lib/compress/zstd_fast.c - dev - githubmirror/zstd

lib/compress/zstd_fast.c
985 строк37 KB
Yann Collet
codemod: ZSTD_matchState_t -> ZSTD_MatchState_t
20 дек 2024, 21:36
20 дек 2024, 21:365df80ac
/*
 * Copyright (c) Meta Platforms, Inc. and affiliates.
 * All rights reserved.
 *
 * This source code is licensed under both the BSD-style license (found in the
 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
 * in the COPYING file in the root directory of this source tree).
 * You may select, at your option, one of the above-listed licenses.
 */

#include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
#include "zstd_fast.h"

static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
void ZSTD_fillHashTableForCDict(ZSTD_MatchState_t* ms,
                        const void* const end,
                        ZSTD_dictTableLoadMethod_e dtlm)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32  const hBits = cParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;
    U32  const mls = cParams->minMatch;
    const BYTE* const base = ms->window.base;
    const BYTE* ip = base + ms->nextToUpdate;
    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
    const U32 fastHashFillStep = 3;

    /* Currently, we always use ZSTD_dtlm_full for filling CDict tables.
     * Feel free to remove this assert if there's a good reason! */
    assert(dtlm == ZSTD_dtlm_full);

    /* Always insert every fastHashFillStep position into the hash table.
     * Insert the other positions if their hash entry is empty.
     */
    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
        U32 const curr = (U32)(ip - base);
        {   size_t const hashAndTag = ZSTD_hashPtr(ip, hBits, mls);
            ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr);   }

        if (dtlm == ZSTD_dtlm_fast) continue;
        /* Only load extra positions for ZSTD_dtlm_full */
        {   U32 p;
            for (p = 1; p < fastHashFillStep; ++p) {
                size_t const hashAndTag = ZSTD_hashPtr(ip + p, hBits, mls);
                if (hashTable[hashAndTag >> ZSTD_SHORT_CACHE_TAG_BITS] == 0) {  /* not yet filled */
                    ZSTD_writeTaggedIndex(hashTable, hashAndTag, curr + p);
    }   }   }   }
}

static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
void ZSTD_fillHashTableForCCtx(ZSTD_MatchState_t* ms,
                        const void* const end,
                        ZSTD_dictTableLoadMethod_e dtlm)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32  const hBits = cParams->hashLog;
    U32  const mls = cParams->minMatch;
    const BYTE* const base = ms->window.base;
    const BYTE* ip = base + ms->nextToUpdate;
    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
    const U32 fastHashFillStep = 3;

    /* Currently, we always use ZSTD_dtlm_fast for filling CCtx tables.
     * Feel free to remove this assert if there's a good reason! */
    assert(dtlm == ZSTD_dtlm_fast);

    /* Always insert every fastHashFillStep position into the hash table.
     * Insert the other positions if their hash entry is empty.
     */
    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
        U32 const curr = (U32)(ip - base);
        size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
        hashTable[hash0] = curr;
        if (dtlm == ZSTD_dtlm_fast) continue;
        /* Only load extra positions for ZSTD_dtlm_full */
        {   U32 p;
            for (p = 1; p < fastHashFillStep; ++p) {
                size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
                if (hashTable[hash] == 0) {  /* not yet filled */
                    hashTable[hash] = curr + p;
    }   }   }   }
}

void ZSTD_fillHashTable(ZSTD_MatchState_t* ms,
                        const void* const end,
                        ZSTD_dictTableLoadMethod_e dtlm,
                        ZSTD_tableFillPurpose_e tfp)
{
    if (tfp == ZSTD_tfp_forCDict) {
        ZSTD_fillHashTableForCDict(ms, end, dtlm);
    } else {
        ZSTD_fillHashTableForCCtx(ms, end, dtlm);
    }
}


typedef int (*ZSTD_match4Found) (const BYTE* currentPtr, const BYTE* matchAddress, U32 matchIdx, U32 idxLowLimit);

static int
ZSTD_match4Found_cmov(const BYTE* currentPtr, const BYTE* matchAddress, U32 matchIdx, U32 idxLowLimit)
{
    /* Array of ~random data, should have low probability of matching data.
     * Load from here if the index is invalid.
     * Used to avoid unpredictable branches. */
    static const BYTE dummy[] = {0x12,0x34,0x56,0x78};

