zstdlib 0.3.0-x86-mingw32 → 0.8.0-x86-mingw32

Sign up to get free protection for your applications and to get access to all the features.
Files changed (88) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGES.md +30 -1
  3. data/README.md +2 -2
  4. data/Rakefile +1 -1
  5. data/ext/zstdlib/extconf.rb +3 -3
  6. data/ext/zstdlib/ruby/zlib-2.7/zstdlib.c +4895 -0
  7. data/ext/zstdlib/ruby/zlib-3.0/zstdlib.c +4994 -0
  8. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/bitstream.h +59 -51
  9. data/ext/zstdlib/zstd-1.5.0/lib/common/compiler.h +289 -0
  10. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/cpu.h +1 -3
  11. data/ext/zstdlib/zstd-1.5.0/lib/common/debug.c +24 -0
  12. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/debug.h +22 -49
  13. data/ext/zstdlib/zstd-1.5.0/lib/common/entropy_common.c +362 -0
  14. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/error_private.c +3 -1
  15. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/error_private.h +8 -4
  16. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/fse.h +50 -42
  17. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/fse_decompress.c +149 -55
  18. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/huf.h +43 -39
  19. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/mem.h +69 -25
  20. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/pool.c +30 -20
  21. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/pool.h +3 -3
  22. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/threading.c +51 -4
  23. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/threading.h +36 -4
  24. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/xxhash.c +40 -92
  25. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/xxhash.h +12 -32
  26. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/common/zstd_common.c +10 -10
  27. data/ext/zstdlib/zstd-1.5.0/lib/common/zstd_deps.h +111 -0
  28. data/ext/zstdlib/zstd-1.5.0/lib/common/zstd_internal.h +490 -0
  29. data/ext/zstdlib/zstd-1.5.0/lib/common/zstd_trace.h +154 -0
  30. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/fse_compress.c +47 -63
  31. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/hist.c +41 -63
  32. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/hist.h +13 -33
  33. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/huf_compress.c +332 -193
  34. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_compress.c +6393 -0
  35. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_compress_internal.h +522 -86
  36. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_compress_literals.c +25 -16
  37. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_compress_literals.h +2 -2
  38. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_compress_sequences.c +50 -24
  39. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_compress_sequences.h +11 -4
  40. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_compress_superblock.c +572 -0
  41. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_compress_superblock.h +32 -0
  42. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_cwksp.h +662 -0
  43. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_double_fast.c +43 -41
  44. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_double_fast.h +2 -2
  45. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_fast.c +85 -80
  46. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_fast.h +2 -2
  47. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_lazy.c +2184 -0
  48. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_lazy.h +125 -0
  49. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_ldm.c +333 -208
  50. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_ldm.h +15 -3
  51. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstd_ldm_geartab.h +103 -0
  52. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_opt.c +228 -129
  53. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstd_opt.h +1 -1
  54. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/compress/zstdmt_compress.c +151 -440
  55. data/ext/zstdlib/zstd-1.5.0/lib/compress/zstdmt_compress.h +110 -0
  56. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/huf_decompress.c +395 -276
  57. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/zstd_ddict.c +20 -16
  58. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/zstd_ddict.h +3 -3
  59. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/zstd_decompress.c +628 -231
  60. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/zstd_decompress_block.c +606 -380
  61. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/zstd_decompress_block.h +8 -5
  62. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/decompress/zstd_decompress_internal.h +39 -9
  63. data/ext/zstdlib/zstd-1.5.0/lib/zdict.h +452 -0
  64. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/lib/zstd.h +740 -153
  65. data/ext/zstdlib/{zstd-1.4.2/lib/common → zstd-1.5.0/lib}/zstd_errors.h +3 -1
  66. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/gzclose.c +1 -1
  67. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/gzcompatibility.h +1 -1
  68. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/gzguts.h +0 -0
  69. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/gzlib.c +9 -9
  70. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/gzread.c +16 -8
  71. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/gzwrite.c +8 -8
  72. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/zstd_zlibwrapper.c +131 -45
  73. data/ext/zstdlib/{zstd-1.4.2 → zstd-1.5.0}/zlibWrapper/zstd_zlibwrapper.h +1 -1
  74. data/lib/2.2/zstdlib.so +0 -0
  75. data/lib/2.3/zstdlib.so +0 -0
  76. data/lib/2.4/zstdlib.so +0 -0
  77. data/lib/2.5/zstdlib.so +0 -0
  78. data/lib/2.6/zstdlib.so +0 -0
  79. data/lib/2.7/zstdlib.so +0 -0
  80. metadata +76 -67
  81. data/ext/zstdlib/zstd-1.4.2/lib/common/compiler.h +0 -147
  82. data/ext/zstdlib/zstd-1.4.2/lib/common/debug.c +0 -44
  83. data/ext/zstdlib/zstd-1.4.2/lib/common/entropy_common.c +0 -236
  84. data/ext/zstdlib/zstd-1.4.2/lib/common/zstd_internal.h +0 -371
  85. data/ext/zstdlib/zstd-1.4.2/lib/compress/zstd_compress.c +0 -3904
  86. data/ext/zstdlib/zstd-1.4.2/lib/compress/zstd_lazy.c +0 -1111
  87. data/ext/zstdlib/zstd-1.4.2/lib/compress/zstd_lazy.h +0 -67
  88. data/ext/zstdlib/zstd-1.4.2/lib/compress/zstdmt_compress.h +0 -192
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) 2016-present, 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
@@ -15,10 +15,10 @@
15
15
 
