zstdlib 0.10.0-x64-mingw-ucrt → 0.12.0-x64-mingw-ucrt

Sign up to get free protection for your applications and to get access to all the features.
Files changed (108) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGES.md +17 -0
  3. data/ext/zstdlib_c/extconf.rb +9 -4
  4. data/ext/zstdlib_c/ruby/zlib-3.2/zstdlib.c +5090 -0
  5. data/ext/zstdlib_c/ruby/zlib-3.3/zstdlib.c +5090 -0
  6. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/adler32.c +5 -27
  7. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/compress.c +5 -16
  8. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/crc32.c +94 -161
  9. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/deflate.c +362 -434
  10. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/deflate.h +43 -12
  11. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/gzclose.c +1 -3
  12. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/gzguts.h +13 -18
  13. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/gzlib.c +28 -85
  14. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/gzread.c +23 -73
  15. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/gzwrite.c +19 -65
  16. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/infback.c +17 -30
  17. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inffast.c +1 -4
  18. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inffast.h +1 -1
  19. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inflate.c +36 -102
  20. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inftrees.c +6 -11
  21. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inftrees.h +6 -6
  22. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/trees.c +290 -355
  23. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/uncompr.c +4 -12
  24. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/zconf.h +23 -14
  25. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/zlib.h +202 -199
  26. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/zutil.c +18 -44
  27. data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/zutil.h +13 -33
  28. data/ext/zstdlib_c/zstd-1.5.5/lib/common/allocations.h +55 -0
  29. data/ext/zstdlib_c/zstd-1.5.5/lib/common/bits.h +200 -0
  30. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/bitstream.h +19 -60
  31. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/compiler.h +26 -3
  32. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/cpu.h +1 -1
  33. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/debug.c +1 -1
  34. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/debug.h +1 -1
  35. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/entropy_common.c +12 -40
  36. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/error_private.c +9 -2
  37. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/error_private.h +1 -1
  38. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/fse.h +5 -83
  39. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/fse_decompress.c +7 -99
  40. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/huf.h +65 -156
  41. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/mem.h +39 -46
  42. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/pool.c +26 -10
  43. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/pool.h +7 -1
  44. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/portability_macros.h +22 -3
  45. data/ext/zstdlib_c/zstd-1.5.5/lib/common/threading.c +176 -0
  46. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/threading.h +5 -10
  47. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/xxhash.c +2 -2
  48. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/xxhash.h +8 -8
  49. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/zstd_common.c +1 -36
  50. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/zstd_deps.h +1 -1
  51. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/zstd_internal.h +17 -118
  52. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/common/zstd_trace.h +3 -3
  53. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/clevels.h +1 -1
  54. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/fse_compress.c +7 -124
  55. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/hist.c +1 -1
  56. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/hist.h +1 -1
  57. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/huf_compress.c +234 -169
  58. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress.c +1243 -538
  59. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress_internal.h +225 -151
  60. data/ext/zstdlib_c/zstd-1.5.5/lib/compress/zstd_compress_literals.c +235 -0
  61. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress_literals.h +16 -8
  62. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress_sequences.c +3 -3
  63. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress_sequences.h +1 -1
  64. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress_superblock.c +25 -21
  65. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_compress_superblock.h +1 -1
  66. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_cwksp.h +128 -62
  67. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_double_fast.c +95 -33
  68. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_double_fast.h +3 -2
  69. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_fast.c +433 -148
  70. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_fast.h +3 -2
  71. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_lazy.c +398 -345
  72. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_lazy.h +4 -2
  73. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_ldm.c +5 -5
  74. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_ldm.h +1 -1
  75. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_ldm_geartab.h +1 -1
  76. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_opt.c +106 -80
  77. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstd_opt.h +1 -1
  78. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstdmt_compress.c +17 -9
  79. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/compress/zstdmt_compress.h +1 -1
  80. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/huf_decompress.c +434 -441
  81. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/huf_decompress_amd64.S +30 -39
  82. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/zstd_ddict.c +4 -4
  83. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/zstd_ddict.h +1 -1
  84. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/zstd_decompress.c +205 -80
  85. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/zstd_decompress_block.c +201 -81
  86. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/zstd_decompress_block.h +6 -1
  87. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/decompress/zstd_decompress_internal.h +4 -2
  88. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/zdict.h +53 -31
  89. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/zstd.h +580 -135
  90. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/lib/zstd_errors.h +27 -8
  91. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/gzclose.c +1 -1
  92. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/gzcompatibility.h +8 -8
  93. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/gzguts.h +10 -10
  94. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/gzlib.c +3 -3
  95. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/gzread.c +10 -10
  96. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/gzwrite.c +5 -5
  97. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/zstd_zlibwrapper.c +46 -44
  98. data/ext/zstdlib_c/{zstd-1.5.2 → zstd-1.5.5}/zlibWrapper/zstd_zlibwrapper.h +4 -1
  99. data/lib/3.1/zstdlib_c.so +0 -0
  100. data/lib/3.2/zstdlib_c.so +0 -0
  101. data/lib/3.3/zstdlib_c.so +0 -0
  102. metadata +111 -105
  103. data/ext/zstdlib_c/zstd-1.5.2/lib/common/threading.c +0 -122
  104. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/zstd_compress_literals.c +0 -159
  105. /data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/crc32.h +0 -0
  106. /data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inffixed.h +0 -0
  107. /data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/inflate.h +0 -0
  108. /data/ext/zstdlib_c/{zlib-1.2.12 → zlib-1.3.1}/trees.h +0 -0
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) Yann Collet, Facebook, Inc.
2
+ * Copyright (c) Meta Platforms, Inc. and affiliates.
3
3
  * All rights reserved.
4
4
  *
5
5
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,8 +11,42 @@
11
11
  #include "zstd_compress_internal.h" /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
12
12
  #include "zstd_fast.h"
13
13
 
