zstd-ruby 1.4.2.0 → 1.4.4.0

data/ext/zstdruby/libzstd/compress/zstd_ldm.c

@@ -49,9 +49,9 @@ size_t ZSTD_ldm_getTableSize(ldmParams_t params)
 {
     size_t const ldmHSize = ((size_t)1) << params.hashLog;
     size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
-    size_t const ldmBucketSize =
-        ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
-    size_t const totalSize = ldmBucketSize + ldmHSize * sizeof(ldmEntry_t);
+    size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
+    size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+                           + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
     return params.enableLdm ? totalSize : 0;
 }
 
@@ -583,7 +583,7 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
                 rep[i] = rep[i-1];
             rep[0] = sequence.offset;
             /* Store the sequence */
-            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength,
+            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
                           sequence.offset + ZSTD_REP_MOVE,
                           sequence.matchLength - MINMATCH);
             ip += sequence.matchLength;

data/ext/zstdruby/libzstd/compress/zstd_opt.c

@@ -552,7 +552,6 @@ U32 ZSTD_insertBtAndGetAllMatches (
 {
     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;
@@ -569,8 +568,7 @@ U32 ZSTD_insertBtAndGetAllMatches (
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
     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 windowLow = ZSTD_getLowestMatchIndex(ms, current, cParams->windowLog);
     U32 const matchLow = windowLow ? windowLow : 1;
     U32* smallerPtr = bt + 2*(current&btMask);
     U32* largerPtr  = bt + 2*(current&btMask) + 1;
@@ -674,19 +672,21 @@ U32 ZSTD_insertBtAndGetAllMatches (
 
     while (nbCompares-- && (matchIndex >= matchLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
-        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         const BYTE* match;
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
         assert(current > matchIndex);
 
         if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
             assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
             match = base + matchIndex;
+            if (matchIndex >= dictLimit) assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
             matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
         } else {
             match = dictBase + matchIndex;
+            assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
             matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
             if (matchIndex+matchLength >= dictLimit)
-                match = base + matchIndex;   /* prepare for match[matchLength] */
+                match = base + matchIndex;   /* prepare for match[matchLength] read */
         }
 
         if (matchLength > bestLength) {
@@ -1098,7 +1098,7 @@ _shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
 
                     assert(anchor + llen <= iend);
                     ZSTD_updateStats(optStatePtr, llen, anchor, offCode, mlen);
-                    ZSTD_storeSeq(seqStore, llen, anchor, offCode, mlen-MINMATCH);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offCode, mlen-MINMATCH);
                     anchor += advance;
                     ip = anchor;
             }   }

data/ext/zstdruby/libzstd/compress/zstdmt_compress.c

@@ -668,7 +668,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
 
     /* init */
     if (job->cdict) {
-        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast, job->cdict, jobParams, job->fullFrameSize);
+        size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast, job->cdict, &jobParams, job->fullFrameSize);
         assert(job->firstJob);  /* only allowed for first job */
         if (ZSTD_isError(initError)) JOB_ERROR(initError);
     } else {  /* srcStart points at reloaded section */
@@ -680,7 +680,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
                                         job->prefix.start, job->prefix.size, ZSTD_dct_rawContent, /* load dictionary in "content-only" mode (no header analysis) */
                                         ZSTD_dtlm_fast,
                                         NULL, /*cdict*/
-                                        jobParams, pledgedSrcSize);
+                                        &jobParams, pledgedSrcSize);
             if (ZSTD_isError(initError)) JOB_ERROR(initError);
     }   }
 
@@ -927,12 +927,18 @@ static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
     unsigned jobID;
     DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
     for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
+        /* Copy the mutex/cond out */
+        ZSTD_pthread_mutex_t const mutex = mtctx->jobs[jobID].job_mutex;
+        ZSTD_pthread_cond_t const cond = mtctx->jobs[jobID].job_cond;
+
         DEBUGLOG(4, "job%02u: release dst address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].dstBuff.start);
         ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
-        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
-        mtctx->jobs[jobID].cSize = 0;
+
+        /* Clear the job description, but keep the mutex/cond */
+        memset(&mtctx->jobs[jobID], 0, sizeof(mtctx->jobs[jobID]));
+        mtctx->jobs[jobID].job_mutex = mutex;
+        mtctx->jobs[jobID].job_cond = cond;
     }
-    memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
     mtctx->inBuff.buffer = g_nullBuffer;
     mtctx->inBuff.filled = 0;
     mtctx->allJobsCompleted = 1;
@@ -1028,9 +1034,9 @@ size_t ZSTDMT_getMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter,
 
