10c16b537SWarner Losh /*
237f1f268SConrad Meyer  * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
30c16b537SWarner Losh  * All rights reserved.
40c16b537SWarner Losh  *
50c16b537SWarner Losh  * This source code is licensed under both the BSD-style license (found in the
60c16b537SWarner Losh  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
70c16b537SWarner Losh  * in the COPYING file in the root directory of this source tree).
80c16b537SWarner Losh  * You may select, at your option, one of the above-listed licenses.
90c16b537SWarner Losh  */
100c16b537SWarner Losh 
11052d3c12SConrad Meyer #include "zstd_compress_internal.h"
120c16b537SWarner Losh #include "zstd_double_fast.h"
130c16b537SWarner Losh 
140c16b537SWarner Losh 
1519fcbaf1SConrad Meyer void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
160f743729SConrad Meyer                               void const* end, ZSTD_dictTableLoadMethod_e dtlm)
170c16b537SWarner Losh {
180f743729SConrad Meyer     const ZSTD_compressionParameters* const cParams = &ms->cParams;
1919fcbaf1SConrad Meyer     U32* const hashLarge = ms->hashTable;
2019fcbaf1SConrad Meyer     U32  const hBitsL = cParams->hashLog;
21a0483764SConrad Meyer     U32  const mls = cParams->minMatch;
2219fcbaf1SConrad Meyer     U32* const hashSmall = ms->chainTable;
2319fcbaf1SConrad Meyer     U32  const hBitsS = cParams->chainLog;
2419fcbaf1SConrad Meyer     const BYTE* const base = ms->window.base;
2519fcbaf1SConrad Meyer     const BYTE* ip = base + ms->nextToUpdate;
260c16b537SWarner Losh     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
2719fcbaf1SConrad Meyer     const U32 fastHashFillStep = 3;
280c16b537SWarner Losh 
2919fcbaf1SConrad Meyer     /* Always insert every fastHashFillStep position into the hash tables.
3019fcbaf1SConrad Meyer      * Insert the other positions into the large hash table if their entry
3119fcbaf1SConrad Meyer      * is empty.
3219fcbaf1SConrad Meyer      */
3319fcbaf1SConrad Meyer     for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
34f7cd7fe5SConrad Meyer         U32 const curr = (U32)(ip - base);
3519fcbaf1SConrad Meyer         U32 i;
3619fcbaf1SConrad Meyer         for (i = 0; i < fastHashFillStep; ++i) {
3719fcbaf1SConrad Meyer             size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
3819fcbaf1SConrad Meyer             size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
3919fcbaf1SConrad Meyer             if (i == 0)
40f7cd7fe5SConrad Meyer                 hashSmall[smHash] = curr + i;
4119fcbaf1SConrad Meyer             if (i == 0 || hashLarge[lgHash] == 0)
42f7cd7fe5SConrad Meyer                 hashLarge[lgHash] = curr + i;
430f743729SConrad Meyer             /* Only load extra positions for ZSTD_dtlm_full */
440f743729SConrad Meyer             if (dtlm == ZSTD_dtlm_fast)
450f743729SConrad Meyer                 break;
464d3f1eafSConrad Meyer     }   }
470c16b537SWarner Losh }
480c16b537SWarner Losh 
490c16b537SWarner Losh 
500c16b537SWarner Losh FORCE_INLINE_TEMPLATE
5119fcbaf1SConrad Meyer size_t ZSTD_compressBlock_doubleFast_generic(
5219fcbaf1SConrad Meyer         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
530f743729SConrad Meyer         void const* src, size_t srcSize,
540f743729SConrad Meyer         U32 const mls /* template */, ZSTD_dictMode_e const dictMode)
