extzstd 0.3.1 → 0.3.2

data/contrib/zstd/lib/compress/zstd_compress_internal.h

@@ -28,7 +28,6 @@
 extern "C" {
 #endif
 
-
 /*-*************************************
 *  Constants
 ***************************************/
@@ -64,7 +63,7 @@ typedef struct {
 } ZSTD_localDict;
 
 typedef struct {
-    U32 CTable[HUF_CTABLE_SIZE_U32(255)];
+    HUF_CElt CTable[HUF_CTABLE_SIZE_U32(255)];
     HUF_repeat repeatMode;
 } ZSTD_hufCTables_t;
 
@@ -83,10 +82,27 @@ typedef struct {
 } ZSTD_entropyCTables_t;
 
 typedef struct {
-    U32 off;
-    U32 len;
+    U32 off;            /* Offset code (offset + ZSTD_REP_MOVE) for the match */
+    U32 len;            /* Raw length of match */
 } ZSTD_match_t;
 
+typedef struct {
+    U32 offset;         /* Offset of sequence */
+    U32 litLength;      /* Length of literals prior to match */
+    U32 matchLength;    /* Raw length of match */
+} rawSeq;
+
+typedef struct {
+  rawSeq* seq;          /* The start of the sequences */
+  size_t pos;           /* The index in seq where reading stopped. pos <= size. */
+  size_t posInSequence; /* The position within the sequence at seq[pos] where reading
+                           stopped. posInSequence <= seq[pos].litLength + seq[pos].matchLength */
+  size_t size;          /* The number of sequences. <= capacity. */
+  size_t capacity;      /* The capacity starting from `seq` pointer */
+} rawSeqStore_t;
+
+UNUSED_ATTR static const rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0, 0};
+
 typedef struct {
     int price;
     U32 off;
@@ -147,9 +163,13 @@ struct ZSTD_matchState_t {
     U32* hashTable;
     U32* hashTable3;
     U32* chainTable;
+    int dedicatedDictSearch;  /* Indicates whether this matchState is using the
+                               * dedicated dictionary search structure.
+                               */
     optState_t opt;         /* optimal parser state */
     const ZSTD_matchState_t* dictMatchState;
     ZSTD_compressionParameters cParams;
+    const rawSeqStore_t* ldmSeqStore;
 };
 
 typedef struct {
@@ -181,19 +201,6 @@ typedef struct {
     U32 windowLog;          /* Window log for the LDM */
 } ldmParams_t;
 
-typedef struct {
-    U32 offset;
-    U32 litLength;
-    U32 matchLength;
-} rawSeq;
-
-typedef struct {
-  rawSeq* seq;     /* The start of the sequences */
-  size_t pos;      /* The position where reading stopped. <= size. */
-  size_t size;     /* The number of sequences. <= capacity. */
-  size_t capacity; /* The capacity starting from `seq` pointer */
-} rawSeqStore_t;
-
 typedef struct {
     int collectSequences;
     ZSTD_Sequence* seqStart;
@@ -228,10 +235,34 @@ struct ZSTD_CCtx_params_s {
     /* Long distance matching parameters */
     ldmParams_t ldmParams;
 
+    /* Dedicated dict search algorithm trigger */
+    int enableDedicatedDictSearch;
+
+    /* Input/output buffer modes */
+    ZSTD_bufferMode_e inBufferMode;
+    ZSTD_bufferMode_e outBufferMode;
+
+    /* Sequence compression API */
+    ZSTD_sequenceFormat_e blockDelimiters;
+    int validateSequences;
+
     /* Internal use, for createCCtxParams() and freeCCtxParams() only */
     ZSTD_customMem customMem;
 };  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
 
+#define COMPRESS_SEQUENCES_WORKSPACE_SIZE (sizeof(unsigned) * (MaxSeq + 2))
+#define ENTROPY_WORKSPACE_SIZE (HUF_WORKSPACE_SIZE + COMPRESS_SEQUENCES_WORKSPACE_SIZE)
+
+/**
+ * Indicates whether this compression proceeds directly from user-provided
+ * source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
+ * whether the context needs to buffer the input/output (ZSTDb_buffered).
+ */
+typedef enum {
+    ZSTDb_not_buffered,
+    ZSTDb_buffered
+} ZSTD_buffered_policy_e;
+
 struct ZSTD_CCtx_s {
     ZSTD_compressionStage_e stage;
     int cParamsChanged;                  /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
@@ -247,6 +278,7 @@ struct ZSTD_CCtx_s {
     unsigned long long producedCSize;
     XXH64_state_t xxhState;
     ZSTD_customMem customMem;
+    ZSTD_threadPool* pool;
     size_t staticSize;
     SeqCollector seqCollector;
     int isFirstBlock;
@@ -258,7 +290,10 @@ struct ZSTD_CCtx_s {
     size_t maxNbLdmSequences;
     rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
     ZSTD_blockState_t blockState;
-    U32* entropyWorkspace;  /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+    U32* entropyWorkspace;  /* entropy workspace of ENTROPY_WORKSPACE_SIZE bytes */
+
+    /* Wether we are streaming or not */
+    ZSTD_buffered_policy_e bufferedPolicy;
 
