Add long distance matcher

Move last literals section to ZSTD_block_internal

lib/common/zstd_internal.h

@@ -274,6 +274,20 @@ typedef struct {
     const BYTE* cachedLiterals;
 } optState_t;
 
+typedef struct {
+    U32 offset;
+    U32 checksum;
+} ldmEntry_t;
+
+typedef struct {
+    ldmEntry_t* hashTable;
+    BYTE* bucketOffsets;
+    U32 ldmEnable;          /* 1 if enable long distance matching */
+    U32 hashLog;            /* log size of hashTable */
+    U32 bucketLog;          /* log number of buckets, at most 4 */
+    U32 hashEveryLog;
+} ldmState_t;
+
 typedef struct {
     U32 hufCTable[HUF_CTABLE_SIZE_U32(255)];
     FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];

lib/compress/zstd_opt.h

@@ -413,8 +413,9 @@ static U32 ZSTD_BtGetAllMatches_selectMLS_extDict (
 *  Optimal parser
 *********************************/
 FORCE_INLINE
-void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
+size_t ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
-                                    const void* src, size_t srcSize, const int ultra)
+                                      const void* src, size_t srcSize,
+                                      const int ultra)
 {
     seqStore_t* seqStorePtr = &(ctx->seqStore);
     optState_t* optStatePtr = &(ctx->optState);
@@ -654,17 +655,15 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
     /* Save reps for next block */
     { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->repToConfirm[i] = rep[i]; }
 
-    /* Last Literals */
+    /* Return the last literals size */
-    {   size_t const lastLLSize = iend - anchor;
+    return iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
 }
 
 
 FORCE_INLINE
-void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
+size_t ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize, const int ultra)
+                                              const void* src, size_t srcSize,
+                                              const int ultra)
 {
     seqStore_t* seqStorePtr = &(ctx->seqStore);
     optState_t* optStatePtr = &(ctx->optState);
@@ -928,11 +927,8 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
     /* Save reps for next block */
     { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->repToConfirm[i] = rep[i]; }
 
-    /* Last Literals */
+    /* Return the last literals size */
-    {   size_t lastLLSize = iend - anchor;
+    return iend - anchor;
-        memcpy(seqStorePtr->lit, anchor, lastLLSize);
-        seqStorePtr->lit += lastLLSize;
-    }
 }
 
 #endif  /* ZSTD_OPT_H_91842398743 */

lib/zstd.h

@@ -978,7 +978,9 @@ typedef enum {
     /* advanced parameters - may not remain available after API update */
     ZSTD_p_forceMaxWindow=1100, /* Force back-reference distances to remain < windowSize,
                               * even when referencing into Dictionary content (default:0) */
-
+    ZSTD_p_longDistanceMatching,    /* Enable long distance matching.
+                                     * This increases the memory usage as well as the
+                                     * window size. */
 } ZSTD_cParameter;
 
 

programs/bench.c

@@ -129,7 +129,10 @@ void BMK_setNbThreads(unsigned nbThreads) {
 #endif
     g_nbThreads = nbThreads;
 }
-
+static U32 g_ldmFlag = 0;
+void BMK_setLdmFlag(unsigned ldmFlag) {
+    g_ldmFlag = ldmFlag;
+}
 
 /* ********************************************************
 *  Bench functions
@@ -271,6 +274,7 @@ static int BMK_benchMem(const void* srcBuffer, size_t srcSize,
                     ZSTD_CCtx_setParameter(ctx, ZSTD_p_searchLog, comprParams->searchLog);
                     ZSTD_CCtx_setParameter(ctx, ZSTD_p_minMatch, comprParams->searchLength);
                     ZSTD_CCtx_setParameter(ctx, ZSTD_p_targetLength, comprParams->targetLength);
+                    ZSTD_CCtx_setParameter(ctx, ZSTD_p_longDistanceMatching, g_ldmFlag);
                     ZSTD_CCtx_setParameter(ctx, ZSTD_p_compressionStrategy, comprParams->strategy);
                     ZSTD_CCtx_loadDictionary(ctx, dictBuffer, dictBufferSize);
 #else

programs/bench.h

@@ -25,5 +25,6 @@ void BMK_setNbThreads(unsigned nbThreads);
 void BMK_setNotificationLevel(unsigned level);
 void BMK_setAdditionalParam(int additionalParam);
 void BMK_setDecodeOnlyMode(unsigned decodeFlag);
+void BMK_setLdmFlag(unsigned ldmFlag);
 
