Scott Baker | 8461e15 | 2019-10-01 14:44:30 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2016-present, 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 | /* zstd_decompress_block : |
| 12 | * this module takes care of decompressing _compressed_ block */ |
| 13 | |
| 14 | /*-******************************************************* |
| 15 | * Dependencies |
| 16 | *********************************************************/ |
| 17 | #include <string.h> /* memcpy, memmove, memset */ |
| 18 | #include "compiler.h" /* prefetch */ |
| 19 | #include "cpu.h" /* bmi2 */ |
| 20 | #include "mem.h" /* low level memory routines */ |
| 21 | #define FSE_STATIC_LINKING_ONLY |
| 22 | #include "fse.h" |
| 23 | #define HUF_STATIC_LINKING_ONLY |
| 24 | #include "huf.h" |
| 25 | #include "zstd_internal.h" |
| 26 | #include "zstd_decompress_internal.h" /* ZSTD_DCtx */ |
| 27 | #include "zstd_ddict.h" /* ZSTD_DDictDictContent */ |
| 28 | #include "zstd_decompress_block.h" |
| 29 | |
| 30 | /*_******************************************************* |
| 31 | * Macros |
| 32 | **********************************************************/ |
| 33 | |
| 34 | /* These two optional macros force the use one way or another of the two |
| 35 | * ZSTD_decompressSequences implementations. You can't force in both directions |
| 36 | * at the same time. |
| 37 | */ |
| 38 | #if defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ |
| 39 | defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) |
| 40 | #error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!" |
| 41 | #endif |
| 42 | |
| 43 | |
| 44 | /*_******************************************************* |
| 45 | * Memory operations |
| 46 | **********************************************************/ |
| 47 | static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); } |
| 48 | |
| 49 | |
| 50 | /*-************************************************************* |
| 51 | * Block decoding |
| 52 | ***************************************************************/ |
| 53 | |
| 54 | /*! ZSTD_getcBlockSize() : |
| 55 | * Provides the size of compressed block from block header `src` */ |
| 56 | size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, |
| 57 | blockProperties_t* bpPtr) |
| 58 | { |
| 59 | RETURN_ERROR_IF(srcSize < ZSTD_blockHeaderSize, srcSize_wrong); |
| 60 | |
| 61 | { U32 const cBlockHeader = MEM_readLE24(src); |
| 62 | U32 const cSize = cBlockHeader >> 3; |
| 63 | bpPtr->lastBlock = cBlockHeader & 1; |
| 64 | bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3); |
| 65 | bpPtr->origSize = cSize; /* only useful for RLE */ |
| 66 | if (bpPtr->blockType == bt_rle) return 1; |
| 67 | RETURN_ERROR_IF(bpPtr->blockType == bt_reserved, corruption_detected); |
| 68 | return cSize; |
| 69 | } |
| 70 | } |
| 71 | |
| 72 | |
| 73 | /* Hidden declaration for fullbench */ |
| 74 | size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx, |
| 75 | const void* src, size_t srcSize); |
| 76 | /*! ZSTD_decodeLiteralsBlock() : |
| 77 | * @return : nb of bytes read from src (< srcSize ) |
| 78 | * note : symbol not declared but exposed for fullbench */ |
| 79 | size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx, |
| 80 | const void* src, size_t srcSize) /* note : srcSize < BLOCKSIZE */ |
| 81 | { |
| 82 | RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected); |
| 83 | |
| 84 | { const BYTE* const istart = (const BYTE*) src; |
| 85 | symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3); |
| 86 | |
| 87 | switch(litEncType) |
| 88 | { |
| 89 | case set_repeat: |
| 90 | RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted); |
| 91 | /* fall-through */ |
| 92 | |
| 93 | case set_compressed: |
| 94 | RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3"); |
| 95 | { size_t lhSize, litSize, litCSize; |
| 96 | U32 singleStream=0; |
| 97 | U32 const lhlCode = (istart[0] >> 2) & 3; |
| 98 | U32 const lhc = MEM_readLE32(istart); |
| 99 | size_t hufSuccess; |
| 100 | switch(lhlCode) |
| 101 | { |
| 102 | case 0: case 1: default: /* note : default is impossible, since lhlCode into [0..3] */ |
| 103 | /* 2 - 2 - 10 - 10 */ |
| 104 | singleStream = !lhlCode; |
| 105 | lhSize = 3; |
| 106 | litSize = (lhc >> 4) & 0x3FF; |
| 107 | litCSize = (lhc >> 14) & 0x3FF; |
| 108 | break; |
| 109 | case 2: |
| 110 | /* 2 - 2 - 14 - 14 */ |
| 111 | lhSize = 4; |
| 112 | litSize = (lhc >> 4) & 0x3FFF; |
| 113 | litCSize = lhc >> 18; |
| 114 | break; |
| 115 | case 3: |
| 116 | /* 2 - 2 - 18 - 18 */ |
| 117 | lhSize = 5; |
| 118 | litSize = (lhc >> 4) & 0x3FFFF; |
| 119 | litCSize = (lhc >> 22) + (istart[4] << 10); |
| 120 | break; |
| 121 | } |
| 122 | RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected); |
| 123 | RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected); |
| 124 | |
| 125 | /* prefetch huffman table if cold */ |
| 126 | if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) { |
| 127 | PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable)); |
| 128 | } |
| 129 | |
| 130 | if (litEncType==set_repeat) { |
| 131 | if (singleStream) { |
| 132 | hufSuccess = HUF_decompress1X_usingDTable_bmi2( |
| 133 | dctx->litBuffer, litSize, istart+lhSize, litCSize, |
| 134 | dctx->HUFptr, dctx->bmi2); |
| 135 | } else { |
| 136 | hufSuccess = HUF_decompress4X_usingDTable_bmi2( |
| 137 | dctx->litBuffer, litSize, istart+lhSize, litCSize, |
| 138 | dctx->HUFptr, dctx->bmi2); |
| 139 | } |
| 140 | } else { |
| 141 | if (singleStream) { |
| 142 | #if defined(HUF_FORCE_DECOMPRESS_X2) |
| 143 | hufSuccess = HUF_decompress1X_DCtx_wksp( |
| 144 | dctx->entropy.hufTable, dctx->litBuffer, litSize, |
| 145 | istart+lhSize, litCSize, dctx->workspace, |
| 146 | sizeof(dctx->workspace)); |
| 147 | #else |
| 148 | hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2( |
| 149 | dctx->entropy.hufTable, dctx->litBuffer, litSize, |
| 150 | istart+lhSize, litCSize, dctx->workspace, |
| 151 | sizeof(dctx->workspace), dctx->bmi2); |
| 152 | #endif |
| 153 | } else { |
| 154 | hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2( |
| 155 | dctx->entropy.hufTable, dctx->litBuffer, litSize, |
| 156 | istart+lhSize, litCSize, dctx->workspace, |
| 157 | sizeof(dctx->workspace), dctx->bmi2); |
| 158 | } |
| 159 | } |
| 160 | |
| 161 | RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected); |
| 162 | |
| 163 | dctx->litPtr = dctx->litBuffer; |
| 164 | dctx->litSize = litSize; |
| 165 | dctx->litEntropy = 1; |
| 166 | if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable; |
| 167 | memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH); |
| 168 | return litCSize + lhSize; |
| 169 | } |
| 170 | |
| 171 | case set_basic: |
| 172 | { size_t litSize, lhSize; |
| 173 | U32 const lhlCode = ((istart[0]) >> 2) & 3; |
| 174 | switch(lhlCode) |
| 175 | { |
| 176 | case 0: case 2: default: /* note : default is impossible, since lhlCode into [0..3] */ |
| 177 | lhSize = 1; |
| 178 | litSize = istart[0] >> 3; |