     /* streaming */
     char*  inBuff;
@@ -273,6 +308,10 @@ struct ZSTD_CCtx_s {
     ZSTD_cStreamStage streamStage;
     U32    frameEnded;
 
+    /* Stable in/out buffer verification */
+    ZSTD_inBuffer expectedInBuffer;
+    size_t expectedOutBufferSize;
+
     /* Dictionary */
     ZSTD_localDict localDict;
     const ZSTD_CDict* cdict;
@@ -286,8 +325,32 @@ struct ZSTD_CCtx_s {
 
 typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
 
-typedef enum { ZSTD_noDict = 0, ZSTD_extDict = 1, ZSTD_dictMatchState = 2 } ZSTD_dictMode_e;
-
+typedef enum {
+    ZSTD_noDict = 0,
+    ZSTD_extDict = 1,
+    ZSTD_dictMatchState = 2,
+    ZSTD_dedicatedDictSearch = 3
+} ZSTD_dictMode_e;
+
+typedef enum {
+    ZSTD_cpm_noAttachDict = 0,  /* Compression with ZSTD_noDict or ZSTD_extDict.
+                                 * In this mode we use both the srcSize and the dictSize
+                                 * when selecting and adjusting parameters.
+                                 */
+    ZSTD_cpm_attachDict = 1,    /* Compression with ZSTD_dictMatchState or ZSTD_dedicatedDictSearch.
+                                 * In this mode we only take the srcSize into account when selecting
+                                 * and adjusting parameters.
+                                 */
+    ZSTD_cpm_createCDict = 2,   /* Creating a CDict.
+                                 * In this mode we take both the source size and the dictionary size
+                                 * into account when selecting and adjusting the parameters.
+                                 */
+    ZSTD_cpm_unknown = 3,       /* ZSTD_getCParams, ZSTD_getParams, ZSTD_adjustParams.
+                                 * We don't know what these parameters are for. We default to the legacy
+                                 * behavior of taking both the source size and the dict size into account
+                                 * when selecting and adjusting parameters.
+                                 */
+} ZSTD_cParamMode_e;
 
 typedef size_t (*ZSTD_blockCompressor) (
         ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
@@ -345,7 +408,7 @@ MEM_STATIC repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 con
             newReps.rep[1] = rep[0];
             newReps.rep[0] = currentOffset;
         } else {   /* repCode == 0 */
-            memcpy(&newReps, rep, sizeof(newReps));
+            ZSTD_memcpy(&newReps, rep, sizeof(newReps));
         }
     }
     return newReps;
@@ -372,7 +435,7 @@ MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const voi
     RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
                     dstSize_tooSmall, "dst buf too small for uncompressed block");
     MEM_writeLE24(dst, cBlockHeader24);
-    memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
+    ZSTD_memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
     return ZSTD_blockHeaderSize + srcSize;
 }
 
@@ -498,8 +561,12 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
     if (MEM_isLittleEndian()) {
         if (MEM_64bits()) {
 #       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
+#           if STATIC_BMI2
+                return _tzcnt_u64(val) >> 3;
+#           else
+                unsigned long r = 0;
+                return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
+#           endif
 #       elif defined(__GNUC__) && (__GNUC__ >= 4)
             return (__builtin_ctzll((U64)val) >> 3);
 #       else
@@ -530,8 +597,12 @@ static unsigned ZSTD_NbCommonBytes (size_t val)
     } else {  /* Big Endian CPU */
         if (MEM_64bits()) {
 #       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            return _BitScanReverse64( &r, val ) ? (unsigned)(r >> 3) : 0;
+#           if STATIC_BMI2
+			    return _lzcnt_u64(val) >> 3;
+#           else
+			    unsigned long r = 0;
+			    return _BitScanReverse64(&r, (U64)val) ? (unsigned)(r >> 3) : 0;
+#           endif
 #       elif defined(__GNUC__) && (__GNUC__ >= 4)
             return (__builtin_clzll(val) >> 3);
 #       else
@@ -626,7 +697,8 @@ static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
 static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
 static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
 
