vendor/zstd/1.4.1

28 files changed, 1046 insertions, 594 deletions

diff --git a/lib/Makefile b/lib/Makefile
index 404f5b6921017..87a396c53ebd8 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -17,6 +17,7 @@ LIBVER_MINOR := $(shell echo $(LIBVER_MINOR_SCRIPT))
 LIBVER_PATCH := $(shell echo $(LIBVER_PATCH_SCRIPT))
 LIBVER := $(shell echo $(LIBVER_SCRIPT))
 VERSION?= $(LIBVER)
+CCVER := $(shell $(CC) --version)
 
 CPPFLAGS+= -I. -I./common -DXXH_NAMESPACE=ZSTD_
 ifeq ($(OS),Windows_NT)   # MinGW assumed
@@ -45,6 +46,10 @@ ZDICT_FILES := $(sort $(wildcard dictBuilder/*.c))
 ZDEPR_FILES := $(sort $(wildcard deprecated/*.c))
 ZSTD_FILES := $(ZSTDCOMMON_FILES)
 
+ifeq ($(findstring GCC,$(CCVER)),GCC)
+decompress/zstd_decompress_block.o :	CFLAGS+=-fno-tree-vectorize
+endif
+
 ZSTD_LEGACY_SUPPORT ?= 5
 ZSTD_LIB_COMPRESSION ?= 1
 ZSTD_LIB_DECOMPRESSION ?= 1
diff --git a/lib/common/compiler.h b/lib/common/compiler.h
index 0836e3ed27a39..87bf51ae8c896 100644
--- a/lib/common/compiler.h
+++ b/lib/common/compiler.h
@@ -127,6 +127,13 @@
     }                                     \
 }
 
+/* vectorization */
+#if !defined(__clang__) && defined(__GNUC__)
+#  define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
+#else
+#  define DONT_VECTORIZE
+#endif
+
 /* disable warnings */
 #ifdef _MSC_VER    /* Visual Studio */
 #  include <intrin.h>                    /* For Visual 2005 */
diff --git a/lib/common/zstd_internal.h b/lib/common/zstd_internal.h
index 31f756ab58123..81b16eac2ea62 100644
--- a/lib/common/zstd_internal.h
+++ b/lib/common/zstd_internal.h
@@ -34,7 +34,6 @@
 #endif
 #include "xxhash.h"                /* XXH_reset, update, digest */
 
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
@@ -193,19 +192,72 @@ static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 *  Shared functions to include for inlining
 *********************************************/
 static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+
 #define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+static void ZSTD_copy16(void* dst, const void* src) { memcpy(dst, src, 16); }
+#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
+
+#define WILDCOPY_OVERLENGTH 8
+#define VECLEN 16
+
+typedef enum {
+    ZSTD_no_overlap,
+    ZSTD_overlap_src_before_dst,
+    /*  ZSTD_overlap_dst_before_src, */
+} ZSTD_overlap_e;
 
 /*! ZSTD_wildcopy() :
  *  custom version of memcpy(), can overwrite up to WILDCOPY_OVERLENGTH bytes (if length==0) */
-#define WILDCOPY_OVERLENGTH 8
-MEM_STATIC void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+MEM_STATIC FORCE_INLINE_ATTR DONT_VECTORIZE
+void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e ovtype)
 {
+    ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
     const BYTE* ip = (const BYTE*)src;
     BYTE* op = (BYTE*)dst;
     BYTE* const oend = op + length;
-    do
-        COPY8(op, ip)
-    while (op < oend);
+
+    assert(diff >= 8 || (ovtype == ZSTD_no_overlap && diff < -8));
+    if (length < VECLEN || (ovtype == ZSTD_overlap_src_before_dst && diff < VECLEN)) {
+      do
+          COPY8(op, ip)
+      while (op < oend);
+    }
+    else {
+      if ((length & 8) == 0)
+        COPY8(op, ip);
+      do {
+        COPY16(op, ip);
+      }
+      while (op < oend);
+    }
+}
+
+/*! ZSTD_wildcopy_16min() :
+ *  same semantics as ZSTD_wilcopy() except guaranteed to be able to copy 16 bytes at the start */
+MEM_STATIC FORCE_INLINE_ATTR DONT_VECTORIZE
+void ZSTD_wildcopy_16min(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e ovtype)
+{
+    ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    BYTE* const oend = op + length;
+
+    assert(length >= 8);
+    assert(diff >= 8 || (ovtype == ZSTD_no_overlap && diff < -8));
+
+    if (ovtype == ZSTD_overlap_src_before_dst && diff < VECLEN) {
+      do
+          COPY8(op, ip)
+      while (op < oend);
+    }
+    else {
+      if ((length & 8) == 0)
+        COPY8(op, ip);
+      do {
+        COPY16(op, ip);
+      }
+      while (op < oend);
+    }
 }
 
 MEM_STATIC void ZSTD_wildcopy_e(void* dst, const void* src, void* dstEnd)   /* should be faster for decoding, but strangely, not verified on all platform */
diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c
index 2e163c8bf3d4e..1476512580b03 100644
--- a/lib/compress/zstd_compress.c
+++ b/lib/compress/zstd_compress.c
@@ -385,6 +385,11 @@ ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
         bounds.upperBound = ZSTD_lcm_uncompressed;
         return bounds;
 
+    case ZSTD_c_targetCBlockSize:
+        bounds.lowerBound = ZSTD_TARGETCBLOCKSIZE_MIN;
+        bounds.upperBound = ZSTD_TARGETCBLOCKSIZE_MAX;
+        return bounds;
+
     default:
         {   ZSTD_bounds const boundError = { ERROR(parameter_unsupported), 0, 0 };
             return boundError;
@@ -452,6 +457,7 @@ static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
     case ZSTD_c_ldmHashRateLog:
     case ZSTD_c_forceAttachDict:
     case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_targetCBlockSize:
     default:
         return 0;
     }
@@ -497,6 +503,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
     case ZSTD_c_ldmHashLog:
     case ZSTD_c_ldmMinMatch:
     case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_targetCBlockSize:
         break;
 
     default: RETURN_ERROR(parameter_unsupported);
@@ -671,6 +678,12 @@ size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
         CCtxParams->ldmParams.hashRateLog = value;
         return CCtxParams->ldmParams.hashRateLog;
 
+    case ZSTD_c_targetCBlockSize :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
+        CCtxParams->targetCBlockSize = value;
+        return CCtxParams->targetCBlockSize;
+
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
 }
@@ -692,13 +705,13 @@ size_t ZSTD_CCtxParams_getParameter(
         *value = CCtxParams->compressionLevel;
         break;
     case ZSTD_c_windowLog :
-        *value = CCtxParams->cParams.windowLog;
+        *value = (int)CCtxParams->cParams.windowLog;
         break;
     case ZSTD_c_hashLog :
-        *value = CCtxParams->cParams.hashLog;
+        *value = (int)CCtxParams->cParams.hashLog;
         break;
     case ZSTD_c_chainLog :
-        *value = CCtxParams->cParams.chainLog;
+        *value = (int)CCtxParams->cParams.chainLog;
         break;
     case ZSTD_c_searchLog :
         *value = CCtxParams->cParams.searchLog;
@@ -773,6 +786,9 @@ size_t ZSTD_CCtxParams_getParameter(
     case ZSTD_c_ldmHashRateLog :
         *value = CCtxParams->ldmParams.hashRateLog;
         break;
+    case ZSTD_c_targetCBlockSize :
+        *value = (int)CCtxParams->targetCBlockSize;
+        break;
     default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
     return 0;
@@ -930,12 +946,12 @@ size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
     @return : 0, or an error code if one value is beyond authorized range */
 size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
 {
-    BOUNDCHECK(ZSTD_c_windowLog, cParams.windowLog);
-    BOUNDCHECK(ZSTD_c_chainLog,  cParams.chainLog);
-    BOUNDCHECK(ZSTD_c_hashLog,   cParams.hashLog);
-    BOUNDCHECK(ZSTD_c_searchLog, cParams.searchLog);
-    BOUNDCHECK(ZSTD_c_minMatch,  cParams.minMatch);
-    BOUNDCHECK(ZSTD_c_targetLength,cParams.targetLength);
+    BOUNDCHECK(ZSTD_c_windowLog, (int)cParams.windowLog);
+    BOUNDCHECK(ZSTD_c_chainLog,  (int)cParams.chainLog);
+    BOUNDCHECK(ZSTD_c_hashLog,   (int)cParams.hashLog);
+    BOUNDCHECK(ZSTD_c_searchLog, (int)cParams.searchLog);
+    BOUNDCHECK(ZSTD_c_minMatch,  (int)cParams.minMatch);
+    BOUNDCHECK(ZSTD_c_targetLength,(int)cParams.targetLength);
     BOUNDCHECK(ZSTD_c_strategy,  cParams.strategy);
     return 0;
 }
@@ -951,7 +967,7 @@ ZSTD_clampCParams(ZSTD_compressionParameters cParams)
         if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound;      \
         else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
     }
-#   define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, int)
+#   define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, unsigned)
     CLAMP(ZSTD_c_windowLog, cParams.windowLog);
     CLAMP(ZSTD_c_chainLog,  cParams.chainLog);
     CLAMP(ZSTD_c_hashLog,   cParams.hashLog);
@@ -1282,15 +1298,14 @@ static void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
 }
 
 /*! ZSTD_invalidateMatchState()
- * Invalidate all the matches in the match finder tables.
- * Requires nextSrc and base to be set (can be NULL).
+ *  Invalidate all the matches in the match finder tables.
+ *  Requires nextSrc and base to be set (can be NULL).
  */
 static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
 {
     ZSTD_window_clear(&ms->window);
 
     ms->nextToUpdate = ms->window.dictLimit;
-    ms->nextToUpdate3 = ms->window.dictLimit;
     ms->loadedDictEnd = 0;
     ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
     ms->dictMatchState = NULL;
@@ -1327,15 +1342,17 @@ static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pl
 
 typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e;
 
+typedef enum { ZSTD_resetTarget_CDict, ZSTD_resetTarget_CCtx } ZSTD_resetTarget_e;
+
 static void*
 ZSTD_reset_matchState(ZSTD_matchState_t* ms,
                       void* ptr,
                 const ZSTD_compressionParameters* cParams,
-                      ZSTD_compResetPolicy_e const crp, U32 const forCCtx)
+                      ZSTD_compResetPolicy_e const crp, ZSTD_resetTarget_e const forWho)
 {
     size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
     size_t const hSize = ((size_t)1) << cParams->hashLog;
-    U32    const hashLog3 = (forCCtx && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    U32    const hashLog3 = ((forWho == ZSTD_resetTarget_CCtx) && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
     size_t const h3Size = ((size_t)1) << hashLog3;
     size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
 
@@ -1349,7 +1366,7 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
     ZSTD_invalidateMatchState(ms);
 
     /* opt parser space */
-    if (forCCtx && (cParams->strategy >= ZSTD_btopt)) {
+    if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
         DEBUGLOG(4, "reserving optimal parser space");
         ms->opt.litFreq = (unsigned*)ptr;
         ms->opt.litLengthFreq = ms->opt.litFreq + (1<<Litbits);
@@ -1377,6 +1394,19 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
     return ptr;
 }
 
+/* ZSTD_indexTooCloseToMax() :
+ * minor optimization : prefer memset() rather than reduceIndex()
+ * which is measurably slow in some circumstances (reported for Visual Studio).
+ * Works when re-using a context for a lot of smallish inputs :
+ * if all inputs are smaller than ZSTD_INDEXOVERFLOW_MARGIN,
+ * memset() will be triggered before reduceIndex().
+ */
+#define ZSTD_INDEXOVERFLOW_MARGIN (16 MB)
+static int ZSTD_indexTooCloseToMax(ZSTD_window_t w)
+{
+    return (size_t)(w.nextSrc - w.base) > (ZSTD_CURRENT_MAX - ZSTD_INDEXOVERFLOW_MARGIN);
+}
+
 #define ZSTD_WORKSPACETOOLARGE_FACTOR 3 /* define "workspace is too large" as this number of times larger than needed */
 #define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128  /* when workspace is continuously too large
                                          * during at least this number of times,
@@ -1388,7 +1418,7 @@ ZSTD_reset_matchState(ZSTD_matchState_t* ms,
     note : `params` are assumed fully validated at this stage */
 static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
                                       ZSTD_CCtx_params params,
-                                      U64 pledgedSrcSize,
+                                      U64 const pledgedSrcSize,
                                       ZSTD_compResetPolicy_e const crp,
                                       ZSTD_buffered_policy_e const zbuff)
 {
@@ -1400,13 +1430,21 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         if (ZSTD_equivalentParams(zc->appliedParams, params,
                                   zc->inBuffSize,
                                   zc->seqStore.maxNbSeq, zc->seqStore.maxNbLit,
-                                  zbuff, pledgedSrcSize)) {
-            DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> continue mode (wLog1=%u, blockSize1=%zu)",
-                        zc->appliedParams.cParams.windowLog, zc->blockSize);
+                                  zbuff, pledgedSrcSize) ) {
+            DEBUGLOG(4, "ZSTD_equivalentParams()==1 -> consider continue mode");
             zc->workSpaceOversizedDuration += (zc->workSpaceOversizedDuration > 0);   /* if it was too large, it still is */
-            if (zc->workSpaceOversizedDuration <= ZSTD_WORKSPACETOOLARGE_MAXDURATION)
+            if (zc->workSpaceOversizedDuration <= ZSTD_WORKSPACETOOLARGE_MAXDURATION) {
+                DEBUGLOG(4, "continue mode confirmed (wLog1=%u, blockSize1=%zu)",
+                            zc->appliedParams.cParams.windowLog, zc->blockSize);
+                if (ZSTD_indexTooCloseToMax(zc->blockState.matchState.window)) {
+                    /* prefer a reset, faster than a rescale */
+                    ZSTD_reset_matchState(&zc->blockState.matchState,
+                                           zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32,
+                                          &params.cParams,
+                                           crp, ZSTD_resetTarget_CCtx);
+                }
                 return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
-    }   }
+    }   }   }
     DEBUGLOG(4, "ZSTD_equivalentParams()==0 -> reset CCtx");
 
     if (params.ldmParams.enableLdm) {
@@ -1449,7 +1487,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
 
             if (workSpaceTooSmall || workSpaceWasteful) {
-                DEBUGLOG(4, "Need to resize workSpaceSize from %zuKB to %zuKB",
+                DEBUGLOG(4, "Resize workSpaceSize from %zuKB to %zuKB",
                             zc->workSpaceSize >> 10,
                             neededSpace >> 10);
 
@@ -1491,7 +1529,10 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
 
         ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
 
-        ptr = zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32;
+        ptr = ZSTD_reset_matchState(&zc->blockState.matchState,
+                                     zc->entropyWorkspace + HUF_WORKSPACE_SIZE_U32,
+                                    &params.cParams,
+                                     crp, ZSTD_resetTarget_CCtx);
 
         /* ldm hash table */
         /* initialize bucketOffsets table later for pointer alignment */
@@ -1509,8 +1550,6 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
         }
         assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
 
-        ptr = ZSTD_reset_matchState(&zc->blockState.matchState, ptr, &params.cParams, crp, /* forCCtx */ 1);
-
         /* sequences storage */
         zc->seqStore.maxNbSeq = maxNbSeq;
         zc->seqStore.sequencesStart = (seqDef*)ptr;
@@ -1587,15 +1626,14 @@ static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
                                  * handled in _enforceMaxDist */
 }
 
-static size_t ZSTD_resetCCtx_byAttachingCDict(
-    ZSTD_CCtx* cctx,
-    const ZSTD_CDict* cdict,
-    ZSTD_CCtx_params params,
-    U64 pledgedSrcSize,
-    ZSTD_buffered_policy_e zbuff)
+static size_t
+ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
+                        const ZSTD_CDict* cdict,
+                        ZSTD_CCtx_params params,
+                        U64 pledgedSrcSize,
+                        ZSTD_buffered_policy_e zbuff)
 {
-    {
-        const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
+    {   const ZSTD_compressionParameters* const cdict_cParams = &cdict->matchState.cParams;
         unsigned const windowLog = params.cParams.windowLog;
         assert(windowLog != 0);
         /* Resize working context table params for input only, since the dict
@@ -1607,8 +1645,7 @@ static size_t ZSTD_resetCCtx_byAttachingCDict(
         assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
     }
 
