brintos

brintos / linux-shallow public Read only

0
0
Text · 26.5 KiB · a752e6b Raw
676 lines · c
1/*2 * Copyright (c) Yann Collet, Facebook, Inc.3 * All rights reserved.4 *5 * This source code is licensed under both the BSD-style license (found in the6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found7 * 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"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */12#include "zstd_fast.h"13 14 15void ZSTD_fillHashTable(ZSTD_matchState_t* ms,16                        const void* const end,17                        ZSTD_dictTableLoadMethod_e dtlm)18{19    const ZSTD_compressionParameters* const cParams = &ms->cParams;20    U32* const hashTable = ms->hashTable;21    U32  const hBits = cParams->hashLog;22    U32  const mls = cParams->minMatch;23    const BYTE* const base = ms->window.base;24    const BYTE* ip = base + ms->nextToUpdate;25    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;26    const U32 fastHashFillStep = 3;27 28    /* Always insert every fastHashFillStep position into the hash table.29     * Insert the other positions if their hash entry is empty.30     */31    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {32        U32 const curr = (U32)(ip - base);33        size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);34        hashTable[hash0] = curr;35        if (dtlm == ZSTD_dtlm_fast) continue;36        /* Only load extra positions for ZSTD_dtlm_full */37        {   U32 p;38            for (p = 1; p < fastHashFillStep; ++p) {39                size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);40                if (hashTable[hash] == 0) {  /* not yet filled */41                    hashTable[hash] = curr + p;42    }   }   }   }43}44 45 46/*47 * If you squint hard enough (and ignore repcodes), the search operation at any48 * given position is broken into 4 stages:49 *50 * 1. Hash   (map position to hash value via input read)51 * 2. Lookup (map hash val to index via hashtable read)52 * 3. Load   (map index to value at that position via input read)53 * 4. Compare54 *55 * Each of these steps involves a memory read at an address which is computed56 * from the previous step. This means these steps must be sequenced and their57 * latencies are cumulative.58 *59 * Rather than do 1->2->3->4 sequentially for a single position before moving60 * onto the next, this implementation interleaves these operations across the61 * next few positions:62 *63 * R = Repcode Read & Compare64 * H = Hash65 * T = Table Lookup66 * M = Match Read & Compare67 *68 * Pos | Time -->69 * ----+-------------------70 * N   | ... M71 * N+1 | ...   TM72 * N+2 |    R H   T M73 * N+3 |         H    TM74 * N+4 |           R H   T M75 * N+5 |                H   ...76 * N+6 |                  R ...77 *78 * This is very much analogous to the pipelining of execution in a CPU. And just79 * like a CPU, we have to dump the pipeline when we find a match (i.e., take a80 * branch).81 *82 * When this happens, we throw away our current state, and do the following prep83 * to re-enter the loop:84 *85 * Pos | Time -->86 * ----+-------------------87 * N   | H T88 * N+1 |  H89 *90 * This is also the work we do at the beginning to enter the loop initially.91 */92FORCE_INLINE_TEMPLATE size_t93ZSTD_compressBlock_fast_noDict_generic(94        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],95        void const* src, size_t srcSize,96        U32 const mls, U32 const hasStep)97{98    const ZSTD_compressionParameters* const cParams = &ms->cParams;99    U32* const hashTable = ms->hashTable;100    U32 const hlog = cParams->hashLog;101    /* support stepSize of 0 */102    size_t const stepSize = hasStep ? (cParams->targetLength + !(cParams->targetLength) + 1) : 2;103    const BYTE* const base = ms->window.base;104    const BYTE* const istart = (const BYTE*)src;105    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);106    const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);107    const BYTE* const prefixStart = base + prefixStartIndex;108    const BYTE* const iend = istart + srcSize;109    const BYTE* const ilimit = iend - HASH_READ_SIZE;110 111    const BYTE* anchor = istart;112    const BYTE* ip0 = istart;113    const BYTE* ip1;114    const BYTE* ip2;115    const BYTE* ip3;116    U32 current0;117 118    U32 rep_offset1 = rep[0];119    U32 rep_offset2 = rep[1];120    U32 offsetSaved = 0;121 122    size_t hash0; /* hash for ip0 */123    size_t hash1; /* hash for ip1 */124    U32 idx; /* match idx for ip0 */125    U32 mval; /* src value at match idx */126 127    U32 offcode;128    const BYTE* match0;129    size_t mLength;130 131    /* ip0 and ip1 are always adjacent. The targetLength skipping and132     * uncompressibility acceleration is applied to every other position,133     * matching the behavior of #1562. step therefore represents the gap134     * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */135    size_t step;136    const BYTE* nextStep;137    const size_t kStepIncr = (1 << (kSearchStrength - 1));138 139    DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");140    ip0 += (ip0 == prefixStart);141    {   U32 const curr = (U32)(ip0 - base);142        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);143        U32 const maxRep = curr - windowLow;144        