diff --git a/lib/compress/zstd_lazy.c b/lib/compress/zstd_lazy.c index 272ebe0ec..63fa05217 100644 --- a/lib/compress/zstd_lazy.c +++ b/lib/compress/zstd_lazy.c @@ -8,2192 +8,246 @@ * You may select, at your option, one of the above-listed licenses. */ +/* The lazy matchers are implemented in rust/src/zstd_lazy.rs. Keep only the + * private match-state projection and the established C entry points here. */ #include "zstd_compress_internal.h" #include "zstd_lazy.h" -#include "../common/bits.h" /* ZSTD_countTrailingZeros64 */ #if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \ || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \ || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \ || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) -#define kLazySkippingStep 8 +typedef char ZSTD_rust_lazy_seqdef_layout[(sizeof(SeqDef) == 8) ? 1 : -1]; +typedef char ZSTD_rust_lazy_seqstore_long_length_pos[ + (offsetof(SeqStore_t, longLengthPos) == 9 * sizeof(size_t) + 4) ? 1 : -1]; +typedef char ZSTD_rust_lazy_seqstore_layout[ + (sizeof(SeqStore_t) == 9 * sizeof(size_t) + 8) ? 1 : -1]; +typedef char ZSTD_rust_lazy_rep_count[(ZSTD_REP_NUM == 3) ? 1 : -1]; +/* This must remain in lock-step with `ZSTD_RustLazyState` in Rust. The large + * and private `ZSTD_MatchState_t` itself deliberately never crosses FFI. */ +typedef struct ZSTD_RustLazyState_s { + U32* hashTable; + U32* chainTable; + BYTE* tagTable; + U32* hashCache; + const BYTE* base; + const BYTE* dictBase; + const BYTE* nextSrc; + U32 dictLimit; + U32 lowLimit; + U32 loadedDictEnd; + U32* nextToUpdate; + int* lazySkipping; + U32 hashLog; + U32 chainLog; + U32 minMatch; + U32 searchLog; + U32 windowLog; + U32 rowHashLog; + U64 hashSalt; + U32* hashSaltEntropy; + const struct ZSTD_RustLazyState_s* dictMatchState; +} ZSTD_RustLazyState; -/*-************************************* -* Binary Tree search -***************************************/ +U32 ZSTD_rust_lazy_insertAndFindFirstIndex(ZSTD_RustLazyState* state, + const void* ip); +void ZSTD_rust_lazy_row_update(ZSTD_RustLazyState* state, const void* ip); +void ZSTD_rust_lazy_loadDedicatedDict(ZSTD_RustLazyState* state, + const void* ip); +size_t ZSTD_rust_compressBlock_lazy( + ZSTD_RustLazyState* state, void* seqStore, U32 rep[ZSTD_REP_NUM], + const void* src, size_t srcSize, int searchMethod, U32 depth, + int dictMode); -static -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -void ZSTD_updateDUBT(ZSTD_MatchState_t* ms, - const BYTE* ip, const BYTE* iend, - U32 mls) +enum { + ZSTD_rust_search_hashChain = 0, + ZSTD_rust_search_binaryTree = 1, + ZSTD_rust_search_rowHash = 2, +}; + +enum { + ZSTD_rust_dict_noDict = 0, + ZSTD_rust_dict_extDict = 1, + ZSTD_rust_dict_dictMatchState = 2, + ZSTD_rust_dict_dedicatedDictSearch = 3, +}; + +static void ZSTD_rustLazyState_init( + ZSTD_RustLazyState* const out, + ZSTD_MatchState_t* const ms, + const ZSTD_RustLazyState* const dms) { - const ZSTD_compressionParameters* const cParams = &ms->cParams; - U32* const hashTable = ms->hashTable; - U32 const hashLog = cParams->hashLog; - - U32* const bt = ms->chainTable; - U32 const btLog = cParams->chainLog - 1; - U32 const btMask = (1 << btLog) - 1; - - const BYTE* const base = ms->window.base; - U32 const target = (U32)(ip - base); - U32 idx = ms->nextToUpdate; - - if (idx != target) - DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)", - idx, target, ms->window.dictLimit); - assert(ip + 8 <= iend); /* condition for ZSTD_hashPtr */ - (void)iend; - - assert(idx >= ms->window.dictLimit); /* condition for valid base+idx */ - for ( ; idx < target ; idx++) { - size_t const h = ZSTD_hashPtr(base + idx, hashLog, mls); /* assumption : ip + 8 <= iend */ - U32 const matchIndex = hashTable[h]; - - U32* const nextCandidatePtr = bt + 2*(idx&btMask); - U32* const sortMarkPtr = nextCandidatePtr + 1; - - DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx); - hashTable[h] = idx; /* Update Hash Table */ - *nextCandidatePtr = matchIndex; /* update BT like a chain */ - *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK; - } - ms->nextToUpdate = target; + out->hashTable = ms->hashTable; + out->chainTable = ms->chainTable; + out->tagTable = ms->tagTable; + out->hashCache = ms->hashCache; + out->base = ms->window.base; + out->dictBase = ms->window.dictBase; + out->nextSrc = ms->window.nextSrc; + out->dictLimit = ms->window.dictLimit; + out->lowLimit = ms->window.lowLimit; + out->loadedDictEnd = ms->loadedDictEnd; + out->nextToUpdate = &ms->nextToUpdate; + out->lazySkipping = &ms->lazySkipping; + out->hashLog = ms->cParams.hashLog; + out->chainLog = ms->cParams.chainLog; + out->minMatch = ms->cParams.minMatch; + out->searchLog = ms->cParams.searchLog; + out->windowLog = ms->cParams.windowLog; + out->rowHashLog = ms->rowHashLog; + out->hashSalt = ms->hashSalt; + out->hashSaltEntropy = &ms->hashSaltEntropy; + out->dictMatchState = dms; } - -/** ZSTD_insertDUBT1() : - * sort one already inserted but unsorted position - * assumption : curr >= btlow == (curr - btmask) - * doesn't fail */ -static -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -void ZSTD_insertDUBT1(const ZSTD_MatchState_t* ms, - U32 curr, const BYTE* inputEnd, - U32 nbCompares, U32 btLow, - const ZSTD_dictMode_e dictMode) +/* Dictionary match states are searched but never updated by this module. */ +static void ZSTD_rustLazyState_initDict( + ZSTD_RustLazyState* const out, + const ZSTD_MatchState_t* const ms) { - const ZSTD_compressionParameters* const cParams = &ms->cParams; - U32* const bt = ms->chainTable; - U32 const btLog = cParams->chainLog - 1; - U32 const btMask = (1 << btLog) - 1; - size_t commonLengthSmaller=0, commonLengthLarger=0; - const BYTE* const base = ms->window.base; - const BYTE* const dictBase = ms->window.dictBase; - const U32 dictLimit = ms->window.dictLimit; - const BYTE* const ip = (curr>=dictLimit) ? base + curr : dictBase + curr; - const BYTE* const iend = (curr>=dictLimit) ? inputEnd : dictBase + dictLimit; - const BYTE* const dictEnd = dictBase + dictLimit; - const BYTE* const prefixStart = base + dictLimit; - const BYTE* match; - U32* smallerPtr = bt + 2*(curr&btMask); - U32* largerPtr = smallerPtr + 1; - U32 matchIndex = *smallerPtr; /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */ - U32 dummy32; /* to be nullified at the end */ - U32 const windowValid = ms->window.lowLimit; - U32 const maxDistance = 1U << cParams->windowLog; - U32 const windowLow = (curr - windowValid > maxDistance) ? curr - maxDistance : windowValid; - - - DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)", - curr, dictLimit, windowLow); - assert(curr >= btLow); - assert(ip < iend); /* condition for ZSTD_count */ - - for (; nbCompares && (matchIndex > windowLow); --nbCompares) { - U32* const nextPtr = bt + 2*(matchIndex & btMask); - size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */ - assert(matchIndex < curr); - /* note : all candidates are now supposed sorted, - * but it's still possible to have nextPtr[1] == ZSTD_DUBT_UNSORTED_MARK - * when a real index has the same value as ZSTD_DUBT_UNSORTED_MARK */ - - if ( (dictMode != ZSTD_extDict) - || (matchIndex+matchLength >= dictLimit) /* both in current segment*/ - || (curr < dictLimit) /* both in extDict */) { - const BYTE* const mBase = ( (dictMode != ZSTD_extDict) - || (matchIndex+matchLength >= dictLimit)) ? - base : dictBase; - assert( (matchIndex+matchLength >= dictLimit) /* might be wrong if extDict is incorrectly set to 0 */ - || (curr < dictLimit) ); - match = mBase + matchIndex; - matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend); - } else { - match = dictBase + matchIndex; - matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart); - if (matchIndex+matchLength >= dictLimit) - match = base + matchIndex; /* preparation for next read of match[matchLength] */ - } - - DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ", - curr, matchIndex, (U32)matchLength); - - if (ip+matchLength == iend) { /* equal : no way to know if inf or sup */ - break; /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */ - } - - if (match[matchLength] < ip[matchLength]) { /* necessarily within buffer */ - /* match is smaller than current */ - *smallerPtr = matchIndex; /* update smaller idx */ - commonLengthSmaller = matchLength; /* all smaller will now have at least this guaranteed common length */ - if (matchIndex <= btLow) { smallerPtr=&dummy32; break; } /* beyond tree size, stop searching */ - DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u", - matchIndex, btLow, nextPtr[1]); - smallerPtr = nextPtr+1; /* new "candidate" => larger than match, which was smaller than target */ - matchIndex = nextPtr[1]; /* new matchIndex, larger than previous and closer to current */ - } else { - /* match is larger than current */ - *largerPtr = matchIndex; - commonLengthLarger = matchLength; - if (matchIndex <= btLow) { largerPtr=&dummy32; break; } /* beyond tree size, stop searching */ - DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u", - matchIndex, btLow, nextPtr[0]); - largerPtr = nextPtr; - matchIndex = nextPtr[0]; - } } - - *smallerPtr = *largerPtr = 0; + out->hashTable = ms->hashTable; + out->chainTable = ms->chainTable; + out->tagTable = ms->tagTable; + out->hashCache = NULL; + out->base = ms->window.base; + out->dictBase = ms->window.dictBase; + out->nextSrc = ms->window.nextSrc; + out->dictLimit = ms->window.dictLimit; + out->lowLimit = ms->window.lowLimit; + out->loadedDictEnd = ms->loadedDictEnd; + out->nextToUpdate = NULL; + out->lazySkipping = NULL; + out->hashLog = ms->cParams.hashLog; + out->chainLog = ms->cParams.chainLog; + out->minMatch = ms->cParams.minMatch; + out->searchLog = ms->cParams.searchLog; + out->windowLog = ms->cParams.windowLog; + out->rowHashLog = ms->rowHashLog; + out->hashSalt = ms->hashSalt; + out->hashSaltEntropy = NULL; + out->dictMatchState = NULL; } - -static -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_DUBT_findBetterDictMatch ( - const ZSTD_MatchState_t* ms, - const BYTE* const ip, const BYTE* const iend, - size_t* offsetPtr, - size_t bestLength, - U32 nbCompares, - U32 const mls, - const ZSTD_dictMode_e dictMode) +static size_t ZSTD_rust_compressBlock_lazy_call( + ZSTD_MatchState_t* const ms, SeqStore_t* const seqStore, + U32 rep[ZSTD_REP_NUM], const void* const src, size_t const srcSize, + int const searchMethod, U32 const depth, int const dictMode) { - const ZSTD_MatchState_t * const dms = ms->dictMatchState; - const ZSTD_compressionParameters* const dmsCParams = &dms->cParams; - const U32 * const dictHashTable = dms->hashTable; - U32 const hashLog = dmsCParams->hashLog; - size_t const h = ZSTD_hashPtr(ip, hashLog, mls); - U32 dictMatchIndex = dictHashTable[h]; + ZSTD_RustLazyState state; + ZSTD_RustLazyState dictState; + const ZSTD_RustLazyState* dms = NULL; - const BYTE* const base = ms->window.base; - const BYTE* const prefixStart = base + ms->window.dictLimit; - U32 const curr = (U32)(ip-base); - const BYTE* const dictBase = dms->window.base; - const BYTE* const dictEnd = dms->window.nextSrc; - U32 const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base); - U32 const dictLowLimit = dms->window.lowLimit; - U32 const dictIndexDelta = ms->window.lowLimit - dictHighLimit; - - U32* const dictBt = dms->chainTable; - U32 const btLog = dmsCParams->chainLog - 1; - U32 const btMask = (1 << btLog) - 1; - U32 const btLow = (btMask >= dictHighLimit - dictLowLimit) ? dictLowLimit : dictHighLimit - btMask; - - size_t commonLengthSmaller=0, commonLengthLarger=0; - - (void)dictMode; - assert(dictMode == ZSTD_dictMatchState); - - for (; nbCompares && (dictMatchIndex > dictLowLimit); --nbCompares) { - U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask); - size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */ - const BYTE* match = dictBase + dictMatchIndex; - matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart); - if (dictMatchIndex+matchLength >= dictHighLimit) - match = base + dictMatchIndex + dictIndexDelta; /* to prepare for next usage of match[matchLength] */ - - if (matchLength > bestLength) { - U32 matchIndex = dictMatchIndex + dictIndexDelta; - if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) { - DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)", - curr, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, OFFSET_TO_OFFBASE(curr - matchIndex), dictMatchIndex, matchIndex); - bestLength = matchLength, *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex); - } - if (ip+matchLength == iend) { /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */ - break; /* drop, to guarantee consistency (miss a little bit of compression) */ - } - } - - if (match[matchLength] < ip[matchLength]) { - if (dictMatchIndex <= btLow) { break; } /* beyond tree size, stop the search */ - commonLengthSmaller = matchLength; /* all smaller will now have at least this guaranteed common length */ - dictMatchIndex = nextPtr[1]; /* new matchIndex larger than previous (closer to current) */ - } else { - /* match is larger than current */ - if (dictMatchIndex <= btLow) { break; } /* beyond tree size, stop the search */ - commonLengthLarger = matchLength; - dictMatchIndex = nextPtr[0]; - } + if (dictMode == ZSTD_rust_dict_dictMatchState + || dictMode == ZSTD_rust_dict_dedicatedDictSearch) { + const ZSTD_MatchState_t* const rawDms = ms->dictMatchState; + assert(rawDms != NULL); + ZSTD_rustLazyState_initDict(&dictState, rawDms); + dms = &dictState; } - - if (bestLength >= MINMATCH) { - U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offsetPtr); (void)mIndex; - DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)", - curr, (U32)bestLength, (U32)*offsetPtr, mIndex); - } - return bestLength; - + ZSTD_rustLazyState_init(&state, ms, dms); + return ZSTD_rust_compressBlock_lazy( + &state, seqStore, rep, src, srcSize, searchMethod, depth, dictMode); } - -static -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_DUBT_findBestMatch(ZSTD_MatchState_t* ms, - const BYTE* const ip, const BYTE* const iend, - size_t* offBasePtr, - U32 const mls, - const ZSTD_dictMode_e dictMode) +U32 ZSTD_insertAndFindFirstIndex(ZSTD_MatchState_t* ms, const BYTE* ip) { - const ZSTD_compressionParameters* const cParams = &ms->cParams; - U32* const hashTable = ms->hashTable; - U32 const hashLog = cParams->hashLog; - size_t const h = ZSTD_hashPtr(ip, hashLog, mls); - U32 matchIndex = hashTable[h]; - - const BYTE* const base = ms->window.base; - U32 const curr = (U32)(ip-base); - U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog); - - U32* const bt = ms->chainTable; - U32 const btLog = cParams->chainLog - 1; - U32 const btMask = (1 << btLog) - 1; - U32 const btLow = (btMask >= curr) ? 0 : curr - btMask; - U32 const unsortLimit = MAX(btLow, windowLow); - - U32* nextCandidate = bt + 2*(matchIndex&btMask); - U32* unsortedMark = bt + 2*(matchIndex&btMask) + 1; - U32 nbCompares = 1U << cParams->searchLog; - U32 nbCandidates = nbCompares; - U32 previousCandidate = 0; - - DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", curr); - assert(ip <= iend-8); /* required for h calculation */ - assert(dictMode != ZSTD_dedicatedDictSearch); - - /* reach end of unsorted candidates list */ - while ( (matchIndex > unsortLimit) - && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK) - && (nbCandidates > 1) ) { - DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted", - matchIndex); - *unsortedMark = previousCandidate; /* the unsortedMark becomes a reversed chain, to move up back to original position */ - previousCandidate = matchIndex; - matchIndex = *nextCandidate; - nextCandidate = bt + 2*(matchIndex&btMask); - unsortedMark = bt + 2*(matchIndex&btMask) + 1; - nbCandidates --; - } - - /* nullify last candidate if it's still unsorted - * simplification, detrimental to compression ratio, beneficial for speed */ - if ( (matchIndex > unsortLimit) - && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) { - DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u", - matchIndex); - *nextCandidate = *unsortedMark = 0; - } - - /* batch sort stacked candidates */ - matchIndex = previousCandidate; - while (matchIndex) { /* will end on matchIndex == 0 */ - U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1; - U32 const nextCandidateIdx = *nextCandidateIdxPtr; - ZSTD_insertDUBT1(ms, matchIndex, iend, - nbCandidates, unsortLimit, dictMode); - matchIndex = nextCandidateIdx; - nbCandidates++; - } - - /* find longest match */ - { size_t commonLengthSmaller = 0, commonLengthLarger = 0; - const BYTE* const dictBase = ms->window.dictBase; - const U32 dictLimit = ms->window.dictLimit; - const BYTE* const dictEnd = dictBase + dictLimit; - const BYTE* const prefixStart = base + dictLimit; - U32* smallerPtr = bt + 2*(curr&btMask); - U32* largerPtr = bt + 2*(curr&btMask) + 1; - U32 matchEndIdx = curr + 8 + 1; - U32 dummy32; /* to be nullified at the end */ - size_t bestLength = 0; - - matchIndex = hashTable[h]; - hashTable[h] = curr; /* Update Hash Table */ - - for (; nbCompares && (matchIndex > windowLow); --nbCompares) { - U32* const nextPtr = bt + 2*(matchIndex & btMask); - size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger); /* guaranteed minimum nb of common bytes */ - const BYTE* match; - - if ((dictMode != ZSTD_extDict) || (matchIndex+matchLength >= dictLimit)) { - match = base + matchIndex; - matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend); - } else { - match = dictBase + matchIndex; - matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart); - if (matchIndex+matchLength >= dictLimit) - match = base + matchIndex; /* to prepare for next usage of match[matchLength] */ - } - - if (matchLength > bestLength) { - if (matchLength > matchEndIdx - matchIndex) - matchEndIdx = matchIndex + (U32)matchLength; - if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(curr - matchIndex + 1) - ZSTD_highbit32((U32)*offBasePtr)) ) - bestLength = matchLength, *offBasePtr = OFFSET_TO_OFFBASE(curr - matchIndex); - if (ip+matchLength == iend) { /* equal : no way to know if inf or sup */ - if (dictMode == ZSTD_dictMatchState) { - nbCompares = 0; /* in addition to avoiding checking any - * further in this loop, make sure we - * skip checking in the dictionary. */ - } - break; /* drop, to guarantee consistency (miss a little bit of compression) */ - } - } - - if (match[matchLength] < ip[matchLength]) { - /* match is smaller than current */ - *smallerPtr = matchIndex; /* update smaller idx */ - commonLengthSmaller = matchLength; /* all smaller will now have at least this guaranteed common length */ - if (matchIndex <= btLow) { smallerPtr=&dummy32; break; } /* beyond tree size, stop the search */ - smallerPtr = nextPtr+1; /* new "smaller" => larger of match */ - matchIndex = nextPtr[1]; /* new matchIndex larger than previous (closer to current) */ - } else { - /* match is larger than current */ - *largerPtr = matchIndex; - commonLengthLarger = matchLength; - if (matchIndex <= btLow) { largerPtr=&dummy32; break; } /* beyond tree size, stop the search */ - largerPtr = nextPtr; - matchIndex = nextPtr[0]; - } } - - *smallerPtr = *largerPtr = 0; - - assert(nbCompares <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */ - if (dictMode == ZSTD_dictMatchState && nbCompares) { - bestLength = ZSTD_DUBT_findBetterDictMatch( - ms, ip, iend, - offBasePtr, bestLength, nbCompares, - mls, dictMode); - } - - assert(matchEndIdx > curr+8); /* ensure nextToUpdate is increased */ - ms->nextToUpdate = matchEndIdx - 8; /* skip repetitive patterns */ - if (bestLength >= MINMATCH) { - U32 const mIndex = curr - (U32)OFFBASE_TO_OFFSET(*offBasePtr); (void)mIndex; - DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)", - curr, (U32)bestLength, (U32)*offBasePtr, mIndex); - } - return bestLength; - } + ZSTD_RustLazyState state; + ZSTD_rustLazyState_init(&state, ms, NULL); + return ZSTD_rust_lazy_insertAndFindFirstIndex(&state, ip); } - -/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_BtFindBestMatch( ZSTD_MatchState_t* ms, - const BYTE* const ip, const BYTE* const iLimit, - size_t* offBasePtr, - const U32 mls /* template */, - const ZSTD_dictMode_e dictMode) +void ZSTD_row_update(ZSTD_MatchState_t* const ms, const BYTE* const ip) { - DEBUGLOG(7, "ZSTD_BtFindBestMatch"); - if (ip < ms->window.base + ms->nextToUpdate) return 0; /* skipped area */ - ZSTD_updateDUBT(ms, ip, iLimit, mls); - return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offBasePtr, mls, dictMode); + ZSTD_RustLazyState state; + ZSTD_rustLazyState_init(&state, ms, NULL); + ZSTD_rust_lazy_row_update(&state, ip); } -/*********************************** -* Dedicated dict search -***********************************/ - -void ZSTD_dedicatedDictSearch_lazy_loadDictionary(ZSTD_MatchState_t* ms, const BYTE* const ip) +void ZSTD_dedicatedDictSearch_lazy_loadDictionary( + ZSTD_MatchState_t* ms, const BYTE* const ip) { - const BYTE* const base = ms->window.base; - U32 const target = (U32)(ip - base); - U32* const hashTable = ms->hashTable; - U32* const chainTable = ms->chainTable; - U32 const chainSize = 1 << ms->cParams.chainLog; - U32 idx = ms->nextToUpdate; - U32 const minChain = chainSize < target - idx ? target - chainSize : idx; - U32 const bucketSize = 1 << ZSTD_LAZY_DDSS_BUCKET_LOG; - U32 const cacheSize = bucketSize - 1; - U32 const chainAttempts = (1 << ms->cParams.searchLog) - cacheSize; - U32 const chainLimit = chainAttempts > 255 ? 