extzstd 0.3.1 → 0.3.3

Sign up to get free protection for your applications and to get access to all the features.
Files changed (113) hide show
  1. checksums.yaml +4 -4
  2. data/README.md +28 -14
  3. data/contrib/zstd/CHANGELOG +301 -56
  4. data/contrib/zstd/CONTRIBUTING.md +169 -72
  5. data/contrib/zstd/LICENSE +4 -4
  6. data/contrib/zstd/Makefile +116 -87
  7. data/contrib/zstd/Package.swift +36 -0
  8. data/contrib/zstd/README.md +62 -32
  9. data/contrib/zstd/TESTING.md +2 -3
  10. data/contrib/zstd/appveyor.yml +52 -136
  11. data/contrib/zstd/lib/BUCK +5 -7
  12. data/contrib/zstd/lib/Makefile +225 -222
  13. data/contrib/zstd/lib/README.md +51 -6
  14. data/contrib/zstd/lib/common/allocations.h +55 -0
  15. data/contrib/zstd/lib/common/bits.h +200 -0
  16. data/contrib/zstd/lib/common/bitstream.h +45 -62
  17. data/contrib/zstd/lib/common/compiler.h +205 -22
  18. data/contrib/zstd/lib/common/cpu.h +1 -3
  19. data/contrib/zstd/lib/common/debug.c +1 -1
  20. data/contrib/zstd/lib/common/debug.h +12 -19
  21. data/contrib/zstd/lib/common/entropy_common.c +172 -48
  22. data/contrib/zstd/lib/common/error_private.c +10 -2
  23. data/contrib/zstd/lib/common/error_private.h +82 -3
  24. data/contrib/zstd/lib/common/fse.h +37 -86
  25. data/contrib/zstd/lib/common/fse_decompress.c +117 -92
  26. data/contrib/zstd/lib/common/huf.h +99 -166
  27. data/contrib/zstd/lib/common/mem.h +124 -142
  28. data/contrib/zstd/lib/common/pool.c +54 -27
  29. data/contrib/zstd/lib/common/pool.h +10 -4
  30. data/contrib/zstd/lib/common/portability_macros.h +156 -0
  31. data/contrib/zstd/lib/common/threading.c +74 -19
  32. data/contrib/zstd/lib/common/threading.h +5 -10
  33. data/contrib/zstd/lib/common/xxhash.c +7 -847
  34. data/contrib/zstd/lib/common/xxhash.h +5568 -167
  35. data/contrib/zstd/lib/common/zstd_common.c +2 -37
  36. data/contrib/zstd/lib/common/zstd_deps.h +111 -0
  37. data/contrib/zstd/lib/common/zstd_internal.h +132 -187
  38. data/contrib/zstd/lib/common/zstd_trace.h +163 -0
  39. data/contrib/zstd/lib/compress/clevels.h +134 -0
  40. data/contrib/zstd/lib/compress/fse_compress.c +83 -157
  41. data/contrib/zstd/lib/compress/hist.c +27 -29
  42. data/contrib/zstd/lib/compress/hist.h +2 -2
  43. data/contrib/zstd/lib/compress/huf_compress.c +916 -279
  44. data/contrib/zstd/lib/compress/zstd_compress.c +3773 -1019
  45. data/contrib/zstd/lib/compress/zstd_compress_internal.h +610 -203
  46. data/contrib/zstd/lib/compress/zstd_compress_literals.c +119 -42
  47. data/contrib/zstd/lib/compress/zstd_compress_literals.h +16 -6
  48. data/contrib/zstd/lib/compress/zstd_compress_sequences.c +42 -19
  49. data/contrib/zstd/lib/compress/zstd_compress_sequences.h +1 -1
  50. data/contrib/zstd/lib/compress/zstd_compress_superblock.c +49 -317
  51. data/contrib/zstd/lib/compress/zstd_compress_superblock.h +1 -1
  52. data/contrib/zstd/lib/compress/zstd_cwksp.h +320 -103
  53. data/contrib/zstd/lib/compress/zstd_double_fast.c +388 -151
  54. data/contrib/zstd/lib/compress/zstd_double_fast.h +3 -2
  55. data/contrib/zstd/lib/compress/zstd_fast.c +729 -265
  56. data/contrib/zstd/lib/compress/zstd_fast.h +3 -2
  57. data/contrib/zstd/lib/compress/zstd_lazy.c +1270 -251
  58. data/contrib/zstd/lib/compress/zstd_lazy.h +61 -1
  59. data/contrib/zstd/lib/compress/zstd_ldm.c +324 -219
  60. data/contrib/zstd/lib/compress/zstd_ldm.h +9 -2
  61. data/contrib/zstd/lib/compress/zstd_ldm_geartab.h +106 -0
  62. data/contrib/zstd/lib/compress/zstd_opt.c +481 -209
  63. data/contrib/zstd/lib/compress/zstd_opt.h +1 -1
  64. data/contrib/zstd/lib/compress/zstdmt_compress.c +181 -457
  65. data/contrib/zstd/lib/compress/zstdmt_compress.h +34 -113
  66. data/contrib/zstd/lib/decompress/huf_decompress.c +1199 -565
  67. data/contrib/zstd/lib/decompress/huf_decompress_amd64.S +576 -0
  68. data/contrib/zstd/lib/decompress/zstd_ddict.c +12 -12
  69. data/contrib/zstd/lib/decompress/zstd_ddict.h +2 -2
  70. data/contrib/zstd/lib/decompress/zstd_decompress.c +627 -157
  71. data/contrib/zstd/lib/decompress/zstd_decompress_block.c +1086 -326
  72. data/contrib/zstd/lib/decompress/zstd_decompress_block.h +19 -5
  73. data/contrib/zstd/lib/decompress/zstd_decompress_internal.h +62 -13
  74. data/contrib/zstd/lib/deprecated/zbuff.h +1 -1
  75. data/contrib/zstd/lib/deprecated/zbuff_common.c +1 -1
  76. data/contrib/zstd/lib/deprecated/zbuff_compress.c +24 -4
  77. data/contrib/zstd/lib/deprecated/zbuff_decompress.c +3 -1
  78. data/contrib/zstd/lib/dictBuilder/cover.c +73 -52
  79. data/contrib/zstd/lib/dictBuilder/cover.h +7 -6
  80. data/contrib/zstd/lib/dictBuilder/divsufsort.c +1 -1
  81. data/contrib/zstd/lib/dictBuilder/fastcover.c +44 -35
  82. data/contrib/zstd/lib/dictBuilder/zdict.c +103 -111
  83. data/contrib/zstd/lib/legacy/zstd_legacy.h +8 -1
  84. data/contrib/zstd/lib/legacy/zstd_v01.c +21 -54
  85. data/contrib/zstd/lib/legacy/zstd_v01.h +1 -1
  86. data/contrib/zstd/lib/legacy/zstd_v02.c +29 -70
  87. data/contrib/zstd/lib/legacy/zstd_v02.h +1 -1
  88. data/contrib/zstd/lib/legacy/zstd_v03.c +30 -73
  89. data/contrib/zstd/lib/legacy/zstd_v03.h +1 -1
  90. data/contrib/zstd/lib/legacy/zstd_v04.c +29 -71
  91. data/contrib/zstd/lib/legacy/zstd_v04.h +1 -1
  92. data/contrib/zstd/lib/legacy/zstd_v05.c +40 -86
  93. data/contrib/zstd/lib/legacy/zstd_v05.h +1 -1
  94. data/contrib/zstd/lib/legacy/zstd_v06.c +47 -88
  95. data/contrib/zstd/lib/legacy/zstd_v06.h +1 -1
  96. data/contrib/zstd/lib/legacy/zstd_v07.c +40 -83
  97. data/contrib/zstd/lib/legacy/zstd_v07.h +1 -1
  98. data/contrib/zstd/lib/libzstd.mk +214 -0
  99. data/contrib/zstd/lib/libzstd.pc.in +7 -6
  100. data/contrib/zstd/lib/module.modulemap +35 -0
  101. data/contrib/zstd/lib/{dictBuilder/zdict.h → zdict.h} +203 -34
  102. data/contrib/zstd/lib/zstd.h +1217 -287
  103. data/contrib/zstd/lib/{common/zstd_errors.h → zstd_errors.h} +28 -8
  104. data/ext/extconf.rb +7 -6
  105. data/ext/extzstd.c +19 -10
  106. data/ext/extzstd.h +6 -0
  107. data/ext/libzstd_conf.h +0 -1
  108. data/ext/zstd_decompress_asm.S +1 -0
  109. data/gemstub.rb +3 -21
  110. data/lib/extzstd/version.rb +6 -1
  111. data/lib/extzstd.rb +0 -2
  112. data/test/test_basic.rb +0 -5
  113. metadata +18 -6
@@ -1,5 +1,5 @@
1
1
  /*
2
- * Copyright (c) 2016-2020, 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
@@ -20,8 +20,8 @@
20
20
 