if (rep_offset2 > maxRep) offsetSaved = rep_offset2, rep_offset2 = 0;145        if (rep_offset1 > maxRep) offsetSaved = rep_offset1, rep_offset1 = 0;146    }147 148    /* start each op */149_start: /* Requires: ip0 */150 151    step = stepSize;152    nextStep = ip0 + kStepIncr;153 154    /* calculate positions, ip0 - anchor == 0, so we skip step calc */155    ip1 = ip0 + 1;156    ip2 = ip0 + step;157    ip3 = ip2 + 1;158 159    if (ip3 >= ilimit) {160        goto _cleanup;161    }162 163    hash0 = ZSTD_hashPtr(ip0, hlog, mls);164    hash1 = ZSTD_hashPtr(ip1, hlog, mls);165 166    idx = hashTable[hash0];167 168    do {169        /* load repcode match for ip[2]*/170        const U32 rval = MEM_read32(ip2 - rep_offset1);171 172        /* write back hash table entry */173        current0 = (U32)(ip0 - base);174        hashTable[hash0] = current0;175 176        /* check repcode at ip[2] */177        if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {178            ip0 = ip2;179            match0 = ip0 - rep_offset1;180            mLength = ip0[-1] == match0[-1];181            ip0 -= mLength;182            match0 -= mLength;183            offcode = STORE_REPCODE_1;184            mLength += 4;185            goto _match;186        }187 188        /* load match for ip[0] */189        if (idx >= prefixStartIndex) {190            mval = MEM_read32(base + idx);191        } else {192            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */193        }194 195        /* check match at ip[0] */196        if (MEM_read32(ip0) == mval) {197            /* found a match! */198            goto _offset;199        }200 201        /* lookup ip[1] */202        idx = hashTable[hash1];203 204        /* hash ip[2] */205        hash0 = hash1;206        hash1 = ZSTD_hashPtr(ip2, hlog, mls);207 208        /* advance to next positions */209        ip0 = ip1;210        ip1 = ip2;211        ip2 = ip3;212 213        /* write back hash table entry */214        current0 = (U32)(ip0 - base);215        hashTable[hash0] = current0;216 217        /* load match for ip[0] */218        if (idx >= prefixStartIndex) {219            mval = MEM_read32(base + idx);220        } else {221            mval = MEM_read32(ip0) ^ 1; /* guaranteed to not match. */222        }223 224        /* check match at ip[0] */225        if (MEM_read32(ip0) == mval) {226            /* found a match! */227            goto _offset;228        }229 230        /* lookup ip[1] */231        idx = hashTable[hash1];232 233        /* hash ip[2] */234        hash0 = hash1;235        hash1 = ZSTD_hashPtr(ip2, hlog, mls);236 237        /* advance to next positions */238        ip0 = ip1;239        ip1 = ip2;240        ip2 = ip0 + step;241        ip3 = ip1 + step;242 243        /* calculate step */244        if (ip2 >= nextStep) {245            step++;246            PREFETCH_L1(ip1 + 64);247            PREFETCH_L1(ip1 + 128);248            nextStep += kStepIncr;249        }250    } while (ip3 < ilimit);251 252_cleanup:253    /* Note that there are probably still a couple positions we could search.254     * However, it seems to be a meaningful performance hit to try to search255     * them. So let's not. */256 257    /* save reps for next block */258    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved;259    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved;260 261    /* Return the last literals size */262    return (size_t)(iend - anchor);263 264_offset: /* Requires: ip0, idx */265 266    /* Compute the offset code. */267    match0 = base + idx;268    rep_offset2 = rep_offset1;269    rep_offset1 = (U32)(ip0-match0);270    offcode = STORE_OFFSET(rep_offset1);271    mLength = 4;272 273    /* Count the backwards match length. */274    while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {275        ip0--;276        match0--;277        mLength++;278    }279 280_match: /* Requires: ip0, match0, offcode */281 282    /* Count the forward length. */283    mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);284 285    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);286 287    ip0 += mLength;288    anchor = ip0;289 290    /* write next hash table entry */291    if (ip1 < ip0) {292        hashTable[hash1] = (U32)(ip1 - base);293    }294 295    /* Fill table and check for immediate repcode. */296    if (ip0 <= ilimit) {297        /* Fill Table */298        assert(base+current0+2 > istart);  /* check base overflow */299        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */300        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);301 302        if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */303            while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {304                /* store sequence */305                size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;306                { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */307                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);308                ip0 += rLength;309                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, STORE_REPCODE_1, rLength);310                anchor = ip0;311                continue;   /* faster when present (confirmed on gcc-8) ... (?) */312    }   }   }313 314    goto _start;315}316 317#define