| 179 | break; |
| 180 | case 1: |
| 181 | lhSize = 2; |
| 182 | litSize = MEM_readLE16(istart) >> 4; |
| 183 | break; |
| 184 | case 3: |
| 185 | lhSize = 3; |
| 186 | litSize = MEM_readLE24(istart) >> 4; |
| 187 | break; |
| 188 | } |
| 189 | |
| 190 | if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) { /* risk reading beyond src buffer with wildcopy */ |
| 191 | RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected); |
| 192 | memcpy(dctx->litBuffer, istart+lhSize, litSize); |
| 193 | dctx->litPtr = dctx->litBuffer; |
| 194 | dctx->litSize = litSize; |
| 195 | memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH); |
| 196 | return lhSize+litSize; |
| 197 | } |
| 198 | /* direct reference into compressed stream */ |
| 199 | dctx->litPtr = istart+lhSize; |
| 200 | dctx->litSize = litSize; |
| 201 | return lhSize+litSize; |
| 202 | } |
| 203 | |
| 204 | case set_rle: |
| 205 | { U32 const lhlCode = ((istart[0]) >> 2) & 3; |
| 206 | size_t litSize, lhSize; |
| 207 | switch(lhlCode) |
| 208 | { |
| 209 | case 0: case 2: default: /* note : default is impossible, since lhlCode into [0..3] */ |
| 210 | lhSize = 1; |
| 211 | litSize = istart[0] >> 3; |
| 212 | break; |
| 213 | case 1: |
| 214 | lhSize = 2; |
| 215 | litSize = MEM_readLE16(istart) >> 4; |
| 216 | break; |
| 217 | case 3: |
| 218 | lhSize = 3; |
| 219 | litSize = MEM_readLE24(istart) >> 4; |
| 220 | RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4"); |
| 221 | break; |
| 222 | } |
| 223 | RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected); |
| 224 | memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH); |
| 225 | dctx->litPtr = dctx->litBuffer; |
| 226 | dctx->litSize = litSize; |
| 227 | return lhSize+1; |
| 228 | } |
| 229 | default: |
| 230 | RETURN_ERROR(corruption_detected, "impossible"); |
| 231 | } |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | /* Default FSE distribution tables. |
| 236 | * These are pre-calculated FSE decoding tables using default distributions as defined in specification : |
| 237 | * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#default-distributions |
| 238 | * They were generated programmatically with following method : |
| 239 | * - start from default distributions, present in /lib/common/zstd_internal.h |
| 240 | * - generate tables normally, using ZSTD_buildFSETable() |
| 241 | * - printout the content of tables |
| 242 | * - pretify output, report below, test with fuzzer to ensure it's correct */ |
| 243 | |
| 244 | /* Default FSE distribution table for Literal Lengths */ |
| 245 | static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = { |
| 246 | { 1, 1, 1, LL_DEFAULTNORMLOG}, /* header : fastMode, tableLog */ |
| 247 | /* nextState, nbAddBits, nbBits, baseVal */ |
| 248 | { 0, 0, 4, 0}, { 16, 0, 4, 0}, |
| 249 | { 32, 0, 5, 1}, { 0, 0, 5, 3}, |
| 250 | { 0, 0, 5, 4}, { 0, 0, 5, 6}, |
| 251 | { 0, 0, 5, 7}, { 0, 0, 5, 9}, |
| 252 | { 0, 0, 5, 10}, { 0, 0, 5, 12}, |
| 253 | { 0, 0, 6, 14}, { 0, 1, 5, 16}, |
| 254 | { 0, 1, 5, 20}, { 0, 1, 5, 22}, |
| 255 | { 0, 2, 5, 28}, { 0, 3, 5, 32}, |
| 256 | { 0, 4, 5, 48}, { 32, 6, 5, 64}, |
| 257 | { 0, 7, 5, 128}, { 0, 8, 6, 256}, |
| 258 | { 0, 10, 6, 1024}, { 0, 12, 6, 4096}, |
| 259 | { 32, 0, 4, 0}, { 0, 0, 4, 1}, |
| 260 | { 0, 0, 5, 2}, { 32, 0, 5, 4}, |
| 261 | { 0, 0, 5, 5}, { 32, 0, 5, 7}, |
| 262 | { 0, 0, 5, 8}, { 32, 0, 5, 10}, |
| 263 | { 0, 0, 5, 11}, { 0, 0, 6, 13}, |
| 264 | { 32, 1, 5, 16}, { 0, 1, 5, 18}, |
| 265 | { 32, 1, 5, 22}, { 0, 2, 5, 24}, |
| 266 | { 32, 3, 5, 32}, { 0, 3, 5, 40}, |
| 267 | { 0, 6, 4, 64}, { 16, 6, 4, 64}, |
| 268 | { 32, 7, 5, 128}, { 0, 9, 6, 512}, |
| 269 | { 0, 11, 6, 2048}, { 48, 0, 4, 0}, |
| 270 | { 16, 0, 4, 1}, { 32, 0, 5, 2}, |
| 271 | { 32, 0, 5, 3}, { 32, 0, 5, 5}, |
| 272 | { 32, 0, 5, 6}, { 32, 0, 5, 8}, |
| 273 | { 32, 0, 5, 9}, { 32, 0, 5, 11}, |
| 274 | { 32, 0, 5, 12}, { 0, 0, 6, 15}, |
| 275 | { 32, 1, 5, 18}, { 32, 1, 5, 20}, |
| 276 | { 32, 2, 5, 24}, { 32, 2, 5, 28}, |
| 277 | { 32, 3, 5, 40}, { 32, 4, 5, 48}, |
| 278 | { 0, 16, 6,65536}, { 0, 15, 6,32768}, |
| 279 | { 0, 14, 6,16384}, { 0, 13, 6, 8192}, |
| 280 | }; /* LL_defaultDTable */ |
| 281 | |
| 282 | /* Default FSE distribution table for Offset Codes */ |
| 283 | static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = { |
| 284 | { 1, 1, 1, OF_DEFAULTNORMLOG}, /* header : fastMode, tableLog */ |
| 285 | /* nextState, nbAddBits, nbBits, baseVal */ |
| 286 | { 0, 0, 5, 0}, { 0, 6, 4, 61}, |
| 287 | { 0, 9, 5, 509}, { 0, 15, 5,32765}, |
| 288 | { 0, 21, 5,2097149}, { 0, 3, 5, 5}, |
| 289 | { 0, 7, 4, 125}, { 0, 12, 5, 4093}, |
| 290 | { 0, 18, 5,262141}, { 0, 23, 5,8388605}, |
| 291 | { 0, 5, 5, 29}, { 0, 8, 4, 253}, |
| 292 | { 0, 14, 5,16381}, { 0, 20, 5,1048573}, |
| 293 | { 0, 2, 5, 1}, { 16, 7, 4, 125}, |
| 294 | { 0, 11, 5, 2045}, { 0, 17, 5,131069}, |
| 295 | { 0, 22, 5,4194301}, { 0, 4, 5, 13}, |
| 296 | { 16, 8, 4, 253}, { 0, 13, 5, 8189}, |
| 297 | { 0, 19, 5,524285}, { 0, 1, 5, 1}, |
| 298 | { 16, 6, 4, 61}, { 0, 10, 5, 1021}, |
| 299 | { 0, 16, 5,65533}, { 0, 28, 5,268435453}, |
| 300 | { 0, 27, 5,134217725}, { 0, 26, 5,67108861}, |
| 301 | { 0, 25, 5,33554429}, { 0, 24, 5,16777213}, |
| 302 | }; /* OF_defaultDTable */ |
| 303 | |
| 304 | |
| 305 | /* Default FSE distribution table for Match Lengths */ |
| 306 | static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = { |
| 307 | { 1, 1, 1, ML_DEFAULTNORMLOG}, /* header : fastMode, tableLog */ |
| 308 | /* nextState, nbAddBits, nbBits, baseVal */ |
| 309 | { 0, 0, 6, 3}, { 0, 0, 4, 4}, |
| 310 | { 32, 0, 5, 5}, { 0, 0, 5, 6}, |
| 311 | { 0, 0, 5, 8}, { 0, 0, 5, 9}, |
| 312 | { 0, 0, 5, 11}, { 0, 0, 6, 13}, |
| 313 | { 0, 0, 6, 16}, { 0, 0, 6, 19}, |
| 314 | { 0, 0, 6, 22}, { 0, 0, 6, 25}, |
| 315 | { 0, 0, 6, 28}, { 0, 0, 6, 31}, |
| 316 | { 0, 0, 6, 34}, { 0, 1, 6, 37}, |
| 317 | { 0, 1, 6, 41}, { 0, 2, 6, 47}, |
| 318 | { 0, 3, 6, 59}, { 0, 4, 6, 83}, |
| 319 | { 0, 7, 6, 131}, { 0, 9, 6, 515}, |
| 320 | { 16, 0, 4, 4}, { 0, 0, 4, 5}, |
| 321 | { 32, 0, 5, 6}, { 0, 0, 5, 7}, |
| 322 | { 32, 0, 5, 9}, { 0, 0, 5, 10}, |
| 323 | { 0, 0, 6, 12}, { 0, 0, 6, 15}, |
| 324 | { 0, 0, 6, 18}, { 0, 0, 6, 21}, |
| 325 | { 0, 0, 6, 24}, { 0, 0, 6, 27}, |
| 326 | { 0, 0, 6, 30}, { 0, 0, 6, 33}, |
| 327 | { 0, 1, 6, 35}, { 0, 1, 6, 39}, |
| 328 | { 0, 2, 6, 43}, { 0, 3, 6, 51}, |
| 329 | { 0, 4, 6, 67}, { 0, 5, 6, 99}, |
| 330 | { 0, 8, 6, 259}, { 32, 0, 4, 4}, |
| 331 | { 48, 0, 4, 4}, { 16, 0, 4, 5}, |
| 332 | { 32, 0, 5, 7}, { 32, 0, 5, 8}, |
| 333 | { 32, 0, 5, 10}, { 32, 0, 5, 11}, |
| 334 | { 0, 0, 6, 14}, { 0, 0, 6, 17}, |
| 335 | { 0, 0, 6, 20}, { 0, 0, 6, 23}, |
| 336 | { 0, 0, 6, 26}, { 0, 0, 6, 29}, |
| 337 | { 0, 0, 6, 32}, { 0, 16, 6,65539}, |
| 338 | { 0, 15, 6,32771}, { 0, 14, 6,16387}, |
| 339 | { 0, 13, 6, 8195}, { 0, 12, 6, 4099}, |
| 340 | { 0, 11, 6, 2051}, { 0, 10, 6, 1027}, |
| 341 | }; /* ML_defaultDTable */ |
| 342 | |
| 343 | |
