diff --git a/lib/zstd_compress.c b/lib/zstd_compress.c
index 07c72f971..95cd8ff2d 100644
--- a/lib/zstd_compress.c
+++ b/lib/zstd_compress.c
@@ -1060,8 +1060,8 @@ void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* zc,
         if (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)) {   /* note : by construction, offset_1 <= current */
             mlCode = ZSTD_count(ip+1+MINMATCH, ip+1+MINMATCH-offset_1, iend);
             ip++;
-            offset = 0;
-        } else {
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mlCode);
+       } else {
             if ( (matchIndex <= lowIndex) ||
                  (MEM_read32(match) != MEM_read32(ip)) ) {
                 ip += ((ip-anchor) >> g_searchStrength) + 1;
@@ -1072,10 +1072,11 @@ void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* zc,
             while ((ip>anchor) && (match>lowest) && (ip[-1] == match[-1])) { ip--; match--; mlCode++; }  /* catch up */
             offset_2 = offset_1;
             offset_1 = offset;
+
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mlCode);
         }
 
         /* match found */
-        ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset, mlCode);
         ip += mlCode + MINMATCH;
         anchor = ip;
 
@@ -1171,7 +1172,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
             const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
             mlCode = ZSTD_count_2segments(ip+1+MINMATCH, repMatch+MINMATCH, iend, repMatchEnd, lowPrefixPtr);
             ip++;
-            offset = 0;
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mlCode);
         } else {
             if ( (matchIndex < lowLimit) ||
                  (MEM_read32(match) != MEM_read32(ip)) ) {
@@ -1185,10 +1186,10 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
                 offset = current - matchIndex;
                 offset_2 = offset_1;
                 offset_1 = offset;
+                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mlCode);
         }   }
 
         /* found a match : store it */
-        ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset, mlCode);
         ip += mlCode + MINMATCH;
         anchor = ip;
 
@@ -1632,7 +1633,6 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
     const BYTE* const ilimit = iend - 8;
     const BYTE* const base = ctx->base + ctx->dictLimit;
 
-    size_t offset_2=REPCODE_STARTVALUE, offset_1=REPCODE_STARTVALUE;
     const U32 maxSearches = 1 << ctx->params.cParams.searchLog;
     const U32 mls = ctx->params.cParams.searchLength;
 
@@ -1642,6 +1642,10 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
     searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
 
     /* init */
+    U32 rep[ZSTD_REP_INIT];
+    for (int i=0; i<ZSTD_REP_INIT; i++)
+        rep[i]=REPCODE_STARTVALUE;
+
     ZSTD_resetSeqStore(seqStorePtr);
     if ((ip-base) < REPCODE_STARTVALUE) ip = base + REPCODE_STARTVALUE;
 
@@ -1649,20 +1653,30 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
     while (ip < ilimit) {
         size_t matchLength=0;
         size_t offset=0;
-        const BYTE* start=ip+1;
+        const BYTE* start=ip;
 
         /* check repCode */
-        if (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1)) {
+        for (int i=0; i<ZSTD_REP_NUM; i++)
+        if (MEM_read32(ip) == MEM_read32(ip - rep[i])) {
             /* repcode : we take it */
-            matchLength = ZSTD_count(ip+1+MINMATCH, ip+1+MINMATCH-offset_1, iend) + MINMATCH;
-            if (depth==0) goto _storeSequence;
+            if (matchLength==0) {
+                matchLength = ZSTD_count(ip+MINMATCH, ip+MINMATCH-rep[i], iend) + MINMATCH;
+                offset = i;
+                if (depth==0) goto _storeSequence;
+            } else {
+                size_t mlRep = ZSTD_count(ip+MINMATCH, ip+MINMATCH-rep[i], iend) + MINMATCH;
+                int gain2 = (int)(mlRep * 3 /*- ZSTD_highbit((U32)i+1)*/ + (i==1));
+                int gain1 = (int)(matchLength*3 - /*ZSTD_highbit((U32)offset+1)*/ + 1 + (offset==1));
+                if (gain2 > gain1)
+                    matchLength = mlRep, offset = i;
+            }
         }
 
