zstd-ruby 1.4.0.0 → 1.4.9.0

data/ext/zstdruby/libzstd/compress/zstd_ldm.h

@@ -1,10 +1,11 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef ZSTD_LDM_H
@@ -15,7 +16,7 @@ extern "C" {
 #endif
 
 #include "zstd_compress_internal.h"   /* ldmParams_t, U32 */
-#include "zstd.h"   /* ZSTD_CCtx, size_t */
+#include "../zstd.h"   /* ZSTD_CCtx, size_t */
 
 /*-*************************************
 *  Long distance matching
@@ -23,6 +24,10 @@ extern "C" {
 
 #define ZSTD_LDM_DEFAULT_WINDOW_LOG ZSTD_WINDOWLOG_LIMIT_DEFAULT
 
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* state, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params);
+
 /**
  * ZSTD_ldm_generateSequences():
  *
@@ -68,11 +73,17 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
  *
  * Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
  * Avoids emitting matches less than `minMatch` bytes.
- * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ * Must be called for data that is not passed to ZSTD_ldm_blockCompress().
  */
 void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
     U32 const minMatch);
 
+/* ZSTD_ldm_skipRawSeqStoreBytes():
+ * Moves forward in rawSeqStore by nbBytes, updating fields 'pos' and 'posInSequence'.
+ * Not to be used in conjunction with ZSTD_ldm_skipSequences().
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes);
 
 /** ZSTD_ldm_getTableSize() :
  *  Estimate the space needed for long distance matching tables or 0 if LDM is

