1 /*
2  * Copyright (c) Yann Collet, Facebook, Inc.
3  * All rights reserved.
4  *
5  * This source code is licensed under both the BSD-style license (found in the
6  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7  * in the COPYING file in the root directory of this source tree).
8  * You may select, at your option, one of the above-listed licenses.
9  */
10 
11 #include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
12 #include "zstd_fast.h"
13 
14 
ZSTD_fillHashTable(ZSTD_matchState_t * ms,const void * const end,ZSTD_dictTableLoadMethod_e dtlm)15 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
16                         const void* const end,
17                         ZSTD_dictTableLoadMethod_e dtlm)
18 {
19     const ZSTD_compressionParameters* const cParams = &ms->cParams;
20     U32* const hashTable = ms->hashTable;
21     U32  const hBits = cParams->hashLog;
22     U32  const mls = cParams->minMatch;
23     const BYTE* const base = ms->window.base;
24     const BYTE* ip = base + ms->nextToUpdate;
25     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
26     const U32 fastHashFillStep = 3;
27 
28     /* Always insert every fastHashFillStep position into the hash table.
29      * Insert the other positions if their hash entry is empty.
30      */
31     for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
32         U32 const curr = (U32)(ip - base);
33         size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
34         hashTable[hash0] = curr;
35         if (dtlm == ZSTD_dtlm_fast) continue;
36         /* Only load extra positions for ZSTD_dtlm_full */
37         {   U32 p;
38             for (p = 1; p < fastHashFillStep; ++p) {
39                 size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
40                 if (hashTable[hash] == 0) {  /* not yet filled */
41                     hashTable[hash] = curr + p;
42     }   }   }   }
43 }
44 
45 
46 /**
47  * If you squint hard enough (and ignore repcodes), the search operation at any
48  * given position is broken into 4 stages:
49  *
50  * 1. Hash   (map position to hash value via input read)
51  * 2. Lookup (map hash val to index via hashtable read)
52  * 3. Load   (map index to value at that position via input read)
53  * 4. Compare
54  *
55  * Each of these steps involves a memory read at an address which is computed
56  * from the previous step. This means these steps must be sequenced and their
57  * latencies are cumulative.
58  *
59  * Rather than do 1->2->3->4 sequentially for a single position before moving
60  * onto the next, this implementation interleaves these operations across the
61  * next few positions:
62  *
63  * R = Repcode Read & Compare
64  * H = Hash
65  * T = Table Lookup
66  * M = Match Read & Compare
67  *
68  * Pos | Time -->
69  * ----+-------------------
70  * N   | ... M
71  * N+1 | ...   TM
72  * N+2 |    R H   T M
73  * N+3 |         H    TM
74  * N+4 |           R H   T M
75  * N+5 |                H   ...
76  * N+6 |                  R ...
77  *
78  * This is very much analogous to the pipelining of execution in a CPU. And just
79  * like a CPU, we have to dump the pipeline when we find a match (i.e., take a
80  * branch).
81  *
82  * When this happens, we throw away our current state, and do the following prep
83  * to re-enter the loop:
84  *
85  * Pos | Time -->
86  * ----+-------------------
87  * N   | H T
88  * N+1 |  H
89  *
90  * This is also the work we do at the beginning to enter the loop initially.
