[Minor] Update bundled zstd to version 1.3

28 files changed, 5184 insertions, 2381 deletions

diff --git a/contrib/zstd/CMakeLists.txt b/contrib/zstd/CMakeLists.txt
index 953839ba7..e3c283c61 100644
--- a/contrib/zstd/CMakeLists.txt
+++ b/contrib/zstd/CMakeLists.txt
@@ -1,11 +1,14 @@
-SET(ZSTDSRC	divsufsort.c
+SET(ZSTDSRC
+			cover.c
+			divsufsort.c
 			entropy_common.c
+			error_private.c
 			fse_compress.c
 			fse_decompress.c
 			huf_compress.c
 			huf_decompress.c
-			zbuff_compress.c
-			zbuff_decompress.c
+			pool.c
+			threading.c
 			zdict.c
 			zstd_common.c
 			zstd_compress.c
diff --git a/contrib/zstd/bitstream.h b/contrib/zstd/bitstream.h
index 9f33c8865..61f45328f 100644
--- a/contrib/zstd/bitstream.h
+++ b/contrib/zstd/bitstream.h
@@ -2,7 +2,7 @@
    bitstream
    Part of FSE library
    header file (to include)
-   Copyright (C) 2013-2016, Yann Collet.
+   Copyright (C) 2013-2017, Yann Collet.
 
    BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
 
@@ -39,7 +39,6 @@
 extern "C" {
 #endif
 
-
 /*
 *  This API consists of small unitary functions, which must be inlined for best performance.
 *  Since link-time-optimization is not available for all compilers,
@@ -53,6 +52,16 @@ extern "C" {
 #include "error_private.h"  /* error codes and messages */
 
 
+/*-*************************************
+*  Debug
+***************************************/
+#if defined(BIT_DEBUG) && (BIT_DEBUG>=1)
+#  include <assert.h>
+#else
+#  define assert(condition) ((void)0)
+#endif
+
+
 /*=========================================
 *  Target specific
 =========================================*/
@@ -60,6 +69,9 @@ extern "C" {
 #  include <immintrin.h>   /* support for bextr (experimental) */
 #endif
 
+#define STREAM_ACCUMULATOR_MIN_32  25
+#define STREAM_ACCUMULATOR_MIN_64  57
+#define STREAM_ACCUMULATOR_MIN    ((U32)(MEM_32bits() ? STREAM_ACCUMULATOR_MIN_32 : STREAM_ACCUMULATOR_MIN_64))
 
 /*-******************************************
 *  bitStream encoding API (write forward)
@@ -71,7 +83,7 @@ extern "C" {
 typedef struct
 {
     size_t bitContainer;
-    int    bitPos;
+    unsigned bitPos;
     char*  startPtr;
     char*  ptr;
     char*  endPtr;
@@ -109,6 +121,7 @@ typedef struct
     unsigned bitsConsumed;
     const char* ptr;
     const char* start;
+    const char* limitPtr;
 } BIT_DStream_t;
 
 typedef enum { BIT_DStream_unfinished = 0,
@@ -160,7 +173,10 @@ MEM_STATIC unsigned BIT_highbit32 (register U32 val)
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
     return 31 - __builtin_clz (val);
 #   else   /* Software version */
-    static const unsigned DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+    static const unsigned DeBruijnClz[32] = { 0,  9,  1, 10, 13, 21,  2, 29,
+                                             11, 14, 16, 18, 22, 25,  3, 30,
+                                              8, 12, 20, 28, 15, 17, 24,  7,
+                                             19, 27, 23,  6, 26,  5,  4, 31 };
     U32 v = val;
     v |= v >> 1;
     v |= v >> 2;
@@ -172,31 +188,36 @@ MEM_STATIC unsigned BIT_highbit32 (register U32 val)
 }
 
 /*=====    Local Constants   =====*/
-static const unsigned BIT_mask[] = { 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,  0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF };   /* up to 26 bits */
+static const unsigned BIT_mask[] = { 0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F,
+                                    0xFF, 0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF,
+                                    0xFFFF, 0x1FFFF, 0x3FFFF, 0x7FFFF, 0xFFFFF, 0x1FFFFF, 0x3FFFFF, 0x7FFFFF,
+                                    0xFFFFFF, 0x1FFFFFF, 0x3FFFFFF };   /* up to 26 bits */
 
 
 /*-**************************************************************
 *  bitStream encoding
 ****************************************************************/
 /*! BIT_initCStream() :
- *  `dstCapacity` must be > sizeof(void*)
+ *  `dstCapacity` must be > sizeof(size_t)
  *  @return : 0 if success,
               otherwise an error code (can be tested using ERR_isError() ) */
-MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC, void* startPtr, size_t dstCapacity)
+MEM_STATIC size_t BIT_initCStream(BIT_CStream_t* bitC,
+                                  void* startPtr, size_t dstCapacity)
 {
     bitC->bitContainer = 0;
     bitC->bitPos = 0;
     bitC->startPtr = (char*)startPtr;
     bitC->ptr = bitC->startPtr;
-    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->ptr);
-    if (dstCapacity <= sizeof(bitC->ptr)) return ERROR(dstSize_tooSmall);
+    bitC->endPtr = bitC->startPtr + dstCapacity - sizeof(bitC->bitContainer);
+    if (dstCapacity <= sizeof(bitC->bitContainer)) return ERROR(dstSize_tooSmall);
     return 0;
 }
 
 /*! BIT_addBits() :
     can add up to 26 bits into `bitC`.
     Does not check for register overflow ! */
-MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC, size_t value, unsigned nbBits)
+MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
+                            size_t value, unsigned nbBits)
 {
     bitC->bitContainer |= (value & BIT_mask[nbBits]) << bitC->bitPos;
     bitC->bitPos += nbBits;
@@ -204,34 +225,42 @@ MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC, size_t value, unsigned nbBits)
 
 /*! BIT_addBitsFast() :
  *  works only if `value` is _clean_, meaning all high bits above nbBits are 0 */
-MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC, size_t value, unsigned nbBits)
+MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
+                                size_t value, unsigned nbBits)
 {
+    assert((value>>nbBits) == 0);
     bitC->bitContainer |= value << bitC->bitPos;
     bitC->bitPos += nbBits;
 }
 
 /*! BIT_flushBitsFast() :
+ *  assumption : bitContainer has not overflowed
  *  unsafe version; does not check buffer overflow */
 MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC)
 {
     size_t const nbBytes = bitC->bitPos >> 3;
+    assert( bitC->bitPos <= (sizeof(bitC->bitContainer)*8) );
     MEM_writeLEST(bitC->ptr, bitC->bitContainer);
     bitC->ptr += nbBytes;
+    assert(bitC->ptr <= bitC->endPtr);
     bitC->bitPos &= 7;
-    bitC->bitContainer >>= nbBytes*8;   /* if bitPos >= sizeof(bitContainer)*8 --> undefined behavior */
+    bitC->bitContainer >>= nbBytes*8;
 }
 
 /*! BIT_flushBits() :
+ *  assumption : bitContainer has not overflowed
  *  safe version; check for buffer overflow, and prevents it.
- *  note : does not signal buffer overflow. This will be revealed later on using BIT_closeCStream() */
+ *  note : does not signal buffer overflow.
+ *  overflow will be revealed later on using BIT_closeCStream() */
 MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
 {
     size_t const nbBytes = bitC->bitPos >> 3;
+    assert( bitC->bitPos <= (sizeof(bitC->bitContainer)*8) );
     MEM_writeLEST(bitC->ptr, bitC->bitContainer);
     bitC->ptr += nbBytes;
     if (bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
     bitC->bitPos &= 7;
-    bitC->bitContainer >>= nbBytes*8;   /* if bitPos >= sizeof(bitContainer)*8 --> undefined behavior */
+    bitC->bitContainer >>= nbBytes*8;
 }
 
 /*! BIT_closeCStream() :
@@ -241,9 +270,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC)
 {
     BIT_addBitsFast(bitC, 1, 1);   /* endMark */
     BIT_flushBits(bitC);
-
-    if (bitC->ptr >= bitC->endPtr) return 0; /* doesn't fit within authorized budget : cancel */
-
+    if (bitC->ptr >= bitC->endPtr) return 0; /* overflow detected */
     return (bitC->ptr - bitC->startPtr) + (bitC->bitPos > 0);
 }
 
@@ -261,26 +288,39 @@ MEM_STATIC size_t BIT_initDStream(BIT_DStream_t* bitD, const void* srcBuffer, si
 {
     if (srcSize < 1) { memset(bitD, 0, sizeof(*bitD)); return ERROR(srcSize_wrong); }
 
+    bitD->start = (const char*)srcBuffer;
+    bitD->limitPtr = bitD->start + sizeof(bitD->bitContainer);
+
     if (srcSize >=  sizeof(bitD->bitContainer)) {  /* normal case */
-        bitD->start = (const char*)srcBuffer;
         bitD->ptr   = (const char*)srcBuffer + srcSize - sizeof(bitD->bitContainer);
         bitD->bitContainer = MEM_readLEST(bitD->ptr);
         { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
-          bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
+          bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;  /* ensures bitsConsumed is always set */
           if (lastByte == 0) return ERROR(GENERIC); /* endMark not present */ }
     } else {
-        bitD->start = (const char*)srcBuffer;
         bitD->ptr   = bitD->start;
         bitD->bitContainer = *(const BYTE*)(bitD->start);
         switch(srcSize)
         {
-            case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
-            case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
-            case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
-            case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
-            case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
-            case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
-            default:;
+	    case 7: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[6]) << (sizeof(bitD->bitContainer)*8 - 16);
+	            /* fall-through */
+
+	    case 6: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[5]) << (sizeof(bitD->bitContainer)*8 - 24);
+	            /* fall-through */
+
+	    case 5: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[4]) << (sizeof(bitD->bitContainer)*8 - 32);
+	            /* fall-through */
+
+	    case 4: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[3]) << 24;
+	            /* fall-through */
+
+	    case 3: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[2]) << 16;
+	            /* fall-through */
+
+	    case 2: bitD->bitContainer += (size_t)(((const BYTE*)(srcBuffer))[1]) <<  8;
+	            /* fall-through */
+
+            default: break;
         }
         { BYTE const lastByte = ((const BYTE*)srcBuffer)[srcSize-1];
           bitD->bitsConsumed = lastByte ? 8 - BIT_highbit32(lastByte) : 0;
@@ -298,7 +338,7 @@ MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
 
 MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
 {
-#if defined(__BMI__) && defined(__GNUC__) && !defined(__llvm__)  /* experimental */
+#if defined(__BMI__) && defined(__GNUC__) && __GNUC__*1000+__GNUC_MINOR__ >= 4008  /* experimental */
 #  if defined(__x86_64__)
     if (sizeof(bitContainer)==8)
         return _bextr_u64(bitContainer, start, nbBits);
@@ -327,17 +367,18 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
 #if defined(__BMI__) && defined(__GNUC__)   /* experimental; fails if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8 */
     return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
 #else
-    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
-    return ((bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> 1) >> ((bitMask-nbBits) & bitMask);
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
 #endif
 }
 
 /*! BIT_lookBitsFast() :
-*   unsafe version; only works only if nbBits >= 1 */
+ *  unsafe version; only works if nbBits >= 1 */
 MEM_STATIC size_t BIT_lookBitsFast(const BIT_DStream_t* bitD, U32 nbBits)
 {
-    U32 const bitMask = sizeof(bitD->bitContainer)*8 - 1;
-    return (bitD->bitContainer << (bitD->bitsConsumed & bitMask)) >> (((bitMask+1)-nbBits) & bitMask);
+    U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
+    assert(nbBits >= 1);
+    return (bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> (((regMask+1)-nbBits) & regMask);
 }
 
 MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
@@ -362,21 +403,22 @@ MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
 {
     size_t const value = BIT_lookBitsFast(bitD, nbBits);
+    assert(nbBits >= 1);
     BIT_skipBits(bitD, nbBits);
     return value;
 }
 
 /*! BIT_reloadDStream() :
-*   Refill `BIT_DStream_t` from src buffer previously defined (see BIT_initDStream() ).
+*   Refill `bitD` from buffer previously set in BIT_initDStream() .
 *   This function is safe, it guarantees it will not read beyond src buffer.
 *   @return : status of `BIT_DStream_t` internal register.
-              if status == unfinished, internal register is filled with >= (sizeof(bitD->bitContainer)*8 - 7) bits */
+              if status == BIT_DStream_unfinished, internal register is filled with >= (sizeof(bitD->bitContainer)*8 - 7) bits */
 MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
 {
-	if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* should not happen => corruption detected */
-		return BIT_DStream_overflow;
+    if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
+        return BIT_DStream_overflow;
 
-    if (bitD->ptr >= bitD->start + sizeof(bitD->bitContainer)) {
+    if (bitD->ptr >= bitD->limitPtr) {
         bitD->ptr -= bitD->bitsConsumed >> 3;
         bitD->bitsConsumed &= 7;
         bitD->bitContainer = MEM_readLEST(bitD->ptr);
@@ -386,6 +428,7 @@ MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
         if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
         return BIT_DStream_completed;
     }
+    /* start < ptr < limitPtr */
     {   U32 nbBytes = bitD->bitsConsumed >> 3;
         BIT_DStream_status result = BIT_DStream_unfinished;
         if (bitD->ptr - nbBytes < bitD->start) {
@@ -394,7 +437,7 @@ MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
         }
         bitD->ptr -= nbBytes;
         bitD->bitsConsumed -= nbBytes*8;
-        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD) */
+        bitD->bitContainer = MEM_readLEST(bitD->ptr);   /* reminder : srcSize > sizeof(bitD->bitContainer), otherwise bitD->ptr == bitD->start */
         return result;
     }
 }
diff --git a/contrib/zstd/cover.c b/contrib/zstd/cover.c
new file mode 100644
index 000000000..1863c8f34
--- /dev/null
+++ b/contrib/zstd/cover.c
@@ -0,0 +1,1050 @@
+/**
+ * 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.
+ */
+
+/* *****************************************************************************
+ * Constructs a dictionary using a heuristic based on the following paper:
+ *
+ * Liao, Petri, Moffat, Wirth
+ * Effective Construction of Relative Lempel-Ziv Dictionaries
+ * Published in WWW 2016.
+ *
+ * Adapted from code originally written by @ot (Giuseppe Ottaviano).
+ ******************************************************************************/
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include <stdio.h>  /* fprintf */
+#include <stdlib.h> /* malloc, free, qsort */
+#include <string.h> /* memset */
+#include <time.h>   /* clock */
+
+#include "mem.h" /* read */
+#include "pool.h"
+#include "threading.h"
+#include "zstd_internal.h" /* includes zstd.h */
+#ifndef ZDICT_STATIC_LINKING_ONLY
+#define ZDICT_STATIC_LINKING_ONLY
+#endif
+#include "zdict.h"
+
+/*-*************************************
+*  Constants
+***************************************/
+#define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((U32)-1) : ((U32)1 GB))
+
+/*-*************************************
+*  Console display
+***************************************/
+static int g_displayLevel = 2;
+#define DISPLAY(...)                                                           \
+  {                                                                            \
+    fprintf(stderr, __VA_ARGS__);                                              \
+    fflush(stderr);                                                            \
+  }
+#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
+  if (displayLevel >= l) {                                                     \
+    DISPLAY(__VA_ARGS__);                                                      \
+  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
+#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
+
+#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
+  if (displayLevel >= l) {                                                     \
+    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
+      g_time = clock();                                                        \
+      DISPLAY(__VA_ARGS__);                                                    \
+    }                                                                          \
+  }
+#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
+static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
+static clock_t g_time = 0;
+
+/*-*************************************
+* Hash table
+***************************************
+* A small specialized hash map for storing activeDmers.
+* The map does not resize, so if it becomes full it will loop forever.
+* Thus, the map must be large enough to store every value.
+* The map implements linear probing and keeps its load less than 0.5.
+*/
+
+#define MAP_EMPTY_VALUE ((U32)-1)
+typedef struct COVER_map_pair_t_s {
+  U32 key;
+  U32 value;
+} COVER_map_pair_t;
+
+typedef struct COVER_map_s {
+  COVER_map_pair_t *data;
+  U32 sizeLog;
+  U32 size;
+  U32 sizeMask;
+} COVER_map_t;
+
+/**
+ * Clear the map.
+ */
+static void COVER_map_clear(COVER_map_t *map) {
+  memset(map->data, MAP_EMPTY_VALUE, map->size * sizeof(COVER_map_pair_t));
+}
+
+/**
+ * Initializes a map of the given size.
+ * Returns 1 on success and 0 on failure.
+ * The map must be destroyed with COVER_map_destroy().
+ * The map is only guaranteed to be large enough to hold size elements.
+ */
+static int COVER_map_init(COVER_map_t *map, U32 size) {
+  map->sizeLog = ZSTD_highbit32(size) + 2;
+  map->size = (U32)1 << map->sizeLog;
+  map->sizeMask = map->size - 1;
+  map->data = (COVER_map_pair_t *)malloc(map->size * sizeof(COVER_map_pair_t));
+  if (!map->data) {
+    map->sizeLog = 0;
+    map->size = 0;
+    return 0;
+  }
+  COVER_map_clear(map);
+  return 1;
+}
+
+/**
+ * Internal hash function
+ */
+static const U32 prime4bytes = 2654435761U;
+static U32 COVER_map_hash(COVER_map_t *map, U32 key) {
+  return (key * prime4bytes) >> (32 - map->sizeLog);
+}
+
+/**
+ * Helper function that returns the index that a key should be placed into.
+ */
+static U32 COVER_map_index(COVER_map_t *map, U32 key) {
+  const U32 hash = COVER_map_hash(map, key);
+  U32 i;
+  for (i = hash;; i = (i + 1) & map->sizeMask) {
+    COVER_map_pair_t *pos = &map->data[i];
+    if (pos->value == MAP_EMPTY_VALUE) {
+      return i;
+    }
+    if (pos->key == key) {
+      return i;
+    }
+  }
+}
+
+/**
+ * Returns the pointer to the value for key.
+ * If key is not in the map, it is inserted and the value is set to 0.
+ * The map must not be full.
+ */
+static U32 *COVER_map_at(COVER_map_t *map, U32 key) {
+  COVER_map_pair_t *pos = &map->data[COVER_map_index(map, key)];
+  if (pos->value == MAP_EMPTY_VALUE) {
+    pos->key = key;
+    pos->value = 0;
+  }
+  return &pos->value;
+}
+
+/**
+ * Deletes key from the map if present.
+ */
+static void COVER_map_remove(COVER_map_t *map, U32 key) {
+  U32 i = COVER_map_index(map, key);
+  COVER_map_pair_t *del = &map->data[i];
+  U32 shift = 1;
+  if (del->value == MAP_EMPTY_VALUE) {
+    return;
+  }
+  for (i = (i + 1) & map->sizeMask;; i = (i + 1) & map->sizeMask) {
+    COVER_map_pair_t *const pos = &map->data[i];
+    /* If the position is empty we are done */
+    if (pos->value == MAP_EMPTY_VALUE) {
+      del->value = MAP_EMPTY_VALUE;
+      return;
+    }
+    /* If pos can be moved to del do so */
+    if (((i - COVER_map_hash(map, pos->key)) & map->sizeMask) >= shift) {
+      del->key = pos->key;
+      del->value = pos->value;
+      del = pos;
+      shift = 1;
+    } else {
+      ++shift;
+    }
+  }
+}
+
+/**
+ * Destroyes a map that is inited with COVER_map_init().
+ */
+static void COVER_map_destroy(COVER_map_t *map) {
+  if (map->data) {
+    free(map->data);
+  }
+  map->data = NULL;
+  map->size = 0;
+}
+
+/*-*************************************
+* Context
+***************************************/
+
+typedef struct {
+  const BYTE *samples;
+  size_t *offsets;
+  const size_t *samplesSizes;
+  size_t nbSamples;
+  U32 *suffix;
+  size_t suffixSize;
+  U32 *freqs;
+  U32 *dmerAt;
+  unsigned d;
+} COVER_ctx_t;
+
+/* We need a global context for qsort... */
+static COVER_ctx_t *g_ctx = NULL;
+
+/*-*************************************
+*  Helper functions
+***************************************/
+
+/**
+ * Returns the sum of the sample sizes.
+ */
+static size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
+  size_t sum = 0;
+  size_t i;
+  for (i = 0; i < nbSamples; ++i) {
+    sum += samplesSizes[i];
+  }
+  return sum;
+}
+
+/**
+ * Returns -1 if the dmer at lp is less than the dmer at rp.
+ * Return 0 if the dmers at lp and rp are equal.
+ * Returns 1 if the dmer at lp is greater than the dmer at rp.
+ */
+static int COVER_cmp(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U32 const lhs = *(U32 const *)lp;
+  U32 const rhs = *(U32 const *)rp;
+  return memcmp(ctx->samples + lhs, ctx->samples + rhs, ctx->d);
+}
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
+  U64 const mask = (ctx->d == 8) ? (U64)-1 : (((U64)1 << (8 * ctx->d)) - 1);
+  U64 const lhs = MEM_readLE64(ctx->samples + *(U32 const *)lp) & mask;
+  U64 const rhs = MEM_readLE64(ctx->samples + *(U32 const *)rp) & mask;
+  if (lhs < rhs) {
+    return -1;
+  }
+  return (lhs > rhs);
+}
+
+/**
+ * Same as COVER_cmp() except ties are broken by pointer value
+ * NOTE: g_ctx must be set to call this function.  A global is required because
+ * qsort doesn't take an opaque pointer.
+ */
+static int COVER_strict_cmp(const void *lp, const void *rp) {
+  int result = COVER_cmp(g_ctx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
+/**
+ * Faster version for d <= 8.
+ */
+static int COVER_strict_cmp8(const void *lp, const void *rp) {
+  int result = COVER_cmp8(g_ctx, lp, rp);
+  if (result == 0) {
+    result = lp < rp ? -1 : 1;
+  }
+  return result;
+}
+
+/**
+ * Returns the first pointer in [first, last) whose element does not compare
+ * less than value.  If no such element exists it returns last.
+ */
+static const size_t *COVER_lower_bound(const size_t *first, const size_t *last,
+                                       size_t value) {
+  size_t count = last - first;
+  while (count != 0) {
+    size_t step = count / 2;
+    const size_t *ptr = first;
+    ptr += step;
+    if (*ptr < value) {
+      first = ++ptr;
+      count -= step + 1;
+    } else {
+      count = step;
+    }
+  }
+  return first;
+}
+
+/**
+ * Generic groupBy function.
+ * Groups an array sorted by cmp into groups with equivalent values.
+ * Calls grp for each group.
+ */
+static void
+COVER_groupBy(const void *data, size_t count, size_t size, COVER_ctx_t *ctx,
+              int (*cmp)(COVER_ctx_t *, const void *, const void *),
+              void (*grp)(COVER_ctx_t *, const void *, const void *)) {
+  const BYTE *ptr = (const BYTE *)data;
+  size_t num = 0;
+  while (num < count) {
+    const BYTE *grpEnd = ptr + size;
+    ++num;
+    while (num < count && cmp(ctx, ptr, grpEnd) == 0) {
+      grpEnd += size;
+      ++num;
+    }
+    grp(ctx, ptr, grpEnd);
+    ptr = grpEnd;
+  }
+}
+
+/*-*************************************
+*  Cover functions
+***************************************/
+
+/**
+ * Called on each group of positions with the same dmer.
+ * Counts the frequency of each dmer and saves it in the suffix array.
+ * Fills `ctx->dmerAt`.
+ */
+static void COVER_group(COVER_ctx_t *ctx, const void *group,
+                        const void *groupEnd) {
+  /* The group consists of all the positions with the same first d bytes. */
+  const U32 *grpPtr = (const U32 *)group;
+  const U32 *grpEnd = (const U32 *)groupEnd;
+  /* The dmerId is how we will reference this dmer.
+   * This allows us to map the whole dmer space to a much smaller space, the
+   * size of the suffix array.
+   */
+  const U32 dmerId = (U32)(grpPtr - ctx->suffix);
+  /* Count the number of samples this dmer shows up in */
+  U32 freq = 0;
+  /* Details */
+  const size_t *curOffsetPtr = ctx->offsets;
+  const size_t *offsetsEnd = ctx->offsets + ctx->nbSamples;
+  /* Once *grpPtr >= curSampleEnd this occurrence of the dmer is in a
+   * different sample than the last.
+   */
+  size_t curSampleEnd = ctx->offsets[0];
+  for (; grpPtr != grpEnd; ++grpPtr) {
+    /* Save the dmerId for this position so we can get back to it. */
+    ctx->dmerAt[*grpPtr] = dmerId;
+    /* Dictionaries only help for the first reference to the dmer.
+     * After that zstd can reference the match from the previous reference.
+     * So only count each dmer once for each sample it is in.
+     */
+    if (*grpPtr < curSampleEnd) {
+      continue;
+    }
+    freq += 1;
+    /* Binary search to find the end of the sample *grpPtr is in.
+     * In the common case that grpPtr + 1 == grpEnd we can skip the binary
+     * search because the loop is over.
+     */
+    if (grpPtr + 1 != grpEnd) {
+      const size_t *sampleEndPtr =
+          COVER_lower_bound(curOffsetPtr, offsetsEnd, *grpPtr);
+      curSampleEnd = *sampleEndPtr;
+      curOffsetPtr = sampleEndPtr + 1;
+    }
+  }
+  /* At this point we are never going to look at this segment of the suffix
+   * array again.  We take advantage of this fact to save memory.
+   * We store the frequency of the dmer in the first position of the group,
+   * which is dmerId.
+   */
+  ctx->suffix[dmerId] = freq;
+}
+
+/**
+ * A segment is a range in the source as well as the score of the segment.
+ */
+typedef struct {
+  U32 begin;
+  U32 end;
+  double score;
+} COVER_segment_t;
+
+/**
+ * Selects the best segment in an epoch.
+ * Segments of are scored according to the function:
+ *
+ * Let F(d) be the frequency of dmer d.
+ * Let S_i be the dmer at position i of segment S which has length k.
+ *
+ *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
+ *
+ * Once the dmer d is in the dictionay we set F(d) = 0.
+ */
+static COVER_segment_t COVER_selectSegment(const COVER_ctx_t *ctx, U32 *freqs,
+                                           COVER_map_t *activeDmers, U32 begin,
+                                           U32 end, COVER_params_t parameters) {
+  /* Constants */
+  const U32 k = parameters.k;
+  const U32 d = parameters.d;
+  const U32 dmersInK = k - d + 1;
+  /* Try each segment (activeSegment) and save the best (bestSegment) */
+  COVER_segment_t bestSegment = {0, 0, 0};
+  COVER_segment_t activeSegment;
+  /* Reset the activeDmers in the segment */
+  COVER_map_clear(activeDmers);
+  /* The activeSegment starts at the beginning of the epoch. */
+  activeSegment.begin = begin;
+  activeSegment.end = begin;
+  activeSegment.score = 0;
+  /* Slide the activeSegment through the whole epoch.
+   * Save the best segment in bestSegment.
+   */
+  while (activeSegment.end < end) {
+    /* The dmerId for the dmer at the next position */
+    U32 newDmer = ctx->dmerAt[activeSegment.end];
+    /* The entry in activeDmers for this dmerId */
+    U32 *newDmerOcc = COVER_map_at(activeDmers, newDmer);
+    /* If the dmer isn't already present in the segment add its score. */
+    if (*newDmerOcc == 0) {
+      /* The paper suggest using the L-0.5 norm, but experiments show that it
+       * doesn't help.
+       */
+      activeSegment.score += freqs[newDmer];
+    }
+    /* Add the dmer to the segment */
+    activeSegment.end += 1;
+    *newDmerOcc += 1;
+
+    /* If the window is now too large, drop the first position */
+    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
+      U32 delDmer = ctx->dmerAt[activeSegment.begin];
+      U32 *delDmerOcc = COVER_map_at(activeDmers, delDmer);
+      activeSegment.begin += 1;
+      *delDmerOcc -= 1;
+      /* If this is the last occurence of the dmer, subtract its score */
+      if (*delDmerOcc == 0) {
+        COVER_map_remove(activeDmers, delDmer);
+        activeSegment.score -= freqs[delDmer];
+      }
+    }
+
+    /* If this segment is the best so far save it */
+    if (activeSegment.score > bestSegment.score) {
+      bestSegment = activeSegment;
+    }
+  }
+  {
+    /* Trim off the zero frequency head and tail from the segment. */
+    U32 newBegin = bestSegment.end;
+    U32 newEnd = bestSegment.begin;
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      U32 freq = freqs[ctx->dmerAt[pos]];
+      if (freq != 0) {
+        newBegin = MIN(newBegin, pos);
+        newEnd = pos + 1;
+      }
+    }
+    bestSegment.begin = newBegin;
+    bestSegment.end = newEnd;
+  }
+  {
+    /* Zero out the frequency of each dmer covered by the chosen segment. */
+    U32 pos;
+    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
+      freqs[ctx->dmerAt[pos]] = 0;
+    }
+  }
+  return bestSegment;
+}
+
+/**
+ * Check the validity of the parameters.
+ * Returns non-zero if the parameters are valid and 0 otherwise.
+ */
+static int COVER_checkParameters(COVER_params_t parameters) {
+  /* k and d are required parameters */
+  if (parameters.d == 0 || parameters.k == 0) {
+    return 0;
+  }
+  /* d <= k */
+  if (parameters.d > parameters.k) {
+    return 0;
+  }
+  return 1;
+}
+
+/**
+ * Clean up a context initialized with `COVER_ctx_init()`.
+ */
+static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
+  if (!ctx) {
+    return;
+  }
+  if (ctx->suffix) {
+    free(ctx->suffix);
+    ctx->suffix = NULL;
+  }
+  if (ctx->freqs) {
+    free(ctx->freqs);
+    ctx->freqs = NULL;
+  }
+  if (ctx->dmerAt) {
+    free(ctx->dmerAt);
+    ctx->dmerAt = NULL;
+  }
+  if (ctx->offsets) {
+    free(ctx->offsets);
+    ctx->offsets = NULL;
+  }
+}
+
+/**
+ * Prepare a context for dictionary building.
+ * The context is only dependent on the parameter `d` and can used multiple
+ * times.
+ * Returns 1 on success or zero on error.
+ * The context must be destroyed with `COVER_ctx_destroy()`.
+ */
+static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
+                          const size_t *samplesSizes, unsigned nbSamples,
+                          unsigned d) {
+  const BYTE *const samples = (const BYTE *)samplesBuffer;
+  const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
+  /* Checks */
+  if (totalSamplesSize < MAX(d, sizeof(U64)) ||
+      totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
+    DISPLAYLEVEL(1, "Total samples size is too large, maximum size is %u MB\n",
+                 (COVER_MAX_SAMPLES_SIZE >> 20));
+    return 0;
+  }
+  /* Zero the context */
+  memset(ctx, 0, sizeof(*ctx));
+  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbSamples,
+               (U32)totalSamplesSize);
+  ctx->samples = samples;
+  ctx->samplesSizes = samplesSizes;
+  ctx->nbSamples = nbSamples;
+  /* Partial suffix array */
+  ctx->suffixSize = totalSamplesSize - MAX(d, sizeof(U64)) + 1;
+  ctx->suffix = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  /* Maps index to the dmerID */
+  ctx->dmerAt = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  /* The offsets of each file */
+  ctx->offsets = (size_t *)malloc((nbSamples + 1) * sizeof(size_t));
+  if (!ctx->suffix || !ctx->dmerAt || !ctx->offsets) {
+    DISPLAYLEVEL(1, "Failed to allocate scratch buffers\n");
+    COVER_ctx_destroy(ctx);
+    return 0;
+  }
+  ctx->freqs = NULL;
+  ctx->d = d;
+
+  /* Fill offsets from the samlesSizes */
+  {
+    U32 i;
+    ctx->offsets[0] = 0;
+    for (i = 1; i <= nbSamples; ++i) {
+      ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
+    }
+  }
+  DISPLAYLEVEL(2, "Constructing partial suffix array\n");
+  {
+    /* suffix is a partial suffix array.
+     * It only sorts suffixes by their first parameters.d bytes.
+     * The sort is stable, so each dmer group is sorted by position in input.
+     */
+    U32 i;
+    for (i = 0; i < ctx->suffixSize; ++i) {
+      ctx->suffix[i] = i;
+    }
+    /* qsort doesn't take an opaque pointer, so pass as a global */
+    g_ctx = ctx;
+    qsort(ctx->suffix, ctx->suffixSize, sizeof(U32),
+          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
+  }
+  DISPLAYLEVEL(2, "Computing frequencies\n");
+  /* For each dmer group (group of positions with the same first d bytes):
+   * 1. For each position we set dmerAt[position] = dmerID.  The dmerID is
+   *    (groupBeginPtr - suffix).  This allows us to go from position to
+   *    dmerID so we can look up values in freq.
+   * 2. We calculate how many samples the dmer occurs in and save it in
+   *    freqs[dmerId].
+   */
+  COVER_groupBy(ctx->suffix, ctx->suffixSize, sizeof(U32), ctx,
+                (ctx->d <= 8 ? &COVER_cmp8 : &COVER_cmp), &COVER_group);
+  ctx->freqs = ctx->suffix;
+  ctx->suffix = NULL;
+  return 1;
+}
+
+/**
+ * Given the prepared context build the dictionary.
+ */
+static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
+                                    COVER_map_t *activeDmers, void *dictBuffer,
+                                    size_t dictBufferCapacity,
+                                    COVER_params_t parameters) {
+  BYTE *const dict = (BYTE *)dictBuffer;
+  size_t tail = dictBufferCapacity;
+  /* Divide the data up into epochs of equal size.
+   * We will select at least one segment from each epoch.
+   */
+  const U32 epochs = (U32)(dictBufferCapacity / parameters.k);
+  const U32 epochSize = (U32)(ctx->suffixSize / epochs);
+  size_t epoch;
+  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n", epochs,
+               epochSize);
+  /* Loop through the epochs until there are no more segments or the dictionary
+   * is full.
+   */
+  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs) {
+    const U32 epochBegin = (U32)(epoch * epochSize);
+    const U32 epochEnd = epochBegin + epochSize;
+    size_t segmentSize;
+    /* Select a segment */
+    COVER_segment_t segment = COVER_selectSegment(
+        ctx, freqs, activeDmers, epochBegin, epochEnd, parameters);
+    /* Trim the segment if necessary and if it is empty then we are done */
+    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
+    if (segmentSize == 0) {
+      break;
+    }
+    /* We fill the dictionary from the back to allow the best segments to be
+     * referenced with the smallest offsets.
+     */
+    tail -= segmentSize;
+    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
+    DISPLAYUPDATE(
+        2, "\r%u%%       ",
+        (U32)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
+  }
+  DISPLAYLEVEL(2, "\r%79s\r", "");
+  return tail;
+}
+
+/**
+ * Translate from COVER_params_t to ZDICT_params_t required for finalizing the
+ * dictionary.
+ */
+static ZDICT_params_t COVER_translateParams(COVER_params_t parameters) {
+  ZDICT_params_t zdictParams;
+  memset(&zdictParams, 0, sizeof(zdictParams));
+  zdictParams.notificationLevel = 1;
+  zdictParams.dictID = parameters.dictID;
+  zdictParams.compressionLevel = parameters.compressionLevel;
+  return zdictParams;
+}
+
+ZDICTLIB_API size_t COVER_trainFromBuffer(
+    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
+    const size_t *samplesSizes, unsigned nbSamples, COVER_params_t parameters) {
+  BYTE *const dict = (BYTE *)dictBuffer;
+  COVER_ctx_t ctx;
+  COVER_map_t activeDmers;
+  /* Checks */
+  if (!COVER_checkParameters(parameters)) {
+    DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+    return ERROR(GENERIC);
+  }
+  if (nbSamples == 0) {
+    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
+    return ERROR(GENERIC);
+  }
+  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                 ZDICT_DICTSIZE_MIN);
+    return ERROR(dstSize_tooSmall);
+  }
+  /* Initialize global data */
+  g_displayLevel = parameters.notificationLevel;
+  /* Initialize context and activeDmers */
+  if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                      parameters.d)) {
+    return ERROR(GENERIC);
+  }
+  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
+    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
+    COVER_ctx_destroy(&ctx);
+    return ERROR(GENERIC);
+  }
+
+  DISPLAYLEVEL(2, "Building dictionary\n");
+  {
+    const size_t tail =
+        COVER_buildDictionary(&ctx, ctx.freqs, &activeDmers, dictBuffer,
+                              dictBufferCapacity, parameters);
+    ZDICT_params_t zdictParams = COVER_translateParams(parameters);
+    const size_t dictionarySize = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        samplesBuffer, samplesSizes, nbSamples, zdictParams);
+    if (!ZSTD_isError(dictionarySize)) {
+      DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
+                   (U32)dictionarySize);
+    }
+    COVER_ctx_destroy(&ctx);
+    COVER_map_destroy(&activeDmers);
+    return dictionarySize;
+  }
+}
+
+/**
+ * COVER_best_t is used for two purposes:
+ * 1. Synchronizing threads.
+ * 2. Saving the best parameters and dictionary.
+ *
+ * All of the methods except COVER_best_init() are thread safe if zstd is
+ * compiled with multithreaded support.
+ */
+typedef struct COVER_best_s {
+  pthread_mutex_t mutex;
+  pthread_cond_t cond;
+  size_t liveJobs;
+  void *dict;
+  size_t dictSize;
+  COVER_params_t parameters;
+  size_t compressedSize;
+} COVER_best_t;
+
+/**
+ * Initialize the `COVER_best_t`.
+ */
+static void COVER_best_init(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  pthread_mutex_init(&best->mutex, NULL);
+  pthread_cond_init(&best->cond, NULL);
+  best->liveJobs = 0;
+  best->dict = NULL;
+  best->dictSize = 0;
+  best->compressedSize = (size_t)-1;
+  memset(&best->parameters, 0, sizeof(best->parameters));
+}
+
+/**
+ * Wait until liveJobs == 0.
+ */
+static void COVER_best_wait(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  pthread_mutex_lock(&best->mutex);
+  while (best->liveJobs != 0) {
+    pthread_cond_wait(&best->cond, &best->mutex);
+  }
+  pthread_mutex_unlock(&best->mutex);
+}
+
+/**
+ * Call COVER_best_wait() and then destroy the COVER_best_t.
+ */
+static void COVER_best_destroy(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  COVER_best_wait(best);
+  if (best->dict) {
+    free(best->dict);
+  }
+  pthread_mutex_destroy(&best->mutex);
+  pthread_cond_destroy(&best->cond);
+}
+
+/**
+ * Called when a thread is about to be launched.
+ * Increments liveJobs.
+ */
+static void COVER_best_start(COVER_best_t *best) {
+  if (!best) {
+    return;
+  }
+  pthread_mutex_lock(&best->mutex);
+  ++best->liveJobs;
+  pthread_mutex_unlock(&best->mutex);
+}
+
+/**
+ * Called when a thread finishes executing, both on error or success.
+ * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
+ * If this dictionary is the best so far save it and its parameters.
+ */
+static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
+                              COVER_params_t parameters, void *dict,
+                              size_t dictSize) {
+  if (!best) {
+    return;
+  }
+  {
+    size_t liveJobs;
+    pthread_mutex_lock(&best->mutex);
+    --best->liveJobs;
+    liveJobs = best->liveJobs;
+    /* If the new dictionary is better */
+    if (compressedSize < best->compressedSize) {
+      /* Allocate space if necessary */
+      if (!best->dict || best->dictSize < dictSize) {
+        if (best->dict) {
+          free(best->dict);
+        }
+        best->dict = malloc(dictSize);
+        if (!best->dict) {
+          best->compressedSize = ERROR(GENERIC);
+          best->dictSize = 0;
+          return;
+        }
+      }
+      /* Save the dictionary, parameters, and size */
+      memcpy(best->dict, dict, dictSize);
+      best->dictSize = dictSize;
+      best->parameters = parameters;
+      best->compressedSize = compressedSize;
+    }
+    pthread_mutex_unlock(&best->mutex);
+    if (liveJobs == 0) {
+      pthread_cond_broadcast(&best->cond);
+    }
+  }
+}
+
+/**
+ * Parameters for COVER_tryParameters().
+ */
+typedef struct COVER_tryParameters_data_s {
+  const COVER_ctx_t *ctx;
+  COVER_best_t *best;
+  size_t dictBufferCapacity;
+  COVER_params_t parameters;
+} COVER_tryParameters_data_t;
+
+/**
+ * Tries a set of parameters and upates the COVER_best_t with the results.
+ * This function is thread safe if zstd is compiled with multithreaded support.
+ * It takes its parameters as an *OWNING* opaque pointer to support threading.
+ */
+static void COVER_tryParameters(void *opaque) {
+  /* Save parameters as local variables */
+  COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t *)opaque;
+  const COVER_ctx_t *const ctx = data->ctx;
+  const COVER_params_t parameters = data->parameters;
+  size_t dictBufferCapacity = data->dictBufferCapacity;
+  size_t totalCompressedSize = ERROR(GENERIC);
+  /* Allocate space for hash table, dict, and freqs */
+  COVER_map_t activeDmers;
+  BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
+  U32 *freqs = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
+  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
+    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
+    goto _cleanup;
+  }
+  if (!dict || !freqs) {
+    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
+    goto _cleanup;
+  }
+  /* Copy the frequencies because we need to modify them */
+  memcpy(freqs, ctx->freqs, ctx->suffixSize * sizeof(U32));
+  /* Build the dictionary */
+  {
+    const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
+                                              dictBufferCapacity, parameters);
+    const ZDICT_params_t zdictParams = COVER_translateParams(parameters);
+    dictBufferCapacity = ZDICT_finalizeDictionary(
+        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
+        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbSamples, zdictParams);
+    if (ZDICT_isError(dictBufferCapacity)) {
+      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
+      goto _cleanup;
+    }
+  }
+  /* Check total compressed size */
+  {
+    /* Pointers */
+    ZSTD_CCtx *cctx;
+    ZSTD_CDict *cdict;
+    void *dst;
+    /* Local variables */
+    size_t dstCapacity;
+    size_t i;
+    /* Allocate dst with enough space to compress the maximum sized sample */
+    {
+      size_t maxSampleSize = 0;
+      for (i = 0; i < ctx->nbSamples; ++i) {
+        maxSampleSize = MAX(ctx->samplesSizes[i], maxSampleSize);
+      }
+      dstCapacity = ZSTD_compressBound(maxSampleSize);
+      dst = malloc(dstCapacity);
+    }
+    /* Create the cctx and cdict */
+    cctx = ZSTD_createCCtx();
+    cdict =
+        ZSTD_createCDict(dict, dictBufferCapacity, parameters.compressionLevel);
+    if (!dst || !cctx || !cdict) {
+      goto _compressCleanup;
+    }
+    /* Compress each sample and sum their sizes (or error) */
+    totalCompressedSize = 0;
+    for (i = 0; i < ctx->nbSamples; ++i) {
+      const size_t size = ZSTD_compress_usingCDict(
+          cctx, dst, dstCapacity, ctx->samples + ctx->offsets[i],
+          ctx->samplesSizes[i], cdict);
+      if (ZSTD_isError(size)) {
+        totalCompressedSize = ERROR(GENERIC);
+        goto _compressCleanup;
+      }
+      totalCompressedSize += size;
+    }
+  _compressCleanup:
+    ZSTD_freeCCtx(cctx);
+    ZSTD_freeCDict(cdict);
+    if (dst) {
+      free(dst);
+    }
+  }
+
+_cleanup:
+  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
+                    dictBufferCapacity);
+  free(data);
+  COVER_map_destroy(&activeDmers);
+  if (dict) {
+    free(dict);
+  }
+  if (freqs) {
+    free(freqs);
+  }
+}
+
+ZDICTLIB_API size_t COVER_optimizeTrainFromBuffer(void *dictBuffer,
+                                                  size_t dictBufferCapacity,
+                                                  const void *samplesBuffer,
+                                                  const size_t *samplesSizes,
+                                                  unsigned nbSamples,
+                                                  COVER_params_t *parameters) {
+  /* constants */
+  const unsigned nbThreads = parameters->nbThreads;
+  const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
+  const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
+  const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
+  const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
+  const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
+  const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
+  const unsigned kIterations =
+      (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+  /* Local variables */
+  const int displayLevel = parameters->notificationLevel;
+  unsigned iteration = 1;
+  unsigned d;
+  unsigned k;
+  COVER_best_t best;
+  POOL_ctx *pool = NULL;
+  /* Checks */
+  if (kMinK < kMaxD || kMaxK < kMinK) {
+    LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
+    return ERROR(GENERIC);
+  }
+  if (nbSamples == 0) {
+    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
+    return ERROR(GENERIC);
+  }
+  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
+    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
+                 ZDICT_DICTSIZE_MIN);
+    return ERROR(dstSize_tooSmall);
+  }
+  if (nbThreads > 1) {
+    pool = POOL_create(nbThreads, 1);
+    if (!pool) {
+      return ERROR(memory_allocation);
+    }
+  }
+  /* Initialization */
+  COVER_best_init(&best);
+  /* Turn down global display level to clean up display at level 2 and below */
+  g_displayLevel = parameters->notificationLevel - 1;
+  /* Loop through d first because each new value needs a new context */
+  LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
+                    kIterations);
+  for (d = kMinD; d <= kMaxD; d += 2) {
+    /* Initialize the context for this value of d */
+    COVER_ctx_t ctx;
+    LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
+    if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d)) {
+      LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return ERROR(GENERIC);
+    }
+    /* Loop through k reusing the same context */
+    for (k = kMinK; k <= kMaxK; k += kStepSize) {
+      /* Prepare the arguments */
+      COVER_tryParameters_data_t *data = (COVER_tryParameters_data_t *)malloc(
+          sizeof(COVER_tryParameters_data_t));
+      LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
+      if (!data) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
+        COVER_best_destroy(&best);
+        COVER_ctx_destroy(&ctx);
+        POOL_free(pool);
+        return ERROR(GENERIC);
+      }
+      data->ctx = &ctx;
+      data->best = &best;
+      data->dictBufferCapacity = dictBufferCapacity;
+      data->parameters = *parameters;
+      data->parameters.k = k;
+      data->parameters.d = d;
+      data->parameters.steps = kSteps;
+      /* Check the parameters */
+      if (!COVER_checkParameters(data->parameters)) {
+        DISPLAYLEVEL(1, "Cover parameters incorrect\n");
+        free(data);
+        continue;
+      }
+      /* Call the function and pass ownership of data to it */
+      COVER_best_start(&best);
+      if (pool) {
+        POOL_add(pool, &COVER_tryParameters, data);
+      } else {
+        COVER_tryParameters(data);
+      }
+      /* Print status */
+      LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
+                         (U32)((iteration * 100) / kIterations));
+      ++iteration;
+    }
+    COVER_best_wait(&best);
+    COVER_ctx_destroy(&ctx);
+  }
+  LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
+  /* Fill the output buffer and parameters with output of the best parameters */
+  {
+    const size_t dictSize = best.dictSize;
+    if (ZSTD_isError(best.compressedSize)) {
+      const size_t compressedSize = best.compressedSize;
+      COVER_best_destroy(&best);
+      POOL_free(pool);
+      return compressedSize;
+    }
+    *parameters = best.parameters;
+    memcpy(dictBuffer, best.dict, dictSize);
+    COVER_best_destroy(&best);
+    POOL_free(pool);
+    return dictSize;
+  }
+}
diff --git a/contrib/zstd/entropy_common.c b/contrib/zstd/entropy_common.c
index acd966999..b37a082fe 100644
--- a/contrib/zstd/entropy_common.c
+++ b/contrib/zstd/entropy_common.c
@@ -43,27 +43,21 @@
 #include "huf.h"
 