| 344 | static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddBits) |
| 345 | { |
| 346 | void* ptr = dt; |
| 347 | ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr; |
| 348 | ZSTD_seqSymbol* const cell = dt + 1; |
| 349 | |
| 350 | DTableH->tableLog = 0; |
| 351 | DTableH->fastMode = 0; |
| 352 | |
| 353 | cell->nbBits = 0; |
| 354 | cell->nextState = 0; |
| 355 | assert(nbAddBits < 255); |
| 356 | cell->nbAdditionalBits = (BYTE)nbAddBits; |
| 357 | cell->baseValue = baseValue; |
| 358 | } |
| 359 | |
| 360 | |
| 361 | /* ZSTD_buildFSETable() : |
| 362 | * generate FSE decoding table for one symbol (ll, ml or off) |
| 363 | * cannot fail if input is valid => |
| 364 | * all inputs are presumed validated at this stage */ |
| 365 | void |
| 366 | ZSTD_buildFSETable(ZSTD_seqSymbol* dt, |
| 367 | const short* normalizedCounter, unsigned maxSymbolValue, |
| 368 | const U32* baseValue, const U32* nbAdditionalBits, |
| 369 | unsigned tableLog) |
| 370 | { |
| 371 | ZSTD_seqSymbol* const tableDecode = dt+1; |
| 372 | U16 symbolNext[MaxSeq+1]; |
| 373 | |
| 374 | U32 const maxSV1 = maxSymbolValue + 1; |
| 375 | U32 const tableSize = 1 << tableLog; |
| 376 | U32 highThreshold = tableSize-1; |
| 377 | |
| 378 | /* Sanity Checks */ |
| 379 | assert(maxSymbolValue <= MaxSeq); |
| 380 | assert(tableLog <= MaxFSELog); |
| 381 | |
| 382 | /* Init, lay down lowprob symbols */ |
| 383 | { ZSTD_seqSymbol_header DTableH; |
| 384 | DTableH.tableLog = tableLog; |
| 385 | DTableH.fastMode = 1; |
| 386 | { S16 const largeLimit= (S16)(1 << (tableLog-1)); |
| 387 | U32 s; |
| 388 | for (s=0; s<maxSV1; s++) { |
| 389 | if (normalizedCounter[s]==-1) { |
| 390 | tableDecode[highThreshold--].baseValue = s; |
| 391 | symbolNext[s] = 1; |
| 392 | } else { |
| 393 | if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0; |
| 394 | symbolNext[s] = normalizedCounter[s]; |
| 395 | } } } |
| 396 | memcpy(dt, &DTableH, sizeof(DTableH)); |
| 397 | } |
| 398 | |
| 399 | /* Spread symbols */ |
| 400 | { U32 const tableMask = tableSize-1; |
| 401 | U32 const step = FSE_TABLESTEP(tableSize); |
| 402 | U32 s, position = 0; |
| 403 | for (s=0; s<maxSV1; s++) { |
| 404 | int i; |
| 405 | for (i=0; i<normalizedCounter[s]; i++) { |
| 406 | tableDecode[position].baseValue = s; |
| 407 | position = (position + step) & tableMask; |
| 408 | while (position > highThreshold) position = (position + step) & tableMask; /* lowprob area */ |
| 409 | } } |
| 410 | assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */ |
| 411 | } |
| 412 | |
| 413 | /* Build Decoding table */ |
| 414 | { U32 u; |
| 415 | for (u=0; u<tableSize; u++) { |
| 416 | U32 const symbol = tableDecode[u].baseValue; |
| 417 | U32 const nextState = symbolNext[symbol]++; |
| 418 | tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) ); |
| 419 | tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize); |
| 420 | assert(nbAdditionalBits[symbol] < 255); |
| 421 | tableDecode[u].nbAdditionalBits = (BYTE)nbAdditionalBits[symbol]; |
| 422 | tableDecode[u].baseValue = baseValue[symbol]; |
| 423 | } } |
| 424 | } |
| 425 | |
| 426 | |
| 427 | /*! ZSTD_buildSeqTable() : |
| 428 | * @return : nb bytes read from src, |
| 429 | * or an error code if it fails */ |
| 430 | static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr, |
| 431 | symbolEncodingType_e type, unsigned max, U32 maxLog, |
| 432 | const void* src, size_t srcSize, |
| 433 | const U32* baseValue, const U32* nbAdditionalBits, |
| 434 | const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable, |
| 435 | int ddictIsCold, int nbSeq) |
| 436 | { |
| 437 | switch(type) |
| 438 | { |
| 439 | case set_rle : |
| 440 | RETURN_ERROR_IF(!srcSize, srcSize_wrong); |
| 441 | RETURN_ERROR_IF((*(const BYTE*)src) > max, corruption_detected); |
| 442 | { U32 const symbol = *(const BYTE*)src; |
| 443 | U32 const baseline = baseValue[symbol]; |
| 444 | U32 const nbBits = nbAdditionalBits[symbol]; |
| 445 | ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits); |
| 446 | } |
| 447 | *DTablePtr = DTableSpace; |
| 448 | return 1; |
| 449 | case set_basic : |
| 450 | *DTablePtr = defaultTable; |
| 451 | return 0; |
| 452 | case set_repeat: |
| 453 | RETURN_ERROR_IF(!flagRepeatTable, corruption_detected); |
| 454 | /* prefetch FSE table if used */ |
| 455 | if (ddictIsCold && (nbSeq > 24 /* heuristic */)) { |
| 456 | const void* const pStart = *DTablePtr; |
| 457 | size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog)); |
| 458 | PREFETCH_AREA(pStart, pSize); |
| 459 | } |
| 460 | return 0; |
| 461 | case set_compressed : |
| 462 | { unsigned tableLog; |
| 463 | S16 norm[MaxSeq+1]; |
| 464 | size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize); |
| 465 | RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected); |
| 466 | RETURN_ERROR_IF(tableLog > maxLog, corruption_detected); |
| 467 | ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog); |
| 468 | *DTablePtr = DTableSpace; |
| 469 | return headerSize; |
| 470 | } |
| 471 | default : |
| 472 | assert(0); |
| 473 | RETURN_ERROR(GENERIC, "impossible"); |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr, |
| 478 | const void* src, size_t srcSize) |
| 479 | { |
| 480 | const BYTE* const istart = (const BYTE* const)src; |
| 481 | const BYTE* const iend = istart + srcSize; |
| 482 | const BYTE* ip = istart; |
| 483 | int nbSeq; |
| 484 | DEBUGLOG(5, "ZSTD_decodeSeqHeaders"); |
| 485 | |
| 486 | /* check */ |
| 487 | RETURN_ERROR_IF(srcSize < MIN_SEQUENCES_SIZE, srcSize_wrong); |
| 488 | |
| 489 | /* SeqHead */ |
| 490 | nbSeq = *ip++; |
| 491 | if (!nbSeq) { |
| 492 | *nbSeqPtr=0; |
| 493 | RETURN_ERROR_IF(srcSize != 1, srcSize_wrong); |
| 494 | return 1; |
| 495 | } |
| 496 | if (nbSeq > 0x7F) { |
| 497 | if (nbSeq == 0xFF) { |
| 498 | RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong); |
| 499 | nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2; |
| 500 | } else { |
| 501 | RETURN_ERROR_IF(ip >= iend, srcSize_wrong); |
| 502 | nbSeq = ((nbSeq-0x80)<<8) + *ip++; |
| 503 | } |
| 504 | } |
| 505 | *nbSeqPtr = nbSeq; |
| 506 | |
| 507 | /* FSE table descriptors */ |
| 508 | RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong); /* minimum possible size: 1 byte for symbol encoding types */ |
| 509 | { symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6); |
| 510 | symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3); |
| 511 | symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3); |
| 512 | ip++; |
| 513 | |
| 514 | /* Build DTables */ |
| 515 | { size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr, |
| 516 | LLtype, MaxLL, LLFSELog, |
| 517 | ip, iend-ip, |
| 518 | LL_base, LL_bits, |
| 519 | LL_defaultDTable, dctx->fseEntropy, |
| 520 | dctx->ddictIsCold, nbSeq); |
| 521 | RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected); |
| 522 | ip += llhSize; |
| 523 | } |
| 524 | |
| 525 | { size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr, |
| 526 | OFtype, MaxOff, OffFSELog, |
