1512 *  Common parser - lazy strategy
1513 *********************************/
1514 
1515 FORCE_INLINE_TEMPLATE
1516 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_compressBlock_lazy_generic(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],const void * src,size_t srcSize,const searchMethod_e searchMethod,const U32 depth,ZSTD_dictMode_e const dictMode)1517 size_t ZSTD_compressBlock_lazy_generic(
1518                         ZSTD_MatchState_t* ms, SeqStore_t* seqStore,
1519                         U32 rep[ZSTD_REP_NUM],
1520                         const void* src, size_t srcSize,
1521                         const searchMethod_e searchMethod, const U32 depth,
1522                         ZSTD_dictMode_e const dictMode)
1523 {
1524     const BYTE* const istart = (const BYTE*)src;
1525     const BYTE* ip = istart;
1526     const BYTE* anchor = istart;
1527     const BYTE* const iend = istart + srcSize;
1528     const BYTE* const ilimit = (searchMethod == search_rowHash) ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
1529     const BYTE* const base = ms->window.base;
1530     const U32 prefixLowestIndex = ms->window.dictLimit;
1531     const BYTE* const prefixLowest = base + prefixLowestIndex;
1532     const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
1533     const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
1534 
1535     U32 offset_1 = rep[0], offset_2 = rep[1];
1536     U32 offsetSaved1 = 0, offsetSaved2 = 0;
1537 
1538     const int isDMS = dictMode == ZSTD_dictMatchState;
1539     const int isDDS = dictMode == ZSTD_dedicatedDictSearch;
1540     const int isDxS = isDMS || isDDS;
1541     const ZSTD_MatchState_t* const dms = ms->dictMatchState;
1542     const U32 dictLowestIndex      = isDxS ? dms->window.dictLimit : 0;
1543     const BYTE* const dictBase     = isDxS ? dms->window.base : NULL;
1544     const BYTE* const dictLowest   = isDxS ? dictBase + dictLowestIndex : NULL;
1545     const BYTE* const dictEnd      = isDxS ? dms->window.nextSrc : NULL;
1546     const U32 dictIndexDelta       = isDxS ?
1547                                      prefixLowestIndex - (U32)(dictEnd - dictBase) :
1548                                      0;
1549     const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
1550 
1551     DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u) (searchFunc=%u)", (U32)dictMode, (U32)searchMethod);
1552     ip += (dictAndPrefixLength == 0);
1553     if (dictMode == ZSTD_noDict) {
1554         U32 const curr = (U32)(ip - base);
1555         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog);
1556         U32 const maxRep = curr - windowLow;
1557         if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
1558         if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
1559     }
1560     if (isDxS) {
1561         /* dictMatchState repCode checks don't currently handle repCode == 0
1562          * disabling. */
1563         assert(offset_1 <= dictAndPrefixLength);
1564         assert(offset_2 <= dictAndPrefixLength);
1565     }
1566 
1567     /* Reset the lazy skipping state */
1568     ms->lazySkipping = 0;
1569 
1570     if (searchMethod == search_rowHash) {
1571         ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
1572     }
1573 
1574     /* Match Loop */
1575 #if defined(__x86_64__)
1576     /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
1577      * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
1578      */
1579     __asm__(".p2align 5");
1580 #endif
1581     while (ip < ilimit) {
1582         size_t matchLength=0;
1583         size_t offBase = REPCODE1_TO_OFFBASE;
1584         const BYTE* start=ip+1;
1585         DEBUGLOG(7, "search baseline (depth 0)");
1586 
1587         /* check repCode */
1588         if (isDxS) {
1589             const U32 repIndex = (U32)(ip - base) + 1 - offset_1;
1590             const BYTE* repMatch = ((dictMode == ZSTD_dictMatchState || dictMode == ZSTD_dedicatedDictSearch)
1591                                 && repIndex < prefixLowestIndex) ?