16
16
  size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
17
17
  {
18
- BYTE* const ostart = (BYTE* const)dst;
18
+ BYTE* const ostart = (BYTE*)dst;
19
19
  U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
20
20
 
21
- RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall);
21
+ RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
22
22
 
23
23
  switch(flSize)
24
24
  {
@@ -35,13 +35,14 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
35
35
  assert(0);
36
36
  }
37
37
 
38
- memcpy(ostart + flSize, src, srcSize);
38
+ ZSTD_memcpy(ostart + flSize, src, srcSize);
39
+ DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
39
40
  return srcSize + flSize;
40
41
  }
41
42
 
42
43
  size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
43
44
  {
44
- BYTE* const ostart = (BYTE* const)dst;
45
+ BYTE* const ostart = (BYTE*)dst;
45
46
  U32 const flSize = 1 + (srcSize>31) + (srcSize>4095);
46
47
 
47
48
  (void)dstCapacity; /* dstCapacity already guaranteed to be >=4, hence large enough */
@@ -62,6 +63,7 @@ size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void*
62
63
  }
63
64
 
64
65
  ostart[flSize] = *(const BYTE*)src;
66
+ DEBUGLOG(5, "RLE literals: %u -> %u", (U32)srcSize, (U32)flSize + 1);
65
67
  return flSize+1;
66
68
  }
67
69
 
@@ -70,7 +72,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
70
72
  ZSTD_strategy strategy, int disableLiteralCompression,
71
73
  void* dst, size_t dstCapacity,
72
74
  const void* src, size_t srcSize,
73
- void* workspace, size_t wkspSize,
75
+ void* entropyWorkspace, size_t entropyWorkspaceSize,
74
76
  const int bmi2)
75
77
  {
76
78
  size_t const minGain = ZSTD_minGain(srcSize, strategy);
@@ -80,11 +82,11 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
80
82
  symbolEncodingType_e hType = set_compressed;
81
83
  size_t cLitSize;
82
84
 
83
- DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i)",
84
- disableLiteralCompression);
85
+ DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i srcSize=%u)",
86
+ disableLiteralCompression, (U32)srcSize);
85
87
 
86
88
  /* Prepare nextEntropy assuming reusing the existing table */
87
- memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
89
+ ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
88
90
 
89
91
  if (disableLiteralCompression)
90
92
  return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
@@ -99,22 +101,28 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
99
101
  { HUF_repeat repeat = prevHuf->repeatMode;
100
102
  int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
101
103
  if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
102
- cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
103
- workspace, wkspSize, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2)
104
- : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
105
- workspace, wkspSize, (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
104
+ cLitSize = singleStream ?
105
+ HUF_compress1X_repeat(
106
+ ostart+lhSize, dstCapacity-lhSize, src, srcSize,
107
+ HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
108
+ (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2) :
109
+ HUF_compress4X_repeat(
110
+ ostart+lhSize, dstCapacity-lhSize, src, srcSize,
111
+ HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
112
+ (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
106
113
  if (repeat != HUF_repeat_none) {
107
114
  /* reused the existing table */
115
+ DEBUGLOG(5, "Reusing previous huffman table");
108
116
  hType = set_repeat;
109
117
  }
110
118
  }
111
119
 
112
- if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
113
- memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
120
+ if ((cLitSize==0) || (cLitSize >= srcSize - minGain) || ERR_isError(cLitSize)) {
121
+ ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
114
122
  return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
115
123
  }
116
124
  if (cLitSize==1) {
117
- memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
125
+ ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
118
126
  return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
119
127
  }
120
128
 
@@ -145,5 +153,6 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
145
153
  default: /* not possible : lhSize is {3,4,5} */
146
154
  assert(0);
147
155
  }
156
+ DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)srcSize, (U32)(lhSize+cLitSize));
148
157
  return lhSize+cLitSize;
149
158
  }
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) 2016-present, 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
@@ -23,7 +23,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
23
23
  ZSTD_strategy strategy, int disableLiteralCompression,
24
24
  void* dst, size_t dstCapacity,
25
25
  const void* src, size_t srcSize,
26
- void* workspace, size_t wkspSize,
26
+ void* entropyWorkspace, size_t entropyWorkspaceSize,
27
27
  const int bmi2);
28
28
 
29
29
  #endif /* ZSTD_COMPRESS_LITERALS_H */
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) 2016-present, 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
@@ -50,6 +50,19 @@ static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
50
50
  return maxSymbolValue;
51
51
  }
52
52
 
53
+ /**
54
+ * Returns true if we should use ncount=-1 else we should
55
+ * use ncount=1 for low probability symbols instead.
56
+ */
57
+ static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
58
+ {
59
+ /* Heuristic: This should cover most blocks <= 16K and
60
+ * start to fade out after 16K to about 32K depending on
61
+ * comprssibility.
62
+ */
63
+ return nbSeq >= 2048;
64
+ }
65
+
53
66
  /**
54
67
  * Returns the cost in bytes of encoding the normalized count header.
55
68
  * Returns an error if any of the helper functions return an error.
@@ -60,7 +73,7 @@ static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
60
73
  BYTE wksp[FSE_NCOUNTBOUND];
61
74
  S16 norm[MaxSeq + 1];
62
75
  const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
63
- FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max));
76
+ FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
64
77
  return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
65
78
  }
66
79
 
