zstdlib 0.7.0 → 0.10.0

Sign up to get free protection for your applications and to get access to all the features.
Files changed (126) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGES.md +20 -0
  3. data/README.md +7 -1
  4. data/Rakefile +38 -8
  5. data/ext/{zstdlib → zstdlib_c}/extconf.rb +11 -6
  6. data/ext/{zstdlib → zstdlib_c}/ruby/zlib-2.2/zstdlib.c +2 -2
  7. data/ext/{zstdlib → zstdlib_c}/ruby/zlib-2.3/zstdlib.c +2 -2
  8. data/ext/{zstdlib → zstdlib_c}/ruby/zlib-2.4/zstdlib.c +2 -2
  9. data/ext/{zstdlib → zstdlib_c}/ruby/zlib-2.5/zstdlib.c +2 -2
  10. data/ext/{zstdlib → zstdlib_c}/ruby/zlib-2.6/zstdlib.c +2 -2
  11. data/ext/{zstdlib → zstdlib_c}/ruby/zlib-2.7/zstdlib.c +2 -2
  12. data/ext/zstdlib_c/ruby/zlib-3.0/zstdlib.c +4994 -0
  13. data/ext/zstdlib_c/ruby/zlib-3.1/zstdlib.c +5076 -0
  14. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/adler32.c +0 -0
  15. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/compress.c +0 -0
  16. data/ext/zstdlib_c/zlib-1.2.12/crc32.c +1116 -0
  17. data/ext/zstdlib_c/zlib-1.2.12/crc32.h +9446 -0
  18. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/deflate.c +78 -30
  19. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/deflate.h +12 -15
  20. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/gzclose.c +0 -0
  21. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/gzguts.h +3 -2
  22. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/gzlib.c +5 -3
  23. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/gzread.c +5 -7
  24. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/gzwrite.c +25 -13
  25. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/infback.c +2 -1
  26. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inffast.c +14 -14
  27. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inffast.h +0 -0
  28. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inffixed.h +0 -0
  29. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inflate.c +39 -8
  30. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inflate.h +3 -2
  31. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inftrees.c +3 -3
  32. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/inftrees.h +0 -0
  33. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/trees.c +27 -48
  34. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/trees.h +0 -0
  35. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/uncompr.c +0 -0
  36. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/zconf.h +0 -0
  37. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/zlib.h +123 -100
  38. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/zutil.c +2 -2
  39. data/ext/{zstdlib/zlib-1.2.11 → zstdlib_c/zlib-1.2.12}/zutil.h +12 -9
  40. data/ext/{zstdlib → zstdlib_c}/zlib.mk +0 -0
  41. data/ext/{zstdlib → zstdlib_c}/zlibwrapper/zlibwrapper.c +1 -5
  42. data/ext/{zstdlib → zstdlib_c}/zlibwrapper.mk +0 -0
  43. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/bitstream.h +46 -22
  44. data/ext/zstdlib_c/zstd-1.5.2/lib/common/compiler.h +335 -0
  45. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/cpu.h +1 -3
  46. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/debug.c +1 -1
  47. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/debug.h +12 -19
  48. data/ext/zstdlib_c/zstd-1.5.2/lib/common/entropy_common.c +368 -0
  49. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/error_private.c +2 -1
  50. data/ext/zstdlib_c/zstd-1.5.2/lib/common/error_private.h +159 -0
  51. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/fse.h +41 -12
  52. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/fse_decompress.c +139 -22
  53. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/huf.h +47 -23
  54. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/mem.h +87 -98
  55. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/pool.c +34 -23
  56. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/pool.h +4 -4
  57. data/ext/zstdlib_c/zstd-1.5.2/lib/common/portability_macros.h +137 -0
  58. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/threading.c +6 -5
  59. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/threading.h +0 -0
  60. data/ext/zstdlib_c/zstd-1.5.2/lib/common/xxhash.c +24 -0
  61. data/ext/zstdlib_c/zstd-1.5.2/lib/common/xxhash.h +5686 -0
  62. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/zstd_common.c +10 -10
  63. data/ext/zstdlib_c/zstd-1.5.2/lib/common/zstd_deps.h +111 -0
  64. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/common/zstd_internal.h +191 -145
  65. data/ext/zstdlib_c/zstd-1.5.2/lib/common/zstd_trace.h +163 -0
  66. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/clevels.h +134 -0
