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 */
41static 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, dstCapacity: dstSize, src: literals, srcSize: litSize);
64 } else if (hufMetadata->hType == set_rle) {
65 DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
66 return ZSTD_compressRleLiteralsBlock(dst, dstCapacity: dstSize, src: literals, srcSize: 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(dst: op, dstSize: oend-op, src: literals, srcSize: litSize, CTable: hufTable)
81 : HUF_compress4X_usingCTable(dst: op, dstSize: oend-op, src: literals, srcSize: litSize, CTable: hufTable);
82 op += cSize;
83 cLitSize += cSize;
84 if (cSize == 0 || ERR_isError(code: 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, dstCapacity: dstSize, src: literals, srcSize: 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, dstCapacity: dstSize, src: literals, srcSize: 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(memPtr: ostart, val: 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(memPtr: ostart, val32: 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(memPtr: ostart, val32: 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
129static 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 (void)(litLengthSum); /* suppress unused variable warning on some environments */
136 while (send-sp > 0) {
137 ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, seq: sp);
138 litLengthSum += seqLen.litLength;
139 matchLengthSum += seqLen.matchLength;
140 sp++;
141 }
142 assert(litLengthSum <= litSize);
143 if (!lastSequence) {
144 assert(litLengthSum == litSize);
145 }
146 return matchLengthSum + litSize;
147}
148
149/* ZSTD_compressSubBlock_sequences() :
150 * Compresses sequences section for a sub-block.
151 * fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
152 * symbol compression modes for the super-block.
153 * The first successfully compressed block will have these in its header.
154 * We set entropyWritten=1 when we succeed in compressing the sequences.
155 * The following sub-blocks will always have repeat mode.
156 * @return : compressed size of sequences section of a sub-block
157 * Or 0 if it is unable to compress
158 * Or error code. */
159static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
160 const ZSTD_fseCTablesMetadata_t* fseMetadata,
161 const seqDef* sequences, size_t nbSeq,
162 const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
163 const ZSTD_CCtx_params* cctxParams,
164 void* dst, size_t dstCapacity,
165 const int bmi2, int writeEntropy, int* entropyWritten)
166{
167 const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
168 BYTE* const ostart = (BYTE*)dst;
169 BYTE* const oend = ostart + dstCapacity;
170 BYTE* op = ostart;
171 BYTE* seqHead;
172
173 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
174
175 *entropyWritten = 0;
176 /* Sequences Header */
177 RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
178 dstSize_tooSmall, "");
179 if (nbSeq < 0x7F)
180 *op++ = (BYTE)nbSeq;
181 else if (nbSeq < LONGNBSEQ)
182 op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
183 else
184 op[0]=0xFF, MEM_writeLE16(memPtr: op+1, val: (U16)(nbSeq - LONGNBSEQ)), op+=3;
185 if (nbSeq==0) {
186 return op - ostart;
187 }
188
189 /* seqHead : flags for FSE encoding type */
190 seqHead = op++;
191
192 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
193
194 if (writeEntropy) {
195 const U32 LLtype = fseMetadata->llType;
196 const U32 Offtype = fseMetadata->ofType;
197 const U32 MLtype = fseMetadata->mlType;
198 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
199 *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
200 ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
201 op += fseMetadata->fseTablesSize;
202 } else {
203 const U32 repeat = set_repeat;
204 *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
205 }
206
207 { size_t const bitstreamSize = ZSTD_encodeSequences(
208 dst: op, dstCapacity: oend - op,
209 CTable_MatchLength: fseTables->matchlengthCTable, mlCodeTable: mlCode,
210 CTable_OffsetBits: fseTables->offcodeCTable, ofCodeTable: ofCode,
211 CTable_LitLength: fseTables->litlengthCTable, llCodeTable: llCode,
212 sequences, nbSeq,
213 longOffsets, bmi2);
214 FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
215 op += bitstreamSize;
216 /* zstd versions <= 1.3.4 mistakenly report corruption when
217 * FSE_readNCount() receives a buffer < 4 bytes.
218 * Fixed by https://github.com/facebook/zstd/pull/1146.
219 * This can happen when the last set_compressed table present is 2
220 * bytes and the bitstream is only one byte.
221 * In this exceedingly rare case, we will simply emit an uncompressed
222 * block, since it isn't worth optimizing.
223 */
224#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
225 if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
226 /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
227 assert(fseMetadata->lastCountSize + bitstreamSize == 3);
228 DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
229 "emitting an uncompressed block.");
230 return 0;
231 }
232#endif
233 DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
234 }
235
236 /* zstd versions <= 1.4.0 mistakenly report error when
237 * sequences section body size is less than 3 bytes.
238 * Fixed by https://github.com/facebook/zstd/pull/1664.