 
-/*-****************************************
-*  FSE Error Management
-******************************************/
-unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+/*===   Version   ===*/
+unsigned FSE_versionNumber(void) { return FSE_VERSION_NUMBER; }
 
-const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
+/*===   Error Management   ===*/
+unsigned FSE_isError(size_t code) { return ERR_isError(code); }
+const char* FSE_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
-/* **************************************************************
-*  HUF Error Management
-****************************************************************/
 unsigned HUF_isError(size_t code) { return ERR_isError(code); }
-
 const char* HUF_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
 
 /*-**************************************************************
 *  FSE NCount encoding-decoding
 ****************************************************************/
-static short FSE_abs(short a) { return (short)(a<0 ? -a : a); }
-
 size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* tableLogPtr,
                  const void* headerBuffer, size_t hbSize)
 {
@@ -117,21 +111,21 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
             } else {
                 bitStream >>= 2;
         }   }
-        {   short const max = (short)((2*threshold-1)-remaining);
-            short count;
+        {   int const max = (2*threshold-1) - remaining;
+            int count;
 
             if ((bitStream & (threshold-1)) < (U32)max) {
-                count = (short)(bitStream & (threshold-1));
-                bitCount   += nbBits-1;
+                count = bitStream & (threshold-1);
+                bitCount += nbBits-1;
             } else {
-                count = (short)(bitStream & (2*threshold-1));
+                count = bitStream & (2*threshold-1);
                 if (count >= threshold) count -= max;
-                bitCount   += nbBits;
+                bitCount += nbBits;
             }
 
             count--;   /* extra accuracy */
-            remaining -= FSE_abs(count);
-            normalizedCounter[charnum++] = count;
+            remaining -= count < 0 ? -count : count;   /* -1 means +1 */
+            normalizedCounter[charnum++] = (short)count;
             previous0 = !count;
             while (remaining < threshold) {
                 nbBits--;
@@ -159,6 +153,7 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
 /*! HUF_readStats() :
     Read compact Huffman tree, saved by HUF_writeCTable().
     `huffWeight` is destination buffer.
+    `rankStats` is assumed to be a table of at least HUF_TABLELOG_MAX U32.
     @return : size read from `src` , or an error Code .
     Note : Needed by HUF_readCTable() and HUF_readDTableX?() .
 */
@@ -168,9 +163,11 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
 {
     U32 weightTotal;
     const BYTE* ip = (const BYTE*) src;
-    size_t iSize = ip[0];
+    size_t iSize;
     size_t oSize;
 
+    if (!srcSize) return ERROR(srcSize_wrong);
+    iSize = ip[0];
     /* memset(huffWeight, 0, hwSize);   *//* is not necessary, even though some analyzer complain ... */
 
     if (iSize >= 128) {  /* special header */
@@ -185,23 +182,25 @@ size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
                 huffWeight[n+1] = ip[n/2] & 15;
     }   }   }
     else  {   /* header compressed with FSE (normal case) */
+        FSE_DTable fseWorkspace[FSE_DTABLE_SIZE_U32(6)];  /* 6 is max possible tableLog for HUF header (maybe even 5, to be tested) */
         if (iSize+1 > srcSize) return ERROR(srcSize_wrong);
-        oSize = FSE_decompress(huffWeight, hwSize-1, ip+1, iSize);   /* max (hwSize-1) values decoded, as last one is implied */
+        oSize = FSE_decompress_wksp(huffWeight, hwSize-1, ip+1, iSize, fseWorkspace, 6);   /* max (hwSize-1) values decoded, as last one is implied */
         if (FSE_isError(oSize)) return oSize;
     }
 
     /* collect weight stats */
-    memset(rankStats, 0, (HUF_TABLELOG_ABSOLUTEMAX + 1) * sizeof(U32));
+    memset(rankStats, 0, (HUF_TABLELOG_MAX + 1) * sizeof(U32));
     weightTotal = 0;
     {   U32 n; for (n=0; n<oSize; n++) {
-            if (huffWeight[n] >= HUF_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+            if (huffWeight[n] >= HUF_TABLELOG_MAX) return ERROR(corruption_detected);
             rankStats[huffWeight[n]]++;
             weightTotal += (1 << huffWeight[n]) >> 1;
     }   }
+    if (weightTotal == 0) return ERROR(corruption_detected);
 
     /* get last non-null symbol weight (implied, total must be 2^n) */
     {   U32 const tableLog = BIT_highbit32(weightTotal) + 1;
-        if (tableLog > HUF_TABLELOG_ABSOLUTEMAX) return ERROR(corruption_detected);
+        if (tableLog > HUF_TABLELOG_MAX) return ERROR(corruption_detected);
         *tableLogPtr = tableLog;
         /* determine last weight */
         {   U32 const total = 1 << tableLog;
diff --git a/contrib/zstd/error_private.c b/contrib/zstd/error_private.c
new file mode 100644
index 000000000..6bc86da7a
--- /dev/null
+++ b/contrib/zstd/error_private.c
@@ -0,0 +1,46 @@
+/**
+ * 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.
+ */
+
+/* The purpose of this file is to have a single list of error strings embedded in binary */
+
+#include "error_private.h"
+
+const char* ERR_getErrorString(ERR_enum code)
+{
+    static const char* const notErrorCode = "Unspecified error code";
+    switch( code )
+    {
+    case PREFIX(no_error): return "No error detected";
+    case PREFIX(GENERIC):  return "Error (generic)";
+    case PREFIX(prefix_unknown): return "Unknown frame descriptor";
+    case PREFIX(version_unsupported): return "Version not supported";
+    case PREFIX(parameter_unknown): return "Unknown parameter type";
+    case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
+    case PREFIX(frameParameter_unsupportedBy32bits): return "Frame parameter unsupported in 32-bits mode";
+    case PREFIX(frameParameter_windowTooLarge): return "Frame requires too much memory for decoding";
+    case PREFIX(compressionParameter_unsupported): return "Compression parameter is out of bound";
+    case PREFIX(init_missing): return "Context should be init first";
+    case PREFIX(memory_allocation): return "Allocation error : not enough memory";
+    case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
+    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
+    case PREFIX(srcSize_wrong): return "Src size is incorrect";
+    case PREFIX(corruption_detected): return "Corrupted block detected";
+    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
+    case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
+    case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
+    case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
+    case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
+    case PREFIX(dictionary_wrong): return "Dictionary mismatch";
+    case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
+    case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
+    case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
+    case PREFIX(maxCode):
+    default: return notErrorCode;
+    }
+}
diff --git a/contrib/zstd/error_private.h b/contrib/zstd/error_private.h
index d27e15af8..1bc2e4954 100644
--- a/contrib/zstd/error_private.h
+++ b/contrib/zstd/error_private.h
@@ -21,7 +21,7 @@ extern "C" {
 *  Dependencies
 ******************************************/
 #include <stddef.h>        /* size_t */
-#include "error_public.h"  /* enum list */
+#include "zstd_errors.h"  /* enum list */
 
 
 /* ****************************************
@@ -62,35 +62,7 @@ ERR_STATIC ERR_enum ERR_getErrorCode(size_t code) { if (!ERR_isError(code)) retu
 *  Error Strings
 ******************************************/
 
-ERR_STATIC const char* ERR_getErrorString(ERR_enum code)
-{
-    static const char* notErrorCode = "Unspecified error code";
-    switch( code )
-    {
-    case PREFIX(no_error): return "No error detected";
-    case PREFIX(GENERIC):  return "Error (generic)";
-    case PREFIX(prefix_unknown): return "Unknown frame descriptor";
-    case PREFIX(version_unsupported): return "Version not supported";
-    case PREFIX(parameter_unknown): return "Unknown parameter type";
-    case PREFIX(frameParameter_unsupported): return "Unsupported frame parameter";
-    case PREFIX(frameParameter_unsupportedBy32bits): return "Frame parameter unsupported in 32-bits mode";
-    case PREFIX(compressionParameter_unsupported): return "Compression parameter is out of bound";
-    case PREFIX(init_missing): return "Context should be init first";
-    case PREFIX(memory_allocation): return "Allocation error : not enough memory";
-    case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
-    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
-    case PREFIX(srcSize_wrong): return "Src size incorrect";
-    case PREFIX(corruption_detected): return "Corrupted block detected";
-    case PREFIX(checksum_wrong): return "Restored data doesn't match checksum";
-    case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
-    case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
-    case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
-    case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
-    case PREFIX(dictionary_wrong): return "Dictionary mismatch";
-    case PREFIX(maxCode):
-    default: return notErrorCode;
-    }
-}
+const char* ERR_getErrorString(ERR_enum code);   /* error_private.c */
 
 ERR_STATIC const char* ERR_getErrorName(size_t code)
 {
diff --git a/contrib/zstd/fse.h b/contrib/zstd/fse.h
index 720d54b11..6d5d41def 100644
--- a/contrib/zstd/fse.h
+++ b/contrib/zstd/fse.h
@@ -45,6 +45,32 @@ extern "C" {
 #include <stddef.h>    /* size_t, ptrdiff_t */
 
 
+/*-*****************************************
+*  FSE_PUBLIC_API : control library symbols visibility
+******************************************/
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define FSE_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define FSE_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define FSE_PUBLIC_API
+#endif
+
+/*------   Version   ------*/
+#define FSE_VERSION_MAJOR    0
+#define FSE_VERSION_MINOR    9
+#define FSE_VERSION_RELEASE  0
+
+#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE
+#define FSE_QUOTE(str) #str
+#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)
+#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)
+
+#define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
+FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
+
 /*-****************************************
 *  FSE simple functions
 ******************************************/
@@ -56,8 +82,8 @@ extern "C" {
                      if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression instead.
                      if FSE_isError(return), compression failed (more details using FSE_getErrorName())
 */
-size_t FSE_compress(void* dst, size_t dstCapacity,
-              const void* src, size_t srcSize);
+FSE_PUBLIC_API size_t FSE_compress(void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
 
 /*! FSE_decompress():
     Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',
@@ -69,18 +95,18 @@ size_t FSE_compress(void* dst, size_t dstCapacity,
     Why ? : making this distinction requires a header.
     Header management is intentionally delegated to the user layer, which can better manage special cases.
 */
-size_t FSE_decompress(void* dst,  size_t dstCapacity,
-                const void* cSrc, size_t cSrcSize);
+FSE_PUBLIC_API size_t FSE_decompress(void* dst,  size_t dstCapacity,
+                               const void* cSrc, size_t cSrcSize);
 
 
 /*-*****************************************
 *  Tool functions
 ******************************************/
-size_t FSE_compressBound(size_t size);       /* maximum compressed size */
+FSE_PUBLIC_API size_t FSE_compressBound(size_t size);       /* maximum compressed size */
 
 /* Error Management */
-unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
-const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
+FSE_PUBLIC_API unsigned    FSE_isError(size_t code);        /* tells if a return value is an error code */
+FSE_PUBLIC_API const char* FSE_getErrorName(size_t code);   /* provides error code string (useful for debugging) */
 
 
 /*-*****************************************
@@ -94,7 +120,7 @@ const char* FSE_getErrorName(size_t code);   /* provides error code string (usef
                      if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression.
                      if FSE_isError(return), it's an error code.
 */
-size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
 
 
 /*-*****************************************
@@ -127,50 +153,50 @@ or to save and provide normalized distribution using external method.
     @return : the count of the most frequent symbol (which is not identified).
               if return == srcSize, there is only one symbol.
               Can also return an error code, which can be tested with FSE_isError(). */
-size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+FSE_PUBLIC_API size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
 
 /*! FSE_optimalTableLog():
     dynamically downsize 'tableLog' when conditions are met.
     It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
     @return : recommended tableLog (necessarily <= 'maxTableLog') */
-unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
+FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
 
 /*! FSE_normalizeCount():
     normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)
     'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
     @return : tableLog,
               or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
 
 /*! FSE_NCountWriteBound():
     Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
     Typically useful for allocation purpose. */
-size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);
 
 /*! FSE_writeNCount():
     Compactly save 'normalizedCounter' into 'buffer'.
     @return : size of the compressed table,
               or an errorCode, which can be tested using FSE_isError(). */
-size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
 
 
 /*! Constructor and Destructor of FSE_CTable.
     Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
 typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
-FSE_CTable* FSE_createCTable (unsigned tableLog, unsigned maxSymbolValue);
-void        FSE_freeCTable (FSE_CTable* ct);
+FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned tableLog, unsigned maxSymbolValue);
+FSE_PUBLIC_API void        FSE_freeCTable (FSE_CTable* ct);
 
 /*! FSE_buildCTable():
     Builds `ct`, which must be already allocated, using FSE_createCTable().
     @return : 0, or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
 
 /*! FSE_compress_usingCTable():
     Compress `src` using `ct` into `dst` which must be already allocated.
     @return : size of compressed data (<= `dstCapacity`),
               or 0 if compressed data could not fit into `dst`,
               or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
+FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);
 
 /*!
 Tutorial :
@@ -223,25 +249,25 @@ If there is an error, the function will return an ErrorCode (which can be tested
     @return : size read from 'rBuffer',
               or an errorCode, which can be tested using FSE_isError().
               maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
-size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
 
 /*! Constructor and Destructor of FSE_DTable.
     Note that its size depends on 'tableLog' */
 typedef unsigned FSE_DTable;   /* don't allocate that. It's just a way to be more restrictive than void* */
-FSE_DTable* FSE_createDTable(unsigned tableLog);
-void        FSE_freeDTable(FSE_DTable* dt);
+FSE_PUBLIC_API FSE_DTable* FSE_createDTable(unsigned tableLog);
+FSE_PUBLIC_API void        FSE_freeDTable(FSE_DTable* dt);
 
 /*! FSE_buildDTable():
     Builds 'dt', which must be already allocated, using FSE_createDTable().
     return : 0, or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
+FSE_PUBLIC_API size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
 
 /*! FSE_decompress_usingDTable():
     Decompress compressed source `cSrc` of size `cSrcSize` using `dt`
     into `dst` which must be already allocated.
     @return : size of regenerated data (necessarily <= `dstCapacity`),
               or an errorCode, which can be tested using FSE_isError() */
-size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
+FSE_PUBLIC_API size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);
 
 /*!
 Tutorial :
@@ -286,45 +312,84 @@ If there is an error, the function will return an error code, which can be teste
 #define FSE_BLOCKBOUND(size) (size + (size>>7))
 #define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
 
-/* It is possible to statically allocate FSE CTable/DTable as a table of unsigned using below macros */
+/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
 #define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue)   (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))
 #define FSE_DTABLE_SIZE_U32(maxTableLog)                   (1 + (1<<maxTableLog))
 
+/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */
+#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue)   (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))
+#define FSE_DTABLE_SIZE(maxTableLog)                   (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))
+
 
 /* *****************************************
 *  FSE advanced API
 *******************************************/
+/* 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);
+
+/** FSE_countFast() :
+ *  same as FSE_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr
+ */
 size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-/**< same as FSE_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr  */
+
+/* FSE_countFast_wksp() :
+ * Same as FSE_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` must be a table of minimum `1024` unsigned
+ */
+size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* workSpace);
+
+/*! FSE_count_simple
+ * Same as FSE_countFast(), but does not use any additional memory (not even on stack).
+ * This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr` (presuming it's also the size of `count`).
+*/
+size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
+
+
 
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
 /**< same as FSE_optimalTableLog(), which used `minus==2` */
 
+/* FSE_compress_wksp() :
+ * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
+ * FSE_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.
+ */
+#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue)   ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )
+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);
+
 size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);
-/**< build a fake FSE_CTable, designed to not compress an input, where each symbol uses nbBits */
+/**< build a fake FSE_CTable, designed for a flat distribution, where each symbol uses nbBits */
 
 size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);
 /**< build a fake FSE_CTable, designed to compress always the same symbolValue */
 
+/* FSE_buildCTable_wksp() :
+ * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
+ * `wkspSize` must be >= `(1<<tableLog)`.
+ */
+size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
+
 size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);
-/**< build a fake FSE_DTable, designed to read an uncompressed bitstream where each symbol uses nbBits */
+/**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */
 
 size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);
 /**< build a fake FSE_DTable, designed to always generate the same symbolValue */
 
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);
+/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */
+
 
 /* *****************************************
 *  FSE symbol compression API
 *******************************************/
 /*!
    This API consists of small unitary functions, which highly benefit from being inlined.
-   You will want to enable link-time-optimization to ensure these functions are properly inlined in your binary.
-   Visual seems to do it automatically.
-   For gcc or clang, you'll need to add -flto flag at compilation and linking stages.
-   If none of these solutions is applicable, include "fse.c" directly.
+   Hence their body are included in next section.
 */
-typedef struct
-{
+typedef struct {
     ptrdiff_t   value;
     const void* stateTable;
     const void* symbolTT;
@@ -384,8 +449,7 @@ If there is an error, it returns an errorCode (which can be tested using FSE_isE
 /* *****************************************
 *  FSE symbol decompression API
 *******************************************/
-typedef struct
-{
+typedef struct {
     size_t      state;
     const void* table;   /* precise table may vary, depending on U16 */
 } FSE_DState_t;
@@ -490,9 +554,9 @@ MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U3
 
 MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, U32 symbol)
 {
-    const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
+    FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
     const U16* const stateTable = (const U16*)(statePtr->stateTable);
-    U32 nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
+    U32 const nbBitsOut  = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);
     BIT_addBits(bitC, statePtr->value, nbBitsOut);
     statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];
 }
@@ -503,6 +567,7 @@ MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePt
     BIT_flushBits(bitC);
 }
 
+
 /* ======    Decompression    ====== */
 
 typedef struct {
@@ -581,14 +646,19 @@ MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)
 *  Increasing memory usage improves compression ratio
 *  Reduced memory usage can improve speed, due to cache effect
 *  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
-#define FSE_MAX_MEMORY_USAGE 14
-#define FSE_DEFAULT_MEMORY_USAGE 13
+#ifndef FSE_MAX_MEMORY_USAGE
+#  define FSE_MAX_MEMORY_USAGE 14
+#endif
+#ifndef FSE_DEFAULT_MEMORY_USAGE
+#  define FSE_DEFAULT_MEMORY_USAGE 13
+#endif
 
 /*!FSE_MAX_SYMBOL_VALUE :
 *  Maximum symbol value authorized.
 *  Required for proper stack allocation */
-#define FSE_MAX_SYMBOL_VALUE 255
-
+#ifndef FSE_MAX_SYMBOL_VALUE
+#  define FSE_MAX_SYMBOL_VALUE 255
+#endif
 
 /* **************************************************************
 *  template functions type & suffix
diff --git a/contrib/zstd/fse_compress.c b/contrib/zstd/fse_compress.c
index 386b2c010..26e8052dd 100644
--- a/contrib/zstd/fse_compress.c
+++ b/contrib/zstd/fse_compress.c
@@ -41,12 +41,15 @@
 #  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
 #  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
 #else
-#  ifdef __GNUC__
-#    define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-#    define FORCE_INLINE static inline __attribute__((always_inline))
+#  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 inline
-#  endif
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
 #endif
 
 
@@ -68,12 +71,6 @@
 
 
 /* **************************************************************
-*  Complex types
-****************************************************************/
-typedef U32 CTable_max_t[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
-
-
-/* **************************************************************
 *  Templates
 ****************************************************************/
 /*
@@ -97,7 +94,13 @@ typedef U32 CTable_max_t[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VA
 
 
 /* Function templates */
-size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+
+/* 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;
@@ -108,10 +111,11 @@ size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned
     U32 const step = FSE_TABLESTEP(tableSize);
     U32 cumul[FSE_MAX_SYMBOL_VALUE+2];
 
-    FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE]; /* memset() is not necessary, even if static analyzer complain about it */
+    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;
 
@@ -178,6 +182,13 @@ size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned
 }
 
 
+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
 
@@ -186,12 +197,10 @@ size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned
 ****************************************************************/
 size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
 {
-    size_t maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
+    size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
-static short FSE_abs(short a) { return (short)(a<0 ? -a : a); }
-
 static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
                                        const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
                                        unsigned writeIsSafe)
@@ -247,16 +256,16 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
                 bitStream >>= 16;
                 bitCount -= 16;
         }   }
-        {   short count = normalizedCounter[charnum++];
-            const short max = (short)((2*threshold-1)-remaining);
-            remaining -= FSE_abs(count);
-            if (remaining<1) return ERROR(GENERIC);
+        {   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) {
@@ -282,7 +291,7 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
 
 size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
 {
-    if (tableLog > FSE_MAX_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
+    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))
@@ -297,21 +306,20 @@ size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalized
 *  Counting histogram
 ****************************************************************/
 /*! FSE_count_simple
-    This function just counts byte values within `src`,
-    and store the histogram into table `count`.
-    This function is unsafe : it doesn't check that all values within `src` can fit into `count`.
+    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
 */
-static size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
-                               const void* src, size_t srcSize)
+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; }
 
@@ -326,20 +334,24 @@ static size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
 }
 
 
-static size_t FSE_count_parallel(unsigned* count, unsigned* maxSymbolValuePtr,
+/* 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 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;
 
-
-    U32 Counting1[256] = { 0 };
-    U32 Counting2[256] = { 0 };
-    U32 Counting3[256] = { 0 };
-    U32 Counting4[256] = { 0 };
+    memset(Counting1, 0, 4*256*sizeof(unsigned));
 
     /* safety checks */
     if (!sourceSize) {
@@ -385,31 +397,51 @@ static size_t FSE_count_parallel(unsigned* count, unsigned* maxSymbolValuePtr,
             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];
-    }}
+    {   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)
 {
-    if (sourceSize < 1500) return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
-    return FSE_count_parallel(count, maxSymbolValuePtr, source, sourceSize, 0);
+    unsigned tmpCounters[1024];
+    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters);
 }
 
-size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* source, size_t sourceSize)
+/* 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(count, maxSymbolValuePtr, source, sourceSize, 1);
+    if (*maxSymbolValuePtr < 255)
+        return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 1, workSpace);
     *maxSymbolValuePtr = 255;
-    return FSE_countFast(count, maxSymbolValuePtr, source, sourceSize);
+    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);
 }
 
 
@@ -425,14 +457,10 @@ size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr,
     `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)
 {
-    size_t size;
-    FSE_STATIC_ASSERT((size_t)FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)*4 >= sizeof(CTable_max_t));   /* A compilation error here means FSE_CTABLE_SIZE_U32 is not large enough */
-    if (tableLog > FSE_MAX_TABLELOG) return ERROR(GENERIC);
-    size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
-    return size;
+    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)
@@ -448,20 +476,20 @@ 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;
+    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 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
     U32 tableLog = maxTableLog;
-	U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
+    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 (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;
@@ -478,12 +506,13 @@ unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
 
 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 lowThreshold = (U32)(total >> tableLog);
+    U32 const lowThreshold = (U32)(total >> tableLog);
     U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
 
     for (s=0; s<=maxSymbolValue; s++) {
@@ -503,7 +532,8 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
             total -= count[s];
             continue;
         }
-        norm[s]=-2;
+
+        norm[s]=NOT_YET_ASSIGNED;
     }
     ToDistribute = (1 << tableLog) - distributed;
 
@@ -511,7 +541,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
         /* risk of rounding to zero */
         lowOne = (U32)((total * 3) / (ToDistribute * 2));
         for (s=0; s<=maxSymbolValue; s++) {
-            if ((norm[s] == -2) && (count[s] <= lowOne)) {
+            if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
                 norm[s] = 1;
                 distributed++;
                 total -= count[s];
@@ -531,17 +561,23 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
         return 0;
     }
 
-    {
-        U64 const vStepLog = 62 - tableLog;
+    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]==-2) {
-                U64 end = tmpTotal + (count[s] * rStep);
-                U32 sStart = (U32)(tmpTotal >> vStepLog);
-                U32 sEnd = (U32)(end >> vStepLog);
-                U32 weight = sEnd - sStart;
+            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;
@@ -563,7 +599,6 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
     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);
@@ -591,7 +626,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
         }   }
         if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
             /* corner case, need another normalization method */
-            size_t errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
+            size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
             if (FSE_isError(errorCode)) return errorCode;
         }
         else normalizedCounter[largest] += (short)stillToDistribute;
@@ -640,17 +675,15 @@ size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
 
     /* 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 (100% always same symbol) input */
+/* fake FSE_CTable, for rle input (always same symbol) */
 size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
 {
     void* ptr = ct;
@@ -682,14 +715,13 @@ static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
     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 errorCode = BIT_initCStream(&bitC, dst, dstSize);
-      if (FSE_isError(errorCode)) 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))
 
@@ -712,7 +744,7 @@ static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
     }
 
     /* 2 or 4 encoding per loop */
-    for ( ; ip>istart ; ) {
+    while ( ip>istart ) {
 
         FSE_encodeSymbol(&bitC, &CState2, *--ip);
 
@@ -738,7 +770,7 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
                            const void* src, size_t srcSize,
                            const FSE_CTable* ct)
 {
-    const unsigned fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
+    unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
 
     if (fast)
         return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
@@ -749,58 +781,76 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
 
 size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
 
-size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
-{
-    const BYTE* const istart = (const BYTE*) src;
-    const BYTE* ip = istart;
+#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];
-    CTable_max_t ct;
-    size_t errorCode;
+    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 (srcSize <= 1) return 0;  /* Uncompressible */
+    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 */
-    errorCode = FSE_count (count, &maxSymbolValue, ip, srcSize);
-    if (FSE_isError(errorCode)) return errorCode;
-    if (errorCode == srcSize) return 1;
-    if (errorCode == 1) return 0;   /* each symbol only present once */
-    if (errorCode < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
+    {   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);
-    errorCode = FSE_normalizeCount (norm, tableLog, count, srcSize, maxSymbolValue);
-    if (FSE_isError(errorCode)) return errorCode;
+    CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue) );
 
     /* Write table description header */
-    errorCode = FSE_writeNCount (op, oend-op, norm, maxSymbolValue, tableLog);
-    if (FSE_isError(errorCode)) return errorCode;
-    op += errorCode;
+    {   CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
+        op += nc_err;
+    }
 
     /* Compress */
-    errorCode = FSE_buildCTable (ct, norm, maxSymbolValue, tableLog);
-    if (FSE_isError(errorCode)) return errorCode;
-    errorCode = FSE_compress_usingCTable(op, oend - op, ip, srcSize, ct);
-    if (errorCode == 0) return 0;   /* not enough space for compressed data */
-    op += errorCode;
+    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;
+    if ( (size_t)(op-ostart) >= srcSize-1 ) return 0;
 
     return op-ostart;
 }
 
-size_t FSE_compress (void* dst, size_t dstSize, const void* src, size_t srcSize)
+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, dstSize, src, (U32)srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
+    return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
 }
 
 
diff --git a/contrib/zstd/fse_decompress.c b/contrib/zstd/fse_decompress.c
index 032e65771..8474a4c07 100644
--- a/contrib/zstd/fse_decompress.c
+++ b/contrib/zstd/fse_decompress.c
@@ -42,12 +42,15 @@
 #  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
 #  pragma warning(disable : 4214)        /* disable: C4214: non-int bitfields */
 #else
-#  ifdef __GNUC__
-#    define GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-#    define FORCE_INLINE static inline __attribute__((always_inline))
+#  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 inline
-#  endif
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
 #endif
 
 
@@ -56,7 +59,6 @@
 ****************************************************************/
 #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"
@@ -73,12 +75,6 @@
 
 
 /* **************************************************************
-*  Complex types
-****************************************************************/
-typedef U32 DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
-
-
-/* **************************************************************
 *  Templates
 ****************************************************************/
 /*
@@ -297,28 +293,34 @@ size_t FSE_decompress_usingDTable(void* dst, size_t originalSize,
 }
 
 
-size_t FSE_decompress(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize)
+size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog)
 {
     const BYTE* const istart = (const BYTE*)cSrc;
     const BYTE* ip = istart;
     short counting[FSE_MAX_SYMBOL_VALUE+1];
-    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
     unsigned tableLog;
     unsigned maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
 
-    if (cSrcSize<2) return ERROR(srcSize_wrong);   /* too small input size */
-
     /* normal FSE decoding mode */
-    {   size_t const NCountLength = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
-        if (FSE_isError(NCountLength)) return NCountLength;
-        if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size */
-        ip += NCountLength;
-        cSrcSize -= NCountLength;
-    }
+    size_t const NCountLength = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
+    if (FSE_isError(NCountLength)) return NCountLength;
+    //if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size; supposed to be already checked in NCountLength, only remaining case : NCountLength==cSrcSize */
+    if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
+    ip += NCountLength;
+    cSrcSize -= NCountLength;
 
-    CHECK_F( FSE_buildDTable (dt, counting, maxSymbolValue, tableLog) );
+    CHECK_F( FSE_buildDTable (workSpace, counting, maxSymbolValue, tableLog) );
 
-    return FSE_decompress_usingDTable (dst, maxDstSize, ip, cSrcSize, dt);   /* always return, even if it is an error code */
+    return FSE_decompress_usingDTable (dst, dstCapacity, ip, cSrcSize, workSpace);   /* always return, even if it is an error code */
+}
+
+
+typedef FSE_DTable DTable_max_t[FSE_DTABLE_SIZE_U32(FSE_MAX_TABLELOG)];
+
+size_t FSE_decompress(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize)
+{
+    DTable_max_t dt;   /* Static analyzer seems unable to understand this table will be properly initialized later */
+    return FSE_decompress_wksp(dst, dstCapacity, cSrc, cSrcSize, dt, FSE_MAX_TABLELOG);
 }
 
 
diff --git a/contrib/zstd/huf.h b/contrib/zstd/huf.h
index 29bab4b76..7873ca3d4 100644
--- a/contrib/zstd/huf.h
+++ b/contrib/zstd/huf.h
@@ -43,6 +43,21 @@ extern "C" {
 #include <stddef.h>    /* size_t */
 
 
+/* *** library symbols visibility *** */
+/* Note : when linking with -fvisibility=hidden on gcc, or by default on Visual,
+ *        HUF symbols remain "private" (internal symbols for library only).
+ *        Set macro FSE_DLL_EXPORT to 1 if you want HUF symbols visible on DLL interface */
+#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)
+#  define HUF_PUBLIC_API __attribute__ ((visibility ("default")))
+#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1)   /* Visual expected */
+#  define HUF_PUBLIC_API __declspec(dllexport)
+#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)
+#  define HUF_PUBLIC_API __declspec(dllimport)  /* not required, just to generate faster code (saves a function pointer load from IAT and an indirect jump) */
+#else
+#  define HUF_PUBLIC_API
+#endif
+
+
 /* *** simple functions *** */
 /**
 HUF_compress() :
@@ -55,42 +70,56 @@ HUF_compress() :
                      if return == 1, srcData is a single repeated byte symbol (RLE compression).
                      if HUF_isError(return), compression failed (more details using HUF_getErrorName())
 */
-size_t HUF_compress(void* dst, size_t dstCapacity,
-              const void* src, size_t srcSize);
+HUF_PUBLIC_API size_t HUF_compress(void* dst, size_t dstCapacity,
+                             const void* src, size_t srcSize);
 
 /**
 HUF_decompress() :
     Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
     into already allocated buffer 'dst', of minimum size 'dstSize'.
-    `dstSize` : **must** be the ***exact*** size of original (uncompressed) data.
+    `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
     Note : in contrast with FSE, HUF_decompress can regenerate
            RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
            because it knows size to regenerate.
-    @return : size of regenerated data (== dstSize),
+    @return : size of regenerated data (== originalSize),
               or an error code, which can be tested using HUF_isError()
 */
-size_t HUF_decompress(void* dst,  size_t dstSize,
-                const void* cSrc, size_t cSrcSize);
+HUF_PUBLIC_API size_t HUF_decompress(void* dst,  size_t originalSize,
+                               const void* cSrc, size_t cSrcSize);
 
 
-/* ****************************************
-*  Tool functions
-******************************************/
-#define HUF_BLOCKSIZE_MAX (128 * 1024)
-size_t HUF_compressBound(size_t size);       /**< maximum compressed size (worst case) */
+/* ***   Tool functions *** */
+#define HUF_BLOCKSIZE_MAX (128 * 1024)                  /**< maximum input size for a single block compressed with HUF_compress */
+HUF_PUBLIC_API size_t HUF_compressBound(size_t size);   /**< maximum compressed size (worst case) */
 
 /* Error Management */
-unsigned    HUF_isError(size_t code);        /**< tells if a return value is an error code */
-const char* HUF_getErrorName(size_t code);   /**< provides error code string (useful for debugging) */
+HUF_PUBLIC_API unsigned    HUF_isError(size_t code);       /**< tells if a return value is an error code */
+HUF_PUBLIC_API const char* HUF_getErrorName(size_t code);  /**< provides error code string (useful for debugging) */
 
 
-/* *** Advanced function *** */
+/* ***   Advanced function   *** */
 
 /** HUF_compress2() :
-*   Same as HUF_compress(), but offers direct control over `maxSymbolValue` and `tableLog` */
-size_t HUF_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+ *  Same as HUF_compress(), but offers direct control over `maxSymbolValue` and `tableLog`.
+ *  `tableLog` must be `<= HUF_TABLELOG_MAX` . */
+HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
 
+/** HUF_compress4X_wksp() :
+ *  Same as HUF_compress2(), but uses externally allocated `workSpace`.
+ *  `workspace` must have minimum alignment of 4, and be at least as large as following macro */
+#define HUF_WORKSPACE_SIZE (6 << 10)
+#define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
+HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
 
+
+
+/* ******************************************************************
+ *  WARNING !!
+ *  The following section contains advanced and experimental definitions
+ *  which shall never be used in the context of dll
+ *  because they are not guaranteed to remain stable in the future.
+ *  Only consider them in association with static linking.
+ *******************************************************************/
 #ifdef HUF_STATIC_LINKING_ONLY
 
 /* *** Dependencies *** */
@@ -98,10 +127,11 @@ size_t HUF_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize
 
 
 /* *** Constants *** */
-#define HUF_TABLELOG_ABSOLUTEMAX  16   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
-#define HUF_TABLELOG_MAX  12           /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_TABLELOG_MAX      12       /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
 #define HUF_TABLELOG_DEFAULT  11       /* tableLog by default, when not specified */
-#define HUF_SYMBOLVALUE_MAX 255
+#define HUF_SYMBOLVALUE_MAX  255
+
+#define HUF_TABLELOG_ABSOLUTEMAX  15   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
 #if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
 #  error "HUF_TABLELOG_MAX is too large !"
 #endif
@@ -112,12 +142,14 @@ size_t HUF_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize
 ******************************************/
 /* HUF buffer bounds */
 #define HUF_CTABLEBOUND 129
-#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true if incompressible pre-filtered with fast heuristic */
+#define HUF_BLOCKBOUND(size) (size + (size>>8) + 8)   /* only true when incompressible is pre-filtered with fast heuristic */
 #define HUF_COMPRESSBOUND(size) (HUF_CTABLEBOUND + HUF_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
 
 /* static allocation of HUF's Compression Table */
+#define HUF_CTABLE_SIZE_U32(maxSymbolValue)   ((maxSymbolValue)+1)   /* Use tables of U32, for proper alignment */
+#define HUF_CTABLE_SIZE(maxSymbolValue)       (HUF_CTABLE_SIZE_U32(maxSymbolValue) * sizeof(U32))
 #define HUF_CREATE_STATIC_CTABLE(name, maxSymbolValue) \
-    U32 name##hb[maxSymbolValue+1]; \
+    U32 name##hb[HUF_CTABLE_SIZE_U32(maxSymbolValue)]; \
     void* name##hv = &(name##hb); \
     HUF_CElt* name = (HUF_CElt*)(name##hv)   /* no final ; */
 
@@ -125,9 +157,9 @@ size_t HUF_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize
 typedef U32 HUF_DTable;
 #define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
 #define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
-        HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1)*0x1000001) }
+        HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
 #define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
-        HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog)*0x1000001) }
+        HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
 
 
 /* ****************************************
@@ -141,10 +173,6 @@ size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, con
 size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
 size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
 
-size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
-size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress1X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-
 
 /* ****************************************
 *  HUF detailed API
@@ -168,6 +196,23 @@ size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSym
 size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
 size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
 
+typedef enum {
+   HUF_repeat_none,  /**< Cannot use the previous table */
+   HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
+   HUF_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+ } HUF_repeat;
+/** HUF_compress4X_repeat() :
+*   Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+*   If it uses hufTable it does not modify hufTable or repeat.
+*   If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+*   If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress4X_repeat(void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize, HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
+
+/** 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.
+ */
+size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize);
 
 /*! HUF_readStats() :
     Read compact Huffman tree, saved by HUF_writeCTable().
@@ -208,16 +253,26 @@ size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* c
 /* single stream variants */
 
 size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
 size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+/** HUF_compress1X_repeat() :
+*   Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+*   If it uses hufTable it does not modify hufTable or repeat.
+*   If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+*   If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress1X_repeat(void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize, HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
 
 size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
 size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
 
-size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
+size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+
+size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /**< automatic selection of sing or double symbol decoder, based on DTable */
 size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 size_t HUF_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
 
-
 #endif /* HUF_STATIC_LINKING_ONLY */
 
 
diff --git a/contrib/zstd/huf_compress.c b/contrib/zstd/huf_compress.c
index c2dd13c87..7248c2513 100644
--- a/contrib/zstd/huf_compress.c
+++ b/contrib/zstd/huf_compress.c
@@ -35,24 +35,8 @@
 /* **************************************************************
 *  Compiler specifics
 ****************************************************************/
-#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-/* inline is defined */
-#elif defined(_MSC_VER)
-#  define inline __inline
-#else
-#  define inline /* disable inline */
-#endif
-
-
 #ifdef _MSC_VER    /* Visual Studio */
-#  define FORCE_INLINE static __forceinline
 #  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
-#else
-#  ifdef __GNUC__
-#    define FORCE_INLINE static inline __attribute__((always_inline))
-#  else
-#    define FORCE_INLINE static inline
-#  endif
 #endif
 
 
@@ -72,6 +56,8 @@
 *  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); }
 
 
 /* **************************************************************
@@ -86,31 +72,73 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
 /* *******************************************************
 *  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" */
 
-typedef struct nodeElt_s {
-    U32 count;
-    U16 parent;
-    BYTE byte;
-    BYTE nbBits;
-} nodeElt;
-
 /*! HUF_writeCTable() :
-    `CTable` : huffman tree to save, using huf representation.
+    `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];
+    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(GENERIC);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
 
     /* convert to weight */
     bitsToWeight[0] = 0;
@@ -119,38 +147,33 @@ size_t HUF_writeCTable (void* dst, size_t maxDstSize,
     for (n=0; n<maxSymbolValue; n++)
         huffWeight[n] = bitsToWeight[CTable[n].nbBits];
 
-    {   size_t const size = FSE_compress(op+1, maxDstSize-1, huffWeight, maxSymbolValue);
-        if (FSE_isError(size)) return size;
-        if ((size>1) & (size < maxSymbolValue/2)) {   /* FSE compressed */
-            op[0] = (BYTE)size;
-            return size+1;
-        }
-    }
+    /* 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;
+    }   }
 
-    /* raw values */
-    if (maxSymbolValue > (256-128)) return ERROR(GENERIC);   /* should not happen */
+    /* 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 issue in final combination */
+    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];
+    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;
-    size_t readSize;
     U32 nbSymbols = 0;
-    /*memset(huffWeight, 0, sizeof(huffWeight));*/   /* is not necessary, even though some analyzer complain ... */
 
     /* get symbol weights */
-    readSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX+1, rankVal, &nbSymbols, &tableLog, src, srcSize);
-    if (HUF_isError(readSize)) return readSize;
+    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);
@@ -171,13 +194,14 @@ size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, si
     }   }
 
     /* fill val */
-    {   U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
-        U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+    {   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=HUF_TABLELOG_MAX; n>0; n--) {
-                valPerRank[n] = min;      /* get starting value within each rank */
+            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;
         }   }
@@ -189,6 +213,13 @@ size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, si
 }
 
 
+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;
@@ -294,20 +325,26 @@ static void HUF_sort(nodeElt* huffNode, const U32* count, U32 maxSymbolValue)
 }
 
 
+/** 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)
-size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits)
+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 huffNode0[2*HUF_SYMBOLVALUE_MAX+1 +1];
-    nodeElt* huffNode = huffNode0 + 1;
+    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(huffNode0));
+    memset(huffNode0, 0, sizeof(huffNodeTable));
 
     /* sort, decreasing order */
     HUF_sort(huffNode, count, maxSymbolValue);
@@ -320,7 +357,7 @@ size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U3
     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);
+    huffNode0[0].count = (U32)(1U<<31);  /* fake entry, strong barrier */
 
     /* create parents */
     while (nodeNb <= nodeRoot) {
@@ -363,6 +400,34 @@ size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U3
     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);
@@ -390,20 +455,23 @@ size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, si
 
     /* init */
     if (dstSize < 8) return 0;   /* not enough space to compress */
-    { size_t const errorCode = BIT_initCStream(&bitC, op, oend-op);
-      if (HUF_isError(errorCode)) return 0; }
+    { 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);
+		 /* fall-through */
         case 2 : HUF_encodeSymbol(&bitC, ip[n+ 1], CTable);
                  HUF_FLUSHBITS_1(&bitC);
+		 /* fall-through */
         case 1 : HUF_encodeSymbol(&bitC, ip[n+ 0], CTable);
                  HUF_FLUSHBITS(&bitC);
-        case 0 :
-        default: ;
+		 /* fall-through */
+        case 0 : /* fall-through */
+        default: break;
     }
 
     for (; n>0; n-=4) {  /* note : n&3==0 at this stage */
@@ -434,32 +502,28 @@ size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, si
     if (srcSize < 12) return 0;   /* no saving possible : too small input */
     op += 6;   /* jumpTable */
 
-    {   size_t const cSize = HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable);
-        if (HUF_isError(cSize)) return cSize;
+    {   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;
-    {   size_t const cSize = HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable);
-        if (HUF_isError(cSize)) return cSize;
+    {   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;
-    {   size_t const cSize = HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable);
-        if (HUF_isError(cSize)) return cSize;
+    {   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;
-    {   size_t const cSize = HUF_compress1X_usingCTable(op, oend-op, ip, iend-ip, CTable);
-        if (HUF_isError(cSize)) return cSize;
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, iend-ip, CTable) );
         if (cSize==0) return 0;
         op += cSize;
     }
@@ -468,20 +532,43 @@ size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, si
 }
 
 
+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)
+                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[HUF_SYMBOLVALUE_MAX+1];
-    HUF_CElt CTable[HUF_SYMBOLVALUE_MAX+1];
+    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 */
@@ -489,59 +576,111 @@ static size_t HUF_compress_internal (
     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 */
-    {   size_t const largest = FSE_count (count, &maxSymbolValue, (const BYTE*)src, srcSize);
-        if (HUF_isError(largest)) return largest;
+    {   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);
-    {   size_t const maxBits = HUF_buildCTable (CTable, count, maxSymbolValue, huffLog);
-        if (HUF_isError(maxBits)) return maxBits;
+    {   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 */
-    {   size_t const hSize = HUF_writeCTable (op, dstSize, CTable, maxSymbolValue, huffLog);
-        if (HUF_isError(hSize)) return hSize;
-        if (hSize + 12 >= srcSize) return 0;   /* not useful to try compression */
+    {   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);
+}
 
-    /* Compress */
-    {   size_t const cSize = (singleStream) ?
-                            HUF_compress1X_usingCTable(op, oend - op, src, srcSize, CTable) :   /* single segment */
-                            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;
+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)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 1);
+    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)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 0);
+    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/contrib/zstd/huf_decompress.c b/contrib/zstd/huf_decompress.c
index a5521bd36..ea35c3620 100644
--- a/contrib/zstd/huf_decompress.c
+++ b/contrib/zstd/huf_decompress.c
@@ -35,24 +35,19 @@
 /* **************************************************************
 *  Compiler specifics
 ****************************************************************/
-#if defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
-/* inline is defined */
-#elif defined(_MSC_VER) || defined(__GNUC__)
-#  define inline __inline
-#else
-#  define inline /* disable inline */
-#endif
-
-
 #ifdef _MSC_VER    /* Visual Studio */
 #  define FORCE_INLINE static __forceinline
 #  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
 #else
-#  ifdef __GNUC__
-#    define FORCE_INLINE static inline __attribute__((always_inline))
+#  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 inline
-#  endif
+#    define FORCE_INLINE static
+#  endif /* __STDC_VERSION__ */
 #endif
 
 
@@ -110,16 +105,16 @@ size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize)
 
     /* Table header */
     {   DTableDesc dtd = HUF_getDTableDesc(DTable);
-        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, huffman tree cannot fit in */
+        if (tableLog > (U32)(dtd.maxTableLog+1)) return ERROR(tableLog_tooLarge);   /* DTable too small, Huffman tree cannot fit in */
         dtd.tableType = 0;
         dtd.tableLog = (BYTE)tableLog;
         memcpy(DTable, &dtd, sizeof(dtd));
     }
 