  67. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/fse_compress.c +89 -46
  68. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/hist.c +27 -29
  69. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/hist.h +2 -2
  70. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/huf_compress.c +1370 -0
  71. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress.c +2917 -868
  72. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_internal.h +458 -125
  73. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_literals.c +12 -11
  74. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_literals.h +4 -2
  75. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_sequences.c +41 -18
  76. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_sequences.h +1 -1
  77. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_superblock.c +26 -298
  78. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_compress_superblock.h +1 -1
  79. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_cwksp.h +234 -83
  80. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_double_fast.c +313 -138
  81. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_double_fast.h +1 -1
  82. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_fast.c +329 -150
  83. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_fast.h +1 -1
  84. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/zstd_lazy.c +2104 -0
  85. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/zstd_lazy.h +125 -0
  86. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_ldm.c +321 -216
  87. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_ldm.h +9 -2
  88. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/zstd_ldm_geartab.h +106 -0
  89. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_opt.c +412 -166
  90. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstd_opt.h +1 -1
  91. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/compress/zstdmt_compress.c +169 -453
  92. data/ext/zstdlib_c/zstd-1.5.2/lib/compress/zstdmt_compress.h +113 -0
  93. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/huf_decompress.c +1044 -403
  94. data/ext/zstdlib_c/zstd-1.5.2/lib/decompress/huf_decompress_amd64.S +585 -0
  95. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/zstd_ddict.c +9 -9
  96. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/zstd_ddict.h +2 -2
  97. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/zstd_decompress.c +450 -105
  98. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/zstd_decompress_block.c +913 -273
  99. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/zstd_decompress_block.h +14 -5
  100. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/decompress/zstd_decompress_internal.h +59 -12
  101. data/ext/zstdlib_c/zstd-1.5.2/lib/zdict.h +452 -0
  102. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/lib/zstd.h +699 -214
  103. data/ext/{zstdlib/zstd-1.4.5/lib/common → zstdlib_c/zstd-1.5.2/lib}/zstd_errors.h +2 -1
  104. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/gzclose.c +0 -0
  105. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/gzcompatibility.h +1 -1
  106. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/gzguts.h +0 -0
  107. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/gzlib.c +0 -0
  108. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/gzread.c +0 -0
  109. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/gzwrite.c +0 -0
  110. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/zstd_zlibwrapper.c +133 -44
  111. data/ext/{zstdlib/zstd-1.4.5 → zstdlib_c/zstd-1.5.2}/zlibWrapper/zstd_zlibwrapper.h +1 -1
  112. data/ext/zstdlib_c/zstd.mk +15 -0
  113. data/lib/zstdlib.rb +2 -2
  114. metadata +118 -109
  115. data/ext/zstdlib/zlib-1.2.11/crc32.c +0 -442
  116. data/ext/zstdlib/zlib-1.2.11/crc32.h +0 -441
  117. data/ext/zstdlib/zstd-1.4.5/lib/common/compiler.h +0 -175
  118. data/ext/zstdlib/zstd-1.4.5/lib/common/entropy_common.c +0 -216
  119. data/ext/zstdlib/zstd-1.4.5/lib/common/error_private.h +0 -80
  120. data/ext/zstdlib/zstd-1.4.5/lib/common/xxhash.c +0 -864
  121. data/ext/zstdlib/zstd-1.4.5/lib/common/xxhash.h +0 -285
  122. data/ext/zstdlib/zstd-1.4.5/lib/compress/huf_compress.c +0 -798
  123. data/ext/zstdlib/zstd-1.4.5/lib/compress/zstd_lazy.c +0 -1138
  124. data/ext/zstdlib/zstd-1.4.5/lib/compress/zstd_lazy.h +0 -67
  125. data/ext/zstdlib/zstd-1.4.5/lib/compress/zstdmt_compress.h +0 -192
  126. data/ext/zstdlib/zstd.mk +0 -14
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
2
+ * Copyright (c) Yann Collet, Facebook, Inc.
3
3
  * All rights reserved.
4
4
  *
5
5
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,13 +11,126 @@
11
11
  #include "zstd_ldm.h"
12
12
 
