skips the need to create a temporary cdict for --patch-from

thus saving a bit of memory and a little bit of cpu time

lib/compress/zstdmt_compress.c

@@ -714,7 +714,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
         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 */
+    } else {
         U64 const pledgedSrcSize = job->firstJob ? job->fullFrameSize : job->src.size;
         {   size_t const forceWindowError = ZSTD_CCtxParams_setParameter(&jobParams, ZSTD_c_forceMaxWindow, !job->firstJob);
             if (ZSTD_isError(forceWindowError)) JOB_ERROR(forceWindowError);
@@ -741,7 +741,7 @@ static void ZSTDMT_compressionJob(void* jobDescription)
         ZSTD_invalidateRepCodes(cctx);
     }
 
-    /* compress */
+    /* compress the entire job by smaller chunks, for better granularity */
     {   size_t const chunkSize = 4*ZSTD_BLOCKSIZE_MAX;
         int const nbChunks = (int)((job->src.size + (chunkSize-1)) / chunkSize);
         const BYTE* ip = (const BYTE*) job->src.start;
@@ -1258,18 +1258,6 @@ size_t ZSTDMT_initCStream_internal(
 
     mtctx->params = params;
     mtctx->frameContentSize = pledgedSrcSize;
-    ZSTD_freeCDict(mtctx->cdictLocal);
-    if (dict) {
-        mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
-                                                    dictLoadMethod, dictContentType, /* note : a loadPrefix becomes an internal CDict */
-                                                    params.cParams, mtctx->cMem);
-        mtctx->cdict = mtctx->cdictLocal;
-        if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
-    } else {
-        mtctx->cdictLocal = NULL;
-        mtctx->cdict = cdict;
-    }
-
     mtctx->targetPrefixSize = ZSTDMT_computeOverlapSize(&params);
     DEBUGLOG(4, "overlapLog=%i => %u KB", params.overlapLog, (U32)(mtctx->targetPrefixSize>>10));
     mtctx->targetSectionSize = params.jobSize;
@@ -1331,9 +1319,31 @@ size_t ZSTDMT_initCStream_internal(
     mtctx->allJobsCompleted = 0;
     mtctx->consumed = 0;
     mtctx->produced = 0;
+
+    /* update dictionary */
+    ZSTD_freeCDict(mtctx->cdictLocal);
+    mtctx->cdictLocal = NULL;
+    if (dict) {
+        if (dictContentType == ZSTD_dct_rawContent) {
+            mtctx->inBuff.prefix.start = (const BYTE*)dict;
+            mtctx->inBuff.prefix.size = dictSize;
+        } else {
+            /* note : a loadPrefix becomes an internal CDict */
+            mtctx->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
+                                                        dictLoadMethod, dictContentType,
+                                                        params.cParams, mtctx->cMem);
+            mtctx->cdict = mtctx->cdictLocal;
+            if (mtctx->cdictLocal == NULL) return ERROR(memory_allocation);
+        }
+    } else {
+        mtctx->cdict = cdict;
+    }
+
     if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params, mtctx->targetSectionSize,
                                  dict, dictSize, dictContentType))
         return ERROR(memory_allocation);
+
+
     return 0;
 }
 
@@ -1543,6 +1553,11 @@ static Range ZSTDMT_getInputDataInUse(ZSTDMT_CCtx* mtctx)
     unsigned const lastJobID = mtctx->nextJobID;
     unsigned jobID;
 
+    /* no need to check during first round */
+    size_t roundBuffCapacity = mtctx->roundBuff.capacity;
+    size_t nbJobs1stRoundMin = roundBuffCapacity / mtctx->targetSectionSize;
+    if (lastJobID < nbJobs1stRoundMin) return kNullRange;
+
     for (jobID = firstJobID; jobID < lastJobID; ++jobID) {
         unsigned const wJobID = jobID & mtctx->jobIDMask;
         size_t consumed;
@@ -1637,14 +1652,14 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
 {
     Range const inUse = ZSTDMT_getInputDataInUse(mtctx);
     size_t const spaceLeft = mtctx->roundBuff.capacity - mtctx->roundBuff.pos;
-    size_t const target = mtctx->targetSectionSize;
+    size_t const spaceNeeded = mtctx->targetSectionSize;
     Buffer buffer;
 
     DEBUGLOG(5, "ZSTDMT_tryGetInputRange");
     assert(mtctx->inBuff.buffer.start == NULL);
-    assert(mtctx->roundBuff.capacity >= target);
+    assert(mtctx->roundBuff.capacity >= spaceNeeded);
 
-    if (spaceLeft < target) {
+    if (spaceLeft < spaceNeeded) {
         /* ZSTD_invalidateRepCodes() doesn't work for extDict variants.
          * Simply copy the prefix to the beginning in that case.
          */
@@ -1663,7 +1678,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
         mtctx->roundBuff.pos = prefixSize;
     }
     buffer.start = mtctx->roundBuff.buffer + mtctx->roundBuff.pos;
-    buffer.capacity = target;
+    buffer.capacity = spaceNeeded;
 
     if (ZSTDMT_isOverlapped(buffer, inUse)) {
         DEBUGLOG(5, "Waiting for buffer...");