2 * Copyright (c) Meta Platforms, Inc. and affiliates.
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.
11 /* This header contains definitions
12 * that shall **only** be used by modules within lib/compress.
15 #ifndef ZSTD_COMPRESS_H
16 #define ZSTD_COMPRESS_H
18 /*-*************************************
20 ***************************************/
21 #include "../common/zstd_internal.h"
22 #include "zstd_cwksp.h"
23 #ifdef ZSTD_MULTITHREAD
24 # include "zstdmt_compress.h"
26 #include "../common/bits.h" /* ZSTD_highbit32, ZSTD_NbCommonBytes */
28 #if defined (__cplusplus)
32 /*-*************************************
34 ***************************************/
35 #define kSearchStrength 8
36 #define HASH_READ_SIZE 8
37 #define ZSTD_DUBT_UNSORTED_MARK 1 /* For btlazy2 strategy, index ZSTD_DUBT_UNSORTED_MARK==1 means "unsorted".
38 It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
39 It's not a big deal though : candidate will just be sorted again.
40 Additionally, candidate position 1 will be lost.
41 But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
42 The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy.
43 This constant is required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
46 /*-*************************************
47 * Context memory management
48 ***************************************/
49 typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
50 typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
52 typedef struct ZSTD_prefixDict_s {
55 ZSTD_dictContentType_e dictContentType;
62 ZSTD_dictContentType_e dictContentType;
67 HUF_CElt CTable[HUF_CTABLE_SIZE_ST(255)];
68 HUF_repeat repeatMode;
72 FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
73 FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
74 FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
75 FSE_repeat offcode_repeatMode;
76 FSE_repeat matchlength_repeatMode;
77 FSE_repeat litlength_repeatMode;
81 ZSTD_hufCTables_t huf;
82 ZSTD_fseCTables_t fse;
83 } ZSTD_entropyCTables_t;
85 /***********************************************
86 * Entropy buffer statistics structs and funcs *
87 ***********************************************/
88 /** ZSTD_hufCTablesMetadata_t :
89 * Stores Literals Block Type for a super-block in hType, and
90 * huffman tree description in hufDesBuffer.
91 * hufDesSize refers to the size of huffman tree description in bytes.
92 * This metadata is populated in ZSTD_buildBlockEntropyStats_literals() */
94 symbolEncodingType_e hType;
95 BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
97 } ZSTD_hufCTablesMetadata_t;
99 /** ZSTD_fseCTablesMetadata_t :
100 * Stores symbol compression modes for a super-block in {ll, ol, ml}Type, and
101 * fse tables in fseTablesBuffer.
102 * fseTablesSize refers to the size of fse tables in bytes.
103 * This metadata is populated in ZSTD_buildBlockEntropyStats_sequences() */
105 symbolEncodingType_e llType;
106 symbolEncodingType_e ofType;
107 symbolEncodingType_e mlType;
108 BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
109 size_t fseTablesSize;
110 size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_entropyCompressSeqStore_internal() */
111 } ZSTD_fseCTablesMetadata_t;
114 ZSTD_hufCTablesMetadata_t hufMetadata;
115 ZSTD_fseCTablesMetadata_t fseMetadata;
116 } ZSTD_entropyCTablesMetadata_t;
118 /** ZSTD_buildBlockEntropyStats() :
119 * Builds entropy for the block.
120 * @return : 0 on success or error code */
121 size_t ZSTD_buildBlockEntropyStats(
122 const seqStore_t* seqStorePtr,
123 const ZSTD_entropyCTables_t* prevEntropy,
124 ZSTD_entropyCTables_t* nextEntropy,
125 const ZSTD_CCtx_params* cctxParams,
126 ZSTD_entropyCTablesMetadata_t* entropyMetadata,
127 void* workspace, size_t wkspSize);
129 /*********************************
130 * Compression internals structs *
131 *********************************/
134 U32 off; /* Offset sumtype code for the match, using ZSTD_storeSeq() format */
135 U32 len; /* Raw length of match */
139 U32 offset; /* Offset of sequence */
140 U32 litLength; /* Length of literals prior to match */
141 U32 matchLength; /* Raw length of match */
145 rawSeq* seq; /* The start of the sequences */
146 size_t pos; /* The index in seq where reading stopped. pos <= size. */
147 size_t posInSequence; /* The position within the sequence at seq[pos] where reading
148 stopped. posInSequence <= seq[pos].litLength + seq[pos].matchLength */
149 size_t size; /* The number of sequences. <= capacity. */
150 size_t capacity; /* The capacity starting from `seq` pointer */
154 U32 idx; /* Index in array of ZSTD_Sequence */
155 U32 posInSequence; /* Position within sequence at idx */
156 size_t posInSrc; /* Number of bytes given by sequences provided so far */
157 } ZSTD_sequencePosition;
159 UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
166 U32 rep[ZSTD_REP_NUM];
169 typedef enum { zop_dynamic=0, zop_predef } ZSTD_OptPrice_e;
172 /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
173 unsigned* litFreq; /* table of literals statistics, of size 256 */
174 unsigned* litLengthFreq; /* table of litLength statistics, of size (MaxLL+1) */
175 unsigned* matchLengthFreq; /* table of matchLength statistics, of size (MaxML+1) */
176 unsigned* offCodeFreq; /* table of offCode statistics, of size (MaxOff+1) */
177 ZSTD_match_t* matchTable; /* list of found matches, of size ZSTD_OPT_NUM+1 */
178 ZSTD_optimal_t* priceTable; /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
180 U32 litSum; /* nb of literals */
181 U32 litLengthSum; /* nb of litLength codes */
182 U32 matchLengthSum; /* nb of matchLength codes */
183 U32 offCodeSum; /* nb of offset codes */
184 U32 litSumBasePrice; /* to compare to log2(litfreq) */
185 U32 litLengthSumBasePrice; /* to compare to log2(llfreq) */
186 U32 matchLengthSumBasePrice;/* to compare to log2(mlfreq) */
187 U32 offCodeSumBasePrice; /* to compare to log2(offreq) */
188 ZSTD_OptPrice_e priceType; /* prices can be determined dynamically, or follow a pre-defined cost structure */
189 const ZSTD_entropyCTables_t* symbolCosts; /* pre-calculated dictionary statistics */
190 ZSTD_paramSwitch_e literalCompressionMode;
194 ZSTD_entropyCTables_t entropy;
195 U32 rep[ZSTD_REP_NUM];
196 } ZSTD_compressedBlockState_t;
199 BYTE const* nextSrc; /* next block here to continue on current prefix */
200 BYTE const* base; /* All regular indexes relative to this position */
201 BYTE const* dictBase; /* extDict indexes relative to this position */
202 U32 dictLimit; /* below that point, need extDict */
203 U32 lowLimit; /* below that point, no more valid data */
204 U32 nbOverflowCorrections; /* Number of times overflow correction has run since
205 * ZSTD_window_init(). Useful for debugging coredumps
206 * and for ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY.
