Add zstd compression support.

zstd has much better compression speed and ratio, and better decompression speed than currently available methods.
Also set zstd as the default compression method for Compression as well as FileAccessCompressed functions.

6 files changed, 6889 insertions, 0 deletions

diff --git a/thirdparty/zstd/compress/fse_compress.c b/thirdparty/zstd/compress/fse_compress.c
new file mode 100644
index 0000000000..26e8052ddc
--- /dev/null
+++ b/thirdparty/zstd/compress/fse_compress.c
@@ -0,0 +1,857 @@
+/* ******************************************************************
+   FSE : Finite State Entropy encoder
+   Copyright (C) 2013-2015, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
+#else
+#  if defined (__cplusplus) || defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+#    ifdef __GNUC__
+#      define FORCE_INLINE static inline __attribute__((always_inline))
+#    else
+#      define FORCE_INLINE static inline
+#    endif
+#  else
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <stdlib.h>     /* malloc, free, qsort */
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+#include "bitstream.h"
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+
+
+/* **************************************************************
+*  Templates
+****************************************************************/
+/*
+  designed to be included
+  for type-specific functions (template emulation in C)
+  Objective is to write these functions only once, for improved maintenance
+*/
+
+/* safety checks */
+#ifndef FSE_FUNCTION_EXTENSION
+#  error "FSE_FUNCTION_EXTENSION must be defined"
+#endif
+#ifndef FSE_FUNCTION_TYPE
+#  error "FSE_FUNCTION_TYPE must be defined"
+#endif
+
+/* Function names */
+#define FSE_CAT(X,Y) X##Y
+#define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
+#define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
+
+
+/* Function templates */
+
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
+ * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
+ */
+size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    U32 const tableSize = 1 << tableLog;
+    U32 const tableMask = tableSize - 1;
+    void* const ptr = ct;
+    U16* const tableU16 = ( (U16*) ptr) + 2;
+    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
+    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
+    U32 const step = FSE_TABLESTEP(tableSize);
+    U32 cumul[FSE_MAX_SYMBOL_VALUE+2];
+
+    FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)workSpace;
+    U32 highThreshold = tableSize-1;
+
+    /* CTable header */
+    if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
+    tableU16[-2] = (U16) tableLog;
+    tableU16[-1] = (U16) maxSymbolValue;
+
+    /* For explanations on how to distribute symbol values over the table :
+    *  http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+    /* symbol start positions */
+    {   U32 u;
+        cumul[0] = 0;
+        for (u=1; u<=maxSymbolValue+1; u++) {
+            if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
+                cumul[u] = cumul[u-1] + 1;
+                tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
+            } else {
+                cumul[u] = cumul[u-1] + normalizedCounter[u-1];
+        }   }
+        cumul[maxSymbolValue+1] = tableSize+1;
+    }
+
+    /* Spread symbols */
+    {   U32 position = 0;
+        U32 symbol;
+        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+            int nbOccurences;
+            for (nbOccurences=0; nbOccurences<normalizedCounter[symbol]; nbOccurences++) {
+                tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* Low proba area */
+        }   }
+
+        if (position!=0) return ERROR(GENERIC);   /* Must have gone through all positions */
+    }
+
+    /* Build table */
+    {   U32 u; for (u=0; u<tableSize; u++) {
+        FSE_FUNCTION_TYPE s = tableSymbol[u];   /* note : static analyzer may not understand tableSymbol is properly initialized */
+        tableU16[cumul[s]++] = (U16) (tableSize+u);   /* TableU16 : sorted by symbol order; gives next state value */
+    }   }
+
+    /* Build Symbol Transformation Table */
+    {   unsigned total = 0;
+        unsigned s;
+        for (s=0; s<=maxSymbolValue; s++) {
+            switch (normalizedCounter[s])
+            {
+            case  0: break;
+
+            case -1:
+            case  1:
+                symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
+                symbolTT[s].deltaFindState = total - 1;
+                total ++;
+                break;
+            default :
+                {
+                    U32 const maxBitsOut = tableLog - BIT_highbit32 (normalizedCounter[s]-1);
+                    U32 const minStatePlus = normalizedCounter[s] << maxBitsOut;
+                    symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
+                    symbolTT[s].deltaFindState = total - normalizedCounter[s];
+                    total +=  normalizedCounter[s];
+    }   }   }   }
+
+    return 0;
+}
+
+
+size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE];   /* memset() is not necessary, even if static analyzer complain about it */
+    return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
+}
+
+
+
+#ifndef FSE_COMMONDEFS_ONLY
+
+/*-**************************************************************
+*  FSE NCount encoding-decoding
+****************************************************************/
+size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
+{
+    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
+    return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
+}
+
+static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                                       const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                                       unsigned writeIsSafe)
+{
+    BYTE* const ostart = (BYTE*) header;
+    BYTE* out = ostart;
+    BYTE* const oend = ostart + headerBufferSize;
+    int nbBits;
+    const int tableSize = 1 << tableLog;
+    int remaining;
+    int threshold;
+    U32 bitStream;
+    int bitCount;
+    unsigned charnum = 0;
+    int previous0 = 0;
+
+    bitStream = 0;
+    bitCount  = 0;
+    /* Table Size */
+    bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
+    bitCount  += 4;
+
+    /* Init */
+    remaining = tableSize+1;   /* +1 for extra accuracy */
+    threshold = tableSize;
+    nbBits = tableLog+1;
+
+    while (remaining>1) {  /* stops at 1 */
+        if (previous0) {
+            unsigned start = charnum;
+            while (!normalizedCounter[charnum]) charnum++;
+            while (charnum >= start+24) {
+                start+=24;
+                bitStream += 0xFFFFU << bitCount;
+                if ((!writeIsSafe) && (out > oend-2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE) bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out+=2;
+                bitStream>>=16;
+            }
+            while (charnum >= start+3) {
+                start+=3;
+                bitStream += 3 << bitCount;
+                bitCount += 2;
+            }
+            bitStream += (charnum-start) << bitCount;
+            bitCount += 2;
+            if (bitCount>16) {
+                if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                out[0] = (BYTE)bitStream;
+                out[1] = (BYTE)(bitStream>>8);
+                out += 2;
+                bitStream >>= 16;
+                bitCount -= 16;
+        }   }
+        {   int count = normalizedCounter[charnum++];
+            int const max = (2*threshold-1)-remaining;
+            remaining -= count < 0 ? -count : count;
+            count++;   /* +1 for extra accuracy */
+            if (count>=threshold) count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            bitStream += count << bitCount;
+            bitCount  += nbBits;
+            bitCount  -= (count<max);
+            previous0  = (count==1);
+            if (remaining<1) return ERROR(GENERIC);
+            while (remaining<threshold) nbBits--, threshold>>=1;
+        }
+        if (bitCount>16) {
+            if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            out[0] = (BYTE)bitStream;
+            out[1] = (BYTE)(bitStream>>8);
+            out += 2;
+            bitStream >>= 16;
+            bitCount -= 16;
+    }   }
+
+    /* flush remaining bitStream */
+    if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    out[0] = (BYTE)bitStream;
+    out[1] = (BYTE)(bitStream>>8);
+    out+= (bitCount+7) /8;
+
+    if (charnum > maxSymbolValue + 1) return ERROR(GENERIC);
+
+    return (out-ostart);
+}
+
+
+size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+{
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
+
+    if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
+        return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
+
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1);
+}
+
+
+
+/*-**************************************************************
+*  Counting histogram
+****************************************************************/
+/*! FSE_count_simple
+    This function counts byte values within `src`, and store the histogram into table `count`.
+    It doesn't use any additional memory.
+    But this function is unsafe : it doesn't check that all values within `src` can fit into `count`.
+    For this reason, prefer using a table `count` with 256 elements.
+    @return : count of most numerous element
+*/
+size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                        const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned max=0;
+
+    memset(count, 0, (maxSymbolValue+1)*sizeof(*count));
+    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
+
+    while (ip<end) count[*ip++]++;
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+
+    { U32 s; for (s=0; s<=maxSymbolValue; s++) if (count[s] > max) max = count[s]; }
+
+    return (size_t)max;
+}
+
+
+/* FSE_count_parallel_wksp() :
+ * Same as FSE_count_parallel(), but using an externally provided scratch buffer.
+ * `workSpace` size must be a minimum of `1024 * sizeof(unsigned)`` */
+static size_t FSE_count_parallel_wksp(
+                                unsigned* count, unsigned* maxSymbolValuePtr,
+                                const void* source, size_t sourceSize,
+                                unsigned checkMax, unsigned* const workSpace)
+{
+    const BYTE* ip = (const BYTE*)source;
+    const BYTE* const iend = ip+sourceSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned max=0;
+    U32* const Counting1 = workSpace;
+    U32* const Counting2 = Counting1 + 256;
+    U32* const Counting3 = Counting2 + 256;
+    U32* const Counting4 = Counting3 + 256;
+
+    memset(Counting1, 0, 4*256*sizeof(unsigned));
+
+    /* safety checks */
+    if (!sourceSize) {
+        memset(count, 0, maxSymbolValue + 1);
+        *maxSymbolValuePtr = 0;
+        return 0;
+    }
+    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
+
+    /* by stripes of 16 bytes */
+    {   U32 cached = MEM_read32(ip); ip += 4;
+        while (ip < iend-15) {
+            U32 c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+        }
+        ip-=4;
+    }
+
+    /* finish last symbols */
+    while (ip<iend) Counting1[*ip++]++;
+
+    if (checkMax) {   /* verify stats will fit into destination table */
+        U32 s; for (s=255; s>maxSymbolValue; s--) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
+    }   }
+
+    {   U32 s; for (s=0; s<=maxSymbolValue; s++) {
+            count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
+            if (count[s] > max) max = count[s];
+    }   }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+    return (size_t)max;
+}
+
+/* FSE_countFast_wksp() :
+ * Same as FSE_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= `1024` unsigned */
+size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize, unsigned* workSpace)
+{
+    if (sourceSize < 1500) return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 0, workSpace);
+}
+
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[1024];
+    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters);
+}
+
+/* FSE_count_wksp() :
+ * Same as FSE_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= `1024` unsigned */
+size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* source, size_t sourceSize, unsigned* workSpace)
+{
+    if (*maxSymbolValuePtr < 255)
+        return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 1, workSpace);
+    *maxSymbolValuePtr = 255;
+    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace);
+}
+
+size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* src, size_t srcSize)
+{
+    unsigned tmpCounters[1024];
+    return FSE_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters);
+}
+
+
+
+/*-**************************************************************
+*  FSE Compression Code
+****************************************************************/
+/*! FSE_sizeof_CTable() :
+    FSE_CTable is a variable size structure which contains :
+    `U16 tableLog;`
+    `U16 maxSymbolValue;`
+    `U16 nextStateNumber[1 << tableLog];`                         // This size is variable
+    `FSE_symbolCompressionTransform symbolTT[maxSymbolValue+1];`  // This size is variable
+Allocation is manual (C standard does not support variable-size structures).
+*/
+size_t FSE_sizeof_CTable (unsigned maxSymbolValue, unsigned tableLog)
+{
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    return FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
+}
+
+FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
+{
+    size_t size;
+    if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
+    size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
+    return (FSE_CTable*)malloc(size);
+}
+
+void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
+
+/* provides the minimum logSize to safely represent a distribution */
+static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
+{
+    U32 minBitsSrc = BIT_highbit32((U32)(srcSize - 1)) + 1;
+    U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
+    U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
+    return minBits;
+}
+
+unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
+{
+    U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
+    U32 tableLog = maxTableLog;
+    U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
+    if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
+    if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
+    if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
+    return tableLog;
+}
+
+unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+{
+    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
+}
+
+
+/* Secondary normalization method.
+   To be used when primary method fails. */
+
+static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue)
+{
+    short const NOT_YET_ASSIGNED = -2;
+    U32 s;
+    U32 distributed = 0;
+    U32 ToDistribute;
+
+    /* Init */
+    U32 const lowThreshold = (U32)(total >> tableLog);
+    U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
+
+    for (s=0; s<=maxSymbolValue; s++) {
+        if (count[s] == 0) {
+            norm[s]=0;
+            continue;
+        }
+        if (count[s] <= lowThreshold) {
+            norm[s] = -1;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+        if (count[s] <= lowOne) {
+            norm[s] = 1;
+            distributed++;
+            total -= count[s];
+            continue;
+        }
+
+        norm[s]=NOT_YET_ASSIGNED;
+    }
+    ToDistribute = (1 << tableLog) - distributed;
+
+    if ((total / ToDistribute) > lowOne) {
+        /* risk of rounding to zero */
+        lowOne = (U32)((total * 3) / (ToDistribute * 2));
+        for (s=0; s<=maxSymbolValue; s++) {
+            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
+                norm[s] = 1;
+                distributed++;
+                total -= count[s];
+                continue;
+        }   }
+        ToDistribute = (1 << tableLog) - distributed;
+    }
+
+    if (distributed == maxSymbolValue+1) {
+        /* all values are pretty poor;
+           probably incompressible data (should have already been detected);
+           find max, then give all remaining points to max */
+        U32 maxV = 0, maxC = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > maxC) maxV=s, maxC=count[s];
+        norm[maxV] += (short)ToDistribute;
+        return 0;
+    }
+
+    if (total == 0) {
+        /* all of the symbols were low enough for the lowOne or lowThreshold */
+        for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
+            if (norm[s] > 0) ToDistribute--, norm[s]++;
+        return 0;
+    }
+
+    {   U64 const vStepLog = 62 - tableLog;
+        U64 const mid = (1ULL << (vStepLog-1)) - 1;
+        U64 const rStep = ((((U64)1<<vStepLog) * ToDistribute) + mid) / total;   /* scale on remaining */
+        U64 tmpTotal = mid;
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (norm[s]==NOT_YET_ASSIGNED) {
+                U64 const end = tmpTotal + (count[s] * rStep);
+                U32 const sStart = (U32)(tmpTotal >> vStepLog);
+                U32 const sEnd = (U32)(end >> vStepLog);
+                U32 const weight = sEnd - sStart;
+                if (weight < 1)
+                    return ERROR(GENERIC);
+                norm[s] = (short)weight;
+                tmpTotal = end;
+    }   }   }
+
+    return 0;
+}
+
+
+size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
+                           const unsigned* count, size_t total,
+                           unsigned maxSymbolValue)
+{
+    /* Sanity checks */
+    if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
+    if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported size */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
+    if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
+
+    {   U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+        U64 const scale = 62 - tableLog;
+        U64 const step = ((U64)1<<62) / total;   /* <== here, one division ! */
+        U64 const vStep = 1ULL<<(scale-20);
+        int stillToDistribute = 1<<tableLog;
+        unsigned s;
+        unsigned largest=0;
+        short largestP=0;
+        U32 lowThreshold = (U32)(total >> tableLog);
+
+        for (s=0; s<=maxSymbolValue; s++) {
+            if (count[s] == total) return 0;   /* rle special case */
+            if (count[s] == 0) { normalizedCounter[s]=0; continue; }
+            if (count[s] <= lowThreshold) {
+                normalizedCounter[s] = -1;
+                stillToDistribute--;
+            } else {
+                short proba = (short)((count[s]*step) >> scale);
+                if (proba<8) {
+                    U64 restToBeat = vStep * rtbTable[proba];
+                    proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
+                }
+                if (proba > largestP) largestP=proba, largest=s;
+                normalizedCounter[s] = proba;
+                stillToDistribute -= proba;
+        }   }
+        if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
+            /* corner case, need another normalization method */
+            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
+            if (FSE_isError(errorCode)) return errorCode;
+        }
+        else normalizedCounter[largest] += (short)stillToDistribute;
+    }
+
+#if 0
+    {   /* Print Table (debug) */
+        U32 s;
+        U32 nTotal = 0;
+        for (s=0; s<=maxSymbolValue; s++)
+            printf("%3i: %4i \n", s, normalizedCounter[s]);
+        for (s=0; s<=maxSymbolValue; s++)
+            nTotal += abs(normalizedCounter[s]);
+        if (nTotal != (1U<<tableLog))
+            printf("Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
+        getchar();
+    }
+#endif
+
+    return tableLog;
+}
+
+
+/* fake FSE_CTable, for raw (uncompressed) input */
+size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
+{
+    const unsigned tableSize = 1 << nbBits;
+    const unsigned tableMask = tableSize - 1;
+    const unsigned maxSymbolValue = tableMask;
+    void* const ptr = ct;
+    U16* const tableU16 = ( (U16*) ptr) + 2;
+    void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableSize>>1);   /* assumption : tableLog >= 1 */
+    FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
+    unsigned s;
+
+    /* Sanity checks */
+    if (nbBits < 1) return ERROR(GENERIC);             /* min size */
+
+    /* header */
+    tableU16[-2] = (U16) nbBits;
+    tableU16[-1] = (U16) maxSymbolValue;
+
+    /* Build table */
+    for (s=0; s<tableSize; s++)
+        tableU16[s] = (U16)(tableSize + s);
+
+    /* Build Symbol Transformation Table */
+    {   const U32 deltaNbBits = (nbBits << 16) - (1 << nbBits);
+        for (s=0; s<=maxSymbolValue; s++) {
+            symbolTT[s].deltaNbBits = deltaNbBits;
+            symbolTT[s].deltaFindState = s-1;
+    }   }
+
+    return 0;
+}
+
+/* fake FSE_CTable, for rle input (always same symbol) */
+size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
+{
+    void* ptr = ct;
+    U16* tableU16 = ( (U16*) ptr) + 2;
+    void* FSCTptr = (U32*)ptr + 2;
+    FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
+
+    /* header */
+    tableU16[-2] = (U16) 0;
+    tableU16[-1] = (U16) symbolValue;
+
+    /* Build table */
+    tableU16[0] = 0;
+    tableU16[1] = 0;   /* just in case */
+
+    /* Build Symbol Transformation Table */
+    symbolTT[symbolValue].deltaNbBits = 0;
+    symbolTT[symbolValue].deltaFindState = 0;
+
+    return 0;
+}
+
+
+static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct, const unsigned fast)
+{
+    const BYTE* const istart = (const BYTE*) src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip=iend;
+
+    BIT_CStream_t bitC;
+    FSE_CState_t CState1, CState2;
+
+    /* init */
+    if (srcSize <= 2) return 0;
+    { size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
+      if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
+
+#define FSE_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+
+    if (srcSize & 1) {
+        FSE_initCState2(&CState1, ct, *--ip);
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    } else {
+        FSE_initCState2(&CState2, ct, *--ip);
+        FSE_initCState2(&CState1, ct, *--ip);
+    }
+
+    /* join to mod 4 */
+    srcSize -= 2;
+    if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) {  /* test bit 2 */
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    /* 2 or 4 encoding per loop */
+    while ( ip>istart ) {
+
+        FSE_encodeSymbol(&bitC, &CState2, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 )   /* this test must be static */
+            FSE_FLUSHBITS(&bitC);
+
+        FSE_encodeSymbol(&bitC, &CState1, *--ip);
+
+        if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) {  /* this test must be static */
+            FSE_encodeSymbol(&bitC, &CState2, *--ip);
+            FSE_encodeSymbol(&bitC, &CState1, *--ip);
+        }
+
+        FSE_FLUSHBITS(&bitC);
+    }
+
+    FSE_flushCState(&bitC, &CState2);
+    FSE_flushCState(&bitC, &CState1);
+    return BIT_closeCStream(&bitC);
+}
+
+size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
+                           const void* src, size_t srcSize,
+                           const FSE_CTable* ct)
+{
+    unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
+
+    if (fast)
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
+    else
+        return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
+}
+
+
+size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
+
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return f
+#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+
+/* FSE_compress_wksp() :
+ * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
+ * `wkspSize` size must be `(1<<tableLog)`.
+ */
+size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstSize;
+
+    U32   count[FSE_MAX_SYMBOL_VALUE+1];
+    S16   norm[FSE_MAX_SYMBOL_VALUE+1];
+    FSE_CTable* CTable = (FSE_CTable*)workSpace;
+    size_t const CTableSize = FSE_CTABLE_SIZE_U32(tableLog, maxSymbolValue);
+    void* scratchBuffer = (void*)(CTable + CTableSize);
+    size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
+
+    /* init conditions */
+    if (wkspSize < FSE_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
+    if (srcSize <= 1) return 0;  /* Not compressible */
+    if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
+    if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(maxCount, FSE_count_wksp(count, &maxSymbolValue, src, srcSize, (unsigned*)scratchBuffer) );
+        if (maxCount == srcSize) return 1;   /* only a single symbol in src : rle */
+        if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
+        if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
+    }
+
+    tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue) );
+
+    /* Write table description header */
+    {   CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
+        op += nc_err;
+    }
+
+    /* Compress */
+    CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, scratchBufferSize) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, src, srcSize, CTable) );
+        if (cSize == 0) return 0;   /* not enough space for compressed data */
+        op += cSize;
+    }
+
+    /* check compressibility */
+    if ( (size_t)(op-ostart) >= srcSize-1 ) return 0;
+
+    return op-ostart;
+}
+
+typedef struct {
+    FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
+    BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
+} fseWkspMax_t;
+
+size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
+{
+    fseWkspMax_t scratchBuffer;
+    FSE_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
+    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
+    return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
+}
+
+size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
+}
+
+
+#endif   /* FSE_COMMONDEFS_ONLY */
diff --git a/thirdparty/zstd/compress/huf_compress.c b/thirdparty/zstd/compress/huf_compress.c
new file mode 100644
index 0000000000..fe11aafb8f
--- /dev/null
+++ b/thirdparty/zstd/compress/huf_compress.c
@@ -0,0 +1,684 @@
+/* ******************************************************************
+   Huffman encoder, part of New Generation Entropy library
+   Copyright (C) 2013-2016, Yann Collet.
+
+   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are
+   met:
+
+       * Redistributions of source code must retain the above copyright
+   notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above
+   copyright notice, this list of conditions and the following disclaimer
+   in the documentation and/or other materials provided with the
+   distribution.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+    You can contact the author at :
+    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+****************************************************************** */
+
+/* **************************************************************
+*  Compiler specifics
+****************************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#endif
+
+
+/* **************************************************************
+*  Includes
+****************************************************************/
+#include <string.h>     /* memcpy, memset */
+#include <stdio.h>      /* printf (debug) */
+#include "bitstream.h"
+#define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
+#include "fse.h"        /* header compression */
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+
+
+/* **************************************************************
+*  Error Management
+****************************************************************/
+#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return f
+#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+
+
+/* **************************************************************
+*  Utils
+****************************************************************/
+unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
+{
+    return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+}
+
+
+/* *******************************************************
+*  HUF : Huffman block compression
+*********************************************************/
+/* HUF_compressWeights() :
+ * Same as FSE_compress(), but dedicated to huff0's weights compression.
+ * The use case needs much less stack memory.
+ * Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
+ */
+#define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
+size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
+{
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* op = ostart;
+    BYTE* const oend = ostart + dstSize;
+
+    U32 maxSymbolValue = HUF_TABLELOG_MAX;
+    U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
+
+    FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
+    BYTE scratchBuffer[1<<MAX_FSE_TABLELOG_FOR_HUFF_HEADER];
+
+    U32 count[HUF_TABLELOG_MAX+1];
+    S16 norm[HUF_TABLELOG_MAX+1];
+
+    /* init conditions */
+    if (wtSize <= 1) return 0;  /* Not compressible */
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(maxCount, FSE_count_simple(count, &maxSymbolValue, weightTable, wtSize) );
+        if (maxCount == wtSize) return 1;   /* only a single symbol in src : rle */
+        if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
+    }
+
+    tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, wtSize, maxSymbolValue) );
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
+        op += hSize;
+    }
+
+    /* Compress */
+    CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, sizeof(scratchBuffer)) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, weightTable, wtSize, CTable) );
+        if (cSize == 0) return 0;   /* not enough space for compressed data */
+        op += cSize;
+    }
+
+    return op-ostart;
+}
+
+
+struct HUF_CElt_s {
+  U16  val;
+  BYTE nbBits;
+};   /* typedef'd to HUF_CElt within "huf.h" */
+
+/*! HUF_writeCTable() :
+    `CTable` : Huffman tree to save, using huf representation.