13
13
  #include "../common/debug.h"
14
+ #include "../common/xxhash.h"
14
15
  #include "zstd_fast.h" /* ZSTD_fillHashTable() */
15
16
  #include "zstd_double_fast.h" /* ZSTD_fillDoubleHashTable() */
17
+ #include "zstd_ldm_geartab.h"
16
18
 
17
19
  #define LDM_BUCKET_SIZE_LOG 3
18
20
  #define LDM_MIN_MATCH_LENGTH 64
19
21
  #define LDM_HASH_RLOG 7
20
- #define LDM_HASH_CHAR_OFFSET 10
22
+
23
+ typedef struct {
24
+ U64 rolling;
25
+ U64 stopMask;
26
+ } ldmRollingHashState_t;
27
+
28
+ /** ZSTD_ldm_gear_init():
29
+ *
30
+ * Initializes the rolling hash state such that it will honor the
31
+ * settings in params. */
32
+ static void ZSTD_ldm_gear_init(ldmRollingHashState_t* state, ldmParams_t const* params)
33
+ {
34
+ unsigned maxBitsInMask = MIN(params->minMatchLength, 64);
35
+ unsigned hashRateLog = params->hashRateLog;
36
+
37
+ state->rolling = ~(U32)0;
38
+
39
+ /* The choice of the splitting criterion is subject to two conditions:
40
+ * 1. it has to trigger on average every 2^(hashRateLog) bytes;
41
+ * 2. ideally, it has to depend on a window of minMatchLength bytes.
42
+ *
43
+ * In the gear hash algorithm, bit n depends on the last n bytes;
44
+ * so in order to obtain a good quality splitting criterion it is
45
+ * preferable to use bits with high weight.
46
+ *
47
+ * To match condition 1 we use a mask with hashRateLog bits set
48
+ * and, because of the previous remark, we make sure these bits
49
+ * have the highest possible weight while still respecting
50
+ * condition 2.
51
+ */
52
+ if (hashRateLog > 0 && hashRateLog <= maxBitsInMask) {
53
+ state->stopMask = (((U64)1 << hashRateLog) - 1) << (maxBitsInMask - hashRateLog);
54
+ } else {
55
+ /* In this degenerate case we simply honor the hash rate. */
56
+ state->stopMask = ((U64)1 << hashRateLog) - 1;
57
+ }
58
+ }
59
+
60
+ /** ZSTD_ldm_gear_reset()
61
+ * Feeds [data, data + minMatchLength) into the hash without registering any
62
+ * splits. This effectively resets the hash state. This is used when skipping
63
+ * over data, either at the beginning of a block, or skipping sections.
64
+ */
65
+ static void ZSTD_ldm_gear_reset(ldmRollingHashState_t* state,
66
+ BYTE const* data, size_t minMatchLength)
67
+ {
68
+ U64 hash = state->rolling;
69
+ size_t n = 0;
70
+
71
+ #define GEAR_ITER_ONCE() do { \
72
+ hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
73
+ n += 1; \
74
+ } while (0)
75
+ while (n + 3 < minMatchLength) {
76
+ GEAR_ITER_ONCE();
77
+ GEAR_ITER_ONCE();
78
+ GEAR_ITER_ONCE();
79
+ GEAR_ITER_ONCE();
80
+ }
81
+ while (n < minMatchLength) {
82
+ GEAR_ITER_ONCE();
83
+ }
84
+ #undef GEAR_ITER_ONCE
85
+ }
86
+
87
+ /** ZSTD_ldm_gear_feed():
88
+ *
89
+ * Registers in the splits array all the split points found in the first
90
+ * size bytes following the data pointer. This function terminates when
91
+ * either all the data has been processed or LDM_BATCH_SIZE splits are
92
+ * present in the splits array.
93
+ *
94
+ * Precondition: The splits array must not be full.
95
+ * Returns: The number of bytes processed. */
96
+ static size_t ZSTD_ldm_gear_feed(ldmRollingHashState_t* state,
97
+ BYTE const* data, size_t size,
98
+ size_t* splits, unsigned* numSplits)
99
+ {
100
+ size_t n;
101
+ U64 hash, mask;
102
+
103
+ hash = state->rolling;
104
+ mask = state->stopMask;
105
+ n = 0;
106
+
107
+ #define GEAR_ITER_ONCE() do { \
108
+ hash = (hash << 1) + ZSTD_ldm_gearTab[data[n] & 0xff]; \
109
+ n += 1; \
110
+ if (UNLIKELY((hash & mask) == 0)) { \
111
+ splits[*numSplits] = n; \
112
+ *numSplits += 1; \
113
+ if (*numSplits == LDM_BATCH_SIZE) \
114
+ goto done; \
115
+ } \
116
+ } while (0)
117
+
118
+ while (n + 3 < size) {
119
+ GEAR_ITER_ONCE();
120
+ GEAR_ITER_ONCE();
121
+ GEAR_ITER_ONCE();
122
+ GEAR_ITER_ONCE();
123
+ }
124
+ while (n < size) {
125
+ GEAR_ITER_ONCE();
126
+ }
127
+
128
+ #undef GEAR_ITER_ONCE
129
+
130
+ done:
131
+ state->rolling = hash;
132
+ return n;
133
+ }
21
134
 
22
135
  void ZSTD_ldm_adjustParameters(ldmParams_t* params,
23
136
  ZSTD_compressionParameters const* cParams)
@@ -27,13 +140,6 @@ void ZSTD_ldm_adjustParameters(ldmParams_t* params,
27
140
  DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
28
141
  if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
29
142
  if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
30
- if (cParams->strategy >= ZSTD_btopt) {
31
- /* Get out of the way of the optimal parser */
32
- U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
33
- assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
34
- assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
35
- params->minMatchLength = minMatch;
36
- }
37
143
  if (params->hashLog == 0) {
38
144
  params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
39
145
  assert(params->hashLog <= ZSTD_HASHLOG_MAX);
@@ -53,47 +159,12 @@ size_t ZSTD_ldm_getTableSize(ldmParams_t params)
53
159
  size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
54
160
  size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
55
161
  + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
56
- return params.enableLdm ? totalSize : 0;
162
+ return params.enableLdm == ZSTD_ps_enable ? totalSize : 0;
57
163
  }
58
164
 