14
+ static void ZSTD_fillHashTableForCDict(ZSTD_matchState_t* ms,
15
+ const void* const end,
16
+ ZSTD_dictTableLoadMethod_e dtlm)
17
+ {
18
+ const ZSTD_compressionParameters* const cParams = &ms->cParams;
19
+ U32* const hashTable = ms->hashTable;
20
+ U32 const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
21
+ U32 const mls = cParams->minMatch;
22
+ const BYTE* const base = ms->window.base;
23
+ const BYTE* ip = base + ms->nextToUpdate;
24
+ const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
25
+ const U32 fastHashFillStep = 3;
14
26
 
15
- void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
27
+ /* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
28
+ * Feel free to remove this assert if there's a good reason! */
29
+ assert(dtlm == ZSTD_dtlm_full);
30
+
31
+ /* Always insert every fastHashFillStep position into the hash table.
32
+ * Insert the other positions if their hash entry is empty.
33
+ */
34
+ for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
35
+ U32 const curr = (U32)(ip - base);
36
+ { size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
37
+ ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr); }
38
+
39
+ if (dtlm == ZSTD_dtlm_fast) continue;
40
+ /* Only load extra positions for ZSTD_dtlm_full */
41
+ { U32 p;
42
+ for (p = 1; p < fastHashFillStep; ++p) {
43
+ size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
44
+ if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) { /* not yet filled */
45
+ ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
46
+ } } } }
47
+ }
48
+
49
+ static void ZSTD_fillHashTableForCCtx(ZSTD_matchState_t* ms,
16
50
  const void* const end,
17
51
  ZSTD_dictTableLoadMethod_e dtlm)
18
52
  {
@@ -25,6 +59,10 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
25
59
  const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
26
60
  const U32 fastHashFillStep = 3;
27
61
 