-    /* Prepare ranks */
+    /* Calculate starting value for each rank */
     {   U32 n, nextRankStart = 0;
         for (n=1; n<tableLog+1; n++) {
-            U32 current = nextRankStart;
+            U32 const current = nextRankStart;
             nextRankStart += (rankVal[n] << (n-1));
             rankVal[n] = current;
     }   }
@@ -129,11 +124,11 @@ size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize)
         for (n=0; n<nbSymbols; n++) {
             U32 const w = huffWeight[n];
             U32 const length = (1 << w) >> 1;
-            U32 i;
+            U32 u;
             HUF_DEltX2 D;
             D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
-            for (i = rankVal[w]; i < rankVal[w] + length; i++)
-                dt[i] = D;
+            for (u = rankVal[w]; u < rankVal[w] + length; u++)
+                dt[u] = D;
             rankVal[w] += length;
     }   }
 
@@ -160,7 +155,7 @@ static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, con
     if (MEM_64bits()) \
         HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
 
-static inline size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+FORCE_INLINE size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
 {
     BYTE* const pStart = p;
 
@@ -366,13 +361,15 @@ typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* doubl
 
 typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
 
+/* HUF_fillDTableX4Level2() :
+ * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
 static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
                            const U32* rankValOrigin, const int minWeight,
                            const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
                            U32 nbBitsBaseline, U16 baseSeq)
 {
     HUF_DEltX4 DElt;
-    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
+    U32 rankVal[HUF_TABLELOG_MAX + 1];
 
     /* get pre-calculated rankVal */
     memcpy(rankVal, rankValOrigin, sizeof(rankVal));
@@ -406,14 +403,14 @@ static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 co
     }   }
 }
 
-typedef U32 rankVal_t[HUF_TABLELOG_ABSOLUTEMAX][HUF_TABLELOG_ABSOLUTEMAX + 1];
+typedef U32 rankVal_t[HUF_TABLELOG_MAX][HUF_TABLELOG_MAX + 1];
 
 static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
                            const sortedSymbol_t* sortedList, const U32 sortedListSize,
                            const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
                            const U32 nbBitsBaseline)
 {
-    U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];
+    U32 rankVal[HUF_TABLELOG_MAX + 1];
     const int scaleLog = nbBitsBaseline - targetLog;   /* note : targetLog >= srcLog, hence scaleLog <= 1 */
     const U32 minBits  = nbBitsBaseline - maxWeight;
     U32 s;
@@ -454,8 +451,8 @@ size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize)
 {
     BYTE weightList[HUF_SYMBOLVALUE_MAX + 1];
     sortedSymbol_t sortedSymbol[HUF_SYMBOLVALUE_MAX + 1];
-    U32 rankStats[HUF_TABLELOG_ABSOLUTEMAX + 1] = { 0 };
-    U32 rankStart0[HUF_TABLELOG_ABSOLUTEMAX + 2] = { 0 };
+    U32 rankStats[HUF_TABLELOG_MAX + 1] = { 0 };
+    U32 rankStart0[HUF_TABLELOG_MAX + 2] = { 0 };
     U32* const rankStart = rankStart0+1;
     rankVal_t rankVal;
     U32 tableLog, maxW, sizeOfSort, nbSymbols;
@@ -465,8 +462,8 @@ size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize)
     void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
     HUF_DEltX4* const dt = (HUF_DEltX4*)dtPtr;
 
-    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compilation fails here, assertion is false */
-    if (maxTableLog > HUF_TABLELOG_ABSOLUTEMAX) return ERROR(tableLog_tooLarge);
+    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
+    if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     /* memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
     iSize = HUF_readStats(weightList, HUF_SYMBOLVALUE_MAX + 1, rankStats, &nbSymbols, &tableLog, src, srcSize);
@@ -565,7 +562,7 @@ static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DE
     if (MEM_64bits()) \
         ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
 
-static inline size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+FORCE_INLINE size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
 {
     BYTE* const pStart = p;
 
diff --git a/contrib/zstd/mem.h b/contrib/zstd/mem.h
index 681dd35d2..4773a8b93 100644
--- a/contrib/zstd/mem.h
+++ b/contrib/zstd/mem.h
@@ -39,7 +39,7 @@ extern "C" {
 #endif
 
 /* code only tested on 32 and 64 bits systems */
-#define MEM_STATIC_ASSERT(c)   { enum { XXH_static_assert = 1/(int)(!!(c)) }; }
+#define MEM_STATIC_ASSERT(c)   { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
 MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
 
 
@@ -48,21 +48,25 @@ MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (size
 *****************************************************************/
 #if  !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
 # include <stdint.h>
-  typedef  uint8_t BYTE;
-  typedef uint16_t U16;
-  typedef  int16_t S16;
-  typedef uint32_t U32;
-  typedef  int32_t S32;
-  typedef uint64_t U64;
-  typedef  int64_t S64;
+  typedef   uint8_t BYTE;
+  typedef  uint16_t U16;
+  typedef   int16_t S16;
+  typedef  uint32_t U32;
+  typedef   int32_t S32;
+  typedef  uint64_t U64;
+  typedef   int64_t S64;
+  typedef  intptr_t iPtrDiff;
+  typedef uintptr_t uPtrDiff;
 #else
-  typedef unsigned char       BYTE;
+  typedef unsigned char      BYTE;
   typedef unsigned short      U16;
   typedef   signed short      S16;
   typedef unsigned int        U32;
   typedef   signed int        S32;
   typedef unsigned long long  U64;
   typedef   signed long long  S64;
+  typedef ptrdiff_t      iPtrDiff;
+  typedef size_t         uPtrDiff;
 #endif
 
 
@@ -74,19 +78,18 @@ MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (size
  * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
  * The below switch allow to select different access method for improved performance.
  * Method 0 (default) : use `memcpy()`. Safe and portable.
- * Method 1 : `__packed` statement. It depends on compiler extension (ie, not portable).
+ * Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable).
  *            This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
  * Method 2 : direct access. This method is portable but violate C standard.
  *            It can generate buggy code on targets depending on alignment.
- *            In some circumstances, it's the only known way to get the most performance (ie GCC + ARMv6)
+ *            In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6)
  * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
  * Prefer these methods in priority order (0 > 1 > 2)
  */
 #ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
 #  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
 #    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif defined(__INTEL_COMPILER) /*|| defined(_MSC_VER)*/ || \
-  (defined(__GNUC__) && ( defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7S__) ))
+#  elif defined(__INTEL_COMPILER) || defined(__GNUC__)
 #    define MEM_FORCE_MEMORY_ACCESS 1
 #  endif
 #endif
@@ -118,7 +121,7 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
 /* currently only defined for gcc and icc */
 #if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
-	__pragma( pack(push, 1) )
+    __pragma( pack(push, 1) )
     typedef union { U16 u16; U32 u32; U64 u64; size_t st; } unalign;
     __pragma( pack(pop) )
 #else
@@ -180,7 +183,7 @@ MEM_STATIC U32 MEM_swap32(U32 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_ulong(in);
-#elif defined (__GNUC__)
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
     return __builtin_bswap32(in);
 #else
     return  ((in << 24) & 0xff000000 ) |
@@ -194,7 +197,7 @@ MEM_STATIC U64 MEM_swap64(U64 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_uint64(in);
-#elif defined (__GNUC__)
+#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
     return __builtin_bswap64(in);
 #else
     return  ((in << 56) & 0xff00000000000000ULL) |
diff --git a/contrib/zstd/pool.c b/contrib/zstd/pool.c
new file mode 100644
index 000000000..e439fe1b0
--- /dev/null
+++ b/contrib/zstd/pool.c
@@ -0,0 +1,194 @@
+/**
+ * Copyright (c) 2016-present, 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 <stddef.h>  /* size_t */
+#include <stdlib.h>  /* malloc, calloc, free */
+#include "pool.h"
+
+/* ======   Compiler specifics   ====== */
+#if defined(_MSC_VER)
+#  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */
+#endif
+
+
+#ifdef ZSTD_MULTITHREAD
+
+#include "threading.h"   /* pthread adaptation */
+
+/* A job is a function and an opaque argument */
+typedef struct POOL_job_s {
+  POOL_function function;
+  void *opaque;
+} POOL_job;
+
+struct POOL_ctx_s {
+    /* Keep track of the threads */
+    pthread_t *threads;
+    size_t numThreads;
+
+    /* The queue is a circular buffer */
+    POOL_job *queue;
+    size_t queueHead;
+    size_t queueTail;
+    size_t queueSize;
+    /* The mutex protects the queue */
+    pthread_mutex_t queueMutex;
+    /* Condition variable for pushers to wait on when the queue is full */
+    pthread_cond_t queuePushCond;
+    /* Condition variables for poppers to wait on when the queue is empty */
+    pthread_cond_t queuePopCond;
+    /* Indicates if the queue is shutting down */
+    int shutdown;
+};
+
+/* POOL_thread() :
+   Work thread for the thread pool.
+   Waits for jobs and executes them.
+   @returns : NULL on failure else non-null.
+*/
+static void* POOL_thread(void* opaque) {
+    POOL_ctx* const ctx = (POOL_ctx*)opaque;
+    if (!ctx) { return NULL; }
+    for (;;) {
+        /* Lock the mutex and wait for a non-empty queue or until shutdown */
+        pthread_mutex_lock(&ctx->queueMutex);
+        while (ctx->queueHead == ctx->queueTail && !ctx->shutdown) {
+            pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
+        }
+        /* empty => shutting down: so stop */
+        if (ctx->queueHead == ctx->queueTail) {
+            pthread_mutex_unlock(&ctx->queueMutex);
+            return opaque;
+        }
+        /* Pop a job off the queue */
+        {   POOL_job const job = ctx->queue[ctx->queueHead];
+            ctx->queueHead = (ctx->queueHead + 1) % ctx->queueSize;
+            /* Unlock the mutex, signal a pusher, and run the job */
+            pthread_mutex_unlock(&ctx->queueMutex);
+            pthread_cond_signal(&ctx->queuePushCond);
+            job.function(job.opaque);
+        }
+    }
+    /* Unreachable */
+}
+
+POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
+    POOL_ctx *ctx;
+    /* Check the parameters */
+    if (!numThreads || !queueSize) { return NULL; }
+    /* Allocate the context and zero initialize */
+    ctx = (POOL_ctx *)calloc(1, sizeof(POOL_ctx));
+    if (!ctx) { return NULL; }
+    /* Initialize the job queue.
+     * It needs one extra space since one space is wasted to differentiate empty
+     * and full queues.
+     */
+    ctx->queueSize = queueSize + 1;
+    ctx->queue = (POOL_job *)malloc(ctx->queueSize * sizeof(POOL_job));
+    ctx->queueHead = 0;
+    ctx->queueTail = 0;
+    pthread_mutex_init(&ctx->queueMutex, NULL);
+    pthread_cond_init(&ctx->queuePushCond, NULL);
+    pthread_cond_init(&ctx->queuePopCond, NULL);
+    ctx->shutdown = 0;
+    /* Allocate space for the thread handles */
+    ctx->threads = (pthread_t *)malloc(numThreads * sizeof(pthread_t));
+    ctx->numThreads = 0;
+    /* Check for errors */
+    if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
+    /* Initialize the threads */
+    {   size_t i;
+        for (i = 0; i < numThreads; ++i) {
+            if (pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
+                ctx->numThreads = i;
+                POOL_free(ctx);
+                return NULL;
+        }   }
+        ctx->numThreads = numThreads;
+    }
+    return ctx;
+}
+
+/*! POOL_join() :
+    Shutdown the queue, wake any sleeping threads, and join all of the threads.
+*/
+static void POOL_join(POOL_ctx *ctx) {
+    /* Shut down the queue */
+    pthread_mutex_lock(&ctx->queueMutex);
+    ctx->shutdown = 1;
+    pthread_mutex_unlock(&ctx->queueMutex);
+    /* Wake up sleeping threads */
+    pthread_cond_broadcast(&ctx->queuePushCond);
+    pthread_cond_broadcast(&ctx->queuePopCond);
+    /* Join all of the threads */
+    {   size_t i;
+        for (i = 0; i < ctx->numThreads; ++i) {
+            pthread_join(ctx->threads[i], NULL);
+    }   }
+}
+
+void POOL_free(POOL_ctx *ctx) {
+    if (!ctx) { return; }
+    POOL_join(ctx);
+    pthread_mutex_destroy(&ctx->queueMutex);
+    pthread_cond_destroy(&ctx->queuePushCond);
+    pthread_cond_destroy(&ctx->queuePopCond);
+    if (ctx->queue) free(ctx->queue);
+    if (ctx->threads) free(ctx->threads);
+    free(ctx);
+}
+
+void POOL_add(void *ctxVoid, POOL_function function, void *opaque) {
+    POOL_ctx *ctx = (POOL_ctx *)ctxVoid;
+    if (!ctx) { return; }
+
+    pthread_mutex_lock(&ctx->queueMutex);
+    {   POOL_job const job = {function, opaque};
+        /* Wait until there is space in the queue for the new job */
+        size_t newTail = (ctx->queueTail + 1) % ctx->queueSize;
+        while (ctx->queueHead == newTail && !ctx->shutdown) {
+          pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
+          newTail = (ctx->queueTail + 1) % ctx->queueSize;
+        }
+        /* The queue is still going => there is space */
+        if (!ctx->shutdown) {
+            ctx->queue[ctx->queueTail] = job;
+            ctx->queueTail = newTail;
+        }
+    }
+    pthread_mutex_unlock(&ctx->queueMutex);
+    pthread_cond_signal(&ctx->queuePopCond);
+}
+
+#else  /* ZSTD_MULTITHREAD  not defined */
+/* No multi-threading support */
+
+/* We don't need any data, but if it is empty malloc() might return NULL. */
+struct POOL_ctx_s {
+  int data;
+};
+
+POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
+  (void)numThreads;
+  (void)queueSize;
+  return (POOL_ctx *)malloc(sizeof(POOL_ctx));
+}
+
+void POOL_free(POOL_ctx *ctx) {
+  if (ctx) free(ctx);
+}
+
+void POOL_add(void *ctx, POOL_function function, void *opaque) {
+  (void)ctx;
+  function(opaque);
+}
+
+#endif  /* ZSTD_MULTITHREAD */
diff --git a/contrib/zstd/pool.h b/contrib/zstd/pool.h
new file mode 100644
index 000000000..50cb25b12
--- /dev/null
+++ b/contrib/zstd/pool.h
@@ -0,0 +1,56 @@
+/**
+ * Copyright (c) 2016-present, 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 POOL_H
+#define POOL_H
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+#include <stddef.h>   /* size_t */
+
+typedef struct POOL_ctx_s POOL_ctx;
+
+/*! POOL_create() :
+    Create a thread pool with at most `numThreads` threads.
+    `numThreads` must be at least 1.
+    The maximum number of queued jobs before blocking is `queueSize`.
+    `queueSize` must be at least 1.
+    @return : The POOL_ctx pointer on success else NULL.
+*/
+POOL_ctx *POOL_create(size_t numThreads, size_t queueSize);
+
+/*! POOL_free() :
+    Free a thread pool returned by POOL_create().
+*/
+void POOL_free(POOL_ctx *ctx);
+
+/*! POOL_function :
+    The function type that can be added to a thread pool.
+*/
+typedef void (*POOL_function)(void *);
+/*! POOL_add_function :
+    The function type for a generic thread pool add function.
+*/
+typedef void (*POOL_add_function)(void *, POOL_function, void *);
+
+/*! POOL_add() :
+    Add the job `function(opaque)` to the thread pool.
+    Possibly blocks until there is room in the queue.
+    Note : The function may be executed asynchronously, so `opaque` must live until the function has been completed.
+*/
+void POOL_add(void *ctx, POOL_function function, void *opaque);
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif
diff --git a/contrib/zstd/threading.c b/contrib/zstd/threading.c
new file mode 100644
index 000000000..141376c56
--- /dev/null
+++ b/contrib/zstd/threading.c
@@ -0,0 +1,79 @@
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * 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.
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ */
+
+/**
+ * This file will hold wrapper for systems, which do not support pthreads
+ */
+
+/* When ZSTD_MULTITHREAD is not defined, this file would become an empty translation unit.
+* Include some ISO C header code to prevent this and portably avoid related warnings.
+* (Visual C++: C4206 / GCC: -Wpedantic / Clang: -Wempty-translation-unit)
+*/
+#include <stddef.h>
+
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper, based on :
+ * http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ */
+
+
+/* ===  Dependencies  === */
+#include <process.h>
+#include <errno.h>
+#include "threading.h"
+
+
+/* ===  Implementation  === */
+
+static unsigned __stdcall worker(void *arg)
+{
+    pthread_t* const thread = (pthread_t*) arg;
+    thread->arg = thread->start_routine(thread->arg);
+    return 0;
+}
+
+int pthread_create(pthread_t* thread, const void* unused,
+            void* (*start_routine) (void*), void* arg)
+{
+    (void)unused;
+    thread->arg = arg;
+    thread->start_routine = start_routine;
+    thread->handle = (HANDLE) _beginthreadex(NULL, 0, worker, thread, 0, NULL);
+
+    if (!thread->handle)
+        return errno;
+    else
+        return 0;
+}
+
+int _pthread_join(pthread_t * thread, void **value_ptr)
+{
+    DWORD result;
+
+    if (!thread->handle) return 0;
+
+    result = WaitForSingleObject(thread->handle, INFINITE);
+    switch (result) {
+    case WAIT_OBJECT_0:
+        if (value_ptr) *value_ptr = thread->arg;
+        return 0;
+    case WAIT_ABANDONED:
+        return EINVAL;
+    default:
+        return GetLastError();
+    }
+}
+
+#endif   /* ZSTD_MULTITHREAD */
diff --git a/contrib/zstd/threading.h b/contrib/zstd/threading.h
new file mode 100644
index 000000000..c0086139e
--- /dev/null
+++ b/contrib/zstd/threading.h
@@ -0,0 +1,104 @@
+
+/**
+ * Copyright (c) 2016 Tino Reichardt
+ * 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.
+ *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ */
+
+#ifndef THREADING_H_938743
+#define THREADING_H_938743
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
+
+/**
+ * Windows minimalist Pthread Wrapper, based on :
+ * http://www.cse.wustl.edu/~schmidt/win32-cv-1.html
+ */
+#ifdef WINVER
+#  undef WINVER
+#endif
+#define WINVER       0x0600
+
+#ifdef _WIN32_WINNT
+#  undef _WIN32_WINNT
+#endif
+#define _WIN32_WINNT 0x0600
+
+#ifndef WIN32_LEAN_AND_MEAN
+#  define WIN32_LEAN_AND_MEAN
+#endif
+
+#include <windows.h>
+
+/* mutex */
+#define pthread_mutex_t           CRITICAL_SECTION
+#define pthread_mutex_init(a,b)   InitializeCriticalSection((a))
+#define pthread_mutex_destroy(a)  DeleteCriticalSection((a))
+#define pthread_mutex_lock(a)     EnterCriticalSection((a))
+#define pthread_mutex_unlock(a)   LeaveCriticalSection((a))
+
+/* condition variable */
+#define pthread_cond_t             CONDITION_VARIABLE
+#define pthread_cond_init(a, b)    InitializeConditionVariable((a))
+#define pthread_cond_destroy(a)    /* No delete */
+#define pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
+#define pthread_cond_signal(a)     WakeConditionVariable((a))
+#define pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
+
+/* pthread_create() and pthread_join() */
+typedef struct {
+    HANDLE handle;
+    void* (*start_routine)(void*);
+    void* arg;
+} pthread_t;
+
+int pthread_create(pthread_t* thread, const void* unused,
+                   void* (*start_routine) (void*), void* arg);
+
+#define pthread_join(a, b) _pthread_join(&(a), (b))
+int _pthread_join(pthread_t* thread, void** value_ptr);
+
+/**
+ * add here more wrappers as required
+ */
+
+
+#elif defined(ZSTD_MULTITHREAD)   /* posix assumed ; need a better detection method */
+/* ===   POSIX Systems   === */
+#  include <pthread.h>
+
+#else  /* ZSTD_MULTITHREAD not defined */
+/* No multithreading support */
+
+#define pthread_mutex_t int   /* #define rather than typedef, as sometimes pthread support is implicit, resulting in duplicated symbols */
+#define pthread_mutex_init(a,b)
+#define pthread_mutex_destroy(a)
+#define pthread_mutex_lock(a)
+#define pthread_mutex_unlock(a)
+
+#define pthread_cond_t int
+#define pthread_cond_init(a,b)
+#define pthread_cond_destroy(a)
+#define pthread_cond_wait(a,b)
+#define pthread_cond_signal(a)
+#define pthread_cond_broadcast(a)
+
+/* do not use pthread_t */
+
+#endif /* ZSTD_MULTITHREAD */
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* THREADING_H_938743 */
diff --git a/contrib/zstd/zbuff.h b/contrib/zstd/zbuff.h
deleted file mode 100644
index f99e06197..000000000
--- a/contrib/zstd/zbuff.h
+++ /dev/null
@@ -1,191 +0,0 @@
-/**
- * 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.
- */
-
-/* ***************************************************************
-*  NOTES/WARNINGS
-*****************************************************************/
-/* The streaming API defined here will soon be deprecated by the
-* new one in 'zstd.h'; consider migrating towards newer streaming
-* API. See 'lib/README.md'.
-*****************************************************************/
-
-#ifndef ZSTD_BUFFERED_H_23987
-#define ZSTD_BUFFERED_H_23987
-
-#if defined (__cplusplus)
-extern "C" {
-#endif
-
-/* *************************************
-*  Dependencies
-***************************************/
-#include <stddef.h>      /* size_t */
-
-
-/* ***************************************************************
-*  Compiler specifics
-*****************************************************************/
-/* ZSTD_DLL_EXPORT :
-*  Enable exporting of functions when building a Windows DLL */
-#if defined(_WIN32) && defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZSTDLIB_API __declspec(dllexport)
-#else
-#  define ZSTDLIB_API
-#endif
-
-
-/* *************************************
-*  Streaming functions
-***************************************/
-/* This is the easier "buffered" streaming API,
-*  using an internal buffer to lift all restrictions on user-provided buffers
-*  which can be any size, any place, for both input and output.
-*  ZBUFF and ZSTD are 100% interoperable,
-*  frames created by one can be decoded by the other one */
-
-typedef struct ZBUFF_CCtx_s ZBUFF_CCtx;
-ZSTDLIB_API ZBUFF_CCtx* ZBUFF_createCCtx(void);
-ZSTDLIB_API size_t      ZBUFF_freeCCtx(ZBUFF_CCtx* cctx);
-
-ZSTDLIB_API size_t ZBUFF_compressInit(ZBUFF_CCtx* cctx, int compressionLevel);
-ZSTDLIB_API size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
-
-ZSTDLIB_API size_t ZBUFF_compressContinue(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr, const void* src, size_t* srcSizePtr);
-ZSTDLIB_API size_t ZBUFF_compressFlush(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr);
-ZSTDLIB_API size_t ZBUFF_compressEnd(ZBUFF_CCtx* cctx, void* dst, size_t* dstCapacityPtr);
-
-/*-*************************************************
-*  Streaming compression - howto
-*
-*  A ZBUFF_CCtx object is required to track streaming operation.
-*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
-*  ZBUFF_CCtx objects can be reused multiple times.
-*
-*  Start by initializing ZBUF_CCtx.
-*  Use ZBUFF_compressInit() to start a new compression operation.
-*  Use ZBUFF_compressInitDictionary() for a compression which requires a dictionary.
-*
-*  Use ZBUFF_compressContinue() repetitively to consume input stream.
-*  *srcSizePtr and *dstCapacityPtr can be any size.
-*  The function will report how many bytes were read or written within *srcSizePtr and *dstCapacityPtr.
-*  Note that it may not consume the entire input, in which case it's up to the caller to present again remaining data.
-*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each call, so save its content if it matters or change @dst .
-*  @return : a hint to preferred nb of bytes to use as input for next function call (it's just a hint, to improve latency)
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  At any moment, it's possible to flush whatever data remains within buffer, using ZBUFF_compressFlush().
-*  The nb of bytes written into `dst` will be reported into *dstCapacityPtr.
-*  Note that the function cannot output more than *dstCapacityPtr,
-*  therefore, some content might still be left into internal buffer if *dstCapacityPtr is too small.
-*  @return : nb of bytes still present into internal buffer (0 if it's empty)
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  ZBUFF_compressEnd() instructs to finish a frame.
-*  It will perform a flush and write frame epilogue.
-*  The epilogue is required for decoders to consider a frame completed.
-*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
-*  In which case, call again ZBUFF_compressFlush() to complete the flush.
-*  @return : nb of bytes still present into internal buffer (0 if it's empty)
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  Hint : _recommended buffer_ sizes (not compulsory) : ZBUFF_recommendedCInSize() / ZBUFF_recommendedCOutSize()
-*  input : ZBUFF_recommendedCInSize==128 KB block size is the internal unit, use this value to reduce intermediate stages (better latency)
-*  output : ZBUFF_recommendedCOutSize==ZSTD_compressBound(128 KB) + 3 + 3 : ensures it's always possible to write/flush/end a full block. Skip some buffering.
-*  By using both, it ensures that input will be entirely consumed, and output will always contain the result, reducing intermediate buffering.
-* **************************************************/
-
-
-typedef struct ZBUFF_DCtx_s ZBUFF_DCtx;
-ZSTDLIB_API ZBUFF_DCtx* ZBUFF_createDCtx(void);
-ZSTDLIB_API size_t      ZBUFF_freeDCtx(ZBUFF_DCtx* dctx);
-
-ZSTDLIB_API size_t ZBUFF_decompressInit(ZBUFF_DCtx* dctx);
-ZSTDLIB_API size_t ZBUFF_decompressInitDictionary(ZBUFF_DCtx* dctx, const void* dict, size_t dictSize);
-
-ZSTDLIB_API size_t ZBUFF_decompressContinue(ZBUFF_DCtx* dctx,
-                                            void* dst, size_t* dstCapacityPtr,
-                                      const void* src, size_t* srcSizePtr);
-
-/*-***************************************************************************
-*  Streaming decompression howto
-*
-*  A ZBUFF_DCtx object is required to track streaming operations.
-*  Use ZBUFF_createDCtx() and ZBUFF_freeDCtx() to create/release resources.
-*  Use ZBUFF_decompressInit() to start a new decompression operation,
-*   or ZBUFF_decompressInitDictionary() if decompression requires a dictionary.
-*  Note that ZBUFF_DCtx objects can be re-init multiple times.
-*
-*  Use ZBUFF_decompressContinue() repetitively to consume your input.
-*  *srcSizePtr and *dstCapacityPtr can be any size.
-*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
-*  Note that it may not consume the entire input, in which case it's up to the caller to present remaining input again.
-*  The content of `dst` will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters, or change `dst`.
-*  @return : 0 when a frame is completely decoded and fully flushed,
-*            1 when there is still some data left within internal buffer to flush,
-*            >1 when more data is expected, with value being a suggested next input size (it's just a hint, which helps latency),
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  Hint : recommended buffer sizes (not compulsory) : ZBUFF_recommendedDInSize() and ZBUFF_recommendedDOutSize()
-*  output : ZBUFF_recommendedDOutSize== 128 KB block size is the internal unit, it ensures it's always possible to write a full block when decoded.
-*  input  : ZBUFF_recommendedDInSize == 128KB + 3;
-*           just follow indications from ZBUFF_decompressContinue() to minimize latency. It should always be <= 128 KB + 3 .
-* *******************************************************************************/
-
-
-/* *************************************
-*  Tool functions
-***************************************/
-ZSTDLIB_API unsigned ZBUFF_isError(size_t errorCode);
-ZSTDLIB_API const char* ZBUFF_getErrorName(size_t errorCode);
-
-/** Functions below provide recommended buffer sizes for Compression or Decompression operations.
-*   These sizes are just hints, they tend to offer better latency */
-ZSTDLIB_API size_t ZBUFF_recommendedCInSize(void);
-ZSTDLIB_API size_t ZBUFF_recommendedCOutSize(void);
-ZSTDLIB_API size_t ZBUFF_recommendedDInSize(void);
-ZSTDLIB_API size_t ZBUFF_recommendedDOutSize(void);
-
-
-#ifdef ZBUFF_STATIC_LINKING_ONLY
-
-/* ====================================================================================
- * The definitions in this section are considered experimental.
- * They should never be used in association with a dynamic library, as they may change in the future.
- * They are provided for advanced usages.
- * Use them only in association with static linking.
- * ==================================================================================== */
-
-/*--- Dependency ---*/
-#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters, ZSTD_customMem */
-#include "zstd.h"
-
-
-/*--- Custom memory allocator ---*/
-/*! ZBUFF_createCCtx_advanced() :
- *  Create a ZBUFF compression context using external alloc and free functions */
-ZSTDLIB_API ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem);
-
-/*! ZBUFF_createDCtx_advanced() :
- *  Create a ZBUFF decompression context using external alloc and free functions */
-ZSTDLIB_API ZBUFF_DCtx* ZBUFF_createDCtx_advanced(ZSTD_customMem customMem);
-
-
-/*--- Advanced Streaming Initialization ---*/
-ZSTDLIB_API size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
-                                               const void* dict, size_t dictSize,
-                                               ZSTD_parameters params, unsigned long long pledgedSrcSize);
-
-#endif /* ZBUFF_STATIC_LINKING_ONLY */
-
-
-#if defined (__cplusplus)
-}
-#endif
-
-#endif  /* ZSTD_BUFFERED_H_23987 */
diff --git a/contrib/zstd/zbuff_compress.c b/contrib/zstd/zbuff_compress.c
deleted file mode 100644
index 5095b43e6..000000000
--- a/contrib/zstd/zbuff_compress.c
+++ /dev/null
@@ -1,319 +0,0 @@
-/**
- * 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 <stdlib.h>
-#include "error_private.h"
-#include "zstd_internal.h"  /* MIN, ZSTD_BLOCKHEADERSIZE, defaultCustomMem */
-#define ZBUFF_STATIC_LINKING_ONLY
-#include "zbuff.h"
-
-
-/* *************************************
-*  Constants
-***************************************/
-static size_t const ZBUFF_endFrameSize = ZSTD_BLOCKHEADERSIZE;
-
-
-/*-***********************************************************
-*  Streaming compression
-*
-*  A ZBUFF_CCtx object is required to track streaming operation.
-*  Use ZBUFF_createCCtx() and ZBUFF_freeCCtx() to create/release resources.
-*  Use ZBUFF_compressInit() to start a new compression operation.
-*  ZBUFF_CCtx objects can be reused multiple times.
-*
-*  Use ZBUFF_compressContinue() repetitively to consume your input.
-*  *srcSizePtr and *dstCapacityPtr can be any size.
-*  The function will report how many bytes were read or written by modifying *srcSizePtr and *dstCapacityPtr.
-*  Note that it may not consume the entire input, in which case it's up to the caller to call again the function with remaining input.
-*  The content of dst will be overwritten (up to *dstCapacityPtr) at each function call, so save its content if it matters or change dst .
-*  @return : a hint to preferred nb of bytes to use as input for next function call (it's only a hint, to improve latency)
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  ZBUFF_compressFlush() can be used to instruct ZBUFF to compress and output whatever remains within its buffer.
-*  Note that it will not output more than *dstCapacityPtr.
-*  Therefore, some content might still be left into its internal buffer if dst buffer is too small.
-*  @return : nb of bytes still present into internal buffer (0 if it's empty)
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  ZBUFF_compressEnd() instructs to finish a frame.
-*  It will perform a flush and write frame epilogue.
-*  Similar to ZBUFF_compressFlush(), it may not be able to output the entire internal buffer content if *dstCapacityPtr is too small.
-*  @return : nb of bytes still present into internal buffer (0 if it's empty)
-*            or an error code, which can be tested using ZBUFF_isError().
-*
-*  Hint : recommended buffer sizes (not compulsory)
-*  input : ZSTD_BLOCKSIZE_MAX (128 KB), internal unit size, it improves latency to use this value.
-*  output : ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + ZBUFF_endFrameSize : ensures it's always possible to write/flush/end a full block at best speed.
-* ***********************************************************/
-
-typedef enum { ZBUFFcs_init, ZBUFFcs_load, ZBUFFcs_flush, ZBUFFcs_final } ZBUFF_cStage;
-
-/* *** Resources *** */
-struct ZBUFF_CCtx_s {
-    ZSTD_CCtx* zc;
-    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;
-    ZBUFF_cStage stage;
-    U32    checksum;
-    U32    frameEnded;
-    ZSTD_customMem customMem;
-};   /* typedef'd tp ZBUFF_CCtx within "zbuff.h" */
-
-ZBUFF_CCtx* ZBUFF_createCCtx(void)
-{
-    return ZBUFF_createCCtx_advanced(defaultCustomMem);
-}
-
-ZBUFF_CCtx* ZBUFF_createCCtx_advanced(ZSTD_customMem customMem)
-{
-    ZBUFF_CCtx* zbc;
-
-    if (!customMem.customAlloc && !customMem.customFree)
-        customMem = defaultCustomMem;
-
-    if (!customMem.customAlloc || !customMem.customFree)
-        return NULL;
-
-    zbc = (ZBUFF_CCtx*)customMem.customAlloc(customMem.opaque, sizeof(ZBUFF_CCtx));
-    if (zbc==NULL) return NULL;
-    memset(zbc, 0, sizeof(ZBUFF_CCtx));
-    memcpy(&zbc->customMem, &customMem, sizeof(ZSTD_customMem));
-    zbc->zc = ZSTD_createCCtx_advanced(customMem);
-    if (zbc->zc == NULL) { ZBUFF_freeCCtx(zbc); return NULL; }
-    return zbc;
-}
-
-size_t ZBUFF_freeCCtx(ZBUFF_CCtx* zbc)
-{
-    if (zbc==NULL) return 0;   /* support free on NULL */
-    ZSTD_freeCCtx(zbc->zc);
-    if (zbc->inBuff) zbc->customMem.customFree(zbc->customMem.opaque, zbc->inBuff);
-    if (zbc->outBuff) zbc->customMem.customFree(zbc->customMem.opaque, zbc->outBuff);
-    zbc->customMem.customFree(zbc->customMem.opaque, zbc);
-    return 0;
-}
-
-
-/* ======   Initialization   ====== */
-
-size_t ZBUFF_compressInit_advanced(ZBUFF_CCtx* zbc,
-                                   const void* dict, size_t dictSize,
-                                   ZSTD_parameters params, unsigned long long pledgedSrcSize)
-{
-    /* allocate buffers */
-    {   size_t const neededInBuffSize = (size_t)1 << params.cParams.windowLog;
-        if (zbc->inBuffSize < neededInBuffSize) {
-            zbc->inBuffSize = neededInBuffSize;
-            zbc->customMem.customFree(zbc->customMem.opaque, zbc->inBuff);   /* should not be necessary */
-            zbc->inBuff = (char*)zbc->customMem.customAlloc(zbc->customMem.opaque, neededInBuffSize);
-            if (zbc->inBuff == NULL) return ERROR(memory_allocation);
-        }
-        zbc->blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, neededInBuffSize);
-    }
-    if (zbc->outBuffSize < ZSTD_compressBound(zbc->blockSize)+1) {
-        zbc->outBuffSize = ZSTD_compressBound(zbc->blockSize)+1;
-        zbc->customMem.customFree(zbc->customMem.opaque, zbc->outBuff);   /* should not be necessary */
-        zbc->outBuff = (char*)zbc->customMem.customAlloc(zbc->customMem.opaque, zbc->outBuffSize);
-        if (zbc->outBuff == NULL) return ERROR(memory_allocation);
-    }
-
-    { size_t const errorCode = ZSTD_compressBegin_advanced(zbc->zc, dict, dictSize, params, pledgedSrcSize);
-      if (ZSTD_isError(errorCode)) return errorCode; }
-
-    zbc->inToCompress = 0;
-    zbc->inBuffPos = 0;
-    zbc->inBuffTarget = zbc->blockSize;
-    zbc->outBuffContentSize = zbc->outBuffFlushedSize = 0;
-    zbc->stage = ZBUFFcs_load;
-    zbc->checksum = params.fParams.checksumFlag > 0;
-    zbc->frameEnded = 0;
-    return 0;   /* ready to go */
-}
-
-
-size_t ZBUFF_compressInitDictionary(ZBUFF_CCtx* zbc, const void* dict, size_t dictSize, int compressionLevel)
-{
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    return ZBUFF_compressInit_advanced(zbc, dict, dictSize, params, 0);
-}
-
-size_t ZBUFF_compressInit(ZBUFF_CCtx* zbc, int compressionLevel)
-{
-    return ZBUFF_compressInitDictionary(zbc, NULL, 0, compressionLevel);
-}
-
-
-/* internal util function */
-MEM_STATIC size_t ZBUFF_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;
-}
-
-
-/* ======   Compression   ====== */
-
-typedef enum { zbf_gather, zbf_flush, zbf_end } ZBUFF_flush_e;
-
-static size_t ZBUFF_compressContinue_generic(ZBUFF_CCtx* zbc,
-                              void* dst, size_t* dstCapacityPtr,
-                        const void* src, size_t* srcSizePtr,
-                              ZBUFF_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(zbc->stage)
-        {
-        case ZBUFFcs_init: return ERROR(init_missing);   /* call ZBUFF_compressInit() first ! */
-
-        case ZBUFFcs_load:
-            /* complete inBuffer */
-            {   size_t const toLoad = zbc->inBuffTarget - zbc->inBuffPos;
-                size_t const loaded = ZBUFF_limitCopy(zbc->inBuff + zbc->inBuffPos, toLoad, ip, iend-ip);
-                zbc->inBuffPos += loaded;
-                ip += loaded;
-                if ( (zbc->inBuffPos==zbc->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 = zbc->inBuffPos - zbc->inToCompress;
-                size_t oSize = oend-op;
-                if (oSize >= ZSTD_compressBound(iSize))
-                    cDst = op;   /* compress directly into output buffer (avoid flush stage) */
-                else
-                    cDst = zbc->outBuff, oSize = zbc->outBuffSize;
-                cSize = (flush == zbf_end) ?
-                        ZSTD_compressEnd(zbc->zc, cDst, oSize, zbc->inBuff + zbc->inToCompress, iSize) :
-                        ZSTD_compressContinue(zbc->zc, cDst, oSize, zbc->inBuff + zbc->inToCompress, iSize);
-                if (ZSTD_isError(cSize)) return cSize;
-                if (flush == zbf_end) zbc->frameEnded = 1;
-                /* prepare next block */
-                zbc->inBuffTarget = zbc->inBuffPos + zbc->blockSize;
-                if (zbc->inBuffTarget > zbc->inBuffSize)
-                    zbc->inBuffPos = 0, zbc->inBuffTarget = zbc->blockSize;   /* note : inBuffSize >= blockSize */
-                zbc->inToCompress = zbc->inBuffPos;
-                if (cDst == op) { op += cSize; break; }   /* no need to flush */
-                zbc->outBuffContentSize = cSize;
-                zbc->outBuffFlushedSize = 0;
-                zbc->stage = ZBUFFcs_flush;   /* continue to flush stage */
-            }
-
-        case ZBUFFcs_flush:
-            {   size_t const toFlush = zbc->outBuffContentSize - zbc->outBuffFlushedSize;
-                size_t const flushed = ZBUFF_limitCopy(op, oend-op, zbc->outBuff + zbc->outBuffFlushedSize, toFlush);
-                op += flushed;
-                zbc->outBuffFlushedSize += flushed;
-                if (toFlush!=flushed) { someMoreWork = 0; break; } /* dst too small to store flushed data : stop there */
-                zbc->outBuffContentSize = zbc->outBuffFlushedSize = 0;
-                zbc->stage = ZBUFFcs_load;
-                break;
-            }
-
-        case ZBUFFcs_final:
-            someMoreWork = 0;   /* do nothing */
-            break;
-
-        default:
-            return ERROR(GENERIC);   /* impossible */
-        }
-    }
-
-    *srcSizePtr = ip - istart;
-    *dstCapacityPtr = op - ostart;
-    if (zbc->frameEnded) return 0;
-    {   size_t hintInSize = zbc->inBuffTarget - zbc->inBuffPos;
-        if (hintInSize==0) hintInSize = zbc->blockSize;
-        return hintInSize;
-    }
-}
-
-size_t ZBUFF_compressContinue(ZBUFF_CCtx* zbc,
-                              void* dst, size_t* dstCapacityPtr,
-                        const void* src, size_t* srcSizePtr)
-{
-    return ZBUFF_compressContinue_generic(zbc, dst, dstCapacityPtr, src, srcSizePtr, zbf_gather);
-}
-
-
-
-/* ======   Finalize   ====== */
-
-size_t ZBUFF_compressFlush(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
-{
-    size_t srcSize = 0;
-    ZBUFF_compressContinue_generic(zbc, dst, dstCapacityPtr, &srcSize, &srcSize, zbf_flush);  /* use a valid src address instead of NULL */
-    return zbc->outBuffContentSize - zbc->outBuffFlushedSize;
-}
-
-
-size_t ZBUFF_compressEnd(ZBUFF_CCtx* zbc, void* dst, size_t* dstCapacityPtr)
-{
-    BYTE* const ostart = (BYTE*)dst;
-    BYTE* const oend = ostart + *dstCapacityPtr;
-    BYTE* op = ostart;
-
-    if (zbc->stage != ZBUFFcs_final) {
-        /* flush whatever remains */
-        size_t outSize = *dstCapacityPtr;
-        size_t srcSize = 0;
-        size_t const notEnded = ZBUFF_compressContinue_generic(zbc, dst, &outSize, &srcSize, &srcSize, zbf_end);  /* use a valid address instead of NULL */
-        size_t const remainingToFlush = zbc->outBuffContentSize - zbc->outBuffFlushedSize;
-        op += outSize;
-        if (remainingToFlush) {
-            *dstCapacityPtr = op-ostart;
-            return remainingToFlush + ZBUFF_endFrameSize + (zbc->checksum * 4);
-        }
-        /* create epilogue */
-        zbc->stage = ZBUFFcs_final;
-        zbc->outBuffContentSize = !notEnded ? 0 :
-            ZSTD_compressEnd(zbc->zc, zbc->outBuff, zbc->outBuffSize, NULL, 0);  /* write epilogue into outBuff */
-    }
-
-    /* flush epilogue */
-    {   size_t const toFlush = zbc->outBuffContentSize - zbc->outBuffFlushedSize;
-        size_t const flushed = ZBUFF_limitCopy(op, oend-op, zbc->outBuff + zbc->outBuffFlushedSize, toFlush);
-        op += flushed;
-        zbc->outBuffFlushedSize += flushed;
-        *dstCapacityPtr = op-ostart;
-        if (toFlush==flushed) zbc->stage = ZBUFFcs_init;  /* end reached */
-        return toFlush - flushed;
-    }
-}
-
-
-
-/* *************************************
-*  Tool functions
-***************************************/
-size_t ZBUFF_recommendedCInSize(void)  { return ZSTD_BLOCKSIZE_ABSOLUTEMAX; }
-size_t ZBUFF_recommendedCOutSize(void) { return ZSTD_compressBound(ZSTD_BLOCKSIZE_ABSOLUTEMAX) + ZSTD_blockHeaderSize + ZBUFF_endFrameSize; }
diff --git a/contrib/zstd/zbuff_decompress.c b/contrib/zstd/zbuff_decompress.c
deleted file mode 100644
index b20ee9705..000000000
--- a/contrib/zstd/zbuff_decompress.c
+++ /dev/null
@@ -1,252 +0,0 @@
-/**
- * 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 <stdlib.h>
-#include "error_private.h"
-#include "zstd_internal.h"  /* MIN, ZSTD_blockHeaderSize, ZSTD_BLOCKSIZE_MAX */
-#define ZBUFF_STATIC_LINKING_ONLY
-#include "zbuff.h"
-
-
-typedef enum { ZBUFFds_init, ZBUFFds_loadHeader,
-               ZBUFFds_read, ZBUFFds_load, ZBUFFds_flush } ZBUFF_dStage;
-
-/* *** Resource management *** */
-struct ZBUFF_DCtx_s {
-    ZSTD_DCtx* zd;
-    ZSTD_frameParams fParams;
-    ZBUFF_dStage stage;
-    char*  inBuff;
-    size_t inBuffSize;
-    size_t inPos;
-    char*  outBuff;
-    size_t outBuffSize;
-    size_t outStart;
-    size_t outEnd;
-    size_t blockSize;
-    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
-    size_t lhSize;
-    ZSTD_customMem customMem;
-};   /* typedef'd to ZBUFF_DCtx within "zbuff.h" */
-
-
-ZBUFF_DCtx* ZBUFF_createDCtx(void)
-{
-    return ZBUFF_createDCtx_advanced(defaultCustomMem);
-}
-
-ZBUFF_DCtx* ZBUFF_createDCtx_advanced(ZSTD_customMem customMem)
-{
-    ZBUFF_DCtx* zbd;
-
-    if (!customMem.customAlloc && !customMem.customFree)
-        customMem = defaultCustomMem;
-
-    if (!customMem.customAlloc || !customMem.customFree)
-        return NULL;
-
-    zbd = (ZBUFF_DCtx*)customMem.customAlloc(customMem.opaque, sizeof(ZBUFF_DCtx));
-    if (zbd==NULL) return NULL;
-    memset(zbd, 0, sizeof(ZBUFF_DCtx));
-    memcpy(&zbd->customMem, &customMem, sizeof(ZSTD_customMem));
-    zbd->zd = ZSTD_createDCtx_advanced(customMem);
-    if (zbd->zd == NULL) { ZBUFF_freeDCtx(zbd); return NULL; }
-    zbd->stage = ZBUFFds_init;
-    return zbd;
-}
-
-size_t ZBUFF_freeDCtx(ZBUFF_DCtx* zbd)
-{
-    if (zbd==NULL) return 0;   /* support free on null */
-    ZSTD_freeDCtx(zbd->zd);
-    if (zbd->inBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
-    if (zbd->outBuff) zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
-    zbd->customMem.customFree(zbd->customMem.opaque, zbd);
-    return 0;
-}
-
-
-/* *** Initialization *** */
-
-size_t ZBUFF_decompressInitDictionary(ZBUFF_DCtx* zbd, const void* dict, size_t dictSize)
-{
-    zbd->stage = ZBUFFds_loadHeader;
-    zbd->lhSize = zbd->inPos = zbd->outStart = zbd->outEnd = 0;
-    return ZSTD_decompressBegin_usingDict(zbd->zd, dict, dictSize);
-}
-
-size_t ZBUFF_decompressInit(ZBUFF_DCtx* zbd)
-{
-    return ZBUFF_decompressInitDictionary(zbd, NULL, 0);
-}
-
-
-/* internal util function */
-MEM_STATIC size_t ZBUFF_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;
-}
-
-
-/* *** Decompression *** */
-
-size_t ZBUFF_decompressContinue(ZBUFF_DCtx* zbd,
-                                void* dst, size_t* dstCapacityPtr,
-                          const void* src, size_t* srcSizePtr)
-{
-    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;
-    U32 someMoreWork = 1;
-
-    while (someMoreWork) {
-        switch(zbd->stage)
-        {
-        case ZBUFFds_init :
-            return ERROR(init_missing);
-
-        case ZBUFFds_loadHeader :
-            {   size_t const hSize = ZSTD_getFrameParams(&(zbd->fParams), zbd->headerBuffer, zbd->lhSize);
-                if (ZSTD_isError(hSize)) return hSize;
-                if (hSize != 0) {   /* need more input */
-                    size_t const toLoad = hSize - zbd->lhSize;   /* if hSize!=0, hSize > zbd->lhSize */
-                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
-                        memcpy(zbd->headerBuffer + zbd->lhSize, ip, iend-ip);
-                        zbd->lhSize += iend-ip;
-                        *dstCapacityPtr = 0;
-                        return (hSize - zbd->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
-                    }
-                    memcpy(zbd->headerBuffer + zbd->lhSize, ip, toLoad); zbd->lhSize = hSize; ip += toLoad;
-                    break;
-            }   }
-
-            /* Consume header */
-            {   size_t const h1Size = ZSTD_nextSrcSizeToDecompress(zbd->zd);  /* == ZSTD_frameHeaderSize_min */
-                size_t const h1Result = ZSTD_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer, h1Size);
-                if (ZSTD_isError(h1Result)) return h1Result;   /* should not happen : already checked */
-                if (h1Size < zbd->lhSize) {   /* long header */
-                    size_t const h2Size = ZSTD_nextSrcSizeToDecompress(zbd->zd);
-                    size_t const h2Result = ZSTD_decompressContinue(zbd->zd, NULL, 0, zbd->headerBuffer+h1Size, h2Size);
-                    if (ZSTD_isError(h2Result)) return h2Result;
-            }   }
-
-            zbd->fParams.windowSize = MAX(zbd->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
-
-            /* Frame header instruct buffer sizes */
-            {   size_t const blockSize = MIN(zbd->fParams.windowSize, ZSTD_BLOCKSIZE_ABSOLUTEMAX);
-                size_t const neededOutSize = zbd->fParams.windowSize + blockSize;
-                zbd->blockSize = blockSize;
-                if (zbd->inBuffSize < blockSize) {
-                    zbd->customMem.customFree(zbd->customMem.opaque, zbd->inBuff);
-                    zbd->inBuffSize = blockSize;
-                    zbd->inBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, blockSize);
-                    if (zbd->inBuff == NULL) return ERROR(memory_allocation);
-                }
-                if (zbd->outBuffSize < neededOutSize) {
-                    zbd->customMem.customFree(zbd->customMem.opaque, zbd->outBuff);
-                    zbd->outBuffSize = neededOutSize;
-                    zbd->outBuff = (char*)zbd->customMem.customAlloc(zbd->customMem.opaque, neededOutSize);
-                    if (zbd->outBuff == NULL) return ERROR(memory_allocation);
-            }   }
-            zbd->stage = ZBUFFds_read;
-            /* pass-through */
-
-        case ZBUFFds_read:
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zbd->zd);
-                if (neededInSize==0) {  /* end of frame */
-                    zbd->stage = ZBUFFds_init;
-                    someMoreWork = 0;
-                    break;
-                }
-                if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
-                    const int isSkipFrame = ZSTD_isSkipFrame(zbd->zd);
-                    size_t const decodedSize = ZSTD_decompressContinue(zbd->zd,
-                        zbd->outBuff + zbd->outStart, (isSkipFrame ? 0 : zbd->outBuffSize - zbd->outStart),
-                        ip, neededInSize);
-                    if (ZSTD_isError(decodedSize)) return decodedSize;
-                    ip += neededInSize;
-                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
-                    zbd->outEnd = zbd->outStart +  decodedSize;
-                    zbd->stage = ZBUFFds_flush;
-                    break;
-                }
-                if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
-                zbd->stage = ZBUFFds_load;
-                /* pass-through */
-            }
-
-        case ZBUFFds_load:
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zbd->zd);
-                size_t const toLoad = neededInSize - zbd->inPos;   /* should always be <= remaining space within inBuff */
-                size_t loadedSize;
-                if (toLoad > zbd->inBuffSize - zbd->inPos) return ERROR(corruption_detected);   /* should never happen */
-                loadedSize = ZBUFF_limitCopy(zbd->inBuff + zbd->inPos, toLoad, ip, iend-ip);
-                ip += loadedSize;
-                zbd->inPos += loadedSize;
-                if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
-
-                /* decode loaded input */
-                {  const int isSkipFrame = ZSTD_isSkipFrame(zbd->zd);
-                   size_t const decodedSize = ZSTD_decompressContinue(zbd->zd,
-                        zbd->outBuff + zbd->outStart, zbd->outBuffSize - zbd->outStart,
-                        zbd->inBuff, neededInSize);
-                    if (ZSTD_isError(decodedSize)) return decodedSize;
-                    zbd->inPos = 0;   /* input is consumed */
-                    if (!decodedSize && !isSkipFrame) { zbd->stage = ZBUFFds_read; break; }   /* this was just a header */
-                    zbd->outEnd = zbd->outStart +  decodedSize;
-                    zbd->stage = ZBUFFds_flush;
-                    /* pass-through */
-            }   }
-
-        case ZBUFFds_flush:
-            {   size_t const toFlushSize = zbd->outEnd - zbd->outStart;
-                size_t const flushedSize = ZBUFF_limitCopy(op, oend-op, zbd->outBuff + zbd->outStart, toFlushSize);
-                op += flushedSize;
-                zbd->outStart += flushedSize;
-                if (flushedSize == toFlushSize) {  /* flush completed */
-                    zbd->stage = ZBUFFds_read;
-                    if (zbd->outStart + zbd->blockSize > zbd->outBuffSize)
-                        zbd->outStart = zbd->outEnd = 0;
-                    break;
-                }
-                /* cannot flush everything */
-                someMoreWork = 0;
-                break;
-            }
-        default: return ERROR(GENERIC);   /* impossible */
-    }   }
-
-    /* result */
-    *srcSizePtr = ip-istart;
-    *dstCapacityPtr = op-ostart;
-    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zbd->zd);
-        if (!nextSrcSizeHint) return (zbd->outEnd != zbd->outStart);   /* return 0 only if fully flushed too */
-        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zbd->zd) == ZSTDnit_block);
-        if (zbd->inPos > nextSrcSizeHint) return ERROR(GENERIC);   /* should never happen */
-        nextSrcSizeHint -= zbd->inPos;   /* already loaded*/
-        return nextSrcSizeHint;
-    }
-}
-
-
-/* *************************************
-*  Tool functions
-***************************************/
-size_t ZBUFF_recommendedDInSize(void)  { return ZSTD_BLOCKSIZE_ABSOLUTEMAX + ZSTD_blockHeaderSize /* block header size*/ ; }
-size_t ZBUFF_recommendedDOutSize(void) { return ZSTD_BLOCKSIZE_ABSOLUTEMAX; }
diff --git a/contrib/zstd/zdict.c b/contrib/zstd/zdict.c
index adfe55cf7..179e02eff 100644
--- a/contrib/zstd/zdict.c
+++ b/contrib/zstd/zdict.c
@@ -11,8 +11,9 @@
 /*-**************************************
 *  Tuning parameters
 ****************************************/
+#define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
 #define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
-#define ZDICT_MIN_SAMPLES_SIZE 512
+#define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
 