         /* first search (depth 0) */
         {   size_t offsetFound = 99999999;
             size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
             if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offset=offsetFound;
+                matchLength = ml2, start = ip, offset=offsetFound + ZSTD_REP_MOVE;
         }
 
         if (matchLength < MINMATCH) {
@@ -1674,70 +1688,81 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
         if (depth>=1)
         while (ip<ilimit) {
             ip ++;
-            if ((offset) && (MEM_read32(ip) == MEM_read32(ip - offset_1))) {
-                size_t const mlRep = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_1, iend) + MINMATCH;
+            for (int i=0; i<ZSTD_REP_NUM; i++)
+            if (MEM_read32(ip) == MEM_read32(ip - rep[i])) {
+                size_t const mlRep = ZSTD_count(ip+MINMATCH, ip+MINMATCH-rep[i], iend) + MINMATCH;
                 int const gain2 = (int)(mlRep * 3);
-                int const gain1 = (int)(matchLength*3 - ZSTD_highbit((U32)offset+1) + 1);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit((U32)offset+1) + 1 + (offset<ZSTD_REP_NUM));
                 if ((mlRep >= MINMATCH) && (gain2 > gain1))
-                    matchLength = mlRep, offset = 0, start = ip;
+                    matchLength = mlRep, offset = i, start = ip;
             }
             {   size_t offset2=99999999;
                 size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
                 int const gain2 = (int)(ml2*4 - ZSTD_highbit((U32)offset2+1));   /* raw approx */
                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 4);
                 if ((ml2 >= MINMATCH) && (gain2 > gain1)) {
-                    matchLength = ml2, offset = offset2, start = ip;
+                    matchLength = ml2, offset = offset2 + ZSTD_REP_MOVE, start = ip;
                     continue;   /* search a better one */
             }   }
 
             /* let's find an even better one */
             if ((depth==2) && (ip<ilimit)) {
                 ip ++;
-                if ((offset) && (MEM_read32(ip) == MEM_read32(ip - offset_1))) {
-                    size_t const ml2 = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_1, iend) + MINMATCH;
+                for (int i=0; i<ZSTD_REP_NUM; i++)
+                if (MEM_read32(ip) == MEM_read32(ip - rep[i])) {
+                    size_t const ml2 = ZSTD_count(ip+MINMATCH, ip+MINMATCH-rep[i], iend) + MINMATCH;
                     int const gain2 = (int)(ml2 * 4);
-                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 1);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 1 + (offset<ZSTD_REP_NUM));
                     if ((ml2 >= MINMATCH) && (gain2 > gain1))
-                        matchLength = ml2, offset = 0, start = ip;
+                        matchLength = ml2, offset = i, start = ip;
                 }
                 {   size_t offset2=99999999;
                     size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
                     int const gain2 = (int)(ml2*4 - ZSTD_highbit((U32)offset2+1));   /* raw approx */
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 7);
                     if ((ml2 >= MINMATCH) && (gain2 > gain1)) {
-                        matchLength = ml2, offset = offset2, start = ip;
+                        matchLength = ml2, offset = offset2 + ZSTD_REP_MOVE, start = ip;
                         continue;
             }   }   }
             break;  /* nothing found : store previous solution */
         }
 
         /* catch up */
-        if (offset) {
-            while ((start>anchor) && (start>base+offset) && (start[-1] == start[-1-offset]))   /* only search for offset within prefix */
+        if (offset >= ZSTD_REP_NUM) {
+            while ((start>anchor) && (start>base+offset-ZSTD_REP_MOVE) && (start[-1] == start[-1-offset+ZSTD_REP_MOVE]))   /* only search for offset within prefix */
                 { start--; matchLength++; }
-            offset_2 = offset_1; offset_1 = offset;
         }
 
         /* store sequence */
 _storeSequence:
-        {   size_t const litLength = start - anchor;
+        {
+            if (offset >= ZSTD_REP_NUM) {
+#if ZSTD_REP_NUM > 3
+                rep[3] = rep[2];
+#endif
+                rep[2] = rep[1];
+                rep[1] = rep[0];
+                rep[0] = offset - ZSTD_REP_MOVE;
+            } else {
+                if (offset != 0) {
+                    size_t temp = rep[offset];
+#if ZSTD_REP_NUM > 3
+                    if (offset > 2) rep[3] = rep[2];
+#endif
+                    if (offset > 1) rep[2] = rep[1];
+                    if (offset > 0) rep[1] = rep[0];
+                    rep[0] = temp;
+                }
+
+                if (offset<=1 && start==anchor) offset = 1-offset;
+            }
+
+            size_t const litLength = start - anchor;
+
             ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, matchLength-MINMATCH);
             anchor = ip = start + matchLength;
         }
-
-        /* check immediate repcode */
-        while ( (ip <= ilimit)
-             && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
-            /* store sequence */
-            matchLength = ZSTD_count(ip+MINMATCH, ip+MINMATCH-offset_2, iend);
-            offset = offset_2;
-            offset_2 = offset_1;
-            offset_1 = offset;
-            ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength);
-            ip += matchLength+MINMATCH;
-            anchor = ip;
-            continue;   /* faster when present ... (?) */
-    }   }
+    }
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1792,7 +1817,6 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
     const BYTE* const dictEnd  = dictBase + dictLimit;
     const BYTE* const dictStart  = dictBase + ctx->lowLimit;
 