    /* currentIdx >= lowLimit is a (somewhat) unpredictable branch.
     * However expression below compiles into conditional move.
     */
    const BYTE* mvalAddr = ZSTD_selectAddr(matchIdx, idxLowLimit, matchAddress, dummy);
    /* Note: this used to be written as : return test1 && test2;
     * Unfortunately, once inlined, these tests become branches,
     * in which case it becomes critical that they are executed in the right order (test1 then test2).
     * So we have to write these tests in a specific manner to ensure their ordering.
     */
    if (MEM_read32(currentPtr) != MEM_read32(mvalAddr)) return 0;
    /* force ordering of these tests, which matters once the function is inlined, as they become branches */
#if defined(__GNUC__)
    __asm__("");
#endif
    return matchIdx >= idxLowLimit;
}

static int
ZSTD_match4Found_branch(const BYTE* currentPtr, const BYTE* matchAddress, U32 matchIdx, U32 idxLowLimit)
{
    /* using a branch instead of a cmov,
     * because it's faster in scenarios where matchIdx >= idxLowLimit is generally true,
     * aka almost all candidates are within range */
    U32 mval;
    if (matchIdx >= idxLowLimit) {
        mval = MEM_read32(matchAddress);
    } else {
        mval = MEM_read32(currentPtr) ^ 1; /* guaranteed to not match. */
    }

    return (MEM_read32(currentPtr) == mval);
}


/**
 * If you squint hard enough (and ignore repcodes), the search operation at any
 * given position is broken into 4 stages:
 *
 * 1. Hash   (map position to hash value via input read)
 * 2. Lookup (map hash val to index via hashtable read)
 * 3. Load   (map index to value at that position via input read)
 * 4. Compare
 *
 * Each of these steps involves a memory read at an address which is computed
 * from the previous step. This means these steps must be sequenced and their
 * latencies are cumulative.
 *
 * Rather than do 1->2->3->4 sequentially for a single position before moving
 * onto the next, this implementation interleaves these operations across the
 * next few positions:
 *
 * R = Repcode Read & Compare
 * H = Hash
 * T = Table Lookup
 * M = Match Read & Compare
 *
 * Pos | Time -->
 * ----+-------------------
 * N   | ... M
 * N+1 | ...   TM
 * N+2 |    R H   T M
 * N+3 |         H    TM
 * N+4 |           R H   T M
 * N+5 |                H   ...
 * N+6 |                  R ...
 *
 * This is very much analogous to the pipelining of execution in a CPU. And just
 * like a CPU, we have to dump the pipeline when we find a match (i.e., take a
 * branch).
 *
 * When this happens, we throw away our current state, and do the following prep
 * to re-enter the loop:
 *
 * Pos | Time -->
 * ----+-------------------
 * N   | H T
 * N+1 |  H
 *
 * This is also the work we do at the beginning to enter the loop initially.
 */
FORCE_INLINE_TEMPLATE
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_fast_noDict_generic(
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize,
        U32 const mls, int useCmov)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32 const hlog = cParams->hashLog;
    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1; /* min 2 */
    const BYTE* const base = ms->window.base;
    const BYTE* const istart = (const BYTE*)src;
    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
    const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
    const BYTE* const prefixStart = base + prefixStartIndex;
    const BYTE* const iend = istart + srcSize;
    const BYTE* const ilimit = iend - HASH_READ_SIZE;

    const BYTE* anchor = istart;
    const BYTE* ip0 = istart;
    const BYTE* ip1;
    const BYTE* ip2;
    const BYTE* ip3;
    U32 current0;

    U32 rep_offset1 = rep[0];
    U32 rep_offset2 = rep[1];
    U32 offsetSaved1 = 0, offsetSaved2 = 0;

    size_t hash0; /* hash for ip0 */
    size_t hash1; /* hash for ip1 */
    U32 matchIdx; /* match idx for ip0 */

    U32 offcode;
    const BYTE* match0;
    size_t mLength;

    /* ip0 and ip1 are always adjacent. The targetLength skipping and
     * uncompressibility acceleration is applied to every other position,
     * matching the behavior of #1562. step therefore represents the gap
     * between pairs of positions, from ip0 to ip2 or ip1 to ip3. */
    size_t step;
    const BYTE* nextStep;
    const size_t kStepIncr = (1 << (kSearchStrength - 1));
    const ZSTD_match4Found matchFound = useCmov ? ZSTD_match4Found_cmov : ZSTD_match4Found_branch;

    DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
    ip0 += (ip0 == prefixStart);
    {   U32 const curr = (U32)(ip0 - base);
        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, cParams->windowLog);
        U32 const maxRep = curr - windowLow;
        if (rep_offset2 > maxRep) offsetSaved2 = rep_offset2, rep_offset2 = 0;
        if (rep_offset1 > maxRep) offsetSaved1 = rep_offset1, rep_offset1 = 0;
    }

    /* start each op */
_start: /* Requires: ip0 */

    step = stepSize;
    nextStep = ip0 + kStepIncr;

    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
    ip1 = ip0 + 1;
    ip2 = ip0 + step;
    ip3 = ip2 + 1;

    if (ip3 >= ilimit) {
        goto _cleanup;
    }

    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
    hash1 = ZSTD_hashPtr(ip1, hlog, mls);

    matchIdx = hashTable[hash0];

    do {
        /* load repcode match for ip[2]*/
        const U32 rval = MEM_read32(ip2 - rep_offset1);

        /* write back hash table entry */
        current0 = (U32)(ip0 - base);
        hashTable[hash0] = current0;

        /* check repcode at ip[2] */
        if ((MEM_read32(ip2) == rval) & (rep_offset1 > 0)) {
            ip0 = ip2;
            match0 = ip0 - rep_offset1;
            mLength = ip0[-1] == match0[-1];
            ip0 -= mLength;
            match0 -= mLength;
            offcode = REPCODE1_TO_OFFBASE;
            mLength += 4;

            /* Write next hash table entry: it's already calculated.
             * This write is known to be safe because ip1 is before the
             * repcode (ip2). */
            hashTable[hash1] = (U32)(ip1 - base);

            goto _match;
        }

         if (matchFound(ip0, base + matchIdx, matchIdx, prefixStartIndex)) {
            /* Write next hash table entry (it's already calculated).
            * This write is known to be safe because the ip1 == ip0 + 1,
            * so searching will resume after ip1 */
            hashTable[hash1] = (U32)(ip1 - base);

            goto _offset;
        }

        /* lookup ip[1] */
        matchIdx = hashTable[hash1];

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip3;

        /* write back hash table entry */
        current0 = (U32)(ip0 - base);
        hashTable[hash0] = current0;

         if (matchFound(ip0, base + matchIdx, matchIdx, prefixStartIndex)) {
            /* Write next hash table entry, since it's already calculated */
            if (step <= 4) {
                /* Avoid writing an index if it's >= position where search will resume.
                * The minimum possible match has length 4, so search can resume at ip0 + 4.
                */
                hashTable[hash1] = (U32)(ip1 - base);
            }
            goto _offset;
        }

        /* lookup ip[1] */
        matchIdx = hashTable[hash1];

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip0 + step;
        ip3 = ip1 + step;

        /* calculate step */
        if (ip2 >= nextStep) {
            step++;
            PREFETCH_L1(ip1 + 64);
            PREFETCH_L1(ip1 + 128);
            nextStep += kStepIncr;
        }
    } while (ip3 < ilimit);

_cleanup:
    /* Note that there are probably still a couple positions one could search.
     * However, it seems to be a meaningful performance hit to try to search
     * them. So let's not. */

    /* When the repcodes are outside of the prefix, we set them to zero before the loop.
     * When the offsets are still zero, we need to restore them after the block to have a correct
     * repcode history. If only one offset was invalid, it is easy. The tricky case is when both
     * offsets were invalid. We need to figure out which offset to refill with.
     *     - If both offsets are zero they are in the same order.
     *     - If both offsets are non-zero, we won't restore the offsets from `offsetSaved[12]`.
     *     - If only one is zero, we need to decide which offset to restore.
     *         - If rep_offset1 is non-zero, then rep_offset2 must be offsetSaved1.
     *         - It is impossible for rep_offset2 to be non-zero.
     *
     * So if rep_offset1 started invalid (offsetSaved1 != 0) and became valid (rep_offset1 != 0), then
     * set rep[0] = rep_offset1 and rep[1] = offsetSaved1.
     */
    offsetSaved2 = ((offsetSaved1 != 0) && (rep_offset1 != 0)) ? offsetSaved1 : offsetSaved2;

    /* save reps for next block */
    rep[0] = rep_offset1 ? rep_offset1 : offsetSaved1;
    rep[1] = rep_offset2 ? rep_offset2 : offsetSaved2;

    /* Return the last literals size */
    return (size_t)(iend - anchor);

_offset: /* Requires: ip0, idx */

    /* Compute the offset code. */
    match0 = base + matchIdx;
    rep_offset2 = rep_offset1;
    rep_offset1 = (U32)(ip0-match0);
    offcode = OFFSET_TO_OFFBASE(rep_offset1);
    mLength = 4;