550c16b537SWarner Losh {
560f743729SConrad Meyer     ZSTD_compressionParameters const* cParams = &ms->cParams;
5719fcbaf1SConrad Meyer     U32* const hashLong = ms->hashTable;
5819fcbaf1SConrad Meyer     const U32 hBitsL = cParams->hashLog;
5919fcbaf1SConrad Meyer     U32* const hashSmall = ms->chainTable;
6019fcbaf1SConrad Meyer     const U32 hBitsS = cParams->chainLog;
6119fcbaf1SConrad Meyer     const BYTE* const base = ms->window.base;
620c16b537SWarner Losh     const BYTE* const istart = (const BYTE*)src;
630c16b537SWarner Losh     const BYTE* ip = istart;
640c16b537SWarner Losh     const BYTE* anchor = istart;
654d3f1eafSConrad Meyer     const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
669cbefe25SConrad Meyer     /* presumes that, if there is a dictionary, it must be using Attach mode */
6737f1f268SConrad Meyer     const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
680f743729SConrad Meyer     const BYTE* const prefixLowest = base + prefixLowestIndex;
690c16b537SWarner Losh     const BYTE* const iend = istart + srcSize;
700c16b537SWarner Losh     const BYTE* const ilimit = iend - HASH_READ_SIZE;
7119fcbaf1SConrad Meyer     U32 offset_1=rep[0], offset_2=rep[1];
720c16b537SWarner Losh     U32 offsetSaved = 0;
730c16b537SWarner Losh 
740f743729SConrad Meyer     const ZSTD_matchState_t* const dms = ms->dictMatchState;
750f743729SConrad Meyer     const ZSTD_compressionParameters* const dictCParams =
760f743729SConrad Meyer                                      dictMode == ZSTD_dictMatchState ?
770f743729SConrad Meyer                                      &dms->cParams : NULL;
780f743729SConrad Meyer     const U32* const dictHashLong  = dictMode == ZSTD_dictMatchState ?
790f743729SConrad Meyer                                      dms->hashTable : NULL;
800f743729SConrad Meyer     const U32* const dictHashSmall = dictMode == ZSTD_dictMatchState ?
810f743729SConrad Meyer                                      dms->chainTable : NULL;
820f743729SConrad Meyer     const U32 dictStartIndex       = dictMode == ZSTD_dictMatchState ?
830f743729SConrad Meyer                                      dms->window.dictLimit : 0;
840f743729SConrad Meyer     const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
850f743729SConrad Meyer                                      dms->window.base : NULL;
860f743729SConrad Meyer     const BYTE* const dictStart    = dictMode == ZSTD_dictMatchState ?
870f743729SConrad Meyer                                      dictBase + dictStartIndex : NULL;
880f743729SConrad Meyer     const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
890f743729SConrad Meyer                                      dms->window.nextSrc : NULL;
900f743729SConrad Meyer     const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
910f743729SConrad Meyer                                      prefixLowestIndex - (U32)(dictEnd - dictBase) :
920f743729SConrad Meyer                                      0;
930f743729SConrad Meyer     const U32 dictHBitsL           = dictMode == ZSTD_dictMatchState ?
940f743729SConrad Meyer                                      dictCParams->hashLog : hBitsL;
950f743729SConrad Meyer     const U32 dictHBitsS           = dictMode == ZSTD_dictMatchState ?