-    size_t offset_2=REPCODE_STARTVALUE, offset_1=REPCODE_STARTVALUE;
     const U32 maxSearches = 1 << ctx->params.cParams.searchLog;
     const U32 mls = ctx->params.cParams.searchLength;
 
@@ -1801,7 +1825,12 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                         U32 maxNbAttempts, U32 matchLengthSearch);
     searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
 
+    printf("ZSTD_compressBlock_lazy_extDict_generic reps not implemented!\n"); exit(1);
+
     /* init */
+    U32 rep[ZSTD_REP_INIT];
+    for (int i=0; i<ZSTD_REP_INIT; i++)
+        rep[i]=REPCODE_STARTVALUE;
     ZSTD_resetSeqStore(seqStorePtr);
     if ((ip - prefixStart) < REPCODE_STARTVALUE) ip += REPCODE_STARTVALUE;
 
@@ -1814,7 +1843,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
 
         /* check repCode */
         {
-            const U32 repIndex = (U32)(current+1 - offset_1);
+            const U32 repIndex = (U32)(current+1 - rep[0]);
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
             if ((U32)((dictLimit-1) - repIndex) >= 3)   /* intentional overflow */
@@ -1829,7 +1858,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
         {   size_t offsetFound = 99999999;
             size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
             if (ml2 > matchLength)
-                matchLength = ml2, start = ip, offset=offsetFound;
+                matchLength = ml2, start = ip, offset=offsetFound + ZSTD_REP_MOVE;
         }
 
          if (matchLength < MINMATCH) {
@@ -1843,8 +1872,8 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
             ip ++;
             current++;
             /* check repCode */
-            if (offset) {
-                const U32 repIndex = (U32)(current - offset_1);
+            if (offset >= ZSTD_REP_NUM) {
+                const U32 repIndex = (U32)(current - rep[0]);
                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                 const BYTE* const repMatch = repBase + repIndex;
                 if ((U32)((dictLimit-1) - repIndex) >= 3)   /* intentional overflow */
@@ -1864,7 +1893,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 int const gain2 = (int)(ml2*4 - ZSTD_highbit((U32)offset2+1));   /* raw approx */
                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 4);
                 if ((ml2 >= MINMATCH) && (gain2 > gain1)) {
-                    matchLength = ml2, offset = offset2, start = ip;
+                    matchLength = ml2, offset = offset2 + ZSTD_REP_MOVE, start = ip;
                     continue;   /* search a better one */
             }   }
 
@@ -1873,8 +1902,8 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 ip ++;
                 current++;
                 /* check repCode */
-                if (offset) {
-                    const U32 repIndex = (U32)(current - offset_1);
+                if (offset >= ZSTD_REP_NUM) {
+                    const U32 repIndex = (U32)(current - rep[0]);
                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                     const BYTE* const repMatch = repBase + repIndex;
                     if ((U32)((dictLimit-1) - repIndex) >= 3)   /* intentional overflow */
@@ -1894,19 +1923,19 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                     int const gain2 = (int)(ml2*4 - ZSTD_highbit((U32)offset2+1));   /* raw approx */
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit((U32)offset+1) + 7);
                     if ((ml2 >= MINMATCH) && (gain2 > gain1)) {
-                        matchLength = ml2, offset = offset2, start = ip;
+                        matchLength = ml2, offset = offset2 + ZSTD_REP_MOVE, start = ip;
                         continue;
             }   }   }
             break;  /* nothing found : store previous solution */
         }
 