59
165
  size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
60
166
  {
61
- return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
62
- }
63
-
64
- /** ZSTD_ldm_getSmallHash() :
65
- * numBits should be <= 32
66
- * If numBits==0, returns 0.
67
- * @return : the most significant numBits of value. */
68
- static U32 ZSTD_ldm_getSmallHash(U64 value, U32 numBits)
69
- {
70
- assert(numBits <= 32);
71
- return numBits == 0 ? 0 : (U32)(value >> (64 - numBits));
72
- }
73
-
74
- /** ZSTD_ldm_getChecksum() :
75
- * numBitsToDiscard should be <= 32
76
- * @return : the next most significant 32 bits after numBitsToDiscard */
77
- static U32 ZSTD_ldm_getChecksum(U64 hash, U32 numBitsToDiscard)
78
- {
79
- assert(numBitsToDiscard <= 32);
80
- return (hash >> (64 - 32 - numBitsToDiscard)) & 0xFFFFFFFF;
81
- }
82
-
83
- /** ZSTD_ldm_getTag() ;
84
- * Given the hash, returns the most significant numTagBits bits
85
- * after (32 + hbits) bits.
86
- *
87
- * If there are not enough bits remaining, return the last
88
- * numTagBits bits. */
89
- static U32 ZSTD_ldm_getTag(U64 hash, U32 hbits, U32 numTagBits)
90
- {
91
- assert(numTagBits < 32 && hbits <= 32);
92
- if (32 - hbits < numTagBits) {
93
- return hash & (((U32)1 << numTagBits) - 1);
94
- } else {
95
- return (hash >> (32 - hbits - numTagBits)) & (((U32)1 << numTagBits) - 1);
96
- }
167
+ return params.enableLdm == ZSTD_ps_enable ? (maxChunkSize / params.minMatchLength) : 0;
97
168
  }
98
169
 
99
170
  /** ZSTD_ldm_getBucket() :
@@ -110,38 +181,12 @@ static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
110
181
  size_t const hash, const ldmEntry_t entry,
111
182
  ldmParams_t const ldmParams)
112
183
  {
113
- BYTE* const bucketOffsets = ldmState->bucketOffsets;
114
- *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + bucketOffsets[hash]) = entry;
115
- bucketOffsets[hash]++;
116
- bucketOffsets[hash] &= ((U32)1 << ldmParams.bucketSizeLog) - 1;
117
- }
184
+ BYTE* const pOffset = ldmState->bucketOffsets + hash;
185
+ unsigned const offset = *pOffset;
186
+
187
+ *(ZSTD_ldm_getBucket(ldmState, hash, ldmParams) + offset) = entry;
188
+ *pOffset = (BYTE)((offset + 1) & ((1u << ldmParams.bucketSizeLog) - 1));
118
189
 
119
- /** ZSTD_ldm_makeEntryAndInsertByTag() :
120
- *
121
- * Gets the small hash, checksum, and tag from the rollingHash.
122
- *
123
- * If the tag matches (1 << ldmParams.hashRateLog)-1, then
124
- * creates an ldmEntry from the offset, and inserts it into the hash table.
125
- *
126
- * hBits is the length of the small hash, which is the most significant hBits
127
- * of rollingHash. The checksum is the next 32 most significant bits, followed
128
- * by ldmParams.hashRateLog bits that make up the tag. */
129
- static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
130
- U64 const rollingHash,
131
- U32 const hBits,
132
- U32 const offset,
133
- ldmParams_t const ldmParams)
134
- {
135
- U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashRateLog);
136
- U32 const tagMask = ((U32)1 << ldmParams.hashRateLog) - 1;
137
- if (tag == tagMask) {
138
- U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
139
- U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
140
- ldmEntry_t entry;
141
- entry.offset = offset;
142
- entry.checksum = checksum;
143
- ZSTD_ldm_insertEntry(ldmState, hash, entry, ldmParams);
144
- }
145
190
  }
146
191
 
147
192
  /** ZSTD_ldm_countBackwardsMatch() :
@@ -150,10 +195,10 @@ static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
150
195
  * We count only bytes where pMatch >= pBase and pIn >= pAnchor. */
151
196
  static size_t ZSTD_ldm_countBackwardsMatch(
152
197
  const BYTE* pIn, const BYTE* pAnchor,
153
- const BYTE* pMatch, const BYTE* pBase)
198
+ const BYTE* pMatch, const BYTE* pMatchBase)
154
199
  {
155
200
  size_t matchLength = 0;
156
- while (pIn > pAnchor && pMatch > pBase && pIn[-1] == pMatch[-1]) {
201
+ while (pIn > pAnchor && pMatch > pMatchBase && pIn[-1] == pMatch[-1]) {
157
202
  pIn--;
158
203
  pMatch--;
159
204
  matchLength++;
@@ -161,6 +206,27 @@ static size_t ZSTD_ldm_countBackwardsMatch(
161
206
  return matchLength;
162
207
  }
163
208
 
