extlz4 0.2.5 → 0.3.4

data/contrib/lz4/lib/lz4hc.c

@@ -1,6 +1,6 @@
 /*
     LZ4 HC - High Compression Mode of LZ4
-    Copyright (C) 2011-2017, Yann Collet.
+    Copyright (C) 2011-2020, Yann Collet.
 
     BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -42,7 +42,7 @@
  *  Select how default compression function will allocate workplace memory,
  *  in stack (0:fastest), or in heap (1:requires malloc()).
  *  Since workplace is rather large, heap mode is recommended.
- */
+**/
 #ifndef LZ4HC_HEAPMODE
 #  define LZ4HC_HEAPMODE 1
 #endif
@@ -53,7 +53,7 @@
 #include "lz4hc.h"
 
 
-/*===   Common LZ4 definitions   ===*/
+/*===   Common definitions   ===*/
 #if defined(__GNUC__)
 #  pragma GCC diagnostic ignored "-Wunused-function"
 #endif
@@ -62,11 +62,18 @@
 #endif
 
 #define LZ4_COMMONDEFS_ONLY
+#ifndef LZ4_SRC_INCLUDED
 #include "lz4.c"   /* LZ4_count, constants, mem */
+#endif
+
+
+/*===   Enums   ===*/
+typedef enum { noDictCtx, usingDictCtxHc } dictCtx_directive;
 
 
 /*===   Constants   ===*/
 #define OPTIMAL_ML (int)((ML_MASK-1)+MINMATCH)
+#define LZ4_OPT_NUM   (1<<12)
 
 
 /*===   Macros   ===*/
@@ -75,24 +82,37 @@
 #define HASH_FUNCTION(i)         (((i) * 2654435761U) >> ((MINMATCH*8)-LZ4HC_HASH_LOG))
 #define DELTANEXTMAXD(p)         chainTable[(p) & LZ4HC_MAXD_MASK]    /* flexible, LZ4HC_MAXD dependent */
 #define DELTANEXTU16(table, pos) table[(U16)(pos)]   /* faster */
+/* Make fields passed to, and updated by LZ4HC_encodeSequence explicit */
+#define UPDATABLE(ip, op, anchor) &ip, &op, &anchor
 
 static U32 LZ4HC_hashPtr(const void* ptr) { return HASH_FUNCTION(LZ4_read32(ptr)); }
 
 
-
 /**************************************
 *  HC Compression
 **************************************/
-static void LZ4HC_init (LZ4HC_CCtx_internal* hc4, const BYTE* start)
+static void LZ4HC_clearTables (LZ4HC_CCtx_internal* hc4)
 {
-    MEM_INIT((void*)hc4->hashTable, 0, sizeof(hc4->hashTable));
+    MEM_INIT(hc4->hashTable, 0, sizeof(hc4->hashTable));
     MEM_INIT(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
-    hc4->nextToUpdate = 64 KB;
-    hc4->base = start - 64 KB;
+}
+
+static void LZ4HC_init_internal (LZ4HC_CCtx_internal* hc4, const BYTE* start)
+{
+    size_t const bufferSize = (size_t)(hc4->end - hc4->prefixStart);
+    size_t newStartingOffset = bufferSize + hc4->dictLimit;
+    assert(newStartingOffset >= bufferSize);  /* check overflow */
+    if (newStartingOffset > 1 GB) {
+        LZ4HC_clearTables(hc4);
+        newStartingOffset = 0;
+    }
+    newStartingOffset += 64 KB;
+    hc4->nextToUpdate = (U32)newStartingOffset;
+    hc4->prefixStart = start;
     hc4->end = start;
-    hc4->dictBase = start - 64 KB;
-    hc4->dictLimit = 64 KB;
-    hc4->lowLimit = 64 KB;
+    hc4->dictStart = start;
+    hc4->dictLimit = (U32)newStartingOffset;
+    hc4->lowLimit = (U32)newStartingOffset;
 }
 
 
@@ -101,14 +121,17 @@ LZ4_FORCE_INLINE void LZ4HC_Insert (LZ4HC_CCtx_internal* hc4, const BYTE* ip)
 {
     U16* const chainTable = hc4->chainTable;
     U32* const hashTable  = hc4->hashTable;
-    const BYTE* const base = hc4->base;
-    U32 const target = (U32)(ip - base);
+    const BYTE* const prefixPtr = hc4->prefixStart;
+    U32 const prefixIdx = hc4->dictLimit;
+    U32 const target = (U32)(ip - prefixPtr) + prefixIdx;
     U32 idx = hc4->nextToUpdate;
+    assert(ip >= prefixPtr);
+    assert(target >= prefixIdx);
 
     while (idx < target) {
-        U32 const h = LZ4HC_hashPtr(base+idx);
+        U32 const h = LZ4HC_hashPtr(prefixPtr+idx-prefixIdx);
         size_t delta = idx - hashTable[h];
-        if (delta>MAX_DISTANCE) delta = MAX_DISTANCE;
+        if (delta>LZ4_DISTANCE_MAX) delta = LZ4_DISTANCE_MAX;
         DELTANEXTU16(chainTable, idx) = (U16)delta;
         hashTable[h] = idx;
         idx++;
@@ -123,20 +146,39 @@ LZ4_FORCE_INLINE
 int LZ4HC_countBack(const BYTE* const ip, const BYTE* const match,
                     const BYTE* const iMin, const BYTE* const mMin)
 {
-    int back=0;
-    while ( (ip+back > iMin)
-         && (match+back > mMin)
-         && (ip[back-1] == match[back-1]))
+    int back = 0;
+    int const min = (int)MAX(iMin - ip, mMin - match);
+    assert(min <= 0);
+    assert(ip >= iMin); assert((size_t)(ip-iMin) < (1U<<31));
+    assert(match >= mMin); assert((size_t)(match - mMin) < (1U<<31));
+    while ( (back > min)
+         && (ip[back-1] == match[back-1]) )
             back--;
     return back;
 }
 
+#if defined(_MSC_VER)
+#  define LZ4HC_rotl32(x,r) _rotl(x,r)
+#else
+#  define LZ4HC_rotl32(x,r) ((x << r) | (x >> (32 - r)))
+#endif
+
+
+static U32 LZ4HC_rotatePattern(size_t const rotate, U32 const pattern)
+{
+    size_t const bitsToRotate = (rotate & (sizeof(pattern) - 1)) << 3;
+    if (bitsToRotate == 0) return pattern;
+    return LZ4HC_rotl32(pattern, (int)bitsToRotate);
+}
+
 /* LZ4HC_countPattern() :
  * pattern32 must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!) */
-static unsigned LZ4HC_countPattern(const BYTE* ip, const BYTE* const iEnd, U32 const pattern32)
+static unsigned
+LZ4HC_countPattern(const BYTE* ip, const BYTE* const iEnd, U32 const pattern32)
 {
     const BYTE* const iStart = ip;
-    reg_t const pattern = (sizeof(pattern)==8) ? (reg_t)pattern32 + (((reg_t)pattern32) << 32) : pattern32;
+    reg_t const pattern = (sizeof(pattern)==8) ?
+        (reg_t)pattern32 + (((reg_t)pattern32) << (sizeof(pattern)*4)) : pattern32;
 
     while (likely(ip < iEnd-(sizeof(pattern)-1))) {
         reg_t const diff = LZ4_read_ARCH(ip) ^ pattern;
@@ -156,16 +198,16 @@ static unsigned LZ4HC_countPattern(const BYTE* ip, const BYTE* const iEnd, U32 c
             BYTE const byte = (BYTE)(pattern >> bitOffset);
             if (*ip != byte) break;
             ip ++; bitOffset -= 8;
-        }
-    }
+    }   }
 
     return (unsigned)(ip - iStart);
 }
 
 /* LZ4HC_reverseCountPattern() :
  * pattern must be a sample of repetitive pattern of length 1, 2 or 4 (but not 3!)
- * read using natural platform endianess */
-static unsigned LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow, U32 pattern)
+ * read using natural platform endianness */
+static unsigned
+LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow, U32 pattern)
 {
     const BYTE* const iStart = ip;
 
@@ -173,7 +215,7 @@ static unsigned LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow
         if (LZ4_read32(ip-4) != pattern) break;
         ip -= 4;
     }
-    {   const BYTE* bytePtr = (const BYTE*)(&pattern) + 3; /* works for any endianess */
+    {   const BYTE* bytePtr = (const BYTE*)(&pattern) + 3; /* works for any endianness */
         while (likely(ip>iLow)) {
             if (ip[-1] != *bytePtr) break;
             ip--; bytePtr--;
@@ -181,28 +223,46 @@ static unsigned LZ4HC_reverseCountPattern(const BYTE* ip, const BYTE* const iLow
     return (unsigned)(iStart - ip);
 }
 
-typedef enum { rep_untested, rep_not, rep_confirmed } repeat_state_e;
+/* LZ4HC_protectDictEnd() :
+ * Checks if the match is in the last 3 bytes of the dictionary, so reading the
+ * 4 byte MINMATCH would overflow.
+ * @returns true if the match index is okay.
+ */
+static int LZ4HC_protectDictEnd(U32 const dictLimit, U32 const matchIndex)
+{
+    return ((U32)((dictLimit - 1) - matchIndex) >= 3);
+}
 