         /* catch up */
-        if (offset) {
-            U32 matchIndex = (U32)((start-base) - offset);
+        if (offset >= ZSTD_REP_NUM) {
+            U32 matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
             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 = offset;
+            rep[1] = rep[0]; rep[0] = offset - ZSTD_REP_MOVE;
         }
 
         /* store sequence */
@@ -1918,7 +1947,7 @@ _storeSequence:
 
         /* check immediate repcode */
         while (ip <= ilimit) {
-            const U32 repIndex = (U32)((ip-base) - offset_2);
+            const U32 repIndex = (U32)((ip-base) - rep[1]);
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
             if ((U32)((dictLimit-1) - repIndex) >= 3)   /* intentional overflow */
@@ -1926,7 +1955,7 @@ _storeSequence:
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                 matchLength = ZSTD_count_2segments(ip+MINMATCH, repMatch+MINMATCH, iend, repEnd, prefixStart) + MINMATCH;
-                offset = offset_2; offset_2 = offset_1; offset_1 = offset;   /* swap offset history */
+                offset = rep[1]; rep[1] = rep[0]; rep[0] = offset;   /* swap offset history */
                 ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
                 ip += matchLength;
                 anchor = ip;
@@ -2117,7 +2146,7 @@ size_t ZSTD_compressBlock(ZSTD_CCtx* zc, void* dst, size_t dstCapacity, const vo
 {
     if (srcSize > ZSTD_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
     zc->params.cParams.searchLength = MINMATCH; /* force ZSTD_btopt to MINMATCH in block mode */
-    ZSTD_LOG_BLOCK("%p: ZSTD_compressBlock searchLength=%d\n", zc->base, zc->params.searchLength);
+    ZSTD_LOG_BLOCK("%p: ZSTD_compressBlock searchLength=%d\n", zc->base, zc->params.cParams.searchLength);
     return ZSTD_compressContinue_internal(zc, dst, dstCapacity, src, srcSize, 0);
 }
 
diff --git a/lib/zstd_decompress.c b/lib/zstd_decompress.c
index 436058b52..037b6cb2e 100644
--- a/lib/zstd_decompress.c
+++ b/lib/zstd_decompress.c
@@ -621,10 +621,11 @@ typedef struct {
     FSE_DState_t stateLL;
     FSE_DState_t stateOffb;
     FSE_DState_t stateML;
-    size_t prevOffset;
+    size_t prevOffset[ZSTD_REP_INIT];
 } seqState_t;
 
 
+
 static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState, const U32 mls)
 {
     /* Literal length */
@@ -655,10 +656,40 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState, const U32 mls)
                  0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF, /*fake*/ 1, 1, 1, 1, 1 };
 
     /* sequence */
-    {   size_t const offset = ofCode ? OF_base[ofCode] + BIT_readBits(&(seqState->DStream), ofBits) :   /* <=  26 bits */
-                                       llCode ? seq->offset : seqState->prevOffset;
-        if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
-        if (ofCode | !llCode) seqState->prevOffset = seq->offset;   /* cmove */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            offset = OF_base[ofCode] + BIT_readBits(&(seqState->DStream), ofBits);   /* <=  26 bits */
+            if (MEM_32bits()) BIT_reloadDStream(&(seqState->DStream));
+        }
+        
+        if (offset < ZSTD_REP_NUM) {
+            if (llCode == 0 && offset <= 1) offset = 1-offset;
+
+            if (offset != 0) {
+                size_t temp = seqState->prevOffset[offset];
+                if (offset != 1) {
+#if ZSTD_REP_NUM > 3
+                    if (offset == 3) seqState->prevOffset[3] = seqState->prevOffset[2];
+#endif
+                    seqState->prevOffset[2] = seqState->prevOffset[1];
+                }
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            offset -= ZSTD_REP_MOVE;
+#if ZSTD_REP_NUM > 3
+            seqState->prevOffset[3] = seqState->prevOffset[2];
+#endif
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];               
+            seqState->prevOffset[0] = offset;
+        }
         seq->offset = offset;
     }
 
@@ -785,7 +816,8 @@ static size_t ZSTD_decompressSequences(
 
         memset(&sequence, 0, sizeof(sequence));
         sequence.offset = REPCODE_STARTVALUE;
-        seqState.prevOffset = REPCODE_STARTVALUE;
+        for (int i=0; i<ZSTD_REP_INIT; i++)
+            seqState.prevOffset[i] = REPCODE_STARTVALUE;
         { size_t const errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
           if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
         FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
diff --git a/lib/zstd_internal.h b/lib/zstd_internal.h
index ce36aad3b..e7f07f9b7 100644
--- a/lib/zstd_internal.h
+++ b/lib/zstd_internal.h
@@ -65,6 +65,10 @@
 #define ZSTD_OPT_NUM    (1<<12)
 #define ZSTD_DICT_MAGIC  0xEC30A435
 