ZSTD_GEN_FAST_FN(dictMode, mls, step)                                                            \318    static size_t ZSTD_compressBlock_fast_##dictMode##_##mls##_##step(                                      \319            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \320            void const* src, size_t srcSize)                                                       \321    {                                                                                              \322        return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mls, step); \323    }324 325ZSTD_GEN_FAST_FN(noDict, 4, 1)326ZSTD_GEN_FAST_FN(noDict, 5, 1)327ZSTD_GEN_FAST_FN(noDict, 6, 1)328ZSTD_GEN_FAST_FN(noDict, 7, 1)329 330ZSTD_GEN_FAST_FN(noDict, 4, 0)331ZSTD_GEN_FAST_FN(noDict, 5, 0)332ZSTD_GEN_FAST_FN(noDict, 6, 0)333ZSTD_GEN_FAST_FN(noDict, 7, 0)334 335size_t ZSTD_compressBlock_fast(336        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],337        void const* src, size_t srcSize)338{339    U32 const mls = ms->cParams.minMatch;340    assert(ms->dictMatchState == NULL);341    if (ms->cParams.targetLength > 1) {342        switch(mls)343        {344        default: /* includes case 3 */345        case 4 :346            return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);347        case 5 :348            return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);349        case 6 :350            return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);351        case 7 :352            return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);353        }354    } else {355        switch(mls)356        {357        default: /* includes case 3 */358        case 4 :359            return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);360        case 5 :361            return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);362        case 6 :363            return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);364        case 7 :365            return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);366        }367 368    }369}370 371FORCE_INLINE_TEMPLATE372size_t ZSTD_compressBlock_fast_dictMatchState_generic(373        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],374        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)375{376    const ZSTD_compressionParameters* const cParams = &ms->cParams;377    U32* const hashTable = ms->hashTable;378    U32 const hlog = cParams->hashLog;379    /* support stepSize of 0 */380    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);381    const BYTE* const base = ms->window.base;382    const BYTE* const istart = (const BYTE*)src;383    const BYTE* ip = istart;384    const BYTE* anchor = istart;385    const U32   prefixStartIndex = ms->window.dictLimit;386    const BYTE* const prefixStart = base + prefixStartIndex;387    const BYTE* const iend = istart + srcSize;388    const BYTE* const ilimit = iend - HASH_READ_SIZE;389    U32 offset_1=rep[0], offset_2=rep[1];390    U32 offsetSaved = 0;391 392    const ZSTD_matchState_t* const dms = ms->dictMatchState;393    const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;394    const U32* const dictHashTable = dms->hashTable;395    const U32 dictStartIndex       = dms->window.dictLimit;396    const BYTE* const dictBase     = dms->window.base;397    const BYTE* const dictStart    = dictBase + dictStartIndex;398    const BYTE* const dictEnd      = dms->window.nextSrc;399    const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);400    const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);401    const U32 dictHLog             = dictCParams->hashLog;402 403    /* if a dictionary is still attached, it necessarily means that404     * it is within window size. So we just check it. */405    const U32 maxDistance = 1U << cParams->windowLog;406    const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);407    assert(endIndex - prefixStartIndex <= maxDistance);408    (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */409 410    (void)hasStep; /* not currently specialized on whether it's accelerated */411 412    /* ensure there will be no underflow413     * when translating a dict index into a local index */414    assert(prefixStartIndex >= (U32)(dictEnd - dictBase));415 416    /* init */417    DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");418    ip += (dictAndPrefixLength == 0);419    /* dictMatchState repCode checks don't currently handle repCode == 0420     * disabling. */421    assert(offset_1 <= dictAndPrefixLength);422    assert(offset_2 <= dictAndPrefixLength);423 424    /* Main Search Loop */425    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */426        size_t mLength;427        size_t const h = ZSTD_hashPtr(ip, hlog, mls);428        U32 const curr = (U32)(ip-base);429        U32 const matchIndex = hashTable[h];430        const BYTE* match = base + matchIndex;431        const U32 repIndex = curr + 1 - offset_1;432        const BYTE* repMatch = (repIndex < prefixStartIndex) ?433                               dictBase + (repIndex - dictIndexDelta) :434                               base + repIndex;435        hashTable[h] = curr;   /* update hash table */436 437        if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */438          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {439            