-LZ4_FORCE_INLINE int LZ4HC_InsertAndGetWiderMatch (
-    LZ4HC_CCtx_internal* hc4,
-    const BYTE* const ip,
-    const BYTE* const iLowLimit,
-    const BYTE* const iHighLimit,
-    int longest,
-    const BYTE** matchpos,
-    const BYTE** startpos,
-    const int maxNbAttempts,
-    const int patternAnalysis)
+typedef enum { rep_untested, rep_not, rep_confirmed } repeat_state_e;
+typedef enum { favorCompressionRatio=0, favorDecompressionSpeed } HCfavor_e;
+
+LZ4_FORCE_INLINE int
+LZ4HC_InsertAndGetWiderMatch (
+        LZ4HC_CCtx_internal* const hc4,
+        const BYTE* const ip,
+        const BYTE* const iLowLimit, const BYTE* const iHighLimit,
+        int longest,
+        const BYTE** matchpos,
+        const BYTE** startpos,
+        const int maxNbAttempts,
+        const int patternAnalysis, const int chainSwap,
+        const dictCtx_directive dict,
+        const HCfavor_e favorDecSpeed)
 {
     U16* const chainTable = hc4->chainTable;
     U32* const HashTable = hc4->hashTable;
-    const BYTE* const base = hc4->base;
-    const U32 dictLimit = hc4->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const U32 lowLimit = (hc4->lowLimit + 64 KB > (U32)(ip-base)) ? hc4->lowLimit : (U32)(ip - base) - MAX_DISTANCE;
-    const BYTE* const dictBase = hc4->dictBase;
-    int const delta = (int)(ip-iLowLimit);
+    const LZ4HC_CCtx_internal * const dictCtx = hc4->dictCtx;
+    const BYTE* const prefixPtr = hc4->prefixStart;
+    const U32 prefixIdx = hc4->dictLimit;
+    const U32 ipIndex = (U32)(ip - prefixPtr) + prefixIdx;
+    const int withinStartDistance = (hc4->lowLimit + (LZ4_DISTANCE_MAX + 1) > ipIndex);
+    const U32 lowestMatchIndex = (withinStartDistance) ? hc4->lowLimit : ipIndex - LZ4_DISTANCE_MAX;
+    const BYTE* const dictStart = hc4->dictStart;
+    const U32 dictIdx = hc4->lowLimit;
+    const BYTE* const dictEnd = dictStart + prefixIdx - dictIdx;
+    int const lookBackLength = (int)(ip-iLowLimit);
     int nbAttempts = maxNbAttempts;
+    U32 matchChainPos = 0;
     U32 const pattern = LZ4_read32(ip);
     U32 matchIndex;
     repeat_state_e repeat = rep_untested;
@@ -212,200 +272,297 @@ LZ4_FORCE_INLINE int LZ4HC_InsertAndGetWiderMatch (
     /* First Match */
     LZ4HC_Insert(hc4, ip);
     matchIndex = HashTable[LZ4HC_hashPtr(ip)];
-    DEBUGLOG(7, "First match at index %u / %u (lowLimit)",
-                matchIndex, lowLimit);
+    DEBUGLOG(7, "First match at index %u / %u (lowestMatchIndex)",
+                matchIndex, lowestMatchIndex);
 
-    while ((matchIndex>=lowLimit) && (nbAttempts)) {
-        DEBUGLOG(7, "remaining attempts : %i", nbAttempts);
+    while ((matchIndex>=lowestMatchIndex) && (nbAttempts>0)) {
+        int matchLength=0;
         nbAttempts--;
-        if (matchIndex >= dictLimit) {
-            const BYTE* const matchPtr = base + matchIndex;
-            if (*(iLowLimit + longest) == *(matchPtr - delta + longest)) {
+        assert(matchIndex < ipIndex);
+        if (favorDecSpeed && (ipIndex - matchIndex < 8)) {
+            /* do nothing */
+        } else if (matchIndex >= prefixIdx) {   /* within current Prefix */
+            const BYTE* const matchPtr = prefixPtr + matchIndex - prefixIdx;
+            assert(matchPtr < ip);
+            assert(longest >= 1);
+            if (LZ4_read16(iLowLimit + longest - 1) == LZ4_read16(matchPtr - lookBackLength + longest - 1)) {
                 if (LZ4_read32(matchPtr) == pattern) {
-                    int mlt = MINMATCH + LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
-    #if 0
-                    /* more generic but unfortunately slower on clang */
-                    int const back = LZ4HC_countBack(ip, matchPtr, iLowLimit, lowPrefixPtr);
-    #else
-                    int back = 0;
-                    while ( (ip+back > iLowLimit)
-                         && (matchPtr+back > lowPrefixPtr)
-                         && (ip[back-1] == matchPtr[back-1])) {
-                            back--;
-                    }
-    #endif
-                    mlt -= back;
-
-                    if (mlt > longest) {
-                        longest = mlt;
-                        *matchpos = matchPtr+back;
-                        *startpos = ip+back;
-                }   }
-            }
-        } else {   /* matchIndex < dictLimit */
-            const BYTE* const matchPtr = dictBase + matchIndex;
-            if (LZ4_read32(matchPtr) == pattern) {
-                int mlt;
+                    int const back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, prefixPtr) : 0;
+                    matchLength = MINMATCH + (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
+                    matchLength -= back;
+                    if (matchLength > longest) {
+                        longest = matchLength;
+                        *matchpos = matchPtr + back;
+                        *startpos = ip + back;
+            }   }   }
+        } else {   /* lowestMatchIndex <= matchIndex < dictLimit */
+            const BYTE* const matchPtr = dictStart + (matchIndex - dictIdx);
+            assert(matchIndex >= dictIdx);
+            if ( likely(matchIndex <= prefixIdx - 4)
+              && (LZ4_read32(matchPtr) == pattern) ) {
                 int back = 0;
-                const BYTE* vLimit = ip + (dictLimit - matchIndex);
+                const BYTE* vLimit = ip + (prefixIdx - matchIndex);
                 if (vLimit > iHighLimit) vLimit = iHighLimit;
-                mlt = LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
-                if ((ip+mlt == vLimit) && (vLimit < iHighLimit))
-                    mlt += LZ4_count(ip+mlt, base+dictLimit, iHighLimit);
-                while ( (ip+back > iLowLimit)
-                     && (matchIndex+back > lowLimit)
-                     && (ip[back-1] == matchPtr[back-1]))
-                        back--;
-                mlt -= back;
-                if (mlt > longest) {
-                    longest = mlt;
-                    *matchpos = base + matchIndex + back;
+                matchLength = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                if ((ip+matchLength == vLimit) && (vLimit < iHighLimit))
+                    matchLength += LZ4_count(ip+matchLength, prefixPtr, iHighLimit);
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictStart) : 0;
+                matchLength -= back;
+                if (matchLength > longest) {
+                    longest = matchLength;
+                    *matchpos = prefixPtr - prefixIdx + matchIndex + back;   /* virtual pos, relative to ip, to retrieve offset */
                     *startpos = ip + back;
         }   }   }
 