239 * This can happen when the previous sequences section block is compressed
240 * with rle mode and the current block's sequences section is compressed
241 * with repeat mode where sequences section body size can be 1 byte.
242 */
243#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
244 if (op-seqHead < 4) {
245 DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
246 "an uncompressed block when sequences are < 4 bytes");
247 return 0;
248 }
249#endif
250
251 *entropyWritten = 1;
252 return op - ostart;
253}
254
255/* ZSTD_compressSubBlock() :
256 * Compresses a single sub-block.
257 * @return : compressed size of the sub-block
258 * Or 0 if it failed to compress. */
259static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
260 const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
261 const seqDef* sequences, size_t nbSeq,
262 const BYTE* literals, size_t litSize,
263 const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
264 const ZSTD_CCtx_params* cctxParams,
265 void* dst, size_t dstCapacity,
266 const int bmi2,
267 int writeLitEntropy, int writeSeqEntropy,
268 int* litEntropyWritten, int* seqEntropyWritten,
269 U32 lastBlock)
270{
271 BYTE* const ostart = (BYTE*)dst;
272 BYTE* const oend = ostart + dstCapacity;
273 BYTE* op = ostart + ZSTD_blockHeaderSize;
274 DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
275 litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
276 { size_t cLitSize = ZSTD_compressSubBlock_literal(hufTable: (const HUF_CElt*)entropy->huf.CTable,
277 hufMetadata: &entropyMetadata->hufMetadata, literals, litSize,
278 dst: op, dstSize: oend-op, bmi2, writeEntropy: writeLitEntropy, entropyWritten: litEntropyWritten);
279 FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
280 if (cLitSize == 0) return 0;
281 op += cLitSize;
282 }
283 { size_t cSeqSize = ZSTD_compressSubBlock_sequences(fseTables: &entropy->fse,
284 fseMetadata: &entropyMetadata->fseMetadata,
285 sequences, nbSeq,
286 llCode, mlCode, ofCode,
287 cctxParams,
288 dst: op, dstCapacity: oend-op,
289 bmi2, writeEntropy: writeSeqEntropy, entropyWritten: seqEntropyWritten);
290 FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
291 if (cSeqSize == 0) return 0;
292 op += cSeqSize;
293 }
294 /* Write block header */
295 { size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
296 U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
297 MEM_writeLE24(memPtr: ostart, val: cBlockHeader24);
298 }
299 return op-ostart;
300}
301
302static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
303 const ZSTD_hufCTables_t* huf,
304 const ZSTD_hufCTablesMetadata_t* hufMetadata,
305 void* workspace, size_t wkspSize,
306 int writeEntropy)
307{
308 unsigned* const countWksp = (unsigned*)workspace;
309 unsigned maxSymbolValue = 255;
310 size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
311
312 if (hufMetadata->hType == set_basic) return litSize;
313 else if (hufMetadata->hType == set_rle) return 1;
314 else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
315 size_t const largest = HIST_count_wksp (count: countWksp, maxSymbolValuePtr: &maxSymbolValue, src: (const BYTE*)literals, srcSize: litSize, workSpace: workspace, workSpaceSize: wkspSize);
316 if (ZSTD_isError(code: largest)) return litSize;
317 { size_t cLitSizeEstimate = HUF_estimateCompressedSize(CTable: (const HUF_CElt*)huf->CTable, count: countWksp, maxSymbolValue);
318 if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
319 return cLitSizeEstimate + literalSectionHeaderSize;
320 } }
321 assert(0); /* impossible */
322 return 0;
323}
324
325static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
326 const BYTE* codeTable, unsigned maxCode,
327 size_t nbSeq, const FSE_CTable* fseCTable,
328 const U8* additionalBits,
329 short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
330 void* workspace, size_t wkspSize)
331{
332 unsigned* const countWksp = (unsigned*)workspace;
333 const BYTE* ctp = codeTable;
334 const BYTE* const ctStart = ctp;
335 const BYTE* const ctEnd = ctStart + nbSeq;
336 size_t cSymbolTypeSizeEstimateInBits = 0;
337 unsigned max = maxCode;
338
339 HIST_countFast_wksp(count: countWksp, maxSymbolValuePtr: &max, src: codeTable, srcSize: nbSeq, workSpace: workspace, workSpaceSize: wkspSize); /* can't fail */
340 if (type == set_basic) {
341 /* We selected this encoding type, so it must be valid. */
342 assert(max <= defaultMax);
343 cSymbolTypeSizeEstimateInBits = max <= defaultMax
344 ? ZSTD_crossEntropyCost(norm: defaultNorm, accuracyLog: defaultNormLog, count: countWksp, max)
345 : ERROR(GENERIC);
346 } else if (type == set_rle) {
347 cSymbolTypeSizeEstimateInBits = 0;
348 } else if (type == set_compressed || type == set_repeat) {
349 cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(ctable: fseCTable, count: countWksp, max);
350 }
351 if (ZSTD_isError(code: cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
352 while (ctp < ctEnd) {
353 if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
