zstd-ruby 1.5.0.0 → 1.5.2.0

Sign up to get free protection for your applications and to get access to all the features.
Files changed (65) hide show
  1. checksums.yaml +4 -4
  2. data/.github/workflows/ruby.yml +2 -2
  3. data/README.md +1 -1
  4. data/ext/zstdruby/extconf.rb +2 -1
  5. data/ext/zstdruby/libzstd/Makefile +50 -175
  6. data/ext/zstdruby/libzstd/README.md +7 -1
  7. data/ext/zstdruby/libzstd/common/bitstream.h +24 -9
  8. data/ext/zstdruby/libzstd/common/compiler.h +89 -43
  9. data/ext/zstdruby/libzstd/common/entropy_common.c +11 -5
  10. data/ext/zstdruby/libzstd/common/error_private.h +79 -0
  11. data/ext/zstdruby/libzstd/common/fse.h +2 -1
  12. data/ext/zstdruby/libzstd/common/fse_decompress.c +1 -1
  13. data/ext/zstdruby/libzstd/common/huf.h +24 -22
  14. data/ext/zstdruby/libzstd/common/mem.h +18 -0
  15. data/ext/zstdruby/libzstd/common/pool.c +11 -6
  16. data/ext/zstdruby/libzstd/common/pool.h +2 -2
  17. data/ext/zstdruby/libzstd/common/portability_macros.h +137 -0
  18. data/ext/zstdruby/libzstd/common/xxhash.c +5 -805
  19. data/ext/zstdruby/libzstd/common/xxhash.h +5568 -167
  20. data/ext/zstdruby/libzstd/common/zstd_internal.h +95 -92
  21. data/ext/zstdruby/libzstd/common/zstd_trace.h +12 -3
  22. data/ext/zstdruby/libzstd/compress/clevels.h +134 -0
  23. data/ext/zstdruby/libzstd/compress/fse_compress.c +63 -27
  24. data/ext/zstdruby/libzstd/compress/huf_compress.c +537 -104
  25. data/ext/zstdruby/libzstd/compress/zstd_compress.c +307 -373
  26. data/ext/zstdruby/libzstd/compress/zstd_compress_internal.h +174 -83
  27. data/ext/zstdruby/libzstd/compress/zstd_compress_literals.c +4 -3
  28. data/ext/zstdruby/libzstd/compress/zstd_compress_literals.h +3 -1
  29. data/ext/zstdruby/libzstd/compress/zstd_compress_sequences.c +15 -14
  30. data/ext/zstdruby/libzstd/compress/zstd_compress_superblock.c +4 -3
  31. data/ext/zstdruby/libzstd/compress/zstd_cwksp.h +41 -27
  32. data/ext/zstdruby/libzstd/compress/zstd_double_fast.c +295 -120
  33. data/ext/zstdruby/libzstd/compress/zstd_fast.c +309 -130
  34. data/ext/zstdruby/libzstd/compress/zstd_lazy.c +482 -562
  35. data/ext/zstdruby/libzstd/compress/zstd_ldm.c +9 -7
  36. data/ext/zstdruby/libzstd/compress/zstd_ldm.h +1 -1
  37. data/ext/zstdruby/libzstd/compress/zstd_ldm_geartab.h +4 -1
  38. data/ext/zstdruby/libzstd/compress/zstd_opt.c +249 -148
  39. data/ext/zstdruby/libzstd/compress/zstdmt_compress.c +76 -38
  40. data/ext/zstdruby/libzstd/compress/zstdmt_compress.h +4 -1
  41. data/ext/zstdruby/libzstd/decompress/huf_decompress.c +727 -189
  42. data/ext/zstdruby/libzstd/decompress/huf_decompress_amd64.S +585 -0
  43. data/ext/zstdruby/libzstd/decompress/zstd_decompress.c +85 -22
  44. data/ext/zstdruby/libzstd/decompress/zstd_decompress_block.c +744 -220
  45. data/ext/zstdruby/libzstd/decompress/zstd_decompress_block.h +8 -2
  46. data/ext/zstdruby/libzstd/decompress/zstd_decompress_internal.h +34 -3
  47. data/ext/zstdruby/libzstd/deprecated/zbuff_compress.c +23 -3
  48. data/ext/zstdruby/libzstd/dictBuilder/cover.c +9 -2
  49. data/ext/zstdruby/libzstd/dictBuilder/fastcover.c +11 -4
  50. data/ext/zstdruby/libzstd/dictBuilder/zdict.c +101 -30
  51. data/ext/zstdruby/libzstd/legacy/zstd_v01.c +2 -6
  52. data/ext/zstdruby/libzstd/legacy/zstd_v02.c +3 -7
  53. data/ext/zstdruby/libzstd/legacy/zstd_v03.c +3 -7
  54. data/ext/zstdruby/libzstd/legacy/zstd_v04.c +3 -7
  55. data/ext/zstdruby/libzstd/legacy/zstd_v05.c +3 -7
  56. data/ext/zstdruby/libzstd/legacy/zstd_v06.c +3 -7
  57. data/ext/zstdruby/libzstd/legacy/zstd_v07.c +3 -7
  58. data/ext/zstdruby/libzstd/libzstd.mk +203 -0
  59. data/ext/zstdruby/libzstd/libzstd.pc.in +1 -0
  60. data/ext/zstdruby/libzstd/module.modulemap +25 -0
  61. data/ext/zstdruby/libzstd/zdict.h +4 -4
  62. data/ext/zstdruby/libzstd/zstd.h +179 -136
  63. data/ext/zstdruby/zstdruby.c +2 -2
  64. data/lib/zstd-ruby/version.rb +1 -1
  65. metadata +11 -6
@@ -48,10 +48,216 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
48
48
 