-        {   U32 const nextOffset = DELTANEXTU16(chainTable, matchIndex);
-            matchIndex -= nextOffset;
-            if (patternAnalysis && nextOffset==1) {
+        if (chainSwap && matchLength==longest) {   /* better match => select a better chain */
+            assert(lookBackLength==0);   /* search forward only */
+            if (matchIndex + (U32)longest <= ipIndex) {
+                int const kTrigger = 4;
+                U32 distanceToNextMatch = 1;
+                int const end = longest - MINMATCH + 1;
+                int step = 1;
+                int accel = 1 << kTrigger;
+                int pos;
+                for (pos = 0; pos < end; pos += step) {
+                    U32 const candidateDist = DELTANEXTU16(chainTable, matchIndex + (U32)pos);
+                    step = (accel++ >> kTrigger);
+                    if (candidateDist > distanceToNextMatch) {
+                        distanceToNextMatch = candidateDist;
+                        matchChainPos = (U32)pos;
+                        accel = 1 << kTrigger;
+                }   }
+                if (distanceToNextMatch > 1) {
+                    if (distanceToNextMatch > matchIndex) break;   /* avoid overflow */
+                    matchIndex -= distanceToNextMatch;
+                    continue;
+        }   }   }
+
+        {   U32 const distNextMatch = DELTANEXTU16(chainTable, matchIndex);
+            if (patternAnalysis && distNextMatch==1 && matchChainPos==0) {
+                U32 const matchCandidateIdx = matchIndex-1;
                 /* may be a repeated pattern */
                 if (repeat == rep_untested) {
                     if ( ((pattern & 0xFFFF) == (pattern >> 16))
                       &  ((pattern & 0xFF)   == (pattern >> 24)) ) {
                         repeat = rep_confirmed;
-                        srcPatternLength = LZ4HC_countPattern(ip+4, iHighLimit, pattern) + 4;
+                        srcPatternLength = LZ4HC_countPattern(ip+sizeof(pattern), iHighLimit, pattern) + sizeof(pattern);
                     } else {
                         repeat = rep_not;
                 }   }
-                if ( (repeat == rep_confirmed)
-                  && (matchIndex >= dictLimit) ) {   /* same segment only */
-                    const BYTE* const matchPtr = base + matchIndex;
+                if ( (repeat == rep_confirmed) && (matchCandidateIdx >= lowestMatchIndex)
+                  && LZ4HC_protectDictEnd(prefixIdx, matchCandidateIdx) ) {
+                    const int extDict = matchCandidateIdx < prefixIdx;
+                    const BYTE* const matchPtr = (extDict ? dictStart - dictIdx : prefixPtr - prefixIdx) + matchCandidateIdx;
                     if (LZ4_read32(matchPtr) == pattern) {  /* good candidate */
-                        size_t const forwardPatternLength = LZ4HC_countPattern(matchPtr+sizeof(pattern), iHighLimit, pattern) + sizeof(pattern);
-                        const BYTE* const maxLowPtr = (lowPrefixPtr + MAX_DISTANCE >= ip) ? lowPrefixPtr : ip - MAX_DISTANCE;
-                        size_t const backLength = LZ4HC_reverseCountPattern(matchPtr, maxLowPtr, pattern);
-                        size_t const currentSegmentLength = backLength + forwardPatternLength;
-
-                        if ( (currentSegmentLength >= srcPatternLength)   /* current pattern segment large enough to contain full srcPatternLength */
-                          && (forwardPatternLength <= srcPatternLength) ) { /* haven't reached this position yet */
-                            matchIndex += (U32)forwardPatternLength - (U32)srcPatternLength;  /* best position, full pattern, might be followed by more match */
-                        } else {
-                            matchIndex -= (U32)backLength;   /* let's go to farthest segment position, will find a match of length currentSegmentLength + maybe some back */
+                        const BYTE* const iLimit = extDict ? dictEnd : iHighLimit;
+                        size_t forwardPatternLength = LZ4HC_countPattern(matchPtr+sizeof(pattern), iLimit, pattern) + sizeof(pattern);
+                        if (extDict && matchPtr + forwardPatternLength == iLimit) {
+                            U32 const rotatedPattern = LZ4HC_rotatePattern(forwardPatternLength, pattern);
+                            forwardPatternLength += LZ4HC_countPattern(prefixPtr, iHighLimit, rotatedPattern);
                         }
-        }   }   }   }
-    }  /* while ((matchIndex>=lowLimit) && (nbAttempts)) */
+                        {   const BYTE* const lowestMatchPtr = extDict ? dictStart : prefixPtr;
+                            size_t backLength = LZ4HC_reverseCountPattern(matchPtr, lowestMatchPtr, pattern);
+                            size_t currentSegmentLength;
+                            if (!extDict
+                              && matchPtr - backLength == prefixPtr
+                              && dictIdx < prefixIdx) {
+                                U32 const rotatedPattern = LZ4HC_rotatePattern((U32)(-(int)backLength), pattern);
+                                backLength += LZ4HC_reverseCountPattern(dictEnd, dictStart, rotatedPattern);
+                            }
+                            /* Limit backLength not go further than lowestMatchIndex */
+                            backLength = matchCandidateIdx - MAX(matchCandidateIdx - (U32)backLength, lowestMatchIndex);
+                            assert(matchCandidateIdx - backLength >= lowestMatchIndex);
+                            currentSegmentLength = backLength + forwardPatternLength;
+                            /* Adjust to end of pattern if the source pattern fits, otherwise the beginning of the pattern */
+                            if ( (currentSegmentLength >= srcPatternLength)   /* current pattern segment large enough to contain full srcPatternLength */
+                              && (forwardPatternLength <= srcPatternLength) ) { /* haven't reached this position yet */
+                                U32 const newMatchIndex = matchCandidateIdx + (U32)forwardPatternLength - (U32)srcPatternLength;  /* best position, full pattern, might be followed by more match */
+                                if (LZ4HC_protectDictEnd(prefixIdx, newMatchIndex))
+                                    matchIndex = newMatchIndex;
+                                else {
+                                    /* Can only happen if started in the prefix */
+                                    assert(newMatchIndex >= prefixIdx - 3 && newMatchIndex < prefixIdx && !extDict);
+                                    matchIndex = prefixIdx;
+                                }
+                            } else {
+                                U32 const newMatchIndex = matchCandidateIdx - (U32)backLength;   /* farthest position in current segment, will find a match of length currentSegmentLength + maybe some back */
+                                if (!LZ4HC_protectDictEnd(prefixIdx, newMatchIndex)) {
+                                    assert(newMatchIndex >= prefixIdx - 3 && newMatchIndex < prefixIdx && !extDict);
+                                    matchIndex = prefixIdx;
+                                } else {
+                                    matchIndex = newMatchIndex;
+                                    if (lookBackLength==0) {  /* no back possible */
+                                        size_t const maxML = MIN(currentSegmentLength, srcPatternLength);
+                                        if ((size_t)longest < maxML) {
+                                            assert(prefixPtr - prefixIdx + matchIndex != ip);
+                                            if ((size_t)(ip - prefixPtr) + prefixIdx - matchIndex > LZ4_DISTANCE_MAX) break;
+                                            assert(maxML < 2 GB);
+                                            longest = (int)maxML;
+                                            *matchpos = prefixPtr - prefixIdx + matchIndex;   /* virtual pos, relative to ip, to retrieve offset */
+                                            *startpos = ip;
+                                        }
+                                        {   U32 const distToNextPattern = DELTANEXTU16(chainTable, matchIndex);
+                                            if (distToNextPattern > matchIndex) break;  /* avoid overflow */
+                                            matchIndex -= distToNextPattern;
+                        }   }   }   }   }
+                        continue;
+                }   }
+        }   }   /* PA optimization */
+
+        /* follow current chain */
+        matchIndex -= DELTANEXTU16(chainTable, matchIndex + matchChainPos);
+
+    }  /* while ((matchIndex>=lowestMatchIndex) && (nbAttempts)) */
+
+    if ( dict == usingDictCtxHc
+      && nbAttempts > 0
+      && ipIndex - lowestMatchIndex < LZ4_DISTANCE_MAX) {
+        size_t const dictEndOffset = (size_t)(dictCtx->end - dictCtx->prefixStart) + dictCtx->dictLimit;
+        U32 dictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
+        assert(dictEndOffset <= 1 GB);
+        matchIndex = dictMatchIndex + lowestMatchIndex - (U32)dictEndOffset;
+        while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
+            const BYTE* const matchPtr = dictCtx->prefixStart - dictCtx->dictLimit + dictMatchIndex;
+
+            if (LZ4_read32(matchPtr) == pattern) {
+                int mlt;
+                int back = 0;
+                const BYTE* vLimit = ip + (dictEndOffset - dictMatchIndex);
+                if (vLimit > iHighLimit) vLimit = iHighLimit;
+                mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->prefixStart) : 0;
+                mlt -= back;
+                if (mlt > longest) {
+                    longest = mlt;
+                    *matchpos = prefixPtr - prefixIdx + matchIndex + back;
+                    *startpos = ip + back;
+            }   }
+
+            {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex);
+                dictMatchIndex -= nextOffset;
+                matchIndex -= nextOffset;
+    }   }   }
 
     return longest;
 }
 
-LZ4_FORCE_INLINE
-int LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4,   /* Index table will be updated */
-                                 const BYTE* const ip, const BYTE* const iLimit,
-                                 const BYTE** matchpos,
-                                 const int maxNbAttempts,
-                                 const int patternAnalysis)
+LZ4_FORCE_INLINE int
+LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4,   /* Index table will be updated */
+                       const BYTE* const ip, const BYTE* const iLimit,
+                       const BYTE** matchpos,
+                       const int maxNbAttempts,
+                       const int patternAnalysis,
+                       const dictCtx_directive dict)
 {
     const BYTE* uselessPtr = ip;
     /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
      * but this won't be the case here, as we define iLowLimit==ip,
      * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
-    return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, matchpos, &uselessPtr, maxNbAttempts, patternAnalysis);
+    return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, matchpos, &uselessPtr, maxNbAttempts, patternAnalysis, 0 /*chainSwap*/, dict, favorCompressionRatio);
 }
 
-
-
-typedef enum {
-    noLimit = 0,
-    limitedOutput = 1,
-    limitedDestSize = 2,
-} limitedOutput_directive;
-
 /* LZ4HC_encodeSequence() :
  * @return : 0 if ok,
  *           1 if buffer issue detected */
 LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
-    const BYTE** ip,
-    BYTE** op,
-    const BYTE** anchor,
+    const BYTE** _ip,
+    BYTE** _op,
+    const BYTE** _anchor,
     int matchLength,
     const BYTE* const match,
     limitedOutput_directive limit,
     BYTE* oend)
 {
+#define ip      (*_ip)
+#define op      (*_op)
+#define anchor  (*_anchor)
+
     size_t length;
-    BYTE* const token = (*op)++;
+    BYTE* const token = op++;
 
-#if defined(LZ4_DEBUG) && (LZ4_DEBUG >= 2)
+#if defined(LZ4_DEBUG) && (LZ4_DEBUG >= 6)
     static const BYTE* start = NULL;
     static U32 totalCost = 0;
-    U32 const pos = (start==NULL) ? 0 : (U32)(*anchor - start);
-    U32 const ll = (U32)(*ip - *anchor);
+    U32 const pos = (start==NULL) ? 0 : (U32)(anchor - start);
+    U32 const ll = (U32)(ip - anchor);
     U32 const llAdd = (ll>=15) ? ((ll-15) / 255) + 1 : 0;
     U32 const mlAdd = (matchLength>=19) ? ((matchLength-19) / 255) + 1 : 0;
     U32 const cost = 1 + llAdd + ll + 2 + mlAdd;
-    if (start==NULL) start = *anchor;  /* only works for single segment */
-    //g_debuglog_enable = (pos >= 2228) & (pos <= 2262);
-    DEBUGLOG(2, "pos:%7u -- literals:%3u, match:%4i, offset:%5u, cost:%3u + %u",
+    if (start==NULL) start = anchor;  /* only works for single segment */
+    /* g_debuglog_enable = (pos >= 2228) & (pos <= 2262); */
+    DEBUGLOG(6, "pos:%7u -- literals:%4u, match:%4i, offset:%5u, cost:%4u + %5u",
                 pos,
-                (U32)(*ip - *anchor), matchLength, (U32)(*ip-match),
+                (U32)(ip - anchor), matchLength, (U32)(ip-match),
                 cost, totalCost);
     totalCost += cost;
 #endif
 
     /* Encode Literal length */
-    length = (size_t)(*ip - *anchor);
-    if ((limit) && ((*op + (length >> 8) + length + (2 + 1 + LASTLITERALS)) > oend)) return 1;   /* Check output limit */
+    length = (size_t)(ip - anchor);
+    LZ4_STATIC_ASSERT(notLimited == 0);
+    /* Check output limit */
+    if (limit && ((op + (length / 255) + length + (2 + 1 + LASTLITERALS)) > oend)) {
+        DEBUGLOG(6, "Not enough room to write %i literals (%i bytes remaining)",
+                (int)length, (int)(oend - op));
+        return 1;
+    }
     if (length >= RUN_MASK) {
         size_t len = length - RUN_MASK;
         *token = (RUN_MASK << ML_BITS);
-        for(; len >= 255 ; len -= 255) *(*op)++ = 255;
-        *(*op)++ = (BYTE)len;
+        for(; len >= 255 ; len -= 255) *op++ = 255;
+        *op++ = (BYTE)len;
     } else {
         *token = (BYTE)(length << ML_BITS);
     }
 
     /* Copy Literals */
-    LZ4_wildCopy(*op, *anchor, (*op) + length);
-    *op += length;
+    LZ4_wildCopy8(op, anchor, op + length);
+    op += length;
 
     /* Encode Offset */
-    LZ4_writeLE16(*op, (U16)(*ip-match)); *op += 2;
+    assert( (ip - match) <= LZ4_DISTANCE_MAX );   /* note : consider providing offset as a value, rather than as a pointer difference */
+    LZ4_writeLE16(op, (U16)(ip - match)); op += 2;
 