-    {
-        const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
+    {   const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
                                   - cdict->matchState.window.base);
         const U32 cdictLen = cdictEnd - cdict->matchState.window.dictLimit;
         if (cdictLen == 0) {
@@ -1625,9 +1662,9 @@ static size_t ZSTD_resetCCtx_byAttachingCDict(
                     cctx->blockState.matchState.window.base + cdictEnd;
                 ZSTD_window_clear(&cctx->blockState.matchState.window);
             }
+            /* loadedDictEnd is expressed within the referential of the active context */
             cctx->blockState.matchState.loadedDictEnd = cctx->blockState.matchState.window.dictLimit;
-        }
-    }
+    }   }
 
     cctx->dictID = cdict->dictID;
 
@@ -1681,7 +1718,6 @@ static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
         ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
         dstMatchState->window       = srcMatchState->window;
         dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
-        dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
         dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
     }
 
@@ -1761,7 +1797,6 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
         ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
         dstMatchState->window       = srcMatchState->window;
         dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
-        dstMatchState->nextToUpdate3= srcMatchState->nextToUpdate3;
         dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
     }
     dstCCtx->dictID = srcCCtx->dictID;
@@ -1831,16 +1866,15 @@ static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const
 
 /*! ZSTD_reduceIndex() :
 *   rescale all indexes to avoid future overflow (indexes are U32) */
-static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
+static void ZSTD_reduceIndex (ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const U32 reducerValue)
 {
-    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
-    {   U32 const hSize = (U32)1 << zc->appliedParams.cParams.hashLog;
+    {   U32 const hSize = (U32)1 << params->cParams.hashLog;
         ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
     }
 
-    if (zc->appliedParams.cParams.strategy != ZSTD_fast) {
-        U32 const chainSize = (U32)1 << zc->appliedParams.cParams.chainLog;
-        if (zc->appliedParams.cParams.strategy == ZSTD_btlazy2)
+    if (params->cParams.strategy != ZSTD_fast) {
+        U32 const chainSize = (U32)1 << params->cParams.chainLog;
+        if (params->cParams.strategy == ZSTD_btlazy2)
             ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
         else
             ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
@@ -2524,6 +2558,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
         op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
     else
         op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    assert(op <= oend);
     if (nbSeq==0) {
         /* Copy the old tables over as if we repeated them */
         memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
@@ -2532,6 +2567,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
 
     /* seqHead : flags for FSE encoding type */
     seqHead = op++;
+    assert(op <= oend);
 
     /* convert length/distances into codes */
     ZSTD_seqToCodes(seqStorePtr);
@@ -2555,6 +2591,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
             if (LLtype == set_compressed)
                 lastNCount = op;
             op += countSize;
+            assert(op <= oend);
     }   }
     /* build CTable for Offsets */
     {   unsigned max = MaxOff;
@@ -2577,6 +2614,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
             if (Offtype == set_compressed)
                 lastNCount = op;
             op += countSize;
+            assert(op <= oend);
     }   }
     /* build CTable for MatchLengths */
     {   unsigned max = MaxML;
@@ -2597,6 +2635,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
             if (MLtype == set_compressed)
                 lastNCount = op;
             op += countSize;
+            assert(op <= oend);
     }   }
 
     *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
@@ -2610,6 +2649,7 @@ ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
                                         longOffsets, bmi2);
         FORWARD_IF_ERROR(bitstreamSize);
         op += bitstreamSize;
+        assert(op <= oend);
         /* zstd versions <= 1.3.4 mistakenly report corruption when
          * FSE_readNCount() receives a buffer < 4 bytes.
          * Fixed by https://github.com/facebook/zstd/pull/1146.
@@ -2721,30 +2761,24 @@ void ZSTD_resetSeqStore(seqStore_t* ssPtr)
     ssPtr->longLengthID = 0;
 }
 
-static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
-                                        void* dst, size_t dstCapacity,
-                                        const void* src, size_t srcSize)
+typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_buildSeqStore_e;
+
+static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
 {
     ZSTD_matchState_t* const ms = &zc->blockState.matchState;
-    size_t cSize;
-    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
-                (unsigned)dstCapacity, (unsigned)ms->window.dictLimit, (unsigned)ms->nextToUpdate);
+    DEBUGLOG(5, "ZSTD_buildSeqStore (srcSize=%zu)", srcSize);
     assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
-
     /* Assert that we have correctly flushed the ctx params into the ms's copy */
     ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
-
     if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
         ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
-        cSize = 0;
-        goto out;  /* don't even attempt compression below a certain srcSize */
+        return ZSTDbss_noCompress; /* don't even attempt compression below a certain srcSize */
     }
     ZSTD_resetSeqStore(&(zc->seqStore));
     /* required for optimal parser to read stats from dictionary */
     ms->opt.symbolCosts = &zc->blockState.prevCBlock->entropy;
     /* tell the optimal parser how we expect to compress literals */
     ms->opt.literalCompressionMode = zc->appliedParams.literalCompressionMode;
-
     /* a gap between an attached dict and the current window is not safe,
      * they must remain adjacent,
      * and when that stops being the case, the dict must be unset */
@@ -2798,6 +2832,21 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
         {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
             ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
     }   }
+    return ZSTDbss_compress;
+}
+
+static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                                        void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize)
+{
+    size_t cSize;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit, (unsigned)zc->blockState.matchState.nextToUpdate);
+
+    {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+        FORWARD_IF_ERROR(bss);
+        if (bss == ZSTDbss_noCompress) { cSize = 0; goto out; }
+    }
 
     /* encode sequences and literals */
     cSize = ZSTD_compressSequences(&zc->seqStore,
@@ -2826,6 +2875,25 @@ out:
 }
 
 
+static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, void const* ip, void const* iend)
+{
+    if (ZSTD_window_needOverflowCorrection(ms->window, iend)) {
+        U32 const maxDist = (U32)1 << params->cParams.windowLog;
+        U32 const cycleLog = ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy);
+        U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
+        ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        ZSTD_reduceIndex(ms, params, correction);
+        if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+        else ms->nextToUpdate -= correction;
+        /* invalidate dictionaries on overflow correction */
+        ms->loadedDictEnd = 0;
+        ms->dictMatchState = NULL;
+    }
+}
+
+
 /*! ZSTD_compress_frameChunk() :
 *   Compress a chunk of data into one or multiple blocks.
 *   All blocks will be terminated, all input will be consumed.
@@ -2844,7 +2912,7 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
     BYTE* const ostart = (BYTE*)dst;
     BYTE* op = ostart;
     U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
-    assert(cctx->appliedParams.cParams.windowLog <= 31);
+    assert(cctx->appliedParams.cParams.windowLog <= ZSTD_WINDOWLOG_MAX);
 
     DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (unsigned)blockSize);
     if (cctx->appliedParams.fParams.checksumFlag && srcSize)
@@ -2859,19 +2927,10 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
                         "not enough space to store compressed block");
         if (remaining < blockSize) blockSize = remaining;
 
-        if (ZSTD_window_needOverflowCorrection(ms->window, ip + blockSize)) {
-            U32 const cycleLog = ZSTD_cycleLog(cctx->appliedParams.cParams.chainLog, cctx->appliedParams.cParams.strategy);
-            U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
-            ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
-            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
-            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
-            ZSTD_reduceIndex(cctx, correction);
-            if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
-            else ms->nextToUpdate -= correction;
-            ms->loadedDictEnd = 0;
-            ms->dictMatchState = NULL;
-        }
-        ZSTD_window_enforceMaxDist(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+        ZSTD_overflowCorrectIfNeeded(ms, &cctx->appliedParams, ip, ip + blockSize);
+        ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
+
+        /* Ensure hash/chain table insertion resumes no sooner than lowlimit */
         if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
 
         {   size_t cSize = ZSTD_compressBlock_internal(cctx,
@@ -2899,7 +2958,7 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
     }   }
 
     if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
 
@@ -2991,6 +3050,7 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
         fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams,
                                        cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
         FORWARD_IF_ERROR(fhSize);
+        assert(fhSize <= dstCapacity);
         dstCapacity -= fhSize;
         dst = (char*)dst + fhSize;
         cctx->stage = ZSTDcs_ongoing;
@@ -3007,18 +3067,7 @@ static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
 
     if (!frame) {
         /* overflow check and correction for block mode */
-        if (ZSTD_window_needOverflowCorrection(ms->window, (const char*)src + srcSize)) {
-            U32 const cycleLog = ZSTD_cycleLog(cctx->appliedParams.cParams.chainLog, cctx->appliedParams.cParams.strategy);
-            U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, 1 << cctx->appliedParams.cParams.windowLog, src);
-            ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
-            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
-            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
-            ZSTD_reduceIndex(cctx, correction);
-            if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
-            else ms->nextToUpdate -= correction;
-            ms->loadedDictEnd = 0;
-            ms->dictMatchState = NULL;
-        }
+        ZSTD_overflowCorrectIfNeeded(ms, &cctx->appliedParams, src, (BYTE const*)src + srcSize);
     }
 
     DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (unsigned)cctx->blockSize);
@@ -3074,7 +3123,7 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
                                          const void* src, size_t srcSize,
                                          ZSTD_dictTableLoadMethod_e dtlm)
 {
-    const BYTE* const ip = (const BYTE*) src;
+    const BYTE* ip = (const BYTE*) src;
     const BYTE* const iend = ip + srcSize;
 
     ZSTD_window_update(&ms->window, src, srcSize);
@@ -3085,32 +3134,42 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
 
     if (srcSize <= HASH_READ_SIZE) return 0;
 
-    switch(params->cParams.strategy)
-    {
-    case ZSTD_fast:
-        ZSTD_fillHashTable(ms, iend, dtlm);
-        break;
-    case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(ms, iend, dtlm);
-        break;
+    while (iend - ip > HASH_READ_SIZE) {
+        size_t const remaining = iend - ip;
+        size_t const chunk = MIN(remaining, ZSTD_CHUNKSIZE_MAX);
+        const BYTE* const ichunk = ip + chunk;
 
-    case ZSTD_greedy:
-    case ZSTD_lazy:
-    case ZSTD_lazy2:
-        if (srcSize >= HASH_READ_SIZE)
-            ZSTD_insertAndFindFirstIndex(ms, iend-HASH_READ_SIZE);
-        break;
+        ZSTD_overflowCorrectIfNeeded(ms, params, ip, ichunk);
 
-    case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
-    case ZSTD_btopt:
-    case ZSTD_btultra:
-    case ZSTD_btultra2:
-        if (srcSize >= HASH_READ_SIZE)
-            ZSTD_updateTree(ms, iend-HASH_READ_SIZE, iend);
-        break;
+        switch(params->cParams.strategy)
+        {
+        case ZSTD_fast:
+            ZSTD_fillHashTable(ms, ichunk, dtlm);
+            break;
+        case ZSTD_dfast:
+            ZSTD_fillDoubleHashTable(ms, ichunk, dtlm);
+            break;
 
-    default:
-        assert(0);  /* not possible : not a valid strategy id */
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            if (chunk >= HASH_READ_SIZE)
+                ZSTD_insertAndFindFirstIndex(ms, ichunk-HASH_READ_SIZE);
+            break;
+
+        case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            if (chunk >= HASH_READ_SIZE)
+                ZSTD_updateTree(ms, ichunk-HASH_READ_SIZE, ichunk);
+            break;
+
+        default:
+            assert(0);  /* not possible : not a valid strategy id */
+        }
+
+        ip = ichunk;
     }
 
     ms->nextToUpdate = (U32)(iend - ms->window.base);
@@ -3297,12 +3356,11 @@ static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
 
     FORWARD_IF_ERROR( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
                                      ZSTDcrp_continue, zbuff) );
-    {
-        size_t const dictID = ZSTD_compress_insertDictionary(
+    {   size_t const dictID = ZSTD_compress_insertDictionary(
                 cctx->blockState.prevCBlock, &cctx->blockState.matchState,
                 &params, dict, dictSize, dictContentType, dtlm, cctx->entropyWorkspace);
         FORWARD_IF_ERROR(dictID);
-        assert(dictID <= (size_t)(U32)-1);
+        assert(dictID <= UINT_MAX);
         cctx->dictID = (U32)dictID;
     }
     return 0;
@@ -3555,10 +3613,10 @@ static size_t ZSTD_initCDict_internal(
 
     /* Reset the state to no dictionary */
     ZSTD_reset_compressedBlockState(&cdict->cBlockState);
-    {   void* const end = ZSTD_reset_matchState(
-                &cdict->matchState,
-                (U32*)cdict->workspace + HUF_WORKSPACE_SIZE_U32,
-                &cParams, ZSTDcrp_continue, /* forCCtx */ 0);
+    {   void* const end = ZSTD_reset_matchState(&cdict->matchState,
+                            (U32*)cdict->workspace + HUF_WORKSPACE_SIZE_U32,
+                            &cParams,
+                             ZSTDcrp_continue, ZSTD_resetTarget_CDict);
         assert(end == (char*)cdict->workspace + cdict->workspaceSize);
         (void)end;
     }
@@ -4068,7 +4126,7 @@ static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
         case zcss_flush:
             DEBUGLOG(5, "flush stage");
             {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
-                size_t const flushed = ZSTD_limitCopy(op, oend-op,
+                size_t const flushed = ZSTD_limitCopy(op, (size_t)(oend-op),
                             zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
                 DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
                             (unsigned)toFlush, (unsigned)(oend-op), (unsigned)flushed);
@@ -4262,7 +4320,7 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
     if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush;   /* minimal estimation */
     /* single thread mode : attempt to calculate remaining to flush more precisely */
     {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
-        size_t const checksumSize = zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4;
+        size_t const checksumSize = (size_t)(zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4);
         size_t const toFlush = remainingToFlush + lastBlockSize + checksumSize;
         DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (unsigned)toFlush);
         return toFlush;
diff --git a/lib/compress/zstd_compress_internal.h b/lib/compress/zstd_compress_internal.h
index cc3cbb9da9ff5..5495899be3546 100644
--- a/lib/compress/zstd_compress_internal.h
+++ b/lib/compress/zstd_compress_internal.h
@@ -33,13 +33,13 @@ extern "C" {
 ***************************************/
 #define kSearchStrength      8
 #define HASH_READ_SIZE       8
-#define ZSTD_DUBT_UNSORTED_MARK 1   /* For btlazy2 strategy, index 1 now means "unsorted".
+#define ZSTD_DUBT_UNSORTED_MARK 1   /* For btlazy2 strategy, index ZSTD_DUBT_UNSORTED_MARK==1 means "unsorted".
                                        It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
                                        It's not a big deal though : candidate will just be sorted again.
                                        Additionally, candidate position 1 will be lost.
                                        But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
-                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy
-                                       Constant required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy.
+                                       This constant is required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
 
 
 /*-*************************************
@@ -128,21 +128,20 @@ typedef struct {
     BYTE const* base;       /* All regular indexes relative to this position */
     BYTE const* dictBase;   /* extDict indexes relative to this position */
     U32 dictLimit;          /* below that point, need extDict */
-    U32 lowLimit;           /* below that point, no more data */
+    U32 lowLimit;           /* below that point, no more valid data */
 } ZSTD_window_t;
 
 typedef struct ZSTD_matchState_t ZSTD_matchState_t;
 struct ZSTD_matchState_t {
     ZSTD_window_t window;   /* State for window round buffer management */
-    U32 loadedDictEnd;      /* index of end of dictionary */
+    U32 loadedDictEnd;      /* index of end of dictionary, within context's referential. When dict referential is copied into active context (i.e. not attached), effectively same value as dictSize, since referential starts from zero */
     U32 nextToUpdate;       /* index from which to continue table update */
-    U32 nextToUpdate3;      /* index from which to continue table update */
     U32 hashLog3;           /* dispatch table : larger == faster, more memory */
     U32* hashTable;
     U32* hashTable3;
     U32* chainTable;
     optState_t opt;         /* optimal parser state */
-    const ZSTD_matchState_t * dictMatchState;
+    const ZSTD_matchState_t* dictMatchState;
     ZSTD_compressionParameters cParams;
 };
 
@@ -195,6 +194,9 @@ struct ZSTD_CCtx_params_s {
     int compressionLevel;
     int forceWindow;           /* force back-references to respect limit of
                                 * 1<<wLog, even for dictionary */
+    size_t targetCBlockSize;   /* Tries to fit compressed block size to be around targetCBlockSize.
+                                * No target when targetCBlockSize == 0.
+                                * There is no guarantee on compressed block size */
 
     ZSTD_dictAttachPref_e attachDictPref;
     ZSTD_literalCompressionMode_e literalCompressionMode;
@@ -324,7 +326,7 @@ MEM_STATIC void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const v
     /* copy Literals */
     assert(seqStorePtr->maxNbLit <= 128 KB);
     assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
-    ZSTD_wildcopy(seqStorePtr->lit, literals, litLength);
+    ZSTD_wildcopy(seqStorePtr->lit, literals, litLength, ZSTD_no_overlap);
     seqStorePtr->lit += litLength;
 