49
49
 
50
50
  FORCE_INLINE_TEMPLATE
51
- size_t ZSTD_compressBlock_doubleFast_generic(
51
+ size_t ZSTD_compressBlock_doubleFast_noDict_generic(
52
+ ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
53
+ void const* src, size_t srcSize, U32 const mls /* template */)
54
+ {
55
+ ZSTD_compressionParameters const* cParams = &ms->cParams;
56
+ U32* const hashLong = ms->hashTable;
57
+ const U32 hBitsL = cParams->hashLog;
58
+ U32* const hashSmall = ms->chainTable;
59
+ const U32 hBitsS = cParams->chainLog;
60
+ const BYTE* const base = ms->window.base;
61
+ const BYTE* const istart = (const BYTE*)src;
62
+ const BYTE* anchor = istart;
63
+ const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
64
+ /* presumes that, if there is a dictionary, it must be using Attach mode */
65
+ const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
66
+ const BYTE* const prefixLowest = base + prefixLowestIndex;
67
+ const BYTE* const iend = istart + srcSize;
68
+ const BYTE* const ilimit = iend - HASH_READ_SIZE;
69
+ U32 offset_1=rep[0], offset_2=rep[1];
70
+ U32 offsetSaved = 0;
71
+
72
+ size_t mLength;
73
+ U32 offset;
74
+ U32 curr;
75
+
76
+ /* how many positions to search before increasing step size */
77
+ const size_t kStepIncr = 1 << kSearchStrength;
78
+ /* the position at which to increment the step size if no match is found */
79
+ const BYTE* nextStep;
80
+ size_t step; /* the current step size */
81
+
82
+ size_t hl0; /* the long hash at ip */
83
+ size_t hl1; /* the long hash at ip1 */
84
+
85
+ U32 idxl0; /* the long match index for ip */
86
+ U32 idxl1; /* the long match index for ip1 */
87
+
88
+ const BYTE* matchl0; /* the long match for ip */
89
+ const BYTE* matchs0; /* the short match for ip */
90
+ const BYTE* matchl1; /* the long match for ip1 */
91
+
92
+ const BYTE* ip = istart; /* the current position */
93
+ const BYTE* ip1; /* the next position */
94
+
95
+ DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_noDict_generic");
96
+
97
+ /* init */
98
+ ip += ((ip - prefixLowest) == 0);
99
+ {
100
+ U32 const current = (U32)(ip - base);
101
+ U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
102
+ U32 const maxRep = current - windowLow;
103
+ if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
104
+ if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
105
+ }
106
+
107
+ /* Outer Loop: one iteration per match found and stored */
108
+ while (1) {
109
+ step = 1;
110
+ nextStep = ip + kStepIncr;
111
+ ip1 = ip + step;
112
+
113
+ if (ip1 > ilimit) {
114
+ goto _cleanup;
115
+ }
116
+
117
+ hl0 = ZSTD_hashPtr(ip, hBitsL, 8);
118
+ idxl0 = hashLong[hl0];
119
+ matchl0 = base + idxl0;
120
+
121
+ /* Inner Loop: one iteration per search / position */
122
+ do {
123
+ const size_t hs0 = ZSTD_hashPtr(ip, hBitsS, mls);
124
+ const U32 idxs0 = hashSmall[hs0];
125
+ curr = (U32)(ip-base);
126
+ matchs0 = base + idxs0;
127
+
128
+ hashLong[hl0] = hashSmall[hs0] = curr; /* update hash tables */
129
+
130
+ /* check noDict repcode */
131
+ if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
132
+ mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
133
+ ip++;
134
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
135
+ goto _match_stored;
136
+ }
137
+
138
+ hl1 = ZSTD_hashPtr(ip1, hBitsL, 8);
139
+
140
+ if (idxl0 > prefixLowestIndex) {
141
+ /* check prefix long match */
142
+ if (MEM_read64(matchl0) == MEM_read64(ip)) {
143
+ mLength = ZSTD_count(ip+8, matchl0+8, iend) + 8;
144
+ offset = (U32)(ip-matchl0);
145
+ while (((ip>anchor) & (matchl0>prefixLowest)) && (ip[-1] == matchl0[-1])) { ip--; matchl0--; mLength++; } /* catch up */
146
+ goto _match_found;
147
+ }
148
+ }
149
+
150
+ idxl1 = hashLong[hl1];
151
+ matchl1 = base + idxl1;
152
+
153
+ if (idxs0 > prefixLowestIndex) {
154
+ /* check prefix short match */
155
+ if (MEM_read32(matchs0) == MEM_read32(ip)) {
156