1592                                    dictBase + (repIndex - dictIndexDelta) :
1593                                    base + repIndex;
1594             if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
1595                 && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
1596                 const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
1597                 matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
1598                 if (depth==0) goto _storeSequence;
1599             }
1600         }
1601         if ( dictMode == ZSTD_noDict
1602           && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
1603             matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
1604             if (depth==0) goto _storeSequence;
1605         }
1606 
1607         /* first search (depth 0) */
1608         {   size_t offbaseFound = 999999999;
1609             size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offbaseFound, mls, rowLog, searchMethod, dictMode);
1610             if (ml2 > matchLength)
1611                 matchLength = ml2, start = ip, offBase = offbaseFound;
1612         }
1613 
1614         if (matchLength < 4) {
1615             size_t const step = ((size_t)(ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */;
1616             ip += step;
1617             /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
1618              * In this mode we stop inserting every position into our tables, and only insert
1619              * positions that we search, which is one in step positions.
1620              * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
1621              * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
1622              * triggered once we've gone 2KB without finding any matches.
1623              */
1624             ms->lazySkipping = step > kLazySkippingStep;
1625             continue;
1626         }
1627 
1628         /* let's try to find a better solution */
1629         if (depth>=1)
1630         while (ip<ilimit) {
1631             DEBUGLOG(7, "search depth 1");
1632             ip ++;
1633             if ( (dictMode == ZSTD_noDict)
1634               && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
1635                 size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
1636                 int const gain2 = (int)(mlRep * 3);
1637                 int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
1638                 if ((mlRep >= 4) && (gain2 > gain1))
1639                     matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
1640             }
1641             if (isDxS) {
1642                 const U32 repIndex = (U32)(ip - base) - offset_1;
1643                 const BYTE* repMatch = repIndex < prefixLowestIndex ?
1644                                dictBase + (repIndex - dictIndexDelta) :
1645                                base + repIndex;
1646                 if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
1647                     && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
1648                     const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
1649                     size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
1650                     int const gain2 = (int)(mlRep * 3);
1651                     int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
1652                     if ((mlRep >= 4) && (gain2 > gain1))
1653                         matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
1654                 }
1655             }
1656             {   size_t ofbCandidate=999999999;
1657                 size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
1658                 int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
1659                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
1660                 if ((ml2 >= 4) && (gain2 > gain1)) {
1661                     matchLength = ml2, offBase = ofbCandidate, start = ip;
1662                     continue;   /* search a better one */
1663             }   }
1664 
1665             /* let's find an even better one */
1666             if ((depth==2) && (ip<ilimit)) {
1667                 DEBUGLOG(7, "search depth 2");
1668                 ip ++;
1669                 if ( (dictMode == ZSTD_noDict)
1670                   && (offBase) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
1671                     size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
1672                     int const gain2 = (int)(mlRep * 4);
1673                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
1674                     if ((mlRep >= 4) && (gain2 > gain1))
1675                         matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
1676                 }
1677                 if (isDxS) {
1678                     const U32 repIndex = (U32)(ip - base) - offset_1;
1679                     const BYTE* repMatch = repIndex < prefixLowestIndex ?
1680                                    dictBase + (repIndex - dictIndexDelta) :
1681                                    base + repIndex;
1682                     if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
1683                         && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
1684                         const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
1685                         size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
1686                         int const gain2 = (int)(mlRep * 4);
1687                         int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
1688                         if ((mlRep >= 4) && (gain2 > gain1))
1689                             matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip;
1690                     }
1691                 }
1692                 {   size_t ofbCandidate=999999999;
1693                     size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode);
1694                     int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
1695                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
1696                     if ((ml2 >= 4) && (gain2 > gain1)) {
1697                         matchLength = ml2, offBase = ofbCandidate, start = ip;
1698                         continue;
1699             }   }   }
1700             break;  /* nothing found : store previous solution */
1701         }
1702 
1703         /* NOTE:
1704          * Pay attention that `start[-value]` can lead to strange undefined behavior
1705          * notably if `value` is unsigned, resulting in a large positive `-value`.