 
 /*-**************************************
@@ -36,12 +37,11 @@
 #include <time.h>          /* clock */
 
 #include "mem.h"           /* read */
-#include "error_private.h"
 #include "fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
 #define HUF_STATIC_LINKING_ONLY
-#include "huf.h"
+#include "huf.h"           /* HUF_buildCTable, HUF_writeCTable */
 #include "zstd_internal.h" /* includes zstd.h */
-#include "xxhash.h"
+#include "xxhash.h"        /* XXH64 */
 #include "divsufsort.h"
 #ifndef ZDICT_STATIC_LINKING_ONLY
 #  define ZDICT_STATIC_LINKING_ONLY
@@ -60,37 +60,26 @@
 
 #define NOISELENGTH 32
 
-#define MINRATIO 4
-static const int g_compressionLevel_default = 5;
+static const int g_compressionLevel_default = 6;
 static const U32 g_selectivity_default = 9;
-static const size_t g_provision_entropySize = 200;
-static const size_t g_min_fast_dictContent = 192;
 
 
 /*-*************************************
 *  Console display
 ***************************************/
 #define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
-#define DISPLAYLEVEL(l, ...) if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); }
-static unsigned g_displayLevel = 0;   /* 0 : no display;   1: errors;   2: default;  4: full information */
-
-#define DISPLAYUPDATE(l, ...) if (g_displayLevel>=l) { \
-            if (ZDICT_clockSpan(g_time) > refreshRate)  \
-            { g_time = clock(); DISPLAY(__VA_ARGS__); \
-            if (g_displayLevel>=4) fflush(stdout); } }
-static const clock_t refreshRate = CLOCKS_PER_SEC * 3 / 10;
-static clock_t g_time = 0;
+#define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); }    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
 
 static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
 
-static void ZDICT_printHex(U32 dlevel, const void* ptr, size_t length)
+static void ZDICT_printHex(const void* ptr, size_t length)
 {
     const BYTE* const b = (const BYTE*)ptr;
     size_t u;
     for (u=0; u<length; u++) {
         BYTE c = b[u];
         if (c<32 || c>126) c = '.';   /* non-printable char */
-        DISPLAYLEVEL(dlevel, "%c", c);
+        DISPLAY("%c", c);
     }
 }
 
@@ -211,7 +200,7 @@ static void ZDICT_initDictItem(dictItem* d)
 static dictItem ZDICT_analyzePos(
                        BYTE* doneMarks,
                        const int* suffix, U32 start,
-                       const void* buffer, U32 minRatio)
+                       const void* buffer, U32 minRatio, U32 notificationLevel)
 {
     U32 lengthList[LLIMIT] = {0};
     U32 cumulLength[LLIMIT] = {0};
@@ -315,13 +304,13 @@ static dictItem ZDICT_analyzePos(
         } while (length >=MINMATCHLENGTH);
 
         /* look backward */
-		length = MINMATCHLENGTH;
-		while ((length >= MINMATCHLENGTH) & (start > 0)) {
-			length = ZDICT_count(b + pos, b + suffix[start - 1]);
-			if (length >= LLIMIT) length = LLIMIT - 1;
-			lengthList[length]++;
-			if (length >= MINMATCHLENGTH) start--;
-		}
+        length = MINMATCHLENGTH;
+        while ((length >= MINMATCHLENGTH) & (start > 0)) {
+            length = ZDICT_count(b + pos, b + suffix[start - 1]);
+            if (length >= LLIMIT) length = LLIMIT - 1;
+            lengthList[length]++;
+            if (length >= MINMATCHLENGTH) start--;
+        }
 
         /* largest useful length */
         memset(cumulLength, 0, sizeof(cumulLength));
@@ -372,28 +361,43 @@ static dictItem ZDICT_analyzePos(
 }
 
 
+static int isIncluded(const void* in, const void* container, size_t length)
+{
+    const char* const ip = (const char*) in;
+    const char* const into = (const char*) container;
+    size_t u;
+
+    for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
+        if (ip[u] != into[u]) break;
+    }
+
+    return u==length;
+}
+
 /*! ZDICT_checkMerge
     check if dictItem can be merged, do it if possible
     @return : id of destination elt, 0 if not merged
 */
-static U32 ZDICT_checkMerge(dictItem* table, dictItem elt, U32 eltNbToSkip)
+static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
 {
     const U32 tableSize = table->pos;
-    const U32 max = elt.pos + (elt.length-1);
+    const U32 eltEnd = elt.pos + elt.length;
+    const char* const buf = (const char*) buffer;
 
     /* tail overlap */
     U32 u; for (u=1; u<tableSize; u++) {
         if (u==eltNbToSkip) continue;
-        if ((table[u].pos > elt.pos) && (table[u].pos < max)) {  /* overlap */
+        if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
             /* append */
-            U32 addedLength = table[u].pos - elt.pos;
+            U32 const addedLength = table[u].pos - elt.pos;
             table[u].length += addedLength;
             table[u].pos = elt.pos;
             table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
-            table[u].savings += elt.length / 8;    /* rough approx */
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
             elt = table[u];
+            /* sort : improve rank */
             while ((u>1) && (table[u-1].savings < elt.savings))
-                table[u] = table[u-1], u--;
+            table[u] = table[u-1], u--;
             table[u] = elt;
             return u;
     }   }
@@ -401,20 +405,33 @@ static U32 ZDICT_checkMerge(dictItem* table, dictItem elt, U32 eltNbToSkip)
     /* front overlap */
     for (u=1; u<tableSize; u++) {
         if (u==eltNbToSkip) continue;
-        if ((table[u].pos + table[u].length > elt.pos) && (table[u].pos < elt.pos)) {  /* overlap */
+
+        if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
             /* append */
-            int addedLength = (elt.pos + elt.length) - (table[u].pos + table[u].length);
-            table[u].savings += elt.length / 8;    /* rough approx */
-            if (addedLength > 0) {   /* otherwise, already included */
+            int const addedLength = (int)eltEnd - (table[u].pos + table[u].length);
+            table[u].savings += elt.length / 8;    /* rough approx bonus */
+            if (addedLength > 0) {   /* otherwise, elt fully included into existing */
                 table[u].length += addedLength;
                 table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
             }
+            /* sort : improve rank */
             elt = table[u];
             while ((u>1) && (table[u-1].savings < elt.savings))
                 table[u] = table[u-1], u--;
             table[u] = elt;
             return u;
-    }   }
+        }
+
+        if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
+            if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
+                size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 );
+                table[u].pos = elt.pos;
+                table[u].savings += (U32)(elt.savings * addedLength / elt.length);
+                table[u].length = MIN(elt.length, table[u].length + 1);
+                return u;
+            }
+        }
+    }
 
     return 0;
 }
@@ -432,14 +449,14 @@ static void ZDICT_removeDictItem(dictItem* table, U32 id)
 }
 
 
-static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt)
+static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
 {
     /* merge if possible */
-    U32 mergeId = ZDICT_checkMerge(table, elt, 0);
+    U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
     if (mergeId) {
         U32 newMerge = 1;
         while (newMerge) {
-            newMerge = ZDICT_checkMerge(table, table[mergeId], mergeId);
+            newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
             if (newMerge) ZDICT_removeDictItem(table, mergeId);
             mergeId = newMerge;
         }
@@ -473,7 +490,7 @@ static U32 ZDICT_dictSize(const dictItem* dictList)
 static size_t ZDICT_trainBuffer(dictItem* dictList, U32 dictListSize,
                             const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
                             const size_t* fileSizes, unsigned nbFiles,
-                            U32 minRatio)
+                            U32 minRatio, U32 notificationLevel)
 {
     int* const suffix0 = (int*)malloc((bufferSize+2)*sizeof(*suffix0));
     int* const suffix = suffix0+1;
@@ -481,6 +498,13 @@ static size_t ZDICT_trainBuffer(dictItem* dictList, U32 dictListSize,
     BYTE* doneMarks = (BYTE*)malloc((bufferSize+16)*sizeof(*doneMarks));   /* +16 for overflow security */
     U32* filePos = (U32*)malloc(nbFiles * sizeof(*filePos));
     size_t result = 0;
+    clock_t displayClock = 0;
+    clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
+
+#   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
+            if (ZDICT_clockSpan(displayClock) > refreshRate)  \
+            { displayClock = clock(); DISPLAY(__VA_ARGS__); \
+            if (notificationLevel>=4) fflush(stderr); } }
 
     /* init */
     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
@@ -506,7 +530,8 @@ static size_t ZDICT_trainBuffer(dictItem* dictList, U32 dictListSize,
     {   size_t pos;
         for (pos=0; pos < bufferSize; pos++)
             reverseSuffix[suffix[pos]] = (U32)pos;
-        /* build file pos */
+        /* note filePos tracks borders between samples.
+           It's not used at this stage, but planned to become useful in a later update */
         filePos[0] = 0;
         for (pos=1; pos<nbFiles; pos++)
             filePos[pos] = (U32)(filePos[pos-1] + fileSizes[pos-1]);
@@ -518,9 +543,9 @@ static size_t ZDICT_trainBuffer(dictItem* dictList, U32 dictListSize,
     {   U32 cursor; for (cursor=0; cursor < bufferSize; ) {
             dictItem solution;
             if (doneMarks[cursor]) { cursor++; continue; }
-            solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio);
+            solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
             if (solution.length==0) { cursor++; continue; }
-            ZDICT_insertDictItem(dictList, dictListSize, solution);
+            ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
             cursor += solution.length;
             DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / bufferSize * 100);
     }   }
@@ -558,17 +583,17 @@ typedef struct
 
 static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
                             U32* countLit, U32* offsetcodeCount, U32* matchlengthCount, U32* litlengthCount, U32* repOffsets,
-                            const void* src, size_t srcSize)
+                            const void* src, size_t srcSize, U32 notificationLevel)
 {
     size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_ABSOLUTEMAX, 1 << params.cParams.windowLog);
     size_t cSize;
 
     if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
-	{	size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref);
-		if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
-	}
+    {  size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref, 0);
+            if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
+    }
     cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_ABSOLUTEMAX, src, srcSize);
-    if (ZSTD_isError(cSize)) { DISPLAYLEVEL(1, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
+    if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
 
     if (cSize) {  /* if == 0; block is not compressible */
         const seqStore_t* seqStorePtr = ZSTD_getSeqStore(esr.zc);
@@ -647,9 +672,10 @@ static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val,
 
 #define OFFCODE_MAX 30  /* only applicable to first block */
 static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
-                                 unsigned compressionLevel,
-                           const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
-                           const void* dictBuffer, size_t  dictBufferSize)
+                                   unsigned compressionLevel,
+                             const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
+                             const void* dictBuffer, size_t  dictBufferSize,
+                                   unsigned notificationLevel)
 {
     U32 countLit[256];
     HUF_CREATE_STATIC_CTABLE(hufTable, 255);
@@ -681,27 +707,28 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
         goto _cleanup;
     }
     if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionary_wrong); goto _cleanup; }   /* too large dictionary */
-    for (u=0; u<256; u++) countLit[u]=1;   /* any character must be described */
-    for (u=0; u<=offcodeMax; u++) offcodeCount[u]=1;
-    for (u=0; u<=MaxML; u++) matchLengthCount[u]=1;
-    for (u=0; u<=MaxLL; u++) litLengthCount[u]=1;
+    for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
+    for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
+    for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
+    for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
     memset(repOffset, 0, sizeof(repOffset));
     repOffset[1] = repOffset[4] = repOffset[8] = 1;
     memset(bestRepOffset, 0, sizeof(bestRepOffset));
-    if (compressionLevel==0) compressionLevel=g_compressionLevel_default;
+    if (compressionLevel==0) compressionLevel = g_compressionLevel_default;
     params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
-	{	size_t const beginResult = ZSTD_compressBegin_advanced(esr.ref, dictBuffer, dictBufferSize, params, 0);
-		if (ZSTD_isError(beginResult)) {
-			eSize = ERROR(GENERIC);
-			DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced failed \n");
-			goto _cleanup;
-	}	}
+    {   size_t const beginResult = ZSTD_compressBegin_advanced(esr.ref, dictBuffer, dictBufferSize, params, 0);
+        if (ZSTD_isError(beginResult)) {
+            DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced() failed : %s \n", ZSTD_getErrorName(beginResult));
+            eSize = ERROR(GENERIC);
+            goto _cleanup;
+    }   }
 
     /* collect stats on all files */
     for (u=0; u<nbFiles; u++) {
         ZDICT_countEStats(esr, params,
-                        countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
-           (const char*)srcBuffer + pos, fileSizes[u]);
+                          countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
+                         (const char*)srcBuffer + pos, fileSizes[u],
+                          notificationLevel);
         pos += fileSizes[u];
     }
 
@@ -748,7 +775,6 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     }
     llLog = (U32)errorCode;
 
-
     /* write result to buffer */
     {   size_t const hhSize = HUF_writeCTable(dstPtr, maxDstSize, hufTable, 255, huffLog);
         if (HUF_isError(hhSize)) {
@@ -810,7 +836,6 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     MEM_writeLE32(dstPtr+4, repStartValue[1]);
     MEM_writeLE32(dstPtr+8, repStartValue[2]);
 #endif
-    dstPtr += 12;
     eSize += 12;
 
 _cleanup:
@@ -822,33 +847,82 @@ _cleanup:
 }
 
 
-size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                                                 ZDICT_params_t params)
+
+size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
+                          const void* customDictContent, size_t dictContentSize,
+                          const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                          ZDICT_params_t params)
 {
     size_t hSize;
+#define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
+    BYTE header[HBUFFSIZE];
     int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+
+    /* check conditions */
+    if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
+    if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
+    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
 
     /* dictionary header */
-    MEM_writeLE32(dictBuffer, ZSTD_DICT_MAGIC);
-    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
+    MEM_writeLE32(header, ZSTD_DICT_MAGIC);
+    {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
         U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
         U32 const dictID = params.dictID ? params.dictID : compliantID;
-        MEM_writeLE32((char*)dictBuffer+4, dictID);
+        MEM_writeLE32(header+4, dictID);
     }
     hSize = 8;
 
     /* entropy tables */
     DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
     DISPLAYLEVEL(2, "statistics ... \n");
+    {   size_t const eSize = ZDICT_analyzeEntropy(header+hSize, HBUFFSIZE-hSize,
+                                  compressionLevel,
+                                  samplesBuffer, samplesSizes, nbSamples,
+                                  customDictContent, dictContentSize,
+                                  notificationLevel);
+        if (ZDICT_isError(eSize)) return eSize;
+        hSize += eSize;
+    }
+
+    /* copy elements in final buffer ; note : src and dst buffer can overlap */
+    if (hSize + dictContentSize > dictBufferCapacity) dictContentSize = dictBufferCapacity - hSize;
+    {   size_t const dictSize = hSize + dictContentSize;
+        char* dictEnd = (char*)dictBuffer + dictSize;
+        memmove(dictEnd - dictContentSize, customDictContent, dictContentSize);
+        memcpy(dictBuffer, header, hSize);
+        return dictSize;
+    }
+}
+
+
+size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                                                 ZDICT_params_t params)
+{
+    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
+    U32 const notificationLevel = params.notificationLevel;
+    size_t hSize = 8;
+
+    /* calculate entropy tables */
+    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
+    DISPLAYLEVEL(2, "statistics ... \n");
     {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
                                   compressionLevel,
                                   samplesBuffer, samplesSizes, nbSamples,
-                                  (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
+                                  (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize,
+                                  notificationLevel);
         if (ZDICT_isError(eSize)) return eSize;
         hSize += eSize;
     }
 
+    /* add dictionary header (after entropy tables) */
+    MEM_writeLE32(dictBuffer, ZSTD_DICT_MAGIC);
+    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
+        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
+        U32 const dictID = params.dictID ? params.dictID : compliantID;
+        MEM_writeLE32((char*)dictBuffer+4, dictID);
+    }
 
     if (hSize + dictContentSize < dictBufferCapacity)
         memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
@@ -872,58 +946,61 @@ size_t ZDICT_trainFromBuffer_unsafe(
     size_t const targetDictSize = maxDictSize;
     size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
     size_t dictSize = 0;
+    U32 const notificationLevel = params.notificationLevel;
 
     /* checks */
     if (!dictList) return ERROR(memory_allocation);
-    if (maxDictSize <= g_provision_entropySize + g_min_fast_dictContent) { free(dictList); return ERROR(dstSize_tooSmall); }
-    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return 0; }   /* not enough source to create dictionary */
+    if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
+    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
 
     /* init */
     ZDICT_initDictItem(dictList);
-    g_displayLevel = params.notificationLevel;
 
     /* build dictionary */
     ZDICT_trainBuffer(dictList, dictListSize,
                     samplesBuffer, samplesBuffSize,
                     samplesSizes, nbSamples,
-                    minRep);
+                    minRep, notificationLevel);
 
     /* display best matches */
-    if (g_displayLevel>= 3) {
+    if (params.notificationLevel>= 3) {
         U32 const nb = MIN(25, dictList[0].pos);
         U32 const dictContentSize = ZDICT_dictSize(dictList);
         U32 u;
-        DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", dictList[0].pos, dictContentSize);
-        DISPLAYLEVEL(3, "list %u best segments \n", nb);
-        for (u=1; u<=nb; u++) {
-            U32 pos = dictList[u].pos;
-            U32 length = dictList[u].length;
-            U32 printedLength = MIN(40, length);
+        DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", dictList[0].pos-1, dictContentSize);
+        DISPLAYLEVEL(3, "list %u best segments \n", nb-1);
+        for (u=1; u<nb; u++) {
+            U32 const pos = dictList[u].pos;
+            U32 const length = dictList[u].length;
+            U32 const printedLength = MIN(40, length);
+            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize))
+                return ERROR(GENERIC);   /* should never happen */
             DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
                          u, length, pos, dictList[u].savings);
-            ZDICT_printHex(3, (const char*)samplesBuffer+pos, printedLength);
+            ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
             DISPLAYLEVEL(3, "| \n");
     }   }
 
 
     /* create dictionary */
     {   U32 dictContentSize = ZDICT_dictSize(dictList);
-        if (dictContentSize < targetDictSize/2) {
+        if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
+        if (dictContentSize < targetDictSize/4) {
             DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (U32)maxDictSize);
+            if (samplesBuffSize < 10 * targetDictSize)
+                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (U32)(samplesBuffSize>>20));
             if (minRep > MINRATIO) {
                 DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
                 DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
             }
-            if (samplesBuffSize < 10 * targetDictSize)
-                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (U32)(samplesBuffSize>>20));
         }
 
-        if ((dictContentSize > targetDictSize*2) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
+        if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
             U32 proposedSelectivity = selectivity-1;
             while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
             DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (U32)maxDictSize);
-            DISPLAYLEVEL(2, "!  you may consider decreasing selectivity to produce denser dictionary (-s%u) \n", proposedSelectivity);
-            DISPLAYLEVEL(2, "!  but test its efficiency on samples \n");
+            DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
+            DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
         }
 
         /* limit dictionary size */
diff --git a/contrib/zstd/zdict.h b/contrib/zstd/zdict.h
index c84aedd1f..9b53de346 100644
--- a/contrib/zstd/zdict.h
+++ b/contrib/zstd/zdict.h
@@ -19,15 +19,18 @@ extern "C" {
 #include <stddef.h>  /* size_t */
 
 
-/*======  Export for Windows  ======*/
-/*!
-*  ZSTD_DLL_EXPORT :
-*  Enable exporting of functions when building a Windows DLL
-*/
-#if defined(_WIN32) && defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZDICTLIB_API __declspec(dllexport)
+/* =====   ZDICTLIB_API : control library symbols visibility   ===== */
+#if defined(__GNUC__) && (__GNUC__ >= 4)
+#  define ZDICTLIB_VISIBILITY __attribute__ ((visibility ("default")))
+#else
+#  define ZDICTLIB_VISIBILITY
+#endif
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBILITY
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
 #else
-#  define ZDICTLIB_API
+#  define ZDICTLIB_API ZDICTLIB_VISIBILITY
 #endif
 
 
@@ -68,7 +71,7 @@ typedef struct {
     int      compressionLevel;   /* 0 means default; target a specific zstd compression level */
     unsigned notificationLevel;  /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
     unsigned dictID;             /* 0 means auto mode (32-bits random value); other : force dictID value */
-    unsigned reserved[2];        /* space for future parameters */
+    unsigned reserved[2];        /* reserved space for future parameters */
 } ZDICT_params_t;
 
 
@@ -79,29 +82,116 @@ typedef struct {
               or an error code, which can be tested by ZDICT_isError().
     note : ZDICT_trainFromBuffer_advanced() will send notifications into stderr if instructed to, using notificationLevel>0.
 */
-size_t ZDICT_trainFromBuffer_advanced(void* dictBuffer, size_t dictBufferCapacity,
+ZDICTLIB_API size_t ZDICT_trainFromBuffer_advanced(void* dictBuffer, size_t dictBufferCapacity,
                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
                                 ZDICT_params_t parameters);
 
+/*! COVER_params_t :
+    For all values 0 means default.
+    k and d are the only required parameters.
+*/
+typedef struct {
+    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
+    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
+    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (32) : Higher means more parameters checked */
+
+    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
+    unsigned notificationLevel;  /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
+    unsigned dictID;             /* 0 means auto mode (32-bits random value); other : force dictID value */
+    int      compressionLevel;   /* 0 means default; target a specific zstd compression level */
+} COVER_params_t;
 
-/*! ZDICT_addEntropyTablesFromBuffer() :
 
-    Given a content-only dictionary (built using any 3rd party algorithm),
-    add entropy tables computed from an array of samples.
+/*! COVER_trainFromBuffer() :
+    Train a dictionary from an array of samples using the COVER algorithm.
+    Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
+    supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
+    The resulting dictionary will be saved into `dictBuffer`.
+    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+              or an error code, which can be tested with ZDICT_isError().
+    Note : COVER_trainFromBuffer() requires about 9 bytes of memory for each input byte.
+    Tips : In general, a reasonable dictionary has a size of ~ 100 KB.
+           It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
+           In general, it's recommended to provide a few thousands samples, but this can vary a lot.
+           It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
+*/
+ZDICTLIB_API size_t COVER_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                              COVER_params_t parameters);
+
+/*! COVER_optimizeTrainFromBuffer() :
+    The same requirements as above hold for all the parameters except `parameters`.
+    This function tries many parameter combinations and picks the best parameters.
+    `*parameters` is filled with the best parameters found, and the dictionary
+    constructed with those parameters is stored in `dictBuffer`.
+
+    All of the parameters d, k, steps are optional.
+    If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
+    if steps is zero it defaults to its default value.
+    If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
+
+    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
+              or an error code, which can be tested with ZDICT_isError().
+              On success `*parameters` contains the parameters selected.
+    Note : COVER_optimizeTrainFromBuffer() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
+*/
+ZDICTLIB_API size_t COVER_optimizeTrainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
+                                     const void* samplesBuffer, const size_t *samplesSizes, unsigned nbSamples,
+                                     COVER_params_t *parameters);
+
+/*! ZDICT_finalizeDictionary() :
+
+    Given a custom content as a basis for dictionary, and a set of samples,
+    finalize dictionary by adding headers and statistics.
+
     Samples must be stored concatenated in a flat buffer `samplesBuffer`,
     supplied with an array of sizes `samplesSizes`, providing the size of each sample in order.
 
-    The input dictionary content must be stored *at the end* of `dictBuffer`.
-    Its size is `dictContentSize`.
-    The resulting dictionary with added entropy tables will be *written back to `dictBuffer`*,
-    starting from its beginning.
-    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`).
+    dictContentSize must be >= ZDICT_CONTENTSIZE_MIN bytes.
+    maxDictSize must be >= dictContentSize, and must be >= ZDICT_DICTSIZE_MIN bytes.
+
+    @return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`),
+              or an error code, which can be tested by ZDICT_isError().
+    note : ZDICT_finalizeDictionary() will push notifications into stderr if instructed to, using notificationLevel>0.
+    note 2 : dictBuffer and dictContent can overlap
 */
-size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                        const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+#define ZDICT_CONTENTSIZE_MIN 128
+#define ZDICT_DICTSIZE_MIN    256
+ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
+                                const void* dictContent, size_t dictContentSize,
+                                const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
+                                ZDICT_params_t parameters);
 
 
 
+/* Deprecation warnings */
+/* It is generally possible to disable deprecation warnings from compiler,
+   for example with -Wno-deprecated-declarations for gcc
+   or _CRT_SECURE_NO_WARNINGS in Visual.
+   Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
+#ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
+#  define ZDICT_DEPRECATED(message) ZDICTLIB_API   /* disable deprecation warnings */
+#else
+#  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
+#    define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
+#  elif (ZDICT_GCC_VERSION >= 405) || defined(__clang__)
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
+#  elif (ZDICT_GCC_VERSION >= 301)
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated))
+#  elif defined(_MSC_VER)
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __declspec(deprecated(message))
+#  else
+#    pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
+#    define ZDICT_DEPRECATED(message) ZDICTLIB_API
+#  endif
+#endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
+
+ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
+size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
+                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
+
+
 #endif   /* ZDICT_STATIC_LINKING_ONLY */
 
 #if defined (__cplusplus)
diff --git a/contrib/zstd/zstd.h b/contrib/zstd/zstd.h
index bd5d68b27..eef2b0b5e 100644
--- a/contrib/zstd/zstd.h
+++ b/contrib/zstd/zstd.h
@@ -1,4 +1,4 @@
-/**
+/*
  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
@@ -7,32 +7,55 @@
  * of patent rights can be found in the PATENTS file in the same directory.
  */
 
-#ifndef ZSTD_H_235446
-#define ZSTD_H_235446
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-/*======   Dependency   ======*/
+#ifndef ZSTD_H_235446
+#define ZSTD_H_235446
+#define ZSTD_STATIC_LINKING_ONLY
+/* ======   Dependency   ======*/
 #include <stddef.h>   /* size_t */
 
 
-/*======  Export for Windows  ======*/
-/*!
-*  ZSTD_DLL_EXPORT :
-*  Enable exporting of functions when building a Windows DLL
-*/
-#if defined(_WIN32) && defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZSTDLIB_API __declspec(dllexport)
+/* =====   ZSTDLIB_API : control library symbols visibility   ===== */
+#if defined(__GNUC__) && (__GNUC__ >= 4)
+#  define ZSTDLIB_VISIBILITY __attribute__ ((visibility ("default")))
 #else
-#  define ZSTDLIB_API
+#  define ZSTDLIB_VISIBILITY
 #endif
+#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
+#  define ZSTDLIB_API __declspec(dllexport) ZSTDLIB_VISIBILITY
+#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
+#  define ZSTDLIB_API __declspec(dllimport) ZSTDLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
+#else
+#  define ZSTDLIB_API ZSTDLIB_VISIBILITY
+#endif
+
+
+/*******************************************************************************************************
+  Introduction
 
+  zstd, short for Zstandard, is a fast lossless compression algorithm, targeting real-time compression scenarios
+  at zlib-level and better compression ratios. The zstd compression library provides in-memory compression and
+  decompression functions. The library supports compression levels from 1 up to ZSTD_maxCLevel() which is 22.
+  Levels >= 20, labeled `--ultra`, should be used with caution, as they require more memory.
+  Compression can be done in:
+    - a single step (described as Simple API)
+    - a single step, reusing a context (described as Explicit memory management)
+    - unbounded multiple steps (described as Streaming compression)
+  The compression ratio achievable on small data can be highly improved using compression with a dictionary in:
+    - a single step (described as Simple dictionary API)
+    - a single step, reusing a dictionary (described as Fast dictionary API)
 
-/*=======   Version   =======*/
+  Advanced experimental functions can be accessed using #define ZSTD_STATIC_LINKING_ONLY before including zstd.h.
+  These APIs shall never be used with a dynamic library.
+  They are not "stable", their definition may change in the future. Only static linking is allowed.
+*********************************************************************************************************/
+
+/*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
-#define ZSTD_VERSION_MINOR    0
+#define ZSTD_VERSION_MINOR    3
 #define ZSTD_VERSION_RELEASE  0
 
 #define ZSTD_LIB_VERSION ZSTD_VERSION_MAJOR.ZSTD_VERSION_MINOR.ZSTD_VERSION_RELEASE
@@ -41,47 +64,57 @@ extern "C" {
 #define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
 
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
-ZSTDLIB_API unsigned ZSTD_versionNumber (void);
+ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< library version number; to be used when checking dll version */
 
 
-/* *************************************
+/***************************************
 *  Simple API
 ***************************************/
 /*! ZSTD_compress() :
-    Compresses `src` buffer into already allocated `dst`.
-    Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
-    @return : the number of bytes written into `dst` (<= `dstCapacity),
-              or an error code if it fails (which can be tested using ZSTD_isError()) */
+ *  Compresses `src` content as a single zstd compressed frame into already allocated `dst`.
+ *  Hint : compression runs faster if `dstCapacity` >=  `ZSTD_compressBound(srcSize)`.
+ *  @return : compressed size written into `dst` (<= `dstCapacity),
+ *            or an error code if it fails (which can be tested using ZSTD_isError()). */
 ZSTDLIB_API size_t ZSTD_compress( void* dst, size_t dstCapacity,
                             const void* src, size_t srcSize,
                                   int compressionLevel);
 
-/*! ZSTD_getDecompressedSize() :
-*   @return : decompressed size as a 64-bits value _if known_, 0 otherwise.
-*    note 1 : decompressed size can be very large (64-bits value),
-*             potentially larger than what local system can handle as a single memory segment.
-*             In which case, it's necessary to use streaming mode to decompress data.
-*    note 2 : decompressed size is an optional field, that may not be present.
-*             When `return==0`, data to decompress can have any size.
-*             In which case, it's necessary to use streaming mode to decompress data.
-*             Optionally, application may rely on its own implied limits.
-*             (For example, application data could be necessarily cut into blocks <= 16 KB).
-*    note 3 : decompressed size could be wrong or intentionally modified !
-*             Always ensure result fits within application's authorized limits !
-*             Each application can set its own limits.
-*    note 4 : when `return==0`, if precise failure cause is needed, use ZSTD_getFrameParams() to know more. */
-ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
-
 /*! ZSTD_decompress() :
-    `compressedSize` : must be the _exact_ size of compressed input, otherwise decompression will fail.
-    `dstCapacity` must be equal or larger than originalSize (see ZSTD_getDecompressedSize() ).
-    If originalSize is unknown, and if there is no implied application-specific limitations,
-    it's preferable to use streaming mode to decompress data.
-    @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
-              or an errorCode if it fails (which can be tested using ZSTD_isError()) */
+ *  `compressedSize` : must be the _exact_ size of some number of compressed and/or skippable frames.
+ *  `dstCapacity` is an upper bound of originalSize.
+ *  If user cannot imply a maximum upper bound, it's better to use streaming mode to decompress data.
+ *  @return : the number of bytes decompressed into `dst` (<= `dstCapacity`),
+ *            or an errorCode if it fails (which can be tested using ZSTD_isError()). */
 ZSTDLIB_API size_t ZSTD_decompress( void* dst, size_t dstCapacity,
                               const void* src, size_t compressedSize);
 
+/*! ZSTD_getDecompressedSize() :
+ *  NOTE: This function is planned to be obsolete, in favour of ZSTD_getFrameContentSize.
+ *  ZSTD_getFrameContentSize functions the same way, returning the decompressed size of a single
+ *  frame, but distinguishes empty frames from frames with an unknown size, or errors.
+ *
+ *  Additionally, ZSTD_findDecompressedSize can be used instead.  It can handle multiple
+ *  concatenated frames in one buffer, and so is more general.
+ *  As a result however, it requires more computation and entire frames to be passed to it,
+ *  as opposed to ZSTD_getFrameContentSize which requires only a single frame's header.
+ *
+ *  'src' is the start of a zstd compressed frame.
+ *  @return : content size to be decompressed, as a 64-bits value _if known_, 0 otherwise.
+ *   note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
+ *            When `return==0`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *            Optionally, application can still use ZSTD_decompress() while relying on implied limits.
+ *            (For example, data may be necessarily cut into blocks <= 16 KB).
+ *   note 2 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 3 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure result fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 5 : when `return==0`, if precise failure cause is needed, use ZSTD_getFrameParams() to know more. */
+ZSTDLIB_API unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize);
+
 
 /*======  Helper functions  ======*/
 ZSTDLIB_API int         ZSTD_maxCLevel(void);               /*!< maximum compression level available */
@@ -90,34 +123,43 @@ ZSTDLIB_API unsigned    ZSTD_isError(size_t code);          /*!< tells if a `siz
 ZSTDLIB_API const char* ZSTD_getErrorName(size_t code);     /*!< provides readable string from an error code */
 
 
-/*-*************************************
+/***************************************
 *  Explicit memory management
 ***************************************/
-/** Compression context */
+/*= Compression context
+ *  When compressing many times,
+ *  it is recommended to allocate a context just once, and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution in multi-threaded environments. */
 typedef struct ZSTD_CCtx_s ZSTD_CCtx;
 ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx(void);
 ZSTDLIB_API size_t     ZSTD_freeCCtx(ZSTD_CCtx* cctx);
 
-/** ZSTD_compressCCtx() :
-    Same as ZSTD_compress(), requires an allocated ZSTD_CCtx (see ZSTD_createCCtx()) */
+/*! ZSTD_compressCCtx() :
+ *  Same as ZSTD_compress(), requires an allocated ZSTD_CCtx (see ZSTD_createCCtx()). */
 ZSTDLIB_API size_t ZSTD_compressCCtx(ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel);
 
-/** Decompression context */
+/*= Decompression context
+ *  When decompressing many times,
+ *  it is recommended to allocate a context just once, and re-use it for each successive compression operation.
+ *  This will make workload friendlier for system's memory.
+ *  Use one context per thread for parallel execution in multi-threaded environments. */
 typedef struct ZSTD_DCtx_s ZSTD_DCtx;
 ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx(void);
 ZSTDLIB_API size_t     ZSTD_freeDCtx(ZSTD_DCtx* dctx);
 
-/** ZSTD_decompressDCtx() :
-*   Same as ZSTD_decompress(), requires an allocated ZSTD_DCtx (see ZSTD_createDCtx()) */
+/*! ZSTD_decompressDCtx() :
+ *  Same as ZSTD_decompress(), requires an allocated ZSTD_DCtx (see ZSTD_createDCtx()). */
 ZSTDLIB_API size_t ZSTD_decompressDCtx(ZSTD_DCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
 
-/*-************************
+/**************************
 *  Simple dictionary API
 ***************************/
 /*! ZSTD_compress_usingDict() :
 *   Compression using a predefined Dictionary (see dictBuilder/zdict.h).
-*   Note : This function load the dictionary, resulting in significant startup delay. */
+*   Note : This function loads the dictionary, resulting in significant startup delay.
+*   Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
 ZSTDLIB_API size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx,
                                            void* dst, size_t dstCapacity,
                                      const void* src, size_t srcSize,
@@ -127,41 +169,54 @@ ZSTDLIB_API size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx,
 /*! ZSTD_decompress_usingDict() :
 *   Decompression using a predefined Dictionary (see dictBuilder/zdict.h).
 *   Dictionary must be identical to the one used during compression.
-*   Note : This function load the dictionary, resulting in significant startup delay */
+*   Note : This function loads the dictionary, resulting in significant startup delay.
+*   Note : When `dict == NULL || dictSize < 8` no dictionary is used. */
 ZSTDLIB_API size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
                                              void* dst, size_t dstCapacity,
                                        const void* src, size_t srcSize,
                                        const void* dict,size_t dictSize);
 
 
-/*-**************************
-*  Fast Dictionary API
+/****************************
+*  Fast dictionary API
 ****************************/
-/*! ZSTD_createCDict() :
-*   Create a digested dictionary, ready to start compression operation without startup delay.
-*   `dict` can be released after ZSTD_CDict creation */
 typedef struct ZSTD_CDict_s ZSTD_CDict;
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel);
+
+/*! ZSTD_createCDict() :
+*   When compressing multiple messages / blocks with the same dictionary, it's recommended to load it just once.
+*   ZSTD_createCDict() will create a digested dictionary, ready to start future compression operations without startup delay.
+*   ZSTD_CDict can be created once and used by multiple threads concurrently, as its usage is read-only.
+*   `dictBuffer` can be released after ZSTD_CDict creation, as its content is copied within CDict */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict(const void* dictBuffer, size_t dictSize, int compressionLevel);
+
+/*! ZSTD_freeCDict() :
+*   Function frees memory allocated by ZSTD_createCDict(). */
 ZSTDLIB_API size_t      ZSTD_freeCDict(ZSTD_CDict* CDict);
 
 /*! 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 level is decided during dictionary creation */
+ *  Compression using a digested Dictionary.
+ *  Faster startup than ZSTD_compress_usingDict(), recommended when same dictionary is used multiple times.
+ *  Note that compression level is decided during dictionary creation.
+ *  Frame parameters are hardcoded (dictID=yes, contentSize=yes, checksum=no) */
 ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
                                             void* dst, size_t dstCapacity,
                                       const void* src, size_t srcSize,
                                       const ZSTD_CDict* cdict);
 
+
+typedef struct ZSTD_DDict_s ZSTD_DDict;
+
 /*! ZSTD_createDDict() :
 *   Create a digested dictionary, ready to start decompression operation without startup delay.
-*   `dict` can be released after creation */
-typedef struct ZSTD_DDict_s ZSTD_DDict;
-ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize);
+*   dictBuffer can be released after DDict creation, as its content is copied inside DDict */
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict(const void* dictBuffer, size_t dictSize);
+
+/*! ZSTD_freeDDict() :
+*   Function frees memory allocated with ZSTD_createDDict() */
 ZSTDLIB_API size_t      ZSTD_freeDDict(ZSTD_DDict* ddict);
 
 /*! ZSTD_decompress_usingDDict() :
-*   Decompression using a digested Dictionary
+*   Decompression using a digested Dictionary.
 *   Faster startup than ZSTD_decompress_usingDict(), recommended when same dictionary is used multiple times. */
 ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
                                               void* dst, size_t dstCapacity,
@@ -169,7 +224,7 @@ ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
                                         const ZSTD_DDict* ddict);
 
 
-/*-**************************
+/****************************
 *  Streaming
 ****************************/
 