+    @return : size of saved CTable */
+size_t HUF_writeCTable (void* dst, size_t maxDstSize,
+                        const HUF_CElt* CTable, U32 maxSymbolValue, U32 huffLog)
+{
+    BYTE bitsToWeight[HUF_TABLELOG_MAX + 1];   /* precomputed conversion table */
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
+    BYTE* op = (BYTE*)dst;
+    U32 n;
+
+     /* check conditions */
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
+
+    /* convert to weight */
+    bitsToWeight[0] = 0;
+    for (n=1; n<huffLog+1; n++)
+        bitsToWeight[n] = (BYTE)(huffLog + 1 - n);
+    for (n=0; n<maxSymbolValue; n++)
+        huffWeight[n] = bitsToWeight[CTable[n].nbBits];
+
+    /* attempt weights compression by FSE */
+    {   CHECK_V_F(hSize, HUF_compressWeights(op+1, maxDstSize-1, huffWeight, maxSymbolValue) );
+        if ((hSize>1) & (hSize < maxSymbolValue/2)) {   /* FSE compressed */
+            op[0] = (BYTE)hSize;
+            return hSize+1;
+    }   }
+
+    /* write raw values as 4-bits (max : 15) */
+    if (maxSymbolValue > (256-128)) return ERROR(GENERIC);   /* should not happen : likely means source cannot be compressed */
+    if (((maxSymbolValue+1)/2) + 1 > maxDstSize) return ERROR(dstSize_tooSmall);   /* not enough space within dst buffer */
+    op[0] = (BYTE)(128 /*special case*/ + (maxSymbolValue-1));
+    huffWeight[maxSymbolValue] = 0;   /* to be sure it doesn't cause msan issue in final combination */
+    for (n=0; n<maxSymbolValue; n+=2)
+        op[(n/2)+1] = (BYTE)((huffWeight[n] << 4) + huffWeight[n+1]);
+    return ((maxSymbolValue+1)/2) + 1;
+}
+
+
+size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, size_t srcSize)
+{
+    BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];   /* init not required, even though some static analyzer may complain */
+    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
+    U32 tableLog = 0;
+    U32 nbSymbols = 0;
+
+    /* get symbol weights */
+    CHECK_V_F(readSize, HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize));
+
+    /* check result */
+    if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (nbSymbols > maxSymbolValue+1) return ERROR(maxSymbolValue_tooSmall);
+
+    /* Prepare base value per rank */
+    {   U32 n, nextRankStart = 0;
+        for (n=1; n<=tableLog; n++) {
+            U32 current = nextRankStart;
+            nextRankStart += (rankVal[n] << (n-1));
+            rankVal[n] = current;
+    }   }
+
+    /* fill nbBits */
+    {   U32 n; for (n=0; n<nbSymbols; n++) {
+            const U32 w = huffWeight[n];
+            CTable[n].nbBits = (BYTE)(tableLog + 1 - w);
+    }   }
+
+    /* fill val */
+    {   U16 nbPerRank[HUF_TABLELOG_MAX+2]  = {0};  /* support w=0=>n=tableLog+1 */
+        U16 valPerRank[HUF_TABLELOG_MAX+2] = {0};
+        { U32 n; for (n=0; n<nbSymbols; n++) nbPerRank[CTable[n].nbBits]++; }
+        /* determine stating value per rank */
+        valPerRank[tableLog+1] = 0;   /* for w==0 */
+        {   U16 min = 0;
+            U32 n; for (n=tableLog; n>0; n--) {  /* start at n=tablelog <-> w=1 */
+                valPerRank[n] = min;     /* get starting value within each rank */
+                min += nbPerRank[n];
+                min >>= 1;
+        }   }
+        /* assign value within rank, symbol order */
+        { U32 n; for (n=0; n<=maxSymbolValue; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
+    }
+
+    return readSize;
+}
+
+
+typedef struct nodeElt_s {
+    U32 count;
+    U16 parent;
+    BYTE byte;
+    BYTE nbBits;
+} nodeElt;
+
+static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
+{
+    const U32 largestBits = huffNode[lastNonNull].nbBits;
+    if (largestBits <= maxNbBits) return largestBits;   /* early exit : no elt > maxNbBits */
+
+    /* there are several too large elements (at least >= 2) */
+    {   int totalCost = 0;
+        const U32 baseCost = 1 << (largestBits - maxNbBits);
+        U32 n = lastNonNull;
+
+        while (huffNode[n].nbBits > maxNbBits) {
+            totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
+            huffNode[n].nbBits = (BYTE)maxNbBits;
+            n --;
+        }  /* n stops at huffNode[n].nbBits <= maxNbBits */
+        while (huffNode[n].nbBits == maxNbBits) n--;   /* n end at index of smallest symbol using < maxNbBits */
+
+        /* renorm totalCost */
+        totalCost >>= (largestBits - maxNbBits);  /* note : totalCost is necessarily a multiple of baseCost */
+
+        /* repay normalized cost */
+        {   U32 const noSymbol = 0xF0F0F0F0;
+            U32 rankLast[HUF_TABLELOG_MAX+2];
+            int pos;
+
+            /* Get pos of last (smallest) symbol per rank */
+            memset(rankLast, 0xF0, sizeof(rankLast));
+            {   U32 currentNbBits = maxNbBits;
+                for (pos=n ; pos >= 0; pos--) {
+                    if (huffNode[pos].nbBits >= currentNbBits) continue;
+                    currentNbBits = huffNode[pos].nbBits;   /* < maxNbBits */
+                    rankLast[maxNbBits-currentNbBits] = pos;
+            }   }
+
+            while (totalCost > 0) {
+                U32 nBitsToDecrease = BIT_highbit32(totalCost) + 1;
+                for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
+                    U32 highPos = rankLast[nBitsToDecrease];
+                    U32 lowPos = rankLast[nBitsToDecrease-1];
+                    if (highPos == noSymbol) continue;
+                    if (lowPos == noSymbol) break;
+                    {   U32 const highTotal = huffNode[highPos].count;
+                        U32 const lowTotal = 2 * huffNode[lowPos].count;
+                        if (highTotal <= lowTotal) break;
+                }   }
+                /* only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !) */
+                while ((nBitsToDecrease<=HUF_TABLELOG_MAX) && (rankLast[nBitsToDecrease] == noSymbol))  /* HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary */
+                    nBitsToDecrease ++;
+                totalCost -= 1 << (nBitsToDecrease-1);
+                if (rankLast[nBitsToDecrease-1] == noSymbol)
+                    rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease];   /* this rank is no longer empty */
+                huffNode[rankLast[nBitsToDecrease]].nbBits ++;
+                if (rankLast[nBitsToDecrease] == 0)    /* special case, reached largest symbol */
+                    rankLast[nBitsToDecrease] = noSymbol;
+                else {
+                    rankLast[nBitsToDecrease]--;
+                    if (huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease)
+                        rankLast[nBitsToDecrease] = noSymbol;   /* this rank is now empty */
+            }   }   /* while (totalCost > 0) */
+
+            while (totalCost < 0) {  /* Sometimes, cost correction overshoot */
+                if (rankLast[1] == noSymbol) {  /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
+                    while (huffNode[n].nbBits == maxNbBits) n--;
+                    huffNode[n+1].nbBits--;
+                    rankLast[1] = n+1;
+                    totalCost++;
+                    continue;
+                }
+                huffNode[ rankLast[1] + 1 ].nbBits--;
+                rankLast[1]++;
+                totalCost ++;
+    }   }   }   /* there are several too large elements (at least >= 2) */
+
+    return maxNbBits;
+}
+
+
+typedef struct {
+    U32 base;
+    U32 current;
+} rankPos;
+
+static void HUF_sort(nodeElt* huffNode, const U32* count, U32 maxSymbolValue)
+{
+    rankPos rank[32];
+    U32 n;
+
+    memset(rank, 0, sizeof(rank));
+    for (n=0; n<=maxSymbolValue; n++) {
+        U32 r = BIT_highbit32(count[n] + 1);
+        rank[r].base ++;
+    }
+    for (n=30; n>0; n--) rank[n-1].base += rank[n].base;
+    for (n=0; n<32; n++) rank[n].current = rank[n].base;
+    for (n=0; n<=maxSymbolValue; n++) {
+        U32 const c = count[n];
+        U32 const r = BIT_highbit32(c+1) + 1;
+        U32 pos = rank[r].current++;
+        while ((pos > rank[r].base) && (c > huffNode[pos-1].count)) huffNode[pos]=huffNode[pos-1], pos--;
+        huffNode[pos].count = c;
+        huffNode[pos].byte  = (BYTE)n;
+    }
+}
+
+
+/** HUF_buildCTable_wksp() :
+ *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
+ *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of 1024 unsigned.
+ */
+#define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
+typedef nodeElt huffNodeTable[2*HUF_SYMBOLVALUE_MAX+1 +1];
+size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
+{
+    nodeElt* const huffNode0 = (nodeElt*)workSpace;
+    nodeElt* const huffNode = huffNode0+1;
+    U32 n, nonNullRank;
+    int lowS, lowN;
+    U16 nodeNb = STARTNODE;
+    U32 nodeRoot;
+
+    /* safety checks */
+    if (wkspSize < sizeof(huffNodeTable)) return ERROR(GENERIC);   /* workSpace is not large enough */
+    if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(GENERIC);
+    memset(huffNode0, 0, sizeof(huffNodeTable));
+
+    /* sort, decreasing order */
+    HUF_sort(huffNode, count, maxSymbolValue);
+
+    /* init for parents */
+    nonNullRank = maxSymbolValue;
+    while(huffNode[nonNullRank].count == 0) nonNullRank--;
+    lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
+    huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
+    huffNode[lowS].parent = huffNode[lowS-1].parent = nodeNb;
+    nodeNb++; lowS-=2;
+    for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
+    huffNode0[0].count = (U32)(1U<<31);  /* fake entry, strong barrier */
+
+    /* create parents */
+    while (nodeNb <= nodeRoot) {
+        U32 n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        U32 n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
+        huffNode[n1].parent = huffNode[n2].parent = nodeNb;
+        nodeNb++;
+    }
+
+    /* distribute weights (unlimited tree height) */
+    huffNode[nodeRoot].nbBits = 0;
+    for (n=nodeRoot-1; n>=STARTNODE; n--)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+    for (n=0; n<=nonNullRank; n++)
+        huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
+
+    /* enforce maxTableLog */
+    maxNbBits = HUF_setMaxHeight(huffNode, nonNullRank, maxNbBits);
+
+    /* fill result into tree (val, nbBits) */
+    {   U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
+        U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+        if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
+        for (n=0; n<=nonNullRank; n++)
+            nbPerRank[huffNode[n].nbBits]++;
+        /* determine stating value per rank */
+        {   U16 min = 0;
+            for (n=maxNbBits; n>0; n--) {
+                valPerRank[n] = min;      /* get starting value within each rank */
+                min += nbPerRank[n];
+                min >>= 1;
+        }   }
+        for (n=0; n<=maxSymbolValue; n++)
+            tree[huffNode[n].byte].nbBits = huffNode[n].nbBits;   /* push nbBits per symbol, symbol order */
+        for (n=0; n<=maxSymbolValue; n++)
+            tree[n].val = valPerRank[tree[n].nbBits]++;   /* assign value within rank, symbol order */
+    }
+
+    return maxNbBits;
+}
+
+/** HUF_buildCTable() :
+ *  Note : count is used before tree is written, so they can safely overlap
+ */
+size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits)
+{
+    huffNodeTable nodeTable;
+    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, nodeTable, sizeof(nodeTable));
+}
+
+static size_t HUF_estimateCompressedSize(HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
+{
+    size_t nbBits = 0;
+    int s;
+    for (s = 0; s <= (int)maxSymbolValue; ++s) {
+        nbBits += CTable[s].nbBits * count[s];
+    }
+    return nbBits >> 3;
+}
+
+static int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+  int bad = 0;
+  int s;
+  for (s = 0; s <= (int)maxSymbolValue; ++s) {
+    bad |= (count[s] != 0) & (CTable[s].nbBits == 0);
+  }
+  return !bad;
+}
+
+static void HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
+{
+    BIT_addBitsFast(bitCPtr, CTable[symbol].val, CTable[symbol].nbBits);
+}
+
+size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
+
+#define HUF_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
+
+#define HUF_FLUSHBITS_1(stream) \
+    if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*2+7) HUF_FLUSHBITS(stream)
+
+#define HUF_FLUSHBITS_2(stream) \
+    if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*4+7) HUF_FLUSHBITS(stream)
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    const BYTE* ip = (const BYTE*) src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+    size_t n;
+    const unsigned fast = (dstSize >= HUF_BLOCKBOUND(srcSize));
+    BIT_CStream_t bitC;
+
+    /* init */
+    if (dstSize < 8) return 0;   /* not enough space to compress */
+    { size_t const initErr = BIT_initCStream(&bitC, op, oend-op);
+      if (HUF_isError(initErr)) return 0; }
+
+    n = srcSize & ~3;  /* join to mod 4 */
+    switch (srcSize & 3)
+    {
+        case 3 : HUF_encodeSymbol(&bitC, ip[n+ 2], CTable);
+                 HUF_FLUSHBITS_2(&bitC);
+        case 2 : HUF_encodeSymbol(&bitC, ip[n+ 1], CTable);
+                 HUF_FLUSHBITS_1(&bitC);
+        case 1 : HUF_encodeSymbol(&bitC, ip[n+ 0], CTable);
+                 HUF_FLUSHBITS(&bitC);
+        case 0 :
+        default: ;
+    }
+
+    for (; n>0; n-=4) {  /* note : n&3==0 at this stage */
+        HUF_encodeSymbol(&bitC, ip[n- 1], CTable);
+        HUF_FLUSHBITS_1(&bitC);
+        HUF_encodeSymbol(&bitC, ip[n- 2], CTable);
+        HUF_FLUSHBITS_2(&bitC);
+        HUF_encodeSymbol(&bitC, ip[n- 3], CTable);
+        HUF_FLUSHBITS_1(&bitC);
+        HUF_encodeSymbol(&bitC, ip[n- 4], CTable);
+        HUF_FLUSHBITS(&bitC);
+    }
+
+    return BIT_closeCStream(&bitC);
+}
+
+
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
+    const BYTE* ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*) dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    if (dstSize < 6 + 1 + 1 + 1 + 8) return 0;   /* minimum space to compress successfully */
+    if (srcSize < 12) return 0;   /* no saving possible : too small input */
+    op += 6;   /* jumpTable */
+
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+        if (cSize==0) return 0;
+        MEM_writeLE16(ostart, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+        if (cSize==0) return 0;
+        MEM_writeLE16(ostart+2, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+        if (cSize==0) return 0;
+        MEM_writeLE16(ostart+4, (U16)cSize);
+        op += cSize;
+    }
+
+    ip += segmentSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, iend-ip, CTable) );
+        if (cSize==0) return 0;
+        op += cSize;
+    }
+
+    return op-ostart;
+}
+
+
+static size_t HUF_compressCTable_internal(
+                BYTE* const ostart, BYTE* op, BYTE* const oend,
+                const void* src, size_t srcSize,
+                unsigned singleStream, const HUF_CElt* CTable)
+{
+    size_t const cSize = singleStream ?
+                         HUF_compress1X_usingCTable(op, oend - op, src, srcSize, CTable) :
+                         HUF_compress4X_usingCTable(op, oend - op, src, srcSize, CTable);
+    if (HUF_isError(cSize)) { return cSize; }
+    if (cSize==0) { return 0; }   /* uncompressible */
+    op += cSize;
+    /* check compressibility */
+    if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
+    return op-ostart;
+}
+
+
+/* `workSpace` must a table of at least 1024 unsigned */
+static size_t HUF_compress_internal (
+                void* dst, size_t dstSize,
+                const void* src, size_t srcSize,
+                unsigned maxSymbolValue, unsigned huffLog,
+                unsigned singleStream,
+                void* workSpace, size_t wkspSize,
+                HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart;
+
+    U32* count;
+    size_t const countSize = sizeof(U32) * (HUF_SYMBOLVALUE_MAX + 1);
+    HUF_CElt* CTable;
+    size_t const CTableSize = sizeof(HUF_CElt) * (HUF_SYMBOLVALUE_MAX + 1);
+
+    /* checks & inits */
+    if (wkspSize < sizeof(huffNodeTable) + countSize + CTableSize) return ERROR(GENERIC);
+    if (!srcSize) return 0;  /* Uncompressed (note : 1 means rle, so first byte must be correct) */
+    if (!dstSize) return 0;  /* cannot fit within dst budget */
+    if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
+    if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
+    if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
+
+    count = (U32*)workSpace;
+    workSpace = (BYTE*)workSpace + countSize;
+    wkspSize -= countSize;
+    CTable = (HUF_CElt*)workSpace;
+    workSpace = (BYTE*)workSpace + CTableSize;
+    wkspSize -= CTableSize;
+
+    /* Heuristic : If we don't need to check the validity of the old table use the old table for small inputs */
+    if (preferRepeat && repeat && *repeat == HUF_repeat_valid) {
+        return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, oldHufTable);
+    }
+
+    /* Scan input and build symbol stats */
+    {   CHECK_V_F(largest, FSE_count_wksp (count, &maxSymbolValue, (const BYTE*)src, srcSize, (U32*)workSpace) );
+        if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
+        if (largest <= (srcSize >> 7)+1) return 0;   /* Fast heuristic : not compressible enough */
+    }
+
+    /* Check validity of previous table */
+    if (repeat && *repeat == HUF_repeat_check && !HUF_validateCTable(oldHufTable, count, maxSymbolValue)) {
+        *repeat = HUF_repeat_none;
+    }
+    /* Heuristic : use existing table for small inputs */
+    if (preferRepeat && repeat && *repeat != HUF_repeat_none) {
+        return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, oldHufTable);
+    }
+
+    /* Build Huffman Tree */
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    {   CHECK_V_F(maxBits, HUF_buildCTable_wksp (CTable, count, maxSymbolValue, huffLog, workSpace, wkspSize) );
+        huffLog = (U32)maxBits;
+        /* Zero the unused symbols so we can check it for validity */
+        memset(CTable + maxSymbolValue + 1, 0, CTableSize - (maxSymbolValue + 1) * sizeof(HUF_CElt));
+    }
+
+    /* Write table description header */
+    {   CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, CTable, maxSymbolValue, huffLog) );
+        /* Check if using the previous table will be beneficial */
+        if (repeat && *repeat != HUF_repeat_none) {
+            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, count, maxSymbolValue);
+            size_t const newSize = HUF_estimateCompressedSize(CTable, count, maxSymbolValue);
+            if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
+                return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, oldHufTable);
+            }
+        }
+        /* Use the new table */
+        if (hSize + 12ul >= srcSize) { return 0; }
+        op += hSize;
+        if (repeat) { *repeat = HUF_repeat_none; }
+        if (oldHufTable) { memcpy(oldHufTable, CTable, CTableSize); } /* Save the new table */
+    }
+    return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, CTable);
+}
+
+
+size_t HUF_compress1X_wksp (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 1 /* single stream */, workSpace, wkspSize, NULL, NULL, 0);
+}
+
+size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 1 /* single stream */, workSpace, wkspSize, hufTable, repeat, preferRepeat);
+}
+
+size_t HUF_compress1X (void* dst, size_t dstSize,
+                 const void* src, size_t srcSize,
+                 unsigned maxSymbolValue, unsigned huffLog)
+{
+    unsigned workSpace[1024];
+    return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
+}
+
+size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 0 /* 4 streams */, workSpace, wkspSize, NULL, NULL, 0);
+}
+
+size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
+                      const void* src, size_t srcSize,
+                      unsigned maxSymbolValue, unsigned huffLog,
+                      void* workSpace, size_t wkspSize,
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat)
+{
+    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 0 /* 4 streams */, workSpace, wkspSize, hufTable, repeat, preferRepeat);
+}
+
+size_t HUF_compress2 (void* dst, size_t dstSize,
+                const void* src, size_t srcSize,
+                unsigned maxSymbolValue, unsigned huffLog)
+{
+    unsigned workSpace[1024];
+    return HUF_compress4X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
+}
+
+size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSize)
+{
+    return HUF_compress2(dst, maxDstSize, src, (U32)srcSize, 255, HUF_TABLELOG_DEFAULT);
+}
diff --git a/thirdparty/zstd/compress/zstd_compress.c b/thirdparty/zstd/compress/zstd_compress.c
new file mode 100644
index 0000000000..c08b315dab
--- /dev/null
+++ b/thirdparty/zstd/compress/zstd_compress.c
@@ -0,0 +1,3598 @@
+/**
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree. An additional grant
+ * of patent rights can be found in the PATENTS file in the same directory.
+ */
+
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <string.h>         /* memset */
+#include "mem.h"
+#define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "zstd_internal.h"  /* includes zstd.h */
+
+
+/*-*************************************
+*  Debug
+***************************************/
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
+#  include <assert.h>
+#else
+#  define assert(condition) ((void)0)
+#endif
+
+#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
+
+#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
+#  include <stdio.h>
+   static unsigned g_debugLevel = ZSTD_DEBUG;
+#  define DEBUGLOG(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __FILE__ ": "); fprintf(stderr, __VA_ARGS__); fprintf(stderr, " \n"); }
+#else
+#  define DEBUGLOG(l, ...)      {}    /* disabled */
+#endif
+
+
+/*-*************************************
+*  Constants
+***************************************/
+static const U32 g_searchStrength = 8;   /* control skip over incompressible data */
+#define HASH_READ_SIZE 8
+typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
+
+/* entropy tables always have same size */
+static size_t const hufCTable_size = HUF_CTABLE_SIZE(255);
+static size_t const litlengthCTable_size = FSE_CTABLE_SIZE(LLFSELog, MaxLL);
+static size_t const offcodeCTable_size = FSE_CTABLE_SIZE(OffFSELog, MaxOff);
+static size_t const matchlengthCTable_size = FSE_CTABLE_SIZE(MLFSELog, MaxML);
+static size_t const entropyScratchSpace_size = HUF_WORKSPACE_SIZE;
+
+
+/*-*************************************
+*  Helper functions
+***************************************/
+size_t ZSTD_compressBound(size_t srcSize) {
+    size_t const lowLimit = 256 KB;
+    size_t const margin = (srcSize < lowLimit) ? (lowLimit-srcSize) >> 12 : 0;  /* from 64 to 0 */
+    return srcSize + (srcSize >> 8) + margin;
+}
+
+
+/*-*************************************
+*  Sequence storage
+***************************************/
+static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
+{
+    ssPtr->lit = ssPtr->litStart;
+    ssPtr->sequences = ssPtr->sequencesStart;
+    ssPtr->longLengthID = 0;
+}
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+struct ZSTD_CCtx_s {
+    const BYTE* nextSrc;    /* next block here to continue on current prefix */
+    const BYTE* base;       /* All regular indexes relative to this position */
+    const BYTE* dictBase;   /* extDict indexes relative to this position */
+    U32   dictLimit;        /* below that point, need extDict */
+    U32   lowLimit;         /* below that point, no more data */
+    U32   nextToUpdate;     /* index from which to continue dictionary update */
+    U32   nextToUpdate3;    /* index from which to continue dictionary update */
+    U32   hashLog3;         /* dispatch table : larger == faster, more memory */
+    U32   loadedDictEnd;    /* index of end of dictionary */
+    U32   forceWindow;      /* force back-references to respect limit of 1<<wLog, even for dictionary */
+    U32   forceRawDict;     /* Force loading dictionary in "content-only" mode (no header analysis) */
+    ZSTD_compressionStage_e stage;
+    U32   rep[ZSTD_REP_NUM];
+    U32   repToConfirm[ZSTD_REP_NUM];
+    U32   dictID;
+    ZSTD_parameters params;
+    void* workSpace;
+    size_t workSpaceSize;
+    size_t blockSize;
+    U64 frameContentSize;
+    U64 consumedSrcSize;
+    XXH64_state_t xxhState;
+    ZSTD_customMem customMem;
+
+    seqStore_t seqStore;    /* sequences storage ptrs */
+    U32* hashTable;
+    U32* hashTable3;
+    U32* chainTable;
+    HUF_repeat hufCTable_repeatMode;
+    HUF_CElt* hufCTable;
+    U32 fseCTables_ready;
+    FSE_CTable* offcodeCTable;
+    FSE_CTable* matchlengthCTable;
+    FSE_CTable* litlengthCTable;
+    unsigned* entropyScratchSpace;
+};
+
+ZSTD_CCtx* ZSTD_createCCtx(void)
+{
+    return ZSTD_createCCtx_advanced(defaultCustomMem);
+}
+
+ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
+{
+    ZSTD_CCtx* cctx;
+
+    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
+    if (!customMem.customAlloc || !customMem.customFree) return NULL;
+
+    cctx = (ZSTD_CCtx*) ZSTD_malloc(sizeof(ZSTD_CCtx), customMem);
+    if (!cctx) return NULL;
+    memset(cctx, 0, sizeof(ZSTD_CCtx));
+    cctx->customMem = customMem;
+    return cctx;
+}
+
+size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support free on NULL */
+    ZSTD_free(cctx->workSpace, cctx->customMem);
+    ZSTD_free(cctx, cctx->customMem);
+    return 0;   /* reserved as a potential error code in the future */
+}
+
+size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*cctx) + cctx->workSpaceSize;
+}
+
+size_t ZSTD_setCCtxParameter(ZSTD_CCtx* cctx, ZSTD_CCtxParameter param, unsigned value)
+{
+    switch(param)
+    {
+    case ZSTD_p_forceWindow : cctx->forceWindow = value>0; cctx->loadedDictEnd = 0; return 0;
+    case ZSTD_p_forceRawDict : cctx->forceRawDict = value>0; return 0;
+    default: return ERROR(parameter_unknown);
+    }
+}
+
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx)   /* hidden interface */
+{
+    return &(ctx->seqStore);
+}
+
+static ZSTD_parameters ZSTD_getParamsFromCCtx(const ZSTD_CCtx* cctx)
+{
+    return cctx->params;
+}
+
+
+/** ZSTD_checkParams() :
+    ensure param values remain within authorized range.
+    @return : 0, or an error code if one value is beyond authorized range */
+size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
+{
+#   define CLAMPCHECK(val,min,max) { if ((val<min) | (val>max)) return ERROR(compressionParameter_unsupported); }
+    CLAMPCHECK(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
+    CLAMPCHECK(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
+    CLAMPCHECK(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
+    CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
+    CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
+    CLAMPCHECK(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btopt2) return ERROR(compressionParameter_unsupported);
+    return 0;
+}
+
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
+{
+    U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
+    return hashLog - btScale;
+}
+
+/** ZSTD_adjustCParams() :
+    optimize `cPar` for a given input (`srcSize` and `dictSize`).
+    mostly downsizing to reduce memory consumption and initialization.
+    Both `srcSize` and `dictSize` are optional (use 0 if unknown),
+    but if both are 0, no optimization can be done.