210 #define ZSTD_WINDOW_START_INDEX 2
212 typedef struct ZSTD_matchState_t ZSTD_matchState_t;
214 #define ZSTD_ROW_HASH_CACHE_SIZE 8 /* Size of prefetching hash cache for row-based matchfinder */
216 struct ZSTD_matchState_t {
217 ZSTD_window_t window; /* State for window round buffer management */
218 U32 loadedDictEnd; /* index of end of dictionary, within context's referential.
219 * When loadedDictEnd != 0, a dictionary is in use, and still valid.
220 * This relies on a mechanism to set loadedDictEnd=0 when dictionary is no longer within distance.
221 * Such mechanism is provided within ZSTD_window_enforceMaxDist() and ZSTD_checkDictValidity().
222 * When dict referential is copied into active context (i.e. not attached),
223 * loadedDictEnd == dictSize, since referential starts from zero.
225 U32 nextToUpdate; /* index from which to continue table update */
226 U32 hashLog3; /* dispatch table for matches of len==3 : larger == faster, more memory */
228 U32 rowHashLog; /* For row-based matchfinder: Hashlog based on nb of rows in the hashTable.*/
229 BYTE* tagTable; /* For row-based matchFinder: A row-based table containing the hashes and head index. */
230 U32 hashCache[ZSTD_ROW_HASH_CACHE_SIZE]; /* For row-based matchFinder: a cache of hashes to improve speed */
231 U64 hashSalt; /* For row-based matchFinder: salts the hash for re-use of tag table */
232 U32 hashSaltEntropy; /* For row-based matchFinder: collects entropy for salt generation */
238 U32 forceNonContiguous; /* Non-zero if we should force non-contiguous load for the next window update. */
240 int dedicatedDictSearch; /* Indicates whether this matchState is using the
241 * dedicated dictionary search structure.
243 optState_t opt; /* optimal parser state */
244 const ZSTD_matchState_t* dictMatchState;
245 ZSTD_compressionParameters cParams;
246 const rawSeqStore_t* ldmSeqStore;
248 /* Controls prefetching in some dictMatchState matchfinders.
249 * This behavior is controlled from the cctx ms.
250 * This parameter has no effect in the cdict ms. */
251 int prefetchCDictTables;
253 /* When == 0, lazy match finders insert every position.
254 * When != 0, lazy match finders only insert positions they search.
255 * This allows them to skip much faster over incompressible data,
256 * at a small cost to compression ratio.
262 ZSTD_compressedBlockState_t* prevCBlock;
263 ZSTD_compressedBlockState_t* nextCBlock;
264 ZSTD_matchState_t matchState;
277 } ldmMatchCandidate_t;
279 #define LDM_BATCH_SIZE 64
282 ZSTD_window_t window; /* State for the window round buffer management */
283 ldmEntry_t* hashTable;
285 BYTE* bucketOffsets; /* Next position in bucket to insert entry */
286 size_t splitIndices[LDM_BATCH_SIZE];
287 ldmMatchCandidate_t matchCandidates[LDM_BATCH_SIZE];
291 ZSTD_paramSwitch_e enableLdm; /* ZSTD_ps_enable to enable LDM. ZSTD_ps_auto by default */
292 U32 hashLog; /* Log size of hashTable */
293 U32 bucketSizeLog; /* Log bucket size for collision resolution, at most 8 */
294 U32 minMatchLength; /* Minimum match length */
295 U32 hashRateLog; /* Log number of entries to skip */
296 U32 windowLog; /* Window log for the LDM */
300 int collectSequences;
301 ZSTD_Sequence* seqStart;
306 struct ZSTD_CCtx_params_s {
307 ZSTD_format_e format;
308 ZSTD_compressionParameters cParams;
309 ZSTD_frameParameters fParams;
311 int compressionLevel;
312 int forceWindow; /* force back-references to respect limit of
313 * 1<<wLog, even for dictionary */
314 size_t targetCBlockSize; /* Tries to fit compressed block size to be around targetCBlockSize.
315 * No target when targetCBlockSize == 0.
316 * There is no guarantee on compressed block size */
317 int srcSizeHint; /* User's best guess of source size.
318 * Hint is not valid when srcSizeHint == 0.
319 * There is no guarantee that hint is close to actual source size */
321 ZSTD_dictAttachPref_e attachDictPref;
322 ZSTD_paramSwitch_e literalCompressionMode;
324 /* Multithreading: used to pass parameters to mtctx */
330 /* Long distance matching parameters */
331 ldmParams_t ldmParams;
333 /* Dedicated dict search algorithm trigger */
334 int enableDedicatedDictSearch;
336 /* Input/output buffer modes */
337 ZSTD_bufferMode_e inBufferMode;
338 ZSTD_bufferMode_e outBufferMode;
340 /* Sequence compression API */
341 ZSTD_sequenceFormat_e blockDelimiters;
342 int validateSequences;
344 /* Block splitting */
345 ZSTD_paramSwitch_e useBlockSplitter;
347 /* Param for deciding whether to use row-based matchfinder */
348 ZSTD_paramSwitch_e useRowMatchFinder;
350 /* Always load a dictionary in ext-dict mode (not prefix mode)? */
351 int deterministicRefPrefix;
353 /* Internal use, for createCCtxParams() and freeCCtxParams() only */
354 ZSTD_customMem customMem;
356 /* Controls prefetching in some dictMatchState matchfinders */
357 ZSTD_paramSwitch_e prefetchCDictTables;
359 /* Controls whether zstd will fall back to an internal matchfinder
360 * if the external matchfinder returns an error code. */
361 int enableMatchFinderFallback;
363 /* Indicates whether an external matchfinder has been referenced.
364 * Users can't set this externally.
365 * It is set internally in ZSTD_registerSequenceProducer(). */
366 int useSequenceProducer;
368 /* Adjust the max block size*/
371 /* Controls repcode search in external sequence parsing */
372 ZSTD_paramSwitch_e searchForExternalRepcodes;
373 }; /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
375 #define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
376 #define ENTROPY_WORKSPACE_SIZE (HUF_WORKSPACE_SIZE + COMPRESS_SEQUENCES_WORKSPACE_SIZE)
379 * Indicates whether this compression proceeds directly from user-provided
380 * source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
381 * whether the context needs to buffer the input/output (ZSTDb_buffered).