1706          */
1707         /* catch up */
1708         if (OFFBASE_IS_OFFSET(offBase)) {
1709             if (dictMode == ZSTD_noDict) {
1710                 while ( ((start > anchor) & (start - OFFBASE_TO_OFFSET(offBase) > prefixLowest))
1711                      && (start[-1] == (start-OFFBASE_TO_OFFSET(offBase))[-1]) )  /* only search for offset within prefix */
1712                     { start--; matchLength++; }
1713             }
1714             if (isDxS) {
1715                 U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
1716                 const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
1717                 const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
1718                 while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
1719             }
1720             offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
1721         }
1722         /* store sequence */
1723 _storeSequence:
1724         {   size_t const litLength = (size_t)(start - anchor);
1725             ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
1726             anchor = ip = start + matchLength;
1727         }
1728         if (ms->lazySkipping) {
1729             /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
1730             if (searchMethod == search_rowHash) {
1731                 ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
1732             }
1733             ms->lazySkipping = 0;
1734         }
1735 
1736         /* check immediate repcode */
1737         if (isDxS) {
1738             while (ip <= ilimit) {
1739                 U32 const current2 = (U32)(ip-base);
1740                 U32 const repIndex = current2 - offset_2;
1741                 const BYTE* repMatch = repIndex < prefixLowestIndex ?
1742                         dictBase - dictIndexDelta + repIndex :
1743                         base + repIndex;
1744                 if ( (ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
1745                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
1746                     const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
1747                     matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
1748                     offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset_2 <=> offset_1 */
1749                     ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
1750                     ip += matchLength;
1751                     anchor = ip;
1752                     continue;
1753                 }
1754                 break;
1755             }
1756         }
1757 
1758         if (dictMode == ZSTD_noDict) {
1759             while ( ((ip <= ilimit) & (offset_2>0))
1760                  && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
1761                 /* store sequence */
1762                 matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
1763                 offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap repcodes */
1764                 ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
1765                 ip += matchLength;
1766                 anchor = ip;
1767                 continue;   /* faster when present ... (?) */
1768     }   }   }
1769 
1770     /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
1771      * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
1772     offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
1773 
1774     /* save reps for next block */
1775     rep[0] = offset_1 ? offset_1 : offsetSaved1;
1776     rep[1] = offset_2 ? offset_2 : offsetSaved2;
1777 
1778     /* Return the last literals size */
1779     return (size_t)(iend - anchor);
1780 }
1781 #endif /* build exclusions */
1782 
1783 
1784 #ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
ZSTD_compressBlock_greedy(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1785 size_t ZSTD_compressBlock_greedy(
1786         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1787         void const* src, size_t srcSize)
1788 {
1789     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict);
1790 }
1791 
ZSTD_compressBlock_greedy_dictMatchState(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1792 size_t ZSTD_compressBlock_greedy_dictMatchState(
1793         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1794         void const* src, size_t srcSize)
1795 {
1796     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState);
1797 }
1798 
ZSTD_compressBlock_greedy_dedicatedDictSearch(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1799 size_t ZSTD_compressBlock_greedy_dedicatedDictSearch(
1800         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1801         void const* src, size_t srcSize)
1802 {
1803     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch);
1804 }
1805 
ZSTD_compressBlock_greedy_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1806 size_t ZSTD_compressBlock_greedy_row(
1807         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1808         void const* src, size_t srcSize)
1809 {
1810     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_noDict);
1811 }
1812 
ZSTD_compressBlock_greedy_dictMatchState_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1813 size_t ZSTD_compressBlock_greedy_dictMatchState_row(
1814         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1815         void const* src, size_t srcSize)
1816 {
1817     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dictMatchState);
1818 }
1819 
ZSTD_compressBlock_greedy_dedicatedDictSearch_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1820 size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row(
1821         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1822         void const* src, size_t srcSize)
1823 {
1824     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dedicatedDictSearch);
1825 }
1826 #endif
1827 
1828 #ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
ZSTD_compressBlock_lazy(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1829 size_t ZSTD_compressBlock_lazy(
1830         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1831         void const* src, size_t srcSize)