+ goto _search_next_long;
157
+ }
158
+ }
159
+
160
+ if (ip1 >= nextStep) {
161
+ PREFETCH_L1(ip1 + 64);
162
+ PREFETCH_L1(ip1 + 128);
163
+ step++;
164
+ nextStep += kStepIncr;
165
+ }
166
+ ip = ip1;
167
+ ip1 += step;
168
+
169
+ hl0 = hl1;
170
+ idxl0 = idxl1;
171
+ matchl0 = matchl1;
172
+ #if defined(__aarch64__)
173
+ PREFETCH_L1(ip+256);
174
+ #endif
175
+ } while (ip1 <= ilimit);
176
+
177
+ _cleanup:
178
+ /* save reps for next block */
179
+ rep[0] = offset_1 ? offset_1 : offsetSaved;
180
+ rep[1] = offset_2 ? offset_2 : offsetSaved;
181
+
182
+ /* Return the last literals size */
183
+ return (size_t)(iend - anchor);
184
+
185
+ _search_next_long:
186
+
187
+ /* check prefix long +1 match */
188
+ if (idxl1 > prefixLowestIndex) {
189
+ if (MEM_read64(matchl1) == MEM_read64(ip1)) {
190
+ ip = ip1;
191
+ mLength = ZSTD_count(ip+8, matchl1+8, iend) + 8;
192
+ offset = (U32)(ip-matchl1);
193
+ while (((ip>anchor) & (matchl1>prefixLowest)) && (ip[-1] == matchl1[-1])) { ip--; matchl1--; mLength++; } /* catch up */
194
+ goto _match_found;
195
+ }
196
+ }
197
+
198
+ /* if no long +1 match, explore the short match we found */
199
+ mLength = ZSTD_count(ip+4, matchs0+4, iend) + 4;
200
+ offset = (U32)(ip - matchs0);
201
+ while (((ip>anchor) & (matchs0>prefixLowest)) && (ip[-1] == matchs0[-1])) { ip--; matchs0--; mLength++; } /* catch up */
202
+
203
+ /* fall-through */
204
+
205
+ _match_found: /* requires ip, offset, mLength */
206
+ offset_2 = offset_1;
207
+ offset_1 = offset;
208
+
209
+ if (step < 4) {
210
+ /* It is unsafe to write this value back to the hashtable when ip1 is
211
+ * greater than or equal to the new ip we will have after we're done
212
+ * processing this match. Rather than perform that test directly
213
+ * (ip1 >= ip + mLength), which costs speed in practice, we do a simpler
214
+ * more predictable test. The minmatch even if we take a short match is
215
+ * 4 bytes, so as long as step, the distance between ip and ip1
216
+ * (initially) is less than 4, we know ip1 < new ip. */
217
+ hashLong[hl1] = (U32)(ip1 - base);
218
+ }
219
+
220
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
221
+
222
+ _match_stored:
223
+ /* match found */
224
+ ip += mLength;
225
+ anchor = ip;
226
+
227
+ if (ip <= ilimit) {
228
+ /* Complementary insertion */
229
+ /* done after iLimit test, as candidates could be > iend-8 */
230
+ { U32 const indexToInsert = curr+2;
231
+ hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
232
+ hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
233
+ hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
234
+ hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
235
+ }
236
+
237
+ /* check immediate repcode */
238
+ while ( (ip <= ilimit)
239
+ && ( (offset_2>0)
240
+ & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
241
+ /* store sequence */
242
+ size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
243
+ U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; /* swap offset_2 <=> offset_1 */
244
+ hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
245
+ hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
246
+ ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, rLength);
247
+ ip += rLength;
248
+ anchor = ip;
249
+ continue; /* faster when present ... (?) */
250
+ }
251
+ }
252
+ }
253
+ }
254
+
255
+
256
+ FORCE_INLINE_TEMPLATE
257
+ size_t ZSTD_compressBlock_doubleFast_dictMatchState_generic(
52
258
  ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
53
259
  void const* src, size_t srcSize,
54
- U32 const mls /* template */, ZSTD_dictMode_e const dictMode)
260
+ U32 const mls /* template */)
55
261
  {
56
262
  ZSTD_compressionParameters const* cParams = &ms->cParams;
57
263
  U32* const hashLong = ms->hashTable;
@@ -72,54 +278,30 @@ size_t ZSTD_compressBlock_doubleFast_generic(
72
278
  U32 offsetSaved = 0;
73
279
 
