1 /*
2 * Copyright (c) Yann Collet, Facebook, Inc.
3 * All rights reserved.
4 *
5 * This source code is licensed under both the BSD-style license (found in the
6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7 * in the COPYING file in the root directory of this source tree).
8 * You may select, at your option, one of the above-listed licenses.
9 */
10
11 /*-*************************************
12 * Dependencies
13 ***************************************/
14 #include "zstd_compress_superblock.h"
15
16 #include "../common/zstd_internal.h" /* ZSTD_getSequenceLength */
17 #include "hist.h" /* HIST_countFast_wksp */
18 #include "zstd_compress_internal.h" /* ZSTD_[huf|fse|entropy]CTablesMetadata_t */
19 #include "zstd_compress_sequences.h"
20 #include "zstd_compress_literals.h"
21
22 /** ZSTD_compressSubBlock_literal() :
23 * Compresses literals section for a sub-block.
24 * When we have to write the Huffman table we will sometimes choose a header
25 * size larger than necessary. This is because we have to pick the header size
26 * before we know the table size + compressed size, so we have a bound on the
27 * table size. If we guessed incorrectly, we fall back to uncompressed literals.
28 *
29 * We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
30 * in writing the header, otherwise it is set to 0.
31 *
32 * hufMetadata->hType has literals block type info.
33 * If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
34 * If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
35 * If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
36 * If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
37 * and the following sub-blocks' literals sections will be Treeless_Literals_Block.
38 * @return : compressed size of literals section of a sub-block
39 * Or 0 if it unable to compress.
40 * Or error code */
ZSTD_compressSubBlock_literal(const HUF_CElt * hufTable,const ZSTD_hufCTablesMetadata_t * hufMetadata,const BYTE * literals,size_t litSize,void * dst,size_t dstSize,const int bmi2,int writeEntropy,int * entropyWritten)41 static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
42 const ZSTD_hufCTablesMetadata_t* hufMetadata,
43 const BYTE* literals, size_t litSize,
44 void* dst, size_t dstSize,
45 const int bmi2, int writeEntropy, int* entropyWritten)
46 {
47 size_t const header = writeEntropy ? 200 : 0;
48 size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
49 BYTE* const ostart = (BYTE*)dst;
50 BYTE* const oend = ostart + dstSize;
51 BYTE* op = ostart + lhSize;
52 U32 const singleStream = lhSize == 3;
53 symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
54 size_t cLitSize = 0;
55
56 (void)bmi2; /* TODO bmi2... */
57
58 DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
59
60 *entropyWritten = 0;
61 if (litSize == 0 || hufMetadata->hType == set_basic) {
62 DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
63 return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
64 } else if (hufMetadata->hType == set_rle) {
65 DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
66 return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
67 }
68
69 assert(litSize > 0);
70 assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
71
72 if (writeEntropy && hufMetadata->hType == set_compressed) {
73 ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
74 op += hufMetadata->hufDesSize;
75 cLitSize += hufMetadata->hufDesSize;
76 DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
77 }
78
79 /* TODO bmi2 */
80 { const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable)
81 : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable);
82 op += cSize;
83 cLitSize += cSize;
84 if (cSize == 0 || ERR_isError(cSize)) {
85 DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
86 return 0;
87 }
88 /* If we expand and we aren't writing a header then emit uncompressed */
89 if (!writeEntropy && cLitSize >= litSize) {
90 DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
91 return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
92 }
93 /* If we are writing headers then allow expansion that doesn't change our header size. */
94 if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
95 assert(cLitSize > litSize);
96 DEBUGLOG(5, "Literals expanded beyond allowed header size");
97 return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
98 }
99 DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
100 }
101
102 /* Build header */
103 switch(lhSize)
104 {
105 case 3: /* 2 - 2 - 10 - 10 */
106 { U32 const lhc = hType + ((!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
107 MEM_writeLE24(ostart, lhc);
108 break;
109 }
110 case 4: /* 2 - 2 - 14 - 14 */
111 { U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
112 MEM_writeLE32(ostart, lhc);
113 break;
114 }
115 case 5: /* 2 - 2 - 18 - 18 */
116 { U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
117 MEM_writeLE32(ostart, lhc);
118 ostart[4] = (BYTE)(cLitSize >> 10);
119 break;
120 }
121 default: /* not possible : lhSize is {3,4,5} */
122 assert(0);
123 }
124 *entropyWritten = 1;
125 DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
126 return op-ostart;
127 }
128
ZSTD_seqDecompressedSize(seqStore_t const * seqStore,const seqDef * sequences,size_t nbSeq,size_t litSize,int lastSequence)129 static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef* sequences, size_t nbSeq, size_t litSize, int lastSequence) {
130 const seqDef* const sstart = sequences;
131 const seqDef* const send = sequences + nbSeq;
132 const seqDef* sp = sstart;
133 size_t matchLengthSum = 0;
134 size_t litLengthSum = 0;
135 while (send-sp > 0) {
136 ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
137 litLengthSum += seqLen.litLength;
138 matchLengthSum += seqLen.matchLength;
139 sp++;
140 }
141 assert(litLengthSum <= litSize);
142 if (!lastSequence) {
143 assert(litLengthSum == litSize);
144 }
145 return matchLengthSum + litSize;
146 }
147
148 /** ZSTD_compressSubBlock_sequences() :
149 * Compresses sequences section for a sub-block.
