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 #include "zstd_compress_internal.h" 12 #include "zstd_fast.h" 13 14 15 void ZSTD_fillHashTable(ZSTD_matchState_t* ms, 16 void const* end, ZSTD_dictTableLoadMethod_e dtlm) 17 { 18 const ZSTD_compressionParameters* const cParams = &ms->cParams; 19 U32* const hashTable = ms->hashTable; 20 U32 const hBits = cParams->hashLog; 21 U32 const mls = cParams->minMatch; 22 const BYTE* const base = ms->window.base; 23 const BYTE* ip = base + ms->nextToUpdate; 24 const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE; 25 const U32 fastHashFillStep = 3; 26 27 /* Always insert every fastHashFillStep position into the hash table. 28 * Insert the other positions if their hash entry is empty. 29 */ 30 for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) { 31 U32 const current = (U32)(ip - base); 32 size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls); 33 hashTable[hash0] = current; 34 if (dtlm == ZSTD_dtlm_fast) continue; 35 /* Only load extra positions for ZSTD_dtlm_full */ 36 { U32 p; 37 for (p = 1; p < fastHashFillStep; ++p) { 38 size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls); 39 if (hashTable[hash] == 0) { /* not yet filled */ 40 hashTable[hash] = current + p; 41 } } } } 42 } 43 44 FORCE_INLINE_TEMPLATE 45 size_t ZSTD_compressBlock_fast_generic( 46 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], 47 void const* src, size_t srcSize, 48 U32 const mls, ZSTD_dictMode_e const dictMode) 49 { 50 const ZSTD_compressionParameters* const cParams = &ms->cParams; 51 U32* const hashTable = ms->hashTable; 52 U32 const hlog = cParams->hashLog; 53 /* support stepSize of 0 */ 54 U32 const stepSize = cParams->targetLength + !(cParams->targetLength); 55 const BYTE* const base = ms->window.base; 56 const BYTE* const istart = (const BYTE*)src; 57 const BYTE* ip = istart; 58 const BYTE* anchor = istart; 59 const U32 prefixStartIndex = ms->window.dictLimit; 60 const BYTE* const prefixStart = base + prefixStartIndex; 61 const BYTE* const iend = istart + srcSize; 62 const BYTE* const ilimit = iend - HASH_READ_SIZE; 63 U32 offset_1=rep[0], offset_2=rep[1]; 64 U32 offsetSaved = 0; 65 66 const ZSTD_matchState_t* const dms = ms->dictMatchState; 67 const ZSTD_compressionParameters* const dictCParams = 68 dictMode == ZSTD_dictMatchState ? 69 &dms->cParams : NULL; 70 const U32* const dictHashTable = dictMode == ZSTD_dictMatchState ? 71 dms->hashTable : NULL; 72 const U32 dictStartIndex = dictMode == ZSTD_dictMatchState ? 73 dms->window.dictLimit : 0; 74 const BYTE* const dictBase = dictMode == ZSTD_dictMatchState ? 75 dms->window.base : NULL; 76 const BYTE* const dictStart = dictMode == ZSTD_dictMatchState ? 77 dictBase + dictStartIndex : NULL; 78 const BYTE* const dictEnd = dictMode == ZSTD_dictMatchState ? 79 dms->window.nextSrc : NULL; 80 const U32 dictIndexDelta = dictMode == ZSTD_dictMatchState ? 81 prefixStartIndex - (U32)(dictEnd - dictBase) : 82 0; 83 const U32 dictAndPrefixLength = (U32)(ip - prefixStart + dictEnd - dictStart); 84 const U32 dictHLog = dictMode == ZSTD_dictMatchState ? 85 dictCParams->hashLog : hlog; 86 87 assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState); 88 89 /* otherwise, we would get index underflow when translating a dict index 90 * into a local index */ 91 assert(dictMode != ZSTD_dictMatchState 92 || prefixStartIndex >= (U32)(dictEnd - dictBase)); 93 94 /* init */ 95 ip += (dictAndPrefixLength == 0); 96 if (dictMode == ZSTD_noDict) { 97 U32 const maxRep = (U32)(ip - prefixStart); 98 if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0; 99 if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0; 100 } 101 if (dictMode == ZSTD_dictMatchState) { 102 /* dictMatchState repCode checks don't currently handle repCode == 0 103 * disabling. */ 104 assert(offset_1 <= dictAndPrefixLength); 105 assert(offset_2 <= dictAndPrefixLength); 106 } 107 108 /* Main Search Loop */ 109 while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */ 110 size_t mLength; 111 size_t const h = ZSTD_hashPtr(ip, hlog, mls); 112 U32 const current = (U32)(ip-base); 113 U32 const matchIndex = hashTable[h]; 114 const BYTE* match = base + matchIndex; 115 const U32 repIndex = current + 1 - offset_1; 116 const BYTE* repMatch = (dictMode == ZSTD_dictMatchState 117 && repIndex < prefixStartIndex) ? 