255 : chainAttempts; - - /* We know the hashtable is oversized by a factor of `bucketSize`. - * We are going to temporarily pretend `bucketSize == 1`, keeping only a - * single entry. We will use the rest of the space to construct a temporary - * chaintable. - */ - U32 const hashLog = ms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG; - U32* const tmpHashTable = hashTable; - U32* const tmpChainTable = hashTable + ((size_t)1 << hashLog); - U32 const tmpChainSize = (U32)((1 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1) << hashLog; - U32 const tmpMinChain = tmpChainSize < target ? target - tmpChainSize : idx; - U32 hashIdx; - - assert(ms->cParams.chainLog <= 24); - assert(ms->cParams.hashLog > ms->cParams.chainLog); - assert(idx != 0); - assert(tmpMinChain <= minChain); - - /* fill conventional hash table and conventional chain table */ - for ( ; idx < target; idx++) { - U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch); - if (idx >= tmpMinChain) { - tmpChainTable[idx - tmpMinChain] = hashTable[h]; - } - tmpHashTable[h] = idx; - } - - /* sort chains into ddss chain table */ - { - U32 chainPos = 0; - for (hashIdx = 0; hashIdx < (1U << hashLog); hashIdx++) { - U32 count; - U32 countBeyondMinChain = 0; - U32 i = tmpHashTable[hashIdx]; - for (count = 0; i >= tmpMinChain && count < cacheSize; count++) { - /* skip through the chain to the first position that won't be - * in the hash cache bucket */ - if (i < minChain) { - countBeyondMinChain++; - } - i = tmpChainTable[i - tmpMinChain]; - } - if (count == cacheSize) { - for (count = 0; count < chainLimit;) { - if (i < minChain) { - if (!i || ++countBeyondMinChain > cacheSize) { - /* only allow pulling `cacheSize` number of entries - * into the cache or chainTable beyond `minChain`, - * to replace the entries pulled out of the - * chainTable into the cache. This lets us reach - * back further without increasing the total number - * of entries in the chainTable, guaranteeing the - * DDSS chain table will fit into the space - * allocated for the regular one. */ - break; - } - } - chainTable[chainPos++] = i; - count++; - if (i < tmpMinChain) { - break; - } - i = tmpChainTable[i - tmpMinChain]; - } - } else { - count = 0; - } - if (count) { - tmpHashTable[hashIdx] = ((chainPos - count) << 8) + count; - } else { - tmpHashTable[hashIdx] = 0; - } - } - assert(chainPos <= chainSize); /* I believe this is guaranteed... */ - } - - /* move chain pointers into the last entry of each hash bucket */ - for (hashIdx = (1 << hashLog); hashIdx; ) { - U32 const bucketIdx = --hashIdx << ZSTD_LAZY_DDSS_BUCKET_LOG; - U32 const chainPackedPointer = tmpHashTable[hashIdx]; - U32 i; - for (i = 0; i < cacheSize; i++) { - hashTable[bucketIdx + i] = 0; - } - hashTable[bucketIdx + bucketSize - 1] = chainPackedPointer; - } - - /* fill the buckets of the hash table */ - for (idx = ms->nextToUpdate; idx < target; idx++) { - U32 const h = (U32)ZSTD_hashPtr(base + idx, hashLog, ms->cParams.minMatch) - << ZSTD_LAZY_DDSS_BUCKET_LOG; - U32 i; - /* Shift hash cache down 1. */ - for (i = cacheSize - 1; i; i--) - hashTable[h + i] = hashTable[h + i - 1]; - hashTable[h] = idx; - } - - ms->nextToUpdate = target; + ZSTD_RustLazyState state; + ZSTD_rustLazyState_init(&state, ms, NULL); + ZSTD_rust_lazy_loadDedicatedDict(&state, ip); } -/* Returns the longest match length found in the dedicated dict search structure. - * If none are longer than the argument ml, then ml will be returned. - */ -FORCE_INLINE_TEMPLATE -size_t ZSTD_dedicatedDictSearch_lazy_search(size_t* offsetPtr, size_t ml, U32 nbAttempts, - const ZSTD_MatchState_t* const dms, - const BYTE* const ip, const BYTE* const iLimit, - const BYTE* const prefixStart, const U32 curr, - const U32 dictLimit, const size_t ddsIdx) { - const U32 ddsLowestIndex = dms->window.dictLimit; - const BYTE* const ddsBase = dms->window.base; - const BYTE* const ddsEnd = dms->window.nextSrc; - const U32 ddsSize = (U32)(ddsEnd - ddsBase); - const U32 ddsIndexDelta = dictLimit - ddsSize; - const U32 bucketSize = (1 << ZSTD_LAZY_DDSS_BUCKET_LOG); - const U32 bucketLimit = nbAttempts < bucketSize - 1 ? nbAttempts : bucketSize - 1; - U32 ddsAttempt; - U32 matchIndex; - - for (ddsAttempt = 0; ddsAttempt < bucketSize - 1; ddsAttempt++) { - PREFETCH_L1(ddsBase + dms->hashTable[ddsIdx + ddsAttempt]); +#define ZSTD_RUST_LAZY_WRAPPER(name, method, parserDepth, mode) \ + size_t name( \ + ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], \ + const void* src, size_t srcSize) \ + { \ + return ZSTD_rust_compressBlock_lazy_call( \ + ms, seqStore, rep, src, srcSize, method, parserDepth, mode); \ } - { - U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1]; - U32 const chainIndex = chainPackedPointer >> 8; - - PREFETCH_L1(&dms->chainTable[chainIndex]); - } - - for (ddsAttempt = 0; ddsAttempt < bucketLimit; ddsAttempt++) { - size_t currentMl=0; - const BYTE* match; - matchIndex = dms->hashTable[ddsIdx + ddsAttempt]; - match = ddsBase + matchIndex; - - if (!matchIndex) { - return ml; - } - - /* guaranteed by table construction */ - (void)ddsLowestIndex; - assert(matchIndex >= ddsLowestIndex); - assert(match+4 <= ddsEnd); - if (MEM_read32(match) == MEM_read32(ip)) { - /* assumption : matchIndex <= dictLimit-4 (by table construction) */ - currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4; - } - - /* save best solution */ - if (currentMl > ml) { - ml = currentMl; - *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta)); - if (ip+currentMl == iLimit) { - /* best possible, avoids read overflow on next attempt */ - return ml; - } - } - } - - { - U32 const chainPackedPointer = dms->hashTable[ddsIdx + bucketSize - 1]; - U32 chainIndex = chainPackedPointer >> 8; - U32 const chainLength = chainPackedPointer & 0xFF; - U32 const chainAttempts = nbAttempts - ddsAttempt; - U32 const chainLimit = chainAttempts > chainLength ? chainLength : chainAttempts; - U32 chainAttempt; - - for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++) { - PREFETCH_L1(ddsBase + dms->chainTable[chainIndex + chainAttempt]); - } - - for (chainAttempt = 0 ; chainAttempt < chainLimit; chainAttempt++, chainIndex++) { - size_t currentMl=0; - const BYTE* match; - matchIndex = dms->chainTable[chainIndex]; - match = ddsBase + matchIndex; - - /* guaranteed by table construction */ - assert(matchIndex >= ddsLowestIndex); - assert(match+4 <= ddsEnd); - if (MEM_read32(match) == MEM_read32(ip)) { - /* assumption : matchIndex <= dictLimit-4 (by table construction) */ - currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, ddsEnd, prefixStart) + 4; - } - - /* save best solution */ - if (currentMl > ml) { - ml = currentMl; - *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + ddsIndexDelta)); - if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */ - } - } - } - return ml; -} - - -/* ********************************* -* Hash Chain -***********************************/ -#define NEXT_IN_CHAIN(d, mask) chainTable[(d) & (mask)] - -/* Update chains up to ip (excluded) - Assumption : always within prefix (i.e. not within extDict) */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -U32 ZSTD_insertAndFindFirstIndex_internal( - ZSTD_MatchState_t* ms, - const ZSTD_compressionParameters* const cParams, - const BYTE* ip, U32 const mls, U32 const lazySkipping) -{ - U32* const hashTable = ms->hashTable; - const U32 hashLog = cParams->hashLog; - U32* const chainTable = ms->chainTable; - const U32 chainMask = (1 << cParams->chainLog) - 1; - const BYTE* const base = ms->window.base; - const U32 target = (U32)(ip - base); - U32 idx = ms->nextToUpdate; - - while(idx < target) { /* catch up */ - size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls); - NEXT_IN_CHAIN(idx, chainMask) = hashTable[h]; - hashTable[h] = idx; - idx++; - /* Stop inserting every position when in the lazy skipping mode. */ - if (lazySkipping) - break; - } - - ms->nextToUpdate = target; - return hashTable[ZSTD_hashPtr(ip, hashLog, mls)]; -} - -U32 ZSTD_insertAndFindFirstIndex(ZSTD_MatchState_t* ms, const BYTE* ip) { - const ZSTD_compressionParameters* const cParams = &ms->cParams; - return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch, /* lazySkipping*/ 0); -} - -/* inlining is important to hardwire a hot branch (template emulation) */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_HcFindBestMatch( - ZSTD_MatchState_t* ms, - const BYTE* const ip, const BYTE* const iLimit, - size_t* offsetPtr, - const U32 mls, const ZSTD_dictMode_e dictMode) -{ - const ZSTD_compressionParameters* const cParams = &ms->cParams; - U32* const chainTable = ms->chainTable; - const U32 chainSize = (1 << cParams->chainLog); - const U32 chainMask = chainSize-1; - const BYTE* const base = ms->window.base; - const BYTE* const dictBase = ms->window.dictBase; - const U32 dictLimit = ms->window.dictLimit; - const BYTE* const prefixStart = base + dictLimit; - const BYTE* const dictEnd = dictBase + dictLimit; - const U32 curr = (U32)(ip-base); - const U32 maxDistance = 1U << cParams->windowLog; - const U32 lowestValid = ms->window.lowLimit; - const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid; - const U32 isDictionary = (ms->loadedDictEnd != 0); - const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance; - const U32 minChain = curr > chainSize ? curr - chainSize : 0; - U32 nbAttempts = 1U << cParams->searchLog; - size_t ml=4-1; - - const ZSTD_MatchState_t* const dms = ms->dictMatchState; - const U32 ddsHashLog = dictMode == ZSTD_dedicatedDictSearch - ? dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG : 0; - const size_t ddsIdx = dictMode == ZSTD_dedicatedDictSearch - ? ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG : 0; - - U32 matchIndex; - - if (dictMode == ZSTD_dedicatedDictSearch) { - const U32* entry = &dms->hashTable[ddsIdx]; - PREFETCH_L1(entry); - } - - /* HC4 match finder */ - matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls, ms->lazySkipping); - - for ( ; (matchIndex>=lowLimit) & (nbAttempts>0) ; nbAttempts--) { - size_t currentMl=0; - if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) { - const BYTE* const match = base + matchIndex; - assert(matchIndex >= dictLimit); /* ensures this is true if dictMode != ZSTD_extDict */ - /* read 4B starting from (match + ml + 1 - sizeof(U32)) */ - if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3)) /* potentially better */ - currentMl = ZSTD_count(ip, match, iLimit); - } else { - const BYTE* const match = dictBase + matchIndex; - assert(match+4 <= dictEnd); - if (MEM_read32(match) == MEM_read32(ip)) /* assumption : matchIndex <= dictLimit-4 (by table construction) */ - currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4; - } - - /* save best solution */ - if (currentMl > ml) { - ml = currentMl; - *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex); - if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */ - } - - if (matchIndex <= minChain) break; - matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask); - } - - assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */ - if (dictMode == ZSTD_dedicatedDictSearch) { - ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts, dms, - ip, iLimit, prefixStart, curr, dictLimit, ddsIdx); - } else if (dictMode == ZSTD_dictMatchState) { - const U32* const dmsChainTable = dms->chainTable; - const U32 dmsChainSize = (1 << dms->cParams.chainLog); - const U32 dmsChainMask = dmsChainSize - 1; - const U32 dmsLowestIndex = dms->window.dictLimit; - const BYTE* const dmsBase = dms->window.base; - const BYTE* const dmsEnd = dms->window.nextSrc; - const U32 dmsSize = (U32)(dmsEnd - dmsBase); - const U32 dmsIndexDelta = dictLimit - dmsSize; - const U32 dmsMinChain = dmsSize > dmsChainSize ? dmsSize - dmsChainSize : 0; - - matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)]; - - for ( ; (matchIndex>=dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) { - size_t currentMl=0; - const BYTE* const match = dmsBase + matchIndex; - assert(match+4 <= dmsEnd); - if (MEM_read32(match) == MEM_read32(ip)) /* assumption : matchIndex <= dictLimit-4 (by table construction) */ - currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4; - - /* save best solution */ - if (currentMl > ml) { - ml = currentMl; - assert(curr > matchIndex + dmsIndexDelta); - *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta)); - if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */ - } - - if (matchIndex <= dmsMinChain) break; - - matchIndex = dmsChainTable[matchIndex & dmsChainMask]; - } - } - - return ml; -} - -/* ********************************* -* (SIMD) Row-based matchfinder -***********************************/ -/* Constants for row-based hash */ -#define ZSTD_ROW_HASH_TAG_MASK ((1u << ZSTD_ROW_HASH_TAG_BITS) - 1) -#define ZSTD_ROW_HASH_MAX_ENTRIES 64 /* absolute maximum number of entries per row, for all configurations */ - -#define ZSTD_ROW_HASH_CACHE_MASK (ZSTD_ROW_HASH_CACHE_SIZE - 1) - -typedef U64 ZSTD_VecMask; /* Clarifies when we are interacting with a U64 representing a mask of matches */ - -/* ZSTD_VecMask_next(): - * Starting from the LSB, returns the idx of the next non-zero bit. - * Basically counting the nb of trailing zeroes. - */ -MEM_STATIC U32 ZSTD_VecMask_next(ZSTD_VecMask val) { - return ZSTD_countTrailingZeros64(val); -} - -/* ZSTD_row_nextIndex(): - * Returns the next index to insert at within a tagTable row, and updates the "head" - * value to reflect the update. Essentially cycles backwards from [1, {entries per row}) - */ -FORCE_INLINE_TEMPLATE U32 ZSTD_row_nextIndex(BYTE* const tagRow, U32 const rowMask) { - U32 next = (*tagRow-1) & rowMask; - next += (next == 0) ? rowMask : 0; /* skip first position */ - *tagRow = (BYTE)next; - return next; -} - -/* ZSTD_isAligned(): - * Checks that a pointer is aligned to "align" bytes which must be a power of 2. - */ -MEM_STATIC int ZSTD_isAligned(void const* ptr, size_t align) { - assert((align & (align - 1)) == 0); - return (((size_t)ptr) & (align - 1)) == 0; -} - -/* ZSTD_row_prefetch(): - * Performs prefetching for the hashTable and tagTable at a given row. - */ -FORCE_INLINE_TEMPLATE void ZSTD_row_prefetch(U32 const* hashTable, BYTE const* tagTable, U32 const relRow, U32 const rowLog) { - PREFETCH_L1(hashTable + relRow); - if (rowLog >= 5) { - PREFETCH_L1(hashTable + relRow + 16); - /* Note: prefetching more of the hash table does not appear to be beneficial for 128-entry rows */ - } - PREFETCH_L1(tagTable + relRow); - if (rowLog == 6) { - PREFETCH_L1(tagTable + relRow + 32); - } - assert(rowLog == 4 || rowLog == 5 || rowLog == 6); - assert(ZSTD_isAligned(hashTable + relRow, 64)); /* prefetched hash row always 64-byte aligned */ - assert(ZSTD_isAligned(tagTable + relRow, (size_t)1 << rowLog)); /* prefetched tagRow sits on correct multiple of bytes (32,64,128) */ -} - -/* ZSTD_row_fillHashCache(): - * Fill up the hash cache starting at idx, prefetching up to ZSTD_ROW_HASH_CACHE_SIZE entries, - * but not beyond iLimit. - */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -void ZSTD_row_fillHashCache(ZSTD_MatchState_t* ms, const BYTE* base, - U32 const rowLog, U32 const mls, - U32 idx, const BYTE* const iLimit) -{ - U32 const* const hashTable = ms->hashTable; - BYTE const* const tagTable = ms->tagTable; - U32 const hashLog = ms->rowHashLog; - U32 const maxElemsToPrefetch = (base + idx) > iLimit ? 