diff --git a/lib/zstd_compress.c b/lib/zstd_compress.c index 864d0ed6a..47f655946 100644 --- a/lib/zstd_compress.c +++ b/lib/zstd_compress.c @@ -240,7 +240,8 @@ static size_t ZSTD_resetCCtx_advanced (ZSTD_CCtx* zc, zc->seqStore.litLengthStart = zc->seqStore.litStart + blockSize; zc->seqStore.matchLengthStart = zc->seqStore.litLengthStart + (blockSize>>2); zc->seqStore.dumpsStart = zc->seqStore.matchLengthStart + (blockSize>>2); - zc->seqStore.litFreq = (U32*) (zc->seqStore.dumpsStart + (blockSize>>2)); + BYTE* dumpsEnd= zc->seqStore.dumpsStart + (blockSize>>2); + zc->seqStore.litFreq = (U32*)(dumpsEnd); zc->seqStore.litLengthFreq = zc->seqStore.litFreq + (1<seqStore.matchLengthFreq = zc->seqStore.litLengthFreq + (1<seqStore.offCodeFreq = zc->seqStore.matchLengthFreq + (1<params.searchLength); break; case ZSTD_btlazy2: + case ZSTD_opt_bt: ZSTD_updateTree(zc, iend-8, iend, 1 << zc->params.searchLog, zc->params.searchLength); break; diff --git a/lib/zstd_opt.c b/lib/zstd_opt.c index d36b61a31..8a9b1e2b1 100644 --- a/lib/zstd_opt.c +++ b/lib/zstd_opt.c @@ -3,19 +3,19 @@ typedef struct { - int off; - int len; - int back; + U32 off; + U32 len; + U32 back; } ZSTD_match_t; typedef struct { - int price; - int off; - int mlen; - int litlen; - int rep; - int rep2; + U32 price; + U32 off; + U32 mlen; + U32 litlen; + U32 rep; + U32 rep2; } ZSTD_optimal_t; @@ -38,34 +38,34 @@ typedef struct // log2_32 is from http://stackoverflow.com/questions/11376288/fast-computing-of-log2-for-64-bit-integers -const int tab32[32] = { +const U32 tab32[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31}; -int log2_32 (uint32_t value) +U32 log2_32 (U32 value) { value |= value >> 1; value |= value >> 2; value |= value >> 4; value |= value >> 8; value |= value >> 16; - return tab32[(uint32_t)(value*0x07C4ACDD) >> 27]; + return tab32[(U32)(value*0x07C4ACDD) >> 27]; } -FORCE_INLINE size_t ZSTD_getLiteralPriceReal(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals) +FORCE_INLINE U32 ZSTD_getLiteralPriceReal(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals) { - size_t price = 0; + U32 price = 0; if (litLength > 0) { /* literals */ - for (int i=litLength-1; i>=0; i--) + for (U32 i=0; i < litLength; i++) price += log2_32(seqStorePtr->litSum) - log2_32(seqStorePtr->litFreq[literals[i]]); /* literal Length */ - size_t freq; + U32 freq; if (litLength >= MaxLL) { freq = seqStorePtr->litLengthFreq[MaxLL]; if (litLength<255 + MaxLL) { @@ -84,7 +84,7 @@ FORCE_INLINE size_t ZSTD_getLiteralPriceReal(seqStore_t* seqStorePtr, size_t lit -FORCE_INLINE size_t ZSTD_getLiteralPrice(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals) +FORCE_INLINE U32 ZSTD_getLiteralPrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals) { if (seqStorePtr->litSum > ZSTD_FREQ_THRESHOLD) return ZSTD_getLiteralPriceReal(seqStorePtr, litLength, literals); @@ -94,17 +94,17 @@ FORCE_INLINE size_t ZSTD_getLiteralPrice(seqStore_t* seqStorePtr, size_t litLeng -FORCE_INLINE size_t ZSTD_getMatchPriceReal(seqStore_t* seqStorePtr, size_t offset, size_t matchLength) +FORCE_INLINE U32 ZSTD_getMatchPriceReal(seqStore_t* seqStorePtr, U32 offset, U32 matchLength) { /* match offset */ BYTE offCode = (BYTE)ZSTD_highbit(offset) + 1; if (offset==0) offCode = 0; - size_t price = log2_32(seqStorePtr->offCodeSum) - log2_32(seqStorePtr->offCodeFreq[offCode]); + U32 price = log2_32(seqStorePtr->offCodeSum) - log2_32(seqStorePtr->offCodeFreq[offCode]); price += offCode; /* match Length */ - size_t freq; + U32 freq; if (matchLength >= MaxML) { freq = seqStorePtr->matchLengthFreq[MaxML]; if (matchLength < 255+MaxML) { @@ -120,7 +120,7 @@ FORCE_INLINE size_t ZSTD_getMatchPriceReal(seqStore_t* seqStorePtr, size_t offse } -FORCE_INLINE size_t ZSTD_getPrice(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals, size_t offset, size_t matchLength) +FORCE_INLINE U32 ZSTD_getPrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength) { if (seqStorePtr->litSum > ZSTD_FREQ_THRESHOLD) return ZSTD_getLiteralPriceReal(seqStorePtr, litLength, literals) + ZSTD_getMatchPriceReal(seqStorePtr, offset, matchLength); @@ -130,7 +130,7 @@ FORCE_INLINE size_t ZSTD_getPrice(seqStore_t* seqStorePtr, size_t litLength, con -MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals, size_t