• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) Meta Platforms, Inc. and affiliates.
3  * All rights reserved.
4  *
5  * This source code is licensed under both the BSD-style license (found in the
6  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7  * in the COPYING file in the root directory of this source tree).
8  * You may select, at your option, one of the above-listed licenses.
9  */
10 
11 #include "zstd_compress_internal.h"
12 #include "zstd_double_fast.h"
13 
14 #ifndef ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR
15 
16 static
17 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_fillDoubleHashTableForCDict(ZSTD_MatchState_t * ms,void const * end,ZSTD_dictTableLoadMethod_e dtlm)18 void ZSTD_fillDoubleHashTableForCDict(ZSTD_MatchState_t* ms,
19                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
20 {
21     const ZSTD_compressionParameters* const cParams = &ms->cParams;
22     U32* const hashLarge = ms->hashTable;
23     U32  const hBitsL = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
24     U32  const mls = cParams->minMatch;
25     U32* const hashSmall = ms->chainTable;
26     U32  const hBitsS = cParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
27     const BYTE* const base = ms->window.base;
28     const BYTE* ip = base + ms->nextToUpdate;
29     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
30     const U32 fastHashFillStep = 3;
31 
32     /* Always insert every fastHashFillStep position into the hash tables.
33      * Insert the other positions into the large hash table if their entry
34      * is empty.
35      */
36     for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
37         U32 const curr = (U32)(ip - base);
38         U32 i;
39         for (i = 0; i < fastHashFillStep; ++i) {
40             size_t const smHashAndTag = ZSTD_hashPtr(ip + i, hBitsS, mls);
41             size_t const lgHashAndTag = ZSTD_hashPtr(ip + i, hBitsL, 8);
42             if (i == 0) {
43                 ZSTD_writeTaggedIndex(hashSmall, smHashAndTag, curr + i);
44             }
45             if (i == 0 || hashLarge[lgHashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {
46                 ZSTD_writeTaggedIndex(hashLarge, lgHashAndTag, curr + i);
47             }
48             /* Only load extra positions for ZSTD_dtlm_full */
49             if (dtlm == ZSTD_dtlm_fast)
50                 break;
51     }   }
52 }
53 
54 static
55 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_fillDoubleHashTableForCCtx(ZSTD_MatchState_t * ms,void const * end,ZSTD_dictTableLoadMethod_e dtlm)56 void ZSTD_fillDoubleHashTableForCCtx(ZSTD_MatchState_t* ms,
57                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
58 {
59     const ZSTD_compressionParameters* const cParams = &ms->cParams;
60     U32* const hashLarge = ms->hashTable;
61     U32  const hBitsL = cParams->hashLog;
62     U32  const mls = cParams->minMatch;
63     U32* const hashSmall = ms->chainTable;
64     U32  const hBitsS = cParams->chainLog;
65     const BYTE* const base = ms->window.base;
66     const BYTE* ip = base + ms->nextToUpdate;
67     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
68     const U32 fastHashFillStep = 3;
69 
70     /* Always insert every fastHashFillStep position into the hash tables.
71      * Insert the other positions into the large hash table if their entry
72      * is empty.