62
+ /* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
63
+ * Feel free to remove this assert if there's a good reason! */
64
+ assert(dtlm == ZSTD_dtlm_fast);
65
+
28
66
  /* Always insert every fastHashFillStep position into the hash table.
29
67
  * Insert the other positions if their hash entry is empty.
30
68
  */
@@ -42,6 +80,18 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
42
80
  } } } }
43
81
  }
44
82
 
83
+ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
84
+ const void* const end,
85
+ ZSTD_dictTableLoadMethod_e dtlm,
86
+ ZSTD_tableFillPurpose_e tfp)
87
+ {
88
+ if (tfp == ZSTD_tfp_forCDict) {
89
+ ZSTD_fillHashTableForCDict(ms, end, dtlm);
90
+ } else {
91
+ ZSTD_fillHashTableForCCtx(ms, end, dtlm);
92
+ }
93
+ }
94
+
45
95
 
46
96
  /**
47
97
  * If you squint hard enough (and ignore repcodes), the search operation at any
@@ -117,7 +167,7 @@ ZSTD_compressBlock_fast_noDict_generic(
117
167
 
118
168
  U32 rep_offset1 = rep[0];
119
169
  U32 rep_offset2 = rep[1];
120
- U32 offsetSaved = 0;
170
+ U32 offsetSaved1 = 0, offsetSaved2 = 0;
121
171
 
122
172
  size_t hash0; /* hash for ip0 */
123
173
  size_t hash1; /* hash for ip1 */
@@ -141,8 +191,8 @@ ZSTD_compressBlock_fast_noDict_generic(
141
191
  { U32 const curr = (U32)(ip0 - base);
142
192
  U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
143
193
  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;
194
+ if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
195
+ if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
146
196
  }
147
197
 
148
198
  /* start each op */
@@ -180,8 +230,14 @@ _start: /* Requires: ip0 */
180
230
  mLength = ip0[-1] == match0[-1];
181
231
  ip0 -= mLength;
182
232
  match0 -= mLength;
183
- offcode = STORE_REPCODE_1;
233
+ offcode = REPCODE1_TO_OFFBASE;
184
234
  mLength += 4;
235
+
236
+ /* First write next hash table entry; we've already calculated it.
237
+ * This write is known to be safe because the ip1 is before the
238
+ * repcode (ip2). */
239
+ hashTable[hash1] = (U32)(ip1 - base);
240
+
185
241
  goto _match;
186
242
  }
187
243
 
@@ -195,6 +251,12 @@ _start: /* Requires: ip0 */
195
251
  /* check match at ip[0] */
196
252
  if (MEM_read32(ip0) == mval) {
197
253
  /* found a match! */
254
+
255
+ /* First write next hash table entry; we've already calculated it.
256
+ * This write is known to be safe because the ip1 == ip0 + 1, so
257
+ * we know we will resume searching after ip1 */
258
+ hashTable[hash1] = (U32)(ip1 - base);
259
+
198
260
  goto _offset;
199
261
  }
200
262
 
@@ -224,6 +286,21 @@ _start: /* Requires: ip0 */
224
286
  /* check match at ip[0] */
225
287
  if (MEM_read32(ip0) == mval) {
226
288
  /* found a match! */
289
+
290
+ /* first write next hash table entry; we've already calculated it */
291
+ if (step <= 4) {
292
+ /* We need to avoid writing an index into the hash table >= the
293
+ * position at which we will pick up our searching after we've
294
+ * taken this match.
295
+ *
296
+ * The minimum possible match has length 4, so the earliest ip0
297
+ * can be after we take this match will be the current ip0 + 4.
298
+ * ip1 is ip0 + step - 1. If ip1 is >= ip0 + 4, we can't safely
299
+ * write this position.
300
+ */
301
+ hashTable[hash1] = (U32)(ip1 - base);
302
+ }
303
+
227
304
  goto _offset;
228
305
  }
229
306
 
@@ -254,9 +331,24 @@ _cleanup:
254
331
  * However, it seems to be a meaningful performance hit to try to search
255
332
  * them. So let's not. */
256
333
 