 #endif   /* BENCH_H_121279284357 */

programs/fileio.c

@@ -213,6 +213,10 @@ void FIO_setOverlapLog(unsigned overlapLog){
         DISPLAYLEVEL(2, "Setting overlapLog is useless in single-thread mode \n");
     g_overlapLog = overlapLog;
 }
+static U32 g_ldmFlag = 0;
+void FIO_setLdmFlag(unsigned ldmFlag) {
+    g_ldmFlag = (ldmFlag>0);
+}
 
 
 /*-*************************************
@@ -407,6 +411,8 @@ static cRess_t FIO_createCResources(const char* dictFileName, int cLevel,
             CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_p_minMatch, comprParams->searchLength) );
             CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_p_targetLength, comprParams->targetLength) );
             CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_p_compressionStrategy, (U32)comprParams->strategy) );
+            /* long distance matching */
+            CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_p_longDistanceMatching, g_ldmFlag) );
             /* multi-threading */
             CHECK( ZSTD_CCtx_setParameter(ress.cctx, ZSTD_p_nbThreads, g_nbThreads) );
             /* dictionary */

programs/fileio.h

@@ -56,6 +56,7 @@ void FIO_setMemLimit(unsigned memLimit);
 void FIO_setNbThreads(unsigned nbThreads);
 void FIO_setBlockSize(unsigned blockSize);
 void FIO_setOverlapLog(unsigned overlapLog);
+void FIO_setLdmFlag(unsigned ldmFlag);
 
 
 /*-*************************************

programs/zstdcli.c

@@ -152,6 +152,7 @@ static int usage_advanced(const char* programName)
 #endif
     DISPLAY( " -M#    : Set a memory usage limit for decompression \n");
     DISPLAY( "--list  : list information about a zstd compressed file \n");
+    DISPLAY( "--long  : enable long distance matching\n");
     DISPLAY( "--      : All arguments after \"--\" are treated as files \n");
 #ifndef ZSTD_NODICT
     DISPLAY( "\n");
@@ -333,7 +334,8 @@ int main(int argCount, const char* argv[])
         ultra=0,
         lastCommand = 0,
         nbThreads = 1,
-        setRealTimePrio = 0;
+        setRealTimePrio = 0,
+        ldmFlag = 0;
     unsigned bench_nbSeconds = 3;   /* would be better if this value was synchronized from bench */
     size_t blockSize = 0;
     zstd_operation_mode operation = zom_compress;
@@ -440,6 +442,7 @@ int main(int argCount, const char* argv[])
 #ifdef ZSTD_LZ4COMPRESS
                     if (!strcmp(argument, "--format=lz4")) { suffix = LZ4_EXTENSION; FIO_setCompressionType(FIO_lz4Compression);  continue; }
 #endif
+                    if (!strcmp(argument, "--long")) { ldmFlag = 1; continue; }
 
                     /* long commands with arguments */
 #ifndef ZSTD_NODICT
@@ -690,6 +693,7 @@ int main(int argCount, const char* argv[])
         BMK_setBlockSize(blockSize);
         BMK_setNbThreads(nbThreads);
         BMK_setNbSeconds(bench_nbSeconds);
+        BMK_setLdmFlag(ldmFlag);
         BMK_benchFiles(filenameTable, filenameIdx, dictFileName, cLevel, cLevelLast, &compressionParams, setRealTimePrio);
 #endif
         (void)bench_nbSeconds;
@@ -757,6 +761,7 @@ int main(int argCount, const char* argv[])
 #ifndef ZSTD_NOCOMPRESS
         FIO_setNbThreads(nbThreads);
         FIO_setBlockSize((U32)blockSize);
+        FIO_setLdmFlag(ldmFlag);
         if (g_overlapLog!=OVERLAP_LOG_DEFAULT) FIO_setOverlapLog(g_overlapLog);
         if ((filenameIdx==1) && outFileName)
           operationResult = FIO_compressFilename(outFileName, filenameTable[0], dictFileName, cLevel, &compressionParams);

tests/fuzzer.c

@@ -440,8 +440,6 @@ static int basicUnitTests(U32 seed, double compressibility)
         free(staticDCtxBuffer);
     }
 
