| 1 | |
| 2 | |
| 3 | * Copyright (c) Meta Platforms, Inc. and affiliates.
|
| 4 | * All rights reserved.
|
| 5 | *
|
| 6 | * This source code is licensed under both the BSD-style license (found in the
|
| 7 | * LICENSE file in the root directory of this source tree) and the GPLv2 (found
|
| 8 | * in the COPYING file in the root directory of this source tree).
|
| 9 | * You may select, at your option, one of the above-listed licenses.
|
| 10 |
|
| 11 |
|
| 12 | |
| 13 | * Dependencies
|
| 14 | **************************************
|
| 15 | #include "zstd_compress_superblock.h"
|
| 16 |
|
| 17 | #include "../common/zstd_internal.h"
|
| 18 | #include "hist.h"
|
| 19 | #include "zstd_compress_internal.h"
|
| 20 | #include "zstd_compress_sequences.h"
|
| 21 | #include "zstd_compress_literals.h"
|
| 22 |
|
| 23 | |
| 24 | * Compresses literals section for a sub-block.
|
| 25 | * When we have to write the Huffman table we will sometimes choose a header
|
| 26 | * size larger than necessary. This is because we have to pick the header size
|
| 27 | * before we know the table size + compressed size, so we have a bound on the
|
| 28 | * table size. If we guessed incorrectly, we fall back to uncompressed literals.
|
| 29 | *
|
| 30 | * We write the header when writeEntropy=1 and set entropyWritten=1 when we succeeded
|
| 31 | * in writing the header, otherwise it is set to 0.
|
| 32 | *
|
| 33 | * hufMetadata->hType has literals block type info.
|
| 34 | * If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
|
| 35 | * If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
|
| 36 | * If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
|
| 37 | * If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
|
| 38 | * and the following sub-blocks' literals sections will be Treeless_Literals_Block.
|
| 39 | * @return : compressed size of literals section of a sub-block
|
| 40 | * Or 0 if unable to compress.
|
| 41 | * Or error code
|
| 42 | static size_t
|
| 43 | ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
|
| 44 | const ZSTD_hufCTablesMetadata_t* hufMetadata,
|
| 45 | const BYTE* literals, size_t litSize,
|
| 46 | void* dst, size_t dstSize,
|
| 47 | const int bmi2, int writeEntropy, int* entropyWritten)
|
| 48 | {
|
| 49 | size_t const header = writeEntropy ? 200 : 0;
|
| 50 | size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
|
| 51 | BYTE* const ostart = (BYTE*)dst;
|
| 52 | BYTE* const oend = ostart + dstSize;
|
| 53 | BYTE* op = ostart + lhSize;
|
| 54 | U32 const singleStream = lhSize == 3;
|
| 55 | SymbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
|
| 56 | size_t cLitSize = 0;
|
| 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 | { int const flags = bmi2 ? HUF_flags_bmi2 : 0;
|
| 80 | const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags)
|
| 81 | : HUF_compress4X_usingCTable(op, (size_t)(oend-op), literals, litSize, hufTable, flags);
|
| 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 |
|
| 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 |
|
| 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 |
|
| 103 | switch(lhSize)
|
| 104 | {
|
| 105 | case 3:
|
| 106 | { U32 const lhc = hType + ((U32)(!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
|
| 107 | MEM_writeLE24(ostart, lhc);
|
| 108 | break;
|
| 109 | }
|
| 110 | case 4:
|
| 111 | { U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
|
| 112 | MEM_writeLE32(ostart, lhc);
|
| 113 | break;
|
| 114 | }
|
| 115 | case 5:
|
| 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:
|
| 122 | assert(0);
|
| 123 | }
|
| 124 | *entropyWritten = 1;
|
| 125 | DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
|
| 126 | return (size_t)(op-ostart);
|
| 127 | }
|
| 128 |
|
| 129 | static size_t
|
| 130 | ZSTD_seqDecompressedSize(SeqStore_t const* seqStore,
|
| 131 | const SeqDef* sequences, size_t nbSeqs,
|
| 132 | size_t litSize, int lastSubBlock)
|
| 133 | {
|
| 134 | size_t matchLengthSum = 0;
|
| 135 | size_t litLengthSum = 0;
|
| 136 | size_t n;
|
| 137 | for (n=0; n<nbSeqs; n++) {
|
| 138 | const ZSTD_SequenceLength seqLen = ZSTD_getSequenceLength(seqStore, sequences+n);
|
| 139 | litLengthSum += seqLen.litLength;
|
| 140 | matchLengthSum += seqLen.matchLength;
|
| 141 | }
|
| 142 | DEBUGLOG(5, "ZSTD_seqDecompressedSize: %u sequences from %p: %u literals + %u matchlength",
|
| 143 | (unsigned)nbSeqs, (const void*)sequences,
|
| 144 | (unsigned)litLengthSum, (unsigned)matchLengthSum);
|
| 145 | if (!lastSubBlock)
|
| 146 | assert(litLengthSum == litSize);
|
| 147 | else
|
| 148 | assert(litLengthSum <= litSize);
|
| 149 | (void)litLengthSum;
|
| 150 | return matchLengthSum + litSize;
|
| 151 | }
|
| 152 |
|
| 153 | |
| 154 | * Compresses sequences section for a sub-block.
