Merge pull request #1059 from terrelln/mt-ldm

Integrate ldm with zstdmt

lib/compress/zstd_compress.c

@@ -922,6 +922,7 @@ static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pl
     cctx->dictID = 0;
     if (params.ldmParams.enableLdm)
         ZSTD_window_clear(&cctx->ldmState.window);
+    ZSTD_referenceExternalSequences(cctx, NULL, 0);
     ZSTD_invalidateMatchState(&cctx->blockState.matchState);
     ZSTD_reset_compressedBlockState(cctx->blockState.prevCBlock);
     XXH64_reset(&cctx->xxhState, 0);
@@ -1108,6 +1109,7 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
             ptr = zc->ldmState.bucketOffsets + ldmBucketSize;
             ZSTD_window_clear(&zc->ldmState.window);
         }
+        ZSTD_referenceExternalSequences(zc, NULL, 0);
 
         /* buffers */
         zc->inBuffSize = buffInSize;
@@ -1925,8 +1927,10 @@ static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
     ZSTD_matchState_t* const ms = &zc->blockState.matchState;
     DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
                 (U32)dstCapacity, ms->window.dictLimit, ms->nextToUpdate);
-    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1)
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+        ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.searchLength);
         return 0;   /* don't even attempt compression below a certain srcSize */
+    }
     ZSTD_resetSeqStore(&(zc->seqStore));
 
     /* limited update after a very long match */

lib/compress/zstd_ldm.c

@@ -536,6 +536,34 @@ size_t ZSTD_ldm_generateSequences(
     return 0;
 }
 
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
+    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
+        if (srcSize <= seq->litLength) {
+            /* Skip past srcSize literals */
+            seq->litLength -= (U32)srcSize;
+            return;
+        }
+        srcSize -= seq->litLength;
+        seq->litLength = 0;
+        if (srcSize < seq->matchLength) {
+            /* Skip past the first srcSize of the match */
+            seq->matchLength -= (U32)srcSize;
+            if (seq->matchLength < minMatch) {
+                /* The match is too short, omit it */
+                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
+                    seq[1].litLength += seq[0].matchLength;
+                }
+                rawSeqStore->pos++;
+            }
+            return;
+        }
+        srcSize -= seq->matchLength;
+        seq->matchLength = 0;
+        rawSeqStore->pos++;
+    }
+}
+
 /**
  * If the sequence length is longer than remaining then the sequence is split
  * between this block and the next.
@@ -546,51 +574,24 @@ size_t ZSTD_ldm_generateSequences(
 static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
                                  U32 const remaining, U32 const minMatch)
 {
-    size_t const pos = rawSeqStore->pos;
     rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
     assert(sequence.offset > 0);
-    /* Handle partial sequences */
+    /* Likely: No partial sequence */
+    if (remaining >= sequence.litLength + sequence.matchLength) {
+        rawSeqStore->pos++;
+        return sequence;
+    }
+    /* Cut the sequence short (offset == 0 ==> rest is literals). */
     if (remaining <= sequence.litLength) {
-        /* Split the literals that we have out of the sequence.
-         * They will become the last literals of this block.
-         * The next block starts off with the remaining literals.
-         */
-        rawSeqStore->seq[pos].litLength -= remaining;
         sequence.offset = 0;
     } else if (remaining < sequence.litLength + sequence.matchLength) {
-        /* Split the match up into two sequences. One in this block, and one
+        sequence.matchLength = remaining - sequence.litLength;
-         * in the next with no literals. If either match would be shorter
-         * than searchLength we omit it.
-         */
-        U32 const matchPrefix = remaining - sequence.litLength;
-        U32 const matchSuffix = sequence.matchLength - matchPrefix;
-
-        assert(remaining > sequence.litLength);
-        assert(matchPrefix < sequence.matchLength);
-        assert(matchPrefix + matchSuffix == sequence.matchLength);
-        /* Update the first sequence */
-        sequence.matchLength = matchPrefix;
-        /* Update the second sequence */
-        if (matchSuffix >= minMatch) {
-            /* Update the second sequence, since the suffix is long enough */
-            rawSeqStore->seq[pos].litLength = 0;
-            rawSeqStore->seq[pos].matchLength = matchSuffix;
-        } else {
-            /* Omit the second sequence since the match suffix is too short.
-             * Add to the next sequences literals (if any).
-             */
-            if (pos + 1 < rawSeqStore->size)
-                rawSeqStore->seq[pos + 1].litLength += matchSuffix;
-            rawSeqStore->pos++; /* Consume the sequence */
-        }
         if (sequence.matchLength < minMatch) {
-            /* Skip the current sequence if it is too short */
             sequence.offset = 0;
         }
-    } else {
-      /* No partial sequence */
-      rawSeqStore->pos++; /* Consume the sequence */
     }
+    /* Skip past `remaining` bytes for the future sequences. */
+    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
     return sequence;
 }
 