73      */
74     for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
75         U32 const curr = (U32)(ip - base);
76         U32 i;
77         for (i = 0; i < fastHashFillStep; ++i) {
78             size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
79             size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
80             if (i == 0)
81                 hashSmall[smHash] = curr + i;
82             if (i == 0 || hashLarge[lgHash] == 0)
83                 hashLarge[lgHash] = curr + i;
84             /* Only load extra positions for ZSTD_dtlm_full */
85             if (dtlm == ZSTD_dtlm_fast)
86                 break;
87         }   }
88 }
89 
ZSTD_fillDoubleHashTable(ZSTD_MatchState_t * ms,const void * const end,ZSTD_dictTableLoadMethod_e dtlm,ZSTD_tableFillPurpose_e tfp)90 void ZSTD_fillDoubleHashTable(ZSTD_MatchState_t* ms,
91                         const void* const end,
92                         ZSTD_dictTableLoadMethod_e dtlm,
93                         ZSTD_tableFillPurpose_e tfp)
94 {
95     if (tfp == ZSTD_tfp_forCDict) {
96         ZSTD_fillDoubleHashTableForCDict(ms, end, dtlm);
97     } else {
98         ZSTD_fillDoubleHashTableForCCtx(ms, end, dtlm);
99     }
100 }
101 
102 
103 FORCE_INLINE_TEMPLATE
104 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)105 size_t ZSTD_compressBlock_doubleFast_noDict_generic(
106         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
107         void const* src, size_t srcSize, U32 const mls /* template */)
108 {
109     ZSTD_compressionParameters const* cParams = &ms->cParams;
110     U32* const hashLong = ms->hashTable;
111     const U32 hBitsL = cParams->hashLog;
112     U32* const hashSmall = ms->chainTable;
113     const U32 hBitsS = cParams->chainLog;
114     const BYTE* const base = ms->window.base;
115     const BYTE* const istart = (const BYTE*)src;
116     const BYTE* anchor = istart;
117     const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
118     /* presumes that, if there is a dictionary, it must be using Attach mode */
119     const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
120     const BYTE* const prefixLowest = base + prefixLowestIndex;
121     const BYTE* const iend = istart + srcSize;
122     const BYTE* const ilimit = iend - HASH_READ_SIZE;
123     U32 offset_1=rep[0], offset_2=rep[1];
124     U32 offsetSaved1 = 0, offsetSaved2 = 0;
125 
126     size_t mLength;
127     U32 offset;
128     U32 curr;
129 
130     /* how many positions to search before increasing step size */
131     const size_t kStepIncr = 1 << kSearchStrength;
132     /* the position at which to increment the step size if no match is found */
133     const BYTE* nextStep;
134     size_t step; /* the current step size */
135 
136     size_t hl0; /* the long hash at ip */
137     size_t hl1; /* the long hash at ip1 */
138 
139     U32 idxl0; /* the long match index for ip */
140     U32 idxl1; /* the long match index for ip1 */
141 
142     const BYTE* matchl0; /* the long match for ip */
143     const BYTE* matchs0; /* the short match for ip */
144     const BYTE* matchl1; /* the long match for ip1 */
145     const BYTE* matchs0_safe; /* matchs0 or safe address */
146 
147     const BYTE* ip = istart; /* the current position */
148     const BYTE* ip1; /* the next position */
149     /* Array of ~random data, should have low probability of matching data
150      * we load from here instead of from tables, if matchl0/matchl1 are
151      * invalid indices. Used to avoid unpredictable branches. */
152     const BYTE dummy[] = {0x12,0x34,0x56,0x78,0x9a,0xbc,0xde,0xf0,0xe2,0xb4};
153 
154     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
155 
156     /* init */
157     ip += ((ip - prefixLowest) == 0);
158     {
159         U32 const current = (U32)(ip - base);
160         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
161         U32 const maxRep = current - windowLow;
162         if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0;
163         if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0;
164     }
165 
166     /* Outer Loop: one iteration per match found and stored */
167     while (1) {
168         step = 1;
169         nextStep = ip + kStepIncr;
170         ip1 = ip + step;
171 
172         if (ip1 > ilimit) {
173             goto _cleanup;
174         }
175 
176         hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
177         idxl0 = hashLong[hl0];
178         matchl0 = base + idxl0;
179 
180         /* Inner Loop: one iteration per search / position */
181         do {
182             const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
183             const U32 idxs0 = hashSmall[hs0];
184             curr = (U32)(ip-base);
185             matchs0 = base + idxs0;
186 
187             hashLong[hl0] = hashSmall[hs0] = curr;   /* update hash tables */
188 
189             /* check noDict repcode */
190             if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
191                 mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
192                 ip++;
193                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
194                 goto _match_stored;
195             }
196 
197             hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
198 
199             /* idxl0 > prefixLowestIndex is a (somewhat) unpredictable branch.