0 : (U32)(iLimit - (base + idx) + 1); - U32 const lim = idx + MIN(ZSTD_ROW_HASH_CACHE_SIZE, maxElemsToPrefetch); - - for (; idx < lim; ++idx) { - U32 const hash = (U32)ZSTD_hashPtrSalted(base + idx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt); - U32 const row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog; - ZSTD_row_prefetch(hashTable, tagTable, row, rowLog); - ms->hashCache[idx & ZSTD_ROW_HASH_CACHE_MASK] = hash; - } - - DEBUGLOG(6, "ZSTD_row_fillHashCache(): [%u %u %u %u %u %u %u %u]", ms->hashCache[0], ms->hashCache[1], - ms->hashCache[2], ms->hashCache[3], ms->hashCache[4], - ms->hashCache[5], ms->hashCache[6], ms->hashCache[7]); -} - -/* ZSTD_row_nextCachedHash(): - * Returns the hash of base + idx, and replaces the hash in the hash cache with the byte at - * base + idx + ZSTD_ROW_HASH_CACHE_SIZE. Also prefetches the appropriate rows from hashTable and tagTable. - */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -U32 ZSTD_row_nextCachedHash(U32* cache, U32 const* hashTable, - BYTE const* tagTable, BYTE const* base, - U32 idx, U32 const hashLog, - U32 const rowLog, U32 const mls, - U64 const hashSalt) -{ - U32 const newHash = (U32)ZSTD_hashPtrSalted(base+idx+ZSTD_ROW_HASH_CACHE_SIZE, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt); - U32 const row = (newHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog; - ZSTD_row_prefetch(hashTable, tagTable, row, rowLog); - { U32 const hash = cache[idx & ZSTD_ROW_HASH_CACHE_MASK]; - cache[idx & ZSTD_ROW_HASH_CACHE_MASK] = newHash; - return hash; - } -} - -/* ZSTD_row_update_internalImpl(): - * Updates the hash table with positions starting from updateStartIdx until updateEndIdx. - */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -void ZSTD_row_update_internalImpl(ZSTD_MatchState_t* ms, - U32 updateStartIdx, U32 const updateEndIdx, - U32 const mls, U32 const rowLog, - U32 const rowMask, U32 const useCache) -{ - U32* const hashTable = ms->hashTable; - BYTE* const tagTable = ms->tagTable; - U32 const hashLog = ms->rowHashLog; - const BYTE* const base = ms->window.base; - - DEBUGLOG(6, "ZSTD_row_update_internalImpl(): updateStartIdx=%u, updateEndIdx=%u", updateStartIdx, updateEndIdx); - for (; updateStartIdx < updateEndIdx; ++updateStartIdx) { - U32 const hash = useCache ? ZSTD_row_nextCachedHash(ms->hashCache, hashTable, tagTable, base, updateStartIdx, hashLog, rowLog, mls, ms->hashSalt) - : (U32)ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt); - U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog; - U32* const row = hashTable + relRow; - BYTE* tagRow = tagTable + relRow; - U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask); - - assert(hash == ZSTD_hashPtrSalted(base + updateStartIdx, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, ms->hashSalt)); - tagRow[pos] = hash & ZSTD_ROW_HASH_TAG_MASK; - row[pos] = updateStartIdx; - } -} - -/* ZSTD_row_update_internal(): - * Inserts the byte at ip into the appropriate position in the hash table, and updates ms->nextToUpdate. - * Skips sections of long matches as is necessary. - */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -void ZSTD_row_update_internal(ZSTD_MatchState_t* ms, const BYTE* ip, - U32 const mls, U32 const rowLog, - U32 const rowMask, U32 const useCache) -{ - U32 idx = ms->nextToUpdate; - const BYTE* const base = ms->window.base; - const U32 target = (U32)(ip - base); - const U32 kSkipThreshold = 384; - const U32 kMaxMatchStartPositionsToUpdate = 96; - const U32 kMaxMatchEndPositionsToUpdate = 32; - - if (useCache) { - /* Only skip positions when using hash cache, i.e. - * if we are loading a dict, don't skip anything. - * If we decide to skip, then we only update a set number - * of positions at the beginning and end of the match. - */ - if (UNLIKELY(target - idx > kSkipThreshold)) { - U32 const bound = idx + kMaxMatchStartPositionsToUpdate; - ZSTD_row_update_internalImpl(ms, idx, bound, mls, rowLog, rowMask, useCache); - idx = target - kMaxMatchEndPositionsToUpdate; - ZSTD_row_fillHashCache(ms, base, rowLog, mls, idx, ip+1); - } - } - assert(target >= idx); - ZSTD_row_update_internalImpl(ms, idx, target, mls, rowLog, rowMask, useCache); - ms->nextToUpdate = target; -} - -/* ZSTD_row_update(): - * External wrapper for ZSTD_row_update_internal(). Used for filling the hashtable during dictionary - * processing. - */ -void ZSTD_row_update(ZSTD_MatchState_t* const ms, const BYTE* ip) { - const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6); - const U32 rowMask = (1u << rowLog) - 1; - const U32 mls = MIN(ms->cParams.minMatch, 6 /* mls caps out at 6 */); - - DEBUGLOG(5, "ZSTD_row_update(), rowLog=%u", rowLog); - ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 0 /* don't use cache */); -} - -/* Returns the mask width of bits group of which will be set to 1. Given not all - * architectures have easy movemask instruction, this helps to iterate over - * groups of bits easier and faster. - */ -FORCE_INLINE_TEMPLATE U32 -ZSTD_row_matchMaskGroupWidth(const U32 rowEntries) -{ - assert((rowEntries == 16) || (rowEntries == 32) || rowEntries == 64); - assert(rowEntries <= ZSTD_ROW_HASH_MAX_ENTRIES); - (void)rowEntries; -#if defined(ZSTD_ARCH_ARM_NEON) - /* NEON path only works for little endian */ - if (!MEM_isLittleEndian()) { - return 1; - } - if (rowEntries == 16) { - return 4; - } - if (rowEntries == 32) { - return 2; - } - if (rowEntries == 64) { - return 1; - } -#endif - return 1; -} - -#if defined(ZSTD_ARCH_X86_SSE2) -FORCE_INLINE_TEMPLATE ZSTD_VecMask -ZSTD_row_getSSEMask(int nbChunks, const BYTE* const src, const BYTE tag, const U32 head) -{ - const __m128i comparisonMask = _mm_set1_epi8((char)tag); - int matches[4] = {0}; - int i; - assert(nbChunks == 1 || nbChunks == 2 || nbChunks == 4); - for (i=0; i> chunkSize; - do { - size_t chunk = MEM_readST(&src[i]); - chunk ^= splatChar; - chunk = (((chunk | x80) - x01) | chunk) & x80; - matches <<= chunkSize; - matches |= (chunk * extractMagic) >> shiftAmount; - i -= chunkSize; - } while (i >= 0); - } else { /* big endian: reverse bits during extraction */ - const size_t msb = xFF ^ (xFF >> 1); - const size_t extractMagic = (msb / 0x1FF) | msb; - do { - size_t chunk = MEM_readST(&src[i]); - chunk ^= splatChar; - chunk = (((chunk | x80) - x01) | chunk) & x80; - matches <<= chunkSize; - matches |= ((chunk >> 7) * extractMagic) >> shiftAmount; - i -= chunkSize; - } while (i >= 0); - } - matches = ~matches; - if (rowEntries == 16) { - return ZSTD_rotateRight_U16((U16)matches, headGrouped); - } else if (rowEntries == 32) { - return ZSTD_rotateRight_U32((U32)matches, headGrouped); - } else { - return ZSTD_rotateRight_U64((U64)matches, headGrouped); - } - } -#endif -} - -/* The high-level approach of the SIMD row based match finder is as follows: - * - Figure out where to insert the new entry: - * - Generate a hash for current input position and split it into a one byte of tag and `rowHashLog` bits of index. - * - The hash is salted by a value that changes on every context reset, so when the same table is used - * we will avoid collisions that would otherwise slow us down by introducing phantom matches. - * - The hashTable is effectively split into groups or "rows" of 15 or 31 entries of U32, and the index determines - * which row to insert into. - * - Determine the correct position within the row to insert the entry into. Each row of 15 or 31 can - * be considered as a circular buffer with a "head" index that resides in the tagTable (overall 16 or 32 bytes - * per row). - * - Use SIMD to efficiently compare the tags in the tagTable to the 1-byte tag calculated for the position and - * generate a bitfield that we can cycle through to check the collisions in the hash table. - * - Pick the longest match. - * - Insert the tag into the equivalent row and position in the tagTable. - */ -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_RowFindBestMatch( - ZSTD_MatchState_t* ms, - const BYTE* const ip, const BYTE* const iLimit, - size_t* offsetPtr, - const U32 mls, const ZSTD_dictMode_e dictMode, - const U32 rowLog) -{ - U32* const hashTable = ms->hashTable; - BYTE* const tagTable = ms->tagTable; - U32* const hashCache = ms->hashCache; - const U32 hashLog = ms->rowHashLog; - const ZSTD_compressionParameters* const cParams = &ms->cParams; - const BYTE* const base = ms->window.base; - const BYTE* const dictBase = ms->window.dictBase; - const U32 dictLimit = ms->window.dictLimit; - const BYTE* const prefixStart = base + dictLimit; - const BYTE* const dictEnd = dictBase + dictLimit; - const U32 curr = (U32)(ip-base); - const U32 maxDistance = 1U << cParams->windowLog; - const U32 lowestValid = ms->window.lowLimit; - const U32 withinMaxDistance = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid; - const U32 isDictionary = (ms->loadedDictEnd != 0); - const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance; - const U32 rowEntries = (1U << rowLog); - const U32 rowMask = rowEntries - 1; - const U32 cappedSearchLog = MIN(cParams->searchLog, rowLog); /* nb of searches is capped at nb entries per row */ - const U32 groupWidth = ZSTD_row_matchMaskGroupWidth(rowEntries); - const U64 hashSalt = ms->hashSalt; - U32 nbAttempts = 1U << cappedSearchLog; - size_t ml=4-1; - U32 hash; - - /* DMS/DDS variables that may be referenced laster */ - const ZSTD_MatchState_t* const dms = ms->dictMatchState; - - /* Initialize the following variables to satisfy static analyzer */ - size_t ddsIdx = 0; - U32 ddsExtraAttempts = 0; /* cctx hash tables are limited in searches, but allow extra searches into DDS */ - U32 dmsTag = 0; - U32* dmsRow = NULL; - BYTE* dmsTagRow = NULL; - - if (dictMode == ZSTD_dedicatedDictSearch) { - const U32 ddsHashLog = dms->cParams.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG; - { /* Prefetch DDS hashtable entry */ - ddsIdx = ZSTD_hashPtr(ip, ddsHashLog, mls) << ZSTD_LAZY_DDSS_BUCKET_LOG; - PREFETCH_L1(&dms->hashTable[ddsIdx]); - } - ddsExtraAttempts = cParams->searchLog > rowLog ? 1U << (cParams->searchLog - rowLog) : 0; - } - - if (dictMode == ZSTD_dictMatchState) { - /* Prefetch DMS rows */ - U32* const dmsHashTable = dms->hashTable; - BYTE* const dmsTagTable = dms->tagTable; - U32 const dmsHash = (U32)ZSTD_hashPtr(ip, dms->rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls); - U32 const dmsRelRow = (dmsHash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog; - dmsTag = dmsHash & ZSTD_ROW_HASH_TAG_MASK; - dmsTagRow = (BYTE*)(dmsTagTable + dmsRelRow); - dmsRow = dmsHashTable + dmsRelRow; - ZSTD_row_prefetch(dmsHashTable, dmsTagTable, dmsRelRow, rowLog); - } - - /* Update the hashTable and tagTable up to (but not including) ip */ - if (!ms->lazySkipping) { - ZSTD_row_update_internal(ms, ip, mls, rowLog, rowMask, 1 /* useCache */); - hash = ZSTD_row_nextCachedHash(hashCache, hashTable, tagTable, base, curr, hashLog, rowLog, mls, hashSalt); - } else { - /* Stop inserting every position when in the lazy skipping mode. - * The hash cache is also not kept up to date in this mode. - */ - hash = (U32)ZSTD_hashPtrSalted(ip, hashLog + ZSTD_ROW_HASH_TAG_BITS, mls, hashSalt); - ms->nextToUpdate = curr; - } - ms->hashSaltEntropy += hash; /* collect salt entropy */ - - { /* Get the hash for ip, compute the appropriate row */ - U32 const relRow = (hash >> ZSTD_ROW_HASH_TAG_BITS) << rowLog; - U32 const tag = hash & ZSTD_ROW_HASH_TAG_MASK; - U32* const row = hashTable + relRow; - BYTE* tagRow = (BYTE*)(tagTable + relRow); - U32 const headGrouped = (*tagRow & rowMask) * groupWidth; - U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES]; - size_t numMatches = 0; - size_t currMatch = 0; - ZSTD_VecMask matches = ZSTD_row_getMatchMask(tagRow, (BYTE)tag, headGrouped, rowEntries); - - /* Cycle through the matches and prefetch */ - for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) { - U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask; - U32 const matchIndex = row[matchPos]; - if(matchPos == 0) continue; - assert(numMatches < rowEntries); - if (matchIndex < lowLimit) - break; - if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) { - PREFETCH_L1(base + matchIndex); - } else { - PREFETCH_L1(dictBase + matchIndex); - } - matchBuffer[numMatches++] = matchIndex; - --nbAttempts; - } - - /* Speed opt: insert current byte into hashtable too. This allows us to avoid one iteration of the loop - in ZSTD_row_update_internal() at the next search. */ - { - U32 const pos = ZSTD_row_nextIndex(tagRow, rowMask); - tagRow[pos] = (BYTE)tag; - row[pos] = ms->nextToUpdate++; - } - - /* Return the longest match */ - for (; currMatch < numMatches; ++currMatch) { - U32 const matchIndex = matchBuffer[currMatch]; - size_t currentMl=0; - assert(matchIndex < curr); - assert(matchIndex >= lowLimit); - - if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) { - const BYTE* const match = base + matchIndex; - assert(matchIndex >= dictLimit); /* ensures this is true if dictMode != ZSTD_extDict */ - /* read 4B starting from (match + ml + 1 - sizeof(U32)) */ - if (MEM_read32(match + ml - 3) == MEM_read32(ip + ml - 3)) /* potentially better */ - currentMl = ZSTD_count(ip, match, iLimit); - } else { - const BYTE* const match = dictBase + matchIndex; - assert(match+4 <= dictEnd); - if (MEM_read32(match) == MEM_read32(ip)) /* assumption : matchIndex <= dictLimit-4 (by table construction) */ - currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4; - } - - /* Save best solution */ - if (currentMl > ml) { - ml = currentMl; - *offsetPtr = OFFSET_TO_OFFBASE(curr - matchIndex); - if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */ - } - } - } - - assert(nbAttempts <= (1U << ZSTD_SEARCHLOG_MAX)); /* Check we haven't underflowed. */ - if (dictMode == ZSTD_dedicatedDictSearch) { - ml = ZSTD_dedicatedDictSearch_lazy_search(offsetPtr, ml, nbAttempts + ddsExtraAttempts, dms, - ip, iLimit, prefixStart, curr, dictLimit, ddsIdx); - } else if (dictMode == ZSTD_dictMatchState) { - /* TODO: Measure and potentially add prefetching to DMS */ - const U32 dmsLowestIndex = dms->window.dictLimit; - const BYTE* const dmsBase = dms->window.base; - const BYTE* const dmsEnd = dms->window.nextSrc; - const U32 dmsSize = (U32)(dmsEnd - dmsBase); - const U32 dmsIndexDelta = dictLimit - dmsSize; - - { U32 const headGrouped = (*dmsTagRow & rowMask) * groupWidth; - U32 matchBuffer[ZSTD_ROW_HASH_MAX_ENTRIES]; - size_t numMatches = 0; - size_t currMatch = 0; - ZSTD_VecMask matches = ZSTD_row_getMatchMask(dmsTagRow, (BYTE)dmsTag, headGrouped, rowEntries); - - for (; (matches > 0) && (nbAttempts > 0); matches &= (matches - 1)) { - U32 const matchPos = ((headGrouped + ZSTD_VecMask_next(matches)) / groupWidth) & rowMask; - U32 const matchIndex = dmsRow[matchPos]; - if(matchPos == 0) continue; - if (matchIndex < dmsLowestIndex) - break; - PREFETCH_L1(dmsBase + matchIndex); - matchBuffer[numMatches++] = matchIndex; - --nbAttempts; - } - - /* Return the longest match */ - for (; currMatch < numMatches; ++currMatch) { - U32 const matchIndex = matchBuffer[currMatch]; - size_t currentMl=0; - assert(matchIndex >= dmsLowestIndex); - assert(matchIndex < curr); - - { const BYTE* const match = dmsBase + matchIndex; - assert(match+4 <= dmsEnd); - if (MEM_read32(match) == MEM_read32(ip)) - currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4; - } - - if (currentMl > ml) { - ml = currentMl; - assert(curr > matchIndex + dmsIndexDelta); - *offsetPtr = OFFSET_TO_OFFBASE(curr - (matchIndex + dmsIndexDelta)); - if (ip+currentMl == iLimit) break; - } - } - } - } - return ml; -} - - -/** - * Generate search functions templated on (dictMode, mls, rowLog). - * These functions are outlined for code size & compilation time. - * ZSTD_searchMax() dispatches to the correct implementation function. - * - * TODO: The start of the search function involves loading and calculating a - * bunch of constants from the ZSTD_MatchState_t. These computations could be - * done in an initialization function, and saved somewhere in the match state. - * Then we could pass a pointer to the saved state instead of the match state, - * and avoid duplicate computations. - * - * TODO: Move the match re-winding into searchMax. This improves compression - * ratio, and unlocks further simplifications with the next TODO. - * - * TODO: Try moving the repcode search into searchMax. After the re-winding - * and repcode search are in searchMax, there is no more logic in the match - * finder loop that requires knowledge about the dictMode. So we should be - * able to avoid force inlining it, and we can join the extDict loop with - * the single segment loop. It should go in searchMax instead of its own - * function to avoid having multiple virtual function calls per search. - */ - -#define ZSTD_BT_SEARCH_FN(dictMode, mls) ZSTD_BtFindBestMatch_##dictMode##_##mls -#define ZSTD_HC_SEARCH_FN(dictMode, mls) ZSTD_HcFindBestMatch_##dictMode##_##mls -#define ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog) ZSTD_RowFindBestMatch_##dictMode##_##mls##_##rowLog - -#define ZSTD_SEARCH_FN_ATTRS FORCE_NOINLINE - -#define GEN_ZSTD_BT_SEARCH_FN(dictMode, mls) \ - ZSTD_SEARCH_FN_ATTRS size_t ZSTD_BT_SEARCH_FN(dictMode, mls)( \ - ZSTD_MatchState_t* ms, \ - const BYTE* ip, const BYTE* const iLimit, \ - size_t* offBasePtr) \ - { \ - assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls); \ - return ZSTD_BtFindBestMatch(ms, ip, iLimit, offBasePtr, mls, ZSTD_##dictMode); \ - } \ - -#define GEN_ZSTD_HC_SEARCH_FN(dictMode, mls) \ - ZSTD_SEARCH_FN_ATTRS size_t ZSTD_HC_SEARCH_FN(dictMode, mls)( \ - ZSTD_MatchState_t* ms, \ - const BYTE* ip, const BYTE* const iLimit, \ - size_t* offsetPtr) \ - { \ - assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls); \ - return ZSTD_HcFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode); \ - } \ - -#define GEN_ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog) \ - ZSTD_SEARCH_FN_ATTRS size_t ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)( \ - ZSTD_MatchState_t* ms, \ - const BYTE* ip, const BYTE* const iLimit, \ - size_t* offsetPtr) \ - { \ - assert(MAX(4, MIN(6, ms->cParams.minMatch)) == mls); \ - assert(MAX(4, MIN(6, ms->cParams.searchLog)) == rowLog); \ - return ZSTD_RowFindBestMatch(ms, ip, iLimit, offsetPtr, mls, ZSTD_##dictMode, rowLog); \ - } \ - -#define ZSTD_FOR_EACH_ROWLOG(X, dictMode, mls) \ - X(dictMode, mls, 4) \ - X(dictMode, mls, 5) \ - X(dictMode, mls, 6) - -#define ZSTD_FOR_EACH_MLS_ROWLOG(X, dictMode) \ - ZSTD_FOR_EACH_ROWLOG(X, dictMode, 4) \ - ZSTD_FOR_EACH_ROWLOG(X, dictMode, 5) \ - ZSTD_FOR_EACH_ROWLOG(X, dictMode, 6) - -#define ZSTD_FOR_EACH_MLS(X, dictMode) \ - X(dictMode, 4) \ - X(dictMode, 5) \ - X(dictMode, 6) - -#define ZSTD_FOR_EACH_DICT_MODE(X, ...) \ - X(__VA_ARGS__, noDict) \ - X(__VA_ARGS__, extDict) \ - X(__VA_ARGS__, dictMatchState) \ - X(__VA_ARGS__, dedicatedDictSearch) - -/* Generate row search fns for each combination of (dictMode, mls, rowLog) */ -ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS_ROWLOG, GEN_ZSTD_ROW_SEARCH_FN) -/* Generate binary Tree search fns for each combination of (dictMode, mls) */ -ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_BT_SEARCH_FN) -/* Generate hash chain search fns for each combination of (dictMode, mls) */ -ZSTD_FOR_EACH_DICT_MODE(ZSTD_FOR_EACH_MLS, GEN_ZSTD_HC_SEARCH_FN) - -typedef enum { search_hashChain=0, search_binaryTree=1, search_rowHash=2 } searchMethod_e; - -#define GEN_ZSTD_CALL_BT_SEARCH_FN(dictMode, mls) \ - case mls: \ - return ZSTD_BT_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr); -#define GEN_ZSTD_CALL_HC_SEARCH_FN(dictMode, mls) \ - case mls: \ - return ZSTD_HC_SEARCH_FN(dictMode, mls)(ms, ip, iend, offsetPtr); -#define GEN_ZSTD_CALL_ROW_SEARCH_FN(dictMode, mls, rowLog) \ - case rowLog: \ - return ZSTD_ROW_SEARCH_FN(dictMode, mls, rowLog)(ms, ip, iend, offsetPtr); - -#define ZSTD_SWITCH_MLS(X, dictMode) \ - switch (mls) { \ - ZSTD_FOR_EACH_MLS(X, dictMode) \ - } - -#define ZSTD_SWITCH_ROWLOG(dictMode, mls) \ - case mls: \ - switch (rowLog) { \ - ZSTD_FOR_EACH_ROWLOG(GEN_ZSTD_CALL_ROW_SEARCH_FN, dictMode, mls) \ - } \ - ZSTD_UNREACHABLE; \ - break; - -#define ZSTD_SWITCH_SEARCH_METHOD(dictMode) \ - switch (searchMethod) { \ - case search_hashChain: \ - ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_HC_SEARCH_FN, dictMode) \ - break; \ - case search_binaryTree: \ - ZSTD_SWITCH_MLS(GEN_ZSTD_CALL_BT_SEARCH_FN, dictMode) \ - break; \ - case search_rowHash: \ - ZSTD_SWITCH_MLS(ZSTD_SWITCH_ROWLOG, dictMode) \ - break; \ - } \ - ZSTD_UNREACHABLE; - -/** - * Searches for the longest match at @p ip. - * Dispatches to the correct implementation function based on the - * (searchMethod, dictMode, mls, rowLog). We use switch statements - * here instead of using an indirect function