|
| 155 | * fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
|
| 156 | * symbol compression modes for the super-block.
|
| 157 | * The first successfully compressed block will have these in its header.
|
| 158 | * We set entropyWritten=1 when we succeed in compressing the sequences.
|
| 159 | * The following sub-blocks will always have repeat mode.
|
| 160 | * @return : compressed size of sequences section of a sub-block
|
| 161 | * Or 0 if it is unable to compress
|
| 162 | * Or error code.
|
| 163 | static size_t
|
| 164 | ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
|
| 165 | const ZSTD_fseCTablesMetadata_t* fseMetadata,
|
| 166 | const SeqDef* sequences, size_t nbSeq,
|
| 167 | const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
|
| 168 | const ZSTD_CCtx_params* cctxParams,
|
| 169 | void* dst, size_t dstCapacity,
|
| 170 | const int bmi2, int writeEntropy, int* entropyWritten)
|
| 171 | {
|
| 172 | const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
|
| 173 | BYTE* const ostart = (BYTE*)dst;
|
| 174 | BYTE* const oend = ostart + dstCapacity;
|
| 175 | BYTE* op = ostart;
|
| 176 | BYTE* seqHead;
|
| 177 |
|
| 178 | DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
|
| 179 |
|
| 180 | *entropyWritten = 0;
|
| 181 |
|
| 182 | RETURN_ERROR_IF((oend-op) < 3 + 1 ,
|
| 183 | dstSize_tooSmall, "");
|
| 184 | if (nbSeq < 128)
|
| 185 | *op++ = (BYTE)nbSeq;
|
| 186 | else if (nbSeq < LONGNBSEQ)
|
| 187 | op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
|
| 188 | else
|
| 189 | op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
|
| 190 | if (nbSeq==0) {
|
| 191 | return (size_t)(op - ostart);
|
| 192 | }
|
| 193 |
|
| 194 |
|
| 195 | seqHead = op++;
|
| 196 |
|
| 197 | DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
|
| 198 |
|
| 199 | if (writeEntropy) {
|
| 200 | const U32 LLtype = fseMetadata->llType;
|
| 201 | const U32 Offtype = fseMetadata->ofType;
|
| 202 | const U32 MLtype = fseMetadata->mlType;
|
| 203 | DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
|
| 204 | *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
|
| 205 | ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
|
| 206 | op += fseMetadata->fseTablesSize;
|
| 207 | } else {
|
| 208 | const U32 repeat = set_repeat;
|
| 209 | *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
|
| 210 | }
|
| 211 |
|
| 212 | { size_t const bitstreamSize = ZSTD_encodeSequences(
|
| 213 | op, (size_t)(oend - op),
|
| 214 | fseTables->matchlengthCTable, mlCode,
|
| 215 | fseTables->offcodeCTable, ofCode,
|
| 216 | fseTables->litlengthCTable, llCode,
|
| 217 | sequences, nbSeq,
|
| 218 | longOffsets, bmi2);
|
| 219 | FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
|
| 220 | op += bitstreamSize;
|
| 221 | |
| 222 | * FSE_readNCount() receives a buffer < 4 bytes.