@@ -72,6 +85,8 @@ static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t
72
85
  {
73
86
  unsigned cost = 0;
74
87
  unsigned s;
88
+
89
+ assert(total > 0);
75
90
  for (s = 0; s <= max; ++s) {
76
91
  unsigned norm = (unsigned)((256 * count[s]) / total);
77
92
  if (count[s] != 0 && norm == 0)
@@ -86,7 +101,7 @@ static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t
86
101
  * Returns the cost in bits of encoding the distribution in count using ctable.
87
102
  * Returns an error if ctable cannot represent all the symbols in count.
88
103
  */
89
- static size_t ZSTD_fseBitCost(
104
+ size_t ZSTD_fseBitCost(
90
105
  FSE_CTable const* ctable,
91
106
  unsigned const* count,
92
107
  unsigned const max)
@@ -96,18 +111,22 @@ static size_t ZSTD_fseBitCost(
96
111
  unsigned s;
97
112
  FSE_CState_t cstate;
98
113
  FSE_initCState(&cstate, ctable);
99
- RETURN_ERROR_IF(ZSTD_getFSEMaxSymbolValue(ctable) < max, GENERIC,
100
- "Repeat FSE_CTable has maxSymbolValue %u < %u",
114
+ if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
115
+ DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
101
116
  ZSTD_getFSEMaxSymbolValue(ctable), max);
117
+ return ERROR(GENERIC);
118
+ }
102
119
  for (s = 0; s <= max; ++s) {
103
120
  unsigned const tableLog = cstate.stateLog;
104
121
  unsigned const badCost = (tableLog + 1) << kAccuracyLog;
105
122
  unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
106
123
  if (count[s] == 0)
107
124
  continue;
108
- RETURN_ERROR_IF(bitCost >= badCost, GENERIC,
109
- "Repeat FSE_CTable has Prob[%u] == 0", s);
110
- cost += count[s] * bitCost;
125
+ if (bitCost >= badCost) {
126
+ DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
127
+ return ERROR(GENERIC);
128
+ }
129
+ cost += (size_t)count[s] * bitCost;
111
130
  }
112
131
  return cost >> kAccuracyLog;
113
132
  }
@@ -117,15 +136,15 @@ static size_t ZSTD_fseBitCost(
117
136
  * table described by norm. The max symbol support by norm is assumed >= max.
118
137
  * norm must be valid for every symbol with non-zero probability in count.
119
138
  */
120
- static size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
121
- unsigned const* count, unsigned const max)
139
+ size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
140
+ unsigned const* count, unsigned const max)
122
141
  {
123
142
  unsigned const shift = 8 - accuracyLog;
124
143
  size_t cost = 0;
125
144
  unsigned s;
126
145
  assert(accuracyLog <= 8);
127
146
  for (s = 0; s <= max; ++s) {
128
- unsigned const normAcc = norm[s] != -1 ? norm[s] : 1;
147
+ unsigned const normAcc = (norm[s] != -1) ? (unsigned)norm[s] : 1;
129
148
  unsigned const norm256 = normAcc << shift;
130
149
  assert(norm256 > 0);
131
150
  assert(norm256 < 256);
@@ -215,6 +234,11 @@ ZSTD_selectEncodingType(
215
234
  return set_compressed;
216
235
  }
217
236
 
237
+ typedef struct {
238
+ S16 norm[MaxSeq + 1];
239
+ U32 wksp[FSE_BUILD_CTABLE_WORKSPACE_SIZE_U32(MaxSeq, MaxFSELog)];
240
+ } ZSTD_BuildCTableWksp;
241
+
218
242
  size_t
219
243
  ZSTD_buildCTable(void* dst, size_t dstCapacity,
220
244
  FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
@@ -222,7 +246,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
222
246
  const BYTE* codeTable, size_t nbSeq,
223
247
  const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
224
248
  const FSE_CTable* prevCTable, size_t prevCTableSize,
225
- void* workspace, size_t workspaceSize)
249
+ void* entropyWorkspace, size_t entropyWorkspaceSize)
226
250
  {
227
251
  BYTE* op = (BYTE*)dst;
228
252
  const BYTE* const oend = op + dstCapacity;
@@ -230,18 +254,18 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
230
254
 