386 } ZSTD_buffered_policy_e;
389 * Struct that contains all elements of block splitter that should be allocated
392 #define ZSTD_MAX_NB_BLOCK_SPLITS 196
394 seqStore_t fullSeqStoreChunk;
395 seqStore_t firstHalfSeqStore;
396 seqStore_t secondHalfSeqStore;
397 seqStore_t currSeqStore;
398 seqStore_t nextSeqStore;
400 U32 partitions[ZSTD_MAX_NB_BLOCK_SPLITS];
401 ZSTD_entropyCTablesMetadata_t entropyMetadata;
402 } ZSTD_blockSplitCtx;
404 /* Context for block-level external matchfinder API */
407 ZSTD_sequenceProducer_F* mFinder;
408 ZSTD_Sequence* seqBuffer;
409 size_t seqBufferCapacity;
410 } ZSTD_externalMatchCtx;
413 ZSTD_compressionStage_e stage;
414 int cParamsChanged; /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
415 int bmi2; /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
416 ZSTD_CCtx_params requestedParams;
417 ZSTD_CCtx_params appliedParams;
418 ZSTD_CCtx_params simpleApiParams; /* Param storage used by the simple API - not sticky. Must only be used in top-level simple API functions for storage. */
420 size_t dictContentSize;
422 ZSTD_cwksp workspace; /* manages buffer for dynamic allocations */
424 unsigned long long pledgedSrcSizePlusOne; /* this way, 0 (default) == unknown */
425 unsigned long long consumedSrcSize;
426 unsigned long long producedCSize;
427 XXH64_state_t xxhState;
428 ZSTD_customMem customMem;
429 ZSTD_threadPool* pool;
431 SeqCollector seqCollector;
435 seqStore_t seqStore; /* sequences storage ptrs */
436 ldmState_t ldmState; /* long distance matching state */
437 rawSeq* ldmSequences; /* Storage for the ldm output sequences */
438 size_t maxNbLdmSequences;
439 rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
440 ZSTD_blockState_t blockState;
441 U32* entropyWorkspace; /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
443 /* Whether we are streaming or not */
444 ZSTD_buffered_policy_e bufferedPolicy;
454 size_t outBuffContentSize;
455 size_t outBuffFlushedSize;
456 ZSTD_cStreamStage streamStage;
459 /* Stable in/out buffer verification */
460 ZSTD_inBuffer expectedInBuffer;
461 size_t stableIn_notConsumed; /* nb bytes within stable input buffer that are said to be consumed but are not */
462 size_t expectedOutBufferSize;
465 ZSTD_localDict localDict;
466 const ZSTD_CDict* cdict;
467 ZSTD_prefixDict prefixDict; /* single-usage dictionary */
469 /* Multi-threading */
470 #ifdef ZSTD_MULTITHREAD
476 ZSTD_TraceCtx traceCtx;
479 /* Workspace for block splitter */
480 ZSTD_blockSplitCtx blockSplitCtx;
482 /* Workspace for external matchfinder */
483 ZSTD_externalMatchCtx externalMatchCtx;
486 typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
487 typedef enum { ZSTD_tfp_forCCtx, ZSTD_tfp_forCDict } ZSTD_tableFillPurpose_e;
492 ZSTD_dictMatchState = 2,
493 ZSTD_dedicatedDictSearch = 3
497 ZSTD_cpm_noAttachDict = 0, /* Compression with ZSTD_noDict or ZSTD_extDict.
498 * In this mode we use both the srcSize and the dictSize
499 * when selecting and adjusting parameters.
501 ZSTD_cpm_attachDict = 1, /* Compression with ZSTD_dictMatchState or ZSTD_dedicatedDictSearch.
502 * In this mode we only take the srcSize into account when selecting
503 * and adjusting parameters.
505 ZSTD_cpm_createCDict = 2, /* Creating a CDict.
506 * In this mode we take both the source size and the dictionary size
507 * into account when selecting and adjusting the parameters.
509 ZSTD_cpm_unknown = 3 /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
510 * We don't know what these parameters are for. We default to the legacy
511 * behavior of taking both the source size and the dict size into account
512 * when selecting and adjusting parameters.
516 typedef size_t (*ZSTD_blockCompressor) (
517 ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
518 void const* src, size_t srcSize);
519 ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_paramSwitch_e rowMatchfinderMode, ZSTD_dictMode_e dictMode);
522 MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
524 static const BYTE LL_Code[64] = { 0, 1, 2, 3, 4, 5, 6, 7,
525 8, 9, 10, 11, 12, 13, 14, 15,
526 16, 16, 17, 17, 18, 18, 19, 19,
527 20, 20, 20, 20, 21, 21, 21, 21,
528 22, 22, 22, 22, 22, 22, 22, 22,
529 23, 23, 23, 23, 23, 23, 23, 23,
530 24, 24, 24, 24, 24, 24, 24, 24,
531 24, 24, 24, 24, 24, 24, 24, 24 };
532 static const U32 LL_deltaCode = 19;
533 return (litLength > 63) ? ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
537 * note : mlBase = matchLength - MINMATCH;
538 * because it's the format it's stored in seqStore->sequences */
539 MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
541 static const BYTE ML_Code[128] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
542 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
543 32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
544 38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
545 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
546 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
547 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
548 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
549 static const U32 ML_deltaCode = 36;
550 return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
553 /* ZSTD_cParam_withinBounds:
554 * @return 1 if value is within cParam bounds,
556 MEM_STATIC int ZSTD_cParam_withinBounds(ZSTD_cParameter cParam, int value)
558 ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
559 if (ZSTD_isError(bounds.error)) return 0;
560 if (value < bounds.lowerBound) return 0;
561 if (value > bounds.upperBound) return 0;
565 /* ZSTD_noCompressBlock() :
566 * Writes uncompressed block to dst buffer from given src.
567 * Returns the size of the block */
569 ZSTD_noCompressBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
571 U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
572 DEBUGLOG(5, "ZSTD_noCompressBlock (srcSize=%zu, dstCapacity=%zu)", srcSize, dstCapacity);
573 RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
574 dstSize_tooSmall, "dst buf too small for uncompressed block");
575 MEM_writeLE24(dst, cBlockHeader24);
576 ZSTD_memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
577 return ZSTD_blockHeaderSize + srcSize;
581 ZSTD_rleCompressBlock(void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
583 BYTE* const op = (BYTE*)dst;
584 U32 const cBlockHeader = lastBlock + (((U32)bt_rle)<<1) + (U32)(srcSize << 3);
585 RETURN_ERROR_IF(dstCapacity < 4, dstSize_tooSmall, "");
586 MEM_writeLE24(op, cBlockHeader);
593 * minimum compression required
594 * to generate a compress block or a compressed literals section.
595 * note : use same formula for both situations */
596 MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
598 U32 const minlog = (strat>=ZSTD_btultra) ? (U32)(strat) - 1 : 6;
599 ZSTD_STATIC_ASSERT(ZSTD_btultra == 8);
600 assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, (int)strat));
601 return (srcSize >> minlog) + 2;
604 MEM_STATIC int ZSTD_literalsCompressionIsDisabled(const ZSTD_CCtx_params* cctxParams)
606 switch (cctxParams->literalCompressionMode) {
609 case ZSTD_ps_disable:
612 assert(0 /* impossible: pre-validated */);
615 return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
619 /*! ZSTD_safecopyLiterals() :
620 * memcpy() function that won't read beyond more than WILDCOPY_OVERLENGTH bytes past ilimit_w.