 /* Sets parameters relevant to the compression job,
  * initializing others to default values. */
-static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(ZSTD_CCtx_params const params)
+static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(const ZSTD_CCtx_params* params)
 {
-    ZSTD_CCtx_params jobParams = params;
+    ZSTD_CCtx_params jobParams = *params;
     /* Clear parameters related to multithreading */
     jobParams.forceWindow = 0;
     jobParams.nbWorkers = 0;
@@ -1151,16 +1157,16 @@ size_t ZSTDMT_toFlushNow(ZSTDMT_CCtx* mtctx)
 /* =====   Multi-threaded compression   ===== */
 /* ------------------------------------------ */
 
-static unsigned ZSTDMT_computeTargetJobLog(ZSTD_CCtx_params const params)
+static unsigned ZSTDMT_computeTargetJobLog(const ZSTD_CCtx_params* params)
 {
     unsigned jobLog;
-    if (params.ldmParams.enableLdm) {
+    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. */
-        jobLog = MAX(21, params.cParams.chainLog + 4);
+        jobLog = MAX(21, params->cParams.chainLog + 4);
     } else {
-        jobLog = MAX(20, params.cParams.windowLog + 2);
+        jobLog = MAX(20, params->cParams.windowLog + 2);
     }
     return MIN(jobLog, (unsigned)ZSTDMT_JOBLOG_MAX);
 }
@@ -1193,27 +1199,27 @@ static int ZSTDMT_overlapLog(int ovlog, ZSTD_strategy strat)
     return ovlog;
 }
 
-static size_t ZSTDMT_computeOverlapSize(ZSTD_CCtx_params const params)
+static size_t ZSTDMT_computeOverlapSize(const ZSTD_CCtx_params* params)
 {
-    int const overlapRLog = 9 - ZSTDMT_overlapLog(params.overlapLog, params.cParams.strategy);
-    int ovLog = (overlapRLog >= 8) ? 0 : (params.cParams.windowLog - overlapRLog);
+    int const overlapRLog = 9 - ZSTDMT_overlapLog(params->overlapLog, params->cParams.strategy);
+    int ovLog = (overlapRLog >= 8) ? 0 : (params->cParams.windowLog - overlapRLog);
     assert(0 <= overlapRLog && overlapRLog <= 8);
-    if (params.ldmParams.enableLdm) {
+    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.
          * Then, ovLog becomes a fraction of the jobSize, rather than windowSize */
-        ovLog = MIN(params.cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2)
+        ovLog = MIN(params->cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2)
                 - overlapRLog;
     }
     assert(0 <= ovLog && ovLog <= ZSTD_WINDOWLOG_MAX);
-    DEBUGLOG(4, "overlapLog : %i", params.overlapLog);
+    DEBUGLOG(4, "overlapLog : %i", params->overlapLog);
     DEBUGLOG(4, "overlap size : %i", 1 << ovLog);
     return (ovLog==0) ? 0 : (size_t)1 << ovLog;
 }
 