200              * However expression below complies into conditional move. Since
201              * match is unlikely and we only *branch* on idxl0 > prefixLowestIndex
202              * if there is a match, all branches become predictable. */
203             {   const BYTE*  const matchl0_safe = ZSTD_selectAddr(idxl0, prefixLowestIndex, matchl0, &dummy[0]);
204 
205                 /* check prefix long match */
206                 if (MEM_read64(matchl0_safe) == MEM_read64(ip) && matchl0_safe == matchl0) {
207                     mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
208                     offset = (U32)(ip-matchl0);
209                     while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
210                     goto _match_found;
211             }   }
212 
213             idxl1 = hashLong[hl1];
214             matchl1 = base + idxl1;
215 
216             /* Same optimization as matchl0 above */
217             matchs0_safe = ZSTD_selectAddr(idxs0, prefixLowestIndex, matchs0, &dummy[0]);
218 
219             /* check prefix short match */
220             if(MEM_read32(matchs0_safe) == MEM_read32(ip) && matchs0_safe == matchs0) {
221                   goto _search_next_long;
222             }
223 
224             if (ip1 >= nextStep) {
225                 PREFETCH_L1(ip1 + 64);
226                 PREFETCH_L1(ip1 + 128);
227                 step++;
228                 nextStep += kStepIncr;
229             }
230             ip = ip1;
231             ip1 += step;
232 
233             hl0 = hl1;
234             idxl0 = idxl1;
235             matchl0 = matchl1;
236     #if defined(__aarch64__)
237             PREFETCH_L1(ip+256);
238     #endif
239         } while (ip1 <= ilimit);
240 
241 _cleanup:
242         /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
243          * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
244         offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
245 
246         /* save reps for next block */
247         rep[0] = offset_1 ? offset_1 : offsetSaved1;
248         rep[1] = offset_2 ? offset_2 : offsetSaved2;
249 
250         /* Return the last literals size */
251         return (size_t)(iend - anchor);
252 
253 _search_next_long:
254 
255         /* short match found: let's check for a longer one */
256         mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
257         offset = (U32)(ip - matchs0);
258 
259         /* check long match at +1 position */
260         if ((idxl1 > prefixLowestIndex) && (MEM_read64(matchl1) == MEM_read64(ip1))) {
261             size_t const l1len = ZSTD_count(ip1+8, matchl1+8, iend) + 8;
262             if (l1len > mLength) {
263                 /* use the long match instead */
264                 ip = ip1;
265                 mLength = l1len;
266                 offset = (U32)(ip-matchl1);
267                 matchs0 = matchl1;
268             }
269         }
270 
271         while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* complete backward */
272 
273         /* fall-through */
274 
275 _match_found: /* requires ip, offset, mLength */
276         offset_2 = offset_1;
277         offset_1 = offset;
278 
279         if (step < 4) {
280             /* It is unsafe to write this value back to the hashtable when ip1 is
281              * greater than or equal to the new ip we will have after we're done
282              * processing this match. Rather than perform that test directly
283              * (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
284              * more predictable test. The minmatch even if we take a short match is
285              * 4 bytes, so as long as step, the distance between ip and ip1
286              * (initially) is less than 4, we know ip1 < new ip. */
287             hashLong[hl1] = (U32)(ip1 - base);
288         }
289 
290         ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
291 
292 _match_stored:
293         /* match found */
294         ip += mLength;
295         anchor = ip;
296 
297         if (ip <= ilimit) {
298             /* Complementary insertion */
299             /* done after iLimit test, as candidates could be > iend-8 */
300             {   U32 const indexToInsert = curr+2;
301                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
302                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
303                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
304                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
305             }
306 
307             /* check immediate repcode */
308             while ( (ip <= ilimit)
309                  && ( (offset_2>0)
310                     & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
311                 /* store sequence */
312                 size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
313                 U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
314                 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
315                 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
316                 ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
317                 ip += rLength;
318                 anchor = ip;
319                 continue;   /* faster when present ... (?) */
320             }
321         }
322     }
323 }
324 
325 
326 FORCE_INLINE_TEMPLATE
327 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)328 size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