74
280
  const ZSTD_matchState_t* const dms = ms->dictMatchState;
75
- const ZSTD_compressionParameters* const dictCParams =
76
- dictMode == ZSTD_dictMatchState ?
77
- &dms->cParams : NULL;
78
- const U32* const dictHashLong = dictMode == ZSTD_dictMatchState ?
79
- dms->hashTable : NULL;
80
- const U32* const dictHashSmall = dictMode == ZSTD_dictMatchState ?
81
- dms->chainTable : NULL;
82
- const U32 dictStartIndex = dictMode == ZSTD_dictMatchState ?
83
- dms->window.dictLimit : 0;
84
- const BYTE* const dictBase = dictMode == ZSTD_dictMatchState ?
85
- dms->window.base : NULL;
86
- const BYTE* const dictStart = dictMode == ZSTD_dictMatchState ?
87
- dictBase + dictStartIndex : NULL;
88
- const BYTE* const dictEnd = dictMode == ZSTD_dictMatchState ?
89
- dms->window.nextSrc : NULL;
90
- const U32 dictIndexDelta = dictMode == ZSTD_dictMatchState ?
91
- prefixLowestIndex - (U32)(dictEnd - dictBase) :
92
- 0;
93
- const U32 dictHBitsL = dictMode == ZSTD_dictMatchState ?
94
- dictCParams->hashLog : hBitsL;
95
- const U32 dictHBitsS = dictMode == ZSTD_dictMatchState ?
96
- dictCParams->chainLog : hBitsS;
281
+ const ZSTD_compressionParameters* const dictCParams = &dms->cParams;
282
+ const U32* const dictHashLong = dms->hashTable;
283
+ const U32* const dictHashSmall = dms->chainTable;
284
+ const U32 dictStartIndex = dms->window.dictLimit;
285
+ const BYTE* const dictBase = dms->window.base;
286
+ const BYTE* const dictStart = dictBase + dictStartIndex;
287
+ const BYTE* const dictEnd = dms->window.nextSrc;
288
+ const U32 dictIndexDelta = prefixLowestIndex - (U32)(dictEnd - dictBase);
289
+ const U32 dictHBitsL = dictCParams->hashLog;
290
+ const U32 dictHBitsS = dictCParams->chainLog;
97
291
  const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
98
292
 