1832 {
1833     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict);
1834 }
1835 
ZSTD_compressBlock_lazy_dictMatchState(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1836 size_t ZSTD_compressBlock_lazy_dictMatchState(
1837         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1838         void const* src, size_t srcSize)
1839 {
1840     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState);
1841 }
1842 
ZSTD_compressBlock_lazy_dedicatedDictSearch(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1843 size_t ZSTD_compressBlock_lazy_dedicatedDictSearch(
1844         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1845         void const* src, size_t srcSize)
1846 {
1847     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch);
1848 }
1849 
ZSTD_compressBlock_lazy_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1850 size_t ZSTD_compressBlock_lazy_row(
1851         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1852         void const* src, size_t srcSize)
1853 {
1854     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_noDict);
1855 }
1856 
ZSTD_compressBlock_lazy_dictMatchState_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1857 size_t ZSTD_compressBlock_lazy_dictMatchState_row(
1858         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1859         void const* src, size_t srcSize)
1860 {
1861     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dictMatchState);
1862 }
1863 
ZSTD_compressBlock_lazy_dedicatedDictSearch_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1864 size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row(
1865         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1866         void const* src, size_t srcSize)
1867 {
1868     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dedicatedDictSearch);
1869 }
1870 #endif
1871 
1872 #ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
ZSTD_compressBlock_lazy2(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1873 size_t ZSTD_compressBlock_lazy2(
1874         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1875         void const* src, size_t srcSize)
1876 {
1877     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict);
1878 }
1879 
ZSTD_compressBlock_lazy2_dictMatchState(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1880 size_t ZSTD_compressBlock_lazy2_dictMatchState(
1881         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1882         void const* src, size_t srcSize)
1883 {
1884     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState);
1885 }
1886 
ZSTD_compressBlock_lazy2_dedicatedDictSearch(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1887 size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch(
1888         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1889         void const* src, size_t srcSize)
1890 {
1891     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch);
1892 }
1893 
ZSTD_compressBlock_lazy2_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1894 size_t ZSTD_compressBlock_lazy2_row(
1895         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1896         void const* src, size_t srcSize)
1897 {
1898     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_noDict);
1899 }
1900 
ZSTD_compressBlock_lazy2_dictMatchState_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1901 size_t ZSTD_compressBlock_lazy2_dictMatchState_row(
1902         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1903         void const* src, size_t srcSize)
1904 {
1905     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dictMatchState);
1906 }
1907 
ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1908 size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row(
1909         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1910         void const* src, size_t srcSize)
1911 {
1912     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dedicatedDictSearch);
1913 }
1914 #endif
1915 
1916 #ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
ZSTD_compressBlock_btlazy2(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1917 size_t ZSTD_compressBlock_btlazy2(
1918         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1919         void const* src, size_t srcSize)
1920 {
1921     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict);
1922 }
1923 
ZSTD_compressBlock_btlazy2_dictMatchState(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)1924 size_t ZSTD_compressBlock_btlazy2_dictMatchState(
1925         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
1926         void const* src, size_t srcSize)
1927 {
1928     return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState);
1929 }
1930 #endif
1931 
1932 #if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \
1933  || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \
1934  || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \
1935  || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR)
1936 FORCE_INLINE_TEMPLATE
1937 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_compressBlock_lazy_extDict_generic(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],const void * src,size_t srcSize,const searchMethod_e searchMethod,const U32 depth)1938 size_t ZSTD_compressBlock_lazy_extDict_generic(
1939                         ZSTD_MatchState_t* ms, SeqStore_t* seqStore,
1940                         U32 rep[ZSTD_REP_NUM],
1941                         const void* src, size_t srcSize,
1942                         const searchMethod_e searchMethod, const U32 depth)
1943 {
1944     const BYTE* const istart = (const BYTE*)src;
1945     const BYTE* ip = istart;
1946     const BYTE* anchor = istart;
1947     const BYTE* const iend = istart + srcSize;
1948     const BYTE* const ilimit = searchMethod == search_rowHash ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8;
1949     const BYTE* const base = ms->window.base;
1950     const U32 dictLimit = ms->window.dictLimit;
1951     const BYTE* const prefixStart = base + dictLimit;
1952     const BYTE* const dictBase = ms->window.dictBase;
1953     const BYTE* const dictEnd  = dictBase + dictLimit;