     /* Encode MatchLength */
     assert(matchLength >= MINMATCH);
-    length = (size_t)(matchLength - MINMATCH);
-    if ((limit) && (*op + (length >> 8) + (1 + LASTLITERALS) > oend)) return 1;   /* Check output limit */
+    length = (size_t)matchLength - MINMATCH;
+    if (limit && (op + (length / 255) + (1 + LASTLITERALS) > oend)) {
+        DEBUGLOG(6, "Not enough room to write match length");
+        return 1;   /* Check output limit */
+    }
     if (length >= ML_MASK) {
         *token += ML_MASK;
         length -= ML_MASK;
-        for(; length >= 510 ; length -= 510) { *(*op)++ = 255; *(*op)++ = 255; }
-        if (length >= 255) { length -= 255; *(*op)++ = 255; }
-        *(*op)++ = (BYTE)length;
+        for(; length >= 510 ; length -= 510) { *op++ = 255; *op++ = 255; }
+        if (length >= 255) { length -= 255; *op++ = 255; }
+        *op++ = (BYTE)length;
     } else {
         *token += (BYTE)(length);
     }
 
     /* Prepare next loop */
-    *ip += matchLength;
-    *anchor = *ip;
+    ip += matchLength;
+    anchor = ip;
 
     return 0;
 }
+#undef ip
+#undef op
+#undef anchor
 
-/* btopt */
-#include "lz4opt.h"
-
-
-static int LZ4HC_compress_hashChain (
+LZ4_FORCE_INLINE int LZ4HC_compress_hashChain (
     LZ4HC_CCtx_internal* const ctx,
     const char* const source,
     char* const dest,
     int* srcSizePtr,
     int const maxOutputSize,
-    unsigned maxNbAttempts,
-    limitedOutput_directive limit
+    int maxNbAttempts,
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
     )
 {
     const int inputSize = *srcSizePtr;
-    const int patternAnalysis = (maxNbAttempts > 64);   /* levels 8+ */
+    const int patternAnalysis = (maxNbAttempts > 128);   /* levels 9+ */
 
     const BYTE* ip = (const BYTE*) source;
     const BYTE* anchor = ip;
@@ -417,51 +574,47 @@ static int LZ4HC_compress_hashChain (
     BYTE* op = (BYTE*) dest;
     BYTE* oend = op + maxOutputSize;
 
-    int   ml, ml2, ml3, ml0;
+    int   ml0, ml, ml2, ml3;
+    const BYTE* start0;
+    const BYTE* ref0;
     const BYTE* ref = NULL;
     const BYTE* start2 = NULL;
     const BYTE* ref2 = NULL;
     const BYTE* start3 = NULL;
     const BYTE* ref3 = NULL;
-    const BYTE* start0;
-    const BYTE* ref0;
 
     /* init */
     *srcSizePtr = 0;
-    if (limit == limitedDestSize) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
-    if (inputSize < LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
+    if (limit == fillOutput) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
+    if (inputSize < LZ4_minLength) goto _last_literals;             /* Input too small, no compression (all literals) */
 
     /* Main Loop */
-    while (ip < mflimit) {
-        ml = LZ4HC_InsertAndFindBestMatch (ctx, ip, matchlimit, &ref, maxNbAttempts, patternAnalysis);
+    while (ip <= mflimit) {
+        ml = LZ4HC_InsertAndFindBestMatch(ctx, ip, matchlimit, &ref, maxNbAttempts, patternAnalysis, dict);
         if (ml<MINMATCH) { ip++; continue; }
 
         /* saved, in case we would skip too much */
-        start0 = ip;
-        ref0 = ref;
-        ml0 = ml;
+        start0 = ip; ref0 = ref; ml0 = ml;
 
 _Search2:
-        if (ip+ml < mflimit)
+        if (ip+ml <= mflimit) {
             ml2 = LZ4HC_InsertAndGetWiderMatch(ctx,
                             ip + ml - 2, ip + 0, matchlimit, ml, &ref2, &start2,
-                            maxNbAttempts, patternAnalysis);
-        else
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+        } else {
             ml2 = ml;
+        }
 
-        if (ml2 == ml) { /* No better match */
+        if (ml2 == ml) { /* No better match => encode ML1 */
             optr = op;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
             continue;
         }
 
-        if (start0 < ip) {
-            if (start2 < ip + ml0) {  /* empirical */
-                ip = start0;
-                ref = ref0;
-                ml = ml0;
-            }
-        }
+        if (start0 < ip) {   /* first match was skipped at least once */
+            if (start2 < ip + ml0) {  /* squeezing ML1 between ML0(original ML1) and ML2 */
+                ip = start0; ref = ref0; ml = ml0;  /* restore initial ML1 */
+        }   }
 
         /* Here, start0==ip */
         if ((start2 - ip) < 3) {  /* First Match too small : removed */
@@ -489,22 +642,27 @@ _Search3:
         }
         /* Now, we have start2 = ip+new_ml, with new_ml = min(ml, OPTIMAL_ML=18) */
 
-        if (start2 + ml2 < mflimit)
+        if (start2 + ml2 <= mflimit) {
             ml3 = LZ4HC_InsertAndGetWiderMatch(ctx,
                             start2 + ml2 - 3, start2, matchlimit, ml2, &ref3, &start3,
-                            maxNbAttempts, patternAnalysis);
-        else
+                            maxNbAttempts, patternAnalysis, 0, dict, favorCompressionRatio);
+        } else {
             ml3 = ml2;
+        }
 
-        if (ml3 == ml2) {  /* No better match : 2 sequences to encode */
+        if (ml3 == ml2) {  /* No better match => encode ML1 and ML2 */
             /* ip & ref are known; Now for ml */
             if (start2 < ip+ml)  ml = (int)(start2 - ip);
             /* Now, encode 2 sequences */
             optr = op;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
             ip = start2;
             optr = op;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml2, ref2, limit, oend)) goto _dest_overflow;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml2, ref2, limit, oend)) {
+                ml  = ml2;
+                ref = ref2;
+                goto _dest_overflow;
+            }
             continue;
         }
 
@@ -523,7 +681,7 @@ _Search3:
                 }
 
                 optr = op;
-                if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+                if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
                 ip  = start3;
                 ref = ref3;
                 ml  = ml3;
@@ -541,11 +699,12 @@ _Search3:
         }
 
         /*
-        * OK, now we have 3 ascending matches; let's write at least the first one
-        * ip & ref are known; Now for ml
+        * OK, now we have 3 ascending matches;
+        * let's write the first one ML1.
+        * ip & ref are known; Now decide ml.
         */
         if (start2 < ip+ml) {
-            if ((start2 - ip) < (int)ML_MASK) {
+            if ((start2 - ip) < OPTIMAL_ML) {
                 int correction;
                 if (ml > OPTIMAL_ML) ml = OPTIMAL_ML;
                 if (ip + ml > start2 + ml2 - MINMATCH) ml = (int)(start2 - ip) + ml2 - MINMATCH;
@@ -560,33 +719,33 @@ _Search3:
             }
         }
         optr = op;
-        if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+        if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
 
-        ip = start2;
-        ref = ref2;
-        ml = ml2;
+        /* ML2 becomes ML1 */
+        ip = start2; ref = ref2; ml = ml2;
 
-        start2 = start3;
-        ref2 = ref3;
-        ml2 = ml3;
+        /* ML3 becomes ML2 */
+        start2 = start3; ref2 = ref3; ml2 = ml3;
 
+        /* let's find a new ML3 */
         goto _Search3;
     }
 
 _last_literals:
     /* Encode Last Literals */
     {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
-        size_t litLength = (lastRunSize + 255 - RUN_MASK) / 255;
-        size_t const totalSize = 1 + litLength + lastRunSize;
-        if (limit == limitedDestSize) oend += LASTLITERALS;  /* restore correct value */
+        size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+        size_t const totalSize = 1 + llAdd + lastRunSize;
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
         if (limit && (op + totalSize > oend)) {
-            if (limit == limitedOutput) return 0;  /* Check output limit */
+            if (limit == limitedOutput) return 0;
             /* adapt lastRunSize to fill 'dest' */
-            lastRunSize  = (size_t)(oend - op) - 1;
-            litLength = (lastRunSize + 255 - RUN_MASK) / 255;
-            lastRunSize -= litLength;
+            lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+            llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+            lastRunSize -= llAdd;
         }
-        ip = anchor + lastRunSize;
+        DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+        ip = anchor + lastRunSize;  /* can be != iend if limit==fillOutput */
 
         if (lastRunSize >= RUN_MASK) {
             size_t accumulator = lastRunSize - RUN_MASK;
@@ -596,7 +755,7 @@ _last_literals:
         } else {
             *op++ = (BYTE)(lastRunSize << ML_BITS);
         }
-        memcpy(op, anchor, lastRunSize);
+        LZ4_memcpy(op, anchor, lastRunSize);
         op += lastRunSize;
     }
 
@@ -605,103 +764,222 @@ _last_literals:
     return (int) (((char*)op)-dest);
 
 _dest_overflow:
-    if (limit == limitedDestSize) {
+    if (limit == fillOutput) {
+        /* Assumption : ip, anchor, ml and ref must be set correctly */
+        size_t const ll = (size_t)(ip - anchor);
+        size_t const ll_addbytes = (ll + 240) / 255;
+        size_t const ll_totalCost = 1 + ll_addbytes + ll;
+        BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+        DEBUGLOG(6, "Last sequence overflowing");
         op = optr;  /* restore correct out pointer */
+        if (op + ll_totalCost <= maxLitPos) {
+            /* ll validated; now adjust match length */
+            size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+            size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+            assert(maxMlSize < INT_MAX); assert(ml >= 0);
+            if ((size_t)ml > maxMlSize) ml = (int)maxMlSize;
+            if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + ml >= MFLIMIT) {
+                LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, notLimited, oend);
+        }   }
         goto _last_literals;
     }
+    /* compression failed */
     return 0;
 }
 
