diff options
author | Yann Collet <cyan@fb.com> | 2018-04-26 20:01:59 (GMT) |
---|---|---|
committer | Yann Collet <cyan@fb.com> | 2018-04-26 20:01:59 (GMT) |
commit | 1148173c5dd1ad9b672c63fd0da110e3c2d66274 (patch) | |
tree | 04e18205e58733cd9d1d2c53b6cfd5c19356d777 /lib/lz4hc.c | |
parent | cadf5cd5f9289a9dc8bc60d2357452573908e729 (diff) | |
download | lz4-1148173c5dd1ad9b672c63fd0da110e3c2d66274.zip lz4-1148173c5dd1ad9b672c63fd0da110e3c2d66274.tar.gz lz4-1148173c5dd1ad9b672c63fd0da110e3c2d66274.tar.bz2 |
introduced ability to parse for decompression speed
triggered through an enum.
Now, it's still necessary to properly expose this capability
all the way up to the cli.
Diffstat (limited to 'lib/lz4hc.c')
-rw-r--r-- | lib/lz4hc.c | 53 |
1 files changed, 34 insertions, 19 deletions
diff --git a/lib/lz4hc.c b/lib/lz4hc.c index 4126ef8..3593da7 100644 --- a/lib/lz4hc.c +++ b/lib/lz4hc.c @@ -199,6 +199,7 @@ LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow, U32 pattern) } typedef enum { rep_untested, rep_not, rep_confirmed } repeat_state_e; +typedef enum { favorCompressionRatio=0, favorDecompressionSpeed } HCfavor_e; LZ4_FORCE_INLINE int LZ4HC_InsertAndGetWiderMatch ( @@ -211,7 +212,8 @@ LZ4HC_InsertAndGetWiderMatch ( const BYTE** startpos, const int maxNbAttempts, const int patternAnalysis, - const dictCtx_directive dict) + const dictCtx_directive dict, + const HCfavor_e favorDecSpeed) { U16* const chainTable = hc4->chainTable; U32* const HashTable = hc4->hashTable; @@ -240,7 +242,10 @@ LZ4HC_InsertAndGetWiderMatch ( while ((matchIndex>=lowLimit) && (nbAttempts)) { DEBUGLOG(7, "remaining attempts : %i", nbAttempts); nbAttempts--; - if (matchIndex >= dictLimit) { + assert(matchIndex < ipIndex); + if (favorDecSpeed && (ipIndex - matchIndex < 8)) { + /* do nothing */ + } else if (matchIndex >= dictLimit) { const BYTE* const matchPtr = base + matchIndex; assert(longest >= 1); if (LZ4_read16(iLowLimit + longest - 1) == LZ4_read16(matchPtr - delta + longest - 1)) { @@ -326,14 +331,12 @@ LZ4HC_InsertAndGetWiderMatch ( } } - { - U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex); + { U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex); dictMatchIndex -= nextOffset; matchIndex -= nextOffset; } } } - return longest; } @@ -349,7 +352,7 @@ int LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4, /* Index tabl /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos), * but this won't be the case here, as we define iLowLimit==ip, * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */ - return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, matchpos, &uselessPtr, maxNbAttempts, patternAnalysis, dict); + return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, matchpos, &uselessPtr, maxNbAttempts, patternAnalysis, dict, favorCompressionRatio); } @@ -484,7 +487,7 @@ _Search2: if (ip+ml <= mflimit) ml2 = LZ4HC_InsertAndGetWiderMatch(ctx, ip + ml - 2, ip + 0, matchlimit, ml, &ref2, &start2, - maxNbAttempts, patternAnalysis, dict); + maxNbAttempts, patternAnalysis, dict, favorCompressionRatio); else ml2 = ml; @@ -531,7 +534,7 @@ _Search3: if (start2 + ml2 <= mflimit) ml3 = LZ4HC_InsertAndGetWiderMatch(ctx, start2 + ml2 - 3, start2, matchlimit, ml2, &ref3, &start3, - maxNbAttempts, patternAnalysis, dict); + maxNbAttempts, patternAnalysis, dict, favorCompressionRatio); else ml3 = ml2; @@ -651,12 +654,14 @@ _dest_overflow: return 0; } + static int LZ4HC_compress_optimal( LZ4HC_CCtx_internal* ctx, const char* const source, char* dst, int* srcSizePtr, int dstCapacity, int const nbSearches, size_t