+#define ZSTD_REP_NUM    3
+#define ZSTD_REP_INIT   ZSTD_REP_NUM
+#define ZSTD_REP_MOVE   (ZSTD_REP_NUM-1)
+
 #define KB *(1 <<10)
 #define MB *(1 <<20)
 #define GB *(1U<<30)
@@ -188,8 +192,7 @@ typedef struct {
     U32 off;
     U32 mlen;
     U32 litlen;
-    U32 rep;
-    U32 rep2;
+    U32 rep[ZSTD_REP_INIT];
 } ZSTD_optimal_t;
 
 #if ZSTD_OPT_DEBUG == 3
diff --git a/lib/zstd_opt.h b/lib/zstd_opt.h
index 145570031..003823230 100644
--- a/lib/zstd_opt.h
+++ b/lib/zstd_opt.h
@@ -284,7 +284,7 @@ static U32 ZSTD_insertBtAndGetAllMatches (
             /* save best solution */
             if (currentMl > bestLength) {
                 bestLength = currentMl;
-                matches[mnum].off = current - matchIndex3;
+                matches[mnum].off = ZSTD_REP_MOVE + current - matchIndex3;
                 matches[mnum].len = (U32)currentMl;
                 mnum++;
                 if (currentMl > ZSTD_OPT_NUM) goto update;
@@ -329,7 +329,7 @@ static U32 ZSTD_insertBtAndGetAllMatches (
         if (matchLength > bestLength) {
             if (matchLength > matchEndIdx - matchIndex) matchEndIdx = matchIndex + (U32)matchLength;
             bestLength = matchLength;
-            matches[mnum].off = current - matchIndex;
+            matches[mnum].off = ZSTD_REP_MOVE + current - matchIndex;
             matches[mnum].len = (U32)matchLength;
             mnum++;
             if (matchLength > ZSTD_OPT_NUM) break;
@@ -434,7 +434,6 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
     const BYTE* const base = ctx->base;
     const BYTE* const prefixStart = base + ctx->dictLimit;
 
-    U32 rep_2=REPCODE_STARTVALUE, rep_1=REPCODE_STARTVALUE;
     const U32 maxSearches = 1U << ctx->params.cParams.searchLog;
     const U32 sufficient_len = ctx->params.cParams.targetLength;
     const U32 mls = ctx->params.cParams.searchLength;
@@ -446,6 +445,10 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
     U32 cur, match_num, last_pos, litlen, price;
 
     /* init */
+    U32 rep[ZSTD_REP_INIT];
+    for (int i=0; i<ZSTD_REP_INIT; i++)
+        rep[i]=REPCODE_STARTVALUE;
+
     ctx->nextToUpdate3 = ctx->nextToUpdate;
     ZSTD_resetSeqStore(seqStorePtr);
     ZSTD_rescaleFreqs(seqStorePtr);
@@ -466,21 +469,23 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
         opt[0].litlen = (U32)(ip - litstart);
 
         /* check repCode */
-        if (MEM_readMINMATCH(ip+1, minMatch) == MEM_readMINMATCH(ip+1 - rep_1, minMatch)) {
+        for (int i=0; i<ZSTD_REP_NUM; i++)
+        if (MEM_readMINMATCH(ip, minMatch) == MEM_readMINMATCH(ip - rep[i], minMatch)) {
             /* repcode : we take it */
-            mlen = (U32)ZSTD_count(ip+1+minMatch, ip+1+minMatch-rep_1, iend) + minMatch;
+            mlen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-rep[i], iend) + minMatch;
+            best_off = (i<=1 && ip == anchor) ? 1-i : i;
 
-            ZSTD_LOG_PARSER("%d: start try REP rep=%d mlen=%d\n", (int)(ip-base), (int)rep_1, (int)mlen);
+            ZSTD_LOG_PARSER("%d: start try REP rep=%d mlen=%d\n", (int)(ip-base), (int)rep[i], (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;
+                best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
                 goto _storeSequence;
             }
 