     /* literal Length */
@@ -564,6 +566,9 @@ MEM_STATIC U64 ZSTD_rollingHash_rotate(U64 hash, BYTE toRemove, BYTE toAdd, U64
 /*-*************************************
 *  Round buffer management
 ***************************************/
+#if (ZSTD_WINDOWLOG_MAX_64 > 31)
+# error "ZSTD_WINDOWLOG_MAX is too large : would overflow ZSTD_CURRENT_MAX"
+#endif
 /* Max current allowed */
 #define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
 /* Maximum chunk size before overflow correction needs to be called again */
@@ -675,31 +680,49 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
  * Updates lowLimit so that:
  *    (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
  *
- * This allows a simple check that index >= lowLimit to see if index is valid.
- * This must be called before a block compression call, with srcEnd as the block
- * source end.
+ * It ensures index is valid as long as index >= lowLimit.
+ * This must be called before a block compression call.
+ *
+ * loadedDictEnd is only defined if a dictionary is in use for current compression.
+ * As the name implies, loadedDictEnd represents the index at end of dictionary.
+ * The value lies within context's referential, it can be directly compared to blockEndIdx.
  *
- * If loadedDictEndPtr is not NULL, we set it to zero once we update lowLimit.
- * This is because dictionaries are allowed to be referenced as long as the last
- * byte of the dictionary is in the window, but once they are out of range,
- * they cannot be referenced. If loadedDictEndPtr is NULL, we use
- * loadedDictEnd == 0.
+ * If loadedDictEndPtr is NULL, no dictionary is in use, and we use loadedDictEnd == 0.
+ * If loadedDictEndPtr is not NULL, we set it to zero after updating lowLimit.
+ * This is because dictionaries are allowed to be referenced fully
+ * as long as the last byte of the dictionary is in the window.
+ * Once input has progressed beyond window size, dictionary cannot be referenced anymore.
  *
- * In normal dict mode, the dict is between lowLimit and dictLimit. In
- * dictMatchState mode, lowLimit and dictLimit are the same, and the dictionary
- * is below them. forceWindow and dictMatchState are therefore incompatible.
+ * In normal dict mode, the dictionary lies between lowLimit and dictLimit.
+ * In dictMatchState mode, lowLimit and dictLimit are the same,
+ * and the dictionary is below them.
+ * forceWindow and dictMatchState are therefore incompatible.
  */
 MEM_STATIC void
 ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
-                           void const* srcEnd,
-                           U32 maxDist,
-                           U32* loadedDictEndPtr,
+                     const void* blockEnd,
+                           U32   maxDist,
+                           U32*  loadedDictEndPtr,
                      const ZSTD_matchState_t** dictMatchStatePtr)
 {
-    U32 const blockEndIdx = (U32)((BYTE const*)srcEnd - window->base);
-    U32 loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
-    DEBUGLOG(5, "ZSTD_window_enforceMaxDist: blockEndIdx=%u, maxDist=%u",
-                (unsigned)blockEndIdx, (unsigned)maxDist);
+    U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
+    U32 const loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
+    DEBUGLOG(5, "ZSTD_window_enforceMaxDist: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
+                (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
+
+    /* - When there is no dictionary : loadedDictEnd == 0.
+         In which case, the test (blockEndIdx > maxDist) is merely to avoid
+         overflowing next operation `newLowLimit = blockEndIdx - maxDist`.
+       - When there is a standard dictionary :
+         Index referential is copied from the dictionary,
+         which means it starts from 0.
+         In which case, loadedDictEnd == dictSize,
+         and it makes sense to compare `blockEndIdx > maxDist + dictSize`
+         since `blockEndIdx` also starts from zero.
+       - When there is an attached dictionary :
+         loadedDictEnd is expressed within the referential of the context,
+         so it can be directly compared against blockEndIdx.
+    */
     if (blockEndIdx > maxDist + loadedDictEnd) {
         U32 const newLowLimit = blockEndIdx - maxDist;
         if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
@@ -708,10 +731,31 @@ ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
                         (unsigned)window->dictLimit, (unsigned)window->lowLimit);
             window->dictLimit = window->lowLimit;
         }
-        if (loadedDictEndPtr)
-            *loadedDictEndPtr = 0;
-        if (dictMatchStatePtr)
-            *dictMatchStatePtr = NULL;
+        /* On reaching window size, dictionaries are invalidated */
+        if (loadedDictEndPtr) *loadedDictEndPtr = 0;
+        if (dictMatchStatePtr) *dictMatchStatePtr = NULL;
+    }
+}
+
+/* Similar to ZSTD_window_enforceMaxDist(),
+ * but only invalidates dictionary
+ * when input progresses beyond window size. */
+MEM_STATIC void
+ZSTD_checkDictValidity(ZSTD_window_t* window,
+                       const void* blockEnd,
+                             U32   maxDist,
+                             U32*  loadedDictEndPtr,
+                       const ZSTD_matchState_t** dictMatchStatePtr)
+{
+    U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
+    U32 const loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
+    DEBUGLOG(5, "ZSTD_checkDictValidity: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
+                (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
+
+    if (loadedDictEnd && (blockEndIdx > maxDist + loadedDictEnd)) {
+        /* On reaching window size, dictionaries are invalidated */
+        if (loadedDictEndPtr) *loadedDictEndPtr = 0;
+        if (dictMatchStatePtr) *dictMatchStatePtr = NULL;
     }
 }
 
diff --git a/lib/compress/zstd_double_fast.c b/lib/compress/zstd_double_fast.c
index 47faf6d641db5..5957255d9041c 100644
--- a/lib/compress/zstd_double_fast.c
+++ b/lib/compress/zstd_double_fast.c
@@ -43,8 +43,7 @@ void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
             /* Only load extra positions for ZSTD_dtlm_full */
             if (dtlm == ZSTD_dtlm_fast)
                 break;
-        }
-    }
+    }   }
 }
 
 
@@ -63,7 +62,10 @@ size_t ZSTD_compressBlock_doubleFast_generic(
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32 prefixLowestIndex = ms->window.dictLimit;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32 lowestValid = ms->window.dictLimit;
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 prefixLowestIndex = (endIndex - lowestValid > maxDistance) ? endIndex - maxDistance : lowestValid;
     const BYTE* const prefixLowest = base + prefixLowestIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
@@ -95,8 +97,15 @@ size_t ZSTD_compressBlock_doubleFast_generic(
                                      dictCParams->chainLog : hBitsS;
     const U32 dictAndPrefixLength  = (U32)(ip - prefixLowest + dictEnd - dictStart);
 
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
+
     assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
 
+    /* if a dictionary is attached, it must be within window range */
+    if (dictMode == ZSTD_dictMatchState) {
+        assert(lowestValid + maxDistance >= endIndex);
+    }
+
     /* init */
     ip += (dictAndPrefixLength == 0);
     if (dictMode == ZSTD_noDict) {
@@ -138,7 +147,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
             const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, 0, mLength-MINMATCH);
             goto _match_stored;
         }
 
@@ -147,7 +156,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
           && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
             mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, 0, mLength-MINMATCH);
             goto _match_stored;
         }
 
@@ -170,8 +179,7 @@ size_t ZSTD_compressBlock_doubleFast_generic(
                 offset = (U32)(current - dictMatchIndexL - dictIndexDelta);
                 while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
                 goto _match_found;
-            }
-        }
+        }   }
 
         if (matchIndexS > prefixLowestIndex) {
             /* check prefix short match */
@@ -186,16 +194,14 @@ size_t ZSTD_compressBlock_doubleFast_generic(
 
             if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
                 goto _search_next_long;
-            }
-        }
+        }   }
 
         ip += ((ip-anchor) >> kSearchStrength) + 1;
         continue;
 
 _search_next_long:
 
-        {
-            size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+        {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
             size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
             U32 const matchIndexL3 = hashLong[hl3];
             const BYTE* matchL3 = base + matchIndexL3;
@@ -221,9 +227,7 @@ _search_next_long:
                     offset = (U32)(current + 1 - dictMatchIndexL3 - dictIndexDelta);
                     while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
                     goto _match_found;
-                }
-            }
-        }
+        }   }   }
 
         /* if no long +1 match, explore the short match we found */
         if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
@@ -242,7 +246,7 @@ _match_found:
         offset_2 = offset_1;
         offset_1 = offset;
 
-        ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
 _match_stored:
         /* match found */
@@ -250,11 +254,14 @@ _match_stored:
         anchor = ip;
 
         if (ip <= ilimit) {
-            /* Fill Table */
-            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] =
-                hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;  /* here because current+2 could be > iend-8 */
-            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] =
-                hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = current+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
 
             /* check immediate repcode */
             if (dictMode == ZSTD_dictMatchState) {
@@ -278,8 +285,7 @@ _match_stored:
                         continue;
                     }
                     break;
-                }
-            }
+            }   }
 
             if (dictMode == ZSTD_noDict) {
                 while ( (ip <= ilimit)
@@ -294,14 +300,15 @@ _match_stored:
                     ip += rLength;
                     anchor = ip;
                     continue;   /* faster when present ... (?) */
-    }   }   }   }
+        }   }   }
+    }   /* while (ip < ilimit) */
 
     /* save reps for next block */
     rep[0] = offset_1 ? offset_1 : offsetSaved;
     rep[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
@@ -360,10 +367,15 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
     const BYTE* anchor = istart;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    const U32   prefixStartIndex = ms->window.dictLimit;
     const BYTE* const base = ms->window.base;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   maxDistance = 1U << cParams->windowLog;
+    const U32   lowestValid = ms->window.lowLimit;
+    const U32   lowLimit = (endIndex - lowestValid > maxDistance) ? endIndex - maxDistance : lowestValid;
+    const U32   dictStartIndex = lowLimit;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
     const BYTE* const prefixStart = base + prefixStartIndex;
-    const U32   dictStartIndex = ms->window.lowLimit;
     const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const dictStart = dictBase + dictStartIndex;
     const BYTE* const dictEnd = dictBase + prefixStartIndex;
@@ -371,6 +383,10 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
 
     DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
 
+    /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, mls, ZSTD_noDict);
+
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
         const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
@@ -396,7 +412,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, 0, mLength-MINMATCH);
         } else {
             if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
                 const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
@@ -407,7 +423,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
             } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
@@ -432,23 +448,27 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
                 }
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
             } else {
                 ip += ((ip-anchor) >> kSearchStrength) + 1;
                 continue;
         }   }
 
-        /* found a match : store it */
+        /* move to next sequence start */
         ip += mLength;
         anchor = ip;
 
         if (ip <= ilimit) {
-            /* Fill Table */
-            hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;
-            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2;
-            hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
-            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = current+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
             /* check immediate repcode */
             while (ip <= ilimit) {
                 U32 const current2 = (U32)(ip-base);
@@ -475,7 +495,7 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
     rep[1] = offset_2;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
diff --git a/lib/compress/zstd_fast.c b/lib/compress/zstd_fast.c
index ed997b441c1e9..a05b8a47f14ea 100644
--- a/lib/compress/zstd_fast.c
+++ b/lib/compress/zstd_fast.c
@@ -13,7 +13,8 @@
 
 
 void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
-                        void const* end, ZSTD_dictTableLoadMethod_e dtlm)
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32* const hashTable = ms->hashTable;
@@ -41,6 +42,7 @@ void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
     }   }   }   }
 }
 
+
 FORCE_INLINE_TEMPLATE
 size_t ZSTD_compressBlock_fast_generic(
         ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -58,7 +60,10 @@ size_t ZSTD_compressBlock_fast_generic(
     const BYTE* ip0 = istart;
     const BYTE* ip1;
     const BYTE* anchor = istart;
-    const U32   prefixStartIndex = ms->window.dictLimit;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   maxDistance = 1U << cParams->windowLog;
+    const U32   validStartIndex = ms->window.dictLimit;
+    const U32   prefixStartIndex = (endIndex - validStartIndex > maxDistance) ? endIndex - maxDistance : validStartIndex;
     const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
@@ -165,7 +170,7 @@ _match: /* Requires: ip0, match0, offcode */
     rep[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
@@ -222,8 +227,15 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);
     const U32 dictHLog             = dictCParams->hashLog;
 
-    /* otherwise, we would get index underflow when translating a dict index
-     * into a local index */
+    /* if a dictionary is still attached, it necessarily means that
+     * it is within window size. So we just check it. */
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);
+    assert(endIndex - prefixStartIndex <= maxDistance);
+    (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */
+
+    /* ensure there will be no no underflow
+     * when translating a dict index into a local index */
     assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
 
     /* init */
@@ -251,7 +263,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
             const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, 0, mLength-MINMATCH);
         } else if ( (matchIndex <= prefixStartIndex) ) {
             size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);
             U32 const dictMatchIndex = dictHashTable[dictHash];
@@ -271,7 +283,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
                 } /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
             }
         } else if (MEM_read32(match) != MEM_read32(ip)) {
             /* it's not a match, and we're not going to check the dictionary */
@@ -286,7 +298,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
                  && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
             offset_2 = offset_1;
             offset_1 = offset;
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
         }
 
         /* match found */
@@ -327,7 +339,7 @@ size_t ZSTD_compressBlock_fast_dictMatchState_generic(
     rep[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 size_t ZSTD_compressBlock_fast_dictMatchState(
@@ -366,15 +378,24 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32   dictStartIndex = ms->window.lowLimit;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   maxDistance = 1U << cParams->windowLog;
+    const U32   validLow = ms->window.lowLimit;
+    const U32   lowLimit = (endIndex - validLow > maxDistance) ? endIndex - maxDistance : validLow;
+    const U32   dictStartIndex = lowLimit;
     const BYTE* const dictStart = dictBase + dictStartIndex;
-    const U32   prefixStartIndex = ms->window.dictLimit;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
     const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const dictEnd = dictBase + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
     U32 offset_1=rep[0], offset_2=rep[1];
 
+    /* switch to "regular" variant if extDict is invalidated due to maxDistance */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, mls);
+
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
         const size_t h = ZSTD_hashPtr(ip, hlog, mls);
@@ -394,7 +415,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
             const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
             mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStore, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, 0, mLength-MINMATCH);
         } else {
             if ( (matchIndex < dictStartIndex) ||
                  (MEM_read32(match) != MEM_read32(ip)) ) {
@@ -410,7 +431,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
                 offset = current - matchIndex;
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStore, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
         }   }
 
         /* found a match : store it */
@@ -445,7 +466,7 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(
     rep[1] = offset_2;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
diff --git a/lib/compress/zstd_lazy.c b/lib/compress/zstd_lazy.c
index 53f998a4374b1..94d906c01f1c8 100644
--- a/lib/compress/zstd_lazy.c
+++ b/lib/compress/zstd_lazy.c
@@ -83,7 +83,10 @@ ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
     U32* largerPtr  = smallerPtr + 1;
     U32 matchIndex = *smallerPtr;   /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */
     U32 dummy32;   /* to be nullified at the end */
-    U32 const windowLow = ms->window.lowLimit;
+    U32 const windowValid = ms->window.lowLimit;
+    U32 const maxDistance = 1U << cParams->windowLog;
+    U32 const windowLow = (current - windowValid > maxDistance) ? current - maxDistance : windowValid;
+
 
     DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
                 current, dictLimit, windowLow);
@@ -239,7 +242,9 @@ ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
 
     const BYTE* const base = ms->window.base;
     U32    const current = (U32)(ip-base);
-    U32    const windowLow = ms->window.lowLimit;
+    U32    const maxDistance = 1U << cParams->windowLog;
+    U32    const windowValid = ms->window.lowLimit;
+    U32    const windowLow = (current - windowValid > maxDistance) ? current - maxDistance : windowValid;
 
     U32*   const bt = ms->chainTable;
     U32    const btLog  = cParams->chainLog - 1;
@@ -490,8 +495,10 @@ size_t ZSTD_HcFindBestMatch_generic (
     const U32 dictLimit = ms->window.dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
-    const U32 lowLimit = ms->window.lowLimit;
     const U32 current = (U32)(ip-base);
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 lowValid = ms->window.lowLimit;
+    const U32 lowLimit = (current - lowValid > maxDistance) ? current - maxDistance : lowValid;
     const U32 minChain = current > chainSize ? current - chainSize : 0;
     U32 nbAttempts = 1U << cParams->searchLog;
     size_t ml=4-1;
@@ -653,7 +660,6 @@ size_t ZSTD_compressBlock_lazy_generic(
 