@@ -186,31 +241,33 @@ typedef struct ZSTD_outBuffer_s {
 } ZSTD_outBuffer;
 
 
-/*======   streaming compression   ======*/
 
 /*-***********************************************************************
-*  Streaming compression - howto
+*  Streaming compression - HowTo
 *
 *  A ZSTD_CStream object is required to track streaming operation.
 *  Use ZSTD_createCStream() and ZSTD_freeCStream() to create/release resources.
 *  ZSTD_CStream objects can be reused multiple times on consecutive compression operations.
+*  It is recommended to re-use ZSTD_CStream in situations where many streaming operations will be achieved consecutively,
+*  since it will play nicer with system's memory, by re-using already allocated memory.
+*  Use one separate ZSTD_CStream per thread for parallel execution.
 *
-*  Start by initializing ZSTD_CStream.
+*  Start a new compression by initializing ZSTD_CStream.
 *  Use ZSTD_initCStream() to start a new compression operation.
-*  Use ZSTD_initCStream_usingDict() for a compression which requires a dictionary.
+*  Use ZSTD_initCStream_usingDict() or ZSTD_initCStream_usingCDict() for a compression which requires a dictionary (experimental section)
 *
 *  Use ZSTD_compressStream() repetitively to consume input stream.
-*  The function will automatically update both `pos`.
+*  The function will automatically update both `pos` fields.
 *  Note that it may not consume the entire input, in which case `pos < size`,
 *  and it's up to the caller to present again remaining data.
 *  @return : a size hint, preferred nb of bytes to use as input for next function call
-*           (it's just a hint, to help latency a little, any other value will work fine)
-*           (note : the size hint is guaranteed to be <= ZSTD_CStreamInSize() )
 *            or an error code, which can be tested using ZSTD_isError().
+*            Note 1 : it's just a hint, to help latency a little, any other value will work fine.
+*            Note 2 : size hint is guaranteed to be <= ZSTD_CStreamInSize()
 *
-*  At any moment, it's possible to flush whatever data remains within buffer, using ZSTD_flushStream().
+*  At any moment, it's possible to flush whatever data remains within internal buffer, using ZSTD_flushStream().
 *  `output->pos` will be updated.
-*  Note some content might still be left within internal buffer if `output->size` is too small.
+*  Note that some content might still be left within internal buffer if `output->size` is too small.
 *  @return : nb of bytes still present within internal buffer (0 if it's empty)
 *            or an error code, which can be tested using ZSTD_isError().
 *
@@ -219,29 +276,30 @@ typedef struct ZSTD_outBuffer_s {
 *  The epilogue is required for decoders to consider a frame completed.
 *  Similar to ZSTD_flushStream(), it may not be able to flush the full content if `output->size` is too small.
 *  In which case, call again ZSTD_endStream() to complete the flush.
-*  @return : nb of bytes still present within internal buffer (0 if it's empty)
+*  @return : nb of bytes still present within internal buffer (0 if it's empty, hence compression completed)
 *            or an error code, which can be tested using ZSTD_isError().
 *
 * *******************************************************************/
 
-typedef struct ZSTD_CStream_s ZSTD_CStream;
+typedef ZSTD_CCtx ZSTD_CStream;  /**< CCtx and CStream are effectively same object */
+                                 /* Continue due distinghish them for compatibility with versions <= v1.2.0 */
+/*===== ZSTD_CStream management functions =====*/
 ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream(void);
 ZSTDLIB_API size_t ZSTD_freeCStream(ZSTD_CStream* zcs);
 
-ZSTDLIB_API size_t ZSTD_CStreamInSize(void);    /**< recommended size for input buffer */
-ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output buffer */
-
+/*===== Streaming compression functions =====*/
 ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel);
 ZSTDLIB_API size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
 ZSTDLIB_API size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
 ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
 
+ZSTDLIB_API size_t ZSTD_CStreamInSize(void);    /**< recommended size for input buffer */
+ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output buffer. Guarantee to successfully flush at least one complete compressed block in all circumstances. */
+
 
-/*======   decompression   ======*/
 
 /*-***************************************************************************
-*  Streaming decompression howto
+*  Streaming decompression - HowTo
 *
 *  A ZSTD_DStream object is required to track streaming operations.
 *  Use ZSTD_createDStream() and ZSTD_freeDStream() to create/release resources.
@@ -249,68 +307,75 @@ ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
 *
 *  Use ZSTD_initDStream() to start a new decompression operation,
 *   or ZSTD_initDStream_usingDict() if decompression requires a dictionary.
+*   @return : recommended first input size
 *
 *  Use ZSTD_decompressStream() repetitively to consume your input.
-*  The function will update both `pos`.
-*  Note that it may not consume the entire input (pos < size),
-*  in which case it's up to the caller to present remaining input again.
+*  The function will update both `pos` fields.
+*  If `input.pos < input.size`, some input has not been consumed.
+*  It's up to the caller to present again remaining data.
+*  If `output.pos < output.size`, decoder has flushed everything it could.
 *  @return : 0 when a frame is completely decoded and fully flushed,
-*            1 when there is still some data left within internal buffer to flush,
-*            >1 when more data is expected, with value being a suggested next input size (it's just a hint, which helps latency, any size is accepted),
-*            or an error code, which can be tested using ZSTD_isError().
-*
+*            an error code, which can be tested using ZSTD_isError(),
+*            any other value > 0, which means there is still some decoding to do to complete current frame.
+*            The return value is a suggested next input size (a hint to improve latency) that will never load more than the current frame.
 * *******************************************************************************/
 
 typedef struct ZSTD_DStream_s ZSTD_DStream;
+/*===== ZSTD_DStream management functions =====*/
 ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream(void);
 ZSTDLIB_API size_t ZSTD_freeDStream(ZSTD_DStream* zds);
 
-ZSTDLIB_API size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
-ZSTDLIB_API size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output buffer */
-
+/*===== Streaming decompression functions =====*/
 ZSTDLIB_API size_t ZSTD_initDStream(ZSTD_DStream* zds);
 ZSTDLIB_API size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
-ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize);
 
+ZSTDLIB_API size_t ZSTD_DStreamInSize(void);    /*!< recommended size for input buffer */
+ZSTDLIB_API size_t ZSTD_DStreamOutSize(void);   /*!< recommended size for output buffer. Guarantee to successfully flush at least one complete block in all circumstances. */
+
+#endif  /* ZSTD_H_235446 */
 
 
-#ifdef ZSTD_STATIC_LINKING_ONLY
+#if defined(ZSTD_STATIC_LINKING_ONLY) && !defined(ZSTD_H_ZSTD_STATIC_LINKING_ONLY)
+#define ZSTD_H_ZSTD_STATIC_LINKING_ONLY
 
-/* ====================================================================================
+/****************************************************************************************
+ * START OF ADVANCED AND EXPERIMENTAL FUNCTIONS
  * The definitions in this section are considered experimental.
  * They should never be used with a dynamic library, as they may change in the future.
  * They are provided for advanced usages.
  * Use them only in association with static linking.
- * ==================================================================================== */
+ * ***************************************************************************************/
 
-/*--- Constants ---*/
-#define ZSTD_MAGICNUMBER            0xFD2FB528   /* v0.8 */
+/* --- Constants ---*/
+#define ZSTD_MAGICNUMBER            0xFD2FB528   /* >= v0.8.0 */
 #define ZSTD_MAGIC_SKIPPABLE_START  0x184D2A50U
 
-#define ZSTD_WINDOWLOG_MAX_32  25
+#define ZSTD_WINDOWLOG_MAX_32  27
 #define ZSTD_WINDOWLOG_MAX_64  27
-#define ZSTD_WINDOWLOG_MAX    ((U32)(MEM_32bits() ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
-#define ZSTD_WINDOWLOG_MIN     18
-#define ZSTD_CHAINLOG_MAX     (ZSTD_WINDOWLOG_MAX+1)
-#define ZSTD_CHAINLOG_MIN       4
+#define ZSTD_WINDOWLOG_MAX    ((unsigned)(sizeof(size_t) == 4 ? ZSTD_WINDOWLOG_MAX_32 : ZSTD_WINDOWLOG_MAX_64))
+#define ZSTD_WINDOWLOG_MIN     10
 #define ZSTD_HASHLOG_MAX       ZSTD_WINDOWLOG_MAX
-#define ZSTD_HASHLOG_MIN       12
+#define ZSTD_HASHLOG_MIN        6
+#define ZSTD_CHAINLOG_MAX     (ZSTD_WINDOWLOG_MAX+1)
+#define ZSTD_CHAINLOG_MIN      ZSTD_HASHLOG_MIN
 #define ZSTD_HASHLOG3_MAX      17
 #define ZSTD_SEARCHLOG_MAX    (ZSTD_WINDOWLOG_MAX-1)
 #define ZSTD_SEARCHLOG_MIN      1
-#define ZSTD_SEARCHLENGTH_MAX   7
-#define ZSTD_SEARCHLENGTH_MIN   3
+#define ZSTD_SEARCHLENGTH_MAX   7   /* only for ZSTD_fast, other strategies are limited to 6 */
+#define ZSTD_SEARCHLENGTH_MIN   3   /* only for ZSTD_btopt, other strategies are limited to 4 */
 #define ZSTD_TARGETLENGTH_MIN   4
 #define ZSTD_TARGETLENGTH_MAX 999
 
 #define ZSTD_FRAMEHEADERSIZE_MAX 18    /* for static allocation */
-static const size_t ZSTD_frameHeaderSize_min = 5;
+#define ZSTD_FRAMEHEADERSIZE_MIN  6
+static const size_t ZSTD_frameHeaderSize_prefix = 5;
+static const size_t ZSTD_frameHeaderSize_min = ZSTD_FRAMEHEADERSIZE_MIN;
 static const size_t ZSTD_frameHeaderSize_max = ZSTD_FRAMEHEADERSIZE_MAX;
 static const size_t ZSTD_skippableHeaderSize = 8;  /* magic number + skippable frame length */
 
 
-/*--- Types ---*/
-typedef enum { ZSTD_fast, ZSTD_dfast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2, ZSTD_btopt } ZSTD_strategy;   /* from faster to stronger */
+/*--- Advanced types ---*/
+typedef enum { ZSTD_fast, ZSTD_dfast, ZSTD_greedy, ZSTD_lazy, ZSTD_lazy2, ZSTD_btlazy2, ZSTD_btopt, ZSTD_btultra } ZSTD_strategy;   /* from faster to stronger */
 
 typedef struct {
     unsigned windowLog;      /**< largest match distance : larger == more compression, more memory needed during decompression */
@@ -323,9 +388,9 @@ typedef struct {
 } ZSTD_compressionParameters;
 
 typedef struct {
-    unsigned contentSizeFlag; /**< 1: content size will be in frame header (if known). */
-    unsigned checksumFlag;    /**< 1: will generate a 22-bits checksum at end of frame, to be used for error detection by decompressor */
-    unsigned noDictIDFlag;    /**< 1: no dict ID will be saved into frame header (if dictionary compression) */
+    unsigned contentSizeFlag; /**< 1: content size will be in frame header (when known) */
+    unsigned checksumFlag;    /**< 1: generate a 32-bits checksum at end of frame, for error detection */
+    unsigned noDictIDFlag;    /**< 1: no dictID will be saved into frame header (if dictionary compression) */
 } ZSTD_frameParameters;
 
 typedef struct {
@@ -333,42 +398,138 @@ typedef struct {
     ZSTD_frameParameters fParams;
 } ZSTD_parameters;
 
-/* custom memory allocation functions */
+typedef struct {
+    unsigned long long frameContentSize;
+    unsigned windowSize;
+    unsigned dictID;
+    unsigned checksumFlag;
+} ZSTD_frameHeader;
+
+/*= Custom memory allocation functions */
 typedef void* (*ZSTD_allocFunction) (void* opaque, size_t size);
 typedef void  (*ZSTD_freeFunction) (void* opaque, void* address);
 typedef struct { ZSTD_allocFunction customAlloc; ZSTD_freeFunction customFree; void* opaque; } ZSTD_customMem;
 
+/***************************************
+*  Frame size functions
+***************************************/
 
-/*-*************************************
-*  Advanced compression functions
+/*! ZSTD_findFrameCompressedSize() :
+ *  `src` should point to the start of a ZSTD encoded frame or skippable frame
+ *  `srcSize` must be at least as large as the frame
+ *  @return : the compressed size of the frame pointed to by `src`,
+ *            suitable to pass to `ZSTD_decompress` or similar,
+ *            or an error code if given invalid input. */
+ZSTDLIB_API size_t ZSTD_findFrameCompressedSize(const void* src, size_t srcSize);
+
+/*! ZSTD_getFrameContentSize() :
+ *  `src` should point to the start of a ZSTD encoded frame.
+ *  `srcSize` must be at least as large as the frame header.
+ *       A value >= `ZSTD_frameHeaderSize_max` is guaranteed to be large enough.
+ *  @return : - decompressed size of the frame pointed to be `src` if known
+ *            - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+ *            - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small) */
+#define ZSTD_CONTENTSIZE_UNKNOWN (0ULL - 1)
+#define ZSTD_CONTENTSIZE_ERROR   (0ULL - 2)
+ZSTDLIB_API unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize);
+
+/*! ZSTD_findDecompressedSize() :
+ *  `src` should point the start of a series of ZSTD encoded and/or skippable frames
+ *  `srcSize` must be the _exact_ size of this series
+ *       (i.e. there should be a frame boundary exactly `srcSize` bytes after `src`)
+ *  @return : - decompressed size of all data in all successive frames
+ *            - if the decompressed size cannot be determined: ZSTD_CONTENTSIZE_UNKNOWN
+ *            - if an error occurred: ZSTD_CONTENTSIZE_ERROR
+ *
+ *   note 1 : decompressed size is an optional field, that may not be present, especially in streaming mode.
+ *            When `return==ZSTD_CONTENTSIZE_UNKNOWN`, data to decompress could be any size.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *            Optionally, application can still use ZSTD_decompress() while relying on implied limits.
+ *            (For example, data may be necessarily cut into blocks <= 16 KB).
+ *   note 2 : decompressed size is always present when compression is done with ZSTD_compress()
+ *   note 3 : decompressed size can be very large (64-bits value),
+ *            potentially larger than what local system can handle as a single memory segment.
+ *            In which case, it's necessary to use streaming mode to decompress data.
+ *   note 4 : If source is untrusted, decompressed size could be wrong or intentionally modified.
+ *            Always ensure result fits within application's authorized limits.
+ *            Each application can set its own limits.
+ *   note 5 : ZSTD_findDecompressedSize handles multiple frames, and so it must traverse the input to
+ *            read each contained frame header.  This is efficient as most of the data is skipped,
+ *            however it does mean that all frame data must be present and valid. */
+ZSTDLIB_API unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize);
+
+
+/***************************************
+*  Context memory usage
 ***************************************/
-/*! ZSTD_estimateCCtxSize() :
- *  Gives the amount of memory allocated for a ZSTD_CCtx given a set of compression parameters.
- *  `frameContentSize` is an optional parameter, provide `0` if unknown */
+
+/*! ZSTD_sizeof_*() :
+ *  These functions give the current memory usage of selected object.
+ *  Object memory usage can evolve if it's re-used multiple times. */
+ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
+ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
+ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
+ZSTDLIB_API size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict);
+ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
+
+/*! ZSTD_estimate*() :
+ *  These functions make it possible to estimate memory usage
+ *  of a future target object, before its allocation,
+ *  given a set of parameters, which vary depending on target object.
+ *  The objective is to guide decision before allocation. */
 ZSTDLIB_API size_t ZSTD_estimateCCtxSize(ZSTD_compressionParameters cParams);
+ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
+
+/*! ZSTD_estimate?StreamSize() :
+ *  Note : if streaming is init with function ZSTD_init?Stream_usingDict(),
+ *         an internal ?Dict will be created, which size is not estimated.
+ *         In this case, get additional size by using ZSTD_estimate?DictSize */
+ZSTDLIB_API size_t ZSTD_estimateCStreamSize(ZSTD_compressionParameters cParams);
+ZSTDLIB_API size_t ZSTD_estimateDStreamSize(ZSTD_frameHeader fHeader);
 
+/*! ZSTD_estimate?DictSize() :
+ *  Note : if dictionary is created "byReference", reduce estimation by dictSize */
+ZSTDLIB_API size_t ZSTD_estimateCDictSize(ZSTD_compressionParameters cParams, size_t dictSize);
+ZSTDLIB_API size_t ZSTD_estimateDDictSize(size_t dictSize);
+
+
+/***************************************
+*  Advanced compression functions
+***************************************/
 /*! ZSTD_createCCtx_advanced() :
  *  Create a ZSTD compression context using external alloc and free functions */
 ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem);
 
+typedef enum {
+    ZSTD_p_forceWindow,   /* Force back-references to remain < windowSize, even when referencing Dictionary content (default:0) */
+    ZSTD_p_forceRawDict   /* Force loading dictionary in "content-only" mode (no header analysis) */
+} ZSTD_CCtxParameter;
+/*! ZSTD_setCCtxParameter() :
+ *  Set advanced parameters, selected through enum ZSTD_CCtxParameter
+ *  @result : 0, or an error code (which can be tested with ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTD_setCCtxParameter(ZSTD_CCtx* cctx, ZSTD_CCtxParameter param, unsigned value);
+
+/*! ZSTD_createCDict_byReference() :
+ *  Create a digested dictionary for compression
+ *  Dictionary content is simply referenced, and therefore stays in dictBuffer.
+ *  It is important that dictBuffer outlives CDict, it must remain read accessible throughout the lifetime of CDict */
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_byReference(const void* dictBuffer, size_t dictSize, int compressionLevel);
+
 /*! ZSTD_createCDict_advanced() :
  *  Create a ZSTD_CDict using external alloc and free, and customized compression parameters */
-ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize,
-                                                  ZSTD_parameters params, ZSTD_customMem customMem);
+ZSTDLIB_API ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize, unsigned byReference,
+                                                  ZSTD_compressionParameters cParams, ZSTD_customMem customMem);
 
-/*! ZSTD_sizeofCCtx() :
- *  Gives the amount of memory used by a given ZSTD_CCtx */
-ZSTDLIB_API size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx);
+/*! ZSTD_getCParams() :
+*   @return ZSTD_compressionParameters structure for a selected compression level and estimated srcSize.
+*   `estimatedSrcSize` value is optional, select 0 if not known */
+ZSTDLIB_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
 /*! ZSTD_getParams() :
-*   same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of a `ZSTD_compressionParameters`.
+*   same as ZSTD_getCParams(), but @return a full `ZSTD_parameters` object instead of sub-component `ZSTD_compressionParameters`.
 *   All fields of `ZSTD_frameParameters` are set to default (0) */
-ZSTDLIB_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSize, size_t dictSize);
-
-/*! ZSTD_getCParams() :
-*   @return ZSTD_compressionParameters structure for a selected compression level and srcSize.
-*   `srcSize` value is optional, select 0 if not known */
-ZSTDLIB_API ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSize, size_t dictSize);
+ZSTDLIB_API ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long estimatedSrcSize, size_t dictSize);
 
 /*! ZSTD_checkCParams() :
 *   Ensure param values remain within authorized range */
@@ -380,67 +541,113 @@ ZSTDLIB_API size_t ZSTD_checkCParams(ZSTD_compressionParameters params);
 ZSTDLIB_API ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize);
 
 /*! ZSTD_compress_advanced() :
-*   Same as ZSTD_compress_usingDict(), with fine-tune control of each compression parameter */
-ZSTDLIB_API 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);
+*   Same as ZSTD_compress_usingDict(), with fine-tune control over each compression parameter */
+ZSTDLIB_API size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const void* dict,size_t dictSize,
+                                  ZSTD_parameters params);
 
+/*! ZSTD_compress_usingCDict_advanced() :
+*   Same as ZSTD_compress_usingCDict(), with fine-tune control over frame parameters */
+ZSTDLIB_API 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);
 
-/*--- Advanced Decompression functions ---*/
 
-/*! ZSTD_estimateDCtxSize() :
- *  Gives the potential amount of memory allocated to create a ZSTD_DCtx */
-ZSTDLIB_API size_t ZSTD_estimateDCtxSize(void);
+/*--- Advanced decompression functions ---*/
+
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+ZSTDLIB_API unsigned ZSTD_isFrame(const void* buffer, size_t size);
 
 /*! ZSTD_createDCtx_advanced() :
  *  Create a ZSTD decompression context using external alloc and free functions */
 ZSTDLIB_API ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem);
 
-/*! ZSTD_sizeofDCtx() :
- *  Gives the amount of memory used by a given ZSTD_DCtx */
-ZSTDLIB_API size_t ZSTD_sizeof_DCtx(const ZSTD_DCtx* dctx);
-
-
-/* ******************************************************************
-*  Advanced Streaming functions
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, ready to start decompression operation without startup delay.
+ *  Dictionary content is simply referenced, and therefore stays in dictBuffer.
+ *  It is important that dictBuffer outlives DDict, it must remain read accessible throughout the lifetime of DDict */
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize);
+
+/*! ZSTD_createDDict_advanced() :
+ *  Create a ZSTD_DDict using external alloc and free, optionally by reference */
+ZSTDLIB_API ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                                  unsigned byReference, ZSTD_customMem customMem);
+
+/*! ZSTD_getDictID_fromDict() :
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize);
+
+/*! ZSTD_getDictID_fromDDict() :
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict);
+
+/*! ZSTD_getDictID_fromFrame() :
+ *  Provides the dictID required to decompressed the frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary to be decoded (most common case).
+ *  - The frame was built with dictID intentionally removed. Whatever dictionary is necessary is a hidden information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, the frame header could not be decoded (only possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`).
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use ZSTD_getFrameParams(), which will provide a more precise error code. */
+ZSTDLIB_API unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize);
+
+
+/********************************************************************
+*  Advanced streaming functions
 ********************************************************************/
 
-/*======   compression   ======*/
-
+/*=====   Advanced Streaming compression functions  =====*/
 ZSTDLIB_API ZSTD_CStream* ZSTD_createCStream_advanced(ZSTD_customMem customMem);
+ZSTDLIB_API size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize);   /**< pledgedSrcSize must be correct, a size of 0 means unknown.  for a frame size of 0 use initCStream_advanced */
+ZSTDLIB_API size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel); /**< note: a dict will not be used if dict == NULL or dictSize < 8. This result in the creation of an internal CDict */
 ZSTDLIB_API size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs, const void* dict, size_t dictSize,
-                                 ZSTD_parameters params, unsigned long long pledgedSrcSize);
-ZSTDLIB_API size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs);
-
-
-/*======   decompression   ======*/
-
-typedef enum { ZSTDdsp_maxWindowSize } ZSTD_DStreamParameter_e;
-
+                                             ZSTD_parameters params, unsigned long long pledgedSrcSize);  /**< pledgedSrcSize is optional and can be 0 (meaning unknown). note: if the contentSizeFlag is set, pledgedSrcSize == 0 means the source size is actually 0 */
+ZSTDLIB_API size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict);  /**< note : cdict will just be referenced, and must outlive compression session */
+ZSTDLIB_API size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs, const ZSTD_CDict* cdict, unsigned long long pledgedSrcSize, ZSTD_frameParameters fParams);  /**< same as ZSTD_initCStream_usingCDict(), with control over frame parameters */
+
+/*! ZSTD_resetCStream() :
+ *  start a new compression job, using same parameters from previous job.
+ *  This is typically useful to skip dictionary loading stage, since it will re-use it in-place..
+ *  Note that zcs must be init at least once before using ZSTD_resetCStream().
+ *  pledgedSrcSize==0 means "srcSize unknown".
+ *  If pledgedSrcSize > 0, its value must be correct, as it will be written in header, and controlled at the end.
+ *  @return : 0, or an error code (which can be tested using ZSTD_isError()) */
+ZSTDLIB_API size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize);
+
+
+/*=====   Advanced Streaming decompression functions  =====*/
+typedef enum { DStream_p_maxWindowSize } ZSTD_DStreamParameter_e;
 ZSTDLIB_API ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem);
-
 ZSTDLIB_API size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds, ZSTD_DStreamParameter_e paramType, unsigned paramValue);
-ZSTDLIB_API size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds);
+ZSTDLIB_API size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize); /**< note: a dict will not be used if dict == NULL or dictSize < 8 */
+ZSTDLIB_API size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict);  /**< note : ddict will just be referenced, and must outlive decompression session */
+ZSTDLIB_API size_t ZSTD_resetDStream(ZSTD_DStream* zds);  /**< re-use decompression parameters from previous init; saves dictionary loading */
 
 
-/* ******************************************************************
+/*********************************************************************
 *  Buffer-less and synchronous inner streaming functions
-********************************************************************/
-/* This is an advanced API, giving full control over buffer management, for users which need direct control over memory.
+*
+*  This is an advanced API, giving full control over buffer management, for users which need direct control over memory.
 *  But it's also a complex one, with many restrictions (documented below).
-*  Prefer using normal streaming API for an easier experience */
+*  Prefer using normal streaming API for an easier experience
+********************************************************************* */
 
-ZSTDLIB_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
-ZSTDLIB_API size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize);
-ZSTDLIB_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx);
-
-ZSTDLIB_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
-ZSTDLIB_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+/**
+  Buffer-less streaming compression (synchronous mode)
 
-/*
   A ZSTD_CCtx object is required to track streaming operations.
   Use ZSTD_createCCtx() / ZSTD_freeCCtx() to manage resource.
   ZSTD_CCtx object can be re-used multiple times within successive compression operations.
@@ -463,32 +670,26 @@ ZSTDLIB_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapaci
     In which case, it will "discard" the relevant memory section from its history.
 
   Finish a frame with ZSTD_compressEnd(), which will write the last block(s) and optional checksum.
-  It's possible to use a NULL,0 src content, in which case, it will write a final empty block to end the frame,
-  Without last block mark, frames will be considered unfinished (broken) by decoders.
+  It's possible to use srcSize==0, in which case, it will write a final empty block to end the frame.
+  Without last block mark, frames will be considered unfinished (corrupted) by decoders.
 
-  You can then reuse `ZSTD_CCtx` (ZSTD_compressBegin()) to compress some new frame.
+  `ZSTD_CCtx` object can be re-used (ZSTD_compressBegin()) to compress some new frame.
 */
 
-typedef struct {
-    unsigned long long frameContentSize;
-    unsigned windowSize;
-    unsigned dictID;
-    unsigned checksumFlag;
-} ZSTD_frameParams;
+/*=====   Buffer-less streaming compression functions  =====*/
+ZSTDLIB_API size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel);
+ZSTDLIB_API size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, ZSTD_parameters params, unsigned long long pledgedSrcSize); /**< pledgedSrcSize is optional and can be 0 (meaning unknown). note: if the contentSizeFlag is set, pledgedSrcSize == 0 means the source size is actually 0 */
+ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict); /**< note: fails if cdict==NULL */
+ZSTDLIB_API size_t ZSTD_compressBegin_usingCDict_advanced(ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict, ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize);   /* compression parameters are already set within cdict. pledgedSrcSize=0 means null-size */
+ZSTDLIB_API size_t ZSTD_copyCCtx(ZSTD_CCtx* cctx, const ZSTD_CCtx* preparedCCtx, unsigned long long pledgedSrcSize); /**<  note: if pledgedSrcSize can be 0, indicating unknown size.  if it is non-zero, it must be accurate.  for 0 size frames, use compressBegin_advanced */
 
-ZSTDLIB_API size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t srcSize);   /**< doesn't consume input, see details below */
+ZSTDLIB_API size_t ZSTD_compressContinue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+ZSTDLIB_API size_t ZSTD_compressEnd(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
-ZSTDLIB_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
-ZSTDLIB_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
-ZSTDLIB_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
 
-ZSTDLIB_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
-ZSTDLIB_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 
-typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
-ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
-
-/*
+/*-
   Buffer-less streaming decompression (synchronous mode)
 
   A ZSTD_DCtx object is required to track streaming operations.
@@ -542,14 +743,26 @@ ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
   c) Frame Content - any content (User Data) of length equal to Frame Size
   For skippable frames ZSTD_decompressContinue() always returns 0.
   For skippable frames ZSTD_getFrameParams() returns fparamsPtr->windowLog==0 what means that a frame is skippable.
+    Note : If fparamsPtr->frameContentSize==0, it is ambiguous: the frame might actually be a Zstd encoded frame with no content.
+           For purposes of decompression, it is valid in both cases to skip the frame using
+           ZSTD_findFrameCompressedSize to find its size in bytes.
   It also returns Frame Size as fparamsPtr->frameContentSize.
 */
 
+/*=====   Buffer-less streaming decompression functions  =====*/
+ZSTDLIB_API size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize);   /**< doesn't consume input, see details below */
+ZSTDLIB_API size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize);
+ZSTDLIB_API void   ZSTD_copyDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* preparedDCtx);
+ZSTDLIB_API size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx);
+ZSTDLIB_API size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+typedef enum { ZSTDnit_frameHeader, ZSTDnit_blockHeader, ZSTDnit_block, ZSTDnit_lastBlock, ZSTDnit_checksum, ZSTDnit_skippableFrame } ZSTD_nextInputType_e;
+ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
 
-/* **************************************
-*  Block functions
-****************************************/
-/*! Block functions produce and decode raw zstd blocks, without frame metadata.
+/**
+    Block functions
+
+    Block functions produce and decode raw zstd blocks, without frame metadata.
     Frame metadata cost is typically ~18 bytes, which can be non-negligible for very small blocks (< 100 bytes).
     User will have to take in charge required information to regenerate data, such as compressed and content sizes.
 
@@ -557,32 +770,32 @@ ZSTDLIB_API ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx);
     - Compressing and decompressing require a context structure
       + Use ZSTD_createCCtx() and ZSTD_createDCtx()
     - It is necessary to init context before starting
-      + compression : ZSTD_compressBegin()
-      + decompression : ZSTD_decompressBegin()
-      + variants _usingDict() are also allowed
-      + copyCCtx() and copyDCtx() work too
-    - Block size is limited, it must be <= ZSTD_getBlockSizeMax()
-      + If you need to compress more, cut data into multiple blocks
-      + Consider using the regular ZSTD_compress() instead, as frame metadata costs become negligible when source size is large.
+      + compression : any ZSTD_compressBegin*() variant, including with dictionary
+      + decompression : any ZSTD_decompressBegin*() variant, including with dictionary
+      + copyCCtx() and copyDCtx() can be used too
+    - Block size is limited, it must be <= ZSTD_getBlockSizeMax() <= ZSTD_BLOCKSIZE_ABSOLUTEMAX
+      + If input is larger than a block size, it's necessary to split input data into multiple blocks
+      + For inputs larger than a single block size, consider using the regular ZSTD_compress() instead.
+        Frame metadata is not that costly, and quickly becomes negligible as source size grows larger.
     - When a block is considered not compressible enough, ZSTD_compressBlock() result will be zero.
       In which case, nothing is produced into `dst`.
       + User must test for such outcome and deal directly with uncompressed data
       + ZSTD_decompressBlock() doesn't accept uncompressed data as input !!!
-      + In case of multiple successive blocks, decoder must be informed of uncompressed block existence to follow proper history.
-        Use ZSTD_insertBlock() in such a case.
+      + In case of multiple successive blocks, should some of them be uncompressed,
+        decoder must be informed of their existence in order to follow proper history.
+        Use ZSTD_insertBlock() for such a case.
 */
 
 #define ZSTD_BLOCKSIZE_ABSOLUTEMAX (128 * 1024)   /* define, for static allocation */
+/*=====   Raw zstd block functions  =====*/
 ZSTDLIB_API size_t ZSTD_getBlockSizeMax(ZSTD_CCtx* cctx);
 ZSTDLIB_API size_t ZSTD_compressBlock  (ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 ZSTDLIB_API size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize);
 ZSTDLIB_API size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize);  /**< insert block into `dctx` history. Useful for uncompressed blocks */
 
 
-#endif   /* ZSTD_STATIC_LINKING_ONLY */
+#endif   /* ZSTD_H_ZSTD_STATIC_LINKING_ONLY */
 
 #if defined (__cplusplus)
 }
 #endif
-
-#endif  /* ZSTD_H_235446 */
diff --git a/contrib/zstd/zstd_common.c b/contrib/zstd/zstd_common.c
index 54bc91c89..8408a589a 100644
--- a/contrib/zstd/zstd_common.c
+++ b/contrib/zstd/zstd_common.c
@@ -16,7 +16,6 @@
 #include "error_private.h"
 #define ZSTD_STATIC_LINKING_ONLY
 #include "zstd.h"           /* declaration of ZSTD_isError, ZSTD_getErrorName, ZSTD_getErrorCode, ZSTD_getErrorString, ZSTD_versionNumber */
-#include "zbuff.h"          /* declaration of ZBUFF_isError, ZBUFF_getErrorName */
 
 
 /*-****************************************
@@ -42,16 +41,7 @@ ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
 
 /*! ZSTD_getErrorString() :
 *   provides error code string from enum */
-const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorName(code); }
-
-
-/* **************************************************************
-*  ZBUFF Error Management
-****************************************************************/
-unsigned ZBUFF_isError(size_t errorCode) { return ERR_isError(errorCode); }
-
-const char* ZBUFF_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
-
+const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
 
 
 /*=**************************************************************
diff --git a/contrib/zstd/zstd_compress.c b/contrib/zstd/zstd_compress.c
index 0116136c0..edf4609f7 100644
--- a/contrib/zstd/zstd_compress.c
+++ b/contrib/zstd/zstd_compress.c
@@ -8,30 +8,11 @@
  */
 
 
-
-/*-*******************************************************
-*  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 */
-#else
-#  ifdef __GNUC__
-#    define FORCE_INLINE static inline __attribute__((always_inline))
-#  else
-#    define FORCE_INLINE static inline
-#  endif
-#endif
-
-
 /*-*************************************
 *  Dependencies
 ***************************************/
 #include <string.h>         /* memset */
 #include "mem.h"
-#define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */
-#include "xxhash.h"         /* XXH_reset, update, digest */
 #define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
@@ -40,17 +21,48 @@
 
 
 /*-*************************************
+*  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) { return FSE_compressBound(srcSize) + 12; }
+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;
+}
 
 
 /*-*************************************
@@ -67,8 +79,9 @@ static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
 /*-*************************************
 *  Context memory management
 ***************************************/
-struct ZSTD_CCtx_s
-{
+typedef enum { zcss_init, zcss_load, zcss_flush, zcss_final } ZSTD_cStreamStage;
+
+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 */
@@ -77,16 +90,19 @@ struct ZSTD_CCtx_s
     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;
+    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   savedRep[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;
 
@@ -94,11 +110,29 @@ struct ZSTD_CCtx_s
     U32* hashTable;
     U32* hashTable3;
     U32* chainTable;
-    HUF_CElt* hufTable;
-    U32 flagStaticTables;
-    FSE_CTable offcodeCTable  [FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
-    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
-    FSE_CTable litlengthCTable  [FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
+    HUF_repeat hufCTable_repeatMode;
+    HUF_CElt* hufCTable;
+    U32 fseCTables_ready;
+    FSE_CTable* offcodeCTable;
+    FSE_CTable* matchlengthCTable;
+    FSE_CTable* litlengthCTable;
+    unsigned* entropyScratchSpace;
+
+    /* streaming */
+    ZSTD_CDict* cdictLocal;
+    const ZSTD_CDict* cdict;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inToCompress;
+    size_t inBuffPos;
+    size_t inBuffTarget;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outBuffContentSize;
+    size_t outBuffFlushedSize;
+    ZSTD_cStreamStage streamStage;
+    U32    frameEnded;
+    U64    pledgedSrcSize;
 };
 
 ZSTD_CCtx* ZSTD_createCCtx(void)
@@ -116,7 +150,7 @@ ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
     cctx = (ZSTD_CCtx*) ZSTD_malloc(sizeof(ZSTD_CCtx), customMem);
     if (!cctx) return NULL;
     memset(cctx, 0, sizeof(ZSTD_CCtx));
-    memcpy(&(cctx->customMem), &customMem, sizeof(ZSTD_customMem));
+    cctx->customMem = customMem;
     return cctx;
 }
 
@@ -124,13 +158,33 @@ size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
 {
     if (cctx==NULL) return 0;   /* support free on NULL */
     ZSTD_free(cctx->workSpace, cctx->customMem);
+    cctx->workSpace = NULL;
+    ZSTD_freeCDict(cctx->cdictLocal);
+    cctx->cdictLocal = NULL;
+    ZSTD_free(cctx->inBuff, cctx->customMem);
+    cctx->inBuff = NULL;
+    ZSTD_free(cctx->outBuff, cctx->customMem);
+    cctx->outBuff = NULL;
     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)
 {
-    return sizeof(*cctx) + cctx->workSpaceSize;
+    if (cctx==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*cctx) + cctx->workSpaceSize
+           + ZSTD_sizeof_CDict(cctx->cdictLocal)
+           + cctx->outBuffSize + cctx->inBuffSize;
+}
+
+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 */
@@ -138,43 +192,39 @@ 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;
+}
 
-#define CLAMP(val,min,max) { if (val<min) val=min; else if (val>max) val=max; }
-#define CLAMPCHECK(val,min,max) { if ((val<min) || (val>max)) return ERROR(compressionParameter_unsupported); }
 
 /** 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);
-    { U32 const searchLengthMin = (cParams.strategy == ZSTD_fast || cParams.strategy == ZSTD_greedy) ? ZSTD_SEARCHLENGTH_MIN+1 : ZSTD_SEARCHLENGTH_MIN;
-      U32 const searchLengthMax = (cParams.strategy == ZSTD_fast) ? ZSTD_SEARCHLENGTH_MAX : ZSTD_SEARCHLENGTH_MAX-1;
-      CLAMPCHECK(cParams.searchLength, searchLengthMin, searchLengthMax); }
+    CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
     CLAMPCHECK(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
-    if ((U32)(cParams.strategy) > (U32)ZSTD_btopt) return ERROR(compressionParameter_unsupported);
+    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) return ERROR(compressionParameter_unsupported);
     return 0;
 }
 
 
-/** ZSTD_checkCParams_advanced() :
-    temporary work-around, while the compressor compatibility remains limited regarding windowLog < 18 */
-size_t ZSTD_checkCParams_advanced(ZSTD_compressionParameters cParams, U64 srcSize)
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
 {
-    if (srcSize > (1ULL << ZSTD_WINDOWLOG_MIN)) return ZSTD_checkCParams(cParams);
-    if (cParams.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN) return ERROR(compressionParameter_unsupported);
-    if (srcSize <= (1ULL << cParams.windowLog)) cParams.windowLog = ZSTD_WINDOWLOG_MIN; /* fake value - temporary work around */
-    if (srcSize <= (1ULL << cParams.chainLog)) cParams.chainLog = ZSTD_CHAINLOG_MIN;    /* fake value - temporary work around */
-    if ((srcSize <= (1ULL << cParams.hashLog)) && ((U32)cParams.strategy < (U32)ZSTD_btlazy2)) cParams.hashLog = ZSTD_HASHLOG_MIN;       /* fake value - temporary work around */
-    return ZSTD_checkCParams(cParams);
+    U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
+    return hashLog - btScale;
 }
 
-
 /** ZSTD_adjustCParams() :
-    optimize cPar for a given input (`srcSize` and `dictSize`).
+    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.
@@ -187,16 +237,15 @@ ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, u
     {   U32 const minSrcSize = (srcSize==0) ? 500 : 0;
         U64 const rSize = srcSize + dictSize + minSrcSize;
         if (rSize < ((U64)1<<ZSTD_WINDOWLOG_MAX)) {
-            U32 const srcLog = ZSTD_highbit32((U32)(rSize)-1) + 1;
+            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 btPlus = (cPar.strategy == ZSTD_btlazy2) || (cPar.strategy == ZSTD_btopt);
-        U32 const maxChainLog = cPar.windowLog+btPlus;
-        if (cPar.chainLog > maxChainLog) cPar.chainLog = maxChainLog; }   /* <= ZSTD_CHAINLOG_MAX */
+    {   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 */
-    if ((cPar.hashLog  < ZSTD_HASHLOG_MIN) && ( (U32)cPar.strategy >= (U32)ZSTD_btlazy2)) cPar.hashLog = ZSTD_HASHLOG_MIN;  /* required to ensure collision resistance in bt */
 
     return cPar;
 }
@@ -213,112 +262,210 @@ size_t ZSTD_estimateCCtxSize(ZSTD_compressionParameters cParams)
     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)
+    size_t const optBudget = ((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 = tableSpace + (256*sizeof(U32)) /* huffTable */ + tokenSpace
-                             + ((cParams.strategy == ZSTD_btopt) ? optSpace : 0);
+    size_t const optSpace = ((cParams.strategy == ZSTD_btopt) || (cParams.strategy == ZSTD_btultra)) ? optBudget : 0;
+    size_t const neededSpace = entropySpace + tableSpace + tokenSpace + optSpace;
 
     return sizeof(ZSTD_CCtx) + neededSpace;
 }
 
-/*! ZSTD_resetCCtx_advanced() :
-    note : 'params' is expected to be validated */
-static size_t ZSTD_resetCCtx_advanced (ZSTD_CCtx* zc,
+
+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)   /* opt parser space */
+         & ((param1.cParams.searchLength==3) == (param2.cParams.searchLength==3));  /* hashlog3 space */
+}
+
+/*! 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,
-                                       U32 reset)
-{   /* note : params considered validated here */
-    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 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 = tableSpace + (256*sizeof(U32)) /* huffTable */ + tokenSpace
-                              + ((params.cParams.strategy == ZSTD_btopt) ? optSpace : 0);
-        if (zc->workSpaceSize < neededSpace) {
-            ZSTD_free(zc->workSpace, zc->customMem);
-            zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
-            if (zc->workSpace == NULL) return ERROR(memory_allocation);
-            zc->workSpaceSize = neededSpace;
-    }   }
+                                       ZSTD_compResetPolicy_e const crp)
+{
+    DEBUGLOG(5, "ZSTD_resetCCtx_internal \n");
+
+    if (crp == ZSTDcrp_continue)
+        if (ZSTD_equivalentParams(params, zc->params)) {
+            DEBUGLOG(5, "ZSTD_equivalentParams()==1 \n");
+            zc->fseCTables_ready = 0;
+            zc->hufCTable_repeatMode = HUF_repeat_none;
+            return ZSTD_continueCCtx(zc, params, frameContentSize);
+        }
 
-    if (reset) memset(zc->workSpace, 0, tableSpace );   /* reset only tables */
-    XXH64_reset(&zc->xxhState, 0);
-    zc->hashLog3 = hashLog3;
-    zc->hashTable = (U32*)(zc->workSpace);
-    zc->chainTable = zc->hashTable + hSize;
-    zc->hashTable3 = zc->chainTable + chainSize;
-    ptr = zc->hashTable3 + h3Size;
-    zc->hufTable = (HUF_CElt*)ptr;
-    zc->flagStaticTables = 0;
-    ptr = ((U32*)ptr) + 256;  /* note : HUF_CElt* is incomplete type, size is simulated using U32 */
-
-    zc->nextToUpdate = 1;
-    zc->nextSrc = NULL;
-    zc->base = NULL;
-    zc->dictBase = NULL;
-    zc->dictLimit = 0;
-    zc->lowLimit = 0;
-    zc->params = params;
-    zc->blockSize = blockSize;
-    zc->frameContentSize = frameContentSize;
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->rep[i] = repStartValue[i]; }
-
-    if (params.cParams.strategy == ZSTD_btopt) {
-        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;
+    {   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_btultra)) ? optPotentialSpace : 0;
+            size_t const neededSpace = entropySpace + optSpace + tableSpace + tokenSpace;
+            if (zc->workSpaceSize < neededSpace) {
+                DEBUGLOG(5, "Need to update workSpaceSize from %uK to %uK \n",
+                            (unsigned)zc->workSpaceSize>>10, (unsigned)neededSpace>>10);
+                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;
+        DEBUGLOG(5, "blockSize = %uK \n", (U32)blockSize>>10);
+        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;
-    }
-    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;
 
-    zc->stage = ZSTDcs_init;
-    zc->dictID = 0;
-    zc->loadedDictEnd = 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_btultra)) {
+            DEBUGLOG(5, "reserving optimal parser space ");
+            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;
+        }
 
-    return 0;
+        /* 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() :
-*   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()).
-*   @return : 0, or an error code */
-size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx)
+
+/*! 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)
 {
+    DEBUGLOG(5, "ZSTD_copyCCtx_internal \n");
     if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
 
     memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
-    ZSTD_resetCCtx_advanced(dstCCtx, srcCCtx->params, srcCCtx->frameContentSize, 0);
-    dstCCtx->params.fParams.contentSizeFlag = 0;   /* content size different from the one set during srcCCtx init */
+    {   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 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);
-        memcpy(dstCCtx->workSpace, srcCCtx->workSpace, tableSpace);
+        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 */
@@ -333,20 +480,36 @@ size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx)
     dstCCtx->dictID       = srcCCtx->dictID;
 
     /* copy entropy tables */
-    dstCCtx->flagStaticTables = srcCCtx->flagStaticTables;
-    if (srcCCtx->flagStaticTables) {
-        memcpy(dstCCtx->hufTable, srcCCtx->hufTable, 256*4);
-        memcpy(dstCCtx->litlengthCTable, srcCCtx->litlengthCTable, sizeof(dstCCtx->litlengthCTable));
-        memcpy(dstCCtx->matchlengthCTable, srcCCtx->matchlengthCTable, sizeof(dstCCtx->matchlengthCTable));
-        memcpy(dstCCtx->offcodeCTable, srcCCtx->offcodeCTable, sizeof(dstCCtx->offcodeCTable));
+    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` */
+ *  reduce table indexes by `reducerValue` */
 static void ZSTD_reduceTable (U32* const table, U32 const size, U32 const reducerValue)
 {
     U32 u;
@@ -375,7 +538,7 @@ static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
 *  Block entropic compression
 *********************************************************/
 
-/* See zstd_compression_format.md for detailed format description */
+/* 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)
 {
@@ -453,24 +616,30 @@ static size_t ZSTD_compressLiterals (ZSTD_CCtx* zc,
 