lib/compress/zstd_ldm.h

@@ -65,6 +65,16 @@ size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
     ZSTD_compressionParameters const* cParams, void const* src, size_t srcSize,
     int const extDict);
 
+/**
+ * ZSTD_ldm_skipSequences():
+ *
+ * Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
+ * Avoids emitting matches less than `minMatch` bytes.
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
+    U32 const minMatch);
+
 
 /** ZSTD_ldm_initializeParameters() :
  *  Initialize the long distance matching parameters to their default values. */

lib/compress/zstdmt_compress.c

@@ -27,8 +27,14 @@
 #include "pool.h"        /* threadpool */
 #include "threading.h"   /* mutex */
 #include "zstd_compress_internal.h"  /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
+#include "zstd_ldm.h"
 #include "zstdmt_compress.h"
 
+/* Guards code to support resizing the SeqPool.
+ * We will want to resize the SeqPool to save memory in the future.
+ * Until then, comment the code out since it is unused.
+ */
+#define ZSTD_RESIZE_SEQPOOL 0
 
 /* ======   Debug   ====== */
 #if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
@@ -194,6 +200,32 @@ static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
     }
 }
 
+#if ZSTD_RESIZE_SEQPOOL
+/** ZSTDMT_resizeBuffer() :
+ * assumption : bufPool must be valid
+ * @return : a buffer that is at least the buffer pool buffer size.
+ *           If a reallocation happens, the data in the input buffer is copied.
+ */
+static buffer_t ZSTDMT_resizeBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buffer)
+{
+    size_t const bSize = bufPool->bufferSize;
+    if (buffer.capacity < bSize) {
+        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
+        buffer_t newBuffer;
+        newBuffer.start = start;
+        newBuffer.capacity = start == NULL ? 0 : bSize;
+        if (start != NULL) {
+            assert(newBuffer.capacity >= buffer.capacity);
+            memcpy(newBuffer.start, buffer.start, buffer.capacity);
+            DEBUGLOG(5, "ZSTDMT_resizeBuffer: created buffer of size %u", (U32)bSize);
+            return newBuffer;
+        }
+        DEBUGLOG(5, "ZSTDMT_resizeBuffer: buffer allocation failure !!");
+    }
+    return buffer;
+}
+#endif
+
 /* store buffer for later re-use, up to pool capacity */
 static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
 {
@@ -214,6 +246,72 @@ static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
 }
 
 
+/* =====   Seq Pool Wrapper   ====== */
+
+static rawSeqStore_t kNullRawSeqStore = {NULL, 0, 0, 0};
+
+typedef ZSTDMT_bufferPool ZSTDMT_seqPool;
+
+static size_t ZSTDMT_sizeof_seqPool(ZSTDMT_seqPool* seqPool)
+{
+    return ZSTDMT_sizeof_bufferPool(seqPool);
+}
+
+static rawSeqStore_t bufferToSeq(buffer_t buffer)
+{
+    rawSeqStore_t seq = {NULL, 0, 0, 0};
+    seq.seq = (rawSeq*)buffer.start;
+    seq.capacity = buffer.capacity / sizeof(rawSeq);
+    return seq;
+}
+
+static buffer_t seqToBuffer(rawSeqStore_t seq)
+{
+    buffer_t buffer;
+    buffer.start = seq.seq;
+    buffer.capacity = seq.capacity * sizeof(rawSeq);
+    return buffer;
+}
+
+static rawSeqStore_t ZSTDMT_getSeq(ZSTDMT_seqPool* seqPool)
+{
+    if (seqPool->bufferSize == 0) {
+        return kNullRawSeqStore;
+    }
+    return bufferToSeq(ZSTDMT_getBuffer(seqPool));
+}
+
+#if ZSTD_RESIZE_SEQPOOL
+static rawSeqStore_t ZSTDMT_resizeSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  return bufferToSeq(ZSTDMT_resizeBuffer(seqPool, seqToBuffer(seq)));
+}
+#endif
+
+static void ZSTDMT_releaseSeq(ZSTDMT_seqPool* seqPool, rawSeqStore_t seq)
+{
+  ZSTDMT_releaseBuffer(seqPool, seqToBuffer(seq));
+}
+
+static void ZSTDMT_setNbSeq(ZSTDMT_seqPool* const seqPool, size_t const nbSeq)
+{
+  ZSTDMT_setBufferSize(seqPool, nbSeq * sizeof(rawSeq));
+}
+
+static ZSTDMT_seqPool* ZSTDMT_createSeqPool(unsigned nbWorkers, ZSTD_customMem cMem)
+{
+    ZSTDMT_seqPool* seqPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
+    ZSTDMT_setNbSeq(seqPool, 0);
+    return seqPool;
+}
+
+static void ZSTDMT_freeSeqPool(ZSTDMT_seqPool* seqPool)
+{
+    ZSTDMT_freeBufferPool(seqPool);
+}
+
+
+
 /* =====   CCtx Pool   ===== */
 /* a single CCtx Pool can be invoked from multiple threads in parallel */
 