118 dictBase + (repIndex - dictIndexDelta) : 119 base + repIndex; 120 hashTable[h] = current; /* update hash table */ 121 122 if ( (dictMode == ZSTD_dictMatchState) 123 && ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */ 124 && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { 125 const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend; 126 mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4; 127 ip++; 128 ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH); 129 } else if ( dictMode == ZSTD_noDict 130 && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) { 131 mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4; 132 ip++; 133 ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH); 134 } else if ( (matchIndex <= prefixStartIndex) ) { 135 if (dictMode == ZSTD_dictMatchState) { 136 size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls); 137 U32 const dictMatchIndex = dictHashTable[dictHash]; 138 const BYTE* dictMatch = dictBase + dictMatchIndex; 139 if (dictMatchIndex <= dictStartIndex || 140 MEM_read32(dictMatch) != MEM_read32(ip)) { 141 assert(stepSize >= 1); 142 ip += ((ip-anchor) >> kSearchStrength) + stepSize; 143 continue; 144 } else { 145 /* found a dict match */ 146 U32 const offset = (U32)(current-dictMatchIndex-dictIndexDelta); 147 mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4; 148 while (((ip>anchor) & (dictMatch>dictStart)) 149 && (ip[-1] == dictMatch[-1])) { 150 ip--; dictMatch--; mLength++; 151 } /* catch up */ 152 offset_2 = offset_1; 153 offset_1 = offset; 154 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); 155 } 156 } else { 157 assert(stepSize >= 1); 158 ip += ((ip-anchor) >> kSearchStrength) + stepSize; 159 continue; 160 } 161 } else if (MEM_read32(match) != MEM_read32(ip)) { 162 /* it's not a match, and we're not going to check the dictionary */ 163 assert(stepSize >= 1); 164 ip += ((ip-anchor) >> kSearchStrength) + stepSize; 165 continue; 166 } else { 167 /* found a regular match */ 168 U32 const offset = (U32)(ip-match); 169 mLength = ZSTD_count(ip+4, match+4, iend) + 4; 170 while (((ip>anchor) & (match>prefixStart)) 171 && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ 172 offset_2 = offset_1; 173 offset_1 = offset; 174 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); 175 } 176 177 /* match found */ 178 ip += mLength; 179 anchor = ip; 180 181 if (ip <= ilimit) { 182 /* Fill Table */ 183 assert(base+current+2 > istart); /* check base overflow */ 184 hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2; /* here because current+2 could be > iend-8 */ 185 hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base); 186 187 /* check immediate repcode */ 188 if (dictMode == ZSTD_dictMatchState) { 189 while (ip <= ilimit) { 190 U32 const current2 = (U32)(ip-base); 191 U32 const repIndex2 = current2 - offset_2; 192 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? 193 dictBase - dictIndexDelta + repIndex2 : 194 base + repIndex2; 195 if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */) 196 && (MEM_read32(repMatch2) == MEM_read32(ip)) ) { 197 const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend; 198 size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4; 199 U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */ 200 ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH); 201 hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2; 202 ip += repLength2; 203 anchor = ip; 204 continue; 205 } 206 break; 207 } 208 } 209 210 if (dictMode == ZSTD_noDict) { 211 while ( (ip <= ilimit) 212 && ( (offset_2>0) 213 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) { 214 /* store sequence */ 215 size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4; 216 U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; /* swap offset_2 <=> offset_1 */ 217 hashTable[ZSTD_hashPtr(ip, hlog, mls)] = (U32)(ip-base); 218 ZSTD_storeSeq(seqStore, 0, anchor, 0, rLength-MINMATCH); 219 ip += rLength; 220 anchor = ip; 221 continue; /* faster when present ... (?) */ 222 } } } } 223 224 /* save reps for next block */ 225 rep[0] = offset_1 ? offset_1 : offsetSaved; 226 rep[1] = offset_2 ? offset_2 : offsetSaved; 227 228 /* Return the last literals size */ 229 return iend - anchor; 230 } 231 232 233 size_t ZSTD_compressBlock_fast( 234 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], 235 void const* src, size_t srcSize) 236 { 237 ZSTD_compressionParameters const* cParams = &ms->cParams; 238 U32 const mls = cParams->minMatch; 239 assert(ms->dictMatchState == NULL); 240 switch(mls) 241 { 242 default: /* includes case 3 */ 243 case 4 : 244 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict); 245 case 5 : 246 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict); 247 case 6 : 248 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict); 249 case 7 : 250 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict); 251 } 252 } 253 254 size_t ZSTD_compressBlock_fast_dictMatchState( 255 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], 256 void const* src, size_t srcSize) 257 { 258 ZSTD_compressionParameters const* cParams = &ms->cParams; 259 U32 const mls = cParams->minMatch; 260 assert(ms->dictMatchState != NULL); 261 switch(mls) 262 { 263 default: /* includes case 3 */ 264 case 4 : 265 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState); 266 case 5 : 267 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState); 268 case 6 : 269 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState); 270 case 7 : 271 return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState); 272 } 273 } 274 275 276 static size_t ZSTD_compressBlock_fast_extDict_generic( 277 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], 278 void const* src, size_t srcSize, U32 const mls) 279 { 280 const ZSTD_compressionParameters* const cParams = &ms->cParams; 281 U32* const hashTable = ms->hashTable; 282 U32 const hlog = cParams->hashLog; 283 /* support stepSize of 0 */ 284 U32 const stepSize = cParams->targetLength + !