data/ext/zstdruby/libzstd/compress/zstd_ldm_geartab.h

@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2016-2021, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_LDM_GEARTAB_H
+#define ZSTD_LDM_GEARTAB_H
+
+static U64 ZSTD_ldm_gearTab[256] = {
+    0xf5b8f72c5f77775c, 0x84935f266b7ac412, 0xb647ada9ca730ccc,
+    0xb065bb4b114fb1de, 0x34584e7e8c3a9fd0, 0x4e97e17c6ae26b05,
+    0x3a03d743bc99a604, 0xcecd042422c4044f, 0x76de76c58524259e,
+    0x9c8528f65badeaca, 0x86563706e2097529, 0x2902475fa375d889,
+    0xafb32a9739a5ebe6, 0xce2714da3883e639, 0x21eaf821722e69e,
+    0x37b628620b628,    0x49a8d455d88caf5,  0x8556d711e6958140,
+    0x4f7ae74fc605c1f,  0x829f0c3468bd3a20, 0x4ffdc885c625179e,
+    0x8473de048a3daf1b, 0x51008822b05646b2, 0x69d75d12b2d1cc5f,
+    0x8c9d4a19159154bc, 0xc3cc10f4abbd4003, 0xd06ddc1cecb97391,
+    0xbe48e6e7ed80302e, 0x3481db31cee03547, 0xacc3f67cdaa1d210,
+    0x65cb771d8c7f96cc, 0x8eb27177055723dd, 0xc789950d44cd94be,
+    0x934feadc3700b12b, 0x5e485f11edbdf182, 0x1e2e2a46fd64767a,
+    0x2969ca71d82efa7c, 0x9d46e9935ebbba2e, 0xe056b67e05e6822b,
+    0x94d73f55739d03a0, 0xcd7010bdb69b5a03, 0x455ef9fcd79b82f4,
+    0x869cb54a8749c161, 0x38d1a4fa6185d225, 0xb475166f94bbe9bb,
+    0xa4143548720959f1, 0x7aed4780ba6b26ba, 0xd0ce264439e02312,
+    0x84366d746078d508, 0xa8ce973c72ed17be, 0x21c323a29a430b01,
+    0x9962d617e3af80ee, 0xab0ce91d9c8cf75b, 0x530e8ee6d19a4dbc,
+    0x2ef68c0cf53f5d72, 0xc03a681640a85506, 0x496e4e9f9c310967,
+    0x78580472b59b14a0, 0x273824c23b388577, 0x66bf923ad45cb553,
+    0x47ae1a5a2492ba86, 0x35e304569e229659, 0x4765182a46870b6f,
+    0x6cbab625e9099412, 0xddac9a2e598522c1, 0x7172086e666624f2,
+    0xdf5003ca503b7837, 0x88c0c1db78563d09, 0x58d51865acfc289d,
+    0x177671aec65224f1, 0xfb79d8a241e967d7, 0x2be1e101cad9a49a,
+    0x6625682f6e29186b, 0x399553457ac06e50, 0x35dffb4c23abb74,
+    0x429db2591f54aade, 0xc52802a8037d1009, 0x6acb27381f0b25f3,
+    0xf45e2551ee4f823b, 0x8b0ea2d99580c2f7, 0x3bed519cbcb4e1e1,
+    0xff452823dbb010a,  0x9d42ed614f3dd267, 0x5b9313c06257c57b,
+    0xa114b8008b5e1442, 0xc1fe311c11c13d4b, 0x66e8763ea34c5568,
+    0x8b982af1c262f05d, 0xee8876faaa75fbb7, 0x8a62a4d0d172bb2a,
+    0xc13d94a3b7449a97, 0x6dbbba9dc15d037c, 0xc786101f1d92e0f1,
+    0xd78681a907a0b79b, 0xf61aaf2962c9abb9, 0x2cfd16fcd3cb7ad9,
+    0x868c5b6744624d21, 0x25e650899c74ddd7, 0xba042af4a7c37463,
+    0x4eb1a539465a3eca, 0xbe09dbf03b05d5ca, 0x774e5a362b5472ba,
+    0x47a1221229d183cd, 0x504b0ca18ef5a2df, 0xdffbdfbde2456eb9,
+    0x46cd2b2fbee34634, 0xf2aef8fe819d98c3, 0x357f5276d4599d61,
+    0x24a5483879c453e3, 0x88026889192b4b9,  0x28da96671782dbec,
+    0x4ef37c40588e9aaa, 0x8837b90651bc9fb3, 0xc164f741d3f0e5d6,
+    0xbc135a0a704b70ba, 0x69cd868f7622ada,  0xbc37ba89e0b9c0ab,
+    0x47c14a01323552f6, 0x4f00794bacee98bb, 0x7107de7d637a69d5,
+    0x88af793bb6f2255e, 0xf3c6466b8799b598, 0xc288c616aa7f3b59,
+    0x81ca63cf42fca3fd, 0x88d85ace36a2674b, 0xd056bd3792389e7,
+    0xe55c396c4e9dd32d, 0xbefb504571e6c0a6, 0x96ab32115e91e8cc,
+    0xbf8acb18de8f38d1, 0x66dae58801672606, 0x833b6017872317fb,
+    0xb87c16f2d1c92864, 0xdb766a74e58b669c, 0x89659f85c61417be,
+    0xc8daad856011ea0c, 0x76a4b565b6fe7eae, 0xa469d085f6237312,
+    0xaaf0365683a3e96c, 0x4dbb746f8424f7b8, 0x638755af4e4acc1,
+    0x3d7807f5bde64486, 0x17be6d8f5bbb7639, 0x903f0cd44dc35dc,
+    0x67b672eafdf1196c, 0xa676ff93ed4c82f1, 0x521d1004c5053d9d,
+    0x37ba9ad09ccc9202, 0x84e54d297aacfb51, 0xa0b4b776a143445,
+    0x820d471e20b348e,  0x1874383cb83d46dc, 0x97edeec7a1efe11c,
+    0xb330e50b1bdc42aa, 0x1dd91955ce70e032, 0xa514cdb88f2939d5,
+    0x2791233fd90db9d3, 0x7b670a4cc50f7a9b, 0x77c07d2a05c6dfa5,
+    0xe3778b6646d0a6fa, 0xb39c8eda47b56749, 0x933ed448addbef28,
+    0xaf846af6ab7d0bf4, 0xe5af208eb666e49,  0x5e6622f73534cd6a,
+    0x297daeca42ef5b6e, 0x862daef3d35539a6, 0xe68722498f8e1ea9,
+    0x981c53093dc0d572, 0xfa09b0bfbf86fbf5, 0x30b1e96166219f15,
+    0x70e7d466bdc4fb83, 0x5a66736e35f2a8e9, 0xcddb59d2b7c1baef,
+    0xd6c7d247d26d8996, 0xea4e39eac8de1ba3, 0x539c8bb19fa3aff2,
+    0x9f90e4c5fd508d8,  0xa34e5956fbaf3385, 0x2e2f8e151d3ef375,
+    0x173691e9b83faec1, 0xb85a8d56bf016379, 0x8382381267408ae3,
+    0xb90f901bbdc0096d, 0x7c6ad32933bcec65, 0x76bb5e2f2c8ad595,
+    0x390f851a6cf46d28, 0xc3e6064da1c2da72, 0xc52a0c101cfa5389,
+    0xd78eaf84a3fbc530, 0x3781b9e2288b997e, 0x73c2f6dea83d05c4,
+    0x4228e364c5b5ed7,  0x9d7a3edf0da43911, 0x8edcfeda24686756,
+    0x5e7667a7b7a9b3a1, 0x4c4f389fa143791d, 0xb08bc1023da7cddc,
+    0x7ab4be3ae529b1cc, 0x754e6132dbe74ff9, 0x71635442a839df45,
+    0x2f6fb1643fbe52de, 0x961e0a42cf7a8177, 0xf3b45d83d89ef2ea,
+    0xee3de4cf4a6e3e9b, 0xcd6848542c3295e7, 0xe4cee1664c78662f,
+    0x9947548b474c68c4, 0x25d73777a5ed8b0b, 0xc915b1d636b7fc,
+    0x21c2ba75d9b0d2da, 0x5f6b5dcf608a64a1, 0xdcf333255ff9570c,
+    0x633b922418ced4ee, 0xc136dde0b004b34a, 0x58cc83b05d4b2f5a,
+    0x5eb424dda28e42d2, 0x62df47369739cd98, 0xb4e0b42485e4ce17,
+    0x16e1f0c1f9a8d1e7, 0x8ec3916707560ebf, 0x62ba6e2df2cc9db3,
+    0xcbf9f4ff77d83a16, 0x78d9d7d07d2bbcc4, 0xef554ce1e02c41f4,
+    0x8d7581127eccf94d, 0xa9b53336cb3c8a05, 0x38c42c0bf45c4f91,
+    0x640893cdf4488863, 0x80ec34bc575ea568, 0x39f324f5b48eaa40,
+    0xe9d9ed1f8eff527f, 0x9224fc058cc5a214, 0xbaba00b04cfe7741,
+    0x309a9f120fcf52af, 0xa558f3ec65626212, 0x424bec8b7adabe2f,
+    0x41622513a6aea433, 0xb88da2d5324ca798, 0xd287733b245528a4,
+    0x9a44697e6d68aec3, 0x7b1093be2f49bb28, 0x50bbec632e3d8aad,
+    0x6cd90723e1ea8283, 0x897b9e7431b02bf3, 0x219efdcb338a7047,
+    0x3b0311f0a27c0656, 0xdb17bf91c0db96e7, 0x8cd4fd6b4e85a5b2,
+    0xfab071054ba6409d, 0x40d6fe831fa9dfd9, 0xaf358debad7d791e,
+    0xeb8d0e25a65e3e58, 0xbbcbd3df14e08580, 0xcf751f27ecdab2b,
+    0x2b4da14f2613d8f4
+};
+
+#endif /* ZSTD_LDM_GEARTAB_H */