334
+ /* When the repcodes are outside of the prefix, we set them to zero before the loop.
335
+ * When the offsets are still zero, we need to restore them after the block to have a correct
336
+ * repcode history. If only one offset was invalid, it is easy. The tricky case is when both
337
+ * offsets were invalid. We need to figure out which offset to refill with.
338
+ * - If both offsets are zero they are in the same order.
339
+ * - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
340
+ * - If only one is zero, we need to decide which offset to restore.
341
+ * - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
342
+ * - It is impossible for rep_offset2 to be non-zero.
343
+ *
344
+ * So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
345
+ * set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
346
+ */
347
+ offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;
348
+
257
349
  /* save reps for next block */
258
- rep[0] = rep_offset1 ? rep_offset1 : offsetSaved;
259
- rep[1] = rep_offset2 ? rep_offset2 : offsetSaved;
350
+ rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
351
+ rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;
260
352
 
261
353
  /* Return the last literals size */
262
354
  return (size_t)(iend - anchor);
@@ -267,7 +359,7 @@ _offset: /* Requires: ip0, idx */
267
359
  match0 = base + idx;
268
360
  rep_offset2 = rep_offset1;
269
361
  rep_offset1 = (U32)(ip0-match0);
270
- offcode = STORE_OFFSET(rep_offset1);
362
+ offcode = OFFSET_TO_OFFBASE(rep_offset1);
271
363
  mLength = 4;
272
364
 
273
365
  /* Count the backwards match length. */
@@ -287,11 +379,6 @@ _match: /* Requires: ip0, match0, offcode */
287
379
  ip0 += mLength;
288
380
  anchor = ip0;
289
381
 
290
- /* write next hash table entry */
291
- if (ip1 < ip0) {
292
- hashTable[hash1] = (U32)(ip1 - base);
293
- }
294
-
295
382
  /* Fill table and check for immediate repcode. */
296
383
  if (ip0 <= ilimit) {
297
384
  /* Fill Table */
@@ -306,7 +393,7 @@ _match: /* Requires: ip0, match0, offcode */
306
393
  { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
307
394
  hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
308
395
  ip0 += rLength;
309
- ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, STORE_REPCODE_1, rLength);
396
+ ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
310
397
  anchor = ip0;
311
398
  continue; /* faster when present (confirmed on gcc-8) ... (?) */
312
399
  } } }
@@ -380,14 +467,14 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
380
467
  U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
381
468
  const BYTE* const base = ms->window.base;
382
469
  const BYTE* const istart = (const BYTE*)src;
383
- const BYTE* ip = istart;
470
+ const BYTE* ip0 = istart;
471
+ const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
384
472
  const BYTE* anchor = istart;
385
473
  const U32 prefixStartIndex = ms->window.dictLimit;
386
474
  const BYTE* const prefixStart = base + prefixStartIndex;
387
475
  const BYTE* const iend = istart + srcSize;
388
476
  const BYTE* const ilimit = iend - HASH_READ_SIZE;
389
477
  U32 offset_1=rep[0], offset_2=rep[1];
390
- U32 offsetSaved = 0;
391
478
 
392
479
  const ZSTD_matchState_t* const dms = ms->dictMatchState;
393
480
  const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
@@ -397,13 +484,13 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
397
484
  const BYTE* const dictStart = dictBase + dictStartIndex;
398
485
  const BYTE* const dictEnd = dms->window.nextSrc;
399
486
  const U32 dictIndexDelta = prefixStartIndex - (U32)(dictEnd - dictBase);
400
- const U32 dictAndPrefixLength = (U32)(ip - prefixStart + dictEnd - dictStart);
401
- const U32 dictHLog = dictCParams->hashLog;
487
+ const U32 dictAndPrefixLength = (U32)(istart - prefixStart + dictEnd - dictStart);
488
+ const U32 dictHBits = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
402
489
 