|
| 223 | * Fixed by https://github.com/facebook/zstd/pull/1146.
|
| 224 | * This can happen when the last set_compressed table present is 2
|
| 225 | * bytes and the bitstream is only one byte.
|
| 226 | * In this exceedingly rare case, we will simply emit an uncompressed
|
| 227 | * block, since it isn't worth optimizing.
|
| 228 |
|
| 229 | #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
|
| 230 | if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
|
| 231 |
|
| 232 | assert(fseMetadata->lastCountSize + bitstreamSize == 3);
|
| 233 | DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
|
| 234 | "emitting an uncompressed block.");
|
| 235 | return 0;
|
| 236 | }
|
| 237 | #endif
|
| 238 | DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
|
| 239 | }
|
| 240 |
|
| 241 | |
| 242 | * sequences section body size is less than 3 bytes.
|
| 243 | * Fixed by https://github.com/facebook/zstd/pull/1664.
|
| 244 | * This can happen when the previous sequences section block is compressed
|
| 245 | * with rle mode and the current block's sequences section is compressed
|
| 246 | * with repeat mode where sequences section body size can be 1 byte.
|
| 247 |
|
| 248 | #ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
|
| 249 | if (op-seqHead < 4) {
|
| 250 | DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
|
| 251 | "an uncompressed block when sequences are < 4 bytes");
|
| 252 | return 0;
|
| 253 | }
|
| 254 | #endif
|
| 255 |
|
| 256 | *entropyWritten = 1;
|
| 257 | return (size_t)(op - ostart);
|
| 258 | }
|
| 259 |
|
| 260 | |
| 261 | * Compresses a single sub-block.
|
| 262 | * @return : compressed size of the sub-block
|
| 263 | * Or 0 if it failed to compress.
|
| 264 | static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
|
| 265 | const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
|
| 266 | const SeqDef* sequences, size_t nbSeq,
|
| 267 | const BYTE* literals, size_t litSize,
|
| 268 | const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
|
| 269 | const ZSTD_CCtx_params* cctxParams,
|
| 270 | void* dst, size_t dstCapacity,
|
| 271 | const int bmi2,
|
| 272 | int writeLitEntropy, int writeSeqEntropy,
|
| 273 | int* litEntropyWritten, int* seqEntropyWritten,
|
| 274 | U32 lastBlock)
|
| 275 | {
|
| 276 | BYTE* const ostart = (BYTE*)dst;
|
| 277 | BYTE* const oend = ostart + dstCapacity;
|
| 278 | BYTE* op = ostart + ZSTD_blockHeaderSize;
|
| 279 | DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
|
| 280 | litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
|
| 281 | { size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
|
| 282 | &entropyMetadata->hufMetadata, literals, litSize,
|
| 283 | op, (size_t)(oend-op),
|
| 284 | bmi2, writeLitEntropy, litEntropyWritten);
|
| 285 | FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
|
| 286 | if (cLitSize == 0) return 0;
|
| 287 | op += cLitSize;
|
| 288 | }
|
| 289 | { size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
|
| 290 | &entropyMetadata->fseMetadata,
|
| 291 | sequences, nbSeq,
|
| 292 | llCode, mlCode, ofCode,
|
| 293 | cctxParams,
|
| 294 | op, (size_t)(oend-op),
|
| 295 | bmi2, writeSeqEntropy, seqEntropyWritten);
|
| 296 | FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
|
| 297 | if (cSeqSize == 0) return 0;
|
| 298 | op += cSeqSize;
|
| 299 | }
|
| 300 |
|
| 301 | { size_t cSize = (size_t)(op-ostart) - ZSTD_blockHeaderSize;
|
| 302 | U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
