This commit is contained in:
Yann Collet 2015-11-09 03:19:33 +01:00
parent d1ade5a61d
commit 4baee50728
2 changed files with 20 additions and 18 deletions

View File

@ -447,7 +447,6 @@ size_t ZSTD_HC_insertBtAndFindBestMatch (
*smallerPtr = *largerPtr = 0; *smallerPtr = *largerPtr = 0;
zc->nextToUpdate = current+1; /* current has been inserted */ zc->nextToUpdate = current+1; /* current has been inserted */
if (bestLength < MINMATCH) return 0;
return bestLength; return bestLength;
} }
@ -562,22 +561,21 @@ size_t ZSTD_HC_HcFindBestMatch (
if (matchIndex >= dictLimit) if (matchIndex >= dictLimit)
{ {
match = base + matchIndex; match = base + matchIndex;
if ( (match[ml] == ip[ml]) if (match[ml] == ip[ml]) /* potentially better */
&& (MEM_read32(match) == MEM_read32(ip)) ) /* ensures minimum match of 4 */
{ {
const size_t mlt = ZSTD_count(ip+MINMATCH, match+MINMATCH, iLimit) + MINMATCH; const size_t mlt = ZSTD_count(ip, match, iLimit);
if (mlt > ml) if (mlt > ml)
//if (((int)(4*mlt) - (int)ZSTD_highbit((U32)(ip-match)+1)) > ((int)(4*ml) - (int)ZSTD_highbit((U32)((*offsetPtr)+1)))) //if (((int)(4*mlt) - (int)ZSTD_highbit((U32)(ip-match)+1)) > ((int)(4*ml) - (int)ZSTD_highbit((U32)((*offsetPtr)+1))))
{ {
ml = mlt; *offsetPtr = ip-match; ml = mlt; *offsetPtr = ip-match;
if (ip+ml >= iLimit) break; if (ip+mlt >= iLimit) break;
} }
} }
} }
else else
{ {
match = dictBase + matchIndex; match = dictBase + matchIndex;
if (MEM_read32(match) == MEM_read32(ip)) if (MEM_read32(match) == MEM_read32(ip)) /* beware of end of dict */
{ {
size_t mlt; size_t mlt;
const BYTE* vLimit = ip + (dictLimit - matchIndex); const BYTE* vLimit = ip + (dictLimit - matchIndex);
@ -663,7 +661,7 @@ size_t ZSTD_HC_compressBlock_lazy_generic(ZSTD_HC_CCtx* ctx,
offset_2 = offset_1; offset_2 = offset_1;
matchLength = searchMax(ctx, ip, iend, &offset, maxSearches, mls); matchLength = searchMax(ctx, ip, iend, &offset, maxSearches, mls);
if (!matchLength) if (matchLength < MINMATCH)
{ {
ip += ((ip-anchor) >> g_searchStrength) + 1; /* jump faster over incompressible sections */ ip += ((ip-anchor) >> g_searchStrength) + 1; /* jump faster over incompressible sections */
continue; continue;
@ -680,7 +678,7 @@ size_t ZSTD_HC_compressBlock_lazy_generic(ZSTD_HC_CCtx* ctx,
size_t ml2 = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_1, iend) + MINMATCH; size_t ml2 = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_1, iend) + MINMATCH;
int gain2 = (int)(ml2 * 3); int gain2 = (int)(ml2 * 3);
int gain1 = (int)(matchLength*3 - ZSTD_highbit((U32)offset+1) + 1); int gain1 = (int)(matchLength*3 - ZSTD_highbit((U32)offset+1) + 1);
if (gain2 > gain1) if ((ml2 >= MINMATCH) && (gain2 > gain1))
matchLength = ml2, offset = 0, start = ip; matchLength = ml2, offset = 0, start = ip;
} }
{ {
@ -688,7 +686,7 @@ size_t ZSTD_HC_compressBlock_lazy_generic(ZSTD_HC_CCtx* ctx,
size_t ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls); size_t ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
int gain2 = (int)(ml2*(3+deep) - ZSTD_highbit((U32)offset2+1)); /* raw approx */ int gain2 = (int)(ml2*(3+deep) - ZSTD_highbit((U32)offset2+1)); /* raw approx */
int gain1 = (int)(matchLength*(3+deep) - ZSTD_highbit((U32)offset+1) + (3+deep)); int gain1 = (int)(matchLength*(3+deep) - ZSTD_highbit((U32)offset+1) + (3+deep));
if (gain2 > gain1) if ((ml2 >= MINMATCH) && (gain2 > gain1))
{ {
matchLength = ml2, offset = offset2, start = ip; matchLength = ml2, offset = offset2, start = ip;
continue; /* search a better one */ continue; /* search a better one */
@ -704,7 +702,7 @@ size_t ZSTD_HC_compressBlock_lazy_generic(ZSTD_HC_CCtx* ctx,
size_t ml2 = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_1, iend) + MINMATCH; size_t ml2 = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_1, iend) + MINMATCH;
int gain2 = (int)(ml2 * 4); int gain2 = (int)(ml2 * 4);
int gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 1); int gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 1);
if (gain2 > gain1) if ((ml2 >= MINMATCH) && (gain2 > gain1))
matchLength = ml2, offset = 0, start = ip; matchLength = ml2, offset = 0, start = ip;
} }
{ {
@ -712,7 +710,7 @@ size_t ZSTD_HC_compressBlock_lazy_generic(ZSTD_HC_CCtx* ctx,
size_t ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls); size_t ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
int gain2 = (int)(ml2*4 - ZSTD_highbit((U32)offset2+1)); /* raw approx */ int gain2 = (int)(ml2*4 - ZSTD_highbit((U32)offset2+1)); /* raw approx */
int gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 7); int gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 7);
if (gain2 > gain1) if ((ml2 >= MINMATCH) && (gain2 > gain1))
{ {
matchLength = ml2, offset = offset2, start = ip; matchLength = ml2, offset = offset2, start = ip;
continue; continue;
@ -722,15 +720,19 @@ size_t ZSTD_HC_compressBlock_lazy_generic(ZSTD_HC_CCtx* ctx,
break; /* nothing found : store previous solution */ break; /* nothing found : store previous solution */
} }
/* store sequence */ /* catch up */
if (offset) if (offset)
while ((start>anchor) && (start-offset>ctx->base) && (start[-1] == start[-1-offset])) { start--; matchLength++; } /* catch up */ {
while ((start>anchor) && (start>ctx->base+offset) && (start[-1] == start[-1-offset]))
{ start--; matchLength++; }
}
/* store sequence */
{ {
size_t litLength = start - anchor; size_t litLength = start - anchor;
if (offset) offset_1 = offset; if (offset) offset_1 = offset;
ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, matchLength-MINMATCH); ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, matchLength-MINMATCH);
ip = start + matchLength; anchor = ip = start + matchLength;
anchor = ip;
} }
} }
@ -815,7 +817,7 @@ size_t ZSTD_HC_compressBlock_greedy(ZSTD_HC_CCtx* ctx, void* dst, size_t maxDstS
{ {
size_t offset=999999; size_t offset=999999;
size_t matchLength = ZSTD_HC_HcFindBestMatch_selectMLS(ctx, ip, iend, &offset, maxSearches, mls); size_t matchLength = ZSTD_HC_HcFindBestMatch_selectMLS(ctx, ip, iend, &offset, maxSearches, mls);
if (!matchLength) if (matchLength < MINMATCH)
{ {
ip += ((ip-anchor) >> g_searchStrength) + 1; /* jump faster over incompressible sections */ ip += ((ip-anchor) >> g_searchStrength) + 1; /* jump faster over incompressible sections */
continue; continue;