| 527 | ip, iend-ip, |
| 528 | OF_base, OF_bits, |
| 529 | OF_defaultDTable, dctx->fseEntropy, |
| 530 | dctx->ddictIsCold, nbSeq); |
| 531 | RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected); |
| 532 | ip += ofhSize; |
| 533 | } |
| 534 | |
| 535 | { size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr, |
| 536 | MLtype, MaxML, MLFSELog, |
| 537 | ip, iend-ip, |
| 538 | ML_base, ML_bits, |
| 539 | ML_defaultDTable, dctx->fseEntropy, |
| 540 | dctx->ddictIsCold, nbSeq); |
| 541 | RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected); |
| 542 | ip += mlhSize; |
| 543 | } |
| 544 | } |
| 545 | |
| 546 | return ip-istart; |
| 547 | } |
| 548 | |
| 549 | |
| 550 | typedef struct { |
| 551 | size_t litLength; |
| 552 | size_t matchLength; |
| 553 | size_t offset; |
| 554 | const BYTE* match; |
| 555 | } seq_t; |
| 556 | |
| 557 | typedef struct { |
| 558 | size_t state; |
| 559 | const ZSTD_seqSymbol* table; |
| 560 | } ZSTD_fseState; |
| 561 | |
| 562 | typedef struct { |
| 563 | BIT_DStream_t DStream; |
| 564 | ZSTD_fseState stateLL; |
| 565 | ZSTD_fseState stateOffb; |
| 566 | ZSTD_fseState stateML; |
| 567 | size_t prevOffset[ZSTD_REP_NUM]; |
| 568 | const BYTE* prefixStart; |
| 569 | const BYTE* dictEnd; |
| 570 | size_t pos; |
| 571 | } seqState_t; |
| 572 | |
| 573 | |
| 574 | /* ZSTD_execSequenceLast7(): |
| 575 | * exceptional case : decompress a match starting within last 7 bytes of output buffer. |
| 576 | * requires more careful checks, to ensure there is no overflow. |
| 577 | * performance does not matter though. |
| 578 | * note : this case is supposed to be never generated "naturally" by reference encoder, |
| 579 | * since in most cases it needs at least 8 bytes to look for a match. |
| 580 | * but it's allowed by the specification. */ |
| 581 | FORCE_NOINLINE |
| 582 | size_t ZSTD_execSequenceLast7(BYTE* op, |
| 583 | BYTE* const oend, seq_t sequence, |
| 584 | const BYTE** litPtr, const BYTE* const litLimit, |
| 585 | const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd) |
| 586 | { |
| 587 | BYTE* const oLitEnd = op + sequence.litLength; |
| 588 | size_t const sequenceLength = sequence.litLength + sequence.matchLength; |
| 589 | BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */ |
| 590 | const BYTE* const iLitEnd = *litPtr + sequence.litLength; |
| 591 | const BYTE* match = oLitEnd - sequence.offset; |
| 592 | |
| 593 | /* check */ |
| 594 | RETURN_ERROR_IF(oMatchEnd>oend, dstSize_tooSmall, "last match must fit within dstBuffer"); |
| 595 | RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "try to read beyond literal buffer"); |
| 596 | |
| 597 | /* copy literals */ |
| 598 | while (op < oLitEnd) *op++ = *(*litPtr)++; |
| 599 | |
| 600 | /* copy Match */ |
| 601 | if (sequence.offset > (size_t)(oLitEnd - base)) { |
| 602 | /* offset beyond prefix */ |
| 603 | RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - vBase),corruption_detected); |
| 604 | match = dictEnd - (base-match); |
| 605 | if (match + sequence.matchLength <= dictEnd) { |
| 606 | memmove(oLitEnd, match, sequence.matchLength); |
| 607 | return sequenceLength; |
| 608 | } |
| 609 | /* span extDict & currentPrefixSegment */ |
| 610 | { size_t const length1 = dictEnd - match; |
| 611 | memmove(oLitEnd, match, length1); |
| 612 | op = oLitEnd + length1; |
| 613 | sequence.matchLength -= length1; |
| 614 | match = base; |
| 615 | } } |
| 616 | while (op < oMatchEnd) *op++ = *match++; |
| 617 | return sequenceLength; |
| 618 | } |
| 619 | |
| 620 | |
| 621 | HINT_INLINE |
| 622 | size_t ZSTD_execSequence(BYTE* op, |
| 623 | BYTE* const oend, seq_t sequence, |
| 624 | const BYTE** litPtr, const BYTE* const litLimit, |
| 625 | const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd) |
| 626 | { |
| 627 | BYTE* const oLitEnd = op + sequence.litLength; |
| 628 | size_t const sequenceLength = sequence.litLength + sequence.matchLength; |
| 629 | BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */ |
| 630 | BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH; |
| 631 | const BYTE* const iLitEnd = *litPtr + sequence.litLength; |
| 632 | const BYTE* match = oLitEnd - sequence.offset; |
| 633 | |
| 634 | /* check */ |
| 635 | RETURN_ERROR_IF(oMatchEnd>oend, dstSize_tooSmall, "last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend"); |
| 636 | RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "over-read beyond lit buffer"); |
| 637 | if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd); |
| 638 | |
| 639 | /* copy Literals */ |
| 640 | if (sequence.litLength > 8) |
| 641 | ZSTD_wildcopy_16min(op, (*litPtr), sequence.litLength, ZSTD_no_overlap); /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */ |
| 642 | else |
| 643 | ZSTD_copy8(op, *litPtr); |
| 644 | op = oLitEnd; |
| 645 | *litPtr = iLitEnd; /* update for next sequence */ |
| 646 | |
| 647 | /* copy Match */ |
| 648 | if (sequence.offset > (size_t)(oLitEnd - prefixStart)) { |
| 649 | /* offset beyond prefix -> go into extDict */ |
| 650 | RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected); |
| 651 | match = dictEnd + (match - prefixStart); |
| 652 | if (match + sequence.matchLength <= dictEnd) { |
| 653 | memmove(oLitEnd, match, sequence.matchLength); |
| 654 | return sequenceLength; |
| 655 | } |
| 656 | /* span extDict & currentPrefixSegment */ |
| 657 | { size_t const length1 = dictEnd - match; |
| 658 | memmove(oLitEnd, match, length1); |
| 659 | op = oLitEnd + length1; |
| 660 | sequence.matchLength -= length1; |
| 661 | match = prefixStart; |
| 662 | if (op > oend_w || sequence.matchLength < MINMATCH) { |
| 663 | U32 i; |
| 664 | for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i]; |
| 665 | return sequenceLength; |
| 666 | } |
| 667 | } } |
| 668 | /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */ |
| 669 | |
| 670 | /* match within prefix */ |
| 671 | if (sequence.offset < 8) { |
| 672 | /* close range match, overlap */ |
| 673 | static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 }; /* added */ |
| 674 | static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 }; /* subtracted */ |
| 675 | int const sub2 = dec64table[sequence.offset]; |
| 676 | op[0] = match[0]; |
| 677 | op[1] = match[1]; |
| 678 | op[2] = match[2]; |
| 679 | op[3] = match[3]; |
| 680 | match += dec32table[sequence.offset]; |
| 681 | ZSTD_copy4(op+4, match); |
| 682 | match -= sub2; |
| 683 | } else { |
| 684 | ZSTD_copy8(op, match); |
| 685 | } |
| 686 | op += 8; match += 8; |
| 687 | |
| 688 | if (oMatchEnd > oend-(16-MINMATCH)) { |
| 689 | if (op < oend_w) { |
| 690 | ZSTD_wildcopy(op, match, oend_w - op, ZSTD_overlap_src_before_dst); |
| 691 | match += oend_w - op; |
| 692 | op = oend_w; |
| 693 | } |
| 694 | while (op < oMatchEnd) *op++ = *match++; |
| 695 | } else { |
| 696 | ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst); /* works even if matchLength < 8 */ |
| 697 | } |
| 698 | return sequenceLength; |
| 699 | } |