|
| 303 | MEM_writeLE24(ostart, cBlockHeader24);
|
| 304 | }
|
| 305 | return (size_t)(op-ostart);
|
| 306 | }
|
| 307 |
|
| 308 | static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
|
| 309 | const ZSTD_hufCTables_t* huf,
|
| 310 | const ZSTD_hufCTablesMetadata_t* hufMetadata,
|
| 311 | void* workspace, size_t wkspSize,
|
| 312 | int writeEntropy)
|
| 313 | {
|
| 314 | unsigned* const countWksp = (unsigned*)workspace;
|
| 315 | unsigned maxSymbolValue = 255;
|
| 316 | size_t literalSectionHeaderSize = 3;
|
| 317 |
|
| 318 | if (hufMetadata->hType == set_basic) return litSize;
|
| 319 | else if (hufMetadata->hType == set_rle) return 1;
|
| 320 | else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
|
| 321 | size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
|
| 322 | if (ZSTD_isError(largest)) return litSize;
|
| 323 | { size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
|
| 324 | if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
|
| 325 | return cLitSizeEstimate + literalSectionHeaderSize;
|
| 326 | } }
|
| 327 | assert(0);
|
| 328 | return 0;
|
| 329 | }
|
| 330 |
|
| 331 | static size_t ZSTD_estimateSubBlockSize_symbolType(SymbolEncodingType_e type,
|
| 332 | const BYTE* codeTable, unsigned maxCode,
|
| 333 | size_t nbSeq, const FSE_CTable* fseCTable,
|
| 334 | const U8* additionalBits,
|
| 335 | short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
|
| 336 | void* workspace, size_t wkspSize)
|
| 337 | {
|
| 338 | unsigned* const countWksp = (unsigned*)workspace;
|
| 339 | const BYTE* ctp = codeTable;
|
| 340 | const BYTE* const ctStart = ctp;
|
| 341 | const BYTE* const ctEnd = ctStart + nbSeq;
|
| 342 | size_t cSymbolTypeSizeEstimateInBits = 0;
|
| 343 | unsigned max = maxCode;
|
| 344 |
|
| 345 | HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);
|
| 346 | if (type == set_basic) {
|
| 347 |
|
| 348 | assert(max <= defaultMax);
|
| 349 | cSymbolTypeSizeEstimateInBits = max <= defaultMax
|
| 350 | ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
|
| 351 | : ERROR(GENERIC);
|
| 352 | } else if (type == set_rle) {
|
| 353 | cSymbolTypeSizeEstimateInBits = 0;
|
| 354 | } else if (type == set_compressed || type == set_repeat) {
|
| 355 | cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
|
| 356 | }
|
| 357 | if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
|
| 358 | while (ctp < ctEnd) {
|
| 359 | if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
|
| 360 | else cSymbolTypeSizeEstimateInBits += *ctp;
|
| 361 | ctp++;
|
| 362 | }
|
| 363 | return cSymbolTypeSizeEstimateInBits / 8;
|
| 364 | }
|
| 365 |
|
| 366 | static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
|
| 367 | const BYTE* llCodeTable,
|
| 368 | const BYTE* mlCodeTable,
|
| 369 | size_t nbSeq,
|
| 370 | const ZSTD_fseCTables_t* fseTables,
|
| 371 | const ZSTD_fseCTablesMetadata_t* fseMetadata,
|
| 372 | void* workspace, size_t wkspSize,
|
| 373 | int writeEntropy)
|
| 374 | {
|
| 375 | size_t const sequencesSectionHeaderSize = 3;
|
| 376 | size_t cSeqSizeEstimate = 0;
|
| 377 | if (nbSeq == 0) return sequencesSectionHeaderSize;
|
| 378 | cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
|
| 379 | nbSeq, fseTables->offcodeCTable, NULL,
|
| 380 | OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
|
| 381 | workspace, wkspSize);
|
| 382 | cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
|
| 383 | nbSeq, fseTables->litlengthCTable, LL_bits,
|