    /* Count the backwards match length. */
    while (((ip0>anchor) & (match0>prefixStart)) && (ip0[-1] == match0[-1])) {
        ip0--;
        match0--;
        mLength++;
    }

_match: /* Requires: ip0, match0, offcode */

    /* Count the forward length. */
    mLength += ZSTD_count(ip0 + mLength, match0 + mLength, iend);

    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);

    ip0 += mLength;
    anchor = ip0;

    /* Fill table and check for immediate repcode. */
    if (ip0 <= ilimit) {
        /* Fill Table */
        assert(base+current0+2 > istart);  /* check base overflow */
        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);

        if (rep_offset2 > 0) { /* rep_offset2==0 means rep_offset2 is invalidated */
            while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - rep_offset2)) ) {
                /* store sequence */
                size_t const rLength = ZSTD_count(ip0+4, ip0+4-rep_offset2, iend) + 4;
                { U32 const tmpOff = rep_offset2; rep_offset2 = rep_offset1; rep_offset1 = tmpOff; } /* swap rep_offset2 <=> rep_offset1 */
                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
                ip0 += rLength;
                ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, REPCODE1_TO_OFFBASE, rLength);
                anchor = ip0;
                continue;   /* faster when present (confirmed on gcc-8) ... (?) */
    }   }   }

    goto _start;
}

#define ZSTD_GEN_FAST_FN(dictMode, mml, cmov)                                                       \
    static size_t ZSTD_compressBlock_fast_##dictMode##_##mml##_##cmov(                              \
            ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],                    \
            void const* src, size_t srcSize)                                                       \
    {                                                                                              \
        return ZSTD_compressBlock_fast_##dictMode##_generic(ms, seqStore, rep, src, srcSize, mml, cmov); \
    }

ZSTD_GEN_FAST_FN(noDict, 4, 1)
ZSTD_GEN_FAST_FN(noDict, 5, 1)
ZSTD_GEN_FAST_FN(noDict, 6, 1)
ZSTD_GEN_FAST_FN(noDict, 7, 1)

ZSTD_GEN_FAST_FN(noDict, 4, 0)
ZSTD_GEN_FAST_FN(noDict, 5, 0)
ZSTD_GEN_FAST_FN(noDict, 6, 0)
ZSTD_GEN_FAST_FN(noDict, 7, 0)

size_t ZSTD_compressBlock_fast(
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize)
{
    U32 const mml = ms->cParams.minMatch;
    /* use cmov when "candidate in range" branch is likely unpredictable */
    int const useCmov = ms->cParams.windowLog < 19;
    assert(ms->dictMatchState == NULL);
    if (useCmov) {
        switch(mml)
        {
        default: /* includes case 3 */
        case 4 :
            return ZSTD_compressBlock_fast_noDict_4_1(ms, seqStore, rep, src, srcSize);
        case 5 :
            return ZSTD_compressBlock_fast_noDict_5_1(ms, seqStore, rep, src, srcSize);
        case 6 :
            return ZSTD_compressBlock_fast_noDict_6_1(ms, seqStore, rep, src, srcSize);
        case 7 :
            return ZSTD_compressBlock_fast_noDict_7_1(ms, seqStore, rep, src, srcSize);
        }
    } else {
        /* use a branch instead */
        switch(mml)
        {
        default: /* includes case 3 */
        case 4 :
            return ZSTD_compressBlock_fast_noDict_4_0(ms, seqStore, rep, src, srcSize);
        case 5 :
            return ZSTD_compressBlock_fast_noDict_5_0(ms, seqStore, rep, src, srcSize);
        case 6 :
            return ZSTD_compressBlock_fast_noDict_6_0(ms, seqStore, rep, src, srcSize);
        case 7 :
            return ZSTD_compressBlock_fast_noDict_7_0(ms, seqStore, rep, src, srcSize);
        }
    }
}