| 700 | |
| 701 | |
| 702 | HINT_INLINE |
| 703 | size_t ZSTD_execSequenceLong(BYTE* op, |
| 704 | BYTE* const oend, seq_t sequence, |
| 705 | const BYTE** litPtr, const BYTE* const litLimit, |
| 706 | const BYTE* const prefixStart, const BYTE* const dictStart, const BYTE* const dictEnd) |
| 707 | { |
| 708 | BYTE* const oLitEnd = op + sequence.litLength; |
| 709 | size_t const sequenceLength = sequence.litLength + sequence.matchLength; |
| 710 | BYTE* const oMatchEnd = op + sequenceLength; /* risk : address space overflow (32-bits) */ |
| 711 | BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH; |
| 712 | const BYTE* const iLitEnd = *litPtr + sequence.litLength; |
| 713 | const BYTE* match = sequence.match; |
| 714 | |
| 715 | /* check */ |
| 716 | RETURN_ERROR_IF(oMatchEnd > oend, dstSize_tooSmall, "last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend"); |
| 717 | RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "over-read beyond lit buffer"); |
| 718 | if (oLitEnd > oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, dictStart, dictEnd); |
| 719 | |
| 720 | /* copy Literals */ |
| 721 | if (sequence.litLength > 8) |
| 722 | ZSTD_wildcopy_16min(op, *litPtr, sequence.litLength, ZSTD_no_overlap); /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */ |
| 723 | else |
| 724 | ZSTD_copy8(op, *litPtr); /* note : op <= oLitEnd <= oend_w == oend - 8 */ |
| 725 | |
| 726 | op = oLitEnd; |
| 727 | *litPtr = iLitEnd; /* update for next sequence */ |
| 728 | |
| 729 | /* copy Match */ |
| 730 | if (sequence.offset > (size_t)(oLitEnd - prefixStart)) { |
| 731 | /* offset beyond prefix */ |
| 732 | RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - dictStart), corruption_detected); |
| 733 | if (match + sequence.matchLength <= dictEnd) { |
| 734 | memmove(oLitEnd, match, sequence.matchLength); |
| 735 | return sequenceLength; |
| 736 | } |
| 737 | /* span extDict & currentPrefixSegment */ |
| 738 | { size_t const length1 = dictEnd - match; |
| 739 | memmove(oLitEnd, match, length1); |
| 740 | op = oLitEnd + length1; |
| 741 | sequence.matchLength -= length1; |
| 742 | match = prefixStart; |
| 743 | if (op > oend_w || sequence.matchLength < MINMATCH) { |
| 744 | U32 i; |
| 745 | for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i]; |
| 746 | return sequenceLength; |
| 747 | } |
| 748 | } } |
| 749 | assert(op <= oend_w); |
| 750 | assert(sequence.matchLength >= MINMATCH); |
| 751 | |
| 752 | /* match within prefix */ |
| 753 | if (sequence.offset < 8) { |
| 754 | /* close range match, overlap */ |
| 755 | static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 }; /* added */ |
| 756 | static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 }; /* subtracted */ |
| 757 | int const sub2 = dec64table[sequence.offset]; |
| 758 | op[0] = match[0]; |
| 759 | op[1] = match[1]; |
| 760 | op[2] = match[2]; |
| 761 | op[3] = match[3]; |
| 762 | match += dec32table[sequence.offset]; |
| 763 | ZSTD_copy4(op+4, match); |
| 764 | match -= sub2; |
| 765 | } else { |
| 766 | ZSTD_copy8(op, match); |
| 767 | } |
| 768 | op += 8; match += 8; |
| 769 | |
| 770 | if (oMatchEnd > oend-(16-MINMATCH)) { |
| 771 | if (op < oend_w) { |
| 772 | ZSTD_wildcopy(op, match, oend_w - op, ZSTD_overlap_src_before_dst); |
| 773 | match += oend_w - op; |
| 774 | op = oend_w; |
| 775 | } |
| 776 | while (op < oMatchEnd) *op++ = *match++; |
| 777 | } else { |
| 778 | ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst); /* works even if matchLength < 8 */ |
| 779 | } |
| 780 | return sequenceLength; |
| 781 | } |
| 782 | |
| 783 | static void |
| 784 | ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt) |
| 785 | { |
| 786 | const void* ptr = dt; |
| 787 | const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr; |
| 788 | DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog); |
| 789 | DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits", |
| 790 | (U32)DStatePtr->state, DTableH->tableLog); |
| 791 | BIT_reloadDStream(bitD); |
| 792 | DStatePtr->table = dt + 1; |
| 793 | } |
| 794 | |
| 795 | FORCE_INLINE_TEMPLATE void |
| 796 | ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD) |
| 797 | { |
| 798 | ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state]; |
| 799 | U32 const nbBits = DInfo.nbBits; |
| 800 | size_t const lowBits = BIT_readBits(bitD, nbBits); |
| 801 | DStatePtr->state = DInfo.nextState + lowBits; |
| 802 | } |
| 803 | |
| 804 | /* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum |
| 805 | * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1) |
| 806 | * bits before reloading. This value is the maximum number of bytes we read |
| 807 | * after reloading when we are decoding long offsets. |
| 808 | */ |
| 809 | #define LONG_OFFSETS_MAX_EXTRA_BITS_32 \ |
| 810 | (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32 \ |
| 811 | ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32 \ |
| 812 | : 0) |
| 813 | |
| 814 | typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e; |
| 815 | |
| 816 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG |
| 817 | FORCE_INLINE_TEMPLATE seq_t |
| 818 | ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets) |
| 819 | { |
| 820 | seq_t seq; |
| 821 | U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits; |
| 822 | U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits; |
| 823 | U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits; |
| 824 | U32 const totalBits = llBits+mlBits+ofBits; |
| 825 | U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue; |
| 826 | U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue; |
| 827 | U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue; |
| 828 | |
| 829 | /* sequence */ |
| 830 | { size_t offset; |
| 831 | if (!ofBits) |
| 832 | offset = 0; |
| 833 | else { |
| 834 | ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1); |
| 835 | ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5); |
| 836 | assert(ofBits <= MaxOff); |
| 837 | if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) { |
| 838 | U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed); |
| 839 | offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits); |
| 840 | BIT_reloadDStream(&seqState->DStream); |
| 841 | if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits); |
| 842 | assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32); /* to avoid another reload */ |
| 843 | } else { |
| 844 | offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */ |
| 845 | if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); |
| 846 | } |
| 847 | } |
| 848 | |
| 849 | if (ofBits <= 1) { |
| 850 | offset += (llBase==0); |
| 851 | if (offset) { |
| 852 | size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset]; |
| 853 | temp += !temp; /* 0 is not valid; input is corrupted; force offset to 1 */ |
| 854 | if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1]; |
| 855 | seqState->prevOffset[1] = seqState->prevOffset[0]; |