209
+ /** ZSTD_ldm_countBackwardsMatch_2segments() :
210
+ * Returns the number of bytes that match backwards from pMatch,
211
+ * even with the backwards match spanning 2 different segments.
212
+ *
213
+ * On reaching `pMatchBase`, start counting from mEnd */
214
+ static size_t ZSTD_ldm_countBackwardsMatch_2segments(
215
+ const BYTE* pIn, const BYTE* pAnchor,
216
+ const BYTE* pMatch, const BYTE* pMatchBase,
217
+ const BYTE* pExtDictStart, const BYTE* pExtDictEnd)
218
+ {
219
+ size_t matchLength = ZSTD_ldm_countBackwardsMatch(pIn, pAnchor, pMatch, pMatchBase);
220
+ if (pMatch - matchLength != pMatchBase || pMatchBase == pExtDictStart) {
221
+ /* If backwards match is entirely in the extDict or prefix, immediately return */
222
+ return matchLength;
223
+ }
224
+ DEBUGLOG(7, "ZSTD_ldm_countBackwardsMatch_2segments: found 2-parts backwards match (length in prefix==%zu)", matchLength);
225
+ matchLength += ZSTD_ldm_countBackwardsMatch(pIn - matchLength, pAnchor, pExtDictEnd, pExtDictStart);
226
+ DEBUGLOG(7, "final backwards match length = %zu", matchLength);
227
+ return matchLength;
228
+ }
229
+
164
230
  /** ZSTD_ldm_fillFastTables() :
165
231
  *
166
232
  * Fills the relevant tables for the ZSTD_fast and ZSTD_dfast strategies.
@@ -198,43 +264,42 @@ static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
198
264
  return 0;
199
265
  }
200
266
 
201
- /** ZSTD_ldm_fillLdmHashTable() :
202
- *
203
- * Fills hashTable from (lastHashed + 1) to iend (non-inclusive).
204
- * lastHash is the rolling hash that corresponds to lastHashed.
205
- *
206
- * Returns the rolling hash corresponding to position iend-1. */
207
- static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
208
- U64 lastHash, const BYTE* lastHashed,
209
- const BYTE* iend, const BYTE* base,
210
- U32 hBits, ldmParams_t const ldmParams)
211
- {
212
- U64 rollingHash = lastHash;
213
- const BYTE* cur = lastHashed + 1;
214
-
215
- while (cur < iend) {
216
- rollingHash = ZSTD_rollingHash_rotate(rollingHash, cur[-1],
217
- cur[ldmParams.minMatchLength-1],
218
- state->hashPower);
219
- ZSTD_ldm_makeEntryAndInsertByTag(state,
220
- rollingHash, hBits,
221
- (U32)(cur - base), ldmParams);
222
- ++cur;
223
- }
224
- return rollingHash;
225
- }
226
-
227
267
  void ZSTD_ldm_fillHashTable(
228
- ldmState_t* state, const BYTE* ip,
268
+ ldmState_t* ldmState, const BYTE* ip,
229
269
  const BYTE* iend, ldmParams_t const* params)
230
270
  {
271
+ U32 const minMatchLength = params->minMatchLength;
272
+ U32 const hBits = params->hashLog - params->bucketSizeLog;
273
+ BYTE const* const base = ldmState->window.base;
274
+ BYTE const* const istart = ip;
275
+ ldmRollingHashState_t hashState;
276
+ size_t* const splits = ldmState->splitIndices;
277
+ unsigned numSplits;
278
+
231
279
  DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
232
- if ((size_t)(iend - ip) >= params->minMatchLength) {
233
- U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
234
- ZSTD_ldm_fillLdmHashTable(
235
- state, startingHash, ip, iend - params->minMatchLength, state->window.base,
236
- params->hashLog - params->bucketSizeLog,
237
- *params);
280
+
281
+ ZSTD_ldm_gear_init(&hashState, params);
282
+ while (ip < iend) {
283
+ size_t hashed;
284
+ unsigned n;
285
+
286
+ numSplits = 0;
287
+ hashed = ZSTD_ldm_gear_feed(&hashState, ip, iend - ip, splits, &numSplits);
288
+
289
+ for (n = 0; n < numSplits; n++) {
290
+ if (ip + splits[n] >= istart + minMatchLength) {
291
+ BYTE const* const split = ip + splits[n] - minMatchLength;
292
+ U64 const xxhash = XXH64(split, minMatchLength, 0);
293
+ U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
294
+ ldmEntry_t entry;
295
+
296
+ entry.offset = (U32)(split - base);
297
+ entry.checksum = (U32)(xxhash >> 32);
298
+ ZSTD_ldm_insertEntry(ldmState, hash, entry, *params);
299
+ }
300
+ }
301
+
302
+ ip += hashed;
238
303
  }
239
304
  }
240
305
 
@@ -246,10 +311,10 @@ void ZSTD_ldm_fillHashTable(
246
311
  * (after a long match, only update tables a limited amount). */
247
312
  static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
248
313
  {
249
- U32 const current = (U32)(anchor - ms->window.base);
250
- if (current > ms->nextToUpdate + 1024) {
314
+ U32 const curr = (U32)(anchor - ms->window.base);
315
+ if (curr > ms->nextToUpdate + 1024) {
251
316
  ms->nextToUpdate =
252
- current - MIN(512, current - ms->nextToUpdate - 1024);
317
+ curr - MIN(512, curr - ms->nextToUpdate - 1024);
253
318
  }
254
319
  }
255
320
 