-MEM_STATIC size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
+MEM_STATIC FORCE_INLINE_ATTR
+size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
 {
     switch(mls)
     {
@@ -742,7 +814,7 @@ MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
     return ZSTD_window_hasExtDict(ms->window) ?
         ZSTD_extDict :
         ms->dictMatchState != NULL ?
-            ZSTD_dictMatchState :
+            (ms->dictMatchState->dedicatedDictSearch ? ZSTD_dedicatedDictSearch : ZSTD_dictMatchState) :
             ZSTD_noDict;
 }
 
@@ -754,8 +826,8 @@ MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
 MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
                                                   void const* srcEnd)
 {
-    U32 const current = (U32)((BYTE const*)srcEnd - window.base);
-    return current > ZSTD_CURRENT_MAX;
+    U32 const curr = (U32)((BYTE const*)srcEnd - window.base);
+    return curr > ZSTD_CURRENT_MAX;
 }
 
 /**
@@ -791,14 +863,14 @@ MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
      *    windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
      */
     U32 const cycleMask = (1U << cycleLog) - 1;
-    U32 const current = (U32)((BYTE const*)src - window->base);
-    U32 const currentCycle0 = current & cycleMask;
+    U32 const curr = (U32)((BYTE const*)src - window->base);
+    U32 const currentCycle0 = curr & cycleMask;
     /* Exclude zero so that newCurrent - maxDist >= 1. */
     U32 const currentCycle1 = currentCycle0 == 0 ? (1U << cycleLog) : currentCycle0;
     U32 const newCurrent = currentCycle1 + maxDist;
-    U32 const correction = current - newCurrent;
+    U32 const correction = curr - newCurrent;
     assert((maxDist & cycleMask) == 0);
-    assert(current > newCurrent);
+    assert(curr > newCurrent);
     /* Loose bound, should be around 1<<29 (see above) */
     assert(correction > 1<<28);
 
@@ -919,7 +991,7 @@ ZSTD_checkDictValidity(const ZSTD_window_t* window,
 }
 
 MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
-    memset(window, 0, sizeof(*window));
+    ZSTD_memset(window, 0, sizeof(*window));
     window->base = (BYTE const*)"";
     window->dictBase = (BYTE const*)"";
     window->dictLimit = 1;    /* start from 1, so that 1st position is valid */
@@ -973,12 +1045,16 @@ MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
 /**
  * Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
  */
-MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
+MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
 {
     U32    const maxDistance = 1U << windowLog;
     U32    const lowestValid = ms->window.lowLimit;
-    U32    const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    U32    const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
     U32    const isDictionary = (ms->loadedDictEnd != 0);
+    /* When using a dictionary the entire dictionary is valid if a single byte of the dictionary
+     * is within the window. We invalidate the dictionary (and set loadedDictEnd to 0) when it isn't
+     * valid for the entire block. So this check is sufficient to find the lowest valid match index.
+     */
     U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
     return matchLowest;
 }
@@ -986,12 +1062,15 @@ MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current
 /**
  * Returns the lowest allowed match index in the prefix.
  */
-MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
+MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 curr, unsigned windowLog)
 {
     U32    const maxDistance = 1U << windowLog;
     U32    const lowestValid = ms->window.dictLimit;
-    U32    const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    U32    const withinWindow = (curr - lowestValid > maxDistance) ? curr - maxDistance : lowestValid;
     U32    const isDictionary = (ms->loadedDictEnd != 0);
+    /* When computing the lowest prefix index we need to take the dictionary into account to handle
+     * the edge case where the dictionary and the source are contiguous in memory.
+     */
     U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
     return matchLowest;
 }
@@ -1045,7 +1124,6 @@ MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
  * assumptions : magic number supposed already checked
  *               and dictSize >= 8 */
 size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
-                         short* offcodeNCount, unsigned* offcodeMaxValue,
                          const void* const dict, size_t dictSize);
 
 void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
@@ -1061,7 +1139,7 @@ void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
  * Note: srcSizeHint == 0 means 0!
  */
 ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
-        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize);
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize, ZSTD_cParamMode_e mode);
 