 static unsigned
-ZSTDMT_computeNbJobs(ZSTD_CCtx_params params, size_t srcSize, unsigned nbWorkers)
+ZSTDMT_computeNbJobs(const ZSTD_CCtx_params* params, size_t srcSize, unsigned nbWorkers)
 {
     assert(nbWorkers>0);
     {   size_t const jobSizeTarget = (size_t)1 << ZSTDMT_computeTargetJobLog(params);
@@ -1236,9 +1242,9 @@ static size_t ZSTDMT_compress_advanced_internal(
           const ZSTD_CDict* cdict,
                 ZSTD_CCtx_params params)
 {
-    ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(params);
-    size_t const overlapSize = ZSTDMT_computeOverlapSize(params);
-    unsigned const nbJobs = ZSTDMT_computeNbJobs(params, srcSize, params.nbWorkers);
+    ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(&params);
+    size_t const overlapSize = ZSTDMT_computeOverlapSize(&params);
+    unsigned const nbJobs = ZSTDMT_computeNbJobs(&params, srcSize, params.nbWorkers);
     size_t const proposedJobSize = (srcSize + (nbJobs-1)) / nbJobs;
     size_t const avgJobSize = (((proposedJobSize-1) & 0x1FFFF) < 0x7FFF) ? proposedJobSize + 0xFFFF : proposedJobSize;   /* avoid too small last block */
     const char* const srcStart = (const char*)src;
@@ -1256,7 +1262,7 @@ static size_t ZSTDMT_compress_advanced_internal(
         ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
         DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: fallback to single-thread mode");
         if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
-        return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, jobParams);
+        return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, &jobParams);
     }
 
     assert(avgJobSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), required to compress directly into Dst (no additional buffer) */
@@ -1404,12 +1410,12 @@ size_t ZSTDMT_initCStream_internal(
 
     mtctx->singleBlockingThread = (pledgedSrcSize <= ZSTDMT_JOBSIZE_MIN);  /* do not trigger multi-threading when srcSize is too small */
     if (mtctx->singleBlockingThread) {
-        ZSTD_CCtx_params const singleThreadParams = ZSTDMT_initJobCCtxParams(params);
+        ZSTD_CCtx_params const singleThreadParams = ZSTDMT_initJobCCtxParams(&params);
         DEBUGLOG(5, "ZSTDMT_initCStream_internal: switch to single blocking thread mode");
         assert(singleThreadParams.nbWorkers == 0);
         return ZSTD_initCStream_internal(mtctx->cctxPool->cctx[0],
                                          dict, dictSize, cdict,
-                                         singleThreadParams, pledgedSrcSize);
+                                         &singleThreadParams, pledgedSrcSize);
     }
 
     DEBUGLOG(4, "ZSTDMT_initCStream_internal: %u workers", params.nbWorkers);
@@ -1435,11 +1441,11 @@ size_t ZSTDMT_initCStream_internal(
         mtctx->cdict = cdict;
     }
 
-    mtctx->targetPrefixSize = ZSTDMT_computeOverlapSize(params);
+    mtctx->targetPrefixSize = ZSTDMT_computeOverlapSize(&params);
     DEBUGLOG(4, "overlapLog=%i => %u KB", params.overlapLog, (U32)(mtctx->targetPrefixSize>>10));
     mtctx->targetSectionSize = params.jobSize;
     if (mtctx->targetSectionSize == 0) {
-        mtctx->targetSectionSize = 1ULL << ZSTDMT_computeTargetJobLog(params);
+        mtctx->targetSectionSize = 1ULL << ZSTDMT_computeTargetJobLog(&params);
     }
     assert(mtctx->targetSectionSize <= (size_t)ZSTDMT_JOBSIZE_MAX);
 

data/ext/zstdruby/libzstd/decompress/huf_decompress.c

@@ -61,7 +61,9 @@
 *  Error Management
 ****************************************************************/
 #define HUF_isError ERR_isError
+#ifndef CHECK_F
 #define CHECK_F(f) { size_t const err_ = (f); if (HUF_isError(err_)) return err_; }
+#endif
 
 
 /* **************************************************************

data/ext/zstdruby/libzstd/decompress/zstd_decompress.c

@@ -88,10 +88,7 @@ size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 
 static size_t ZSTD_startingInputLength(ZSTD_format_e format)
 {
-    size_t const startingInputLength = (format==ZSTD_f_zstd1_magicless) ?
-                    ZSTD_FRAMEHEADERSIZE_PREFIX - ZSTD_FRAMEIDSIZE :
-                    ZSTD_FRAMEHEADERSIZE_PREFIX;
-    ZSTD_STATIC_ASSERT(ZSTD_FRAMEHEADERSIZE_PREFIX >= ZSTD_FRAMEIDSIZE);
+    size_t const startingInputLength = ZSTD_FRAMEHEADERSIZE_PREFIX(format);
     /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
     assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
     return startingInputLength;
@@ -376,7 +373,7 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
 {
     unsigned long long totalDstSize = 0;
 