| 856 | seqState->prevOffset[0] = offset = temp; |
| 857 | } else { /* offset == 0 */ |
| 858 | offset = seqState->prevOffset[0]; |
| 859 | } |
| 860 | } else { |
| 861 | seqState->prevOffset[2] = seqState->prevOffset[1]; |
| 862 | seqState->prevOffset[1] = seqState->prevOffset[0]; |
| 863 | seqState->prevOffset[0] = offset; |
| 864 | } |
| 865 | seq.offset = offset; |
| 866 | } |
| 867 | |
| 868 | seq.matchLength = mlBase |
| 869 | + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/) : 0); /* <= 16 bits */ |
| 870 | if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32)) |
| 871 | BIT_reloadDStream(&seqState->DStream); |
| 872 | if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog))) |
| 873 | BIT_reloadDStream(&seqState->DStream); |
| 874 | /* Ensure there are enough bits to read the rest of data in 64-bit mode. */ |
| 875 | ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64); |
| 876 | |
| 877 | seq.litLength = llBase |
| 878 | + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits/*>0*/) : 0); /* <= 16 bits */ |
| 879 | if (MEM_32bits()) |
| 880 | BIT_reloadDStream(&seqState->DStream); |
| 881 | |
| 882 | DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u", |
| 883 | (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset); |
| 884 | |
| 885 | /* ANS state update */ |
| 886 | ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream); /* <= 9 bits */ |
| 887 | ZSTD_updateFseState(&seqState->stateML, &seqState->DStream); /* <= 9 bits */ |
| 888 | if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */ |
| 889 | ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream); /* <= 8 bits */ |
| 890 | |
| 891 | return seq; |
| 892 | } |
| 893 | |
| 894 | FORCE_INLINE_TEMPLATE size_t |
| 895 | DONT_VECTORIZE |
| 896 | ZSTD_decompressSequences_body( ZSTD_DCtx* dctx, |
| 897 | void* dst, size_t maxDstSize, |
| 898 | const void* seqStart, size_t seqSize, int nbSeq, |
| 899 | const ZSTD_longOffset_e isLongOffset) |
| 900 | { |
| 901 | const BYTE* ip = (const BYTE*)seqStart; |
| 902 | const BYTE* const iend = ip + seqSize; |
| 903 | BYTE* const ostart = (BYTE* const)dst; |
| 904 | BYTE* const oend = ostart + maxDstSize; |
| 905 | BYTE* op = ostart; |
| 906 | const BYTE* litPtr = dctx->litPtr; |
| 907 | const BYTE* const litEnd = litPtr + dctx->litSize; |
| 908 | const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart); |
| 909 | const BYTE* const vBase = (const BYTE*) (dctx->virtualStart); |
| 910 | const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd); |
| 911 | DEBUGLOG(5, "ZSTD_decompressSequences_body"); |
| 912 | |
| 913 | /* Regen sequences */ |
| 914 | if (nbSeq) { |
| 915 | seqState_t seqState; |
| 916 | dctx->fseEntropy = 1; |
| 917 | { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; } |
| 918 | RETURN_ERROR_IF( |
| 919 | ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)), |
| 920 | corruption_detected); |
| 921 | ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr); |
| 922 | ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr); |
| 923 | ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr); |
| 924 | |
| 925 | ZSTD_STATIC_ASSERT( |
| 926 | BIT_DStream_unfinished < BIT_DStream_completed && |
| 927 | BIT_DStream_endOfBuffer < BIT_DStream_completed && |
| 928 | BIT_DStream_completed < BIT_DStream_overflow); |
| 929 | |
| 930 | for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) { |
| 931 | nbSeq--; |
| 932 | { seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset); |
| 933 | size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd); |
| 934 | DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize); |
| 935 | if (ZSTD_isError(oneSeqSize)) return oneSeqSize; |
| 936 | op += oneSeqSize; |
| 937 | } } |
| 938 | |
| 939 | /* check if reached exact end */ |
| 940 | DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq); |
| 941 | RETURN_ERROR_IF(nbSeq, corruption_detected); |
| 942 | RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected); |
| 943 | /* save reps for next block */ |
| 944 | { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); } |
| 945 | } |
| 946 | |
| 947 | /* last literal segment */ |
| 948 | { size_t const lastLLSize = litEnd - litPtr; |
| 949 | RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall); |
| 950 | memcpy(op, litPtr, lastLLSize); |
| 951 | op += lastLLSize; |
| 952 | } |
| 953 | |
| 954 | return op-ostart; |
| 955 | } |
| 956 | |
| 957 | static size_t |
| 958 | ZSTD_decompressSequences_default(ZSTD_DCtx* dctx, |
| 959 | void* dst, size_t maxDstSize, |
| 960 | const void* seqStart, size_t seqSize, int nbSeq, |
| 961 | const ZSTD_longOffset_e isLongOffset) |
| 962 | { |
| 963 | return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 964 | } |
| 965 | #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */ |
| 966 | |
| 967 | |
| 968 | |
| 969 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT |
| 970 | FORCE_INLINE_TEMPLATE seq_t |
| 971 | ZSTD_decodeSequenceLong(seqState_t* seqState, ZSTD_longOffset_e const longOffsets) |
| 972 | { |
| 973 | seq_t seq; |
| 974 | U32 const llBits = seqState->stateLL.table[seqState->stateLL.state].nbAdditionalBits; |
| 975 | U32 const mlBits = seqState->stateML.table[seqState->stateML.state].nbAdditionalBits; |
| 976 | U32 const ofBits = seqState->stateOffb.table[seqState->stateOffb.state].nbAdditionalBits; |
| 977 | U32 const totalBits = llBits+mlBits+ofBits; |
| 978 | U32 const llBase = seqState->stateLL.table[seqState->stateLL.state].baseValue; |
| 979 | U32 const mlBase = seqState->stateML.table[seqState->stateML.state].baseValue; |
| 980 | U32 const ofBase = seqState->stateOffb.table[seqState->stateOffb.state].baseValue; |
| 981 | |
| 982 | /* sequence */ |
| 983 | { size_t offset; |
| 984 | if (!ofBits) |
| 985 | offset = 0; |
| 986 | else { |
| 987 | ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1); |
| 988 | ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5); |
| 989 | assert(ofBits <= MaxOff); |
| 990 | if (MEM_32bits() && longOffsets) { |
| 991 | U32 const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN_32-1); |
| 992 | offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits); |
| 993 | if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream); |
| 994 | if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits); |
| 995 | } else { |
| 996 | offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits); /* <= (ZSTD_WINDOWLOG_MAX-1) bits */ |
| 997 | if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); |
| 998 | } |
| 999 | } |
| 1000 | |
| 1001 | if (ofBits <= 1) { |
| 1002 | offset += (llBase==0); |
| 1003 | if (offset) { |
| 1004 | size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset]; |
| 1005 | temp += !temp; /* 0 is not valid; input is corrupted; force offset to 1 */ |
| 1006 | if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1]; |
| 1007 | seqState->prevOffset[1] = seqState->prevOffset[0]; |
| 1008 | seqState->prevOffset[0] = offset = temp; |