(cParams->targetLength); 285 const BYTE* const base = ms->window.base; 286 const BYTE* const dictBase = ms->window.dictBase; 287 const BYTE* const istart = (const BYTE*)src; 288 const BYTE* ip = istart; 289 const BYTE* anchor = istart; 290 const U32 dictStartIndex = ms->window.lowLimit; 291 const BYTE* const dictStart = dictBase + dictStartIndex; 292 const U32 prefixStartIndex = ms->window.dictLimit; 293 const BYTE* const prefixStart = base + prefixStartIndex; 294 const BYTE* const dictEnd = dictBase + prefixStartIndex; 295 const BYTE* const iend = istart + srcSize; 296 const BYTE* const ilimit = iend - 8; 297 U32 offset_1=rep[0], offset_2=rep[1]; 298 299 /* Search Loop */ 300 while (ip < ilimit) { /* < instead of <=, because (ip+1) */ 301 const size_t h = ZSTD_hashPtr(ip, hlog, mls); 302 const U32 matchIndex = hashTable[h]; 303 const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base; 304 const BYTE* match = matchBase + matchIndex; 305 const U32 current = (U32)(ip-base); 306 const U32 repIndex = current + 1 - offset_1; 307 const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base; 308 const BYTE* const repMatch = repBase + repIndex; 309 size_t mLength; 310 hashTable[h] = current; /* update hash table */ 311 assert(offset_1 <= current +1); /* check repIndex */ 312 313 if ( (((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > dictStartIndex)) 314 && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { 315 const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend; 316 mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4; 317 ip++; 318 ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH); 319 } else { 320 if ( (matchIndex < dictStartIndex) || 321 (MEM_read32(match) != MEM_read32(ip)) ) { 322 assert(stepSize >= 1); 323 ip += ((ip-anchor) >> kSearchStrength) + stepSize; 324 continue; 325 } 326 { const BYTE* matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend; 327 const BYTE* lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart; 328 U32 offset; 329 mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4; 330 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ 331 offset = current - matchIndex; 332 offset_2 = offset_1; 333 offset_1 = offset; 334 ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); 335 } } 336 337 /* found a match : store it */ 338 ip += mLength; 339 anchor = ip; 340 341 if (ip <= ilimit) { 342 /* Fill Table */ 343 hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2; 344 hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base); 345 /* check immediate repcode */ 346 while (ip <= ilimit) { 347 U32 const current2 = (U32)(ip-base); 348 U32 const repIndex2 = current2 - offset_2; 349 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2; 350 if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (repIndex2 > dictStartIndex)) /* intentional overflow */ 351 && (MEM_read32(repMatch2) == MEM_read32(ip)) ) { 352 const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend; 353 size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4; 354 U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */ 355 ZSTD_storeSeq(seqStore, 0, anchor, 0, repLength2-MINMATCH); 356 hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2; 357 ip += repLength2; 358 anchor = ip; 359 continue; 360 } 361 break; 362 } } } 363 364 /* save reps for next block */ 365 rep[0] = offset_1; 366 rep[1] = offset_2; 367 368 /* Return the last literals size */ 369 return iend - anchor; 370 } 371 372 373 size_t ZSTD_compressBlock_fast_extDict( 374 ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], 375 void const* src, size_t srcSize) 376 { 377 ZSTD_compressionParameters const* cParams = &ms->cParams; 378 U32 const mls = cParams->minMatch; 379 switch(mls) 380 { 381 default: /* includes case 3 */ 382 case 4 : 383 return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 4); 384 case 5 : 385 return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 5); 386 case 6 : 387 return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 6); 388 case 7 : 389 return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 7); 390 } 391 } 392