     /* small ? don't even attempt compression (speed opt) */
 #   define LITERAL_NOENTROPY 63
-    {   size_t const minLitSize = zc->flagStaticTables ? 6 : LITERAL_NOENTROPY;
+    {   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 */
-    if (zc->flagStaticTables && (lhSize==3)) {
-        hType = set_repeat;
-        singleStream = 1;
-        cLitSize = HUF_compress1X_usingCTable(ostart+lhSize, dstCapacity-lhSize, src, srcSize, zc->hufTable);
-    } else {
-        cLitSize = singleStream ? HUF_compress1X(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11)
-                                : HUF_compress2 (ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11);
+    {   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))
+    if ((cLitSize==0) | (cLitSize >= srcSize - minGain)) {
+        zc->hufCTable_repeatMode = HUF_repeat_none;
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    if (cLitSize==1)
+    }
+    if (cLitSize==1) {
+        zc->hufCTable_repeatMode = HUF_repeat_none;
         return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }
 
     /* Build header */
     switch(lhSize)
@@ -538,11 +707,11 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
 }
 
-
-size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
+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];
@@ -559,6 +728,7 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
     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;
@@ -586,15 +756,15 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
 
     /* CTable for Literal Lengths */
     {   U32 max = MaxLL;
-        size_t const mostFrequent = FSE_countFast(count, &max, llCodeTable, nbSeq);
+        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->flagStaticTables) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+        } 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(CTable_LitLength, LL_defaultNorm, MaxLL, LL_defaultNormLog);
+            FSE_buildCTable_wksp(CTable_LitLength, LL_defaultNorm, MaxLL, LL_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
             LLtype = set_basic;
         } else {
             size_t nbSeq_1 = nbSeq;
@@ -602,23 +772,23 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
             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 ERROR(GENERIC);
+              if (FSE_isError(NCountSize)) return NCountSize;
               op += NCountSize; }
-            FSE_buildCTable(CTable_LitLength, norm, max, tableLog);
+            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(count, &max, ofCodeTable, nbSeq);
+        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->flagStaticTables) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+        } 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(CTable_OffsetBits, OF_defaultNorm, MaxOff, OF_defaultNormLog);
+            FSE_buildCTable_wksp(CTable_OffsetBits, OF_defaultNorm, MaxOff, OF_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
             Offtype = set_basic;
         } else {
             size_t nbSeq_1 = nbSeq;
@@ -626,23 +796,23 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
             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 ERROR(GENERIC);
+              if (FSE_isError(NCountSize)) return NCountSize;
               op += NCountSize; }
-            FSE_buildCTable(CTable_OffsetBits, norm, max, tableLog);
+            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(count, &max, mlCodeTable, nbSeq);
+        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->flagStaticTables) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
+        } 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(CTable_MatchLength, ML_defaultNorm, MaxML, ML_defaultNormLog);
+            FSE_buildCTable_wksp(CTable_MatchLength, ML_defaultNorm, MaxML, ML_defaultNormLog, scratchBuffer, sizeof(scratchBuffer));
             MLtype = set_basic;
         } else {
             size_t nbSeq_1 = nbSeq;
@@ -650,14 +820,14 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
             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 ERROR(GENERIC);
+              if (FSE_isError(NCountSize)) return NCountSize;
               op += NCountSize; }
-            FSE_buildCTable(CTable_MatchLength, norm, max, tableLog);
+            FSE_buildCTable_wksp(CTable_MatchLength, norm, max, tableLog, scratchBuffer, sizeof(scratchBuffer));
             MLtype = set_compressed;
     }   }
 
     *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
-    zc->flagStaticTables = 0;
+    zc->fseCTables_ready = 0;
 
     /* Encoding Sequences */
     {   BIT_CStream_t blockStream;
@@ -665,8 +835,7 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
         FSE_CState_t  stateOffsetBits;
         FSE_CState_t  stateLitLength;
 
-        { size_t const errorCode = BIT_initCStream(&blockStream, op, oend-op);
-          if (ERR_isError(errorCode)) return ERROR(dstSize_tooSmall); }   /* not enough space remaining */
+        CHECK_E(BIT_initCStream(&blockStream, op, oend-op), dstSize_tooSmall); /* not enough space remaining */
 
         /* first symbols */
         FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
@@ -676,7 +845,18 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
         if (MEM_32bits()) BIT_flushBits(&blockStream);
         BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
         if (MEM_32bits()) BIT_flushBits(&blockStream);
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+        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;
@@ -698,7 +878,17 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
                 if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
                 BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
                 if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
-                BIT_addBits(&blockStream, sequences[n].offset, ofBits);         /* 31 */
+                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)*/
         }   }
 
@@ -713,16 +903,25 @@ size_t ZSTD_compressSequences(ZSTD_CCtx* zc,
 
     /* check compressibility */
 _check_compressibility:
-    { size_t const minGain = ZSTD_minGain(srcSize);
-      size_t const maxCSize = srcSize - minGain;
-      if ((size_t)(op-ostart) >= maxCSize) return 0; }
+    {   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->savedRep[i]; }
+    { 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.
@@ -731,27 +930,34 @@ _check_compressibility:
 */
 MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const void* literals, U32 offsetCode, size_t matchCode)
 {
-#if 0  /* for debug */
-    static const BYTE* g_start = NULL;
-    const U32 pos = (U32)(literals - g_start);
-    if (g_start==NULL) g_start = literals;
-    //if ((pos > 1) && (pos < 50000))
-        printf("Cpos %6u :%5u literals & match %3u bytes at distance %6u \n",
-               pos, (U32)litLength, (U32)matchCode+MINMATCH, (U32)offsetCode);
+#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); }
+    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); }
+    if (matchCode>0xFFFF) {
+        seqStorePtr->longLengthID = 2;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
     seqStorePtr->sequences[0].matchLength = (U16)matchCode;
 
     seqStorePtr->sequences++;
@@ -772,7 +978,14 @@ static unsigned ZSTD_NbCommonBytes (register size_t val)
 #       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 };
+            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 */
@@ -783,7 +996,10 @@ static unsigned ZSTD_NbCommonBytes (register size_t val)
 #       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 };
+            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
         }
@@ -855,7 +1071,7 @@ static size_t ZSTD_count_2segments(const BYTE* ip, const BYTE* match, const BYTE
 ***************************************/
 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 */
+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) ; }
@@ -987,8 +1203,8 @@ void ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
     }   }   }
 
     /* save reps for next block */
-    cctx->savedRep[0] = offset_1 ? offset_1 : offsetSaved;
-    cctx->savedRep[1] = offset_2 ? offset_2 : offsetSaved;
+    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
+    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1004,7 +1220,7 @@ static void ZSTD_compressBlock_fast(ZSTD_CCtx* ctx,
     const U32 mls = ctx->params.cParams.searchLength;
     switch(mls)
     {
-    default:
+    default: /* includes case 3 */
     case 4 :
         ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 4); return;
     case 5 :
@@ -1054,7 +1270,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
         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+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repMatchEnd, lowPrefixPtr) + EQUAL_READ32;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
             ip++;
             ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
         } else {
@@ -1066,7 +1282,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
             {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
                 const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
                 U32 offset;
-                mLength = ZSTD_count_2segments(ip+EQUAL_READ32, match+EQUAL_READ32, iend, matchEnd, lowPrefixPtr) + EQUAL_READ32;
+                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;
@@ -1080,7 +1296,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
 
         if (ip <= ilimit) {
             /* Fill Table */
-			hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;
+            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) {
@@ -1090,7 +1306,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
                 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 repLength2 = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch2+EQUAL_READ32, iend, repEnd2, lowPrefixPtr) + EQUAL_READ32;
+                    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;
@@ -1102,7 +1318,7 @@ static void ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
     }   }   }
 
     /* save reps for next block */
-    ctx->savedRep[0] = offset_1; ctx->savedRep[1] = offset_2;
+    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1118,7 +1334,7 @@ static void ZSTD_compressBlock_fast_extDict(ZSTD_CCtx* ctx,
     U32 const mls = ctx->params.cParams.searchLength;
     switch(mls)
     {
-    default:
+    default: /* includes case 3 */
     case 4 :
         ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 4); return;
     case 5 :
@@ -1193,7 +1409,9 @@ void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
         const BYTE* match = base + matchIndexS;
         hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
 
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) { /* note : by construction, offset_1 <= current */
+        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);
@@ -1204,15 +1422,15 @@ void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
                 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 h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-                U32 const matchIndex3 = hashLong[h3];
-                const BYTE* match3 = base + matchIndex3;
-                hashLong[h3] = current + 1;
-                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
-                    mLength = ZSTD_count(ip+9, match3+8, iend) + 8;
+                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-match3);
-                    while (((ip>anchor) & (match3>lowest)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
+                    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);
@@ -1256,8 +1474,8 @@ void ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
     }   }   }
 
     /* save reps for next block */
-    cctx->savedRep[0] = offset_1 ? offset_1 : offsetSaved;
-    cctx->savedRep[1] = offset_2 ? offset_2 : offsetSaved;
+    cctx->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
+    cctx->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1272,7 +1490,7 @@ static void ZSTD_compressBlock_doubleFast(ZSTD_CCtx* ctx, const void* src, size_
     const U32 mls = ctx->params.cParams.searchLength;
     switch(mls)
     {
-    default:
+    default: /* includes case 3 */
     case 4 :
         ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 4); return;
     case 5 :
@@ -1381,8 +1599,8 @@ static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
 
         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(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 */
@@ -1393,7 +1611,7 @@ static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
                 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+EQUAL_READ32, repMatch2+EQUAL_READ32, iend, repEnd2, lowPrefixPtr) + EQUAL_READ32;
+                    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;
@@ -1406,7 +1624,7 @@ static void ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
     }   }   }
 
     /* save reps for next block */
-    ctx->savedRep[0] = offset_1; ctx->savedRep[1] = offset_2;
+    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -1422,7 +1640,7 @@ static void ZSTD_compressBlock_doubleFast_extDict(ZSTD_CCtx* ctx,
     U32 const mls = ctx->params.cParams.searchLength;
     switch(mls)
     {
-    default:
+    default: /* includes case 3 */
     case 4 :
         ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 4); return;
     case 5 :
@@ -1457,7 +1675,7 @@ static U32 ZSTD_insertBt1(ZSTD_CCtx* zc, const BYTE* const ip, const U32 mls, co
     const U32 dictLimit = zc->dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* match = base + matchIndex;
+    const BYTE* match;
     const U32 current = (U32)(ip-base);
     const U32 btLow = btMask >= current ? 0 : current - btMask;
     U32* smallerPtr = bt + 2*(current&btMask);
@@ -1476,8 +1694,9 @@ static U32 ZSTD_insertBt1(ZSTD_CCtx* zc, const BYTE* const ip, const U32 mls, co
     hashTable[h] = current;   /* Update Hash Table */
 
     while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* nextPtr = bt + 2*(matchIndex & btMask);
+        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) {
@@ -1506,7 +1725,7 @@ static U32 ZSTD_insertBt1(ZSTD_CCtx* zc, const BYTE* const ip, const U32 mls, co
             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] */
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
         }
 
         if (matchLength > bestLength) {
@@ -1573,7 +1792,7 @@ static size_t ZSTD_insertBtAndFindBestMatch (
     hashTable[h] = current;   /* Update Hash Table */
 
     while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* nextPtr = bt + 2*(matchIndex & btMask);
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         const BYTE* match;
 
@@ -1585,7 +1804,7 @@ static size_t ZSTD_insertBtAndFindBestMatch (
             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] */
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
         }
 
         if (matchLength > bestLength) {
@@ -1651,9 +1870,10 @@ static size_t ZSTD_BtFindBestMatch_selectMLS (
 {
     switch(matchLengthSearch)
     {
-    default :
+    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);
     }
 }
@@ -1690,9 +1910,10 @@ static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
 {
     switch(matchLengthSearch)
     {
-    default :
+    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);
     }
 }
@@ -1705,7 +1926,7 @@ static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
 #define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & mask]
 
 /* Update chains up to ip (excluded)
-   Assumption : always within prefix (ie. not within extDict) */
+   Assumption : always within prefix (i.e. not within extDict) */
 FORCE_INLINE
 U32 ZSTD_insertAndFindFirstIndex (ZSTD_CCtx* zc, const BYTE* ip, U32 mls)
 {
@@ -1749,7 +1970,7 @@ size_t ZSTD_HcFindBestMatch_generic (
     const U32 current = (U32)(ip-base);
     const U32 minChain = current > chainSize ? current - chainSize : 0;
     int nbAttempts=maxNbAttempts;
-    size_t ml=EQUAL_READ32-1;
+    size_t ml=4-1;
 
     /* HC4 match finder */
     U32 matchIndex = ZSTD_insertAndFindFirstIndex (zc, ip, mls);
@@ -1764,11 +1985,15 @@ size_t ZSTD_HcFindBestMatch_generic (
         } else {
             match = dictBase + matchIndex;
             if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
-                currentMl = ZSTD_count_2segments(ip+EQUAL_READ32, match+EQUAL_READ32, iLimit, dictEnd, prefixStart) + EQUAL_READ32;
+                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, and avoid read overflow*/ }
+        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);
@@ -1786,9 +2011,10 @@ FORCE_INLINE size_t ZSTD_HcFindBestMatch_selectMLS (
 {
     switch(matchLengthSearch)
     {
-    default :
+    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);
     }
 }
@@ -1802,9 +2028,10 @@ FORCE_INLINE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
 {
     switch(matchLengthSearch)
     {
-    default :
+    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);
     }
 }
@@ -1852,7 +2079,7 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
         /* 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+EQUAL_READ32, ip+1+EQUAL_READ32-offset_1, iend) + EQUAL_READ32;
+            matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             if (depth==0) goto _storeSequence;
         }
 
@@ -1863,7 +2090,7 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
                 matchLength = ml2, start = ip, offset=offsetFound;
         }
 
-        if (matchLength < EQUAL_READ32) {
+        if (matchLength < 4) {
             ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
             continue;
         }
@@ -1873,17 +2100,17 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
         while (ip<ilimit) {
             ip ++;
             if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
-                size_t const mlRep = ZSTD_count(ip+EQUAL_READ32, ip+EQUAL_READ32-offset_1, iend) + EQUAL_READ32;
+                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 >= EQUAL_READ32) && (gain2 > gain1))
+                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 >= EQUAL_READ32) && (gain2 > gain1)) {
+                if ((ml2 >= 4) && (gain2 > gain1)) {
                     matchLength = ml2, offset = offset2, start = ip;
                     continue;   /* search a better one */
             }   }
@@ -1892,17 +2119,17 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
             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+EQUAL_READ32, ip+EQUAL_READ32-offset_1, iend) + EQUAL_READ32;
+                    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 >= EQUAL_READ32) && (gain2 > gain1))
+                    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 >= EQUAL_READ32) && (gain2 > gain1)) {
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
                         matchLength = ml2, offset = offset2, start = ip;
                         continue;
             }   }   }
@@ -1911,7 +2138,9 @@ void ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
 
         /* 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 */
+            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);
         }
@@ -1928,7 +2157,7 @@ _storeSequence:
              && ((offset_2>0)
              & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
             /* store sequence */
-            matchLength = ZSTD_count(ip+EQUAL_READ32, ip+EQUAL_READ32-offset_2, iend) + EQUAL_READ32;
+            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;
@@ -1937,8 +2166,8 @@ _storeSequence:
     }   }
 
     /* Save reps for next block */
-    ctx->savedRep[0] = offset_1 ? offset_1 : savedOffset;
-    ctx->savedRep[1] = offset_2 ? offset_2 : savedOffset;
+    ctx->repToConfirm[0] = offset_1 ? offset_1 : savedOffset;
+    ctx->repToConfirm[1] = offset_2 ? offset_2 : savedOffset;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -2017,7 +2246,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
             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+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repEnd, prefixStart) + 4;
                 if (depth==0) goto _storeSequence;
         }   }
 
@@ -2028,7 +2257,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 matchLength = ml2, start = ip, offset=offsetFound;
         }
 
-         if (matchLength < EQUAL_READ32) {
+         if (matchLength < 4) {
             ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
             continue;
         }
@@ -2047,10 +2276,10 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 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+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
+                    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 >= EQUAL_READ32) && (gain2 > gain1))
+                    if ((repLength >= 4) && (gain2 > gain1))
                         matchLength = repLength, offset = 0, start = ip;
             }   }
 
@@ -2059,7 +2288,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 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 >= EQUAL_READ32) && (gain2 > gain1)) {
+                if ((ml2 >= 4) && (gain2 > gain1)) {
                     matchLength = ml2, offset = offset2, start = ip;
                     continue;   /* search a better one */
             }   }
@@ -2077,10 +2306,10 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                     if (MEM_read32(ip) == MEM_read32(repMatch)) {
                         /* repcode detected */
                         const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                        size_t repLength = ZSTD_count_2segments(ip+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
-                        int gain2 = (int)(repLength * 4);
-                        int gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
-                        if ((repLength >= EQUAL_READ32) && (gain2 > gain1))
+                        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;
                 }   }
 
@@ -2089,7 +2318,7 @@ void ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                     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 >= EQUAL_READ32) && (gain2 > gain1)) {
+                    if ((ml2 >= 4) && (gain2 > gain1)) {
                         matchLength = ml2, offset = offset2, start = ip;
                         continue;
             }   }   }
@@ -2121,7 +2350,7 @@ _storeSequence:
             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+EQUAL_READ32, repMatch+EQUAL_READ32, iend, repEnd, prefixStart) + EQUAL_READ32;
+                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;
@@ -2132,7 +2361,7 @@ _storeSequence:
     }   }
 
     /* Save reps for next block */
-    ctx->savedRep[0] = offset_1; ctx->savedRep[1] = offset_2;
+    ctx->repToConfirm[0] = offset_1; ctx->repToConfirm[1] = offset_2;
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -2169,7 +2398,17 @@ static void ZSTD_compressBlock_btlazy2_extDict(ZSTD_CCtx* ctx, const void* src,
 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);
+    ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 0);
+#else
+    (void)ctx; (void)src; (void)srcSize;
+    return;
+#endif
+}
+
+static void ZSTD_compressBlock_btultra(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;
@@ -2179,7 +2418,17 @@ static void ZSTD_compressBlock_btopt(ZSTD_CCtx* ctx, const void* src, size_t src
 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);
+    ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 0);
+#else
+    (void)ctx; (void)src; (void)srcSize;
+    return;
+#endif
+}
+
+static void ZSTD_compressBlock_btultra_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;
@@ -2191,9 +2440,13 @@ typedef void (*ZSTD_blockCompressor) (ZSTD_CCtx* ctx, const void* src, size_t sr
 
 static ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
 {
-    static const ZSTD_blockCompressor blockCompressor[2][7] = {
-        { ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy, ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2, ZSTD_compressBlock_btopt },
-        { 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 }
+    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_btultra },
+        { 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_btultra_extDict }
     };
 
     return blockCompressor[extDict][(U32)strat];
@@ -2209,7 +2462,7 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCa
     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));   /* update tree not updated after finding very long rep matches */
+        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);
 }
@@ -2234,27 +2487,28 @@ static size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
     BYTE* op = ostart;
     U32 const maxDist = 1 << cctx->params.cParams.windowLog;
 
-    if (cctx->params.fParams.checksumFlag)
+    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 (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 > (1<<30)) {
-            U32 const btplus = (cctx->params.cParams.strategy == ZSTD_btlazy2) | (cctx->params.cParams.strategy == ZSTD_btopt);
-            U32 const chainMask = (1 << (cctx->params.cParams.chainLog - btplus)) - 1;
-            U32 const supLog = MAX(cctx->params.cParams.chainLog, 17 /* blockSize */);
-            U32 const newLowLimit = (cctx->lowLimit & chainMask) + (1 << supLog);   /* preserve position % chainSize, ensure current-repcode doesn't underflow */
-            U32 const correction = cctx->lowLimit - newLowLimit;
+        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 = newLowLimit;
+            cctx->lowLimit -= correction;
             cctx->dictLimit -= correction;
             if (cctx->nextToUpdate < correction) cctx->nextToUpdate = 0;
             else cctx->nextToUpdate -= correction;
@@ -2296,10 +2550,11 @@ static size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
 static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
                                     ZSTD_parameters params, U64 pledgedSrcSize, U32 dictID)
 {   BYTE* const op = (BYTE*)dst;
-    U32   const dictIDSizeCode = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
+    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-1));
+    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 */
@@ -2308,6 +2563,8 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
     size_t pos;
 
     if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
+    DEBUGLOG(5, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u \n",
+                !params.fParams.noDictIDFlag, dictID,  dictIDSizeCode);
 
     MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
     op[4] = frameHeaderDecriptionByte; pos=5;
@@ -2371,12 +2628,15 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
 
     cctx->nextSrc = ip + srcSize;
 
-    {   size_t const cSize = frame ?
+    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;
 }
 
 
@@ -2384,7 +2644,7 @@ 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, 0);
+    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
 }
 
 
@@ -2397,10 +2657,12 @@ size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const
 {
     size_t const blockSizeMax = ZSTD_getBlockSizeMax(cctx);
     if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
-    return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0, 0);
+    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;
@@ -2412,7 +2674,7 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_CCtx* zc, const void* src, size_t
     zc->dictBase = zc->base;
     zc->base += ip - zc->nextSrc;
     zc->nextToUpdate = zc->dictLimit;
-    zc->loadedDictEnd = (U32)(iend - zc->base);
+    zc->loadedDictEnd = zc->forceWindow ? 0 : (U32)(iend - zc->base);
 
     zc->nextSrc = iend;
     if (srcSize <= HASH_READ_SIZE) return 0;
@@ -2430,111 +2692,151 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_CCtx* zc, const void* src, size_t
     case ZSTD_greedy:
     case ZSTD_lazy:
     case ZSTD_lazy2:
-        ZSTD_insertAndFindFirstIndex (zc, iend-HASH_READ_SIZE, zc->params.cParams.searchLength);
+        if (srcSize >= HASH_READ_SIZE)
+            ZSTD_insertAndFindFirstIndex(zc, iend-HASH_READ_SIZE, zc->params.cParams.searchLength);
         break;
 
     case ZSTD_btlazy2:
     case ZSTD_btopt:
-        ZSTD_updateTree(zc, iend-HASH_READ_SIZE, iend, 1 << zc->params.cParams.searchLog, zc->params.cParams.searchLength);
+    case ZSTD_btultra:
+        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 = zc->loadedDictEnd;
+    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 :
-     Magic == ZSTD_DICT_MAGIC (4 bytes)
-     HUF_writeCTable(256)
-     FSE_writeNCount(off)
-     FSE_writeNCount(ml)
-     FSE_writeNCount(ll)
-     RepOffsets
-     Dictionary content
-*/
-/*! ZSTD_loadDictEntropyStats() :
-    @return : size read from dictionary
-    note : magic number supposed already checked */
-static size_t ZSTD_loadDictEntropyStats(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+ * 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)];
 
-    {   size_t const hufHeaderSize = HUF_readCTable(cctx->hufTable, 255, dict, dictSize);
+    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;
     }
 
-    {   short offcodeNCount[MaxOff+1];
-        unsigned offcodeMaxValue = MaxOff, offcodeLog = OffFSELog;
+    {   unsigned offcodeLog;
         size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
-        { size_t const errorCode = FSE_buildCTable(cctx->offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog);
-          if (FSE_isError(errorCode)) 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 = MLFSELog;
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
         size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        { size_t const errorCode = FSE_buildCTable(cctx->matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
-          if (FSE_isError(errorCode)) 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 = LLFSELog;
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
         size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        { size_t const errorCode = FSE_buildCTable(cctx->litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog);
-          if (FSE_isError(errorCode)) 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); if (cctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
-    cctx->rep[1] = MEM_readLE32(dictPtr+4); if (cctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
-    cctx->rep[2] = MEM_readLE32(dictPtr+8); if (cctx->rep[2] >= dictSize) 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;
 
-    cctx->flagStaticTables = 1;
-    return dictPtr - (const BYTE*)dict;
+    {   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* zc, const void* dict, size_t dictSize)
+static size_t ZSTD_compress_insertDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
 {
     if ((dict==NULL) || (dictSize<=8)) return 0;
 
-    /* default : dict is pure content */
-    if (MEM_readLE32(dict) != ZSTD_DICT_MAGIC) return ZSTD_loadDictionaryContent(zc, dict, dictSize);
-    zc->dictID = zc->params.fParams.noDictIDFlag ? 0 :  MEM_readLE32((const char*)dict+4);
+    /* dict as pure content */
+    if ((MEM_readLE32(dict) != ZSTD_DICT_MAGIC) || (cctx->forceRawDict))
+        return ZSTD_loadDictionaryContent(cctx, dict, dictSize);
 
-    /* known magic number : dict is parsed for entropy stats and content */
-    {   size_t const eSize_8 = ZSTD_loadDictEntropyStats(zc, (const char*)dict+8 /* skip dictHeader */, dictSize-8);
-        size_t const eSize = eSize_8 + 8;
-        if (ZSTD_isError(eSize_8)) return eSize_8;
-        return ZSTD_loadDictionaryContent(zc, (const char*)dict+eSize, dictSize-eSize);
-    }
+    /* 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* zc,
+static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
                              const void* dict, size_t dictSize,
                                    ZSTD_parameters params, U64 pledgedSrcSize)
 {
-    size_t const resetError = ZSTD_resetCCtx_advanced(zc, params, pledgedSrcSize, 1);
-    if (ZSTD_isError(resetError)) return resetError;
-
-    return ZSTD_compress_insertDictionary(zc, dict, dictSize);
+    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);
 }
 
 
@@ -2545,9 +2847,7 @@ size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
                                    ZSTD_parameters params, unsigned long long pledgedSrcSize)
 {
     /* compression parameters verification and optimization */
-    { size_t const errorCode = ZSTD_checkCParams_advanced(params.cParams, pledgedSrcSize);
-      if (ZSTD_isError(errorCode)) return errorCode; }
-
+    CHECK_F(ZSTD_checkCParams(params.cParams));
     return ZSTD_compressBegin_internal(cctx, dict, dictSize, params, pledgedSrcSize);
 }
 
@@ -2559,9 +2859,9 @@ size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t di
 }
 
 
-size_t ZSTD_compressBegin(ZSTD_CCtx* zc, int compressionLevel)
+size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
 {
-    return ZSTD_compressBegin_usingDict(zc, NULL, 0, compressionLevel);
+    return ZSTD_compressBegin_usingDict(cctx, NULL, 0, compressionLevel);
 }
 
 
@@ -2574,6 +2874,7 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
     BYTE* op = ostart;
     size_t fhSize = 0;
 
+    DEBUGLOG(5, "ZSTD_writeEpilogue \n");
     if (cctx->stage == ZSTDcs_created) return ERROR(stage_wrong);  /* init missing */
 
     /* special case : empty frame */
@@ -2611,10 +2912,15 @@ size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
                    const void* src, size_t srcSize)
 {
     size_t endResult;
-    size_t const cSize = ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1, 1);
+    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;
 }
 
@@ -2625,9 +2931,7 @@ static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
                          const void* dict,size_t dictSize,
                                ZSTD_parameters params)
 {
-    size_t const errorCode = ZSTD_compressBegin_internal(cctx, dict, dictSize, params, srcSize);
-    if(ZSTD_isError(errorCode)) return errorCode;
-
+    CHECK_F(ZSTD_compressBegin_internal(cctx, dict, dictSize, params, srcSize));
     return ZSTD_compressEnd(cctx, dst,  dstCapacity, src, srcSize);
 }
 
@@ -2637,14 +2941,14 @@ size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
                          const void* dict,size_t dictSize,
                                ZSTD_parameters params)
 {
-    size_t const errorCode = ZSTD_checkCParams_advanced(params.cParams, srcSize);
-    if (ZSTD_isError(errorCode)) return errorCode;
+    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)
+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, dictSize);
+    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);
 }
@@ -2669,39 +2973,72 @@ size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcS
 /* =====  Dictionary API  ===== */
 
 struct ZSTD_CDict_s {
-    void* dictContent;
+    void* dictBuffer;
+    const void* dictContent;
     size_t dictContentSize;
     ZSTD_CCtx* refContext;
 };  /* typedef'd tp ZSTD_CDict within "zstd.h" */
 
-ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize, ZSTD_parameters params, ZSTD_customMem customMem)
+/*! ZSTD_estimateCDictSize() :
+ *  Estimate amount of memory that will be needed to create a dictionary with following arguments */
+size_t ZSTD_estimateCDictSize(ZSTD_compressionParameters cParams, size_t dictSize)
+{
+    cParams = ZSTD_adjustCParams(cParams, 0, dictSize);
+    return sizeof(ZSTD_CDict) + dictSize + ZSTD_estimateCCtxSize(cParams);
+}
+
+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);
-        void* const dictContent = ZSTD_malloc(dictSize, customMem);
         ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(customMem);
 
-        if (!dictContent || !cdict || !cctx) {
-            ZSTD_free(dictContent, customMem);
+        if (!cdict || !cctx) {
             ZSTD_free(cdict, customMem);
-            ZSTD_free(cctx, customMem);
+            ZSTD_freeCCtx(cctx);
             return NULL;
         }
 
-        memcpy(dictContent, dict, dictSize);
-        {   size_t const errorCode = ZSTD_compressBegin_advanced(cctx, dictContent, dictSize, params, 0);
+        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(dictContent, customMem);
+                ZSTD_free(cdict->dictBuffer, customMem);
                 ZSTD_free(cdict, customMem);
-                ZSTD_free(cctx, customMem);
+                ZSTD_freeCCtx(cctx);
                 return NULL;
         }   }
 
-        cdict->dictContent = dictContent;
-        cdict->dictContentSize = dictSize;
         cdict->refContext = cctx;
+        cdict->dictContentSize = dictSize;
         return cdict;
     }
 }
@@ -2709,68 +3046,89 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dict, size_t dictSize, ZSTD_pa
 ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
 {
     ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTD_createCDict_advanced(dict, dictSize, params, allocator);
+    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 cMem = cdict->refContext->customMem;
+    {   ZSTD_customMem const cMem = cdict->refContext->customMem;
         ZSTD_freeCCtx(cdict->refContext);
-        ZSTD_free(cdict->dictContent, cMem);
+        ZSTD_free(cdict->dictBuffer, cMem);
         ZSTD_free(cdict, cMem);
         return 0;
     }
 }
 
-/*! 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 level is decided during dictionary creation */
-ZSTDLIB_API size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
-                                           void* dst, size_t dstCapacity,
-                                     const void* src, size_t srcSize,
-                                     const ZSTD_CDict* cdict)
-{
-    size_t const errorCode = ZSTD_copyCCtx(cctx, cdict->refContext);
-    if (ZSTD_isError(errorCode)) return errorCode;
+static ZSTD_parameters ZSTD_getParamsFromCDict(const ZSTD_CDict* cdict) {
+    return ZSTD_getParamsFromCCtx(cdict->refContext);
+}
 
-    if (cdict->refContext->params.fParams.contentSizeFlag==1) {
-        cctx->params.fParams.contentSizeFlag = 1;
-        cctx->frameContentSize = srcSize;
+/* 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* zc;
-    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;
-    ZSTD_customMem customMem;
-};   /* typedef'd to ZSTD_CStream within "zstd.h" */
-
 ZSTD_CStream* ZSTD_createCStream(void)
 {
     return ZSTD_createCStream_advanced(defaultCustomMem);
@@ -2778,86 +3136,166 @@ ZSTD_CStream* ZSTD_createCStream(void)
 
 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->zc = ZSTD_createCCtx_advanced(customMem);
-    if (zcs->zc == NULL) { ZSTD_freeCStream(zcs); return NULL; }
-    return zcs;
+    /* CStream and CCtx are now same object */
+    return ZSTD_createCCtx_advanced(customMem);
 }
 
 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->zc);
-        ZSTD_free(zcs->inBuff, cMem);
-        ZSTD_free(zcs->outBuff, cMem);
-        ZSTD_free(zcs, cMem);
-        return 0;
-    }
+    return ZSTD_freeCCtx(zcs);   /* same object */
+}
+
+size_t ZSTD_estimateCStreamSize(ZSTD_compressionParameters cParams)
+{
+    size_t const CCtxSize = ZSTD_estimateCCtxSize(cParams);
+    size_t const blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, (size_t)1 << cParams.windowLog);
+    size_t const inBuffSize = ((size_t)1 << cParams.windowLog) + blockSize;
+    size_t const outBuffSize = ZSTD_compressBound(blockSize) + 1;
+    size_t const streamingSize = inBuffSize + outBuffSize;
+
+    return CCtxSize + streamingSize;
 }
 
 
 /*======   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 */ ; }
 
-size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
-                                 const void* dict, size_t dictSize,
-                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)
+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, ZSTD_parameters params, 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, zcs->cdict, params.fParams, pledgedSrcSize))
+    else CHECK_F(ZSTD_compressBegin_internal(zcs, NULL, 0, params, pledgedSrcSize));
+
+    zcs->inToCompress = 0;
+    zcs->inBuffPos = 0;
+    zcs->inBuffTarget = zcs->blockSize;
+    zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
+    zcs->streamStage = zcss_load;
+    zcs->frameEnded = 0;
+    zcs->pledgedSrcSize = pledgedSrcSize;
+    return 0;   /* ready to go */
+}
+
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
+{
+
+    ZSTD_parameters params = zcs->params;
+    params.fParams.contentSizeFlag = (pledgedSrcSize > 0);
+    DEBUGLOG(5, "ZSTD_resetCStream : dictIDFlag == %u \n", !zcs->params.fParams.noDictIDFlag);
+    return ZSTD_resetCStream_internal(zcs, params, 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)));
+    zcs->blockSize = MIN(ZSTD_BLOCKSIZE_ABSOLUTEMAX, (size_t)1 << params.cParams.windowLog);
+
     /* allocate buffers */
-    {   size_t const neededInBuffSize = (size_t)1 << params.cParams.windowLog;
+    {   size_t const neededInBuffSize = ((size_t)1 << params.cParams.windowLog) + zcs->blockSize;
         if (zcs->inBuffSize < neededInBuffSize) {
-            zcs->inBuffSize = neededInBuffSize;
-            ZSTD_free(zcs->inBuff, zcs->customMem);  /* should not be necessary */
-            zcs->inBuff = (char*) ZSTD_malloc(neededInBuffSize, zcs->customMem);
+            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) {
-        zcs->outBuffSize = ZSTD_compressBound(zcs->blockSize)+1;
-        ZSTD_free(zcs->outBuff, zcs->customMem);   /* should not be necessary */
-        zcs->outBuff = (char*) ZSTD_malloc(zcs->outBuffSize, zcs->customMem);
+        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;
     }
 
-    { size_t const errorCode = ZSTD_compressBegin_advanced(zcs->zc, dict, dictSize, params, pledgedSrcSize);
-      if (ZSTD_isError(errorCode)) return errorCode; }
+    DEBUGLOG(5, "ZSTD_initCStream_stage2 : dictIDFlag == %u \n", !params.fParams.noDictIDFlag);
+    return ZSTD_resetCStream_internal(zcs, params, pledgedSrcSize);
+}
 
-    zcs->inToCompress = 0;
-    zcs->inBuffPos = 0;
-    zcs->inBuffTarget = zcs->blockSize;
-    zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
-    zcs->stage = zcss_load;
-    zcs->checksum = params.fParams.checksumFlag > 0;
-    zcs->frameEnded = 0;
-    return 0;   /* ready to go */
+/* 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 : pledgedSrcSize == %u \n", (U32)pledgedSrcSize);
+    DEBUGLOG(5, "wlog %u \n", params.cParams.windowLog);
+    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_advanced(zcs, dict, dictSize, params, 0);
+    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)
 {
-    return ZSTD_initCStream_usingDict(zcs, NULL, 0, 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)
 {
-    return sizeof(zcs) + ZSTD_sizeof_CCtx(zcs->zc) + zcs->outBuffSize + zcs->inBuffSize;
+    return ZSTD_sizeof_CCtx(zcs);   /* same object */
 }
 
 /*======   Compression   ======*/
@@ -2884,8 +3322,9 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
     char* const oend = ostart + *dstCapacityPtr;
     char* op = ostart;
 
+    DEBUGLOG(5, "ZSTD_compressStream_generic \n");
     while (someMoreWork) {
-        switch(zcs->stage)
+        switch(zcs->streamStage)
         {
         case zcss_init: return ERROR(init_missing);   /* call ZBUFF_compressInit() first ! */
 
@@ -2893,12 +3332,14 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
             /* complete inBuffer */
             {   size_t const toLoad = zcs->inBuffTarget - zcs->inBuffPos;
                 size_t const loaded = ZSTD_limitCopy(zcs->inBuff + zcs->inBuffPos, toLoad, ip, iend-ip);
+                DEBUGLOG(5, "loading %u/%u \n", (U32)loaded, (U32)toLoad);
                 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) */
+            DEBUGLOG(5, "stream compression stage (flush==%u)\n", flush);
             {   void* cDst;
                 size_t cSize;
                 size_t const iSize = zcs->inBuffPos - zcs->inToCompress;
@@ -2908,29 +3349,33 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 else
                     cDst = zcs->outBuff, oSize = zcs->outBuffSize;
                 cSize = (flush == zsf_end) ?
-                        ZSTD_compressEnd(zcs->zc, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize) :
-                        ZSTD_compressContinue(zcs->zc, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize);
+                        ZSTD_compressEnd(zcs, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize) :
+                        ZSTD_compressContinue(zcs, cDst, oSize, zcs->inBuff + zcs->inToCompress, iSize);
                 if (ZSTD_isError(cSize)) return cSize;
+                DEBUGLOG(5, "cSize = %u \n", (U32)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->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;   /* note : inBuffTarget == blockSize <= inBuffSize */
+                assert(zcs->inBuffTarget <= zcs->inBuffSize);
                 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 */
+                zcs->streamStage = zcss_flush;   /* pass-through to flush stage */
             }
-
+	    /* fall-through */
         case zcss_flush:
+            DEBUGLOG(5, "flush stage \n");
             {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
                 size_t const flushed = ZSTD_limitCopy(op, oend-op, zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
+                DEBUGLOG(5, "toFlush: %u  ; flushed: %u \n", (U32)toFlush, (U32)flushed);
                 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;
+                zcs->streamStage = zcss_load;
                 break;
             }
 
@@ -2974,8 +3419,8 @@ 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 */
+                                                     (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;
@@ -2989,21 +3434,26 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
     BYTE* const oend = (BYTE*)(output->dst) + output->size;
     BYTE* op = ostart;
 
-    if (zcs->stage != zcss_final) {
+    DEBUGLOG(5, "ZSTD_endStream (dstCapacity : %u) \n", (U32)(oend-op));
+    if (zcs->streamStage != 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, &srcSize, zsf_end);  /* use a valid src address instead of NULL */
+        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);
+            return remainingToFlush + ZSTD_BLOCKHEADERSIZE /* final empty block */
+                   + ((zcs->params.fParams.checksumFlag > 0) * 4) /* optional 32-bits checksum */;
         }
         /* create epilogue */
-        zcs->stage = zcss_final;
+        zcs->streamStage = zcss_final;
         zcs->outBuffContentSize = !notEnded ? 0 :
-            ZSTD_compressEnd(zcs->zc, zcs->outBuff, zcs->outBuffSize, NULL, 0);  /* write epilogue, including final empty block, into outBuff */
+            /* write epilogue, including final empty block, into outBuff */
+            ZSTD_compressEnd(zcs, zcs->outBuff, zcs->outBuffSize, NULL, 0);
+        if (ZSTD_isError(zcs->outBuffContentSize)) return zcs->outBuffContentSize;
     }
 
     /* flush epilogue */
@@ -3012,7 +3462,7 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
         op += flushed;
         zcs->outBuffFlushedSize += flushed;
         output->pos += op-ostart;
-        if (toFlush==flushed) zcs->stage = zcss_init;  /* end reached */
+        if (toFlush==flushed) zcs->streamStage = zcss_init;  /* end reached */
         return toFlush - flushed;
     }
 }
@@ -3046,11 +3496,11 @@ static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEV
     { 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, 23, 22,  6,  5, 32, ZSTD_btopt   },  /* level 18 */
+    { 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_btopt   },  /* level 20 */
-    { 26, 26, 23,  7,  3,256, ZSTD_btopt   },  /* level 21 */
-    { 27, 27, 25,  9,  3,512, ZSTD_btopt   },  /* level 22 */
+    { 25, 25, 23,  7,  3, 64, ZSTD_btultra },  /* level 20 */
+    { 26, 26, 23,  7,  3,256, ZSTD_btultra },  /* level 21 */
+    { 27, 27, 25,  9,  3,512, ZSTD_btultra },  /* level 22 */
 },
 {   /* for srcSize <= 256 KB */
     /* W,  C,  H,  S,  L,  T, strat */
@@ -3074,9 +3524,9 @@ static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEV
     { 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_btopt   },  /* level 20.*/
-    { 18, 19, 18, 12,  3,512, ZSTD_btopt   },  /* level 21.*/
-    { 18, 19, 18, 13,  3,512, ZSTD_btopt   },  /* level 22.*/
+    { 18, 19, 18, 11,  3,512, ZSTD_btultra },  /* level 20.*/
+    { 18, 19, 18, 12,  3,512, ZSTD_btultra },  /* level 21.*/
+    { 18, 19, 18, 13,  3,512, ZSTD_btultra },  /* level 22.*/
 },
 {   /* for srcSize <= 128 KB */
     /* W,  C,  H,  S,  L,  T, strat */
@@ -3100,9 +3550,9 @@ static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEV
     { 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_btopt   },  /* level 20.*/
-    { 17, 18, 17, 10,  3,256, ZSTD_btopt   },  /* level 21.*/
-    { 17, 18, 17, 11,  3,512, ZSTD_btopt   },  /* level 22.*/
+    { 17, 18, 17,  9,  3,256, ZSTD_btultra },  /* level 20.*/
+    { 17, 18, 17, 10,  3,256, ZSTD_btultra },  /* level 21.*/
+    { 17, 18, 17, 11,  3,512, ZSTD_btultra },  /* level 22.*/
 },
 {   /* for srcSize <= 16 KB */
     /* W,  C,  H,  S,  L,  T, strat */
@@ -3126,9 +3576,9 @@ static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEV
     { 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_btopt   },  /* level 20.*/
-    { 14, 15, 15,  9,  3,256, ZSTD_btopt   },  /* level 21.*/
-    { 14, 15, 15, 10,  3,256, ZSTD_btopt   },  /* level 22.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btultra },  /* level 20.*/
+    { 14, 15, 15,  9,  3,256, ZSTD_btultra },  /* level 21.*/
+    { 14, 15, 15, 10,  3,256, ZSTD_btultra },  /* level 22.*/
 },
 };
 
diff --git a/contrib/zstd/zstd_decompress.c b/contrib/zstd/zstd_decompress.c
index fb1ee35a0..379842b57 100644
--- a/contrib/zstd/zstd_decompress.c
+++ b/contrib/zstd/zstd_decompress.c
@@ -28,14 +28,13 @@
 #  define ZSTD_LEGACY_SUPPORT 0
 #endif
 
-
 /*!
 *  MAXWINDOWSIZE_DEFAULT :
 *  maximum window size accepted by DStream, by default.
 *  Frames requiring more memory will be rejected.
 */
 #ifndef ZSTD_MAXWINDOWSIZE_DEFAULT
-#  define ZSTD_MAXWINDOWSIZE_DEFAULT (257 << 20)   /* 257 MB */
+#  define ZSTD_MAXWINDOWSIZE_DEFAULT ((1 << ZSTD_WINDOWLOG_MAX) + 1)   /* defined within zstd.h */
 #endif
 
 
@@ -44,8 +43,6 @@
 *********************************************************/
 #include <string.h>      /* memcpy, memmove, memset */
 #include "mem.h"         /* low level memory routines */
-#define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */
-#include "xxhash.h"      /* XXH64_* */
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
@@ -57,24 +54,15 @@
 #endif
 
 
-/*-*******************************************************
-*  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 : 4324)        /* disable: C4324: padded structure */
-#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#if defined(_MSC_VER)
+#  include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#  define ZSTD_PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
+#elif defined(__GNUC__)
+#  define ZSTD_PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
 #else
-#  ifdef __GNUC__
-#    define FORCE_INLINE static inline __attribute__((always_inline))
-#  else
-#    define FORCE_INLINE static inline
-#  endif
+#  define ZSTD_PREFETCH(ptr)   /* disabled */
 #endif
 
-
 /*-*************************************
 *  Macros
 ***************************************/
@@ -97,19 +85,27 @@ typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
                ZSTDds_decompressLastBlock, ZSTDds_checkChecksum,
                ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTD_dStage;
 
-struct ZSTD_DCtx_s
-{
+typedef struct {
     FSE_DTable LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
-    FSE_DTable OffTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
+    FSE_DTable OFTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
     FSE_DTable MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
     HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
-    const void* previousDstEnd;
-    const void* base;
-    const void* vBase;
-    const void* dictEnd;
-    size_t expected;
     U32 rep[ZSTD_REP_NUM];
-    ZSTD_frameParams fParams;
+} ZSTD_entropyTables_t;
+
+struct ZSTD_DCtx_s
+{
+    const FSE_DTable* LLTptr;
+    const FSE_DTable* MLTptr;
+    const FSE_DTable* OFTptr;
+    const HUF_DTable* HUFptr;
+    ZSTD_entropyTables_t entropy;
+    const void* previousDstEnd;   /* detect continuity */
+    const void* base;             /* start of current segment */
+    const void* vBase;            /* virtual start of previous segment if it was just before current one */
+    const void* dictEnd;          /* end of previous segment */
+    size_t expected;
+    ZSTD_frameHeader fParams;
     blockType_e bType;   /* used in ZSTD_decompressContinue(), to transfer blockType between header decoding and block decoding stages */
     ZSTD_dStage stage;
     U32 litEntropy;
@@ -119,30 +115,33 @@ struct ZSTD_DCtx_s
     U32 dictID;
     const BYTE* litPtr;
     ZSTD_customMem customMem;
-    size_t litBufSize;
     size_t litSize;
     size_t rleSize;
     BYTE litBuffer[ZSTD_BLOCKSIZE_ABSOLUTEMAX + WILDCOPY_OVERLENGTH];
     BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
 };  /* typedef'd to ZSTD_DCtx within "zstd.h" */
 