-    while (srcSize >= ZSTD_FRAMEHEADERSIZE_PREFIX) {
+    while (srcSize >= ZSTD_startingInputLength(ZSTD_f_zstd1)) {
         U32 const magicNumber = MEM_readLE32(src);
 
         if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
@@ -574,9 +571,10 @@ void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
 }
 
 /** ZSTD_insertBlock() :
-    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+ *  insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
 size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
 {
+    DEBUGLOG(5, "ZSTD_insertBlock: %u bytes", (unsigned)blockSize);
     ZSTD_checkContinuity(dctx, blockStart);
     dctx->previousDstEnd = (const char*)blockStart + blockSize;
     return blockSize;
@@ -628,11 +626,12 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
 
     /* check */
     RETURN_ERROR_IF(
-        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN+ZSTD_blockHeaderSize,
+        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN(dctx->format)+ZSTD_blockHeaderSize,
         srcSize_wrong);
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(ip, ZSTD_FRAMEHEADERSIZE_PREFIX);
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize_internal(
+                ip, ZSTD_FRAMEHEADERSIZE_PREFIX(dctx->format), dctx->format);
         if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
         RETURN_ERROR_IF(remainingSrcSize < frameHeaderSize+ZSTD_blockHeaderSize,
                         srcSize_wrong);
@@ -713,7 +712,7 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
         dictSize = ZSTD_DDict_dictSize(ddict);
     }
 
-    while (srcSize >= ZSTD_FRAMEHEADERSIZE_PREFIX) {
+    while (srcSize >= ZSTD_startingInputLength(dctx->format)) {
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
         if (ZSTD_isLegacy(src, srcSize)) {
@@ -1097,7 +1096,7 @@ ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
         size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
         for (i=0; i<3; i++) {
             U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
-            RETURN_ERROR_IF(rep==0 || rep >= dictContentSize,
+            RETURN_ERROR_IF(rep==0 || rep > dictContentSize,
                             dictionary_corrupted);
             entropy->rep[i] = rep;
     }   }
@@ -1266,7 +1265,7 @@ size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
 {
     RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong);
     ZSTD_clearDict(dctx);
-    if (dict && dictSize >= 8) {
+    if (dict && dictSize != 0) {
         dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
         RETURN_ERROR_IF(dctx->ddictLocal == NULL, memory_allocation);
         dctx->ddict = dctx->ddictLocal;
@@ -1299,14 +1298,14 @@ size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSiz
 
 
 /* ZSTD_initDStream_usingDict() :
- * return : expected size, aka ZSTD_FRAMEHEADERSIZE_PREFIX.
+ * return : expected size, aka ZSTD_startingInputLength().
  * this function cannot fail */
 size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
 {
     DEBUGLOG(4, "ZSTD_initDStream_usingDict");
     FORWARD_IF_ERROR( ZSTD_DCtx_reset(zds, ZSTD_reset_session_only) );
     FORWARD_IF_ERROR( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) );
-    return ZSTD_FRAMEHEADERSIZE_PREFIX;
+    return ZSTD_startingInputLength(zds->format);
 }
 
 /* note : this variant can't fail */
@@ -1323,16 +1322,16 @@ size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
 {
     FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) );
     FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) );
-    return ZSTD_FRAMEHEADERSIZE_PREFIX;
+    return ZSTD_startingInputLength(dctx->format);
 }
 
 /* ZSTD_resetDStream() :
- * return : expected size, aka ZSTD_FRAMEHEADERSIZE_PREFIX.
+ * return : expected size, aka ZSTD_startingInputLength().
  * this function cannot fail */
 size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
 {
     FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only));
-    return ZSTD_FRAMEHEADERSIZE_PREFIX;
+    return ZSTD_startingInputLength(dctx->format);
 }
 