+    Note : cPar is considered validated at this stage. Use ZSTD_checkParams() to ensure that. */
+ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+{
+    if (srcSize+dictSize == 0) return cPar;   /* no size information available : no adjustment */
+
+    /* resize params, to use less memory when necessary */
+    {   U32 const minSrcSize = (srcSize==0) ? 500 : 0;
+        U64 const rSize = srcSize + dictSize + minSrcSize;
+        if (rSize < ((U64)1<<ZSTD_WINDOWLOG_MAX)) {
+            U32 const srcLog = MAX(ZSTD_HASHLOG_MIN, ZSTD_highbit32((U32)(rSize)-1) + 1);
+            if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
+    }   }
+    if (cPar.hashLog > cPar.windowLog) cPar.hashLog = cPar.windowLog;
+    {   U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
+        if (cycleLog > cPar.windowLog) cPar.chainLog -= (cycleLog - cPar.windowLog);
+    }
+
+    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN) cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* required for frame header */
+
+    return cPar;
+}
+
+
+size_t ZSTD_estimateCCtxSize(ZSTD_compressionParameters cParams)
+{
+    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, (size_t)1 << cParams.windowLog);
+    U32    const divider = (cParams.searchLength==3) ? 3 : 4;
+    size_t const maxNbSeq = blockSize / divider;
+    size_t const tokenSpace = blockSize + 11*maxNbSeq;
+
+    size_t const chainSize = (cParams.strategy == ZSTD_fast) ? 0 : (1 << cParams.chainLog);
+    size_t const hSize = ((size_t)1) << cParams.hashLog;
+    U32    const hashLog3 = (cParams.searchLength>3) ? 0 : MIN(ZSTD_HASHLOG3_MAX, cParams.windowLog);
+    size_t const h3Size = ((size_t)1) << hashLog3;
+    size_t const entropySpace = hufCTable_size + litlengthCTable_size
+                              + offcodeCTable_size + matchlengthCTable_size
+                              + entropyScratchSpace_size;
+    size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+
+    size_t const optSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits))*sizeof(U32)
+                          + (ZSTD_OPT_NUM+1)*(sizeof(ZSTD_match_t) + sizeof(ZSTD_optimal_t));
+    size_t const neededSpace = entropySpace + tableSpace + tokenSpace
+                             + (((cParams.strategy == ZSTD_btopt) || (cParams.strategy == ZSTD_btopt2)) ? optSpace : 0);
+
+    return sizeof(ZSTD_CCtx) + neededSpace;
+}
+
+
+static U32 ZSTD_equivalentParams(ZSTD_parameters param1, ZSTD_parameters param2)
+{
+    return (param1.cParams.hashLog  == param2.cParams.hashLog)
+         & (param1.cParams.chainLog == param2.cParams.chainLog)
+         & (param1.cParams.strategy == param2.cParams.strategy)
+         & ((param1.cParams.searchLength==3) == (param2.cParams.searchLength==3));
+}
+
+/*! ZSTD_continueCCtx() :
+    reuse CCtx without reset (note : requires no dictionary) */
+static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_parameters params, U64 frameContentSize)
+{
+    U32 const end = (U32)(cctx->nextSrc - cctx->base);
+    cctx->params = params;
+    cctx->frameContentSize = frameContentSize;
+    cctx->consumedSrcSize = 0;
+    cctx->lowLimit = end;
+    cctx->dictLimit = end;
+    cctx->nextToUpdate = end+1;
+    cctx->stage = ZSTDcs_init;
+    cctx->dictID = 0;
+    cctx->loadedDictEnd = 0;
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) cctx->rep[i] = repStartValue[i]; }
+    cctx->seqStore.litLengthSum = 0;  /* force reset of btopt stats */
+    XXH64_reset(&cctx->xxhState, 0);
+    return 0;
+}
+
+typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset, ZSTDcrp_fullReset } ZSTD_compResetPolicy_e;
+
+/*! ZSTD_resetCCtx_internal() :
+    note : `params` must be validated */
+static size_t ZSTD_resetCCtx_internal (ZSTD_CCtx* zc,
+                                       ZSTD_parameters params, U64 frameContentSize,
+                                       ZSTD_compResetPolicy_e const crp)
+{
+    if (crp == ZSTDcrp_continue)
+        if (ZSTD_equivalentParams(params, zc->params)) {
+            zc->fseCTables_ready = 0;
+            zc->hufCTable_repeatMode = HUF_repeat_none;
+            return ZSTD_continueCCtx(zc, params, frameContentSize);
+        }
+
+    {   size_t const blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, (size_t)1 << params.cParams.windowLog);
+        U32    const divider = (params.cParams.searchLength==3) ? 3 : 4;
+        size_t const maxNbSeq = blockSize / divider;
+        size_t const tokenSpace = blockSize + 11*maxNbSeq;
+        size_t const chainSize = (params.cParams.strategy == ZSTD_fast) ? 0 : (1 << params.cParams.chainLog);
+        size_t const hSize = ((size_t)1) << params.cParams.hashLog;
+        U32    const hashLog3 = (params.cParams.searchLength>3) ? 0 : MIN(ZSTD_HASHLOG3_MAX, params.cParams.windowLog);
+        size_t const h3Size = ((size_t)1) << hashLog3;
+        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+        void* ptr;
+
+        /* Check if workSpace is large enough, alloc a new one if needed */
+        {   size_t const entropySpace = hufCTable_size + litlengthCTable_size
+                                  + offcodeCTable_size + matchlengthCTable_size
+                                  + entropyScratchSpace_size;
+            size_t const optPotentialSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits)) * sizeof(U32)
+                                  + (ZSTD_OPT_NUM+1) * (sizeof(ZSTD_match_t)+sizeof(ZSTD_optimal_t));
+            size_t const optSpace = ((params.cParams.strategy == ZSTD_btopt) || (params.cParams.strategy == ZSTD_btopt2)) ? optPotentialSpace : 0;
+            size_t const neededSpace = entropySpace + optSpace + tableSpace + tokenSpace;
+            if (zc->workSpaceSize < neededSpace) {
+                zc->workSpaceSize = 0;
+                ZSTD_free(zc->workSpace, zc->customMem);
+                zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
+                if (zc->workSpace == NULL) return ERROR(memory_allocation);
+                zc->workSpaceSize = neededSpace;
+                ptr = zc->workSpace;
+
+                /* entropy space */
+                zc->hufCTable = (HUF_CElt*)ptr;
+                ptr = (char*)zc->hufCTable + hufCTable_size;  /* note : HUF_CElt* is incomplete type, size is estimated via macro */
+                zc->offcodeCTable = (FSE_CTable*) ptr;
+                ptr = (char*)ptr + offcodeCTable_size;
+                zc->matchlengthCTable = (FSE_CTable*) ptr;
+                ptr = (char*)ptr + matchlengthCTable_size;
+                zc->litlengthCTable = (FSE_CTable*) ptr;
+                ptr = (char*)ptr + litlengthCTable_size;
+                assert(((size_t)ptr & 3) == 0);   /* ensure correct alignment */
+                zc->entropyScratchSpace = (unsigned*) ptr;
+        }   }
+
+        /* init params */
+        zc->params = params;
+        zc->blockSize = blockSize;
+        zc->frameContentSize = frameContentSize;
+        zc->consumedSrcSize = 0;
+
+        XXH64_reset(&zc->xxhState, 0);
+        zc->stage = ZSTDcs_init;
+        zc->dictID = 0;
+        zc->loadedDictEnd = 0;
+        zc->fseCTables_ready = 0;
+        zc->hufCTable_repeatMode = HUF_repeat_none;
+        zc->nextToUpdate = 1;
+        zc->nextSrc = NULL;
+        zc->base = NULL;
+        zc->dictBase = NULL;
+        zc->dictLimit = 0;
+        zc->lowLimit = 0;
+        { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = repStartValue[i]; }
+        zc->hashLog3 = hashLog3;
+        zc->seqStore.litLengthSum = 0;
+
+        /* ensure entropy tables are close together at the beginning */
+        assert((void*)zc->hufCTable == zc->workSpace);
+        assert((char*)zc->offcodeCTable == (char*)zc->hufCTable + hufCTable_size);
+        assert((char*)zc->matchlengthCTable == (char*)zc->offcodeCTable + offcodeCTable_size);
+        assert((char*)zc->litlengthCTable == (char*)zc->matchlengthCTable + matchlengthCTable_size);
+        assert((char*)zc->entropyScratchSpace == (char*)zc->litlengthCTable + litlengthCTable_size);
+        ptr = (char*)zc->entropyScratchSpace + entropyScratchSpace_size;
+
+        /* opt parser space */
+        if ((params.cParams.strategy == ZSTD_btopt) || (params.cParams.strategy == ZSTD_btopt2)) {
+            assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
+            zc->seqStore.litFreq = (U32*)ptr;
+            zc->seqStore.litLengthFreq = zc->seqStore.litFreq + (1<<Litbits);
+            zc->seqStore.matchLengthFreq = zc->seqStore.litLengthFreq + (MaxLL+1);
+            zc->seqStore.offCodeFreq = zc->seqStore.matchLengthFreq + (MaxML+1);
+            ptr = zc->seqStore.offCodeFreq + (MaxOff+1);
+            zc->seqStore.matchTable = (ZSTD_match_t*)ptr;
+            ptr = zc->seqStore.matchTable + ZSTD_OPT_NUM+1;
+            zc->seqStore.priceTable = (ZSTD_optimal_t*)ptr;
+            ptr = zc->seqStore.priceTable + ZSTD_OPT_NUM+1;
+        }
+
+        /* table Space */
+        if (crp!=ZSTDcrp_noMemset) memset(ptr, 0, tableSpace);   /* reset tables only */
+        assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
+        zc->hashTable = (U32*)(ptr);
+        zc->chainTable = zc->hashTable + hSize;
+        zc->hashTable3 = zc->chainTable + chainSize;
+        ptr = zc->hashTable3 + h3Size;
+
+        /* sequences storage */
+        zc->seqStore.sequencesStart = (seqDef*)ptr;
+        ptr = zc->seqStore.sequencesStart + maxNbSeq;
+        zc->seqStore.llCode = (BYTE*) ptr;
+        zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
+        zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
+        zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
+
+        return 0;
+    }
+}
+
+/* ZSTD_invalidateRepCodes() :
+ * ensures next compression will not use repcodes from previous block.
+ * Note : only works with regular variant;
+ *        do not use with extDict variant ! */
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
+    int i;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->rep[i] = 0;
+}
+
+
+/*! ZSTD_copyCCtx_internal() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  pledgedSrcSize=0 means "empty" if fParams.contentSizeFlag=1
+ *  @return : 0, or an error code */
+size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx,
+                              ZSTD_frameParameters fParams, unsigned long long pledgedSrcSize)
+{
+    if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
+
+    memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
+    {   ZSTD_parameters params = srcCCtx->params;
+        params.fParams = fParams;
+        DEBUGLOG(5, "ZSTD_resetCCtx_internal : dictIDFlag : %u \n", !fParams.noDictIDFlag);
+        ZSTD_resetCCtx_internal(dstCCtx, params, pledgedSrcSize, ZSTDcrp_noMemset);
+    }
+
+    /* copy tables */
+    {   size_t const chainSize = (srcCCtx->params.cParams.strategy == ZSTD_fast) ? 0 : (1 << srcCCtx->params.cParams.chainLog);
+        size_t const hSize =  (size_t)1 << srcCCtx->params.cParams.hashLog;
+        size_t const h3Size = (size_t)1 << srcCCtx->hashLog3;
+        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+        assert((U32*)dstCCtx->chainTable == (U32*)dstCCtx->hashTable + hSize);  /* chainTable must follow hashTable */
+        assert((U32*)dstCCtx->hashTable3 == (U32*)dstCCtx->chainTable + chainSize);
+        memcpy(dstCCtx->hashTable, srcCCtx->hashTable, tableSpace);   /* presumes all tables follow each other */
+    }
+
+    /* copy dictionary offsets */
+    dstCCtx->nextToUpdate = srcCCtx->nextToUpdate;
+    dstCCtx->nextToUpdate3= srcCCtx->nextToUpdate3;
+    dstCCtx->nextSrc      = srcCCtx->nextSrc;
+    dstCCtx->base         = srcCCtx->base;
+    dstCCtx->dictBase     = srcCCtx->dictBase;
+    dstCCtx->dictLimit    = srcCCtx->dictLimit;
+    dstCCtx->lowLimit     = srcCCtx->lowLimit;
+    dstCCtx->loadedDictEnd= srcCCtx->loadedDictEnd;
+    dstCCtx->dictID       = srcCCtx->dictID;
+
+    /* copy entropy tables */
+    dstCCtx->fseCTables_ready = srcCCtx->fseCTables_ready;
+    if (srcCCtx->fseCTables_ready) {
+        memcpy(dstCCtx->litlengthCTable, srcCCtx->litlengthCTable, litlengthCTable_size);
+        memcpy(dstCCtx->matchlengthCTable, srcCCtx->matchlengthCTable, matchlengthCTable_size);
+        memcpy(dstCCtx->offcodeCTable, srcCCtx->offcodeCTable, offcodeCTable_size);
+    }
+    dstCCtx->hufCTable_repeatMode = srcCCtx->hufCTable_repeatMode;
+    if (srcCCtx->hufCTable_repeatMode) {
+        memcpy(dstCCtx->hufCTable, srcCCtx->hufCTable, hufCTable_size);
+    }
+
+    return 0;
+}
+
+/*! ZSTD_copyCCtx() :
+ *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
+ *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
+ *  pledgedSrcSize==0 means "unknown".
+*   @return : 0, or an error code */
+size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long long pledgedSrcSize)
+{
+    ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    fParams.contentSizeFlag = pledgedSrcSize>0;
+
+    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx, fParams, pledgedSrcSize);
+}
+
+
+/*! ZSTD_reduceTable() :
+ *  reduce table indexes by `reducerValue` */
+static void ZSTD_reduceTable (U32* const table, U32 const size, U32 const reducerValue)
+{
+    U32 u;
+    for (u=0 ; u < size ; u++) {
+        if (table[u] < reducerValue) table[u] = 0;
+        else table[u] -= reducerValue;
+    }
+}
+
+/*! ZSTD_reduceIndex() :
+*   rescale all indexes to avoid future overflow (indexes are U32) */
+static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
+{
+    { U32 const hSize = 1 << zc->params.cParams.hashLog;
+      ZSTD_reduceTable(zc->hashTable, hSize, reducerValue); }
+
+    { U32 const chainSize = (zc->params.cParams.strategy == ZSTD_fast) ? 0 : (1 << zc->params.cParams.chainLog);
+      ZSTD_reduceTable(zc->chainTable, chainSize, reducerValue); }
+
+    { U32 const h3Size = (zc->hashLog3) ? 1 << zc->hashLog3 : 0;
+      ZSTD_reduceTable(zc->hashTable3, h3Size, reducerValue); }
+}
+
+
+/*-*******************************************************
+*  Block entropic compression
+*********************************************************/
+
+/* See doc/zstd_compression_format.md for detailed format description */
+
+size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    if (srcSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
+    memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
+    MEM_writeLE24(dst, (U32)(srcSize << 2) + (U32)bt_raw);
+    return ZSTD_blockHeaderSize+srcSize;
+}
+
+
+static size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE* const)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    if (srcSize + flSize > dstCapacity) return ERROR(dstSize_tooSmall);
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
+            break;
+        default:   /*note : should not be necessary : flSize is within {1,2,3} */
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
+            break;
+    }
+
+    memcpy(ostart + flSize, src, srcSize);
+    return srcSize + flSize;
+}
+
+static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE* const)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    (void)dstCapacity;  /* dstCapacity already guaranteed to be >=4, hence large enough */
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
+            break;
+        default:   /*note : should not be necessary : flSize is necessarily within {1,2,3} */
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
+            break;
+    }
+
+    ostart[flSize] = *(const BYTE*)src;
+    return flSize+1;
+}
+
+
+static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
+
+static size_t ZSTD_compressLiterals (ZSTD_CCtx* zc,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize)
+{
+    size_t const minGain = ZSTD_minGain(srcSize);
+    size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
+    BYTE*  const ostart = (BYTE*)dst;
+    U32 singleStream = srcSize < 256;
+    symbolEncodingType_e hType = set_compressed;
+    size_t cLitSize;
+
+
+    /* small ? don't even attempt compression (speed opt) */
+#   define LITERAL_NOENTROPY 63
+    {   size_t const minLitSize = zc->hufCTable_repeatMode == HUF_repeat_valid ? 6 : LITERAL_NOENTROPY;
+        if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }
+
+    if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall);   /* not enough space for compression */
+    {   HUF_repeat repeat = zc->hufCTable_repeatMode;
+        int const preferRepeat = zc->params.cParams.strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
+        cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
+                                      zc->entropyScratchSpace, entropyScratchSpace_size, zc->hufCTable, &repeat, preferRepeat)
+                                : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
+                                      zc->entropyScratchSpace, entropyScratchSpace_size, zc->hufCTable, &repeat, preferRepeat);
+        if (repeat != HUF_repeat_none) { hType = set_repeat; }    /* reused the existing table */
+        else { zc->hufCTable_repeatMode = HUF_repeat_check; }       /* now have a table to reuse */
+    }
+
+    if ((cLitSize==0) | (cLitSize >= srcSize - minGain)) {
+        zc->hufCTable_repeatMode = HUF_repeat_none;
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }
+    if (cLitSize==1) {
+        zc->hufCTable_repeatMode = HUF_repeat_none;
+        return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }
+
+    /* Build header */
+    switch(lhSize)
+    {
+    case 3: /* 2 - 2 - 10 - 10 */
+        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
+            MEM_writeLE24(ostart, lhc);
+            break;
+        }
+    case 4: /* 2 - 2 - 14 - 14 */
+        {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
+            MEM_writeLE32(ostart, lhc);
+            break;
+        }
+    default:   /* should not be necessary, lhSize is only {3,4,5} */
+    case 5: /* 2 - 2 - 18 - 18 */
+        {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
+            MEM_writeLE32(ostart, lhc);
+            ostart[4] = (BYTE)(cLitSize >> 10);
+            break;
+        }
+    }
+    return lhSize+cLitSize;
+}
+
+static const BYTE LL_Code[64] = {  0,  1,  2,  3,  4,  5,  6,  7,
+                                   8,  9, 10, 11, 12, 13, 14, 15,
+                                  16, 16, 17, 17, 18, 18, 19, 19,
+                                  20, 20, 20, 20, 21, 21, 21, 21,
+                                  22, 22, 22, 22, 22, 22, 22, 22,
+                                  23, 23, 23, 23, 23, 23, 23, 23,
+                                  24, 24, 24, 24, 24, 24, 24, 24,
+                                  24, 24, 24, 24, 24, 24, 24, 24 };
+
+static const BYTE ML_Code[128] = { 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,
+                                  16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
+                                  32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
+                                  38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
+                                  40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
+                                  41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
+                                  42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
+                                  42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
+
+
+void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
+{
+    BYTE const LL_deltaCode = 19;
+    BYTE const ML_deltaCode = 36;
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    BYTE* const llCodeTable = seqStorePtr->llCode;
+    BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    U32 u;
+    for (u=0; u<nbSeq; u++) {
+        U32 const llv = sequences[u].litLength;
+        U32 const mlv = sequences[u].matchLength;
+        llCodeTable[u] = (llv> 63) ? (BYTE)ZSTD_highbit32(llv) + LL_deltaCode : LL_Code[llv];
+        ofCodeTable[u] = (BYTE)ZSTD_highbit32(sequences[u].offset);
+        mlCodeTable[u] = (mlv>127) ? (BYTE)ZSTD_highbit32(mlv) + ML_deltaCode : ML_Code[mlv];
+    }
+    if (seqStorePtr->longLengthID==1)
+        llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
+    if (seqStorePtr->longLengthID==2)
+        mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
+}
+
+MEM_STATIC size_t ZSTD_compressSequences (ZSTD_CCtx* zc,
+                              void* dst, size_t dstCapacity,
+                              size_t srcSize)
+{
+    const int longOffsets = zc->params.cParams.windowLog > STREAM_ACCUMULATOR_MIN;
+    const seqStore_t* seqStorePtr = &(zc->seqStore);
+    U32 count[MaxSeq+1];
+    S16 norm[MaxSeq+1];
+    FSE_CTable* CTable_LitLength = zc->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = zc->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = zc->matchlengthCTable;
+    U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
+    const seqDef* const sequences = seqStorePtr->sequencesStart;
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    BYTE* seqHead;
+    BYTE scratchBuffer[1<<MAX(MLFSELog,LLFSELog)];
+
+    /* Compress literals */
+    {   const BYTE* const literals = seqStorePtr->litStart;
+        size_t const litSize = seqStorePtr->lit - literals;
+        size_t const cSize = ZSTD_compressLiterals(zc, op, dstCapacity, literals, litSize);
+        if (ZSTD_isError(cSize)) return cSize;
+        op += cSize;
+    }
+
+    /* Sequences Header */
+    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead */) return ERROR(dstSize_tooSmall);
+    if (nbSeq < 0x7F) *op++ = (BYTE)nbSeq;
+    else if (nbSeq < LONGNBSEQ) op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
+    else op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    if (nbSeq==0) goto _check_compressibility;
+
+    /* seqHead : flags for FSE encoding type */
+    seqHead = op++;
+
+#define MIN_SEQ_FOR_DYNAMIC_FSE   64
+#define MAX_SEQ_FOR_STATIC_FSE  1000
+
+    /* convert length/distances into codes */
+    ZSTD_seqToCodes(seqStorePtr);
+
+    /* CTable for Literal Lengths */
+    {   U32 max = MaxLL;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, zc->entropyScratchSpace);
+        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
+            *op++ = llCodeTable[0];
+            FSE_buildCTable_rle(CTable_LitLength, (BYTE)max);
+            LLtype = set_rle;
+        } else if ((zc->fseCTables_ready) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+            LLtype = set_repeat;
+        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (LL_defaultNormLog-1)))) {
+            FSE_buildCTable_wksp(CTable_LitLength, LL_defaultNorm, MaxLL, LL_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
+            LLtype = set_basic;
+        } else {
+            size_t nbSeq_1 = nbSeq;
+            const U32 tableLog = FSE_optimalTableLog(LLFSELog, nbSeq, max);
+            if (count[llCodeTable[nbSeq-1]]>1) { count[llCodeTable[nbSeq-1]]--; nbSeq_1--; }
+            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
+            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
+              if (FSE_isError(NCountSize)) return NCountSize;
+              op += NCountSize; }
+            FSE_buildCTable_wksp(CTable_LitLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
+            LLtype = set_compressed;
+    }   }
+
+    /* CTable for Offsets */
+    {   U32 max = MaxOff;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, zc->entropyScratchSpace);
+        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
+            *op++ = ofCodeTable[0];
+            FSE_buildCTable_rle(CTable_OffsetBits, (BYTE)max);
+            Offtype = set_rle;
+        } else if ((zc->fseCTables_ready) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+            Offtype = set_repeat;
+        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (OF_defaultNormLog-1)))) {
+            FSE_buildCTable_wksp(CTable_OffsetBits, OF_defaultNorm, MaxOff, OF_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
+            Offtype = set_basic;
+        } else {
+            size_t nbSeq_1 = nbSeq;
+            const U32 tableLog = FSE_optimalTableLog(OffFSELog, nbSeq, max);
+            if (count[ofCodeTable[nbSeq-1]]>1) { count[ofCodeTable[nbSeq-1]]--; nbSeq_1--; }
+            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
+            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
+              if (FSE_isError(NCountSize)) return NCountSize;
+              op += NCountSize; }
+            FSE_buildCTable_wksp(CTable_OffsetBits, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
+            Offtype = set_compressed;
+    }   }
+
+    /* CTable for MatchLengths */
+    {   U32 max = MaxML;
+        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, zc->entropyScratchSpace);
+        if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
+            *op++ = *mlCodeTable;
+            FSE_buildCTable_rle(CTable_MatchLength, (BYTE)max);
+            MLtype = set_rle;
+        } else if ((zc->fseCTables_ready) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+            MLtype = set_repeat;
+        } else if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (ML_defaultNormLog-1)))) {
+            FSE_buildCTable_wksp(CTable_MatchLength, ML_defaultNorm, MaxML, ML_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
+            MLtype = set_basic;
+        } else {
+            size_t nbSeq_1 = nbSeq;
+            const U32 tableLog = FSE_optimalTableLog(MLFSELog, nbSeq, max);
+            if (count[mlCodeTable[nbSeq-1]]>1) { count[mlCodeTable[nbSeq-1]]--; nbSeq_1--; }
+            FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max);
+            { size_t const NCountSize = FSE_writeNCount(op, oend-op, norm, max, tableLog);   /* overflow protected */
+              if (FSE_isError(NCountSize)) return NCountSize;
+              op += NCountSize; }
+            FSE_buildCTable_wksp(CTable_MatchLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
+            MLtype = set_compressed;
+    }   }
+
+    *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
+    zc->fseCTables_ready = 0;
+
+    /* Encoding Sequences */
+    {   BIT_CStream_t blockStream;
+        FSE_CState_t  stateMatchLength;
+        FSE_CState_t  stateOffsetBits;
+        FSE_CState_t  stateLitLength;
+
+        CHECK_E(BIT_initCStream(&blockStream, op, oend-op), dstSize_tooSmall); /* not enough space remaining */
+
+        /* first symbols */
+        FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
+        FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
+        FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
+        BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
+        if (MEM_32bits()) BIT_flushBits(&blockStream);
+        BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
+        if (MEM_32bits()) BIT_flushBits(&blockStream);
+        if (longOffsets) {
+            U32 const ofBits = ofCodeTable[nbSeq-1];
+            int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+            if (extraBits) {
+                BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
+                BIT_flushBits(&blockStream);
+            }
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
+                        ofBits - extraBits);
+        } else {
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+        }
+        BIT_flushBits(&blockStream);
+
+        {   size_t n;
+            for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
+                BYTE const llCode = llCodeTable[n];
+                BYTE const ofCode = ofCodeTable[n];
+                BYTE const mlCode = mlCodeTable[n];
+                U32  const llBits = LL_bits[llCode];
+                U32  const ofBits = ofCode;                                     /* 32b*/  /* 64b*/
+                U32  const mlBits = ML_bits[mlCode];
+                                                                                /* (7)*/  /* (7)*/
+                FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
+                FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
+                if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+                FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
+                if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
+                    BIT_flushBits(&blockStream);                                /* (7)*/
+                BIT_addBits(&blockStream, sequences[n].litLength, llBits);
+                if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
+                BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
+                if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+                if (longOffsets) {
+                    int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+                    if (extraBits) {
+                        BIT_addBits(&blockStream, sequences[n].offset, extraBits);
+                        BIT_flushBits(&blockStream);                            /* (7)*/
+                    }
+                    BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
+                                ofBits - extraBits);                            /* 31 */
+                } else {
+                    BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
+                }
+                BIT_flushBits(&blockStream);                                    /* (7)*/
+        }   }
+
+        FSE_flushCState(&blockStream, &stateMatchLength);
+        FSE_flushCState(&blockStream, &stateOffsetBits);
+        FSE_flushCState(&blockStream, &stateLitLength);
+
+        {   size_t const streamSize = BIT_closeCStream(&blockStream);
+            if (streamSize==0) return ERROR(dstSize_tooSmall);   /* not enough space */
+            op += streamSize;
+    }   }
+
+    /* check compressibility */
+_check_compressibility:
+    {   size_t const minGain = ZSTD_minGain(srcSize);
+        size_t const maxCSize = srcSize - minGain;
+        if ((size_t)(op-ostart) >= maxCSize) {
+            zc->hufCTable_repeatMode = HUF_repeat_none;
+            return 0;
+    }   }
+
+    /* confirm repcodes */
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = zc->repToConfirm[i]; }
+
+    return op - ostart;
+}
+
+#if 0 /* for debug */
+#  define STORESEQ_DEBUG
+#include <stdio.h>   /* fprintf */
+U32 g_startDebug = 0;
+const BYTE* g_start = NULL;
+#endif
+
+/*! ZSTD_storeSeq() :
+    Store a sequence (literal length, literals, offset code and match length code) into seqStore_t.
+    `offsetCode` : distance to match, or 0 == repCode.
+    `matchCode` : matchLength - MINMATCH
+*/
+MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const void* literals, U32 offsetCode, size_t matchCode)
+{
+#ifdef STORESEQ_DEBUG
+    if (g_startDebug) {
+        const U32 pos = (U32)((const BYTE*)literals - g_start);
+        if (g_start==NULL) g_start = (const BYTE*)literals;
+        if ((pos > 1895000) && (pos < 1895300))
+            DEBUGLOG(5, "Cpos %6u :%5u literals & match %3u bytes at distance %6u \n",
+                   pos, (U32)litLength, (U32)matchCode+MINMATCH, (U32)offsetCode);
+    }
+#endif
+    /* copy Literals */
+    ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
+    seqStorePtr->lit += litLength;
+
+    /* literal Length */
+    if (litLength>0xFFFF) {
+        seqStorePtr->longLengthID = 1;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].litLength = (U16)litLength;
+
+    /* match offset */
+    seqStorePtr->sequences[0].offset = offsetCode + 1;
+
+    /* match Length */
+    if (matchCode>0xFFFF) {
+        seqStorePtr->longLengthID = 2;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].matchLength = (U16)matchCode;
+
+    seqStorePtr->sequences++;
+}
+
+
+/*-*************************************
+*  Match length counter
+***************************************/
+static unsigned ZSTD_NbCommonBytes (register size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanForward64( &r, (U64)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctzll((U64)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
+                                                     0, 3, 1, 3, 1, 4, 2, 7,
+                                                     0, 2, 3, 6, 1, 5, 3, 5,
+                                                     1, 3, 4, 4, 2, 5, 6, 7,
+                                                     7, 0, 1, 2, 3, 3, 4, 6,
+                                                     2, 6, 5, 5, 3, 4, 5, 6,
+                                                     7, 1, 2, 4, 6, 4, 4, 5,
+                                                     7, 2, 6, 5, 7, 6, 7, 7 };
+            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r=0;
+            _BitScanForward( &r, (U32)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctz((U32)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
+                                                     3, 2, 2, 1, 3, 2, 0, 1,
+                                                     3, 3, 1, 2, 2, 2, 2, 0,
+                                                     3, 1, 2, 0, 1, 0, 1, 1 };
+            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#       endif
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            _BitScanReverse64( &r, val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clzll(val) >> 3);
+#       else
+            unsigned r;
+            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
+            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r = 0;
+            _BitScanReverse( &r, (unsigned long)val );
+            return (unsigned)(r>>3);
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clz((U32)val) >> 3);
+#       else
+            unsigned r;
+            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+    }   }
+}
+
+
+static size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
+{
+    const BYTE* const pStart = pIn;
+    const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
+
+    while (pIn < pInLoopLimit) {
+        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+        if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
+        pIn += ZSTD_NbCommonBytes(diff);
+        return (size_t)(pIn - pStart);
+    }
+    if (MEM_64bits()) if ((pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (size_t)(pIn - pStart);
+}
+
+/** ZSTD_count_2segments() :
+*   can count match length with `ip` & `match` in 2 different segments.