FORCE_INLINE_TEMPLATE
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_fast_dictMatchState_generic(
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32 const hlog = cParams->hashLog;
    /* support stepSize of 0 */
    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
    const BYTE* const base = ms->window.base;
    const BYTE* const istart = (const BYTE*)src;
    const BYTE* ip0 = istart;
    const BYTE* ip1 = ip0 + stepSize; /* we assert below that stepSize >= 1 */
    const BYTE* anchor = istart;
    const U32   prefixStartIndex = ms->window.dictLimit;
    const BYTE* const prefixStart = base + prefixStartIndex;
    const BYTE* const iend = istart + srcSize;
    const BYTE* const ilimit = iend - HASH_READ_SIZE;
    U32 offset_1=rep[0], offset_2=rep[1];

    const ZSTD_MatchState_t* const dms = ms->dictMatchState;
    const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
    const U32* const dictHashTable = dms->hashTable;
    const U32 dictStartIndex       = dms->window.dictLimit;
    const BYTE* const dictBase     = dms->window.base;
    const BYTE* const dictStart    = dictBase + dictStartIndex;
    const BYTE* const dictEnd      = dms->window.nextSrc;
    const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
    const U32 dictAndPrefixLength  = (U32)(istart - prefixStart + dictEnd - dictStart);
    const U32 dictHBits            = dictCParams->hashLog + ZSTD_SHORT_CACHE_TAG_BITS;

    /* if a dictionary is still attached, it necessarily means that
     * it is within window size. So we just check it. */
    const U32 maxDistance = 1U << cParams->windowLog;
    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
    assert(endIndex - prefixStartIndex <= maxDistance);
    (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */

    (void)hasStep; /* not currently specialized on whether it's accelerated */

    /* ensure there will be no underflow
     * when translating a dict index into a local index */
    assert(prefixStartIndex >= (U32)(dictEnd - dictBase));

    if (ms->prefetchCDictTables) {
        size_t const hashTableBytes = (((size_t)1) << dictCParams->hashLog) * sizeof(U32);
        PREFETCH_AREA(dictHashTable, hashTableBytes);
    }

    /* init */
    DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
    ip0 += (dictAndPrefixLength == 0);
    /* dictMatchState repCode checks don't currently handle repCode == 0
     * disabling. */
    assert(offset_1 <= dictAndPrefixLength);
    assert(offset_2 <= dictAndPrefixLength);

    /* Outer search loop */
    assert(stepSize >= 1);
    while (ip1 <= ilimit) {   /* repcode check at (ip0 + 1) is safe because ip0 < ip1 */
        size_t mLength;
        size_t hash0 = ZSTD_hashPtr(ip0, hlog, mls);

        size_t const dictHashAndTag0 = ZSTD_hashPtr(ip0, dictHBits, mls);
        U32 dictMatchIndexAndTag = dictHashTable[dictHashAndTag0 >> ZSTD_SHORT_CACHE_TAG_BITS];
        int dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag0);

        U32 matchIndex = hashTable[hash0];
        U32 curr = (U32)(ip0 - base);
        size_t step = stepSize;
        const size_t kStepIncr = 1 << kSearchStrength;
        const BYTE* nextStep = ip0 + kStepIncr;

        /* Inner search loop */
        while (1) {
            const BYTE* match = base + matchIndex;
            const U32 repIndex = curr + 1 - offset_1;
            const BYTE* repMatch = (repIndex < prefixStartIndex) ?
                                   dictBase + (repIndex - dictIndexDelta) :
                                   base + repIndex;
            const size_t hash1 = ZSTD_hashPtr(ip1, hlog, mls);
            size_t const dictHashAndTag1 = ZSTD_hashPtr(ip1, dictHBits, mls);
            hashTable[hash0] = curr;   /* update hash table */

            if ((ZSTD_index_overlap_check(prefixStartIndex, repIndex))
                && (MEM_read32(repMatch) == MEM_read32(ip0 + 1))) {
                const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
                mLength = ZSTD_count_2segments(ip0 + 1 + 4, repMatch + 4, iend, repMatchEnd, prefixStart) + 4;
                ip0++;
                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, REPCODE1_TO_OFFBASE, mLength);
                break;
            }

            if (dictTagsMatch) {
                /* Found a possible dict match */
                const U32 dictMatchIndex = dictMatchIndexAndTag >> ZSTD_SHORT_CACHE_TAG_BITS;
                const BYTE* dictMatch = dictBase + dictMatchIndex;
                if (dictMatchIndex > dictStartIndex &&
                    MEM_read32(dictMatch) == MEM_read32(ip0)) {
                    /* To replicate extDict parse behavior, we only use dict matches when the normal matchIndex is invalid */
                    if (matchIndex <= prefixStartIndex) {
                        U32 const offset = (U32) (curr - dictMatchIndex - dictIndexDelta);
                        mLength = ZSTD_count_2segments(ip0 + 4, dictMatch + 4, iend, dictEnd, prefixStart) + 4;
                        while (((ip0 > anchor) & (dictMatch > dictStart))
                            && (ip0[-1] == dictMatch[-1])) {
                            ip0--;
                            dictMatch--;
                            mLength++;
                        } /* catch up */
                        offset_2 = offset_1;
                        offset_1 = offset;
                        ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
                        break;
                    }
                }
            }