960f743729SConrad Meyer                                      dictCParams->chainLog : hBitsS;
9737f1f268SConrad Meyer     const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
980f743729SConrad Meyer 
994d3f1eafSConrad Meyer     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
1004d3f1eafSConrad Meyer 
1010f743729SConrad Meyer     assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
1020f743729SConrad Meyer 
1034d3f1eafSConrad Meyer     /* if a dictionary is attached, it must be within window range */
1044d3f1eafSConrad Meyer     if (dictMode == ZSTD_dictMatchState) {
10537f1f268SConrad Meyer         assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
1064d3f1eafSConrad Meyer     }
1074d3f1eafSConrad Meyer 
1080c16b537SWarner Losh     /* init */
1090f743729SConrad Meyer     ip += (dictAndPrefixLength == 0);
1100f743729SConrad Meyer     if (dictMode == ZSTD_noDict) {
111f7cd7fe5SConrad Meyer         U32 const curr = (U32)(ip - base);
112f7cd7fe5SConrad Meyer         U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
113f7cd7fe5SConrad Meyer         U32 const maxRep = curr - windowLow;
1140c16b537SWarner Losh         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
1150c16b537SWarner Losh         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
1160c16b537SWarner Losh     }
1170f743729SConrad Meyer     if (dictMode == ZSTD_dictMatchState) {
1180f743729SConrad Meyer         /* dictMatchState repCode checks don't currently handle repCode == 0
1190f743729SConrad Meyer          * disabling. */
1200f743729SConrad Meyer         assert(offset_1 <= dictAndPrefixLength);
1210f743729SConrad Meyer         assert(offset_2 <= dictAndPrefixLength);
1220f743729SConrad Meyer     }
1230c16b537SWarner Losh 
1240c16b537SWarner Losh     /* Main Search Loop */
1250c16b537SWarner Losh     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
1260c16b537SWarner Losh         size_t mLength;
1270f743729SConrad Meyer         U32 offset;
1280c16b537SWarner Losh         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
1290c16b537SWarner Losh         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
1300f743729SConrad Meyer         size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
1310f743729SConrad Meyer         size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
132f7cd7fe5SConrad Meyer         U32 const curr = (U32)(ip-base);
1330c16b537SWarner Losh         U32 const matchIndexL = hashLong[h2];
1340f743729SConrad Meyer         U32 matchIndexS = hashSmall[h];
1350c16b537SWarner Losh         const BYTE* matchLong = base + matchIndexL;
1360c16b537SWarner Losh         const BYTE* match = base + matchIndexS;
137f7cd7fe5SConrad Meyer         const U32 repIndex = curr + 1 - offset_1;
1380f743729SConrad Meyer         const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
1390f743729SConrad Meyer                             && repIndex < prefixLowestIndex) ?
1400f743729SConrad Meyer                                dictBase + (repIndex - dictIndexDelta) :
1410f743729SConrad Meyer                                base + repIndex;
142f7cd7fe5SConrad Meyer         hashLong[h2] = hashSmall[h] = curr;   /* update hash tables */
1430c16b537SWarner Losh 
1440f743729SConrad Meyer         /* check dictMatchState repcode */
1450f743729SConrad Meyer         if (dictMode == ZSTD_dictMatchState
1460f743729SConrad Meyer             && ((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
1470f743729SConrad Meyer             && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
1480f743729SConrad Meyer             const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
1490f743729SConrad Meyer             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
1500f743729SConrad Meyer             ip++;
1519cbefe25SConrad Meyer             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
1520f743729SConrad Meyer             goto _match_stored;
1530f743729SConrad Meyer         }
1540f743729SConrad Meyer 
1550f743729SConrad Meyer         /* check noDict repcode */
1560f743729SConrad Meyer         if ( dictMode == ZSTD_noDict
1570f743729SConrad Meyer           && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
1580c16b537SWarner Losh             mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
1590c16b537SWarner Losh             ip++;
1609cbefe25SConrad Meyer             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
1610f743729SConrad Meyer             goto _match_stored;
1620f743729SConrad Meyer         }
1630f743729SConrad Meyer 
1640f743729SConrad Meyer         if (matchIndexL > prefixLowestIndex) {
1650f743729SConrad Meyer             /* check prefix long match */