+*   convention : on reaching mEnd, match count continue starting from iStart
+*/
+static size_t ZSTD_count_2segments(const BYTE* ip, const BYTE* match, const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
+{
+    const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
+    size_t const matchLength = ZSTD_count(ip, match, vEnd);
+    if (match + matchLength != mEnd) return matchLength;
+    return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
+}
+
+
+/*-*************************************
+*  Hashes
+***************************************/
+static const U32 prime3bytes = 506832829U;
+static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
+MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); } /* only in zstd_opt.h */
+
+static const U32 prime4bytes = 2654435761U;
+static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
+
+static const U64 prime5bytes = 889523592379ULL;
+static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
+static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
+
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+
+static const U64 prime7bytes = 58295818150454627ULL;
+static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
+static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+
+static size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
+{
+    switch(mls)
+    {
+    default:
+    case 4: return ZSTD_hash4Ptr(p, hBits);
+    case 5: return ZSTD_hash5Ptr(p, hBits);
+    case 6: return ZSTD_hash6Ptr(p, hBits);
+    case 7: return ZSTD_hash7Ptr(p, hBits);
+    case 8: return ZSTD_hash8Ptr(p, hBits);
+    }
+}
+
+
+/*-*************************************
+*  Fast Scan
+***************************************/
+static void ZSTD_fillHashTable (ZSTD_CCtx* zc, const void* end, const U32 mls)
+{
+    U32* const hashTable = zc->hashTable;
+    U32  const hBits = zc->params.cParams.hashLog;
+    const BYTE* const base = zc->base;
+    const BYTE* ip = base + zc->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const size_t fastHashFillStep = 3;
+
+    while(ip <= iend) {
+        hashTable[ZSTD_hashPtr(ip, hBits, mls)] = (U32)(ip - base);
+        ip += fastHashFillStep;
+    }
+}
+
+
+FORCE_INLINE
+void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
+                               const void* src, size_t srcSize,
+                               const U32 mls)
+{
+    U32* const hashTable = cctx->hashTable;
+    U32  const hBits = cctx->params.cParams.hashLog;
+    seqStore_t* seqStorePtr = &(cctx->seqStore);
+    const BYTE* const base = cctx->base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = cctx->dictLimit;
+    const BYTE* const lowest = base + lowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=cctx->rep[0], offset_2=cctx->rep[1];
+    U32 offsetSaved = 0;
+
+    /* init */
+    ip += (ip==lowest);
+    {   U32 const maxRep = (U32)(ip-lowest);
+        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    }
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        size_t const h = ZSTD_hashPtr(ip, hBits, mls);
+        U32 const current = (U32)(ip-base);
+        U32 const matchIndex = hashTable[h];
+        const BYTE* match = base + matchIndex;
+        hashTable[h] = current;   /* update hash table */
+
+        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            U32 offset;
+            if ( (matchIndex <= lowestIndex) || (MEM_read32(match) != MEM_read32(ip)) ) {
+                ip += ((ip-anchor) >> g_searchStrength) + 1;
+                continue;
+            }
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            offset = (U32)(ip-match);
+            while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+            offset_2 = offset_1;
+            offset_1 = offset;
+
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }
+
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashTable[ZSTD_hashPtr(ip-2, hBits, mls)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; }  /* swap offset_2 <=> offset_1 */
+                hashTable[ZSTD_hashPtr(ip, hBits, mls)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* save reps for next block */
+    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
+    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+static void ZSTD_compressBlock_fast(ZSTD_CCtx* ctx,
+                       const void* src, size_t srcSize)
+{
+    const U32 mls = ctx->params.cParams.searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 4); return;
+    case 5 :
+        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 5); return;
+    case 6 :
+        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 6); return;
+    case 7 :
+        ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 7); return;
+    }
+}
+
+
+static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
+                                 const void* src, size_t srcSize,
+                                 const U32 mls)
+{
+    U32* hashTable = ctx->hashTable;
+    const U32 hBits = ctx->params.cParams.hashLog;
+    seqStore_t* seqStorePtr = &(ctx->seqStore);
+    const BYTE* const base = ctx->base;
+    const BYTE* const dictBase = ctx->dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ctx->lowLimit;
+    const BYTE* const dictStart = dictBase + lowestIndex;
+    const U32   dictLimit = ctx->dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=ctx->rep[0], offset_2=ctx->rep[1];
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t h = ZSTD_hashPtr(ip, hBits, mls);
+        const U32 matchIndex = hashTable[h];
+        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+        const U32 current = (U32)(ip-base);
+        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
+        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
+        const BYTE* repMatch = repBase + repIndex;
+        size_t mLength;
+        hashTable[h] = current;   /* update hash table */
+
+        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ( (matchIndex < lowestIndex) ||
+                 (MEM_read32(match) != MEM_read32(ip)) ) {
+                ip += ((ip-anchor) >> g_searchStrength) + 1;
+                continue;
+            }
+            {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                offset = current - matchIndex;
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }   }
+
+        /* found a match : store it */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;
+            hashTable[ZSTD_hashPtr(ip-2, hBits, mls)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
+                    hashTable[ZSTD_hashPtr(ip, hBits, mls)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+static void ZSTD_compressBlock_fast_extDict(ZSTD_CCtx* ctx,
+                         const void* src, size_t srcSize)
+{
+    U32 const mls = ctx->params.cParams.searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 4); return;
+    case 5 :
+        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 5); return;
+    case 6 :
+        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 6); return;
+    case 7 :
+        ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 7); return;
+    }
+}
+
+
+/*-*************************************
+*  Double Fast
+***************************************/
+static void ZSTD_fillDoubleHashTable (ZSTD_CCtx* cctx, const void* end, const U32 mls)
+{
+    U32* const hashLarge = cctx->hashTable;
+    U32  const hBitsL = cctx->params.cParams.hashLog;
+    U32* const hashSmall = cctx->chainTable;
+    U32  const hBitsS = cctx->params.cParams.chainLog;
+    const BYTE* const base = cctx->base;
+    const BYTE* ip = base + cctx->nextToUpdate;
+    const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
+    const size_t fastHashFillStep = 3;
+
+    while(ip <= iend) {
+        hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip - base);
+        hashLarge[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip - base);
+        ip += fastHashFillStep;
+    }
+}
+
+
+FORCE_INLINE
+void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
+                                 const void* src, size_t srcSize,
+                                 const U32 mls)
+{
+    U32* const hashLong = cctx->hashTable;
+    const U32 hBitsL = cctx->params.cParams.hashLog;
+    U32* const hashSmall = cctx->chainTable;
+    const U32 hBitsS = cctx->params.cParams.chainLog;
+    seqStore_t* seqStorePtr = &(cctx->seqStore);
+    const BYTE* const base = cctx->base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32 lowestIndex = cctx->dictLimit;
+    const BYTE* const lowest = base + lowestIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=cctx->rep[0], offset_2=cctx->rep[1];
+    U32 offsetSaved = 0;
+
+    /* init */
+    ip += (ip==lowest);
+    {   U32 const maxRep = (U32)(ip-lowest);
+        if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
+    }
+
+    /* Main Search Loop */
+    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+        size_t mLength;
+        size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
+        size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        U32 const current = (U32)(ip-base);
+        U32 const matchIndexL = hashLong[h2];
+        U32 const matchIndexS = hashSmall[h];
+        const BYTE* matchLong = base + matchIndexL;
+        const BYTE* match = base + matchIndexS;
+        hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
+
+        assert(offset_1 <= current);   /* supposed guaranteed by construction */
+        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
+            /* favor repcode */
+            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            U32 offset;
+            if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) {
+                mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
+                offset = (U32)(ip-matchLong);
+                while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+            } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) {
+                size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndexL3 = hashLong[hl3];
+                const BYTE* matchL3 = base + matchIndexL3;
+                hashLong[hl3] = current + 1;
+                if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) {
+                    mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
+                    ip++;
+                    offset = (U32)(ip-matchL3);
+                    while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                } else {
+                    mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+                    offset = (U32)(ip-match);
+                    while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                }
+            } else {
+                ip += ((ip-anchor) >> g_searchStrength) + 1;
+                continue;
+            }
+
+            offset_2 = offset_1;
+            offset_1 = offset;
+
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        }
+
+        /* match found */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] =
+                hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] =
+                hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+
+            /* check immediate repcode */
+            while ( (ip <= ilimit)
+                 && ( (offset_2>0)
+                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                /* store sequence */
+                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
+                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
+                ip += rLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
+
+    /* save reps for next block */
+    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
+    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+static void ZSTD_compressBlock_doubleFast(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    const U32 mls = ctx->params.cParams.searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 4); return;
+    case 5 :
+        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 5); return;
+    case 6 :
+        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 6); return;
+    case 7 :
+        ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 7); return;
+    }
+}
+
+
+static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
+                                 const void* src, size_t srcSize,
+                                 const U32 mls)
+{
+    U32* const hashLong = ctx->hashTable;
+    U32  const hBitsL = ctx->params.cParams.hashLog;
+    U32* const hashSmall = ctx->chainTable;
+    U32  const hBitsS = ctx->params.cParams.chainLog;
+    seqStore_t* seqStorePtr = &(ctx->seqStore);
+    const BYTE* const base = ctx->base;
+    const BYTE* const dictBase = ctx->dictBase;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   lowestIndex = ctx->lowLimit;
+    const BYTE* const dictStart = dictBase + lowestIndex;
+    const U32   dictLimit = ctx->dictLimit;
+    const BYTE* const lowPrefixPtr = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    U32 offset_1=ctx->rep[0], offset_2=ctx->rep[1];
+
+    /* Search Loop */
+    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
+        const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
+        const U32 matchIndex = hashSmall[hSmall];
+        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* match = matchBase + matchIndex;
+
+        const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
+        const U32 matchLongIndex = hashLong[hLong];
+        const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base;
+        const BYTE* matchLong = matchLongBase + matchLongIndex;
+
+        const U32 current = (U32)(ip-base);
+        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
+        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
+        const BYTE* repMatch = repBase + repIndex;
+        size_t mLength;
+        hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
+
+        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
+        } else {
+            if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend;
+                const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr;
+                U32 offset;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8;
+                offset = current - matchLongIndex;
+                while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+            } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+                size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+                U32 const matchIndex3 = hashLong[h3];
+                const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base;
+                const BYTE* match3 = match3Base + matchIndex3;
+                U32 offset;
+                hashLong[h3] = current + 1;
+                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend;
+                    const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8;
+                    ip++;
+                    offset = current+1 - matchIndex3;
+                    while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
+                } else {
+                    const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
+                    const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                    offset = current - matchIndex;
+                    while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
+                }
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+
+            } else {
+                ip += ((ip-anchor) >> g_searchStrength) + 1;
+                continue;
+        }   }
+
+        /* found a match : store it */
+        ip += mLength;
+        anchor = ip;
+
+        if (ip <= ilimit) {
+            /* Fill Table */
+            hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;
+            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2;
+            hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+            /* check immediate repcode */
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+    }   }   }
+
+    /* save reps for next block */
+    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+static void ZSTD_compressBlock_doubleFast_extDict(ZSTD_CCtx* ctx,
+                         const void* src, size_t srcSize)
+{
+    U32 const mls = ctx->params.cParams.searchLength;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 4); return;
+    case 5 :
+        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 5); return;
+    case 6 :
+        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 6); return;
+    case 7 :
+        ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 7); return;
+    }
+}
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+/** ZSTD_insertBt1() : add one or multiple positions to tree.
+*   ip : assumed <= iend-8 .
+*   @return : nb of positions added */
+static U32 ZSTD_insertBt1(ZSTD_CCtx* zc, const BYTE* const ip, const U32 mls, const BYTE* const iend, U32 nbCompares,
+                          U32 extDict)
+{
+    U32*   const hashTable = zc->hashTable;
+    U32    const hashLog = zc->params.cParams.hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = zc->chainTable;
+    U32    const btLog  = zc->params.cParams.chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = zc->base;
+    const BYTE* const dictBase = zc->dictBase;
+    const U32 dictLimit = zc->dictLimit;
+    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);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowLow = zc->lowLimit;
+    U32 matchEndIdx = current+8;
+    size_t bestLength = 8;
+#ifdef ZSTD_C_PREDICT
+    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
+    predictedSmall += (predictedSmall>0);
+    predictedLarge += (predictedLarge>0);
+#endif /* ZSTD_C_PREDICT */
+
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+
+#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 */
+        if (matchIndex == predictedSmall) {
+            /* no need to check length, result known */
+            *smallerPtr = matchIndex;
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
+            continue;
+        }
+        if (matchIndex == predictedLarge) {
+            *largerPtr = matchIndex;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
+            continue;
+        }
+#endif
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            match = base + matchIndex;
+            if (match[matchLength] == ip[matchLength])
+                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iend) +1;
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            bestLength = matchLength;
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+        }
+
+        if (ip+matchLength == iend)   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt the tree */
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within correct buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+    if (matchEndIdx > current + 8) return matchEndIdx - current - 8;
+    return 1;
+}
+
+
+static size_t ZSTD_insertBtAndFindBestMatch (
+                        ZSTD_CCtx* zc,
+                        const BYTE* const ip, const BYTE* const iend,
+                        size_t* offsetPtr,
+                        U32 nbCompares, const U32 mls,
+                        U32 extDict)
+{
+    U32*   const hashTable = zc->hashTable;
+    U32    const hashLog = zc->params.cParams.hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = zc->chainTable;
+    U32    const btLog  = zc->params.cParams.chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex  = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = zc->base;
+    const BYTE* const dictBase = zc->dictBase;
+    const U32 dictLimit = zc->dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const U32 current = (U32)(ip-base);
+    const U32 btLow = btMask >= current ? 0 : current - btMask;
+    const U32 windowLow = zc->lowLimit;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = bt + 2*(current&btMask) + 1;
+    U32 matchEndIdx = current+8;
+    U32 dummy32;   /* to be nullified at the end */
+    size_t bestLength = 0;
+
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match;
+
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            match = base + matchIndex;
+            if (match[matchLength] == ip[matchLength])
+                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iend) +1;
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
+                bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
+            if (ip+matchLength == iend)   /* equal : no way to know if inf or sup */
+                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+
+    zc->nextToUpdate = (matchEndIdx > current + 8) ? matchEndIdx - 8 : current+1;
+    return bestLength;
+}
+
+
+static void ZSTD_updateTree(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls)
+{
+    const BYTE* const base = zc->base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = zc->nextToUpdate;
+
+    while(idx < target)
+        idx += ZSTD_insertBt1(zc, base+idx, mls, iend, nbCompares, 0);
+}
+
+/** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
+static size_t ZSTD_BtFindBestMatch (
+                        ZSTD_CCtx* zc,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 mls)
+{
+    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls);
+    return ZSTD_insertBtAndFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, mls, 0);
+}
+
+
+static size_t ZSTD_BtFindBestMatch_selectMLS (
+                        ZSTD_CCtx* zc,   /* Index table will be updated */
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+{
+    switch(matchLengthSearch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4);
+    case 5 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6);
+    }
+}
+
+
+static void ZSTD_updateTree_extDict(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls)
+{
+    const BYTE* const base = zc->base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = zc->nextToUpdate;
+
+    while (idx < target) idx += ZSTD_insertBt1(zc, base+idx, mls, iend, nbCompares, 1);
+}
+
+
+/** Tree updater, providing best match */
+static size_t ZSTD_BtFindBestMatch_extDict (
+                        ZSTD_CCtx* zc,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 mls)
+{
+    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls);
+    return ZSTD_insertBtAndFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, mls, 1);
+}
+
+
+static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
+                        ZSTD_CCtx* zc,   /* Index table will be updated */
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+{
+    switch(matchLengthSearch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4);
+    case 5 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6);
+    }
+}
+
+
+
+/* *********************************
+*  Hash Chain
+***********************************/
+#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & mask]
+
+/* Update chains up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+FORCE_INLINE
+U32 ZSTD_insertAndFindFirstIndex (ZSTD_CCtx* zc, const BYTE* ip, U32 mls)
+{
+    U32* const hashTable  = zc->hashTable;
+    const U32 hashLog = zc->params.cParams.hashLog;
+    U32* const chainTable = zc->chainTable;
+    const U32 chainMask = (1 << zc->params.cParams.chainLog) - 1;
+    const BYTE* const base = zc->base;
+    const U32 target = (U32)(ip - base);
+    U32 idx = zc->nextToUpdate;
+
+    while(idx < target) { /* catch up */
+        size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
+        NEXT_IN_CHAIN(idx, chainMask) = hashTable[h];
+        hashTable[h] = idx;
+        idx++;
+    }
+
+    zc->nextToUpdate = target;
+    return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
+}
+
+
+
+FORCE_INLINE /* inlining is important to hardwire a hot branch (template emulation) */
+size_t ZSTD_HcFindBestMatch_generic (
+                        ZSTD_CCtx* zc,   /* Index table will be updated */
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 mls, const U32 extDict)
+{
+    U32* const chainTable = zc->chainTable;
+    const U32 chainSize = (1 << zc->params.cParams.chainLog);
+    const U32 chainMask = chainSize-1;
+    const BYTE* const base = zc->base;
+    const BYTE* const dictBase = zc->dictBase;
+    const U32 dictLimit = zc->dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32 lowLimit = zc->lowLimit;
+    const U32 current = (U32)(ip-base);
+    const U32 minChain = current > chainSize ? current - chainSize : 0;
+    int nbAttempts=maxNbAttempts;
+    size_t ml=4-1;
+
+    /* HC4 match finder */
+    U32 matchIndex = ZSTD_insertAndFindFirstIndex (zc, ip, mls);
+
+    for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
+        const BYTE* match;
+        size_t currentMl=0;
+        if ((!extDict) || matchIndex >= dictLimit) {
+            match = base + matchIndex;
+            if (match[ml] == ip[ml])   /* potentially better */
+                currentMl = ZSTD_count(ip, match, iLimit);
+        } else {
+            match = dictBase + matchIndex;
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
+        }
+
+        /* save best solution */
+        if (currentMl > ml) {
+            ml = currentMl;
+            *offsetPtr = current - matchIndex + ZSTD_REP_MOVE;
+            if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+        }
+
+        if (matchIndex <= minChain) break;
+        matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
+    }
+
+    return ml;
+}
+
+
+FORCE_INLINE size_t ZSTD_HcFindBestMatch_selectMLS (
+                        ZSTD_CCtx* zc,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+{
+    switch(matchLengthSearch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4, 0);
+    case 5 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5, 0);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6, 0);
+    }
+}
+
+
+FORCE_INLINE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
+                        ZSTD_CCtx* zc,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr,
+                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+{
+    switch(matchLengthSearch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4, 1);
+    case 5 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5, 1);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6, 1);
+    }
+}
+
+
+/* *******************************
+*  Common parser - lazy strategy
+*********************************/
+FORCE_INLINE
+void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
+                                     const void* src, size_t srcSize,
+                                     const U32 searchMethod, const U32 depth)
+{
+    seqStore_t* seqStorePtr = &(ctx->seqStore);
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ctx->base + ctx->dictLimit;
+
+    U32 const maxSearches = 1 << ctx->params.cParams.searchLog;
+    U32 const mls = ctx->params.cParams.searchLength;
+
+    typedef size_t (*searchMax_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLimit,
+                        size_t* offsetPtr,
+                        U32 maxNbAttempts, U32 matchLengthSearch);
+    searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
+    U32 offset_1 = ctx->rep[0], offset_2 = ctx->rep[1], savedOffset=0;
+
+    /* init */
+    ip += (ip==base);
+    ctx->nextToUpdate3 = ctx->nextToUpdate;
+    {   U32 const maxRep = (U32)(ip-base);
+        if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
+        if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
+    }
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offset=0;
+        const BYTE* start=ip+1;
+
+        /* check repCode */
+        if ((offset_1>0) & (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1))) {
+            /* repcode : we take it */
+            matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+            if (depth==0) goto _storeSequence;
+        }
+
+        /* first search (depth 0) */
+        {   size_t offsetFound = 99999999;
+            size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offset=offsetFound;
+        }
+
+        if (matchLength < 4) {
+            ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                int const gain2 = (int)(mlRep * 3);
+                int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                if ((mlRep >= 4) && (gain2 > gain1))
+                    matchLength = mlRep, offset = 0, start = ip;
+            }
+            {   size_t offset2=99999999;
+                size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offset = offset2, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const ml2 = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(ml2 * 4);
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((ml2 >= 4) && (gain2 > gain1))
+                        matchLength = ml2, offset = 0, start = ip;
+                }
+                {   size_t offset2=99999999;
+                    size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offset = offset2, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* catch up */
+        if (offset) {
+            while ( (start > anchor)
+                 && (start > base+offset-ZSTD_REP_MOVE)
+                 && (start[-1] == start[-1-offset+ZSTD_REP_MOVE]) )  /* only search for offset within prefix */
+                { start--; matchLength++; }
+            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+        }
+
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = start - anchor;
+            ZSTD_storeSeq(seqStorePtr, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            anchor = ip = start + matchLength;
+        }
+
+        /* check immediate repcode */
+        while ( (ip <= ilimit)
+             && ((offset_2>0)
+             & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+            /* store sequence */
+            matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+            offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
+            ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
+            ip += matchLength;
+            anchor = ip;
+            continue;   /* faster when present ... (?) */
+    }   }
+
+    /* Save reps for next block */
+    ctx->repToConfirm[0] = offset_1 ? offset_1 : savedOffset;
+    ctx->repToConfirm[1] = offset_2 ? offset_2 : savedOffset;
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+static void ZSTD_compressBlock_btlazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 1, 2);
+}
+
+static void ZSTD_compressBlock_lazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 2);
+}
+
+static void ZSTD_compressBlock_lazy(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 1);
+}
+
+static void ZSTD_compressBlock_greedy(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 0);
+}
+
+
+FORCE_INLINE
+void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
+                                     const void* src, size_t srcSize,
+                                     const U32 searchMethod, const U32 depth)
+{
+    seqStore_t* seqStorePtr = &(ctx->seqStore);
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ctx->base;
+    const U32 dictLimit = ctx->dictLimit;
+    const U32 lowestIndex = ctx->lowLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictBase = ctx->dictBase;
+    const BYTE* const dictEnd  = dictBase + dictLimit;
+    const BYTE* const dictStart  = dictBase + ctx->lowLimit;
+
+    const U32 maxSearches = 1 << ctx->params.cParams.searchLog;
+    const U32 mls = ctx->params.cParams.searchLength;
+
+    typedef size_t (*searchMax_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLimit,
+                        size_t* offsetPtr,
+                        U32 maxNbAttempts, U32 matchLengthSearch);
+    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
+
+    U32 offset_1 = ctx->rep[0], offset_2 = ctx->rep[1];
+
+    /* init */
+    ctx->nextToUpdate3 = ctx->nextToUpdate;
+    ip += (ip == prefixStart);
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        size_t matchLength=0;
+        size_t offset=0;
+        const BYTE* start=ip+1;
+        U32 current = (U32)(ip-base);
+
+        /* check repCode */
+        {   const U32 repIndex = (U32)(current+1 - offset_1);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))   /* intentional overflow */
+            if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                if (depth==0) goto _storeSequence;
+        }   }
+
+        /* first search (depth 0) */
+        {   size_t offsetFound = 99999999;
+            size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
+            if (ml2 > matchLength)
+                matchLength = ml2, start = ip, offset=offsetFound;
+        }
+
+         if (matchLength < 4) {
+            ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
+            continue;
+        }
+
+        /* let's try to find a better solution */
+        if (depth>=1)
+        while (ip<ilimit) {
+            ip ++;
+            current++;
+            /* check repCode */
+            if (offset) {
+                const U32 repIndex = (U32)(current - offset_1);
+                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                const BYTE* const repMatch = repBase + repIndex;
+                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                    /* repcode detected */
+                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                    size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                    int const gain2 = (int)(repLength * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((repLength >= 4) && (gain2 > gain1))
+                        matchLength = repLength, offset = 0, start = ip;
+            }   }
+
+            /* search match, depth 1 */
+            {   size_t offset2=99999999;
+                size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+                int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
+                if ((ml2 >= 4) && (gain2 > gain1)) {
+                    matchLength = ml2, offset = offset2, start = ip;
+                    continue;   /* search a better one */
+            }   }
+
+            /* let's find an even better one */
+            if ((depth==2) && (ip<ilimit)) {
+                ip ++;
+                current++;
+                /* check repCode */
+                if (offset) {
+                    const U32 repIndex = (U32)(current - offset_1);
+                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                    const BYTE* const repMatch = repBase + repIndex;
+                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                    if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                        /* repcode detected */
+                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                        size_t const repLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                        int const gain2 = (int)(repLength * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        if ((repLength >= 4) && (gain2 > gain1))
+                            matchLength = repLength, offset = 0, start = ip;
+                }   }
+
+                /* search match, depth 2 */
+                {   size_t offset2=99999999;
+                    size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+                    int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
+                    int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
+                        matchLength = ml2, offset = offset2, start = ip;
+                        continue;
+            }   }   }
+            break;  /* nothing found : store previous solution */
+        }
+
+        /* catch up */
+        if (offset) {
+            U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+            const BYTE* match = (matchIndex < dictLimit) ? dictBase + matchIndex : base + matchIndex;
+            const BYTE* const mStart = (matchIndex < dictLimit) ? dictStart : prefixStart;
+            while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
+        }
+
+        /* store sequence */
+_storeSequence:
+        {   size_t const litLength = start - anchor;
+            ZSTD_storeSeq(seqStorePtr, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            anchor = ip = start + matchLength;
+        }
+
+        /* check immediate repcode */
+        while (ip <= ilimit) {
+            const U32 repIndex = (U32)((ip-base) - offset_2);
+            const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+            const BYTE* const repMatch = repBase + repIndex;
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+            if (MEM_read32(ip) == MEM_read32(repMatch)) {
+                /* repcode detected we should take it */
+                const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
+                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset history */
+                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+            }
+            break;
+    }   }
+
+    /* Save reps for next block */
+    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+void ZSTD_compressBlock_greedy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 0);
+}
+
+static void ZSTD_compressBlock_lazy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 1);
+}
+
+static void ZSTD_compressBlock_lazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 2);
+}
+
+static void ZSTD_compressBlock_btlazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+    ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 1, 2);
+}
+
+
+/* The optimal parser */
+#include "zstd_opt.h"
+
+static void ZSTD_compressBlock_btopt(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+#ifdef ZSTD_OPT_H_91842398743
+    ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 0);
+#else
+    (void)ctx; (void)src; (void)srcSize;
+    return;
+#endif
+}
+
+static void ZSTD_compressBlock_btopt2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+#ifdef ZSTD_OPT_H_91842398743
+    ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 1);
+#else
+    (void)ctx; (void)src; (void)srcSize;
+    return;
+#endif
+}
+
+static void ZSTD_compressBlock_btopt_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+#ifdef ZSTD_OPT_H_91842398743
+    ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 0);
+#else
+    (void)ctx; (void)src; (void)srcSize;
+    return;
+#endif
+}
+
+static void ZSTD_compressBlock_btopt2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+{
+#ifdef ZSTD_OPT_H_91842398743
+    ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 1);
+#else
+    (void)ctx; (void)src; (void)srcSize;
+    return;
+#endif
+}
+
+
+typedef void (*ZSTD_blockCompressor) (ZSTD_CCtx* ctx, const void* src, size_t srcSize);
+
+static ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
+{
+    static const ZSTD_blockCompressor blockCompressor[2][8] = {
+        { ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy,
+          ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2,
+          ZSTD_compressBlock_btopt, ZSTD_compressBlock_btopt2 },
+        { ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict,
+          ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict,
+          ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btopt2_extDict }
+    };
+
+    return blockCompressor[extDict][(U32)strat];
+}
+
+
+static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->params.cParams.strategy, zc->lowLimit < zc->dictLimit);
+    const BYTE* const base = zc->base;
+    const BYTE* const istart = (const BYTE*)src;
+    const U32 current = (U32)(istart-base);
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) return 0;   /* don't even attempt compression below a certain srcSize */
+    ZSTD_resetSeqStore(&(zc->seqStore));
+    if (current > zc->nextToUpdate + 384)
+        zc->nextToUpdate = current - MIN(192, (U32)(current - zc->nextToUpdate - 384));   /* limited update after finding a very long match */
+    blockCompressor(zc, src, srcSize);
+    return ZSTD_compressSequences(zc, dst, dstCapacity, srcSize);
+}
+
+
+/*! ZSTD_compress_generic() :
+*   Compress a chunk of data into one or multiple blocks.
+*   All blocks will be terminated, all input will be consumed.
+*   Function will issue an error if there is not enough `dstCapacity` to hold the compressed content.