call through a function - * pointer because after Spectre and Meltdown mitigations, indirect - * function calls can be very costly, especially in the kernel. - * - * NOTE: dictMode and searchMethod should be templated, so those switch - * statements should be optimized out. Only the mls & rowLog switches - * should be left. - * - * @param ms The match state. - * @param ip The position to search at. - * @param iend The end of the input data. - * @param[out] offsetPtr Stores the match offset into this pointer. - * @param mls The minimum search length, in the range [4, 6]. - * @param rowLog The row log (if applicable), in the range [4, 6]. - * @param searchMethod The search method to use (templated). - * @param dictMode The dictMode (templated). - * - * @returns The length of the longest match found, or < mls if no match is found. - * If a match is found its offset is stored in @p offsetPtr. - */ -FORCE_INLINE_TEMPLATE size_t ZSTD_searchMax( - ZSTD_MatchState_t* ms, - const BYTE* ip, - const BYTE* iend, - size_t* offsetPtr, - U32 const mls, - U32 const rowLog, - searchMethod_e const searchMethod, - ZSTD_dictMode_e const dictMode) -{ - if (dictMode == ZSTD_noDict) { - ZSTD_SWITCH_SEARCH_METHOD(noDict) - } else if (dictMode == ZSTD_extDict) { - ZSTD_SWITCH_SEARCH_METHOD(extDict) - } else if (dictMode == ZSTD_dictMatchState) { - ZSTD_SWITCH_SEARCH_METHOD(dictMatchState) - } else if (dictMode == ZSTD_dedicatedDictSearch) { - ZSTD_SWITCH_SEARCH_METHOD(dedicatedDictSearch) - } - ZSTD_UNREACHABLE; - return 0; -} - -/* ******************************* -* Common parser - lazy strategy -*********************************/ - -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_compressBlock_lazy_generic( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, - U32 rep[ZSTD_REP_NUM], - const void* src, size_t srcSize, - const searchMethod_e searchMethod, const U32 depth, - ZSTD_dictMode_e const dictMode) -{ - const BYTE* const istart = (const BYTE*)src; - const BYTE* ip = istart; - const BYTE* anchor = istart; - const BYTE* const iend = istart + srcSize; - const BYTE* const ilimit = (searchMethod == search_rowHash) ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8; - const BYTE* const base = ms->window.base; - const U32 prefixLowestIndex = ms->window.dictLimit; - const BYTE* const prefixLowest = base + prefixLowestIndex; - const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6); - const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6); - - U32 offset_1 = rep[0], offset_2 = rep[1]; - U32 offsetSaved1 = 0, offsetSaved2 = 0; - - const int isDMS = dictMode == ZSTD_dictMatchState; - const int isDDS = dictMode == ZSTD_dedicatedDictSearch; - const int isDxS = isDMS || isDDS; - const ZSTD_MatchState_t* const dms = ms->dictMatchState; - const U32 dictLowestIndex = isDxS ? dms->window.dictLimit : 0; - const BYTE* const dictBase = isDxS ? dms->window.base : NULL; - const BYTE* const dictLowest = isDxS ? dictBase + dictLowestIndex : NULL; - const BYTE* const dictEnd = isDxS ? dms->window.nextSrc : NULL; - const U32 dictIndexDelta = isDxS ? - prefixLowestIndex - (U32)(dictEnd - dictBase) : - 0; - const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest)); - - DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u) (searchFunc=%u)", (U32)dictMode, (U32)searchMethod); - ip += (dictAndPrefixLength == 0); - if (dictMode == ZSTD_noDict) { - U32 const curr = (U32)(ip - base); - U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, curr, ms->cParams.windowLog); - U32 const maxRep = curr - windowLow; - if (offset_2 > maxRep) offsetSaved2 = offset_2, offset_2 = 0; - if (offset_1 > maxRep) offsetSaved1 = offset_1, offset_1 = 0; - } - if (isDxS) { - /* dictMatchState repCode checks don't currently handle repCode == 0 - * disabling. */ - assert(offset_1 <= dictAndPrefixLength); - assert(offset_2 <= dictAndPrefixLength); - } - - /* Reset the lazy skipping state */ - ms->lazySkipping = 0; - - if (searchMethod == search_rowHash) { - ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit); - } - - /* Match Loop */ -#if defined(__GNUC__) && defined(__x86_64__) - /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the - * code alignment is perturbed. To fix the instability align the loop on 32-bytes. - */ - __asm__(".p2align 5"); -#endif - while (ip < ilimit) { - size_t matchLength=0; - size_t offBase = REPCODE1_TO_OFFBASE; - const BYTE* start=ip+1; - DEBUGLOG(7, "search baseline (depth 0)"); - - /* check repCode */ - if (isDxS) { - const U32 repIndex = (U32)(ip - base) + 1 - offset_1; - const BYTE* repMatch = ((dictMode == ZSTD_dictMatchState || dictMode == ZSTD_dedicatedDictSearch) - && repIndex < prefixLowestIndex) ? - dictBase + (repIndex - dictIndexDelta) : - base + repIndex; - if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex)) - && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) { - const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend; - matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4; - if (depth==0) goto _storeSequence; - } - } - if ( dictMode == ZSTD_noDict - && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) { - matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4; - if (depth==0) goto _storeSequence; - } - - /* first search (depth 0) */ - { size_t offbaseFound = 999999999; - size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &offbaseFound, mls, rowLog, searchMethod, dictMode); - if (ml2 > matchLength) - matchLength = ml2, start = ip, offBase = offbaseFound; - } - - if (matchLength < 4) { - size_t const step = ((size_t)(ip-anchor) >> kSearchStrength) + 1; /* jump faster over incompressible sections */; - ip += step; - /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time. - * In this mode we stop inserting every position into our tables, and only insert - * positions that we search, which is one in step positions. - * The exact cutoff is flexible, I've just chosen a number that is reasonably high, - * so we minimize the compression ratio loss in "normal" scenarios. This mode gets - * triggered once we've gone 2KB without finding any matches. - */ - ms->lazySkipping = step > kLazySkippingStep; - continue; - } - - /* let's try to find a better solution */ - if (depth>=1) - while (ip0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) { - size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4; - int const gain2 = (int)(mlRep * 3); - int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1); - if ((mlRep >= 4) && (gain2 > gain1)) - matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip; - } - if (isDxS) { - const U32 repIndex = (U32)(ip - base) - offset_1; - const BYTE* repMatch = repIndex < prefixLowestIndex ? - dictBase + (repIndex - dictIndexDelta) : - base + repIndex; - if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex)) - && (MEM_read32(repMatch) == MEM_read32(ip)) ) { - const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend; - size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4; - int const gain2 = (int)(mlRep * 3); - int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1); - if ((mlRep >= 4) && (gain2 > gain1)) - matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip; - } - } - { size_t ofbCandidate=999999999; - size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode); - int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate)); /* raw approx */ - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4); - if ((ml2 >= 4) && (gain2 > gain1)) { - matchLength = ml2, offBase = ofbCandidate, start = ip; - continue; /* search a better one */ - } } - - /* let's find an even better one */ - if ((depth==2) && (ip0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) { - size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4; - int const gain2 = (int)(mlRep * 4); - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1); - if ((mlRep >= 4) && (gain2 > gain1)) - matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip; - } - if (isDxS) { - const U32 repIndex = (U32)(ip - base) - offset_1; - const BYTE* repMatch = repIndex < prefixLowestIndex ? - dictBase + (repIndex - dictIndexDelta) : - base + repIndex; - if ((ZSTD_index_overlap_check(prefixLowestIndex, repIndex)) - && (MEM_read32(repMatch) == MEM_read32(ip)) ) { - const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend; - size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4; - int const gain2 = (int)(mlRep * 4); - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1); - if ((mlRep >= 4) && (gain2 > gain1)) - matchLength = mlRep, offBase = REPCODE1_TO_OFFBASE, start = ip; - } - } - { size_t ofbCandidate=999999999; - size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, dictMode); - int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate)); /* raw approx */ - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7); - if ((ml2 >= 4) && (gain2 > gain1)) { - matchLength = ml2, offBase = ofbCandidate, start = ip; - continue; - } } } - break; /* nothing found : store previous solution */ - } - - /* NOTE: - * Pay attention that `start[-value]` can lead to strange undefined behavior - * notably if `value` is unsigned, resulting in a large positive `-value`. - */ - /* catch up */ - if (OFFBASE_IS_OFFSET(offBase)) { - if (dictMode == ZSTD_noDict) { - while ( ((start > anchor) & (start - OFFBASE_TO_OFFSET(offBase) > prefixLowest)) - && (start[-1] == (start-OFFBASE_TO_OFFSET(offBase))[-1]) ) /* only search for offset within prefix */ - { start--; matchLength++; } - } - if (isDxS) { - U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase)); - const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex; - const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest; - while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; } /* catch up */ - } - offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase); - } - /* store sequence */ -_storeSequence: - { size_t const litLength = (size_t)(start - anchor); - ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength); - anchor = ip = start + matchLength; - } - if (ms->lazySkipping) { - /* We've found a match, disable lazy skipping mode, and refill the hash cache. */ - if (searchMethod == search_rowHash) { - ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit); - } - ms->lazySkipping = 0; - } - - /* check immediate repcode */ - if (isDxS) { - while (ip <= ilimit) { - U32 const current2 = (U32)(ip-base); - U32 const repIndex = current2 - offset_2; - const BYTE* repMatch = repIndex < prefixLowestIndex ? - dictBase - dictIndexDelta + repIndex : - base + repIndex; - if ( (ZSTD_index_overlap_check(prefixLowestIndex, repIndex)) - && (MEM_read32(repMatch) == MEM_read32(ip)) ) { - const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend; - matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4; - offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap offset_2 <=> offset_1 */ - ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength); - ip += matchLength; - anchor = ip; - continue; - } - break; - } - } - - if (dictMode == ZSTD_noDict) { - while ( ((ip <= ilimit) & (offset_2>0)) - && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) { - /* store sequence */ - matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4; - offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap repcodes */ - ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength); - ip += matchLength; - anchor = ip; - continue; /* faster when present ... (?) */ - } } } - - /* 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); -} -#endif /* build exclusions */ - - #ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_greedy( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_greedy_dictMatchState( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState); -} - -size_t ZSTD_compressBlock_greedy_dedicatedDictSearch( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dedicatedDictSearch); -} - -size_t ZSTD_compressBlock_greedy_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_greedy_dictMatchState_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dictMatchState); -} - -size_t ZSTD_compressBlock_greedy_dedicatedDictSearch_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0, ZSTD_dedicatedDictSearch); -} +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy, + ZSTD_rust_search_hashChain, 0, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_dictMatchState, + ZSTD_rust_search_hashChain, 0, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_dedicatedDictSearch, + ZSTD_rust_search_hashChain, 0, ZSTD_rust_dict_dedicatedDictSearch) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_row, + ZSTD_rust_search_rowHash, 0, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_dictMatchState_row, + ZSTD_rust_search_rowHash, 0, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_dedicatedDictSearch_row, + ZSTD_rust_search_rowHash, 0, ZSTD_rust_dict_dedicatedDictSearch) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_extDict, + ZSTD_rust_search_hashChain, 0, ZSTD_rust_dict_extDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_greedy_extDict_row, + ZSTD_rust_search_rowHash, 0, ZSTD_rust_dict_extDict) #endif #ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_lazy( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_lazy_dictMatchState( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState); -} - -size_t ZSTD_compressBlock_lazy_dedicatedDictSearch( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dedicatedDictSearch); -} - -size_t ZSTD_compressBlock_lazy_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_lazy_dictMatchState_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dictMatchState); -} - -size_t ZSTD_compressBlock_lazy_dedicatedDictSearch_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1, ZSTD_dedicatedDictSearch); -} +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy, + ZSTD_rust_search_hashChain, 1, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_dictMatchState, + ZSTD_rust_search_hashChain, 1, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_dedicatedDictSearch, + ZSTD_rust_search_hashChain, 1, ZSTD_rust_dict_dedicatedDictSearch) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_row, + ZSTD_rust_search_rowHash, 1, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_dictMatchState_row, + ZSTD_rust_search_rowHash, 1, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_dedicatedDictSearch_row, + ZSTD_rust_search_rowHash, 1, ZSTD_rust_dict_dedicatedDictSearch) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_extDict, + ZSTD_rust_search_hashChain, 1, ZSTD_rust_dict_extDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy_extDict_row, + ZSTD_rust_search_rowHash, 1, ZSTD_rust_dict_extDict) #endif #ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_lazy2( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_lazy2_dictMatchState( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState); -} - -size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dedicatedDictSearch); -} - -size_t ZSTD_compressBlock_lazy2_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_lazy2_dictMatchState_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dictMatchState); -} - -size_t ZSTD_compressBlock_lazy2_dedicatedDictSearch_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2, ZSTD_dedicatedDictSearch); -} +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2, + ZSTD_rust_search_hashChain, 2, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_dictMatchState, + ZSTD_rust_search_hashChain, 2, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_dedicatedDictSearch, + ZSTD_rust_search_hashChain, 2, ZSTD_rust_dict_dedicatedDictSearch) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_row, + ZSTD_rust_search_rowHash, 2, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_dictMatchState_row, + ZSTD_rust_search_rowHash, 2, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_dedicatedDictSearch_row, + ZSTD_rust_search_rowHash, 2, ZSTD_rust_dict_dedicatedDictSearch) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_extDict, + ZSTD_rust_search_hashChain, 2, ZSTD_rust_dict_extDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_lazy2_extDict_row, + ZSTD_rust_search_rowHash, 2, ZSTD_rust_dict_extDict) #endif #ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_btlazy2( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict); -} - -size_t ZSTD_compressBlock_btlazy2_dictMatchState( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState); -} +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_btlazy2, + ZSTD_rust_search_binaryTree, 2, ZSTD_rust_dict_noDict) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_btlazy2_dictMatchState, + ZSTD_rust_search_binaryTree, 2, ZSTD_rust_dict_dictMatchState) +ZSTD_RUST_LAZY_WRAPPER(ZSTD_compressBlock_btlazy2_extDict, + ZSTD_rust_search_binaryTree, 2, ZSTD_rust_dict_extDict) #endif -#if !defined(ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR) \ - || !defined(ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR) \ - || !defined(ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR) \ - || !defined(ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR) -FORCE_INLINE_TEMPLATE -ZSTD_ALLOW_POINTER_OVERFLOW_ATTR -size_t ZSTD_compressBlock_lazy_extDict_generic( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, - U32 rep[ZSTD_REP_NUM], - const void* src, size_t srcSize, - const searchMethod_e searchMethod, const U32 depth) -{ - const BYTE* const istart = (const BYTE*)src; - const BYTE* ip = istart; - const BYTE* anchor = istart; - const BYTE* const iend = istart + srcSize; - const BYTE* const ilimit = searchMethod == search_rowHash ? iend - 8 - ZSTD_ROW_HASH_CACHE_SIZE : iend - 8; - const BYTE* const base = ms->window.base; - const U32 dictLimit = ms->window.dictLimit; - const BYTE* const prefixStart = base + dictLimit; - const BYTE* const dictBase = ms->window.dictBase; - const BYTE* const dictEnd = dictBase + dictLimit; - const BYTE* const dictStart = dictBase + ms->window.lowLimit; - const U32 windowLog = ms->cParams.windowLog; - const U32 mls = BOUNDED(4, ms->cParams.minMatch, 6); - const U32 rowLog = BOUNDED(4, ms->cParams.searchLog, 6); +#undef ZSTD_RUST_LAZY_WRAPPER - U32 offset_1 = rep[0], offset_2 = rep[1]; - - DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic (searchFunc=%u)", (U32)searchMethod); - - /* Reset the lazy skipping state */ - ms->lazySkipping = 0; - - /* init */ - ip += (ip == prefixStart); - if (searchMethod == search_rowHash) { - ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit); - } - - /* Match Loop */ -#if defined(__GNUC__) && defined(__x86_64__) - /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the - * code alignment is perturbed. To fix the instability align the loop on 32-bytes. - */ - __asm__(".p2align 5"); -#endif - while (ip < ilimit) { - size_t matchLength=0; - size_t offBase = REPCODE1_TO_OFFBASE; - const BYTE* start=ip+1; - U32 curr = (U32)(ip-base); - - /* check repCode */ - { const U32 windowLow = ZSTD_getLowestMatchIndex(ms, curr+1, windowLog); - const U32 repIndex = (U32)(curr+1 - offset_1); - const BYTE* const repBase = repIndex < dictLimit ? dictBase : base; - const