            if (ZSTD_match4Found_cmov(ip0, match, matchIndex, prefixStartIndex)) {
                /* found a regular match of size >= 4 */
                U32 const offset = (U32) (ip0 - match);
                mLength = ZSTD_count(ip0 + 4, match + 4, iend) + 4;
                while (((ip0 > anchor) & (match > prefixStart))
                       && (ip0[-1] == match[-1])) {
                    ip0--;
                    match--;
                    mLength++;
                } /* catch up */
                offset_2 = offset_1;
                offset_1 = offset;
                ZSTD_storeSeq(seqStore, (size_t) (ip0 - anchor), anchor, iend, OFFSET_TO_OFFBASE(offset), mLength);
                break;
            }

            /* Prepare for next iteration */
            dictMatchIndexAndTag = dictHashTable[dictHashAndTag1 >> ZSTD_SHORT_CACHE_TAG_BITS];
            dictTagsMatch = ZSTD_comparePackedTags(dictMatchIndexAndTag, dictHashAndTag1);
            matchIndex = hashTable[hash1];

            if (ip1 >= nextStep) {
                step++;
                nextStep += kStepIncr;
            }
            ip0 = ip1;
            ip1 = ip1 + step;
            if (ip1 > ilimit) goto _cleanup;

            curr = (U32)(ip0 - base);
            hash0 = hash1;
        }   /* end inner search loop */

        /* match found */
        assert(mLength);
        ip0 += mLength;
        anchor = ip0;

        if (ip0 <= ilimit) {
            /* Fill Table */
            assert(base+curr+2 > istart);  /* check base overflow */
            hashTable[ZSTD_hashPtr(base+curr+2, hlog, mls)] = curr+2;  /* here because curr+2 could be > iend-8 */
            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);

            /* check immediate repcode */
            while (ip0 <= ilimit) {
                U32 const current2 = (U32)(ip0-base);
                U32 const repIndex2 = current2 - offset_2;
                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
                        dictBase - dictIndexDelta + repIndex2 :
                        base + repIndex2;
                if ( (ZSTD_index_overlap_check(prefixStartIndex, repIndex2))
                   && (MEM_read32(repMatch2) == MEM_read32(ip0))) {
                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
                    size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
                    ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = current2;
                    ip0 += repLength2;
                    anchor = ip0;
                    continue;
                }
                break;
            }
        }

        /* Prepare for next iteration */
        assert(ip0 == anchor);
        ip1 = ip0 + stepSize;
    }

_cleanup:
    /* save reps for next block */
    rep[0] = offset_1;
    rep[1] = offset_2;

    /* Return the last literals size */
    return (size_t)(iend - anchor);
}


ZSTD_GEN_FAST_FN(dictMatchState, 4, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 5, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 6, 0)
ZSTD_GEN_FAST_FN(dictMatchState, 7, 0)

size_t ZSTD_compressBlock_fast_dictMatchState(
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize)
{
    U32 const mls = ms->cParams.minMatch;
    assert(ms->dictMatchState != NULL);
    switch(mls)
    {
    default: /* includes case 3 */
    case 4 :
        return ZSTD_compressBlock_fast_dictMatchState_4_0(ms, seqStore, rep, src, srcSize);
    case 5 :
        return ZSTD_compressBlock_fast_dictMatchState_5_0(ms, seqStore, rep, src, srcSize);
    case 6 :
        return ZSTD_compressBlock_fast_dictMatchState_6_0(ms, seqStore, rep, src, srcSize);
    case 7 :
        return ZSTD_compressBlock_fast_dictMatchState_7_0(ms, seqStore, rep, src, srcSize);
    }
}


static
ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
size_t ZSTD_compressBlock_fast_extDict_generic(
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize, U32 const mls, U32 const hasStep)
{
    const ZSTD_compressionParameters* const cParams = &ms->cParams;
    U32* const hashTable = ms->hashTable;
    U32 const hlog = cParams->hashLog;
    /* support stepSize of 0 */
    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
    const BYTE* const base = ms->window.base;
    const BYTE* const dictBase = ms->window.dictBase;
    const BYTE* const istart = (const BYTE*)src;
    const BYTE* anchor = istart;
    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
    const U32   dictStartIndex = lowLimit;
    const BYTE* const dictStart = dictBase + dictStartIndex;
    const U32   dictLimit = ms->window.dictLimit;
    const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
    const BYTE* const prefixStart = base + prefixStartIndex;
    const BYTE* const dictEnd = dictBase + prefixStartIndex;
    const BYTE* const iend = istart + srcSize;
    const BYTE* const ilimit = iend - 8;
    U32 offset_1=rep[0], offset_2=rep[1];
    U32 offsetSaved1 = 0, offsetSaved2 = 0;