| 384 | LL_defaultNorm, LL_defaultNormLog, MaxLL,
|
| 385 | workspace, wkspSize);
|
| 386 | cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
|
| 387 | nbSeq, fseTables->matchlengthCTable, ML_bits,
|
| 388 | ML_defaultNorm, ML_defaultNormLog, MaxML,
|
| 389 | workspace, wkspSize);
|
| 390 | if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
|
| 391 | return cSeqSizeEstimate + sequencesSectionHeaderSize;
|
| 392 | }
|
| 393 |
|
| 394 | typedef struct {
|
| 395 | size_t estLitSize;
|
| 396 | size_t estBlockSize;
|
| 397 | } EstimatedBlockSize;
|
| 398 | static EstimatedBlockSize ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
|
| 399 | const BYTE* ofCodeTable,
|
| 400 | const BYTE* llCodeTable,
|
| 401 | const BYTE* mlCodeTable,
|
| 402 | size_t nbSeq,
|
| 403 | const ZSTD_entropyCTables_t* entropy,
|
| 404 | const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
|
| 405 | void* workspace, size_t wkspSize,
|
| 406 | int writeLitEntropy, int writeSeqEntropy)
|
| 407 | {
|
| 408 | EstimatedBlockSize ebs;
|
| 409 | ebs.estLitSize = ZSTD_estimateSubBlockSize_literal(literals, litSize,
|
| 410 | &entropy->huf, &entropyMetadata->hufMetadata,
|
| 411 | workspace, wkspSize, writeLitEntropy);
|
| 412 | ebs.estBlockSize = ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
|
| 413 | nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
|
| 414 | workspace, wkspSize, writeSeqEntropy);
|
| 415 | ebs.estBlockSize += ebs.estLitSize + ZSTD_blockHeaderSize;
|
| 416 | return ebs;
|
| 417 | }
|
| 418 |
|
| 419 | static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
|
| 420 | {
|
| 421 | if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
|
| 422 | return 1;
|
| 423 | if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
|
| 424 | return 1;
|
| 425 | if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
|
| 426 | return 1;
|
| 427 | return 0;
|
| 428 | }
|
| 429 |
|
| 430 | static size_t countLiterals(SeqStore_t const* seqStore, const SeqDef* sp, size_t seqCount)
|
| 431 | {
|
| 432 | size_t n, total = 0;
|
| 433 | assert(sp != NULL);
|
| 434 | for (n=0; n<seqCount; n++) {
|
| 435 | total += ZSTD_getSequenceLength(seqStore, sp+n).litLength;
|
| 436 | }
|
| 437 | DEBUGLOG(6, "countLiterals for %zu sequences from %p => %zu bytes", seqCount, (const void*)sp, total);
|
| 438 | return total;
|
| 439 | }
|
| 440 |
|
| 441 | #define BYTESCALE 256
|
| 442 |
|
| 443 | static size_t sizeBlockSequences(const SeqDef* sp, size_t nbSeqs,
|
| 444 | size_t targetBudget, size_t avgLitCost, size_t avgSeqCost,
|
| 445 | int firstSubBlock)
|
| 446 | {
|
| 447 | size_t n, budget = 0, inSize=0;
|
| 448 |
|
| 449 | size_t const headerSize = (size_t)firstSubBlock * 120 * BYTESCALE;
|
| 450 | assert(firstSubBlock==0 || firstSubBlock==1);
|
| 451 | budget += headerSize;
|
| 452 |
|
| 453 |
|
| 454 | budget += sp[0].litLength * avgLitCost + avgSeqCost;
|
| 455 | if (budget > targetBudget) return 1;
|
| 456 | inSize = sp[0].litLength + (sp[0].mlBase+MINMATCH);
|
| 457 |
|
| 458 |
|
| 459 | for (n=1; n<nbSeqs; n++) {
|
| 460 | size_t currentCost = sp[n].litLength * avgLitCost + avgSeqCost;
|
| 461 | budget += currentCost;
|
| 462 | inSize += sp[n].litLength + (sp[n].mlBase+MINMATCH);
|
| 463 |
|
| 464 | if ( (budget > targetBudget)
|
| 465 |
|
| 466 | && (budget < inSize * BYTESCALE) )
|
| 467 | break;
|
| 468 | }
|
| 469 |
|
| 470 | return n;
|
| 471 | }
|
| 472 |
|
| 473 | |
| 474 | * Breaks super-block into multiple sub-blocks and compresses them.