150 * fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
151 * symbol compression modes for the super-block.
152 * The first successfully compressed block will have these in its header.
153 * We set entropyWritten=1 when we succeed in compressing the sequences.
154 * The following sub-blocks will always have repeat mode.
155 * @return : compressed size of sequences section of a sub-block
156 * Or 0 if it is unable to compress
157 * Or error code. */
ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t * fseTables,const ZSTD_fseCTablesMetadata_t * fseMetadata,const seqDef * sequences,size_t nbSeq,const BYTE * llCode,const BYTE * mlCode,const BYTE * ofCode,const ZSTD_CCtx_params * cctxParams,void * dst,size_t dstCapacity,const int bmi2,int writeEntropy,int * entropyWritten)158 static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
159 const ZSTD_fseCTablesMetadata_t* fseMetadata,
160 const seqDef* sequences, size_t nbSeq,
161 const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
162 const ZSTD_CCtx_params* cctxParams,
163 void* dst, size_t dstCapacity,
164 const int bmi2, int writeEntropy, int* entropyWritten)
165 {
166 const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
167 BYTE* const ostart = (BYTE*)dst;
168 BYTE* const oend = ostart + dstCapacity;
169 BYTE* op = ostart;
170 BYTE* seqHead;
171
172 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
173
174 *entropyWritten = 0;
175 /* Sequences Header */
176 RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
177 dstSize_tooSmall, "");
178 if (nbSeq < 0x7F)
179 *op++ = (BYTE)nbSeq;
180 else if (nbSeq < LONGNBSEQ)
181 op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
182 else
183 op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
184 if (nbSeq==0) {
185 return op - ostart;
186 }
187
188 /* seqHead : flags for FSE encoding type */
189 seqHead = op++;
190
191 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
192
193 if (writeEntropy) {
194 const U32 LLtype = fseMetadata->llType;
195 const U32 Offtype = fseMetadata->ofType;
196 const U32 MLtype = fseMetadata->mlType;
197 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
198 *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
199 ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
200 op += fseMetadata->fseTablesSize;
201 } else {
202 const U32 repeat = set_repeat;
203 *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
204 }
205
206 { size_t const bitstreamSize = ZSTD_encodeSequences(
207 op, oend - op,
208 fseTables->matchlengthCTable, mlCode,
209 fseTables->offcodeCTable, ofCode,
210 fseTables->litlengthCTable, llCode,
211 sequences, nbSeq,
212 longOffsets, bmi2);
213 FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
214 op += bitstreamSize;
215 /* zstd versions <= 1.3.4 mistakenly report corruption when
216 * FSE_readNCount() receives a buffer < 4 bytes.
217 * Fixed by https://github.com/facebook/zstd/pull/1146.
218 * This can happen when the last set_compressed table present is 2
219 * bytes and the bitstream is only one byte.
220 * In this exceedingly rare case, we will simply emit an uncompressed
221 * block, since it isn't worth optimizing.
222 */
223 #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
224 if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
225 /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
226 assert(fseMetadata->lastCountSize + bitstreamSize == 3);
227 DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
228 "emitting an uncompressed block.");
229 return 0;
230 }
231 #endif
232 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
233 }
234
235 /* zstd versions <= 1.4.0 mistakenly report error when
236 * sequences section body size is less than 3 bytes.
237 * Fixed by https://github.com/facebook/zstd/pull/1664.
238 * This can happen when the previous sequences section block is compressed
239 * with rle mode and the current block's sequences section is compressed
240 * with repeat mode where sequences section body size can be 1 byte.
241 */
242 #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
243 if (op-seqHead < 4) {
244 DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
245 "an uncompressed block when sequences are < 4 bytes");
246 return 0;
247 }
248 #endif
249
250 *entropyWritten = 1;
251 return op - ostart;
252 }
253
254 /** ZSTD_compressSubBlock() :
255 * Compresses a single sub-block.