621 * Only called when the sequence ends past ilimit_w, so it only needs to be optimized for single
625 ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE const* ilimit_w)
627 assert(iend > ilimit_w);
628 if (ip <= ilimit_w) {
629 ZSTD_wildcopy(op, ip, ilimit_w - ip, ZSTD_no_overlap);
633 while (ip < iend) *op++ = *ip++;
637 #define REPCODE1_TO_OFFBASE REPCODE_TO_OFFBASE(1)
638 #define REPCODE2_TO_OFFBASE REPCODE_TO_OFFBASE(2)
639 #define REPCODE3_TO_OFFBASE REPCODE_TO_OFFBASE(3)
640 #define REPCODE_TO_OFFBASE(r) (assert((r)>=1), assert((r)<=ZSTD_REP_NUM), (r)) /* accepts IDs 1,2,3 */
641 #define OFFSET_TO_OFFBASE(o) (assert((o)>0), o + ZSTD_REP_NUM)
642 #define OFFBASE_IS_OFFSET(o) ((o) > ZSTD_REP_NUM)
643 #define OFFBASE_IS_REPCODE(o) ( 1 <= (o) && (o) <= ZSTD_REP_NUM)
644 #define OFFBASE_TO_OFFSET(o) (assert(OFFBASE_IS_OFFSET(o)), (o) - ZSTD_REP_NUM)
645 #define OFFBASE_TO_REPCODE(o) (assert(OFFBASE_IS_REPCODE(o)), (o)) /* returns ID 1,2,3 */
647 /*! ZSTD_storeSeq() :
648 * Store a sequence (litlen, litPtr, offBase and matchLength) into seqStore_t.
649 * @offBase : Users should employ macros REPCODE_TO_OFFBASE() and OFFSET_TO_OFFBASE().
650 * @matchLength : must be >= MINMATCH
651 * Allowed to over-read literals up to litLimit.
653 HINT_INLINE UNUSED_ATTR void
654 ZSTD_storeSeq(seqStore_t* seqStorePtr,
655 size_t litLength, const BYTE* literals, const BYTE* litLimit,
659 BYTE const* const litLimit_w = litLimit - WILDCOPY_OVERLENGTH;
660 BYTE const* const litEnd = literals + litLength;
661 #if defined(DEBUGLEVEL) && (DEBUGLEVEL >= 6)
662 static const BYTE* g_start = NULL;
663 if (g_start==NULL) g_start = (const BYTE*)literals; /* note : index only works for compression within a single segment */
664 { U32 const pos = (U32)((const BYTE*)literals - g_start);
665 DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offBase%7u",
666 pos, (U32)litLength, (U32)matchLength, (U32)offBase);
669 assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
671 assert(seqStorePtr->maxNbLit <= 128 KB);
672 assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
673 assert(literals + litLength <= litLimit);
674 if (litEnd <= litLimit_w) {
675 /* Common case we can use wildcopy.
676 * First copy 16 bytes, because literals are likely short.
678 ZSTD_STATIC_ASSERT(WILDCOPY_OVERLENGTH >= 16);
679 ZSTD_copy16(seqStorePtr->lit, literals);
680 if (litLength > 16) {
681 ZSTD_wildcopy(seqStorePtr->lit+16, literals+16, (ptrdiff_t)litLength-16, ZSTD_no_overlap);
684 ZSTD_safecopyLiterals(seqStorePtr->lit, literals, litEnd, litLimit_w);
686 seqStorePtr->lit += litLength;
689 if (litLength>0xFFFF) {
690 assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
691 seqStorePtr->longLengthType = ZSTD_llt_literalLength;
692 seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
694 seqStorePtr->sequences[0].litLength = (U16)litLength;
697 seqStorePtr->sequences[0].offBase = offBase;
700 assert(matchLength >= MINMATCH);
701 { size_t const mlBase = matchLength - MINMATCH;
703 assert(seqStorePtr->longLengthType == ZSTD_llt_none); /* there can only be a single long length */
704 seqStorePtr->longLengthType = ZSTD_llt_matchLength;
705 seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
707 seqStorePtr->sequences[0].mlBase = (U16)mlBase;
710 seqStorePtr->sequences++;
713 /* ZSTD_updateRep() :
714 * updates in-place @rep (array of repeat offsets)
715 * @offBase : sum-type, using numeric representation of ZSTD_storeSeq()
718 ZSTD_updateRep(U32 rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
720 if (OFFBASE_IS_OFFSET(offBase)) { /* full offset */
723 rep[0] = OFFBASE_TO_OFFSET(offBase);
724 } else { /* repcode */
725 U32 const repCode = OFFBASE_TO_REPCODE(offBase) - 1 + ll0;
726 if (repCode > 0) { /* note : if repCode==0, no change */
727 U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
728 rep[2] = (repCode >= 2) ? rep[1] : rep[2];
730 rep[0] = currentOffset;
731 } else { /* repCode == 0 */
737 typedef struct repcodes_s {
741 MEM_STATIC repcodes_t
742 ZSTD_newRep(U32 const rep[ZSTD_REP_NUM], U32 const offBase, U32 const ll0)
745 ZSTD_memcpy(&newReps, rep, sizeof(newReps));
746 ZSTD_updateRep(newReps.rep, offBase, ll0);
751 /*-*************************************
752 * Match length counter
753 ***************************************/
754 MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
756 const BYTE* const pStart = pIn;
757 const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
759 if (pIn < pInLoopLimit) {
760 { size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
761 if (diff) return ZSTD_NbCommonBytes(diff); }
762 pIn+=sizeof(size_t); pMatch+=sizeof(size_t);
763 while (pIn < pInLoopLimit) {
764 size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
765 if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
766 pIn += ZSTD_NbCommonBytes(diff);
767 return (size_t)(pIn - pStart);
769 if (MEM_64bits() && (pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
770 if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
771 if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
772 return (size_t)(pIn - pStart);
775 /** ZSTD_count_2segments() :
776 * can count match length with `ip` & `match` in 2 different segments.
777 * convention : on reaching mEnd, match count continue starting from iStart
780 ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
781 const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
783 const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
784 size_t const matchLength = ZSTD_count(ip, match, vEnd);
785 if (match + matchLength != mEnd) return matchLength;
786 DEBUGLOG(7, "ZSTD_count_2segments: found a 2-parts match (current length==%zu)", matchLength);
787 DEBUGLOG(7, "distance from match beginning to end dictionary = %zi", mEnd - match);
788 DEBUGLOG(7, "distance from current pos to end buffer = %zi", iEnd - ip);
789 DEBUGLOG(7, "next byte : ip==%02X, istart==%02X", ip[matchLength], *iStart);
790 DEBUGLOG(7, "final match length = %zu", matchLength + ZSTD_count(ip+matchLength, iStart, iEnd));
791 return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
795 /*-*************************************
797 ***************************************/
798 static const U32 prime3bytes = 506832829U;
799 static U32 ZSTD_hash3(U32 u, U32 h, U32 s) { assert(h <= 32); return (((u << (32-24)) * prime3bytes) ^ s) >> (32-h) ; }
800 MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h, 0); } /* only in zstd_opt.h */
801 MEM_STATIC size_t ZSTD_hash3PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash3(MEM_readLE32(ptr), h, s); }
803 static const U32 prime4bytes = 2654435761U;
804 static U32 ZSTD_hash4(U32 u, U32 h, U32 s) { assert(h <= 32); return ((u * prime4bytes) ^ s) >> (32-h) ; }
805 static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_readLE32(ptr), h, 0); }
806 static size_t ZSTD_hash4PtrS(const void* ptr, U32 h, U32 s) { return ZSTD_hash4(MEM_readLE32(ptr), h, s); }
808 static const U64 prime5bytes = 889523592379ULL;
809 static size_t ZSTD_hash5(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u << (64-40)) * prime5bytes) ^ s) >> (64-h)) ; }
810 static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h, 0); }
811 static size_t ZSTD_hash5PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash5(MEM_readLE64(p), h, s); }
813 static const U64 prime6bytes = 227718039650203ULL;
814 static size_t ZSTD_hash6(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u << (64-48)) * prime6bytes) ^ s) >> (64-h)) ; }
815 static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h, 0); }
816 static size_t ZSTD_hash6PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash6(MEM_readLE64(p), h, s); }
818 static const U64 prime7bytes = 58295818150454627ULL;
819 static size_t ZSTD_hash7(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u << (64-56)) * prime7bytes) ^ s) >> (64-h)) ; }
820 static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h, 0); }
821 static size_t ZSTD_hash7PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash7(MEM_readLE64(p), h, s); }
823 static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
824 static size_t ZSTD_hash8(U64 u, U32 h, U64 s) { assert(h <= 64); return (size_t)((((u) * prime8bytes) ^ s) >> (64-h)) ; }
825 static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h, 0); }
826 static size_t ZSTD_hash8PtrS(const void* p, U32 h, U64 s) { return ZSTD_hash8(MEM_readLE64(p), h, s); }
829 MEM_STATIC FORCE_INLINE_ATTR
830 size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
832 /* Although some of these hashes do support hBits up to 64, some do not.