|
| 475 | * Entropy will be written into the first block.
|
| 476 | * The following blocks use repeat_mode to compress.
|
| 477 | * Sub-blocks are all compressed, except the last one when beneficial.
|
| 478 | * @return : compressed size of the super block (which features multiple ZSTD blocks)
|
| 479 | * or 0 if it failed to compress.
|
| 480 | static size_t ZSTD_compressSubBlock_multi(const SeqStore_t* seqStorePtr,
|
| 481 | const ZSTD_compressedBlockState_t* prevCBlock,
|
| 482 | ZSTD_compressedBlockState_t* nextCBlock,
|
| 483 | const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
|
| 484 | const ZSTD_CCtx_params* cctxParams,
|
| 485 | void* dst, size_t dstCapacity,
|
| 486 | const void* src, size_t srcSize,
|
| 487 | const int bmi2, U32 lastBlock,
|
| 488 | void* workspace, size_t wkspSize)
|
| 489 | {
|
| 490 | const SeqDef* const sstart = seqStorePtr->sequencesStart;
|
| 491 | const SeqDef* const send = seqStorePtr->sequences;
|
| 492 | const SeqDef* sp = sstart;
|
| 493 | size_t const nbSeqs = (size_t)(send - sstart);
|
| 494 | const BYTE* const lstart = seqStorePtr->litStart;
|
| 495 | const BYTE* const lend = seqStorePtr->lit;
|
| 496 | const BYTE* lp = lstart;
|
| 497 | size_t const nbLiterals = (size_t)(lend - lstart);
|
| 498 | BYTE const* ip = (BYTE const*)src;
|
| 499 | BYTE const* const iend = ip + srcSize;
|
| 500 | BYTE* const ostart = (BYTE*)dst;
|
| 501 | BYTE* const oend = ostart + dstCapacity;
|
| 502 | BYTE* op = ostart;
|
| 503 | const BYTE* llCodePtr = seqStorePtr->llCode;
|
| 504 | const BYTE* mlCodePtr = seqStorePtr->mlCode;
|
| 505 | const BYTE* ofCodePtr = seqStorePtr->ofCode;
|
| 506 | size_t const minTarget = ZSTD_TARGETCBLOCKSIZE_MIN;
|
| 507 | size_t const targetCBlockSize = MAX(minTarget, cctxParams->targetCBlockSize);
|
| 508 | int writeLitEntropy = (entropyMetadata->hufMetadata.hType == set_compressed);
|
| 509 | int writeSeqEntropy = 1;
|
| 510 |
|
| 511 | DEBUGLOG(5, "ZSTD_compressSubBlock_multi (srcSize=%u, litSize=%u, nbSeq=%u)",
|
| 512 | (unsigned)srcSize, (unsigned)(lend-lstart), (unsigned)(send-sstart));
|
| 513 |
|
| 514 |
|
| 515 | if (nbSeqs > 0) {
|
| 516 | EstimatedBlockSize const ebs =
|
| 517 | ZSTD_estimateSubBlockSize(lp, nbLiterals,
|
| 518 | ofCodePtr, llCodePtr, mlCodePtr, nbSeqs,
|
| 519 | &nextCBlock->entropy, entropyMetadata,
|
| 520 | workspace, wkspSize,
|
| 521 | writeLitEntropy, writeSeqEntropy);
|
| 522 |
|
| 523 | size_t const avgLitCost = nbLiterals ? (ebs.estLitSize * BYTESCALE) / nbLiterals : BYTESCALE;
|
| 524 | size_t const avgSeqCost = ((ebs.estBlockSize - ebs.estLitSize) * BYTESCALE) / nbSeqs;
|
| 525 | const size_t nbSubBlocks = MAX((ebs.estBlockSize + (targetCBlockSize/2)) / targetCBlockSize, 1);
|
| 526 | size_t n, avgBlockBudget, blockBudgetSupp=0;
|
| 527 | avgBlockBudget = (ebs.estBlockSize * BYTESCALE) / nbSubBlocks;
|
| 528 | DEBUGLOG(5, "estimated fullblock size=%u bytes ; avgLitCost=%.2f ; avgSeqCost=%.2f ; targetCBlockSize=%u, nbSubBlocks=%u ; avgBlockBudget=%.0f bytes",
|
| 529 | (unsigned)ebs.estBlockSize, (double)avgLitCost/BYTESCALE, (double)avgSeqCost/BYTESCALE,
|
| 530 | (unsigned)targetCBlockSize, (unsigned)nbSubBlocks, (double)avgBlockBudget/BYTESCALE);
|
| 531 | |
| 532 | * this will result in the production of a single uncompressed block covering @srcSize.