21
21
 
22
22
  /* ====== Dependencies ====== */
23
- #include <string.h> /* memcpy, memset */
24
- #include <limits.h> /* INT_MAX, UINT_MAX */
23
+ #include "../common/allocations.h" /* ZSTD_customMalloc, ZSTD_customCalloc, ZSTD_customFree */
24
+ #include "../common/zstd_deps.h" /* ZSTD_memcpy, ZSTD_memset, INT_MAX, UINT_MAX */
25
25
  #include "../common/mem.h" /* MEM_STATIC */
26
26
  #include "../common/pool.h" /* threadpool */
27
27
  #include "../common/threading.h" /* mutex */
@@ -103,14 +103,13 @@ typedef struct ZSTDMT_bufferPool_s {
103
103
  buffer_t bTable[1]; /* variable size */
104
104
  } ZSTDMT_bufferPool;
105
105
 
106
- static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbWorkers, ZSTD_customMem cMem)
106
+ static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned maxNbBuffers, ZSTD_customMem cMem)
107
107
  {
108
- unsigned const maxNbBuffers = 2*nbWorkers + 3;
109
- ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
108
+ ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_customCalloc(
110
109
  sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
111
110
  if (bufPool==NULL) return NULL;
112
111
  if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
113
- ZSTD_free(bufPool, cMem);
112
+ ZSTD_customFree(bufPool, cMem);
114
113
  return NULL;
115
114
  }
116
115
  bufPool->bufferSize = 64 KB;
@@ -127,10 +126,10 @@ static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
127
126
  if (!bufPool) return; /* compatibility with free on NULL */
128
127
  for (u=0; u<bufPool->totalBuffers; u++) {
129
128
  DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->bTable[u].start);
130
- ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
129
+ ZSTD_customFree(bufPool->bTable[u].start, bufPool->cMem);
131
130
  }
132
131
  ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
133
- ZSTD_free(bufPool, bufPool->cMem);
132
+ ZSTD_customFree(bufPool, bufPool->cMem);
134
133
  }
135
134
 
136
135
  /* only works at initialization, not during compression */
@@ -161,9 +160,8 @@ static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* const bufPool, size_t const
161
160
  }
162
161
 
163
162
 
164
- static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, U32 nbWorkers)
163
+ static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool, unsigned maxNbBuffers)
165
164
  {
166
- unsigned const maxNbBuffers = 2*nbWorkers + 3;
167
165
  if (srcBufPool==NULL) return NULL;
168
166
  if (srcBufPool->totalBuffers >= maxNbBuffers) /* good enough */
169
167
  return srcBufPool;
@@ -172,7 +170,7 @@ static ZSTDMT_bufferPool* ZSTDMT_expandBufferPool(ZSTDMT_bufferPool* srcBufPool,
172
170
  size_t const bSize = srcBufPool->bufferSize; /* forward parameters */
173
171
  ZSTDMT_bufferPool* newBufPool;
174
172
  ZSTDMT_freeBufferPool(srcBufPool);
175
- newBufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
173
+ newBufPool = ZSTDMT_createBufferPool(maxNbBuffers, cMem);
176
174
  if (newBufPool==NULL) return newBufPool;
177
175
  ZSTDMT_setBufferSize(newBufPool, bSize);
178
176
  return newBufPool;
@@ -201,13 +199,13 @@ static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
201
199
  }
202
200
  /* size conditions not respected : scratch this buffer, create new one */
203
201
  DEBUGLOG(5, "ZSTDMT_getBuffer: existing buffer does not meet size conditions => freeing");
204
- ZSTD_free(buf.start, bufPool->cMem);
202
+ ZSTD_customFree(buf.start, bufPool->cMem);
205
203
  }
206
204
  ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
207
205
  /* create new buffer */
208
206
  DEBUGLOG(5, "ZSTDMT_getBuffer: create a new buffer");
209
207
  { buffer_t buffer;
210
- void* const start = ZSTD_malloc(bSize, bufPool->cMem);
208
+ void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
211
209
  buffer.start = start; /* note : start can be NULL if malloc fails ! */
212
210
  buffer.capacity = (start==NULL) ? 0 : bSize;
213
211
  if (start==NULL) {
@@ -229,13 +227,13 @@ static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
229
227
  {
230
228
  size_t const bSize = bufPool->bufferSize;
231
229
  if (buffer.capacity < bSize) {
232
- void* const start = ZSTD_malloc(bSize, bufPool->cMem);
230
+ void* const start = ZSTD_customMalloc(bSize, bufPool->cMem);
233
231
  buffer_t newBuffer;
234
232
  newBuffer.start = start;
235
233
  newBuffer.capacity = start == NULL ? 0 : bSize;
236
234
  if (start != NULL) {
237
235
  assert(newBuffer.capacity >= buffer.capacity);
238
- memcpy(newBuffer.start, buffer.start, buffer.capacity);
236
+ ZSTD_memcpy(newBuffer.start, buffer.start, buffer.capacity);
239
237
  DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
240
238
  return newBuffer;
241
239
  }
@@ -261,13 +259,21 @@ static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
261
259
  ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
262
260
  /* Reached bufferPool capacity (should not happen) */
263
261
  DEBUGLOG(5, "ZSTDMT_releaseBuffer: pool capacity reached => freeing ");
264
- ZSTD_free(buf.start, bufPool->cMem);
262
+ ZSTD_customFree(buf.start, bufPool->cMem);
265
263
  }
266
264
 
265
+ /* We need 2 output buffers per worker since each dstBuff must be flushed after it is released.
266
+ * The 3 additional buffers are as follows:
267
+ * 1 buffer for input loading
268
+ * 1 buffer for "next input" when submitting current one
269
+ * 1 buffer stuck in queue */
270
+ #define BUF_POOL_MAX_NB_BUFFERS(nbWorkers) (2*(nbWorkers) + 3)
267
271
 
268
- /* ===== Seq Pool Wrapper ====== */
272
+ /* After a worker releases its rawSeqStore, it is immediately ready for reuse.
273
+ * So we only need one seq buffer per worker. */
274
+ #define SEQ_POOL_MAX_NB_BUFFERS(nbWorkers) (nbWorkers)
269
275
 
270
- static rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0};
276
+ /* ===== Seq Pool Wrapper ====== */
271
277
 
272
278
  typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
273
279
 
@@ -278,7 +284,7 @@ static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
278
284
 
279
285
  static rawSeqStore_t bufferToSeq(buffer_t buffer)
280
286
  {
281
- rawSeqStore_t seq = {NULL, 0, 0, 0};
287
+ rawSeqStore_t seq = kNullRawSeqStore;
282
288
  seq.seq = (rawSeq*)buffer.start;
283
289
  seq.capacity = buffer.capacity / sizeof(rawSeq);
284
290
  return seq;
@@ -319,7 +325,7 @@ static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
319
325
 
320
326
  static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
321
327
  {
322
- ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
328
+ ZSTDMT_seqPool* const seqPool = ZSTDMT_createBufferPool(SEQ_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
323
329
  if (seqPool == NULL) return NULL;
324
330
  ZSTDMT_setNbSeq(seqPool, 0);
325
331
  return seqPool;
@@ -332,7 +338,7 @@ static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
332
338
 
333
339
  static ZSTDMT_seqPool* ZSTDMT_expandSeqPool(ZSTDMT_seqPool* pool, U32 nbWorkers)
334
340
  {
335
- return ZSTDMT_expandBufferPool(pool, nbWorkers);
341
+ return ZSTDMT_expandBufferPool(pool, SEQ_POOL_MAX_NB_BUFFERS(nbWorkers));
336
342
  }
337
343
 
338
344
 
@@ -354,7 +360,7 @@ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
354
360
  for (cid=0; cid<pool->totalCCtx; cid++)
355
361
  ZSTD_freeCCtx(pool->cctx[cid]); /* note : compatible with free on NULL */
356
362
  ZSTD_pthread_mutex_destroy(&pool->poolMutex);
357
- ZSTD_free(pool, pool->cMem);
363
+ ZSTD_customFree(pool, pool->cMem);
358
364
  }
359
365
 