+*   Frame is supposed already started (header already produced)
+*   @return : compressed size, or an error code
+*/
+static size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
+                                     void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                                     U32 lastFrameChunk)
+{
+    size_t blockSize = cctx->blockSize;
+    size_t remaining = srcSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    U32 const maxDist = 1 << cctx->params.cParams.windowLog;
+
+    if (cctx->params.fParams.checksumFlag && srcSize)
+        XXH64_update(&cctx->xxhState, src, srcSize);
+
+    while (remaining) {
+        U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
+        size_t cSize;
+
+        if (dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE)
+            return ERROR(dstSize_tooSmall);   /* not enough space to store compressed block */
+        if (remaining < blockSize) blockSize = remaining;
+
+        /* preemptive overflow correction */
+        if (cctx->lowLimit > (3U<<29)) {
+            U32 const cycleMask = (1 << ZSTD_cycleLog(cctx->params.cParams.hashLog, cctx->params.cParams.strategy)) - 1;
+            U32 const current = (U32)(ip - cctx->base);
+            U32 const newCurrent = (current & cycleMask) + (1 << cctx->params.cParams.windowLog);
+            U32 const correction = current - newCurrent;
+            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_64 <= 30);
+            ZSTD_reduceIndex(cctx, correction);
+            cctx->base += correction;
+            cctx->dictBase += correction;
+            cctx->lowLimit -= correction;
+            cctx->dictLimit -= correction;
+            if (cctx->nextToUpdate < correction) cctx->nextToUpdate = 0;
+            else cctx->nextToUpdate -= correction;
+        }
+
+        if ((U32)(ip+blockSize - cctx->base) > cctx->loadedDictEnd + maxDist) {
+            /* enforce maxDist */
+            U32 const newLowLimit = (U32)(ip+blockSize - cctx->base) - maxDist;
+            if (cctx->lowLimit < newLowLimit) cctx->lowLimit = newLowLimit;
+            if (cctx->dictLimit < cctx->lowLimit) cctx->dictLimit = cctx->lowLimit;
+        }
+
+        cSize = ZSTD_compressBlock_internal(cctx, op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize, ip, blockSize);
+        if (ZSTD_isError(cSize)) return cSize;
+
+        if (cSize == 0) {  /* block is not compressible */
+            U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
+            if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
+            MEM_writeLE32(op, cBlockHeader24);   /* no pb, 4th byte will be overwritten */
+            memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
+            cSize = ZSTD_blockHeaderSize+blockSize;
+        } else {
+            U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+            MEM_writeLE24(op, cBlockHeader24);
+            cSize += ZSTD_blockHeaderSize;
+        }
+
+        remaining -= blockSize;
+        dstCapacity -= cSize;
+        ip += blockSize;
+        op += cSize;
+    }
+
+    if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
+    return op-ostart;
+}
+
+
+static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
+                                    ZSTD_parameters params, U64 pledgedSrcSize, U32 dictID)
+{   BYTE* const op = (BYTE*)dst;
+    U32   const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    U32   const dictIDSizeCode = params.fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
+    U32   const checksumFlag = params.fParams.checksumFlag>0;
+    U32   const windowSize = 1U << params.cParams.windowLog;
+    U32   const singleSegment = params.fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
+    BYTE  const windowLogByte = (BYTE)((params.cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
+    U32   const fcsCode = params.fParams.contentSizeFlag ?
+                     (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) :   /* 0-3 */
+                      0;
+    BYTE  const frameHeaderDecriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
+    size_t pos;
+
+    if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
+    DEBUGLOG(5, "ZSTD_writeFrameHeader : dictIDFlag : %u \n", !params.fParams.noDictIDFlag);
+    DEBUGLOG(5, "ZSTD_writeFrameHeader : dictID : %u \n", dictID);
+    DEBUGLOG(5, "ZSTD_writeFrameHeader : dictIDSizeCode : %u \n", dictIDSizeCode);
+
+    MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
+    op[4] = frameHeaderDecriptionByte; pos=5;
+    if (!singleSegment) op[pos++] = windowLogByte;
+    switch(dictIDSizeCode)
+    {
+        default:   /* impossible */
+        case 0 : break;
+        case 1 : op[pos] = (BYTE)(dictID); pos++; break;
+        case 2 : MEM_writeLE16(op+pos, (U16)dictID); pos+=2; break;
+        case 3 : MEM_writeLE32(op+pos, dictID); pos+=4; break;
+    }
+    switch(fcsCode)
+    {
+        default:   /* impossible */
+        case 0 : if (singleSegment) op[pos++] = (BYTE)(pledgedSrcSize); break;
+        case 1 : MEM_writeLE16(op+pos, (U16)(pledgedSrcSize-256)); pos+=2; break;
+        case 2 : MEM_writeLE32(op+pos, (U32)(pledgedSrcSize)); pos+=4; break;
+        case 3 : MEM_writeLE64(op+pos, (U64)(pledgedSrcSize)); pos+=8; break;
+    }
+    return pos;
+}
+
+
+static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                               U32 frame, U32 lastFrameChunk)
+{
+    const BYTE* const ip = (const BYTE*) src;
+    size_t fhSize = 0;
+
+    if (cctx->stage==ZSTDcs_created) return ERROR(stage_wrong);   /* missing init (ZSTD_compressBegin) */
+
+    if (frame && (cctx->stage==ZSTDcs_init)) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->params, cctx->frameContentSize, cctx->dictID);
+        if (ZSTD_isError(fhSize)) return fhSize;
+        dstCapacity -= fhSize;
+        dst = (char*)dst + fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    /* Check if blocks follow each other */
+    if (src != cctx->nextSrc) {
+        /* not contiguous */
+        ptrdiff_t const delta = cctx->nextSrc - ip;
+        cctx->lowLimit = cctx->dictLimit;
+        cctx->dictLimit = (U32)(cctx->nextSrc - cctx->base);
+        cctx->dictBase = cctx->base;
+        cctx->base -= delta;
+        cctx->nextToUpdate = cctx->dictLimit;
+        if (cctx->dictLimit - cctx->lowLimit < HASH_READ_SIZE) cctx->lowLimit = cctx->dictLimit;   /* too small extDict */
+    }
+
+    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
+    if ((ip+srcSize > cctx->dictBase + cctx->lowLimit) & (ip < cctx->dictBase + cctx->dictLimit)) {
+        ptrdiff_t const highInputIdx = (ip + srcSize) - cctx->dictBase;
+        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)cctx->dictLimit) ? cctx->dictLimit : (U32)highInputIdx;
+        cctx->lowLimit = lowLimitMax;
+    }
+
+    cctx->nextSrc = ip + srcSize;
+
+    if (srcSize) {
+        size_t const cSize = frame ?
+                             ZSTD_compress_generic (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
+                             ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize);
+        if (ZSTD_isError(cSize)) return cSize;
+        cctx->consumedSrcSize += srcSize;
+        return cSize + fhSize;
+    } else
+        return fhSize;
+}
+
+
+size_t ZSTD_compressContinue (ZSTD_CCtx* cctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize)
+{
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
+}
+
+
+size_t ZSTD_getBlockSizeMax(ZSTD_CCtx* cctx)
+{
+    return MIN (ZSTD_BLOCKSIZE_ABSOLUTEMAX, 1 << cctx->params.cParams.windowLog);
+}
+
+size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const blockSizeMax = ZSTD_getBlockSizeMax(cctx);
+    if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
+}
+
+/*! ZSTD_loadDictionaryContent() :
+ *  @return : 0, or an error code
+ */
+static size_t ZSTD_loadDictionaryContent(ZSTD_CCtx* zc, const void* src, size_t srcSize)
+{
+    const BYTE* const ip = (const BYTE*) src;
+    const BYTE* const iend = ip + srcSize;
+
+    /* input becomes current prefix */
+    zc->lowLimit = zc->dictLimit;
+    zc->dictLimit = (U32)(zc->nextSrc - zc->base);
+    zc->dictBase = zc->base;
+    zc->base += ip - zc->nextSrc;
+    zc->nextToUpdate = zc->dictLimit;
+    zc->loadedDictEnd = zc->forceWindow ? 0 : (U32)(iend - zc->base);
+
+    zc->nextSrc = iend;
+    if (srcSize <= HASH_READ_SIZE) return 0;
+
+    switch(zc->params.cParams.strategy)
+    {
+    case ZSTD_fast:
+        ZSTD_fillHashTable (zc, iend, zc->params.cParams.searchLength);
+        break;
+
+    case ZSTD_dfast:
+        ZSTD_fillDoubleHashTable (zc, iend, zc->params.cParams.searchLength);
+        break;
+
+    case ZSTD_greedy:
+    case ZSTD_lazy:
+    case ZSTD_lazy2:
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_insertAndFindFirstIndex(zc, iend-HASH_READ_SIZE, zc->params.cParams.searchLength);
+        break;
+
+    case ZSTD_btlazy2:
+    case ZSTD_btopt:
+    case ZSTD_btopt2:
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_updateTree(zc, iend-HASH_READ_SIZE, iend, 1 << zc->params.cParams.searchLog, zc->params.cParams.searchLength);
+        break;
+
+    default:
+        return ERROR(GENERIC);   /* strategy doesn't exist; impossible */
+    }
+
+    zc->nextToUpdate = (U32)(iend - zc->base);
+    return 0;
+}
+
+
+/* Dictionaries that assign zero probability to symbols that show up causes problems
+   when FSE encoding.  Refuse dictionaries that assign zero probability to symbols
+   that we may encounter during compression.
+   NOTE: This behavior is not standard and could be improved in the future. */
+static size_t ZSTD_checkDictNCount(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue) {
+    U32 s;
+    if (dictMaxSymbolValue < maxSymbolValue) return ERROR(dictionary_corrupted);
+    for (s = 0; s <= maxSymbolValue; ++s) {
+        if (normalizedCounter[s] == 0) return ERROR(dictionary_corrupted);
+    }
+    return 0;
+}
+
+
+/* Dictionary format :
+ * See :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
+ */
+/*! ZSTD_loadZstdDictionary() :
+ * @return : 0, or an error code
+ *  assumptions : magic number supposed already checked
+ *                dictSize supposed > 8
+ */
+static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue = MaxOff;
+    BYTE scratchBuffer[1<<MAX(MLFSELog,LLFSELog)];
+
+    dictPtr += 4;   /* skip magic number */
+    cctx->dictID = cctx->params.fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr);
+    dictPtr += 4;
+
+    {   size_t const hufHeaderSize = HUF_readCTable(cctx->hufCTable, 255, dictPtr, dictEnd-dictPtr);
+        if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
+        dictPtr += hufHeaderSize;
+    }
+
+    {   unsigned offcodeLog;
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
+        CHECK_E( FSE_buildCTable_wksp(cctx->offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, scratchBuffer, sizeof(scratchBuffer)),
+                 dictionary_corrupted);
+        dictPtr += offcodeHeaderSize;
+    }
+
+    {   short matchlengthNCount[MaxML+1];
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
+        size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        /* Every match length code must have non-zero probability */
+        CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
+        CHECK_E( FSE_buildCTable_wksp(cctx->matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, scratchBuffer, sizeof(scratchBuffer)),
+                 dictionary_corrupted);
+        dictPtr += matchlengthHeaderSize;
+    }
+
+    {   short litlengthNCount[MaxLL+1];
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
+        size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
+        if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        /* Every literal length code must have non-zero probability */
+        CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
+        CHECK_E( FSE_buildCTable_wksp(cctx->litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, scratchBuffer, sizeof(scratchBuffer)),
+                 dictionary_corrupted);
+        dictPtr += litlengthHeaderSize;
+    }
+
+    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    cctx->rep[0] = MEM_readLE32(dictPtr+0);
+    cctx->rep[1] = MEM_readLE32(dictPtr+4);
+    cctx->rep[2] = MEM_readLE32(dictPtr+8);
+    dictPtr += 12;
+
+    {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
+        U32 offcodeMax = MaxOff;
+        if (dictContentSize <= ((U32)-1) - 128 KB) {
+            U32 const maxOffset = (U32)dictContentSize + 128 KB; /* The maximum offset that must be supported */
+            offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
+        }
+        /* All offset values <= dictContentSize + 128 KB must be representable */
+        CHECK_F (ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)));
+        /* All repCodes must be <= dictContentSize and != 0*/
+        {   U32 u;
+            for (u=0; u<3; u++) {
+                if (cctx->rep[u] == 0) return ERROR(dictionary_corrupted);
+                if (cctx->rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
+        }   }
+
+        cctx->fseCTables_ready = 1;
+        cctx->hufCTable_repeatMode = HUF_repeat_valid;
+        return ZSTD_loadDictionaryContent(cctx, dictPtr, dictContentSize);
+    }
+}
+
+/** ZSTD_compress_insertDictionary() :
+*   @return : 0, or an error code */
+static size_t ZSTD_compress_insertDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+{
+    if ((dict==NULL) || (dictSize<=8)) return 0;
+
+    /* dict as pure content */
+    if ((MEM_readLE32(dict) != ZSTD_DICT_MAGIC) || (cctx->forceRawDict))
+        return ZSTD_loadDictionaryContent(cctx, dict, dictSize);
+
+    /* dict as zstd dictionary */
+    return ZSTD_loadZstdDictionary(cctx, dict, dictSize);
+}
+
+/*! ZSTD_compressBegin_internal() :
+*   @return : 0, or an error code */
+static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, U64 pledgedSrcSize)
+{
+    ZSTD_compResetPolicy_e const crp = dictSize ? ZSTDcrp_fullReset : ZSTDcrp_continue;
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    CHECK_F(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize, crp));
+    return ZSTD_compress_insertDictionary(cctx, dict, dictSize);
+}
+
+
+/*! ZSTD_compressBegin_advanced() :
+*   @return : 0, or an error code */
+size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
+                             const void* dict, size_t dictSize,
+                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    /* compression parameters verification and optimization */
+    CHECK_F(ZSTD_checkCParams(params.cParams));
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, params, pledgedSrcSize);
+}
+
+
+size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, params, 0);
+}
+
+
+size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
+{
+    return ZSTD_compressBegin_usingDict(cctx, NULL, 0, compressionLevel);
+}
+
+
+/*! ZSTD_writeEpilogue() :
+*   Ends a frame.
+*   @return : nb of bytes written into dst (or an error code) */
+static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* op = ostart;
+    size_t fhSize = 0;
+
+    if (cctx->stage == ZSTDcs_created) return ERROR(stage_wrong);  /* init missing */
+
+    /* special case : empty frame */
+    if (cctx->stage == ZSTDcs_init) {
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->params, 0, 0);
+        if (ZSTD_isError(fhSize)) return fhSize;
+        dstCapacity -= fhSize;
+        op += fhSize;
+        cctx->stage = ZSTDcs_ongoing;
+    }
+
+    if (cctx->stage != ZSTDcs_ending) {
+        /* write one last empty block, make it the "last" block */
+        U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
+        if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        MEM_writeLE32(op, cBlockHeader24);
+        op += ZSTD_blockHeaderSize;
+        dstCapacity -= ZSTD_blockHeaderSize;
+    }
+
+    if (cctx->params.fParams.checksumFlag) {
+        U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
+        if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        MEM_writeLE32(op, checksum);
+        op += 4;
+    }
+
+    cctx->stage = ZSTDcs_created;  /* return to "created but no init" status */
+    return op-ostart;
+}
+
+
+size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
+                         void* dst, size_t dstCapacity,
+                   const void* src, size_t srcSize)
+{
+    size_t endResult;
+    size_t const cSize = ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 1 /* last chunk */);
+    if (ZSTD_isError(cSize)) return cSize;
+    endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
+    if (ZSTD_isError(endResult)) return endResult;
+    if (cctx->params.fParams.contentSizeFlag) {  /* control src size */
+        if (cctx->frameContentSize != cctx->consumedSrcSize) return ERROR(srcSize_wrong);
+    }
+    return cSize + endResult;
+}
+
+
+static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict,size_t dictSize,
+                               ZSTD_parameters params)
+{
+    CHECK_F(ZSTD_compressBegin_internal(cctx, dict, dictSize, params, srcSize));
+    return ZSTD_compressEnd(cctx, dst,  dstCapacity, src, srcSize);
+}
+
+size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict,size_t dictSize,
+                               ZSTD_parameters params)
+{
+    CHECK_F(ZSTD_checkCParams(params.cParams));
+    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+}
+
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize,
+                               const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, dict ? dictSize : 0);
+    params.fParams.contentSizeFlag = 1;
+    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+}
+
+size_t ZSTD_compressCCtx (ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+{
+    return ZSTD_compress_usingDict(ctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
+}
+
+size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+{
+    size_t result;
+    ZSTD_CCtx ctxBody;
+    memset(&ctxBody, 0, sizeof(ctxBody));
+    memcpy(&ctxBody.customMem, &defaultCustomMem, sizeof(ZSTD_customMem));
+    result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
+    ZSTD_free(ctxBody.workSpace, defaultCustomMem);  /* can't free ctxBody itself, as it's on stack; free only heap content */
+    return result;
+}
+
+
+/* =====  Dictionary API  ===== */
+
+struct ZSTD_CDict_s {
+    void* dictBuffer;
+    const void* dictContent;
+    size_t dictContentSize;
+    ZSTD_CCtx* refContext;
+};  /* typedef'd tp ZSTD_CDict within "zstd.h" */
+
+size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support sizeof on NULL */
+    return ZSTD_sizeof_CCtx(cdict->refContext) + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict);
+}
+
+static ZSTD_parameters ZSTD_makeParams(ZSTD_compressionParameters cParams, ZSTD_frameParameters fParams)
+{
+    ZSTD_parameters params;
+    params.cParams = cParams;
+    params.fParams = fParams;
+    return params;
+}
+
+ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize, unsigned byReference,
+                                      ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
+{
+    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
+    if (!customMem.customAlloc || !customMem.customFree) return NULL;
+
+    {   ZSTD_CDict* const cdict = (ZSTD_CDict*) ZSTD_malloc(sizeof(ZSTD_CDict), customMem);
+        ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(customMem);
+
+        if (!cdict || !cctx) {
+            ZSTD_free(cdict, customMem);
+            ZSTD_freeCCtx(cctx);
+            return NULL;
+        }
+
+        if ((byReference) || (!dictBuffer) || (!dictSize)) {
+            cdict->dictBuffer = NULL;
+            cdict->dictContent = dictBuffer;
+        } else {
+            void* const internalBuffer = ZSTD_malloc(dictSize, customMem);
+            if (!internalBuffer) { ZSTD_free(cctx, customMem); ZSTD_free(cdict, customMem); return NULL; }
+            memcpy(internalBuffer, dictBuffer, dictSize);
+            cdict->dictBuffer = internalBuffer;
+            cdict->dictContent = internalBuffer;
+        }
+
+        {   ZSTD_frameParameters const fParams = { 0 /* contentSizeFlag */, 0 /* checksumFlag */, 0 /* noDictIDFlag */ };   /* dummy */
+            ZSTD_parameters const params = ZSTD_makeParams(cParams, fParams);
+            size_t const errorCode = ZSTD_compressBegin_advanced(cctx, cdict->dictContent, dictSize, params, 0);
+            if (ZSTD_isError(errorCode)) {
+                ZSTD_free(cdict->dictBuffer, customMem);
+                ZSTD_free(cdict, customMem);
+                ZSTD_freeCCtx(cctx);
+                return NULL;
+        }   }
+
+        cdict->refContext = cctx;
+        cdict->dictContentSize = dictSize;
+        return cdict;
+    }
+}
+
+ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_createCDict_advanced(dict, dictSize, 0, cParams, allocator);
+}
+
+ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    return ZSTD_createCDict_advanced(dict, dictSize, 1, cParams, allocator);
+}
+
+size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
+{
+    if (cdict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = cdict->refContext->customMem;
+        ZSTD_freeCCtx(cdict->refContext);
+        ZSTD_free(cdict->dictBuffer, cMem);
+        ZSTD_free(cdict, cMem);
+        return 0;
+    }
+}
+
+static ZSTD_parameters ZSTD_getParamsFromCDict(const ZSTD_CDict* cdict) {
+    return ZSTD_getParamsFromCCtx(cdict->refContext);
+}
+
+/* ZSTD_compressBegin_usingCDict_advanced() :
+ * cdict must be != NULL */
+size_t ZSTD_compressBegin_usingCDict_advanced(
+    ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
+    ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
+{
+    if (cdict==NULL) return ERROR(GENERIC);  /* does not support NULL cdict */
+    DEBUGLOG(5, "ZSTD_compressBegin_usingCDict_advanced : dictIDFlag == %u \n", !fParams.noDictIDFlag);
+    if (cdict->dictContentSize)
+        CHECK_F( ZSTD_copyCCtx_internal(cctx, cdict->refContext, fParams, pledgedSrcSize) )
+    else {
+        ZSTD_parameters params = cdict->refContext->params;
+        params.fParams = fParams;
+        CHECK_F(ZSTD_compressBegin_internal(cctx, NULL, 0, params, pledgedSrcSize));
+    }
+    return 0;
+}
+
+/* ZSTD_compressBegin_usingCDict() :
+ * pledgedSrcSize=0 means "unknown"
+ * if pledgedSrcSize>0, it will enable contentSizeFlag */
+size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    DEBUGLOG(5, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u \n", !fParams.noDictIDFlag);
+    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, 0);
+}
+
+size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
+{
+    CHECK_F (ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, srcSize));   /* will check if cdict != NULL */
+    return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
+}
+
+/*! ZSTD_compress_usingCDict() :
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression parameters are decided at CDict creation time
+ *  while frame parameters are hardcoded */
+size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
+                                void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
+    return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, fParams);
+}
+
+
+
+/* ******************************************************************
+*  Streaming
+********************************************************************/
+
+typedef enum { zcss_init, zcss_load, zcss_flush, zcss_final } ZSTD_cStreamStage;
+
+struct ZSTD_CStream_s {
+    ZSTD_CCtx* cctx;
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inToCompress;
+    size_t inBuffPos;
+    size_t inBuffTarget;
+    size_t blockSize;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outBuffContentSize;
+    size_t outBuffFlushedSize;
+    ZSTD_cStreamStage stage;
+    U32    checksum;
+    U32    frameEnded;
+    U64    pledgedSrcSize;
+    ZSTD_parameters params;
+    ZSTD_customMem customMem;
+};   /* typedef'd to ZSTD_CStream within "zstd.h" */
+
+ZSTD_CStream* ZSTD_createCStream(void)
+{
+    return ZSTD_createCStream_advanced(defaultCustomMem);
+}
+
+ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem)
+{
+    ZSTD_CStream* zcs;
+
+    if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
+    if (!customMem.customAlloc || !customMem.customFree) return NULL;
+
+    zcs = (ZSTD_CStream*)ZSTD_malloc(sizeof(ZSTD_CStream), customMem);
+    if (zcs==NULL) return NULL;
+    memset(zcs, 0, sizeof(ZSTD_CStream));
+    memcpy(&zcs->customMem, &customMem, sizeof(ZSTD_customMem));
+    zcs->cctx = ZSTD_createCCtx_advanced(customMem);
+    if (zcs->cctx == NULL) { ZSTD_freeCStream(zcs); return NULL; }
+    return zcs;
+}
+
+size_t ZSTD_freeCStream(ZSTD_CStream* zcs)
+{
+    if (zcs==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = zcs->customMem;
+        ZSTD_freeCCtx(zcs->cctx);
+        zcs->cctx = NULL;
+        ZSTD_freeCDict(zcs->cdictLocal);
+        zcs->cdictLocal = NULL;
+        ZSTD_free(zcs->inBuff, cMem);
+        zcs->inBuff = NULL;
+        ZSTD_free(zcs->outBuff, cMem);
+        zcs->outBuff = NULL;
+        ZSTD_free(zcs, cMem);
+        return 0;
+    }
+}
+
+
+/*======   Initialization   ======*/
+
+size_t ZSTD_CStreamInSize(void)  { return ZSTD_BLOCKSIZE_ABSOLUTEMAX; }
+
+size_t ZSTD_CStreamOutSize(void)
+{
+    return ZSTD_compressBound(ZSTD_BLOCKSIZE_ABSOLUTEMAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
+}
+
+static size_t ZSTD_resetCStream_internal(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
+{
+    if (zcs->inBuffSize==0) return ERROR(stage_wrong);   /* zcs has not been init at least once => can't reset */
+
+    DEBUGLOG(5, "ZSTD_resetCStream_internal : dictIDFlag == %u \n", !zcs->params.fParams.noDictIDFlag);
+
+    if (zcs->cdict) CHECK_F(ZSTD_compressBegin_usingCDict_advanced(zcs->cctx, zcs->cdict, zcs->params.fParams, pledgedSrcSize))
+    else CHECK_F(ZSTD_compressBegin_internal(zcs->cctx, NULL, 0, zcs->params, pledgedSrcSize));
+
+    zcs->inToCompress = 0;
+    zcs->inBuffPos = 0;
+    zcs->inBuffTarget = zcs->blockSize;
+    zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
+    zcs->stage = zcss_load;
+    zcs->frameEnded = 0;
+    zcs->pledgedSrcSize = pledgedSrcSize;
+    return 0;   /* ready to go */
+}
+
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
+{
+
+    zcs->params.fParams.contentSizeFlag = (pledgedSrcSize > 0);
+    DEBUGLOG(5, "ZSTD_resetCStream : dictIDFlag == %u \n", !zcs->params.fParams.noDictIDFlag);
+    return ZSTD_resetCStream_internal(zcs, pledgedSrcSize);
+}
+
+/* ZSTD_initCStream_internal() :
+ * params are supposed validated at this stage
+ * and zcs->cdict is supposed to be correct */
+static size_t ZSTD_initCStream_stage2(ZSTD_CStream* zcs,
+                                const ZSTD_parameters params,
+                                unsigned long long pledgedSrcSize)
+{
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+
+    /* allocate buffers */
+    {   size_t const neededInBuffSize = (size_t)1 << params.cParams.windowLog;
+        if (zcs->inBuffSize < neededInBuffSize) {
+            zcs->inBuffSize = 0;
+            ZSTD_free(zcs->inBuff, zcs->customMem);
+            zcs->inBuff = (char*) ZSTD_malloc(neededInBuffSize, zcs->customMem);
+            if (zcs->inBuff == NULL) return ERROR(memory_allocation);
+            zcs->inBuffSize = neededInBuffSize;
+        }
+        zcs->blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, neededInBuffSize);
+    }
+    if (zcs->outBuffSize < ZSTD_compressBound(zcs->blockSize)+1) {
+        size_t const outBuffSize = ZSTD_compressBound(zcs->blockSize)+1;
+        zcs->outBuffSize = 0;
+        ZSTD_free(zcs->outBuff, zcs->customMem);
+        zcs->outBuff = (char*) ZSTD_malloc(outBuffSize, zcs->customMem);
+        if (zcs->outBuff == NULL) return ERROR(memory_allocation);
+        zcs->outBuffSize = outBuffSize;
+    }
+
+    zcs->checksum = params.fParams.checksumFlag > 0;
+    zcs->params = params;
+
+    DEBUGLOG(5, "ZSTD_initCStream_stage2 : dictIDFlag == %u \n", !params.fParams.noDictIDFlag);
+    return ZSTD_resetCStream_internal(zcs, pledgedSrcSize);
+}
+
+/* ZSTD_initCStream_usingCDict_advanced() :
+ * same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
+size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs, const ZSTD_CDict* cdict, unsigned long long pledgedSrcSize, ZSTD_frameParameters fParams)
+{
+    if (!cdict) return ERROR(GENERIC);   /* cannot handle NULL cdict (does not know what to do) */
+    {   ZSTD_parameters params = ZSTD_getParamsFromCDict(cdict);
+        params.fParams = fParams;
+        zcs->cdict = cdict;
+        return ZSTD_initCStream_stage2(zcs, params, pledgedSrcSize);
+    }
+}
+
+/* note : cdict must outlive compression session */
+size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
+{
+    ZSTD_frameParameters const fParams = { 0 /* content */, 0 /* checksum */, 0 /* noDictID */ };
+    return ZSTD_initCStream_usingCDict_advanced(zcs, cdict, 0, fParams);
+}
+
+static size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                                const void* dict, size_t dictSize,
+                                ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    zcs->cdict = NULL;
+
+    if (dict && dictSize >= 8) {
+        ZSTD_freeCDict(zcs->cdictLocal);
+        zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize, 0 /* copy */, params.cParams, zcs->customMem);
+        if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
+        zcs->cdict = zcs->cdictLocal;
+    }
+
+    DEBUGLOG(5, "ZSTD_initCStream_internal : dictIDFlag == %u \n", !params.fParams.noDictIDFlag);
+    return ZSTD_initCStream_stage2(zcs, params, pledgedSrcSize);
+}
+
+size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
+                                 const void* dict, size_t dictSize,
+                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    DEBUGLOG(5, "ZSTD_initCStream_advanced : dictIDFlag == %u \n", !params.fParams.noDictIDFlag);
+    return ZSTD_initCStream_internal(zcs, dict, dictSize, params, pledgedSrcSize);
+}
+
+size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
+{
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
+    return ZSTD_initCStream_internal(zcs, dict, dictSize, params, 0);
+}
+
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize)
+{
+    ZSTD_parameters params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
+    params.fParams.contentSizeFlag = (pledgedSrcSize>0);
+    return ZSTD_initCStream_internal(zcs, NULL, 0, params, pledgedSrcSize);
+}
+
+size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
+{
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, 0);
+    return ZSTD_initCStream_internal(zcs, NULL, 0, params, 0);
+}
+
+size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
+{
+    if (zcs==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*zcs) + ZSTD_sizeof_CCtx(zcs->cctx) + ZSTD_sizeof_CDict(zcs->cdictLocal) + zcs->outBuffSize + zcs->inBuffSize;
+}
+
+/*======   Compression   ======*/
+
+typedef enum { zsf_gather, zsf_flush, zsf_end } ZSTD_flush_e;
+
+MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    size_t const length = MIN(dstCapacity, srcSize);
+    memcpy(dst, src, length);
+    return length;
+}
+
+static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                              void* dst, size_t* dstCapacityPtr,
+                        const void* src, size_t* srcSizePtr,
+                              ZSTD_flush_e const flush)
+{
+    U32 someMoreWork = 1;
+    const char* const istart = (const char*)src;
+    const char* const iend = istart + *srcSizePtr;
+    const char* ip = istart;
+    char* const ostart = (char*)dst;
+    char* const oend = ostart + *dstCapacityPtr;
+    char* op = ostart;
+
+    while (someMoreWork) {
+        switch(zcs->stage)
+        {
+        case zcss_init: return ERROR(init_missing);   /* call ZBUFF_compressInit() first ! */
+
+        case zcss_load:
+            /* complete inBuffer */
+            {   size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
+                size_t const loaded = ZSTD_limitCopy(zcs->inBuff + zcs->inBuffPos, toLoad, ip, iend-ip);
+                zcs->inBuffPos += loaded;
+                ip += loaded;
+                if ( (zcs->inBuffPos==zcs->inToCompress) || (!flush && (toLoad != loaded)) ) {
+                    someMoreWork = 0; break;  /* not enough input to get a full block : stop there, wait for more */
+            }   }
+            /* compress current block (note : this stage cannot be stopped in the middle) */
+            {   void* cDst;
+                size_t cSize;
+                size_t const iSize = zcs->inBuffPos - zcs->inToCompress;
+                size_t oSize = oend-op;
+                if (oSize >= ZSTD_compressBound(iSize))
+                    cDst = op;   /* compress directly into output buffer (avoid flush stage) */
+                else
+                    cDst = zcs->outBuff, oSize = zcs->outBuffSize;
+                cSize = (flush == zsf_end) ?