256 * @return : compressed size of the sub-block
257 * Or 0 if it failed to compress. */
ZSTD_compressSubBlock(const ZSTD_entropyCTables_t * entropy,const ZSTD_entropyCTablesMetadata_t * entropyMetadata,const seqDef * sequences,size_t nbSeq,const BYTE * literals,size_t litSize,const BYTE * llCode,const BYTE * mlCode,const BYTE * ofCode,const ZSTD_CCtx_params * cctxParams,void * dst,size_t dstCapacity,const int bmi2,int writeLitEntropy,int writeSeqEntropy,int * litEntropyWritten,int * seqEntropyWritten,U32 lastBlock)258 static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
259 const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
260 const seqDef* sequences, size_t nbSeq,
261 const BYTE* literals, size_t litSize,
262 const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
263 const ZSTD_CCtx_params* cctxParams,
264 void* dst, size_t dstCapacity,
265 const int bmi2,
266 int writeLitEntropy, int writeSeqEntropy,
267 int* litEntropyWritten, int* seqEntropyWritten,
268 U32 lastBlock)
269 {
270 BYTE* const ostart = (BYTE*)dst;
271 BYTE* const oend = ostart + dstCapacity;
272 BYTE* op = ostart + ZSTD_blockHeaderSize;
273 DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
274 litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
275 { size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
276 &entropyMetadata->hufMetadata, literals, litSize,
277 op, oend-op, bmi2, writeLitEntropy, litEntropyWritten);
278 FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
279 if (cLitSize == 0) return 0;
280 op += cLitSize;
281 }
282 { size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
283 &entropyMetadata->fseMetadata,
284 sequences, nbSeq,
285 llCode, mlCode, ofCode,
286 cctxParams,
287 op, oend-op,
288 bmi2, writeSeqEntropy, seqEntropyWritten);
289 FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
290 if (cSeqSize == 0) return 0;
291 op += cSeqSize;
292 }
293 /* Write block header */
294 { size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
295 U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
296 MEM_writeLE24(ostart, cBlockHeader24);
297 }
298 return op-ostart;
299 }
300
ZSTD_estimateSubBlockSize_literal(const BYTE * literals,size_t litSize,const ZSTD_hufCTables_t * huf,const ZSTD_hufCTablesMetadata_t * hufMetadata,void * workspace,size_t wkspSize,int writeEntropy)301 static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
302 const ZSTD_hufCTables_t* huf,
303 const ZSTD_hufCTablesMetadata_t* hufMetadata,
304 void* workspace, size_t wkspSize,
305 int writeEntropy)
306 {
307 unsigned* const countWksp = (unsigned*)workspace;
308 unsigned maxSymbolValue = 255;
309 size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
310
311 if (hufMetadata->hType == set_basic) return litSize;
312 else if (hufMetadata->hType == set_rle) return 1;
313 else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
314 size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
315 if (ZSTD_isError(largest)) return litSize;
316 { size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
317 if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
318 return cLitSizeEstimate + literalSectionHeaderSize;
319 } }
320 assert(0); /* impossible */
321 return 0;
322 }
323
ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,const BYTE * codeTable,unsigned maxCode,size_t nbSeq,const FSE_CTable * fseCTable,const U32 * additionalBits,short const * defaultNorm,U32 defaultNormLog,U32 defaultMax,void * workspace,size_t wkspSize)324 static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
325 const BYTE* codeTable, unsigned maxCode,
326 size_t nbSeq, const FSE_CTable* fseCTable,
327 const U32* additionalBits,
328 short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
329 void* workspace, size_t wkspSize)
330 {
331 unsigned* const countWksp = (unsigned*)workspace;
332 const BYTE* ctp = codeTable;
333 const BYTE* const ctStart = ctp;
334 const BYTE* const ctEnd = ctStart + nbSeq;
335 size_t cSymbolTypeSizeEstimateInBits = 0;