data/ext/zstdruby/libzstd/compress/zstd_opt.c

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2021, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -249,40 +249,6 @@ static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optP
     }
 }
 
-/* ZSTD_litLengthContribution() :
- * @return ( cost(litlength) - cost(0) )
- * this value can then be added to rawLiteralsCost()
- * to provide a cost which is directly comparable to a match ending at same position */
-static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr, int optLevel)
-{
-    if (optPtr->priceType >= zop_predef) return WEIGHT(litLength, optLevel);
-
-    /* dynamic statistics */
-    {   U32 const llCode = ZSTD_LLcode(litLength);
-        int const contribution = (LL_bits[llCode] * BITCOST_MULTIPLIER)
-                               + WEIGHT(optPtr->litLengthFreq[0], optLevel)   /* note: log2litLengthSum cancel out */
-                               - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
-#if 1
-        return contribution;
-#else
-        return MAX(0, contribution); /* sometimes better, sometimes not ... */
-#endif
-    }
-}
-
-/* ZSTD_literalsContribution() :
- * creates a fake cost for the literals part of a sequence
- * which can be compared to the ending cost of a match
- * should a new match start at this position */
-static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLength,
-                                     const optState_t* const optPtr,
-                                     int optLevel)
-{
-    int const contribution = ZSTD_rawLiteralsCost(literals, litLength, optPtr, optLevel)
-                           + ZSTD_litLengthContribution(litLength, optPtr, optLevel);
-    return contribution;
-}
-
 /* ZSTD_getMatchPrice() :
  * Provides the cost of the match part (offset + matchLength) of a sequence
  * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
@@ -372,13 +338,15 @@ MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
 
 /* Update hashTable3 up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE* const ip)
+static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
+                                              U32* nextToUpdate3,
+                                              const BYTE* const ip)
 {
     U32* const hashTable3 = ms->hashTable3;
     U32 const hashLog3 = ms->hashLog3;
     const BYTE* const base = ms->window.base;
-    U32 idx = ms->nextToUpdate3;
-    U32 const target = ms->nextToUpdate3 = (U32)(ip - base);
+    U32 idx = *nextToUpdate3;
+    U32 const target = (U32)(ip - base);
     size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
     assert(hashLog3 > 0);
 
@@ -387,6 +355,7 @@ static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE*
         idx++;
     }
 
+    *nextToUpdate3 = target;
     return hashTable3[hash3];
 }
 
@@ -417,32 +386,32 @@ static U32 ZSTD_insertBt1(
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
     const BYTE* match;
-    const U32 current = (U32)(ip-base);
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
-    U32* smallerPtr = bt + 2*(current&btMask);
+    const U32 curr = (U32)(ip-base);
+    const U32 btLow = btMask >= curr ? 0 : curr - btMask;
+    U32* smallerPtr = bt + 2*(curr&btMask);
     U32* largerPtr  = smallerPtr + 1;
     U32 dummy32;   /* to be nullified at the end */
     U32 const windowLow = ms->window.lowLimit;
-    U32 matchEndIdx = current+8+1;
+    U32 matchEndIdx = curr+8+1;
     size_t bestLength = 8;
     U32 nbCompares = 1U << cParams->searchLog;
 #ifdef ZSTD_C_PREDICT
-    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
-    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
+    U32 predictedSmall = *(bt + 2*((curr-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((curr-1)&btMask) + 1);
     predictedSmall += (predictedSmall>0);
     predictedLarge += (predictedLarge>0);
 #endif /* ZSTD_C_PREDICT */
 
-    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", curr);
 
     assert(ip <= iend-8);   /* required for h calculation */
-    hashTable[h] = current;   /* Update Hash Table */
+    hashTable[h] = curr;   /* Update Hash Table */
 
     assert(windowLow > 0);
     while (nbCompares-- && (matchIndex >= windowLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-        assert(matchIndex < current);
+        assert(matchIndex < curr);
 
 #ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
         const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
@@ -503,9 +472,11 @@ static U32 ZSTD_insertBt1(
     }   }
 
     *smallerPtr = *largerPtr = 0;
-    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
-    assert(matchEndIdx > current + 8);
-    return matchEndIdx - (current + 8);
+    {   U32 positions = 0;
+        if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+        assert(matchEndIdx > curr + 8);
+        return MAX(positions, matchEndIdx - (curr + 8));
+    }
 }
 