-size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx) { return sizeof(*dctx); }
+size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx) { return (dctx==NULL) ? 0 : sizeof(ZSTD_DCtx); }
 
 size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 
 size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
 {
-    dctx->expected = ZSTD_frameHeaderSize_min;
+    dctx->expected = ZSTD_frameHeaderSize_prefix;
     dctx->stage = ZSTDds_getFrameHeaderSize;
     dctx->previousDstEnd = NULL;
     dctx->base = NULL;
     dctx->vBase = NULL;
     dctx->dictEnd = NULL;
-    dctx->hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);
+    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
     dctx->litEntropy = dctx->fseEntropy = 0;
     dctx->dictID = 0;
-    MEM_STATIC_ASSERT(sizeof(dctx->rep)==sizeof(repStartValue));
-    memcpy(dctx->rep, repStartValue, sizeof(repStartValue));
+    MEM_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
+    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    dctx->LLTptr = dctx->entropy.LLTable;
+    dctx->MLTptr = dctx->entropy.MLTable;
+    dctx->OFTptr = dctx->entropy.OFTable;
+    dctx->HUFptr = dctx->entropy.hufTable;
     return 0;
 }
 
@@ -153,7 +152,7 @@ ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
     if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
     if (!customMem.customAlloc || !customMem.customFree) return NULL;
 
-    dctx = (ZSTD_DCtx*) ZSTD_malloc(sizeof(ZSTD_DCtx), customMem);
+    dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(ZSTD_DCtx), customMem);
     if (!dctx) return NULL;
     memcpy(&dctx->customMem, &customMem, sizeof(customMem));
     ZSTD_decompressBegin(dctx);
@@ -178,45 +177,65 @@ void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
     memcpy(dstDCtx, srcDCtx, sizeof(ZSTD_DCtx) - workSpaceSize);  /* no need to copy workspace */
 }
 
+static void ZSTD_refDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict);
+
 
 /*-*************************************************************
 *   Decompression section
 ***************************************************************/
 
-/* See compression format details in : zstd_compression_format.md */
+/*! ZSTD_isFrame() :
+ *  Tells if the content of `buffer` starts with a valid Frame Identifier.
+ *  Note : Frame Identifier is 4 bytes. If `size < 4`, @return will always be 0.
+ *  Note 2 : Legacy Frame Identifiers are considered valid only if Legacy Support is enabled.
+ *  Note 3 : Skippable Frame Identifiers are considered valid. */
+unsigned ZSTD_isFrame(const void* buffer, size_t size)
+{
+    if (size < 4) return 0;
+    {   U32 const magic = MEM_readLE32(buffer);
+        if (magic == ZSTD_MAGICNUMBER) return 1;
+        if ((magic & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+    }
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(buffer, size)) return 1;
+#endif
+    return 0;
+}
+
 
 /** ZSTD_frameHeaderSize() :
-*   srcSize must be >= ZSTD_frameHeaderSize_min.
+*   srcSize must be >= ZSTD_frameHeaderSize_prefix.
 *   @return : size of the Frame Header */
 static size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
 {
-    if (srcSize < ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);
+    if (srcSize < ZSTD_frameHeaderSize_prefix) return ERROR(srcSize_wrong);
     {   BYTE const fhd = ((const BYTE*)src)[4];
         U32 const dictID= fhd & 3;
         U32 const singleSegment = (fhd >> 5) & 1;
         U32 const fcsId = fhd >> 6;
-        return ZSTD_frameHeaderSize_min + !singleSegment + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
+        return ZSTD_frameHeaderSize_prefix + !singleSegment + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
                 + (singleSegment && !fcsId);
     }
 }
 
 
-/** ZSTD_getFrameParams() :
+/** ZSTD_getFrameHeader() :
 *   decode Frame Header, or require larger `srcSize`.
-*   @return : 0, `fparamsPtr` is correctly filled,
+*   @return : 0, `zfhPtr` is correctly filled,
 *            >0, `srcSize` is too small, result is expected `srcSize`,
 *             or an error code, which can be tested using ZSTD_isError() */
-size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t srcSize)
+size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
 {
     const BYTE* ip = (const BYTE*)src;
+    if (srcSize < ZSTD_frameHeaderSize_prefix) return ZSTD_frameHeaderSize_prefix;
 
-    if (srcSize < ZSTD_frameHeaderSize_min) return ZSTD_frameHeaderSize_min;
     if (MEM_readLE32(src) != ZSTD_MAGICNUMBER) {
         if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+            /* skippable frame */
             if (srcSize < ZSTD_skippableHeaderSize) return ZSTD_skippableHeaderSize; /* magic number + skippable frame length */
-            memset(fparamsPtr, 0, sizeof(*fparamsPtr));
-            fparamsPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
-            fparamsPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
+            memset(zfhPtr, 0, sizeof(*zfhPtr));
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
+            zfhPtr->windowSize = 0; /* windowSize==0 means a frame is skippable */
             return 0;
         }
         return ERROR(prefix_unknown);
@@ -240,7 +259,7 @@ size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t
         if (!singleSegment) {
             BYTE const wlByte = ip[pos++];
             U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
-            if (windowLog > ZSTD_WINDOWLOG_MAX) return ERROR(frameParameter_unsupported);
+            if (windowLog > ZSTD_WINDOWLOG_MAX) return ERROR(frameParameter_windowTooLarge);  /* avoids issue with 1 << windowLog */
             windowSize = (1U << windowLog);
             windowSize += (windowSize >> 3) * (wlByte&7);
         }
@@ -262,15 +281,94 @@ size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t
             case 3 : frameContentSize = MEM_readLE64(ip+pos); break;
         }
         if (!windowSize) windowSize = (U32)frameContentSize;
-        if (windowSize > windowSizeMax) return ERROR(frameParameter_unsupported);
-        fparamsPtr->frameContentSize = frameContentSize;
-        fparamsPtr->windowSize = windowSize;
-        fparamsPtr->dictID = dictID;
-        fparamsPtr->checksumFlag = checksumFlag;
+        if (windowSize > windowSizeMax) return ERROR(frameParameter_windowTooLarge);
+        zfhPtr->frameContentSize = frameContentSize;
+        zfhPtr->windowSize = windowSize;
+        zfhPtr->dictID = dictID;
+        zfhPtr->checksumFlag = checksumFlag;
     }
     return 0;
 }
 
+/** ZSTD_getFrameContentSize() :
+*   compatible with legacy mode
+*   @return : decompressed size of the single frame pointed to be `src` if known, otherwise
+*             - ZSTD_CONTENTSIZE_UNKNOWN if the size cannot be determined
+*             - ZSTD_CONTENTSIZE_ERROR if an error occurred (e.g. invalid magic number, srcSize too small) */
+unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize)) {
+        unsigned long long const ret = ZSTD_getDecompressedSize_legacy(src, srcSize);
+        return ret == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : ret;
+    }
+#endif
+    {   ZSTD_frameHeader fParams;
+        if (ZSTD_getFrameHeader(&fParams, src, srcSize) != 0) return ZSTD_CONTENTSIZE_ERROR;
+        if (fParams.windowSize == 0) {
+            /* Either skippable or empty frame, size == 0 either way */
+            return 0;
+        } else if (fParams.frameContentSize != 0) {
+            return fParams.frameContentSize;
+        } else {
+            return ZSTD_CONTENTSIZE_UNKNOWN;
+        }
+    }
+}
+
+/** ZSTD_findDecompressedSize() :
+ *  compatible with legacy mode
+ *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
+ *      skippable frames
+ *  @return : decompressed size of the frames contained */
+unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
+{
+    {
+        unsigned long long totalDstSize = 0;
+        while (srcSize >= ZSTD_frameHeaderSize_prefix) {
+            const U32 magicNumber = MEM_readLE32(src);
+
+            if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+                size_t skippableSize;
+                if (srcSize < ZSTD_skippableHeaderSize)
+                    return ERROR(srcSize_wrong);
+                skippableSize = MEM_readLE32((const BYTE *)src + 4) +
+                                ZSTD_skippableHeaderSize;
+                if (srcSize < skippableSize) {
+                    return ZSTD_CONTENTSIZE_ERROR;
+                }
+
+                src = (const BYTE *)src + skippableSize;
+                srcSize -= skippableSize;
+                continue;
+            }
+
+            {
+                unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+                if (ret >= ZSTD_CONTENTSIZE_ERROR) return ret;
+
+                /* check for overflow */
+                if (totalDstSize + ret < totalDstSize) return ZSTD_CONTENTSIZE_ERROR;
+                totalDstSize += ret;
+            }
+            {
+                size_t const frameSrcSize = ZSTD_findFrameCompressedSize(src, srcSize);
+                if (ZSTD_isError(frameSrcSize)) {
+                    return ZSTD_CONTENTSIZE_ERROR;
+                }
+
+                src = (const BYTE *)src + frameSrcSize;
+                srcSize -= frameSrcSize;
+            }
+        }
+
+        if (srcSize) {
+            return ZSTD_CONTENTSIZE_ERROR;
+        }
+
+        return totalDstSize;
+    }
+}
 
 /** ZSTD_getDecompressedSize() :
 *   compatible with legacy mode
@@ -281,26 +379,22 @@ size_t ZSTD_getFrameParams(ZSTD_frameParams* fparamsPtr, const void* src, size_t
                    - frame header not complete (`srcSize` too small) */
 unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT==1)
-    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_getDecompressedSize_legacy(src, srcSize);
-#endif
-    {   ZSTD_frameParams fparams;
-        size_t const frResult = ZSTD_getFrameParams(&fparams, src, srcSize);
-        if (frResult!=0) return 0;
-        return fparams.frameContentSize;
-    }
+    unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
+    return ret >= ZSTD_CONTENTSIZE_ERROR ? 0 : ret;
 }
 
 
 /** ZSTD_decodeFrameHeader() :
-*   `srcSize` must be the size provided by ZSTD_frameHeaderSize().
+*   `headerSize` must be the size provided by ZSTD_frameHeaderSize().
 *   @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
-static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t srcSize)
+static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
 {
-    size_t const result = ZSTD_getFrameParams(&(dctx->fParams), src, srcSize);
+    size_t const result = ZSTD_getFrameHeader(&(dctx->fParams), src, headerSize);
+    if (ZSTD_isError(result)) return result;  /* invalid header */
+    if (result>0) return ERROR(srcSize_wrong);   /* headerSize too small */
     if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID)) return ERROR(dictionary_wrong);
     if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
-    return result;
+    return 0;
 }
 
 
@@ -313,7 +407,8 @@ typedef struct
 
 /*! ZSTD_getcBlockSize() :
 *   Provides the size of compressed block from block header `src` */
-size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bpPtr)
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr)
 {
     if (srcSize < ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
     {   U32 const cBlockHeader = MEM_readLE24(src);
@@ -328,7 +423,8 @@ size_t ZSTD_getcBlockSize(const void* src, size_t srcSize, blockProperties_t* bp
 }
 
 
-static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
+                          const void* src, size_t srcSize)
 {
     if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
     memcpy(dst, src, srcSize);
@@ -336,7 +432,9 @@ static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity, const void* src,
 }
 
 
-static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity, const void* src, size_t srcSize, size_t regenSize)
+static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                               size_t regenSize)
 {
     if (srcSize != 1) return ERROR(srcSize_wrong);
     if (regenSize > dstCapacity) return ERROR(dstSize_tooSmall);
@@ -369,43 +467,41 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                 {
                 case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
                     /* 2 - 2 - 10 - 10 */
-                    {   singleStream = !lhlCode;
-                        lhSize = 3;
-                        litSize  = (lhc >> 4) & 0x3FF;
-                        litCSize = (lhc >> 14) & 0x3FF;
-                        break;
-                    }
+                    singleStream = !lhlCode;
+                    lhSize = 3;
+                    litSize  = (lhc >> 4) & 0x3FF;
+                    litCSize = (lhc >> 14) & 0x3FF;
+                    break;
                 case 2:
                     /* 2 - 2 - 14 - 14 */
-                    {   lhSize = 4;
-                        litSize  = (lhc >> 4) & 0x3FFF;
-                        litCSize = lhc >> 18;
-                        break;
-                    }
+                    lhSize = 4;
+                    litSize  = (lhc >> 4) & 0x3FFF;
+                    litCSize = lhc >> 18;
+                    break;
                 case 3:
                     /* 2 - 2 - 18 - 18 */
-                    {   lhSize = 5;
-                        litSize  = (lhc >> 4) & 0x3FFFF;
-                        litCSize = (lhc >> 22) + (istart[4] << 10);
-                        break;
-                    }
+                    lhSize = 5;
+                    litSize  = (lhc >> 4) & 0x3FFFF;
+                    litCSize = (lhc >> 22) + (istart[4] << 10);
+                    break;
                 }
                 if (litSize > ZSTD_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
                 if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
 
                 if (HUF_isError((litEncType==set_repeat) ?
                                     ( singleStream ?
-                                        HUF_decompress1X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTable) :
-                                        HUF_decompress4X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->hufTable) ) :
+                                        HUF_decompress1X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr) :
+                                        HUF_decompress4X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr) ) :
                                     ( singleStream ?
-                                        HUF_decompress1X2_DCtx(dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) :
-                                        HUF_decompress4X_hufOnly (dctx->hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize)) ))
+                                        HUF_decompress1X2_DCtx(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize) :
+                                        HUF_decompress4X_hufOnly (dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize)) ))
                     return ERROR(corruption_detected);
 
                 dctx->litPtr = dctx->litBuffer;
-                dctx->litBufSize = ZSTD_BLOCKSIZE_ABSOLUTEMAX+WILDCOPY_OVERLENGTH;
                 dctx->litSize = litSize;
                 dctx->litEntropy = 1;
+                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                 return litCSize + lhSize;
             }
 
@@ -432,13 +528,12 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
                     memcpy(dctx->litBuffer, istart+lhSize, litSize);
                     dctx->litPtr = dctx->litBuffer;
-                    dctx->litBufSize = ZSTD_BLOCKSIZE_ABSOLUTEMAX+8;
                     dctx->litSize = litSize;
+                    memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
                     return lhSize+litSize;
                 }
                 /* direct reference into compressed stream */
                 dctx->litPtr = istart+lhSize;
-                dctx->litBufSize = srcSize-lhSize;
                 dctx->litSize = litSize;
                 return lhSize+litSize;
             }
@@ -463,37 +558,109 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     break;
                 }
                 if (litSize > ZSTD_BLOCKSIZE_ABSOLUTEMAX) return ERROR(corruption_detected);
-                memset(dctx->litBuffer, istart[lhSize], litSize);
+                memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
                 dctx->litPtr = dctx->litBuffer;
-                dctx->litBufSize = ZSTD_BLOCKSIZE_ABSOLUTEMAX+WILDCOPY_OVERLENGTH;
                 dctx->litSize = litSize;
                 return lhSize+1;
             }
         default:
             return ERROR(corruption_detected);   /* impossible */
         }
-
     }
 }
 
 
+typedef union {
+    FSE_decode_t realData;
+    U32 alignedBy4;
+} FSE_decode_t4;
+
+/* Default FSE distribution table for Literal Lengths */
+static const FSE_decode_t4 LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
+    { { LL_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
+     /* base, symbol, bits */
+    { {  0,  0,  4 } }, { { 16,  0,  4 } }, { { 32,  1,  5 } }, { {  0,  3,  5 } },
+    { {  0,  4,  5 } }, { {  0,  6,  5 } }, { {  0,  7,  5 } }, { {  0,  9,  5 } },
+    { {  0, 10,  5 } }, { {  0, 12,  5 } }, { {  0, 14,  6 } }, { {  0, 16,  5 } },
+    { {  0, 18,  5 } }, { {  0, 19,  5 } }, { {  0, 21,  5 } }, { {  0, 22,  5 } },
+    { {  0, 24,  5 } }, { { 32, 25,  5 } }, { {  0, 26,  5 } }, { {  0, 27,  6 } },
+    { {  0, 29,  6 } }, { {  0, 31,  6 } }, { { 32,  0,  4 } }, { {  0,  1,  4 } },
+    { {  0,  2,  5 } }, { { 32,  4,  5 } }, { {  0,  5,  5 } }, { { 32,  7,  5 } },
+    { {  0,  8,  5 } }, { { 32, 10,  5 } }, { {  0, 11,  5 } }, { {  0, 13,  6 } },
+    { { 32, 16,  5 } }, { {  0, 17,  5 } }, { { 32, 19,  5 } }, { {  0, 20,  5 } },
+    { { 32, 22,  5 } }, { {  0, 23,  5 } }, { {  0, 25,  4 } }, { { 16, 25,  4 } },
+    { { 32, 26,  5 } }, { {  0, 28,  6 } }, { {  0, 30,  6 } }, { { 48,  0,  4 } },
+    { { 16,  1,  4 } }, { { 32,  2,  5 } }, { { 32,  3,  5 } }, { { 32,  5,  5 } },
+    { { 32,  6,  5 } }, { { 32,  8,  5 } }, { { 32,  9,  5 } }, { { 32, 11,  5 } },
+    { { 32, 12,  5 } }, { {  0, 15,  6 } }, { { 32, 17,  5 } }, { { 32, 18,  5 } },
+    { { 32, 20,  5 } }, { { 32, 21,  5 } }, { { 32, 23,  5 } }, { { 32, 24,  5 } },
+    { {  0, 35,  6 } }, { {  0, 34,  6 } }, { {  0, 33,  6 } }, { {  0, 32,  6 } },
+};   /* LL_defaultDTable */
+
+/* Default FSE distribution table for Match Lengths */
+static const FSE_decode_t4 ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
+    { { ML_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
+    /* base, symbol, bits */
+    { {  0,  0,  6 } }, { {  0,  1,  4 } }, { { 32,  2,  5 } }, { {  0,  3,  5 } },
+    { {  0,  5,  5 } }, { {  0,  6,  5 } }, { {  0,  8,  5 } }, { {  0, 10,  6 } },
+    { {  0, 13,  6 } }, { {  0, 16,  6 } }, { {  0, 19,  6 } }, { {  0, 22,  6 } },
+    { {  0, 25,  6 } }, { {  0, 28,  6 } }, { {  0, 31,  6 } }, { {  0, 33,  6 } },
+    { {  0, 35,  6 } }, { {  0, 37,  6 } }, { {  0, 39,  6 } }, { {  0, 41,  6 } },
+    { {  0, 43,  6 } }, { {  0, 45,  6 } }, { { 16,  1,  4 } }, { {  0,  2,  4 } },
+    { { 32,  3,  5 } }, { {  0,  4,  5 } }, { { 32,  6,  5 } }, { {  0,  7,  5 } },
+    { {  0,  9,  6 } }, { {  0, 12,  6 } }, { {  0, 15,  6 } }, { {  0, 18,  6 } },
+    { {  0, 21,  6 } }, { {  0, 24,  6 } }, { {  0, 27,  6 } }, { {  0, 30,  6 } },
+    { {  0, 32,  6 } }, { {  0, 34,  6 } }, { {  0, 36,  6 } }, { {  0, 38,  6 } },
+    { {  0, 40,  6 } }, { {  0, 42,  6 } }, { {  0, 44,  6 } }, { { 32,  1,  4 } },
+    { { 48,  1,  4 } }, { { 16,  2,  4 } }, { { 32,  4,  5 } }, { { 32,  5,  5 } },
+    { { 32,  7,  5 } }, { { 32,  8,  5 } }, { {  0, 11,  6 } }, { {  0, 14,  6 } },
+    { {  0, 17,  6 } }, { {  0, 20,  6 } }, { {  0, 23,  6 } }, { {  0, 26,  6 } },
+    { {  0, 29,  6 } }, { {  0, 52,  6 } }, { {  0, 51,  6 } }, { {  0, 50,  6 } },
+    { {  0, 49,  6 } }, { {  0, 48,  6 } }, { {  0, 47,  6 } }, { {  0, 46,  6 } },
+};   /* ML_defaultDTable */
+
+/* Default FSE distribution table for Offset Codes */
+static const FSE_decode_t4 OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
+    { { OF_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
+    /* base, symbol, bits */
+    { {  0,  0,  5 } }, { {  0,  6,  4 } },
+    { {  0,  9,  5 } }, { {  0, 15,  5 } },
+    { {  0, 21,  5 } }, { {  0,  3,  5 } },
+    { {  0,  7,  4 } }, { {  0, 12,  5 } },
+    { {  0, 18,  5 } }, { {  0, 23,  5 } },
+    { {  0,  5,  5 } }, { {  0,  8,  4 } },
+    { {  0, 14,  5 } }, { {  0, 20,  5 } },
+    { {  0,  2,  5 } }, { { 16,  7,  4 } },
+    { {  0, 11,  5 } }, { {  0, 17,  5 } },
+    { {  0, 22,  5 } }, { {  0,  4,  5 } },
+    { { 16,  8,  4 } }, { {  0, 13,  5 } },
+    { {  0, 19,  5 } }, { {  0,  1,  5 } },
+    { { 16,  6,  4 } }, { {  0, 10,  5 } },
+    { {  0, 16,  5 } }, { {  0, 28,  5 } },
+    { {  0, 27,  5 } }, { {  0, 26,  5 } },
+    { {  0, 25,  5 } }, { {  0, 24,  5 } },
+};   /* OF_defaultDTable */
+
 /*! ZSTD_buildSeqTable() :
     @return : nb bytes read from src,
               or an error code if it fails, testable with ZSTD_isError()
 */
-FORCE_INLINE size_t ZSTD_buildSeqTable(FSE_DTable* DTable, symbolEncodingType_e type, U32 max, U32 maxLog,
+static size_t ZSTD_buildSeqTable(FSE_DTable* DTableSpace, const FSE_DTable** DTablePtr,
+                                 symbolEncodingType_e type, U32 max, U32 maxLog,
                                  const void* src, size_t srcSize,
-                                 const S16* defaultNorm, U32 defaultLog, U32 flagRepeatTable)
+                                 const FSE_decode_t4* defaultTable, U32 flagRepeatTable)
 {
+    const void* const tmpPtr = defaultTable;   /* bypass strict aliasing */
     switch(type)
     {
     case set_rle :
         if (!srcSize) return ERROR(srcSize_wrong);
         if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
-        FSE_buildDTable_rle(DTable, *(const BYTE*)src);   /* if *src > max, data is corrupted */
+        FSE_buildDTable_rle(DTableSpace, *(const BYTE*)src);
+        *DTablePtr = DTableSpace;
         return 1;
     case set_basic :
-        FSE_buildDTable(DTable, defaultNorm, max, defaultLog);
+        *DTablePtr = (const FSE_DTable*)tmpPtr;
         return 0;
     case set_repeat:
         if (!flagRepeatTable) return ERROR(corruption_detected);
@@ -505,14 +672,13 @@ FORCE_INLINE size_t ZSTD_buildSeqTable(FSE_DTable* DTable, symbolEncodingType_e
             size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
             if (FSE_isError(headerSize)) return ERROR(corruption_detected);
             if (tableLog > maxLog) return ERROR(corruption_detected);
-            FSE_buildDTable(DTable, norm, max, tableLog);
+            FSE_buildDTable(DTableSpace, norm, max, tableLog);
+            *DTablePtr = DTableSpace;
             return headerSize;
     }   }
 }
 
-
-size_t ZSTD_decodeSeqHeaders(int* nbSeqPtr,
-                             FSE_DTable* DTableLL, FSE_DTable* DTableML, FSE_DTable* DTableOffb, U32 flagRepeatTable,
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
                              const void* src, size_t srcSize)
 {
     const BYTE* const istart = (const BYTE* const)src;
@@ -526,10 +692,13 @@ size_t ZSTD_decodeSeqHeaders(int* nbSeqPtr,
     {   int nbSeq = *ip++;
         if (!nbSeq) { *nbSeqPtr=0; return 1; }
         if (nbSeq > 0x7F) {
-            if (nbSeq == 0xFF)
+            if (nbSeq == 0xFF) {
+                if (ip+2 > iend) return ERROR(srcSize_wrong);
                 nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
-            else
+            } else {
+                if (ip >= iend) return ERROR(srcSize_wrong);
                 nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+            }
         }
         *nbSeqPtr = nbSeq;
     }
@@ -542,18 +711,25 @@ size_t ZSTD_decodeSeqHeaders(int* nbSeqPtr,
         ip++;
 
         /* Build DTables */
-        {   size_t const llhSize = ZSTD_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
+        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
+                                                      LLtype, MaxLL, LLFSELog,
+                                                      ip, iend-ip, LL_defaultDTable, dctx->fseEntropy);
             if (ZSTD_isError(llhSize)) return ERROR(corruption_detected);
             ip += llhSize;
         }
-        {   size_t const ofhSize = ZSTD_buildSeqTable(DTableOffb, OFtype, MaxOff, OffFSELog, ip, iend-ip, OF_defaultNorm, OF_defaultNormLog, flagRepeatTable);
+        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
+                                                      OFtype, MaxOff, OffFSELog,
+                                                      ip, iend-ip, OF_defaultDTable, dctx->fseEntropy);
             if (ZSTD_isError(ofhSize)) return ERROR(corruption_detected);
             ip += ofhSize;
         }
-        {   size_t const mlhSize = ZSTD_buildSeqTable(DTableML, MLtype, MaxML, MLFSELog, ip, iend-ip, ML_defaultNorm, ML_defaultNormLog, flagRepeatTable);
+        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
+                                                      MLtype, MaxML, MLFSELog,
+                                                      ip, iend-ip, ML_defaultDTable, dctx->fseEntropy);
             if (ZSTD_isError(mlhSize)) return ERROR(corruption_detected);
             ip += mlhSize;
-    }   }
+        }
+    }
 
     return ip-istart;
 }
@@ -563,6 +739,7 @@ typedef struct {
     size_t litLength;
     size_t matchLength;
     size_t offset;
+    const BYTE* match;
 } seq_t;
 
 typedef struct {
@@ -571,9 +748,59 @@ typedef struct {
     FSE_DState_t stateOffb;
     FSE_DState_t stateML;
     size_t prevOffset[ZSTD_REP_NUM];
+    const BYTE* base;
+    size_t pos;
+    uPtrDiff gotoDict;
 } seqState_t;
 
 
+FORCE_NOINLINE
+size_t ZSTD_execSequenceLast7(BYTE* op,
+                              BYTE* const oend, seq_t sequence,
+                              const BYTE** litPtr, const BYTE* const litLimit,
+                              const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    /* check */
+    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd <= oend_w) return ERROR(GENERIC);   /* Precondition */
+
+    /* copy literals */
+    if (op < oend_w) {
+        ZSTD_wildcopy(op, *litPtr, oend_w - op);
+        *litPtr += oend_w - op;
+        op = oend_w;
+    }
+    while (op < oLitEnd) *op++ = *(*litPtr)++;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        match = dictEnd - (base-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+    }   }
+    while (op < oMatchEnd) *op++ = *match++;
+    return sequenceLength;
+}
+
+
 static seq_t ZSTD_decodeSequence(seqState_t* seqState)
 {
     seq_t seq;
@@ -588,35 +815,41 @@ static seq_t ZSTD_decodeSequence(seqState_t* seqState)
     U32 const totalBits = llBits+mlBits+ofBits;
 
     static const U32 LL_base[MaxLL+1] = {
-                             0,  1,  2,  3,  4,  5,  6,  7,  8,  9,   10,    11,    12,    13,    14,     15,
-                            16, 18, 20, 22, 24, 28, 32, 40, 48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                             0,    1,    2,     3,     4,     5,     6,      7,
+                             8,    9,   10,    11,    12,    13,    14,     15,
+                            16,   18,   20,    22,    24,    28,    32,     40,
+                            48,   64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
                             0x2000, 0x4000, 0x8000, 0x10000 };
 
     static const U32 ML_base[MaxML+1] = {
-                             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,    33,    34,
-                            35, 37, 39, 41, 43, 47, 51, 59, 67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                             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,    33,    34,
+                            35, 37, 39,   41,    43,    47,    51,    59,
+                            67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
                             0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
 
     static const U32 OF_base[MaxOff+1] = {
-                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+                     0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                     0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                     0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                     0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
 
     /* sequence */
     {   size_t offset;
         if (!ofCode)
             offset = 0;
         else {
-            offset = OF_base[ofCode] + BIT_readBits(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+            offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
             if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
         }
 
         if (ofCode <= 1) {
             offset += (llCode==0);
             if (offset) {
-                size_t const temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
                 if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
                 seqState->prevOffset[1] = seqState->prevOffset[0];
                 seqState->prevOffset[0] = offset = temp;
@@ -631,10 +864,10 @@ static seq_t ZSTD_decodeSequence(seqState_t* seqState)
         seq.offset = offset;
     }
 
-    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBits(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
     if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
 
-    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBits(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
     if (MEM_32bits() ||
        (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);
 
@@ -650,9 +883,9 @@ static seq_t ZSTD_decodeSequence(seqState_t* seqState)
 
 FORCE_INLINE
 size_t ZSTD_execSequence(BYTE* op,
-                                BYTE* const oend, seq_t sequence,
-                                const BYTE** litPtr, const BYTE* const litLimit_w,
-                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+                         BYTE* const oend, seq_t sequence,
+                         const BYTE** litPtr, const BYTE* const litLimit,
+                         const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
 {
     BYTE* const oLitEnd = op + sequence.litLength;
     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
@@ -662,8 +895,9 @@ size_t ZSTD_execSequence(BYTE* op,
     const BYTE* match = oLitEnd - sequence.offset;
 
     /* check */
-    if ((oLitEnd>oend_w) | (oMatchEnd>oend)) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
-    if (iLitEnd > litLimit_w) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
 
     /* copy Literals */
     ZSTD_copy8(op, *litPtr);
@@ -674,9 +908,9 @@ size_t ZSTD_execSequence(BYTE* op,
 
     /* copy Match */
     if (sequence.offset > (size_t)(oLitEnd - base)) {
-        /* offset beyond prefix */
+        /* offset beyond prefix -> go into extDict */
         if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
-        match = dictEnd - (base-match);
+        match = dictEnd + (match - base);
         if (match + sequence.matchLength <= dictEnd) {
             memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
@@ -687,13 +921,19 @@ size_t ZSTD_execSequence(BYTE* op,
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
             match = base;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              U32 i;
+              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
+              return sequenceLength;
+            }
     }   }
+    /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */
 
     /* match within prefix */
     if (sequence.offset < 8) {
         /* close range match, overlap */
         static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* substracted */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
         int const sub2 = dec64table[sequence.offset];
         op[0] = match[0];
         op[1] = match[1];
@@ -715,7 +955,7 @@ size_t ZSTD_execSequence(BYTE* op,
         }
         while (op < oMatchEnd) *op++ = *match++;
     } else {
-        ZSTD_wildcopy(op, match, sequence.matchLength-8);   /* works even if matchLength < 8 */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
     }
     return sequenceLength;
 }
@@ -732,18 +972,14 @@ static size_t ZSTD_decompressSequences(
     BYTE* const oend = ostart + maxDstSize;
     BYTE* op = ostart;
     const BYTE* litPtr = dctx->litPtr;
-    const BYTE* const litLimit_w = litPtr + dctx->litBufSize - WILDCOPY_OVERLENGTH;
     const BYTE* const litEnd = litPtr + dctx->litSize;
-    FSE_DTable* DTableLL = dctx->LLTable;
-    FSE_DTable* DTableML = dctx->MLTable;
-    FSE_DTable* DTableOffb = dctx->OffTable;
     const BYTE* const base = (const BYTE*) (dctx->base);
     const BYTE* const vBase = (const BYTE*) (dctx->vBase);
     const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
     int nbSeq;
 
     /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(&nbSeq, DTableLL, DTableML, DTableOffb, dctx->fseEntropy, ip, seqSize);
+    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
         if (ZSTD_isError(seqHSize)) return seqHSize;
         ip += seqHSize;
     }
@@ -752,17 +988,16 @@ static size_t ZSTD_decompressSequences(
     if (nbSeq) {
         seqState_t seqState;
         dctx->fseEntropy = 1;
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->rep[i]; }
-        { size_t const errorCode = BIT_initDStream(&(seqState.DStream), ip, iend-ip);
-          if (ERR_isError(errorCode)) return ERROR(corruption_detected); }
-        FSE_initDState(&(seqState.stateLL), &(seqState.DStream), DTableLL);
-        FSE_initDState(&(seqState.stateOffb), &(seqState.DStream), DTableOffb);
-        FSE_initDState(&(seqState.stateML), &(seqState.DStream), DTableML);
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
+        FSE_initDState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        FSE_initDState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        FSE_initDState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
 
         for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) {
             nbSeq--;
             {   seq_t const sequence = ZSTD_decodeSequence(&seqState);
-                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litLimit_w, base, vBase, dictEnd);
+                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
                 if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
                 op += oneSeqSize;
         }   }
@@ -770,7 +1005,7 @@ static size_t ZSTD_decompressSequences(
         /* check if reached exact end */
         if (nbSeq) return ERROR(corruption_detected);
         /* save reps for next block */
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->rep[i] = (U32)(seqState.prevOffset[i]); }
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
     }
 
     /* last literal segment */
@@ -784,14 +1019,268 @@ static size_t ZSTD_decompressSequences(
 }
 
 
-static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+FORCE_INLINE seq_t ZSTD_decodeSequenceLong_generic(seqState_t* seqState, int const longOffsets)
 {
-    if (dst != dctx->previousDstEnd) {   /* not contiguous */
-        dctx->dictEnd = dctx->previousDstEnd;
-        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
-        dctx->base = dst;
-        dctx->previousDstEnd = dst;
+    seq_t seq;
+
+    U32 const llCode = FSE_peekSymbol(&seqState->stateLL);
+    U32 const mlCode = FSE_peekSymbol(&seqState->stateML);
+    U32 const ofCode = FSE_peekSymbol(&seqState->stateOffb);   /* <= maxOff, by table construction */
+
+    U32 const llBits = LL_bits[llCode];
+    U32 const mlBits = ML_bits[mlCode];
+    U32 const ofBits = ofCode;
+    U32 const totalBits = llBits+mlBits+ofBits;
+
+    static const U32 LL_base[MaxLL+1] = {
+                             0,  1,    2,     3,     4,     5,     6,      7,
+                             8,  9,   10,    11,    12,    13,    14,     15,
+                            16, 18,   20,    22,    24,    28,    32,     40,
+                            48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                            0x2000, 0x4000, 0x8000, 0x10000 };
+
+    static const U32 ML_base[MaxML+1] = {
+                             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,    33,    34,
+                            35, 37, 39,   41,    43,    47,    51,    59,
+                            67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+    static const U32 OF_base[MaxOff+1] = {
+                     0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                     0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                     0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                     0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+
+    /* sequence */
+    {   size_t offset;
+        if (!ofCode)
+            offset = 0;
+        else {
+            if (longOffsets) {
+                int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN);
+                offset = OF_base[ofCode] + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream);
+                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+            } else {
+                offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            }
+        }
+
+        if (ofCode <= 1) {
+            offset += (llCode==0);
+            if (offset) {
+                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                seqState->prevOffset[1] = seqState->prevOffset[0];
+                seqState->prevOffset[0] = offset = temp;
+            } else {
+                offset = seqState->prevOffset[0];
+            }
+        } else {
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
+    if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
+
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
+    if (MEM_32bits() ||
+       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);
+
+    {   size_t const pos = seqState->pos + seq.litLength;
+        seq.match = seqState->base + pos - seq.offset;    /* single memory segment */
+        if (seq.offset > pos) seq.match += seqState->gotoDict;   /* separate memory segment */
+        seqState->pos = pos + seq.matchLength;
+    }
+
+    /* ANS state update */
+    FSE_updateState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
+    FSE_updateState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
+    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+    FSE_updateState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
+
+    return seq;
+}
+
+static seq_t ZSTD_decodeSequenceLong(seqState_t* seqState, unsigned const windowSize) {
+    if (ZSTD_highbit32(windowSize) > STREAM_ACCUMULATOR_MIN) {
+        return ZSTD_decodeSequenceLong_generic(seqState, 1);
+    } else {
+        return ZSTD_decodeSequenceLong_generic(seqState, 0);
+    }
+}
+
+FORCE_INLINE
+size_t ZSTD_execSequenceLong(BYTE* op,
+                                BYTE* const oend, seq_t sequence,
+                                const BYTE** litPtr, const BYTE* const litLimit,
+                                const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = sequence.match;
+
+    /* check */
+#if 1
+    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
+    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
+    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
+#endif
+
+    /* copy Literals */
+    ZSTD_copy8(op, *litPtr);
+    if (sequence.litLength > 8)
+        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* copy Match */
+#if 1
+    if (sequence.offset > (size_t)(oLitEnd - base)) {
+        /* offset beyond prefix */
+        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = base;
+            if (op > oend_w || sequence.matchLength < MINMATCH) {
+              U32 i;
+              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
+              return sequenceLength;
+            }
+    }   }
+    /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */
+#endif
+
+    /* match within prefix */
+    if (sequence.offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[sequence.offset];
+        op[0] = match[0];
+        op[1] = match[1];
+        op[2] = match[2];
+        op[3] = match[3];
+        match += dec32table[sequence.offset];
+        ZSTD_copy4(op+4, match);
+        match -= sub2;
+    } else {
+        ZSTD_copy8(op, match);
+    }
+    op += 8; match += 8;
+
+    if (oMatchEnd > oend-(16-MINMATCH)) {
+        if (op < oend_w) {
+            ZSTD_wildcopy(op, match, oend_w - op);
+            match += oend_w - op;
+            op = oend_w;
+        }
+        while (op < oMatchEnd) *op++ = *match++;
+    } else {
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+    }
+    return sequenceLength;
+}
+
+static size_t ZSTD_decompressSequencesLong(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const base = (const BYTE*) (dctx->base);
+    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    unsigned const windowSize = dctx->fParams.windowSize;
+    int nbSeq;
+
+    /* Build Decoding Tables */
+    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+    }
+
+    /* Regen sequences */
+    if (nbSeq) {
+#define STORED_SEQS 4
+#define STOSEQ_MASK (STORED_SEQS-1)
+#define ADVANCED_SEQS 4
+        seq_t sequences[STORED_SEQS];
+        int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
+        seqState_t seqState;
+        int seqNb;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        seqState.base = base;
+        seqState.pos = (size_t)(op-base);
+        seqState.gotoDict = (uPtrDiff)dictEnd - (uPtrDiff)base; /* cast to avoid undefined behaviour */
+        CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
+        FSE_initDState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        FSE_initDState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        FSE_initDState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+
+        /* prepare in advance */
+        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && seqNb<seqAdvance; seqNb++) {
+            sequences[seqNb] = ZSTD_decodeSequenceLong(&seqState, windowSize);
+        }
+        if (seqNb<seqAdvance) return ERROR(corruption_detected);
+
+        /* decode and decompress */
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && seqNb<nbSeq ; seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequenceLong(&seqState, windowSize);
+            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STOSEQ_MASK], &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            ZSTD_PREFETCH(sequence.match);
+            sequences[seqNb&STOSEQ_MASK] = sequence;
+            op += oneSeqSize;
+        }
+        if (seqNb<nbSeq) return ERROR(corruption_detected);
+
+        /* finish queue */
+        seqNb -= seqAdvance;
+        for ( ; seqNb<nbSeq ; seqNb++) {
+            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STOSEQ_MASK], &litPtr, litEnd, base, vBase, dictEnd);
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
+        memcpy(op, litPtr, lastLLSize);
+        op += lastLLSize;
     }
+
+    return op-ostart;
 }
 
 
@@ -803,16 +1292,32 @@ static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
 
     if (srcSize >= ZSTD_BLOCKSIZE_ABSOLUTEMAX) return ERROR(srcSize_wrong);
 
-    /* Decode literals sub-block */
+    /* Decode literals section */
     {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
         if (ZSTD_isError(litCSize)) return litCSize;
         ip += litCSize;
         srcSize -= litCSize;
     }
+    if (sizeof(size_t) > 4)  /* do not enable prefetching on 32-bits x86, as it's performance detrimental */
+                             /* likely because of register pressure */
+                             /* if that's the correct cause, then 32-bits ARM should be affected differently */
+                             /* it would be good to test this on ARM real hardware, to see if prefetch version improves speed */
+        if (dctx->fParams.windowSize > (1<<23))
+            return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize);
     return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize);
 }
 
 
+static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
+        dctx->base = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
 size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
                             void* dst, size_t dstCapacity,
                       const void* src, size_t srcSize)
@@ -842,29 +1347,81 @@ size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t len
     return length;
 }
 
+/** ZSTD_findFrameCompressedSize() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the compressed size of the frame starting at `src` */
+size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+{
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+#endif
+    if (srcSize >= ZSTD_skippableHeaderSize &&
+            (MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + 4);
+    } else {
+        const BYTE* ip = (const BYTE*)src;
+        const BYTE* const ipstart = ip;
+        size_t remainingSize = srcSize;
+        ZSTD_frameHeader fParams;
+
+        size_t const headerSize = ZSTD_frameHeaderSize(ip, remainingSize);
+        if (ZSTD_isError(headerSize)) return headerSize;
+
+        /* Frame Header */
+        {   size_t const ret = ZSTD_getFrameHeader(&fParams, ip, remainingSize);
+            if (ZSTD_isError(ret)) return ret;
+            if (ret > 0) return ERROR(srcSize_wrong);
+        }
+
+        ip += headerSize;
+        remainingSize -= headerSize;
+
+        /* Loop on each block */
+        while (1) {
+            blockProperties_t blockProperties;
+            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+            if (ZSTD_isError(cBlockSize)) return cBlockSize;
+
+            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+
+            ip += ZSTD_blockHeaderSize + cBlockSize;
+            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
+
+            if (blockProperties.lastBlock) break;
+        }
+
+        if (fParams.checksumFlag) {   /* Frame content checksum */
+            if (remainingSize < 4) return ERROR(srcSize_wrong);
+            ip += 4;
+            remainingSize -= 4;
+        }
+
+        return ip - ipstart;
+    }
+}
 
 /*! ZSTD_decompressFrame() :
-*   `dctx` must be properly initialized */
+*   @dctx must be properly initialized */
 static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
                                  void* dst, size_t dstCapacity,
-                                 const void* src, size_t srcSize)
+                                 const void** srcPtr, size_t *srcSizePtr)
 {
-    const BYTE* ip = (const BYTE*)src;
+    const BYTE* ip = (const BYTE*)(*srcPtr);
     BYTE* const ostart = (BYTE* const)dst;
     BYTE* const oend = ostart + dstCapacity;
     BYTE* op = ostart;
-    size_t remainingSize = srcSize;
+    size_t remainingSize = *srcSizePtr;
 
     /* check */
-    if (srcSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+    if (remainingSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(src, ZSTD_frameHeaderSize_min);
-        size_t result;
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(ip, ZSTD_frameHeaderSize_prefix);
         if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
-        if (srcSize < frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
-        result = ZSTD_decodeFrameHeader(dctx, src, frameHeaderSize);
-        if (ZSTD_isError(result)) return result;
+        if (remainingSize < frameHeaderSize+ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
+        CHECK_F(ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize));
         ip += frameHeaderSize; remainingSize -= frameHeaderSize;
     }
 
@@ -909,40 +1466,109 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
         if (remainingSize<4) return ERROR(checksum_wrong);
         checkRead = MEM_readLE32(ip);
         if (checkRead != checkCalc) return ERROR(checksum_wrong);
+        ip += 4;
         remainingSize -= 4;
     }
 
-    if (remainingSize) return ERROR(srcSize_wrong);
+    /* Allow caller to get size read */
+    *srcPtr = ip;
+    *srcSizePtr = remainingSize;
     return op-ostart;
 }
 
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
 
-/*! ZSTD_decompress_usingPreparedDCtx() :
-*   Same as ZSTD_decompress_usingDict, but using a reference context `preparedDCtx`, where dictionary has been loaded.
-*   It avoids reloading the dictionary each time.
-*   `preparedDCtx` must have been properly initialized using ZSTD_decompressBegin_usingDict().
-*   Requires 2 contexts : 1 for reference (preparedDCtx), which will not be modified, and 1 to run the decompression operation (dctx) */
-size_t ZSTD_decompress_usingPreparedDCtx(ZSTD_DCtx* dctx, const ZSTD_DCtx* refDCtx,
-                                         void* dst, size_t dstCapacity,
-                                   const void* src, size_t srcSize)
+static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
+                                        void* dst, size_t dstCapacity,
+                                  const void* src, size_t srcSize,
+                                  const void *dict, size_t dictSize,
+                                  const ZSTD_DDict* ddict)
 {
-    ZSTD_copyDCtx(dctx, refDCtx);
-    ZSTD_checkContinuity(dctx, dst);
-    return ZSTD_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
-}
+    void* const dststart = dst;
+
+    if (ddict) {
+        if (dict) {
+            /* programmer error, these two cases should be mutually exclusive */
+            return ERROR(GENERIC);
+        }
+
+        dict = ZSTD_DDictDictContent(ddict);
+        dictSize = ZSTD_DDictDictSize(ddict);
+    }
+
+    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
+        U32 magicNumber;
 
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+        if (ZSTD_isLegacy(src, srcSize)) {
+            size_t decodedSize;
+            size_t const frameSize = ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
+            if (ZSTD_isError(frameSize)) return frameSize;
+
+            decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
+
+            dst = (BYTE*)dst + decodedSize;
+            dstCapacity -= decodedSize;
+
+            src = (const BYTE*)src + frameSize;
+            srcSize -= frameSize;
+
+            continue;
+        }
+#endif
+
+        magicNumber = MEM_readLE32(src);
+        if (magicNumber != ZSTD_MAGICNUMBER) {
+            if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+                size_t skippableSize;
+                if (srcSize < ZSTD_skippableHeaderSize)
+                    return ERROR(srcSize_wrong);
+                skippableSize = MEM_readLE32((const BYTE *)src + 4) +
+                                ZSTD_skippableHeaderSize;
+                if (srcSize < skippableSize) {
+                    return ERROR(srcSize_wrong);
+                }
+
+                src = (const BYTE *)src + skippableSize;
+                srcSize -= skippableSize;
+                continue;
+            } else {
+                return ERROR(prefix_unknown);
+            }
+        }
+
+        if (ddict) {
+            /* we were called from ZSTD_decompress_usingDDict */
+            ZSTD_refDDict(dctx, ddict);
+        } else {
+            /* this will initialize correctly with no dict if dict == NULL, so
+             * use this in all cases but ddict */
+            CHECK_F(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize));
+        }
+        ZSTD_checkContinuity(dctx, dst);
+
+        {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
+                                                    &src, &srcSize);
+            if (ZSTD_isError(res)) return res;
+            /* don't need to bounds check this, ZSTD_decompressFrame will have
+             * already */
+            dst = (BYTE*)dst + res;
+            dstCapacity -= res;
+        }
+    }
+
+    if (srcSize) return ERROR(srcSize_wrong); /* input not entirely consumed */
+
+    return (BYTE*)dst - (BYTE*)dststart;
+}
 
 size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
                                  void* dst, size_t dstCapacity,
-                                 const void* src, size_t srcSize,
-                                 const void* dict, size_t dictSize)
+                           const void* src, size_t srcSize,
+                           const void* dict, size_t dictSize)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT==1)
-    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_decompressLegacy(dst, dstCapacity, src, srcSize, dict, dictSize);
-#endif
-    ZSTD_decompressBegin_usingDict(dctx, dict, dictSize);
-    ZSTD_checkContinuity(dctx, dst);
-    return ZSTD_decompressFrame(dctx, dst, dstCapacity, src, srcSize);
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize, dict, dictSize, NULL);
 }
 