1660f743729SConrad Meyer             if (MEM_read64(matchLong) == MEM_read64(ip)) {
1670c16b537SWarner Losh                 mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
1680c16b537SWarner Losh                 offset = (U32)(ip-matchLong);
1690f743729SConrad Meyer                 while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
1700f743729SConrad Meyer                 goto _match_found;
1710f743729SConrad Meyer             }
1720f743729SConrad Meyer         } else if (dictMode == ZSTD_dictMatchState) {
1730f743729SConrad Meyer             /* check dictMatchState long match */
1740f743729SConrad Meyer             U32 const dictMatchIndexL = dictHashLong[dictHL];
1750f743729SConrad Meyer             const BYTE* dictMatchL = dictBase + dictMatchIndexL;
1760f743729SConrad Meyer             assert(dictMatchL < dictEnd);
1770f743729SConrad Meyer 
1780f743729SConrad Meyer             if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
1790f743729SConrad Meyer                 mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
180f7cd7fe5SConrad Meyer                 offset = (U32)(curr - dictMatchIndexL - dictIndexDelta);
1810f743729SConrad Meyer                 while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
1820f743729SConrad Meyer                 goto _match_found;
1834d3f1eafSConrad Meyer         }   }
1840f743729SConrad Meyer 
1850f743729SConrad Meyer         if (matchIndexS > prefixLowestIndex) {
1860f743729SConrad Meyer             /* check prefix short match */
1870f743729SConrad Meyer             if (MEM_read32(match) == MEM_read32(ip)) {
1880f743729SConrad Meyer                 goto _search_next_long;
1890f743729SConrad Meyer             }
1900f743729SConrad Meyer         } else if (dictMode == ZSTD_dictMatchState) {
1910f743729SConrad Meyer             /* check dictMatchState short match */
1920f743729SConrad Meyer             U32 const dictMatchIndexS = dictHashSmall[dictHS];
1930f743729SConrad Meyer             match = dictBase + dictMatchIndexS;
1940f743729SConrad Meyer             matchIndexS = dictMatchIndexS + dictIndexDelta;
1950f743729SConrad Meyer 
1960f743729SConrad Meyer             if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
1970f743729SConrad Meyer                 goto _search_next_long;
1984d3f1eafSConrad Meyer         }   }
1990f743729SConrad Meyer 
2000f743729SConrad Meyer         ip += ((ip-anchor) >> kSearchStrength) + 1;
20137f1f268SConrad Meyer #if defined(__aarch64__)
20237f1f268SConrad Meyer         PREFETCH_L1(ip+256);
20337f1f268SConrad Meyer #endif
2040f743729SConrad Meyer         continue;
2050f743729SConrad Meyer 
2060f743729SConrad Meyer _search_next_long:
2070f743729SConrad Meyer 
2084d3f1eafSConrad Meyer         {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
2090f743729SConrad Meyer             size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
2100c16b537SWarner Losh             U32 const matchIndexL3 = hashLong[hl3];
2110c16b537SWarner Losh             const BYTE* matchL3 = base + matchIndexL3;
212f7cd7fe5SConrad Meyer             hashLong[hl3] = curr + 1;
2130f743729SConrad Meyer 
2140f743729SConrad Meyer             /* check prefix long +1 match */
2150f743729SConrad Meyer             if (matchIndexL3 > prefixLowestIndex) {
2160f743729SConrad Meyer                 if (MEM_read64(matchL3) == MEM_read64(ip+1)) {
2170c16b537SWarner Losh                     mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
2180c16b537SWarner Losh                     ip++;
2190c16b537SWarner Losh                     offset = (U32)(ip-matchL3);
2200f743729SConrad Meyer                     while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
2210f743729SConrad Meyer                     goto _match_found;
2220f743729SConrad Meyer                 }
2230f743729SConrad Meyer             } else if (dictMode == ZSTD_dictMatchState) {
2240f743729SConrad Meyer                 /* check dict long +1 match */
2250f743729SConrad Meyer                 U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
2260f743729SConrad Meyer                 const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
2270f743729SConrad Meyer                 assert(dictMatchL3 < dictEnd);
2280f743729SConrad Meyer                 if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
2290f743729SConrad Meyer                     mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
2300f743729SConrad Meyer                     ip++;
231f7cd7fe5SConrad Meyer                     offset = (U32)(curr + 1 - dictMatchIndexL3 - dictIndexDelta);
2320f743729SConrad Meyer                     while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
2330f743729SConrad Meyer                     goto _match_found;