1954     const BYTE* const dictStart  = dictBase + ms->window.lowLimit;
1955     const U32 windowLog = ms->cParams.windowLog;
1956     const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6);
1957     const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6);
1958 
1959     U32 offset_1 = rep[0], offset_2 = rep[1];
1960 
1961     DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod);
1962 
1963     /* Reset the lazy skipping state */
1964     ms->lazySkipping = 0;
1965 
1966     /* init */
1967     ip += (ip == prefixStart);
1968     if (searchMethod == search_rowHash) {
1969         ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
1970     }
1971 
1972     /* Match Loop */
1973 #if defined(__x86_64__)
1974     /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
1975      * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
1976      */
1977     __asm__(".p2align 5");
1978 #endif
1979     while (ip < ilimit) {
1980         size_t matchLength=0;
1981         size_t offBase = REPCODE1_TO_OFFBASE;
1982         const BYTE* start=ip+1;
1983         U32 curr = (U32)(ip-base);
1984 
1985         /* check repCode */
1986         {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr+1, windowLog);
1987             const U32 repIndex = (U32)(curr+1 - offset_1);
1988             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
1989             const BYTE* const repMatch = repBase + repIndex;
1990             if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
1991                & (offset_1 <= curr+1 - windowLow) ) /* note: we are searching at curr+1 */
1992             if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
1993                 /* repcode detected we should take it */
1994                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
1995                 matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
1996                 if (depth==0) goto _storeSequence;
1997         }   }
1998 
1999         /* first search (depth 0) */
2000         {   size_t ofbCandidate = 999999999;
2001             size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
2002             if (ml2 > matchLength)
2003                 matchLength = ml2, start = ip, offBase = ofbCandidate;
2004         }
2005 
2006         if (matchLength < 4) {
2007             size_t const step = ((size_t)(ip-anchor) >> kSearchStrength);
2008             ip += step + 1;   /* jump faster over incompressible sections */
2009             /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time.
2010              * In this mode we stop inserting every position into our tables, and only insert
2011              * positions that we search, which is one in step positions.
2012              * The exact cutoff is flexible, I've just chosen a number that is reasonably high,
2013              * so we minimize the compression ratio loss in "normal" scenarios. This mode gets
2014              * triggered once we've gone 2KB without finding any matches.
2015              */
2016             ms->lazySkipping = step > kLazySkippingStep;
2017             continue;
2018         }
2019 
2020         /* let's try to find a better solution */
2021         if (depth>=1)
2022         while (ip<ilimit) {
2023             ip ++;
2024             curr++;
2025             /* check repCode */
2026             if (offBase) {
2027                 const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
2028                 const U32 repIndex = (U32)(curr - offset_1);
2029                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
2030                 const BYTE* const repMatch = repBase + repIndex;
2031                 if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
2032                    & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
2033                 if (MEM_read32(ip) == MEM_read32(repMatch)) {
2034                     /* repcode detected */
2035                     const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
2036                     size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
2037                     int const gain2 = (int)(repLength * 3);
2038                     int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1);
2039                     if ((repLength >= 4) && (gain2 > gain1))
2040                         matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
2041             }   }
2042 
2043             /* search match, depth 1 */
2044             {   size_t ofbCandidate = 999999999;
2045                 size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
2046                 int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
2047                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4);
2048                 if ((ml2 >= 4) && (gain2 > gain1)) {
2049                     matchLength = ml2, offBase = ofbCandidate, start = ip;
2050                     continue;   /* search a better one */
2051             }   }
2052 
2053             /* let's find an even better one */
2054             if ((depth==2) && (ip<ilimit)) {
2055                 ip ++;
2056                 curr++;
2057                 /* check repCode */
2058                 if (offBase) {
2059                     const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr, windowLog);
2060                     const U32 repIndex = (U32)(curr - offset_1);
2061                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
2062                     const BYTE* const repMatch = repBase + repIndex;
2063                     if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
2064                        & (offset_1 <= curr - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
2065                     if (MEM_read32(ip) == MEM_read32(repMatch)) {
2066                         /* repcode detected */
2067                         const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
2068                         size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
2069                         int const gain2 = (int)(repLength * 4);
2070                         int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1);