     /* init */
     ip += (dictAndPrefixLength == 0);
-    ms->nextToUpdate3 = ms->nextToUpdate;
     if (dictMode == ZSTD_noDict) {
         U32 const maxRep = (U32)(ip - prefixLowest);
         if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
@@ -933,7 +939,6 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(
     U32 offset_1 = rep[0], offset_2 = rep[1];
 
     /* init */
-    ms->nextToUpdate3 = ms->nextToUpdate;
     ip += (ip == prefixStart);
 
     /* Match Loop */
diff --git a/lib/compress/zstd_ldm.c b/lib/compress/zstd_ldm.c
index 784d20f3ab716..3dcf86e6e8a32 100644
--- a/lib/compress/zstd_ldm.c
+++ b/lib/compress/zstd_ldm.c
@@ -447,7 +447,7 @@ size_t ZSTD_ldm_generateSequences(
         if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
             U32 const ldmHSize = 1U << params->hashLog;
             U32 const correction = ZSTD_window_correctOverflow(
-                &ldmState->window, /* cycleLog */ 0, maxDist, src);
+                &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
             ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
         }
         /* 2. We enforce the maximum offset allowed.
diff --git a/lib/compress/zstd_opt.c b/lib/compress/zstd_opt.c
index efb69d3267d1a..e32e542e021a9 100644
--- a/lib/compress/zstd_opt.c
+++ b/lib/compress/zstd_opt.c
@@ -255,13 +255,13 @@ static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optP
  * to provide a cost which is directly comparable to a match ending at same position */
 static int ZSTD_litLengthContribution(U32 const litLength, const optState_t* const optPtr, int optLevel)
 {
-    if (optPtr->priceType >= zop_predef) return WEIGHT(litLength, optLevel);
+    if (optPtr->priceType >= zop_predef) return (int)WEIGHT(litLength, optLevel);
 
     /* dynamic statistics */
     {   U32 const llCode = ZSTD_LLcode(litLength);
-        int const contribution = (LL_bits[llCode] * BITCOST_MULTIPLIER)
-                               + WEIGHT(optPtr->litLengthFreq[0], optLevel)   /* note: log2litLengthSum cancel out */
-                               - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
+        int const contribution = (int)(LL_bits[llCode] * BITCOST_MULTIPLIER)
+                               + (int)WEIGHT(optPtr->litLengthFreq[0], optLevel)   /* note: log2litLengthSum cancel out */
+                               - (int)WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
 #if 1
         return contribution;
 #else
@@ -278,7 +278,7 @@ static int ZSTD_literalsContribution(const BYTE* const literals, U32 const litLe
                                      const optState_t* const optPtr,
                                      int optLevel)
 {
-    int const contribution = ZSTD_rawLiteralsCost(literals, litLength, optPtr, optLevel)
+    int const contribution = (int)ZSTD_rawLiteralsCost(literals, litLength, optPtr, optLevel)
                            + ZSTD_litLengthContribution(litLength, optPtr, optLevel);
     return contribution;
 }
@@ -372,13 +372,15 @@ MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
 
 /* Update hashTable3 up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE* const ip)
+static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
+                                              U32* nextToUpdate3,
+                                              const BYTE* const ip)
 {
     U32* const hashTable3 = ms->hashTable3;
     U32 const hashLog3 = ms->hashLog3;
     const BYTE* const base = ms->window.base;
-    U32 idx = ms->nextToUpdate3;
-    U32 const target = ms->nextToUpdate3 = (U32)(ip - base);
+    U32 idx = *nextToUpdate3;
+    U32 const target = (U32)(ip - base);
     size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
     assert(hashLog3 > 0);
 
@@ -387,6 +389,7 @@ static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms, const BYTE*
         idx++;
     }
 
+    *nextToUpdate3 = target;
     return hashTable3[hash3];
 }
 
@@ -503,9 +506,11 @@ static U32 ZSTD_insertBt1(
     }   }
 
     *smallerPtr = *largerPtr = 0;
-    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
-    assert(matchEndIdx > current + 8);
-    return matchEndIdx - (current + 8);
+    {   U32 positions = 0;
+        if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+        assert(matchEndIdx > current + 8);
+        return MAX(positions, matchEndIdx - (current + 8));
+    }
 }
 
 FORCE_INLINE_TEMPLATE
@@ -520,8 +525,13 @@ void ZSTD_updateTree_internal(
     DEBUGLOG(6, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
                 idx, target, dictMode);
 
-    while(idx < target)
-        idx += ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
+    while(idx < target) {
+        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
+        assert(idx < (U32)(idx + forward));
+        idx += forward;
+    }
+    assert((size_t)(ip - base) <= (size_t)(U32)(-1));
+    assert((size_t)(iend - base) <= (size_t)(U32)(-1));
     ms->nextToUpdate = target;
 }
 
@@ -531,16 +541,18 @@ void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
 
 FORCE_INLINE_TEMPLATE
 U32 ZSTD_insertBtAndGetAllMatches (
+                    ZSTD_match_t* matches,   /* store result (found matches) in this table (presumed large enough) */
                     ZSTD_matchState_t* ms,
+                    U32* nextToUpdate3,
                     const BYTE* const ip, const BYTE* const iLimit, const ZSTD_dictMode_e dictMode,
-                    U32 rep[ZSTD_REP_NUM],
+                    const U32 rep[ZSTD_REP_NUM],
                     U32 const ll0,   /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
-                    ZSTD_match_t* matches,
                     const U32 lengthToBeat,
                     U32 const mls /* template */)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    U32 const maxDistance = 1U << cParams->windowLog;
     const BYTE* const base = ms->window.base;
     U32 const current = (U32)(ip-base);
     U32 const hashLog = cParams->hashLog;
@@ -556,8 +568,9 @@ U32 ZSTD_insertBtAndGetAllMatches (
     U32 const dictLimit = ms->window.dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
-    U32 const btLow = btMask >= current ? 0 : current - btMask;
-    U32 const windowLow = ms->window.lowLimit;
+    U32 const btLow = (btMask >= current) ? 0 : current - btMask;
+    U32 const windowValid = ms->window.lowLimit;
+    U32 const windowLow = ((current - windowValid) > maxDistance) ? current - maxDistance : windowValid;
     U32 const matchLow = windowLow ? windowLow : 1;
     U32* smallerPtr = bt + 2*(current&btMask);
     U32* largerPtr  = bt + 2*(current&btMask) + 1;
@@ -627,7 +640,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
     /* HC3 match finder */
     if ((mls == 3) /*static*/ && (bestLength < mls)) {
-        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, ip);
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
         if ((matchIndex3 >= matchLow)
           & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
             size_t mlen;
@@ -653,9 +666,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
                      (ip+mlen == iLimit) ) {  /* best possible length */
                     ms->nextToUpdate = current+1;  /* skip insertion */
                     return 1;
-                }
-            }
-        }
+        }   }   }
         /* no dictMatchState lookup: dicts don't have a populated HC3 table */
     }
 
@@ -760,10 +771,13 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
 
 FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
+                        ZSTD_match_t* matches,   /* store result (match found, increasing size) in this table */
                         ZSTD_matchState_t* ms,
+                        U32* nextToUpdate3,
                         const BYTE* ip, const BYTE* const iHighLimit, const ZSTD_dictMode_e dictMode,
-                        U32 rep[ZSTD_REP_NUM], U32 const ll0,
-                        ZSTD_match_t* matches, U32 const lengthToBeat)
+                        const U32 rep[ZSTD_REP_NUM],
+                        U32 const ll0,
+                        U32 const lengthToBeat)
 {
     const ZSTD_compressionParameters* const cParams = &ms->cParams;
     U32 const matchLengthSearch = cParams->minMatch;
@@ -772,12 +786,12 @@ FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
     ZSTD_updateTree_internal(ms, ip, iHighLimit, matchLengthSearch, dictMode);
     switch(matchLengthSearch)
     {
-    case 3 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 3);
+    case 3 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 3);
     default :
-    case 4 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 4);
-    case 5 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 5);
+    case 4 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 4);
+    case 5 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 5);
     case 7 :
-    case 6 : return ZSTD_insertBtAndGetAllMatches(ms, ip, iHighLimit, dictMode, rep, ll0, matches, lengthToBeat, 6);
+    case 6 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 6);
     }
 }
 
@@ -853,6 +867,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
 
     U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
     U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
+    U32 nextToUpdate3 = ms->nextToUpdate;
 
     ZSTD_optimal_t* const opt = optStatePtr->priceTable;
     ZSTD_match_t* const matches = optStatePtr->matchTable;
@@ -862,7 +877,6 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
     DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
                 (U32)(ip - base), ms->window.dictLimit, ms->nextToUpdate);
     assert(optLevel <= 2);
-    ms->nextToUpdate3 = ms->nextToUpdate;
     ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
     ip += (ip==prefixStart);
 
@@ -873,7 +887,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
         /* find first match */
         {   U32 const litlen = (U32)(ip - anchor);
             U32 const ll0 = !litlen;
-            U32 const nbMatches = ZSTD_BtGetAllMatches(ms, ip, iend, dictMode, rep, ll0, matches, minMatch);
+            U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
             if (!nbMatches) { ip++; continue; }
 
             /* initialize opt[0] */
@@ -970,7 +984,7 @@ ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
                 U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
                 U32 const previousPrice = opt[cur].price;
                 U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
-                U32 const nbMatches = ZSTD_BtGetAllMatches(ms, inr, iend, dictMode, opt[cur].rep, ll0, matches, minMatch);
+                U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
                 U32 matchNb;
                 if (!nbMatches) {
                     DEBUGLOG(7, "rPos:%u : no match found", cur);
@@ -1094,7 +1108,7 @@ _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
     }   /* while (ip < ilimit) */
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
@@ -1158,7 +1172,6 @@ ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
     ms->window.dictLimit += (U32)srcSize;
     ms->window.lowLimit = ms->window.dictLimit;
     ms->nextToUpdate = ms->window.dictLimit;
-    ms->nextToUpdate3 = ms->window.dictLimit;
 
     /* re-inforce weight of collected statistics */
     ZSTD_upscaleStats(&ms->opt);
diff --git a/lib/compress/zstdmt_compress.c b/lib/compress/zstdmt_compress.c
index 38fbb907685b4..9e537b884853e 100644
--- a/lib/compress/zstdmt_compress.c
+++ b/lib/compress/zstdmt_compress.c
@@ -1129,9 +1129,14 @@ size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
             size_t const produced = ZSTD_isError(cResult) ? 0 : cResult;
             size_t const flushed = ZSTD_isError(cResult) ? 0 : jobPtr->dstFlushed;
             assert(flushed <= produced);
+            assert(jobPtr->consumed <= jobPtr->src.size);
             toFlush = produced - flushed;
-            if (toFlush==0 && (jobPtr->consumed >= jobPtr->src.size)) {
-                /* doneJobID is not-fully-flushed, but toFlush==0 : doneJobID should be compressing some more data */
+            /* if toFlush==0, nothing is available to flush.
+             * However, jobID is expected to still be active:
+             * if jobID was already completed and fully flushed,
+             * ZSTDMT_flushProduced() should have already moved onto next job.
+             * Therefore, some input has not yet been consumed. */
+            if (toFlush==0) {
                 assert(jobPtr->consumed < jobPtr->src.size);
             }
         }
@@ -1148,12 +1153,16 @@ size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
 
 static unsigned ZSTDMT_computeTargetJobLog(ZSTD_CCtx_params const params)
 {
-    if (params.ldmParams.enableLdm)
+    unsigned jobLog;
+    if (params.ldmParams.enableLdm) {
         /* In Long Range Mode, the windowLog is typically oversized.
          * In which case, it's preferable to determine the jobSize
          * based on chainLog instead. */
-        return MAX(21, params.cParams.chainLog + 4);
-    return MAX(20, params.cParams.windowLog + 2);
+        jobLog = MAX(21, params.cParams.chainLog + 4);
+    } else {
+        jobLog = MAX(20, params.cParams.windowLog + 2);
+    }
+    return MIN(jobLog, (unsigned)ZSTDMT_JOBLOG_MAX);
 }
 
 static int ZSTDMT_overlapLog_default(ZSTD_strategy strat)
@@ -1197,7 +1206,7 @@ static size_t ZSTDMT_computeOverlapSize(ZSTD_CCtx_params const params)
         ovLog = MIN(params.cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2)
                 - overlapRLog;
     }
-    assert(0 <= ovLog && ovLog <= 30);
+    assert(0 <= ovLog && ovLog <= ZSTD_WINDOWLOG_MAX);
     DEBUGLOG(4, "overlapLog : %i", params.overlapLog);
     DEBUGLOG(4, "overlap size : %i", 1 << ovLog);
     return (ovLog==0) ? 0 : (size_t)1 << ovLog;
@@ -1391,7 +1400,7 @@ size_t ZSTDMT_initCStream_internal(
         FORWARD_IF_ERROR( ZSTDMT_resize(mtctx, params.nbWorkers) );
 
     if (params.jobSize != 0 && params.jobSize < ZSTDMT_JOBSIZE_MIN) params.jobSize = ZSTDMT_JOBSIZE_MIN;
-    if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = ZSTDMT_JOBSIZE_MAX;
+    if (params.jobSize > (size_t)ZSTDMT_JOBSIZE_MAX) params.jobSize = (size_t)ZSTDMT_JOBSIZE_MAX;
 
     mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN);  /* do not trigger multi-threading when srcSize is too small */
     if (mtctx->singleBlockingThread) {
@@ -1432,6 +1441,8 @@ size_t ZSTDMT_initCStream_internal(
     if (mtctx->targetSectionSize == 0) {
         mtctx->targetSectionSize = 1ULL << ZSTDMT_computeTargetJobLog(params);
     }
+    assert(mtctx->targetSectionSize <= (size_t)ZSTDMT_JOBSIZE_MAX);
+
     if (params.rsyncable) {
         /* Aim for the targetsectionSize as the average job size. */
         U32 const jobSizeMB = (U32)(mtctx->targetSectionSize >> 20);
diff --git a/lib/compress/zstdmt_compress.h b/lib/compress/zstdmt_compress.h
index 12e6bcb3a3389..12a526087db4e 100644
--- a/lib/compress/zstdmt_compress.h
+++ b/lib/compress/zstdmt_compress.h
@@ -50,6 +50,7 @@
 #ifndef ZSTDMT_JOBSIZE_MIN
 #  define ZSTDMT_JOBSIZE_MIN (1 MB)
 #endif
+#define ZSTDMT_JOBLOG_MAX   (MEM_32bits() ? 29 : 30)
 #define ZSTDMT_JOBSIZE_MAX  (MEM_32bits() ? (512 MB) : (1024 MB))
 
 
diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
index 675596f5aa925..e42872ad96abe 100644
--- a/lib/decompress/zstd_decompress.c
+++ b/lib/decompress/zstd_decompress.c
@@ -360,8 +360,11 @@ static size_t readSkippableFrameSize(void const* src, size_t srcSize)
     sizeU32 = MEM_readLE32((BYTE const*)src + ZSTD_FRAMEIDSIZE);
     RETURN_ERROR_IF((U32)(sizeU32 + ZSTD_SKIPPABLEHEADERSIZE) < sizeU32,
                     frameParameter_unsupported);
-
-    return skippableHeaderSize + sizeU32;
+    {
+        size_t const skippableSize = skippableHeaderSize + sizeU32;
+        RETURN_ERROR_IF(skippableSize > srcSize, srcSize_wrong);
+        return skippableSize;
+    }
 }
 
 /** ZSTD_findDecompressedSize() :
@@ -378,11 +381,10 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
 
         if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
             size_t const skippableSize = readSkippableFrameSize(src, srcSize);
-            if (ZSTD_isError(skippableSize))
-                return skippableSize;
-            if (srcSize < skippableSize) {
+            if (ZSTD_isError(skippableSize)) {
                 return ZSTD_CONTENTSIZE_ERROR;
             }
+            assert(skippableSize <= srcSize);
 
             src = (const BYTE *)src + skippableSize;
             srcSize -= skippableSize;
@@ -467,6 +469,8 @@ static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize
     if ((srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
         && (MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
         frameSizeInfo.compressedSize = readSkippableFrameSize(src, srcSize);
+        assert(ZSTD_isError(frameSizeInfo.compressedSize) ||
+               frameSizeInfo.compressedSize <= srcSize);
         return frameSizeInfo;
     } else {
         const BYTE* ip = (const BYTE*)src;
@@ -529,7 +533,6 @@ size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
     return frameSizeInfo.compressedSize;
 }
 