91  */
92 FORCE_INLINE_TEMPLATE size_t
ZSTD_compressBlock_fast_noDict_generic(ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize,U32 const mls,U32 const hasStep)93 ZSTD_compressBlock_fast_noDict_generic(
94         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
95         void const* src, size_t srcSize,
96         U32 const mls, U32 const hasStep)
97 {
98     const ZSTD_compressionParameters* const cParams = &ms->cParams;
99     U32* const hashTable = ms->hashTable;
100     U32 const hlog = cParams->hashLog;
101     /* support stepSize of 0 */
102     size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2;
103     const BYTE* const base = ms->window.base;
104     const BYTE* const istart = (const BYTE*)src;
105     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
106     const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
107     const BYTE* const prefixStart = base + prefixStartIndex;
108     const BYTE* const iend = istart + srcSize;
109     const BYTE* const ilimit = iend - HASH_READ_SIZE;
110 
111     const BYTE* anchor = istart;
112     const BYTE* ip0 = istart;
113     const BYTE* ip1;
114     const BYTE* ip2;
115     const BYTE* ip3;
116     U32 current0;
117 
118     U32 rep_offset1 = rep[0];
119     U32 rep_offset2 = rep[1];
120     U32 offsetSaved = 0;
121 
122     size_t hash0; /* hash for ip0 */
123     size_t hash1; /* hash for ip1 */
124     U32 idx; /* match idx for ip0 */
125     U32 mval; /* src value at match idx */
126 
127     U32 offcode;
128     const BYTE* match0;
129     size_t mLength;
130 
131     /* ip0 and ip1 are always adjacent. The targetLength skipping and
132      * uncompressibility acceleration is applied to every other position,
133      * matching the behavior of #1562. step therefore represents the gap
134      * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */
135     size_t step;
136     const BYTE* nextStep;
137     const size_t kStepIncr = (1 << (kSearchStrength - 1));
138 
139     DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
140     ip0 += (ip0 == prefixStart);
141     {   U32 const curr = (U32)(ip0 - base);
142         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
143         U32 const maxRep = curr - windowLow;
144         if (rep_offset2 > maxRep) offsetSaved = rep_offset2, rep_offset2 = 0;
145         if (rep_offset1 > maxRep) offsetSaved = rep_offset1, rep_offset1 = 0;
146     }
147 
148     /* start each op */
149 _start: /* Requires: ip0 */
150 
151     step = stepSize;
152     nextStep = ip0 + kStepIncr;
153 
154     /* calculate positions, ip0 - anchor == 0, so we skip step calc */
155     ip1 = ip0 + 1;
156     ip2 = ip0 + step;
157     ip3 = ip2 + 1;
158 
159     if (ip3 >= ilimit) {
160         goto _cleanup;
161     }
162 
163     hash0 = ZSTD_hashPtr(ip0, hlog, mls);
164     hash1 = ZSTD_hashPtr(ip1, hlog, mls);
165 
166     idx = hashTable[hash0];
167 
168     do {
169         /* load repcode match for ip[2]*/
170         const U32 rval = MEM_read32(ip2 - rep_offset1);
171 
172         /* write back hash table entry */
173         current0 = (U32)(ip0 - base);
174         hashTable[hash0] = current0;
175 
176         /* check repcode at ip[2] */
177         if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
178             ip0 = ip2;
179             match0 = ip0 - rep_offset1;
180             mLength = ip0[-1] == match0[-1];
181             ip0 -= mLength;
182             match0 -= mLength;
183             offcode = STORE_REPCODE_1;
184             mLength += 4;
185             goto _match;
186         }
187 
188         /* load match for ip[0] */
189         if (idx >= prefixStartIndex) {
190             mval = MEM_read32(base + idx);
191         } else {
192             mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
193         }
194 
195         /* check match at ip[0] */
196         if (MEM_read32(ip0) == mval) {
197             /* found a match! */
198             goto _offset;
199         }
200 
201         /* lookup ip[1] */
202         idx = hashTable[hash1];
203 
204         /* hash ip[2] */
205         hash0 = hash1;
206         hash1 = ZSTD_hashPtr(ip2, hlog, mls);
207 
208         /* advance to next positions */
209         ip0 = ip1;
210         ip1 = ip2;
211         ip2 = ip3;
212 
213         /* write back hash table entry */
214         current0 = (U32)(ip0 - base);
215         hashTable[hash0] = current0;
216 
217         /* load match for ip[0] */
218         if (idx >= prefixStartIndex) {
219             mval = MEM_read32(base + idx);
220         } else {
221             mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */
222         }
223 
224         /* check match at ip[0] */
225         if (MEM_read32(ip0) == mval) {
226             /* found a match! */
227             goto _offset;
228         }
229 
230         /* lookup ip[1] */
231         idx = hashTable[hash1];
232 
233         /* hash ip[2] */
234         hash0 = hash1;
235         hash1 = ZSTD_hashPtr(ip2, hlog, mls);
236 
237         /* advance to next positions */
238         ip0 = ip1;
239         ip1 = ip2;
240         ip2 = ip0 + step;
241         ip3 = ip1 + step;
242 
243         /* calculate step */
244         if (ip2 >= nextStep) {
245             step++;
246             PREFETCH_L1(ip1 + 64);
247             PREFETCH_L1(ip1 + 128);
248             nextStep += kStepIncr;
249         }
250     } while (ip3 < ilimit);
251 
252 _cleanup:
253     /* Note that there are probably still a couple positions we could search.
