xref: /linux/lib/zstd/compress/zstd_double_fast.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_double_fast.h"
14 
15 #ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
16 
17 static
18 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_fillDoubleHashTableForCDict(ZSTD_MatchState_t * ms,void const * end,ZSTD_dictTableLoadMethod_e dtlm)19 void ZSTD_fillDoubleHashTableForCDict(ZSTD_MatchState_t* ms,
20                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
21 {
22     const ZSTD_compressionParameters* const cParams = &ms->cParams;
23     U32* const hashLarge = ms->hashTable;
24     U32  const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
25     U32  const mls = cParams->minMatch;
26     U32* const hashSmall = ms->chainTable;
27     U32  const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
28     const BYTE* const base = ms->window.base;
29     const BYTE* ip = base + ms->nextToUpdate;
30     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
31     const U32 fastHashFillStep = 3;
32 
33     /* Always insert every fastHashFillStep position into the hash tables.
34      * Insert the other positions into the large hash table if their entry
35      * is empty.
36      */
37     for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
38         U32 const curr = (U32)(ip - base);
39         U32 i;
40         for (i = 0; i < fastHashFillStep; ++i) {
41             size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
42             size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
43             if (i == 0) {
44                 ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
45             }
46             if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
47                 ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
48             }
49             /* Only load extra positions for ZSTD_dtlm_full */
50             if (dtlm == ZSTD_dtlm_fast)
51                 break;
52     }   }
53 }
54 
55 static
56 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_fillDoubleHashTableForCCtx(ZSTD_MatchState_t * ms,void const * end,ZSTD_dictTableLoadMethod_e dtlm)57 void ZSTD_fillDoubleHashTableForCCtx(ZSTD_MatchState_t* ms,
58                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
59 {
60     const ZSTD_compressionParameters* const cParams = &ms->cParams;
61     U32* const hashLarge = ms->hashTable;
62     U32  const hBitsL = cParams->hashLog;
63     U32  const mls = cParams->minMatch;
64     U32* const hashSmall = ms->chainTable;
65     U32  const hBitsS = cParams->chainLog;
66     const BYTE* const base = ms->window.base;
67     const BYTE* ip = base + ms->nextToUpdate;
68     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
69     const U32 fastHashFillStep = 3;
70 
71     /* Always insert every fastHashFillStep position into the hash tables.
72      * Insert the other positions into the large hash table if their entry
73      * is empty.
74      */
75     for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
76         U32 const curr = (U32)(ip - base);
77         U32 i;
78         for (i = 0; i < fastHashFillStep; ++i) {
79             size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
80             size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
81             if (i == 0)
82                 hashSmall[smHash] = curr + i;
83             if (i == 0 || hashLarge[lgHash] == 0)
84                 hashLarge[lgHash] = curr + i;
85             /* Only load extra positions for ZSTD_dtlm_full */
86             if (dtlm == ZSTD_dtlm_fast)
87                 break;
88         }   }
89 }
90 
ZSTD_fillDoubleHashTable(ZSTD_MatchState_t * ms,const void * const end,ZSTD_dictTableLoadMethod_e dtlm,ZSTD_tableFillPurpose_e tfp)91 void ZSTD_fillDoubleHashTable(ZSTD_MatchState_t* ms,
92                         const void* const end,
93                         ZSTD_dictTableLoadMethod_e dtlm,
94                         ZSTD_tableFillPurpose_e tfp)
95 {
96     if (tfp == ZSTD_tfp_forCDict) {
97         ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
98     } else {
99         ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
100     }
101 }
102 
103 
104 FORCE_INLINE_TEMPLATE