360
366
  /* ZSTDMT_createCCtxPool() :
@@ -362,12 +368,12 @@ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
362
368
  static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(int nbWorkers,
363
369
  ZSTD_customMem cMem)
364
370
  {
365
- ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
371
+ ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_customCalloc(
366
372
  sizeof(ZSTDMT_CCtxPool) + (nbWorkers-1)*sizeof(ZSTD_CCtx*), cMem);
367
373
  assert(nbWorkers > 0);
368
374
  if (!cctxPool) return NULL;
369
375
  if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
370
- ZSTD_free(cctxPool, cMem);
376
+ ZSTD_customFree(cctxPool, cMem);
371
377
  return NULL;
372
378
  }
373
379
  cctxPool->cMem = cMem;
@@ -470,54 +476,52 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
470
476
  ZSTD_dictContentType_e dictContentType)
471
477
  {
472
478
  /* Adjust parameters */
473
- if (params.ldmParams.enableLdm) {
479
+ if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
474
480
  DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
475
481
  ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
476
482
  assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
477
483
  assert(params.ldmParams.hashRateLog < 32);
478
- serialState->ldmState.hashPower =
479
- ZSTD_rollingHash_primePower(params.ldmParams.minMatchLength);
480
484
  } else {
481
- memset(&params.ldmParams, 0, sizeof(params.ldmParams));
485
+ ZSTD_memset(&params.ldmParams, 0, sizeof(params.ldmParams));
482
486
  }
483
487
  serialState->nextJobID = 0;
484
488
  if (params.fParams.checksumFlag)
485
489
  XXH64_reset(&serialState->xxhState, 0);
486
- if (params.ldmParams.enableLdm) {
490
+ if (params.ldmParams.enableLdm == ZSTD_ps_enable) {
487
491
  ZSTD_customMem cMem = params.customMem;
488
492
  unsigned const hashLog = params.ldmParams.hashLog;
489
493
  size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
490
494
  unsigned const bucketLog =
491
495
  params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
492
- size_t const bucketSize = (size_t)1 << bucketLog;
493
496
  unsigned const prevBucketLog =
494
497
  serialState->params.ldmParams.hashLog -
495
498
  serialState->params.ldmParams.bucketSizeLog;
499
+ size_t const numBuckets = (size_t)1 << bucketLog;
496
500
  /* Size the seq pool tables */
497
501
  ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, jobSize));
498
502
  /* Reset the window */
499
503
  ZSTD_window_init(&serialState->ldmState.window);
500
504
  /* Resize tables and output space if necessary. */
501
505
  if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
502
- ZSTD_free(serialState->ldmState.hashTable, cMem);
503
- serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_malloc(hashSize, cMem);
506
+ ZSTD_customFree(serialState->ldmState.hashTable, cMem);
507
+ serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_customMalloc(hashSize, cMem);
504
508
  }
505
509
  if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
506
- ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
507
- serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_malloc(bucketSize, cMem);
510
+ ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
511
+ serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_customMalloc(numBuckets, cMem);
508
512
  }
509
513
  if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
510
514
  return 1;
511
515
  /* Zero the tables */
512
- memset(serialState->ldmState.hashTable, 0, hashSize);
513
- memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
516
+ ZSTD_memset(serialState->ldmState.hashTable, 0, hashSize);
517
+ ZSTD_memset(serialState->ldmState.bucketOffsets, 0, numBuckets);
514
518
 
515
519
  /* Update window state and fill hash table with dict */
516
520
  serialState->ldmState.loadedDictEnd = 0;
517
521
  if (dictSize > 0) {
518
522
  if (dictContentType == ZSTD_dct_rawContent) {
519
523
  BYTE const* const dictEnd = (const BYTE*)dict + dictSize;
520
- ZSTD_window_update(&serialState->ldmState.window, dict, dictSize);
524
+ ZSTD_window_update(&serialState->ldmState.window, dict, dictSize, /* forceNonContiguous */ 0);
521
525
  ZSTD_ldm_fillHashTable(&serialState->ldmState, (const BYTE*)dict, dictEnd, &params.ldmParams);
522
526
  serialState->ldmState.loadedDictEnd = params.forceWindow ? 0 : (U32)(dictEnd - serialState->ldmState.window.base);
523
527
  } else {
@@ -537,7 +541,7 @@ ZSTDMT_serialState_reset(serialState_t* serialState,
537
541
  static int ZSTDMT_serialState_init(serialState_t* serialState)
538
542
  {
539
543
  int initError = 0;
540
- memset(serialState, 0, sizeof(*serialState));
544
+ ZSTD_memset(serialState, 0, sizeof(*serialState));
541
545
  initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
542
546
  initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
543
547
  initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
@@ -552,8 +556,8 @@ static void ZSTDMT_serialState_free(serialState_t* serialState)
552
556
  ZSTD_pthread_cond_destroy(&serialState->cond);
553
557
  ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
554
558
  ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
555
- ZSTD_free(serialState->ldmState.hashTable, cMem);
556
- ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
559
+ ZSTD_customFree(serialState->ldmState.hashTable, cMem);
560
+ ZSTD_customFree(serialState->ldmState.bucketOffsets, cMem);
557
561
  }
558
562
 
559
563
  static void ZSTDMT_serialState_update(serialState_t* serialState,
@@ -569,12 +573,12 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
569
573
  /* A future job may error and skip our job */
570
574
  if (serialState->nextJobID == jobID) {
571
575
  /* It is now our turn, do any processing necessary */
572
- if (serialState->params.ldmParams.enableLdm) {
576
+ if (serialState->params.ldmParams.enableLdm == ZSTD_ps_enable) {
573
577
  size_t error;
574
578
  assert(seqStore.seq != NULL && seqStore.pos == 0 &&
575
579
  seqStore.size == 0 && seqStore.capacity > 0);
576
580
  assert(src.size <= serialState->params.jobSize);
577
- ZSTD_window_update(&serialState->ldmState.window, src.start, src.size);
581
+ ZSTD_window_update(&serialState->ldmState.window, src.start, src.size, /* forceNonContiguous */ 0);
578
582
  error = ZSTD_ldm_generateSequences(
579
583
  &serialState->ldmState, &seqStore,
580
584
  &serialState->params.ldmParams, src.start, src.size);
@@ -599,7 +603,7 @@ static void ZSTDMT_serialState_update(serialState_t* serialState,
599
603
  if (seqStore.size > 0) {
600
604
  size_t const err = ZSTD_referenceExternalSequences(
601
605
  jobCCtx, seqStore.seq, seqStore.size);
602
- assert(serialState->params.ldmParams.enableLdm);
606
+ assert(serialState->params.ldmParams.enableLdm == ZSTD_ps_enable);
603
607
  assert(!ZSTD_isError(err));
604
608
  (void)err;
605
609
  }
@@ -677,7 +681,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
677
681
  if (dstBuff.start==NULL) JOB_ERROR(ERROR(memory_allocation));
678
682
  job->dstBuff = dstBuff; /* this value can be read in ZSTDMT_flush, when it copies the whole job */
679
683
  }
680
- if (jobParams.ldmParams.enableLdm && rawSeqStore.seq == NULL)
684
+ if (jobParams.ldmParams.enableLdm == ZSTD_ps_enable && rawSeqStore.seq == NULL)
681
685
  JOB_ERROR(ERROR(memory_allocation));
682
686
 
683
687
  /* Don't compute the checksum for chunks, since we compute it externally,
@@ -685,7 +689,9 @@ static void ZSTDMT_compressionJob(void* jobDescription)
685
689
  */
686
690
  if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
687
691
  /* Don't run LDM for the chunks, since we handle it externally */
688
- jobParams.ldmParams.enableLdm = 0;
692
+ jobParams.ldmParams.enableLdm = ZSTD_ps_disable;
693
+ /* Correct nbWorkers to 0. */
694
+ jobParams.nbWorkers = 0;
689
695
 
690
696
 
691
697
  /* init */
@@ -698,6 +704,10 @@ static void ZSTDMT_compressionJob(void* jobDescription)
698
704
  { size_t const forceWindowError = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_forceMaxWindow, !job->firstJob);
699
705
  if (ZSTD_isError(forceWindowError)) JOB_ERROR(forceWindowError);
700
706
  }
707
+ if (!job->firstJob) {
708
+ size_t const err = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_deterministicRefPrefix, 0);
709
+ if (ZSTD_isError(err)) JOB_ERROR(err);
710
+ }
701
711
  { size_t const initError = ZSTD_compressBegin_advanced_internal(cctx,
702
712
  job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
703
713
  ZSTD_dtlm_fast,
@@ -710,7 +720,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
710
720
  ZSTDMT_serialState_update(job->serial, cctx, rawSeqStore, job->src, job->jobID);
711
721
 
712
722
  if (!job->firstJob) { /* flush and overwrite frame header when it's not first job */
713
- size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
723
+ size_t const hSize = ZSTD_compressContinue_public(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
714
724
  if (ZSTD_isError(hSize)) JOB_ERROR(hSize);
715
725
  DEBUGLOG(5, "ZSTDMT_compressionJob: flush and overwrite %u bytes of frame header (not first job)", (U32)hSize);
716
726
  ZSTD_invalidateRepCodes(cctx);
@@ -728,7 +738,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
728
738
  DEBUGLOG(5, "ZSTDMT_compressionJob: compress %u bytes in %i blocks", (U32)job->src.size, nbChunks);
729
739
  assert(job->cSize == 0);
730
740
  for (chunkNb = 1; chunkNb < nbChunks; chunkNb++) {
731
- size_t const cSize = ZSTD_compressContinue(cctx, op, oend-op, ip, chunkSize);
741
+ size_t const cSize = ZSTD_compressContinue_public(cctx, op, oend-op, ip, chunkSize);
732
742
  if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
733
743
  ip += chunkSize;
734
744
  op += cSize; assert(op < oend);
@@ -748,11 +758,18 @@ static void ZSTDMT_compressionJob(void* jobDescription)
748
758
  size_t const lastBlockSize1 = job->src.size & (chunkSize-1);
749
759
  size_t const lastBlockSize = ((lastBlockSize1==0) & (job->src.size>=chunkSize)) ? chunkSize : lastBlockSize1;
750
760
  size_t const cSize = (job->lastJob) ?
751
- ZSTD_compressEnd (cctx, op, oend-op, ip, lastBlockSize) :
752
- ZSTD_compressContinue(cctx, op, oend-op, ip, lastBlockSize);
761
+ ZSTD_compressEnd_public(cctx, op, oend-op, ip, lastBlockSize) :
762
+ ZSTD_compressContinue_public(cctx, op, oend-op, ip, lastBlockSize);
753
763
  if (ZSTD_isError(cSize)) JOB_ERROR(cSize);
754
764
  lastCBlockSize = cSize;
755
765
  } }
766
+ if (!job->firstJob) {
767
+ /* Double check that we don't have an ext-dict, because then our
768
+ * repcode invalidation doesn't work.
769
+ */
770
+ assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
771
+ }
772
+ ZSTD_CCtx_trace(cctx, 0);
756
773
 