-
 /** ZSTD_decompressBound() :
  *  compatible with legacy mode
  *  `src` must point to the start of a ZSTD frame or a skippeable frame
@@ -546,6 +549,7 @@ unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
         unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
         if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
             return ZSTD_CONTENTSIZE_ERROR;
+        assert(srcSize >= compressedSize);
         src = (const BYTE*)src + compressedSize;
         srcSize -= compressedSize;
         bound += decompressedBound;
@@ -738,9 +742,8 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                         (unsigned)magicNumber, ZSTD_MAGICNUMBER);
             if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
                 size_t const skippableSize = readSkippableFrameSize(src, srcSize);
-                if (ZSTD_isError(skippableSize))
-                    return skippableSize;
-                RETURN_ERROR_IF(srcSize < skippableSize, srcSize_wrong);
+                FORWARD_IF_ERROR(skippableSize);
+                assert(skippableSize <= srcSize);
 
                 src = (const BYTE *)src + skippableSize;
                 srcSize -= skippableSize;
diff --git a/lib/decompress/zstd_decompress_block.c b/lib/decompress/zstd_decompress_block.c
index a2a7eedcf218c..24f4859c56c94 100644
--- a/lib/decompress/zstd_decompress_block.c
+++ b/lib/decompress/zstd_decompress_block.c
@@ -505,7 +505,7 @@ size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
     *nbSeqPtr = nbSeq;
 
     /* FSE table descriptors */
-    RETURN_ERROR_IF(ip+4 > iend, srcSize_wrong); /* minimum possible size */
+    RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong); /* minimum possible size: 1 byte for symbol encoding types */
     {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
         symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
         symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
@@ -637,9 +637,10 @@ size_t ZSTD_execSequence(BYTE* op,
     if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
 
     /* 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 */
+        ZSTD_wildcopy_16min(op, (*litPtr), sequence.litLength, ZSTD_no_overlap);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    else
+        ZSTD_copy8(op, *litPtr);
     op = oLitEnd;
     *litPtr = iLitEnd;   /* update for next sequence */
 
@@ -686,13 +687,13 @@ size_t ZSTD_execSequence(BYTE* op,
 
     if (oMatchEnd > oend-(16-MINMATCH)) {
         if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op);
+            ZSTD_wildcopy(op, match, oend_w - op, ZSTD_overlap_src_before_dst);
             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 */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst);   /* works even if matchLength < 8 */
     }
     return sequenceLength;
 }
@@ -717,9 +718,11 @@ size_t ZSTD_execSequenceLong(BYTE* op,
     if (oLitEnd > oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, dictStart, dictEnd);
 
     /* copy Literals */
-    ZSTD_copy8(op, *litPtr);  /* note : op <= oLitEnd <= oend_w == oend - 8 */
     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 */
+        ZSTD_wildcopy_16min(op, *litPtr, sequence.litLength, ZSTD_no_overlap);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
+    else
+        ZSTD_copy8(op, *litPtr);  /* note : op <= oLitEnd <= oend_w == oend - 8 */
+
     op = oLitEnd;
     *litPtr = iLitEnd;   /* update for next sequence */
 
@@ -766,13 +769,13 @@ size_t ZSTD_execSequenceLong(BYTE* op,
 
     if (oMatchEnd > oend-(16-MINMATCH)) {
         if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op);
+            ZSTD_wildcopy(op, match, oend_w - op, ZSTD_overlap_src_before_dst);
             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 */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst);   /* works even if matchLength < 8 */
     }
     return sequenceLength;
 }
@@ -889,6 +892,7 @@ ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
 }
 
 FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
 ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
                                void* dst, size_t maxDstSize,
                          const void* seqStart, size_t seqSize, int nbSeq,
@@ -918,6 +922,11 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
         ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
         ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
 
+        ZSTD_STATIC_ASSERT(
+                BIT_DStream_unfinished < BIT_DStream_completed &&
+                BIT_DStream_endOfBuffer < BIT_DStream_completed &&
+                BIT_DStream_completed < BIT_DStream_overflow);
+
         for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) {
             nbSeq--;
             {   seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
@@ -930,6 +939,7 @@ ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
         /* check if reached exact end */
         DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
         RETURN_ERROR_IF(nbSeq, corruption_detected);
+        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected);
         /* save reps for next block */
         { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
     }
@@ -1131,6 +1141,7 @@ ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
 
 #ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
 static TARGET_ATTRIBUTE("bmi2") size_t
+DONT_VECTORIZE
 ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
                                  void* dst, size_t maxDstSize,
                            const void* seqStart, size_t seqSize, int nbSeq,
diff --git a/lib/dictBuilder/cover.c b/lib/dictBuilder/cover.c
index 21464ad0310ec..621996759b6ae 100644
--- a/lib/dictBuilder/cover.c
+++ b/lib/dictBuilder/cover.c
@@ -526,10 +526,10 @@ static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
  * 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.
+ * Returns 0 on success or error code on error.
  * The context must be destroyed with `COVER_ctx_destroy()`.
  */
-static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
+static size_t COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
                           const size_t *samplesSizes, unsigned nbSamples,
                           unsigned d, double splitPoint) {
   const BYTE *const samples = (const BYTE *)samplesBuffer;
@@ -544,17 +544,17 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
       totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
     DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
                  (unsigned)(totalSamplesSize>>20), (COVER_MAX_SAMPLES_SIZE >> 20));
-    return 0;
+    return ERROR(srcSize_wrong);
   }
   /* Check if there are at least 5 training samples */
   if (nbTrainSamples < 5) {
     DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid.", nbTrainSamples);
-    return 0;
+    return ERROR(srcSize_wrong);
   }
   /* Check if there's testing sample */
   if (nbTestSamples < 1) {
     DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.", nbTestSamples);
-    return 0;
+    return ERROR(srcSize_wrong);
   }
   /* Zero the context */
   memset(ctx, 0, sizeof(*ctx));
@@ -577,7 +577,7 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
   if (!ctx->suffix || !ctx->dmerAt || !ctx->offsets) {
     DISPLAYLEVEL(1, "Failed to allocate scratch buffers\n");
     COVER_ctx_destroy(ctx);
-    return 0;
+    return ERROR(memory_allocation);
   }
   ctx->freqs = NULL;
   ctx->d = d;
@@ -624,7 +624,7 @@ static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
                 (ctx->d <= 8 ? &COVER_cmp8 : &COVER_cmp), &COVER_group);
   ctx->freqs = ctx->suffix;
   ctx->suffix = NULL;
-  return 1;
+  return 0;
 }
 
 void COVER_warnOnSmallCorpus(size_t maxDictSize, size_t nbDmers, int displayLevel)
@@ -729,11 +729,11 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
   /* Checks */
   if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
     DISPLAYLEVEL(1, "Cover parameters incorrect\n");
-    return ERROR(GENERIC);
+    return ERROR(parameter_outOfBound);
   }
   if (nbSamples == 0) {
     DISPLAYLEVEL(1, "Cover must have at least one input file\n");
-    return ERROR(GENERIC);
+    return ERROR(srcSize_wrong);
   }
   if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
     DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
@@ -741,15 +741,18 @@ ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
     return ERROR(dstSize_tooSmall);
   }
   /* Initialize context and activeDmers */
-  if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
-                      parameters.d, parameters.splitPoint)) {
-    return ERROR(GENERIC);
+  {
+    size_t const initVal = COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+                      parameters.d, parameters.splitPoint);
+    if (ZSTD_isError(initVal)) {
+      return initVal;
+    }
   }
   COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.suffixSize, g_displayLevel);
   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);
+    return ERROR(memory_allocation);
   }
 
   DISPLAYLEVEL(2, "Building dictionary\n");
@@ -810,7 +813,7 @@ size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
         cctx, dst, dstCapacity, samples + offsets[i],
         samplesSizes[i], cdict);
     if (ZSTD_isError(size)) {
-      totalCompressedSize = ERROR(GENERIC);
+      totalCompressedSize = size;
       goto _compressCleanup;
     }
     totalCompressedSize += size;
@@ -886,9 +889,11 @@ void COVER_best_start(COVER_best_t *best) {
  * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
  * If this dictionary is the best so far save it and its parameters.
  */
-void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
-                              ZDICT_cover_params_t parameters, void *dict,
-                              size_t dictSize) {
+void COVER_best_finish(COVER_best_t *best, ZDICT_cover_params_t parameters,
+                              COVER_dictSelection_t selection) {
+  void* dict = selection.dictContent;
+  size_t compressedSize = selection.totalCompressedSize;
+  size_t dictSize = selection.dictSize;
   if (!best) {
     return;
   }
@@ -914,6 +919,9 @@ void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
         }
       }
       /* Save the dictionary, parameters, and size */
+      if (!dict) {
+        return;
+      }
       memcpy(best->dict, dict, dictSize);
       best->dictSize = dictSize;
       best->parameters = parameters;
@@ -926,6 +934,111 @@ void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
   }
 }
 
+COVER_dictSelection_t COVER_dictSelectionError(size_t error) {
+    COVER_dictSelection_t selection = { NULL, 0, error };
+    return selection;
+}
+
+unsigned COVER_dictSelectionIsError(COVER_dictSelection_t selection) {
+  return (ZSTD_isError(selection.totalCompressedSize) || !selection.dictContent);
+}
+
+void COVER_dictSelectionFree(COVER_dictSelection_t selection){
+  free(selection.dictContent);
+}
+
+COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
+        size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
+        size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize) {
+
+  size_t largestDict = 0;
+  size_t largestCompressed = 0;
+  BYTE* customDictContentEnd = customDictContent + dictContentSize;
+
+  BYTE * largestDictbuffer = (BYTE *)malloc(dictContentSize);
+  BYTE * candidateDictBuffer = (BYTE *)malloc(dictContentSize);
+  double regressionTolerance = ((double)params.shrinkDictMaxRegression / 100.0) + 1.00;
+
+  if (!largestDictbuffer || !candidateDictBuffer) {
+    free(largestDictbuffer);
+    free(candidateDictBuffer);
+    return COVER_dictSelectionError(dictContentSize);
+  }
+
+  /* Initial dictionary size and compressed size */
+  memcpy(largestDictbuffer, customDictContent, dictContentSize);
+  dictContentSize = ZDICT_finalizeDictionary(
+    largestDictbuffer, dictContentSize, customDictContent, dictContentSize,
+    samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
+
+  if (ZDICT_isError(dictContentSize)) {
+    free(largestDictbuffer);
+    free(candidateDictBuffer);
+    return COVER_dictSelectionError(dictContentSize);
+  }
+
+  totalCompressedSize = COVER_checkTotalCompressedSize(params, samplesSizes,
+                                                       samplesBuffer, offsets,
+                                                       nbCheckSamples, nbSamples,
+                                                       largestDictbuffer, dictContentSize);
+
+  if (ZSTD_isError(totalCompressedSize)) {
+    free(largestDictbuffer);
+    free(candidateDictBuffer);
+    return COVER_dictSelectionError(totalCompressedSize);
+  }
+
+  if (params.shrinkDict == 0) {
+    COVER_dictSelection_t selection = { largestDictbuffer, dictContentSize, totalCompressedSize };
+    free(candidateDictBuffer);
+    return selection;
+  }
+
+  largestDict = dictContentSize;
+  largestCompressed = totalCompressedSize;
+  dictContentSize = ZDICT_DICTSIZE_MIN;
+
+  /* Largest dict is initially at least ZDICT_DICTSIZE_MIN */
+  while (dictContentSize < largestDict) {
+    memcpy(candidateDictBuffer, largestDictbuffer, largestDict);
+    dictContentSize = ZDICT_finalizeDictionary(
+      candidateDictBuffer, dictContentSize, customDictContentEnd - dictContentSize, dictContentSize,
+      samplesBuffer, samplesSizes, nbFinalizeSamples, params.zParams);
+
+    if (ZDICT_isError(dictContentSize)) {
+      free(largestDictbuffer);
+      free(candidateDictBuffer);
+      return COVER_dictSelectionError(dictContentSize);
+
+    }
+
+    totalCompressedSize = COVER_checkTotalCompressedSize(params, samplesSizes,
+                                                         samplesBuffer, offsets,
+                                                         nbCheckSamples, nbSamples,
+                                                         candidateDictBuffer, dictContentSize);
+
+    if (ZSTD_isError(totalCompressedSize)) {
+      free(largestDictbuffer);
+      free(candidateDictBuffer);
+      return COVER_dictSelectionError(totalCompressedSize);
+    }
+
+    if (totalCompressedSize <= largestCompressed * regressionTolerance) {
+      COVER_dictSelection_t selection = { candidateDictBuffer, dictContentSize, totalCompressedSize };
+      free(largestDictbuffer);
+      return selection;
+    }
+    dictContentSize *= 2;
+  }
+  dictContentSize = largestDict;
+  totalCompressedSize = largestCompressed;
+  {
+    COVER_dictSelection_t selection = { largestDictbuffer, dictContentSize, totalCompressedSize };
+    free(candidateDictBuffer);
+    return selection;
+  }
+}
+
 /**
  * Parameters for COVER_tryParameters().
  */
@@ -951,6 +1064,7 @@ static void COVER_tryParameters(void *opaque) {
   /* Allocate space for hash table, dict, and freqs */
   COVER_map_t activeDmers;
   BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
+  COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
   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");
@@ -966,29 +1080,21 @@ static void COVER_tryParameters(void *opaque) {
   {
     const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
                                               dictBufferCapacity, parameters);
-    dictBufferCapacity = ZDICT_finalizeDictionary(
-        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbTrainSamples,
-        parameters.zParams);
-    if (ZDICT_isError(dictBufferCapacity)) {
-      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
+        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbTrainSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
+        totalCompressedSize);
+
+    if (COVER_dictSelectionIsError(selection)) {
+      DISPLAYLEVEL(1, "Failed to select dictionary\n");
       goto _cleanup;
     }
   }
-  /* Check total compressed size */
-  totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
-                                                       ctx->samples, ctx->offsets,
-                                                       ctx->nbTrainSamples, ctx->nbSamples,
-                                                       dict, dictBufferCapacity);
-
 _cleanup:
-  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
-                    dictBufferCapacity);
+  free(dict);
+  COVER_best_finish(data->best, parameters, selection);
   free(data);
   COVER_map_destroy(&activeDmers);
-  if (dict) {
-    free(dict);
-  }
+  COVER_dictSelectionFree(selection);
   if (freqs) {
     free(freqs);
   }
@@ -1010,6 +1116,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
   const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
   const unsigned kIterations =
       (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
+  const unsigned shrinkDict = 0;
   /* Local variables */
   const int displayLevel = parameters->zParams.notificationLevel;
   unsigned iteration = 1;
@@ -1022,15 +1129,15 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
   /* Checks */
   if (splitPoint <= 0 || splitPoint > 1) {
     LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
-    return ERROR(GENERIC);
+    return ERROR(parameter_outOfBound);
   }
   if (kMinK < kMaxD || kMaxK < kMinK) {
     LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
-    return ERROR(GENERIC);
+    return ERROR(parameter_outOfBound);
   }
   if (nbSamples == 0) {
     DISPLAYLEVEL(1, "Cover must have at least one input file\n");
-    return ERROR(GENERIC);
+    return ERROR(srcSize_wrong);
   }
   if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
     DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
@@ -1054,11 +1161,14 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
     /* 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, splitPoint)) {
-      LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
-      COVER_best_destroy(&best);
-      POOL_free(pool);
-      return ERROR(GENERIC);
+    {
+      const size_t initVal = COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint);
+      if (ZSTD_isError(initVal)) {
+        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+        COVER_best_destroy(&best);
+        POOL_free(pool);
+        return initVal;
+      }
     }
     if (!warned) {
       COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.suffixSize, displayLevel);
@@ -1075,7 +1185,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
         COVER_best_destroy(&best);
         COVER_ctx_destroy(&ctx);
         POOL_free(pool);
-        return ERROR(GENERIC);
+        return ERROR(memory_allocation);
       }
       data->ctx = &ctx;
       data->best = &best;
@@ -1085,6 +1195,7 @@ ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
       data->parameters.d = d;
       data->parameters.splitPoint = splitPoint;
       data->parameters.steps = kSteps;
+      data->parameters.shrinkDict = shrinkDict;
       data->parameters.zParams.notificationLevel = g_displayLevel;
       /* Check the parameters */
       if (!COVER_checkParameters(data->parameters, dictBufferCapacity)) {
diff --git a/lib/dictBuilder/cover.h b/lib/dictBuilder/cover.h
index efb46807c7851..d9e0636a65981 100644
--- a/lib/dictBuilder/cover.h
+++ b/lib/dictBuilder/cover.h
@@ -47,6 +47,15 @@ typedef struct {
 } COVER_epoch_info_t;
 