254      * However, it seems to be a meaningful performance hit to try to search
255      * them. So let's not. */
256 
257     /* save reps for next block */
258     rep[0] = rep_offset1 ? rep_offset1 : offsetSaved;
259     rep[1] = rep_offset2 ? rep_offset2 : offsetSaved;
260 
261     /* Return the last literals size */
262     return (size_t)(iend - anchor);
263 
264 _offset: /* Requires: ip0, idx */
265 
266     /* Compute the offset code. */
267     match0 = base + idx;
268     rep_offset2 = rep_offset1;
269     rep_offset1 = (U32)(ip0-match0);
270     offcode = STORE_OFFSET(rep_offset1);
271     mLength = 4;
272 
273     /* Count the backwards match length. */
274     while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
275         ip0--;
276         match0--;
277         mLength++;
278     }
279 
280 _match: /* Requires: ip0, match0, offcode */
281 
282     /* Count the forward length. */
283     mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);
284 
285     ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
286 
287     ip0 += mLength;
288     anchor = ip0;
289 
290     /* write next hash table entry */
291     if (ip1 < ip0) {
292         hashTable[hash1] = (U32)(ip1 - base);
293     }
294 
295     /* Fill table and check for immediate repcode. */
296     if (ip0 <= ilimit) {
297         /* Fill Table */
298         assert(base+current0+2 > istart);  /* check base overflow */
299         hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
300         hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
301 
302         if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
303             while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
304                 /* store sequence */
305                 size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
306                 { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
307                 hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
308                 ip0 += rLength;
309                 ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, STORE_REPCODE_1, rLength);
310                 anchor = ip0;
311                 continue;   /* faster when present (confirmed on gcc-8) ... (?) */
312     }   }   }
313 
314     goto _start;
315 }
316 
317 #define ZSTD_GEN_FAST_FN(dictMode, mls, step)                                                            \
318     static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step(                                      \
319             ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \
320             void const* src, size_t srcSize)                                                       \
321     {                                                                                              \
322         return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \
323     }
324 
325 ZSTD_GEN_FAST_FN(noDict, 4, 1)
326 ZSTD_GEN_FAST_FN(noDict, 5, 1)
327 ZSTD_GEN_FAST_FN(noDict, 6, 1)
328 ZSTD_GEN_FAST_FN(noDict, 7, 1)
329 
330 ZSTD_GEN_FAST_FN(noDict, 4, 0)
331 ZSTD_GEN_FAST_FN(noDict, 5, 0)
332 ZSTD_GEN_FAST_FN(noDict, 6, 0)
333 ZSTD_GEN_FAST_FN(noDict, 7, 0)
334 
ZSTD_compressBlock_fast(ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)335 size_t ZSTD_compressBlock_fast(
336         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
337         void const* src, size_t srcSize)
338 {
339     U32 const mls = ms->cParams.minMatch;
340     assert(ms->dictMatchState == NULL);
341     if (ms->cParams.targetLength > 1) {
342         switch(mls)
343         {
344         default: /* includes case 3 */
345         case 4 :
346             return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);
347         case 5 :
348             return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);
349         case 6 :
350             return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);
351         case 7 :
352             return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);
353         }
354     } else {
355         switch(mls)
356         {
357         default: /* includes case 3 */
358         case 4 :
359             return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);
360         case 5 :
361             return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);
362         case 6 :
363             return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);
364         case 7 :
365             return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);
366         }
367 
368     }
369 }
370 
371 FORCE_INLINE_TEMPLATE
ZSTD_compressBlock_fast_dictMatchState_generic(ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize,U32 const mls,U32 const hasStep)372 size_t ZSTD_compressBlock_fast_dictMatchState_generic(
373         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
374         void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
375 {
376     const ZSTD_compressionParameters* const cParams = &ms->cParams;
377     U32* const hashTable = ms->hashTable;
378     U32 const hlog = cParams->hashLog;
379     /* support stepSize of 0 */
380     U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
381     const BYTE* const base = ms->window.base;
382     const BYTE* const istart = (const BYTE*)src;
383     const BYTE* ip = istart;
384     const BYTE* anchor = istart;
385     const U32   prefixStartIndex = ms->window.dictLimit;
386     const BYTE* const prefixStart = base + prefixStartIndex;
387     const BYTE* const iend = istart + srcSize;
388     const BYTE* const ilimit = iend - HASH_READ_SIZE;
389     U32 offset_1=rep[0], offset_2=rep[1];
390     U32 offsetSaved = 0;
391 
392     const ZSTD_matchState_t* const dms = ms->dictMatchState;
393     const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
394     const U32* const dictHashTable = dms->hashTable;
395     const U32 dictStartIndex       = dms->window.dictLimit;
396     const BYTE* const dictBase     = dms->window.base;
397     const BYTE* const dictStart    = dictBase + dictStartIndex;
398     const BYTE* const dictEnd      = dms->window.nextSrc;
399     const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
400     const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);
401     const U32 dictHLog             = dictCParams->hashLog;
402 
403     /* if a dictionary is still attached, it necessarily means that
404      * it is within window size. So we just check it. */
405     const U32 maxDistance = 1U << cParams->windowLog;
406     const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);
407     assert(endIndex - prefixStartIndex <= maxDistance);
408     (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */
409 
410     (void)hasStep; /* not currently specialized on whether it's accelerated */
411 
412     /* ensure there will be no underflow
413      * when translating a dict index into a local index */
414     assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
415 
416     /* init */
417     DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
418     ip += (dictAndPrefixLength == 0);
419     /* dictMatchState repCode checks don't currently handle repCode == 0
420      * disabling. */
421     assert(offset_1 <= dictAndPrefixLength);
422     assert(offset_2 <= dictAndPrefixLength);
423 
424     /* Main Search Loop */
425     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
426         size_t mLength;
427         size_t const h = ZSTD_hashPtr(ip, hlog, mls);
428         U32 const curr = (U32)(ip-base);
429         U32 const matchIndex = hashTable[h];
430         const BYTE* match = base + matchIndex;
431         const U32 repIndex = curr + 1 - offset_1;
432         const BYTE* repMatch = (repIndex < prefixStartIndex) ?