757
774
  _endJob:
758
775
  ZSTDMT_serialState_ensureFinished(job->serial, job->jobID, job->cSize);
@@ -799,6 +816,15 @@ typedef struct {
799
816
  static const roundBuff_t kNullRoundBuff = {NULL, 0, 0};
800
817
 
801
818
  #define RSYNC_LENGTH 32
819
+ /* Don't create chunks smaller than the zstd block size.
820
+ * This stops us from regressing compression ratio too much,
821
+ * and ensures our output fits in ZSTD_compressBound().
822
+ *
823
+ * If this is shrunk < ZSTD_BLOCKSIZELOG_MIN then
824
+ * ZSTD_COMPRESSBOUND() will need to be updated.
825
+ */
826
+ #define RSYNC_MIN_BLOCK_LOG ZSTD_BLOCKSIZELOG_MAX
827
+ #define RSYNC_MIN_BLOCK_SIZE (1<<RSYNC_MIN_BLOCK_LOG)
802
828
 
803
829
  typedef struct {
804
830
  U64 hash;
@@ -820,7 +846,6 @@ struct ZSTDMT_CCtx_s {
820
846
  roundBuff_t roundBuff;
821
847
  serialState_t serial;
822
848
  rsyncState_t rsync;
823
- unsigned singleBlockingThread;
824
849
  unsigned jobIDMask;
825
850
  unsigned doneJobID;
826
851
  unsigned nextJobID;
@@ -832,6 +857,7 @@ struct ZSTDMT_CCtx_s {
832
857
  ZSTD_customMem cMem;
833
858
  ZSTD_CDict* cdictLocal;
834
859
  const ZSTD_CDict* cdict;
860
+ unsigned providedFactory: 1;
835
861
  };
836
862
 
837
863
  static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZSTD_customMem cMem)
@@ -842,7 +868,7 @@ static void ZSTDMT_freeJobsTable(ZSTDMT_jobDescription* jobTable, U32 nbJobs, ZS
842
868
  ZSTD_pthread_mutex_destroy(&jobTable[jobNb].job_mutex);
843
869
  ZSTD_pthread_cond_destroy(&jobTable[jobNb].job_cond);
844
870
  }
845
- ZSTD_free(jobTable, cMem);
871
+ ZSTD_customFree(jobTable, cMem);
846
872
  }
847
873
 
848
874
  /* ZSTDMT_allocJobsTable()
@@ -854,7 +880,7 @@ static ZSTDMT_jobDescription* ZSTDMT_createJobsTable(U32* nbJobsPtr, ZSTD_custom
854
880
  U32 const nbJobs = 1 << nbJobsLog2;
855
881
  U32 jobNb;
856
882
  ZSTDMT_jobDescription* const jobTable = (ZSTDMT_jobDescription*)
857
- ZSTD_calloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
883
+ ZSTD_customCalloc(nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
858
884
  int initError = 0;
859
885
  if (jobTable==NULL) return NULL;
860
886
  *nbJobsPtr = nbJobs;
@@ -885,12 +911,12 @@ static size_t ZSTDMT_expandJobsTable (ZSTDMT_CCtx* mtctx, U32 nbWorkers) {
885
911
 
886
912
  /* ZSTDMT_CCtxParam_setNbWorkers():
887
913
  * Internal use only */
888
- size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
914
+ static size_t ZSTDMT_CCtxParam_setNbWorkers(ZSTD_CCtx_params* params, unsigned nbWorkers)
889
915
  {
890
916
  return ZSTD_CCtxParams_setParameter(params, ZSTD_c_nbWorkers, (int)nbWorkers);
891
917
  }
892
918
 
893
- MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem)
919
+ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
894
920
  {
895
921
  ZSTDMT_CCtx* mtctx;
896
922
  U32 nbJobs = nbWorkers + 2;
@@ -903,16 +929,23 @@ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers,
903
929
  /* invalid custom allocator */
904
930
  return NULL;
905
931
 
906
- mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
932
+ mtctx = (ZSTDMT_CCtx*) ZSTD_customCalloc(sizeof(ZSTDMT_CCtx), cMem);
907
933
  if (!mtctx) return NULL;
908
934
  ZSTDMT_CCtxParam_setNbWorkers(&mtctx->params, nbWorkers);
909
935
  mtctx->cMem = cMem;
910
936
  mtctx->allJobsCompleted = 1;
911
- mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
937
+ if (pool != NULL) {
938
+ mtctx->factory = pool;
939
+ mtctx->providedFactory = 1;
940
+ }
941
+ else {
942
+ mtctx->factory = POOL_create_advanced(nbWorkers, 0, cMem);
943
+ mtctx->providedFactory = 0;
944
+ }
912
945
  mtctx->jobs = ZSTDMT_createJobsTable(&nbJobs, cMem);
913
946
  assert(nbJobs > 0); assert((nbJobs & (nbJobs - 1)) == 0); /* ensure nbJobs is a power of 2 */
914
947
  mtctx->jobIDMask = nbJobs - 1;
915
- mtctx->bufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
948
+ mtctx->bufPool = ZSTDMT_createBufferPool(BUF_POOL_MAX_NB_BUFFERS(nbWorkers), cMem);
916
949
  mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
917
950
  mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
918
951
  initError = ZSTDMT_serialState_init(&mtctx->serial);
@@ -925,22 +958,18 @@ MEM_STATIC ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced_internal(unsigned nbWorkers,
925
958
  return mtctx;
926
959
  }
927
960
 
928
- ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem)
961
+ ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem, ZSTD_threadPool* pool)
929
962
  {
930
963
  #ifdef ZSTD_MULTITHREAD
931
- return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem);
964
+ return ZSTDMT_createCCtx_advanced_internal(nbWorkers, cMem, pool);
932
965
  #else
933
966
  (void)nbWorkers;
934
967
  (void)cMem;
968
+ (void)pool;
935
969
  return NULL;
936
970
  #endif
937
971
  }
938
972
 
939
- ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbWorkers)
940
- {
941
- return ZSTDMT_createCCtx_advanced(nbWorkers, ZSTD_defaultCMem);
942
- }
943
-
944
973
 
945
974
  /* ZSTDMT_releaseAllJobResources() :
946
975
  * note : ensure all workers are killed first ! */
@@ -957,7 +986,7 @@ static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
957
986
  ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
958
987
 
959
988
  /* Clear the job description, but keep the mutex/cond */
960
- memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
989
+ ZSTD_memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
961
990
  mtctx->jobs[jobID].job_mutex = mutex;
962
991
  mtctx->jobs[jobID].job_cond = cond;
963
992
  }
@@ -984,7 +1013,8 @@ static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* mtctx)
984
1013
  size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
985
1014
  {
986
1015
  if (mtctx==NULL) return 0; /* compatible with free on NULL */
987
- POOL_free(mtctx->factory); /* stop and free worker threads */
1016
+ if (!mtctx->providedFactory)
1017
+ POOL_free(mtctx->factory); /* stop and free worker threads */
988
1018
  ZSTDMT_releaseAllJobResources(mtctx); /* release job resources into pools first */
989
1019
  ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
990
1020
  ZSTDMT_freeBufferPool(mtctx->bufPool);
@@ -993,8 +1023,8 @@ size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
993
1023
  ZSTDMT_serialState_free(&mtctx->serial);
994
1024
  ZSTD_freeCDict(mtctx->cdictLocal);
995
1025
  if (mtctx->roundBuff.buffer)
996
- ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
997
- ZSTD_free(mtctx, mtctx->cMem);
1026
+ ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
1027
+ ZSTD_customFree(mtctx, mtctx->cMem);
998
1028
  return 0;
999
1029
  }
1000
1030
 
@@ -1011,65 +1041,6 @@ size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
1011
1041
  + mtctx->roundBuff.capacity;
1012
1042
  }
1013
1043
 