99
- DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
100
-
101
- assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
293
+ DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_dictMatchState_generic");
102
294
 
103
295
  /* if a dictionary is attached, it must be within window range */
104
- if (dictMode == ZSTD_dictMatchState) {
105
- assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
106
- }
296
+ assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
107
297
 
108
298
  /* init */
109
299
  ip += (dictAndPrefixLength == 0);
110
- if (dictMode == ZSTD_noDict) {
111
- U32 const curr = (U32)(ip - base);
112
- U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
113
- U32 const maxRep = curr - windowLow;
114
- if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
115
- if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
116
- }
117
- if (dictMode == ZSTD_dictMatchState) {
118
- /* dictMatchState repCode checks don't currently handle repCode == 0
119
- * disabling. */
120
- assert(offset_1 <= dictAndPrefixLength);
121
- assert(offset_2 <= dictAndPrefixLength);
122
- }
300
+
301
+ /* dictMatchState repCode checks don't currently handle repCode == 0
302
+ * disabling. */
303
+ assert(offset_1 <= dictAndPrefixLength);
304
+ assert(offset_2 <= dictAndPrefixLength);
123
305
 
124
306
  /* Main Search Loop */
125
307
  while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */
@@ -135,29 +317,18 @@ size_t ZSTD_compressBlock_doubleFast_generic(
135
317
  const BYTE* matchLong = base + matchIndexL;
136
318
  const BYTE* match = base + matchIndexS;
137
319
  const U32 repIndex = curr + 1 - offset_1;
138
- const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
139
- && repIndex < prefixLowestIndex) ?
320
+ const BYTE* repMatch = (repIndex < prefixLowestIndex) ?
140
321
  dictBase + (repIndex - dictIndexDelta) :
141
322
  base + repIndex;
142
323
  hashLong[h2] = hashSmall[h] = curr; /* update hash tables */
143
324
 
144
- /* check dictMatchState repcode */
145
- if (dictMode == ZSTD_dictMatchState
146
- && ((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
325
+ /* check repcode */
326
+ if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
147
327
  && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
148
328
  const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
149
329
  mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
150
330
  ip++;
151
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
152
- goto _match_stored;
153
- }
154
-
155
- /* check noDict repcode */
156
- if ( dictMode == ZSTD_noDict
157
- && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
158
- mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
159
- ip++;
160
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
331
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
161
332
  goto _match_stored;
162
333
  }
163
334
 
@@ -169,7 +340,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
169
340
  while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
170
341
  goto _match_found;
171
342
  }
172
- } else if (dictMode == ZSTD_dictMatchState) {
343
+ } else {
173
344
  /* check dictMatchState long match */
174
345
  U32 const dictMatchIndexL = dictHashLong[dictHL];
175
346
  const BYTE* dictMatchL = dictBase + dictMatchIndexL;
@@ -187,7 +358,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
187
358
  if (MEM_read32(match) == MEM_read32(ip)) {
188
359
  goto _search_next_long;
189
360
  }
190
- } else if (dictMode == ZSTD_dictMatchState) {
361
+ } else {
191
362
  /* check dictMatchState short match */
192
363
  U32 const dictMatchIndexS = dictHashSmall[dictHS];
193
364
  match = dictBase + dictMatchIndexS;
@@ -220,7 +391,7 @@ _search_next_long:
220
391
  while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
221
392
  goto _match_found;
222
393
  }
223
- } else if (dictMode == ZSTD_dictMatchState) {
394
+ } else {
224
395
  /* check dict long +1 match */
225
396
  U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
226
397
  const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
@@ -234,7 +405,7 @@ _search_next_long:
234
405
  } } }
235
406
 
236
407
  /* if no long +1 match, explore the short match we found */
237
- if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
408
+ if (matchIndexS < prefixLowestIndex) {
238
409
  mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
239
410
  offset = (U32)(curr - matchIndexS);
240
411
  while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
@@ -244,13 +415,11 @@ _search_next_long:
244
415
  while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
245
416
  }
246
417
 