+                        ZSTD_compressEnd(zcs->cctx, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize) :
+                        ZSTD_compressContinue(zcs->cctx, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize);
+                if (ZSTD_isError(cSize)) return cSize;
+                if (flush == zsf_end) zcs->frameEnded = 1;
+                /* prepare next block */
+                zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
+                if (zcs->inBuffTarget > zcs->inBuffSize)
+                    zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;   /* note : inBuffSize >= blockSize */
+                zcs->inToCompress = zcs->inBuffPos;
+                if (cDst == op) { op += cSize; break; }   /* no need to flush */
+                zcs->outBuffContentSize = cSize;
+                zcs->outBuffFlushedSize = 0;
+                zcs->stage = zcss_flush;   /* pass-through to flush stage */
+            }
+
+        case zcss_flush:
+            {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
+                size_t const flushed = ZSTD_limitCopy(op, oend-op, zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                op += flushed;
+                zcs->outBuffFlushedSize += flushed;
+                if (toFlush!=flushed) { someMoreWork = 0; break; }  /* dst too small to store flushed data : stop there */
+                zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
+                zcs->stage = zcss_load;
+                break;
+            }
+
+        case zcss_final:
+            someMoreWork = 0;   /* do nothing */
+            break;
+
+        default:
+            return ERROR(GENERIC);   /* impossible */
+        }
+    }
+
+    *srcSizePtr = ip - istart;
+    *dstCapacityPtr = op - ostart;
+    if (zcs->frameEnded) return 0;
+    {   size_t hintInSize = zcs->inBuffTarget - zcs->inBuffPos;
+        if (hintInSize==0) hintInSize = zcs->blockSize;
+        return hintInSize;
+    }
+}
+
+size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    size_t sizeRead = input->size - input->pos;
+    size_t sizeWritten = output->size - output->pos;
+    size_t const result = ZSTD_compressStream_generic(zcs,
+                                                      (char*)(output->dst) + output->pos, &sizeWritten,
+                                                      (const char*)(input->src) + input->pos, &sizeRead, zsf_gather);
+    input->pos += sizeRead;
+    output->pos += sizeWritten;
+    return result;
+}
+
+
+/*======   Finalize   ======*/
+
+/*! ZSTD_flushStream() :
+*   @return : amount of data remaining to flush */
+size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    size_t srcSize = 0;
+    size_t sizeWritten = output->size - output->pos;
+    size_t const result = ZSTD_compressStream_generic(zcs,
+                                                     (char*)(output->dst) + output->pos, &sizeWritten,
+                                                     &srcSize, &srcSize, /* use a valid src address instead of NULL */
+                                                      zsf_flush);
+    output->pos += sizeWritten;
+    if (ZSTD_isError(result)) return result;
+    return zcs->outBuffContentSize - zcs->outBuffFlushedSize;   /* remaining to flush */
+}
+
+
+size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
+{
+    BYTE* const ostart = (BYTE*)(output->dst) + output->pos;
+    BYTE* const oend = (BYTE*)(output->dst) + output->size;
+    BYTE* op = ostart;
+
+    if (zcs->stage != zcss_final) {
+        /* flush whatever remains */
+        size_t srcSize = 0;
+        size_t sizeWritten = output->size - output->pos;
+        size_t const notEnded = ZSTD_compressStream_generic(zcs, ostart, &sizeWritten,
+                                     &srcSize /* use a valid src address instead of NULL */, &srcSize, zsf_end);
+        size_t const remainingToFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
+        op += sizeWritten;
+        if (remainingToFlush) {
+            output->pos += sizeWritten;
+            return remainingToFlush + ZSTD_BLOCKHEADERSIZE /* final empty block */ + (zcs->checksum * 4);
+        }
+        /* create epilogue */
+        zcs->stage = zcss_final;
+        zcs->outBuffContentSize = !notEnded ? 0 :
+            /* write epilogue, including final empty block, into outBuff */
+            ZSTD_compressEnd(zcs->cctx, zcs->outBuff, zcs->outBuffSize, NULL, 0);
+        if (ZSTD_isError(zcs->outBuffContentSize)) return zcs->outBuffContentSize;
+    }
+
+    /* flush epilogue */
+    {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
+        size_t const flushed = ZSTD_limitCopy(op, oend-op, zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+        op += flushed;
+        zcs->outBuffFlushedSize += flushed;
+        output->pos += op-ostart;
+        if (toFlush==flushed) zcs->stage = zcss_init;  /* end reached */
+        return toFlush - flushed;
+    }
+}
+
+
+
+/*-=====  Pre-defined compression levels  =====-*/
+
+#define ZSTD_DEFAULT_CLEVEL 1
+#define ZSTD_MAX_CLEVEL     22
+int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
+
+static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
+{   /* "default" */
+    /* W,  C,  H,  S,  L, TL, strat */
+    { 18, 12, 12,  1,  7, 16, ZSTD_fast    },  /* level  0 - never used */
+    { 19, 13, 14,  1,  7, 16, ZSTD_fast    },  /* level  1 */
+    { 19, 15, 16,  1,  6, 16, ZSTD_fast    },  /* level  2 */
+    { 20, 16, 17,  1,  5, 16, ZSTD_dfast   },  /* level  3.*/
+    { 20, 18, 18,  1,  5, 16, ZSTD_dfast   },  /* level  4.*/
+    { 20, 15, 18,  3,  5, 16, ZSTD_greedy  },  /* level  5 */
+    { 21, 16, 19,  2,  5, 16, ZSTD_lazy    },  /* level  6 */
+    { 21, 17, 20,  3,  5, 16, ZSTD_lazy    },  /* level  7 */
+    { 21, 18, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
+    { 21, 20, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 21, 19, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
+    { 22, 20, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
+    { 22, 20, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
+    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 13 */
+    { 22, 21, 22,  6,  5, 16, ZSTD_lazy2   },  /* level 14 */
+    { 22, 21, 21,  5,  5, 16, ZSTD_btlazy2 },  /* level 15 */
+    { 23, 22, 22,  5,  5, 16, ZSTD_btlazy2 },  /* level 16 */
+    { 23, 21, 22,  4,  5, 24, ZSTD_btopt   },  /* level 17 */
+    { 23, 22, 22,  5,  4, 32, ZSTD_btopt   },  /* level 18 */
+    { 23, 23, 22,  6,  3, 48, ZSTD_btopt   },  /* level 19 */
+    { 25, 25, 23,  7,  3, 64, ZSTD_btopt2  },  /* level 20 */
+    { 26, 26, 23,  7,  3,256, ZSTD_btopt2  },  /* level 21 */
+    { 27, 27, 25,  9,  3,512, ZSTD_btopt2  },  /* level 22 */
+},
+{   /* for srcSize <= 256 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    {  0,  0,  0,  0,  0,  0, ZSTD_fast    },  /* level  0 - not used */
+    { 18, 13, 14,  1,  6,  8, ZSTD_fast    },  /* level  1 */
+    { 18, 14, 13,  1,  5,  8, ZSTD_dfast   },  /* level  2 */
+    { 18, 16, 15,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
+    { 18, 15, 17,  1,  5,  8, ZSTD_greedy  },  /* level  4.*/
+    { 18, 16, 17,  4,  5,  8, ZSTD_greedy  },  /* level  5.*/
+    { 18, 16, 17,  3,  5,  8, ZSTD_lazy    },  /* level  6.*/
+    { 18, 17, 17,  4,  4,  8, ZSTD_lazy    },  /* level  7 */
+    { 18, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 18, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 18, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 18, 18, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 11.*/
+    { 18, 18, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 12.*/
+    { 18, 19, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13 */
+    { 18, 18, 18,  4,  4, 16, ZSTD_btopt   },  /* level 14.*/
+    { 18, 18, 18,  4,  3, 16, ZSTD_btopt   },  /* level 15.*/
+    { 18, 19, 18,  6,  3, 32, ZSTD_btopt   },  /* level 16.*/
+    { 18, 19, 18,  8,  3, 64, ZSTD_btopt   },  /* level 17.*/
+    { 18, 19, 18,  9,  3,128, ZSTD_btopt   },  /* level 18.*/
+    { 18, 19, 18, 10,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 18, 19, 18, 11,  3,512, ZSTD_btopt2  },  /* level 20.*/
+    { 18, 19, 18, 12,  3,512, ZSTD_btopt2  },  /* level 21.*/
+    { 18, 19, 18, 13,  3,512, ZSTD_btopt2  },  /* level 22.*/
+},
+{   /* for srcSize <= 128 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 17, 12, 12,  1,  7,  8, ZSTD_fast    },  /* level  0 - not used */
+    { 17, 12, 13,  1,  6,  8, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 16,  1,  5,  8, ZSTD_fast    },  /* level  2 */
+    { 17, 16, 16,  2,  5,  8, ZSTD_dfast   },  /* level  3 */
+    { 17, 13, 15,  3,  4,  8, ZSTD_greedy  },  /* level  4 */
+    { 17, 15, 17,  4,  4,  8, ZSTD_greedy  },  /* level  5 */
+    { 17, 16, 17,  3,  4,  8, ZSTD_lazy    },  /* level  6 */
+    { 17, 15, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 17, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 17, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 17, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 17, 17, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 11 */
+    { 17, 17, 17,  8,  4,  8, ZSTD_lazy2   },  /* level 12 */
+    { 17, 18, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13.*/
+    { 17, 17, 17,  7,  3,  8, ZSTD_btopt   },  /* level 14.*/
+    { 17, 17, 17,  7,  3, 16, ZSTD_btopt   },  /* level 15.*/
+    { 17, 18, 17,  7,  3, 32, ZSTD_btopt   },  /* level 16.*/
+    { 17, 18, 17,  7,  3, 64, ZSTD_btopt   },  /* level 17.*/
+    { 17, 18, 17,  7,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 17, 18, 17,  9,  3,256, ZSTD_btopt2  },  /* level 20.*/
+    { 17, 18, 17, 10,  3,256, ZSTD_btopt2  },  /* level 21.*/
+    { 17, 18, 17, 11,  3,512, ZSTD_btopt2  },  /* level 22.*/
+},
+{   /* for srcSize <= 16 KB */
+    /* W,  C,  H,  S,  L,  T, strat */
+    { 14, 12, 12,  1,  7,  6, ZSTD_fast    },  /* level  0 - not used */
+    { 14, 14, 14,  1,  6,  6, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 14,  1,  4,  6, ZSTD_fast    },  /* level  2 */
+    { 14, 14, 14,  1,  4,  6, ZSTD_dfast   },  /* level  3.*/
+    { 14, 14, 14,  4,  4,  6, ZSTD_greedy  },  /* level  4.*/
+    { 14, 14, 14,  3,  4,  6, ZSTD_lazy    },  /* level  5.*/
+    { 14, 14, 14,  4,  4,  6, ZSTD_lazy2   },  /* level  6 */
+    { 14, 14, 14,  5,  4,  6, ZSTD_lazy2   },  /* level  7 */
+    { 14, 14, 14,  6,  4,  6, ZSTD_lazy2   },  /* level  8.*/
+    { 14, 15, 14,  6,  4,  6, ZSTD_btlazy2 },  /* level  9.*/
+    { 14, 15, 14,  3,  3,  6, ZSTD_btopt   },  /* level 10.*/
+    { 14, 15, 14,  6,  3,  8, ZSTD_btopt   },  /* level 11.*/
+    { 14, 15, 14,  6,  3, 16, ZSTD_btopt   },  /* level 12.*/
+    { 14, 15, 14,  6,  3, 24, ZSTD_btopt   },  /* level 13.*/
+    { 14, 15, 15,  6,  3, 48, ZSTD_btopt   },  /* level 14.*/
+    { 14, 15, 15,  6,  3, 64, ZSTD_btopt   },  /* level 15.*/
+    { 14, 15, 15,  6,  3, 96, ZSTD_btopt   },  /* level 16.*/
+    { 14, 15, 15,  6,  3,128, ZSTD_btopt   },  /* level 17.*/
+    { 14, 15, 15,  6,  3,256, ZSTD_btopt   },  /* level 18.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btopt   },  /* level 19.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btopt2  },  /* level 20.*/
+    { 14, 15, 15,  9,  3,256, ZSTD_btopt2  },  /* level 21.*/
+    { 14, 15, 15, 10,  3,256, ZSTD_btopt2  },  /* level 22.*/
+},
+};
+
+/*! ZSTD_getCParams() :
+*   @return ZSTD_compressionParameters structure for a selected compression level, `srcSize` and `dictSize`.
+*   Size values are optional, provide 0 if not known or unused */
+ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSize, size_t dictSize)
+{
+    ZSTD_compressionParameters cp;
+    size_t const addedSize = srcSize ? 0 : 500;
+    U64 const rSize = srcSize+dictSize ? srcSize+dictSize+addedSize : (U64)-1;
+    U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);   /* intentional underflow for srcSizeHint == 0 */
+    if (compressionLevel <= 0) compressionLevel = ZSTD_DEFAULT_CLEVEL;   /* 0 == default; no negative compressionLevel yet */
+    if (compressionLevel > ZSTD_MAX_CLEVEL) compressionLevel = ZSTD_MAX_CLEVEL;
+    cp = ZSTD_defaultCParameters[tableID][compressionLevel];
+    if (MEM_32bits()) {   /* auto-correction, for 32-bits mode */
+        if (cp.windowLog > ZSTD_WINDOWLOG_MAX) cp.windowLog = ZSTD_WINDOWLOG_MAX;
+        if (cp.chainLog > ZSTD_CHAINLOG_MAX) cp.chainLog = ZSTD_CHAINLOG_MAX;
+        if (cp.hashLog > ZSTD_HASHLOG_MAX) cp.hashLog = ZSTD_HASHLOG_MAX;
+    }
+    cp = ZSTD_adjustCParams(cp, srcSize, dictSize);
+    return cp;
+}
+
+/*! ZSTD_getParams() :
+*   same as ZSTD_getCParams(), but @return a `ZSTD_parameters` object (instead of `ZSTD_compressionParameters`).
+*   All fields of `ZSTD_frameParameters` are set to default (0) */
+ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSize, size_t dictSize) {
+    ZSTD_parameters params;
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, srcSize, dictSize);
+    memset(&params, 0, sizeof(params));
+    params.cParams = cParams;
+    return params;
+}
diff --git a/thirdparty/zstd/compress/zstd_opt.h b/thirdparty/zstd/compress/zstd_opt.h
new file mode 100644
index 0000000000..5437611912
--- /dev/null
+++ b/thirdparty/zstd/compress/zstd_opt.h
@@ -0,0 +1,921 @@
+/**
+ * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree. An additional grant
+ * of patent rights can be found in the PATENTS file in the same directory.
+ */
+
+
+/* Note : this file is intended to be included within zstd_compress.c */
+
+
+#ifndef ZSTD_OPT_H_91842398743
+#define ZSTD_OPT_H_91842398743
+
+
+#define ZSTD_LITFREQ_ADD    2
+#define ZSTD_FREQ_DIV       4
+#define ZSTD_MAX_PRICE      (1<<30)
+
+/*-*************************************
+*  Price functions for optimal parser
+***************************************/
+FORCE_INLINE void ZSTD_setLog2Prices(seqStore_t* ssPtr)
+{
+    ssPtr->log2matchLengthSum = ZSTD_highbit32(ssPtr->matchLengthSum+1);
+    ssPtr->log2litLengthSum = ZSTD_highbit32(ssPtr->litLengthSum+1);
+    ssPtr->log2litSum = ZSTD_highbit32(ssPtr->litSum+1);
+    ssPtr->log2offCodeSum = ZSTD_highbit32(ssPtr->offCodeSum+1);
+    ssPtr->factor = 1 + ((ssPtr->litSum>>5) / ssPtr->litLengthSum) + ((ssPtr->litSum<<1) / (ssPtr->litSum + ssPtr->matchSum));
+}
+
+
+MEM_STATIC void ZSTD_rescaleFreqs(seqStore_t* ssPtr, const BYTE* src, size_t srcSize)
+{
+    unsigned u;
+
+    ssPtr->cachedLiterals = NULL;
+    ssPtr->cachedPrice = ssPtr->cachedLitLength = 0;
+    ssPtr->staticPrices = 0;
+
+    if (ssPtr->litLengthSum == 0) {
+        if (srcSize <= 1024) ssPtr->staticPrices = 1;
+
+        for (u=0; u<=MaxLit; u++)
+            ssPtr->litFreq[u] = 0;
+        for (u=0; u<srcSize; u++)
+            ssPtr->litFreq[src[u]]++;
+
+        ssPtr->litSum = 0;
+        ssPtr->litLengthSum = MaxLL+1;
+        ssPtr->matchLengthSum = MaxML+1;
+        ssPtr->offCodeSum = (MaxOff+1);
+        ssPtr->matchSum = (ZSTD_LITFREQ_ADD<<Litbits);
+
+        for (u=0; u<=MaxLit; u++) {
+            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>ZSTD_FREQ_DIV);
+            ssPtr->litSum += ssPtr->litFreq[u];
+        }
+        for (u=0; u<=MaxLL; u++)
+            ssPtr->litLengthFreq[u] = 1;
+        for (u=0; u<=MaxML; u++)
+            ssPtr->matchLengthFreq[u] = 1;
+        for (u=0; u<=MaxOff; u++)
+            ssPtr->offCodeFreq[u] = 1;
+    } else {
+        ssPtr->matchLengthSum = 0;
+        ssPtr->litLengthSum = 0;
+        ssPtr->offCodeSum = 0;
+        ssPtr->matchSum = 0;
+        ssPtr->litSum = 0;
+
+        for (u=0; u<=MaxLit; u++) {
+            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>(ZSTD_FREQ_DIV+1));
+            ssPtr->litSum += ssPtr->litFreq[u];
+        }
+        for (u=0; u<=MaxLL; u++) {
+            ssPtr->litLengthFreq[u] = 1 + (ssPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
+            ssPtr->litLengthSum += ssPtr->litLengthFreq[u];
+        }
+        for (u=0; u<=MaxML; u++) {
+            ssPtr->matchLengthFreq[u] = 1 + (ssPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
+            ssPtr->matchLengthSum += ssPtr->matchLengthFreq[u];
+            ssPtr->matchSum += ssPtr->matchLengthFreq[u] * (u + 3);
+        }
+        ssPtr->matchSum *= ZSTD_LITFREQ_ADD;
+        for (u=0; u<=MaxOff; u++) {
+            ssPtr->offCodeFreq[u] = 1 + (ssPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
+            ssPtr->offCodeSum += ssPtr->offCodeFreq[u];
+        }
+    }
+
+    ZSTD_setLog2Prices(ssPtr);
+}
+
+
+FORCE_INLINE U32 ZSTD_getLiteralPrice(seqStore_t* ssPtr, U32 litLength, const BYTE* literals)
+{
+    U32 price, u;
+
+    if (ssPtr->staticPrices)
+        return ZSTD_highbit32((U32)litLength+1) + (litLength*6);
+
+    if (litLength == 0)
+        return ssPtr->log2litLengthSum - ZSTD_highbit32(ssPtr->litLengthFreq[0]+1);
+
+    /* literals */
+    if (ssPtr->cachedLiterals == literals) {
+        U32 const additional = litLength - ssPtr->cachedLitLength;
+        const BYTE* literals2 = ssPtr->cachedLiterals + ssPtr->cachedLitLength;
+        price = ssPtr->cachedPrice + additional * ssPtr->log2litSum;
+        for (u=0; u < additional; u++)
+            price -= ZSTD_highbit32(ssPtr->litFreq[literals2[u]]+1);
+        ssPtr->cachedPrice = price;
+        ssPtr->cachedLitLength = litLength;
+    } else {
+        price = litLength * ssPtr->log2litSum;
+        for (u=0; u < litLength; u++)
+            price -= ZSTD_highbit32(ssPtr->litFreq[literals[u]]+1);
+
+        if (litLength >= 12) {
+            ssPtr->cachedLiterals = literals;
+            ssPtr->cachedPrice = price;
+            ssPtr->cachedLitLength = litLength;
+        }
+    }
+
+    /* literal Length */
+    {   const BYTE LL_deltaCode = 19;
+        const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+        price += LL_bits[llCode] + ssPtr->log2litLengthSum - ZSTD_highbit32(ssPtr->litLengthFreq[llCode]+1);
+    }
+
+    return price;
+}
+
+
+FORCE_INLINE U32 ZSTD_getPrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength, const int ultra)
+{
+    /* offset */
+    U32 price;
+    BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
+
+    if (seqStorePtr->staticPrices)
+        return ZSTD_getLiteralPrice(seqStorePtr, litLength, literals) + ZSTD_highbit32((U32)matchLength+1) + 16 + offCode;
+
+    price = offCode + seqStorePtr->log2offCodeSum - ZSTD_highbit32(seqStorePtr->offCodeFreq[offCode]+1);
+    if (!ultra && offCode >= 20) price += (offCode-19)*2;
+
+    /* match Length */
+    {   const BYTE ML_deltaCode = 36;
+        const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
+        price += ML_bits[mlCode] + seqStorePtr->log2matchLengthSum - ZSTD_highbit32(seqStorePtr->matchLengthFreq[mlCode]+1);
+    }
+
+    return price + ZSTD_getLiteralPrice(seqStorePtr, litLength, literals) + seqStorePtr->factor;
+}
+
+
+MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength)
+{
+    U32 u;
+
+    /* literals */
+    seqStorePtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    for (u=0; u < litLength; u++)
+        seqStorePtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+
+    /* literal Length */
+    {   const BYTE LL_deltaCode = 19;
+        const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+        seqStorePtr->litLengthFreq[llCode]++;
+        seqStorePtr->litLengthSum++;
+    }
+
+    /* match offset */
+    {   BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
+        seqStorePtr->offCodeSum++;
+        seqStorePtr->offCodeFreq[offCode]++;
+    }
+
+    /* match Length */
+    {   const BYTE ML_deltaCode = 36;
+        const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
+        seqStorePtr->matchLengthFreq[mlCode]++;
+        seqStorePtr->matchLengthSum++;
+    }
+
+    ZSTD_setLog2Prices(seqStorePtr);
+}
+
+
+#define SET_PRICE(pos, mlen_, offset_, litlen_, price_)   \
+    {                                                 \
+        while (last_pos < pos)  { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; } \
+        opt[pos].mlen = mlen_;                         \
+        opt[pos].off = offset_;                        \
+        opt[pos].litlen = litlen_;                     \
+        opt[pos].price = price_;                       \
+    }
+
+
+
+/* Update hashTable3 up to ip (excluded)
+   Assumption : always within prefix (i.e. not within extDict) */
+FORCE_INLINE
+U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_CCtx* zc, const BYTE* ip)
+{
+    U32* const hashTable3  = zc->hashTable3;
+    U32 const hashLog3  = zc->hashLog3;
+    const BYTE* const base = zc->base;
+    U32 idx = zc->nextToUpdate3;
+    const U32 target = zc->nextToUpdate3 = (U32)(ip - base);
+    const size_t hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+
+    while(idx < target) {
+        hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
+        idx++;
+    }
+
+    return hashTable3[hash3];
+}
+
+
+/*-*************************************
+*  Binary Tree search
+***************************************/
+static U32 ZSTD_insertBtAndGetAllMatches (
+                        ZSTD_CCtx* zc,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        U32 nbCompares, const U32 mls,
+                        U32 extDict, ZSTD_match_t* matches, const U32 minMatchLen)
+{
+    const BYTE* const base = zc->base;
+    const U32 current = (U32)(ip-base);
+    const U32 hashLog = zc->params.cParams.hashLog;
+    const size_t h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32* const hashTable = zc->hashTable;
+    U32 matchIndex  = hashTable[h];
+    U32* const bt   = zc->chainTable;
+    const U32 btLog = zc->params.cParams.chainLog - 1;
+    const U32 btMask= (1U << btLog) - 1;
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const dictBase = zc->dictBase;
+    const U32 dictLimit = zc->dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const U32 btLow = btMask >= current ? 0 : current - btMask;
+    const U32 windowLow = zc->lowLimit;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = bt + 2*(current&btMask) + 1;
+    U32 matchEndIdx = current+8;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 mnum = 0;
+
+    const U32 minMatch = (mls == 3) ? 3 : 4;
+    size_t bestLength = minMatchLen-1;
+
+    if (minMatch == 3) { /* HC3 match finder */
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3 (zc, ip);
+        if (matchIndex3>windowLow && (current - matchIndex3 < (1<<18))) {
+            const BYTE* match;
+            size_t currentMl=0;
+            if ((!extDict) || matchIndex3 >= dictLimit) {
+                match = base + matchIndex3;
+                if (match[bestLength] == ip[bestLength]) currentMl = ZSTD_count(ip, match, iLimit);
+            } else {
+                match = dictBase + matchIndex3;
+                if (MEM_readMINMATCH(match, MINMATCH) == MEM_readMINMATCH(ip, MINMATCH))    /* assumption : matchIndex3 <= dictLimit-4 (by table construction) */
+                    currentMl = ZSTD_count_2segments(ip+MINMATCH, match+MINMATCH, iLimit, dictEnd, prefixStart) + MINMATCH;
+            }
+
+            /* save best solution */
+            if (currentMl > bestLength) {
+                bestLength = currentMl;
+                matches[mnum].off = ZSTD_REP_MOVE_OPT + current - matchIndex3;
+                matches[mnum].len = (U32)currentMl;
+                mnum++;
+                if (currentMl > ZSTD_OPT_NUM) goto update;
+                if (ip+currentMl == iLimit) goto update; /* best possible, and avoid read overflow*/
+            }
+        }
+    }
+
+    hashTable[h] = current;   /* Update Hash Table */
+
+    while (nbCompares-- && (matchIndex > windowLow)) {
+        U32* nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        const BYTE* match;
+
+        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+            match = base + matchIndex;
+            if (match[matchLength] == ip[matchLength]) {
+                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iLimit) +1;
+            }
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            if (matchLength > matchEndIdx - matchIndex) matchEndIdx = matchIndex + (U32)matchLength;
+            bestLength = matchLength;
+            matches[mnum].off = ZSTD_REP_MOVE_OPT + current - matchIndex;
+            matches[mnum].len = (U32)matchLength;
+            mnum++;
+            if (matchLength > ZSTD_OPT_NUM) break;
+            if (ip+matchLength == iLimit)   /* equal : no way to know if inf or sup */
+                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+
+update:
+    zc->nextToUpdate = (matchEndIdx > current + 8) ? matchEndIdx - 8 : current+1;
+    return mnum;
+}
+
+
+/** Tree updater, providing best match */
+static U32 ZSTD_BtGetAllMatches (
+                        ZSTD_CCtx* zc,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, const U32 minMatchLen)
+{
+    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls);
+    return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 0, matches, minMatchLen);
+}
+
+
+static U32 ZSTD_BtGetAllMatches_selectMLS (
+                        ZSTD_CCtx* zc,   /* Index table will be updated */
+                        const BYTE* ip, const BYTE* const iHighLimit,
+                        const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, const U32 minMatchLen)
+{
+    switch(matchLengthSearch)
+    {
+    case 3 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 3, matches, minMatchLen);
+    default :
+    case 4 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minMatchLen);
+    case 5 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minMatchLen);
+    case 7 :
+    case 6 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minMatchLen);
+    }
+}
+
+/** Tree updater, providing best match */
+static U32 ZSTD_BtGetAllMatches_extDict (
+                        ZSTD_CCtx* zc,
+                        const BYTE* const ip, const BYTE* const iLimit,
+                        const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, const U32 minMatchLen)
+{
+    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls);
+    return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 1, matches, minMatchLen);
+}
+
+
+static U32 ZSTD_BtGetAllMatches_selectMLS_extDict (
+                        ZSTD_CCtx* zc,   /* Index table will be updated */
+                        const BYTE* ip, const BYTE* const iHighLimit,
+                        const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, const U32 minMatchLen)
+{
+    switch(matchLengthSearch)
+    {
+    case 3 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 3, matches, minMatchLen);
+    default :
+    case 4 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minMatchLen);
+    case 5 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minMatchLen);
+    case 7 :
+    case 6 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minMatchLen);
+    }
+}
+
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+FORCE_INLINE
+void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
+                                    const void* src, size_t srcSize, const int ultra)
+{
+    seqStore_t* seqStorePtr = &(ctx->seqStore);
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ctx->base;
+    const BYTE* const prefixStart = base + ctx->dictLimit;
+
+    const U32 maxSearches = 1U << ctx->params.cParams.searchLog;
+    const U32 sufficient_len = ctx->params.cParams.targetLength;
+    const U32 mls = ctx->params.cParams.searchLength;
+    const U32 minMatch = (ctx->params.cParams.searchLength == 3) ? 3 : 4;
+
+    ZSTD_optimal_t* opt = seqStorePtr->priceTable;
+    ZSTD_match_t* matches = seqStorePtr->matchTable;
+    const BYTE* inr;
+    U32 offset, rep[ZSTD_REP_NUM];
+
+    /* init */
+    ctx->nextToUpdate3 = ctx->nextToUpdate;
+    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
+    ip += (ip==prefixStart);
+    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        U32 cur, match_num, last_pos, litlen, price;
+        U32 u, mlen, best_mlen, best_off, litLength;
+        memset(opt, 0, sizeof(ZSTD_optimal_t));
+        last_pos = 0;
+        litlen = (U32)(ip - anchor);
+
+        /* check repCode */
+        {   U32 i, last_i = ZSTD_REP_CHECK + (ip==anchor);
+            for (i=(ip == anchor); i<last_i; i++) {
+                const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : rep[i];
+                if ( (repCur > 0) && (repCur < (S32)(ip-prefixStart))
+                    && (MEM_readMINMATCH(ip, minMatch) == MEM_readMINMATCH(ip - repCur, minMatch))) {
+                    mlen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repCur, iend) + minMatch;
+                    if (mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) {
+                        best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
+                        goto _storeSequence;
+                    }