@@ -260,11 +325,8 @@ static size_t ZSTD_ldm_generateSequences_internal(
260
325
  /* LDM parameters */
261
326
  int const extDict = ZSTD_window_hasExtDict(ldmState->window);
262
327
  U32 const minMatchLength = params->minMatchLength;
263
- U64 const hashPower = ldmState->hashPower;
328
+ U32 const entsPerBucket = 1U << params->bucketSizeLog;
264
329
  U32 const hBits = params->hashLog - params->bucketSizeLog;
265
- U32 const ldmBucketSize = 1U << params->bucketSizeLog;
266
- U32 const hashRateLog = params->hashRateLog;
267
- U32 const ldmTagMask = (1U << params->hashRateLog) - 1;
268
330
  /* Prefix and extDict parameters */
269
331
  U32 const dictLimit = ldmState->window.dictLimit;
270
332
  U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
@@ -276,45 +338,69 @@ static size_t ZSTD_ldm_generateSequences_internal(
276
338
  /* Input bounds */
277
339
  BYTE const* const istart = (BYTE const*)src;
278
340
  BYTE const* const iend = istart + srcSize;
279
- BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
341
+ BYTE const* const ilimit = iend - HASH_READ_SIZE;
280
342
  /* Input positions */
281
343
  BYTE const* anchor = istart;
282
344
  BYTE const* ip = istart;
283
- /* Rolling hash */
284
- BYTE const* lastHashed = NULL;
285
- U64 rollingHash = 0;
286
-
287
- while (ip <= ilimit) {
288
- size_t mLength;
289
- U32 const current = (U32)(ip - base);
290
- size_t forwardMatchLength = 0, backwardMatchLength = 0;
291
- ldmEntry_t* bestEntry = NULL;
292
- if (ip != istart) {
293
- rollingHash = ZSTD_rollingHash_rotate(rollingHash, lastHashed[0],
294
- lastHashed[minMatchLength],
295
- hashPower);
296
- } else {
297
- rollingHash = ZSTD_rollingHash_compute(ip, minMatchLength);
345
+ /* Rolling hash state */
346
+ ldmRollingHashState_t hashState;
347
+ /* Arrays for staged-processing */
348
+ size_t* const splits = ldmState->splitIndices;
349
+ ldmMatchCandidate_t* const candidates = ldmState->matchCandidates;
350
+ unsigned numSplits;
351
+
352
+ if (srcSize < minMatchLength)
353
+ return iend - anchor;
354
+
355
+ /* Initialize the rolling hash state with the first minMatchLength bytes */
356
+ ZSTD_ldm_gear_init(&hashState, params);
357
+ ZSTD_ldm_gear_reset(&hashState, ip, minMatchLength);
358
+ ip += minMatchLength;
359
+
360
+ while (ip < ilimit) {
361
+ size_t hashed;
362
+ unsigned n;
363
+
364
+ numSplits = 0;
365
+ hashed = ZSTD_ldm_gear_feed(&hashState, ip, ilimit - ip,
366
+ splits, &numSplits);
367
+
368
+ for (n = 0; n < numSplits; n++) {
369
+ BYTE const* const split = ip + splits[n] - minMatchLength;
370
+ U64 const xxhash = XXH64(split, minMatchLength, 0);
371
+ U32 const hash = (U32)(xxhash & (((U32)1 << hBits) - 1));
372
+
373
+ candidates[n].split = split;
374
+ candidates[n].hash = hash;
375
+ candidates[n].checksum = (U32)(xxhash >> 32);
376
+ candidates[n].bucket = ZSTD_ldm_getBucket(ldmState, hash, *params);
377
+ PREFETCH_L1(candidates[n].bucket);
298
378
  }
299
- lastHashed = ip;
300
379
 
301
- /* Do not insert and do not look for a match */
302
- if (ZSTD_ldm_getTag(rollingHash, hBits, hashRateLog) != ldmTagMask) {
303
- ip++;
304
- continue;
305
- }
380
+ for (n = 0; n < numSplits; n++) {
381
+ size_t forwardMatchLength = 0, backwardMatchLength = 0,
382
+ bestMatchLength = 0, mLength;
383
+ U32 offset;
384
+ BYTE const* const split = candidates[n].split;
385
+ U32 const checksum = candidates[n].checksum;
386
+ U32 const hash = candidates[n].hash;
387
+ ldmEntry_t* const bucket = candidates[n].bucket;
388
+ ldmEntry_t const* cur;
389
+ ldmEntry_t const* bestEntry = NULL;
390
+ ldmEntry_t newEntry;
391
+
392
+ newEntry.offset = (U32)(split - base);
393
+ newEntry.checksum = checksum;
394
+
395
+ /* If a split point would generate a sequence overlapping with
396
+ * the previous one, we merely register it in the hash table and
397
+ * move on */
398
+ if (split < anchor) {
399
+ ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
400
+ continue;
401
+ }
306
402
 