354 else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
355 ctp++;
356 }
357 return cSymbolTypeSizeEstimateInBits / 8;
358}
359
360static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
361 const BYTE* llCodeTable,
362 const BYTE* mlCodeTable,
363 size_t nbSeq,
364 const ZSTD_fseCTables_t* fseTables,
365 const ZSTD_fseCTablesMetadata_t* fseMetadata,
366 void* workspace, size_t wkspSize,
367 int writeEntropy)
368{
369 size_t const sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
370 size_t cSeqSizeEstimate = 0;
371 if (nbSeq == 0) return sequencesSectionHeaderSize;
372 cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(type: fseMetadata->ofType, codeTable: ofCodeTable, MaxOff,
373 nbSeq, fseCTable: fseTables->offcodeCTable, NULL,
374 defaultNorm: OF_defaultNorm, defaultNormLog: OF_defaultNormLog, DefaultMaxOff,
375 workspace, wkspSize);
376 cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(type: fseMetadata->llType, codeTable: llCodeTable, MaxLL,
377 nbSeq, fseCTable: fseTables->litlengthCTable, additionalBits: LL_bits,
378 defaultNorm: LL_defaultNorm, defaultNormLog: LL_defaultNormLog, MaxLL,
379 workspace, wkspSize);
380 cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(type: fseMetadata->mlType, codeTable: mlCodeTable, MaxML,
381 nbSeq, fseCTable: fseTables->matchlengthCTable, additionalBits: ML_bits,
382 defaultNorm: ML_defaultNorm, defaultNormLog: ML_defaultNormLog, MaxML,
383 workspace, wkspSize);
384 if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
385 return cSeqSizeEstimate + sequencesSectionHeaderSize;
386}
387
388static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
389 const BYTE* ofCodeTable,
390 const BYTE* llCodeTable,
391 const BYTE* mlCodeTable,
392 size_t nbSeq,
393 const ZSTD_entropyCTables_t* entropy,
394 const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
395 void* workspace, size_t wkspSize,
396 int writeLitEntropy, int writeSeqEntropy) {
397 size_t cSizeEstimate = 0;
398 cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
399 huf: &entropy->huf, hufMetadata: &entropyMetadata->hufMetadata,
400 workspace, wkspSize, writeEntropy: writeLitEntropy);
401 cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
402 nbSeq, fseTables: &entropy->fse, fseMetadata: &entropyMetadata->fseMetadata,
403 workspace, wkspSize, writeEntropy: writeSeqEntropy);
404 return cSizeEstimate + ZSTD_blockHeaderSize;
405}
406
407static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
408{
409 if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
410 return 1;
411 if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
412 return 1;
413 if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
414 return 1;
415 return 0;
416}
417
418/* ZSTD_compressSubBlock_multi() :
419 * Breaks super-block into multiple sub-blocks and compresses them.
420 * Entropy will be written to the first block.
421 * The following blocks will use repeat mode to compress.
422 * All sub-blocks are compressed blocks (no raw or rle blocks).
423 * @return : compressed size of the super block (which is multiple ZSTD blocks)
424 * Or 0 if it failed to compress. */
425static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
426 const ZSTD_compressedBlockState_t* prevCBlock,
427 ZSTD_compressedBlockState_t* nextCBlock,
428 const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
429 const ZSTD_CCtx_params* cctxParams,
430 void* dst, size_t dstCapacity,
431 const void* src, size_t srcSize,
432 const int bmi2, U32 lastBlock,
433 void* workspace, size_t wkspSize)
434{
435 const seqDef* const sstart = seqStorePtr->sequencesStart;
436 const seqDef* const send = seqStorePtr->sequences;
437 const seqDef* sp = sstart;
438 const BYTE* const lstart = seqStorePtr->litStart;
439 const BYTE* const lend = seqStorePtr->lit;
440 const BYTE* lp = lstart;
441 BYTE const* ip = (BYTE const*)src;
442 BYTE const* const iend = ip + srcSize;
443 BYTE* const ostart = (BYTE*)dst;
444 BYTE* const oend = ostart + dstCapacity;
445 BYTE* op = ostart;
446 const BYTE* llCodePtr = seqStorePtr->llCode;
447 const BYTE* mlCodePtr = seqStorePtr->mlCode;
448 const BYTE* ofCodePtr = seqStorePtr->ofCode;
449 size_t targetCBlockSize = cctxParams->targetCBlockSize;
450 size_t litSize, seqCount;
451 int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
452 int writeSeqEntropy = 1;
453 int lastSequence = 0;
454
455 DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
456 (unsigned)(lend-lp), (unsigned)(send-sstart));
457
458 litSize = 0;
459 seqCount = 0;
460 do {
461 size_t cBlockSizeEstimate = 0;
462 if (sstart == send) {
463 lastSequence = 1;
464 } else {
465 const seqDef* const sequence = sp + seqCount;
466 lastSequence = sequence == send - 1;
467 litSize += ZSTD_getSequenceLength(seqStore: seqStorePtr, seq: sequence).litLength;
468 seqCount++;
469 }
470 if (lastSequence) {
471 assert(lp <= lend);
472 assert(litSize <= (size_t)(lend - lp));
473 litSize = (size_t)(lend - lp);
474 }
475 /* I think there is an optimization opportunity here.