|
| 533 | if (ebs.estBlockSize > srcSize) return 0;
|
| 534 |
|
| 535 |
|
| 536 | assert(nbSubBlocks>0);
|
| 537 | for (n=0; n < nbSubBlocks-1; n++) {
|
| 538 |
|
| 539 | size_t const seqCount = sizeBlockSequences(sp, (size_t)(send-sp),
|
| 540 | avgBlockBudget + blockBudgetSupp, avgLitCost, avgSeqCost, n==0);
|
| 541 |
|
| 542 | assert(seqCount <= (size_t)(send-sp));
|
| 543 | if (sp + seqCount == send) break;
|
| 544 | assert(seqCount > 0);
|
| 545 |
|
| 546 | { int litEntropyWritten = 0;
|
| 547 | int seqEntropyWritten = 0;
|
| 548 | size_t litSize = countLiterals(seqStorePtr, sp, seqCount);
|
| 549 | const size_t decompressedSize =
|
| 550 | ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 0);
|
| 551 | size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
|
| 552 | sp, seqCount,
|
| 553 | lp, litSize,
|
| 554 | llCodePtr, mlCodePtr, ofCodePtr,
|
| 555 | cctxParams,
|
| 556 | op, (size_t)(oend-op),
|
| 557 | bmi2, writeLitEntropy, writeSeqEntropy,
|
| 558 | &litEntropyWritten, &seqEntropyWritten,
|
| 559 | 0);
|
| 560 | FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
|
| 561 |
|
| 562 |
|
| 563 | if (cSize > 0 && cSize < decompressedSize) {
|
| 564 | DEBUGLOG(5, "Committed sub-block compressing %u bytes => %u bytes",
|
| 565 | (unsigned)decompressedSize, (unsigned)cSize);
|
| 566 | assert(ip + decompressedSize <= iend);
|
| 567 | ip += decompressedSize;
|
| 568 | lp += litSize;
|
| 569 | op += cSize;
|
| 570 | llCodePtr += seqCount;
|
| 571 | mlCodePtr += seqCount;
|
| 572 | ofCodePtr += seqCount;
|
| 573 |
|
| 574 | if (litEntropyWritten) {
|
| 575 | writeLitEntropy = 0;
|
| 576 | }
|
| 577 | if (seqEntropyWritten) {
|
| 578 | writeSeqEntropy = 0;
|
| 579 | }
|
| 580 | sp += seqCount;
|
| 581 | blockBudgetSupp = 0;
|
| 582 | } }
|
| 583 |
|
| 584 | }
|
| 585 | }
|
| 586 |
|
| 587 |
|
| 588 | DEBUGLOG(5, "Generate last sub-block: %u sequences remaining", (unsigned)(send - sp));
|
| 589 | { int litEntropyWritten = 0;
|
| 590 | int seqEntropyWritten = 0;
|
| 591 | size_t litSize = (size_t)(lend - lp);
|
| 592 | size_t seqCount = (size_t)(send - sp);
|
| 593 | const size_t decompressedSize =
|
| 594 | ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, 1);
|
| 595 | size_t const cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
|
| 596 | sp, seqCount,
|
| 597 | lp, litSize,
|
| 598 | llCodePtr, mlCodePtr, ofCodePtr,
|
| 599 | cctxParams,
|
| 600 | op, (size_t)(oend-op),
|
| 601 | bmi2, writeLitEntropy, writeSeqEntropy,
|
| 602 | &litEntropyWritten, &seqEntropyWritten,
|
| 603 | lastBlock);
|
| 604 | FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
|
| 605 |
|
| 606 |
|
| 607 | if (cSize > 0 && cSize < decompressedSize) {
|
| 608 | DEBUGLOG(5, "Last sub-block compressed %u bytes => %u bytes",
|
| 609 | (unsigned)decompressedSize, (unsigned)cSize);