 
-static int LZ4HC_compress_generic (
+static int LZ4HC_compress_optimal( LZ4HC_CCtx_internal* ctx,
+    const char* const source, char* dst,
+    int* srcSizePtr, int dstCapacity,
+    int const nbSearches, size_t sufficient_len,
+    const limitedOutput_directive limit, int const fullUpdate,
+    const dictCtx_directive dict,
+    const HCfavor_e favorDecSpeed);
+
+
+LZ4_FORCE_INLINE int LZ4HC_compress_generic_internal (
     LZ4HC_CCtx_internal* const ctx,
     const char* const src,
     char* const dst,
     int* const srcSizePtr,
     int const dstCapacity,
     int cLevel,
-    limitedOutput_directive limit
+    const limitedOutput_directive limit,
+    const dictCtx_directive dict
     )
 {
     typedef enum { lz4hc, lz4opt } lz4hc_strat_e;
     typedef struct {
         lz4hc_strat_e strat;
-        U32 nbSearches;
+        int nbSearches;
         U32 targetLength;
     } cParams_t;
     static const cParams_t clTable[LZ4HC_CLEVEL_MAX+1] = {
-        { lz4hc,    2, 16 },  /* 0, unused */
-        { lz4hc,    2, 16 },  /* 1, unused */
-        { lz4hc,    2, 16 },  /* 2, unused */
-        { lz4hc,    4, 16 },  /* 3 */
-        { lz4hc,    8, 16 },  /* 4 */
-        { lz4hc,   16, 16 },  /* 5 */
-        { lz4hc,   32, 16 },  /* 6 */
-        { lz4hc,   64, 16 },  /* 7 */
-        { lz4hc,  128, 16 },  /* 8 */
-        { lz4hc,  256, 16 },  /* 9 */
-        { lz4opt,  96, 64 },  /*10==LZ4HC_CLEVEL_OPT_MIN*/
-        { lz4opt, 512,128 },  /*11 */
-        { lz4opt,8192, LZ4_OPT_NUM },  /* 12==LZ4HC_CLEVEL_MAX */
+        { lz4hc,     2, 16 },  /* 0, unused */
+        { lz4hc,     2, 16 },  /* 1, unused */
+        { lz4hc,     2, 16 },  /* 2, unused */
+        { lz4hc,     4, 16 },  /* 3 */
+        { lz4hc,     8, 16 },  /* 4 */
+        { lz4hc,    16, 16 },  /* 5 */
+        { lz4hc,    32, 16 },  /* 6 */
+        { lz4hc,    64, 16 },  /* 7 */
+        { lz4hc,   128, 16 },  /* 8 */
+        { lz4hc,   256, 16 },  /* 9 */
+        { lz4opt,   96, 64 },  /*10==LZ4HC_CLEVEL_OPT_MIN*/
+        { lz4opt,  512,128 },  /*11 */
+        { lz4opt,16384,LZ4_OPT_NUM },  /* 12==LZ4HC_CLEVEL_MAX */
     };
 
-    if (limit == limitedDestSize && dstCapacity < 1) return 0;         /* Impossible to store anything */
-    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;          /* Unsupported input size (too large or negative) */
+    DEBUGLOG(4, "LZ4HC_compress_generic(ctx=%p, src=%p, srcSize=%d, limit=%d)",
+                ctx, src, *srcSizePtr, limit);
+
+    if (limit == fillOutput && dstCapacity < 1) return 0;   /* Impossible to store anything */
+    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;    /* Unsupported input size (too large or negative) */
 
     ctx->end += *srcSizePtr;
     if (cLevel < 1) cLevel = LZ4HC_CLEVEL_DEFAULT;   /* note : convention is different from lz4frame, maybe something to review */
     cLevel = MIN(LZ4HC_CLEVEL_MAX, cLevel);
-    assert(cLevel >= 0);
-    assert(cLevel <= LZ4HC_CLEVEL_MAX);
     {   cParams_t const cParam = clTable[cLevel];
-        if (cParam.strat == lz4hc)
-            return LZ4HC_compress_hashChain(ctx,
+        HCfavor_e const favor = ctx->favorDecSpeed ? favorDecompressionSpeed : favorCompressionRatio;
+        int result;
+
+        if (cParam.strat == lz4hc) {
+            result = LZ4HC_compress_hashChain(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                cParam.nbSearches, limit, dict);
+        } else {
+            assert(cParam.strat == lz4opt);
+            result = LZ4HC_compress_optimal(ctx,
                                 src, dst, srcSizePtr, dstCapacity,
-                                cParam.nbSearches, limit);
-        assert(cParam.strat == lz4opt);
-        return LZ4HC_compress_optimal(ctx,
-                            src, dst, srcSizePtr, dstCapacity,
-                            cParam.nbSearches, cParam.targetLength, limit,
-                            cLevel == LZ4HC_CLEVEL_MAX);  /* ultra mode */
+                                cParam.nbSearches, cParam.targetLength, limit,
+                                cLevel == LZ4HC_CLEVEL_MAX,   /* ultra mode */
+                                dict, favor);
+        }
+        if (result <= 0) ctx->dirty = 1;
+        return result;
     }
 }
 
+static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock);
+
+static int
+LZ4HC_compress_generic_noDictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    assert(ctx->dictCtx == NULL);
+    return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, noDictCtx);
+}
 
-int LZ4_sizeofStateHC(void) { return sizeof(LZ4_streamHC_t); }
+static int
+LZ4HC_compress_generic_dictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    const size_t position = (size_t)(ctx->end - ctx->prefixStart) + (ctx->dictLimit - ctx->lowLimit);
+    assert(ctx->dictCtx != NULL);
+    if (position >= 64 KB) {
+        ctx->dictCtx = NULL;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else if (position == 0 && *srcSizePtr > 4 KB) {
+        LZ4_memcpy(ctx, ctx->dictCtx, sizeof(LZ4HC_CCtx_internal));
+        LZ4HC_setExternalDict(ctx, (const BYTE *)src);
+        ctx->compressionLevel = (short)cLevel;
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, usingDictCtxHc);
+    }
+}
 
-int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+static int
+LZ4HC_compress_generic (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
+{
+    if (ctx->dictCtx == NULL) {
+        return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    } else {
+        return LZ4HC_compress_generic_dictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
+    }
+}
+
+
+int LZ4_sizeofStateHC(void) { return (int)sizeof(LZ4_streamHC_t); }
+
+static size_t LZ4_streamHC_t_alignment(void)
+{
+#if LZ4_ALIGN_TEST
+    typedef struct { char c; LZ4_streamHC_t t; } t_a;
+    return sizeof(t_a) - sizeof(LZ4_streamHC_t);
+#else
+    return 1;  /* effectively disabled */
+#endif
+}
+
+/* state is presumed correctly initialized,
+ * in which case its size and alignment have already been validate */
+int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
 {
     LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
-    if (((size_t)(state)&(sizeof(void*)-1)) != 0) return 0;   /* Error : state is not aligned for pointers (32 or 64 bits) */
-    LZ4HC_init (ctx, (const BYTE*)src);
+    if (!LZ4_isAligned(state, LZ4_streamHC_t_alignment())) return 0;
+    LZ4_resetStreamHC_fast((LZ4_streamHC_t*)state, compressionLevel);
+    LZ4HC_init_internal (ctx, (const BYTE*)src);
     if (dstCapacity < LZ4_compressBound(srcSize))
         return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, limitedOutput);
     else
-        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, noLimit);
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, notLimited);
+}
+
+int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
+{
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    return LZ4_compress_HC_extStateHC_fastReset(state, src, dst, srcSize, dstCapacity, compressionLevel);
 }
 
 int LZ4_compress_HC(const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
 {
+    int cSize;
 #if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
-    LZ4_streamHC_t* const statePtr = (LZ4_streamHC_t*)malloc(sizeof(LZ4_streamHC_t));
+    LZ4_streamHC_t* const statePtr = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
+    if (statePtr==NULL) return 0;
 #else
     LZ4_streamHC_t state;
     LZ4_streamHC_t* const statePtr = &state;
 #endif
-    int const cSize = LZ4_compress_HC_extStateHC(statePtr, src, dst, srcSize, dstCapacity, compressionLevel);
+    cSize = LZ4_compress_HC_extStateHC(statePtr, src, dst, srcSize, dstCapacity, compressionLevel);
 #if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
-    free(statePtr);
+    FREEMEM(statePtr);
 #endif
     return cSize;
 }
 
-/* LZ4_compress_HC_destSize() :
- * only compatible with regular HC parser */
-int LZ4_compress_HC_destSize(void* LZ4HC_Data, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
+/* state is presumed sized correctly (>= sizeof(LZ4_streamHC_t)) */
+int LZ4_compress_HC_destSize(void* state, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
 {
-    LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse;
-    LZ4HC_init(ctx, (const BYTE*) source);
-    return LZ4HC_compress_generic(ctx, source, dest, sourceSizePtr, targetDestSize, cLevel, limitedDestSize);
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    LZ4HC_init_internal(&ctx->internal_donotuse, (const BYTE*) source);
+    LZ4_setCompressionLevel(ctx, cLevel);
+    return LZ4HC_compress_generic(&ctx->internal_donotuse, source, dest, sourceSizePtr, targetDestSize, cLevel, fillOutput);
 }
 
 
@@ -710,87 +988,162 @@ int LZ4_compress_HC_destSize(void* LZ4HC_Data, const char* source, char* dest, i
 *  Streaming Functions
 **************************************/
 /* allocation */
-LZ4_streamHC_t* LZ4_createStreamHC(void) { return (LZ4_streamHC_t*)malloc(sizeof(LZ4_streamHC_t)); }
-int             LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr) {
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+LZ4_streamHC_t* LZ4_createStreamHC(void)
+{
+    LZ4_streamHC_t* const state =
+        (LZ4_streamHC_t*)ALLOC_AND_ZERO(sizeof(LZ4_streamHC_t));
+    if (state == NULL) return NULL;
+    LZ4_setCompressionLevel(state, LZ4HC_CLEVEL_DEFAULT);
+    return state;
+}
+
+int LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr)
+{
+    DEBUGLOG(4, "LZ4_freeStreamHC(%p)", LZ4_streamHCPtr);
     if (!LZ4_streamHCPtr) return 0;  /* support free on NULL */
-    free(LZ4_streamHCPtr);
+    FREEMEM(LZ4_streamHCPtr);
     return 0;
 }
+#endif
 