 
@@ -1009,32 +1635,30 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
     switch (dctx->stage)
     {
     case ZSTDds_getFrameHeaderSize :
-        if (srcSize != ZSTD_frameHeaderSize_min) return ERROR(srcSize_wrong);   /* impossible */
-        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
-            memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_min);
-            dctx->expected = ZSTD_skippableHeaderSize - ZSTD_frameHeaderSize_min; /* magic number + skippable frame length */
+        if (srcSize != ZSTD_frameHeaderSize_prefix) return ERROR(srcSize_wrong);      /* impossible */
+        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
+            memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_prefix);
+            dctx->expected = ZSTD_skippableHeaderSize - ZSTD_frameHeaderSize_prefix;  /* magic number + skippable frame length */
             dctx->stage = ZSTDds_decodeSkippableHeader;
             return 0;
         }
-        dctx->headerSize = ZSTD_frameHeaderSize(src, ZSTD_frameHeaderSize_min);
+        dctx->headerSize = ZSTD_frameHeaderSize(src, ZSTD_frameHeaderSize_prefix);
         if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
-        memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_min);
-        if (dctx->headerSize > ZSTD_frameHeaderSize_min) {
-            dctx->expected = dctx->headerSize - ZSTD_frameHeaderSize_min;
+        memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_prefix);
+        if (dctx->headerSize > ZSTD_frameHeaderSize_prefix) {
+            dctx->expected = dctx->headerSize - ZSTD_frameHeaderSize_prefix;
             dctx->stage = ZSTDds_decodeFrameHeader;
             return 0;
         }
         dctx->expected = 0;   /* not necessary to copy more */
 
     case ZSTDds_decodeFrameHeader:
-        {   size_t result;
-            memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_min, src, dctx->expected);
-            result = ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize);
-            if (ZSTD_isError(result)) return result;
-            dctx->expected = ZSTD_blockHeaderSize;
-            dctx->stage = ZSTDds_decodeBlockHeader;
-            return 0;
-        }
+        memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_prefix, src, dctx->expected);
+        CHECK_F(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize));
+        dctx->expected = ZSTD_blockHeaderSize;
+        dctx->stage = ZSTDds_decodeBlockHeader;
+        return 0;
+
     case ZSTDds_decodeBlockHeader:
         {   blockProperties_t bp;
             size_t const cBlockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
@@ -1106,7 +1730,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             return 0;
         }
     case ZSTDds_decodeSkippableHeader:
-        {   memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_min, src, dctx->expected);
+        {   memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_prefix, src, dctx->expected);
             dctx->expected = MEM_readLE32(dctx->headerBuffer + 4);
             dctx->stage = ZSTDds_skipFrame;
             return 0;
@@ -1131,50 +1755,59 @@ static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dict
     return 0;
 }
 
-static size_t ZSTD_loadEntropy(ZSTD_DCtx* dctx, const void* const dict, size_t const dictSize)
+/* ZSTD_loadEntropy() :
+ * dict : must point at beginning of a valid zstd dictionary
+ * @return : size of entropy tables read */
+static size_t ZSTD_loadEntropy(ZSTD_entropyTables_t* entropy, const void* const dict, size_t const dictSize)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
 
-    {   size_t const hSize = HUF_readDTableX4(dctx->hufTable, dict, dictSize);
+    if (dictSize <= 8) return ERROR(dictionary_corrupted);
+    dictPtr += 8;   /* skip header = magic + dictID */
+
+
+    {   size_t const hSize = HUF_readDTableX4(entropy->hufTable, dictPtr, dictEnd-dictPtr);
         if (HUF_isError(hSize)) return ERROR(dictionary_corrupted);
         dictPtr += hSize;
     }
 
     {   short offcodeNCount[MaxOff+1];
-        U32 offcodeMaxValue=MaxOff, offcodeLog=OffFSELog;
+        U32 offcodeMaxValue = MaxOff, offcodeLog;
         size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
-        { size_t const errorCode = FSE_buildDTable(dctx->OffTable, offcodeNCount, offcodeMaxValue, offcodeLog);
-          if (FSE_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        CHECK_E(FSE_buildDTable(entropy->OFTable, offcodeNCount, offcodeMaxValue, offcodeLog), dictionary_corrupted);
         dictPtr += offcodeHeaderSize;
     }
 
     {   short matchlengthNCount[MaxML+1];
-        unsigned matchlengthMaxValue = MaxML, matchlengthLog = MLFSELog;
+        unsigned matchlengthMaxValue = MaxML, matchlengthLog;
         size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        { size_t const errorCode = FSE_buildDTable(dctx->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog);
-          if (FSE_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        CHECK_E(FSE_buildDTable(entropy->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog), dictionary_corrupted);
         dictPtr += matchlengthHeaderSize;
     }
 
     {   short litlengthNCount[MaxLL+1];
-        unsigned litlengthMaxValue = MaxLL, litlengthLog = LLFSELog;
+        unsigned litlengthMaxValue = MaxLL, litlengthLog;
         size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
         if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        { size_t const errorCode = FSE_buildDTable(dctx->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog);
-          if (FSE_isError(errorCode)) return ERROR(dictionary_corrupted); }
+        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        CHECK_E(FSE_buildDTable(entropy->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog), dictionary_corrupted);
         dictPtr += litlengthHeaderSize;
     }
 
     if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
-    dctx->rep[0] = MEM_readLE32(dictPtr+0); if (dctx->rep[0] >= dictSize) return ERROR(dictionary_corrupted);
-    dctx->rep[1] = MEM_readLE32(dictPtr+4); if (dctx->rep[1] >= dictSize) return ERROR(dictionary_corrupted);
-    dctx->rep[2] = MEM_readLE32(dictPtr+8); if (dctx->rep[2] >= dictSize) return ERROR(dictionary_corrupted);
-    dictPtr += 12;
+    {   int i;
+        size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
+        for (i=0; i<3; i++) {
+            U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
+            if (rep==0 || rep >= dictContentSize) return ERROR(dictionary_corrupted);
+            entropy->rep[i] = rep;
+    }   }
 
-    dctx->litEntropy = dctx->fseEntropy = 1;
     return dictPtr - (const BYTE*)dict;
 }
 
@@ -1188,105 +1821,223 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
     dctx->dictID = MEM_readLE32((const char*)dict + 4);
 
     /* load entropy tables */
-    dict = (const char*)dict + 8;
-    dictSize -= 8;
-    {   size_t const eSize = ZSTD_loadEntropy(dctx, dict, dictSize);
+    {   size_t const eSize = ZSTD_loadEntropy(&dctx->entropy, dict, dictSize);
         if (ZSTD_isError(eSize)) return ERROR(dictionary_corrupted);
         dict = (const char*)dict + eSize;
         dictSize -= eSize;
     }
+    dctx->litEntropy = dctx->fseEntropy = 1;
 
     /* reference dictionary content */
     return ZSTD_refDictContent(dctx, dict, dictSize);
 }
 
-
 size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
 {
-    { size_t const errorCode = ZSTD_decompressBegin(dctx);
-      if (ZSTD_isError(errorCode)) return errorCode; }
-
-    if (dict && dictSize) {
-        size_t const errorCode = ZSTD_decompress_insertDictionary(dctx, dict, dictSize);
-        if (ZSTD_isError(errorCode)) return ERROR(dictionary_corrupted);
-    }
-
+    CHECK_F(ZSTD_decompressBegin(dctx));
+    if (dict && dictSize) CHECK_E(ZSTD_decompress_insertDictionary(dctx, dict, dictSize), dictionary_corrupted);
     return 0;
 }
 
 
+/* ======   ZSTD_DDict   ====== */
+
 struct ZSTD_DDict_s {
-    void* dict;
+    void* dictBuffer;
+    const void* dictContent;
     size_t dictSize;
-    ZSTD_DCtx* refContext;
+    ZSTD_entropyTables_t entropy;
+    U32 dictID;
+    U32 entropyPresent;
+    ZSTD_customMem cMem;
 };  /* typedef'd to ZSTD_DDict within "zstd.h" */
 
-ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize, ZSTD_customMem customMem)
+static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict)
+{
+    return ddict->dictContent;
+}
+
+static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict)
+{
+    return ddict->dictSize;
+}
+
+static void ZSTD_refDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict)
+{
+    ZSTD_decompressBegin(dstDCtx);  /* init */
+    if (ddict) {   /* support refDDict on NULL */
+        dstDCtx->dictID = ddict->dictID;
+        dstDCtx->base = ddict->dictContent;
+        dstDCtx->vBase = ddict->dictContent;
+        dstDCtx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+        dstDCtx->previousDstEnd = dstDCtx->dictEnd;
+        if (ddict->entropyPresent) {
+            dstDCtx->litEntropy = 1;
+            dstDCtx->fseEntropy = 1;
+            dstDCtx->LLTptr = ddict->entropy.LLTable;
+            dstDCtx->MLTptr = ddict->entropy.MLTable;
+            dstDCtx->OFTptr = ddict->entropy.OFTable;
+            dstDCtx->HUFptr = ddict->entropy.hufTable;
+            dstDCtx->entropy.rep[0] = ddict->entropy.rep[0];
+            dstDCtx->entropy.rep[1] = ddict->entropy.rep[1];
+            dstDCtx->entropy.rep[2] = ddict->entropy.rep[2];
+        } else {
+            dstDCtx->litEntropy = 0;
+            dstDCtx->fseEntropy = 0;
+        }
+    }
+}
+
+static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict)
+{
+    ddict->dictID = 0;
+    ddict->entropyPresent = 0;
+    if (ddict->dictSize < 8) return 0;
+    {   U32 const magic = MEM_readLE32(ddict->dictContent);
+        if (magic != ZSTD_DICT_MAGIC) return 0;   /* pure content mode */
+    }
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + 4);
+
+    /* load entropy tables */
+    CHECK_E( ZSTD_loadEntropy(&ddict->entropy, ddict->dictContent, ddict->dictSize), dictionary_corrupted );
+    ddict->entropyPresent = 1;
+    return 0;
+}
+
+
+ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize, unsigned byReference, ZSTD_customMem customMem)
 {
     if (!customMem.customAlloc && !customMem.customFree) customMem = defaultCustomMem;
     if (!customMem.customAlloc || !customMem.customFree) return NULL;
 
     {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
-        void* const dictContent = ZSTD_malloc(dictSize, customMem);
-        ZSTD_DCtx* const dctx = ZSTD_createDCtx_advanced(customMem);
-
-        if (!dictContent || !ddict || !dctx) {
-            ZSTD_free(dictContent, customMem);
-            ZSTD_free(ddict, customMem);
-            ZSTD_free(dctx, customMem);
-            return NULL;
-        }
+        if (!ddict) return NULL;
+        ddict->cMem = customMem;
 
-        memcpy(dictContent, dict, dictSize);
-        {   size_t const errorCode = ZSTD_decompressBegin_usingDict(dctx, dictContent, dictSize);
+        if ((byReference) || (!dict) || (!dictSize)) {
+            ddict->dictBuffer = NULL;
+            ddict->dictContent = dict;
+        } else {
+            void* const internalBuffer = ZSTD_malloc(dictSize, customMem);
+            if (!internalBuffer) { ZSTD_freeDDict(ddict); return NULL; }
+            memcpy(internalBuffer, dict, dictSize);
+            ddict->dictBuffer = internalBuffer;
+            ddict->dictContent = internalBuffer;
+        }
+        ddict->dictSize = dictSize;
+        ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+        /* parse dictionary content */
+        {   size_t const errorCode = ZSTD_loadEntropy_inDDict(ddict);
             if (ZSTD_isError(errorCode)) {
-                ZSTD_free(dictContent, customMem);
-                ZSTD_free(ddict, customMem);
-                ZSTD_free(dctx, customMem);
+                ZSTD_freeDDict(ddict);
                 return NULL;
         }   }
 
-        ddict->dict = dictContent;
-        ddict->dictSize = dictSize;
-        ddict->refContext = dctx;
         return ddict;
     }
 }
 
 /*! ZSTD_createDDict() :
-*   Create a digested dictionary, ready to start decompression without startup delay.
-*   `dict` can be released after `ZSTD_DDict` creation */
+*   Create a digested dictionary, to start decompression without startup delay.
+*   `dict` content is copied inside DDict.
+*   Consequently, `dict` can be released after `ZSTD_DDict` creation */
 ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize)
 {
     ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    return ZSTD_createDDict_advanced(dict, dictSize, allocator);
+    return ZSTD_createDDict_advanced(dict, dictSize, 0, allocator);
 }
 
+
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, to start decompression without startup delay.
+ *  Dictionary content is simply referenced, it will be accessed during decompression.
+ *  Warning : dictBuffer must outlive DDict (DDict must be freed before dictBuffer) */
+ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dictBuffer, dictSize, 1, allocator);
+}
+
+
 size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
 {
     if (ddict==NULL) return 0;   /* support free on NULL */
-    {   ZSTD_customMem const cMem = ddict->refContext->customMem;
-        ZSTD_freeDCtx(ddict->refContext);
-        ZSTD_free(ddict->dict, cMem);
+    {   ZSTD_customMem const cMem = ddict->cMem;
+        ZSTD_free(ddict->dictBuffer, cMem);
         ZSTD_free(ddict, cMem);
         return 0;
     }
 }
 
+/*! ZSTD_estimateDDictSize() :
+ *  Estimate amount of memory that will be needed to create a dictionary for decompression.
+ *  Note : if dictionary is created "byReference", reduce this amount by dictSize */
+size_t ZSTD_estimateDDictSize(size_t dictSize)
+{
+    return dictSize + sizeof(ZSTD_DDict);
+}
+
+size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*ddict) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
+}
+
+/*! ZSTD_getDictID_fromDict() :
+ *  Provides the dictID stored within dictionary.
+ *  if @return == 0, the dictionary is not conformant with Zstandard specification.
+ *  It can still be loaded, but as a content-only dictionary. */
+unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
+{
+    if (dictSize < 8) return 0;
+    if (MEM_readLE32(dict) != ZSTD_DICT_MAGIC) return 0;
+    return MEM_readLE32((const char*)dict + 4);
+}
+
+/*! ZSTD_getDictID_fromDDict() :
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;
+    return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
+}
+
+/*! ZSTD_getDictID_fromFrame() :
+ *  Provides the dictID required to decompresse frame stored within `src`.
+ *  If @return == 0, the dictID could not be decoded.
+ *  This could for one of the following reasons :
+ *  - The frame does not require a dictionary (most common case).
+ *  - The frame was built with dictID intentionally removed.
+ *    Needed dictionary is a hidden information.
+ *    Note : this use case also happens when using a non-conformant dictionary.
+ *  - `srcSize` is too small, and as a result, frame header could not be decoded.
+ *    Note : possible if `srcSize < ZSTD_FRAMEHEADERSIZE_MAX`.
+ *  - This is not a Zstandard frame.
+ *  When identifying the exact failure cause, it's possible to use
+ *  ZSTD_getFrameHeader(), which will provide a more precise error code. */
+unsigned ZSTD_getDictID_fromFrame(const void* src, size_t srcSize)
+{
+    ZSTD_frameHeader zfp = { 0 , 0 , 0 , 0 };
+    size_t const hError = ZSTD_getFrameHeader(&zfp, src, srcSize);
+    if (ZSTD_isError(hError)) return 0;
+    return zfp.dictID;
+}
+
+
 /*! ZSTD_decompress_usingDDict() :
 *   Decompression using a pre-digested Dictionary
 *   Use dictionary without significant overhead. */
-ZSTDLIB_API size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
-                                           void* dst, size_t dstCapacity,
-                                     const void* src, size_t srcSize,
-                                     const ZSTD_DDict* ddict)
+size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const ZSTD_DDict* ddict)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT==1)
-    if (ZSTD_isLegacy(src, srcSize)) return ZSTD_decompressLegacy(dst, dstCapacity, src, srcSize, ddict->dict, ddict->dictSize);
-#endif
-    return ZSTD_decompress_usingPreparedDCtx(dctx, ddict->refContext,
-                                             dst, dstCapacity,
-                                             src, srcSize);
+    /* pass content and size in case legacy frames are encountered */
+    return ZSTD_decompressMultiFrame(dctx, dst, dstCapacity, src, srcSize,
+                                     NULL, 0,
+                                     ddict);
 }
 
 
@@ -1299,8 +2050,10 @@ typedef enum { zdss_init, zdss_loadHeader,
 
 /* *** Resource management *** */
 struct ZSTD_DStream_s {
-    ZSTD_DCtx* zd;
-    ZSTD_frameParams fParams;
+    ZSTD_DCtx* dctx;
+    ZSTD_DDict* ddictLocal;
+    const ZSTD_DDict* ddict;
+    ZSTD_frameHeader fParams;
     ZSTD_dStreamStage stage;
     char*  inBuff;
     size_t inBuffSize;
@@ -1311,15 +2064,13 @@ struct ZSTD_DStream_s {
     size_t outStart;
     size_t outEnd;
     size_t blockSize;
-    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
+    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];   /* tmp buffer to store frame header */
     size_t lhSize;
     ZSTD_customMem customMem;
-    void* dictContent;
-    size_t dictSize;
-    const void* dictSource;
     void* legacyContext;
     U32 previousLegacyVersion;
     U32 legacyVersion;
+    U32 hostageByte;
 };   /* typedef'd to ZSTD_DStream within "zstd.h" */
 
 
@@ -1339,8 +2090,8 @@ ZSTD_DStream* ZSTD_createDStream_advanced(ZSTD_customMem customMem)
     if (zds==NULL) return NULL;
     memset(zds, 0, sizeof(ZSTD_DStream));
     memcpy(&zds->customMem, &customMem, sizeof(ZSTD_customMem));
-    zds->zd = ZSTD_createDCtx_advanced(customMem);
-    if (zds->zd == NULL) { ZSTD_freeDStream(zds); return NULL; }
+    zds->dctx = ZSTD_createDCtx_advanced(customMem);
+    if (zds->dctx == NULL) { ZSTD_freeDStream(zds); return NULL; }
     zds->stage = zdss_init;
     zds->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
     return zds;
@@ -1350,10 +2101,14 @@ size_t ZSTD_freeDStream(ZSTD_DStream* zds)
 {
     if (zds==NULL) return 0;   /* support free on null */
     {   ZSTD_customMem const cMem = zds->customMem;
-        ZSTD_freeDCtx(zds->zd);
+        ZSTD_freeDCtx(zds->dctx);
+        zds->dctx = NULL;
+        ZSTD_freeDDict(zds->ddictLocal);
+        zds->ddictLocal = NULL;
         ZSTD_free(zds->inBuff, cMem);
+        zds->inBuff = NULL;
         ZSTD_free(zds->outBuff, cMem);
-        ZSTD_free(zds->dictContent, cMem);
+        zds->outBuff = NULL;
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
         if (zds->legacyContext)
             ZSTD_freeLegacyStreamContext(zds->legacyContext, zds->previousLegacyVersion);
@@ -1373,17 +2128,15 @@ size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t di
 {
     zds->stage = zdss_loadHeader;
     zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
-    if ((dict != zds->dictSource) | (dictSize != zds->dictSize)) {   /* new dictionary */
-        if (dictSize > zds->dictSize) {
-            ZSTD_free(zds->dictContent, zds->customMem);
-            zds->dictContent = ZSTD_malloc(dictSize, zds->customMem);
-            if (zds->dictContent == NULL) return ERROR(memory_allocation);
-        }
-        memcpy(zds->dictContent, dict, dictSize);
-        zds->dictSize = dictSize;
-    }
+    ZSTD_freeDDict(zds->ddictLocal);
+    if (dict && dictSize >= 8) {
+        zds->ddictLocal = ZSTD_createDDict(dict, dictSize);
+        if (zds->ddictLocal == NULL) return ERROR(memory_allocation);
+    } else zds->ddictLocal = NULL;
+    zds->ddict = zds->ddictLocal;
     zds->legacyVersion = 0;
-    return 0;
+    zds->hostageByte = 0;
+    return ZSTD_frameHeaderSize_prefix;
 }
 
 size_t ZSTD_initDStream(ZSTD_DStream* zds)
@@ -1391,13 +2144,31 @@ size_t ZSTD_initDStream(ZSTD_DStream* zds)
     return ZSTD_initDStream_usingDict(zds, NULL, 0);
 }
 
+/* ZSTD_initDStream_usingDDict() :
+ * ddict will just be referenced, and must outlive decompression session */
+size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict)
+{
+    size_t const initResult = ZSTD_initDStream(zds);
+    zds->ddict = ddict;
+    return initResult;
+}
+
+size_t ZSTD_resetDStream(ZSTD_DStream* zds)
+{
+    zds->stage = zdss_loadHeader;
+    zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
+    zds->legacyVersion = 0;
+    zds->hostageByte = 0;
+    return ZSTD_frameHeaderSize_prefix;
+}
+
 size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds,
                                 ZSTD_DStreamParameter_e paramType, unsigned paramValue)
 {
     switch(paramType)
     {
         default : return ERROR(parameter_unknown);
-        case ZSTDdsp_maxWindowSize : zds->maxWindowSize = paramValue; break;
+        case DStream_p_maxWindowSize : zds->maxWindowSize = paramValue ? paramValue : (U32)(-1); break;
     }
     return 0;
 }
@@ -1405,11 +2176,24 @@ size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds,
 
 size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds)
 {
-    return sizeof(*zds) + ZSTD_sizeof_DCtx(zds->zd) + zds->inBuffSize + zds->outBuffSize;
+    if (zds==NULL) return 0;   /* support sizeof NULL */
+    return sizeof(*zds)
+           + ZSTD_sizeof_DCtx(zds->dctx)
+           + ZSTD_sizeof_DDict(zds->ddictLocal)
+           + zds->inBuffSize + zds->outBuffSize;
+}
+
+size_t ZSTD_estimateDStreamSize(ZSTD_frameHeader fHeader)
+{
+    size_t const windowSize = fHeader.windowSize;
+    size_t const blockSize = MIN(windowSize, ZSTD_BLOCKSIZE_ABSOLUTEMAX);
+    size_t const inBuffSize = blockSize;  /* no block can be larger */
+    size_t const outBuffSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
+    return sizeof(ZSTD_DStream) + ZSTD_estimateDCtxSize() + inBuffSize + outBuffSize;
 }
 
 
-/* *** Decompression *** */
+/* *****   Decompression   ***** */
 
 MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
@@ -1438,25 +2222,26 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
         switch(zds->stage)
         {
         case zdss_init :
-            return ERROR(init_missing);
+            ZSTD_resetDStream(zds);   /* transparent reset on starting decoding a new frame */
+            /* fall-through */
 
         case zdss_loadHeader :
-            {   size_t const hSize = ZSTD_getFrameParams(&zds->fParams, zds->headerBuffer, zds->lhSize);
+            {   size_t const hSize = ZSTD_getFrameHeader(&zds->fParams, zds->headerBuffer, zds->lhSize);
                 if (ZSTD_isError(hSize))
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
                 {   U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
                     if (legacyVersion) {
-                        size_t initResult;
-                        initResult = ZSTD_initLegacyStream(&zds->legacyContext, zds->previousLegacyVersion, legacyVersion,
-                                                           zds->dictContent, zds->dictSize);
-                        if (ZSTD_isError(initResult)) return initResult;
+                        const void* const dict = zds->ddict ? zds->ddict->dictContent : NULL;
+                        size_t const dictSize = zds->ddict ? zds->ddict->dictSize : 0;
+                        CHECK_F(ZSTD_initLegacyStream(&zds->legacyContext, zds->previousLegacyVersion, legacyVersion,
+                                                       dict, dictSize));
                         zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
                         return ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
                     } else {
                         return hSize; /* error */
                 }   }
 #else
-                    return hSize;
+                return hSize;
 #endif
                 if (hSize != 0) {   /* need more input */
                     size_t const toLoad = hSize - zds->lhSize;   /* if hSize!=0, hSize > zds->lhSize */
@@ -1464,61 +2249,75 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                         memcpy(zds->headerBuffer + zds->lhSize, ip, iend-ip);
                         zds->lhSize += iend-ip;
                         input->pos = input->size;
-                        return (hSize - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                        return (MAX(ZSTD_frameHeaderSize_min, hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
                     memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
                     break;
             }   }
 
+            /* check for single-pass mode opportunity */
+            if (zds->fParams.frameContentSize && zds->fParams.windowSize /* skippable frame if == 0 */
+                && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
+                size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
+                if (cSize <= (size_t)(iend-istart)) {
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds->dctx, op, oend-op, istart, cSize, zds->ddict);
+                    if (ZSTD_isError(decompressedSize)) return decompressedSize;
+                    ip = istart + cSize;
+                    op += decompressedSize;
+                    zds->dctx->expected = 0;
+                    zds->stage = zdss_init;
+                    someMoreWork = 0;
+                    break;
+            }   }
+
             /* Consume header */
-            ZSTD_decompressBegin_usingDict(zds->zd, zds->dictContent, zds->dictSize);
-            {   size_t const h1Size = ZSTD_nextSrcSizeToDecompress(zds->zd);  /* == ZSTD_frameHeaderSize_min */
-                size_t const h1Result = ZSTD_decompressContinue(zds->zd, NULL, 0, zds->headerBuffer, h1Size);
-                if (ZSTD_isError(h1Result)) return h1Result;   /* should not happen : already checked */
-                if (h1Size < zds->lhSize) {   /* long header */
-                    size_t const h2Size = ZSTD_nextSrcSizeToDecompress(zds->zd);
-                    size_t const h2Result = ZSTD_decompressContinue(zds->zd, NULL, 0, zds->headerBuffer+h1Size, h2Size);
-                    if (ZSTD_isError(h2Result)) return h2Result;
+            ZSTD_refDDict(zds->dctx, zds->ddict);
+            {   size_t const h1Size = ZSTD_nextSrcSizeToDecompress(zds->dctx);  /* == ZSTD_frameHeaderSize_prefix */
+                CHECK_F(ZSTD_decompressContinue(zds->dctx, NULL, 0, zds->headerBuffer, h1Size));
+                {   size_t const h2Size = ZSTD_nextSrcSizeToDecompress(zds->dctx);
+                    CHECK_F(ZSTD_decompressContinue(zds->dctx, NULL, 0, zds->headerBuffer+h1Size, h2Size));
             }   }
 
             zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
-            if (zds->fParams.windowSize > zds->maxWindowSize) return ERROR(frameParameter_unsupported);
+            if (zds->fParams.windowSize > zds->maxWindowSize) return ERROR(frameParameter_windowTooLarge);
 
-            /* Frame header instruct buffer sizes */
+            /* Adapt buffer sizes to frame header instructions */
             {   size_t const blockSize = MIN(zds->fParams.windowSize, ZSTD_BLOCKSIZE_ABSOLUTEMAX);
-                size_t const neededOutSize = zds->fParams.windowSize + blockSize;
+                size_t const neededOutSize = zds->fParams.windowSize + blockSize + WILDCOPY_OVERLENGTH * 2;
                 zds->blockSize = blockSize;
                 if (zds->inBuffSize < blockSize) {
                     ZSTD_free(zds->inBuff, zds->customMem);
-                    zds->inBuffSize = blockSize;
+                    zds->inBuffSize = 0;
                     zds->inBuff = (char*)ZSTD_malloc(blockSize, zds->customMem);
                     if (zds->inBuff == NULL) return ERROR(memory_allocation);
+                    zds->inBuffSize = blockSize;
                 }
                 if (zds->outBuffSize < neededOutSize) {
                     ZSTD_free(zds->outBuff, zds->customMem);
-                    zds->outBuffSize = neededOutSize;
+                    zds->outBuffSize = 0;
                     zds->outBuff = (char*)ZSTD_malloc(neededOutSize, zds->customMem);
                     if (zds->outBuff == NULL) return ERROR(memory_allocation);
+                    zds->outBuffSize = neededOutSize;
             }   }
             zds->stage = zdss_read;
             /* pass-through */
 
         case zdss_read:
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds->zd);
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds->dctx);
                 if (neededInSize==0) {  /* end of frame */
                     zds->stage = zdss_init;
                     someMoreWork = 0;
                     break;
                 }
                 if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
-                    const int isSkipFrame = ZSTD_isSkipFrame(zds->zd);
-                    size_t const decodedSize = ZSTD_decompressContinue(zds->zd,
+                    const int isSkipFrame = ZSTD_isSkipFrame(zds->dctx);
+                    size_t const decodedSize = ZSTD_decompressContinue(zds->dctx,
                         zds->outBuff + zds->outStart, (isSkipFrame ? 0 : zds->outBuffSize - zds->outStart),
                         ip, neededInSize);
                     if (ZSTD_isError(decodedSize)) return decodedSize;
                     ip += neededInSize;
                     if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
-                    zds->outEnd = zds->outStart +  decodedSize;
+                    zds->outEnd = zds->outStart + decodedSize;
                     zds->stage = zdss_flush;
                     break;
                 }
@@ -1528,7 +2327,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             }
 
         case zdss_load:
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds->zd);
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds->dctx);
                 size_t const toLoad = neededInSize - zds->inPos;   /* should always be <= remaining space within inBuff */
                 size_t loadedSize;
                 if (toLoad > zds->inBuffSize - zds->inPos) return ERROR(corruption_detected);   /* should never happen */
@@ -1538,8 +2337,8 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                 if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
 
                 /* decode loaded input */
-                {  const int isSkipFrame = ZSTD_isSkipFrame(zds->zd);
-                   size_t const decodedSize = ZSTD_decompressContinue(zds->zd,
+                {  const int isSkipFrame = ZSTD_isSkipFrame(zds->dctx);
+                   size_t const decodedSize = ZSTD_decompressContinue(zds->dctx,
                         zds->outBuff + zds->outStart, zds->outBuffSize - zds->outStart,
                         zds->inBuff, neededInSize);
                     if (ZSTD_isError(decodedSize)) return decodedSize;
@@ -1561,7 +2360,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                         zds->outStart = zds->outEnd = 0;
                     break;
                 }
-                /* cannot flush everything */
+                /* cannot complete flush */
                 someMoreWork = 0;
                 break;
             }
@@ -1571,9 +2370,22 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
     /* result */
     input->pos += (size_t)(ip-istart);
     output->pos += (size_t)(op-ostart);
-    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds->zd);
-        if (!nextSrcSizeHint) return (zds->outEnd != zds->outStart);   /* return 0 only if fully flushed too */
-        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds->zd) == ZSTDnit_block);
+    {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds->dctx);
+        if (!nextSrcSizeHint) {   /* frame fully decoded */
+            if (zds->outEnd == zds->outStart) {  /* output fully flushed */
+                if (zds->hostageByte) {
+                    if (input->pos >= input->size) { zds->stage = zdss_read; return 1; }  /* can't release hostage (not present) */
+                    input->pos++;  /* release hostage */
+                }
+                return 0;
+            }
+            if (!zds->hostageByte) { /* output not fully flushed; keep last byte as hostage; will be released when all output is flushed */
+                input->pos--;   /* note : pos > 0, otherwise, impossible to finish reading last block */
+                zds->hostageByte=1;
+            }
+            return 1;
+        }
+        nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds->dctx) == ZSTDnit_block);   /* preload header of next block */
         if (zds->inPos > nextSrcSizeHint) return ERROR(GENERIC);   /* should never happen */
         nextSrcSizeHint -= zds->inPos;   /* already loaded*/
         return nextSrcSizeHint;
diff --git a/contrib/zstd/zstd_internal.h b/contrib/zstd/zstd_internal.h
index 51e7170ec..2533333ba 100644
--- a/contrib/zstd/zstd_internal.h
+++ b/contrib/zstd/zstd_internal.h
@@ -10,6 +10,38 @@
 #ifndef ZSTD_CCOMMON_H_MODULE
 #define ZSTD_CCOMMON_H_MODULE
 
+/*-*******************************************************
+*  Compiler specifics
+*********************************************************/
+#ifdef _MSC_VER    /* Visual Studio */
+#  define FORCE_INLINE static __forceinline
+#  include <intrin.h>                    /* For Visual 2005 */
+#  pragma warning(disable : 4100)        /* disable: C4100: unreferenced formal parameter */
+#  pragma warning(disable : 4127)        /* disable: C4127: conditional expression is constant */
+#  pragma warning(disable : 4324)        /* disable: C4324: padded structure */
+#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
+
+#ifdef _MSC_VER
+#  define FORCE_NOINLINE static __declspec(noinline)
+#else
+#  ifdef __GNUC__
+#    define FORCE_NOINLINE static __attribute__((__noinline__))
+#  else
+#    define FORCE_NOINLINE static
+#  endif
+#endif
+
+
 /*-*************************************
 *  Dependencies
 ***************************************/
@@ -17,13 +49,21 @@
 #include "error_private.h"
 #define ZSTD_STATIC_LINKING_ONLY
 #include "zstd.h"
+#ifndef XXH_STATIC_LINKING_ONLY
+#  define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */
+#endif
+#include "xxhash.h"               /* XXH_reset, update, digest */
 
 
 /*-*************************************
-*  Common macros
+*  shared macros
 ***************************************/
+#undef MIN
+#undef MAX
 #define MIN(a,b) ((a)<(b) ? (a) : (b))
 #define MAX(a,b) ((a)>(b) ? (a) : (b))
+#define CHECK_F(f) { size_t const errcod = f; if (ERR_isError(errcod)) return errcod; }  /* check and Forward error code */
+#define CHECK_E(f, e) { size_t const errcod = f; if (ERR_isError(errcod)) return ERROR(e); }  /* check and send Error code */
 
 
 /*-*************************************
@@ -66,7 +106,6 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
 #define LONGNBSEQ 0x7F00
 
 #define MINMATCH 3
-#define EQUAL_READ32 4
 
 #define Litbits  8
 #define MaxLit ((1<<Litbits) - 1)
@@ -84,7 +123,8 @@ static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
 static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
                                              2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
                                             -1,-1,-1,-1 };
-static const U32 LL_defaultNormLog = 6;
+#define LL_DEFAULTNORMLOG 6  /* for static allocation */
+static const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
 
 static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                                       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
@@ -94,11 +134,13 @@ static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1,
                                              1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
                                              1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
                                             -1,-1,-1,-1,-1 };
-static const U32 ML_defaultNormLog = 6;
+#define ML_DEFAULTNORMLOG 6  /* for static allocation */
+static const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
 
 static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
                                               1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
-static const U32 OF_defaultNormLog = 5;
+#define OF_DEFAULTNORMLOG 5  /* for static allocation */
+static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 
 
 /*-*******************************************
@@ -110,7 +152,7 @@ static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
 /*! ZSTD_wildcopy() :
 *   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
 #define WILDCOPY_OVERLENGTH 8
-MEM_STATIC void ZSTD_wildcopy(void* dst, const void* src, size_t length)
+MEM_STATIC void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
 {
     const BYTE* ip = (const BYTE*)src;
     BYTE* op = (BYTE*)dst;
@@ -185,6 +227,7 @@ typedef struct {
     U32  log2litSum;
     U32  log2offCodeSum;
     U32  factor;
+    U32  staticPrices;
     U32  cachedPrice;
     U32  cachedLitLength;
     const BYTE* cachedLiterals;
@@ -197,7 +240,9 @@ int ZSTD_isSkipFrame(ZSTD_DCtx* dctx);
 /* custom memory allocation functions */
 void* ZSTD_defaultAllocFunction(void* opaque, size_t size);
 void ZSTD_defaultFreeFunction(void* opaque, void* address);
+#ifndef ZSTD_DLL_IMPORT
 static const ZSTD_customMem defaultCustomMem = { ZSTD_defaultAllocFunction, ZSTD_defaultFreeFunction, NULL };
+#endif
 void* ZSTD_malloc(size_t size, ZSTD_customMem customMem);
 void ZSTD_free(void* ptr, ZSTD_customMem customMem);
 
@@ -227,4 +272,13 @@ MEM_STATIC U32 ZSTD_highbit32(U32 val)
 }
 
 
+/* hidden functions */
+
+/* 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);
+
+
 #endif   /* ZSTD_CCOMMON_H_MODULE */
diff --git a/contrib/zstd/zstd_opt.h b/contrib/zstd/zstd_opt.h
index cb5872908..543761191 100644
--- a/contrib/zstd/zstd_opt.h
+++ b/contrib/zstd/zstd_opt.h
@@ -1,5 +1,5 @@
 /**
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * 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
@@ -15,7 +15,9 @@
 #define ZSTD_OPT_H_91842398743
 
 
-#define ZSTD_FREQ_DIV   5
+#define ZSTD_LITFREQ_ADD    2
+#define ZSTD_FREQ_DIV       4
+#define ZSTD_MAX_PRICE      (1<<30)
 
 /*-*************************************
 *  Price functions for optimal parser
@@ -30,22 +32,32 @@ FORCE_INLINE void ZSTD_setLog2Prices(seqStore_t* ssPtr)
 }
 
 
-MEM_STATIC void ZSTD_rescaleFreqs(seqStore_t* ssPtr)
+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) {
-        ssPtr->litSum = (2<<Litbits);
+        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 = (2<<Litbits);
+        ssPtr->matchSum = (ZSTD_LITFREQ_ADD<<Litbits);
 
-        for (u=0; u<=MaxLit; u++)
-            ssPtr->litFreq[u] = 2;
+        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++)
@@ -60,11 +72,11 @@ MEM_STATIC void ZSTD_rescaleFreqs(seqStore_t* ssPtr)
         ssPtr->litSum = 0;
 
         for (u=0; u<=MaxLit; u++) {
-            ssPtr->litFreq[u] = 1 + (ssPtr->litFreq[u]>>ZSTD_FREQ_DIV);
+            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);
+            ssPtr->litLengthFreq[u] = 1 + (ssPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
             ssPtr->litLengthSum += ssPtr->litLengthFreq[u];
         }
         for (u=0; u<=MaxML; u++) {
@@ -72,6 +84,7 @@ MEM_STATIC void ZSTD_rescaleFreqs(seqStore_t* ssPtr)
             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];
@@ -86,6 +99,9 @@ FORCE_INLINE U32 ZSTD_getLiteralPrice(seqStore_t* ssPtr, U32 litLength, const BY
 {
     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);
 
@@ -120,11 +136,17 @@ FORCE_INLINE U32 ZSTD_getLiteralPrice(seqStore_t* ssPtr, U32 litLength, const BY
 }
 
 
-FORCE_INLINE U32 ZSTD_getPrice(seqStore_t* seqStorePtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength)
+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);
-    U32 price = offCode + seqStorePtr->log2offCodeSum - ZSTD_highbit32(seqStorePtr->offCodeFreq[offCode]+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;
@@ -141,9 +163,9 @@ MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const B
     U32 u;
 
     /* literals */
-    seqStorePtr->litSum += litLength;
+    seqStorePtr->litSum += litLength*ZSTD_LITFREQ_ADD;
     for (u=0; u < litLength; u++)
-        seqStorePtr->litFreq[literals[u]]++;
+        seqStorePtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
 
     /* literal Length */
     {   const BYTE LL_deltaCode = 19;
@@ -153,10 +175,10 @@ MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const B
     }
 
     /* match offset */
-	{   BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
-		seqStorePtr->offCodeSum++;
-		seqStorePtr->offCodeFreq[offCode]++;
-	}
+    {   BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
+        seqStorePtr->offCodeSum++;
+        seqStorePtr->offCodeFreq[offCode]++;
+    }
 
     /* match Length */
     {   const BYTE ML_deltaCode = 36;
@@ -171,7 +193,7 @@ MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const B
 
 #define SET_PRICE(pos, mlen_, offset_, litlen_, price_)   \
     {                                                 \
-        while (last_pos < pos)  { opt[last_pos+1].price = 1<<30; last_pos++; } \
+        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_;                     \
@@ -181,7 +203,7 @@ MEM_STATIC void ZSTD_updatePrice(seqStore_t* seqStorePtr, U32 litLength, const B
 
 
 /* Update hashTable3 up to ip (excluded)
-   Assumption : always within prefix (ie. not within extDict) */
+   Assumption : always within prefix (i.e. not within extDict) */
 FORCE_INLINE
 U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_CCtx* zc, const BYTE* ip)
 {
@@ -338,6 +360,7 @@ static U32 ZSTD_BtGetAllMatches_selectMLS (
     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);
     }
 }
@@ -365,6 +388,7 @@ static U32 ZSTD_BtGetAllMatches_selectMLS_extDict (
     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);
     }
 }
@@ -375,7 +399,7 @@ static U32 ZSTD_BtGetAllMatches_selectMLS_extDict (
 *********************************/
 FORCE_INLINE
 void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
-                                    const void* src, size_t srcSize)
+                                    const void* src, size_t srcSize, const int ultra)
 {
     seqStore_t* seqStorePtr = &(ctx->seqStore);
     const BYTE* const istart = (const BYTE*)src;
@@ -398,10 +422,9 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
 
     /* init */
     ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(seqStorePtr);
+    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]; }
-    inr = ip;
 
     /* Match Loop */
     while (ip < ilimit) {
@@ -414,7 +437,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
         /* 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) && (ip==anchor)) ? (rep[0] - 1) : rep[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;
@@ -424,7 +447,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                     }
                     best_off = i - (ip == anchor);
                     do {
-                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH);
+                        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--;
@@ -449,7 +472,7 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
             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);
+                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++;
@@ -496,10 +519,10 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                 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;
+            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].mlen != 1)) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
+                    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;
@@ -510,21 +533,20 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                        }
 
                        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 (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);
-                            } else
-                                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH);
-                        } else {
-                            litlen = 0;
-                            price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH);
-                        }
-
-                        if (mlen > best_mlen) best_mlen = mlen;
-
-                        do {
                             if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
                                 SET_PRICE(cur + mlen, mlen, i, litlen, price);
                             mlen--;
@@ -549,12 +571,12 @@ void ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
                     if (opt[cur].mlen == 1) {
                         litlen = opt[cur].litlen;
                         if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(seqStorePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH);
+                            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);
+                            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);
+                        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))
@@ -600,7 +622,7 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
                 offset--;
             } else {
                 if (offset != 0) {
-                    best_off = ((offset==ZSTD_REP_MOVE_OPT) && (litLength==0)) ? (rep[0] - 1) : (rep[offset]);
+                    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;
@@ -614,7 +636,7 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
     }    }   /* for (cur=0; cur < last_pos; ) */
 
     /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->savedRep[i] = rep[i]; }
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
 
     /* Last Literals */
     {   size_t const lastLLSize = iend - anchor;
@@ -626,7 +648,7 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
 
 FORCE_INLINE
 void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize)
+                                     const void* src, size_t srcSize, const int ultra)
 {
     seqStore_t* seqStorePtr = &(ctx->seqStore);
     const BYTE* const istart = (const BYTE*)src;
@@ -655,9 +677,8 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
     { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=ctx->rep[i]; }
 
     ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(seqStorePtr);
+    ZSTD_rescaleFreqs(seqStorePtr, (const BYTE*)src, srcSize);
     ip += (ip==prefixStart);
-    inr = ip;
 
     /* Match Loop */
     while (ip < ilimit) {
@@ -666,13 +687,12 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
         U32 current = (U32)(ip-base);
         memset(opt, 0, sizeof(ZSTD_optimal_t));
         last_pos = 0;
-        inr = ip;
         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) && (ip==anchor)) ? (rep[0] - 1) : rep[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;
@@ -691,7 +711,7 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                     best_off = i - (ip==anchor);
                     litlen = opt[0].litlen;
                     do {
-                        price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH);
+                        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--;
@@ -721,7 +741,7 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
             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);
+                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++;
@@ -765,11 +785,10 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                 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 = 0;
-
+            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].mlen != 1)) ? (opt[cur].rep[0] - 1) : opt[cur].rep[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;
@@ -786,20 +805,20 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
                         }
 
                         best_off = i - (opt[cur].mlen != 1);
-                        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);
-                            } else
-                                price = ZSTD_getPrice(seqStorePtr, litlen, anchor, best_off, mlen - MINMATCH);
-                        } else {
-                            litlen = 0;
-                            price = opt[cur].price + ZSTD_getPrice(seqStorePtr, 0, NULL, best_off, mlen - MINMATCH);
-                        }
-
-                        best_mlen = mlen;
+                        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--;
@@ -808,30 +827,28 @@ void ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
 
             match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, inr, iend, maxSearches, mls, matches, minMatch);
 
-            if (match_num > 0 && matches[match_num-1].len > sufficient_len) {
+            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;
             }
 
-            best_mlen = (best_mlen > minMatch) ? best_mlen : minMatch;
-
             /* 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 = (cur + matches[u].len < ZSTD_OPT_NUM) ? matches[u].len : ZSTD_OPT_NUM - cur;
+                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);
+                            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);
+                            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);
+                        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))
@@ -877,7 +894,7 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
                 offset--;
             } else {
                 if (offset != 0) {
-                    best_off = ((offset==ZSTD_REP_MOVE_OPT) && (litLength==0)) ? (rep[0] - 1) : (rep[offset]);
+                    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;
@@ -892,7 +909,7 @@ _storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
     }    }   /* for (cur=0; cur < last_pos; ) */
 
     /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->savedRep[i] = rep[i]; }
+    { int i; for (i=0; i<ZSTD_REP_NUM; i++) ctx->repToConfirm[i] = rep[i]; }
 
     /* Last Literals */
     {   size_t lastLLSize = iend - anchor;