 
@@ -1563,7 +1562,7 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                             zds->lhSize += remainingInput;
                         }
                         input->pos = input->size;
-                        return (MAX(ZSTD_FRAMEHEADERSIZE_MIN, hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                        return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
                     assert(ip != NULL);
                     memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;

data/ext/zstdruby/libzstd/decompress/zstd_decompress_block.c

@@ -79,6 +79,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
 size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                           const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
 {
+    DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
     RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected);
 
     {   const BYTE* const istart = (const BYTE*) src;
@@ -87,6 +88,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
         switch(litEncType)
         {
         case set_repeat:
+            DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
             RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted);
             /* fall-through */
 
@@ -116,7 +118,7 @@ size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
                     /* 2 - 2 - 18 - 18 */
                     lhSize = 5;
                     litSize  = (lhc >> 4) & 0x3FFFF;
-                    litCSize = (lhc >> 22) + (istart[4] << 10);
+                    litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
                     break;
                 }
                 RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected);
@@ -391,7 +393,8 @@ ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
                     symbolNext[s] = 1;
                 } else {
                     if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
-                    symbolNext[s] = normalizedCounter[s];
+                    assert(normalizedCounter[s]>=0);
+                    symbolNext[s] = (U16)normalizedCounter[s];
         }   }   }
         memcpy(dt, &DTableH, sizeof(DTableH));
     }
@@ -570,38 +573,118 @@ typedef struct {
     size_t pos;
 } seqState_t;
 
+/*! ZSTD_overlapCopy8() :
+ *  Copies 8 bytes from ip to op and updates op and ip where ip <= op.
+ *  If the offset is < 8 then the offset is spread to at least 8 bytes.
+ *
+ *  Precondition: *ip <= *op
+ *  Postcondition: *op - *op >= 8
+ */
+static void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset) {
+    assert(*ip <= *op);
+    if (offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[offset];
+        (*op)[0] = (*ip)[0];
+        (*op)[1] = (*ip)[1];
+        (*op)[2] = (*ip)[2];
+        (*op)[3] = (*ip)[3];
+        *ip += dec32table[offset];
+        ZSTD_copy4(*op+4, *ip);
+        *ip -= sub2;
+    } else {
+        ZSTD_copy8(*op, *ip);
+    }
+    *ip += 8;
+    *op += 8;
+    assert(*op - *ip >= 8);
+}
+
+/*! ZSTD_safecopy() :
+ *  Specialized version of memcpy() that is allowed to READ up to WILDCOPY_OVERLENGTH past the input buffer
+ *  and write up to 16 bytes past oend_w (op >= oend_w is allowed).
+ *  This function is only called in the uncommon case where the sequence is near the end of the block. It
+ *  should be fast for a single long sequence, but can be slow for several short sequences.
+ *
+ *  @param ovtype controls the overlap detection
+ *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
+ *         - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
+ *           The src buffer must be before the dst buffer.
+ */
+static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
 