1014
- /* Internal only */
1015
- size_t
1016
- ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params,
1017
- ZSTDMT_parameter parameter,
1018
- int value)
1019
- {
1020
- DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter");
1021
- switch(parameter)
1022
- {
1023
- case ZSTDMT_p_jobSize :
1024
- DEBUGLOG(4, "ZSTDMT_CCtxParam_setMTCtxParameter : set jobSize to %i", value);
1025
- return ZSTD_CCtxParams_setParameter(params, ZSTD_c_jobSize, value);
1026
- case ZSTDMT_p_overlapLog :
1027
- DEBUGLOG(4, "ZSTDMT_p_overlapLog : %i", value);
1028
- return ZSTD_CCtxParams_setParameter(params, ZSTD_c_overlapLog, value);
1029
- case ZSTDMT_p_rsyncable :
1030
- DEBUGLOG(4, "ZSTD_p_rsyncable : %i", value);
1031
- return ZSTD_CCtxParams_setParameter(params, ZSTD_c_rsyncable, value);
1032
- default :
1033
- return ERROR(parameter_unsupported);
1034
- }
1035
- }
1036
-
1037
- size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int value)
1038
- {
1039
- DEBUGLOG(4, "ZSTDMT_setMTCtxParameter");
1040
- return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
1041
- }
1042
-
1043
- size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, int* value)
1044
- {
1045
- switch (parameter) {
1046
- case ZSTDMT_p_jobSize:
1047
- return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_jobSize, value);
1048
- case ZSTDMT_p_overlapLog:
1049
- return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_overlapLog, value);
1050
- case ZSTDMT_p_rsyncable:
1051
- return ZSTD_CCtxParams_getParameter(&mtctx->params, ZSTD_c_rsyncable, value);
1052
- default:
1053
- return ERROR(parameter_unsupported);
1054
- }
1055
- }
1056
-
1057
- /* Sets parameters relevant to the compression job,
1058
- * initializing others to default values. */
1059
- static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(const ZSTD_CCtx_params* params)
1060
- {
1061
- ZSTD_CCtx_params jobParams = *params;
1062
- /* Clear parameters related to multithreading */
1063
- jobParams.forceWindow = 0;
1064
- jobParams.nbWorkers = 0;
1065
- jobParams.jobSize = 0;
1066
- jobParams.overlapLog = 0;
1067
- jobParams.rsyncable = 0;
1068
- memset(&jobParams.ldmParams, 0, sizeof(ldmParams_t));
1069
- memset(&jobParams.customMem, 0, sizeof(ZSTD_customMem));
1070
- return jobParams;
1071
- }
1072
-
1073
1044
 
1074
1045
  /* ZSTDMT_resize() :
1075
1046
  * @return : error code if fails, 0 on success */
@@ -1077,7 +1048,7 @@ static size_t ZSTDMT_resize(ZSTDMT_CCtx* mtctx, unsigned nbWorkers)
1077
1048
  {
1078
1049
  if (POOL_resize(mtctx->factory, nbWorkers)) return ERROR(memory_allocation);
1079
1050
  FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbWorkers) , "");
1080
- mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, nbWorkers);
1051
+ mtctx->bufPool = ZSTDMT_expandBufferPool(mtctx->bufPool, BUF_POOL_MAX_NB_BUFFERS(nbWorkers));
1081
1052
  if (mtctx->bufPool == NULL) return ERROR(memory_allocation);
1082
1053
  mtctx->cctxPool = ZSTDMT_expandCCtxPool(mtctx->cctxPool, nbWorkers);
1083
1054
  if (mtctx->cctxPool == NULL) return ERROR(memory_allocation);
@@ -1098,7 +1069,7 @@ void ZSTDMT_updateCParams_whileCompressing(ZSTDMT_CCtx* mtctx, const ZSTD_CCtx_p
1098
1069
  DEBUGLOG(5, "ZSTDMT_updateCParams_whileCompressing (level:%i)",
1099
1070
  compressionLevel);
1100
1071
  mtctx->params.compressionLevel = compressionLevel;
1101
- { ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0);
1072
+ { ZSTD_compressionParameters cParams = ZSTD_getCParamsFromCCtxParams(cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, 0, ZSTD_cpm_noAttachDict);
1102
1073
  cParams.windowLog = saved_wlog;
1103
1074
  mtctx->params.cParams = cParams;
1104
1075
  }
@@ -1182,11 +1153,11 @@ size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
1182
1153
  static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
1183
1154
  {
1184
1155
  unsigned jobLog;
1185
- if (params->ldmParams.enableLdm) {
1156
+ if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
1186
1157
  /* In Long Range Mode, the windowLog is typically oversized.
1187
1158
  * In which case, it's preferable to determine the jobSize
1188
- * based on chainLog instead. */
1189
- jobLog = MAX(21, params->cParams.chainLog + 4);
1159
+ * based on cycleLog instead. */
1160
+ jobLog = MAX(21, ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy) + 3);
1190
1161
  } else {
1191
1162
  jobLog = MAX(20, params->cParams.windowLog + 2);
1192
1163
  }
@@ -1226,7 +1197,7 @@ static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
1226
1197
  int const overlapRLog = 9 - ZSTDMT_overlapLog(params->overlapLog, params->cParams.strategy);
1227
1198
  int ovLog = (overlapRLog >= 8) ? 0 : (params->cParams.windowLog - overlapRLog);
1228
1199
  assert(0 <= overlapRLog && overlapRLog <= 8);
1229
- if (params->ldmParams.enableLdm) {
1200
+ if (params->ldmParams.enableLdm == ZSTD_ps_enable) {
1230
1201
  /* In Long Range Mode, the windowLog is typically oversized.
1231
1202
  * In which case, it's preferable to determine the jobSize
1232
1203
  * based on chainLog instead.
@@ -1240,174 +1211,6 @@ static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
1240
1211
  return (ovLog==0) ? 0 : (size_t)1 << ovLog;
1241
1212
  }
1242
1213
 