 /**
+ * Struct used for the dictionary selection function.
+ */
+typedef struct COVER_dictSelection {
+  BYTE* dictContent;
+  size_t dictSize;
+  size_t totalCompressedSize;
+} COVER_dictSelection_t;
+
+/**
  * Computes the number of epochs and the size of each epoch.
  * We will make sure that each epoch gets at least 10 * k bytes.
  *
@@ -107,6 +116,32 @@ void COVER_best_start(COVER_best_t *best);
  * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
  * If this dictionary is the best so far save it and its parameters.
  */
-void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
-                       ZDICT_cover_params_t parameters, void *dict,
-                       size_t dictSize);
+void COVER_best_finish(COVER_best_t *best, ZDICT_cover_params_t parameters,
+                       COVER_dictSelection_t selection);
+/**
+ * Error function for COVER_selectDict function. Checks if the return
+ * value is an error.
+ */
+unsigned COVER_dictSelectionIsError(COVER_dictSelection_t selection);
+
+ /**
+  * Error function for COVER_selectDict function. Returns a struct where
+  * return.totalCompressedSize is a ZSTD error.
+  */
+COVER_dictSelection_t COVER_dictSelectionError(size_t error);
+
+/**
+ * Always call after selectDict is called to free up used memory from
+ * newly created dictionary.
+ */
+void COVER_dictSelectionFree(COVER_dictSelection_t selection);
+
+/**
+ * Called to finalize the dictionary and select one based on whether or not
+ * the shrink-dict flag was enabled. If enabled the dictionary used is the
+ * smallest dictionary within a specified regression of the compressed size
+ * from the largest dictionary.
+ */
+ COVER_dictSelection_t COVER_selectDict(BYTE* customDictContent,
+                       size_t dictContentSize, const BYTE* samplesBuffer, const size_t* samplesSizes, unsigned nbFinalizeSamples,
+                       size_t nbCheckSamples, size_t nbSamples, ZDICT_cover_params_t params, size_t* offsets, size_t totalCompressedSize);
diff --git a/lib/dictBuilder/fastcover.c b/lib/dictBuilder/fastcover.c
index 5b6b941a90829..941bb5a26ae66 100644
--- a/lib/dictBuilder/fastcover.c
+++ b/lib/dictBuilder/fastcover.c
@@ -287,10 +287,10 @@ FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
  * 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.
+ * Returns 0 on success or error code on error.
  * The context must be destroyed with `FASTCOVER_ctx_destroy()`.
  */
-static int
+static size_t
 FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
                    const void* samplesBuffer,
                    const size_t* samplesSizes, unsigned nbSamples,
@@ -310,19 +310,19 @@ FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
         totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
         DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
                     (unsigned)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
-        return 0;
+        return ERROR(srcSize_wrong);
     }
 
     /* Check if there are at least 5 training samples */
     if (nbTrainSamples < 5) {
         DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
-        return 0;
+        return ERROR(srcSize_wrong);
     }
 
     /* Check if there's testing sample */
     if (nbTestSamples < 1) {
         DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
-        return 0;
+        return ERROR(srcSize_wrong);
     }
 
     /* Zero the context */
@@ -347,7 +347,7 @@ FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
     if (ctx->offsets == NULL) {
         DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
         FASTCOVER_ctx_destroy(ctx);
-        return 0;
+        return ERROR(memory_allocation);
     }
 
     /* Fill offsets from the samplesSizes */
@@ -364,13 +364,13 @@ FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
     if (ctx->freqs == NULL) {
         DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
         FASTCOVER_ctx_destroy(ctx);
-        return 0;
+        return ERROR(memory_allocation);
     }
 
     DISPLAYLEVEL(2, "Computing frequencies\n");
     FASTCOVER_computeFrequency(ctx->freqs, ctx);
 
-    return 1;
+    return 0;
 }
 
 
@@ -435,7 +435,6 @@ FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
   return tail;
 }
 
-
 /**
  * Parameters for FASTCOVER_tryParameters().
  */
@@ -464,6 +463,7 @@ static void FASTCOVER_tryParameters(void *opaque)
   U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
   /* Allocate space for hash table, dict, and freqs */
   BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
+  COVER_dictSelection_t selection = COVER_dictSelectionError(ERROR(GENERIC));
   U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
   if (!segmentFreqs || !dict || !freqs) {
     DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
@@ -473,27 +473,24 @@ static void FASTCOVER_tryParameters(void *opaque)
   memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
   /* Build the dictionary */
   { const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
-                                                  parameters, segmentFreqs);
+                                                    parameters, segmentFreqs);
+
     const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
-    dictBufferCapacity = ZDICT_finalizeDictionary(
-        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        ctx->samples, ctx->samplesSizes, nbFinalizeSamples, parameters.zParams);
-    if (ZDICT_isError(dictBufferCapacity)) {
-      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
+    selection = COVER_selectDict(dict + tail, dictBufferCapacity - tail,
+         ctx->samples, ctx->samplesSizes, nbFinalizeSamples, ctx->nbTrainSamples, ctx->nbSamples, parameters, ctx->offsets,
+         totalCompressedSize);
+
+    if (COVER_dictSelectionIsError(selection)) {
+      DISPLAYLEVEL(1, "Failed to select dictionary\n");
       goto _cleanup;
     }
   }
-  /* Check total compressed size */
-  totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
-                                                       ctx->samples, ctx->offsets,
-                                                       ctx->nbTrainSamples, ctx->nbSamples,
-                                                       dict, dictBufferCapacity);
 _cleanup:
-  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
-                    dictBufferCapacity);
+  free(dict);
+  COVER_best_finish(data->best, parameters, selection);
   free(data);
   free(segmentFreqs);
-  free(dict);
+  COVER_dictSelectionFree(selection);
   free(freqs);
 }
 
@@ -508,6 +505,7 @@ FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
     coverParams->nbThreads = fastCoverParams.nbThreads;
     coverParams->splitPoint = fastCoverParams.splitPoint;
     coverParams->zParams = fastCoverParams.zParams;
+    coverParams->shrinkDict = fastCoverParams.shrinkDict;
 }
 
 
@@ -524,6 +522,7 @@ FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
     fastCoverParams->f = f;
     fastCoverParams->accel = accel;
     fastCoverParams->zParams = coverParams.zParams;
+    fastCoverParams->shrinkDict = coverParams.shrinkDict;
 }
 
 
@@ -550,11 +549,11 @@ ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
     if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
                                    parameters.accel)) {
       DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
-      return ERROR(GENERIC);
+      return ERROR(parameter_outOfBound);
     }
     if (nbSamples == 0) {
       DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
-      return ERROR(GENERIC);
+      return ERROR(srcSize_wrong);
     }
     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
       DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
@@ -564,11 +563,14 @@ ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
     /* Assign corresponding FASTCOVER_accel_t to accelParams*/
     accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
     /* Initialize context */
-    if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
+    {
+      size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
                             coverParams.d, parameters.splitPoint, parameters.f,
-                            accelParams)) {
-      DISPLAYLEVEL(1, "Failed to initialize context\n");
-      return ERROR(GENERIC);
+                            accelParams);
+      if (ZSTD_isError(initVal)) {
+        DISPLAYLEVEL(1, "Failed to initialize context\n");
+        return initVal;
+      }
     }
     COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, g_displayLevel);
     /* Build the dictionary */
@@ -616,6 +618,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
         (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
     const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
     const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
+    const unsigned shrinkDict = 0;
     /* Local variables */
     const int displayLevel = parameters->zParams.notificationLevel;
     unsigned iteration = 1;
@@ -627,19 +630,19 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
     /* Checks */
     if (splitPoint <= 0 || splitPoint > 1) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
-      return ERROR(GENERIC);
+      return ERROR(parameter_outOfBound);
     }
     if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
-      return ERROR(GENERIC);
+      return ERROR(parameter_outOfBound);
     }
     if (kMinK < kMaxD || kMaxK < kMinK) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
-      return ERROR(GENERIC);
+      return ERROR(parameter_outOfBound);
     }
     if (nbSamples == 0) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
-      return ERROR(GENERIC);
+      return ERROR(srcSize_wrong);
     }
     if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
       LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
@@ -666,11 +669,14 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
       /* Initialize the context for this value of d */
       FASTCOVER_ctx_t ctx;
       LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
-      if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams)) {
-        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
-        COVER_best_destroy(&best);
-        POOL_free(pool);
-        return ERROR(GENERIC);
+      {
+        size_t const initVal = FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams);
+        if (ZSTD_isError(initVal)) {
+          LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
+          COVER_best_destroy(&best);
+          POOL_free(pool);
+          return initVal;
+        }
       }
       if (!warned) {
         COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
@@ -687,7 +693,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
           COVER_best_destroy(&best);
           FASTCOVER_ctx_destroy(&ctx);
           POOL_free(pool);
-          return ERROR(GENERIC);
+          return ERROR(memory_allocation);
         }
         data->ctx = &ctx;
         data->best = &best;
@@ -697,6 +703,7 @@ ZDICT_optimizeTrainFromBuffer_fastCover(
         data->parameters.d = d;
         data->parameters.splitPoint = splitPoint;
         data->parameters.steps = kSteps;
+        data->parameters.shrinkDict = shrinkDict;
         data->parameters.zParams.notificationLevel = g_displayLevel;
         /* Check the parameters */
         if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
diff --git a/lib/dictBuilder/zdict.c b/lib/dictBuilder/zdict.c
index c753da0dbbfe7..ee21ee1a9d34b 100644
--- a/lib/dictBuilder/zdict.c
+++ b/lib/dictBuilder/zdict.c
@@ -741,7 +741,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     /* analyze, build stats, starting with literals */
     {   size_t maxNbBits = HUF_buildCTable (hufTable, countLit, 255, huffLog);
         if (HUF_isError(maxNbBits)) {
-            eSize = ERROR(GENERIC);
+            eSize = maxNbBits;
             DISPLAYLEVEL(1, " HUF_buildCTable error \n");
             goto _cleanup;
         }
@@ -764,7 +764,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
     errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax);
     if (FSE_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
+        eSize = errorCode;
         DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
         goto _cleanup;
     }
@@ -773,7 +773,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
     errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML);
     if (FSE_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
+        eSize = errorCode;
         DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
         goto _cleanup;
     }
@@ -782,7 +782,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
     errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL);
     if (FSE_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
+        eSize = errorCode;
         DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
         goto _cleanup;
     }
@@ -791,7 +791,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     /* write result to buffer */
     {   size_t const hhSize = HUF_writeCTable(dstPtr, maxDstSize, hufTable, 255, huffLog);
         if (HUF_isError(hhSize)) {
-            eSize = ERROR(GENERIC);
+            eSize = hhSize;
             DISPLAYLEVEL(1, "HUF_writeCTable error \n");
             goto _cleanup;
         }
@@ -802,7 +802,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
 
     {   size_t const ohSize = FSE_writeNCount(dstPtr, maxDstSize, offcodeNCount, OFFCODE_MAX, Offlog);
         if (FSE_isError(ohSize)) {
-            eSize = ERROR(GENERIC);
+            eSize = ohSize;
             DISPLAYLEVEL(1, "FSE_writeNCount error with offcodeNCount \n");
             goto _cleanup;
         }
@@ -813,7 +813,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
 
     {   size_t const mhSize = FSE_writeNCount(dstPtr, maxDstSize, matchLengthNCount, MaxML, mlLog);
         if (FSE_isError(mhSize)) {
-            eSize = ERROR(GENERIC);
+            eSize = mhSize;
             DISPLAYLEVEL(1, "FSE_writeNCount error with matchLengthNCount \n");
             goto _cleanup;
         }
@@ -824,7 +824,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
 
     {   size_t const lhSize = FSE_writeNCount(dstPtr, maxDstSize, litLengthNCount, MaxLL, llLog);
         if (FSE_isError(lhSize)) {
-            eSize = ERROR(GENERIC);
+            eSize = lhSize;
             DISPLAYLEVEL(1, "FSE_writeNCount error with litlengthNCount \n");
             goto _cleanup;
         }
@@ -834,7 +834,7 @@ static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
     }
 
     if (maxDstSize<12) {
-        eSize = ERROR(GENERIC);
+        eSize = ERROR(dstSize_tooSmall);
         DISPLAYLEVEL(1, "not enough space to write RepOffsets \n");
         goto _cleanup;
     }
diff --git a/lib/dictBuilder/zdict.h b/lib/dictBuilder/zdict.h
index e22973173cb89..37978ecdfb818 100644
--- a/lib/dictBuilder/zdict.h
+++ b/lib/dictBuilder/zdict.h
@@ -94,6 +94,8 @@ typedef struct {
     unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (40) : 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 */
     double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (1.0), 1.0 when all samples are used for both training and testing */
+    unsigned shrinkDict;         /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking  */
+    unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
     ZDICT_params_t zParams;
 } ZDICT_cover_params_t;
 
@@ -105,6 +107,9 @@ typedef struct {
     unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
     double splitPoint;           /* Percentage of samples used for training: Only used for optimization : the first nbSamples * splitPoint samples will be used to training, the last nbSamples * (1 - splitPoint) samples will be used for testing, 0 means default (0.75), 1.0 when all samples are used for both training and testing */
     unsigned accel;              /* Acceleration level: constraint: 0 < accel <= 10, higher means faster and less accurate, 0 means default(1) */
+    unsigned shrinkDict;         /* Train dictionaries to shrink in size starting from the minimum size and selects the smallest dictionary that is shrinkDictMaxRegression% worse than the largest dictionary. 0 means no shrinking and 1 means shrinking  */
+    unsigned shrinkDictMaxRegression; /* Sets shrinkDictMaxRegression so that a smaller dictionary can be at worse shrinkDictMaxRegression% worse than the max dict size dictionary. */
+
     ZDICT_params_t zParams;
 } ZDICT_fastCover_params_t;
 
diff --git a/lib/legacy/zstd_legacy.h b/lib/legacy/zstd_legacy.h
index e5b383ee4c0ce..0dbd3c7a40f5d 100644
--- a/lib/legacy/zstd_legacy.h
+++ b/lib/legacy/zstd_legacy.h
@@ -238,6 +238,10 @@ MEM_STATIC ZSTD_frameSizeInfo ZSTD_findFrameSizeInfoLegacy(const void *src, size
             frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
             break;
     }
+    if (!ZSTD_isError(frameSizeInfo.compressedSize) && frameSizeInfo.compressedSize > srcSize) {
+        frameSizeInfo.compressedSize = ERROR(srcSize_wrong);
+        frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
+    }
     return frameSizeInfo;
 }
 