833 * To be on the safe side, always avoid hBits > 32. */
839 case 4: return ZSTD_hash4Ptr(p, hBits);
840 case 5: return ZSTD_hash5Ptr(p, hBits);
841 case 6: return ZSTD_hash6Ptr(p, hBits);
842 case 7: return ZSTD_hash7Ptr(p, hBits);
843 case 8: return ZSTD_hash8Ptr(p, hBits);
847 MEM_STATIC FORCE_INLINE_ATTR
848 size_t ZSTD_hashPtrSalted(const void* p, U32 hBits, U32 mls, const U64 hashSalt) {
849 /* Although some of these hashes do support hBits up to 64, some do not.
850 * To be on the safe side, always avoid hBits > 32. */
856 case 4: return ZSTD_hash4PtrS(p, hBits, (U32)hashSalt);
857 case 5: return ZSTD_hash5PtrS(p, hBits, hashSalt);
858 case 6: return ZSTD_hash6PtrS(p, hBits, hashSalt);
859 case 7: return ZSTD_hash7PtrS(p, hBits, hashSalt);
860 case 8: return ZSTD_hash8PtrS(p, hBits, hashSalt);
866 * Return base^exponent.
868 static U64 ZSTD_ipow(U64 base, U64 exponent)
872 if (exponent & 1) power *= base;
879 #define ZSTD_ROLL_HASH_CHAR_OFFSET 10
881 /** ZSTD_rollingHash_append() :
882 * Add the buffer to the hash value.
884 static U64 ZSTD_rollingHash_append(U64 hash, void const* buf, size_t size)
886 BYTE const* istart = (BYTE const*)buf;
888 for (pos = 0; pos < size; ++pos) {
890 hash += istart[pos] + ZSTD_ROLL_HASH_CHAR_OFFSET;
895 /** ZSTD_rollingHash_compute() :
896 * Compute the rolling hash value of the buffer.
898 MEM_STATIC U64 ZSTD_rollingHash_compute(void const* buf, size_t size)
900 return ZSTD_rollingHash_append(0, buf, size);
903 /** ZSTD_rollingHash_primePower() :
904 * Compute the primePower to be passed to ZSTD_rollingHash_rotate() for a hash
905 * over a window of length bytes.
907 MEM_STATIC U64 ZSTD_rollingHash_primePower(U32 length)
909 return ZSTD_ipow(prime8bytes, length - 1);
912 /** ZSTD_rollingHash_rotate() :
913 * Rotate the rolling hash by one byte.
915 MEM_STATIC U64 ZSTD_rollingHash_rotate(U64 hash, BYTE toRemove, BYTE toAdd, U64 primePower)
917 hash -= (toRemove + ZSTD_ROLL_HASH_CHAR_OFFSET) * primePower;
919 hash += toAdd + ZSTD_ROLL_HASH_CHAR_OFFSET;
923 /*-*************************************
924 * Round buffer management
925 ***************************************/
926 #if (ZSTD_WINDOWLOG_MAX_64 > 31)
927 # error "ZSTD_WINDOWLOG_MAX is too large : would overflow ZSTD_CURRENT_MAX"
929 /* Max current allowed */
930 #define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
931 /* Maximum chunk size before overflow correction needs to be called again */
932 #define ZSTD_CHUNKSIZE_MAX \
933 ( ((U32)-1) /* Maximum ending current index */ \
934 - ZSTD_CURRENT_MAX) /* Maximum beginning lowLimit */
937 * ZSTD_window_clear():
938 * Clears the window containing the history by simply setting it to empty.
940 MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
942 size_t const endT = (size_t)(window->nextSrc - window->base);
943 U32 const end = (U32)endT;
945 window->lowLimit = end;
946 window->dictLimit = end;
949 MEM_STATIC U32 ZSTD_window_isEmpty(ZSTD_window_t const window)
951 return window.dictLimit == ZSTD_WINDOW_START_INDEX &&
952 window.lowLimit == ZSTD_WINDOW_START_INDEX &&
953 (window.nextSrc - window.base) == ZSTD_WINDOW_START_INDEX;
957 * ZSTD_window_hasExtDict():
958 * Returns non-zero if the window has a non-empty extDict.
960 MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
962 return window.lowLimit < window.dictLimit;
966 * ZSTD_matchState_dictMode():
967 * Inspects the provided matchState and figures out what dictMode should be
968 * passed to the compressor.
970 MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
972 return ZSTD_window_hasExtDict(ms->window) ?
974 ms->dictMatchState != NULL ?
975 (ms->dictMatchState->dedicatedDictSearch ? ZSTD_dedicatedDictSearch : ZSTD_dictMatchState) :
979 /* Defining this macro to non-zero tells zstd to run the overflow correction
980 * code much more frequently. This is very inefficient, and should only be
981 * used for tests and fuzzers.
983 #ifndef ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY
984 # ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
985 # define ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY 1
987 # define ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY 0
992 * ZSTD_window_canOverflowCorrect():
993 * Returns non-zero if the indices are large enough for overflow correction
994 * to work correctly without impacting compression ratio.
996 MEM_STATIC U32 ZSTD_window_canOverflowCorrect(ZSTD_window_t const window,
1002 U32 const cycleSize = 1u << cycleLog;
1003 U32 const curr = (U32)((BYTE const*)src - window.base);
1004 U32 const minIndexToOverflowCorrect = cycleSize
1005 + MAX(maxDist, cycleSize)
1006 + ZSTD_WINDOW_START_INDEX;
1008 /* Adjust the min index to backoff the overflow correction frequency,
1009 * so we don't waste too much CPU in overflow correction. If this
1010 * computation overflows we don't really care, we just need to make
1011 * sure it is at least minIndexToOverflowCorrect.