105 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_compressBlock_doubleFast_noDict_generic(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize,U32 const mls)106 size_t ZSTD_compressBlock_doubleFast_noDict_generic(
107         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
108         void const* src, size_t srcSize, U32 const mls /* template */)
109 {
110     ZSTD_compressionParameters const* cParams = &ms->cParams;
111     U32* const hashLong = ms->hashTable;
112     const U32 hBitsL = cParams->hashLog;
113     U32* const hashSmall = ms->chainTable;
114     const U32 hBitsS = cParams->chainLog;
115     const BYTE* const base = ms->window.base;
116     const BYTE* const istart = (const BYTE*)src;
117     const BYTE* anchor = istart;
118     const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
119     /* presumes that, if there is a dictionary, it must be using Attach mode */
120     const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
121     const BYTE* const prefixLowest = base + prefixLowestIndex;
122     const BYTE* const iend = istart + srcSize;
123     const BYTE* const ilimit = iend - HASH_READ_SIZE;
124     U32 offset_1=rep[0], offset_2=rep[1];
125     U32 offsetSaved1 = 0, offsetSaved2 = 0;
126 
127     size_t mLength;
128     U32 offset;
129     U32 curr;
130 
131     /* how many positions to search before increasing step size */
132     const size_t kStepIncr = 1 << kSearchStrength;
133     /* the position at which to increment the step size if no match is found */
134     const BYTE* nextStep;
135     size_t step; /* the current step size */
136 
137     size_t hl0; /* the long hash at ip */
138     size_t hl1; /* the long hash at ip1 */
139 
140     U32 idxl0; /* the long match index for ip */
141     U32 idxl1; /* the long match index for ip1 */
142 
143     const BYTE* matchl0; /* the long match for ip */
144     const BYTE* matchs0; /* the short match for ip */
145     const BYTE* matchl1; /* the long match for ip1 */
146     const BYTE* matchs0_safe; /* matchs0 or safe address */
147 
148     const BYTE* ip = istart; /* the current position */
149     const BYTE* ip1; /* the next position */
150     /* Array of ~random data, should have low probability of matching data
151      * we load from here instead of from tables, if matchl0/matchl1 are
152      * invalid indices. Used to avoid unpredictable branches. */
153     const BYTE dummy[] = {0x12,0x34,0x56,0x78,0x9a,0xbc,0xde,0xf0,0xe2,0xb4};
154 
155     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
156 
157     /* init */
158     ip += ((ip - prefixLowest) == 0);
159     {
160         U32 const current = (U32)(ip - base);
161         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
162         U32 const maxRep = current - windowLow;
163         if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
164         if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
165     }
166 
167     /* Outer Loop: one iteration per match found and stored */
168     while (1) {
169         step = 1;
170         nextStep = ip + kStepIncr;
171         ip1 = ip + step;
172 
173         if (ip1 > ilimit) {
174             goto _cleanup;
175         }
176 
177         hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
178         idxl0 = hashLong[hl0];
179         matchl0 = base + idxl0;
180 
181         /* Inner Loop: one iteration per search / position */
182         do {
183             const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
184             const U32 idxs0 = hashSmall[hs0];
185             curr = (U32)(ip-base);
186             matchs0 = base + idxs0;
187 
188             hashLong[hl0] = hashSmall[hs0] = curr;   /* update hash tables */
189 
190             /* check noDict repcode */
191             if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
192                 mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
193                 ip++;
194                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
195                 goto _match_stored;
196             }
197 
198             hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
199 
200             /* idxl0 > prefixLowestIndex is a (somewhat) unpredictable branch.