 FORCE_INLINE_TEMPLATE
@@ -520,8 +491,13 @@ void ZSTD_updateTree_internal(
     DEBUGLOG(6, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
                 idx, target, dictMode);
 
-    while(idx < target)
-        idx += ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
+    while(idx < target) {
+        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
+        assert(idx < (U32)(idx + forward));
+        idx += forward;
+    }
+    assert((size_t)(ip - base) <= (size_t)(U32)(-1));
+    assert((size_t)(iend - base) <= (size_t)(U32)(-1));
     ms->nextToUpdate = target;
 }
 
@@ -531,18 +507,19 @@ void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
 
 FORCE_INLINE_TEMPLATE
 U32 ZSTD_insertBtAndGetAllMatches (
+                    ZSTD_match_t* matches,   /* store result (found matches) in this table (presumed large enough) */
                     ZSTD_matchState_t* ms,
+                    U32* nextToUpdate3,
                     const BYTE* const ip, const BYTE* const iLimit, const ZSTD_dictMode_e dictMode,
-                    U32 rep[ZSTD_REP_NUM],
+                    const U32 rep[ZSTD_REP_NUM],
                     U32 const ll0,   /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
-                    ZSTD_match_t* matches,
                     const U32 lengthToBeat,
                     U32 const mls /* template */)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
     const BYTE* const base = ms->window.base;
-    U32 const current = (U32)(ip-base);
+    U32 const curr = (U32)(ip-base);
     U32 const hashLog = cParams->hashLog;
     U32 const minMatch = (mls==3) ? 3 : 4;
     U32* const hashTable = ms->hashTable;
@@ -556,12 +533,12 @@ U32 ZSTD_insertBtAndGetAllMatches (
     U32 const dictLimit = ms->window.dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
-    U32 const btLow = btMask >= current ? 0 : current - btMask;
-    U32 const windowLow = ms->window.lowLimit;
+    U32 const btLow = (btMask >= curr) ? 0 : curr - btMask;
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, curr, cParams->windowLog);
     U32 const matchLow = windowLow ? windowLow : 1;
-    U32* smallerPtr = bt + 2*(current&btMask);
-    U32* largerPtr  = bt + 2*(current&btMask) + 1;
-    U32 matchEndIdx = current+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
+    U32* smallerPtr = bt + 2*(curr&btMask);
+    U32* largerPtr  = bt + 2*(curr&btMask) + 1;
+    U32 matchEndIdx = curr+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
     U32 dummy32;   /* to be nullified at the end */
     U32 mnum = 0;
     U32 nbCompares = 1U << cParams->searchLog;
@@ -580,7 +557,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
     U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
 
     size_t bestLength = lengthToBeat-1;
-    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", current);
+    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", curr);
 
     /* check repCode */
     assert(ll0 <= 1);   /* necessarily 1 or 0 */
@@ -588,26 +565,29 @@ U32 ZSTD_insertBtAndGetAllMatches (
         U32 repCode;
         for (repCode = ll0; repCode < lastR; repCode++) {
             U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
-            U32 const repIndex = current - repOffset;
+            U32 const repIndex = curr - repOffset;
             U32 repLen = 0;
-            assert(current >= dictLimit);
-            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < current-dictLimit) {  /* equivalent to `current > repIndex >= dictLimit` */
-                if (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch)) {
+            assert(curr >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < curr-dictLimit) {  /* equivalent to `curr > repIndex >= dictLimit` */
+                /* We must validate the repcode offset because when we're using a dictionary the
+                 * valid offset range shrinks when the dictionary goes out of bounds.
+                 */
+                if ((repIndex >= windowLow) & (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch))) {
                     repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
                 }
-            } else {  /* repIndex < dictLimit || repIndex >= current */
+            } else {  /* repIndex < dictLimit || repIndex >= curr */
                 const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
                                              dmsBase + repIndex - dmsIndexDelta :
                                              dictBase + repIndex;
-                assert(current >= windowLow);
+                assert(curr >= windowLow);
                 if ( dictMode == ZSTD_extDict
-                  && ( ((repOffset-1) /*intentional overflow*/ < current - windowLow)  /* equivalent to `current > repIndex >= windowLow` */
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - windowLow)  /* equivalent to `curr > repIndex >= windowLow` */
                      & (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
                   && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
                     repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
                 }
                 if (dictMode == ZSTD_dictMatchState
-                  && ( ((repOffset-1) /*intentional overflow*/ < current - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `current > repIndex >= dmsLowLimit` */
+                  && ( ((repOffset-1) /*intentional overflow*/ < curr - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `curr > repIndex >= dmsLowLimit` */
                      & ((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */
                   && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
                     repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
@@ -627,9 +607,9 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
     /* HC3 match finder */
     if ((mls == 3) /*static*/ && (bestLength < mls)) {
-        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, ip);
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
         if ((matchIndex3 >= matchLow)
-          & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+          & (curr - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
             size_t mlen;
             if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
                 const BYTE* const match = base + matchIndex3;
@@ -644,48 +624,48 @@ U32 ZSTD_insertBtAndGetAllMatches (
                 DEBUGLOG(8, "found small match with hlog3, of length %u",
                             (U32)mlen);
                 bestLength = mlen;
-                assert(current > matchIndex3);
+                assert(curr > matchIndex3);
                 assert(mnum==0);  /* no prior solution */
-                matches[0].off = (current - matchIndex3) + ZSTD_REP_MOVE;
+                matches[0].off = (curr - matchIndex3) + ZSTD_REP_MOVE;
                 matches[0].len = (U32)mlen;
                 mnum = 1;
                 if ( (mlen > sufficient_len) |
                      (ip+mlen == iLimit) ) {  /* best possible length */
-                    ms->nextToUpdate = current+1;  /* skip insertion */
+                    ms->nextToUpdate = curr+1;  /* skip insertion */
                     return 1;
-                }
-            }
-        }
+        }   }   }
         /* no dictMatchState lookup: dicts don't have a populated HC3 table */
     }
 