1243
- static unsigned
1244
- ZSTDMT_computeNbJobs(const ZSTD_CCtx_params* params, size_t srcSize, unsigned nbWorkers)
1245
- {
1246
- assert(nbWorkers>0);
1247
- { size_t const jobSizeTarget = (size_t)1 << ZSTDMT_computeTargetJobLog(params);
1248
- size_t const jobMaxSize = jobSizeTarget << 2;
1249
- size_t const passSizeMax = jobMaxSize * nbWorkers;
1250
- unsigned const multiplier = (unsigned)(srcSize / passSizeMax) + 1;
1251
- unsigned const nbJobsLarge = multiplier * nbWorkers;
1252
- unsigned const nbJobsMax = (unsigned)(srcSize / jobSizeTarget) + 1;
1253
- unsigned const nbJobsSmall = MIN(nbJobsMax, nbWorkers);
1254
- return (multiplier>1) ? nbJobsLarge : nbJobsSmall;
1255
- } }
1256
-
1257
- /* ZSTDMT_compress_advanced_internal() :
1258
- * This is a blocking function : it will only give back control to caller after finishing its compression job.
1259
- */
1260
- static size_t
1261
- ZSTDMT_compress_advanced_internal(
1262
- ZSTDMT_CCtx* mtctx,
1263
- void* dst, size_t dstCapacity,
1264
- const void* src, size_t srcSize,
1265
- const ZSTD_CDict* cdict,
1266
- ZSTD_CCtx_params params)
1267
- {
1268
- ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(&params);
1269
- size_t const overlapSize = ZSTDMT_computeOverlapSize(&params);
1270
- unsigned const nbJobs = ZSTDMT_computeNbJobs(&params, srcSize, params.nbWorkers);
1271
- size_t const proposedJobSize = (srcSize + (nbJobs-1)) / nbJobs;
1272
- size_t const avgJobSize = (((proposedJobSize-1) & 0x1FFFF) < 0x7FFF) ? proposedJobSize + 0xFFFF : proposedJobSize; /* avoid too small last block */
1273
- const char* const srcStart = (const char*)src;
1274
- size_t remainingSrcSize = srcSize;
1275
- unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbJobs : (unsigned)(dstCapacity / ZSTD_compressBound(avgJobSize)); /* presumes avgJobSize >= 256 KB, which should be the case */
1276
- size_t frameStartPos = 0, dstBufferPos = 0;
1277
- assert(jobParams.nbWorkers == 0);
1278
- assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
1279
-
1280
- params.jobSize = (U32)avgJobSize;
1281
- DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: nbJobs=%2u (rawSize=%u bytes; fixedSize=%u) ",
1282
- nbJobs, (U32)proposedJobSize, (U32)avgJobSize);
1283
-
1284
- if ((nbJobs==1) | (params.nbWorkers<=1)) { /* fallback to single-thread mode : this is a blocking invocation anyway */
1285
- ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
1286
- DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: fallback to single-thread mode");
1287
- if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
1288
- return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, &jobParams);
1289
- }
1290
-
1291
- assert(avgJobSize >= 256 KB); /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), required to compress directly into Dst (no additional buffer) */
1292
- ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgJobSize) );
1293
- /* LDM doesn't even try to load the dictionary in single-ingestion mode */
1294
- if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, avgJobSize, NULL, 0, ZSTD_dct_auto))
1295
- return ERROR(memory_allocation);
1296
-
1297
- FORWARD_IF_ERROR( ZSTDMT_expandJobsTable(mtctx, nbJobs) , ""); /* only expands if necessary */
1298
-
1299
- { unsigned u;
1300
- for (u=0; u<nbJobs; u++) {
1301
- size_t const jobSize = MIN(remainingSrcSize, avgJobSize);
1302
- size_t const dstBufferCapacity = ZSTD_compressBound(jobSize);
1303
- buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
1304
- buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
1305
- size_t dictSize = u ? overlapSize : 0;
1306
-
1307
- mtctx->jobs[u].prefix.start = srcStart + frameStartPos - dictSize;
1308
- mtctx->jobs[u].prefix.size = dictSize;
1309
- mtctx->jobs[u].src.start = srcStart + frameStartPos;
1310
- mtctx->jobs[u].src.size = jobSize; assert(jobSize > 0); /* avoid job.src.size == 0 */
1311
- mtctx->jobs[u].consumed = 0;
1312
- mtctx->jobs[u].cSize = 0;
1313
- mtctx->jobs[u].cdict = (u==0) ? cdict : NULL;
1314
- mtctx->jobs[u].fullFrameSize = srcSize;
1315
- mtctx->jobs[u].params = jobParams;
1316
- /* do not calculate checksum within sections, but write it in header for first section */
1317
- mtctx->jobs[u].dstBuff = dstBuffer;
1318
- mtctx->jobs[u].cctxPool = mtctx->cctxPool;
1319
- mtctx->jobs[u].bufPool = mtctx->bufPool;
1320
- mtctx->jobs[u].seqPool = mtctx->seqPool;
1321
- mtctx->jobs[u].serial = &mtctx->serial;
1322
- mtctx->jobs[u].jobID = u;
1323
- mtctx->jobs[u].firstJob = (u==0);
1324
- mtctx->jobs[u].lastJob = (u==nbJobs-1);
1325
-
1326
- DEBUGLOG(5, "ZSTDMT_compress_advanced_internal: posting job %u (%u bytes)", u, (U32)jobSize);
1327
- DEBUG_PRINTHEX(6, mtctx->jobs[u].prefix.start, 12);
1328
- POOL_add(mtctx->factory, ZSTDMT_compressionJob, &mtctx->jobs[u]);
1329
-
1330
- frameStartPos += jobSize;
1331
- dstBufferPos += dstBufferCapacity;
1332
- remainingSrcSize -= jobSize;
1333
- } }
1334
-
1335
- /* collect result */
1336
- { size_t error = 0, dstPos = 0;
1337
- unsigned jobID;
1338
- for (jobID=0; jobID<nbJobs; jobID++) {
1339
- DEBUGLOG(5, "waiting for job %u ", jobID);
1340
- ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobs[jobID].job_mutex);
1341
- while (mtctx->jobs[jobID].consumed < mtctx->jobs[jobID].src.size) {
1342
- DEBUGLOG(5, "waiting for jobCompleted signal from job %u", jobID);
1343
- ZSTD_pthread_cond_wait(&mtctx->jobs[jobID].job_cond, &mtctx->jobs[jobID].job_mutex);
1344
- }
1345
- ZSTD_pthread_mutex_unlock(&mtctx->jobs[jobID].job_mutex);
1346
- DEBUGLOG(5, "ready to write job %u ", jobID);
1347
-
1348
- { size_t const cSize = mtctx->jobs[jobID].cSize;
1349
- if (ZSTD_isError(cSize)) error = cSize;
1350
- if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
1351
- if (jobID) { /* note : job 0 is written directly at dst, which is correct position */
1352
- if (!error)
1353
- memmove((char*)dst + dstPos, mtctx->jobs[jobID].dstBuff.start, cSize); /* may overlap when job compressed within dst */
1354
- if (jobID >= compressWithinDst) { /* job compressed into its own buffer, which must be released */
1355
- DEBUGLOG(5, "releasing buffer %u>=%u", jobID, compressWithinDst);
1356
- ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
1357
- } }
1358
- mtctx->jobs[jobID].dstBuff = g_nullBuffer;
1359
- mtctx->jobs[jobID].cSize = 0;
1360
- dstPos += cSize ;
1361
- }
1362
- } /* for (jobID=0; jobID<nbJobs; jobID++) */
1363
-
1364
- DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
1365
- if (params.fParams.checksumFlag) {
1366
- U32 const checksum = (U32)XXH64_digest(&mtctx->serial.xxhState);
1367
- if (dstPos + 4 > dstCapacity) {
1368
- error = ERROR(dstSize_tooSmall);
1369
- } else {
1370
- DEBUGLOG(4, "writing checksum : %08X \n", checksum);
1371
- MEM_writeLE32((char*)dst + dstPos, checksum);
1372
- dstPos += 4;
1373
- } }
1374
-
1375
- if (!error) DEBUGLOG(4, "compressed size : %u ", (U32)dstPos);
1376
- return error ? error : dstPos;
1377
- }
1378
- }
1379
-
1380
- size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
1381
- void* dst, size_t dstCapacity,
1382
- const void* src, size_t srcSize,
1383
- const ZSTD_CDict* cdict,
1384
- ZSTD_parameters params,
1385
- int overlapLog)
1386
- {
1387
- ZSTD_CCtx_params cctxParams = mtctx->params;
1388
- cctxParams.cParams = params.cParams;
1389
- cctxParams.fParams = params.fParams;
1390
- assert(ZSTD_OVERLAPLOG_MIN <= overlapLog && overlapLog <= ZSTD_OVERLAPLOG_MAX);
1391
- cctxParams.overlapLog = overlapLog;
1392
- return ZSTDMT_compress_advanced_internal(mtctx,
1393
- dst, dstCapacity,
1394
- src, srcSize,
1395
- cdict, cctxParams);
1396
- }
1397
-
1398
-
1399
- size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
1400
- void* dst, size_t dstCapacity,
1401
- const void* src, size_t srcSize,
1402
- int compressionLevel)
1403
- {
1404
- ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
1405
- int const overlapLog = ZSTDMT_overlapLog_default(params.cParams.strategy);
1406
- params.fParams.contentSizeFlag = 1;
1407
- return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
1408
- }
1409
-
1410
-
1411
1214
  /* ====================================== */
1412
1215
  /* ======= Streaming API ======= */
1413
1216
  /* ====================================== */
@@ -1432,16 +1235,6 @@ size_t ZSTDMT_initCStream_internal(
1432
1235
  if (params.jobSize != 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
1433
1236
  if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = (size_t)ZSTDMT_JOBSIZE_MAX;
1434
1237
 
1435
- mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN); /* do not trigger multi-threading when srcSize is too small */
1436
- if (mtctx->singleBlockingThread) {
1437
- ZSTD_CCtx_params const singleThreadParams = ZSTDMT_initJobCCtxParams(&params);
1438
- DEBUGLOG(5, "ZSTDMT_initCStream_internal: switch to single blocking thread mode");
1439
- assert(singleThreadParams.nbWorkers == 0);
1440
- return ZSTD_initCStream_internal(mtctx->cctxPool->cctx[0],
1441
- dict, dictSize, cdict,
1442
- &singleThreadParams, pledgedSrcSize);
1443
- }
1444
-
1445
1238
  DEBUGLOG(4, "ZSTDMT_initCStream_internal: %u workers", params.nbWorkers);
1446
1239
 
1447
1240
  if (mtctx->allJobsCompleted == 0) { /* previous compression not correctly finished */
@@ -1475,9 +1268,11 @@ size_t ZSTDMT_initCStream_internal(
1475
1268
 
1476
1269
  if (params.rsyncable) {
1477
1270
  /* Aim for the targetsectionSize as the average job size. */
1478
- U32 const jobSizeMB = (U32)(mtctx->targetSectionSize >> 20);
1479
- U32 const rsyncBits = ZSTD_highbit32(jobSizeMB) + 20;
1480
- assert(jobSizeMB >= 1);
1271
+ U32 const jobSizeKB = (U32)(mtctx->targetSectionSize >> 10);
1272
+ U32 const rsyncBits = (assert(jobSizeKB >= 1), ZSTD_highbit32(jobSizeKB) + 10);
1273
+ /* We refuse to create jobs < RSYNC_MIN_BLOCK_SIZE bytes, so make sure our
1274
+ * expected job size is at least 4x larger. */
1275
+ assert(rsyncBits >= RSYNC_MIN_BLOCK_LOG + 2);
1481
1276
  DEBUGLOG(4, "rsyncLog = %u", rsyncBits);
1482
1277
  mtctx->rsync.hash = 0;
1483
1278
  mtctx->rsync.hitMask = (1ULL << rsyncBits) - 1;
@@ -1489,7 +1284,7 @@ size_t ZSTDMT_initCStream_internal(
1489
1284
  ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
1490
1285
  {
1491
1286
  /* If ldm is enabled we need windowSize space. */
1492
- size_t const windowSize = mtctx->params.ldmParams.enableLdm ? (1U << mtctx->params.cParams.windowLog) : 0;
1287
+ size_t const windowSize = mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable ? (1U << mtctx->params.cParams.windowLog) : 0;
1493
1288
  /* Two buffers of slack, plus extra space for the overlap
1494
1289
  * This is the minimum slack that LDM works with. One extra because
1495
1290
  * flush might waste up to targetSectionSize-1 bytes. Another extra
@@ -1504,8 +1299,8 @@ size_t ZSTDMT_initCStream_internal(
1504
1299
  size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
1505
1300
  if (mtctx->roundBuff.capacity < capacity) {
1506
1301
  if (mtctx->roundBuff.buffer)
1507
- ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
1508
- mtctx->roundBuff.buffer = (BYTE*)ZSTD_malloc(capacity, mtctx->cMem);
1302
+ ZSTD_customFree(mtctx->roundBuff.buffer, mtctx->cMem);
1303
+ mtctx->roundBuff.buffer = (BYTE*)ZSTD_customMalloc(capacity, mtctx->cMem);
1509
1304
  if (mtctx->roundBuff.buffer == NULL) {
1510
1305
  mtctx->roundBuff.capacity = 0;
1511
1306
  return ERROR(memory_allocation);
@@ -1530,53 +1325,6 @@ size_t ZSTDMT_initCStream_internal(
1530
1325
  return 0;
1531
1326
  }
1532
1327
 