231
255
  switch (type) {
232
256
  case set_rle:
233
- FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max));
234
- RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall);
257
+ FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max), "");
258
+ RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall, "not enough space");
235
259
  *op = codeTable[0];
236
260
  return 1;
237
261
  case set_repeat:
238
- memcpy(nextCTable, prevCTable, prevCTableSize);
262
+ ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
239
263
  return 0;
240
264
  case set_basic:
241
- FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize)); /* note : could be pre-calculated */
265
+ FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), ""); /* note : could be pre-calculated */
242
266
  return 0;
243
267
  case set_compressed: {
244
- S16 norm[MaxSeq + 1];
268
+ ZSTD_BuildCTableWksp* wksp = (ZSTD_BuildCTableWksp*)entropyWorkspace;
245
269
  size_t nbSeq_1 = nbSeq;
246
270
  const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
247
271
  if (count[codeTable[nbSeq-1]] > 1) {
@@ -249,14 +273,16 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
249
273
  nbSeq_1--;
250
274
  }
251
275
  assert(nbSeq_1 > 1);
252
- FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
253
- { size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog); /* overflow protected */
254
- FORWARD_IF_ERROR(NCountSize);
255
- FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, workspace, workspaceSize));
276
+ assert(entropyWorkspaceSize >= sizeof(ZSTD_BuildCTableWksp));
277
+ (void)entropyWorkspaceSize;
278
+ FORWARD_IF_ERROR(FSE_normalizeCount(wksp->norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "");
279
+ { size_t const NCountSize = FSE_writeNCount(op, oend - op, wksp->norm, max, tableLog); /* overflow protected */
280
+ FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
281
+ FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, wksp->norm, max, tableLog, wksp->wksp, sizeof(wksp->wksp)), "");
256
282
  return NCountSize;
257
283
  }
258
284
  }
259
- default: assert(0); RETURN_ERROR(GENERIC);
285
+ default: assert(0); RETURN_ERROR(GENERIC, "impossible to reach");
260
286
  }
261
287
  }
262
288
 
@@ -290,7 +316,7 @@ ZSTD_encodeSequences_body(
290
316
  if (MEM_32bits()) BIT_flushBits(&blockStream);
291
317
  if (longOffsets) {
292
318
  U32 const ofBits = ofCodeTable[nbSeq-1];
293
- int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
319
+ unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
294
320
  if (extraBits) {
295
321
  BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
296
322
  BIT_flushBits(&blockStream);
@@ -327,7 +353,7 @@ ZSTD_encodeSequences_body(
327
353
  BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
328
354
  if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
329
355
  if (longOffsets) {
330
- int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
356
+ unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
331
357
  if (extraBits) {
332
358
  BIT_addBits(&blockStream, sequences[n].offset, extraBits);
333
359
  BIT_flushBits(&blockStream); /* (7)*/
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) 2016-present, 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
@@ -11,8 +11,8 @@
11
11
  #ifndef ZSTD_COMPRESS_SEQUENCES_H
12
12
  #define ZSTD_COMPRESS_SEQUENCES_H
13
13
 
14
- #include "fse.h" /* FSE_repeat, FSE_CTable */
15
- #include "zstd_internal.h" /* symbolEncodingType_e, ZSTD_strategy */
14
+ #include "../common/fse.h" /* FSE_repeat, FSE_CTable */
15
+ #include "../common/zstd_internal.h" /* symbolEncodingType_e, ZSTD_strategy */
16
16
 
17
17
  typedef enum {
18
18
  ZSTD_defaultDisallowed = 0,
@@ -35,7 +35,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
35
35
  const BYTE* codeTable, size_t nbSeq,
36
36
  const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
37
37
  const FSE_CTable* prevCTable, size_t prevCTableSize,
38
- void* workspace, size_t workspaceSize);
38
+ void* entropyWorkspace, size_t entropyWorkspaceSize);
39
39
 
40
40
  size_t ZSTD_encodeSequences(
41
41
  void* dst, size_t dstCapacity,
@@ -44,4 +44,11 @@ size_t ZSTD_encodeSequences(
44
44
  FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
45
45
  seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2);
46
46
 