307
- /* Get the best entry and compute the match lengths */
308
- {
309
- ldmEntry_t* const bucket =
310
- ZSTD_ldm_getBucket(ldmState,
311
- ZSTD_ldm_getSmallHash(rollingHash, hBits),
312
- *params);
313
- ldmEntry_t* cur;
314
- size_t bestMatchLength = 0;
315
- U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
316
-
317
- for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
403
+ for (cur = bucket; cur < bucket + entsPerBucket; cur++) {
318
404
  size_t curForwardMatchLength, curBackwardMatchLength,
319
405
  curTotalMatchLength;
320
406
  if (cur->checksum != checksum || cur->offset <= lowestIndex) {
@@ -328,30 +414,23 @@ static size_t ZSTD_ldm_generateSequences_internal(
328
414
  cur->offset < dictLimit ? dictEnd : iend;
329
415
  BYTE const* const lowMatchPtr =
330
416
  cur->offset < dictLimit ? dictStart : lowPrefixPtr;
331
-
332
- curForwardMatchLength = ZSTD_count_2segments(
333
- ip, pMatch, iend,
334
- matchEnd, lowPrefixPtr);
417
+ curForwardMatchLength =
418
+ ZSTD_count_2segments(split, pMatch, iend, matchEnd, lowPrefixPtr);
335
419
  if (curForwardMatchLength < minMatchLength) {
336
420
  continue;
337
421
  }
338
- curBackwardMatchLength =
339
- ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
340
- lowMatchPtr);
341
- curTotalMatchLength = curForwardMatchLength +
342
- curBackwardMatchLength;
422
+ curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch_2segments(
423
+ split, anchor, pMatch, lowMatchPtr, dictStart, dictEnd);
343
424
  } else { /* !extDict */
344
425
  BYTE const* const pMatch = base + cur->offset;
345
- curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
426
+ curForwardMatchLength = ZSTD_count(split, pMatch, iend);
346
427
  if (curForwardMatchLength < minMatchLength) {
347
428
  continue;
348
429
  }
349
430
  curBackwardMatchLength =
350
- ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
351
- lowPrefixPtr);
352
- curTotalMatchLength = curForwardMatchLength +
353
- curBackwardMatchLength;
431
+ ZSTD_ldm_countBackwardsMatch(split, anchor, pMatch, lowPrefixPtr);
354
432
  }
433
+ curTotalMatchLength = curForwardMatchLength + curBackwardMatchLength;
355
434
 
356
435
  if (curTotalMatchLength > bestMatchLength) {
357
436
  bestMatchLength = curTotalMatchLength;
@@ -360,57 +439,54 @@ static size_t ZSTD_ldm_generateSequences_internal(
360
439
  bestEntry = cur;
361
440
  }
362
441
  }
363
- }
364
-
365
- /* No match found -- continue searching */
366
- if (bestEntry == NULL) {
367
- ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
368
- hBits, current,
369
- *params);
370
- ip++;
371
- continue;
372
- }
373
442
 
374
- /* Match found */
375
- mLength = forwardMatchLength + backwardMatchLength;
376
- ip -= backwardMatchLength;
443
+ /* No match found -- insert an entry into the hash table
444
+ * and process the next candidate match */
445
+ if (bestEntry == NULL) {
446
+ ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
447
+ continue;
448
+ }
377
449
 
378
- {
379
- /* Store the sequence:
380
- * ip = current - backwardMatchLength
381
- * The match is at (bestEntry->offset - backwardMatchLength)
382
- */
383
- U32 const matchIndex = bestEntry->offset;
384
- U32 const offset = current - matchIndex;
385
- rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
386
-
387
- /* Out of sequence storage */
388
- if (rawSeqStore->size == rawSeqStore->capacity)
389
- return ERROR(dstSize_tooSmall);
390
- seq->litLength = (U32)(ip - anchor);
391
- seq->matchLength = (U32)mLength;
392
- seq->offset = offset;
393
- rawSeqStore->size++;
394
- }
450
+ /* Match found */
451
+ offset = (U32)(split - base) - bestEntry->offset;
452
+ mLength = forwardMatchLength + backwardMatchLength;
453
+ {
454
+ rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
455
+
456
+ /* Out of sequence storage */
457
+ if (rawSeqStore->size == rawSeqStore->capacity)
458
+ return ERROR(dstSize_tooSmall);
459
+ seq->litLength = (U32)(split - backwardMatchLength - anchor);
460
+ seq->matchLength = (U32)mLength;
461
+ seq->offset = offset;
462
+ rawSeqStore->size++;
463
+ }
395
464
 
396
- /* Insert the current entry into the hash table */
397
- ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
398
- (U32)(lastHashed - base),
399
- *params);
465
+ /* Insert the current entry into the hash table --- it must be
466
+ * done after the previous block to avoid clobbering bestEntry */
467
+ ZSTD_ldm_insertEntry(ldmState, hash, newEntry, *params);
400
468
 
401
- assert(ip + backwardMatchLength == lastHashed);
469
+ anchor = split + forwardMatchLength;
402
470
 