329         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
330         void const* src, size_t srcSize,
331         U32 const mls /* template */)
332 {
333     ZSTD_compressionParameters const* cParams = &ms->cParams;
334     U32* const hashLong = ms->hashTable;
335     const U32 hBitsL = cParams->hashLog;
336     U32* const hashSmall = ms->chainTable;
337     const U32 hBitsS = cParams->chainLog;
338     const BYTE* const base = ms->window.base;
339     const BYTE* const istart = (const BYTE*)src;
340     const BYTE* ip = istart;
341     const BYTE* anchor = istart;
342     const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
343     /* presumes that, if there is a dictionary, it must be using Attach mode */
344     const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
345     const BYTE* const prefixLowest = base + prefixLowestIndex;
346     const BYTE* const iend = istart + srcSize;
347     const BYTE* const ilimit = iend - HASH_READ_SIZE;
348     U32 offset_1=rep[0], offset_2=rep[1];
349 
350     const ZSTD_MatchState_t* const dms = ms->dictMatchState;
351     const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
352     const U32* const dictHashLong  = dms->hashTable;
353     const U32* const dictHashSmall = dms->chainTable;
354     const U32 dictStartIndex       = dms->window.dictLimit;
355     const BYTE* const dictBase     = dms->window.base;
356     const BYTE* const dictStart    = dictBase + dictStartIndex;
357     const BYTE* const dictEnd      = dms->window.nextSrc;
358     const U32 dictIndexDelta       = prefixLowestIndex - (U32)(dictEnd - dictBase);
359     const U32 dictHBitsL           = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
360     const U32 dictHBitsS           = dictCParams->chainLog + ZSTD_SHORT_CACHE_TAG_BITS;
361     const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
362 
363     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
364 
365     /* if a dictionary is attached, it must be within window range */
366     assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
367 
368     if (ms->prefetchCDictTables) {
369         size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
370         size_t const chainTableBytes = (((size_t)1) << dictCParams->chainLog) * sizeof(U32);
371         PREFETCH_AREA(dictHashLong, hashTableBytes);
372         PREFETCH_AREA(dictHashSmall, chainTableBytes);
373     }
374 
375     /* init */
376     ip += (dictAndPrefixLength == 0);
377 
378     /* dictMatchState repCode checks don't currently handle repCode == 0
379      * disabling. */
380     assert(offset_1 <= dictAndPrefixLength);
381     assert(offset_2 <= dictAndPrefixLength);
382 
383     /* Main Search Loop */
384     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
385         size_t mLength;
386         U32 offset;
387         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
388         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
389         size_t const dictHashAndTagL = ZSTD_hashPtr(ip, dictHBitsL, 8);
390         size_t const dictHashAndTagS = ZSTD_hashPtr(ip, dictHBitsS, mls);
391         U32 const dictMatchIndexAndTagL = dictHashLong[dictHashAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS];
392         U32 const dictMatchIndexAndTagS = dictHashSmall[dictHashAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS];
393         int const dictTagsMatchL = ZSTD_comparePackedTags(dictMatchIndexAndTagL, dictHashAndTagL);
394         int const dictTagsMatchS = ZSTD_comparePackedTags(dictMatchIndexAndTagS, dictHashAndTagS);
395         U32 const curr = (U32)(ip-base);
396         U32 const matchIndexL = hashLong[h2];
397         U32 matchIndexS = hashSmall[h];
398         const BYTE* matchLong = base + matchIndexL;
399         const BYTE* match = base + matchIndexS;
400         const U32 repIndex = curr + 1 - offset_1;
401         const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
402                                dictBase + (repIndex - dictIndexDelta) :
403                                base + repIndex;
404         hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
405 
406         /* check repcode */
407         if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex))
408             && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
409             const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
410             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
411             ip++;
412             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
413             goto _match_stored;
414         }
415 
416         if ((matchIndexL >= prefixLowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
417             /* check prefix long match */
418             mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
419             offset = (U32)(ip-matchLong);
420             while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
421             goto _match_found;
422         } else if (dictTagsMatchL) {
423             /* check dictMatchState long match */
424             U32 const dictMatchIndexL = dictMatchIndexAndTagL >> ZSTD_SHORT_CACHE_TAG_BITS;
425             const BYTE* dictMatchL = dictBase + dictMatchIndexL;
426             assert(dictMatchL < dictEnd);
427 
428             if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