offset, size_t matchLength) +MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength) { #if 0 static const BYTE* g_start = NULL; @@ -141,7 +141,7 @@ MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, size_t litLength, cons #endif /* literals */ seqStorePtr->litSum += litLength; - for (int i=litLength-1; i>=0; i--) + for (U32 i=0; i < litLength; i++) seqStorePtr->litFreq[literals[i]]++; /* literal Length */ @@ -178,8 +178,11 @@ MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, size_t litLength, cons +/* ************************************* +* Binary Tree search +***************************************/ FORCE_INLINE /* inlining is important to hardwire a hot branch (template emulation) */ -size_t ZSTD_insertBtAndGetAllMatches ( +U32 ZSTD_insertBtAndGetAllMatches ( ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, U32 nbCompares, const U32 mls, @@ -195,7 +198,7 @@ size_t ZSTD_insertBtAndGetAllMatches ( U32* const bt = zc->contentTable; const U32 btLog = zc->params.contentLog - 1; - const U32 btMask= (1 << btLog) - 1; + const U32 btMask= (1U << btLog) - 1; size_t commonLengthSmaller=0, commonLengthLarger=0; const BYTE* const dictBase = zc->dictBase; const U32 dictLimit = zc->dictLimit; @@ -207,7 +210,7 @@ size_t ZSTD_insertBtAndGetAllMatches ( U32* largerPtr = bt + 2*(current&btMask) + 1; U32 matchEndIdx = current+8; U32 dummy32; /* to be nullified at the end */ - size_t mnum = 0; + U32 mnum = 0; bestLength = 0; hashTable[h] = current; /* Update Hash Table */ @@ -235,7 +238,7 @@ size_t ZSTD_insertBtAndGetAllMatches ( if (matchLength >= MINMATCH) { bestLength = matchLength; matches[mnum].off = current - matchIndex; - matches[mnum].len = matchLength; + matches[mnum].len = (U32)matchLength; matches[mnum].back = 0; mnum++; } @@ -271,10 +274,10 @@ size_t ZSTD_insertBtAndGetAllMatches ( /** Tree updater, providing best match */ FORCE_INLINE /* inlining is important to hardwire a hot branch (template emulation) */ -size_t ZSTD_BtGetAllMatches ( +U32 ZSTD_BtGetAllMatches ( ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iLimit, - const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, size_t minml) + const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, U32 minml) { if (ip < zc->base + zc->nextToUpdate) return 0; /* skipped area */ ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls); @@ -282,10 +285,10 @@ size_t ZSTD_BtGetAllMatches ( } -FORCE_INLINE size_t ZSTD_BtGetAllMatches_selectMLS ( +FORCE_INLINE U32 ZSTD_BtGetAllMatches_selectMLS ( ZSTD_CCtx* zc, /* Index table will be updated */ const BYTE* ip, const BYTE* const iLowLimit, const BYTE* const iHighLimit, - const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, size_t minml) + const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, U32 minml) { if (iLowLimit) {}; // skip warnings @@ -298,28 +301,61 @@ FORCE_INLINE size_t ZSTD_BtGetAllMatches_selectMLS ( } } - +/** Tree updater, providing best match */ FORCE_INLINE /* inlining is important to hardwire a hot branch (template emulation) */ -size_t ZSTD_HcGetAllMatches_generic ( +size_t ZSTD_BtGetAllMatches_extDict ( + ZSTD_CCtx* zc, + const BYTE* const ip, const BYTE* const iLimit, + const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, U32 minml) +{ + if (ip < zc->base + zc->nextToUpdate) return 0; /* skipped area */ + ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls); + return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 1, matches, minml); +} + + +FORCE_INLINE U32 ZSTD_BtGetAllMatches_selectMLS_extDict ( + ZSTD_CCtx* zc, /* Index table will be updated */ + const BYTE* ip, const BYTE* const iLowLimit, const BYTE* const iHighLimit, + const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, U32 minml) +{ + if (iLowLimit) {}; // skip warnings + + switch(matchLengthSearch) + { + default : + case 4 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minml); + case 5 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minml); + case 6 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minml); + } +} + + +/* *********************** +* Hash Chain +*************************/ +FORCE_INLINE /* inlining is important to hardwire a hot branch (template emulation) */ +U32 ZSTD_HcGetAllMatches_generic ( ZSTD_CCtx* zc, /* Index table will be updated */ const BYTE* const ip, const BYTE* const iLowLimit, const BYTE* const iHighLimit, const U32 maxNbAttempts, const U32 mls, const U32 extDict, ZSTD_match_t* matches, size_t minml) { U32* const chainTable = zc->contentTable; - const U32 chainSize = (1 << zc->params.contentLog); + const U32 chainSize = (1U << zc->params.contentLog); const U32 chainMask = chainSize-1; const BYTE* const base = zc->base; const BYTE* const dictBase = zc->dictBase; const U32 dictLimit = zc->dictLimit; const BYTE* const prefixStart = base + dictLimit; const BYTE* const dictEnd = dictBase + dictLimit; + const BYTE* const dictStart = dictBase + zc->lowLimit; const U32 lowLimit = zc->lowLimit; const U32 current = (U32)(ip-base); const U32 minChain = current > chainSize ? current - chainSize : 0; U32 matchIndex; + U32 mnum = 0; const BYTE* match; - int nbAttempts=maxNbAttempts; - size_t mnum = 0; + U32 nbAttempts=maxNbAttempts; minml=MINMATCH-1; /* HC4 match finder */ @@ -328,7 +364,7 @@ size_t ZSTD_HcGetAllMatches_generic ( while ((matchIndex>lowLimit) && (nbAttempts)) { size_t currentMl=0; - size_t back = 0; + int back = 0; nbAttempts--; if ((!extDict) || matchIndex >= dictLimit) { match = base + matchIndex; @@ -343,7 +379,7 @@ size_t ZSTD_HcGetAllMatches_generic ( if (MEM_read32(match) == MEM_read32(ip)) /* assumption : matchIndex <= dictLimit-4 (by table construction) */ currentMl = ZSTD_count_2segments(ip+MINMATCH, match+MINMATCH, iHighLimit, dictEnd, prefixStart) + MINMATCH; if (currentMl > 0) { - while ((match-back > dictBase) && (ip-back > iLowLimit) && (ip[-back-1] == match[-back-1])) back++; /* backward match extension */ + while ((match-back > dictStart) && (ip-back > iLowLimit) && (ip[-back-1] == match[-back-1])) back++; /* backward match extension */ currentMl += back; } } @@ -352,7 +388,7 @@ size_t ZSTD_HcGetAllMatches_generic ( if (currentMl > minml) { minml = currentMl; matches[mnum].off = current - matchIndex; - matches[mnum].len = currentMl; + matches[mnum].len = (U32)currentMl; matches[mnum].back = back; mnum++; if (currentMl > ZSTD_OPT_NUM) break; @@ -367,10 +403,10 @@ size_t ZSTD_HcGetAllMatches_generic ( } -FORCE_INLINE size_t ZSTD_HcGetAllMatches_selectMLS ( +FORCE_INLINE U32 ZSTD_HcGetAllMatches_selectMLS ( ZSTD_CCtx* zc, const BYTE* ip, const BYTE* const iLowLimit, const BYTE* const iHighLimit, - const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, size_t minml) + const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, U32 minml) { switch(matchLengthSearch) { @@ -381,29 +417,21 @@ FORCE_INLINE size_t ZSTD_HcGetAllMatches_selectMLS ( } } - -void print_hex_text(uint8_t* buf, int bufsize, int endline) +FORCE_INLINE U32 ZSTD_HcGetAllMatches_selectMLS_extDict ( + ZSTD_CCtx* zc, + const BYTE* ip, const BYTE* const iLowLimit, const BYTE* const iHighLimit, + const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, U32 minml) { - int i, j; - for (i=0; i32?buf[i+j]:'.'); - printf("\n"); - } - if (endline) printf("\n"); + switch(matchLengthSearch) + { + default : + case 4 : return ZSTD_HcGetAllMatches_generic(zc, ip, iLowLimit, iHighLimit, maxNbAttempts, 4, 1, matches, minml); + case 5 : return ZSTD_HcGetAllMatches_generic(zc, ip, iLowLimit, iHighLimit, maxNbAttempts, 5, 1, matches, minml); + case 6 : return ZSTD_HcGetAllMatches_generic(zc, ip, iLowLimit, iHighLimit, maxNbAttempts, 6, 1, matches, minml); + } } - - /* ******************************* * Optimal parser *********************************/ @@ -420,24 +448,22 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, const BYTE* const ilimit = iend - 8; const BYTE* const base = ctx->base + ctx->dictLimit; - size_t rep_2=REPCODE_STARTVALUE, rep_1=REPCODE_STARTVALUE; - const U32 maxSearches = 1 << ctx->params.searchLog; + U32 rep_2=REPCODE_STARTVALUE, rep_1=REPCODE_STARTVALUE; + const U32 maxSearches = 1U << ctx->params.searchLog; const U32 mls = ctx->params.searchLength; - typedef size_t (*getAllMatches_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLowLimit, const BYTE* iHighLimit, - U32 maxNbAttempts, U32 matchLengthSearch, ZSTD_match_t* matches, size_t minml); + typedef U32 (*getAllMatches_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLowLimit, const BYTE* iHighLimit, + U32 maxNbAttempts, U32 matchLengthSearch, ZSTD_match_t* matches, U32 