    const BYTE* ip0 = istart;
    const BYTE* ip1;
    const BYTE* ip2;
    const BYTE* ip3;
    U32 current0;


    size_t hash0; /* hash for ip0 */
    size_t hash1; /* hash for ip1 */
    U32 idx; /* match idx for ip0 */
    const BYTE* idxBase; /* base pointer for idx */

    U32 offcode;
    const BYTE* match0;
    size_t mLength;
    const BYTE* matchEnd = 0; /* initialize to avoid warning, assert != 0 later */

    size_t step;
    const BYTE* nextStep;
    const size_t kStepIncr = (1 << (kSearchStrength - 1));

    (void)hasStep; /* not currently specialized on whether it's accelerated */

    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);

    /* switch to "regular" variant if extDict is invalidated due to maxDistance */
    if (prefixStartIndex == dictStartIndex)
        return ZSTD_compressBlock_fast(ms, seqStore, rep, src, srcSize);

    {   U32 const curr = (U32)(ip0 - base);
        U32 const maxRep = curr - dictStartIndex;
        if (offset_2 >= maxRep) offsetSaved2 = offset_2, offset_2 = 0;
        if (offset_1 >= maxRep) offsetSaved1 = offset_1, offset_1 = 0;
    }

    /* start each op */
_start: /* Requires: ip0 */

    step = stepSize;
    nextStep = ip0 + kStepIncr;

    /* calculate positions, ip0 - anchor == 0, so we skip step calc */
    ip1 = ip0 + 1;
    ip2 = ip0 + step;
    ip3 = ip2 + 1;

    if (ip3 >= ilimit) {
        goto _cleanup;
    }

    hash0 = ZSTD_hashPtr(ip0, hlog, mls);
    hash1 = ZSTD_hashPtr(ip1, hlog, mls);

    idx = hashTable[hash0];
    idxBase = idx < prefixStartIndex ? dictBase : base;

    do {
        {   /* load repcode match for ip[2] */
            U32 const current2 = (U32)(ip2 - base);
            U32 const repIndex = current2 - offset_1;
            const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
            U32 rval;
            if ( ((U32)(prefixStartIndex - repIndex) >= 4) /* intentional underflow */
                 & (offset_1 > 0) ) {
                rval = MEM_read32(repBase + repIndex);
            } else {
                rval = MEM_read32(ip2) ^ 1; /* guaranteed to not match. */
            }

            /* write back hash table entry */
            current0 = (U32)(ip0 - base);
            hashTable[hash0] = current0;

            /* check repcode at ip[2] */
            if (MEM_read32(ip2) == rval) {
                ip0 = ip2;
                match0 = repBase + repIndex;
                matchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
                assert((match0 != prefixStart) & (match0 != dictStart));
                mLength = ip0[-1] == match0[-1];
                ip0 -= mLength;
                match0 -= mLength;
                offcode = REPCODE1_TO_OFFBASE;
                mLength += 4;
                goto _match;
        }   }

        {   /* load match for ip[0] */
            U32 const mval = idx >= dictStartIndex ?
                    MEM_read32(idxBase + idx) :
                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */

            /* check match at ip[0] */
            if (MEM_read32(ip0) == mval) {
                /* found a match! */
                goto _offset;
        }   }

        /* lookup ip[1] */
        idx = hashTable[hash1];
        idxBase = idx < prefixStartIndex ? dictBase : base;

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip3;

        /* write back hash table entry */
        current0 = (U32)(ip0 - base);
        hashTable[hash0] = current0;

        {   /* load match for ip[0] */
            U32 const mval = idx >= dictStartIndex ?
                    MEM_read32(idxBase + idx) :
                    MEM_read32(ip0) ^ 1; /* guaranteed not to match */

            /* check match at ip[0] */
            if (MEM_read32(ip0) == mval) {
                /* found a match! */
                goto _offset;
        }   }

        /* lookup ip[1] */
        idx = hashTable[hash1];
        idxBase = idx < prefixStartIndex ? dictBase : base;