2344d3f1eafSConrad Meyer         }   }   }
2350f743729SConrad Meyer 
2360f743729SConrad Meyer         /* if no long +1 match, explore the short match we found */
2370f743729SConrad Meyer         if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
2380f743729SConrad Meyer             mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
239f7cd7fe5SConrad Meyer             offset = (U32)(curr - matchIndexS);
2400f743729SConrad Meyer             while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
2410c16b537SWarner Losh         } else {
2420c16b537SWarner Losh             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
2430c16b537SWarner Losh             offset = (U32)(ip - match);
2440f743729SConrad Meyer             while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
2450c16b537SWarner Losh         }
2460c16b537SWarner Losh 
2470f743729SConrad Meyer         /* fall-through */
2480f743729SConrad Meyer 
2490f743729SConrad Meyer _match_found:
2500c16b537SWarner Losh         offset_2 = offset_1;
2510c16b537SWarner Losh         offset_1 = offset;
2520c16b537SWarner Losh 
2539cbefe25SConrad Meyer         ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
2540c16b537SWarner Losh 
2550f743729SConrad Meyer _match_stored:
2560c16b537SWarner Losh         /* match found */
2570c16b537SWarner Losh         ip += mLength;
2580c16b537SWarner Losh         anchor = ip;
2590c16b537SWarner Losh 
2600c16b537SWarner Losh         if (ip <= ilimit) {
2614d3f1eafSConrad Meyer             /* Complementary insertion */
2624d3f1eafSConrad Meyer             /* done after iLimit test, as candidates could be > iend-8 */
263f7cd7fe5SConrad Meyer             {   U32 const indexToInsert = curr+2;
2644d3f1eafSConrad Meyer                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
2654d3f1eafSConrad Meyer                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
2664d3f1eafSConrad Meyer                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
2674d3f1eafSConrad Meyer                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
2684d3f1eafSConrad Meyer             }
2690c16b537SWarner Losh 
2700c16b537SWarner Losh             /* check immediate repcode */
2710f743729SConrad Meyer             if (dictMode == ZSTD_dictMatchState) {
2720f743729SConrad Meyer                 while (ip <= ilimit) {
2730f743729SConrad Meyer                     U32 const current2 = (U32)(ip-base);
2740f743729SConrad Meyer                     U32 const repIndex2 = current2 - offset_2;
2750f743729SConrad Meyer                     const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
2760f743729SConrad Meyer                         && repIndex2 < prefixLowestIndex ?
27737f1f268SConrad Meyer                             dictBase + repIndex2 - dictIndexDelta :
2780f743729SConrad Meyer                             base + repIndex2;
2790f743729SConrad Meyer                     if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
2800f743729SConrad Meyer                        && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
2810f743729SConrad Meyer                         const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
2820f743729SConrad Meyer                         size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
2830f743729SConrad Meyer                         U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
2849cbefe25SConrad Meyer                         ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
2850f743729SConrad Meyer                         hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
2860f743729SConrad Meyer                         hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
2870f743729SConrad Meyer                         ip += repLength2;
2880f743729SConrad Meyer                         anchor = ip;
2890f743729SConrad Meyer                         continue;
2900f743729SConrad Meyer                     }
2910f743729SConrad Meyer                     break;
2924d3f1eafSConrad Meyer             }   }
2930f743729SConrad Meyer 
2940f743729SConrad Meyer             if (dictMode == ZSTD_noDict) {
2950c16b537SWarner Losh                 while ( (ip <= ilimit)
2960c16b537SWarner Losh                      && ( (offset_2>0)
2970c16b537SWarner Losh                         & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
2980c16b537SWarner Losh                     /* store sequence */
2990c16b537SWarner Losh                     size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
3000f743729SConrad Meyer                     U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
3010c16b537SWarner Losh                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