1013 U32 const adjustment = window.nbOverflowCorrections + 1;
1014 U32 const adjustedIndex = MAX(minIndexToOverflowCorrect * adjustment,
1015 minIndexToOverflowCorrect);
1016 U32 const indexLargeEnough = curr > adjustedIndex;
1018 /* Only overflow correct early if the dictionary is invalidated already,
1019 * so we don't hurt compression ratio.
1021 U32 const dictionaryInvalidated = curr > maxDist + loadedDictEnd;
1023 return indexLargeEnough && dictionaryInvalidated;
1027 * ZSTD_window_needOverflowCorrection():
1028 * Returns non-zero if the indices are getting too large and need overflow
1031 MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
1038 U32 const curr = (U32)((BYTE const*)srcEnd - window.base);
1039 if (ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY) {
1040 if (ZSTD_window_canOverflowCorrect(window, cycleLog, maxDist, loadedDictEnd, src)) {
1044 return curr > ZSTD_CURRENT_MAX;
1048 * ZSTD_window_correctOverflow():
1049 * Reduces the indices to protect from index overflow.
1050 * Returns the correction made to the indices, which must be applied to every
1053 * The least significant cycleLog bits of the indices must remain the same,
1054 * which may be 0. Every index up to maxDist in the past must be valid.
1056 MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
1057 U32 maxDist, void const* src)
1059 /* preemptive overflow correction:
1060 * 1. correction is large enough:
1061 * lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog
1062 * 1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
1064 * current - newCurrent
1065 * > (3<<29 + 1<<windowLog) - (1<<windowLog + 1<<chainLog)
1066 * > (3<<29) - (1<<chainLog)
1067 * > (3<<29) - (1<<30) (NOTE: chainLog <= 30)
1070 * 2. (ip+ZSTD_CHUNKSIZE_MAX - cctx->base) doesn't overflow:
1071 * After correction, current is less than (1<<chainLog + 1<<windowLog).
1072 * In 64-bit mode we are safe, because we have 64-bit ptrdiff_t.
1073 * In 32-bit mode we are safe, because (chainLog <= 29), so
1074 * ip+ZSTD_CHUNKSIZE_MAX - cctx->base < 1<<32.
1075 * 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
1076 * windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
1078 U32 const cycleSize = 1u << cycleLog;
1079 U32 const cycleMask = cycleSize - 1;
1080 U32 const curr = (U32)((BYTE const*)src - window->base);
1081 U32 const currentCycle = curr & cycleMask;
1082 /* Ensure newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX. */
1083 U32 const currentCycleCorrection = currentCycle < ZSTD_WINDOW_START_INDEX
1084 ? MAX(cycleSize, ZSTD_WINDOW_START_INDEX)
1086 U32 const newCurrent = currentCycle
1087 + currentCycleCorrection
1088 + MAX(maxDist, cycleSize);
1089 U32 const correction = curr - newCurrent;
1090 /* maxDist must be a power of two so that:
1091 * (newCurrent & cycleMask) == (curr & cycleMask)
1092 * This is required to not corrupt the chains / binary tree.
1094 assert((maxDist & (maxDist - 1)) == 0);
1095 assert((curr & cycleMask) == (newCurrent & cycleMask));
1096 assert(curr > newCurrent);
1097 if (!ZSTD_WINDOW_OVERFLOW_CORRECT_FREQUENTLY) {
1098 /* Loose bound, should be around 1<<29 (see above) */
1099 assert(correction > 1<<28);
1102 window->base += correction;
1103 window->dictBase += correction;
1104 if (window->lowLimit < correction + ZSTD_WINDOW_START_INDEX) {
1105 window->lowLimit = ZSTD_WINDOW_START_INDEX;
1107 window->lowLimit -= correction;
1109 if (window->dictLimit < correction + ZSTD_WINDOW_START_INDEX) {
1110 window->dictLimit = ZSTD_WINDOW_START_INDEX;
1112 window->dictLimit -= correction;
1115 /* Ensure we can still reference the full window. */
1116 assert(newCurrent >= maxDist);
1117 assert(newCurrent - maxDist >= ZSTD_WINDOW_START_INDEX);
1118 /* Ensure that lowLimit and dictLimit didn't underflow. */
1119 assert(window->lowLimit <= newCurrent);
1120 assert(window->dictLimit <= newCurrent);
1122 ++window->nbOverflowCorrections;
1124 DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction,
1130 * ZSTD_window_enforceMaxDist():
1131 * Updates lowLimit so that:
1132 * (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
1134 * It ensures index is valid as long as index >= lowLimit.
1135 * This must be called before a block compression call.
1137 * loadedDictEnd is only defined if a dictionary is in use for current compression.
1138 * As the name implies, loadedDictEnd represents the index at end of dictionary.
1139 * The value lies within context's referential, it can be directly compared to blockEndIdx.
1141 * If loadedDictEndPtr is NULL, no dictionary is in use, and we use loadedDictEnd == 0.
1142 * If loadedDictEndPtr is not NULL, we set it to zero after updating lowLimit.
1143 * This is because dictionaries are allowed to be referenced fully
1144 * as long as the last byte of the dictionary is in the window.
1145 * Once input has progressed beyond window size, dictionary cannot be referenced anymore.
1147 * In normal dict mode, the dictionary lies between lowLimit and dictLimit.
1148 * In dictMatchState mode, lowLimit and dictLimit are the same,
1149 * and the dictionary is below them.
1150 * forceWindow and dictMatchState are therefore incompatible.
1153 ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
1154 const void* blockEnd,
1156 U32* loadedDictEndPtr,
1157 const ZSTD_matchState_t** dictMatchStatePtr)
1159 U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
1160 U32 const loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
1161 DEBUGLOG(5, "ZSTD_window_enforceMaxDist: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
1162 (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
1164 /* - When there is no dictionary : loadedDictEnd == 0.
1165 In which case, the test (blockEndIdx > maxDist) is merely to avoid
1166 overflowing next operation `newLowLimit = blockEndIdx - maxDist`.
1167 - When there is a standard dictionary :
1168 Index referential is copied from the dictionary,
1169 which means it starts from 0.
1170 In which case, loadedDictEnd == dictSize,
1171 and it makes sense to compare `blockEndIdx > maxDist + dictSize`
1172 since `blockEndIdx` also starts from zero.
1173 - When there is an attached dictionary :
1174 loadedDictEnd is expressed within the referential of the context,
1175 so it can be directly compared against blockEndIdx.
1177 if (blockEndIdx > maxDist + loadedDictEnd) {
1178 U32 const newLowLimit = blockEndIdx - maxDist;
1179 if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
1180 if (window->dictLimit < window->lowLimit) {
1181 DEBUGLOG(5, "Update dictLimit to match lowLimit, from %u to %u",
1182 (unsigned)window->dictLimit, (unsigned)window->lowLimit);
1183 window->dictLimit = window->lowLimit;
1185 /* On reaching window size, dictionaries are invalidated */
1186 if (loadedDictEndPtr) *loadedDictEndPtr = 0;
1187 if (dictMatchStatePtr) *dictMatchStatePtr = NULL;
1191 /* Similar to ZSTD_window_enforceMaxDist(),
1192 * but only invalidates dictionary
1193 * when input progresses beyond window size.