-/* ZSTD_execSequenceLast7():
- * exceptional case : decompress a match starting within last 7 bytes of output buffer.
- * requires more careful checks, to ensure there is no overflow.
- * performance does not matter though.
- * note : this case is supposed to be never generated "naturally" by reference encoder,
- *        since in most cases it needs at least 8 bytes to look for a match.
- *        but it's allowed by the specification. */
+    assert((ovtype == ZSTD_no_overlap && (diff <= -8 || diff >= 8 || op >= oend_w)) ||
+           (ovtype == ZSTD_overlap_src_before_dst && diff >= 0));
+
+    if (length < 8) {
+        /* Handle short lengths. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+    if (ovtype == ZSTD_overlap_src_before_dst) {
+        /* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
+        assert(length >= 8);
+        ZSTD_overlapCopy8(&op, &ip, diff);
+        assert(op - ip >= 8);
+        assert(op <= oend);
+    }
+
+    if (oend <= oend_w) {
+        /* No risk of overwrite. */
+        ZSTD_wildcopy(op, ip, length, ovtype);
+        return;
+    }
+    if (op <= oend_w) {
+        /* Wildcopy until we get close to the end. */
+        assert(oend > oend_w);
+        ZSTD_wildcopy(op, ip, oend_w - op, ovtype);
+        ip += oend_w - op;
+        op = oend_w;
+    }
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
+/* ZSTD_execSequenceEnd():
+ * This version handles cases that are near the end of the output buffer. It requires
+ * more careful checks to make sure there is no overflow. By separating out these hard
+ * and unlikely cases, we can speed up the common cases.
+ *
+ * NOTE: This function needs to be fast for a single long sequence, but doesn't need
+ * to be optimized for many small sequences, since those fall into ZSTD_execSequence().
+ */
 FORCE_NOINLINE
-size_t ZSTD_execSequenceLast7(BYTE* op,
-                              BYTE* const oend, seq_t sequence,
-                              const BYTE** litPtr, const BYTE* const litLimit,
-                              const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
+size_t ZSTD_execSequenceEnd(BYTE* op,
+                            BYTE* const oend, seq_t sequence,
+                            const BYTE** litPtr, const BYTE* const litLimit,
+                            const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
 {
     BYTE* const oLitEnd = op + sequence.litLength;
     size_t const sequenceLength = sequence.litLength + sequence.matchLength;
     BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
     const BYTE* match = oLitEnd - sequence.offset;
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
 
-    /* check */
-    RETURN_ERROR_IF(oMatchEnd>oend, dstSize_tooSmall, "last match must fit within dstBuffer");
+    /* bounds checks */
+    assert(oLitEnd < oMatchEnd);
+    RETURN_ERROR_IF(oMatchEnd > oend, dstSize_tooSmall, "last match must fit within dstBuffer");
     RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "try to read beyond literal buffer");
 
     /* copy literals */
-    while (op < oLitEnd) *op++ = *(*litPtr)++;
+    ZSTD_safecopy(op, oend_w, *litPtr, sequence.litLength, ZSTD_no_overlap);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
 
     /* copy Match */
-    if (sequence.offset > (size_t)(oLitEnd - base)) {
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
         /* offset beyond prefix */
-        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - vBase),corruption_detected);
-        match = dictEnd - (base-match);
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected);
+        match = dictEnd - (prefixStart-match);
         if (match + sequence.matchLength <= dictEnd) {
             memmove(oLitEnd, match, sequence.matchLength);
             return sequenceLength;
@@ -611,13 +694,12 @@ size_t ZSTD_execSequenceLast7(BYTE* op,
             memmove(oLitEnd, match, length1);
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
-            match = base;
+            match = prefixStart;
     }   }
-    while (op < oMatchEnd) *op++ = *match++;
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
     return sequenceLength;
 }
 
-
 HINT_INLINE
 size_t ZSTD_execSequence(BYTE* op,
                          BYTE* const oend, seq_t sequence,
@@ -631,20 +713,29 @@ size_t ZSTD_execSequence(BYTE* op,
     const BYTE* const iLitEnd = *litPtr + sequence.litLength;
     const BYTE* match = oLitEnd - sequence.offset;
 
-    /* check */
-    RETURN_ERROR_IF(oMatchEnd>oend, dstSize_tooSmall, "last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend");
-    RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "over-read beyond lit buffer");
-    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
-
-    /* copy Literals */
-    if (sequence.litLength > 8)
-        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);
+    /* Errors and uncommon cases handled here. */
+    assert(oLitEnd < oMatchEnd);
+    if (iLitEnd > litLimit || oMatchEnd > oend_w)
+        return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (sequence.litLength > 16) {
+        ZSTD_wildcopy(op+16, (*litPtr)+16, sequence.litLength-16, ZSTD_no_overlap);
+    }
     op = oLitEnd;
     *litPtr = iLitEnd;   /* update for next sequence */
 