403
490
  /* if a dictionary is still attached, it necessarily means that
404
491
  * it is within window size. So we just check it. */
405
492
  const U32 maxDistance = 1U << cParams->windowLog;
406
- const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);
493
+ const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
407
494
  assert(endIndex - prefixStartIndex <= maxDistance);
408
495
  (void)maxDistance; (void)endIndex; /* these variables are not used when assert() is disabled */
409
496
 
@@ -413,106 +500,155 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
413
500
  * when translating a dict index into a local index */
414
501
  assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
415
502
 
503
+ if (ms->prefetchCDictTables) {
504
+ size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
505
+ PREFETCH_AREA(dictHashTable, hashTableBytes)
506
+ }
507
+
416
508
  /* init */
417
509
  DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
418
- ip += (dictAndPrefixLength == 0);
510
+ ip0 += (dictAndPrefixLength == 0);
419
511
  /* dictMatchState repCode checks don't currently handle repCode == 0
420
512
  * disabling. */
421
513
  assert(offset_1 <= dictAndPrefixLength);
422
514
  assert(offset_2 <= dictAndPrefixLength);
423
515
 
424
- /* Main Search Loop */
425
- while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */
516
+ /* Outer search loop */
517
+ assert(stepSize >= 1);
518
+ while (ip1 <= ilimit) { /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
426
519
  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++;
520
+ size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);
521
+
522
+ size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
523
+ U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
524
+ int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);
525
+
526
+ U32 matchIndex = hashTable[hash0];
527
+ U32 curr = (U32)(ip0 - base);
528
+ size_t step = stepSize;
529
+ const size_t kStepIncr = 1 << kSearchStrength;
530
+ const BYTE* nextStep = ip0 + kStepIncr;
531
+
532
+ /* Inner search loop */
533
+ while (1) {
534
+ const BYTE* match = base + matchIndex;
535
+ const U32 repIndex = curr + 1 - offset_1;
536
+ const BYTE* repMatch = (repIndex < prefixStartIndex) ?
537
+ dictBase + (repIndex - dictIndexDelta) :
538
+ base + repIndex;
539
+ const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
540
+ size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
541
+ hashTable[hash0] = curr; /* update hash table */
542
+
543
+ if (((U32) ((prefixStartIndex - 1) - repIndex) >=
544
+ 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
545
+ && (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
546
+ const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
547
+ mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
548
+ ip0++;
549
+ ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
550
+ break;
551
+ }
552
+
553
+ if (dictTagsMatch) {
554
+ /* Found a possible dict match */
555
+ const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
556
+ const BYTE* dictMatch = dictBase + dictMatchIndex;
557
+ if (dictMatchIndex > dictStartIndex &&
558
+ MEM_read32(dictMatch) == MEM_read32(ip0)) {
559
+ /* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
560
+ if (matchIndex <= prefixStartIndex) {
561
+ U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
562
+ mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
563
+ while (((ip0 > anchor) & (dictMatch > dictStart))
564
+ && (ip0[-1] == dictMatch[-1])) {
565
+ ip0--;
566
+ dictMatch--;
567
+ mLength++;
568
+ } /* catch up */
569
+ offset_2 = offset_1;
570
+ offset_1 = offset;
571
+ ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
572
+ break;
573
+ }
574
+ }
575
+ }
576
+
577
+ if (matchIndex > prefixStartIndex && MEM_read32(match) == MEM_read32(ip0)) {
578
+ /* found a regular match */
579
+ U32 const offset = (U32) (ip0 - match);
580
+ mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
581
+ while (((ip0 > anchor) & (match > prefixStart))
582
+ && (ip0[-1] == match[-1])) {
583
+ ip0--;
584
+ match--;
585
+ mLength++;
459
586
  } /* catch up */
460
587
  offset_2 = offset_1;
461
588
  offset_1 = offset;
462
- ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);
589
+ ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
590
+ break;
463
591
  }
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
- }
592
+
593
+ /* Prepare for next iteration */
594
+ dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
595
+ dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
596
+ matchIndex = hashTable[hash1];
597
+
598
+ if (ip1 >= nextStep) {
599
+ step++;
600
+ nextStep += kStepIncr;
601
+ }
602
+ ip0 = ip1;
603
+ ip1 = ip1 + step;
604
+ if (ip1 > ilimit) goto _cleanup;
605
+
606
+ curr = (U32)(ip0 - base);
607
+ hash0 = hash1;
608
+ } /* end inner search loop */
479
609
 