 /*! ZSTD_initCStream_internal() :
  *  Private use only. Init streaming operation.

data/contrib/zstd/lib/compress/zstd_compress_literals.c

@@ -35,7 +35,7 @@ size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src,
             assert(0);
     }
 
-    memcpy(ostart + flSize, src, srcSize);
+    ZSTD_memcpy(ostart + flSize, src, srcSize);
     DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
     return srcSize + flSize;
 }
@@ -86,7 +86,7 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
                 disableLiteralCompression, (U32)srcSize);
 
     /* Prepare nextEntropy assuming reusing the existing table */
-    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
 
     if (disableLiteralCompression)
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
@@ -118,11 +118,11 @@ size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
     }
 
     if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
-        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
         return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
     }
     if (cLitSize==1) {
-        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+        ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
         return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
     }
 

data/contrib/zstd/lib/compress/zstd_compress_sequences.c

@@ -50,6 +50,19 @@ static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
   return maxSymbolValue;
 }
 
+/**
+ * Returns true if we should use ncount=-1 else we should
+ * use ncount=1 for low probability symbols instead.
+ */
+static unsigned ZSTD_useLowProbCount(size_t const nbSeq)
+{
+    /* Heuristic: This should cover most blocks <= 16K and
+     * start to fade out after 16K to about 32K depending on
+     * comprssibility.
+     */
+    return nbSeq >= 2048;
+}
+
 /**
  * Returns the cost in bytes of encoding the normalized count header.
  * Returns an error if any of the helper functions return an error.
@@ -60,7 +73,7 @@ static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
     BYTE wksp[FSE_NCOUNTBOUND];
     S16 norm[MaxSeq + 1];
     const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
-    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max), "");
+    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max, ZSTD_useLowProbCount(nbSeq)), "");
     return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
 }
 
@@ -239,7 +252,7 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
         *op = codeTable[0];
         return 1;
     case set_repeat:
-        memcpy(nextCTable, prevCTable, prevCTableSize);
+        ZSTD_memcpy(nextCTable, prevCTable, prevCTableSize);
         return 0;
     case set_basic:
         FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), "");  /* note : could be pre-calculated */
@@ -253,7 +266,8 @@ ZSTD_buildCTable(void* dst, size_t dstCapacity,
             nbSeq_1--;
         }
         assert(nbSeq_1 > 1);
-        FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max), "");
+        assert(entropyWorkspaceSize >= FSE_BUILD_CTABLE_WORKSPACE_SIZE(MaxSeq, MaxFSELog));
+        FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max, ZSTD_useLowProbCount(nbSeq_1)), "");
         {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
             FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
             FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, entropyWorkspace, entropyWorkspaceSize), "");

data/contrib/zstd/lib/compress/zstd_compress_superblock.c

@@ -29,7 +29,7 @@
  *  This metadata is populated in ZSTD_buildSuperBlockEntropy_literal() */
 typedef struct {
     symbolEncodingType_e hType;
-    BYTE hufDesBuffer[500]; /* TODO give name to this value */
+    BYTE hufDesBuffer[ZSTD_MAX_HUF_HEADER_SIZE];
     size_t hufDesSize;
 } ZSTD_hufCTablesMetadata_t;
 
@@ -42,7 +42,7 @@ typedef struct {
     symbolEncodingType_e llType;
     symbolEncodingType_e ofType;
     symbolEncodingType_e mlType;
-    BYTE fseTablesBuffer[500]; /* TODO give name to this value */
+    BYTE fseTablesBuffer[ZSTD_MAX_FSE_HEADERS_SIZE];
     size_t fseTablesSize;
     size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_compressSubBlock_sequences() */
 } ZSTD_fseCTablesMetadata_t;
@@ -79,7 +79,7 @@ static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSiz
     DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_literal (srcSize=%zu)", srcSize);
 
     /* Prepare nextEntropy assuming reusing the existing table */
-    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
 
     if (disableLiteralsCompression) {
         DEBUGLOG(5, "set_basic - disabled");
@@ -118,7 +118,7 @@ static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSiz
     }
 