476 * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
477 * since it recalculates estimate from scratch.
478 * For example, it would recount literal distribution and symbol codes every time.
479 */
480 cBlockSizeEstimate = ZSTD_estimateSubBlockSize(literals: lp, litSize, ofCodeTable: ofCodePtr, llCodeTable: llCodePtr, mlCodeTable: mlCodePtr, nbSeq: seqCount,
481 entropy: &nextCBlock->entropy, entropyMetadata,
482 workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
483 if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
484 int litEntropyWritten = 0;
485 int seqEntropyWritten = 0;
486 const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStore: seqStorePtr, sequences: sp, nbSeq: seqCount, litSize, lastSequence);
487 const size_t cSize = ZSTD_compressSubBlock(entropy: &nextCBlock->entropy, entropyMetadata,
488 sequences: sp, nbSeq: seqCount,
489 literals: lp, litSize,
490 llCode: llCodePtr, mlCode: mlCodePtr, ofCode: ofCodePtr,
491 cctxParams,
492 dst: op, dstCapacity: oend-op,
493 bmi2, writeLitEntropy, writeSeqEntropy,
494 litEntropyWritten: &litEntropyWritten, seqEntropyWritten: &seqEntropyWritten,
495 lastBlock: lastBlock && lastSequence);
496 FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
497 if (cSize > 0 && cSize < decompressedSize) {
498 DEBUGLOG(5, "Committed the sub-block");
499 assert(ip + decompressedSize <= iend);
500 ip += decompressedSize;
501 sp += seqCount;
502 lp += litSize;
503 op += cSize;
504 llCodePtr += seqCount;
505 mlCodePtr += seqCount;
506 ofCodePtr += seqCount;
507 litSize = 0;
508 seqCount = 0;
509 /* Entropy only needs to be written once */
510 if (litEntropyWritten) {
511 writeLitEntropy = 0;
512 }
513 if (seqEntropyWritten) {
514 writeSeqEntropy = 0;
515 }
516 }
517 }
518 } while (!lastSequence);
519 if (writeLitEntropy) {
520 DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
521 ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
522 }
523 if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(fseMetadata: &entropyMetadata->fseMetadata)) {
524 /* If we haven't written our entropy tables, then we've violated our contract and
525 * must emit an uncompressed block.
526 */
527 DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
528 return 0;
529 }
530 if (ip < iend) {
531 size_t const cSize = ZSTD_noCompressBlock(dst: op, dstCapacity: oend - op, src: ip, srcSize: iend - ip, lastBlock);
532 DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
533 FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
534 assert(cSize != 0);
535 op += cSize;
536 /* We have to regenerate the repcodes because we've skipped some sequences */
537 if (sp < send) {
538 seqDef const* seq;
539 repcodes_t rep;
540 ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
541 for (seq = sstart; seq < sp; ++seq) {
542 ZSTD_updateRep(rep: rep.rep, offBase_minus1: seq->offBase - 1, ll0: ZSTD_getSequenceLength(seqStore: seqStorePtr, seq).litLength == 0);
543 }
544 ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
545 }
546 }
547 DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
548 return op-ostart;
549}
550
551size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
552 void* dst, size_t dstCapacity,
553 void const* src, size_t srcSize,
554 unsigned lastBlock) {
555 ZSTD_entropyCTablesMetadata_t entropyMetadata;
556
557 FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
558 &zc->blockState.prevCBlock->entropy,
559 &zc->blockState.nextCBlock->entropy,
560 &zc->appliedParams,
561 &entropyMetadata,
562 zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
563
564 return ZSTD_compressSubBlock_multi(seqStorePtr: &zc->seqStore,
565 prevCBlock: zc->blockState.prevCBlock,
566 nextCBlock: zc->blockState.nextCBlock,
567 entropyMetadata: &entropyMetadata,
568 cctxParams: &zc->appliedParams,
569 dst, dstCapacity,
570 src, srcSize,
571 bmi2: zc->bmi2, lastBlock,
572 workspace: zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
573}
574

source code of linux/lib/zstd/compress/zstd_compress_superblock.c