minml); getAllMatches_f getAllMatches = searchMethod ? ZSTD_BtGetAllMatches_selectMLS : ZSTD_HcGetAllMatches_selectMLS; ZSTD_optimal_t opt[ZSTD_OPT_NUM+4]; ZSTD_match_t matches[ZSTD_OPT_NUM+1]; const uint8_t *inr; - int skip_num = 0, cur, cur2, last_pos, litlen, price, match_num; - - const int sufficient_len = ctx->params.sufficientLength; - const size_t faster_get_matches = (ctx->params.strategy == ZSTD_opt); + U32 skip_num, cur, cur2, match_num, last_pos, litlen, price; + const U32 sufficient_len = ctx->params.sufficientLength; + const U32 faster_get_matches = (ctx->params.strategy == ZSTD_opt); - // printf("orig_file="); print_hex_text(ip, srcSize, 0); /* init */ ZSTD_resetSeqStore(seqStorePtr); @@ -446,19 +472,19 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, /* Match Loop */ while (ip < ilimit) { - int mlen=0; - int best_mlen=0; - int best_off=0; + U32 mlen=0; + U32 best_mlen=0; + U32 best_off=0; memset(opt, 0, sizeof(ZSTD_optimal_t)); last_pos = 0; inr = ip; - opt[0].litlen = ip - anchor; + opt[0].litlen = (U32)(ip - anchor); /* check repCode */ if (MEM_read32(ip+1) == MEM_read32(ip+1 - rep_1)) { /* repcode : we take it */ - mlen = ZSTD_count(ip+1+MINMATCH, ip+1+MINMATCH-rep_1, iend) + MINMATCH; + mlen = (U32)ZSTD_count(ip+1+MINMATCH, ip+1+MINMATCH-rep_1, iend) + MINMATCH; ZSTD_LOG_PARSER("%d: start try REP rep=%d mlen=%d\n", (int)(ip-base), (int)rep_1, (int)mlen); if (depth==0 || mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) { @@ -488,9 +514,9 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, ZSTD_LOG_PARSER("%d: match_num=%d last_pos=%d\n", (int)(ip-base), match_num, last_pos); if (!last_pos && !match_num) { ip++; continue; } - opt[0].rep = rep_1; - opt[0].rep2 = rep_2; - opt[0].mlen = 1; + opt[0].rep = rep_1; + opt[0].rep2 = rep_2; + opt[0].mlen = 1; if (match_num && matches[match_num-1].len > sufficient_len) { @@ -502,7 +528,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, } // set prices using matches at position = 0 - for (int i = 0; i < match_num; i++) + for (U32 i = 0; i < match_num; i++) { mlen = (i>0) ? matches[i-1].len+1 : best_mlen; best_mlen = (matches[i].len < ZSTD_OPT_NUM) ? matches[i].len : ZSTD_OPT_NUM; @@ -599,7 +625,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, if (MEM_read32(inr) == MEM_read32(inr - cur_rep)) // check rep { - mlen = ZSTD_count(inr+MINMATCH, inr+MINMATCH - cur_rep, iend) + MINMATCH; + mlen = (U32)ZSTD_count(inr+MINMATCH, inr+MINMATCH - cur_rep, iend) + MINMATCH; ZSTD_LOG_PARSER("%d: Found REP mlen=%d off=%d rep=%d opt[%d].off=%d\n", (int)(inr-base), mlen, 0, opt[cur].rep, cur, opt[cur].off); if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) @@ -670,7 +696,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, // set prices using matches at position = cur - for (int i = 0; i < match_num; i++) + for (U32 i = 0; i < match_num; i++) { mlen = (i>0) ? matches[i-1].len+1 : best_mlen; cur2 = cur - matches[i].back; @@ -715,14 +741,14 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx, /* store sequence */ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set - for (int i = 1; i <= last_pos; i++) + for (U32 i = 1; i <= last_pos; i++) ZSTD_LOG_PARSER("%d: price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+i), i, last_pos, opt[i].price, opt[i].off, opt[i].mlen, opt[i].litlen, opt[i].rep, opt[i].rep2); ZSTD_LOG_PARSER("%d: cur=%d/%d best_mlen=%d best_off=%d rep=%d\n", (int)(ip-base+cur), (int)cur, (int)last_pos, (int)best_mlen, (int)best_off, opt[cur].rep); opt[0].mlen = 1; - size_t offset; + U32 offset; - while (cur >= 0) + while (1) { mlen = opt[cur].mlen; ZSTD_LOG_PARSER("%d: cur=%d mlen=%d\n", (int)(ip-base), cur, mlen); @@ -731,10 +757,12 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set opt[cur].off = best_off; best_mlen = mlen; best_off = offset; + if (mlen > cur) + break; cur -= mlen; } - for (int i = 0; i <= last_pos;) + for (U32 i = 0; i <= last_pos;) { ZSTD_LOG_PARSER("%d: price2[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+i), i, last_pos, opt[i].price, opt[i].off, opt[i].mlen, opt[i].litlen, opt[i].rep, opt[i].rep2); i += opt[i].mlen; @@ -751,7 +779,7 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set cur += mlen; - size_t litLength = ip - anchor; + U32 litLength = (U32)(ip - anchor); ZSTD_LOG_ENCODE("%d/%d: ENCODE1 literals=%d mlen=%d off=%d rep1=%d rep2=%d\n", (int)(ip-base), (int)(iend-base), (int)(litLength), (int)mlen, (int)(offset), (int)rep_1, (int)rep_2); if (offset) @@ -770,9 +798,7 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set } ZSTD_LOG_ENCODE("%d/%d: ENCODE2 literals=%d mlen=%d off=%d rep1=%d rep2=%d\n", (int)(ip-base), (int)(iend-base), (int)(litLength), (int)mlen, (int)(offset), (int)rep_1, (int)rep_2); - // printf("orig="); print_hex_text(ip, mlen, 0); - // printf("match="); print_hex_text(ip-offset, mlen, 0); - + #if ZSTD_OPT_DEBUG >= 5 int ml2; if (offset) @@ -803,7 +829,7 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set while ( (anchor <= ilimit) && (MEM_read32(anchor) == MEM_read32(anchor - rep_2)) ) { /* store sequence */ - best_mlen = ZSTD_count(anchor+MINMATCH, anchor+MINMATCH-rep_2, iend); + best_mlen = (U32)ZSTD_count(anchor+MINMATCH, anchor+MINMATCH-rep_2, iend); best_off = rep_2; rep_2 = rep_1; rep_1 = best_off; @@ -818,7 +844,7 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set /* Last Literals */ { - size_t lastLLSize = iend - anchor; + U32 lastLLSize = (U32)(iend - anchor); ZSTD_LOG_ENCODE("%d: lastLLSize literals=%d\n", (int)(ip-base), (int)(lastLLSize)); memcpy(seqStorePtr->lit, anchor, lastLLSize); seqStorePtr->lit += lastLLSize; @@ -827,3 +853,438 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set +FORCE_INLINE +void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx, + const void* src, size_t srcSize, + const U32 searchMethod, const U32 depth) +{ + seqStore_t* seqStorePtr = &(ctx->seqStore); + 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 = iend - 8; + const BYTE* const base = ctx->base; + const U32 dictLimit = ctx->dictLimit; + const BYTE* const prefixStart = base + dictLimit; + const BYTE* const dictBase = ctx->dictBase; + const BYTE* const dictEnd = dictBase + dictLimit; + + U32 rep_2=REPCODE_STARTVALUE, rep_1=REPCODE_STARTVALUE; + const U32 maxSearches = 1U << ctx->params.searchLog; + const U32 mls = ctx->params.searchLength; + + typedef U32 (*getAllMatches_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLowLimit, const BYTE* iHighLimit, + U32 maxNbAttempts, U32 matchLengthSearch, ZSTD_match_t* matches, U32 minml); + getAllMatches_f getAllMatches = searchMethod ? ZSTD_BtGetAllMatches_selectMLS_extDict : ZSTD_HcGetAllMatches_selectMLS_extDict; + + ZSTD_optimal_t opt[ZSTD_OPT_NUM+4]; + ZSTD_match_t matches[ZSTD_OPT_NUM+1]; + const uint8_t *inr; + U32 skip_num, cur, cur2, match_num, last_pos, litlen, price; + + const U32 sufficient_len = ctx->params.sufficientLength; + const U32 faster_get_matches = (ctx->params.strategy == ZSTD_opt); + + + /* init */ + ZSTD_resetSeqStore(seqStorePtr); + if ((ip - prefixStart) < REPCODE_STARTVALUE) ip += REPCODE_STARTVALUE; + + + /* Match Loop */ + while (ip < ilimit) { + U32 mlen=0; + U32 best_mlen=0; + U32 best_off=0; + U32 current = (U32)(ip-base); + memset(opt, 0, sizeof(ZSTD_optimal_t)); + last_pos = 0; + inr = ip; + opt[0].litlen = (U32)(ip - anchor); + + + /* check repCode */ + { + const U32 repIndex = (U32)(current+1 - rep_1); + const BYTE* const repBase = repIndex < dictLimit ? dictBase : base; + const BYTE* const repMatch = repBase + repIndex; + if ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow */ + if (MEM_read32(ip+1) == MEM_read32(repMatch)) { + /* repcode detected we should take it */ + const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; + mlen = (U32)ZSTD_count_2segments(ip+1+MINMATCH, repMatch+MINMATCH, iend, repEnd, prefixStart) + MINMATCH; + + ZSTD_LOG_PARSER("%d: start try REP rep=%d mlen=%d\n", (int)(ip-base), (int)rep_1, (int)mlen); + if (depth==0 || mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) { + ip+=1; best_mlen = mlen; best_off = 0; cur = 0; last_pos = 1; + goto _storeSequence; + } + + litlen = opt[0].litlen + 1; + do + { + price = ZSTD_getPrice(seqStorePtr, litlen, anchor, 0, mlen - MINMATCH); + if (mlen + 1 > last_pos || price < opt[mlen + 1].price) + SET_PRICE(mlen + 1, mlen, 0, litlen, price); + mlen--; + } + while (mlen >= MINMATCH); + } + } + + best_mlen = (last_pos) ? last_pos : MINMATCH; + + if (faster_get_matches && last_pos) + match_num = 0; + else + match_num = getAllMatches(ctx, ip, ip, iend, maxSearches, mls, matches, best_mlen); /* first search (depth 0) */ + + ZSTD_LOG_PARSER("%d: match_num=%d last_pos=%d\n", (int)(ip-base), match_num, last_pos); + if (!last_pos && !match_num) { ip++; continue; } + + opt[0].rep = rep_1; + opt[0].rep2 = rep_2; + opt[0].mlen = 1; + + if (match_num && matches[match_num-1].len > sufficient_len) + { + best_mlen = matches[match_num-1].len; + best_off = matches[match_num-1].off; + cur = 0; + last_pos = 1; + goto _storeSequence; + } + + // set prices using matches at position = 0 + for (U32 i = 0; i < match_num; i++) + { + mlen = (i>0) ? matches[i-1].len+1 : best_mlen; + best_mlen = (matches[i].len < ZSTD_OPT_NUM) ? matches[i].len : ZSTD_OPT_NUM; + ZSTD_LOG_PARSER("%d: start Found mlen=%d off=%d best_mlen=%d last_pos=%d\n", (int)(ip-base), matches[i].len, matches[i].off, (int)best_mlen, (int)last_pos); + litlen = opt[0].litlen; + while (mlen <= best_mlen) + { + price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[i].off, mlen - MINMATCH); + if (mlen > last_pos || price < opt[mlen].price) + SET_PRICE(mlen, mlen, matches[i].off, litlen, price); + mlen++; + } + } + + if (last_pos < MINMATCH) { + // ip += ((ip-anchor) >> g_searchStrength) + 1; /* jump faster over incompressible sections */ + ip++; continue; + } + + + // check further positions + for (skip_num = 0, cur = 1; cur <= last_pos; cur++) + { + inr = ip + cur; + + if (opt[cur-1].mlen == 1) + { + litlen = opt[cur-1].litlen + 1; + if (cur > litlen) + { + price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen); + ZSTD_LOG_TRY_PRICE("%d: TRY1 opt[%d].price=%d price=%d cur=%d litlen=%d\n", (int)(inr-base), cur - litlen, opt[cur - litlen].price, price, cur, litlen); + } + else + price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor); + } + else + { + litlen = 1; + price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1); + ZSTD_LOG_TRY_PRICE("%d: TRY3 price=%d cur=%d litlen=%d litonly=%d\n", (int)(inr-base), price, cur, litlen, (int)ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1)); + } + + ZSTD_LOG_TRY_PRICE("%d: TRY4 price=%d opt[%d].price=%d\n", (int)(inr-base), price, cur, opt[cur].price); + + if (cur > last_pos || price <= opt[cur].price) // || ((price == opt[cur].price) && (opt[cur-1].mlen == 1) && (cur != litlen))) + SET_PRICE(cur, 1, 0, litlen, price); + + if (cur == last_pos) break; + + if (inr > ilimit) // last match must start at a minimum distance of 8 from oend + continue; + + mlen = opt[cur].mlen; + + if (opt[cur-mlen].off) + { + opt[cur].rep2 = opt[cur-mlen].rep; + opt[cur].rep = opt[cur-mlen].off; + ZSTD_LOG_PARSER("%d: COPYREP1 cur=%d mlen=%d rep=%d rep2=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep, opt[cur].rep2); + } + else + { + if (cur!=mlen && opt[cur-mlen].litlen == 0) + { + opt[cur].rep2 = opt[cur-mlen].rep; + opt[cur].rep = opt[cur-mlen].rep2; + ZSTD_LOG_PARSER("%d: COPYREP2 cur=%d mlen=%d rep=%d rep2=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep, opt[cur].rep2); + } + else + { + opt[cur].rep2 = opt[cur-mlen].rep2; + opt[cur].rep = opt[cur-mlen].rep; + ZSTD_LOG_PARSER("%d: COPYREP3 cur=%d mlen=%d rep=%d rep2=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep, opt[cur].rep2); + } + } + + ZSTD_LOG_PARSER("%d: CURRENT price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(inr-base), cur, last_pos, opt[cur].price, opt[cur].off, opt[cur].mlen, opt[cur].litlen, opt[cur].rep, opt[cur].rep2); + + + + size_t cur_rep; + best_mlen = 0; + + if (!opt[cur].off && opt[cur].mlen != 1) { + cur_rep = opt[cur].rep2; + ZSTD_LOG_PARSER("%d: try REP2 rep2=%d mlen=%d\n", (int)(inr-base), cur_rep, mlen); + } + else { + cur_rep = opt[cur].rep; + ZSTD_LOG_PARSER("%d: try REP1 rep=%d mlen=%d\n", (int)(inr-base), cur_rep, mlen); + } + + const U32 repIndex = (U32)(current+cur - cur_rep); + const BYTE* const repBase = repIndex < dictLimit ? dictBase : base; + const BYTE* const repMatch = repBase + repIndex; + if ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow */ + if (MEM_read32(inr) == MEM_read32(repMatch)) { + /* repcode detected */ + const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; + mlen = (U32)ZSTD_count_2segments(inr+MINMATCH, repMatch+MINMATCH, iend, repEnd, prefixStart) + MINMATCH; + ZSTD_LOG_PARSER("%d: Found REP