433                                dictBase + (repIndex - dictIndexDelta) :
434                                base + repIndex;
435         hashTable[h] = curr;   /* update hash table */
436 
437         if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
438           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
439             const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
440             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
441             ip++;
442             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
443         } else if ( (matchIndex <= prefixStartIndex) ) {
444             size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);
445             U32 const dictMatchIndex = dictHashTable[dictHash];
446             const BYTE* dictMatch = dictBase + dictMatchIndex;
447             if (dictMatchIndex <= dictStartIndex ||
448                 MEM_read32(dictMatch) != MEM_read32(ip)) {
449                 assert(stepSize >= 1);
450                 ip += ((ip-anchor) >> kSearchStrength) + stepSize;
451                 continue;
452             } else {
453                 /* found a dict match */
454                 U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta);
455                 mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
456                 while (((ip>anchor) & (dictMatch>dictStart))
457                      && (ip[-1] == dictMatch[-1])) {
458                     ip--; dictMatch--; mLength++;
459                 } /* catch up */
460                 offset_2 = offset_1;
461                 offset_1 = offset;
462                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
463             }
464         } else if (MEM_read32(match) != MEM_read32(ip)) {
465             /* it's not a match, and we're not going to check the dictionary */
466             assert(stepSize >= 1);
467             ip += ((ip-anchor) >> kSearchStrength) + stepSize;
468             continue;
469         } else {
470             /* found a regular match */
471             U32 const offset = (U32)(ip-match);
472             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
473             while (((ip>anchor) & (match>prefixStart))
474                  && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
475             offset_2 = offset_1;
476             offset_1 = offset;
477             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
478         }
479 
480         /* match found */
481         ip += mLength;
482         anchor = ip;
483 
484         if (ip <= ilimit) {
485             /* Fill Table */
486             assert(base+curr+2 > istart);  /* check base overflow */
487             hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
488             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
489 
490             /* check immediate repcode */
491             while (ip <= ilimit) {
492                 U32 const current2 = (U32)(ip-base);
493                 U32 const repIndex2 = current2 - offset_2;
494                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
495                         dictBase - dictIndexDelta + repIndex2 :
496                         base + repIndex2;
497                 if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
498                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
499                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
500                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
501                     U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
502                     ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);
503                     hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
504                     ip += repLength2;
505                     anchor = ip;
506                     continue;
507                 }
508                 break;
509             }
510         }
511     }
512 
513     /* save reps for next block */
514     rep[0] = offset_1 ? offset_1 : offsetSaved;
515     rep[1] = offset_2 ? offset_2 : offsetSaved;
516 
517     /* Return the last literals size */
518     return (size_t)(iend - anchor);
519 }
520 
521 
522 ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)
523 ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)
524 ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)
525 ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)
526 
ZSTD_compressBlock_fast_dictMatchState(ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)527 size_t ZSTD_compressBlock_fast_dictMatchState(
528         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
529         void const* src, size_t srcSize)
530 {
531     U32 const mls = ms->cParams.minMatch;
532     assert(ms->dictMatchState != NULL);
533     switch(mls)
534     {
535     default: /* includes case 3 */
536     case 4 :
537         return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);
538     case 5 :
539         return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);
540     case 6 :
541         return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);
542     case 7 :
543         return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);
544     }
545 }
546 
547 
ZSTD_compressBlock_fast_extDict_generic(ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize,U32 const mls,U32 const hasStep)548 static size_t ZSTD_compressBlock_fast_extDict_generic(
549         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
550         void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
551 {
552     const ZSTD_compressionParameters* const cParams = &ms->cParams;
553     U32* const hashTable = ms->hashTable;
554     U32 const hlog = cParams->hashLog;
555     /* support stepSize of 0 */
556     U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