-    hashTable[h] = current;   /* Update Hash Table */
+    hashTable[h] = curr;   /* Update Hash Table */
 
     while (nbCompares-- && (matchIndex >= matchLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
-        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         const BYTE* match;
-        assert(current > matchIndex);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(curr > matchIndex);
 
         if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
             assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
             match = base + matchIndex;
+            if (matchIndex >= dictLimit) assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
             matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
         } else {
             match = dictBase + matchIndex;
+            assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
             matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
             if (matchIndex+matchLength >= dictLimit)
-                match = base + matchIndex;   /* prepare for match[matchLength] */
+                match = base + matchIndex;   /* prepare for match[matchLength] read */
         }
 
         if (matchLength > bestLength) {
             DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
-                    (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+                    (U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
             assert(matchEndIdx > matchIndex);
             if (matchLength > matchEndIdx - matchIndex)
                 matchEndIdx = matchIndex + (U32)matchLength;
             bestLength = matchLength;
-            matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+            matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
             matches[mnum].len = (U32)matchLength;
             mnum++;
             if ( (matchLength > ZSTD_OPT_NUM)
@@ -728,11 +708,11 @@ U32 ZSTD_insertBtAndGetAllMatches (
             if (matchLength > bestLength) {
                 matchIndex = dictMatchIndex + dmsIndexDelta;
                 DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
-                        (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+                        (U32)matchLength, curr - matchIndex, curr - matchIndex + ZSTD_REP_MOVE);
                 if (matchLength > matchEndIdx - matchIndex)
                     matchEndIdx = matchIndex + (U32)matchLength;
                 bestLength = matchLength;
-                matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+                matches[mnum].off = (curr - matchIndex) + ZSTD_REP_MOVE;
                 matches[mnum].len = (U32)matchLength;
                 mnum++;
                 if ( (matchLength > ZSTD_OPT_NUM)
@@ -753,17 +733,20 @@ U32 ZSTD_insertBtAndGetAllMatches (
         }
     }
 
-    assert(matchEndIdx > current+8);
+    assert(matchEndIdx > curr+8);
     ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
     return mnum;
 }
 
 
 FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
+                        ZSTD_match_t* matches,   /* store result (match found, increasing size) in this table */
                         ZSTD_matchState_t* ms,
+                        U32* nextToUpdate3,
                         const BYTE* ip, const BYTE* const iHighLimit, const ZSTD_dictMode_e dictMode,
-                        U32 rep[ZSTD_REP_NUM], U32 const ll0,
-                        ZSTD_match_t* matches, U32 const lengthToBeat)
+                        const U32 rep[ZSTD_REP_NUM],
+                        U32 const ll0,
+                        U32 const lengthToBeat)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const matchLengthSearch = cParams->minMatch;
@@ -772,44 +755,154 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
     ZSTD_updateTree_internal(ms, ip, iHighLimit, matchLengthSearch, dictMode);
     switch(matchLengthSearch)
     {
-    case 3 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 3);
+    case 3 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 3);
     default :
-    case 4 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 4);
-    case 5 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 5);
+    case 4 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 4);
+    case 5 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 5);
     case 7 :
-    case 6 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 6);
+    case 6 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 6);
     }
 }
 
+/*************************
+*  LDM helper functions  *
+*************************/
 
-/*-*******************************
-*  Optimal parser
-*********************************/
-typedef struct repcodes_s {
-    U32 rep[3];
-} repcodes_t;
+/* Struct containing info needed to make decision about ldm inclusion */
+typedef struct {
+    rawSeqStore_t seqStore;         /* External match candidates store for this block */
+    U32 startPosInBlock;            /* Start position of the current match candidate */
+    U32 endPosInBlock;              /* End position of the current match candidate */
+    U32 offset;                     /* Offset of the match candidate */
+} ZSTD_optLdm_t;
 