429                 mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
430                 offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
431                 while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
432                 goto _match_found;
433         }   }
434 
435         if (matchIndexS > prefixLowestIndex) {
436             /* short match  candidate */
437             if (MEM_read32(match) == MEM_read32(ip)) {
438                 goto _search_next_long;
439             }
440         } else if (dictTagsMatchS) {
441             /* check dictMatchState short match */
442             U32 const dictMatchIndexS = dictMatchIndexAndTagS >> ZSTD_SHORT_CACHE_TAG_BITS;
443             match = dictBase + dictMatchIndexS;
444             matchIndexS = dictMatchIndexS + dictIndexDelta;
445 
446             if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
447                 goto _search_next_long;
448         }   }
449 
450         ip += ((ip-anchor) >> kSearchStrength) + 1;
451 #if defined(__aarch64__)
452         PREFETCH_L1(ip+256);
453 #endif
454         continue;
455 
456 _search_next_long:
457         {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
458             size_t const dictHashAndTagL3 = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
459             U32 const matchIndexL3 = hashLong[hl3];
460             U32 const dictMatchIndexAndTagL3 = dictHashLong[dictHashAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS];
461             int const dictTagsMatchL3 = ZSTD_comparePackedTags(dictMatchIndexAndTagL3, dictHashAndTagL3);
462             const BYTE* matchL3 = base + matchIndexL3;
463             hashLong[hl3] = curr + 1;
464 
465             /* check prefix long +1 match */
466             if ((matchIndexL3 >= prefixLowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1))) {
467                 mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
468                 ip++;
469                 offset = (U32)(ip-matchL3);
470                 while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
471                 goto _match_found;
472             } else if (dictTagsMatchL3) {
473                 /* check dict long +1 match */
474                 U32 const dictMatchIndexL3 = dictMatchIndexAndTagL3 >> ZSTD_SHORT_CACHE_TAG_BITS;
475                 const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
476                 assert(dictMatchL3 < dictEnd);
477                 if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
478                     mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
479                     ip++;
480                     offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
481                     while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
482                     goto _match_found;
483         }   }   }
484 
485         /* if no long +1 match, explore the short match we found */
486         if (matchIndexS < prefixLowestIndex) {
487             mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
488             offset = (U32)(curr - matchIndexS);
489             while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
490         } else {
491             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
492             offset = (U32)(ip - match);
493             while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
494         }
495 
496 _match_found:
497         offset_2 = offset_1;
498         offset_1 = offset;
499 
500         ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
501 
502 _match_stored:
503         /* match found */
504         ip += mLength;
505         anchor = ip;
506 
507         if (ip <= ilimit) {
508             /* Complementary insertion */
509             /* done after iLimit test, as candidates could be > iend-8 */
510             {   U32 const indexToInsert = curr+2;
511                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
512                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
513                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
514                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
515             }
516 
517             /* check immediate repcode */
518             while (ip <= ilimit) {
519                 U32 const current2 = (U32)(ip-base);
520                 U32 const repIndex2 = current2 - offset_2;
521                 const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
522                         dictBase + repIndex2 - dictIndexDelta :
523                         base + repIndex2;
524                 if ( (ZSTD_index_overlap_check(prefixLowestIndex, repIndex2))
525                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
526                     const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
527                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
528                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
529                     ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
530                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
531                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
532                     ip += repLength2;
533                     anchor = ip;
534                     continue;
535                 }
536                 break;
537             }
538         }
539     }   /* while (ip < ilimit) */
540 
541     /* save reps for next block */
