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_double_fast.h" 12 13 14 void ZSTD_fillDoubleHashTable(ZSTD_CCtx* cctx, const void* end, const U32 mls) 15 { 16 U32* const hashLarge = cctx->hashTable; 17 U32 const hBitsL = cctx->appliedParams.cParams.hashLog; 18 U32* const hashSmall = cctx->chainTable; 19 U32 const hBitsS = cctx->appliedParams.cParams.chainLog; 20 const BYTE* const base = cctx->base; 21 const BYTE* ip = base + cctx->nextToUpdate; 22 const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE; 23 const size_t fastHashFillStep = 3; 24 25 while(ip <= iend) { 26 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip - base); 27 hashLarge[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip - base); 28 ip += fastHashFillStep; 29 } 30 } 31 32 33 FORCE_INLINE_TEMPLATE 34 size_t ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx, 35 const void* src, size_t srcSize, 36 const U32 mls) 37 { 38 U32* const hashLong = cctx->hashTable; 39 const U32 hBitsL = cctx->appliedParams.cParams.hashLog; 40 U32* const hashSmall = cctx->chainTable; 41 const U32 hBitsS = cctx->appliedParams.cParams.chainLog; 42 seqStore_t* seqStorePtr = &(cctx->seqStore); 43 const BYTE* const base = cctx->base; 44 const BYTE* const istart = (const BYTE*)src; 45 const BYTE* ip = istart; 46 const BYTE* anchor = istart; 47 const U32 lowestIndex = cctx->dictLimit; 48 const BYTE* const lowest = base + lowestIndex; 49 const BYTE* const iend = istart + srcSize; 50 const BYTE* const ilimit = iend - HASH_READ_SIZE; 51 U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1]; 52 U32 offsetSaved = 0; 53 54 /* init */ 55 ip += (ip==lowest); 56 { U32 const maxRep = (U32)(ip-lowest); 57 if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0; 58 if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0; 59 } 60 61 /* Main Search Loop */ 62 while (ip < ilimit) { /* < instead of <=, because repcode check at (ip+1) */ 63 size_t mLength; 64 size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8); 65 size_t const h = ZSTD_hashPtr(ip, hBitsS, mls); 66 U32 const current = (U32)(ip-base); 67 U32 const matchIndexL = hashLong[h2]; 68 U32 const matchIndexS = hashSmall[h]; 69 const BYTE* matchLong = base + matchIndexL; 70 const BYTE* match = base + matchIndexS; 71 hashLong[h2] = hashSmall[h] = current; /* update hash tables */ 72 73 assert(offset_1 <= current); /* supposed guaranteed by construction */ 74 if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) { 75 /* favor repcode */ 76 mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4; 77 ip++; 78 ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH); 79 } else { 80 U32 offset; 81 if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) { 82 mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8; 83 offset = (U32)(ip-matchLong); 84 while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */ 85 } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) { 86 size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8); 87 U32 const matchIndexL3 = hashLong[hl3]; 88 const BYTE* matchL3 = base + matchIndexL3; 89 hashLong[hl3] = current + 1; 90 if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) { 91 mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8; 92 ip++; 93 offset = (U32)(ip-matchL3); 94 while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */ 95 } else { 96 mLength = ZSTD_count(ip+4, match+4, iend) + 4; 97 offset = (U32)(ip-match); 98 while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ 99 } 100 } else { 101 ip += ((ip-anchor) >> g_searchStrength) + 1; 102 continue; 103 } 104 105 offset_2 = offset_1; 106 offset_1 = offset; 107 108 ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); 109 } 110 111 /* match found */ 112 ip += mLength; 113 anchor = ip; 114 115 if (ip <= ilimit) { 116 /* Fill Table */ 117 hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = 118 hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2; /* here because current+2 could be > iend-8 */ 119 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = 120 hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base); 121 122 /* check immediate repcode */ 123 while ( (ip <= ilimit) 124 && ( (offset_2>0) 125 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) { 126 /* store sequence */ 127 size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4; 128 { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */ 129 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base); 130 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base); 131 ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH); 132 ip += rLength; 133 anchor = ip; 134 continue; /* faster when present ... (?) */ 135 } } } 136 137 /* save reps for next block */ 138 seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved; 139 seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved; 140 141 /* Return the last literals size */ 142 return iend - anchor; 143 } 144 145 146 size_t ZSTD_compressBlock_doubleFast(ZSTD_CCtx* ctx, const void* src, size_t srcSize) 147 { 148 const U32 mls = ctx->appliedParams.cParams.searchLength; 149 switch(mls) 150 { 151 default: /* includes case 3 */ 152 case 4 : 153 return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 4); 154 case 5 : 155 return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 5); 156 case 6 : 157 return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 6); 158 case 7 : 159 return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 7); 160 } 161 } 162 163 164 static size_t ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx, 165 const void* src, size_t srcSize, 166 const U32 mls) 167 { 168 U32* const hashLong = ctx->hashTable; 169 U32 const hBitsL = ctx->appliedParams.cParams.hashLog; 170 U32* const hashSmall = ctx->chainTable; 171 U32 const hBitsS = ctx->appliedParams.cParams.chainLog; 172 seqStore_t* seqStorePtr = &(ctx->seqStore); 173 const BYTE* const base = ctx->base; 174 const BYTE* const dictBase = ctx->dictBase; 175 const BYTE* const istart = (const BYTE*)src; 176 const BYTE* ip = istart; 177 const BYTE* anchor = istart; 178 const U32 lowestIndex = ctx->lowLimit; 179 const BYTE* const dictStart = dictBase + lowestIndex; 180 const U32 dictLimit = ctx->dictLimit; 181 const BYTE* const lowPrefixPtr = base + dictLimit; 182 const BYTE* const dictEnd = dictBase + dictLimit; 183 const BYTE* const iend = istart + srcSize; 184 const BYTE* const ilimit = iend - 8; 185 U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1]; 186 187 /* Search Loop */ 188 while (ip < ilimit) { /* < instead of <=, because (ip+1) */ 189 const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls); 190 const U32 matchIndex = hashSmall[hSmall]; 191 const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base; 192 const BYTE* match = matchBase + matchIndex; 193 194 const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8); 195 const U32 matchLongIndex = hashLong[hLong]; 196 const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base; 197 const BYTE* matchLong = matchLongBase + matchLongIndex; 198 199 const U32 current = (U32)(ip-base); 200 const U32 repIndex = current + 1 - offset_1; /* offset_1 expected <= current +1 */ 201 const BYTE* repBase = repIndex < dictLimit ? dictBase : base; 202 const BYTE* repMatch = repBase + repIndex; 203 size_t mLength; 204 hashSmall[hSmall] = hashLong[hLong] = current; /* update hash table */ 205 206 if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex)) 207 && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { 208 const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend; 209 mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4; 210 ip++; 211 ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH); 212 } else { 213 if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) { 214 const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend; 215 const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr; 216 U32 offset; 217 mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8; 218 offset = current - matchLongIndex; 219 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */ 220 offset_2 = offset_1; 221 offset_1 = offset; 222 ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); 223 224 } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) { 225 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8); 226 U32 const matchIndex3 = hashLong[h3]; 227 const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base; 228 const BYTE* match3 = match3Base + matchIndex3; 229 U32 offset; 230 hashLong[h3] = current + 1; 231 if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) { 232 const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend; 233 const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr; 234 mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8; 235 ip++; 236 offset = current+1 - matchIndex3; 237 while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */ 238 } else { 239 const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend; 240 const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr; 241 mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4; 242 offset = current - matchIndex; 243 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */ 244 } 245 offset_2 = offset_1; 246 offset_1 = offset; 247 ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH); 248 249 } else { 250 ip += ((ip-anchor) >> g_searchStrength) + 1; 251 continue; 252 } } 253 254 /* found a match : store it */ 255 ip += mLength; 256 anchor = ip; 257 258 if (ip <= ilimit) { 259 /* Fill Table */ 260 hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2; 261 hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2; 262 hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base); 263 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base); 264 /* check immediate repcode */ 265 while (ip <= ilimit) { 266 U32 const current2 = (U32)(ip-base); 267 U32 const repIndex2 = current2 - offset_2; 268 const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2; 269 if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex)) /* intentional overflow */ 270 && (MEM_read32(repMatch2) == MEM_read32(ip)) ) { 271 const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend; 272 size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4; 273 U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; /* swap offset_2 <=> offset_1 */ 274 ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH); 275 hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2; 276 hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2; 277 ip += repLength2; 278 anchor = ip; 279 continue; 280 } 281 break; 282 } } } 283 284 /* save reps for next block */ 285 seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2; 286 287 /* Return the last literals size */ 288 return iend - anchor; 289 } 290 291 292 size_t ZSTD_compressBlock_doubleFast_extDict(ZSTD_CCtx* ctx, 293 const void* src, size_t srcSize) 294 { 295 U32 const mls = ctx->appliedParams.cParams.searchLength; 296 switch(mls) 297 { 298 default: /* includes case 3 */ 299 case 4 : 300 return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 4); 301 case 5 : 302 return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 5); 303 case 6 : 304 return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 6); 305 case 7 : 306 return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 7); 307 } 308 } 309