2071                         if ((repLength >= 4) && (gain2 > gain1))
2072                             matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip;
2073                 }   }
2074 
2075                 /* search match, depth 2 */
2076                 {   size_t ofbCandidate = 999999999;
2077                     size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict);
2078                     int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate));   /* raw approx */
2079                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7);
2080                     if ((ml2 >= 4) && (gain2 > gain1)) {
2081                         matchLength = ml2, offBase = ofbCandidate, start = ip;
2082                         continue;
2083             }   }   }
2084             break;  /* nothing found : store previous solution */
2085         }
2086 
2087         /* catch up */
2088         if (OFFBASE_IS_OFFSET(offBase)) {
2089             U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase));
2090             const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
2091             const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
2092             while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
2093             offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase);
2094         }
2095 
2096         /* store sequence */
2097 _storeSequence:
2098         {   size_t const litLength = (size_t)(start - anchor);
2099             ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength);
2100             anchor = ip = start + matchLength;
2101         }
2102         if (ms->lazySkipping) {
2103             /* We've found a match, disable lazy skipping mode, and refill the hash cache. */
2104             if (searchMethod == search_rowHash) {
2105                 ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit);
2106             }
2107             ms->lazySkipping = 0;
2108         }
2109 
2110         /* check immediate repcode */
2111         while (ip <= ilimit) {
2112             const U32 repCurrent = (U32)(ip-base);
2113             const U32 windowLow = ZSTD_getLowestMatchIndex(ms, repCurrent, windowLog);
2114             const U32 repIndex = repCurrent - offset_2;
2115             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
2116             const BYTE* const repMatch = repBase + repIndex;
2117             if ( (ZSTD_index_overlap_check(dictLimit, repIndex))
2118                & (offset_2 <= repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */
2119             if (MEM_read32(ip) == MEM_read32(repMatch)) {
2120                 /* repcode detected we should take it */
2121                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
2122                 matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
2123                 offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase;   /* swap offset history */
2124                 ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength);
2125                 ip += matchLength;
2126                 anchor = ip;
2127                 continue;   /* faster when present ... (?) */
2128             }
2129             break;
2130     }   }
2131 
2132     /* Save reps for next block */
2133     rep[0] = offset_1;
2134     rep[1] = offset_2;
2135 
2136     /* Return the last literals size */
2137     return (size_t)(iend - anchor);
2138 }
2139 #endif /* build exclusions */
2140 
2141 #ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR
ZSTD_compressBlock_greedy_extDict(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2142 size_t ZSTD_compressBlock_greedy_extDict(
2143         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2144         void const* src, size_t srcSize)
2145 {
2146     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0);
2147 }
2148 
ZSTD_compressBlock_greedy_extDict_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2149 size_t ZSTD_compressBlock_greedy_extDict_row(
2150         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2151         void const* src, size_t srcSize)
2152 {
2153     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0);
2154 }
2155 #endif
2156 
2157 #ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR
ZSTD_compressBlock_lazy_extDict(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2158 size_t ZSTD_compressBlock_lazy_extDict(
2159         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2160         void const* src, size_t srcSize)
2161 
2162 {
2163     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1);
2164 }
2165 
ZSTD_compressBlock_lazy_extDict_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2166 size_t ZSTD_compressBlock_lazy_extDict_row(
2167         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2168         void const* src, size_t srcSize)
2169 
2170 {
2171     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1);
2172 }
2173 #endif
2174 
2175 #ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR
ZSTD_compressBlock_lazy2_extDict(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2176 size_t ZSTD_compressBlock_lazy2_extDict(
2177         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2178         void const* src, size_t srcSize)
2179 
2180 {
2181     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2);
2182 }
2183 
ZSTD_compressBlock_lazy2_extDict_row(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2184 size_t ZSTD_compressBlock_lazy2_extDict_row(
2185         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2186         void const* src, size_t srcSize)
2187 {
2188     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2);
2189 }
2190 #endif
2191 
2192 #ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR
ZSTD_compressBlock_btlazy2_extDict(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)2193 size_t ZSTD_compressBlock_btlazy2_extDict(
2194         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
2195         void const* src, size_t srcSize)
2196 
2197 {
2198     return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2);
2199 }
2200 #endif
2201