diff --git a/lib/legacy/zstd_v01.c b/lib/legacy/zstd_v01.c
index cad2b99b4a732..ae8cba2a3fa20 100644
--- a/lib/legacy/zstd_v01.c
+++ b/lib/legacy/zstd_v01.c
@@ -1073,99 +1073,102 @@ static size_t HUF_decompress_usingDTable(   /* -3% slower when non static */
     const void* cSrc, size_t cSrcSize,
     const U16* DTable)
 {
-    BYTE* const ostart = (BYTE*) dst;
-    BYTE* op = ostart;
-    BYTE* const omax = op + maxDstSize;
-    BYTE* const olimit = omax-15;
-
-    const void* ptr = DTable;
-    const HUF_DElt* const dt = (const HUF_DElt*)(ptr)+1;
-    const U32 dtLog = DTable[0];
-    size_t errorCode;
-    U32 reloadStatus;
-
-    /* Init */
-
-    const U16* jumpTable = (const U16*)cSrc;
-    const size_t length1 = FSE_readLE16(jumpTable);
-    const size_t length2 = FSE_readLE16(jumpTable+1);
-    const size_t length3 = FSE_readLE16(jumpTable+2);
-    const size_t length4 = cSrcSize - 6 - length1 - length2 - length3;   // check coherency !!
-    const char* const start1 = (const char*)(cSrc) + 6;
-    const char* const start2 = start1 + length1;
-    const char* const start3 = start2 + length2;
-    const char* const start4 = start3 + length3;
-    FSE_DStream_t bitD1, bitD2, bitD3, bitD4;
-
-    if (length1+length2+length3+6 >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
-
-    errorCode = FSE_initDStream(&bitD1, start1, length1);
-    if (FSE_isError(errorCode)) return errorCode;
-    errorCode = FSE_initDStream(&bitD2, start2, length2);
-    if (FSE_isError(errorCode)) return errorCode;
-    errorCode = FSE_initDStream(&bitD3, start3, length3);
-    if (FSE_isError(errorCode)) return errorCode;
-    errorCode = FSE_initDStream(&bitD4, start4, length4);
-    if (FSE_isError(errorCode)) return errorCode;
-
-    reloadStatus=FSE_reloadDStream(&bitD2);
-
-    /* 16 symbols per loop */
-    for ( ; (reloadStatus<FSE_DStream_completed) && (op<olimit);  /* D2-3-4 are supposed to be synchronized and finish together */
-        op+=16, reloadStatus = FSE_reloadDStream(&bitD2) | FSE_reloadDStream(&bitD3) | FSE_reloadDStream(&bitD4), FSE_reloadDStream(&bitD1))
+    if (cSrcSize < 6) return (size_t)-FSE_ERROR_srcSize_wrong;
     {
-#define HUF_DECODE_SYMBOL_0(n, Dstream) \
-        op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog);
-
-#define HUF_DECODE_SYMBOL_1(n, Dstream) \
-        op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
-        if (FSE_32bits() && (HUF_MAX_TABLELOG>12)) FSE_reloadDStream(&Dstream)
-
-#define HUF_DECODE_SYMBOL_2(n, Dstream) \
-        op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
-        if (FSE_32bits()) FSE_reloadDStream(&Dstream)
-
-        HUF_DECODE_SYMBOL_1( 0, bitD1);
-        HUF_DECODE_SYMBOL_1( 1, bitD2);
-        HUF_DECODE_SYMBOL_1( 2, bitD3);
-        HUF_DECODE_SYMBOL_1( 3, bitD4);
-        HUF_DECODE_SYMBOL_2( 4, bitD1);
-        HUF_DECODE_SYMBOL_2( 5, bitD2);
-        HUF_DECODE_SYMBOL_2( 6, bitD3);
-        HUF_DECODE_SYMBOL_2( 7, bitD4);
-        HUF_DECODE_SYMBOL_1( 8, bitD1);
-        HUF_DECODE_SYMBOL_1( 9, bitD2);
-        HUF_DECODE_SYMBOL_1(10, bitD3);
-        HUF_DECODE_SYMBOL_1(11, bitD4);
-        HUF_DECODE_SYMBOL_0(12, bitD1);
-        HUF_DECODE_SYMBOL_0(13, bitD2);
-        HUF_DECODE_SYMBOL_0(14, bitD3);
-        HUF_DECODE_SYMBOL_0(15, bitD4);
-    }
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* op = ostart;
+        BYTE* const omax = op + maxDstSize;
+        BYTE* const olimit = omax-15;
+
+        const void* ptr = DTable;
+        const HUF_DElt* const dt = (const HUF_DElt*)(ptr)+1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
+        U32 reloadStatus;
+
+        /* Init */
+
+        const U16* jumpTable = (const U16*)cSrc;
+        const size_t length1 = FSE_readLE16(jumpTable);
+        const size_t length2 = FSE_readLE16(jumpTable+1);
+        const size_t length3 = FSE_readLE16(jumpTable+2);
+        const size_t length4 = cSrcSize - 6 - length1 - length2 - length3;   // check coherency !!
+        const char* const start1 = (const char*)(cSrc) + 6;
+        const char* const start2 = start1 + length1;
+        const char* const start3 = start2 + length2;
+        const char* const start4 = start3 + length3;
+        FSE_DStream_t bitD1, bitD2, bitD3, bitD4;
+
+        if (length1+length2+length3+6 >= cSrcSize) return (size_t)-FSE_ERROR_srcSize_wrong;
+
+        errorCode = FSE_initDStream(&bitD1, start1, length1);
+        if (FSE_isError(errorCode)) return errorCode;
+        errorCode = FSE_initDStream(&bitD2, start2, length2);
+        if (FSE_isError(errorCode)) return errorCode;
+        errorCode = FSE_initDStream(&bitD3, start3, length3);
+        if (FSE_isError(errorCode)) return errorCode;
+        errorCode = FSE_initDStream(&bitD4, start4, length4);
+        if (FSE_isError(errorCode)) return errorCode;
+
+        reloadStatus=FSE_reloadDStream(&bitD2);
+
+        /* 16 symbols per loop */
+        for ( ; (reloadStatus<FSE_DStream_completed) && (op<olimit);  /* D2-3-4 are supposed to be synchronized and finish together */
+            op+=16, reloadStatus = FSE_reloadDStream(&bitD2) | FSE_reloadDStream(&bitD3) | FSE_reloadDStream(&bitD4), FSE_reloadDStream(&bitD1))
+        {
+    #define HUF_DECODE_SYMBOL_0(n, Dstream) \
+            op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog);
+
+    #define HUF_DECODE_SYMBOL_1(n, Dstream) \
+            op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
+            if (FSE_32bits() && (HUF_MAX_TABLELOG>12)) FSE_reloadDStream(&Dstream)
+
+    #define HUF_DECODE_SYMBOL_2(n, Dstream) \
+            op[n] = HUF_decodeSymbol(&Dstream, dt, dtLog); \
+            if (FSE_32bits()) FSE_reloadDStream(&Dstream)
+
+            HUF_DECODE_SYMBOL_1( 0, bitD1);
+            HUF_DECODE_SYMBOL_1( 1, bitD2);
+            HUF_DECODE_SYMBOL_1( 2, bitD3);
+            HUF_DECODE_SYMBOL_1( 3, bitD4);
+            HUF_DECODE_SYMBOL_2( 4, bitD1);
+            HUF_DECODE_SYMBOL_2( 5, bitD2);
+            HUF_DECODE_SYMBOL_2( 6, bitD3);
+            HUF_DECODE_SYMBOL_2( 7, bitD4);
+            HUF_DECODE_SYMBOL_1( 8, bitD1);
+            HUF_DECODE_SYMBOL_1( 9, bitD2);
+            HUF_DECODE_SYMBOL_1(10, bitD3);
+            HUF_DECODE_SYMBOL_1(11, bitD4);
+            HUF_DECODE_SYMBOL_0(12, bitD1);
+            HUF_DECODE_SYMBOL_0(13, bitD2);
+            HUF_DECODE_SYMBOL_0(14, bitD3);
+            HUF_DECODE_SYMBOL_0(15, bitD4);
+        }
 
-    if (reloadStatus!=FSE_DStream_completed)   /* not complete : some bitStream might be FSE_DStream_unfinished */
-        return (size_t)-FSE_ERROR_corruptionDetected;
+        if (reloadStatus!=FSE_DStream_completed)   /* not complete : some bitStream might be FSE_DStream_unfinished */
+            return (size_t)-FSE_ERROR_corruptionDetected;
 
-    /* tail */
-    {
-        // bitTail = bitD1;   // *much* slower : -20% !??!
-        FSE_DStream_t bitTail;
-        bitTail.ptr = bitD1.ptr;
-        bitTail.bitsConsumed = bitD1.bitsConsumed;
-        bitTail.bitContainer = bitD1.bitContainer;   // required in case of FSE_DStream_endOfBuffer
-        bitTail.start = start1;
-        for ( ; (FSE_reloadDStream(&bitTail) < FSE_DStream_completed) && (op<omax) ; op++)
+        /* tail */
         {
-            HUF_DECODE_SYMBOL_0(0, bitTail);
-        }
+            // bitTail = bitD1;   // *much* slower : -20% !??!
+            FSE_DStream_t bitTail;
+            bitTail.ptr = bitD1.ptr;
+            bitTail.bitsConsumed = bitD1.bitsConsumed;
+            bitTail.bitContainer = bitD1.bitContainer;   // required in case of FSE_DStream_endOfBuffer
+            bitTail.start = start1;
+            for ( ; (FSE_reloadDStream(&bitTail) < FSE_DStream_completed) && (op<omax) ; op++)
+            {
+                HUF_DECODE_SYMBOL_0(0, bitTail);
+            }
 
-        if (FSE_endOfDStream(&bitTail))
-            return op-ostart;
-    }
+            if (FSE_endOfDStream(&bitTail))
+                return op-ostart;
+        }
 
-    if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall;   /* dst buffer is full, but cSrc unfinished */
+        if (op==omax) return (size_t)-FSE_ERROR_dstSize_tooSmall;   /* dst buffer is full, but cSrc unfinished */
 
-    return (size_t)-FSE_ERROR_corruptionDetected;
+        return (size_t)-FSE_ERROR_corruptionDetected;
+    }
 }
 
 
@@ -1355,8 +1358,6 @@ static unsigned ZSTD_isLittleEndian(void)
 
 static U16    ZSTD_read16(const void* p) { U16 r; memcpy(&r, p, sizeof(r)); return r; }
 
-static U32    ZSTD_read32(const void* p) { U32 r; memcpy(&r, p, sizeof(r)); return r; }
-
 static void   ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
 
 static void   ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
@@ -1381,16 +1382,9 @@ static U16 ZSTD_readLE16(const void* memPtr)
     }
 }
 
-
-static U32 ZSTD_readLE32(const void* memPtr)
+static U32 ZSTD_readLE24(const void* memPtr)
 {
-    if (ZSTD_isLittleEndian())
-        return ZSTD_read32(memPtr);
-    else
-    {
-        const BYTE* p = (const BYTE*)memPtr;
-        return (U32)((U32)p[0] + ((U32)p[1]<<8) + ((U32)p[2]<<16) + ((U32)p[3]<<24));
-    }
+    return ZSTD_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
 }
 
 static U32 ZSTD_readBE32(const void* memPtr)
@@ -1704,13 +1698,13 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     seqState->prevOffset = seq->offset;
     if (litLength == MaxLL)
     {
-        U32 add = dumps<de ? *dumps++ : 0;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) litLength += add;
         else
         {
             if (dumps<=(de-3))
             {
-                litLength = ZSTD_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+                litLength = ZSTD_readLE24(dumps);
                 dumps += 3;
             }
         }
@@ -1732,13 +1726,13 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
     if (matchLength == MaxML)
     {
-        U32 add = dumps<de ? *dumps++ : 0;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) matchLength += add;
         else
         {
             if (dumps<=(de-3))
             {
-                matchLength = ZSTD_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+                matchLength = ZSTD_readLE24(dumps);
                 dumps += 3;
             }
         }
diff --git a/lib/legacy/zstd_v02.c b/lib/legacy/zstd_v02.c
index 561bc412e9a8e..793df6024bbd0 100644
--- a/lib/legacy/zstd_v02.c
+++ b/lib/legacy/zstd_v02.c
@@ -217,6 +217,11 @@ MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
     }
 }
 
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
 MEM_STATIC U32 MEM_readLE32(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -3043,11 +3048,11 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     seqState->prevOffset = seq->offset;
     if (litLength == MaxLL)
     {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) litLength += add;
-        else
+        else if (dumps + 3 <= de)
         {
-            litLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            litLength = MEM_readLE24(dumps);
             dumps += 3;
         }
         if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
@@ -3073,11 +3078,11 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
     if (matchLength == MaxML)
     {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) matchLength += add;
-        else
+        else if (dumps + 3 <= de)
         {
-            matchLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            matchLength = MEM_readLE24(dumps);
             dumps += 3;
         }
         if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
diff --git a/lib/legacy/zstd_v03.c b/lib/legacy/zstd_v03.c
index a1bf0fa9b5c8b..7a0e7c9b69fb0 100644
--- a/lib/legacy/zstd_v03.c
+++ b/lib/legacy/zstd_v03.c
@@ -219,6 +219,11 @@ MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
     }
 }
 
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
 MEM_STATIC U32 MEM_readLE32(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -2684,11 +2689,11 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     seqState->prevOffset = seq->offset;
     if (litLength == MaxLL)
     {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) litLength += add;
-        else
+        else if (dumps + 3 <= de)
         {
-            litLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            litLength = MEM_readLE24(dumps);
             dumps += 3;
         }
         if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
@@ -2714,11 +2719,11 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
     if (matchLength == MaxML)
     {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) matchLength += add;
-        else
+        else if (dumps + 3 <= de)
         {
-            matchLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+            matchLength = MEM_readLE24(dumps);
             dumps += 3;
         }
         if (dumps >= de) dumps = de-1;   /* late correction, to avoid read overflow (data is now corrupted anyway) */
diff --git a/lib/legacy/zstd_v04.c b/lib/legacy/zstd_v04.c
index 4342330e2553d..645a6e313c05f 100644
--- a/lib/legacy/zstd_v04.c
+++ b/lib/legacy/zstd_v04.c
@@ -189,6 +189,11 @@ MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
     }
 }
 
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
 MEM_STATIC U32 MEM_readLE32(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -2808,13 +2813,12 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     litLength = FSE_decodeSymbol(&(seqState->stateLL), &(seqState->DStream));
     prevOffset = litLength ? seq->offset : seqState->prevOffset;
     if (litLength == MaxLL) {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) litLength += add;
-        else {
-            litLength = dumps[0] + (dumps[1]<<8) + (dumps[2]<<16);
+        else if (dumps + 3 <= de) {
+            litLength = MEM_readLE24(dumps);
             dumps += 3;
         }
-        if (dumps > de) { litLength = MaxLL+255; }  /* late correction, to avoid using uninitialized memory */
         if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
     }
 
@@ -2837,13 +2841,12 @@ static void ZSTD_decodeSequence(seq_t* seq, seqState_t* seqState)
     /* MatchLength */
     matchLength = FSE_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
     if (matchLength == MaxML) {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) matchLength += add;
-        else {
-            matchLength = dumps[0] + (dumps[1]<<8) + (dumps[2]<<16);
+        else if (dumps + 3 <= de){
+            matchLength = MEM_readLE24(dumps);
             dumps += 3;
         }
-        if (dumps > de) { matchLength = MaxML+255; }  /* late correction, to avoid using uninitialized memory */
         if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
     }
     matchLength += MINMATCH;
diff --git a/lib/legacy/zstd_v05.c b/lib/legacy/zstd_v05.c
index caaf15f9bdb1f..a7ea6066fb083 100644
--- a/lib/legacy/zstd_v05.c
+++ b/lib/legacy/zstd_v05.c
@@ -218,6 +218,11 @@ MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
     }
 }
 
+MEM_STATIC U32 MEM_readLE24(const void* memPtr)
+{
+    return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
+}
+
 MEM_STATIC U32 MEM_readLE32(const void* memPtr)
 {
     if (MEM_isLittleEndian())
@@ -1998,91 +2003,92 @@ size_t HUFv05_decompress4X2_usingDTable(
     const void* cSrc, size_t cSrcSize,
     const U16* DTable)
 {
-    const BYTE* const istart = (const BYTE*) cSrc;
-    BYTE* const ostart = (BYTE*) dst;
-    BYTE* const oend = ostart + dstSize;
-    const void* const dtPtr = DTable;
-    const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
-    const U32 dtLog = DTable[0];
-    size_t errorCode;
-
-    /* Init */
-    BITv05_DStream_t bitD1;
-    BITv05_DStream_t bitD2;
-    BITv05_DStream_t bitD3;
-    BITv05_DStream_t bitD4;
-    const size_t length1 = MEM_readLE16(istart);
-    const size_t length2 = MEM_readLE16(istart+2);
-    const size_t length3 = MEM_readLE16(istart+4);
-    size_t length4;
-    const BYTE* const istart1 = istart + 6;  /* jumpTable */
-    const BYTE* const istart2 = istart1 + length1;
-    const BYTE* const istart3 = istart2 + length2;
-    const BYTE* const istart4 = istart3 + length3;
-    const size_t segmentSize = (dstSize+3) / 4;
-    BYTE* const opStart2 = ostart + segmentSize;
-    BYTE* const opStart3 = opStart2 + segmentSize;
-    BYTE* const opStart4 = opStart3 + segmentSize;
-    BYTE* op1 = ostart;
-    BYTE* op2 = opStart2;
-    BYTE* op3 = opStart3;
-    BYTE* op4 = opStart4;
-    U32 endSignal;
-
     /* Check */
     if (cSrcSize < 10) return ERROR(corruption_detected);   /* strict minimum : jump table + 1 byte per stream */
+    {
+        const BYTE* const istart = (const BYTE*) cSrc;
+        BYTE* const ostart = (BYTE*) dst;
+        BYTE* const oend = ostart + dstSize;
+        const void* const dtPtr = DTable;
+        const HUFv05_DEltX2* const dt = ((const HUFv05_DEltX2*)dtPtr) +1;
+        const U32 dtLog = DTable[0];
+        size_t errorCode;
 