-    /* copy Match */
+    /* Copy Match */
     if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
         /* offset beyond prefix -> go into extDict */
         RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected);
@@ -659,123 +750,33 @@ size_t ZSTD_execSequence(BYTE* op,
             op = oLitEnd + length1;
             sequence.matchLength -= length1;
             match = prefixStart;
-            if (op > oend_w || sequence.matchLength < MINMATCH) {
-              U32 i;
-              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
-              return sequenceLength;
-            }
     }   }
-    /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */
-
-    /* match within prefix */
-    if (sequence.offset < 8) {
-        /* close range match, overlap */
-        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
-        int const sub2 = dec64table[sequence.offset];
-        op[0] = match[0];
-        op[1] = match[1];
-        op[2] = match[2];
-        op[3] = match[3];
-        match += dec32table[sequence.offset];
-        ZSTD_copy4(op+4, match);
-        match -= sub2;
-    } else {
-        ZSTD_copy8(op, match);
-    }
-    op += 8; match += 8;
-
-    if (oMatchEnd > oend-(16-MINMATCH)) {
-        if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op, 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, ZSTD_overlap_src_before_dst);   /* works even if matchLength < 8 */
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (sequence.offset >= WILDCOPY_VECLEN) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
     }
-    return sequenceLength;
-}
-
-
-HINT_INLINE
-size_t ZSTD_execSequenceLong(BYTE* op,
-                             BYTE* const oend, seq_t sequence,
-                             const BYTE** litPtr, const BYTE* const litLimit,
-                             const BYTE* const prefixStart, const BYTE* const dictStart, const BYTE* const dictEnd)
-{
-    BYTE* const oLitEnd = op + sequence.litLength;
-    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
-    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
-    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
-    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
-    const BYTE* match = sequence.match;
+    assert(sequence.offset < WILDCOPY_VECLEN);
 
-    /* check */
-    RETURN_ERROR_IF(oMatchEnd > oend, dstSize_tooSmall, "last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend");
-    RETURN_ERROR_IF(iLitEnd > litLimit, corruption_detected, "over-read beyond lit buffer");
-    if (oLitEnd > oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, prefixStart, dictStart, dictEnd);
-
-    /* copy Literals */
-    if (sequence.litLength > 8)
-        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 */
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
 
-    op = oLitEnd;
-    *litPtr = iLitEnd;   /* update for next sequence */
-
-    /* copy Match */
-    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
-        /* offset beyond prefix */
-        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - dictStart), corruption_detected);
-        if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
-            return sequenceLength;
-        }
-        /* span extDict & currentPrefixSegment */
-        {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
-            op = oLitEnd + length1;
-            sequence.matchLength -= length1;
-            match = prefixStart;
-            if (op > oend_w || sequence.matchLength < MINMATCH) {
-              U32 i;
-              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
-              return sequenceLength;
-            }
-    }   }
-    assert(op <= oend_w);
-    assert(sequence.matchLength >= MINMATCH);
-
-    /* match within prefix */
-    if (sequence.offset < 8) {
-        /* close range match, overlap */
-        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
-        int const sub2 = dec64table[sequence.offset];
-        op[0] = match[0];
-        op[1] = match[1];
-        op[2] = match[2];
-        op[3] = match[3];
-        match += dec32table[sequence.offset];
-        ZSTD_copy4(op+4, match);
-        match -= sub2;
-    } else {
-        ZSTD_copy8(op, match);
-    }
-    op += 8; match += 8;
-
-    if (oMatchEnd > oend-(16-MINMATCH)) {
-        if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op, 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, ZSTD_overlap_src_before_dst);   /* works even if matchLength < 8 */
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst);
     }
     return sequenceLength;
 }
@@ -1095,7 +1096,7 @@ ZSTD_decompressSequencesLong_body(
         /* decode and decompress */
         for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) {
             seq_t const sequence = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
-            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
             if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
             PREFETCH_L1(sequence.match); PREFETCH_L1(sequence.match + sequence.matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
             sequences[seqNb & STORED_SEQS_MASK] = sequence;
@@ -1106,7 +1107,7 @@ ZSTD_decompressSequencesLong_body(
         /* finish queue */
         seqNb -= seqAdvance;
         for ( ; seqNb<nbSeq ; seqNb++) {
-            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[seqNb&STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
             if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
             op += oneSeqSize;
         }