+                    best_off = i - (ip == anchor);
+                    do {
+                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                        if (mlen > last_pos || price < opt[mlen].price)
+                            SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
+                        mlen--;
+                    } while (mlen >= minMatch);
+        }   }   }
+
+        match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, ip, iend, maxSearches, mls, matches, minMatch);
+
+        if (!last_pos && !match_num) { ip++; continue; }
+
+        if (match_num && (matches[match_num-1].len > sufficient_len || matches[match_num-1].len >= ZSTD_OPT_NUM)) {
+            best_mlen = matches[match_num-1].len;
+            best_off = matches[match_num-1].off;
+            cur = 0;
+            last_pos = 1;
+            goto _storeSequence;
+        }
+
+        /* set prices using matches at position = 0 */
+        best_mlen = (last_pos) ? last_pos : minMatch;
+        for (u = 0; u < match_num; u++) {
+            mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
+            best_mlen = matches[u].len;
+            while (mlen <= best_mlen) {
+                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                if (mlen > last_pos || price < opt[mlen].price)
+                    SET_PRICE(mlen, mlen, matches[u].off, litlen, price);   /* note : macro modifies last_pos */
+                mlen++;
+        }   }
+
+        if (last_pos < minMatch) { ip++; continue; }
+
+        /* initialize opt[0] */
+        { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
+        opt[0].mlen = 1;
+        opt[0].litlen = litlen;
+
+         /* check further positions */
+        for (cur = 1; cur <= last_pos; cur++) {
+           inr = ip + cur;
+
+           if (opt[cur-1].mlen == 1) {
+                litlen = opt[cur-1].litlen + 1;
+                if (cur > litlen) {
+                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen);
+                } else
+                    price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor);
+           } else {
+                litlen = 1;
+                price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1);
+           }
+
+           if (cur > last_pos || price <= opt[cur].price)
+                SET_PRICE(cur, 1, 0, litlen, price);
+
+           if (cur == last_pos) break;
+
+           if (inr > ilimit)  /* last match must start at a minimum distance of 8 from oend */
+               continue;
+
+           mlen = opt[cur].mlen;
+           if (opt[cur].off > ZSTD_REP_MOVE_OPT) {
+                opt[cur].rep[2] = opt[cur-mlen].rep[1];
+                opt[cur].rep[1] = opt[cur-mlen].rep[0];
+                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE_OPT;
+           } else {
+                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
+                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
+                opt[cur].rep[0] = ((opt[cur].off==ZSTD_REP_MOVE_OPT) && (mlen != 1)) ? (opt[cur-mlen].rep[0] - 1) : (opt[cur-mlen].rep[opt[cur].off]);
+           }
+
+            best_mlen = minMatch;
+            {   U32 i, last_i = ZSTD_REP_CHECK + (mlen != 1);
+                for (i=(opt[cur].mlen != 1); i<last_i; i++) {  /* check rep */
+                    const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
+                    if ( (repCur > 0) && (repCur < (S32)(inr-prefixStart))
+                       && (MEM_readMINMATCH(inr, minMatch) == MEM_readMINMATCH(inr - repCur, minMatch))) {
+                       mlen = (U32)ZSTD_count(inr+minMatch, inr+minMatch - repCur, iend) + minMatch;
+
+                       if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
+                            best_mlen = mlen; best_off = i; last_pos = cur + 1;
+                            goto _storeSequence;
+                       }
+
+                       best_off = i - (opt[cur].mlen != 1);
+                       if (mlen > best_mlen) best_mlen = mlen;
+
+                       do {
+                           if (opt[cur].mlen == 1) {
+                                litlen = opt[cur].litlen;
+                                if (cur > litlen) {
+                                    price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
+                                } else
+                                    price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                            } else {
+                                litlen = 0;
+                                price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
+                            }
+
+                            if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
+                                SET_PRICE(cur + mlen, mlen, i, litlen, price);
+                            mlen--;
+                        } while (mlen >= minMatch);
+            }   }   }
+
+            match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, inr, iend, maxSearches, mls, matches, best_mlen);
+
+            if (match_num > 0 && (matches[match_num-1].len > sufficient_len || cur + matches[match_num-1].len >= ZSTD_OPT_NUM)) {
+                best_mlen = matches[match_num-1].len;
+                best_off = matches[match_num-1].off;
+                last_pos = cur + 1;
+                goto _storeSequence;
+            }
+
+            /* set prices using matches at position = cur */
+            for (u = 0; u < match_num; u++) {
+                mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
+                best_mlen = matches[u].len;
+
+                while (mlen <= best_mlen) {
+                    if (opt[cur].mlen == 1) {
+                        litlen = opt[cur].litlen;
+                        if (cur > litlen)
+                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
+                        else
+                            price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                    } else {
+                        litlen = 0;
+                        price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
+                    }
+
+                    if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
+                        SET_PRICE(cur + mlen, mlen, matches[u].off, litlen, price);
+
+                    mlen++;
+        }   }   }
+
+        best_mlen = opt[last_pos].mlen;
+        best_off = opt[last_pos].off;
+        cur = last_pos - best_mlen;
+
+        /* store sequence */
+_storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
+        opt[0].mlen = 1;
+
+        while (1) {
+            mlen = opt[cur].mlen;
+            offset = opt[cur].off;
+            opt[cur].mlen = best_mlen;
+            opt[cur].off = best_off;
+            best_mlen = mlen;
+            best_off = offset;
+            if (mlen > cur) break;
+            cur -= mlen;
+        }
+
+        for (u = 0; u <= last_pos;) {
+            u += opt[u].mlen;
+        }
+
+        for (cur=0; cur < last_pos; ) {
+            mlen = opt[cur].mlen;
+            if (mlen == 1) { ip++; cur++; continue; }
+            offset = opt[cur].off;
+            cur += mlen;
+            litLength = (U32)(ip - anchor);
+
+            if (offset > ZSTD_REP_MOVE_OPT) {
+                rep[2] = rep[1];
+                rep[1] = rep[0];
+                rep[0] = offset - ZSTD_REP_MOVE_OPT;
+                offset--;
+            } else {
+                if (offset != 0) {
+                    best_off = (offset==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : (rep[offset]);
+                    if (offset != 1) rep[2] = rep[1];
+                    rep[1] = rep[0];
+                    rep[0] = best_off;
+                }
+                if (litLength==0) offset--;
+            }
+
+            ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
+            ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
+            anchor = ip = ip + mlen;
+    }    }   /* for (cur=0; cur < last_pos; ) */
+
+    /* Save reps for next block */
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
+
+    /* Last Literals */
+    {   size_t const lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+
+FORCE_INLINE
+void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
+                                     const void* src, size_t srcSize, const int ultra)
+{
+    seqStore_t* seqStorePtr = &(ctx->seqStore);
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - 8;
+    const BYTE* const base = ctx->base;
+    const U32 lowestIndex = ctx->lowLimit;
+    const U32 dictLimit = ctx->dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* const dictBase = ctx->dictBase;
+    const BYTE* const dictEnd  = dictBase + dictLimit;
+
+    const U32 maxSearches = 1U << ctx->params.cParams.searchLog;
+    const U32 sufficient_len = ctx->params.cParams.targetLength;
+    const U32 mls = ctx->params.cParams.searchLength;
+    const U32 minMatch = (ctx->params.cParams.searchLength == 3) ? 3 : 4;
+
+    ZSTD_optimal_t* opt = seqStorePtr->priceTable;
+    ZSTD_match_t* matches = seqStorePtr->matchTable;
+    const BYTE* inr;
+
+    /* init */
+    U32 offset, rep[ZSTD_REP_NUM];
+    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
+
+    ctx->nextToUpdate3 = ctx->nextToUpdate;
+    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
+    ip += (ip==prefixStart);
+
+    /* Match Loop */
+    while (ip < ilimit) {
+        U32 cur, match_num, last_pos, litlen, price;
+        U32 u, mlen, best_mlen, best_off, litLength;
+        U32 current = (U32)(ip-base);
+        memset(opt, 0, sizeof(ZSTD_optimal_t));
+        last_pos = 0;
+        opt[0].litlen = (U32)(ip - anchor);
+
+        /* check repCode */
+        {   U32 i, last_i = ZSTD_REP_CHECK + (ip==anchor);
+            for (i = (ip==anchor); i<last_i; i++) {
+                const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : rep[i];
+                const U32 repIndex = (U32)(current - repCur);
+                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                const BYTE* const repMatch = repBase + repIndex;
+                if ( (repCur > 0 && repCur <= (S32)current)
+                   && (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex>lowestIndex))  /* intentional overflow */
+                   && (MEM_readMINMATCH(ip, minMatch) == MEM_readMINMATCH(repMatch, minMatch)) ) {
+                    /* repcode detected we should take it */
+                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                    mlen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iend, repEnd, prefixStart) + minMatch;
+
+                    if (mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) {
+                        best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
+                        goto _storeSequence;
+                    }
+
+                    best_off = i - (ip==anchor);
+                    litlen = opt[0].litlen;
+                    do {
+                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                        if (mlen > last_pos || price < opt[mlen].price)
+                            SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
+                        mlen--;
+                    } while (mlen >= minMatch);
+        }   }   }
+
+        match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, ip, iend, maxSearches, mls, matches, minMatch);  /* first search (depth 0) */
+
+        if (!last_pos && !match_num) { ip++; continue; }
+
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
+        opt[0].mlen = 1;
+
+        if (match_num && (matches[match_num-1].len > sufficient_len || matches[match_num-1].len >= ZSTD_OPT_NUM)) {
+            best_mlen = matches[match_num-1].len;
+            best_off = matches[match_num-1].off;
+            cur = 0;
+            last_pos = 1;
+            goto _storeSequence;
+        }
+
+        best_mlen = (last_pos) ? last_pos : minMatch;
+
+        /* set prices using matches at position = 0 */
+        for (u = 0; u < match_num; u++) {
+            mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
+            best_mlen = matches[u].len;
+            litlen = opt[0].litlen;
+            while (mlen <= best_mlen) {
+                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                if (mlen > last_pos || price < opt[mlen].price)
+                    SET_PRICE(mlen, mlen, matches[u].off, litlen, price);
+                mlen++;
+        }   }
+
+        if (last_pos < minMatch) {
+            ip++; continue;
+        }
+
+        /* check further positions */
+        for (cur = 1; cur <= last_pos; cur++) {
+            inr = ip + cur;
+
+            if (opt[cur-1].mlen == 1) {
+                litlen = opt[cur-1].litlen + 1;
+                if (cur > litlen) {
+                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-litlen);
+                } else
+                    price = ZSTD_getLiteralPrice(seqStorePtr, litlen, anchor);
+            } else {
+                litlen = 1;
+                price = opt[cur - 1].price + ZSTD_getLiteralPrice(seqStorePtr, litlen, inr-1);
+            }
+
+            if (cur > last_pos || price <= opt[cur].price)
+                SET_PRICE(cur, 1, 0, litlen, price);
+
+            if (cur == last_pos) break;
+
+            if (inr > ilimit)  /* last match must start at a minimum distance of 8 from oend */
+                continue;
+
+            mlen = opt[cur].mlen;
+            if (opt[cur].off > ZSTD_REP_MOVE_OPT) {
+                opt[cur].rep[2] = opt[cur-mlen].rep[1];
+                opt[cur].rep[1] = opt[cur-mlen].rep[0];
+                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE_OPT;
+            } else {
+                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
+                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
+                opt[cur].rep[0] = ((opt[cur].off==ZSTD_REP_MOVE_OPT) && (mlen != 1)) ? (opt[cur-mlen].rep[0] - 1) : (opt[cur-mlen].rep[opt[cur].off]);
+            }
+
+            best_mlen = minMatch;
+            {   U32 i, last_i = ZSTD_REP_CHECK + (mlen != 1);
+                for (i = (mlen != 1); i<last_i; i++) {
+                    const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
+                    const U32 repIndex = (U32)(current+cur - repCur);
+                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
+                    const BYTE* const repMatch = repBase + repIndex;
+                    if ( (repCur > 0 && repCur <= (S32)(current+cur))
+                      && (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex>lowestIndex))  /* intentional overflow */
+                      && (MEM_readMINMATCH(inr, minMatch) == MEM_readMINMATCH(repMatch, minMatch)) ) {
+                        /* repcode detected */
+                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
+                        mlen = (U32)ZSTD_count_2segments(inr+minMatch, repMatch+minMatch, iend, repEnd, prefixStart) + minMatch;
+
+                        if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
+                            best_mlen = mlen; best_off = i; last_pos = cur + 1;
+                            goto _storeSequence;
+                        }
+
+                        best_off = i - (opt[cur].mlen != 1);
+                        if (mlen > best_mlen) best_mlen = mlen;
+
+                        do {
+                            if (opt[cur].mlen == 1) {
+                                litlen = opt[cur].litlen;
+                                if (cur > litlen) {
+                                    price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
+                                } else
+                                    price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
+                            } else {
+                                litlen = 0;
+                                price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
+                            }
+
+                            if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
+                                SET_PRICE(cur + mlen, mlen, i, litlen, price);
+                            mlen--;
+                        } while (mlen >= minMatch);
+            }   }   }
+
+            match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, inr, iend, maxSearches, mls, matches, minMatch);
+
+            if (match_num > 0 && (matches[match_num-1].len > sufficient_len || cur + matches[match_num-1].len >= ZSTD_OPT_NUM)) {
+                best_mlen = matches[match_num-1].len;
+                best_off = matches[match_num-1].off;
+                last_pos = cur + 1;
+                goto _storeSequence;
+            }
+
+            /* set prices using matches at position = cur */
+            for (u = 0; u < match_num; u++) {
+                mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
+                best_mlen = matches[u].len;
+
+                while (mlen <= best_mlen) {
+                    if (opt[cur].mlen == 1) {
+                        litlen = opt[cur].litlen;
+                        if (cur > litlen)
+                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
+                        else
+                            price = ZSTD_getPrice(seqStorePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
+                    } else {
+                        litlen = 0;
+                        price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
+                    }
+
+                    if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
+                        SET_PRICE(cur + mlen, mlen, matches[u].off, litlen, price);
+
+                    mlen++;
+        }   }   }   /* for (cur = 1; cur <= last_pos; cur++) */
+
+        best_mlen = opt[last_pos].mlen;
+        best_off = opt[last_pos].off;
+        cur = last_pos - best_mlen;
+
+        /* store sequence */
+_storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
+        opt[0].mlen = 1;
+
+        while (1) {
+            mlen = opt[cur].mlen;
+            offset = opt[cur].off;
+            opt[cur].mlen = best_mlen;
+            opt[cur].off = best_off;
+            best_mlen = mlen;
+            best_off = offset;
+            if (mlen > cur) break;
+            cur -= mlen;
+        }
+
+        for (u = 0; u <= last_pos; ) {
+            u += opt[u].mlen;
+        }
+
+        for (cur=0; cur < last_pos; ) {
+            mlen = opt[cur].mlen;
+            if (mlen == 1) { ip++; cur++; continue; }
+            offset = opt[cur].off;
+            cur += mlen;
+            litLength = (U32)(ip - anchor);
+
+            if (offset > ZSTD_REP_MOVE_OPT) {
+                rep[2] = rep[1];
+                rep[1] = rep[0];
+                rep[0] = offset - ZSTD_REP_MOVE_OPT;
+                offset--;
+            } else {
+                if (offset != 0) {
+                    best_off = (offset==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : (rep[offset]);
+                    if (offset != 1) rep[2] = rep[1];
+                    rep[1] = rep[0];
+                    rep[0] = best_off;
+                }
+
+                if (litLength==0) offset--;
+            }
+
+            ZSTD_updatePrice(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
+            ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
+            anchor = ip = ip + mlen;
+    }    }   /* for (cur=0; cur < last_pos; ) */
+
+    /* Save reps for next block */
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
+
+    /* Last Literals */
+    {   size_t lastLLSize = iend - anchor;
+        memcpy(seqStorePtr->lit, anchor, lastLLSize);
+        seqStorePtr->lit += lastLLSize;
+    }
+}
+
+#endif  /* ZSTD_OPT_H_91842398743 */
diff --git a/thirdparty/zstd/compress/zstdmt_compress.c b/thirdparty/zstd/compress/zstdmt_compress.c
new file mode 100644
index 0000000000..fc7f52a290
--- /dev/null
+++ b/thirdparty/zstd/compress/zstdmt_compress.c
@@ -0,0 +1,751 @@
+/**
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree. An additional grant
+ * of patent rights can be found in the PATENTS file in the same directory.
+ */
+
+
+/* ======   Tuning parameters   ====== */
+#define ZSTDMT_NBTHREADS_MAX 128
+
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+/* ======   Dependencies   ====== */
+#include <stdlib.h>   /* malloc */
+#include <string.h>   /* memcpy */
+#include "pool.h"     /* threadpool */
+#include "threading.h"  /* mutex */
+#include "zstd_internal.h"   /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include "zstdmt_compress.h"
+
+
+/* ======   Debug   ====== */
+#if 0
+
+#  include <stdio.h>
+#  include <unistd.h>
+#  include <sys/times.h>
+   static unsigned g_debugLevel = 5;
+#  define DEBUGLOGRAW(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __VA_ARGS__); }
+#  define DEBUGLOG(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __FILE__ ": "); fprintf(stderr, __VA_ARGS__); fprintf(stderr, " \n"); }
+
+#  define DEBUG_PRINTHEX(l,p,n) { \
+    unsigned debug_u;                   \
+    for (debug_u=0; debug_u<(n); debug_u++)           \
+        DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
+    DEBUGLOGRAW(l, " \n");       \
+}
+
+static unsigned long long GetCurrentClockTimeMicroseconds(void)
+{
+   static clock_t _ticksPerSecond = 0;
+   if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
+
+   { struct tms junk; clock_t newTicks = (clock_t) times(&junk);
+     return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond); }
+}
+
+#define MUTEX_WAIT_TIME_DLEVEL 5
+#define PTHREAD_MUTEX_LOCK(mutex) \
+if (g_debugLevel>=MUTEX_WAIT_TIME_DLEVEL) { \
+    unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
+    pthread_mutex_lock(mutex); \
+    {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
+        unsigned long long const elapsedTime = (afterTime-beforeTime); \
+        if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \
+            DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
+               elapsedTime, #mutex); \
+    }   } \
+} else pthread_mutex_lock(mutex);
+
+#else
+
+#  define DEBUGLOG(l, ...)      {}    /* disabled */
+#  define PTHREAD_MUTEX_LOCK(m) pthread_mutex_lock(m)
+#  define DEBUG_PRINTHEX(l,p,n) {}
+
+#endif
+
+
+/* =====   Buffer Pool   ===== */
+
+typedef struct buffer_s {
+    void* start;
+    size_t size;
+} buffer_t;
+
+static const buffer_t g_nullBuffer = { NULL, 0 };
+
+typedef struct ZSTDMT_bufferPool_s {
+    unsigned totalBuffers;
+    unsigned nbBuffers;
+    buffer_t bTable[1];   /* variable size */
+} ZSTDMT_bufferPool;
+
+static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads)
+{
+    unsigned const maxNbBuffers = 2*nbThreads + 2;
+    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)calloc(1, sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t));
+    if (bufPool==NULL) return NULL;
+    bufPool->totalBuffers = maxNbBuffers;
+    bufPool->nbBuffers = 0;
+    return bufPool;
+}
+
+static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
+{
+    unsigned u;
+    if (!bufPool) return;   /* compatibility with free on NULL */
+    for (u=0; u<bufPool->totalBuffers; u++)
+        free(bufPool->bTable[u].start);
+    free(bufPool);
+}
+
+/* assumption : invocation from main thread only ! */
+static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* pool, size_t bSize)
+{
+    if (pool->nbBuffers) {   /* try to use an existing buffer */
+        buffer_t const buf = pool->bTable[--(pool->nbBuffers)];
+        size_t const availBufferSize = buf.size;
+        if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize))   /* large enough, but not too much */
+            return buf;
+        free(buf.start);   /* size conditions not respected : scratch this buffer and create a new one */
+    }
+    /* create new buffer */
+    {   buffer_t buffer;
+        void* const start = malloc(bSize);
+        if (start==NULL) bSize = 0;
+        buffer.start = start;   /* note : start can be NULL if malloc fails ! */
+        buffer.size = bSize;
+        return buffer;
+    }
+}
+
+/* store buffer for later re-use, up to pool capacity */
+static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* pool, buffer_t buf)
+{
+    if (buf.start == NULL) return;   /* release on NULL */
+    if (pool->nbBuffers < pool->totalBuffers) {
+        pool->bTable[pool->nbBuffers++] = buf;   /* store for later re-use */
+        return;
+    }
+    /* Reached bufferPool capacity (should not happen) */
+    free(buf.start);
+}
+
+
+/* =====   CCtx Pool   ===== */
+
+typedef struct {
+    unsigned totalCCtx;
+    unsigned availCCtx;
+    ZSTD_CCtx* cctx[1];   /* variable size */
+} ZSTDMT_CCtxPool;
+
+/* assumption : CCtxPool invocation only from main thread */
+
+/* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
+static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
+{
+    unsigned u;
+    for (u=0; u<pool->totalCCtx; u++)
+        ZSTD_freeCCtx(pool->cctx[u]);  /* note : compatible with free on NULL */
+    free(pool);
+}
+
+/* ZSTDMT_createCCtxPool() :
+ * implies nbThreads >= 1 , checked by caller ZSTDMT_createCCtx() */
+static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads)
+{
+    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) calloc(1, sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*));
+    if (!cctxPool) return NULL;
+    cctxPool->totalCCtx = nbThreads;
+    cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
+    cctxPool->cctx[0] = ZSTD_createCCtx();
+    if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
+    DEBUGLOG(1, "cctxPool created, with %u threads", nbThreads);
+    return cctxPool;
+}
+
+static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* pool)
+{
+    if (pool->availCCtx) {
+        pool->availCCtx--;
+        return pool->cctx[pool->availCCtx];
+    }
+    return ZSTD_createCCtx();   /* note : can be NULL, when creation fails ! */
+}
+
+static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return;   /* compatibility with release on NULL */
+    if (pool->availCCtx < pool->totalCCtx)
+        pool->cctx[pool->availCCtx++] = cctx;
+    else
+        /* pool overflow : should not happen, since totalCCtx==nbThreads */
+        ZSTD_freeCCtx(cctx);
+}
+
+
+/* =====   Thread worker   ===== */
+
+typedef struct {
+    buffer_t buffer;
+    size_t filled;
+} inBuff_t;
+
+typedef struct {
+    ZSTD_CCtx* cctx;
+    buffer_t src;
+    const void* srcStart;
+    size_t   srcSize;
+    size_t   dictSize;
+    buffer_t dstBuff;
+    size_t   cSize;
+    size_t   dstFlushed;
+    unsigned firstChunk;
+    unsigned lastChunk;
+    unsigned jobCompleted;
+    unsigned jobScanned;
+    pthread_mutex_t* jobCompleted_mutex;
+    pthread_cond_t* jobCompleted_cond;
+    ZSTD_parameters params;
+    ZSTD_CDict* cdict;
+    unsigned long long fullFrameSize;
+} ZSTDMT_jobDescription;
+
+/* ZSTDMT_compressChunk() : POOL_function type */
+void ZSTDMT_compressChunk(void* jobDescription)
+{
+    ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
+    const void* const src = (const char*)job->srcStart + job->dictSize;
+    buffer_t const dstBuff = job->dstBuff;
+    DEBUGLOG(3, "job (first:%u) (last:%u) : dictSize %u, srcSize %u",
+                 job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize);
+    if (job->cdict) {  /* should only happen for first segment */
+        size_t const initError = ZSTD_compressBegin_usingCDict_advanced(job->cctx, job->cdict, job->params.fParams, job->fullFrameSize);
+        if (job->cdict) DEBUGLOG(3, "using CDict ");
+        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
+    } else {  /* srcStart points at reloaded section */
+        if (!job->firstChunk) job->params.fParams.contentSizeFlag = 0;  /* ensure no srcSize control */
+        {   size_t const dictModeError = ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceRawDict, 1);  /* Force loading dictionary in "content-only" mode (no header analysis) */
+            size_t const initError = ZSTD_compressBegin_advanced(job->cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize);
+            if (ZSTD_isError(initError) || ZSTD_isError(dictModeError)) { job->cSize = initError; goto _endJob; }
+            ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceWindow, 1);
+    }   }
+    if (!job->firstChunk) {  /* flush and overwrite frame header when it's not first segment */
+        size_t const hSize = ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, 0);
+        if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; }
+        ZSTD_invalidateRepCodes(job->cctx);
+    }
+
+    DEBUGLOG(4, "Compressing : ");
+    DEBUG_PRINTHEX(4, job->srcStart, 12);
+    job->cSize = (job->lastChunk) ?