542     rep[0] = offset_1;
543     rep[1] = offset_2;
544 
545     /* Return the last literals size */
546     return (size_t)(iend - anchor);
547 }
548 
549 #define ZSTD_GEN_DFAST_FN(dictMode, mls)                                                                 \
550     static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls(                                      \
551             ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                          \
552             void const* src, size_t srcSize)                                                             \
553     {                                                                                                    \
554         return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
555     }
556 
557 ZSTD_GEN_DFAST_FN(noDict, 4)
558 ZSTD_GEN_DFAST_FN(noDict, 5)
559 ZSTD_GEN_DFAST_FN(noDict, 6)
560 ZSTD_GEN_DFAST_FN(noDict, 7)
561 
562 ZSTD_GEN_DFAST_FN(dictMatchState, 4)
563 ZSTD_GEN_DFAST_FN(dictMatchState, 5)
564 ZSTD_GEN_DFAST_FN(dictMatchState, 6)
565 ZSTD_GEN_DFAST_FN(dictMatchState, 7)
566 
567 
ZSTD_compressBlock_doubleFast(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)568 size_t ZSTD_compressBlock_doubleFast(
569         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
570         void const* src, size_t srcSize)
571 {
572     const U32 mls = ms->cParams.minMatch;
573     switch(mls)
574     {
575     default: /* includes case 3 */
576     case 4 :
577         return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
578     case 5 :
579         return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
580     case 6 :
581         return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
582     case 7 :
583         return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
584     }
585 }
586 
587 
ZSTD_compressBlock_doubleFast_dictMatchState(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)588 size_t ZSTD_compressBlock_doubleFast_dictMatchState(
589         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
590         void const* src, size_t srcSize)
591 {
592     const U32 mls = ms->cParams.minMatch;
593     switch(mls)
594     {
595     default: /* includes case 3 */
596     case 4 :
597         return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
598     case 5 :
599         return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
600     case 6 :
601         return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
602     case 7 :
603         return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
604     }
605 }
606 
607 
608 static
609 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)610 size_t ZSTD_compressBlock_doubleFast_extDict_generic(
611         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
612         void const* src, size_t srcSize,
613         U32 const mls /* template */)
614 {
615     ZSTD_compressionParameters const* cParams = &ms->cParams;
616     U32* const hashLong = ms->hashTable;
617     U32  const hBitsL = cParams->hashLog;
618     U32* const hashSmall = ms->chainTable;
619     U32  const hBitsS = cParams->chainLog;
620     const BYTE* const istart = (const BYTE*)src;
621     const BYTE* ip = istart;
622     const BYTE* anchor = istart;
623     const BYTE* const iend = istart + srcSize;
624     const BYTE* const ilimit = iend - 8;
625     const BYTE* const base = ms->window.base;
626     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
627     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
628     const U32   dictStartIndex = lowLimit;
629     const U32   dictLimit = ms->window.dictLimit;
630     const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
631     const BYTE* const prefixStart = base + prefixStartIndex;
632     const BYTE* const dictBase = ms->window.dictBase;
633     const BYTE* const dictStart = dictBase + dictStartIndex;
634     const BYTE* const dictEnd = dictBase + prefixStartIndex;
635     U32 offset_1=rep[0], offset_2=rep[1];
636 
637     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
638 
639     /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
640     if (prefixStartIndex == dictStartIndex)
641         return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
642 
643     /* Search Loop */
644     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
645         const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
646         const U32 matchIndex = hashSmall[hSmall];
647         const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
648         const BYTE* match = matchBase + matchIndex;
649 
650         const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
651         const U32 matchLongIndex = hashLong[hLong];
652         const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
653         const BYTE* matchLong = matchLongBase + matchLongIndex;
654 
655         const U32 curr = (U32)(ip-base);
656         const U32 repIndex = curr + 1 - offset_1;   /* offset_1 expected <= curr +1 */
657         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
658         const BYTE* const repMatch = repBase + repIndex;
659         size_t mLength;
660         hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
661 
662         if (((ZSTD_index_overlap_check(prefixStartIndex, repIndex))
663             & (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
664           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