-static repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
-{
-    repcodes_t newReps;
-    if (offset >= ZSTD_REP_NUM) {  /* full offset */
-        newReps.rep[2] = rep[1];
-        newReps.rep[1] = rep[0];
-        newReps.rep[0] = offset - ZSTD_REP_MOVE;
-    } else {   /* repcode */
-        U32 const repCode = offset + ll0;
-        if (repCode > 0) {  /* note : if repCode==0, no change */
-            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
-            newReps.rep[2] = (repCode >= 2) ? rep[1] : rep[2];
-            newReps.rep[1] = rep[0];
-            newReps.rep[0] = currentOffset;
-        } else {   /* repCode == 0 */
-            memcpy(&newReps, rep, sizeof(newReps));
+/* ZSTD_optLdm_skipRawSeqStoreBytes():
+ * Moves forward in rawSeqStore by nbBytes, which will update the fields 'pos' and 'posInSequence'.
+ */
+static void ZSTD_optLdm_skipRawSeqStoreBytes(rawSeqStore_t* rawSeqStore, size_t nbBytes) {
+    U32 currPos = (U32)(rawSeqStore->posInSequence + nbBytes);
+    while (currPos && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq currSeq = rawSeqStore->seq[rawSeqStore->pos];
+        if (currPos >= currSeq.litLength + currSeq.matchLength) {
+            currPos -= currSeq.litLength + currSeq.matchLength;
+            rawSeqStore->pos++;
+        } else {
+            rawSeqStore->posInSequence = currPos;
+            break;
         }
     }
-    return newReps;
+    if (currPos == 0 || rawSeqStore->pos == rawSeqStore->size) {
+        rawSeqStore->posInSequence = 0;
+    }
+}
+
+/* ZSTD_opt_getNextMatchAndUpdateSeqStore():
+ * Calculates the beginning and end of the next match in the current block.
+ * Updates 'pos' and 'posInSequence' of the ldmSeqStore.
+ */
+static void ZSTD_opt_getNextMatchAndUpdateSeqStore(ZSTD_optLdm_t* optLdm, U32 currPosInBlock,
+                                                   U32 blockBytesRemaining) {
+    rawSeq currSeq;
+    U32 currBlockEndPos;
+    U32 literalsBytesRemaining;
+    U32 matchBytesRemaining;
+
+    /* Setting match end position to MAX to ensure we never use an LDM during this block */
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        return;
+    }
+    /* Calculate appropriate bytes left in matchLength and litLength after adjusting
+       based on ldmSeqStore->posInSequence */
+    currSeq = optLdm->seqStore.seq[optLdm->seqStore.pos];
+    assert(optLdm->seqStore.posInSequence <= currSeq.litLength + currSeq.matchLength);
+    currBlockEndPos = currPosInBlock + blockBytesRemaining;
+    literalsBytesRemaining = (optLdm->seqStore.posInSequence < currSeq.litLength) ?
+            currSeq.litLength - (U32)optLdm->seqStore.posInSequence :
+            0;
+    matchBytesRemaining = (literalsBytesRemaining == 0) ?
+            currSeq.matchLength - ((U32)optLdm->seqStore.posInSequence - currSeq.litLength) :
+            currSeq.matchLength;
+
+    /* If there are more literal bytes than bytes remaining in block, no ldm is possible */
+    if (literalsBytesRemaining >= blockBytesRemaining) {
+        optLdm->startPosInBlock = UINT_MAX;
+        optLdm->endPosInBlock = UINT_MAX;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, blockBytesRemaining);
+        return;
+    }
+
+    /* Matches may be < MINMATCH by this process. In that case, we will reject them
+       when we are deciding whether or not to add the ldm */
+    optLdm->startPosInBlock = currPosInBlock + literalsBytesRemaining;
+    optLdm->endPosInBlock = optLdm->startPosInBlock + matchBytesRemaining;
+    optLdm->offset = currSeq.offset;
+
+    if (optLdm->endPosInBlock > currBlockEndPos) {
+        /* Match ends after the block ends, we can't use the whole match */
+        optLdm->endPosInBlock = currBlockEndPos;
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, currBlockEndPos - currPosInBlock);
+    } else {
+        /* Consume nb of bytes equal to size of sequence left */
+        ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, literalsBytesRemaining + matchBytesRemaining);
+    }
+}
+
+/* ZSTD_optLdm_maybeAddMatch():
+ * Adds a match if it's long enough, based on it's 'matchStartPosInBlock'
+ * and 'matchEndPosInBlock', into 'matches'. Maintains the correct ordering of 'matches'
+ */
+static void ZSTD_optLdm_maybeAddMatch(ZSTD_match_t* matches, U32* nbMatches,
+                                      ZSTD_optLdm_t* optLdm, U32 currPosInBlock) {
+    U32 posDiff = currPosInBlock - optLdm->startPosInBlock;
+    /* Note: ZSTD_match_t actually contains offCode and matchLength (before subtracting MINMATCH) */
+    U32 candidateMatchLength = optLdm->endPosInBlock - optLdm->startPosInBlock - posDiff;
+    U32 candidateOffCode = optLdm->offset + ZSTD_REP_MOVE;
+
+    /* Ensure that current block position is not outside of the match */
+    if (currPosInBlock < optLdm->startPosInBlock
+      || currPosInBlock >= optLdm->endPosInBlock
+      || candidateMatchLength < MINMATCH) {
+        return;
+    }
+
+    if (*nbMatches == 0 || ((candidateMatchLength > matches[*nbMatches-1].len) && *nbMatches < ZSTD_OPT_NUM)) {
+        DEBUGLOG(6, "ZSTD_optLdm_maybeAddMatch(): Adding ldm candidate match (offCode: %u matchLength %u) at block position=%u",
+                 candidateOffCode, candidateMatchLength, currPosInBlock);
+        matches[*nbMatches].len = candidateMatchLength;
+        matches[*nbMatches].off = candidateOffCode;
+        (*nbMatches)++;
+    }
 }
 