1533
- size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
1534
- const void* dict, size_t dictSize,
1535
- ZSTD_parameters params,
1536
- unsigned long long pledgedSrcSize)
1537
- {
1538
- ZSTD_CCtx_params cctxParams = mtctx->params; /* retrieve sticky params */
1539
- DEBUGLOG(4, "ZSTDMT_initCStream_advanced (pledgedSrcSize=%u)", (U32)pledgedSrcSize);
1540
- cctxParams.cParams = params.cParams;
1541
- cctxParams.fParams = params.fParams;
1542
- return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, ZSTD_dct_auto, NULL,
1543
- cctxParams, pledgedSrcSize);
1544
- }
1545
-
1546
- size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
1547
- const ZSTD_CDict* cdict,
1548
- ZSTD_frameParameters fParams,
1549
- unsigned long long pledgedSrcSize)
1550
- {
1551
- ZSTD_CCtx_params cctxParams = mtctx->params;
1552
- if (cdict==NULL) return ERROR(dictionary_wrong); /* method incompatible with NULL cdict */
1553
- cctxParams.cParams = ZSTD_getCParamsFromCDict(cdict);
1554
- cctxParams.fParams = fParams;
1555
- return ZSTDMT_initCStream_internal(mtctx, NULL, 0 /*dictSize*/, ZSTD_dct_auto, cdict,
1556
- cctxParams, pledgedSrcSize);
1557
- }
1558
-
1559
-
1560
- /* ZSTDMT_resetCStream() :
1561
- * pledgedSrcSize can be zero == unknown (for the time being)
1562
- * prefer using ZSTD_CONTENTSIZE_UNKNOWN,
1563
- * as `0` might mean "empty" in the future */
1564
- size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize)
1565
- {
1566
- if (!pledgedSrcSize) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
1567
- return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, 0, mtctx->params,
1568
- pledgedSrcSize);
1569
- }
1570
-
1571
- size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel) {
1572
- ZSTD_parameters const params = ZSTD_getParams(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
1573
- ZSTD_CCtx_params cctxParams = mtctx->params; /* retrieve sticky params */
1574
- DEBUGLOG(4, "ZSTDMT_initCStream (cLevel=%i)", compressionLevel);
1575
- cctxParams.cParams = params.cParams;
1576
- cctxParams.fParams = params.fParams;
1577
- return ZSTDMT_initCStream_internal(mtctx, NULL, 0, ZSTD_dct_auto, NULL, cctxParams, ZSTD_CONTENTSIZE_UNKNOWN);
1578
- }
1579
-
1580
1328
 
1581
1329
  /* ZSTDMT_writeLastEmptyBlock()
1582
1330
  * Write a single empty block with an end-of-frame to finish a frame.
@@ -1740,7 +1488,7 @@ static size_t ZSTDMT_flushProduced(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, u
1740
1488
  assert(cSize >= mtctx->jobs[wJobID].dstFlushed);
1741
1489
  assert(mtctx->jobs[wJobID].dstBuff.start != NULL);
1742
1490
  if (toFlush > 0) {
1743
- memcpy((char*)output->dst + output->pos,
1491
+ ZSTD_memcpy((char*)output->dst + output->pos,
1744
1492
  (const char*)mtctx->jobs[wJobID].dstBuff.start + mtctx->jobs[wJobID].dstFlushed,
1745
1493
  toFlush);
1746
1494
  }
@@ -1811,17 +1559,21 @@ static range_t ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
1811
1559
  static int ZSTDMT_isOverlapped(buffer_t buffer, range_t range)
1812
1560
  {
1813
1561
  BYTE const* const bufferStart = (BYTE const*)buffer.start;
1814
- BYTE const* const bufferEnd = bufferStart + buffer.capacity;
1815
1562
  BYTE const* const rangeStart = (BYTE const*)range.start;
1816
- BYTE const* const rangeEnd = range.size != 0 ? rangeStart + range.size : rangeStart;
1817
1563
 
1818
1564
  if (rangeStart == NULL || bufferStart == NULL)
1819
1565
  return 0;
1820
- /* Empty ranges cannot overlap */
1821
- if (bufferStart == bufferEnd || rangeStart == rangeEnd)
1822
- return 0;
1823
1566
 
1824
- return bufferStart < rangeEnd && rangeStart < bufferEnd;
1567
+ {
1568
+ BYTE const* const bufferEnd = bufferStart + buffer.capacity;
1569
+ BYTE const* const rangeEnd = rangeStart + range.size;
1570
+
1571
+ /* Empty ranges cannot overlap */
1572
+ if (bufferStart == bufferEnd || rangeStart == rangeEnd)
1573
+ return 0;
1574
+
1575
+ return bufferStart < rangeEnd && rangeStart < bufferEnd;
1576
+ }
1825
1577
  }
1826
1578
 
1827
1579
  static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
@@ -1848,7 +1600,7 @@ static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
1848
1600
 