1194 * assumption : loadedDictEndPtr and dictMatchStatePtr are valid (non NULL)
1195 * loadedDictEnd uses same referential as window->base
1196 * maxDist is the window size */
1198 ZSTD_checkDictValidity(const ZSTD_window_t* window,
1199 const void* blockEnd,
1201 U32* loadedDictEndPtr,
1202 const ZSTD_matchState_t** dictMatchStatePtr)
1204 assert(loadedDictEndPtr != NULL);
1205 assert(dictMatchStatePtr != NULL);
1206 { U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
1207 U32 const loadedDictEnd = *loadedDictEndPtr;
1208 DEBUGLOG(5, "ZSTD_checkDictValidity: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
1209 (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
1210 assert(blockEndIdx >= loadedDictEnd);
1212 if (blockEndIdx > loadedDictEnd + maxDist || loadedDictEnd != window->dictLimit) {
1213 /* On reaching window size, dictionaries are invalidated.
1214 * For simplification, if window size is reached anywhere within next block,
1215 * the dictionary is invalidated for the full block.
1217 * We also have to invalidate the dictionary if ZSTD_window_update() has detected
1218 * non-contiguous segments, which means that loadedDictEnd != window->dictLimit.
1219 * loadedDictEnd may be 0, if forceWindow is true, but in that case we never use
1220 * dictMatchState, so setting it to NULL is not a problem.
1222 DEBUGLOG(6, "invalidating dictionary for current block (distance > windowSize)");
1223 *loadedDictEndPtr = 0;
1224 *dictMatchStatePtr = NULL;
1226 if (*loadedDictEndPtr != 0) {
1227 DEBUGLOG(6, "dictionary considered valid for current block");
1231 MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
1232 ZSTD_memset(window, 0, sizeof(*window));
1233 window->base = (BYTE const*)" ";
1234 window->dictBase = (BYTE const*)" ";
1235 ZSTD_STATIC_ASSERT(ZSTD_DUBT_UNSORTED_MARK < ZSTD_WINDOW_START_INDEX); /* Start above ZSTD_DUBT_UNSORTED_MARK */
1236 window->dictLimit = ZSTD_WINDOW_START_INDEX; /* start from >0, so that 1st position is valid */
1237 window->lowLimit = ZSTD_WINDOW_START_INDEX; /* it ensures first and later CCtx usages compress the same */
1238 window->nextSrc = window->base + ZSTD_WINDOW_START_INDEX; /* see issue #1241 */
1239 window->nbOverflowCorrections = 0;
1243 * ZSTD_window_update():
1244 * Updates the window by appending [src, src + srcSize) to the window.
1245 * If it is not contiguous, the current prefix becomes the extDict, and we
1246 * forget about the extDict. Handles overlap of the prefix and extDict.
1247 * Returns non-zero if the segment is contiguous.
1249 MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
1250 void const* src, size_t srcSize,
1251 int forceNonContiguous)
1253 BYTE const* const ip = (BYTE const*)src;
1255 DEBUGLOG(5, "ZSTD_window_update");
1258 assert(window->base != NULL);
1259 assert(window->dictBase != NULL);
1260 /* Check if blocks follow each other */
1261 if (src != window->nextSrc || forceNonContiguous) {
1262 /* not contiguous */
1263 size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
1264 DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u", window->dictLimit);
1265 window->lowLimit = window->dictLimit;
1266 assert(distanceFromBase == (size_t)(U32)distanceFromBase); /* should never overflow */
1267 window->dictLimit = (U32)distanceFromBase;
1268 window->dictBase = window->base;
1269 window->base = ip - distanceFromBase;
1270 /* ms->nextToUpdate = window->dictLimit; */
1271 if (window->dictLimit - window->lowLimit < HASH_READ_SIZE) window->lowLimit = window->dictLimit; /* too small extDict */
1274 window->nextSrc = ip + srcSize;
1275 /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
1276 if ( (ip+srcSize > window->dictBase + window->lowLimit)
1277 & (ip < window->dictBase + window->dictLimit)) {
1278 ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
1279 U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
1280 window->lowLimit = lowLimitMax;
1281 DEBUGLOG(5, "Overlapping extDict and input : new lowLimit = %u", window->lowLimit);
1287 * Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
1289 MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
1291 U32 const maxDistance = 1U << windowLog;
1292 U32 const lowestValid = ms->window.lowLimit;
1293 U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
1294 U32 const isDictionary = (ms->loadedDictEnd != 0);
1295 /* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
1296 * is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
1297 * valid for the entire block. So this check is sufficient to find the lowest valid match index.
1299 U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
1304 * Returns the lowest allowed match index in the prefix.
1306 MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
1308 U32 const maxDistance = 1U << windowLog;
1309 U32 const lowestValid = ms->window.dictLimit;
1310 U32 const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
1311 U32 const isDictionary = (ms->loadedDictEnd != 0);
1312 /* When computing the lowest prefix index we need to take the dictionary into account to handle
1313 * the edge case where the dictionary and the source are contiguous in memory.
1315 U32 const matchLowest = isDictionary ? lowestValid : withinWindow;
1321 /* debug functions */
1324 MEM_STATIC double ZSTD_fWeight(U32 rawStat)
1326 U32 const fp_accuracy = 8;
1327 U32 const fp_multiplier = (1 << fp_accuracy);
1328 U32 const newStat = rawStat + 1;
1329 U32 const hb = ZSTD_highbit32(newStat);
1330 U32 const BWeight = hb * fp_multiplier;
1331 U32 const FWeight = (newStat << fp_accuracy) >> hb;
1332 U32 const weight = BWeight + FWeight;
1333 assert(hb + fp_accuracy < 31);
1334 return (double)weight / fp_multiplier;
1337 /* display a table content,
1338 * listing each element, its frequency, and its predicted bit cost */
1339 MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
1342 for (u=0, sum=0; u<=max; u++) sum += table[u];
1343 DEBUGLOG(2, "total nb elts: %u", sum);
1344 for (u=0; u<=max; u++) {
1345 DEBUGLOG(2, "%2u: %5u (%.2f)",
1346 u, table[u], ZSTD_fWeight(sum) - ZSTD_fWeight(table[u]) );
1354 /* Normally, zstd matchfinders follow this flow:
1355 * 1. Compute hash at ip
1356 * 2. Load index from hashTable[hash]
1357 * 3. Check if *ip == *(base + index)
1358 * In dictionary compression, loading *(base + index) is often an L2 or even L3 miss.
1360 * Short cache is an optimization which allows us to avoid step 3 most of the time
1361 * when the data doesn't actually match. With short cache, the flow becomes:
1362 * 1. Compute (hash, currentTag) at ip. currentTag is an 8-bit independent hash at ip.
1363 * 2. Load (index, matchTag) from hashTable[hash]. See ZSTD_writeTaggedIndex to understand how this works.