3020c16b537SWarner Losh                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
3039cbefe25SConrad Meyer                     ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, rLength-MINMATCH);
3040c16b537SWarner Losh                     ip += rLength;
3050c16b537SWarner Losh                     anchor = ip;
3060c16b537SWarner Losh                     continue;   /* faster when present ... (?) */
3074d3f1eafSConrad Meyer         }   }   }
3084d3f1eafSConrad Meyer     }   /* while (ip < ilimit) */
3090c16b537SWarner Losh 
3100c16b537SWarner Losh     /* save reps for next block */
31119fcbaf1SConrad Meyer     rep[0] = offset_1 ? offset_1 : offsetSaved;
31219fcbaf1SConrad Meyer     rep[1] = offset_2 ? offset_2 : offsetSaved;
3130c16b537SWarner Losh 
3140c16b537SWarner Losh     /* Return the last literals size */
3154d3f1eafSConrad Meyer     return (size_t)(iend - anchor);
3160c16b537SWarner Losh }
3170c16b537SWarner Losh 
3180c16b537SWarner Losh 
31919fcbaf1SConrad Meyer size_t ZSTD_compressBlock_doubleFast(
32019fcbaf1SConrad Meyer         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
3210f743729SConrad Meyer         void const* src, size_t srcSize)
3220c16b537SWarner Losh {
323a0483764SConrad Meyer     const U32 mls = ms->cParams.minMatch;
3240c16b537SWarner Losh     switch(mls)
3250c16b537SWarner Losh     {
3260c16b537SWarner Losh     default: /* includes case 3 */
3270c16b537SWarner Losh     case 4 :
3280f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
3290c16b537SWarner Losh     case 5 :
3300f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
3310c16b537SWarner Losh     case 6 :
3320f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
3330c16b537SWarner Losh     case 7 :
3340f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
3350f743729SConrad Meyer     }
3360f743729SConrad Meyer }
3370f743729SConrad Meyer 
3380f743729SConrad Meyer 
3390f743729SConrad Meyer size_t ZSTD_compressBlock_doubleFast_dictMatchState(
3400f743729SConrad Meyer         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
3410f743729SConrad Meyer         void const* src, size_t srcSize)
3420f743729SConrad Meyer {
343a0483764SConrad Meyer     const U32 mls = ms->cParams.minMatch;
3440f743729SConrad Meyer     switch(mls)
3450f743729SConrad Meyer     {
3460f743729SConrad Meyer     default: /* includes case 3 */
3470f743729SConrad Meyer     case 4 :
3480f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
3490f743729SConrad Meyer     case 5 :
3500f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
3510f743729SConrad Meyer     case 6 :
3520f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
3530f743729SConrad Meyer     case 7 :
3540f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
3550c16b537SWarner Losh     }
3560c16b537SWarner Losh }
3570c16b537SWarner Losh 
3580c16b537SWarner Losh 
35919fcbaf1SConrad Meyer static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
36019fcbaf1SConrad Meyer         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
3610f743729SConrad Meyer         void const* src, size_t srcSize,
36219fcbaf1SConrad Meyer         U32 const mls /* template */)
3630c16b537SWarner Losh {
3640f743729SConrad Meyer     ZSTD_compressionParameters const* cParams = &ms->cParams;
36519fcbaf1SConrad Meyer     U32* const hashLong = ms->hashTable;
36619fcbaf1SConrad Meyer     U32  const hBitsL = cParams->hashLog;
36719fcbaf1SConrad Meyer     U32* const hashSmall = ms->chainTable;
36819fcbaf1SConrad Meyer     U32  const hBitsS = cParams->chainLog;
3690c16b537SWarner Losh     const BYTE* const istart = (const BYTE*)src;
3700c16b537SWarner Losh     const BYTE* ip = istart;
3710c16b537SWarner Losh     const BYTE* anchor = istart;
3720c16b537SWarner Losh     const BYTE* const iend = istart + srcSize;
3730c16b537SWarner Losh     const BYTE* const ilimit = iend - 8;
3740f743729SConrad Meyer     const BYTE* const base = ms->window.base;
3754d3f1eafSConrad Meyer     const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
3769cbefe25SConrad Meyer     const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
3774d3f1eafSConrad Meyer     const U32   dictStartIndex = lowLimit;
3784d3f1eafSConrad Meyer     const U32   dictLimit = ms->window.dictLimit;
3794d3f1eafSConrad Meyer     const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
3800f743729SConrad Meyer     const BYTE* const prefixStart = base + prefixStartIndex;
3810f743729SConrad Meyer     const BYTE* const dictBase = ms->window.dictBase;
3820f743729SConrad Meyer     const BYTE* const dictStart = dictBase + dictStartIndex;
3830f743729SConrad Meyer     const BYTE* const dictEnd = dictBase + prefixStartIndex;