47
+ size_t ZSTD_fseBitCost(
48
+ FSE_CTable const* ctable,
49
+ unsigned const* count,
50
+ unsigned const max);
51
+
52
+ size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
53
+ unsigned const* count, unsigned const max);
47
54
  #endif /* ZSTD_COMPRESS_SEQUENCES_H */
@@ -0,0 +1,572 @@
1
+ /*
2
+ * Copyright (c) Yann Collet, Facebook, Inc.
3
+ * All rights reserved.
4
+ *
5
+ * This source code is licensed under both the BSD-style license (found in the
6
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7
+ * in the COPYING file in the root directory of this source tree).
8
+ * You may select, at your option, one of the above-listed licenses.
9
+ */
10
+
11
+ /*-*************************************
12
+ * Dependencies
13
+ ***************************************/
14
+ #include "zstd_compress_superblock.h"
15
+
16
+ #include "../common/zstd_internal.h" /* ZSTD_getSequenceLength */
17
+ #include "hist.h" /* HIST_countFast_wksp */
18
+ #include "zstd_compress_internal.h" /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
19
+ #include "zstd_compress_sequences.h"
20
+ #include "zstd_compress_literals.h"
21
+
22
+ /** ZSTD_compressSubBlock_literal() :
23
+ * Compresses literals section for a sub-block.
24
+ * When we have to write the Huffman table we will sometimes choose a header
25
+ * size larger than necessary. This is because we have to pick the header size
26
+ * before we know the table size + compressed size, so we have a bound on the
27
+ * table size. If we guessed incorrectly, we fall back to uncompressed literals.
28
+ *
29
+ * We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
30
+ * in writing the header, otherwise it is set to 0.
31
+ *
32
+ * hufMetadata->hType has literals block type info.
33
+ * If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
34
+ * If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
35
+ * If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
36
+ * If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
37
+ * and the following sub-blocks' literals sections will be Treeless_Literals_Block.
38
+ * @return : compressed size of literals section of a sub-block
39
+ * Or 0 if it unable to compress.
40
+ * Or error code */
41
+ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
42
+ const ZSTD_hufCTablesMetadata_t* hufMetadata,
43
+ const BYTE* literals, size_t litSize,
44
+ void* dst, size_t dstSize,
45
+ const int bmi2, int writeEntropy, int* entropyWritten)
46
+ {
47
+ size_t const header = writeEntropy ? 200 : 0;
48
+ size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
49
+ BYTE* const ostart = (BYTE*)dst;
50
+ BYTE* const oend = ostart + dstSize;
51
+ BYTE* op = ostart + lhSize;
52
+ U32 const singleStream = lhSize == 3;
53
+ symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
54
+ size_t cLitSize = 0;
55
+
56
+ (void)bmi2; /* TODO bmi2... */
57
+
58
+ DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
59
+
60
+ *entropyWritten = 0;
61
+ if (litSize == 0 || hufMetadata->hType == set_basic) {
62
+ DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
63
+ return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
64
+ } else if (hufMetadata->hType == set_rle) {
65
+ DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
66
+ return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
67
+ }
68
+
69
+ assert(litSize > 0);
70
+ assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
71
+
72
+ if (writeEntropy && hufMetadata->hType == set_compressed) {
73
+ ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
74
+ op += hufMetadata->hufDesSize;
75
+ cLitSize += hufMetadata->hufDesSize;
76
+ DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
77
+ }
78
+
79
+ /* TODO bmi2 */
80
+ { const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable)
81
+ : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable);
82
+ op += cSize;
83
+ cLitSize += cSize;
84
+ if (cSize == 0 || ERR_isError(cSize)) {
85
+ DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
86
+ return 0;
87
+ }
88
+ /* If we expand and we aren't writing a header then emit uncompressed */
89
+ if (!writeEntropy && cLitSize >= litSize) {
90
+ DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
91
+ return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
92
+ }
93
+ /* If we are writing headers then allow expansion that doesn't change our header size. */
94
+ if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
95
+ assert(cLitSize > litSize);
96
+ DEBUGLOG(5, "Literals expanded beyond allowed header size");
97
+ return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
98
+ }
99
+ DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
100
+ }
101
+
102
+ /* Build header */
103
+ switch(lhSize)
104
+ {
105
+ case 3: /* 2 - 2 - 10 - 10 */
106
+ { U32 const lhc = hType + ((!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
107
+ MEM_writeLE24(ostart, lhc);
108
+ break;
109
+ }
110
+ case 4: /* 2 - 2 - 14 - 14 */
111
+ { U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
112
+ MEM_writeLE32(ostart, lhc);
113
+ break;
114
+ }
115
+ case 5: /* 2 - 2 - 18 - 18 */
116
+ { U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
117
+ MEM_writeLE32(ostart, lhc);
118
+ ostart[4] = (BYTE)(cLitSize >> 10);
119
+ break;
120
+ }
121
+ default: /* not possible : lhSize is {3,4,5} */
122
+ assert(0);
123
+ }
124
+ *entropyWritten = 1;
125
+ DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
126
+ return op-ostart;
127
+ }
128
+
129
+ static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef* sequences, size_t nbSeq, size_t litSize, int lastSequence) {
130
+ const seqDef* const sstart = sequences;
131
+ const seqDef* const send = sequences + nbSeq;
132
+ const seqDef* sp = sstart;