480
610
  /* match found */
481
- ip += mLength;
482
- anchor = ip;
611
+ assert(mLength);
612
+ ip0 += mLength;
613
+ anchor = ip0;
483
614
 
484
- if (ip <= ilimit) {
615
+ if (ip0 <= ilimit) {
485
616
  /* Fill Table */
486
617
  assert(base+curr+2 > istart); /* check base overflow */
487
618
  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);
619
+ hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
489
620
 
490
621
  /* check immediate repcode */
491
- while (ip <= ilimit) {
492
- U32 const current2 = (U32)(ip-base);
622
+ while (ip0 <= ilimit) {
623
+ U32 const current2 = (U32)(ip0-base);
493
624
  U32 const repIndex2 = current2 - offset_2;
494
625
  const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
495
626
  dictBase - dictIndexDelta + repIndex2 :
496
627
  base + repIndex2;
497
628
  if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
498
- && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
629
+ && (MEM_read32(repMatch2) == MEM_read32(ip0))) {
499
630
  const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
500
- size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
631
+ size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
501
632
  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;
633
+ ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
634
+ hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
635
+ ip0 += repLength2;
636
+ anchor = ip0;
506
637
  continue;
507
638
  }
508
639
  break;
509
640
  }
510
641
  }
642
+
643
+ /* Prepare for next iteration */
644
+ assert(ip0 == anchor);
645
+ ip1 = ip0 + stepSize;
511
646
  }
512
647
 
648
+ _cleanup:
513
649
  /* save reps for next block */
514
- rep[0] = offset_1 ? offset_1 : offsetSaved;
515
- rep[1] = offset_2 ? offset_2 : offsetSaved;
650
+ rep[0] = offset_1;
651
+ rep[1] = offset_2;
516
652
 
517
653
  /* Return the last literals size */
518
654
  return (size_t)(iend - anchor);
@@ -553,11 +689,10 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
553
689
  U32* const hashTable = ms->hashTable;
554
690
  U32 const hlog = cParams->hashLog;
555
691
  /* support stepSize of 0 */
556
- U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
692
+ size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
557
693
  const BYTE* const base = ms->window.base;
558
694
  const BYTE* const dictBase = ms->window.dictBase;
559
695
  const BYTE* const istart = (const BYTE*)src;
560
- const BYTE* ip = istart;
561
696
  const BYTE* anchor = istart;
562
697
  const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
563
698
  const U32 lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
@@ -570,6 +705,28 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
570
705
  const BYTE* const iend = istart + srcSize;
571
706
  const BYTE* const ilimit = iend - 8;
572
707
  U32 offset_1=rep[0], offset_2=rep[1];
708
+ U32 offsetSaved1 = 0, offsetSaved2 = 0;
709
+
710
+ const BYTE* ip0 = istart;
711
+ const BYTE* ip1;
712
+ const BYTE* ip2;
713
+ const BYTE* ip3;
714
+ U32 current0;
715
+
716
+
717
+ size_t hash0; /* hash for ip0 */
718
+ size_t hash1; /* hash for ip1 */
719
+ U32 idx; /* match idx for ip0 */
720
+ const BYTE* idxBase; /* base pointer for idx */
721
+
722
+ U32 offcode;
723
+ const BYTE* match0;
724
+ size_t mLength;
725
+ const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */
726
+
727
+ size_t step;
728
+ const BYTE* nextStep;
729
+ const size_t kStepIncr = (1 << (kSearchStrength - 1));
573
730
 