 
-/* initialization */
+LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size)
+{
+    LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)buffer;
+    DEBUGLOG(4, "LZ4_initStreamHC(%p, %u)", buffer, (unsigned)size);
+    /* check conditions */
+    if (buffer == NULL) return NULL;
+    if (size < sizeof(LZ4_streamHC_t)) return NULL;
+    if (!LZ4_isAligned(buffer, LZ4_streamHC_t_alignment())) return NULL;
+    /* init */
+    { LZ4HC_CCtx_internal* const hcstate = &(LZ4_streamHCPtr->internal_donotuse);
+      MEM_INIT(hcstate, 0, sizeof(*hcstate)); }
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, LZ4HC_CLEVEL_DEFAULT);
+    return LZ4_streamHCPtr;
+}
+
+/* just a stub */
 void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
 {
-    LZ4_STATIC_ASSERT(sizeof(LZ4HC_CCtx_internal) <= sizeof(size_t) * LZ4_STREAMHCSIZE_SIZET);   /* if compilation fails here, LZ4_STREAMHCSIZE must be increased */
-    LZ4_streamHCPtr->internal_donotuse.base = NULL;
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
+}
+
+void LZ4_resetStreamHC_fast (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    DEBUGLOG(4, "LZ4_resetStreamHC_fast(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (LZ4_streamHCPtr->internal_donotuse.dirty) {
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    } else {
+        /* preserve end - prefixStart : can trigger clearTable's threshold */
+        if (LZ4_streamHCPtr->internal_donotuse.end != NULL) {
+            LZ4_streamHCPtr->internal_donotuse.end -= (uptrval)LZ4_streamHCPtr->internal_donotuse.prefixStart;
+        } else {
+            assert(LZ4_streamHCPtr->internal_donotuse.prefixStart == NULL);
+        }
+        LZ4_streamHCPtr->internal_donotuse.prefixStart = NULL;
+        LZ4_streamHCPtr->internal_donotuse.dictCtx = NULL;
+    }
     LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
 }
 
 void LZ4_setCompressionLevel(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
 {
-    if (compressionLevel < 1) compressionLevel = 1;
+    DEBUGLOG(5, "LZ4_setCompressionLevel(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    if (compressionLevel < 1) compressionLevel = LZ4HC_CLEVEL_DEFAULT;
     if (compressionLevel > LZ4HC_CLEVEL_MAX) compressionLevel = LZ4HC_CLEVEL_MAX;
-    LZ4_streamHCPtr->internal_donotuse.compressionLevel = compressionLevel;
+    LZ4_streamHCPtr->internal_donotuse.compressionLevel = (short)compressionLevel;
 }
 
-int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, const char* dictionary, int dictSize)
+void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor)
+{
+    LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = (favor!=0);
+}
+
+/* LZ4_loadDictHC() :
+ * LZ4_streamHCPtr is presumed properly initialized */
+int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr,
+              const char* dictionary, int dictSize)
 {
     LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(4, "LZ4_loadDictHC(ctx:%p, dict:%p, dictSize:%d)", LZ4_streamHCPtr, dictionary, dictSize);
+    assert(LZ4_streamHCPtr != NULL);
     if (dictSize > 64 KB) {
-        dictionary += dictSize - 64 KB;
+        dictionary += (size_t)dictSize - 64 KB;
         dictSize = 64 KB;
     }
-    LZ4HC_init (ctxPtr, (const BYTE*)dictionary);
+    /* need a full initialization, there are bad side-effects when using resetFast() */
+    {   int const cLevel = ctxPtr->compressionLevel;
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+        LZ4_setCompressionLevel(LZ4_streamHCPtr, cLevel);
+    }
+    LZ4HC_init_internal (ctxPtr, (const BYTE*)dictionary);
     ctxPtr->end = (const BYTE*)dictionary + dictSize;
     if (dictSize >= 4) LZ4HC_Insert (ctxPtr, ctxPtr->end-3);
     return dictSize;
 }
 
+void LZ4_attach_HC_dictionary(LZ4_streamHC_t *working_stream, const LZ4_streamHC_t *dictionary_stream) {
+    working_stream->internal_donotuse.dictCtx = dictionary_stream != NULL ? &(dictionary_stream->internal_donotuse) : NULL;
+}
 
 /* compression */
 
 static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock)
 {
-    if (ctxPtr->end >= ctxPtr->base + 4) LZ4HC_Insert (ctxPtr, ctxPtr->end-3);   /* Referencing remaining dictionary content */
+    DEBUGLOG(4, "LZ4HC_setExternalDict(%p, %p)", ctxPtr, newBlock);
+    if (ctxPtr->end >= ctxPtr->prefixStart + 4)
+        LZ4HC_Insert (ctxPtr, ctxPtr->end-3);   /* Referencing remaining dictionary content */
 
     /* Only one memory segment for extDict, so any previous extDict is lost at this stage */
     ctxPtr->lowLimit  = ctxPtr->dictLimit;
-    ctxPtr->dictLimit = (U32)(ctxPtr->end - ctxPtr->base);
-    ctxPtr->dictBase  = ctxPtr->base;
-    ctxPtr->base = newBlock - ctxPtr->dictLimit;
+    ctxPtr->dictStart  = ctxPtr->prefixStart;
+    ctxPtr->dictLimit += (U32)(ctxPtr->end - ctxPtr->prefixStart);
+    ctxPtr->prefixStart = newBlock;
     ctxPtr->end  = newBlock;
     ctxPtr->nextToUpdate = ctxPtr->dictLimit;   /* match referencing will resume from there */
+
+    /* cannot reference an extDict and a dictCtx at the same time */
+    ctxPtr->dictCtx = NULL;
 }
 
-static int LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
-                                            const char* src, char* dst,
-                                            int* srcSizePtr, int dstCapacity,
-                                            limitedOutput_directive limit)
+static int
+LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
+                                 const char* src, char* dst,
+                                 int* srcSizePtr, int dstCapacity,
+                                 limitedOutput_directive limit)
 {
     LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
+    DEBUGLOG(5, "LZ4_compressHC_continue_generic(ctx=%p, src=%p, srcSize=%d, limit=%d)",
+                LZ4_streamHCPtr, src, *srcSizePtr, limit);
+    assert(ctxPtr != NULL);
     /* auto-init if forgotten */
-    if (ctxPtr->base == NULL) LZ4HC_init (ctxPtr, (const BYTE*) src);
+    if (ctxPtr->prefixStart == NULL) LZ4HC_init_internal (ctxPtr, (const BYTE*) src);
 
     /* Check overflow */
-    if ((size_t)(ctxPtr->end - ctxPtr->base) > 2 GB) {
-        size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->base) - ctxPtr->dictLimit;
+    if ((size_t)(ctxPtr->end - ctxPtr->prefixStart) + ctxPtr->dictLimit > 2 GB) {
+        size_t dictSize = (size_t)(ctxPtr->end - ctxPtr->prefixStart);
         if (dictSize > 64 KB) dictSize = 64 KB;
         LZ4_loadDictHC(LZ4_streamHCPtr, (const char*)(ctxPtr->end) - dictSize, (int)dictSize);
     }
 
     /* Check if blocks follow each other */
-    if ((const BYTE*)src != ctxPtr->end) LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
+    if ((const BYTE*)src != ctxPtr->end)
+        LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
 
     /* Check overlapping input/dictionary space */
     {   const BYTE* sourceEnd = (const BYTE*) src + *srcSizePtr;
-        const BYTE* const dictBegin = ctxPtr->dictBase + ctxPtr->lowLimit;
-        const BYTE* const dictEnd   = ctxPtr->dictBase + ctxPtr->dictLimit;
+        const BYTE* const dictBegin = ctxPtr->dictStart;
+        const BYTE* const dictEnd   = ctxPtr->dictStart + (ctxPtr->dictLimit - ctxPtr->lowLimit);
         if ((sourceEnd > dictBegin) && ((const BYTE*)src < dictEnd)) {
             if (sourceEnd > dictEnd) sourceEnd = dictEnd;
-            ctxPtr->lowLimit = (U32)(sourceEnd - ctxPtr->dictBase);
-            if (ctxPtr->dictLimit - ctxPtr->lowLimit < 4) ctxPtr->lowLimit = ctxPtr->dictLimit;
-        }
-    }
+            ctxPtr->lowLimit += (U32)(sourceEnd - ctxPtr->dictStart);
+            ctxPtr->dictStart += (U32)(sourceEnd - ctxPtr->dictStart);
+            if (ctxPtr->dictLimit - ctxPtr->lowLimit < 4) {
+                ctxPtr->lowLimit = ctxPtr->dictLimit;
+                ctxPtr->dictStart = ctxPtr->prefixStart;
+    }   }   }
 
     return LZ4HC_compress_generic (ctxPtr, src, dst, srcSizePtr, dstCapacity, ctxPtr->compressionLevel, limit);
 }
@@ -800,42 +1153,53 @@ int LZ4_compress_HC_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src,
     if (dstCapacity < LZ4_compressBound(srcSize))
         return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, limitedOutput);
     else
-        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, noLimit);
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, notLimited);
 }
 
 int LZ4_compress_HC_continue_destSize (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int* srcSizePtr, int targetDestSize)
 {
-    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, limitedDestSize);
+    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, fillOutput);
 }
 
 
 
-/* dictionary saving */
-
+/* LZ4_saveDictHC :
+ * save history content
+ * into a user-provided buffer
+ * which is then used to continue compression
+ */
 int LZ4_saveDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, char* safeBuffer, int dictSize)
 {
     LZ4HC_CCtx_internal* const streamPtr = &LZ4_streamHCPtr->internal_donotuse;
-    int const prefixSize = (int)(streamPtr->end - (streamPtr->base + streamPtr->dictLimit));
+    int const prefixSize = (int)(streamPtr->end - streamPtr->prefixStart);
+    DEBUGLOG(5, "LZ4_saveDictHC(%p, %p, %d)", LZ4_streamHCPtr, safeBuffer, dictSize);
+    assert(prefixSize >= 0);
     if (dictSize > 64 KB) dictSize = 64 KB;
     if (dictSize < 4) dictSize = 0;
     if (dictSize > prefixSize) dictSize = prefixSize;
-    memmove(safeBuffer, streamPtr->end - dictSize, dictSize);
-    {   U32 const endIndex = (U32)(streamPtr->end - streamPtr->base);
+    if (safeBuffer == NULL) assert(dictSize == 0);
+    if (dictSize > 0)
+        LZ4_memmove(safeBuffer, streamPtr->end - dictSize, dictSize);
+    {   U32 const endIndex = (U32)(streamPtr->end - streamPtr->prefixStart) + streamPtr->dictLimit;
         streamPtr->end = (const BYTE*)safeBuffer + dictSize;
-        streamPtr->base = streamPtr->end - endIndex;
-        streamPtr->dictLimit = endIndex - dictSize;
-        streamPtr->lowLimit = endIndex - dictSize;
-        if (streamPtr->nextToUpdate < streamPtr->dictLimit) streamPtr->nextToUpdate = streamPtr->dictLimit;
+        streamPtr->prefixStart = streamPtr->end - dictSize;
+        streamPtr->dictLimit = endIndex - (U32)dictSize;
+        streamPtr->lowLimit = endIndex - (U32)dictSize;
+        streamPtr->dictStart = streamPtr->prefixStart;
+        if (streamPtr->nextToUpdate < streamPtr->dictLimit)
+            streamPtr->nextToUpdate = streamPtr->dictLimit;
     }
     return dictSize;
 }
 