        /* hash ip[2] */
        hash0 = hash1;
        hash1 = ZSTD_hashPtr(ip2, hlog, mls);

        /* advance to next positions */
        ip0 = ip1;
        ip1 = ip2;
        ip2 = ip0 + step;
        ip3 = ip1 + step;

        /* calculate step */
        if (ip2 >= nextStep) {
            step++;
            PREFETCH_L1(ip1 + 64);
            PREFETCH_L1(ip1 + 128);
            nextStep += kStepIncr;
        }
    } while (ip3 < ilimit);

_cleanup:
    /* Note that there are probably still a couple positions we could search.
     * However, it seems to be a meaningful performance hit to try to search
     * them. So let's not. */

    /* If offset_1 started invalid (offsetSaved1 != 0) and became valid (offset_1 != 0),
     * rotate saved offsets. See comment in ZSTD_compressBlock_fast_noDict for more context. */
    offsetSaved2 = ((offsetSaved1 != 0) && (offset_1 != 0)) ? offsetSaved1 : offsetSaved2;

    /* save reps for next block */
    rep[0] = offset_1 ? offset_1 : offsetSaved1;
    rep[1] = offset_2 ? offset_2 : offsetSaved2;

    /* Return the last literals size */
    return (size_t)(iend - anchor);

_offset: /* Requires: ip0, idx, idxBase */

    /* Compute the offset code. */
    {   U32 const offset = current0 - idx;
        const BYTE* const lowMatchPtr = idx < prefixStartIndex ? dictStart : prefixStart;
        matchEnd = idx < prefixStartIndex ? dictEnd : iend;
        match0 = idxBase + idx;
        offset_2 = offset_1;
        offset_1 = offset;
        offcode = OFFSET_TO_OFFBASE(offset);
        mLength = 4;

        /* Count the backwards match length. */
        while (((ip0>anchor) & (match0>lowMatchPtr)) && (ip0[-1] == match0[-1])) {
            ip0--;
            match0--;
            mLength++;
    }   }

_match: /* Requires: ip0, match0, offcode, matchEnd */

    /* Count the forward length. */
    assert(matchEnd != 0);
    mLength += ZSTD_count_2segments(ip0 + mLength, match0 + mLength, iend, matchEnd, prefixStart);

    ZSTD_storeSeq(seqStore, (size_t)(ip0 - anchor), anchor, iend, offcode, mLength);

    ip0 += mLength;
    anchor = ip0;

    /* write next hash table entry */
    if (ip1 < ip0) {
        hashTable[hash1] = (U32)(ip1 - base);
    }

    /* Fill table and check for immediate repcode. */
    if (ip0 <= ilimit) {
        /* Fill Table */
        assert(base+current0+2 > istart);  /* check base overflow */
        hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
        hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);

        while (ip0 <= ilimit) {
            U32 const repIndex2 = (U32)(ip0-base) - offset_2;
            const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
            if ( ((ZSTD_index_overlap_check(prefixStartIndex, repIndex2)) & (offset_2 > 0))
                 && (MEM_read32(repMatch2) == MEM_read32(ip0)) ) {
                const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
                size_t const repLength2 = ZSTD_count_2segments(ip0+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
                { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
                ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, REPCODE1_TO_OFFBASE, repLength2);
                hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
                ip0 += repLength2;
                anchor = ip0;
                continue;
            }
            break;
    }   }

    goto _start;
}

ZSTD_GEN_FAST_FN(extDict, 4, 0)
ZSTD_GEN_FAST_FN(extDict, 5, 0)
ZSTD_GEN_FAST_FN(extDict, 6, 0)
ZSTD_GEN_FAST_FN(extDict, 7, 0)

size_t ZSTD_compressBlock_fast_extDict(
        ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
        void const* src, size_t srcSize)
{
    U32 const mls = ms->cParams.minMatch;
    assert(ms->dictMatchState == NULL);
    switch(mls)
    {
    default: /* includes case 3 */
    case 4 :
        return ZSTD_compressBlock_fast_extDict_4_0(ms, seqStore, rep, src, srcSize);
    case 5 :
        return ZSTD_compressBlock_fast_extDict_5_0(ms, seqStore, rep, src, srcSize);
    case 6 :
        return ZSTD_compressBlock_fast_extDict_6_0(ms, seqStore, rep, src, srcSize);
    case 7 :
        return ZSTD_compressBlock_fast_extDict_7_0(ms, seqStore, rep, src, srcSize);
    }
}