665             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
666             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
667             ip++;
668             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
669         } else {
670             if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
671                 const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
672                 const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
673                 U32 offset;
674                 mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
675                 offset = curr - matchLongIndex;
676                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
677                 offset_2 = offset_1;
678                 offset_1 = offset;
679                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
680 
681             } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
682                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
683                 U32 const matchIndex3 = hashLong[h3];
684                 const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
685                 const BYTE* match3 = match3Base + matchIndex3;
686                 U32 offset;
687                 hashLong[h3] = curr + 1;
688                 if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
689                     const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
690                     const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
691                     mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
692                     ip++;
693                     offset = curr+1 - matchIndex3;
694                     while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
695                 } else {
696                     const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
697                     const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
698                     mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
699                     offset = curr - matchIndex;
700                     while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
701                 }
702                 offset_2 = offset_1;
703                 offset_1 = offset;
704                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
705 
706             } else {
707                 ip += ((ip-anchor) >> kSearchStrength) + 1;
708                 continue;
709         }   }
710 
711         /* move to next sequence start */
712         ip += mLength;
713         anchor = ip;
714 
715         if (ip <= ilimit) {
716             /* Complementary insertion */
717             /* done after iLimit test, as candidates could be > iend-8 */
718             {   U32 const indexToInsert = curr+2;
719                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
720                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
721                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
722                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
723             }
724 
725             /* check immediate repcode */
726             while (ip <= ilimit) {
727                 U32 const current2 = (U32)(ip-base);
728                 U32 const repIndex2 = current2 - offset_2;
729                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
730                 if ( ((ZSTD_index_overlap_check(prefixStartIndex, repIndex2))
731                     & (offset_2 <= current2 - dictStartIndex))
732                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
733                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
734                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
735                     U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
736                     ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
737                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
738                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
739                     ip += repLength2;
740                     anchor = ip;
741                     continue;
742                 }
743                 break;
744     }   }   }
745 
746     /* save reps for next block */
747     rep[0] = offset_1;
748     rep[1] = offset_2;
749 
750     /* Return the last literals size */
751     return (size_t)(iend - anchor);
752 }
753 
754 ZSTD_GEN_DFAST_FN(extDict, 4)
755 ZSTD_GEN_DFAST_FN(extDict, 5)
756 ZSTD_GEN_DFAST_FN(extDict, 6)
757 ZSTD_GEN_DFAST_FN(extDict, 7)
758 
ZSTD_compressBlock_doubleFast_extDict(ZSTD_MatchState_t * ms,SeqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)759 size_t ZSTD_compressBlock_doubleFast_extDict(
760         ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
761         void const* src, size_t srcSize)
762 {
763     U32 const mls = ms->cParams.minMatch;
764     switch(mls)
765     {
766     default: /* includes case 3 */
767     case 4 :
768         return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
769     case 5 :
770         return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
771     case 6 :
772         return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
773     case 7 :
774         return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
775     }
776 }
777 
778 #endif /* ZSTD_EXCLUDE_DFAST_BLOCK_COMPRESSOR */
779