View File

@ -106,7 +106,7 @@ static const ZSTD_HC_parameters ZSTD_HC_defaultParameters[ZSTD_HC_MAX_CLEVEL+1]
{ 19, 15, 16, 1, 6, ZSTD_HC_fast }, /* level 2 */ { 19, 15, 16, 1, 6, ZSTD_HC_fast }, /* level 2 */
{ 20, 18, 20, 1, 6, ZSTD_HC_fast }, /* level 3 */ { 20, 18, 20, 1, 6, ZSTD_HC_fast }, /* level 3 */
{ 21, 19, 21, 1, 6, ZSTD_HC_fast }, /* level 4 */ { 21, 19, 21, 1, 6, ZSTD_HC_fast }, /* level 4 */
{ 20, 13, 18, 4, 5, ZSTD_HC_greedy }, /* level 5 */ { 20, 13, 18, 5, 5, ZSTD_HC_greedy }, /* level 5 */
{ 20, 17, 19, 3, 5, ZSTD_HC_greedy }, /* level 6 */ { 20, 17, 19, 3, 5, ZSTD_HC_greedy }, /* level 6 */
{ 21, 17, 20, 3, 5, ZSTD_HC_lazy }, /* level 7 */ { 21, 17, 20, 3, 5, ZSTD_HC_lazy }, /* level 7 */
{ 21, 19, 20, 3, 5, ZSTD_HC_lazy }, /* level 8 */ { 21, 19, 20, 3, 5, ZSTD_HC_lazy }, /* level 8 */
@ -116,7 +116,7 @@ static const ZSTD_HC_parameters ZSTD_HC_defaultParameters[ZSTD_HC_MAX_CLEVEL+1]
{ 22, 20, 22, 5, 5, ZSTD_HC_lazy2 }, /* level 12 */ { 22, 20, 22, 5, 5, ZSTD_HC_lazy2 }, /* level 12 */
{ 22, 21, 22, 5, 5, ZSTD_HC_lazy2 }, /* level 13 */ { 22, 21, 22, 5, 5, ZSTD_HC_lazy2 }, /* level 13 */
{ 22, 22, 23, 5, 5, ZSTD_HC_lazy2 }, /* level 14 */ { 22, 22, 23, 5, 5, ZSTD_HC_lazy2 }, /* level 14 */
{ 22, 21, 22, 6, 5, ZSTD_HC_lazy2 }, /* level 15 */ { 23, 23, 23, 5, 5, ZSTD_HC_lazy2 }, /* level 15 */
{ 23, 21, 22, 5, 5, ZSTD_HC_btlazy2 }, /* level 16 */ { 23, 21, 22, 5, 5, ZSTD_HC_btlazy2 }, /* level 16 */
{ 23, 24, 23, 4, 5, ZSTD_HC_btlazy2 }, /* level 17 */ { 23, 24, 23, 4, 5, ZSTD_HC_btlazy2 }, /* level 17 */
{ 25, 24, 23, 5, 5, ZSTD_HC_btlazy2 }, /* level 18 */ { 25, 24, 23, 5, 5, ZSTD_HC_btlazy2 }, /* level 18 */