@@ -312,36 +410,97 @@ typedef struct {
 } range_t;
 
 typedef struct {
+    /* All variables in the struct are protected by mutex. */
     ZSTD_pthread_mutex_t mutex;
     ZSTD_pthread_cond_t cond;
     ZSTD_CCtx_params params;
+    ldmState_t ldmState;
     XXH64_state_t xxhState;
     unsigned nextJobID;
+    /* Protects ldmWindow.
+     * Must be acquired after the main mutex when acquiring both.
+     */
+    ZSTD_pthread_mutex_t ldmWindowMutex;
+    ZSTD_pthread_cond_t ldmWindowCond;  /* Signaled when ldmWindow is udpated */
+    ZSTD_window_t ldmWindow;  /* A thread-safe copy of ldmState.window */
 } serialState_t;
 
-static void ZSTDMT_serialState_reset(serialState_t* serialState, ZSTD_CCtx_params params)
+static int ZSTDMT_serialState_reset(serialState_t* serialState, ZSTDMT_seqPool* seqPool, ZSTD_CCtx_params params)
 {
+    /* Adjust parameters */
+    if (params.ldmParams.enableLdm) {
+        DEBUGLOG(4, "LDM window size = %u KB", (1U << params.cParams.windowLog) >> 10);
+        params.ldmParams.windowLog = params.cParams.windowLog;
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
+        assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
+        assert(params.ldmParams.hashEveryLog < 32);
+        serialState->ldmState.hashPower =
+                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+    } else {
+        memset(&params.ldmParams, 0, sizeof(params.ldmParams));
+    }
     serialState->nextJobID = 0;
     if (params.fParams.checksumFlag)
         XXH64_reset(&serialState->xxhState, 0);
+    if (params.ldmParams.enableLdm) {
+        ZSTD_customMem cMem = params.customMem;
+        unsigned const hashLog = params.ldmParams.hashLog;
+        size_t const hashSize = ((size_t)1 << hashLog) * sizeof(ldmEntry_t);
+        unsigned const bucketLog =
+            params.ldmParams.hashLog - params.ldmParams.bucketSizeLog;
+        size_t const bucketSize = (size_t)1 << bucketLog;
+        unsigned const prevBucketLog =
+            serialState->params.ldmParams.hashLog -
+            serialState->params.ldmParams.bucketSizeLog;
+        /* Size the seq pool tables */
+        ZSTDMT_setNbSeq(seqPool, ZSTD_ldm_getMaxNbSeq(params.ldmParams, params.jobSize));
+        /* Reset the window */
+        ZSTD_window_clear(&serialState->ldmState.window);
+        serialState->ldmWindow = serialState->ldmState.window;
+        /* Resize tables and output space if necessary. */
+        if (serialState->ldmState.hashTable == NULL || serialState->params.ldmParams.hashLog < hashLog) {
+            ZSTD_free(serialState->ldmState.hashTable, cMem);
+            serialState->ldmState.hashTable = (ldmEntry_t*)ZSTD_malloc(hashSize, cMem);
+        }
+        if (serialState->ldmState.bucketOffsets == NULL || prevBucketLog < bucketLog) {
+            ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
+            serialState->ldmState.bucketOffsets = (BYTE*)ZSTD_malloc(bucketSize, cMem);
+        }
+        if (!serialState->ldmState.hashTable || !serialState->ldmState.bucketOffsets)
+            return 1;
+        /* Zero the tables */
+        memset(serialState->ldmState.hashTable, 0, hashSize);
+        memset(serialState->ldmState.bucketOffsets, 0, bucketSize);
+    }
     serialState->params = params;
+    return 0;
 }
 