201              * However expression below complies into conditional move. Since
202              * match is unlikely and we only *branch* on idxl0 > prefixLowestIndex
203              * if there is a match, all branches become predictable. */
204             {   const BYTE*  const matchl0_safe = ZSTD_selectAddr(idxl0, prefixLowestIndex, matchl0, &dummy[0]);
205 
206                 /* check prefix long match */
207                 if (MEM_read64(matchl0_safe) == MEM_read64(ip) && matchl0_safe == matchl0) {
208                     mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
209                     offset = (U32)(ip-matchl0);
210                     while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
211                     goto _match_found;
212             }   }
213 
214             idxl1 = hashLong[hl1];
215             matchl1 = base + idxl1;
216 
217             /* Same optimization as matchl0 above */
218             matchs0_safe = ZSTD_selectAddr(idxs0, prefixLowestIndex, matchs0, &dummy[0]);
219 
220             /* check prefix short match */
221             if(MEM_read32(matchs0_safe) == MEM_read32(ip) && matchs0_safe == matchs0) {
222                   goto _search_next_long;
223             }
224 
225             if (ip1 >= nextStep) {
226                 PREFETCH_L1(ip1 + 64);
227                 PREFETCH_L1(ip1 + 128);
228                 step++;
229                 nextStep += kStepIncr;
230             }
231             ip = ip1;
232             ip1 += step;
233 
234             hl0 = hl1;
235             idxl0 = idxl1;
236             matchl0 = matchl1;
237     #if defined(__aarch64__)
238             PREFETCH_L1(ip+256);
239     #endif
240         } while (ip1 <= ilimit);
241 
242 _cleanup:
243         /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
244          * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
245         offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
246 
247         /* save reps for next block */
248         rep[0] = offset_1 ? offset_1 : offsetSaved1;
249         rep[1] = offset_2 ? offset_2 : offsetSaved2;
250 
251         /* Return the last literals size */
252         return (size_t)(iend - anchor);
253 
254 _search_next_long:
255 
256         /* short match found: let's check for a longer one */
257         mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
258         offset = (U32)(ip - matchs0);
259 
260         /* check long match at +1 position */
261         if ((idxl1 > prefixLowestIndex) && (MEM_read64(matchl1) == MEM_read64(ip1))) {
262             size_t const l1len = ZSTD_count(ip1+8, matchl1+8, iend) + 8;
263             if (l1len > mLength) {
264                 /* use the long match instead */
265                 ip = ip1;
266                 mLength = l1len;
267                 offset = (U32)(ip-matchl1);
268                 matchs0 = matchl1;
269             }
270         }
271 
272         while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* complete backward */
273 
274         /* fall-through */
275 
276 _match_found: /* requires ip, offset, mLength */
277         offset_2 = offset_1;
278         offset_1 = offset;
279 
280         if (step < 4) {
281             /* It is unsafe to write this value back to the hashtable when ip1 is
282              * greater than or equal to the new ip we will have after we're done
283              * processing this match. Rather than perform that test directly
284              * (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
285              * more predictable test. The minmatch even if we take a short match is
286              * 4 bytes, so as long as step, the distance between ip and ip1
287              * (initially) is less than 4, we know ip1 < new ip. */
288             hashLong[hl1] = (U32)(ip1 - base);
289         }
290 
291         ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
292 
293 _match_stored:
294         /* match found */
295         ip += mLength;
296         anchor = ip;
297 
298         if (ip <= ilimit) {
299             /* Complementary insertion */
300             /* done after iLimit test, as candidates could be > iend-8 */
301             {   U32 const indexToInsert = curr+2;
302                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
303                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
304                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
305                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
306             }
307 
308             /* check immediate repcode */
309             while ( (ip <= ilimit)
310                  && ( (offset_2>0)
311                     & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
312                 /* store sequence */
313                 size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
314                 U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
315                 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
316                 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
317                 ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
318                 ip += rLength;
319                 anchor = ip;
320                 continue;   /* faster when present ... (?) */
321             }
322         }
323     }
324 }
325 
326 
327 FORCE_INLINE_TEMPLATE
328 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_compressBlock_doubleFast_dictMatchState_generic(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize,U32 const mls)329 size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
330         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
331         void const* src, size_t srcSize,
332         U32 const mls /* template */)
333 {
334     ZSTD_compressionParameters const* cParams = &ms->cParams;
335     U32* const hashLong = ms->hashTable;
336     const U32 hBitsL = cParams->hashLog;
337     U32* const hashSmall = ms->chainTable;
338     const U32 hBitsS = cParams->chainLog;
339     const BYTE* const base = ms->window.base;
340     const BYTE* const istart = (const BYTE*)src;
341     const BYTE* ip = istart;
342     const BYTE* anchor = istart;
343     const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
344     /* presumes that, if there is a dictionary, it must be using Attach mode */
345     const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
346     const BYTE* const prefixLowest = base + prefixLowestIndex;
347     const BYTE* const iend = istart + srcSize;
348     const BYTE* const ilimit = iend - HASH_READ_SIZE;
349     U32 offset_1=rep[0], offset_2=rep[1];
350 
351     const ZSTD_MatchState_t* const dms = ms->dictMatchState;