574
731
  (void)hasStep; /* not currently specialized on whether it's accelerated */
575
732
 
@@ -579,75 +736,202 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
579
736
  if (prefixStartIndex == dictStartIndex)
580
737
  return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);
581
738
 
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;
739
+ { U32 const curr = (U32)(ip0 - base);
740
+ U32 const maxRep = curr - dictStartIndex;
741
+ if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
742
+ if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
743
+ }
744
+
745
+ /* start each op */
746
+ _start: /* Requires: ip0 */
747
+
748
+ step = stepSize;
749
+ nextStep = ip0 + kStepIncr;
750
+
751
+ /* calculate positions, ip0 - anchor == 0, so we skip step calc */
752
+ ip1 = ip0 + 1;
753
+ ip2 = ip0 + step;
754
+ ip3 = ip2 + 1;
755
+
756
+ if (ip3 >= ilimit) {
757
+ goto _cleanup;
758
+ }
759
+
760
+ hash0 = ZSTD_hashPtr(ip0, hlog, mls);
761
+ hash1 = ZSTD_hashPtr(ip1, hlog, mls);
762
+
763
+ idx = hashTable[hash0];
764
+ idxBase = idx < prefixStartIndex ? dictBase : base;
765
+
766
+ do {
767
+ { /* load repcode match for ip[2] */
768
+ U32 const current2 = (U32)(ip2 - base);
769
+ U32 const repIndex = current2 - offset_1;
770
+ const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
771
+ U32 rval;
772
+ if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
773
+ & (offset_1 > 0) ) {
774
+ rval = MEM_read32(repBase + repIndex);
775
+ } else {
776
+ rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
610
777
  }
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;
778
+
779
+ /* write back hash table entry */
780
+ current0 = (U32)(ip0 - base);
781
+ hashTable[hash0] = current0;
782
+
783
+ /* check repcode at ip[2] */
784
+ if (MEM_read32(ip2) == rval) {
785
+ ip0 = ip2;
786
+ match0 = repBase + repIndex;
787
+ matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
788
+ assert((match0 != prefixStart) & (match0 != dictStart));
789
+ mLength = ip0[-1] == match0[-1];
790
+ ip0 -= mLength;
791
+ match0 -= mLength;
792
+ offcode = REPCODE1_TO_OFFBASE;
793
+ mLength += 4;
794
+ goto _match;
620
795
  } }
621
796
 
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
- } } }
797
+ { /* load match for ip[0] */
798
+ U32 const mval = idx >= dictStartIndex ?
799
+ MEM_read32(idxBase + idx) :
800
+ MEM_read32(ip0) ^ 1; /* guaranteed not to match */
801
+
802
+ /* check match at ip[0] */
803
+ if (MEM_read32(ip0) == mval) {
804
+ /* found a match! */
805
+ goto _offset;
806
+ } }
807
+
808
+ /* lookup ip[1] */
809
+ idx = hashTable[hash1];
810
+ idxBase = idx < prefixStartIndex ? dictBase : base;
811
+
812
+ /* hash ip[2] */
813
+ hash0 = hash1;
814
+ hash1 = ZSTD_hashPtr(ip2, hlog, mls);
815
+
816
+ /* advance to next positions */
817
+ ip0 = ip1;
818
+ ip1 = ip2;
819
+ ip2 = ip3;
820
+
821
+ /* write back hash table entry */
822
+ current0 = (U32)(ip0 - base);
823
+ hashTable[hash0] = current0;
824
+
825
+ { /* load match for ip[0] */
826
+ U32 const mval = idx >= dictStartIndex ?
827
+ MEM_read32(idxBase + idx) :
828
+ MEM_read32(ip0) ^ 1; /* guaranteed not to match */
829
+
830
+ /* check match at ip[0] */
831
+ if (MEM_read32(ip0) == mval) {
832
+ /* found a match! */
833
+ goto _offset;
834
+ } }
835
+
836
+ /* lookup ip[1] */
837
+ idx = hashTable[hash1];
838
+ idxBase = idx < prefixStartIndex ? dictBase : base;
839
+
840
+ /* hash ip[2] */
841
+ hash0 = hash1;
842
+ hash1 = ZSTD_hashPtr(ip2, hlog, mls);
843
+
844
+ /* advance to next positions */
845
+ ip0 = ip1;
846
+ ip1 = ip2;
847
+ ip2 = ip0 + step;
848
+ ip3 = ip1 + step;
849
+
850
+ /* calculate step */
851
+ if (ip2 >= nextStep) {
852
+ step++;
853
+ PREFETCH_L1(ip1 + 64);
854
+ PREFETCH_L1(ip1 + 128);
855
+ nextStep += kStepIncr;
856
+ }
857
+ } while (ip3 < ilimit);
858
+
859
+ _cleanup:
860
+ /* Note that there are probably still a couple positions we could search.
861
+ * However, it seems to be a meaningful performance hit to try to search
862
+ * them. So let's not. */
863
+
864
+ /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
865
+ * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
866
+ offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;
644
867
 