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;440            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;441            ip++;442            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, mLength);443        } else if ( (matchIndex <= prefixStartIndex) ) {444            size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);445            U32 const dictMatchIndex = dictHashTable[dictHash];446            const BYTE* dictMatch = dictBase + dictMatchIndex;447            if (dictMatchIndex <= dictStartIndex ||448                MEM_read32(dictMatch) != MEM_read32(ip)) {449                assert(stepSize >= 1);450                ip += ((ip-anchor) >> kSearchStrength) + stepSize;451                continue;452            } else {453                /* found a dict match */454                U32 const offset = (U32)(curr-dictMatchIndex-dictIndexDelta);455                mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;456                while (((ip>anchor) & (dictMatch>dictStart))457                     && (ip[-1] == dictMatch[-1])) {458                    ip--; dictMatch--; mLength++;459                } /* catch up */460                offset_2 = offset_1;461                offset_1 = offset;462                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);463            }464        } else if (MEM_read32(match) != MEM_read32(ip)) {465            /* it's not a match, and we're not going to check the dictionary */466            assert(stepSize >= 1);467            ip += ((ip-anchor) >> kSearchStrength) + stepSize;468            continue;469        } else {470            /* found a regular match */471            U32 const offset = (U32)(ip-match);472            mLength = ZSTD_count(ip+4, match+4, iend) + 4;473            while (((ip>anchor) & (match>prefixStart))474                 && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */475            offset_2 = offset_1;476            offset_1 = offset;477            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);478        }479 480        /* match found */481        ip += mLength;482        anchor = ip;483 484        if (ip <= ilimit) {485            /* Fill Table */486            assert(base+curr+2 > istart);  /* check base overflow */487            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */488            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);489 490            /* check immediate repcode */491            while (ip <= ilimit) {492                U32 const current2 = (U32)(ip-base);493                U32 const repIndex2 = current2 - offset_2;494                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?495                        dictBase - dictIndexDelta + repIndex2 :496                        base + repIndex2;497                if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)498                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {499                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;500                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;501                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */502                    ZSTD_storeSeq(seqStore, 0, anchor, iend, STORE_REPCODE_1, repLength2);503                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;504                    ip += repLength2;505                    anchor = ip;506                    continue;507                }508                break;509            }510        }511    }512 513    /* save reps for next block */514    rep[0] = offset_1 ? offset_1 : offsetSaved;515    rep[1] = offset_2 ? offset_2 : offsetSaved;516 517    /* Return the last literals size */518    return (size_t)(iend - anchor);519}520 521 522ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)523ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)524ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)525ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)526 527size_t ZSTD_compressBlock_fast_dictMatchState(528        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],529        void const* src, size_t srcSize)530{531    U32 const mls = ms->cParams.minMatch;532    assert(ms->dictMatchState != NULL);533    switch(mls)534    {535    default: /* includes case 3 */536    case 4 :537        return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);538    case 5 :539        return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);540    case 6 :541        return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);542    case 7 :543        return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);544    }545}546 547 548static size_t ZSTD_compressBlock_fast_extDict_generic(549        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],550        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)551{552    const ZSTD_compressionParameters* const cParams = &ms->cParams;553    U32* const hashTable = ms->hashTable;554    U32 const hlog = cParams->hashLog;555    /* support stepSize of 0 */556    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);557    const BYTE* const base = ms->window.base;558    const BYTE* const dictBase = ms->window.dictBase;559    const BYTE* const istart = (const BYTE*)src;560    const BYTE* ip = istart;561    const