+                 ZSTD_compressEnd     (job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :
+                 ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize);
+    DEBUGLOG(3, "compressed %u bytes into %u bytes   (first:%u) (last:%u)",
+                (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk);
+    DEBUGLOG(5, "dstBuff.size : %u ; => %s", (U32)dstBuff.size, ZSTD_getErrorName(job->cSize));
+
+_endJob:
+    PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex);
+    job->jobCompleted = 1;
+    job->jobScanned = 0;
+    pthread_cond_signal(job->jobCompleted_cond);
+    pthread_mutex_unlock(job->jobCompleted_mutex);
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+struct ZSTDMT_CCtx_s {
+    POOL_ctx* factory;
+    ZSTDMT_bufferPool* buffPool;
+    ZSTDMT_CCtxPool* cctxPool;
+    pthread_mutex_t jobCompleted_mutex;
+    pthread_cond_t jobCompleted_cond;
+    size_t targetSectionSize;
+    size_t marginSize;
+    size_t inBuffSize;
+    size_t dictSize;
+    size_t targetDictSize;
+    inBuff_t inBuff;
+    ZSTD_parameters params;
+    XXH64_state_t xxhState;
+    unsigned nbThreads;
+    unsigned jobIDMask;
+    unsigned doneJobID;
+    unsigned nextJobID;
+    unsigned frameEnded;
+    unsigned allJobsCompleted;
+    unsigned overlapRLog;
+    unsigned long long frameContentSize;
+    size_t sectionSize;
+    ZSTD_CDict* cdict;
+    ZSTD_CStream* cstream;
+    ZSTDMT_jobDescription jobs[1];   /* variable size (must lies at the end) */
+};
+
+ZSTDMT_CCtx *ZSTDMT_createCCtx(unsigned nbThreads)
+{
+    ZSTDMT_CCtx* cctx;
+    U32 const minNbJobs = nbThreads + 2;
+    U32 const nbJobsLog2 = ZSTD_highbit32(minNbJobs) + 1;
+    U32 const nbJobs = 1 << nbJobsLog2;
+    DEBUGLOG(5, "nbThreads : %u  ; minNbJobs : %u ;  nbJobsLog2 : %u ;  nbJobs : %u  \n",
+            nbThreads, minNbJobs, nbJobsLog2, nbJobs);
+    if ((nbThreads < 1) | (nbThreads > ZSTDMT_NBTHREADS_MAX)) return NULL;
+    cctx = (ZSTDMT_CCtx*) calloc(1, sizeof(ZSTDMT_CCtx) + nbJobs*sizeof(ZSTDMT_jobDescription));
+    if (!cctx) return NULL;
+    cctx->nbThreads = nbThreads;
+    cctx->jobIDMask = nbJobs - 1;
+    cctx->allJobsCompleted = 1;
+    cctx->sectionSize = 0;
+    cctx->overlapRLog = 3;
+    cctx->factory = POOL_create(nbThreads, 1);
+    cctx->buffPool = ZSTDMT_createBufferPool(nbThreads);
+    cctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads);
+    if (!cctx->factory | !cctx->buffPool | !cctx->cctxPool) {  /* one object was not created */
+        ZSTDMT_freeCCtx(cctx);
+        return NULL;
+    }
+    if (nbThreads==1) {
+        cctx->cstream = ZSTD_createCStream();
+        if (!cctx->cstream) {
+            ZSTDMT_freeCCtx(cctx); return NULL;
+    }   }
+    pthread_mutex_init(&cctx->jobCompleted_mutex, NULL);   /* Todo : check init function return */
+    pthread_cond_init(&cctx->jobCompleted_cond, NULL);
+    DEBUGLOG(4, "mt_cctx created, for %u threads \n", nbThreads);
+    return cctx;
+}
+
+/* ZSTDMT_releaseAllJobResources() :
+ * Ensure all workers are killed first. */
+static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
+{
+    unsigned jobID;
+    for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
+        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].dstBuff);
+        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
+        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].src);
+        mtctx->jobs[jobID].src = g_nullBuffer;
+        ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[jobID].cctx);
+        mtctx->jobs[jobID].cctx = NULL;
+    }
+    memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
+    ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer);
+    mtctx->inBuff.buffer = g_nullBuffer;
+    mtctx->allJobsCompleted = 1;
+}
+
+size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
+{
+    if (mtctx==NULL) return 0;   /* compatible with free on NULL */
+    POOL_free(mtctx->factory);
+    if (!mtctx->allJobsCompleted) ZSTDMT_releaseAllJobResources(mtctx); /* stop workers first */
+    ZSTDMT_freeBufferPool(mtctx->buffPool);  /* release job resources into pools first */
+    ZSTDMT_freeCCtxPool(mtctx->cctxPool);
+    ZSTD_freeCDict(mtctx->cdict);
+    ZSTD_freeCStream(mtctx->cstream);
+    pthread_mutex_destroy(&mtctx->jobCompleted_mutex);
+    pthread_cond_destroy(&mtctx->jobCompleted_cond);
+    free(mtctx);
+    return 0;
+}
+
+size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter, unsigned value)
+{
+    switch(parameter)
+    {
+    case ZSTDMT_p_sectionSize :
+        mtctx->sectionSize = value;
+        return 0;
+    case ZSTDMT_p_overlapSectionLog :
+    DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);
+        mtctx->overlapRLog = (value >= 9) ? 0 : 9 - value;
+        return 0;
+    default :
+        return ERROR(compressionParameter_unsupported);
+    }
+}
+
+
+/* ------------------------------------------ */
+/* =====   Multi-threaded compression   ===== */
+/* ------------------------------------------ */
+
+size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
+                           void* dst, size_t dstCapacity,
+                     const void* src, size_t srcSize,
+                           int compressionLevel)
+{
+    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
+    U32 const overlapLog = (compressionLevel >= ZSTD_maxCLevel()) ? 0 : 3;
+    size_t const overlapSize = (size_t)1 << (params.cParams.windowLog - overlapLog);
+    size_t const chunkTargetSize = (size_t)1 << (params.cParams.windowLog + 2);
+    unsigned const nbChunksMax = (unsigned)(srcSize / chunkTargetSize) + 1;
+    unsigned nbChunks = MIN(nbChunksMax, mtctx->nbThreads);
+    size_t const proposedChunkSize = (srcSize + (nbChunks-1)) / nbChunks;
+    size_t const avgChunkSize = ((proposedChunkSize & 0x1FFFF) < 0xFFFF) ? proposedChunkSize + 0xFFFF : proposedChunkSize;   /* avoid too small last block */
+    size_t remainingSrcSize = srcSize;
+    const char* const srcStart = (const char*)src;
+    unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbChunks : (unsigned)(dstCapacity / ZSTD_compressBound(avgChunkSize));  /* presumes avgChunkSize >= 256 KB, which should be the case */
+    size_t frameStartPos = 0, dstBufferPos = 0;
+
+    DEBUGLOG(3, "windowLog : %2u => chunkTargetSize : %u bytes  ", params.cParams.windowLog, (U32)chunkTargetSize);
+    DEBUGLOG(2, "nbChunks  : %2u   (chunkSize : %u bytes)   ", nbChunks, (U32)avgChunkSize);
+    params.fParams.contentSizeFlag = 1;
+
+    if (nbChunks==1) {   /* fallback to single-thread mode */
+        ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
+        return ZSTD_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);
+    }
+
+    {   unsigned u;
+        for (u=0; u<nbChunks; u++) {
+            size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize);
+            size_t const dstBufferCapacity = ZSTD_compressBound(chunkSize);
+            buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
+            buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : ZSTDMT_getBuffer(mtctx->buffPool, dstBufferCapacity);
+            ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(mtctx->cctxPool);
+            size_t dictSize = u ? overlapSize : 0;
+
+            if ((cctx==NULL) || (dstBuffer.start==NULL)) {
+                mtctx->jobs[u].cSize = ERROR(memory_allocation);   /* job result */
+                mtctx->jobs[u].jobCompleted = 1;
+                nbChunks = u+1;
+                break;   /* let's wait for previous jobs to complete, but don't start new ones */
+            }
+
+            mtctx->jobs[u].srcStart = srcStart + frameStartPos - dictSize;
+            mtctx->jobs[u].dictSize = dictSize;
+            mtctx->jobs[u].srcSize = chunkSize;
+            mtctx->jobs[u].fullFrameSize = srcSize;
+            mtctx->jobs[u].params = params;
+            mtctx->jobs[u].dstBuff = dstBuffer;
+            mtctx->jobs[u].cctx = cctx;
+            mtctx->jobs[u].firstChunk = (u==0);
+            mtctx->jobs[u].lastChunk = (u==nbChunks-1);
+            mtctx->jobs[u].jobCompleted = 0;
+            mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex;
+            mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond;
+
+            DEBUGLOG(3, "posting job %u   (%u bytes)", u, (U32)chunkSize);
+            DEBUG_PRINTHEX(3, mtctx->jobs[u].srcStart, 12);
+            POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]);
+
+            frameStartPos += chunkSize;
+            dstBufferPos += dstBufferCapacity;
+            remainingSrcSize -= chunkSize;
+    }   }
+    /* note : since nbChunks <= nbThreads, all jobs should be running immediately in parallel */
+
+    {   unsigned chunkID;
+        size_t error = 0, dstPos = 0;
+        for (chunkID=0; chunkID<nbChunks; chunkID++) {
+            DEBUGLOG(3, "waiting for chunk %u ", chunkID);
+            PTHREAD_MUTEX_LOCK(&mtctx->jobCompleted_mutex);
+            while (mtctx->jobs[chunkID].jobCompleted==0) {
+                DEBUGLOG(4, "waiting for jobCompleted signal from chunk %u", chunkID);
+                pthread_cond_wait(&mtctx->jobCompleted_cond, &mtctx->jobCompleted_mutex);
+            }
+            pthread_mutex_unlock(&mtctx->jobCompleted_mutex);
+            DEBUGLOG(3, "ready to write chunk %u ", chunkID);
+
+            ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[chunkID].cctx);
+            mtctx->jobs[chunkID].cctx = NULL;
+            mtctx->jobs[chunkID].srcStart = NULL;
+            {   size_t const cSize = mtctx->jobs[chunkID].cSize;
+                if (ZSTD_isError(cSize)) error = cSize;
+                if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
+                if (chunkID) {   /* note : chunk 0 is already written directly into dst */
+                    if (!error)
+                        memmove((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize);  /* may overlap if chunk decompressed within dst */
+                    if (chunkID >= compressWithinDst)   /* otherwise, it decompresses within dst */
+                        ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[chunkID].dstBuff);
+                    mtctx->jobs[chunkID].dstBuff = g_nullBuffer;
+                }
+                dstPos += cSize ;
+            }
+        }
+        if (!error) DEBUGLOG(3, "compressed size : %u  ", (U32)dstPos);
+        return error ? error : dstPos;
+    }
+
+}
+
+
+/* ====================================== */
+/* =======      Streaming API     ======= */
+/* ====================================== */
+
+static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* zcs) {
+    while (zcs->doneJobID < zcs->nextJobID) {
+        unsigned const jobID = zcs->doneJobID & zcs->jobIDMask;
+        PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
+        while (zcs->jobs[jobID].jobCompleted==0) {
+            DEBUGLOG(4, "waiting for jobCompleted signal from chunk %u", zcs->doneJobID);   /* we want to block when waiting for data to flush */
+            pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);
+        }
+        pthread_mutex_unlock(&zcs->jobCompleted_mutex);
+        zcs->doneJobID++;
+    }
+}
+
+
+static size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
+                                    const void* dict, size_t dictSize, unsigned updateDict,
+                                    ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    ZSTD_customMem const cmem = { NULL, NULL, NULL };
+    DEBUGLOG(3, "Started new compression, with windowLog : %u", params.cParams.windowLog);
+    if (zcs->nbThreads==1) return ZSTD_initCStream_advanced(zcs->cstream, dict, dictSize, params, pledgedSrcSize);
+    if (zcs->allJobsCompleted == 0) {   /* previous job not correctly finished */
+        ZSTDMT_waitForAllJobsCompleted(zcs);
+        ZSTDMT_releaseAllJobResources(zcs);
+        zcs->allJobsCompleted = 1;
+    }
+    zcs->params = params;
+    if (updateDict) {
+        ZSTD_freeCDict(zcs->cdict); zcs->cdict = NULL;
+        if (dict && dictSize) {
+            zcs->cdict = ZSTD_createCDict_advanced(dict, dictSize, 0, params.cParams, cmem);
+            if (zcs->cdict == NULL) return ERROR(memory_allocation);
+    }   }
+    zcs->frameContentSize = pledgedSrcSize;
+    zcs->targetDictSize = (zcs->overlapRLog>=9) ? 0 : (size_t)1 << (zcs->params.cParams.windowLog - zcs->overlapRLog);
+    DEBUGLOG(4, "overlapRLog : %u ", zcs->overlapRLog);
+    DEBUGLOG(3, "overlap Size : %u KB", (U32)(zcs->targetDictSize>>10));
+    zcs->targetSectionSize = zcs->sectionSize ? zcs->sectionSize : (size_t)1 << (zcs->params.cParams.windowLog + 2);
+    zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize);
+    zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize);
+    DEBUGLOG(3, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10));
+    zcs->marginSize = zcs->targetSectionSize >> 2;
+    zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize + zcs->marginSize;
+    zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);
+    if (zcs->inBuff.buffer.start == NULL) return ERROR(memory_allocation);
+    zcs->inBuff.filled = 0;
+    zcs->dictSize = 0;
+    zcs->doneJobID = 0;
+    zcs->nextJobID = 0;
+    zcs->frameEnded = 0;
+    zcs->allJobsCompleted = 0;
+    if (params.fParams.checksumFlag) XXH64_reset(&zcs->xxhState, 0);
+    return 0;
+}
+
+size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* zcs,
+                                const void* dict, size_t dictSize,
+                                ZSTD_parameters params, unsigned long long pledgedSrcSize)
+{
+    return ZSTDMT_initCStream_internal(zcs, dict, dictSize, 1, params, pledgedSrcSize);
+}
+
+/* ZSTDMT_resetCStream() :
+ * pledgedSrcSize is optional and can be zero == unknown */
+size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* zcs, unsigned long long pledgedSrcSize)
+{
+    if (zcs->nbThreads==1) return ZSTD_resetCStream(zcs->cstream, pledgedSrcSize);
+    return ZSTDMT_initCStream_internal(zcs, NULL, 0, 0, zcs->params, pledgedSrcSize);
+}
+
+size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) {
+    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, 0);
+    return ZSTDMT_initCStream_internal(zcs, NULL, 0, 1, params, 0);
+}
+
+
+static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame)
+{
+    size_t const dstBufferCapacity = ZSTD_compressBound(srcSize);
+    buffer_t const dstBuffer = ZSTDMT_getBuffer(zcs->buffPool, dstBufferCapacity);
+    ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(zcs->cctxPool);
+    unsigned const jobID = zcs->nextJobID & zcs->jobIDMask;
+
+    if ((cctx==NULL) || (dstBuffer.start==NULL)) {
+        zcs->jobs[jobID].jobCompleted = 1;
+        zcs->nextJobID++;
+        ZSTDMT_waitForAllJobsCompleted(zcs);
+        ZSTDMT_releaseAllJobResources(zcs);
+        return ERROR(memory_allocation);
+    }
+
+    DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ", zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize);
+    zcs->jobs[jobID].src = zcs->inBuff.buffer;
+    zcs->jobs[jobID].srcStart = zcs->inBuff.buffer.start;
+    zcs->jobs[jobID].srcSize = srcSize;
+    zcs->jobs[jobID].dictSize = zcs->dictSize;   /* note : zcs->inBuff.filled is presumed >= srcSize + dictSize */
+    zcs->jobs[jobID].params = zcs->params;
+    if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0;  /* do not calculate checksum within sections, just keep it in header for first section */
+    zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL;
+    zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize;
+    zcs->jobs[jobID].dstBuff = dstBuffer;
+    zcs->jobs[jobID].cctx = cctx;
+    zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0);
+    zcs->jobs[jobID].lastChunk = endFrame;
+    zcs->jobs[jobID].jobCompleted = 0;
+    zcs->jobs[jobID].dstFlushed = 0;
+    zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex;
+    zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond;
+
+    /* get a new buffer for next input */
+    if (!endFrame) {
+        size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize);
+        zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);
+        if (zcs->inBuff.buffer.start == NULL) {   /* not enough memory to allocate next input buffer */
+            zcs->jobs[jobID].jobCompleted = 1;
+            zcs->nextJobID++;
+            ZSTDMT_waitForAllJobsCompleted(zcs);
+            ZSTDMT_releaseAllJobResources(zcs);
+            return ERROR(memory_allocation);
+        }
+        DEBUGLOG(5, "inBuff filled to %u", (U32)zcs->inBuff.filled);
+        zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize;
+        DEBUGLOG(5, "new job : filled to %u, with %u dict and %u src", (U32)zcs->inBuff.filled, (U32)newDictSize, (U32)(zcs->inBuff.filled - newDictSize));
+        memmove(zcs->inBuff.buffer.start, (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize, zcs->inBuff.filled);
+        DEBUGLOG(5, "new inBuff pre-filled");
+        zcs->dictSize = newDictSize;
+    } else {
+        zcs->inBuff.buffer = g_nullBuffer;
+        zcs->inBuff.filled = 0;
+        zcs->dictSize = 0;
+        zcs->frameEnded = 1;
+        if (zcs->nextJobID == 0)
+            zcs->params.fParams.checksumFlag = 0;   /* single chunk : checksum is calculated directly within worker thread */
+    }
+
+    DEBUGLOG(3, "posting job %u : %u bytes  (end:%u) (note : doneJob = %u=>%u)", zcs->nextJobID, (U32)zcs->jobs[jobID].srcSize, zcs->jobs[jobID].lastChunk, zcs->doneJobID, zcs->doneJobID & zcs->jobIDMask);
+    POOL_add(zcs->factory, ZSTDMT_compressChunk, &zcs->jobs[jobID]);   /* this call is blocking when thread worker pool is exhausted */
+    zcs->nextJobID++;
+    return 0;
+}
+
+
+/* ZSTDMT_flushNextJob() :
+ * output : will be updated with amount of data flushed .
+ * blockToFlush : if >0, the function will block and wait if there is no data available to flush .
+ * @return : amount of data remaining within internal buffer, 1 if unknown but > 0, 0 if no more, or an error code */
+static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned blockToFlush)
+{
+    unsigned const wJobID = zcs->doneJobID & zcs->jobIDMask;
+    if (zcs->doneJobID == zcs->nextJobID) return 0;   /* all flushed ! */
+    PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
+    while (zcs->jobs[wJobID].jobCompleted==0) {
+        DEBUGLOG(5, "waiting for jobCompleted signal from job %u", zcs->doneJobID);
+        if (!blockToFlush) { pthread_mutex_unlock(&zcs->jobCompleted_mutex); return 0; }  /* nothing ready to be flushed => skip */
+        pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);  /* block when nothing available to flush */
+    }
+    pthread_mutex_unlock(&zcs->jobCompleted_mutex);
+    /* compression job completed : output can be flushed */
+    {   ZSTDMT_jobDescription job = zcs->jobs[wJobID];
+        if (!job.jobScanned) {
+            if (ZSTD_isError(job.cSize)) {
+                DEBUGLOG(5, "compression error detected ");
+                ZSTDMT_waitForAllJobsCompleted(zcs);
+                ZSTDMT_releaseAllJobResources(zcs);
+                return job.cSize;
+            }
+            ZSTDMT_releaseCCtx(zcs->cctxPool, job.cctx);
+            zcs->jobs[wJobID].cctx = NULL;
+            DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag);
+            if (zcs->params.fParams.checksumFlag) {
+                XXH64_update(&zcs->xxhState, (const char*)job.srcStart + job.dictSize, job.srcSize);
+                if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) {  /* write checksum at end of last section */
+                    U32 const checksum = (U32)XXH64_digest(&zcs->xxhState);
+                    DEBUGLOG(4, "writing checksum : %08X \n", checksum);
+                    MEM_writeLE32((char*)job.dstBuff.start + job.cSize, checksum);
+                    job.cSize += 4;
+                    zcs->jobs[wJobID].cSize += 4;
+            }   }
+            ZSTDMT_releaseBuffer(zcs->buffPool, job.src);
+            zcs->jobs[wJobID].srcStart = NULL;
+            zcs->jobs[wJobID].src = g_nullBuffer;
+            zcs->jobs[wJobID].jobScanned = 1;
+        }
+        {   size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos);
+            DEBUGLOG(4, "Flushing %u bytes from job %u ", (U32)toWrite, zcs->doneJobID);
+            memcpy((char*)output->dst + output->pos, (const char*)job.dstBuff.start + job.dstFlushed, toWrite);
+            output->pos += toWrite;
+            job.dstFlushed += toWrite;
+        }
+        if (job.dstFlushed == job.cSize) {   /* output buffer fully flushed => move to next one */
+            ZSTDMT_releaseBuffer(zcs->buffPool, job.dstBuff);
+            zcs->jobs[wJobID].dstBuff = g_nullBuffer;
+            zcs->jobs[wJobID].jobCompleted = 0;
+            zcs->doneJobID++;
+        } else {
+            zcs->jobs[wJobID].dstFlushed = job.dstFlushed;
+        }
+        /* return value : how many bytes left in buffer ; fake it to 1 if unknown but >0 */
+        if (job.cSize > job.dstFlushed) return (job.cSize - job.dstFlushed);
+        if (zcs->doneJobID < zcs->nextJobID) return 1;   /* still some buffer to flush */
+        zcs->allJobsCompleted = zcs->frameEnded;   /* frame completed and entirely flushed */
+        return 0;   /* everything flushed */
+}   }
+
+
+size_t ZSTDMT_compressStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
+{
+    size_t const newJobThreshold = zcs->dictSize + zcs->targetSectionSize + zcs->marginSize;
+    if (zcs->frameEnded) return ERROR(stage_wrong);   /* current frame being ended. Only flush is allowed. Restart with init */
+    if (zcs->nbThreads==1) return ZSTD_compressStream(zcs->cstream, output, input);
+
+    /* fill input buffer */
+    {   size_t const toLoad = MIN(input->size - input->pos, zcs->inBuffSize - zcs->inBuff.filled);
+        memcpy((char*)zcs->inBuff.buffer.start + zcs->inBuff.filled, input->src, toLoad);
+        input->pos += toLoad;
+        zcs->inBuff.filled += toLoad;
+    }
+
+    if ( (zcs->inBuff.filled >= newJobThreshold)  /* filled enough : let's compress */
+        && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {   /* avoid overwriting job round buffer */
+        CHECK_F( ZSTDMT_createCompressionJob(zcs, zcs->targetSectionSize, 0) );
+    }
+
+    /* check for data to flush */
+    CHECK_F( ZSTDMT_flushNextJob(zcs, output, (zcs->inBuff.filled == zcs->inBuffSize)) ); /* block if it wasn't possible to create new job due to saturation */
+
+    /* recommended next input size : fill current input buffer */
+    return zcs->inBuffSize - zcs->inBuff.filled;   /* note : could be zero when input buffer is fully filled and no more availability to create new job */
+}
+
+
+static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned endFrame)
+{
+    size_t const srcSize = zcs->inBuff.filled - zcs->dictSize;
+
+    if (srcSize) DEBUGLOG(4, "flushing : %u bytes left to compress", (U32)srcSize);
+    if ( ((srcSize > 0) || (endFrame && !zcs->frameEnded))
+       && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {
+        CHECK_F( ZSTDMT_createCompressionJob(zcs, srcSize, endFrame) );
+    }
+
+    /* check if there is any data available to flush */
+    DEBUGLOG(5, "zcs->doneJobID : %u  ; zcs->nextJobID : %u ", zcs->doneJobID, zcs->nextJobID);
+    return ZSTDMT_flushNextJob(zcs, output, 1);
+}
+
+
+size_t ZSTDMT_flushStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
+{
+    if (zcs->nbThreads==1) return ZSTD_flushStream(zcs->cstream, output);
+    return ZSTDMT_flushStream_internal(zcs, output, 0);
+}
+
+size_t ZSTDMT_endStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
+{
+    if (zcs->nbThreads==1) return ZSTD_endStream(zcs->cstream, output);
+    return ZSTDMT_flushStream_internal(zcs, output, 1);
+}
diff --git a/thirdparty/zstd/compress/zstdmt_compress.h b/thirdparty/zstd/compress/zstdmt_compress.h
new file mode 100644
index 0000000000..27f78ee031
--- /dev/null
+++ b/thirdparty/zstd/compress/zstdmt_compress.h
@@ -0,0 +1,78 @@
+/**
+ * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under the BSD-style license found in the
+ * LICENSE file in the root directory of this source tree. An additional grant
+ * of patent rights can be found in the PATENTS file in the same directory.
+ */
+
+ #ifndef ZSTDMT_COMPRESS_H
+ #define ZSTDMT_COMPRESS_H
+
+ #if defined (__cplusplus)
+ extern "C" {
+ #endif
+
+
+/* Note : All prototypes defined in this file shall be considered experimental.
+ *        There is no guarantee of API continuity (yet) on any of these prototypes */
+
+/* ===   Dependencies   === */
+#include <stddef.h>   /* size_t */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters */
+#include "zstd.h"     /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
+
+
+/* ===   Simple one-pass functions   === */
+
+typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
+ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbThreads);
+ZSTDLIB_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* cctx);
+
+ZSTDLIB_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* cctx,
+                           void* dst, size_t dstCapacity,
+                     const void* src, size_t srcSize,
+                           int compressionLevel);
+
+
+/* ===   Streaming functions   === */
+
+ZSTDLIB_API size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel);
+ZSTDLIB_API size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize);    /**< pledgedSrcSize is optional and can be zero == unknown */
+
+ZSTDLIB_API size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
+
+ZSTDLIB_API size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);   /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);     /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
+
+
+/* ===   Advanced functions and parameters  === */
+
+#ifndef ZSTDMT_SECTION_SIZE_MIN
+#  define ZSTDMT_SECTION_SIZE_MIN (1U << 20)   /* 1 MB - Minimum size of each compression job */
+#endif
+
+ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx, const void* dict, size_t dictSize,  /**< dict can be released after init, a local copy is preserved within zcs */
+                                          ZSTD_parameters params, unsigned long long pledgedSrcSize);  /**< pledgedSrcSize is optional and can be zero == unknown */
+
+/* ZSDTMT_parameter :
+ * List of parameters that can be set using ZSTDMT_setMTCtxParameter() */
+typedef enum {
+    ZSTDMT_p_sectionSize,        /* size of input "section". Each section is compressed in parallel. 0 means default, which is dynamically determined within compression functions */
+    ZSTDMT_p_overlapSectionLog   /* Log of overlapped section; 0 == no overlap, 6(default) == use 1/8th of window, >=9 == use full window */
+} ZSDTMT_parameter;
+
+/* ZSTDMT_setMTCtxParameter() :
+ * allow setting individual parameters, one at a time, among a list of enums defined in ZSTDMT_parameter.
+ * The function must be called typically after ZSTD_createCCtx().
+ * Parameters not explicitly reset by ZSTDMT_init*() remain the same in consecutive compression sessions.
+ * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter, unsigned value);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif   /* ZSTDMT_COMPRESS_H */