mlen=%d off=%d rep=%d opt[%d].off=%d\n", (int)(inr-base), mlen, 0, opt[cur].rep, cur, opt[cur].off); + + if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) + { + best_mlen = mlen; + best_off = 0; + ZSTD_LOG_PARSER("%d: REP sufficient_len=%d best_mlen=%d best_off=%d last_pos=%d\n", (int)(inr-base), sufficient_len, best_mlen, best_off, last_pos); + last_pos = cur + 1; + goto _storeSequence; + } + + if (opt[cur].mlen == 1) + { + litlen = opt[cur].litlen; + + if (cur > litlen) + { + price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, 0, mlen - MINMATCH); + ZSTD_LOG_TRY_PRICE("%d: TRY5 opt[%d].price=%d price=%d cur=%d litlen=%d\n", (int)(inr-base), cur - litlen, opt[cur - litlen].price, price, cur, litlen); + } + else + price = ZSTD_getPrice(seqStorePtr, litlen, anchor, 0, mlen - MINMATCH); + } + else + { + litlen = 0; + price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, 0, mlen - MINMATCH); + ZSTD_LOG_TRY_PRICE("%d: TRY7 price=%d cur=%d litlen=0 getprice=%d\n", (int)(inr-base), price, cur, (int)ZSTD_getPrice(seqStorePtr, 0, NULL, 0, mlen - MINMATCH)); + } + + best_mlen = mlen; + if (faster_get_matches) + skip_num = best_mlen; + + ZSTD_LOG_PARSER("%d: Found REP mlen=%d off=%d price=%d litlen=%d price[%d]=%d\n", (int)(inr-base), mlen, 0, price, litlen, cur - litlen, opt[cur - litlen].price); + + do + { + if (cur + mlen > last_pos || price <= opt[cur + mlen].price) // || ((price == opt[cur + mlen].price) && (opt[cur].mlen == 1) && (cur != litlen))) // at equal price prefer REP instead of MATCH + SET_PRICE(cur + mlen, mlen, 0, litlen, price); + mlen--; + } + while (mlen >= MINMATCH); + } + + + if (faster_get_matches && skip_num > 0) + { + skip_num--; + continue; + } + + + best_mlen = (best_mlen > MINMATCH) ? best_mlen : MINMATCH; + + match_num = getAllMatches(ctx, inr, ip, iend, maxSearches, mls, matches, best_mlen); + ZSTD_LOG_PARSER("%d: ZSTD_GetAllMatches match_num=%d\n", (int)(inr-base), match_num); + + + if (match_num > 0 && matches[match_num-1].len > sufficient_len) + { + cur -= matches[match_num-1].back; + best_mlen = matches[match_num-1].len; + best_off = matches[match_num-1].off; + last_pos = cur + 1; + goto _storeSequence; + } + + + // set prices using matches at position = cur + for (U32 i = 0; i < match_num; i++) + { + mlen = (i>0) ? matches[i-1].len+1 : best_mlen; + cur2 = cur - matches[i].back; + best_mlen = (cur2 + matches[i].len < ZSTD_OPT_NUM) ? matches[i].len : ZSTD_OPT_NUM - cur2; + + ZSTD_LOG_PARSER("%d: Found1 cur=%d cur2=%d mlen=%d off=%d best_mlen=%d last_pos=%d\n", (int)(inr-base), cur, cur2, matches[i].len, matches[i].off, best_mlen, last_pos); + if (mlen < matches[i].back + 1) + mlen = matches[i].back + 1; + + while (mlen <= best_mlen) + { + if (opt[cur2].mlen == 1) + { + litlen = opt[cur2].litlen; + if (cur2 > litlen) + price = opt[cur2 - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur2-litlen, matches[i].off, mlen - MINMATCH); + else + price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[i].off, mlen - MINMATCH); + } + else + { + litlen = 0; + price = opt[cur2].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[i].off, mlen - MINMATCH); + } + + ZSTD_LOG_PARSER("%d: Found2 pred=%d mlen=%d best_mlen=%d off=%d price=%d litlen=%d price[%d]=%d\n", (int)(inr-base), matches[i].back, mlen, best_mlen, matches[i].off, price, litlen, cur - litlen, opt[cur - litlen].price); + ZSTD_LOG_TRY_PRICE("%d: TRY8 price=%d opt[%d].price=%d\n", (int)(inr-base), price, cur2 + mlen, opt[cur2 + mlen].price); + + if (cur2 + mlen > last_pos || (price < opt[cur2 + mlen].price)) + SET_PRICE(cur2 + mlen, mlen, matches[i].off, litlen, price); + + mlen++; + } + } + } // for (skip_num = 0, cur = 1; cur <= last_pos; cur++) + + + best_mlen = opt[last_pos].mlen; + best_off = opt[last_pos].off; + cur = last_pos - best_mlen; + // printf("%d: start=%d best_mlen=%d best_off=%d cur=%d\n", (int)(ip - base), (int)(start - ip), (int)best_mlen, (int)best_off, cur); + + /* store sequence */ +_storeSequence: // cur, last_pos, best_mlen, best_off have to be set + for (U32 i = 1; i <= last_pos; i++) + ZSTD_LOG_PARSER("%d: price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+i), i, last_pos, opt[i].price, opt[i].off, opt[i].mlen, opt[i].litlen, opt[i].rep, opt[i].rep2); + ZSTD_LOG_PARSER("%d: cur=%d/%d best_mlen=%d best_off=%d rep=%d\n", (int)(ip-base+cur), (int)cur, (int)last_pos, (int)best_mlen, (int)best_off, opt[cur].rep); + + opt[0].mlen = 1; + U32 offset; + + while (1) + { + mlen = opt[cur].mlen; + ZSTD_LOG_PARSER("%d: cur=%d mlen=%d\n", (int)(ip-base), cur, mlen); + offset = opt[cur].off; + opt[cur].mlen = best_mlen; + opt[cur].off = best_off; + best_mlen = mlen; + best_off = offset; + if (mlen > cur) + break; + cur -= mlen; + } + + for (U32 i = 0; i <= last_pos;) + { + ZSTD_LOG_PARSER("%d: price2[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+i), i, last_pos, opt[i].price, opt[i].off, opt[i].mlen, opt[i].litlen, opt[i].rep, opt[i].rep2); + i += opt[i].mlen; + } + + cur = 0; + + while (cur < last_pos) + { + ZSTD_LOG_PARSER("%d: price3[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+cur), cur, last_pos, opt[cur].price, opt[cur].off, opt[cur].mlen, opt[cur].litlen, opt[cur].rep, opt[cur].rep2); + mlen = opt[cur].mlen; + if (mlen == 1) { ip++; cur++; continue; } + offset = opt[cur].off; + cur += mlen; + + + U32 litLength = (U32)(ip - anchor); + ZSTD_LOG_ENCODE("%d/%d: ENCODE1 literals=%d mlen=%d off=%d rep1=%d rep2=%d\n", (int)(ip-base), (int)(iend-base), (int)(litLength), (int)mlen, (int)(offset), (int)rep_1, (int)rep_2); + + if (offset) + { + rep_2 = rep_1; + rep_1 = offset; + } + else + { + if (litLength == 0) + { + best_off = rep_2; + rep_2 = rep_1; + rep_1 = best_off; + } + } + + ZSTD_LOG_ENCODE("%d/%d: ENCODE2 literals=%d mlen=%d off=%d rep1=%d rep2=%d\n", (int)(ip-base), (int)(iend-base), (int)(litLength), (int)mlen, (int)(offset), (int)rep_1, (int)rep_2); + +#if ZSTD_OPT_DEBUG >= 5 + int ml2; + if (offset) + ml2 = ZSTD_count(ip, ip-offset, iend); + else + ml2 = ZSTD_count(ip, ip-rep_1, iend); + if (ml2 < mlen && ml2 < MINMATCH) { + printf("%d: ERROR iend=%d mlen=%d offset=%d ml2=%d\n", (int)(ip - base), (int)(iend - ip), (int)mlen, (int)offset, (int)ml2); exit(0); } + if (ip < anchor) { + printf("%d: ERROR ip < anchor iend=%d mlen=%d offset=%d\n", (int)(ip - base), (int)(iend - ip), (int)mlen, (int)offset); exit(0); } + if (ip - offset < ctx->base) { + printf("%d: ERROR ip - offset < base iend=%d mlen=%d offset=%d\n", (int)(ip - base), (int)(iend - ip), (int)mlen, (int)offset); exit(0); } + if ((int)offset >= (1 << ctx->params.windowLog)) { + printf("%d: offset >= (1 << params.windowLog) iend=%d mlen=%d offset=%d\n", (int)(ip - base), (int)(iend - ip), (int)mlen, (int)offset); exit(0); } + if (mlen < MINMATCH) { + printf("%d: ERROR mlen < MINMATCH iend=%d mlen=%d offset=%d\n", (int)(ip - base), (int)(iend - ip), (int)mlen, (int)offset); exit(0); } + if (ip + mlen > iend) { + printf("%d: ERROR ip + mlen >= iend iend=%d mlen=%d offset=%d\n", (int)(ip - base), (int)(iend - ip), (int)mlen, (int)offset); exit(0); } +#endif + + ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH); + ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH); + anchor = ip = ip + mlen; + } + + + // check immediate repcode + while (anchor <= ilimit) { + const U32 repIndex = (U32)((anchor-base) - rep_2); + const BYTE* const repBase = repIndex < dictLimit ? dictBase : base; + const BYTE* const repMatch = repBase + repIndex; + if ((U32)((dictLimit-1) - repIndex) >= 3) /* intentional overflow */ + if (MEM_read32(anchor) == MEM_read32(repMatch)) { + /* repcode detected we should take it */ + const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend; + mlen = (U32)ZSTD_count_2segments(anchor+MINMATCH, repMatch+MINMATCH, iend, repEnd, prefixStart) + MINMATCH; + offset = rep_2; rep_2 = rep_1; rep_1 = offset; /* swap offset history */ + ZSTD_LOG_ENCODE("%d/%d: ENCODE REP literals=%d mlen=%d off=%d rep1=%d rep2=%d\n", (int)(anchor-base), (int)(iend-base), (int)(0), (int)best_mlen, (int)(0), (int)rep_1, (int)rep_2); + ZSTD_updatePrice(seqStorePtr, 0, anchor, 0, mlen-MINMATCH); + ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, mlen-MINMATCH); + anchor += mlen; + ip = anchor; + continue; /* faster when present ... (?) */ + } + break; + } + } + + /* Last Literals */ + { + U32 lastLLSize = (U32)(iend - anchor); + ZSTD_LOG_ENCODE("%d: lastLLSize literals=%d\n", (int)(ip-base), (int)(lastLLSize)); + memcpy(seqStorePtr->lit, anchor, lastLLSize); + seqStorePtr->lit += lastLLSize; + } +}