| 1009 | } else { |
| 1010 | offset = seqState->prevOffset[0]; |
| 1011 | } |
| 1012 | } else { |
| 1013 | seqState->prevOffset[2] = seqState->prevOffset[1]; |
| 1014 | seqState->prevOffset[1] = seqState->prevOffset[0]; |
| 1015 | seqState->prevOffset[0] = offset; |
| 1016 | } |
| 1017 | seq.offset = offset; |
| 1018 | } |
| 1019 | |
| 1020 | seq.matchLength = mlBase + ((mlBits>0) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0); /* <= 16 bits */ |
| 1021 | if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32)) |
| 1022 | BIT_reloadDStream(&seqState->DStream); |
| 1023 | if (MEM_64bits() && (totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog))) |
| 1024 | BIT_reloadDStream(&seqState->DStream); |
| 1025 | /* Verify that there is enough bits to read the rest of the data in 64-bit mode. */ |
| 1026 | ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64); |
| 1027 | |
| 1028 | seq.litLength = llBase + ((llBits>0) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0); /* <= 16 bits */ |
| 1029 | if (MEM_32bits()) |
| 1030 | BIT_reloadDStream(&seqState->DStream); |
| 1031 | |
| 1032 | { size_t const pos = seqState->pos + seq.litLength; |
| 1033 | const BYTE* const matchBase = (seq.offset > pos) ? seqState->dictEnd : seqState->prefixStart; |
| 1034 | seq.match = matchBase + pos - seq.offset; /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted. |
| 1035 | * No consequence though : no memory access will occur, overly large offset will be detected in ZSTD_execSequenceLong() */ |
| 1036 | seqState->pos = pos + seq.matchLength; |
| 1037 | } |
| 1038 | |
| 1039 | /* ANS state update */ |
| 1040 | ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream); /* <= 9 bits */ |
| 1041 | ZSTD_updateFseState(&seqState->stateML, &seqState->DStream); /* <= 9 bits */ |
| 1042 | if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream); /* <= 18 bits */ |
| 1043 | ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream); /* <= 8 bits */ |
| 1044 | |
| 1045 | return seq; |
| 1046 | } |
| 1047 | |
| 1048 | FORCE_INLINE_TEMPLATE size_t |
| 1049 | ZSTD_decompressSequencesLong_body( |
| 1050 | ZSTD_DCtx* dctx, |
| 1051 | void* dst, size_t maxDstSize, |
| 1052 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1053 | const ZSTD_longOffset_e isLongOffset) |
| 1054 | { |
| 1055 | const BYTE* ip = (const BYTE*)seqStart; |
| 1056 | const BYTE* const iend = ip + seqSize; |
| 1057 | BYTE* const ostart = (BYTE* const)dst; |
| 1058 | BYTE* const oend = ostart + maxDstSize; |
| 1059 | BYTE* op = ostart; |
| 1060 | const BYTE* litPtr = dctx->litPtr; |
| 1061 | const BYTE* const litEnd = litPtr + dctx->litSize; |
| 1062 | const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart); |
| 1063 | const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart); |
| 1064 | const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd); |
| 1065 | |
| 1066 | /* Regen sequences */ |
| 1067 | if (nbSeq) { |
| 1068 | #define STORED_SEQS 4 |
| 1069 | #define STORED_SEQS_MASK (STORED_SEQS-1) |
| 1070 | #define ADVANCED_SEQS 4 |
| 1071 | seq_t sequences[STORED_SEQS]; |
| 1072 | int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS); |
| 1073 | seqState_t seqState; |
| 1074 | int seqNb; |
| 1075 | dctx->fseEntropy = 1; |
| 1076 | { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; } |
| 1077 | seqState.prefixStart = prefixStart; |
| 1078 | seqState.pos = (size_t)(op-prefixStart); |
| 1079 | seqState.dictEnd = dictEnd; |
| 1080 | assert(iend >= ip); |
| 1081 | RETURN_ERROR_IF( |
| 1082 | ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)), |
| 1083 | corruption_detected); |
| 1084 | ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr); |
| 1085 | ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr); |
| 1086 | ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr); |
| 1087 | |
| 1088 | /* prepare in advance */ |
| 1089 | for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) { |
| 1090 | sequences[seqNb] = ZSTD_decodeSequenceLong(&seqState, isLongOffset); |
| 1091 | PREFETCH_L1(sequences[seqNb].match); PREFETCH_L1(sequences[seqNb].match + sequences[seqNb].matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */ |
| 1092 | } |
| 1093 | RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected); |
| 1094 | |
| 1095 | /* decode and decompress */ |
| 1096 | for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) { |
| 1097 | seq_t const sequence = ZSTD_decodeSequenceLong(&seqState, isLongOffset); |
| 1098 | size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd); |
| 1099 | if (ZSTD_isError(oneSeqSize)) return oneSeqSize; |
| 1100 | PREFETCH_L1(sequence.match); PREFETCH_L1(sequence.match + sequence.matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */ |
| 1101 | sequences[seqNb & STORED_SEQS_MASK] = sequence; |
| 1102 | op += oneSeqSize; |
| 1103 | } |
| 1104 | RETURN_ERROR_IF(seqNb<nbSeq, corruption_detected); |
| 1105 | |
| 1106 | /* finish queue */ |
| 1107 | seqNb -= seqAdvance; |
| 1108 | for ( ; seqNb<nbSeq ; seqNb++) { |
| 1109 | size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd); |
| 1110 | if (ZSTD_isError(oneSeqSize)) return oneSeqSize; |
| 1111 | op += oneSeqSize; |
| 1112 | } |
| 1113 | |
| 1114 | /* save reps for next block */ |
| 1115 | { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); } |
| 1116 | } |
| 1117 | |
| 1118 | /* last literal segment */ |
| 1119 | { size_t const lastLLSize = litEnd - litPtr; |
| 1120 | RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall); |
| 1121 | memcpy(op, litPtr, lastLLSize); |
| 1122 | op += lastLLSize; |
| 1123 | } |
| 1124 | |
| 1125 | return op-ostart; |
| 1126 | } |
| 1127 | |
| 1128 | static size_t |
| 1129 | ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx, |
| 1130 | void* dst, size_t maxDstSize, |
| 1131 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1132 | const ZSTD_longOffset_e isLongOffset) |
| 1133 | { |
| 1134 | return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1135 | } |
| 1136 | #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */ |
| 1137 | |
| 1138 | |
| 1139 | |
| 1140 | #if DYNAMIC_BMI2 |
| 1141 | |
| 1142 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG |
| 1143 | static TARGET_ATTRIBUTE("bmi2") size_t |
| 1144 | DONT_VECTORIZE |
| 1145 | ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx, |
| 1146 | void* dst, size_t maxDstSize, |
| 1147 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1148 | const ZSTD_longOffset_e isLongOffset) |
| 1149 | { |
| 1150 | return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1151 | } |
| 1152 | #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */ |
| 1153 | |
| 1154 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT |
| 1155 | static TARGET_ATTRIBUTE("bmi2") size_t |
| 1156 | ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx, |
| 1157 | void* dst, size_t maxDstSize, |
| 1158 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1159 | const ZSTD_longOffset_e isLongOffset) |