BYTE* const repMatch = repBase + repIndex; - if ( (ZSTD_index_overlap_check(dictLimit, repIndex)) - & (offset_1 <= curr+1 - windowLow) ) /* note: we are searching at curr+1 */ - if (MEM_read32(ip+1) == MEM_read32(repMatch)) { - /* repcode detected we should take it */ - const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; - matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4; - if (depth==0) goto _storeSequence; - } } - - /* first search (depth 0) */ - { size_t ofbCandidate = 999999999; - size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict); - if (ml2 > matchLength) - matchLength = ml2, start = ip, offBase = ofbCandidate; - } - - if (matchLength < 4) { - size_t const step = ((size_t)(ip-anchor) >> kSearchStrength); - ip += step + 1; /* jump faster over incompressible sections */ - /* Enter the lazy skipping mode once we are skipping more than 8 bytes at a time. - * In this mode we stop inserting every position into our tables, and only insert - * positions that we search, which is one in step positions. - * The exact cutoff is flexible, I've just chosen a number that is reasonably high, - * so we minimize the compression ratio loss in "normal" scenarios. This mode gets - * triggered once we've gone 2KB without finding any matches. - */ - ms->lazySkipping = step > kLazySkippingStep; - continue; - } - - /* let's try to find a better solution */ - if (depth>=1) - while (ip repIndex >= windowLow` */ - if (MEM_read32(ip) == MEM_read32(repMatch)) { - /* repcode detected */ - const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; - size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4; - int const gain2 = (int)(repLength * 3); - int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offBase) + 1); - if ((repLength >= 4) && (gain2 > gain1)) - matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip; - } } - - /* search match, depth 1 */ - { size_t ofbCandidate = 999999999; - size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict); - int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate)); /* raw approx */ - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 4); - if ((ml2 >= 4) && (gain2 > gain1)) { - matchLength = ml2, offBase = ofbCandidate, start = ip; - continue; /* search a better one */ - } } - - /* let's find an even better one */ - if ((depth==2) && (ip repIndex >= windowLow` */ - if (MEM_read32(ip) == MEM_read32(repMatch)) { - /* repcode detected */ - const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; - size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4; - int const gain2 = (int)(repLength * 4); - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 1); - if ((repLength >= 4) && (gain2 > gain1)) - matchLength = repLength, offBase = REPCODE1_TO_OFFBASE, start = ip; - } } - - /* search match, depth 2 */ - { size_t ofbCandidate = 999999999; - size_t const ml2 = ZSTD_searchMax(ms, ip, iend, &ofbCandidate, mls, rowLog, searchMethod, ZSTD_extDict); - int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)ofbCandidate)); /* raw approx */ - int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offBase) + 7); - if ((ml2 >= 4) && (gain2 > gain1)) { - matchLength = ml2, offBase = ofbCandidate, start = ip; - continue; - } } } - break; /* nothing found : store previous solution */ - } - - /* catch up */ - if (OFFBASE_IS_OFFSET(offBase)) { - U32 const matchIndex = (U32)((size_t)(start-base) - OFFBASE_TO_OFFSET(offBase)); - const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex; - const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart; - while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; } /* catch up */ - offset_2 = offset_1; offset_1 = (U32)OFFBASE_TO_OFFSET(offBase); - } - - /* store sequence */ -_storeSequence: - { size_t const litLength = (size_t)(start - anchor); - ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offBase, matchLength); - anchor = ip = start + matchLength; - } - if (ms->lazySkipping) { - /* We've found a match, disable lazy skipping mode, and refill the hash cache. */ - if (searchMethod == search_rowHash) { - ZSTD_row_fillHashCache(ms, base, rowLog, mls, ms->nextToUpdate, ilimit); - } - ms->lazySkipping = 0; - } - - /* check immediate repcode */ - while (ip <= ilimit) { - const U32 repCurrent = (U32)(ip-base); - const U32 windowLow = ZSTD_getLowestMatchIndex(ms, repCurrent, windowLog); - const U32 repIndex = repCurrent - offset_2; - const BYTE* const repBase = repIndex < dictLimit ? dictBase : base; - const BYTE* const repMatch = repBase + repIndex; - if ( (ZSTD_index_overlap_check(dictLimit, repIndex)) - & (offset_2 <= repCurrent - windowLow) ) /* equivalent to `curr > repIndex >= windowLow` */ - if (MEM_read32(ip) == MEM_read32(repMatch)) { - /* repcode detected we should take it */ - const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; - matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4; - offBase = offset_2; offset_2 = offset_1; offset_1 = (U32)offBase; /* swap offset history */ - ZSTD_storeSeq(seqStore, 0, anchor, iend, REPCODE1_TO_OFFBASE, matchLength); - ip += matchLength; - anchor = ip; - continue; /* faster when present ... (?) */ - } - break; - } } - - /* Save reps for next block */ - rep[0] = offset_1; - rep[1] = offset_2; - - /* Return the last literals size */ - return (size_t)(iend - anchor); -} -#endif /* build exclusions */ - -#ifndef ZSTD_EXCLUDE_GREEDY_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_greedy_extDict( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0); -} - -size_t ZSTD_compressBlock_greedy_extDict_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 0); -} -#endif - -#ifndef ZSTD_EXCLUDE_LAZY_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_lazy_extDict( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) - -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1); -} - -size_t ZSTD_compressBlock_lazy_extDict_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) - -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 1); -} -#endif - -#ifndef ZSTD_EXCLUDE_LAZY2_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_lazy2_extDict( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) - -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2); -} - -size_t ZSTD_compressBlock_lazy2_extDict_row( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_rowHash, 2); -} -#endif - -#ifndef ZSTD_EXCLUDE_BTLAZY2_BLOCK_COMPRESSOR -size_t ZSTD_compressBlock_btlazy2_extDict( - ZSTD_MatchState_t* ms, SeqStore_t* seqStore, U32 rep[ZSTD_REP_NUM], - void const* src, size_t srcSize) - -{ - return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2); -} -#endif +#endif /* lazy matcher build exclusions */ diff --git a/rust/README.md b/rust/README.md index d0fe191e9..2ac7d048e 100644 --- a/rust/README.md +++ b/rust/README.md @@ -33,16 +33,18 @@ zstd ABI: while preserving the compressor's Huffman-table repeat state. - `zstd_fast` and `zstd_double_fast` implement the single- and two-table fast block match finders, including attached and external dictionary paths. + - `zstd_lazy` implements greedy, lazy, lazy2, and binary-tree matching, + including row-based and dictionary search variants. - Runtime support - `threading` provides platform pthread wrappers required by zstd headers. - `pool` implements the bounded worker pool used by multithreaded compression. - Dictionary support - `zstd_ddict` owns, loads, copies, and references decode dictionaries. -The lazy and optimal block matchers, general decompression, dictionary-building, -legacy, and CLI translation units are still C. They must move before the -rewrite is complete. Keeping that boundary explicit prevents a passing hybrid -build from being mistaken for the final all-Rust result. +The optimal block matcher, general decompression, dictionary-building, legacy, +and CLI translation units are still C. They must move before the rewrite is +complete. Keeping that boundary explicit prevents a passing hybrid build from +being mistaken for the final all-Rust result. ## Compatibility boundary diff --git a/rust/src/lib.rs b/rust/src/lib.rs index aa2954073..84a3ece17 100644 --- a/rust/src/lib.rs +++ b/rust/src/lib.rs @@ -36,4 +36,6 @@ pub mod zstd_double_fast; #[cfg(feature = "compression")] pub mod zstd_fast; #[cfg(feature = "compression")] +pub mod zstd_lazy; +#[cfg(feature = "compression")] pub mod zstd_presplit; diff --git a/rust/src/zstd_lazy.rs b/rust/src/zstd_lazy.rs new file mode 100644 index 000000000..f61aa69fa --- /dev/null +++ b/rust/src/zstd_lazy.rs @@ -0,0 +1,2623 @@ +#![allow(non_camel_case_types)] +#![allow(non_snake_case)] +#![allow(clippy::missing_safety_doc)] +#![allow(clippy::too_many_arguments)] +#![allow(clippy::not_unsafe_ptr_arg_deref)] + +//! Lazy, lazy2, greedy, and binary-tree block match finders. +//! +//! `ZSTD_MatchState_t` deliberately stays on the C side of the boundary. The +//! small representation below contains only the fields that the match finders +//! use, with mutable scalar fields represented by pointers back into the C +//! match state. This keeps the matching and sequence-generation loops in +//! Rust without making a private C layout part of the Rust ABI. + +use crate::bits::ZSTD_highbit32; +use crate::mem::{ + MEM_64bits, MEM_isLittleEndian, MEM_read16, MEM_read32, MEM_readLE32, MEM_readLE64, MEM_readST, +}; +use std::ffi::c_void; +use std::mem::size_of; +use std::os::raw::c_int; +use std::ptr; + +const ZSTD_REP_NUM: usize = 3; +const MINMATCH: usize = 3; +const HASH_READ_SIZE: usize = 8; +const K_LAZY_SKIPPING_STEP: usize = 8; +const K_SEARCH_STRENGTH: usize = 8; +const ZSTD_DUBT_UNSORTED_MARK: u32 = 1; +const ZSTD_LAZY_DDSS_BUCKET_LOG: u32 = 2; +const ZSTD_ROW_HASH_TAG_BITS: u32 = 8; +const ZSTD_ROW_HASH_TAG_MASK: u32 = (1 << ZSTD_ROW_HASH_TAG_BITS) - 1; +const ZSTD_ROW_HASH_CACHE_SIZE: usize = 8; +const ZSTD_ROW_HASH_MAX_ENTRIES: usize = 64; +const REPCODE1_TO_OFFBASE: u32 = 1; +const OFFSET_OFFBASE: u32 = ZSTD_REP_NUM as u32; + +const SEARCH_HASH_CHAIN: c_int = 0; +const SEARCH_BINARY_TREE: c_int = 1; +const SEARCH_ROW_HASH: c_int = 2; + +const DICT_NO_DICT: c_int = 0; +const DICT_EXT: c_int = 1; +const DICT_MATCH_STATE: c_int = 2; +const DICT_DEDICATED: c_int = 3; + +#[repr(C)] +#[derive(Clone, Copy)] +struct SeqDef { + offBase: u32, + litLength: u16, + mlBase: u16, +} + +/// The C shim verifies the layout of this leaf structure. The matcher only +/// needs its sequence and literal append cursors. +#[repr(C)] +struct SeqStore_t { + sequencesStart: *mut SeqDef, + sequences: *mut SeqDef, + litStart: *mut u8, + lit: *mut u8, + llCode: *mut u8, + mlCode: *mut u8, + ofCode: *mut u8, + maxNbSeq: usize, + maxNbLit: usize, + longLengthType: c_int, + longLengthPos: u32, +} + +/// A field-level view of `ZSTD_MatchState_t` used by the lazy match finders. +/// +/// All pointers remain owned by C. `dictMatchState` points at a second view +/// constructed by the C shim for calls that search an attached dictionary. +#[repr(C)] +pub struct ZSTD_RustLazyState { + hashTable: *mut u32, + chainTable: *mut u32, + tagTable: *mut u8, + hashCache: *mut u32, + base: *const u8, + dictBase: *const u8, + nextSrc: *const u8, + dictLimit: u32, + lowLimit: u32, + loadedDictEnd: u32, + nextToUpdate: *mut u32, + lazySkipping: *mut c_int, + hashLog: u32, + chainLog: u32, + minMatch: u32, + searchLog: u32, + windowLog: u32, + rowHashLog: u32, + hashSalt: u64, + hashSaltEntropy: *mut u32, + dictMatchState: *const ZSTD_RustLazyState, +} + +#[inline] +fn ptr_lt(left: *const u8, right: *const u8) -> bool { + (left as usize) < (right as usize) +} + +#[inline] +fn ptr_le(left: *const u8, right: *const u8) -> bool { + (left as usize) <= (right as usize) +} + +#[inline] +fn ptr_gt(left: *const u8, right: *const u8) -> bool { + (left as usize) > (right as usize) +} + +#[inline] +unsafe fn index_from(base: *const u8, value: *const u8) -> u32 { + (value as usize).wrapping_sub(base as usize) as u32 +} + +#[inline] +unsafe fn read32(value: *const u8) -> u32 { + unsafe { MEM_read32(value.cast::()) } +} + +#[inline] +unsafe fn table_get(table: *const u32, index: usize) -> u32 { + unsafe { *table.add(index) } +} + +#[inline] +unsafe fn table_set(table: *mut u32, index: usize, value: u32) { + unsafe { *table.add(index) = value }; +} + +#[inline] +unsafe fn byte_get(table: *const u8, index: usize) -> u8 { + unsafe { *table.add(index) } +} + +#[inline] +unsafe fn byte_set(table: *mut u8, index: usize, value: u8) { + unsafe { *table.add(index) = value }; +} + +#[inline] +unsafe fn next_to_update(state: &ZSTD_RustLazyState) -> u32 { + unsafe { *state.nextToUpdate } +} + +#[inline] +unsafe fn set_next_to_update(state: &ZSTD_RustLazyState, value: u32) { + unsafe { *state.nextToUpdate = value }; +} + +#[inline] +unsafe fn lazy_skipping(state: &ZSTD_RustLazyState) -> bool { + unsafe { *state.lazySkipping != 0 } +} + +#[inline] +unsafe fn set_lazy_skipping(state: &ZSTD_RustLazyState, value: bool) { + unsafe { *state.lazySkipping = c_int::from(value) }; +} + +#[inline] +unsafe fn add_hash_salt_entropy(state: &ZSTD_RustLazyState, value: u32) { + unsafe { + *state.hashSaltEntropy = (*state.hashSaltEntropy).wrapping_add(value); + } +} + +#[inline] +unsafe fn dict_state<'a>(state: &ZSTD_RustLazyState) -> &'a ZSTD_RustLazyState { + debug_assert!(!state.dictMatchState.is_null()); + unsafe { &*state.dictMatchState } +} + +#[inline] +fn bounded(low: u32, value: u32, high: u32) -> u32 { + value.clamp(low, high) +} + +#[inline] +fn hash_shift32(value: u32, hbits: u32) -> usize { + if hbits == 0 { + 0 + } else { + (value >> (32 - hbits)) as usize + } +} + +#[inline] +fn hash_shift64(value: u64, hbits: u32) -> usize { + if hbits == 0 { + 0 + } else { + (value >> (64 - hbits)) as usize + } +} + +#[inline] +unsafe fn hash_ptr(ptr: *const u8, hbits: u32, mls: u32) -> usize { + match mls { + 5 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64(value.wrapping_shl(24).wrapping_mul(889_523_592_379), hbits) + } + 6 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64( + value.wrapping_shl(16).wrapping_mul(227_718_039_650_203), + hbits, + ) + } + 7 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64( + value.wrapping_shl(8).wrapping_mul(58_295_818_150_454_627), + hbits, + ) + } + 8 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64(value.wrapping_mul(0xCF1B_BCDC_B7A5_6463), hbits) + } + _ => { + let value = unsafe { MEM_readLE32(ptr.cast::()) }; + hash_shift32(value.wrapping_mul(2_654_435_761), hbits) + } + } +} + +#[inline] +unsafe fn hash_ptr_salted(ptr: *const u8, hbits: u32, mls: u32, salt: u64) -> usize { + match mls { + 5 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64( + value.wrapping_shl(24).wrapping_mul(889_523_592_379) ^ salt, + hbits, + ) + } + 6 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64( + value.wrapping_shl(16).wrapping_mul(227_718_039_650_203) ^ salt, + hbits, + ) + } + 7 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64( + value.wrapping_shl(8).wrapping_mul(58_295_818_150_454_627) ^ salt, + hbits, + ) + } + 8 => { + let value = unsafe { MEM_readLE64(ptr.cast::()) }; + hash_shift64(value.wrapping_mul(0xCF1B_BCDC_B7A5_6463) ^ salt, hbits) + } + _ => { + let value = unsafe { MEM_readLE32(ptr.cast::()) }; + hash_shift32(value.wrapping_mul(2_654_435_761) ^ salt as u32, hbits) + } + } +} + +#[inline] +fn common_bytes(word: usize) -> usize { + let zeros = if MEM_isLittleEndian() { + word.trailing_zeros() + } else { + word.leading_zeros() + }; + (zeros / 8) as usize +} + +/// Equivalent to C's `ZSTD_count()`. +unsafe fn count(mut input: *const u8, mut matched: *const u8, input_limit: *const u8) -> usize { + let input_start = input; + let word_size = size_of::(); + + while (input_limit as usize).wrapping_sub(input as usize) >= word_size { + let diff = + unsafe { MEM_readST(matched.cast::()) ^ MEM_readST(input.cast::()) }; + if diff != 0 { + return (input as usize).wrapping_sub(input_start as usize) + common_bytes(diff); + } + input = input.wrapping_add(word_size); + matched = matched.wrapping_add(word_size); + } + + if MEM_64bits() + && (input_limit as usize).wrapping_sub(input as usize) >= 4 + && unsafe { MEM_read32(matched.cast::()) == MEM_read32(input.cast::()) } + { + input = input.wrapping_add(4); + matched = matched.wrapping_add(4); + } + if (input_limit as usize).wrapping_sub(input as usize) >= 2 + && unsafe { MEM_read16(matched.cast::()) == MEM_read16(input.cast::()) } + { + input = input.wrapping_add(2); + matched = matched.wrapping_add(2); + } + if ptr_lt(input, input_limit) && unsafe { *matched == *input } { + input = input.wrapping_add(1); + } + (input as usize).wrapping_sub(input_start as usize) +} + +unsafe fn count_2segments( + input: *const u8, + matched: *const u8, + input_end: *const u8, + match_end: *const u8, + input_start: *const u8, +) -> usize { + let match_remaining = (match_end as usize).wrapping_sub(matched as usize); + let input_remaining = (input_end as usize).wrapping_sub(input as usize); + let first_end = input.wrapping_add(match_remaining.min(input_remaining)); + let first_count = unsafe { count(input, matched, first_end) }; + if matched.wrapping_add(first_count) != match_end { + return first_count; + } + first_count + unsafe { count(input.wrapping_add(first_count), input_start, input_end) } +} + +#[inline] +fn lowest_prefix_index(dict_limit: u32, loaded_dict_end: u32, curr: u32, window_log: u32) -> u32 { + let max_distance = 1u32.wrapping_shl(window_log); + let within_window = if curr.wrapping_sub(dict_limit) > max_distance { + curr.wrapping_sub(max_distance) + } else { + dict_limit + }; + if loaded_dict_end != 0 { + dict_limit + } else { + within_window + } +} + +#[inline] +fn lowest_match_index(low_limit: u32, loaded_dict_end: u32, curr: u32, window_log: u32) -> u32 { + let max_distance = 1u32.wrapping_shl(window_log); + let within_window = if curr.wrapping_sub(low_limit) > max_distance { + curr.wrapping_sub(max_distance) + } else { + low_limit + }; + if loaded_dict_end != 0 { + low_limit + } else { + within_window + } +} + +#[inline] +fn index_overlap_check(prefix_lowest_index: u32, rep_index: u32) -> bool { + prefix_lowest_index.wrapping_sub(1).wrapping_sub(rep_index) >= 3 +} + +/// Stores exactly the observable bytes that C's `ZSTD_storeSeq()` writes. +unsafe fn store_seq( + seq_store: *mut SeqStore_t, + lit_length: usize, + literals: *const u8, + _lit_limit: *const u8, + off_base: u32, + match_length: usize, +) { + let seq_store = unsafe { &mut *seq_store }; + let sequence = seq_store.sequences; + debug_assert!( + (sequence as usize).wrapping_sub(seq_store.sequencesStart as usize) / size_of::() + < seq_store.maxNbSeq + ); + debug_assert!(lit_length <= seq_store.maxNbLit); + debug_assert!(match_length >= MINMATCH); + + if lit_length != 0 { + unsafe { ptr::copy_nonoverlapping(literals, seq_store.lit, lit_length) }; + } + seq_store.lit = seq_store.lit.wrapping_add(lit_length); + + let sequence_index = + (sequence as usize).wrapping_sub(seq_store.sequencesStart as usize) / size_of::(); + if lit_length > u16::MAX as usize { + debug_assert_eq!