38419fcbaf1SConrad Meyer     U32 offset_1=rep[0], offset_2=rep[1];
3850c16b537SWarner Losh 
3860f743729SConrad Meyer     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
3870f743729SConrad Meyer 
3884d3f1eafSConrad Meyer     /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
3894d3f1eafSConrad Meyer     if (prefixStartIndex == dictStartIndex)
3904d3f1eafSConrad Meyer         return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, mls, ZSTD_noDict);
3914d3f1eafSConrad Meyer 
3920c16b537SWarner Losh     /* Search Loop */
3930c16b537SWarner Losh     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
3940c16b537SWarner Losh         const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
3950c16b537SWarner Losh         const U32 matchIndex = hashSmall[hSmall];
3960f743729SConrad Meyer         const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
3970c16b537SWarner Losh         const BYTE* match = matchBase + matchIndex;
3980c16b537SWarner Losh 
3990c16b537SWarner Losh         const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
4000c16b537SWarner Losh         const U32 matchLongIndex = hashLong[hLong];
4010f743729SConrad Meyer         const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
4020c16b537SWarner Losh         const BYTE* matchLong = matchLongBase + matchLongIndex;
4030c16b537SWarner Losh 
404f7cd7fe5SConrad Meyer         const U32 curr = (U32)(ip-base);
405f7cd7fe5SConrad Meyer         const U32 repIndex = curr + 1 - offset_1;   /* offset_1 expected <= curr +1 */
4060f743729SConrad Meyer         const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
4070f743729SConrad Meyer         const BYTE* const repMatch = repBase + repIndex;
4080c16b537SWarner Losh         size_t mLength;
409f7cd7fe5SConrad Meyer         hashSmall[hSmall] = hashLong[hLong] = curr;   /* update hash table */
4100c16b537SWarner Losh 
4110f743729SConrad Meyer         if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
4120f743729SConrad Meyer             & (repIndex > dictStartIndex))
4130c16b537SWarner Losh           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
4140f743729SConrad Meyer             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
4150f743729SConrad Meyer             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
4160c16b537SWarner Losh             ip++;
4179cbefe25SConrad Meyer             ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
4180c16b537SWarner Losh         } else {
4190f743729SConrad Meyer             if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
4200f743729SConrad Meyer                 const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
4210f743729SConrad Meyer                 const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
4220c16b537SWarner Losh                 U32 offset;
4230f743729SConrad Meyer                 mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
424f7cd7fe5SConrad Meyer                 offset = curr - matchLongIndex;
4250c16b537SWarner Losh                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
4260c16b537SWarner Losh                 offset_2 = offset_1;
4270c16b537SWarner Losh                 offset_1 = offset;
4289cbefe25SConrad Meyer                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
4290c16b537SWarner Losh 
4300f743729SConrad Meyer             } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
4310c16b537SWarner Losh                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
4320c16b537SWarner Losh                 U32 const matchIndex3 = hashLong[h3];
4330f743729SConrad Meyer                 const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
4340c16b537SWarner Losh                 const BYTE* match3 = match3Base + matchIndex3;
4350c16b537SWarner Losh                 U32 offset;
436f7cd7fe5SConrad Meyer                 hashLong[h3] = curr + 1;
4370f743729SConrad Meyer                 if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
4380f743729SConrad Meyer                     const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
4390f743729SConrad Meyer                     const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
4400f743729SConrad Meyer                     mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
4410c16b537SWarner Losh                     ip++;
442f7cd7fe5SConrad Meyer                     offset = curr+1 - matchIndex3;
4430c16b537SWarner Losh                     while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
4440c16b537SWarner Losh                 } else {
4450f743729SConrad Meyer                     const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