     /* Build Huffman Tree */
-    memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
+    ZSTD_memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
     huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
     {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
                                                     maxSymbolValue, huffLog,
@@ -137,14 +137,14 @@ static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSiz
                         (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
                 if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
                     DEBUGLOG(5, "set_repeat - smaller");
-                    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                    ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
                     hufMetadata->hType = set_repeat;
                     return 0;
                 }
             }
             if (newCSize + hSize >= srcSize) {
                 DEBUGLOG(5, "set_basic - no gains");
-                memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                ZSTD_memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
                 hufMetadata->hType = set_basic;
                 return 0;
             }
@@ -188,7 +188,7 @@ static size_t ZSTD_buildSuperBlockEntropy_sequences(seqStore_t* seqStorePtr,
 
     assert(cTableWkspSize >= (1 << MaxFSELog) * sizeof(FSE_FUNCTION_TYPE));
     DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_sequences (nbSeq=%zu)", nbSeq);
-    memset(workspace, 0, wkspSize);
+    ZSTD_memset(workspace, 0, wkspSize);
 
     fseMetadata->lastCountSize = 0;
     /* convert length/distances into codes */
@@ -348,7 +348,7 @@ static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
     assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
 
     if (writeEntropy && hufMetadata->hType == set_compressed) {
-        memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
+        ZSTD_memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
         op += hufMetadata->hufDesSize;
         cLitSize += hufMetadata->hufDesSize;
         DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
@@ -474,7 +474,7 @@ static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables
         const U32 MLtype = fseMetadata->mlType;
         DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
         *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
-        memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
+        ZSTD_memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
         op += fseMetadata->fseTablesSize;
     } else {
         const U32 repeat = set_repeat;
@@ -603,7 +603,7 @@ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
                         const BYTE* codeTable, unsigned maxCode,
                         size_t nbSeq, const FSE_CTable* fseCTable,
                         const U32* additionalBits,
-                        short const* defaultNorm, U32 defaultNormLog,
+                        short const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
                         void* workspace, size_t wkspSize)
 {
     unsigned* const countWksp = (unsigned*)workspace;
@@ -615,7 +615,11 @@ static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
 
     HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
     if (type == set_basic) {
-        cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
+        /* We selected this encoding type, so it must be valid. */
+        assert(max <= defaultMax);
+        cSymbolTypeSizeEstimateInBits = max <= defaultMax
+                ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max)
+                : ERROR(GENERIC);
     } else if (type == set_rle) {
         cSymbolTypeSizeEstimateInBits = 0;
     } else if (type == set_compressed || type == set_repeat) {
@@ -643,15 +647,15 @@ static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
     size_t cSeqSizeEstimate = 0;
     cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
                                          nbSeq, fseTables->offcodeCTable, NULL,
-                                         OF_defaultNorm, OF_defaultNormLog,
+                                         OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
                                          workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
                                          nbSeq, fseTables->litlengthCTable, LL_bits,
-                                         LL_defaultNorm, LL_defaultNormLog,
+                                         LL_defaultNorm, LL_defaultNormLog, MaxLL,
                                          workspace, wkspSize);
     cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
                                          nbSeq, fseTables->matchlengthCTable, ML_bits,
-                                         ML_defaultNorm, ML_defaultNormLog,
+                                         ML_defaultNorm, ML_defaultNormLog, MaxML,
                                          workspace, wkspSize);
     if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
     return cSeqSizeEstimate + sequencesSectionHeaderSize;
@@ -790,7 +794,7 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
     } while (!lastSequence);
     if (writeLitEntropy) {
         DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
-        memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
+        ZSTD_memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
     }
     if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
         /* If we haven't written our entropy tables, then we've violated our contract and
@@ -809,11 +813,11 @@ static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
         if (sp < send) {
             seqDef const* seq;
             repcodes_t rep;
-            memcpy(&rep, prevCBlock->rep, sizeof(rep)); 
+            ZSTD_memcpy(&rep, prevCBlock->rep, sizeof(rep));
             for (seq = sstart; seq < sp; ++seq) {
                 rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
             }
-            memcpy(nextCBlock->rep, &rep, sizeof(rep));
+            ZSTD_memcpy(nextCBlock->rep, &rep, sizeof(rep));
         }
     }
     DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
@@ -831,7 +835,7 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
           &zc->blockState.nextCBlock->entropy,
           &zc->appliedParams,
           &entropyMetadata,
-          zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
+          zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
 
     return ZSTD_compressSubBlock_multi(&zc->seqStore,
             zc->blockState.prevCBlock,
@@ -841,5 +845,5 @@ size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
             dst, dstCapacity,
             src, srcSize,
             zc->bmi2, lastBlock,
-            zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
+            zc->entropyWorkspace, ENTROPY_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
 }