336 unsigned max = maxCode;
337
338 HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize); /* can't fail */
339 if (type == set_basic) {
340 /* We selected this encoding type, so it must be valid. */
341 assert(max <= defaultMax);
342 cSymbolTypeSizeEstimateInBits = max <= defaultMax
343 ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
344 : ERROR(GENERIC);
345 } else if (type == set_rle) {
346 cSymbolTypeSizeEstimateInBits = 0;
347 } else if (type == set_compressed || type == set_repeat) {
348 cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
349 }
350 if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
351 while (ctp < ctEnd) {
352 if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
353 else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
354 ctp++;
355 }
356 return cSymbolTypeSizeEstimateInBits / 8;
357 }
358
ZSTD_estimateSubBlockSize_sequences(const BYTE * ofCodeTable,const BYTE * llCodeTable,const BYTE * mlCodeTable,size_t nbSeq,const ZSTD_fseCTables_t * fseTables,const ZSTD_fseCTablesMetadata_t * fseMetadata,void * workspace,size_t wkspSize,int writeEntropy)359 static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
360 const BYTE* llCodeTable,
361 const BYTE* mlCodeTable,
362 size_t nbSeq,
363 const ZSTD_fseCTables_t* fseTables,
364 const ZSTD_fseCTablesMetadata_t* fseMetadata,
365 void* workspace, size_t wkspSize,
366 int writeEntropy)
367 {
368 size_t const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
369 size_t cSeqSizeEstimate = 0;
370 if (nbSeq == 0) return sequencesSectionHeaderSize;
371 cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
372 nbSeq, fseTables->offcodeCTable, NULL,
373 OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
374 workspace, wkspSize);
375 cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
376 nbSeq, fseTables->litlengthCTable, LL_bits,
377 LL_defaultNorm, LL_defaultNormLog, MaxLL,
378 workspace, wkspSize);
379 cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
380 nbSeq, fseTables->matchlengthCTable, ML_bits,
381 ML_defaultNorm, ML_defaultNormLog, MaxML,
382 workspace, wkspSize);
383 if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
384 return cSeqSizeEstimate + sequencesSectionHeaderSize;
385 }
386
ZSTD_estimateSubBlockSize(const BYTE * literals,size_t litSize,const BYTE * ofCodeTable,const BYTE * llCodeTable,const BYTE * mlCodeTable,size_t nbSeq,const ZSTD_entropyCTables_t * entropy,const ZSTD_entropyCTablesMetadata_t * entropyMetadata,void * workspace,size_t wkspSize,int writeLitEntropy,int writeSeqEntropy)387 static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
388 const BYTE* ofCodeTable,
389 const BYTE* llCodeTable,
390 const BYTE* mlCodeTable,
391 size_t nbSeq,
392 const ZSTD_entropyCTables_t* entropy,
393 const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
394 void* workspace, size_t wkspSize,
395 int writeLitEntropy, int writeSeqEntropy) {
396 size_t cSizeEstimate = 0;
397 cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
398 &entropy->huf, &entropyMetadata->hufMetadata,
399 workspace, wkspSize, writeLitEntropy);
400 cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
401 nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
402 workspace, wkspSize, writeSeqEntropy);
403 return cSizeEstimate + ZSTD_blockHeaderSize;
404 }
405
ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const * fseMetadata)406 static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
407 {
408 if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
409 return 1;
410 if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
411 return 1;
412 if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
413 return 1;
414 return 0;
415 }
416
417 /** ZSTD_compressSubBlock_multi() :
418 * Breaks super-block into multiple sub-blocks and compresses them.
419 * Entropy will be written to the first block.
420 * The following blocks will use repeat mode to compress.
421 * All sub-blocks are compressed blocks (no raw or rle blocks).