| 1160 | { |
| 1161 | return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1162 | } |
| 1163 | #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */ |
| 1164 | |
| 1165 | #endif /* DYNAMIC_BMI2 */ |
| 1166 | |
| 1167 | typedef size_t (*ZSTD_decompressSequences_t)( |
| 1168 | ZSTD_DCtx* dctx, |
| 1169 | void* dst, size_t maxDstSize, |
| 1170 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1171 | const ZSTD_longOffset_e isLongOffset); |
| 1172 | |
| 1173 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG |
| 1174 | static size_t |
| 1175 | ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize, |
| 1176 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1177 | const ZSTD_longOffset_e isLongOffset) |
| 1178 | { |
| 1179 | DEBUGLOG(5, "ZSTD_decompressSequences"); |
| 1180 | #if DYNAMIC_BMI2 |
| 1181 | if (dctx->bmi2) { |
| 1182 | return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1183 | } |
| 1184 | #endif |
| 1185 | return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1186 | } |
| 1187 | #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */ |
| 1188 | |
| 1189 | |
| 1190 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT |
| 1191 | /* ZSTD_decompressSequencesLong() : |
| 1192 | * decompression function triggered when a minimum share of offsets is considered "long", |
| 1193 | * aka out of cache. |
| 1194 | * note : "long" definition seems overloaded here, sometimes meaning "wider than bitstream register", and sometimes meaning "farther than memory cache distance". |
| 1195 | * This function will try to mitigate main memory latency through the use of prefetching */ |
| 1196 | static size_t |
| 1197 | ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx, |
| 1198 | void* dst, size_t maxDstSize, |
| 1199 | const void* seqStart, size_t seqSize, int nbSeq, |
| 1200 | const ZSTD_longOffset_e isLongOffset) |
| 1201 | { |
| 1202 | DEBUGLOG(5, "ZSTD_decompressSequencesLong"); |
| 1203 | #if DYNAMIC_BMI2 |
| 1204 | if (dctx->bmi2) { |
| 1205 | return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1206 | } |
| 1207 | #endif |
| 1208 | return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset); |
| 1209 | } |
| 1210 | #endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */ |
| 1211 | |
| 1212 | |
| 1213 | |
| 1214 | #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ |
| 1215 | !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) |
| 1216 | /* ZSTD_getLongOffsetsShare() : |
| 1217 | * condition : offTable must be valid |
| 1218 | * @return : "share" of long offsets (arbitrarily defined as > (1<<23)) |
| 1219 | * compared to maximum possible of (1<<OffFSELog) */ |
| 1220 | static unsigned |
| 1221 | ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable) |
| 1222 | { |
| 1223 | const void* ptr = offTable; |
| 1224 | U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog; |
| 1225 | const ZSTD_seqSymbol* table = offTable + 1; |
| 1226 | U32 const max = 1 << tableLog; |
| 1227 | U32 u, total = 0; |
| 1228 | DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog); |
| 1229 | |
| 1230 | assert(max <= (1 << OffFSELog)); /* max not too large */ |
| 1231 | for (u=0; u<max; u++) { |
| 1232 | if (table[u].nbAdditionalBits > 22) total += 1; |
| 1233 | } |
| 1234 | |
| 1235 | assert(tableLog <= OffFSELog); |
| 1236 | total <<= (OffFSELog - tableLog); /* scale to OffFSELog */ |
| 1237 | |
| 1238 | return total; |
| 1239 | } |
| 1240 | #endif |
| 1241 | |
| 1242 | |
| 1243 | size_t |
| 1244 | ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx, |
| 1245 | void* dst, size_t dstCapacity, |
| 1246 | const void* src, size_t srcSize, const int frame) |
| 1247 | { /* blockType == blockCompressed */ |
| 1248 | const BYTE* ip = (const BYTE*)src; |
| 1249 | /* isLongOffset must be true if there are long offsets. |
| 1250 | * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN. |
| 1251 | * We don't expect that to be the case in 64-bit mode. |
| 1252 | * In block mode, window size is not known, so we have to be conservative. |
| 1253 | * (note: but it could be evaluated from current-lowLimit) |
| 1254 | */ |
| 1255 | ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)))); |
| 1256 | DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize); |
| 1257 | |
| 1258 | RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong); |
| 1259 | |
| 1260 | /* Decode literals section */ |
| 1261 | { size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize); |
| 1262 | DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize); |
| 1263 | if (ZSTD_isError(litCSize)) return litCSize; |
| 1264 | ip += litCSize; |
| 1265 | srcSize -= litCSize; |
| 1266 | } |
| 1267 | |
| 1268 | /* Build Decoding Tables */ |
| 1269 | { |
| 1270 | /* These macros control at build-time which decompressor implementation |
| 1271 | * we use. If neither is defined, we do some inspection and dispatch at |
| 1272 | * runtime. |
| 1273 | */ |
| 1274 | #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ |
| 1275 | !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) |
| 1276 | int usePrefetchDecoder = dctx->ddictIsCold; |
| 1277 | #endif |
| 1278 | int nbSeq; |
| 1279 | size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize); |
| 1280 | if (ZSTD_isError(seqHSize)) return seqHSize; |
| 1281 | ip += seqHSize; |
| 1282 | srcSize -= seqHSize; |
| 1283 | |
| 1284 | #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ |
| 1285 | !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) |
| 1286 | if ( !usePrefetchDecoder |
| 1287 | && (!frame || (dctx->fParams.windowSize > (1<<24))) |
| 1288 | && (nbSeq>ADVANCED_SEQS) ) { /* could probably use a larger nbSeq limit */ |
| 1289 | U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr); |
| 1290 | U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */ |
| 1291 | usePrefetchDecoder = (shareLongOffsets >= minShare); |
| 1292 | } |
| 1293 | #endif |
| 1294 | |
| 1295 | dctx->ddictIsCold = 0; |
| 1296 | |
| 1297 | #if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \ |
| 1298 | !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG) |
| 1299 | if (usePrefetchDecoder) |
| 1300 | #endif |
| 1301 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT |
| 1302 | return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset); |
| 1303 | #endif |
| 1304 | |
| 1305 | #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG |
| 1306 | /* else */ |
| 1307 | return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset); |
| 1308 | #endif |
| 1309 | } |
| 1310 | } |
| 1311 | |
| 1312 | |
| 1313 | size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, |
| 1314 | void* dst, size_t dstCapacity, |
| 1315 | const void* src, size_t srcSize) |
| 1316 | { |
| 1317 | size_t dSize; |
| 1318 | ZSTD_checkContinuity(dctx, dst); |
| 1319 | dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0); |
| 1320 | dctx->previousDstEnd = (char*)dst + dSize; |
| 1321 | return dSize; |
| 1322 | } |