 
-/***********************************
+/***************************************************
 *  Deprecated Functions
-***********************************/
+***************************************************/
+
 /* These functions currently generate deprecation warnings */
-/* Deprecated compression functions */
+
+/* Wrappers for deprecated compression functions */
 int LZ4_compressHC(const char* src, char* dst, int srcSize) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
 int LZ4_compressHC_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, 0); }
 int LZ4_compressHC2(const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
@@ -849,35 +1213,38 @@ int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* ctx, const char* src,
 
 
 /* Deprecated streaming functions */
-int LZ4_sizeofStreamStateHC(void) { return LZ4_STREAMHCSIZE; }
+int LZ4_sizeofStreamStateHC(void) { return sizeof(LZ4_streamHC_t); }
 
+/* state is presumed correctly sized, aka >= sizeof(LZ4_streamHC_t)
+ * @return : 0 on success, !=0 if error */
 int LZ4_resetStreamStateHC(void* state, char* inputBuffer)
 {
-    LZ4HC_CCtx_internal *ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
-    if ((((size_t)state) & (sizeof(void*)-1)) != 0) return 1;   /* Error : pointer is not aligned for pointer (32 or 64 bits) */
-    LZ4HC_init(ctx, (const BYTE*)inputBuffer);
-    ctx->inputBuffer = (BYTE*)inputBuffer;
+    LZ4_streamHC_t* const hc4 = LZ4_initStreamHC(state, sizeof(*hc4));
+    if (hc4 == NULL) return 1;   /* init failed */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
     return 0;
 }
 
-void* LZ4_createHC (char* inputBuffer)
+#if !defined(LZ4_STATIC_LINKING_ONLY_DISABLE_MEMORY_ALLOCATION)
+void* LZ4_createHC (const char* inputBuffer)
 {
-    LZ4_streamHC_t* hc4 = (LZ4_streamHC_t*)ALLOCATOR(1, sizeof(LZ4_streamHC_t));
+    LZ4_streamHC_t* const hc4 = LZ4_createStreamHC();
     if (hc4 == NULL) return NULL;   /* not enough memory */
-    LZ4HC_init (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
-    hc4->internal_donotuse.inputBuffer = (BYTE*)inputBuffer;
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
     return hc4;
 }
 
-int LZ4_freeHC (void* LZ4HC_Data) {
+int LZ4_freeHC (void* LZ4HC_Data)
+{
     if (!LZ4HC_Data) return 0;  /* support free on NULL */
     FREEMEM(LZ4HC_Data);
     return 0;
 }
+#endif
 
 int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int cLevel)
 {
-    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, noLimit);
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, notLimited);
 }
 
 int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int dstCapacity, int cLevel)
@@ -887,7 +1254,378 @@ int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* src, c
 