403
- /* Fill the hash table from lastHashed+1 to ip+mLength*/
404
- /* Heuristic: don't need to fill the entire table at end of block */
405
- if (ip + mLength <= ilimit) {
406
- rollingHash = ZSTD_ldm_fillLdmHashTable(
407
- ldmState, rollingHash, lastHashed,
408
- ip + mLength, base, hBits, *params);
409
- lastHashed = ip + mLength - 1;
471
+ /* If we find a match that ends after the data that we've hashed
472
+ * then we have a repeating, overlapping, pattern. E.g. all zeros.
473
+ * If one repetition of the pattern matches our `stopMask` then all
474
+ * repetitions will. We don't need to insert them all into out table,
475
+ * only the first one. So skip over overlapping matches.
476
+ * This is a major speed boost (20x) for compressing a single byte
477
+ * repeated, when that byte ends up in the table.
478
+ */
479
+ if (anchor > ip + hashed) {
480
+ ZSTD_ldm_gear_reset(&hashState, anchor - minMatchLength, minMatchLength);
481
+ /* Continue the outer loop at anchor (ip + hashed == anchor). */
482
+ ip = anchor - hashed;
483
+ break;
484
+ }
410
485
  }
411
- ip += mLength;
412
- anchor = ip;
486
+
487
+ ip += hashed;
413
488
  }
489
+
414
490
  return iend - anchor;
415
491
  }
416
492
 
@@ -459,7 +535,7 @@ size_t ZSTD_ldm_generateSequences(
459
535
 
460
536
  assert(chunkStart < iend);
461
537
  /* 1. Perform overflow correction if necessary. */
462
- if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
538
+ if (ZSTD_window_needOverflowCorrection(ldmState->window, 0, maxDist, ldmState->loadedDictEnd, chunkStart, chunkEnd)) {
463
539
  U32 const ldmHSize = 1U << params->hashLog;
464
540
  U32 const correction = ZSTD_window_correctOverflow(
465
541
  &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
@@ -503,7 +579,9 @@ size_t ZSTD_ldm_generateSequences(
503
579
  return 0;
504
580
  }
505
581
 
506
- void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
582
+ void
583
+ ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch)
584
+ {
507
585
  while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
508
586
  rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
509
587
  if (srcSize <= seq->litLength) {
@@ -562,14 +640,32 @@ static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
562
640
  return sequence;
563
641
  }
564
642
 
643
+ void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
644
+ U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
645
+ while (currPos && rawSeqStore->pos < rawSeqStore->size) {
646
+ rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
647
+ if (currPos >= currSeq.litLength + currSeq.matchLength) {
648
+ currPos -= currSeq.litLength + currSeq.matchLength;
649
+ rawSeqStore->pos++;
650
+ } else {
651
+ rawSeqStore->posInSequence = currPos;
652
+ break;
653
+ }
654
+ }
655
+ if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
656
+ rawSeqStore->posInSequence = 0;
657
+ }
658
+ }
659
+
565
660
  size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
566
661
  ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
662
+ ZSTD_paramSwitch_e useRowMatchFinder,
567
663
  void const* src, size_t srcSize)
568
664
  {
569
665
  const ZSTD_compressionParameters* const cParams = &ms->cParams;
570
666
  unsigned const minMatch = cParams->minMatch;
571
667
  ZSTD_blockCompressor const blockCompressor =
572
- ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
668
+ ZSTD_selectBlockCompressor(cParams->strategy, useRowMatchFinder, ZSTD_matchState_dictMode(ms));
573
669
  /* Input bounds */
574
670
  BYTE const* const istart = (BYTE const*)src;
575
671
  BYTE const* const iend = istart + srcSize;
@@ -577,9 +673,18 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
577
673
  BYTE const* ip = istart;
578
674
 
579
675
  DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
676
+ /* If using opt parser, use LDMs only as candidates rather than always accepting them */
677
+ if (cParams->strategy >= ZSTD_btopt) {
678
+ size_t lastLLSize;
679
+ ms->ldmSeqStore = rawSeqStore;
680
+ lastLLSize = blockCompressor(ms, seqStore, rep, src, srcSize);
681
+ ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore, srcSize);
682
+ return lastLLSize;
683
+ }
684
+
580
685
  assert(rawSeqStore->pos <= rawSeqStore->size);
581
686
  assert(rawSeqStore->size <= rawSeqStore->capacity);
582
- /* Loop through each sequence and apply the block compressor to the lits */
687
+ /* Loop through each sequence and apply the block compressor to the literals */
583
688
  while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
584
689
  /* maybeSplitSequence updates rawSeqStore->pos */
585
690
  rawSeq const sequence = maybeSplitSequence(rawSeqStore,
@@ -606,8 +711,8 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
606
711
  rep[0] = sequence.offset;
607
712
  /* Store the sequence */
608
713
  ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
609
- sequence.offset + ZSTD_REP_MOVE,
610
- sequence.matchLength - MINMATCH);
714
+ STORE_OFFSET(sequence.offset),
715
+ sequence.matchLength);
611
716
  ip += sequence.matchLength;
612
717
  }
613
718
  }