(seq_store.longLengthType, 0); + seq_store.longLengthType = 1; + seq_store.longLengthPos = sequence_index as u32; + } + unsafe { (*sequence).litLength = lit_length as u16 }; + unsafe { (*sequence).offBase = off_base }; + + let match_base = match_length - MINMATCH; + if match_base > u16::MAX as usize { + debug_assert_eq!(seq_store.longLengthType, 0); + seq_store.longLengthType = 2; + seq_store.longLengthPos = sequence_index as u32; + } + unsafe { (*sequence).mlBase = match_base as u16 }; + seq_store.sequences = sequence.wrapping_add(1); +} + +#[inline] +fn offset_to_offbase(offset: u32) -> u32 { + offset.wrapping_add(OFFSET_OFFBASE) +} + +#[inline] +fn offbase_is_offset(offbase: usize) -> bool { + offbase > OFFSET_OFFBASE as usize +} + +#[inline] +fn offbase_to_offset(offbase: usize) -> u32 { + (offbase as u32).wrapping_sub(OFFSET_OFFBASE) +} + +#[inline] +fn match_improves( + match_length: usize, + best_length: usize, + curr: u32, + match_index: u32, + old_offbase: usize, + old_offbase_plus_one: bool, +) -> bool { + let lhs = (match_length.wrapping_sub(best_length) as i32).wrapping_mul(4); + let new_hb = ZSTD_highbit32(curr.wrapping_sub(match_index).wrapping_add(1)) as i32; + let old_value = (old_offbase as u32).wrapping_add(u32::from(old_offbase_plus_one)); + let old_hb = ZSTD_highbit32(old_value) as i32; + lhs > new_hb.wrapping_sub(old_hb) +} + +/* ------------------------------------------------------------------------- */ +/* Binary tree search */ +/* ------------------------------------------------------------------------- */ + +unsafe fn update_dubt(state: &ZSTD_RustLazyState, ip: *const u8, _iend: *const u8, mls: u32) { + let target = unsafe { index_from(state.base, ip) }; + let mut idx = unsafe { next_to_update(state) }; + let bt_log = state.chainLog - 1; + let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1); + + while idx < target { + let hash = unsafe { hash_ptr(state.base.wrapping_add(idx as usize), state.hashLog, mls) }; + let match_index = unsafe { table_get(state.hashTable, hash) }; + let cell = 2 * (idx & bt_mask) as usize; + unsafe { + table_set(state.hashTable, hash, idx); + table_set(state.chainTable, cell, match_index); + table_set(state.chainTable, cell + 1, ZSTD_DUBT_UNSORTED_MARK); + } + idx = idx.wrapping_add(1); + } + unsafe { set_next_to_update(state, target) }; +} + +unsafe fn insert_dubt1( + state: &ZSTD_RustLazyState, + curr: u32, + input_end: *const u8, + mut nb_compares: u32, + bt_low: u32, + dict_mode: c_int, +) { + let bt_log = state.chainLog - 1; + let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1); + let mut common_smaller = 0usize; + let mut common_larger = 0usize; + let ip = if curr >= state.dictLimit { + state.base.wrapping_add(curr as usize) + } else { + state.dictBase.wrapping_add(curr as usize) + }; + let iend = if curr >= state.dictLimit { + input_end + } else { + state.dictBase.wrapping_add(state.dictLimit as usize) + }; + let dict_end = state.dictBase.wrapping_add(state.dictLimit as usize); + let prefix_start = state.base.wrapping_add(state.dictLimit as usize); + let window_valid = state.lowLimit; + let max_distance = 1u32.wrapping_shl(state.windowLog); + let window_low = if curr.wrapping_sub(window_valid) > max_distance { + curr.wrapping_sub(max_distance) + } else { + window_valid + }; + + let mut smaller_slot = 2 * (curr & bt_mask) as usize; + let mut larger_slot = smaller_slot + 1; + let mut smaller_real = true; + let mut larger_real = true; + let mut match_index = unsafe { table_get(state.chainTable, smaller_slot) }; + + while nb_compares != 0 && match_index > window_low { + let next_cell = 2 * (match_index & bt_mask) as usize; + let mut match_length = common_smaller.min(common_larger); + let mut matched; + + if dict_mode != DICT_EXT || match_index.wrapping_add(match_length as u32) >= state.dictLimit + { + matched = state.base.wrapping_add(match_index as usize); + match_length += unsafe { + count( + ip.wrapping_add(match_length), + matched.wrapping_add(match_length), + iend, + ) + }; + } else { + matched = state.dictBase.wrapping_add(match_index as usize); + match_length += unsafe { + count_2segments( + ip.wrapping_add(match_length), + matched.wrapping_add(match_length), + iend, + dict_end, + prefix_start, + ) + }; + if match_index.wrapping_add(match_length as u32) >= state.dictLimit { + matched = state.base.wrapping_add(match_index as usize); + } + } + + if ip.wrapping_add(match_length) == iend { + break; + } + + if unsafe { *matched.wrapping_add(match_length) < *ip.wrapping_add(match_length) } { + if smaller_real { + unsafe { table_set(state.chainTable, smaller_slot, match_index) }; + } + common_smaller = match_length; + if match_index <= bt_low { + smaller_real = false; + break; + } + smaller_slot = next_cell + 1; + match_index = unsafe { table_get(state.chainTable, next_cell + 1) }; + } else { + if larger_real { + unsafe { table_set(state.chainTable, larger_slot, match_index) }; + } + common_larger = match_length; + if match_index <= bt_low { + larger_real = false; + break; + } + larger_slot = next_cell; + match_index = unsafe { table_get(state.chainTable, next_cell) }; + } + nb_compares = nb_compares.wrapping_sub(1); + } + + if smaller_real { + unsafe { table_set(state.chainTable, smaller_slot, 0) }; + } + if larger_real { + unsafe { table_set(state.chainTable, larger_slot, 0) }; + } +} + +unsafe fn dubt_find_better_dict_match( + state: &ZSTD_RustLazyState, + ip: *const u8, + iend: *const u8, + offbase: &mut usize, + mut best_length: usize, + mut nb_compares: u32, + mls: u32, +) -> usize { + let dms = unsafe { dict_state(state) }; + let dict_hash = unsafe { hash_ptr(ip, dms.hashLog, mls) }; + let mut dict_match_index = unsafe { table_get(dms.hashTable, dict_hash) }; + let prefix_start = state.base.wrapping_add(state.dictLimit as usize); + let curr = unsafe { index_from(state.base, ip) }; + let dict_base = dms.base; + let dict_end = dms.nextSrc; + let dict_high_limit = unsafe { index_from(dms.base, dms.nextSrc) }; + let dict_low_limit = dms.lowLimit; + let dict_index_delta = state.lowLimit.wrapping_sub(dict_high_limit); + let bt_log = dms.chainLog - 1; + let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1); + let bt_low = if bt_mask >= dict_high_limit.wrapping_sub(dict_low_limit) { + dict_low_limit + } else { + dict_high_limit.wrapping_sub(bt_mask) + }; + let mut common_smaller = 0usize; + let mut common_larger = 0usize; + + while nb_compares != 0 && dict_match_index > dict_low_limit { + let next_cell = 2 * (dict_match_index & bt_mask) as usize; + let mut match_length = common_smaller.min(common_larger); + let mut matched = dict_base.wrapping_add(dict_match_index as usize); + match_length += unsafe { + count_2segments( + ip.wrapping_add(match_length), + matched.wrapping_add(match_length), + iend, + dict_end, + prefix_start, + ) + }; + if dict_match_index.wrapping_add(match_length as u32) >= dict_high_limit { + matched = state + .base + .wrapping_add(dict_match_index.wrapping_add(dict_index_delta) as usize); + } + + if match_length > best_length { + let match_index = dict_match_index.wrapping_add(dict_index_delta); + if match_improves(match_length, best_length, curr, match_index, *offbase, true) { + best_length = match_length; + *offbase = offset_to_offbase(curr.wrapping_sub(match_index)) as usize; + } + if ip.wrapping_add(match_length) == iend { + break; + } + } + + if unsafe { *matched.wrapping_add(match_length) < *ip.wrapping_add(match_length) } { + if dict_match_index <= bt_low { + break; + } + common_smaller = match_length; + dict_match_index = unsafe { table_get(dms.chainTable, next_cell + 1) }; + } else { + if dict_match_index <= bt_low { + break; + } + common_larger = match_length; + dict_match_index = unsafe { table_get(dms.chainTable, next_cell) }; + } + nb_compares = nb_compares.wrapping_sub(1); + } + best_length +} + +unsafe fn dubt_find_best_match( + state: &ZSTD_RustLazyState, + ip: *const u8, + iend: *const u8, + offbase: &mut usize, + mls: u32, + dict_mode: c_int, +) -> usize { + let hash = unsafe { hash_ptr(ip, state.hashLog, mls) }; + let mut match_index = unsafe { table_get(state.hashTable, hash) }; + let curr = unsafe { index_from(state.base, ip) }; + let window_low = lowest_match_index(state.lowLimit, state.loadedDictEnd, curr, state.windowLog); + let bt_log = state.chainLog - 1; + let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1); + let bt_low = if bt_mask >= curr { + 0 + } else { + curr.wrapping_sub(bt_mask) + }; + let unsort_limit = bt_low.max(window_low); + let mut next_candidate = 2 * (match_index & bt_mask) as usize; + let mut unsorted_mark = next_candidate + 1; + let mut nb_compares = 1u32.wrapping_shl(state.searchLog); + let mut nb_candidates = nb_compares; + let mut previous_candidate = 0u32; + + while match_index > unsort_limit + && unsafe { table_get(state.chainTable, unsorted_mark) == ZSTD_DUBT_UNSORTED_MARK } + && nb_candidates > 1 + { + unsafe { table_set(state.chainTable, unsorted_mark, previous_candidate) }; + previous_candidate = match_index; + match_index = unsafe { table_get(state.chainTable, next_candidate) }; + next_candidate = 2 * (match_index & bt_mask) as usize; + unsorted_mark = next_candidate + 1; + nb_candidates = nb_candidates.wrapping_sub(1); + } + + if match_index > unsort_limit + && unsafe { table_get(state.chainTable, unsorted_mark) == ZSTD_DUBT_UNSORTED_MARK } + { + unsafe { + table_set(state.chainTable, next_candidate, 0); + table_set(state.chainTable, unsorted_mark, 0); + } + } + + match_index = previous_candidate; + while match_index != 0 { + let next_candidate_idx = + unsafe { table_get(state.chainTable, 2 * (match_index & bt_mask) as usize + 1) }; + unsafe { + insert_dubt1( + state, + match_index, + iend, + nb_candidates, + unsort_limit, + dict_mode, + ) + }; + match_index = next_candidate_idx; + nb_candidates = nb_candidates.wrapping_add(1); + } + + let mut common_smaller = 0usize; + let mut common_larger = 0usize; + let dict_base = state.dictBase; + let dict_limit = state.dictLimit; + let dict_end = dict_base.wrapping_add(dict_limit as usize); + let prefix_start = state.base.wrapping_add(dict_limit as usize); + let mut smaller_slot = 2 * (curr & bt_mask) as usize; + let mut larger_slot = smaller_slot + 1; + let mut smaller_real = true; + let mut larger_real = true; + let mut match_end_idx = curr.wrapping_add(HASH_READ_SIZE as u32 + 1); + let mut best_length = 0usize; + + match_index = unsafe { table_get(state.hashTable, hash) }; + unsafe { table_set(state.hashTable, hash, curr) }; + + while nb_compares != 0 && match_index > window_low { + let next_cell = 2 * (match_index & bt_mask) as usize; + let mut match_length = common_smaller.min(common_larger); + let mut matched; + + if dict_mode != DICT_EXT || match_index.wrapping_add(match_length as u32) >= dict_limit { + matched = state.base.wrapping_add(match_index as usize); + match_length += unsafe { + count( + ip.wrapping_add(match_length), + matched.wrapping_add(match_length), + iend, + ) + }; + } else { + matched = dict_base.wrapping_add(match_index as usize); + match_length += unsafe { + count_2segments( + ip.wrapping_add(match_length), + matched.wrapping_add(match_length), + iend, + dict_end, + prefix_start, + ) + }; + if match_index.wrapping_add(match_length as u32) >= dict_limit { + matched = state.base.wrapping_add(match_index as usize); + } + } + + if match_length > best_length { + if match_length > match_end_idx.wrapping_sub(match_index) as usize { + match_end_idx = match_index.wrapping_add(match_length as u32); + } + if match_improves( + match_length, + best_length, + curr, + match_index, + *offbase, + false, + ) { + best_length = match_length; + *offbase = offset_to_offbase(curr.wrapping_sub(match_index)) as usize; + } + if ip.wrapping_add(match_length) == iend { + if dict_mode == DICT_MATCH_STATE { + nb_compares = 0; + } + break; + } + } + + if unsafe { *matched.wrapping_add(match_length) < *ip.wrapping_add(match_length) } { + if smaller_real { + unsafe { table_set(state.chainTable, smaller_slot, match_index) }; + } + common_smaller = match_length; + if match_index <= bt_low { + smaller_real = false; + break; + } + smaller_slot = next_cell + 1; + match_index = unsafe { table_get(state.chainTable, next_cell + 1) }; + } else { + if larger_real { + unsafe { table_set(state.chainTable, larger_slot, match_index) }; + } + common_larger = match_length; + if match_index <= bt_low { + larger_real = false; + break; + } + larger_slot = next_cell; + match_index = unsafe { table_get(state.chainTable, next_cell) }; + } + nb_compares = nb_compares.wrapping_sub(1); + } + + if smaller_real { + unsafe { table_set(state.chainTable, smaller_slot, 0) }; + } + if larger_real { + unsafe { table_set(state.chainTable, larger_slot, 0) }; + } + + if dict_mode == DICT_MATCH_STATE && nb_compares != 0 { + best_length = unsafe { + dubt_find_better_dict_match(state, ip, iend, offbase, best_length, nb_compares, mls) + }; + } + + unsafe { set_next_to_update(state, match_end_idx.wrapping_sub(HASH_READ_SIZE as u32)) }; + best_length +} + +unsafe fn bt_find_best_match( + state: &ZSTD_RustLazyState, + ip: *const u8, + iend: *const u8, + offbase: &mut usize, + mls: u32, + dict_mode: c_int, +) -> usize { + if ptr_lt( + ip, + state + .base + .wrapping_add(unsafe { next_to_update(state) } as usize), + ) { + return 0; + } + unsafe { update_dubt(state, ip, iend, mls) }; + unsafe { dubt_find_best_match(state, ip, iend, offbase, mls, dict_mode) } +} + +/* ------------------------------------------------------------------------- */ +/* Dedicated dictionary search */ +/* ------------------------------------------------------------------------- */ + +unsafe fn dedicated_dict_search_load_dictionary(state: &ZSTD_RustLazyState, ip: *const u8) { + let target = unsafe { index_from(state.base, ip) }; + let chain_size = 1u32.wrapping_shl(state.chainLog); + let mut idx = unsafe { next_to_update(state) }; + let min_chain = if chain_size < target.wrapping_sub(idx) { + target.wrapping_sub(chain_size) + } else { + idx + }; + let bucket_size = 1u32 << ZSTD_LAZY_DDSS_BUCKET_LOG; + let cache_size = bucket_size - 1; + let chain_attempts = (1u32 << state.searchLog).wrapping_sub(cache_size); + let chain_limit = chain_attempts.min(255); + let hash_log = state.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG; + let tmp_hash_table = state.hashTable; + let tmp_chain_table = state + .hashTable + .wrapping_add((1usize).wrapping_shl(hash_log)); + let tmp_chain_size = ((1u32 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1).wrapping_shl(hash_log); + let tmp_min_chain = if tmp_chain_size < target { + target.wrapping_sub(tmp_chain_size) + } else { + idx + }; + + while idx < target { + let hash = unsafe { + hash_ptr( + state.base.wrapping_add(idx as usize), + hash_log, + state.minMatch, + ) + }; + if idx >= tmp_min_chain { + let previous = unsafe { table_get(state.hashTable, hash) }; + unsafe { + table_set( + tmp_chain_table, + idx.wrapping_sub(tmp_min_chain) as usize, + previous, + ) + }; + } + unsafe { table_set(tmp_hash_table, hash, idx) }; + idx = idx.wrapping_add(1); + } + + let mut chain_pos = 0u32; + let mut hash_idx = 0u32; + while hash_idx < (1u32 << hash_log) { + let mut count = 0u32; + let mut count_beyond_min_chain = 0u32; + let mut chain_index = unsafe { table_get(tmp_hash_table, hash_idx as usize) }; + while chain_index >= tmp_min_chain && count < cache_size { + if chain_index < min_chain { + count_beyond_min_chain = count_beyond_min_chain.wrapping_add(1); + } + chain_index = unsafe { + table_get( + tmp_chain_table, + chain_index.wrapping_sub(tmp_min_chain) as usize, + ) + }; + count = count.wrapping_add(1); + } + if count == cache_size { + count = 0; + while count < chain_limit { + if chain_index < min_chain { + if chain_index == 0 || count_beyond_min_chain.wrapping_add(1) > cache_size { + break; + } + count_beyond_min_chain = count_beyond_min_chain.wrapping_add(1); + } + unsafe { table_set(state.chainTable, chain_pos as usize, chain_index) }; + chain_pos = chain_pos.wrapping_add(1); + count = count.wrapping_add(1); + if chain_index < tmp_min_chain { + break; + } + chain_index = unsafe { + table_get( + tmp_chain_table, + chain_index.wrapping_sub(tmp_min_chain) as usize, + ) + }; + } + } else { + count = 0; + } + if count != 0 { + unsafe { + table_set( + tmp_hash_table, + hash_idx as usize, + ((chain_pos.wrapping_sub(count)) << 8).wrapping_add(count), + ) + }; + } else { + unsafe { table_set(tmp_hash_table, hash_idx as usize, 0) }; + } + hash_idx = hash_idx.wrapping_add(1); + } + + hash_idx = 1u32 << hash_log; + while hash_idx != 0 { + hash_idx = hash_idx.wrapping_sub(1); + let bucket_idx = hash_idx << ZSTD_LAZY_DDSS_BUCKET_LOG; + let packed = unsafe { table_get(tmp_hash_table, hash_idx as usize) }; + let mut i = 0u32; + while i < cache_size { + unsafe { table_set(state.hashTable, bucket_idx.wrapping_add(i) as usize, 0) }; + i = i.wrapping_add(1); + } + unsafe { + table_set( + state.hashTable, + bucket_idx.wrapping_add(bucket_size).wrapping_sub(1) as usize, + packed, + ) + }; + } + + idx = unsafe { next_to_update(state) }; + while idx < target { + let hash = unsafe { + hash_ptr( + state.base.wrapping_add(idx as usize), + hash_log, + state.minMatch, + ) + } << ZSTD_LAZY_DDSS_BUCKET_LOG; + let mut i = cache_size.wrapping_sub(1); + while i != 0 { + let previous = + unsafe { table_get(state.hashTable, hash.wrapping_add((i - 1) as usize)) }; + unsafe { table_set(state.hashTable, hash.wrapping_add(i as usize), previous) }; + i = i.wrapping_sub(1); + } + unsafe { table_set(state.hashTable, hash, idx) }; + idx = idx.wrapping_add(1); + } + unsafe { set_next_to_update(state, target) }; +} + +unsafe fn dedicated_dict_search( + offset: &mut usize, + mut ml: usize, + nb_attempts: u32, + dms: &ZSTD_RustLazyState, + ip: *const u8, + ilimit: *const u8, + prefix_start: *const u8, + curr: u32, + dict_limit: u32, + dds_idx: usize, +) -> usize { + let dds_lowest_index = dms.dictLimit; + let dds_base = dms.base; + let dds_end = dms.nextSrc; + let dds_size = unsafe { index_from(dds_base, dds_end) }; + let dds_index_delta = dict_limit.wrapping_sub(dds_size); + let bucket_size = 1u32 << ZSTD_LAZY_DDSS_BUCKET_LOG; + let bucket_limit = nb_attempts.min(bucket_size - 1); + let mut dds_attempt = 0u32; + + while dds_attempt < bucket_limit { + let match_index = unsafe { table_get(dms.hashTable, dds_idx + dds_attempt as usize) }; + if match_index == 0 { + return ml; + } + debug_assert!