133
+ size_t matchLengthSum = 0;
134
+ size_t litLengthSum = 0;
135
+ while (send-sp > 0) {
136
+ ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
137
+ litLengthSum += seqLen.litLength;
138
+ matchLengthSum += seqLen.matchLength;
139
+ sp++;
140
+ }
141
+ assert(litLengthSum <= litSize);
142
+ if (!lastSequence) {
143
+ assert(litLengthSum == litSize);
144
+ }
145
+ return matchLengthSum + litSize;
146
+ }
147
+
148
+ /** ZSTD_compressSubBlock_sequences() :
149
+ * Compresses sequences section for a sub-block.
150
+ * fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
151
+ * symbol compression modes for the super-block.
152
+ * The first successfully compressed block will have these in its header.
153
+ * We set entropyWritten=1 when we succeed in compressing the sequences.
154
+ * The following sub-blocks will always have repeat mode.
155
+ * @return : compressed size of sequences section of a sub-block
156
+ * Or 0 if it is unable to compress
157
+ * Or error code. */
158
+ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
159
+ const ZSTD_fseCTablesMetadata_t* fseMetadata,
160
+ const seqDef* sequences, size_t nbSeq,
161
+ const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
162
+ const ZSTD_CCtx_params* cctxParams,
163
+ void* dst, size_t dstCapacity,
164
+ const int bmi2, int writeEntropy, int* entropyWritten)
165
+ {
166
+ const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
167
+ BYTE* const ostart = (BYTE*)dst;
168
+ BYTE* const oend = ostart + dstCapacity;
169
+ BYTE* op = ostart;
170
+ BYTE* seqHead;
171
+
172
+ DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
173
+
174
+ *entropyWritten = 0;
175
+ /* Sequences Header */
176
+ RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
177
+ dstSize_tooSmall, "");
178
+ if (nbSeq < 0x7F)
179
+ *op++ = (BYTE)nbSeq;
180
+ else if (nbSeq < LONGNBSEQ)
181
+ op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
182
+ else
183
+ op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
184
+ if (nbSeq==0) {
185
+ return op - ostart;
186
+ }
187
+
188
+ /* seqHead : flags for FSE encoding type */
189
+ seqHead = op++;
190
+
191
+ DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
192
+
193
+ if (writeEntropy) {
194
+ const U32 LLtype = fseMetadata->llType;
195
+ const U32 Offtype = fseMetadata->ofType;
196
+ const U32 MLtype = fseMetadata->mlType;
197
+ DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
198
+ *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
199
+ ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
200
+ op += fseMetadata->fseTablesSize;
201
+ } else {
202
+ const U32 repeat = set_repeat;
203
+ *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
204
+ }
205
+
206
+ { size_t const bitstreamSize = ZSTD_encodeSequences(
207
+ op, oend - op,
208
+ fseTables->matchlengthCTable, mlCode,
209
+ fseTables->offcodeCTable, ofCode,
210
+ fseTables->litlengthCTable, llCode,
211
+ sequences, nbSeq,
212
+ longOffsets, bmi2);
213
+ FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
214
+ op += bitstreamSize;
215
+ /* zstd versions <= 1.3.4 mistakenly report corruption when
216
+ * FSE_readNCount() receives a buffer < 4 bytes.
217
+ * Fixed by https://github.com/facebook/zstd/pull/1146.
218
+ * This can happen when the last set_compressed table present is 2
219
+ * bytes and the bitstream is only one byte.
220
+ * In this exceedingly rare case, we will simply emit an uncompressed
221
+ * block, since it isn't worth optimizing.
222
+ */
223
+ #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
224
+ if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
225
+ /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
226
+ assert(fseMetadata->lastCountSize + bitstreamSize == 3);
227
+ DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
228
+ "emitting an uncompressed block.");
229
+ return 0;
230
+ }
231
+ #endif
232
+ DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
233
+ }
234
+
235
+ /* zstd versions <= 1.4.0 mistakenly report error when
236
+ * sequences section body size is less than 3 bytes.
237
+ * Fixed by https://github.com/facebook/zstd/pull/1664.
238
+ * This can happen when the previous sequences section block is compressed
239
+ * with rle mode and the current block's sequences section is compressed
240
+ * with repeat mode where sequences section body size can be 1 byte.
241
+ */
242
+ #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
243
+ if (op-seqHead < 4) {
244
+ DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
245
+ "an uncompressed block when sequences are < 4 bytes");
246
+ return 0;
247
+ }
248
+ #endif
249
+
250
+ *entropyWritten = 1;
251
+ return op - ostart;
252
+ }
253
+
254
+ /** ZSTD_compressSubBlock() :
255
+ * Compresses a single sub-block.
256
+ * @return : compressed size of the sub-block
257
+ * Or 0 if it failed to compress. */
258
+ static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
259
+ const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
260
+ const seqDef* sequences, size_t nbSeq,
261
+ const BYTE* literals, size_t litSize,
262
+ const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
263
+ const ZSTD_CCtx_params* cctxParams,
264
+ void* dst, size_t dstCapacity,
265
+ const int bmi2,
266
+ int writeLitEntropy, int writeSeqEntropy,
267
+ int* litEntropyWritten, int* seqEntropyWritten,
268
+ U32 lastBlock)
269
+ {
270
+ BYTE* const ostart = (BYTE*)dst;
271
+ BYTE* const oend = ostart + dstCapacity;
272
+ BYTE* op = ostart + ZSTD_blockHeaderSize;
273
+ DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
274
+ litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
275
+ { size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
276