BYTE* anchor = istart;562    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);563    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);564    const U32   dictStartIndex = lowLimit;565    const BYTE* const dictStart = dictBase + dictStartIndex;566    const U32   dictLimit = ms->window.dictLimit;567    const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;568    const BYTE* const prefixStart = base + prefixStartIndex;569    const BYTE* const dictEnd = dictBase + prefixStartIndex;570    const BYTE* const iend = istart + srcSize;571    const BYTE* const ilimit = iend - 8;572    U32 offset_1=rep[0], offset_2=rep[1];573 574    (void)hasStep; /* not currently specialized on whether it's accelerated */575 576    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);577 578    /* switch to "regular" variant if extDict is invalidated due to maxDistance */579    if (prefixStartIndex == dictStartIndex)580        return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);581 582    /* Search Loop */583    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */584        const size_t h = ZSTD_hashPtr(ip, hlog, mls);585        const U32    matchIndex = hashTable[h];586        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;587        const BYTE*  match = matchBase + matchIndex;588        const U32    curr = (U32)(ip-base);589        const U32    repIndex = curr + 1 - offset_1;590        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;591        const BYTE* const repMatch = repBase + repIndex;592        hashTable[h] = curr;   /* update hash table */593        DEBUGLOG(7, "offset_1 = %u , curr = %u", offset_1, curr);594 595        if ( ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */596             & (offset_1 <= curr+1 - dictStartIndex) ) /* note: we are searching at curr+1 */597           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {598            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;599            size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;600            ip++;601            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_REPCODE_1, rLength);602            ip += rLength;603            anchor = ip;604        } else {605            if ( (matchIndex < dictStartIndex) ||606                 (MEM_read32(match) != MEM_read32(ip)) ) {607                assert(stepSize >= 1);608                ip += ((ip-anchor) >> kSearchStrength) + stepSize;609                continue;610            }611            {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;612                const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;613                U32 const offset = curr - matchIndex;614                size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;615                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */616                offset_2 = offset_1; offset_1 = offset;  /* update offset history */617                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, STORE_OFFSET(offset), mLength);618                ip += mLength;619                anchor = ip;620        }   }621 622        if (ip <= ilimit) {623            /* Fill Table */624            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;625            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);626            /* check immediate repcode */627            while (ip <= ilimit) {628                U32 const current2 = (U32)(ip-base);629                U32 const repIndex2 = current2 - offset_2;630                const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;631                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (offset_2 <= curr - dictStartIndex))  /* intentional overflow */632                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {633                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;634                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;635                    { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */636                    ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, STORE_REPCODE_1, repLength2);637                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;638                    ip += repLength2;639                    anchor = ip;640                    continue;641                }642                break;643    }   }   }644 645    /* save reps for next block */646    rep[0] = offset_1;647    rep[1] = offset_2;648 649    /* Return the last literals size */650    return (size_t)(iend - anchor);651}652 653ZSTD_GEN_FAST_FN(extDict, 4, 0)654ZSTD_GEN_FAST_FN(extDict, 5, 0)655ZSTD_GEN_FAST_FN(extDict, 6, 0)656ZSTD_GEN_FAST_FN(extDict, 7, 0)657 658size_t ZSTD_compressBlock_fast_extDict(659        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],660        void const* src, size_t srcSize)661{662    U32 const mls = ms->cParams.minMatch;663    switch(mls)664    {665    default: /* includes case 3 */666    case 4 :667        return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);668    case 5 :669        return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);670    case 6 :671        return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);672    case 7 :673        return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);674    }675}676