645
868
  /* save reps for next block */
646
- rep[0] = offset_1;
647
- rep[1] = offset_2;
869
+ rep[0] = offset_1 ? offset_1 : offsetSaved1;
870
+ rep[1] = offset_2 ? offset_2 : offsetSaved2;
648
871
 
649
872
  /* Return the last literals size */
650
873
  return (size_t)(iend - anchor);
874
+
875
+ _offset: /* Requires: ip0, idx, idxBase */
876
+
877
+ /* Compute the offset code. */
878
+ { U32 const offset = current0 - idx;
879
+ const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
880
+ matchEnd = idx < prefixStartIndex ? dictEnd : iend;
881
+ match0 = idxBase + idx;
882
+ offset_2 = offset_1;
883
+ offset_1 = offset;
884
+ offcode = OFFSET_TO_OFFBASE(offset);
885
+ mLength = 4;
886
+
887
+ /* Count the backwards match length. */
888
+ while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
889
+ ip0--;
890
+ match0--;
891
+ mLength++;
892
+ } }
893
+
894
+ _match: /* Requires: ip0, match0, offcode, matchEnd */
895
+
896
+ /* Count the forward length. */
897
+ assert(matchEnd != 0);
898
+ mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);
899
+
900
+ ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);
901
+
902
+ ip0 += mLength;
903
+ anchor = ip0;
904
+
905
+ /* write next hash table entry */
906
+ if (ip1 < ip0) {
907
+ hashTable[hash1] = (U32)(ip1 - base);
908
+ }
909
+
910
+ /* Fill table and check for immediate repcode. */
911
+ if (ip0 <= ilimit) {
912
+ /* Fill Table */
913
+ assert(base+current0+2 > istart); /* check base overflow */
914
+ hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2; /* here because current+2 could be > iend-8 */
915
+ hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
916
+
917
+ while (ip0 <= ilimit) {
918
+ U32 const repIndex2 = (U32)(ip0-base) - offset_2;
919
+ const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
920
+ if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 > 0)) /* intentional underflow */
921
+ && (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
922
+ const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
923
+ size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
924
+ { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; } /* swap offset_2 <=> offset_1 */
925
+ ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
926
+ hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
927
+ ip0 += repLength2;
928
+ anchor = ip0;
929
+ continue;
930
+ }
931
+ break;
932
+ } }
933
+
934
+ goto _start;
651
935
  }
652
936
 
653
937
  ZSTD_GEN_FAST_FN(extDict, 4, 0)
@@ -660,6 +944,7 @@ size_t ZSTD_compressBlock_fast_extDict(
660
944
  void const* src, size_t srcSize)
661
945
  {
662
946
  U32 const mls = ms->cParams.minMatch;
947
+ assert(ms->dictMatchState == NULL);
663
948
  switch(mls)
664
949
  {
665
950
  default: /* includes case 3 */