+ &entropyMetadata->hufMetadata, literals, litSize,
277
+ op, oend-op, bmi2, writeLitEntropy, litEntropyWritten);
278
+ FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
279
+ if (cLitSize == 0) return 0;
280
+ op += cLitSize;
281
+ }
282
+ { size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
283
+ &entropyMetadata->fseMetadata,
284
+ sequences, nbSeq,
285
+ llCode, mlCode, ofCode,
286
+ cctxParams,
287
+ op, oend-op,
288
+ bmi2, writeSeqEntropy, seqEntropyWritten);
289
+ FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
290
+ if (cSeqSize == 0) return 0;
291
+ op += cSeqSize;
292
+ }
293
+ /* Write block header */
294
+ { size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
295
+ U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
296
+ MEM_writeLE24(ostart, cBlockHeader24);
297
+ }
298
+ return op-ostart;
299
+ }
300
+
301
+ static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
302
+ const ZSTD_hufCTables_t* huf,
303
+ const ZSTD_hufCTablesMetadata_t* hufMetadata,
304
+ void* workspace, size_t wkspSize,
305
+ int writeEntropy)
306
+ {
307
+ unsigned* const countWksp = (unsigned*)workspace;
308
+ unsigned maxSymbolValue = 255;
309
+ size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
310
+
311
+ if (hufMetadata->hType == set_basic) return litSize;
312
+ else if (hufMetadata->hType == set_rle) return 1;
313
+ else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
314
+ size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
315
+ if (ZSTD_isError(largest)) return litSize;
316
+ { size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
317
+ if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
318
+ return cLitSizeEstimate + literalSectionHeaderSize;
319
+ } }
320
+ assert(0); /* impossible */
321
+ return 0;
322
+ }
323
+
324
+ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
325
+ const BYTE* codeTable, unsigned maxCode,
326
+ size_t nbSeq, const FSE_CTable* fseCTable,
327
+ const U32* additionalBits,
328
+ short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
329
+ void* workspace, size_t wkspSize)
330
+ {
331
+ unsigned* const countWksp = (unsigned*)workspace;
332
+ const BYTE* ctp = codeTable;
333
+ const BYTE* const ctStart = ctp;
334
+ const BYTE* const ctEnd = ctStart + nbSeq;
335
+ size_t cSymbolTypeSizeEstimateInBits = 0;
336
+ unsigned max = maxCode;
337
+
338
+ HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize); /* can't fail */
339
+ if (type == set_basic) {
340
+ /* We selected this encoding type, so it must be valid. */
341
+ assert(max <= defaultMax);
342
+ cSymbolTypeSizeEstimateInBits = max <= defaultMax
343
+ ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
344
+ : ERROR(GENERIC);
345
+ } else if (type == set_rle) {
346
+ cSymbolTypeSizeEstimateInBits = 0;
347
+ } else if (type == set_compressed || type == set_repeat) {
348
+ cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
349
+ }
350
+ if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
351
+ while (ctp < ctEnd) {
352
+ if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
353
+ else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
354
+ ctp++;
355
+ }
356
+ return cSymbolTypeSizeEstimateInBits / 8;
357
+ }
358
+
359
+ static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
360
+ const BYTE* llCodeTable,
361
+ const BYTE* mlCodeTable,
362
+ size_t nbSeq,
363
+ const ZSTD_fseCTables_t* fseTables,
364
+ const ZSTD_fseCTablesMetadata_t* fseMetadata,
365
+ void* workspace, size_t wkspSize,
366
+ int writeEntropy)
367
+ {
368
+ size_t const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
369
+ size_t cSeqSizeEstimate = 0;
370
+ if (nbSeq == 0) return sequencesSectionHeaderSize;
371
+ cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
372
+ nbSeq, fseTables->offcodeCTable, NULL,
373
+ OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
374
+ workspace, wkspSize);
375
+ cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
376
+ nbSeq, fseTables->litlengthCTable, LL_bits,
377
+ LL_defaultNorm, LL_defaultNormLog, MaxLL,
378
+ workspace, wkspSize);
379
+ cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
380
+ nbSeq, fseTables->matchlengthCTable, ML_bits,
381
+ ML_defaultNorm, ML_defaultNormLog, MaxML,
382
+ workspace, wkspSize);
383
+ if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
384
+ return cSeqSizeEstimate + sequencesSectionHeaderSize;
385
+ }
386
+
387
+ static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
388
+ const BYTE* ofCodeTable,
389
+ const BYTE* llCodeTable,
390
+ const BYTE* mlCodeTable,
391
+ size_t nbSeq,
392
+ const ZSTD_entropyCTables_t* entropy,
393
+ const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
394
+ void* workspace, size_t wkspSize,
395
+ int writeLitEntropy, int writeSeqEntropy) {
396
+ size_t cSizeEstimate = 0;
397
+ cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
398
+ &entropy->huf, &entropyMetadata->hufMetadata,
399
+ workspace, wkspSize, writeLitEntropy);
400
+ cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
401
+ nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
402
+ workspace, wkspSize, writeSeqEntropy);
403
+ return cSizeEstimate + ZSTD_blockHeaderSize;
404
+ }
405
+
406
+ static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
407
+ {
408
+ if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
409
+ return 1;
410
+ if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
411
+ return 1;
412
+ if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
413
+ return 1;
414
+ return 0;
415
+ }
416
+
417
+ /** ZSTD_compressSubBlock_multi() :
418
+ * Breaks super-block into multiple sub-blocks and compresses them.
419
+ * Entropy will be written to the first block.
420
+ * The following blocks will use repeat mode to compress.