(match_index >= dds_lowest_index); + let matched = dds_base.wrapping_add(match_index as usize); + let mut current_ml = 0usize; + if unsafe { read32(matched) == read32(ip) } { + current_ml = unsafe { + count_2segments( + ip.wrapping_add(4), + matched.wrapping_add(4), + ilimit, + dds_end, + prefix_start, + ) + 4 + }; + } + if current_ml > ml { + ml = current_ml; + *offset = + offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dds_index_delta))) + as usize; + if ip.wrapping_add(current_ml) == ilimit { + return ml; + } + } + dds_attempt = dds_attempt.wrapping_add(1); + } + + let packed = unsafe { table_get(dms.hashTable, dds_idx + (bucket_size - 1) as usize) }; + let mut chain_index = packed >> 8; + let chain_length = packed & 0xff; + let chain_attempts = nb_attempts.wrapping_sub(dds_attempt); + let chain_limit = chain_attempts.min(chain_length); + let mut chain_attempt = 0u32; + while chain_attempt < chain_limit { + let match_index = unsafe { table_get(dms.chainTable, chain_index as usize) }; + let matched = dds_base.wrapping_add(match_index as usize); + let mut current_ml = 0usize; + if unsafe { read32(matched) == read32(ip) } { + current_ml = unsafe { + count_2segments( + ip.wrapping_add(4), + matched.wrapping_add(4), + ilimit, + dds_end, + prefix_start, + ) + 4 + }; + } + if current_ml > ml { + ml = current_ml; + *offset = + offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dds_index_delta))) + as usize; + if ip.wrapping_add(current_ml) == ilimit { + break; + } + } + chain_index = chain_index.wrapping_add(1); + chain_attempt = chain_attempt.wrapping_add(1); + } + ml +} + +/* ------------------------------------------------------------------------- */ +/* Hash-chain search */ +/* ------------------------------------------------------------------------- */ + +unsafe fn insert_and_find_first_index_internal( + state: &ZSTD_RustLazyState, + ip: *const u8, + mls: u32, + lazy_skip: bool, +) -> u32 { + let chain_mask = 1u32.wrapping_shl(state.chainLog).wrapping_sub(1); + let target = unsafe { index_from(state.base, ip) }; + let mut idx = unsafe { next_to_update(state) }; + + while idx < target { + let hash = unsafe { hash_ptr(state.base.wrapping_add(idx as usize), state.hashLog, mls) }; + let previous = unsafe { table_get(state.hashTable, hash) }; + unsafe { + table_set(state.chainTable, (idx & chain_mask) as usize, previous); + table_set(state.hashTable, hash, idx); + } + idx = idx.wrapping_add(1); + if lazy_skip { + break; + } + } + unsafe { set_next_to_update(state, target) }; + let hash = unsafe { hash_ptr(ip, state.hashLog, mls) }; + unsafe { table_get(state.hashTable, hash) } +} + +unsafe fn hc_find_best_match( + state: &ZSTD_RustLazyState, + ip: *const u8, + ilimit: *const u8, + offset: &mut usize, + mls: u32, + dict_mode: c_int, +) -> usize { + let chain_size = 1u32.wrapping_shl(state.chainLog); + let chain_mask = chain_size.wrapping_sub(1); + let base = state.base; + let dict_base = state.dictBase; + let dict_limit = state.dictLimit; + let prefix_start = base.wrapping_add(dict_limit as usize); + let dict_end = dict_base.wrapping_add(dict_limit as usize); + let curr = unsafe { index_from(base, ip) }; + let max_distance = 1u32.wrapping_shl(state.windowLog); + let lowest_valid = state.lowLimit; + let within_max_distance = if curr.wrapping_sub(lowest_valid) > max_distance { + curr.wrapping_sub(max_distance) + } else { + lowest_valid + }; + let low_limit = if state.loadedDictEnd != 0 { + lowest_valid + } else { + within_max_distance + }; + let min_chain = if curr > chain_size { + curr.wrapping_sub(chain_size) + } else { + 0 + }; + let mut nb_attempts = 1u32.wrapping_shl(state.searchLog); + let mut ml = 3usize; + + let dds_hash_log = if dict_mode == DICT_DEDICATED { + unsafe { dict_state(state) }.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG + } else { + 0 + }; + let dds_idx = if dict_mode == DICT_DEDICATED { + (unsafe { hash_ptr(ip, dds_hash_log, mls) }) << ZSTD_LAZY_DDSS_BUCKET_LOG + } else { + 0 + }; + + let mut match_index = + unsafe { insert_and_find_first_index_internal(state, ip, mls, lazy_skipping(state)) }; + while match_index >= low_limit && nb_attempts != 0 { + let mut current_ml = 0usize; + if dict_mode != DICT_EXT || match_index >= dict_limit { + let matched = base.wrapping_add(match_index as usize); + if unsafe { read32(matched.wrapping_add(ml - 3)) == read32(ip.wrapping_add(ml - 3)) } { + current_ml = unsafe { count(ip, matched, ilimit) }; + } + } else { + let matched = dict_base.wrapping_add(match_index as usize); + if unsafe { read32(matched) == read32(ip) } { + current_ml = unsafe { + count_2segments( + ip.wrapping_add(4), + matched.wrapping_add(4), + ilimit, + dict_end, + prefix_start, + ) + 4 + }; + } + } + + if current_ml > ml { + ml = current_ml; + *offset = offset_to_offbase(curr.wrapping_sub(match_index)) as usize; + if ip.wrapping_add(current_ml) == ilimit { + break; + } + } + if match_index <= min_chain { + break; + } + match_index = unsafe { table_get(state.chainTable, (match_index & chain_mask) as usize) }; + nb_attempts = nb_attempts.wrapping_sub(1); + } + + if dict_mode == DICT_DEDICATED { + ml = unsafe { + dedicated_dict_search( + offset, + ml, + nb_attempts, + dict_state(state), + ip, + ilimit, + prefix_start, + curr, + dict_limit, + dds_idx, + ) + }; + } else if dict_mode == DICT_MATCH_STATE { + let dms = unsafe { dict_state(state) }; + let dms_chain_size = 1u32.wrapping_shl(dms.chainLog); + let dms_chain_mask = dms_chain_size.wrapping_sub(1); + let dms_lowest_index = dms.dictLimit; + let dms_base = dms.base; + let dms_end = dms.nextSrc; + let dms_size = unsafe { index_from(dms_base, dms_end) }; + let dms_index_delta = dict_limit.wrapping_sub(dms_size); + let dms_min_chain = if dms_size > dms_chain_size { + dms_size.wrapping_sub(dms_chain_size) + } else { + 0 + }; + + match_index = unsafe { table_get(dms.hashTable, hash_ptr(ip, dms.hashLog, mls)) }; + while match_index >= dms_lowest_index && nb_attempts != 0 { + let matched = dms_base.wrapping_add(match_index as usize); + let mut current_ml = 0usize; + if unsafe { read32(matched) == read32(ip) } { + current_ml = unsafe { + count_2segments( + ip.wrapping_add(4), + matched.wrapping_add(4), + ilimit, + dms_end, + prefix_start, + ) + 4 + }; + } + if current_ml > ml { + ml = current_ml; + *offset = + offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dms_index_delta))) + as usize; + if ip.wrapping_add(current_ml) == ilimit { + break; + } + } + if match_index <= dms_min_chain { + break; + } + match_index = + unsafe { table_get(dms.chainTable, (match_index & dms_chain_mask) as usize) }; + nb_attempts = nb_attempts.wrapping_sub(1); + } + } + ml +} + +/* ------------------------------------------------------------------------- */ +/* Row hash search */ +/* ------------------------------------------------------------------------- */ + +#[inline] +unsafe fn row_next_index(tag_row: *mut u8, row_mask: u32) -> u32 { + let mut next = unsafe { ((*tag_row).wrapping_sub(1) as u32) & row_mask }; + if next == 0 { + next = next.wrapping_add(row_mask); + } + unsafe { *tag_row = next as u8 }; + next +} + +unsafe fn row_fill_hash_cache( + state: &ZSTD_RustLazyState, + base: *const u8, + _row_log: u32, + mls: u32, + mut idx: u32, + ilimit: *const u8, +) { + let hash_log = state.rowHashLog; + let at = base.wrapping_add(idx as usize); + let max_to_fill = if ptr_gt(at, ilimit) { + 0 + } else { + ((ilimit as usize).wrapping_sub(at as usize)).wrapping_add(1) as u32 + }; + let limit = idx.wrapping_add((ZSTD_ROW_HASH_CACHE_SIZE as u32).min(max_to_fill)); + while idx < limit { + let hash = unsafe { + hash_ptr_salted( + base.wrapping_add(idx as usize), + hash_log + ZSTD_ROW_HASH_TAG_BITS, + mls, + state.hashSalt, + ) as u32 + }; + unsafe { + table_set( + state.hashCache, + (idx as usize) & (ZSTD_ROW_HASH_CACHE_SIZE - 1), + hash, + ) + }; + idx = idx.wrapping_add(1); + } +} + +unsafe fn row_next_cached_hash( + cache: *mut u32, + _hash_table: *const u32, + _tag_table: *const u8, + base: *const u8, + idx: u32, + hash_log: u32, + _row_log: u32, + mls: u32, + hash_salt: u64, +) -> u32 { + let new_hash = unsafe { + hash_ptr_salted( + base.wrapping_add(idx as usize + ZSTD_ROW_HASH_CACHE_SIZE), + hash_log + ZSTD_ROW_HASH_TAG_BITS, + mls, + hash_salt, + ) as u32 + }; + let slot = (idx as usize) & (ZSTD_ROW_HASH_CACHE_SIZE - 1); + let hash = unsafe { table_get(cache, slot) }; + unsafe { table_set(cache, slot, new_hash) }; + hash +} + +unsafe fn row_update_internal_impl( + state: &ZSTD_RustLazyState, + mut update_start: u32, + update_end: u32, + mls: u32, + row_log: u32, + row_mask: u32, + use_cache: bool, +) { + while update_start < update_end { + let hash = if use_cache { + unsafe { + row_next_cached_hash( + state.hashCache, + state.hashTable, + state.tagTable, + state.base, + update_start, + state.rowHashLog, + row_log, + mls, + state.hashSalt, + ) + } + } else { + unsafe { + hash_ptr_salted( + state.base.wrapping_add(update_start as usize), + state.rowHashLog + ZSTD_ROW_HASH_TAG_BITS, + mls, + state.hashSalt, + ) as u32 + } + }; + let rel_row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << row_log; + let row = state.hashTable.wrapping_add(rel_row as usize); + let tag_row = state.tagTable.wrapping_add(rel_row as usize); + let pos = unsafe { row_next_index(tag_row, row_mask) }; + unsafe { + byte_set(tag_row, pos as usize, (hash & ZSTD_ROW_HASH_TAG_MASK) as u8); + table_set(row, pos as usize, update_start); + } + update_start = update_start.wrapping_add(1); + } +} + +unsafe fn row_update_internal( + state: &ZSTD_RustLazyState, + ip: *const u8, + mls: u32, + row_log: u32, + row_mask: u32, + use_cache: bool, +) { + let mut idx = unsafe { next_to_update(state) }; + let target = unsafe { index_from(state.base, ip) }; + const SKIP_THRESHOLD: u32 = 384; + const MAX_START: u32 = 96; + const MAX_END: u32 = 32; + + if use_cache && target.wrapping_sub(idx) > SKIP_THRESHOLD { + let bound = idx.wrapping_add(MAX_START); + unsafe { row_update_internal_impl(state, idx, bound, mls, row_log, row_mask, use_cache) }; + idx = target.wrapping_sub(MAX_END); + unsafe { row_fill_hash_cache(state, state.base, row_log, mls, idx, ip.wrapping_add(1)) }; + } + unsafe { row_update_internal_impl(state, idx, target, mls, row_log, row_mask, use_cache) }; + unsafe { set_next_to_update(state, target) }; +} + +/// Returns matching tag positions in the same circular-buffer order as the +/// C SSE/SWAR mask iterator. The scalar version avoids architecture-specific +/// vector intrinsics while preserving the selected candidate order. +unsafe fn row_matching_positions( + tag_row: *const u8, + tag: u8, + head: u32, + row_mask: u32, + row_entries: u32, + out: &mut [u32; ZSTD_ROW_HASH_MAX_ENTRIES], +) -> usize { + let mut len = 0usize; + let mut relative = 0u32; + while relative < row_entries { + let pos = head.wrapping_add(relative) & row_mask; + if unsafe { byte_get(tag_row, pos as usize) } == tag { + out[len] = pos; + len += 1; + } + relative = relative.wrapping_add(1); + } + len +} + +unsafe fn row_find_best_match( + state: &ZSTD_RustLazyState, + ip: *const u8, + ilimit: *const u8, + offset: &mut usize, + mls: u32, + dict_mode: c_int, + row_log: u32, +) -> usize { + let base = state.base; + let dict_base = state.dictBase; + let dict_limit = state.dictLimit; + let prefix_start = base.wrapping_add(dict_limit as usize); + let dict_end = dict_base.wrapping_add(dict_limit as usize); + let curr = unsafe { index_from(base, ip) }; + let max_distance = 1u32.wrapping_shl(state.windowLog); + let lowest_valid = state.lowLimit; + let within_max_distance = if curr.wrapping_sub(lowest_valid) > max_distance { + curr.wrapping_sub(max_distance) + } else { + lowest_valid + }; + let low_limit = if state.loadedDictEnd != 0 { + lowest_valid + } else { + within_max_distance + }; + let row_entries = 1u32 << row_log; + let row_mask = row_entries - 1; + let capped_search_log = state.searchLog.min(row_log); + let mut nb_attempts = 1u32 << capped_search_log; + let mut ml = 3usize; + + let (mut dds_idx, mut dds_extra_attempts) = (0usize, 0u32); + if dict_mode == DICT_DEDICATED { + let dms = unsafe { dict_state(state) }; + let dds_hash_log = dms.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG; + dds_idx = unsafe { hash_ptr(ip, dds_hash_log, mls) } << ZSTD_LAZY_DDSS_BUCKET_LOG; + dds_extra_attempts = if state.searchLog > row_log { + 1u32 << (state.searchLog - row_log) + } else { + 0 + }; + } + + let mut dms_tag = 0u32; + let mut dms_row = ptr::null_mut::(); + let mut dms_tag_row = ptr::null_mut::(); + if dict_mode == DICT_MATCH_STATE { + let dms = unsafe { dict_state(state) }; + let dms_hash = unsafe { hash_ptr(ip, dms.rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls) as u32 }; + let dms_rel_row = (dms_hash >> ZSTD_ROW_HASH_TAG_BITS) << row_log; + dms_tag = dms_hash & ZSTD_ROW_HASH_TAG_MASK; + dms_tag_row = dms.tagTable.wrapping_add(dms_rel_row as usize); + dms_row = dms.hashTable.wrapping_add(dms_rel_row as usize); + } + + let hash = if !unsafe { lazy_skipping(state) } { + unsafe { row_update_internal(state, ip, mls, row_log, row_mask, true) }; + unsafe { + row_next_cached_hash( + state.hashCache, + state.hashTable, + state.tagTable, + base, + curr, + state.rowHashLog, + row_log, + mls, + state.hashSalt, + ) + } + } else { + let h = unsafe { + hash_ptr_salted( + ip, + state.rowHashLog + ZSTD_ROW_HASH_TAG_BITS, + mls, + state.hashSalt, + ) as u32 + }; + unsafe { set_next_to_update(state, curr) }; + h + }; + unsafe { add_hash_salt_entropy(state, hash) }; + + let rel_row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << row_log; + let tag = (hash & ZSTD_ROW_HASH_TAG_MASK) as u8; + let row = state.hashTable.wrapping_add(rel_row as usize); + let tag_row = state.tagTable.wrapping_add(rel_row as usize); + let head = unsafe { byte_get(tag_row, 0) as u32 } & row_mask; + let mut positions = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES]; + let position_len = unsafe { + row_matching_positions(tag_row, tag, head, row_mask, row_entries, &mut positions) + }; + let mut match_buffer = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES]; + let mut num_matches = 0usize; + for pos in positions[..position_len].iter().copied() { + if nb_attempts == 0 { + break; + } + if pos == 0 { + continue; + } + let match_index = unsafe { table_get(row, pos as usize) }; + if match_index < low_limit { + break; + } + match_buffer[num_matches] = match_index; + num_matches += 1; + nb_attempts = nb_attempts.wrapping_sub(1); + } + + let insert_pos = unsafe { row_next_index(tag_row, row_mask) }; + unsafe { + byte_set(tag_row, insert_pos as usize, tag); + let index = next_to_update(state); + table_set(row, insert_pos as usize, index); + set_next_to_update(state, index.wrapping_add(1)); + } + + for match_index in match_buffer[..num_matches].iter().copied() { + let mut current_ml = 0usize; + if dict_mode != DICT_EXT || match_index >= dict_limit { + let matched = base.wrapping_add(match_index as usize); + if unsafe { read32(matched.wrapping_add(ml - 3)) == read32(ip.wrapping_add(ml - 3)) } { + current_ml = unsafe { count(ip, matched, ilimit) }; + } + } else { + let matched = dict_base.wrapping_add(match_index as usize); + if unsafe { read32(matched) == read32(ip) } { + current_ml = unsafe { + count_2segments( + ip.wrapping_add(4), + matched.wrapping_add(4), + ilimit, + dict_end, + prefix_start, + ) + 4 + }; + } + } + if current_ml > ml { + ml = current_ml; + *offset = offset_to_offbase(curr.wrapping_sub(match_index)) as usize; + if ip.wrapping_add(current_ml) == ilimit { + break; + } + } + } + + if dict_mode == DICT_DEDICATED { + ml = unsafe { + dedicated_dict_search( + offset, + ml, + nb_attempts.wrapping_add(dds_extra_attempts), + dict_state(state), + ip, + ilimit, + prefix_start, + curr, + dict_limit, + dds_idx, + ) + }; + } else if dict_mode == DICT_MATCH_STATE { + let dms = unsafe { dict_state(state) }; + let dms_lowest_index = dms.dictLimit; + let dms_base = dms.base; + let dms_end = dms.nextSrc; + let dms_size = unsafe { index_from(dms_base, dms_end) }; + let dms_index_delta = dict_limit.wrapping_sub(dms_size); + let dms_head = unsafe { byte_get(dms_tag_row, 0) as u32 } & row_mask; + let mut dms_positions = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES]; + let dms_position_len = unsafe { + row_matching_positions( + dms_tag_row, + dms_tag as u8, + dms_head, + row_mask, + row_entries, + &mut dms_positions, + ) + }; + let mut dms_matches = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES]; + let mut dms_count = 0usize; + for pos in dms_positions[..dms_position_len].iter().copied() { + if nb_attempts == 0 { + break; + } + if pos == 0 { + continue; + } + let match_index = unsafe { table_get(dms_row, pos as usize) }; + if match_index < dms_lowest_index { + break; + } + dms_matches[dms_count] = match_index; + dms_count += 1; + nb_attempts = nb_attempts.wrapping_sub(1); + } + for match_index in dms_matches[..dms_count].iter().copied() { + let matched = dms_base.wrapping_add(match_index as usize); + let mut current_ml = 0usize; + if unsafe { read32(matched) == read32(ip) } { + current_ml = unsafe { + count_2segments( + ip.wrapping_add(4), + matched.wrapping_add(4), + ilimit, + dms_end, + prefix_start, + ) + 4 + }; + } + if current_ml > ml { + ml = current_ml; + *offset = + offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dms_index_delta))) + as usize; + if ip.wrapping_add(current_ml) == ilimit { + break; + } + } + } + } + ml +} + +unsafe fn search_max( + state: &ZSTD_RustLazyState, + ip: *const u8, + iend: *const u8, + offset: &mut usize, + mls: u32, + row_log: u32, + search_method: c_int, + dict_mode: c_int, +) -> usize { + match search_method { + SEARCH_HASH_CHAIN => unsafe { hc_find_best_match(state, ip, iend, offset, mls, dict_mode) }, + SEARCH_BINARY_TREE => unsafe { + bt_find_best_match(state, ip, iend, offset, mls, dict_mode) + }, + SEARCH_ROW_HASH => unsafe { + row_find_best_match(state, ip, iend, offset, mls, dict_mode, row_log) + }, + _ => 0, + } +} + +#[inline] +fn score(length: usize, multiplier: usize, offbase: usize, adjustment: usize) -> i32 { + let raw = length + .wrapping_mul(multiplier) + .wrapping_sub(ZSTD_highbit32(offbase as u32) as usize) + .wrapping_add(adjustment); + raw as i32 +} + +/* ------------------------------------------------------------------------- */ +/* Lazy parser for no dictionary, dictionary match state, and DDS */ +/* ------------------------------------------------------------------------- */ + +unsafe fn compress_block_lazy_generic( + state: &ZSTD_RustLazyState, + seq_store: *mut SeqStore_t, + reps: *mut u32, + src: *const u8, + src_size: usize, + search_method: c_int, + depth: u32, + dict_mode: c_int, +) -> usize { + let istart = src; + let mut ip = istart; + let mut anchor = istart; + let iend = istart.wrapping_add(src_size); + let required = if search_method == SEARCH_ROW_HASH { + HASH_READ_SIZE + ZSTD_ROW_HASH_CACHE_SIZE + } else { + HASH_READ_SIZE + }; + if src_size < required { + return src_size; + } + let ilimit = iend.wrapping_sub(required); + let base = state.base; + let prefix_lowest_index = state.dictLimit; + let prefix_lowest = base.wrapping_add(prefix_lowest_index as usize); + let mls = bounded(4, state.minMatch, 6); + let row_log = bounded(4, state.searchLog, 6); + + let mut offset_1 = unsafe { *reps }; + let mut offset_2 = unsafe { *reps.add(1) }; + let mut offset_saved_1 = 0u32; + let mut offset_saved_2 = 0u32; + + let is_dms = dict_mode == DICT_MATCH_STATE; + let is_dds = dict_mode == DICT_DEDICATED; + let is_dxs = is_dms || is_dds; + let dms = if is_dxs { + Some(unsafe { dict_state(state) }) + } else { + None + }; + let dict_lowest_index = dms.map_or(0, |value| value.dictLimit); + let dict_base = dms.map_or(ptr::null(), |value| value.base); + let dict_lowest = if is_dxs { + dict_base.wrapping_add(dict_lowest_index as usize) + } else { + ptr::null() + }; + let dict_end = dms.map_or(ptr::null(), |value| value.nextSrc); + let dict_index_delta = if is_dxs { + prefix_lowest_index.wrapping_sub(unsafe { index_from(dict_base, dict_end) }) + } else { + 0 + }; + let dict_and_prefix_length = if is_dxs { + (ip as usize) + .wrapping_sub(prefix_lowest as usize) + .wrapping_add((dict_end as usize).wrapping_sub(dict_lowest as usize)) as u32 + } else { + (ip as usize).wrapping_sub(prefix_lowest as usize) as u32 + }; + + if dict_and_prefix_length == 0 { + ip = ip.wrapping_add(1); + } + if dict_mode == DICT_NO_DICT { + let curr = unsafe { index_from(base, ip) }; + let window_low = + lowest_prefix_index(state.dictLimit, state.loadedDictEnd, curr, state.windowLog); + let max_rep = curr.wrapping_sub(window_low); + if offset_2 > max_rep { + offset_saved_2 = offset_2; + offset_2 = 0; + } + if offset_1 > max_rep { + offset_saved_1 = offset_1; + offset_1 = 0; + } + } + + unsafe { set_lazy_skipping(state, false) }; + if search_method == SEARCH_ROW_HASH { + unsafe { row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit) }; + } + + while ptr_lt(ip, ilimit) { + let mut match_length = 0usize; + let mut offbase = REPCODE1_TO_OFFBASE as usize; + let mut start = ip.wrapping_add(1); + let mut store_directly = false; + + if is_dxs { + let rep_index = unsafe { index_from(base, ip) } + .wrapping_add(1) + .wrapping_sub(offset_1); + let rep_match = if rep_index < prefix_lowest_index { + dict_base.wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize) + } else { + base.wrapping_add(rep_index as usize) + }; + if index_overlap_check(prefix_lowest_index, rep_index) + && unsafe { read32(rep_match) == read32(ip.wrapping_add(1)) } + { + let rep_end = if rep_index < prefix_lowest_index { + dict_end + } else { + iend + }; + match_length = unsafe { + count_2segments( + ip.wrapping_add(5), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_lowest, + ) + 4 + }; + if depth == 0 { + store_directly = true; + } + } + } + if dict_mode == DICT_NO_DICT + && offset_1 > 0 + && unsafe { + read32(ip.wrapping_add(1).wrapping_sub(offset_1 as usize)) + == read32(ip.wrapping_add(1)) + } + { + match_length = unsafe { + count( + ip.wrapping_add(5), + ip.wrapping_add(5).wrapping_sub(offset_1 as usize), + iend, + ) + 4 + }; + if depth == 0 { + store_directly = true; + } + } + + if !store_directly { + let mut candidate = 999_999_999usize; + let found = unsafe { + search_max( + state, + ip, + iend, + &mut candidate, + mls, + row_log, + search_method, + dict_mode, + ) + }; + if found > match_length { + match_length = found; + start = ip; + offbase = candidate; + } + + if match_length < 4 { + let step = (ip as usize).wrapping_sub(anchor as usize) >> K_SEARCH_STRENGTH; + ip = ip.wrapping_add(step.wrapping_add(1)); + unsafe { set_lazy_skipping(state, step.wrapping_add(1) > K_LAZY_SKIPPING_STEP) }; + continue; + } + + if depth >= 1 { + while ptr_lt(ip, ilimit) { + ip = ip.wrapping_add(1); + if dict_mode == DICT_NO_DICT + && offbase != 0 + && offset_1 > 0 + && unsafe { read32(ip) == read32(ip.wrapping_sub(offset_1 as usize)) } + { + let ml_rep = unsafe { + count( + ip.wrapping_add(4), + ip.wrapping_add(4).wrapping_sub(offset_1 as usize), + iend, + ) + 4 + }; + if ml_rep >= 4 + && score(ml_rep, 3, REPCODE1_TO_OFFBASE as usize, 0) + > score(match_length, 3, offbase, 1) + { + match_length = ml_rep; + offbase = REPCODE1_TO_OFFBASE as usize; + start = ip; + } + } + if is_dxs { + let rep_index = unsafe { index_from(base, ip) }.wrapping_sub(offset_1); + let rep_match = if rep_index < prefix_lowest_index { + dict_base + .wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize) + } else { + base.wrapping_add(rep_index as usize) + }; + if index_overlap_check(prefix_lowest_index, rep_index) + && unsafe { read32(rep_match) == read32(ip) } + { + let rep_end = if rep_index < prefix_lowest_index { + dict_end + } else { + iend + }; + let ml_rep = unsafe { + count_2segments( + ip.wrapping_add(4), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_lowest, + ) + 4 + }; + if ml_rep >= 4 + && score(ml_rep, 3, REPCODE1_TO_OFFBASE as usize, 0) + > score(match_length, 3, offbase, 1) + { + match_length = ml_rep; + offbase = REPCODE1_TO_OFFBASE as usize; + start = ip; + } + } + } + + let mut candidate = 999_999_999usize; + let found = unsafe { + search_max( + state, + ip, + iend, + &mut candidate, + mls, + row_log, + search_method, + dict_mode, + ) + }; + if found >= 4 + && score(found, 4, candidate, 0) > score(match_length, 4, offbase, 4) + { + match_length = found; + offbase = candidate; + start = ip; + continue; + } + + if depth == 2 && ptr_lt(ip, ilimit) { + ip = ip.wrapping_add(1); + if dict_mode == DICT_NO_DICT + && offbase != 0 + && offset_1 > 0 + && unsafe { read32(ip) == read32(ip.wrapping_sub(offset_1 as usize)) } + { + let ml_rep = unsafe { + count( + ip.wrapping_add(4), + ip.wrapping_add(4).wrapping_sub(offset_1 as usize), + iend, + ) + 4 + }; + if ml_rep >= 4 + && score(ml_rep, 4, REPCODE1_TO_OFFBASE as usize, 0) + > score(match_length, 4, offbase, 1) + { + match_length = ml_rep; + offbase = REPCODE1_TO_OFFBASE as usize; + start = ip; + } + } + if is_dxs { + let rep_index = unsafe { index_from(base, ip) }.wrapping_sub(offset_1); + let rep_match = if rep_index < prefix_lowest_index { + dict_base + .wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize) + } else { + base.wrapping_add(rep_index as usize) + }; + if index_overlap_check(prefix_lowest_index, rep_index) + && unsafe { read32(rep_match) == read32(ip) } + { + let rep_end = if rep_index < prefix_lowest_index { + dict_end + } else { + iend + }; + let ml_rep = unsafe { + count_2segments( + ip.wrapping_add(4), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_lowest, + ) + 4 + }; + if ml_rep >= 4 + && score(ml_rep, 4, REPCODE1_TO_OFFBASE as usize, 0) + > score(match_length, 4, offbase, 1) + { + match_length = ml_rep; + offbase = REPCODE1_TO_OFFBASE as usize; + start = ip; + } + } + } + + let mut second_candidate = 999_999_999usize; + let second_found = unsafe { + search_max( + state, + ip, + iend, + &mut second_candidate, + mls, + row_log, + search_method, + dict_mode, + ) + }; + if second_found >= 4 + && score(second_found, 4, second_candidate, 0) + > score(match_length, 4, offbase, 7) + { + match_length = second_found; + offbase = second_candidate; + start = ip; + continue; + } + } + break; + } + } + + if offbase_is_offset(offbase) { + let offset = offbase_to_offset(offbase); + if dict_mode == DICT_NO_DICT { + while ptr_gt(start, anchor) + && ptr_gt(start.wrapping_sub(offset as usize), prefix_lowest) + && unsafe { + *start.wrapping_sub(1) == *start.wrapping_sub(offset as usize + 1) + } + { + start = start.wrapping_sub(1); + match_length += 1; + } + } + if is_dxs { + let match_index = unsafe { index_from(base, start) }.wrapping_sub(offset); + let mut matched = if match_index < prefix_lowest_index { + dict_base.wrapping_add(match_index.wrapping_sub(dict_index_delta) as usize) + } else { + base.wrapping_add(match_index as usize) + }; + let match_start = if match_index < prefix_lowest_index { + dict_lowest + } else { + prefix_lowest + }; + while ptr_gt(start, anchor) + && ptr_gt(matched, match_start) + && unsafe { *start.wrapping_sub(1) == *matched.wrapping_sub(1) } + { + start = start.wrapping_sub(1); + matched = matched.wrapping_sub(1); + match_length += 1; + } + } + offset_2 = offset_1; + offset_1 = offset; + } + } + + let lit_length = (start as usize).wrapping_sub(anchor as usize); + unsafe { + store_seq( + seq_store, + lit_length, + anchor, + iend, + offbase as u32, + match_length, + ) + }; + anchor = start.wrapping_add(match_length); + ip = anchor; + + if unsafe { lazy_skipping(state) } { + if search_method == SEARCH_ROW_HASH { + unsafe { + row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit) + }; + } + unsafe { set_lazy_skipping(state, false) }; + } + + if is_dxs { + while ptr_le(ip, ilimit) { + let current = unsafe { index_from(base, ip) }; + let rep_index = current.wrapping_sub(offset_2); + let rep_match = if rep_index < prefix_lowest_index { + dict_base.wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize) + } else { + base.wrapping_add(rep_index as usize) + }; + if index_overlap_check(prefix_lowest_index, rep_index) + && unsafe { read32(rep_match) == read32(ip) } + { + let rep_end = if rep_index < prefix_lowest_index { + dict_end + } else { + iend + }; + match_length = unsafe { + count_2segments( + ip.wrapping_add(4), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_lowest, + ) + 4 + }; + std::mem::swap(&mut offset_2, &mut offset_1); + unsafe { + store_seq( + seq_store, + 0, + anchor, + iend, + REPCODE1_TO_OFFBASE, + match_length, + ) + }; + ip = ip.wrapping_add(match_length); + anchor = ip; + continue; + } + break; + } + } + + if dict_mode == DICT_NO_DICT { + while ptr_le(ip, ilimit) + && offset_2 > 0 + && unsafe { read32(ip) == read32(ip.wrapping_sub(offset_2 as usize)) } + { + match_length = unsafe { + count( + ip.wrapping_add(4), + ip.wrapping_add(4).wrapping_sub(offset_2 as usize), + iend, + ) + 4 + }; + std::mem::swap(&mut offset_2, &mut offset_1); + unsafe { + store_seq( + seq_store, + 0, + anchor, + iend, + REPCODE1_TO_OFFBASE, + match_length, + ) + }; + ip = ip.wrapping_add(match_length); + anchor = ip; + } + } + } + + if offset_saved_1 != 0 && offset_1 != 0 { + offset_saved_2 = offset_saved_1; + } + unsafe { + *reps = if offset_1 != 0 { + offset_1 + } else { + offset_saved_1 + }; + *reps.add(1) = if offset_2 != 0 { + offset_2 + } else { + offset_saved_2 + }; + } + (iend as usize).wrapping_sub(anchor as usize) +} + +/* ------------------------------------------------------------------------- */ +/* Lazy parser for external dictionaries */ +/* ------------------------------------------------------------------------- */ + +unsafe fn compress_block_lazy_ext_dict_generic( + state: &ZSTD_RustLazyState, + seq_store: *mut SeqStore_t, + reps: *mut u32, + src: *const u8, + src_size: usize, + search_method: c_int, + depth: u32, +) -> usize { + let istart = src; + let mut ip = istart; + let mut anchor = istart; + let iend = istart.wrapping_add(src_size); + let required = if search_method == SEARCH_ROW_HASH { + HASH_READ_SIZE + ZSTD_ROW_HASH_CACHE_SIZE + } else { + HASH_READ_SIZE + }; + if src_size < required { + return src_size; + } + let ilimit = iend.wrapping_sub(required); + let base = state.base; + let dict_limit = state.dictLimit; + let prefix_start = base.wrapping_add(dict_limit as usize); + let dict_base = state.dictBase; + let dict_end = dict_base.wrapping_add(dict_limit as usize); + let dict_start = dict_base.wrapping_add(state.lowLimit as usize); + let mls = bounded(4, state.minMatch, 6); + let row_log = bounded(4, state.searchLog, 6); + let mut offset_1 = unsafe { *reps }; + let mut offset_2 = unsafe { *reps.add(1) }; + + if ip == prefix_start { + ip = ip.wrapping_add(1); + } + unsafe { set_lazy_skipping(state, false) }; + if search_method == SEARCH_ROW_HASH { + unsafe { row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit) }; + } + + while ptr_lt(ip, ilimit) { + let mut match_length = 0usize; + let mut offbase = REPCODE1_TO_OFFBASE as usize; + let mut start = ip.wrapping_add(1); + let mut curr = unsafe { index_from(base, ip) }; + let mut store_directly = false; + + let window_low = lowest_match_index( + state.lowLimit, + state.loadedDictEnd, + curr.wrapping_add(1), + state.windowLog, + ); + let rep_index = curr.wrapping_add(1).wrapping_sub(offset_1); + let rep_base = if rep_index < dict_limit { + dict_base + } else { + base + }; + let rep_match = rep_base.wrapping_add(rep_index as usize); + if index_overlap_check(dict_limit, rep_index) + && offset_1 <= curr.wrapping_add(1).wrapping_sub(window_low) + && unsafe { read32(ip.wrapping_add(1)) == read32(rep_match) } + { + let rep_end = if rep_index < dict_limit { + dict_end + } else { + iend + }; + match_length = unsafe { + count_2segments( + ip.wrapping_add(5), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_start, + ) + 4 + }; + if depth == 0 { + store_directly = true; + } + } + + if !store_directly { + let mut candidate = 999_999_999usize; + let found = unsafe { + search_max( + state, + ip, + iend, + &mut candidate, + mls, + row_log, + search_method, + DICT_EXT, + ) + }; + if found > match_length { + match_length = found; + start = ip; + offbase = candidate; + } + if match_length < 4 { + let step = (ip as usize).wrapping_sub(anchor as usize) >> K_SEARCH_STRENGTH; + ip = ip.wrapping_add(step.wrapping_add(1)); + unsafe { set_lazy_skipping(state, step > K_LAZY_SKIPPING_STEP) }; + continue; + } + + if depth >= 1 { + while ptr_lt(ip, ilimit) { + ip = ip.wrapping_add(1); + curr = curr.wrapping_add(1); + if offbase != 0 { + let window_low = lowest_match_index( + state.lowLimit, + state.loadedDictEnd, + curr, + state.windowLog, + ); + let rep_index = curr.wrapping_sub(offset_1); + let rep_base = if rep_index < dict_limit { + dict_base + } else { + base + }; + let rep_match = rep_base.wrapping_add(rep_index as usize); + if index_overlap_check(dict_limit, rep_index) + && offset_1 <= curr.wrapping_sub(window_low) + && unsafe { read32(ip) == read32(rep_match) } + { + let rep_end = if rep_index < dict_limit { + dict_end + } else { + iend + }; + let rep_length = unsafe { + count_2segments( + ip.wrapping_add(4), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_start, + ) + 4 + }; + if rep_length >= 4 + && score(rep_length, 3, REPCODE1_TO_OFFBASE as usize, 0) + > score(match_length, 3, offbase, 1) + { + match_length = rep_length; + offbase = REPCODE1_TO_OFFBASE as usize; + start = ip; + } + } + } + + let mut candidate = 999_999_999usize; + let found = unsafe { + search_max( + state, + ip, + iend, + &mut candidate, + mls, + row_log, + search_method, + DICT_EXT, + ) + }; + if found >= 4 + && score(found, 4, candidate, 0) > score(match_length, 4, offbase, 4) + { + match_length = found; + offbase = candidate; + start = ip; + continue; + } + + if depth == 2 && ptr_lt(ip, ilimit) { + ip = ip.wrapping_add(1); + curr = curr.wrapping_add(1); + if offbase != 0 { + let window_low = lowest_match_index( + state.lowLimit, + state.loadedDictEnd, + curr, + state.windowLog, + ); + let rep_index = curr.wrapping_sub(offset_1); + let rep_base = if rep_index < dict_limit { + dict_base + } else { + base + }; + let rep_match = rep_base.wrapping_add(rep_index as usize); + if index_overlap_check(dict_limit, rep_index) + && offset_1 <= curr.wrapping_sub(window_low) + && unsafe { read32(ip) == read32(rep_match) } + { + let rep_end = if rep_index < dict_limit { + dict_end + } else { + iend + }; + let rep_length = unsafe { + count_2segments( + ip.wrapping_add(4), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_start, + ) + 4 + }; + if rep_length >= 4 + && score(rep_length, 4, REPCODE1_TO_OFFBASE as usize, 0) + > score(match_length, 4, offbase, 1) + { + match_length = rep_length; + offbase = REPCODE1_TO_OFFBASE as usize; + start = ip; + } + } + } + + let mut second_candidate = 999_999_999usize; + let second_found = unsafe { + search_max( + state, + ip, + iend, + &mut second_candidate, + mls, + row_log, + search_method, + DICT_EXT, + ) + }; + if second_found >= 4 + && score(second_found, 4, second_candidate, 0) + > score(match_length, 4, offbase, 7) + { + match_length = second_found; + offbase = second_candidate; + start = ip; + continue; + } + } + break; + } + } + + if offbase_is_offset(offbase) { + let offset = offbase_to_offset(offbase); + let match_index = unsafe { index_from(base, start) }.wrapping_sub(offset); + let mut matched = if match_index < dict_limit { + dict_base.wrapping_add(match_index as usize) + } else { + base.wrapping_add(match_index as usize) + }; + let match_start = if match_index < dict_limit { + dict_start + } else { + prefix_start + }; + while ptr_gt(start, anchor) + && ptr_gt(matched, match_start) + && unsafe { *start.wrapping_sub(1) == *matched.wrapping_sub(1) } + { + start = start.wrapping_sub(1); + matched = matched.wrapping_sub(1); + match_length += 1; + } + offset_2 = offset_1; + offset_1 = offset; + } + } + + let lit_length = (start as usize).wrapping_sub(anchor as usize); + unsafe { + store_seq( + seq_store, + lit_length, + anchor, + iend, + offbase as u32, + match_length, + ) + }; + anchor = start.wrapping_add(match_length); + ip = anchor; + + if unsafe { lazy_skipping(state) } { + if search_method == SEARCH_ROW_HASH { + unsafe { + row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit) + }; + } + unsafe { set_lazy_skipping(state, false) }; + } + + while ptr_le(ip, ilimit) { + let rep_current = unsafe { index_from(base, ip) }; + let window_low = lowest_match_index( + state.lowLimit, + state.loadedDictEnd, + rep_current, + state.windowLog, + ); + let rep_index = rep_current.wrapping_sub(offset_2); + let rep_base = if rep_index < dict_limit { + dict_base + } else { + base + }; + let rep_match = rep_base.wrapping_add(rep_index as usize); + if index_overlap_check(dict_limit, rep_index) + && offset_2 <= rep_current.wrapping_sub(window_low) + && unsafe { read32(ip) == read32(rep_match) } + { + let rep_end = if rep_index < dict_limit { + dict_end + } else { + iend + }; + match_length = unsafe { + count_2segments( + ip.wrapping_add(4), + rep_match.wrapping_add(4), + iend, + rep_end, + prefix_start, + ) + 4 + }; + std::mem::swap(&mut offset_2, &mut offset_1); + unsafe { + store_seq( + seq_store, + 0, + anchor, + iend, + REPCODE1_TO_OFFBASE, + match_length, + ) + }; + ip = ip.wrapping_add(match_length); + anchor = ip; + continue; + } + break; + } + } + + unsafe { + *reps = offset_1; + *reps.add(1) = offset_2; + } + (iend as usize).wrapping_sub(anchor as usize) +} + +/* ------------------------------------------------------------------------- */ +/* C ABI */ +/* ------------------------------------------------------------------------- */ + +#[no_mangle] +pub unsafe extern "C" fn ZSTD_rust_lazy_insertAndFindFirstIndex( + state: *mut ZSTD_RustLazyState, + ip: *const c_void, +) -> u32 { + let state = unsafe { &*state }; + unsafe { insert_and_find_first_index_internal(state, ip.cast::(), state.minMatch, false) } +} + +#[no_mangle] +pub unsafe extern "C" fn ZSTD_rust_lazy_row_update( + state: *mut ZSTD_RustLazyState, + ip: *const c_void, +) { + let state = unsafe { &*state }; + let row_log = bounded(4, state.searchLog, 6); + let row_mask = (1u32 << row_log) - 1; + let mls = state.minMatch.min(6); + unsafe { row_update_internal(state, ip.cast::(), mls, row_log, row_mask, false) }; +} + +#[no_mangle] +pub unsafe extern "C" fn ZSTD_rust_lazy_loadDedicatedDict( + state: *mut ZSTD_RustLazyState, + ip: *const c_void, +) { + let state = unsafe { &*state }; + unsafe { dedicated_dict_search_load_dictionary(state, ip.cast::()) }; +} + +#[no_mangle] +pub unsafe extern "C" fn ZSTD_rust_compressBlock_lazy( + state: *mut ZSTD_RustLazyState, + seq_store: *mut c_void, + reps: *mut u32, + src: *const c_void, + src_size: usize, + search_method: c_int, + depth: u32, + dict_mode: c_int, +) -> usize { + let state = unsafe { &*state }; + if dict_mode == DICT_EXT { + unsafe { + compress_block_lazy_ext_dict_generic( + state, + seq_store.cast::(), + reps, + src.cast::(), + src_size, + search_method, + depth, + ) + } + } else { + unsafe { + compress_block_lazy_generic( + state, + seq_store.cast::(), + reps, + src.cast::(), + src_size, + search_method, + depth, + dict_mode, + ) + } + } +}