 char* LZ4_slideInputBufferHC(void* LZ4HC_Data)
 {
-    LZ4HC_CCtx_internal* const hc4 = &((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse;
-    int const dictSize = LZ4_saveDictHC((LZ4_streamHC_t*)LZ4HC_Data, (char*)(hc4->inputBuffer), 64 KB);
-    return (char*)(hc4->inputBuffer + dictSize);
+    LZ4_streamHC_t* const ctx = (LZ4_streamHC_t*)LZ4HC_Data;
+    const BYTE* bufferStart = ctx->internal_donotuse.prefixStart - ctx->internal_donotuse.dictLimit + ctx->internal_donotuse.lowLimit;
+    LZ4_resetStreamHC_fast(ctx, ctx->internal_donotuse.compressionLevel);
+    /* avoid const char * -> char * conversion warning :( */
+    return (char*)(uptrval)bufferStart;
+}
+
+
+/* ================================================
+ *  LZ4 Optimal parser (levels [LZ4HC_CLEVEL_OPT_MIN - LZ4HC_CLEVEL_MAX])
+ * ===============================================*/
+typedef struct {
+    int price;
+    int off;
+    int mlen;
+    int litlen;
+} LZ4HC_optimal_t;
+
+/* price in bytes */
+LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen)
+{
+    int price = litlen;
+    assert(litlen >= 0);
+    if (litlen >= (int)RUN_MASK)
+        price += 1 + ((litlen-(int)RUN_MASK) / 255);
+    return price;
+}
+
+
+/* requires mlen >= MINMATCH */
+LZ4_FORCE_INLINE int LZ4HC_sequencePrice(int litlen, int mlen)
+{
+    int price = 1 + 2 ; /* token + 16-bit offset */
+    assert(litlen >= 0);
+    assert(mlen >= MINMATCH);
+
+    price += LZ4HC_literalsPrice(litlen);
+
+    if (mlen >= (int)(ML_MASK+MINMATCH))
+        price += 1 + ((mlen-(int)(ML_MASK+MINMATCH)) / 255);
+
+    return price;
+}
+
+
+typedef struct {
+    int off;
+    int len;
+} LZ4HC_match_t;
+
+LZ4_FORCE_INLINE LZ4HC_match_t
+LZ4HC_FindLongerMatch(LZ4HC_CCtx_internal* const ctx,
+                      const BYTE* ip, const BYTE* const iHighLimit,
+                      int minLen, int nbSearches,
+                      const dictCtx_directive dict,
+                      const HCfavor_e favorDecSpeed)
+{
+    LZ4HC_match_t match = { 0 , 0 };
+    const BYTE* matchPtr = NULL;
+    /* note : LZ4HC_InsertAndGetWiderMatch() is able to modify the starting position of a match (*startpos),
+     * but this won't be the case here, as we define iLowLimit==ip,
+     * so LZ4HC_InsertAndGetWiderMatch() won't be allowed to search past ip */
+    int matchLength = LZ4HC_InsertAndGetWiderMatch(ctx, ip, ip, iHighLimit, minLen, &matchPtr, &ip, nbSearches, 1 /*patternAnalysis*/, 1 /*chainSwap*/, dict, favorDecSpeed);
+    if (matchLength <= minLen) return match;
+    if (favorDecSpeed) {
+        if ((matchLength>18) & (matchLength<=36)) matchLength=18;   /* favor shortcut */
+    }
+    match.len = matchLength;
+    match.off = (int)(ip-matchPtr);
+    return match;
+}
+
+
+static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
+                                    const char* const source,
+                                    char* dst,
+                                    int* srcSizePtr,
+                                    int dstCapacity,
+                                    int const nbSearches,
+                                    size_t sufficient_len,
+                                    const limitedOutput_directive limit,
+                                    int const fullUpdate,
+                                    const dictCtx_directive dict,
+                                    const HCfavor_e favorDecSpeed)
+{
+    int retval = 0;
+#define TRAILING_LITERALS 3
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    LZ4HC_optimal_t* const opt = (LZ4HC_optimal_t*)ALLOC(sizeof(LZ4HC_optimal_t) * (LZ4_OPT_NUM + TRAILING_LITERALS));
+#else
+    LZ4HC_optimal_t opt[LZ4_OPT_NUM + TRAILING_LITERALS];   /* ~64 KB, which is a bit large for stack... */
+#endif
+
+    const BYTE* ip = (const BYTE*) source;
+    const BYTE* anchor = ip;
+    const BYTE* const iend = ip + *srcSizePtr;
+    const BYTE* const mflimit = iend - MFLIMIT;
+    const BYTE* const matchlimit = iend - LASTLITERALS;
+    BYTE* op = (BYTE*) dst;
+    BYTE* opSaved = (BYTE*) dst;
+    BYTE* oend = op + dstCapacity;
+    int ovml = MINMATCH;  /* overflow - last sequence */
+    const BYTE* ovref = NULL;
+
+    /* init */
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+    if (opt == NULL) goto _return_label;
+#endif
+    DEBUGLOG(5, "LZ4HC_compress_optimal(dst=%p, dstCapa=%u)", dst, (unsigned)dstCapacity);
+    *srcSizePtr = 0;
+    if (limit == fillOutput) oend -= LASTLITERALS;   /* Hack for support LZ4 format restriction */
+    if (sufficient_len >= LZ4_OPT_NUM) sufficient_len = LZ4_OPT_NUM-1;
+
+    /* Main Loop */
+    while (ip <= mflimit) {
+         int const llen = (int)(ip - anchor);
+         int best_mlen, best_off;
+         int cur, last_match_pos = 0;
+
+         LZ4HC_match_t const firstMatch = LZ4HC_FindLongerMatch(ctx, ip, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+         if (firstMatch.len==0) { ip++; continue; }
+
+         if ((size_t)firstMatch.len > sufficient_len) {
+             /* good enough solution : immediate encoding */
+             int const firstML = firstMatch.len;
+             const BYTE* const matchPos = ip - firstMatch.off;
+             opSaved = op;
+             if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), firstML, matchPos, limit, oend) ) {  /* updates ip, op and anchor */
+                 ovml = firstML;
+                 ovref = matchPos;
+                 goto _dest_overflow;
+             }
+             continue;
+         }
+
+         /* set prices for first positions (literals) */
+         {   int rPos;
+             for (rPos = 0 ; rPos < MINMATCH ; rPos++) {
+                 int const cost = LZ4HC_literalsPrice(llen + rPos);
+                 opt[rPos].mlen = 1;
+                 opt[rPos].off = 0;
+                 opt[rPos].litlen = llen + rPos;
+                 opt[rPos].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             rPos, cost, opt[rPos].litlen);
+         }   }
+         /* set prices using initial match */
+         {   int mlen = MINMATCH;
+             int const matchML = firstMatch.len;   /* necessarily < sufficient_len < LZ4_OPT_NUM */
+             int const offset = firstMatch.off;
+             assert(matchML < LZ4_OPT_NUM);
+             for ( ; mlen <= matchML ; mlen++) {
+                 int const cost = LZ4HC_sequencePrice(llen, mlen);
+                 opt[mlen].mlen = mlen;
+                 opt[mlen].off = offset;
+                 opt[mlen].litlen = llen;
+                 opt[mlen].price = cost;
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i) -- initial setup",
+                             mlen, cost, mlen);
+         }   }
+         last_match_pos = firstMatch.len;
+         {   int addLit;
+             for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                 opt[last_match_pos+addLit].mlen = 1; /* literal */
+                 opt[last_match_pos+addLit].off = 0;
+                 opt[last_match_pos+addLit].litlen = addLit;
+                 opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
+                             last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+         }   }
+
+         /* check further positions */
+         for (cur = 1; cur < last_match_pos; cur++) {
+             const BYTE* const curPtr = ip + cur;
+             LZ4HC_match_t newMatch;
+
+             if (curPtr > mflimit) break;
+             DEBUGLOG(7, "rPos:%u[%u] vs [%u]%u",
+                     cur, opt[cur].price, opt[cur+1].price, cur+1);
+             if (fullUpdate) {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if ( (opt[cur+1].price <= opt[cur].price)
+                   /* in some cases, next position has same cost, but cost rises sharply after, so a small match would still be beneficial */
+                   && (opt[cur+MINMATCH].price < opt[cur].price + 3/*min seq price*/) )
+                     continue;
+             } else {
+                 /* not useful to search here if next position has same (or lower) cost */
+                 if (opt[cur+1].price <= opt[cur].price) continue;
+             }
+
+             DEBUGLOG(7, "search at rPos:%u", cur);
+             if (fullUpdate)
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, MINMATCH-1, nbSearches, dict, favorDecSpeed);
+             else
+                 /* only test matches of minimum length; slightly faster, but misses a few bytes */
+                 newMatch = LZ4HC_FindLongerMatch(ctx, curPtr, matchlimit, last_match_pos - cur, nbSearches, dict, favorDecSpeed);
+             if (!newMatch.len) continue;
+
+             if ( ((size_t)newMatch.len > sufficient_len)
+               || (newMatch.len + cur >= LZ4_OPT_NUM) ) {
+                 /* immediate encoding */
+                 best_mlen = newMatch.len;
+                 best_off = newMatch.off;
+                 last_match_pos = cur + 1;
+                 goto encode;
+             }
+
+             /* before match : set price with literals at beginning */
+             {   int const baseLitlen = opt[cur].litlen;
+                 int litlen;
+                 for (litlen = 1; litlen < MINMATCH; litlen++) {
+                     int const price = opt[cur].price - LZ4HC_literalsPrice(baseLitlen) + LZ4HC_literalsPrice(baseLitlen+litlen);
+                     int const pos = cur + litlen;
+                     if (price < opt[pos].price) {
+                         opt[pos].mlen = 1; /* literal */
+                         opt[pos].off = 0;
+                         opt[pos].litlen = baseLitlen+litlen;
+                         opt[pos].price = price;
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)",
+                                     pos, price, opt[pos].litlen);
+             }   }   }
+
+             /* set prices using match at position = cur */
+             {   int const matchML = newMatch.len;
+                 int ml = MINMATCH;
+
+                 assert(cur + newMatch.len < LZ4_OPT_NUM);
+                 for ( ; ml <= matchML ; ml++) {
+                     int const pos = cur + ml;
+                     int const offset = newMatch.off;
+                     int price;
+                     int ll;
+                     DEBUGLOG(7, "testing price rPos %i (last_match_pos=%i)",
+                                 pos, last_match_pos);
+                     if (opt[cur].mlen == 1) {
+                         ll = opt[cur].litlen;
+                         price = ((cur > ll) ? opt[cur - ll].price : 0)
+                               + LZ4HC_sequencePrice(ll, ml);
+                     } else {
+                         ll = 0;
+                         price = opt[cur].price + LZ4HC_sequencePrice(0, ml);
+                     }
+
+                    assert((U32)favorDecSpeed <= 1);
+                     if (pos > last_match_pos+TRAILING_LITERALS
+                      || price <= opt[pos].price - (int)favorDecSpeed) {
+                         DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i)",
+                                     pos, price, ml);
+                         assert(pos < LZ4_OPT_NUM);
+                         if ( (ml == matchML)  /* last pos of last match */
+                           && (last_match_pos < pos) )
+                             last_match_pos = pos;
+                         opt[pos].mlen = ml;
+                         opt[pos].off = offset;
+                         opt[pos].litlen = ll;
+                         opt[pos].price = price;
+             }   }   }
+             /* complete following positions with literals */
+             {   int addLit;
+                 for (addLit = 1; addLit <= TRAILING_LITERALS; addLit ++) {
+                     opt[last_match_pos+addLit].mlen = 1; /* literal */
+                     opt[last_match_pos+addLit].off = 0;
+                     opt[last_match_pos+addLit].litlen = addLit;
+                     opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
+                     DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)", last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
+             }   }
+         }  /* for (cur = 1; cur <= last_match_pos; cur++) */
+
+         assert(last_match_pos < LZ4_OPT_NUM + TRAILING_LITERALS);
+         best_mlen = opt[last_match_pos].mlen;
+         best_off = opt[last_match_pos].off;
+         cur = last_match_pos - best_mlen;
+
+encode: /* cur, last_match_pos, best_mlen, best_off must be set */
+         assert(cur < LZ4_OPT_NUM);
+         assert(last_match_pos >= 1);  /* == 1 when only one candidate */
+         DEBUGLOG(6, "reverse traversal, looking for shortest path (last_match_pos=%i)", last_match_pos);
+         {   int candidate_pos = cur;
+             int selected_matchLength = best_mlen;
+             int selected_offset = best_off;
+             while (1) {  /* from end to beginning */
+                 int const next_matchLength = opt[candidate_pos].mlen;  /* can be 1, means literal */
+                 int const next_offset = opt[candidate_pos].off;
+                 DEBUGLOG(7, "pos %i: sequence length %i", candidate_pos, selected_matchLength);
+                 opt[candidate_pos].mlen = selected_matchLength;
+                 opt[candidate_pos].off = selected_offset;
+                 selected_matchLength = next_matchLength;
+                 selected_offset = next_offset;
+                 if (next_matchLength > candidate_pos) break; /* last match elected, first match to encode */
+                 assert(next_matchLength > 0);  /* can be 1, means literal */
+                 candidate_pos -= next_matchLength;
+         }   }
+
+         /* encode all recorded sequences in order */
+         {   int rPos = 0;  /* relative position (to ip) */
+             while (rPos < last_match_pos) {
+                 int const ml = opt[rPos].mlen;
+                 int const offset = opt[rPos].off;
+                 if (ml == 1) { ip++; rPos++; continue; }  /* literal; note: can end up with several literals, in which case, skip them */
+                 rPos += ml;
+                 assert(ml >= MINMATCH);
+                 assert((offset >= 1) && (offset <= LZ4_DISTANCE_MAX));
+                 opSaved = op;
+                 if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ip - offset, limit, oend) ) {  /* updates ip, op and anchor */
+                     ovml = ml;
+                     ovref = ip - offset;
+                     goto _dest_overflow;
+         }   }   }
+     }  /* while (ip <= mflimit) */
+
+_last_literals:
+     /* Encode Last Literals */
+     {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
+         size_t llAdd = (lastRunSize + 255 - RUN_MASK) / 255;
+         size_t const totalSize = 1 + llAdd + lastRunSize;
+         if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
+         if (limit && (op + totalSize > oend)) {
+             if (limit == limitedOutput) { /* Check output limit */
+                retval = 0;
+                goto _return_label;
+             }
+             /* adapt lastRunSize to fill 'dst' */
+             lastRunSize  = (size_t)(oend - op) - 1 /*token*/;
+             llAdd = (lastRunSize + 256 - RUN_MASK) / 256;
+             lastRunSize -= llAdd;
+         }
+         DEBUGLOG(6, "Final literal run : %i literals", (int)lastRunSize);
+         ip = anchor + lastRunSize; /* can be != iend if limit==fillOutput */
+
+         if (lastRunSize >= RUN_MASK) {
+             size_t accumulator = lastRunSize - RUN_MASK;
+             *op++ = (RUN_MASK << ML_BITS);
+             for(; accumulator >= 255 ; accumulator -= 255) *op++ = 255;
+             *op++ = (BYTE) accumulator;
+         } else {
+             *op++ = (BYTE)(lastRunSize << ML_BITS);
+         }
+         LZ4_memcpy(op, anchor, lastRunSize);
+         op += lastRunSize;
+     }
+
+     /* End */
+     *srcSizePtr = (int) (((const char*)ip) - source);
+     retval = (int) ((char*)op-dst);
+     goto _return_label;
+
+_dest_overflow:
+if (limit == fillOutput) {
+     /* Assumption : ip, anchor, ovml and ovref must be set correctly */
+     size_t const ll = (size_t)(ip - anchor);
+     size_t const ll_addbytes = (ll + 240) / 255;
+     size_t const ll_totalCost = 1 + ll_addbytes + ll;
+     BYTE* const maxLitPos = oend - 3; /* 2 for offset, 1 for token */
+     DEBUGLOG(6, "Last sequence overflowing (only %i bytes remaining)", (int)(oend-1-opSaved));
+     op = opSaved;  /* restore correct out pointer */
+     if (op + ll_totalCost <= maxLitPos) {
+         /* ll validated; now adjust match length */
+         size_t const bytesLeftForMl = (size_t)(maxLitPos - (op+ll_totalCost));
+         size_t const maxMlSize = MINMATCH + (ML_MASK-1) + (bytesLeftForMl * 255);
+         assert(maxMlSize < INT_MAX); assert(ovml >= 0);
+         if ((size_t)ovml > maxMlSize) ovml = (int)maxMlSize;
+         if ((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1 + ovml >= MFLIMIT) {
+             DEBUGLOG(6, "Space to end : %i + ml (%i)", (int)((oend + LASTLITERALS) - (op + ll_totalCost + 2) - 1), ovml);
+             DEBUGLOG(6, "Before : ip = %p, anchor = %p", ip, anchor);
+             LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ovml, ovref, notLimited, oend);
+             DEBUGLOG(6, "After : ip = %p, anchor = %p", ip, anchor);
+     }   }
+     goto _last_literals;
+}
+_return_label:
+#if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
+     FREEMEM(opt);
+#endif
+     return retval;
 }