sufficient_len, const limitedOutput_directive limit, int const fullUpdate, - const dictCtx_directive dict); + const dictCtx_directive dict, + HCfavor_e favorDecSpeed); LZ4_FORCE_INLINE int LZ4HC_compress_generic_internal ( @@ -711,7 +716,9 @@ LZ4_FORCE_INLINE int LZ4HC_compress_generic_internal ( return LZ4HC_compress_optimal(ctx, src, dst, srcSizePtr, dstCapacity, cParam.nbSearches, cParam.targetLength, limit, - cLevel == LZ4HC_CLEVEL_MAX, dict); /* ultra mode */ + cLevel == LZ4HC_CLEVEL_MAX, /* ultra mode */ + dict, + favorDecompressionSpeed); } } @@ -1082,23 +1089,26 @@ LZ4_FORCE_INLINE LZ4HC_match_t LZ4HC_FindLongerMatch(LZ4HC_CCtx_internal* const ctx, const BYTE* ip, const BYTE* const iHighLimit, int minLen, int nbSearches, - const dictCtx_directive dict) + const dictCtx_directive dict, + const HCfavor_e favorDecSpeed) { LZ4HC_match_t match = { 0 , 0 }; const BYTE* matchPtr = NULL; /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos), * but this won't be the case here, as we define iLowLimit==ip, * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */ - int const matchLength = LZ4HC_InsertAndGetWiderMatch(ctx, + int matchLength = LZ4HC_InsertAndGetWiderMatch(ctx, ip, ip, iHighLimit, minLen, &matchPtr, &ip, - nbSearches, 1 /* patternAnalysis */, dict); + nbSearches, 1 /* patternAnalysis */, dict, favorDecSpeed); if (matchLength <= minLen) return match; + if (favorDecSpeed) { + if ((matchLength>18) & (matchLength<=36)) matchLength=18; /* favor shortcut */ + } match.len = matchLength; match.off = (int)(ip-matchPtr); return match; } - static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx, const char* const source, @@ -1109,7 +1119,8 @@ static int LZ4HC_compress_optimal ( size_t sufficient_len, const limitedOutput_directive limit, int const fullUpdate, - const dictCtx_directive dict + const dictCtx_directive dict, + HCfavor_e favorDecSpeed ) { #define TRAILING_LITERALS 3 @@ -1125,6 +1136,7 @@ static int LZ4HC_compress_optimal ( BYTE* oend = op + dstCapacity; /* init */ + favorDecSpeed = favorCompressionRatio; DEBUGLOG(5, "LZ4HC_compress_optimal"); *srcSizePtr = 0; if (limit == limitedDestSize) oend -= LASTLITERALS; /* Hack for support LZ4 format restriction */ @@ -1137,7 +1149,7 @@ static int LZ4HC_compress_optimal ( int best_mlen, best_off; int cur, last_match_pos = 0; - LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches, dict); + LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed); if (firstMatch.len==0) { ip++; continue; } if ((size_t)firstMatch.len > sufficient_len) { @@ -1207,10 +1219,10 @@ static int LZ4HC_compress_optimal ( DEBUGLOG(7, "search at rPos:%u", cur); if (fullUpdate) - newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches, dict); + newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed); else /* only test matches of minimum length; slightly faster, but misses a few bytes */ - newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches, dict); + newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches, dict, favorDecSpeed); if (!newMatch.len) continue; if ( ((size_t)newMatch.len > sufficient_len) @@ -1258,7 +1270,10 @@ static int LZ4HC_compress_optimal ( price = opt[cur].price + LZ4HC_sequencePrice(0, ml); } - if (pos > last_match_pos+TRAILING_LITERALS || price <= opt[pos].price) { + assert(opt[pos].price > 1); + assert((U32)favorDecSpeed <= 1); + if (pos > last_match_pos+TRAILING_LITERALS + || price <= opt[pos].price - (int)favorDecSpeed) { DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i)", pos, price, ml); assert(pos < LZ4_OPT_NUM); |