-            litlen = opt[0].litlen + 1;
+            litlen = opt[0].litlen;
             do {
-                price = ZSTD_getPrice(seqStorePtr, litlen, litstart, 0, mlen - minMatch);
-                if (mlen + 1 > last_pos || price < opt[mlen + 1].price)
-                    SET_PRICE(mlen + 1, mlen, 0, litlen, price);   /* note : macro modifies last_pos */
+                price = ZSTD_getPrice(seqStorePtr, litlen, litstart, best_off, mlen - minMatch);
+                if (mlen > last_pos || price < opt[mlen].price)
+                    SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
                 mlen--;
             } while (mlen >= minMatch);
         }
@@ -490,8 +495,8 @@ 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;
+        for (int i=0; i<ZSTD_REP_INIT; i++)
+            opt[0].rep[i] = rep[i];
         opt[0].mlen = 1;
 
         if (match_num && matches[match_num-1].len > sufficient_len) {
@@ -521,8 +526,8 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
 
          /* check further positions */
         for (cur = 1; cur <= last_pos; cur++) {
-           size_t cur_rep;
            inr = ip + cur;
+           ZSTD_LOG_PARSER("%d: START_NoExt price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep[1]=%d\n", (int)(inr-base), cur, last_pos, opt[cur].price, opt[cur].off, opt[cur].mlen, opt[cur].litlen, opt[cur].rep[0], opt[cur].rep[1]);
 
            if (opt[cur-1].mlen == 1) {
                 litlen = opt[cur-1].litlen + 1;
@@ -545,40 +550,38 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
 
            mlen = opt[cur].mlen;
 
-           if (opt[cur].off) {
-                opt[cur].rep2 = opt[cur-mlen].rep;
-                opt[cur].rep = opt[cur].off;
-                ZSTD_LOG_ENCODE("%d: COPYREP_OFF cur=%d mlen=%d rep=%d rep2=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep, opt[cur].rep2);
+           if (opt[cur].off >= ZSTD_REP_NUM) {
+#if ZSTD_REP_NUM > 3
+                opt[cur].rep[3] = opt[cur-mlen].rep[2];
+#endif
+                opt[cur].rep[2] = opt[cur-mlen].rep[1];
+                opt[cur].rep[1] = opt[cur-mlen].rep[0];
+                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE;               
+                ZSTD_LOG_ENCODE("%d: COPYREP_OFF cur=%d mlen=%d rep=%d rep[1]=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep[0], opt[cur].rep[1]);
            } else {
-                if (cur!=mlen && opt[cur].litlen == 0) {
-                    opt[cur].rep2 = opt[cur-mlen].rep;
-                    opt[cur].rep = opt[cur-mlen].rep2;
-                    ZSTD_LOG_ENCODE("%d: COPYREP_SWI 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_ENCODE("%d: COPYREP_NOR cur=%d mlen=%d rep=%d rep2=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep, opt[cur].rep2);
-           }   }
+#if ZSTD_REP_NUM > 3
+                opt[cur].rep[3] = (opt[cur].off > 2) ? opt[cur-mlen].rep[2] : opt[cur-mlen].rep[3];
+#endif
+                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
+                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
+                opt[cur].rep[0] = opt[cur-mlen].rep[opt[cur].off];
+                ZSTD_LOG_ENCODE("%d: COPYREP_NOR cur=%d mlen=%d rep=%d rep[1]=%d\n", (int)(inr-base), cur, mlen, opt[cur].rep[0], opt[cur].rep[1]);
+           }
 
-           ZSTD_LOG_PARSER("%d: CURRENT_NoExt 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);
+           ZSTD_LOG_PARSER("%d: CURRENT_NoExt price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep[1]=%d\n", (int)(inr-base), cur, last_pos, opt[cur].price, opt[cur].off, opt[cur].mlen, opt[cur].litlen, opt[cur].rep[0], opt[cur].rep[1]);
 
            best_mlen = 0;
 