352     const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
353     const U32* const dictHashLong  = dms->hashTable;
354     const U32* const dictHashSmall = dms->chainTable;
355     const U32 dictStartIndex       = dms->window.dictLimit;
356     const BYTE* const dictBase     = dms->window.base;
357     const BYTE* const dictStart    = dictBase + dictStartIndex;
358     const BYTE* const dictEnd      = dms->window.nextSrc;
359     const U32 dictIndexDelta       = prefixLowestIndex - (U32)(dictEnd - dictBase);
360     const U32 dictHBitsL           = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
361     const U32 dictHBitsS           = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
362     const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
363 
364     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
365 
366     /* if a dictionary is attached, it must be within window range */
367     assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
368 
369     if (ms->prefetchCDictTables) {
370         size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
371         size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
372         PREFETCH_AREA(dictHashLong, hashTableBytes);
373         PREFETCH_AREA(dictHashSmall, chainTableBytes);
374     }
375 
376     /* init */
377     ip += (dictAndPrefixLength == 0);
378 
379     /* dictMatchState repCode checks don't currently handle repCode == 0
380      * disabling. */
381     assert(offset_1 <= dictAndPrefixLength);
382     assert(offset_2 <= dictAndPrefixLength);
383 
384     /* Main Search Loop */
385     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
386         size_t mLength;
387         U32 offset;
388         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
389         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
390         size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
391         size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
392         U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
393         U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
394         int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
395         int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
396         U32 const curr = (U32)(ip-base);
397         U32 const matchIndexL = hashLong[h2];
398         U32 matchIndexS = hashSmall[h];
399         const BYTE* matchLong = base + matchIndexL;
400         const BYTE* match = base + matchIndexS;
401         const U32 repIndex = curr + 1 - offset_1;
402         const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
403                                dictBase + (repIndex - dictIndexDelta) :
404                                base + repIndex;
405         hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
406 
407         /* check repcode */
408         if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
409             && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
410             const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
411             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
412             ip++;
413             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
414             goto _match_stored;
415         }
416 
417         if ((matchIndexL >= prefixLowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
418             /* check prefix long match */
419             mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
420             offset = (U32)(ip-matchLong);
421             while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
422             goto _match_found;
423         } else if (dictTagsMatchL) {
424             /* check dictMatchState long match */
425             U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
426             const BYTE* dictMatchL = dictBase + dictMatchIndexL;
427             assert(dictMatchL < dictEnd);
428 
429             if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
430                 mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
431                 offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
432                 while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
433                 goto _match_found;
434         }   }
435 
436         if (matchIndexS > prefixLowestIndex) {
437             /* short match  candidate */
438             if (MEM_read32(match) == MEM_read32(ip)) {
439                 goto _search_next_long;
440             }
441         } else if (dictTagsMatchS) {
442             /* check dictMatchState short match */
443             U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
444             match = dictBase + dictMatchIndexS;
445             matchIndexS = dictMatchIndexS + dictIndexDelta;
446 
447             if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
448                 goto _search_next_long;
449         }   }
450 
451         ip += ((ip-anchor) >> kSearchStrength) + 1;
452 #if defined(__aarch64__)
453         PREFETCH_L1(ip+256);
454 #endif
455         continue;
456 
457 _search_next_long:
458         {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
459             size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
460             U32 const matchIndexL3 = hashLong[hl3];
461             U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
462             int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
463             const BYTE* matchL3 = base + matchIndexL3;
464             hashLong[hl3] = curr + 1;
465 
466             /* check prefix long +1 match */
467             if ((matchIndexL3 >= prefixLowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1))) {
468                 mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
469                 ip++;
470                 offset = (U32)(ip-matchL3);
471                 while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
472                 goto _match_found;
473             } else if (dictTagsMatchL3) {
474                 /* check dict long +1 match */
475                 U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
476                 const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
477                 assert(dictMatchL3 < dictEnd);
478                 if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
479                     mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
480                     ip++;
481                     offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
482                     while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
483                     goto _match_found;
484         }   }   }
485 