247
- /* fall-through */
248
-
249
418
  _match_found:
250
419
  offset_2 = offset_1;
251
420
  offset_1 = offset;
252
421
 
253
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
422
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
254
423
 
255
424
  _match_stored:
256
425
  /* match found */
@@ -268,43 +437,27 @@ _match_stored:
268
437
  }
269
438
 
270
439
  /* check immediate repcode */
271
- if (dictMode == ZSTD_dictMatchState) {
272
- while (ip <= ilimit) {
273
- U32 const current2 = (U32)(ip-base);
274
- U32 const repIndex2 = current2 - offset_2;
275
- const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
276
- && repIndex2 < prefixLowestIndex ?
277
- dictBase + repIndex2 - dictIndexDelta :
278
- base + repIndex2;
279
- if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
280
- && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
281
- const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
282
- size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
283
- U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
284
- ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
285
- hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
286
- hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
287
- ip += repLength2;
288
- anchor = ip;
289
- continue;
290
- }
291
- break;
292
- } }
293
-
294
- if (dictMode == ZSTD_noDict) {
295
- while ( (ip <= ilimit)
296
- && ( (offset_2>0)
297
- & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
298
- /* store sequence */
299
- size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
300
- U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; /* swap offset_2 <=> offset_1 */
301
- hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
302
- hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
303
- ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, rLength-MINMATCH);
304
- ip += rLength;
440
+ while (ip <= ilimit) {
441
+ U32 const current2 = (U32)(ip-base);
442
+ U32 const repIndex2 = current2 - offset_2;
443
+ const BYTE* repMatch2 = repIndex2 < prefixLowestIndex ?
444
+ dictBase + repIndex2 - dictIndexDelta :
445
+ base + repIndex2;
446
+ if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
447
+ && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
448
+ const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
449
+ size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
450
+ U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
451
+ ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);
452
+ hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
453
+ hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
454
+ ip += repLength2;
305
455
  anchor = ip;
306
- continue; /* faster when present ... (?) */
307
- } } }
456
+ continue;
457
+ }
458
+ break;
459
+ }
460
+ }
308
461
  } /* while (ip < ilimit) */
309
462
 
310
463
  /* save reps for next block */
@@ -315,6 +468,24 @@ _match_stored:
315
468
  return (size_t)(iend - anchor);
316
469
  }
317
470
 
471
+ #define ZSTD_GEN_DFAST_FN(dictMode, mls) \
472
+ static size_t ZSTD_compressBlock_doubleFast_##dictMode##_##mls( \
473
+ ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], \
474
+ void const* src, size_t srcSize) \
475
+ { \
476
+ return ZSTD_compressBlock_doubleFast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls); \
477
+ }
478
+
479
+ ZSTD_GEN_DFAST_FN(noDict, 4)
480
+ ZSTD_GEN_DFAST_FN(noDict, 5)
481
+ ZSTD_GEN_DFAST_FN(noDict, 6)
482
+ ZSTD_GEN_DFAST_FN(noDict, 7)
483
+
484
+ ZSTD_GEN_DFAST_FN(dictMatchState, 4)
485
+ ZSTD_GEN_DFAST_FN(dictMatchState, 5)
486
+ ZSTD_GEN_DFAST_FN(dictMatchState, 6)
487
+ ZSTD_GEN_DFAST_FN(dictMatchState, 7)
488
+
318
489
 
319
490
  size_t ZSTD_compressBlock_doubleFast(
320
491
  ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -325,13 +496,13 @@ size_t ZSTD_compressBlock_doubleFast(
325
496
  {
326
497
  default: /* includes case 3 */
327
498
  case 4 :
328
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
499
+ return ZSTD_compressBlock_doubleFast_noDict_4(ms, seqStore, rep, src, srcSize);
329
500
  case 5 :
330
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
501
+ return ZSTD_compressBlock_doubleFast_noDict_5(ms, seqStore, rep, src, srcSize);
331
502
  case 6 :
332
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
503
+ return ZSTD_compressBlock_doubleFast_noDict_6(ms, seqStore, rep, src, srcSize);
333
504
  case 7 :
334
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
505
+ return ZSTD_compressBlock_doubleFast_noDict_7(ms, seqStore, rep, src, srcSize);
335
506
  }
336
507
  }
337
508
 