421
+ * All sub-blocks are compressed blocks (no raw or rle blocks).
422
+ * @return : compressed size of the super block (which is multiple ZSTD blocks)
423
+ * Or 0 if it failed to compress. */
424
+ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
425
+ const ZSTD_compressedBlockState_t* prevCBlock,
426
+ ZSTD_compressedBlockState_t* nextCBlock,
427
+ const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
428
+ const ZSTD_CCtx_params* cctxParams,
429
+ void* dst, size_t dstCapacity,
430
+ const void* src, size_t srcSize,
431
+ const int bmi2, U32 lastBlock,
432
+ void* workspace, size_t wkspSize)
433
+ {
434
+ const seqDef* const sstart = seqStorePtr->sequencesStart;
435
+ const seqDef* const send = seqStorePtr->sequences;
436
+ const seqDef* sp = sstart;
437
+ const BYTE* const lstart = seqStorePtr->litStart;
438
+ const BYTE* const lend = seqStorePtr->lit;
439
+ const BYTE* lp = lstart;
440
+ BYTE const* ip = (BYTE const*)src;
441
+ BYTE const* const iend = ip + srcSize;
442
+ BYTE* const ostart = (BYTE*)dst;
443
+ BYTE* const oend = ostart + dstCapacity;
444
+ BYTE* op = ostart;
445
+ const BYTE* llCodePtr = seqStorePtr->llCode;
446
+ const BYTE* mlCodePtr = seqStorePtr->mlCode;
447
+ const BYTE* ofCodePtr = seqStorePtr->ofCode;
448
+ size_t targetCBlockSize = cctxParams->targetCBlockSize;
449
+ size_t litSize, seqCount;
450
+ int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
451
+ int writeSeqEntropy = 1;
452
+ int lastSequence = 0;
453
+
454
+ DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
455
+ (unsigned)(lend-lp), (unsigned)(send-sstart));
456
+
457
+ litSize = 0;
458
+ seqCount = 0;
459
+ do {
460
+ size_t cBlockSizeEstimate = 0;
461
+ if (sstart == send) {
462
+ lastSequence = 1;
463
+ } else {
464
+ const seqDef* const sequence = sp + seqCount;
465
+ lastSequence = sequence == send - 1;
466
+ litSize += ZSTD_getSequenceLength(seqStorePtr, sequence).litLength;
467
+ seqCount++;
468
+ }
469
+ if (lastSequence) {
470
+ assert(lp <= lend);
471
+ assert(litSize <= (size_t)(lend - lp));
472
+ litSize = (size_t)(lend - lp);
473
+ }
474
+ /* I think there is an optimization opportunity here.
475
+ * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
476
+ * since it recalculates estimate from scratch.
477
+ * For example, it would recount literal distribution and symbol codes everytime.
478
+ */
479
+ cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
480
+ &nextCBlock->entropy, entropyMetadata,
481
+ workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
482
+ if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
483
+ int litEntropyWritten = 0;
484
+ int seqEntropyWritten = 0;
485
+ const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, lastSequence);
486
+ const size_t cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
487
+ sp, seqCount,
488
+ lp, litSize,
489
+ llCodePtr, mlCodePtr, ofCodePtr,
490
+ cctxParams,
491
+ op, oend-op,
492
+ bmi2, writeLitEntropy, writeSeqEntropy,
493
+ &litEntropyWritten, &seqEntropyWritten,
494
+ lastBlock && lastSequence);
495
+ FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
496
+ if (cSize > 0 && cSize < decompressedSize) {
497
+ DEBUGLOG(5, "Committed the sub-block");
498
+ assert(ip + decompressedSize <= iend);
499
+ ip += decompressedSize;
500
+ sp += seqCount;
501
+ lp += litSize;
502
+ op += cSize;
503
+ llCodePtr += seqCount;
504
+ mlCodePtr += seqCount;
505
+ ofCodePtr += seqCount;
506
+ litSize = 0;
507
+ seqCount = 0;
508
+ /* Entropy only needs to be written once */
509
+ if (litEntropyWritten) {
510
+ writeLitEntropy = 0;
511
+ }
512
+ if (seqEntropyWritten) {
513
+ writeSeqEntropy = 0;
514
+ }
515
+ }
516
+ }
517
+ } while (!lastSequence);
518
+ if (writeLitEntropy) {
519
+ DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
520
+ ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
521
+ }
522
+ if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
523
+ /* If we haven't written our entropy tables, then we've violated our contract and
524
+ * must emit an uncompressed block.
525
+ */
526
+ DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
527
+ return 0;
528
+ }
529
+ if (ip < iend) {
530
+ size_t const cSize = ZSTD_noCompressBlock(op, oend - op, ip, iend - ip, lastBlock);
531
+ DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
532
+ FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
533
+ assert(cSize != 0);
534
+ op += cSize;
535
+ /* We have to regenerate the repcodes because we've skipped some sequences */
536
+ if (sp < send) {
537
+ seqDef const* seq;
538
+ repcodes_t rep;
539
+ ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
540
+ for (seq = sstart; seq < sp; ++seq) {
541
+ rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
542
+ }
543
+ ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
544
+ }
545
+ }
546
+ DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
547
+ return op-ostart;
548
+ }
549
+
550
+ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
551
+ void* dst, size_t dstCapacity,
552
+ void const* src, size_t srcSize,
553
+ unsigned lastBlock) {
554
+ ZSTD_entropyCTablesMetadata_t entropyMetadata;
555
+
556
+ FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
557
+ &zc->blockState.prevCBlock->entropy,
558
+ &zc->blockState.nextCBlock->entropy,
559
+ &zc->appliedParams,
560
+ &entropyMetadata,
561
+ zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
562
+
563
+ return ZSTD_compressSubBlock_multi(&zc->seqStore,
564
+ zc->blockState.prevCBlock,
565
+ zc->blockState.nextCBlock,
566
+ &entropyMetadata,
567
+ &zc->appliedParams,
568
+ dst, dstCapacity,
569
+ src, srcSize,
570
+ zc->bmi2, lastBlock,
571
+ zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
572
+ }