+/* ZSTD_optLdm_processMatchCandidate():
+ * Wrapper function to update ldm seq store and call ldm functions as necessary.
+ */
+static void ZSTD_optLdm_processMatchCandidate(ZSTD_optLdm_t* optLdm, ZSTD_match_t* matches, U32* nbMatches,
+                                              U32 currPosInBlock, U32 remainingBytes) {
+    if (optLdm->seqStore.size == 0 || optLdm->seqStore.pos >= optLdm->seqStore.size) {
+        return;
+    }
+
+    if (currPosInBlock >= optLdm->endPosInBlock) {
+        if (currPosInBlock > optLdm->endPosInBlock) {
+            /* The position at which ZSTD_optLdm_processMatchCandidate() is called is not necessarily
+             * at the end of a match from the ldm seq store, and will often be some bytes
+             * over beyond matchEndPosInBlock. As such, we need to correct for these "overshoots"
+             */
+            U32 posOvershoot = currPosInBlock - optLdm->endPosInBlock;
+            ZSTD_optLdm_skipRawSeqStoreBytes(&optLdm->seqStore, posOvershoot);
+        } 
+        ZSTD_opt_getNextMatchAndUpdateSeqStore(optLdm, currPosInBlock, remainingBytes);
+    }
+    ZSTD_optLdm_maybeAddMatch(matches, nbMatches, optLdm, currPosInBlock);
+}
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+
 
 static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
 {
@@ -825,7 +918,7 @@ listStats(const U32* table, int lastEltID)
     int enb;
     for (enb=0; enb < nbElts; enb++) {
         (void)table;
-        //RAWLOG(2, "%3i:%3i,  ", enb, table[enb]);
+        /* RAWLOG(2, "%3i:%3i,  ", enb, table[enb]); */
         RAWLOG(2, "%4i,", table[enb]);
     }
     RAWLOG(2, " \n");
@@ -853,16 +946,21 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
 
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
     U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
+    U32 nextToUpdate3 = ms->nextToUpdate;
 
     ZSTD_optimal_t* const opt = optStatePtr->priceTable;
     ZSTD_match_t* const matches = optStatePtr->matchTable;
     ZSTD_optimal_t lastSequence;
+    ZSTD_optLdm_t optLdm;
+
+    optLdm.seqStore = ms->ldmSeqStore ? *ms->ldmSeqStore : kNullRawSeqStore;
+    optLdm.endPosInBlock = optLdm.startPosInBlock = optLdm.offset = 0;
+    ZSTD_opt_getNextMatchAndUpdateSeqStore(&optLdm, (U32)(ip-istart), (U32)(iend-ip));
 
     /* init */
     DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
                 (U32)(ip - base), ms->window.dictLimit, ms->nextToUpdate);
     assert(optLevel <= 2);
-    ms->nextToUpdate3 = ms->nextToUpdate;
     ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
     ip += (ip==prefixStart);
 
@@ -873,14 +971,21 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
         /* find first match */
         {   U32 const litlen = (U32)(ip - anchor);
             U32 const ll0 = !litlen;
-            U32 const nbMatches = ZSTD_BtGetAllMatches(ms, ip, iend, dictMode, rep, ll0, matches, minMatch);
+            U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
+            ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                              (U32)(ip-istart), (U32)(iend - ip));
             if (!nbMatches) { ip++; continue; }
 
             /* initialize opt[0] */
             { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
             opt[0].mlen = 0;  /* means is_a_literal */
             opt[0].litlen = litlen;
-            opt[0].price = ZSTD_literalsContribution(anchor, litlen, optStatePtr, optLevel);
+            /* We don't need to include the actual price of the literals because
+             * it is static for the duration of the forward pass, and is included
+             * in every price. We include the literal length to avoid negative
+             * prices when we subtract the previous literal length.
+             */
+            opt[0].price = ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
 
             /* large match -> immediate encoding */
             {   U32 const maxML = matches[nbMatches-1].len;
@@ -909,7 +1014,6 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 for (matchNb = 0; matchNb < nbMatches; matchNb++) {
                     U32 const offset = matches[matchNb].off;
                     U32 const end = matches[matchNb].len;
-                    repcodes_t const repHistory = ZSTD_updateRep(rep, offset, ll0);
                     for ( ; pos <= end ; pos++ ) {
                         U32 const matchPrice = ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
                         U32 const sequencePrice = literalsPrice + matchPrice;
@@ -919,8 +1023,6 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                         opt[pos].off = offset;
                         opt[pos].litlen = litlen;
                         opt[pos].price = sequencePrice;
-                        ZSTD_STATIC_ASSERT(sizeof(opt[pos].rep) == sizeof(repHistory));
-                        memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
                 }   }
                 last_pos = pos-1;
             }
@@ -947,7 +1049,6 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                     opt[cur].off = 0;
                     opt[cur].litlen = litlen;
                     opt[cur].price = price;
-                    memcpy(opt[cur].rep, opt[cur-1].rep, sizeof(opt[cur].rep));
                 } else {
                     DEBUGLOG(7, "cPos:%zi==rPos:%u : literal would cost more (%.2f>%.2f) (hist:%u,%u,%u)",
                                 inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price),
@@ -955,6 +1056,21 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 }
             }
 