1364 * 3. Only if currentTag == matchTag, check *ip == *(base + index). Otherwise, continue.
1366 * Currently, short cache is only implemented in CDict hashtables. Thus, its use is limited to
1367 * dictMatchState matchfinders.
1369 #define ZSTD_SHORT_CACHE_TAG_BITS 8
1370 #define ZSTD_SHORT_CACHE_TAG_MASK ((1u << ZSTD_SHORT_CACHE_TAG_BITS) - 1)
1372 /* Helper function for ZSTD_fillHashTable and ZSTD_fillDoubleHashTable.
1373 * Unpacks hashAndTag into (hash, tag), then packs (index, tag) into hashTable[hash]. */
1374 MEM_STATIC void ZSTD_writeTaggedIndex(U32* const hashTable, size_t hashAndTag, U32 index) {
1375 size_t const hash = hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
1376 U32 const tag = (U32)(hashAndTag & ZSTD_SHORT_CACHE_TAG_MASK);
1377 assert(index >> (32 - ZSTD_SHORT_CACHE_TAG_BITS) == 0);
1378 hashTable[hash] = (index << ZSTD_SHORT_CACHE_TAG_BITS) | tag;
1381 /* Helper function for short cache matchfinders.
1382 * Unpacks tag1 and tag2 from lower bits of packedTag1 and packedTag2, then checks if the tags match. */
1383 MEM_STATIC int ZSTD_comparePackedTags(size_t packedTag1, size_t packedTag2) {
1384 U32 const tag1 = packedTag1 & ZSTD_SHORT_CACHE_TAG_MASK;
1385 U32 const tag2 = packedTag2 & ZSTD_SHORT_CACHE_TAG_MASK;
1386 return tag1 == tag2;
1389 #if defined (__cplusplus)
1393 /* ===============================================================
1394 * Shared internal declarations
1395 * These prototypes may be called from sources not in lib/compress
1396 * =============================================================== */
1398 /* ZSTD_loadCEntropy() :
1399 * dict : must point at beginning of a valid zstd dictionary.
1400 * return : size of dictionary header (size of magic number + dict ID + entropy tables)
1401 * assumptions : magic number supposed already checked
1402 * and dictSize >= 8 */
1403 size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
1404 const void* const dict, size_t dictSize);
1406 void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
1408 /* ==============================================================
1409 * Private declarations
1410 * These prototypes shall only be called from within lib/compress
1411 * ============================================================== */
1413 /* ZSTD_getCParamsFromCCtxParams() :
1414 * cParams are built depending on compressionLevel, src size hints,
1415 * LDM and manually set compression parameters.
1416 * Note: srcSizeHint == 0 means 0!
1418 ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
1419 const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
1421 /*! ZSTD_initCStream_internal() :
1422 * Private use only. Init streaming operation.
1423 * expects params to be valid.
1424 * must receive dict, or cdict, or none, but not both.
1425 * @return : 0, or an error code */
1426 size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
1427 const void* dict, size_t dictSize,
1428 const ZSTD_CDict* cdict,
1429 const ZSTD_CCtx_params* params, unsigned long long pledgedSrcSize);
1431 void ZSTD_resetSeqStore(seqStore_t* ssPtr);
1433 /*! ZSTD_getCParamsFromCDict() :
1434 * as the name implies */
1435 ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
1437 /* ZSTD_compressBegin_advanced_internal() :
1438 * Private use only. To be called from zstdmt_compress.c. */
1439 size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
1440 const void* dict, size_t dictSize,
1441 ZSTD_dictContentType_e dictContentType,
1442 ZSTD_dictTableLoadMethod_e dtlm,
1443 const ZSTD_CDict* cdict,
1444 const ZSTD_CCtx_params* params,
1445 unsigned long long pledgedSrcSize);
1447 /* ZSTD_compress_advanced_internal() :
1448 * Private use only. To be called from zstdmt_compress.c. */
1449 size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
1450 void* dst, size_t dstCapacity,
1451 const void* src, size_t srcSize,
1452 const void* dict,size_t dictSize,
1453 const ZSTD_CCtx_params* params);
1456 /* ZSTD_writeLastEmptyBlock() :
1457 * output an empty Block with end-of-frame mark to complete a frame
1458 * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
1459 * or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
1461 size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
1464 /* ZSTD_referenceExternalSequences() :
1465 * Must be called before starting a compression operation.
1466 * seqs must parse a prefix of the source.
1467 * This cannot be used when long range matching is enabled.
1468 * Zstd will use these sequences, and pass the literals to a secondary block
1470 * @return : An error code on failure.
1471 * NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
1472 * access and data corruption.
1474 size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
1476 /** ZSTD_cycleLog() :
1477 * condition for correct operation : hashLog > 1 */
1478 U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat);
1480 /** ZSTD_CCtx_trace() :
1481 * Trace the end of a compression call.
1483 void ZSTD_CCtx_trace(ZSTD_CCtx* cctx, size_t extraCSize);
1485 /* Returns 0 on success, and a ZSTD_error otherwise. This function scans through an array of
1486 * ZSTD_Sequence, storing the sequences it finds, until it reaches a block delimiter.
1487 * Note that the block delimiter must include the last literals of the block.
1490 ZSTD_copySequencesToSeqStoreExplicitBlockDelim(ZSTD_CCtx* cctx,
1491 ZSTD_sequencePosition* seqPos,
1492 const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
1493 const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
1495 /* Returns the number of bytes to move the current read position back by.
1496 * Only non-zero if we ended up splitting a sequence.
1497 * Otherwise, it may return a ZSTD error if something went wrong.
1499 * This function will attempt to scan through blockSize bytes
1500 * represented by the sequences in @inSeqs,
1501 * storing any (partial) sequences.
1503 * Occasionally, we may want to change the actual number of bytes we consumed from inSeqs to
1504 * avoid splitting a match, or to avoid splitting a match such that it would produce a match
1505 * smaller than MINMATCH. In this case, we return the number of bytes that we didn't read from this block.
1508 ZSTD_copySequencesToSeqStoreNoBlockDelim(ZSTD_CCtx* cctx, ZSTD_sequencePosition* seqPos,
1509 const ZSTD_Sequence* const inSeqs, size_t inSeqsSize,
1510 const void* src, size_t blockSize, ZSTD_paramSwitch_e externalRepSearch);
1513 /* ===============================================================
1514 * Deprecated definitions that are still used internally to avoid
1515 * deprecation warnings. These functions are exactly equivalent to
1516 * their public variants, but avoid the deprecation warnings.
1517 * =============================================================== */
1519 size_t ZSTD_compressBegin_usingCDict_deprecated(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict);
1521 size_t ZSTD_compressContinue_public(ZSTD_CCtx* cctx,
1522 void* dst, size_t dstCapacity,
1523 const void* src, size_t srcSize);
1525 size_t ZSTD_compressEnd_public(ZSTD_CCtx* cctx,
1526 void* dst, size_t dstCapacity,
1527 const void* src, size_t srcSize);
1529 size_t ZSTD_compressBlock_deprecated(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
1532 #endif /* ZSTD_COMPRESS_H */