-           if (opt[cur].mlen != 1) {
-               cur_rep = opt[cur].rep2;
-               ZSTD_LOG_PARSER("%d: tryNoExt REP2 rep2=%u mlen=%u\n", (int)(inr-base), (U32)cur_rep, mlen);
-           } else {
-               cur_rep = opt[cur].rep;
-               ZSTD_LOG_PARSER("%d: tryNoExt REP1 rep=%u mlen=%u\n", (int)(inr-base), (U32)cur_rep, mlen);
-           }
+           for (int i=0; i<ZSTD_REP_NUM; i++)
+           if (MEM_readMINMATCH(inr, minMatch) == MEM_readMINMATCH(inr - opt[cur].rep[i], minMatch)) {  // check rep
+               mlen = (U32)ZSTD_count(inr+minMatch, inr+minMatch - opt[cur].rep[i], iend) + minMatch;
+               ZSTD_LOG_PARSER("%d: Found REP %d/%d mlen=%d off=%d rep=%d opt[%d].off=%d\n", (int)(inr-base), i, ZSTD_REP_NUM, mlen, i, opt[cur].rep[i], cur, opt[cur].off);
 
-           if (MEM_readMINMATCH(inr, minMatch) == MEM_readMINMATCH(inr - cur_rep, minMatch)) {  // check rep
-               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);
+               best_off = (i<=1 && opt[cur].mlen != 1) ? 1-i : i;
 
                if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
-                    best_mlen = mlen;
-                    best_off = 0;
+                    best_mlen = mlen;     
+                    best_off = i;
                     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;
@@ -587,24 +590,25 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                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);
+                        price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - minMatch);
                     } else
-                        price = ZSTD_getPrice(seqStorePtr, litlen, litstart, 0, mlen - minMatch);
+                        price = ZSTD_getPrice(seqStorePtr, litlen, litstart, best_off, mlen - minMatch);
                 } else {
                     litlen = 0;
-                    price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, 0, mlen - minMatch);
+                    price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - minMatch);
                 }
 
                 best_mlen = mlen;
-                 ZSTD_LOG_PARSER("%d: Found REP mlen=%d off=%d price=%d litlen=%d\n", (int)(inr-base), mlen, 0, price, litlen);
+                ZSTD_LOG_PARSER("%d: Found REP mlen=%d off=%d price=%d litlen=%d\n", (int)(inr-base), mlen, best_off, price, litlen);
 
                 do {
                     if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
-                        SET_PRICE(cur + mlen, mlen, 0, litlen, price);
+                        SET_PRICE(cur + mlen, mlen, i, litlen, price);
                     mlen--;
                 } while (mlen >= minMatch);
             }
 
+
             match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, inr, iend, maxSearches, mls, matches);
             ZSTD_LOG_PARSER("%d: ZSTD_GetAllMatches match_num=%d\n", (int)(inr-base), match_num);
 
@@ -651,8 +655,8 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
         /* store sequence */
 _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
         for (u = 1; u <= last_pos; u++)
-            ZSTD_LOG_PARSER("%d: price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+u), u, last_pos, opt[u].price, opt[u].off, opt[u].mlen, opt[u].litlen, opt[u].rep, opt[u].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);
+            ZSTD_LOG_PARSER("%d: price[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep[1]=%d\n", (int)(ip-base+u), u, last_pos, opt[u].price, opt[u].off, opt[u].mlen, opt[u].litlen, opt[u].rep[0], opt[u].rep[1]);
+        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[0]);
 
         opt[0].mlen = 1;
         U32 offset;
@@ -669,38 +673,51 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
         }
 
         for (u = 0; u <= last_pos;) {
-            ZSTD_LOG_PARSER("%d: price2[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep2=%d\n", (int)(ip-base+u), u, last_pos, opt[u].price, opt[u].off, opt[u].mlen, opt[u].litlen, opt[u].rep, opt[u].rep2);
+            ZSTD_LOG_PARSER("%d: price2[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep[1]=%d\n", (int)(ip-base+u), u, last_pos, opt[u].price, opt[u].off, opt[u].mlen, opt[u].litlen, opt[u].rep[0], opt[u].rep[1]);
             u += opt[u].mlen;
         }
 
         for (cur=0; 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);
+            ZSTD_LOG_PARSER("%d: price3[%d/%d]=%d off=%d mlen=%d litlen=%d rep=%d rep[1]=%d\n", (int)(ip-base+cur), cur, last_pos, opt[cur].price, opt[cur].off, opt[cur].mlen, opt[cur].litlen, opt[cur].rep[0], opt[cur].rep[1]);
             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: ENCODE 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);
+            ZSTD_LOG_ENCODE("%d/%d: ENCODE literals=%d mlen=%d off=%d rep1=%d rep[1]=%d\n", (int)(ip-base), (int)(iend-base), (int)(litLength), (int)mlen, (int)(offset), (int)rep[0], (int)rep[1]);
 