-
-
     /* ZSTDMT simple MT compression test */
     DISPLAYLEVEL(4, "test%3i : create ZSTDMT CCtx : ", testNb++);
     {   ZSTDMT_CCtx* mtctx = ZSTDMT_createCCtx(2);
@@ -1342,6 +1340,7 @@ static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, U32 const maxD
             dictSize = FUZ_rLogLength(&lseed, dictLog);   /* needed also for decompression */
             dict = srcBuffer + (FUZ_rand(&lseed) % (srcBufferSize - dictSize));
 
+
             if (FUZ_rand(&lseed) & 0xF) {
                 CHECK_Z ( ZSTD_compressBegin_usingDict(refCtx, dict, dictSize, cLevel) );
             } else {
@@ -1350,6 +1349,7 @@ static int fuzzerTests(U32 seed, U32 nbTests, unsigned startTest, U32 const maxD
                                                     !(FUZ_rand(&lseed)&3) /* contentChecksumFlag*/,
                                                     0 /*NodictID*/ };   /* note : since dictionary is fake, dictIDflag has no impact */
                 ZSTD_parameters const p = FUZ_makeParams(cPar, fPar);
+
                 CHECK_Z ( ZSTD_compressBegin_advanced(refCtx, dict, dictSize, p, 0) );
             }
             CHECK_Z( ZSTD_copyCCtx(ctx, refCtx, 0) );

tests/playTests.sh

@@ -544,6 +544,15 @@ roundTripTest -g516K 19   # btopt
 
 fileRoundTripTest -g500K
 
+$ECHO "\n**** zstd long distance matching round-trip tests **** "
+roundTripTest -g0 "2 --long"
+roundTripTest -g1000K "1 --long"
+roundTripTest -g517K "6 --long"
+roundTripTest -g516K "16 --long"
+roundTripTest -g518K "19 --long"
+fileRoundTripTest -g5M "3 --long"
+
+
 if [ -n "$hasMT" ]
 then
     $ECHO "\n**** zstdmt round-trip tests **** "
@@ -551,6 +560,9 @@ then
     roundTripTest -g8M "3 -T2"
     roundTripTest -g8000K "2 --threads=2"
     fileRoundTripTest -g4M "19 -T2 -B1M"
+
+    $ECHO "\n**** zstdmt long distance matching round-trip tests **** "
+    roundTripTest -g8M "3 --long -T2"
 else
     $ECHO "\n**** no multithreading, skipping zstdmt tests **** "
 fi
@@ -639,6 +651,15 @@ roundTripTest -g6000000000 -P99 1
 
 fileRoundTripTest -g4193M -P99 1
 
+$ECHO "\n**** zstd long, long distance matching round-trip tests **** "
+roundTripTest -g0 "2 --long"
+roundTripTest -g270000000 "1 --long"
+roundTripTest -g140000000 -P60 "5 --long"
+roundTripTest -g70000000 -P70 "8 --long"
+roundTripTest -g18000001 -P80  "18 --long"
+fileRoundTripTest -g4100M -P99 "1 --long"
+
+
 if [ -n "$hasMT" ]
 then
     $ECHO "\n**** zstdmt long round-trip tests **** "
@@ -646,6 +667,7 @@ then
     roundTripTest -g6000000000 -P99 "1 -T2"
     roundTripTest -g1500000000 -P97 "1 -T999"
     fileRoundTripTest -g4195M -P98 " -T0"
+    roundTripTest -g1500000000 -P97 "1 --long -T999"
 else
     $ECHO "\n**** no multithreading, skipping zstdmt tests **** "
 fi

tests/zstreamtest.c

@@ -1380,6 +1380,8 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double
                 if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_minMatch, cParams.searchLength, useOpaqueAPI) );
                 if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_targetLength, cParams.targetLength, useOpaqueAPI) );
 
+                if (FUZ_rand(&lseed) & 1) CHECK_Z( ZSTD_CCtx_setParameter(zc, ZSTD_p_longDistanceMatching, FUZ_rand(&lseed) & 63) );
+
                 /* unconditionally set, to be sync with decoder */
                 /* mess with frame parameters */
                 if (FUZ_rand(&lseed) & 1) CHECK_Z( setCCtxParameter(zc, cctxParams, ZSTD_p_checksumFlag, FUZ_rand(&lseed) & 1, useOpaqueAPI) );