1849
1601
  static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
1850
1602
  {
1851
- if (mtctx->params.ldmParams.enableLdm) {
1603
+ if (mtctx->params.ldmParams.enableLdm == ZSTD_ps_enable) {
1852
1604
  ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
1853
1605
  DEBUGLOG(5, "ZSTDMT_waitForLdmComplete");
1854
1606
  DEBUGLOG(5, "source [0x%zx, 0x%zx)",
@@ -1894,7 +1646,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
1894
1646
  return 0;
1895
1647
  }
1896
1648
  ZSTDMT_waitForLdmComplete(mtctx, buffer);
1897
- memmove(start, mtctx->inBuff.prefix.start, prefixSize);
1649
+ ZSTD_memmove(start, mtctx->inBuff.prefix.start, prefixSize);
1898
1650
  mtctx->inBuff.prefix.start = start;
1899
1651
  mtctx->roundBuff.pos = prefixSize;
1900
1652
  }
@@ -1951,6 +1703,11 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
1951
1703
  if (!mtctx->params.rsyncable)
1952
1704
  /* Rsync is disabled. */
1953
1705
  return syncPoint;
1706
+ if (mtctx->inBuff.filled + input.size - input.pos < RSYNC_MIN_BLOCK_SIZE)
1707
+ /* We don't emit synchronization points if it would produce too small blocks.
1708
+ * We don't have enough input to find a synchronization point, so don't look.
1709
+ */
1710
+ return syncPoint;
1954
1711
  if (mtctx->inBuff.filled + syncPoint.toLoad < RSYNC_LENGTH)
1955
1712
  /* Not enough to compute the hash.
1956
1713
  * We will miss any synchronization points in this RSYNC_LENGTH byte
@@ -1961,23 +1718,41 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
1961
1718
  */
1962
1719
  return syncPoint;
1963
1720
  /* Initialize the loop variables. */
1964
- if (mtctx->inBuff.filled >= RSYNC_LENGTH) {
1965
- /* We have enough bytes buffered to initialize the hash.
1721
+ if (mtctx->inBuff.filled < RSYNC_MIN_BLOCK_SIZE) {
1722
+ /* We don't need to scan the first RSYNC_MIN_BLOCK_SIZE positions
1723
+ * because they can't possibly be a sync point. So we can start
1724
+ * part way through the input buffer.
1725
+ */
1726
+ pos = RSYNC_MIN_BLOCK_SIZE - mtctx->inBuff.filled;
1727
+ if (pos >= RSYNC_LENGTH) {
1728
+ prev = istart + pos - RSYNC_LENGTH;
1729
+ hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
1730
+ } else {
1731
+ assert(mtctx->inBuff.filled >= RSYNC_LENGTH);
1732
+ prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
1733
+ hash = ZSTD_rollingHash_compute(prev + pos, (RSYNC_LENGTH - pos));
1734
+ hash = ZSTD_rollingHash_append(hash, istart, pos);
1735
+ }
1736
+ } else {
1737
+ /* We have enough bytes buffered to initialize the hash,
1738
+ * and have processed enough bytes to find a sync point.
1966
1739
  * Start scanning at the beginning of the input.
1967
1740
  */
1741
+ assert(mtctx->inBuff.filled >= RSYNC_MIN_BLOCK_SIZE);
1742
+ assert(RSYNC_MIN_BLOCK_SIZE >= RSYNC_LENGTH);
1968
1743
  pos = 0;
1969
1744
  prev = (BYTE const*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled - RSYNC_LENGTH;
1970
1745
  hash = ZSTD_rollingHash_compute(prev, RSYNC_LENGTH);
1971
- } else {
1972
- /* We don't have enough bytes buffered to initialize the hash, but
1973
- * we know we have at least RSYNC_LENGTH bytes total.
1974
- * Start scanning after the first RSYNC_LENGTH bytes less the bytes
1975
- * already buffered.
1976
- */
1977
- pos = RSYNC_LENGTH - mtctx->inBuff.filled;
1978
- prev = (BYTE const*)mtctx->inBuff.buffer.start - pos;
1979
- hash = ZSTD_rollingHash_compute(mtctx->inBuff.buffer.start, mtctx->inBuff.filled);
1980
- hash = ZSTD_rollingHash_append(hash, istart, pos);
1746
+ if ((hash & hitMask) == hitMask) {
1747
+ /* We're already at a sync point so don't load any more until
1748
+ * we're able to flush this sync point.
1749
+ * This likely happened because the job table was full so we
1750
+ * couldn't add our job.
1751
+ */
1752
+ syncPoint.toLoad = 0;
1753
+ syncPoint.flush = 1;
1754
+ return syncPoint;
1755
+ }
1981
1756
  }
1982
1757
  /* Starting with the hash of the previous RSYNC_LENGTH bytes, roll
1983
1758
  * through the input. If we hit a synchronization point, then cut the
@@ -1987,16 +1762,24 @@ findSynchronizationPoint(ZSTDMT_CCtx const* mtctx, ZSTD_inBuffer const input)
1987
1762
  * then a block will be emitted anyways, but this is okay, since if we
1988
1763
  * are already synchronized we will remain synchronized.
1989
1764
  */
1765
+ assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
1990
1766
  for (; pos < syncPoint.toLoad; ++pos) {
1991
1767
  BYTE const toRemove = pos < RSYNC_LENGTH ? prev[pos] : istart[pos - RSYNC_LENGTH];
1992
- /* if (pos >= RSYNC_LENGTH) assert(ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash); */
1768
+ /* This assert is very expensive, and Debian compiles with asserts enabled.
1769
+ * So disable it for now. We can get similar coverage by checking it at the
1770
+ * beginning & end of the loop.
1771
+ * assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
1772
+ */
1993
1773
  hash = ZSTD_rollingHash_rotate(hash, toRemove, istart[pos], primePower);
1774
+ assert(mtctx->inBuff.filled + pos >= RSYNC_MIN_BLOCK_SIZE);
1994
1775
  if ((hash & hitMask) == hitMask) {
1995
1776
  syncPoint.toLoad = pos + 1;
1996
1777
  syncPoint.flush = 1;
1778
+ ++pos; /* for assert */
1997
1779
  break;
1998
1780
  }
1999
1781
  }
1782
+ assert(pos < RSYNC_LENGTH || ZSTD_rollingHash_compute(istart + pos - RSYNC_LENGTH, RSYNC_LENGTH) == hash);
2000
1783
  return syncPoint;
2001
1784
  }
2002
1785
 
@@ -2022,34 +1805,11 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
2022
1805
  assert(output->pos <= output->size);
2023
1806
  assert(input->pos <= input->size);
2024
1807
 
2025
- if (mtctx->singleBlockingThread) { /* delegate to single-thread (synchronous) */
2026
- return ZSTD_compressStream2(mtctx->cctxPool->cctx[0], output, input, endOp);
2027
- }
2028
-
2029
1808
  if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
2030
1809
  /* current frame being ended. Only flush/end are allowed */
2031
1810
  return ERROR(stage_wrong);
2032
1811
  }
2033
1812
 
2034
- /* single-pass shortcut (note : synchronous-mode) */
2035
- if ( (!mtctx->params.rsyncable) /* rsyncable mode is disabled */
2036
- && (mtctx->nextJobID == 0) /* just started */
2037
- && (mtctx->inBuff.filled == 0) /* nothing buffered */
2038
- && (!mtctx->jobReady) /* no job already created */
2039
- && (endOp == ZSTD_e_end) /* end order */
2040
- && (output->size - output->pos >= ZSTD_compressBound(input->size - input->pos)) ) { /* enough space in dst */
2041
- size_t const cSize = ZSTDMT_compress_advanced_internal(mtctx,
2042
- (char*)output->dst + output->pos, output->size - output->pos,
2043
- (const char*)input->src + input->pos, input->size - input->pos,
2044
- mtctx->cdict, mtctx->params);
2045
- if (ZSTD_isError(cSize)) return cSize;
2046
- input->pos = input->size;
2047
- output->pos += cSize;
2048
- mtctx->allJobsCompleted = 1;
2049
- mtctx->frameEnded = 1;
2050
- return 0;
2051
- }
2052
-
2053
1813
  /* fill input buffer */
2054
1814
  if ( (!mtctx->jobReady)
2055
1815
  && (input->size > input->pos) ) { /* support NULL input */
@@ -2072,13 +1832,21 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
2072
1832
  assert(mtctx->inBuff.buffer.capacity >= mtctx->targetSectionSize);
2073
1833
  DEBUGLOG(5, "ZSTDMT_compressStream_generic: adding %u bytes on top of %u to buffer of size %u",
2074
1834
  (U32)syncPoint.toLoad, (U32)mtctx->inBuff.filled, (U32)mtctx->targetSectionSize);
2075
- memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
1835
+ ZSTD_memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, syncPoint.toLoad);
2076
1836
  input->pos += syncPoint.toLoad;
2077
1837
  mtctx->inBuff.filled += syncPoint.toLoad;
2078
1838
  forwardInputProgress = syncPoint.toLoad>0;
2079
1839
  }
2080
- if ((input->pos < input->size) && (endOp == ZSTD_e_end))
2081
- endOp = ZSTD_e_flush; /* can't end now : not all input consumed */
1840
+ }
1841
+ if ((input->pos < input->size) && (endOp == ZSTD_e_end)) {
1842
+ /* Can't end yet because the input is not fully consumed.
1843
+ * We are in one of these cases:
1844
+ * - mtctx->inBuff is NULL & empty: we couldn't get an input buffer so don't create a new job.
1845
+ * - We filled the input buffer: flush this job but don't end the frame.
1846
+ * - We hit a synchronization point: flush this job but don't end the frame.
1847
+ */
1848
+ assert(mtctx->inBuff.filled == 0 || mtctx->inBuff.filled == mtctx->targetSectionSize || mtctx->params.rsyncable);
1849
+ endOp = ZSTD_e_flush;
2082
1850
  }
2083
1851
 
2084
1852
  if ( (mtctx->jobReady)
@@ -2097,47 +1865,3 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
2097
1865
  return remainingToFlush;
2098
1866
  }
2099
1867
  }
2100
-
2101
-
2102
- size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
2103
- {
2104
- FORWARD_IF_ERROR( ZSTDMT_compressStream_generic(mtctx, output, input, ZSTD_e_continue) , "");
2105
-
2106
- /* recommended next input size : fill current input buffer */
2107
- return mtctx->targetSectionSize - mtctx->inBuff.filled; /* note : could be zero when input buffer is fully filled and no more availability to create new job */
2108
- }
2109
-
2110
-
2111
- static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_EndDirective endFrame)
2112
- {
2113
- size_t const srcSize = mtctx->inBuff.filled;
2114
- DEBUGLOG(5, "ZSTDMT_flushStream_internal");
2115
-
2116
- if ( mtctx->jobReady /* one job ready for a worker to pick up */
2117
- || (srcSize > 0) /* still some data within input buffer */
2118
- || ((endFrame==ZSTD_e_end) && !mtctx->frameEnded)) { /* need a last 0-size block to end frame */
2119
- DEBUGLOG(5, "ZSTDMT_flushStream_internal : create a new job (%u bytes, end:%u)",
2120
- (U32)srcSize, (U32)endFrame);
2121
- FORWARD_IF_ERROR( ZSTDMT_createCompressionJob(mtctx, srcSize, endFrame) , "");
2122
- }
2123
-
2124
- /* check if there is any data available to flush */
2125
- return ZSTDMT_flushProduced(mtctx, output, 1 /* blockToFlush */, endFrame);
2126
- }
2127
-
2128
-
2129
- size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
2130
- {
2131
- DEBUGLOG(5, "ZSTDMT_flushStream");
2132
- if (mtctx->singleBlockingThread)
2133
- return ZSTD_flushStream(mtctx->cctxPool->cctx[0], output);
2134
- return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_flush);
2135
- }
2136
-
2137
- size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output)
2138
- {
2139
- DEBUGLOG(4, "ZSTDMT_endStream");
2140
- if (mtctx->singleBlockingThread)
2141
- return ZSTD_endStream(mtctx->cctxPool->cctx[0], output);
2142
- return ZSTDMT_flushStream_internal(mtctx, output, ZSTD_e_end);
2143
- }