4460f743729SConrad Meyer                     const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
4470f743729SConrad Meyer                     mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
448f7cd7fe5SConrad Meyer                     offset = curr - matchIndex;
4490c16b537SWarner Losh                     while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
4500c16b537SWarner Losh                 }
4510c16b537SWarner Losh                 offset_2 = offset_1;
4520c16b537SWarner Losh                 offset_1 = offset;
4539cbefe25SConrad Meyer                 ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
4540c16b537SWarner Losh 
4550c16b537SWarner Losh             } else {
45619fcbaf1SConrad Meyer                 ip += ((ip-anchor) >> kSearchStrength) + 1;
4570c16b537SWarner Losh                 continue;
4580c16b537SWarner Losh         }   }
4590c16b537SWarner Losh 
4604d3f1eafSConrad Meyer         /* move to next sequence start */
4610c16b537SWarner Losh         ip += mLength;
4620c16b537SWarner Losh         anchor = ip;
4630c16b537SWarner Losh 
4640c16b537SWarner Losh         if (ip <= ilimit) {
4654d3f1eafSConrad Meyer             /* Complementary insertion */
4664d3f1eafSConrad Meyer             /* done after iLimit test, as candidates could be > iend-8 */
467f7cd7fe5SConrad Meyer             {   U32 const indexToInsert = curr+2;
4684d3f1eafSConrad Meyer                 hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
4690c16b537SWarner Losh                 hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
4704d3f1eafSConrad Meyer                 hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
4714d3f1eafSConrad Meyer                 hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
4724d3f1eafSConrad Meyer             }
4734d3f1eafSConrad Meyer 
4740c16b537SWarner Losh             /* check immediate repcode */
4750c16b537SWarner Losh             while (ip <= ilimit) {
4760c16b537SWarner Losh                 U32 const current2 = (U32)(ip-base);
4770c16b537SWarner Losh                 U32 const repIndex2 = current2 - offset_2;
4780f743729SConrad Meyer                 const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
4790f743729SConrad Meyer                 if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3)   /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
4800f743729SConrad Meyer                     & (repIndex2 > dictStartIndex))
4810c16b537SWarner Losh                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
4820f743729SConrad Meyer                     const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
4830f743729SConrad Meyer                     size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
4840f743729SConrad Meyer                     U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
4859cbefe25SConrad Meyer                     ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
4860c16b537SWarner Losh                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
4870c16b537SWarner Losh                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
4880c16b537SWarner Losh                     ip += repLength2;
4890c16b537SWarner Losh                     anchor = ip;
4900c16b537SWarner Losh                     continue;
4910c16b537SWarner Losh                 }
4920c16b537SWarner Losh                 break;
4930c16b537SWarner Losh     }   }   }
4940c16b537SWarner Losh 
4950c16b537SWarner Losh     /* save reps for next block */
49619fcbaf1SConrad Meyer     rep[0] = offset_1;
49719fcbaf1SConrad Meyer     rep[1] = offset_2;
4980c16b537SWarner Losh 
4990c16b537SWarner Losh     /* Return the last literals size */
5004d3f1eafSConrad Meyer     return (size_t)(iend - anchor);
5010c16b537SWarner Losh }
5020c16b537SWarner Losh 
5030c16b537SWarner Losh 
50419fcbaf1SConrad Meyer size_t ZSTD_compressBlock_doubleFast_extDict(
50519fcbaf1SConrad Meyer         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
5060f743729SConrad Meyer         void const* src, size_t srcSize)
5070c16b537SWarner Losh {
508a0483764SConrad Meyer     U32 const mls = ms->cParams.minMatch;
5090c16b537SWarner Losh     switch(mls)
5100c16b537SWarner Losh     {
5110c16b537SWarner Losh     default: /* includes case 3 */
5120c16b537SWarner Losh     case 4 :
5130f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
5140c16b537SWarner Losh     case 5 :
5150f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
5160c16b537SWarner Losh     case 6 :
5170f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
5180c16b537SWarner Losh     case 7 :
5190f743729SConrad Meyer         return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
5200c16b537SWarner Losh     }
5210c16b537SWarner Losh }
522