-            if (offset) {
-                rep_2 = rep_1;
-                rep_1 = offset;
+            if (offset >= ZSTD_REP_NUM) {
+#if ZSTD_REP_NUM > 3
+                rep[3] = rep[2];
+#endif
+                rep[2] = rep[1];
+                rep[1] = rep[0];               
+                rep[0] = offset - ZSTD_REP_MOVE;               
             } else {
-                if (litLength == 0) {
-                    best_off = rep_2;
-                    rep_2 = rep_1;
-                    rep_1 = best_off;
-            }   }
+                if (offset != 0) {
+                    size_t temp = rep[offset];
+                    if (offset != 1) {
+#if ZSTD_REP_NUM > 3
+                        if (offset == 3) rep[3] = rep[2];
+#endif
+                        rep[2] = rep[1];
+                    }
+                    rep[1] = rep[0];
+                    rep[0] = temp;
+                }
+                
+                if (litLength == 0 && offset<=1) offset = 1-offset;
+           }
 
-           // 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);
+           // ZSTD_LOG_ENCODE("%d/%d: ENCODE2 literals=%d mlen=%d off=%d rep1=%d rep[1]=%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
             U32 ml2;
             if (offset)
                 ml2 = (U32)ZSTD_count(ip, ip-offset, iend);
             else
-                ml2 = (U32)ZSTD_count(ip, ip-rep_1, iend);
+                ml2 = (U32)ZSTD_count(ip, ip-rep[0], iend);
             if ((offset >= 8) && (ml2 < mlen || ml2 < minMatch)) {
                 printf("%d: ERROR_NoExt 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) {
@@ -713,22 +730,6 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
             ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-minMatch);
             anchor = ip = ip + mlen;
         }   /* for (cur=0; cur < last_pos; ) */
-
-        /* check immediate repcode */
-        while ((anchor >= prefixStart + rep_2) && (anchor <= ilimit)
-             && (MEM_readMINMATCH(anchor, minMatch) == MEM_readMINMATCH(anchor - rep_2, minMatch)) ) {
-            /* store sequence */
-            best_mlen = (U32)ZSTD_count(anchor+minMatch, anchor+minMatch-rep_2, iend);
-            best_off = rep_2;
-            rep_2 = rep_1;
-            rep_1 = best_off;
-            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, best_mlen);
-            ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, best_mlen);
-            anchor += best_mlen+minMatch;
-            continue;   /* faster when present ... (?) */
-        }
-        if (anchor > ip) ip = anchor;
     }
 
     {   /* Last Literals */
@@ -745,6 +746,9 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                                      const void* src, size_t srcSize,
                                      const U32 depth)
 {
+    printf("NOT IMPLEMENTED: ZSTD_compressBlock_opt_extDict_generic\n"), exit(0);
+    (void)ctx; (void)src; (void)srcSize; (void)depth; (void)ZSTD_BtGetAllMatches_selectMLS_extDict;
+#if 0
     seqStore_t* seqStorePtr = &(ctx->seqStore);
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
@@ -759,7 +763,6 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
     const BYTE* const dictEnd  = dictBase + dictLimit;
     const U32 lowLimit = ctx->lowLimit;
 
-    U32 rep_2=REPCODE_STARTVALUE, rep_1=REPCODE_STARTVALUE;
     const U32 maxSearches = 1U << ctx->params.cParams.searchLog;
     const U32 sufficient_len = ctx->params.cParams.targetLength;
     const U32 mls = ctx->params.cParams.searchLength;
@@ -771,6 +774,10 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
     U32 cur, match_num, last_pos, litlen, price;
 
     /* init */
+    U32 rep[ZSTD_REP_INIT];
+    for (int i=0; i<ZSTD_REP_INIT; i++)
+        rep[i]=REPCODE_STARTVALUE;
+
     ctx->nextToUpdate3 = ctx->nextToUpdate;
     ZSTD_resetSeqStore(seqStorePtr);
     ZSTD_rescaleFreqs(seqStorePtr);
@@ -1112,4 +1119,5 @@ _storeSequence: // cur, last_pos, best_mlen, best_off have to be set
         memcpy(seqStorePtr->lit, anchor, lastLLSize);
         seqStorePtr->lit += lastLLSize;
     }
+#endif
 }