486         /* if no long +1 match, explore the short match we found */
487         if (matchIndexS < prefixLowestIndex) {
488             mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
489             offset = (U32)(curr - matchIndexS);
490             while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
491         } else {
492             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
493             offset = (U32)(ip - match);
494             while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
495         }
496 
497 _match_found:
498         offset_2 = offset_1;
499         offset_1 = offset;
500 
501         ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
502 
503 _match_stored:
504         /* match found */
505         ip += mLength;
506         anchor = ip;
507 
508         if (ip <= ilimit) {
509             /* Complementary insertion */
510             /* done after iLimit test, as candidates could be > iend-8 */
511             {   U32 const indexToInsert = curr+2;
512                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
513                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
514                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
515                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
516             }
517 
518             /* check immediate repcode */
519             while (ip <= ilimit) {
520                 U32 const current2 = (U32)(ip-base);
521                 U32 const repIndex2 = current2 - offset_2;
522                 const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
523                         dictBase + repIndex2 - dictIndexDelta :
524                         base + repIndex2;
525                 if ( (ZSTD_index_overlap_check(prefixLowestIndex, repIndex2))
526                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
527                     const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
528                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
529                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
530                     ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
531                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
532                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
533                     ip += repLength2;
534                     anchor = ip;
535                     continue;
536                 }
537                 break;
538             }
539         }
540     }   /* while (ip < ilimit) */
541 
542     /* save reps for next block */
543     rep[0] = offset_1;
544     rep[1] = offset_2;
545 
546     /* Return the last literals size */
547     return (size_t)(iend - anchor);
548 }
549 
550 #define ZSTD_GEN_DFAST_FN(dictMode, mls)                                                                 \
551     static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls(                                      \
552             ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                          \
553             void const* src, size_t srcSize)                                                             \
554     {                                                                                                    \
555         return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
556     }
557 
558 ZSTD_GEN_DFAST_FN(noDict, 4)
559 ZSTD_GEN_DFAST_FN(noDict, 5)
560 ZSTD_GEN_DFAST_FN(noDict, 6)
561 ZSTD_GEN_DFAST_FN(noDict, 7)
562 
563 ZSTD_GEN_DFAST_FN(dictMatchState, 4)
564 ZSTD_GEN_DFAST_FN(dictMatchState, 5)
565 ZSTD_GEN_DFAST_FN(dictMatchState, 6)
566 ZSTD_GEN_DFAST_FN(dictMatchState, 7)
567 
568 
ZSTD_compressBlock_doubleFast(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)569 size_t ZSTD_compressBlock_doubleFast(
570         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
571         void const* src, size_t srcSize)
572 {
573     const U32 mls = ms->cParams.minMatch;
574     switch(mls)
575     {
576     default: /* includes case 3 */
577     case 4 :
578         return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
579     case 5 :
580         return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
581     case 6 :
582         return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
583     case 7 :
584         return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
585     }
586 }
587 
588 
ZSTD_compressBlock_doubleFast_dictMatchState(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)589 size_t ZSTD_compressBlock_doubleFast_dictMatchState(
590         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
591         void const* src, size_t srcSize)
592 {
593     const U32 mls = ms->cParams.minMatch;
594     switch(mls)
595     {
596     default: /* includes case 3 */
597     case 4 :
598         return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
599     case 5 :
600         return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
601     case 6 :
602         return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
603     case 7 :
604         return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
605     }
606 }
607 
608 
609 static
610 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize,U32 const mls)611 size_t ZSTD_compressBlock_doubleFast_extDict_generic(
612         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
613         void const* src, size_t srcSize,
614         U32 const mls /* template */)
615 {
616     ZSTD_compressionParameters const* cParams = &ms->cParams;
617     U32* const hashLong = ms->hashTable;
618     U32  const hBitsL = cParams->hashLog;
619     U32* const hashSmall = ms->chainTable;
620     U32  const hBitsS = cParams->chainLog;
621     const BYTE* const istart = (const BYTE*)src;
622     const BYTE* ip = istart;
623     const BYTE* anchor = istart;
624     const BYTE* const iend = istart + srcSize;
625     const BYTE* const ilimit = iend - 8;
626     const BYTE* const base = ms->window.base;
627     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
628     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
629     const U32   dictStartIndex = lowLimit;
630     const U32   dictLimit = ms->window.dictLimit;
631     const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
632     const BYTE* const prefixStart = base + prefixStartIndex;
633     const BYTE* const dictBase = ms->window.dictBase;
634     const BYTE* const dictStart = dictBase + dictStartIndex;
635     const BYTE* const dictEnd = dictBase + prefixStartIndex;
636     U32 offset_1=rep[0], offset_2=rep[1];
637 
638     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
639 
640     /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