557     const BYTE* const base = ms->window.base;
558     const BYTE* const dictBase = ms->window.dictBase;
559     const BYTE* const istart = (const BYTE*)src;
560     const BYTE* ip = istart;
561     const BYTE* anchor = istart;
562     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
563     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
564     const U32   dictStartIndex = lowLimit;
565     const BYTE* const dictStart = dictBase + dictStartIndex;
566     const U32   dictLimit = ms->window.dictLimit;
567     const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
568     const BYTE* const prefixStart = base + prefixStartIndex;
569     const BYTE* const dictEnd = dictBase + prefixStartIndex;
570     const BYTE* const iend = istart + srcSize;
571     const BYTE* const ilimit = iend - 8;
572     U32 offset_1=rep[0], offset_2=rep[1];
573 
574     (void)hasStep; /* not currently specialized on whether it's accelerated */
575 
576     DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);
577 
578     /* switch to "regular" variant if extDict is invalidated due to maxDistance */
579     if (prefixStartIndex == dictStartIndex)
580         return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
581 
582     /* Search Loop */
583     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
584         const size_t h = ZSTD_hashPtr(ip, hlog, mls);
585         const U32    matchIndex = hashTable[h];
586         const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
587         const BYTE*  match = matchBase + matchIndex;
588         const U32    curr = (U32)(ip-base);
589         const U32    repIndex = curr + 1 - offset_1;
590         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
591         const BYTE* const repMatch = repBase + repIndex;
592         hashTable[h] = curr;   /* update hash table */
593         DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);
594 
595         if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */
596              & (offset_1 <= curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */
597            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
598             const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
599             size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;
600             ip++;
601             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, rLength);
602             ip += rLength;
603             anchor = ip;
604         } else {
605             if ( (matchIndex < dictStartIndex) ||
606                  (MEM_read32(match) != MEM_read32(ip)) ) {
607                 assert(stepSize >= 1);
608                 ip += ((ip-anchor) >> kSearchStrength) + stepSize;
609                 continue;
610             }
611             {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
612                 const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
613                 U32 const offset = curr - matchIndex;
614                 size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
615                 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
616                 offset_2 = offset_1; offset_1 = offset;  /* update offset history */
617                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
618                 ip += mLength;
619                 anchor = ip;
620         }   }
621 
622         if (ip <= ilimit) {
623             /* Fill Table */
624             hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;
625             hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
626             /* check immediate repcode */
627             while (ip <= ilimit) {
628                 U32 const current2 = (U32)(ip-base);
629                 U32 const repIndex2 = current2 - offset_2;
630                 const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
631                 if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 <= curr - dictStartIndex))  /* intentional overflow */
632                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
633                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
634                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
635                     { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
636                     ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, STORE_REPCODE_1, repLength2);
637                     hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
638                     ip += repLength2;
639                     anchor = ip;
640                     continue;
641                 }
642                 break;
643     }   }   }
644 
645     /* save reps for next block */
646     rep[0] = offset_1;
647     rep[1] = offset_2;
648 
649     /* Return the last literals size */
650     return (size_t)(iend - anchor);
651 }
652 
653 ZSTD_GEN_FAST_FN(extDict, 4, 0)
654 ZSTD_GEN_FAST_FN(extDict, 5, 0)
655 ZSTD_GEN_FAST_FN(extDict, 6, 0)
656 ZSTD_GEN_FAST_FN(extDict, 7, 0)
657 
ZSTD_compressBlock_fast_extDict(ZSTD_matchState_t * ms,seqStore_t * seqStore,U32 rep[ZSTD_REP_NUM],void const * src,size_t srcSize)658 size_t ZSTD_compressBlock_fast_extDict(
659         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
660         void const* src, size_t srcSize)
661 {
662     U32 const mls = ms->cParams.minMatch;
663     switch(mls)
664     {
665     default: /* includes case 3 */
666     case 4 :
667         return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);
668     case 5 :
669         return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);
670     case 6 :
671         return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);
672     case 7 :
673         return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);
674     }
675 }
676