422 * @return : compressed size of the super block (which is multiple ZSTD blocks)
423 * Or 0 if it failed to compress. */
ZSTD_compressSubBlock_multi(const seqStore_t * seqStorePtr,const ZSTD_compressedBlockState_t * prevCBlock,ZSTD_compressedBlockState_t * nextCBlock,const ZSTD_entropyCTablesMetadata_t * entropyMetadata,const ZSTD_CCtx_params * cctxParams,void * dst,size_t dstCapacity,const void * src,size_t srcSize,const int bmi2,U32 lastBlock,void * workspace,size_t wkspSize)424 static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
425 const ZSTD_compressedBlockState_t* prevCBlock,
426 ZSTD_compressedBlockState_t* nextCBlock,
427 const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
428 const ZSTD_CCtx_params* cctxParams,
429 void* dst, size_t dstCapacity,
430 const void* src, size_t srcSize,
431 const int bmi2, U32 lastBlock,
432 void* workspace, size_t wkspSize)
433 {
434 const seqDef* const sstart = seqStorePtr->sequencesStart;
435 const seqDef* const send = seqStorePtr->sequences;
436 const seqDef* sp = sstart;
437 const BYTE* const lstart = seqStorePtr->litStart;
438 const BYTE* const lend = seqStorePtr->lit;
439 const BYTE* lp = lstart;
440 BYTE const* ip = (BYTE const*)src;
441 BYTE const* const iend = ip + srcSize;
442 BYTE* const ostart = (BYTE*)dst;
443 BYTE* const oend = ostart + dstCapacity;
444 BYTE* op = ostart;
445 const BYTE* llCodePtr = seqStorePtr->llCode;
446 const BYTE* mlCodePtr = seqStorePtr->mlCode;
447 const BYTE* ofCodePtr = seqStorePtr->ofCode;
448 size_t targetCBlockSize = cctxParams->targetCBlockSize;
449 size_t litSize, seqCount;
450 int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
451 int writeSeqEntropy = 1;
452 int lastSequence = 0;
453
454 DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
455 (unsigned)(lend-lp), (unsigned)(send-sstart));
456
457 litSize = 0;
458 seqCount = 0;
459 do {
460 size_t cBlockSizeEstimate = 0;
461 if (sstart == send) {
462 lastSequence = 1;
463 } else {
464 const seqDef* const sequence = sp + seqCount;
465 lastSequence = sequence == send - 1;
466 litSize += ZSTD_getSequenceLength(seqStorePtr, sequence).litLength;
467 seqCount++;
468 }
469 if (lastSequence) {
470 assert(lp <= lend);
471 assert(litSize <= (size_t)(lend - lp));
472 litSize = (size_t)(lend - lp);
473 }
474 /* I think there is an optimization opportunity here.
475 * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
476 * since it recalculates estimate from scratch.
477 * For example, it would recount literal distribution and symbol codes everytime.
478 */
479 cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
480 &nextCBlock->entropy, entropyMetadata,
481 workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
482 if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
483 int litEntropyWritten = 0;
484 int seqEntropyWritten = 0;
485 const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, lastSequence);
486 const size_t cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
487 sp, seqCount,
488 lp, litSize,
489 llCodePtr, mlCodePtr, ofCodePtr,
490 cctxParams,
491 op, oend-op,
492 bmi2, writeLitEntropy, writeSeqEntropy,
493 &litEntropyWritten, &seqEntropyWritten,
494 lastBlock && lastSequence);
495 FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
496 if (cSize > 0 && cSize < decompressedSize) {
497 DEBUGLOG(5, "Committed the sub-block");
498 assert(ip + decompressedSize <= iend);
499 ip += decompressedSize;
500 sp += seqCount;
501 lp += litSize;
502 op += cSize;
503 llCodePtr += seqCount;
504 mlCodePtr += seqCount;
505 ofCodePtr += seqCount;
506 litSize = 0;
507 seqCount = 0;
508 /* Entropy only needs to be written once */
509 if (litEntropyWritten) {
510 writeLitEntropy = 0;
511 }
512 if (seqEntropyWritten) {
513 writeSeqEntropy = 0;
514 }
515 }
516 }
517 } while (!lastSequence);
518 if (writeLitEntropy) {
519 DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
520 ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
521 }
522 if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
523 /* If we haven't written our entropy tables, then we've violated our contract and
524 * must emit an uncompressed block.
525 */
526 DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
527 return 0;
528 }
529 if (ip < iend) {
530 size_t const cSize = ZSTD_noCompressBlock(op, oend - op, ip, iend - ip, lastBlock);
531 DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
532 FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
533 assert(cSize != 0);
534 op += cSize;
535 /* We have to regenerate the repcodes because we've skipped some sequences */
536 if (sp < send) {
537 seqDef const* seq;
538 repcodes_t rep;
539 ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
540 for (seq = sstart; seq < sp; ++seq) {
541 rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
542 }
543 ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
544 }
545 }
546 DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
547 return op-ostart;
548 }
549
ZSTD_compressSuperBlock(ZSTD_CCtx * zc,void * dst,size_t dstCapacity,void const * src,size_t srcSize,unsigned lastBlock)550 size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
551 void* dst, size_t dstCapacity,
552 void const* src, size_t srcSize,
553 unsigned lastBlock) {
554 ZSTD_entropyCTablesMetadata_t entropyMetadata;
555
556 FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
557 &zc->blockState.prevCBlock->entropy,
558 &zc->blockState.nextCBlock->entropy,
559 &zc->appliedParams,
560 &entropyMetadata,
561 zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
562
563 return ZSTD_compressSubBlock_multi(&zc->seqStore,
564 zc->blockState.prevCBlock,
565 zc->blockState.nextCBlock,
566 &entropyMetadata,
567 &zc->appliedParams,
568 dst, dstCapacity,
569 src, srcSize,
570 zc->bmi2, lastBlock,
571 zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
572 }
573