+            /* Set the repcodes of the current position. We must do it here
+             * because we rely on the repcodes of the 2nd to last sequence being
+             * correct to set the next chunks repcodes during the backward
+             * traversal.
+             */
+            ZSTD_STATIC_ASSERT(sizeof(opt[cur].rep) == sizeof(repcodes_t));
+            assert(cur >= opt[cur].mlen);
+            if (opt[cur].mlen != 0) {
+                U32 const prev = cur - opt[cur].mlen;
+                repcodes_t newReps = ZSTD_updateRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
+                ZSTD_memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
+            } else {
+                ZSTD_memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
+            }
+
             /* last match must start at a minimum distance of 8 from oend */
             if (inr > ilimit) continue;
 
@@ -970,8 +1086,12 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
                 U32 const previousPrice = opt[cur].price;
                 U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
-                U32 const nbMatches = ZSTD_BtGetAllMatches(ms, inr, iend, dictMode, opt[cur].rep, ll0, matches, minMatch);
+                U32 nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
                 U32 matchNb;
+
+                ZSTD_optLdm_processMatchCandidate(&optLdm, matches, &nbMatches,
+                                                  (U32)(inr-istart), (U32)(iend-inr));
+
                 if (!nbMatches) {
                     DEBUGLOG(7, "rPos:%u : no match found", cur);
                     continue;
@@ -995,7 +1115,6 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 /* set prices using matches found at position == cur */
                 for (matchNb = 0; matchNb < nbMatches; matchNb++) {
                     U32 const offset = matches[matchNb].off;
-                    repcodes_t const repHistory = ZSTD_updateRep(opt[cur].rep, offset, ll0);
                     U32 const lastML = matches[matchNb].len;
                     U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
                     U32 mlen;
@@ -1015,8 +1134,6 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                             opt[pos].off = offset;
                             opt[pos].litlen = litlen;
                             opt[pos].price = price;
-                            ZSTD_STATIC_ASSERT(sizeof(opt[pos].rep) == sizeof(repHistory));
-                            memcpy(opt[pos].rep, &repHistory, sizeof(repHistory));
                         } else {
                             DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
                                         pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
@@ -1032,6 +1149,17 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
 _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
         assert(opt[0].mlen == 0);
 
+        /* Set the next chunk's repcodes based on the repcodes of the beginning
+         * of the last match, and the last sequence. This avoids us having to
+         * update them while traversing the sequences.
+         */
+        if (lastSequence.mlen != 0) {
+            repcodes_t reps = ZSTD_updateRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
+            ZSTD_memcpy(rep, &reps, sizeof(reps));
+        } else {
+            ZSTD_memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
+        }
+
         {   U32 const storeEnd = cur + 1;
             U32 storeStart = storeEnd;
             U32 seqPos = cur;
@@ -1068,33 +1196,18 @@ _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
                         continue;   /* will finish */
                     }
 
-                    /* repcodes update : like ZSTD_updateRep(), but update in place */
-                    if (offCode >= ZSTD_REP_NUM) {  /* full offset */
-                        rep[2] = rep[1];
-                        rep[1] = rep[0];
-                        rep[0] = offCode - ZSTD_REP_MOVE;
-                    } else {   /* repcode */
-                        U32 const repCode = offCode + (llen==0);
-                        if (repCode) {  /* note : if repCode==0, no change */
-                            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
-                            if (repCode >= 2) rep[2] = rep[1];
-                            rep[1] = rep[0];
-                            rep[0] = currentOffset;
-                    }   }
-
                     assert(anchor + llen <= iend);
                     ZSTD_updateStats(optStatePtr, llen, anchor, offCode, mlen);
-                    ZSTD_storeSeq(seqStore, llen, anchor, offCode, mlen-MINMATCH);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offCode, mlen-MINMATCH);
                     anchor += advance;
                     ip = anchor;
             }   }
             ZSTD_setBasePrices(optStatePtr, optLevel);
         }
-
     }   /* while (ip < ilimit) */
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
@@ -1142,7 +1255,7 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
                const void* src, size_t srcSize)
 {
     U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
-    memcpy(tmpRep, rep, sizeof(tmpRep));
+    ZSTD_memcpy(tmpRep, rep, sizeof(tmpRep));
 
     DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
     assert(ms->opt.litLengthSum == 0);    /* first block */
@@ -1158,7 +1271,6 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
     ms->window.dictLimit += (U32)srcSize;
     ms->window.lowLimit = ms->window.dictLimit;
     ms->nextToUpdate = ms->window.dictLimit;
-    ms->nextToUpdate3 = ms->window.dictLimit;
 
     /* re-inforce weight of collected statistics */
     ZSTD_upscaleStats(&ms->opt);
@@ -1176,7 +1288,7 @@ size_t ZSTD_compressBlock_btultra2(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
         const void* src, size_t srcSize)
 {
-    U32 const current = (U32)((const BYTE*)src - ms->window.base);
+    U32 const curr = (U32)((const BYTE*)src - ms->window.base);
     DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
 
     /* 2-pass strategy:
@@ -1191,7 +1303,7 @@ size_t ZSTD_compressBlock_btultra2(
     if ( (ms->opt.litLengthSum==0)   /* first block */
       && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
       && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
-      && (current == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
+      && (curr == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
       && (srcSize > ZSTD_PREDEF_THRESHOLD)
       ) {
         ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);