@@ -345,13 +516,13 @@ size_t ZSTD_compressBlock_doubleFast_dictMatchState(
345
516
  {
346
517
  default: /* includes case 3 */
347
518
  case 4 :
348
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
519
+ return ZSTD_compressBlock_doubleFast_dictMatchState_4(ms, seqStore, rep, src, srcSize);
349
520
  case 5 :
350
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
521
+ return ZSTD_compressBlock_doubleFast_dictMatchState_5(ms, seqStore, rep, src, srcSize);
351
522
  case 6 :
352
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
523
+ return ZSTD_compressBlock_doubleFast_dictMatchState_6(ms, seqStore, rep, src, srcSize);
353
524
  case 7 :
354
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
525
+ return ZSTD_compressBlock_doubleFast_dictMatchState_7(ms, seqStore, rep, src, srcSize);
355
526
  }
356
527
  }
357
528
 
@@ -387,7 +558,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
387
558
 
388
559
  /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
389
560
  if (prefixStartIndex == dictStartIndex)
390
- return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, mls, ZSTD_noDict);
561
+ return ZSTD_compressBlock_doubleFast(ms, seqStore, rep, src, srcSize);
391
562
 
392
563
  /* Search Loop */
393
564
  while (ip < ilimit) { /* < instead of <=, because (ip+1) */
@@ -409,12 +580,12 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
409
580
  hashSmall[hSmall] = hashLong[hLong] = curr; /* update hash table */
410
581
 
411
582
  if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
412
- & (offset_1 < curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
583
+ & (offset_1 <= curr+1 - dictStartIndex)) /* note: we are searching at curr+1 */
413
584
  && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
414
585
  const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
415
586
  mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
416
587
  ip++;
417
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
588
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);
418
589
  } else {
419
590
  if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
420
591
  const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
@@ -425,7 +596,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
425
596
  while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
426
597
  offset_2 = offset_1;
427
598
  offset_1 = offset;
428
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
599
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
429
600
 
430
601
  } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
431
602
  size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
@@ -450,7 +621,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
450
621
  }
451
622
  offset_2 = offset_1;
452
623
  offset_1 = offset;
453
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
624
+ ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
454
625
 
455
626
  } else {
456
627
  ip += ((ip-anchor) >> kSearchStrength) + 1;
@@ -477,12 +648,12 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
477
648
  U32 const repIndex2 = current2 - offset_2;
478
649
  const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
479
650
  if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
480
- & (offset_2 < current2 - dictStartIndex))
651
+ & (offset_2 <= current2 - dictStartIndex))
481
652
  && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
482
653
  const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
483
654
  size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
484
655
  U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */
485
- ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
656
+ ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);
486
657
  hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
487
658
  hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
488
659
  ip += repLength2;
@@ -500,6 +671,10 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
500
671
  return (size_t)(iend - anchor);
501
672
  }
502
673
 
674
+ ZSTD_GEN_DFAST_FN(extDict, 4)
675
+ ZSTD_GEN_DFAST_FN(extDict, 5)
676
+ ZSTD_GEN_DFAST_FN(extDict, 6)
677
+ ZSTD_GEN_DFAST_FN(extDict, 7)
503
678
 
504
679
  size_t ZSTD_compressBlock_doubleFast_extDict(
505
680
  ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -510,12 +685,12 @@ size_t ZSTD_compressBlock_doubleFast_extDict(
510
685
  {
511
686
  default: /* includes case 3 */
512
687
  case 4 :
513
- return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
688
+ return ZSTD_compressBlock_doubleFast_extDict_4(ms, seqStore, rep, src, srcSize);
514
689
  case 5 :
515
- return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
690
+ return ZSTD_compressBlock_doubleFast_extDict_5(ms, seqStore, rep, src, srcSize);
516
691
  case 6 :
517
- return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
692
+ return ZSTD_compressBlock_doubleFast_extDict_6(ms, seqStore, rep, src, srcSize);
518
693
  case 7 :
519
- return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
694
+ return ZSTD_compressBlock_doubleFast_extDict_7(ms, seqStore, rep, src, srcSize);
520
695
  }
521
696
  }