-    length4 = cSrcSize - (length1 + length2 + length3 + 6);
-    if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-    errorCode = BITv05_initDStream(&bitD1, istart1, length1);
-    if (HUFv05_isError(errorCode)) return errorCode;
-    errorCode = BITv05_initDStream(&bitD2, istart2, length2);
-    if (HUFv05_isError(errorCode)) return errorCode;
-    errorCode = BITv05_initDStream(&bitD3, istart3, length3);
-    if (HUFv05_isError(errorCode)) return errorCode;
-    errorCode = BITv05_initDStream(&bitD4, istart4, length4);
-    if (HUFv05_isError(errorCode)) return errorCode;
+        /* Init */
+        BITv05_DStream_t bitD1;
+        BITv05_DStream_t bitD2;
+        BITv05_DStream_t bitD3;
+        BITv05_DStream_t bitD4;
+        const size_t length1 = MEM_readLE16(istart);
+        const size_t length2 = MEM_readLE16(istart+2);
+        const size_t length3 = MEM_readLE16(istart+4);
+        size_t length4;
+        const BYTE* const istart1 = istart + 6;  /* jumpTable */
+        const BYTE* const istart2 = istart1 + length1;
+        const BYTE* const istart3 = istart2 + length2;
+        const BYTE* const istart4 = istart3 + length3;
+        const size_t segmentSize = (dstSize+3) / 4;
+        BYTE* const opStart2 = ostart + segmentSize;
+        BYTE* const opStart3 = opStart2 + segmentSize;
+        BYTE* const opStart4 = opStart3 + segmentSize;
+        BYTE* op1 = ostart;
+        BYTE* op2 = opStart2;
+        BYTE* op3 = opStart3;
+        BYTE* op4 = opStart4;
+        U32 endSignal;
 
-    /* 16-32 symbols per loop (4-8 symbols per stream) */
-    endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
-    for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
-        HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
-        HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
-        HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
-        HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
-        HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
-        HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
-        HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
-        HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
-        HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
-        HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
-        HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
-        HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
-        HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
-        HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
-        HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
-        HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
+        length4 = cSrcSize - (length1 + length2 + length3 + 6);
+        if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
+        errorCode = BITv05_initDStream(&bitD1, istart1, length1);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD2, istart2, length2);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD3, istart3, length3);
+        if (HUFv05_isError(errorCode)) return errorCode;
+        errorCode = BITv05_initDStream(&bitD4, istart4, length4);
+        if (HUFv05_isError(errorCode)) return errorCode;
+
+        /* 16-32 symbols per loop (4-8 symbols per stream) */
         endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
-    }
+        for ( ; (endSignal==BITv05_DStream_unfinished) && (op4<(oend-7)) ; ) {
+            HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUFv05_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUFv05_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUFv05_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUFv05_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = BITv05_reloadDStream(&bitD1) | BITv05_reloadDStream(&bitD2) | BITv05_reloadDStream(&bitD3) | BITv05_reloadDStream(&bitD4);
+        }
 
-    /* check corruption */
-    if (op1 > opStart2) return ERROR(corruption_detected);
-    if (op2 > opStart3) return ERROR(corruption_detected);
-    if (op3 > opStart4) return ERROR(corruption_detected);
-    /* note : op4 supposed already verified within main loop */
+        /* check corruption */
+        if (op1 > opStart2) return ERROR(corruption_detected);
+        if (op2 > opStart3) return ERROR(corruption_detected);
+        if (op3 > opStart4) return ERROR(corruption_detected);
+        /* note : op4 supposed already verified within main loop */
 
-    /* finish bitStreams one by one */
-    HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
-    HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
-    HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
-    HUFv05_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+        /* finish bitStreams one by one */
+        HUFv05_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUFv05_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUFv05_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUFv05_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
 
-    /* check */
-    endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
-    if (!endSignal) return ERROR(corruption_detected);
+        /* check */
+        endSignal = BITv05_endOfDStream(&bitD1) & BITv05_endOfDStream(&bitD2) & BITv05_endOfDStream(&bitD3) & BITv05_endOfDStream(&bitD4);
+        if (!endSignal) return ERROR(corruption_detected);
 
-    /* decoded size */
-    return dstSize;
+        /* decoded size */
+        return dstSize;
+    }
 }
 
 
@@ -3150,14 +3156,13 @@ static void ZSTDv05_decodeSequence(seq_t* seq, seqState_t* seqState)
     litLength = FSEv05_peakSymbol(&(seqState->stateLL));
     prevOffset = litLength ? seq->offset : seqState->prevOffset;
     if (litLength == MaxLL) {
-        U32 add = *dumps++;
+        const U32 add = *dumps++;
         if (add < 255) litLength += add;
-        else {
-            litLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no risk : dumps is always followed by seq tables > 1 byte */
+        else if (dumps + 3 <= de) {
+            litLength = MEM_readLE24(dumps);
             if (litLength&1) litLength>>=1, dumps += 3;
             else litLength = (U16)(litLength)>>1, dumps += 2;
         }
-        if (dumps > de) { litLength = MaxLL+255; }  /* late correction, to avoid using uninitialized memory */
         if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
     }
 
@@ -3184,14 +3189,13 @@ static void ZSTDv05_decodeSequence(seq_t* seq, seqState_t* seqState)
     /* MatchLength */
     matchLength = FSEv05_decodeSymbol(&(seqState->stateML), &(seqState->DStream));
     if (matchLength == MaxML) {
-        U32 add = *dumps++;
+        const U32 add = dumps<de ? *dumps++ : 0;
         if (add < 255) matchLength += add;
-        else {
-            matchLength = MEM_readLE32(dumps) & 0xFFFFFF;  /* no pb : dumps is always followed by seq tables > 1 byte */
+        else if (dumps + 3 <= de) {
+            matchLength = MEM_readLE24(dumps);
             if (matchLength&1) matchLength>>=1, dumps += 3;
             else matchLength = (U16)(matchLength)>>1, dumps += 2;
         }
-        if (dumps > de) { matchLength = MaxML+255; }  /* late correction, to avoid using uninitialized memory */
         if (dumps >= de) { dumps = de-1; }  /* late correction, to avoid read overflow (data is now corrupted anyway) */
     }
     matchLength += MINMATCH;
diff --git a/lib/legacy/zstd_v06.c b/lib/legacy/zstd_v06.c
index a695cbb8a6dde..f907a3a712261 100644
--- a/lib/legacy/zstd_v06.c
+++ b/lib/legacy/zstd_v06.c
@@ -3242,14 +3242,12 @@ static size_t ZSTDv06_decodeSeqHeaders(int* nbSeqPtr,
     }
 
     /* FSE table descriptors */
+    if (ip + 4 > iend) return ERROR(srcSize_wrong); /* min : header byte + all 3 are "raw", hence no header, but at least xxLog bits per type */
     {   U32 const LLtype  = *ip >> 6;
         U32 const Offtype = (*ip >> 4) & 3;
         U32 const MLtype  = (*ip >> 2) & 3;
         ip++;
 
-        /* check */
-        if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
-
         /* Build DTables */
         {   size_t const bhSize = ZSTDv06_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
             if (ZSTDv06_isError(bhSize)) return ERROR(corruption_detected);
@@ -3672,7 +3670,7 @@ void ZSTDv06_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cS
     blockProperties_t blockProperties = { bt_compressed, 0 };
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, ZSTDv06_frameHeaderSize_min);
+    {   size_t const frameHeaderSize = ZSTDv06_frameHeaderSize(src, srcSize);
         if (ZSTDv06_isError(frameHeaderSize)) {
             ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, frameHeaderSize);
             return;
diff --git a/lib/legacy/zstd_v07.c b/lib/legacy/zstd_v07.c
index 6b9488931f35c..a83ddc9a68b0a 100644
--- a/lib/legacy/zstd_v07.c
+++ b/lib/legacy/zstd_v07.c
@@ -3470,14 +3470,12 @@ static size_t ZSTDv07_decodeSeqHeaders(int* nbSeqPtr,
     }
 
     /* FSE table descriptors */
+    if (ip + 4 > iend) return ERROR(srcSize_wrong); /* min : header byte + all 3 are "raw", hence no header, but at least xxLog bits per type */
     {   U32 const LLtype  = *ip >> 6;
         U32 const OFtype = (*ip >> 4) & 3;
         U32 const MLtype  = (*ip >> 2) & 3;
         ip++;
 
-        /* check */
-        if (ip > iend-3) return ERROR(srcSize_wrong); /* min : all 3 are "raw", hence no header, but at least xxLog bits per type */
-
         /* Build DTables */
         {   size_t const llhSize = ZSTDv07_buildSeqTable(DTableLL, LLtype, MaxLL, LLFSELog, ip, iend-ip, LL_defaultNorm, LL_defaultNormLog, flagRepeatTable);
             if (ZSTDv07_isError(llhSize)) return ERROR(corruption_detected);
@@ -3918,7 +3916,7 @@ void ZSTDv07_findFrameSizeInfoLegacy(const void *src, size_t srcSize, size_t* cS
     }
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, ZSTDv07_frameHeaderSize_min);
+    {   size_t const frameHeaderSize = ZSTDv07_frameHeaderSize(src, srcSize);
         if (ZSTDv07_isError(frameHeaderSize)) {
             ZSTD_errorFrameSizeInfoLegacy(cSize, dBound, frameHeaderSize);
             return;
diff --git a/lib/zstd.h b/lib/zstd.h
index 53470c18f3675..a1910ee223e92 100644
--- a/lib/zstd.h
+++ b/lib/zstd.h
@@ -71,7 +71,7 @@ extern "C" {
 /*------   Version   ------*/
 #define ZSTD_VERSION_MAJOR    1
 #define ZSTD_VERSION_MINOR    4
-#define ZSTD_VERSION_RELEASE  0
+#define ZSTD_VERSION_RELEASE  1
 
 #define ZSTD_VERSION_NUMBER  (ZSTD_VERSION_MAJOR *100*100 + ZSTD_VERSION_MINOR *100 + ZSTD_VERSION_RELEASE)
 ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< to check runtime library version */
@@ -82,16 +82,16 @@ ZSTDLIB_API unsigned ZSTD_versionNumber(void);   /**< to check runtime library v
 #define ZSTD_VERSION_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_LIB_VERSION)
 ZSTDLIB_API const char* ZSTD_versionString(void);   /* requires v1.3.0+ */
 
-/***************************************
-*  Default constant
-***************************************/
+/* *************************************
+ *  Default constant
+ ***************************************/
 #ifndef ZSTD_CLEVEL_DEFAULT
 #  define ZSTD_CLEVEL_DEFAULT 3
 #endif
 
-/***************************************
-*  Constants
-***************************************/
+/* *************************************
+ *  Constants
+ ***************************************/
 
 /* All magic numbers are supposed read/written to/from files/memory using little-endian convention */
 #define ZSTD_MAGICNUMBER            0xFD2FB528    /* valid since v0.8.0 */
@@ -183,9 +183,14 @@ ZSTDLIB_API int         ZSTD_maxCLevel(void);               /*!< maximum compres
 ***************************************/
 /*= Compression context
  *  When compressing many times,
- *  it is recommended to allocate a context just once, and re-use it for each successive compression operation.
+ *  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. */
+ *  Note : re-using context is just a speed / resource optimization.
+ *         It doesn't change the compression ratio, which remains identical.
+ *  Note 2 : In multi-threaded environments,
+ *         use one different context per thread for parallel execution.
+ */
 typedef struct ZSTD_CCtx_s ZSTD_CCtx;
 ZSTDLIB_API ZSTD_CCtx* ZSTD_createCCtx(void);
 ZSTDLIB_API size_t     ZSTD_freeCCtx(ZSTD_CCtx* cctx);
@@ -380,6 +385,7 @@ typedef enum {
      * ZSTD_c_forceMaxWindow
      * ZSTD_c_forceAttachDict
      * ZSTD_c_literalCompressionMode
+     * ZSTD_c_targetCBlockSize
      * Because they are not stable, it's necessary to define ZSTD_STATIC_LINKING_ONLY to access them.
      * note : never ever use experimentalParam? names directly;
      *        also, the enums values themselves are unstable and can still change.
@@ -389,6 +395,7 @@ typedef enum {
      ZSTD_c_experimentalParam3=1000,
      ZSTD_c_experimentalParam4=1001,
      ZSTD_c_experimentalParam5=1002,
+     ZSTD_c_experimentalParam6=1003,
 } ZSTD_cParameter;
 
 typedef struct {
@@ -657,17 +664,33 @@ ZSTDLIB_API size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
                                          ZSTD_inBuffer* input,
                                          ZSTD_EndDirective endOp);
 
+
+/* These buffer sizes are softly recommended.
+ * They are not required : ZSTD_compressStream*() happily accepts any buffer size, for both input and output.
+ * Respecting the recommended size just makes it a bit easier for ZSTD_compressStream*(),
+ * reducing the amount of memory shuffling and buffering, resulting in minor performance savings.
+ *
+ * However, note that these recommendations are from the perspective of a C caller program.
+ * If the streaming interface is invoked from some other language,
+ * especially managed ones such as Java or Go, through a foreign function interface such as jni or cgo,
+ * a major performance rule is to reduce crossing such interface to an absolute minimum.
+ * It's not rare that performance ends being spent more into the interface, rather than compression itself.
+ * In which cases, prefer using large buffers, as large as practical,
+ * for both input and output, to reduce the nb of roundtrips.
+ */
 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. */
+ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output buffer. Guarantee to successfully flush at least one complete compressed block. */
 
-/*******************************************************************************
- * This is a legacy streaming API, and can be replaced by ZSTD_CCtx_reset() and
- * ZSTD_compressStream2(). It is redundant, but is still fully supported.
+
+/* *****************************************************************************
+ * This following is a legacy streaming API.
+ * It can be replaced by ZSTD_CCtx_reset() and ZSTD_compressStream2().
+ * It is redundant, but remains fully supported.
  * Advanced parameters and dictionary compression can only be used through the
  * new API.
  ******************************************************************************/
 
-/**
+/*!
  * Equivalent to:
  *
  *     ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
@@ -675,16 +698,16 @@ ZSTDLIB_API size_t ZSTD_CStreamOutSize(void);   /**< recommended size for output
  *     ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel);
  */
 ZSTDLIB_API size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel);
-/**
+/*!
  * Alternative for ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue).
  * NOTE: The return value is different. ZSTD_compressStream() returns a hint for
  * the next read size (if non-zero and not an error). ZSTD_compressStream2()
- * returns the number of bytes left to flush (if non-zero and not an error).
+ * returns the minimum nb of bytes left to flush (if non-zero and not an error).
  */
 ZSTDLIB_API size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
-/** Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_flush). */
+/*! Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_flush). */
 ZSTDLIB_API size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
-/** Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_end). */
+/*! Equivalent to ZSTD_compressStream2(zcs, output, &emptyInput, ZSTD_e_end). */
 ZSTDLIB_API size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output);
 
 
@@ -969,7 +992,7 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #endif  /* ZSTD_H_235446 */
 
 
-/****************************************************************************************
+/* **************************************************************************************
  *   ADVANCED AND EXPERIMENTAL FUNCTIONS
  ****************************************************************************************
  * The definitions in the following section are considered experimental.
@@ -1037,6 +1060,10 @@ ZSTDLIB_API size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict);
 #define ZSTD_LDM_HASHRATELOG_MIN     0
 #define ZSTD_LDM_HASHRATELOG_MAX (ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN)
 
+/* Advanced parameter bounds */
+#define ZSTD_TARGETCBLOCKSIZE_MIN   64
+#define ZSTD_TARGETCBLOCKSIZE_MAX   ZSTD_BLOCKSIZE_MAX
+
 /* internal */
 #define ZSTD_HASHLOG3_MAX           17
 
@@ -1162,7 +1189,7 @@ typedef enum {
  *            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);
 
-/** ZSTD_decompressBound() :
+/*! ZSTD_decompressBound() :
  *  `src` should point to 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 at `src + srcSize`)
@@ -1409,6 +1436,11 @@ ZSTDLIB_API size_t ZSTD_CCtx_refPrefix_advanced(ZSTD_CCtx* cctx, const void* pre
  */
 #define ZSTD_c_literalCompressionMode ZSTD_c_experimentalParam5
 
+/* Tries to fit compressed block size to be around targetCBlockSize.
+ * No target when targetCBlockSize == 0.
+ * There is no guarantee on compressed block size (default:0) */
+#define ZSTD_c_targetCBlockSize ZSTD_c_experimentalParam6
+
 /*! ZSTD_CCtx_getParameter() :
  *  Get the requested compression parameter value, selected by enum ZSTD_cParameter,
  *  and store it into int* value.
@@ -1843,7 +1875,7 @@ typedef struct {
     unsigned checksumFlag;
 } ZSTD_frameHeader;
 
-/** ZSTD_getFrameHeader() :
+/*! ZSTD_getFrameHeader() :
  *  decode Frame Header, or requires larger `srcSize`.
  * @return : 0, `zfhPtr` is correctly filled,
  *          >0, `srcSize` is too small, value is wanted `srcSize` amount,