|
| 610 | assert(ip + decompressedSize <= iend);
|
| 611 | ip += decompressedSize;
|
| 612 | lp += litSize;
|
| 613 | op += cSize;
|
| 614 | llCodePtr += seqCount;
|
| 615 | mlCodePtr += seqCount;
|
| 616 | ofCodePtr += seqCount;
|
| 617 |
|
| 618 | if (litEntropyWritten) {
|
| 619 | writeLitEntropy = 0;
|
| 620 | }
|
| 621 | if (seqEntropyWritten) {
|
| 622 | writeSeqEntropy = 0;
|
| 623 | }
|
| 624 | sp += seqCount;
|
| 625 | }
|
| 626 | }
|
| 627 |
|
| 628 |
|
| 629 | if (writeLitEntropy) {
|
| 630 | DEBUGLOG(5, "Literal entropy tables were never written");
|
| 631 | ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
|
| 632 | }
|
| 633 | if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
|
| 634 | |
| 635 | * must emit an uncompressed block.
|
| 636 |
|
| 637 | DEBUGLOG(5, "Sequence entropy tables were never written => cancel, emit an uncompressed block");
|
| 638 | return 0;
|
| 639 | }
|
| 640 |
|
| 641 | if (ip < iend) {
|
| 642 |
|
| 643 | size_t const rSize = (size_t)((iend - ip));
|
| 644 | size_t const cSize = ZSTD_noCompressBlock(op, (size_t)(oend - op), ip, rSize, lastBlock);
|
| 645 | DEBUGLOG(5, "Generate last uncompressed sub-block of %u bytes", (unsigned)(rSize));
|
| 646 | FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
|
| 647 | assert(cSize != 0);
|
| 648 | op += cSize;
|
| 649 |
|
| 650 | if (sp < send) {
|
| 651 | const SeqDef* seq;
|
| 652 | Repcodes_t rep;
|
| 653 | ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
|
| 654 | for (seq = sstart; seq < sp; ++seq) {
|
| 655 | ZSTD_updateRep(rep.rep, seq->offBase, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
|
| 656 | }
|
| 657 | ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
|
| 658 | }
|
| 659 | }
|
| 660 |
|
| 661 | DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed all subBlocks: total compressed size = %u",
|
| 662 | (unsigned)(op-ostart));
|
| 663 | return (size_t)(op-ostart);
|
| 664 | }
|
| 665 |
|
| 666 | size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
|
| 667 | void* dst, size_t dstCapacity,
|
| 668 | const void* src, size_t srcSize,
|
| 669 | unsigned lastBlock)
|
| 670 | {
|
| 671 | ZSTD_entropyCTablesMetadata_t entropyMetadata;
|
| 672 |
|
| 673 | FORWARD_IF_ERROR(ZSTD_buildBlockEntropyStats(&zc->seqStore,
|
| 674 | &zc->blockState.prevCBlock->entropy,
|
| 675 | &zc->blockState.nextCBlock->entropy,
|
| 676 | &zc->appliedParams,
|
| 677 | &entropyMetadata,
|
| 678 | zc->tmpWorkspace, zc->tmpWkspSize ), "");
|
| 679 |
|
| 680 | return ZSTD_compressSubBlock_multi(&zc->seqStore,
|
| 681 | zc->blockState.prevCBlock,
|
| 682 | zc->blockState.nextCBlock,
|
| 683 | &entropyMetadata,
|
| 684 | &zc->appliedParams,
|
| 685 | dst, dstCapacity,
|
| 686 | src, srcSize,
|
| 687 | zc->bmi2, lastBlock,
|
| 688 | zc->tmpWorkspace, zc->tmpWkspSize );
|
| 689 | }
|