641     if (prefixStartIndex == dictStartIndex)
642         return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
643 
644     /* Search Loop */
645     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
646         const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
647         const U32 matchIndex = hashSmall[hSmall];
648         const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
649         const BYTE* match = matchBase + matchIndex;
650 
651         const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
652         const U32 matchLongIndex = hashLong[hLong];
653         const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
654         const BYTE* matchLong = matchLongBase + matchLongIndex;
655 
656         const U32 curr = (U32)(ip-base);
657         const U32 repIndex = curr + 1 - offset_1;   /* offset_1 expected <= curr +1 */
658         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
659         const BYTE* const repMatch = repBase + repIndex;
660         size_t mLength;
661         hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
662 
663         if (((ZSTD_index_overlap_check(prefixStartIndex, repIndex))
664             & (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
665           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
666             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
667             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
668             ip++;
669             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
670         } else {
671             if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
672                 const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
673                 const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
674                 U32 offset;
675                 mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
676                 offset = curr - matchLongIndex;
677                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
678                 offset_2 = offset_1;
679                 offset_1 = offset;
680                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
681 
682             } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
683                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
684                 U32 const matchIndex3 = hashLong[h3];
685                 const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
686                 const BYTE* match3 = match3Base + matchIndex3;
687                 U32 offset;
688                 hashLong[h3] = curr + 1;
689                 if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
690                     const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
691                     const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
692                     mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
693                     ip++;
694                     offset = curr+1 - matchIndex3;
695                     while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
696                 } else {
697                     const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
698                     const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
699                     mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
700                     offset = curr - matchIndex;
701                     while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
702                 }
703                 offset_2 = offset_1;
704                 offset_1 = offset;
705                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
706 
707             } else {
708                 ip += ((ip-anchor) >> kSearchStrength) + 1;
709                 continue;
710         }   }
711 
712         /* move to next sequence start */
713         ip += mLength;
714         anchor = ip;
715 
716         if (ip <= ilimit) {
717             /* Complementary insertion */
718             /* done after iLimit test, as candidates could be > iend-8 */
719             {   U32 const indexToInsert = curr+2;
720                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
721                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
722                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
723                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
724             }
725 
726             /* check immediate repcode */
727             while (ip <= ilimit) {
728                 U32 const current2 = (U32)(ip-base);
729                 U32 const repIndex2 = current2 - offset_2;
730                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
731                 if ( ((ZSTD_index_overlap_check(prefixStartIndex, repIndex2))
732                     & (offset_2 <= current2 - dictStartIndex))
733                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
734                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
735                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
736                     U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
737                     ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
738                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
739                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
740                     ip += repLength2;
741                     anchor = ip;
742                     continue;
743                 }
744                 break;
745     }   }   }
746 
747     /* save reps for next block */
748     rep[0] = offset_1;
749     rep[1] = offset_2;
750 
751     /* Return the last literals size */
752     return (size_t)(iend - anchor);
753 }
754 
755 ZSTD_GEN_DFAST_FN(extDict, 4)
756 ZSTD_GEN_DFAST_FN(extDict, 5)
757 ZSTD_GEN_DFAST_FN(extDict, 6)
758 ZSTD_GEN_DFAST_FN(extDict, 7)
759 
ZSTD_compressBlock_doubleFast_extDict(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)760 size_t ZSTD_compressBlock_doubleFast_extDict(
761         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
762         void const* src, size_t srcSize)
763 {
764     U32 const mls = ms->cParams.minMatch;
765     switch(mls)
766     {
767     default: /* includes case 3 */
768     case 4 :
769         return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
770     case 5 :
771         return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
772     case 6 :
773         return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
774     case 7 :
775         return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
776     }
777 }
778 
779 #endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
780