 static int ZSTDMT_serialState_init(serialState_t* serialState)
 {
     int initError = 0;
+    memset(serialState, 0, sizeof(*serialState));
     initError |= ZSTD_pthread_mutex_init(&serialState->mutex, NULL);
     initError |= ZSTD_pthread_cond_init(&serialState->cond, NULL);
+    initError |= ZSTD_pthread_mutex_init(&serialState->ldmWindowMutex, NULL);
+    initError |= ZSTD_pthread_cond_init(&serialState->ldmWindowCond, NULL);
     return initError;
 }
 
 static void ZSTDMT_serialState_free(serialState_t* serialState)
 {
+    ZSTD_customMem cMem = serialState->params.customMem;
     ZSTD_pthread_mutex_destroy(&serialState->mutex);
     ZSTD_pthread_cond_destroy(&serialState->cond);
+    ZSTD_pthread_mutex_destroy(&serialState->ldmWindowMutex);
+    ZSTD_pthread_cond_destroy(&serialState->ldmWindowCond);
+    ZSTD_free(serialState->ldmState.hashTable, cMem);
+    ZSTD_free(serialState->ldmState.bucketOffsets, cMem);
 }
 
-static void ZSTDMT_serialState_update(serialState_t* serialState, range_t src, unsigned jobID)
+static void ZSTDMT_serialState_update(serialState_t* serialState,
+                                      ZSTD_CCtx* jobCCtx, rawSeqStore_t seqStore,
+                                      range_t src, unsigned jobID)
 {
     /* Wait for our turn */
     ZSTD_PTHREAD_MUTEX_LOCK(&serialState->mutex);
@@ -351,6 +510,24 @@ static void ZSTDMT_serialState_update(serialState_t* serialState, range_t src, u
     /* A future job may error and skip our job */
     if (serialState->nextJobID == jobID) {
         /* It is now our turn, do any processing necessary */
+        if (serialState->params.ldmParams.enableLdm) {
+            size_t error;
+            assert(seqStore.seq != NULL && seqStore.pos == 0 &&
+                   seqStore.size == 0 && seqStore.capacity > 0);
+            ZSTD_window_update(&serialState->ldmState.window, src.start, src.size);
+            error = ZSTD_ldm_generateSequences(
+                &serialState->ldmState, &seqStore,
+                &serialState->params.ldmParams, src.start, src.size);
+            /* We provide a large enough buffer to never fail. */
+            assert(!ZSTD_isError(error)); (void)error;
+            /* Update ldmWindow to match the ldmState.window and signal the main
+             * thread if it is waiting for a buffer.
+             */
+            ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+            serialState->ldmWindow = serialState->ldmState.window;
+            ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+            ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
+        }
         if (serialState->params.fParams.checksumFlag && src.size > 0)
             XXH64_update(&serialState->xxhState, src.start, src.size);
     }
@@ -358,6 +535,14 @@ static void ZSTDMT_serialState_update(serialState_t* serialState, range_t src, u
     serialState->nextJobID++;
     ZSTD_pthread_cond_broadcast(&serialState->cond);
     ZSTD_pthread_mutex_unlock(&serialState->mutex);
+
+    if (seqStore.size > 0) {
+        size_t const err = ZSTD_referenceExternalSequences(
+            jobCCtx, seqStore.seq, seqStore.size);
+        assert(serialState->params.ldmParams.enableLdm);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
 }
 
 static void ZSTDMT_serialState_ensureFinished(serialState_t* serialState,
@@ -369,6 +554,11 @@ static void ZSTDMT_serialState_ensureFinished(serialState_t* serialState,
         DEBUGLOG(5, "Skipping past job %u because of error", jobID);
         serialState->nextJobID = jobID + 1;
         ZSTD_pthread_cond_broadcast(&serialState->cond);
+
+        ZSTD_PTHREAD_MUTEX_LOCK(&serialState->ldmWindowMutex);
+        ZSTD_window_clear(&serialState->ldmWindow);
+        ZSTD_pthread_cond_signal(&serialState->ldmWindowCond);
+        ZSTD_pthread_mutex_unlock(&serialState->ldmWindowMutex);
     }
     ZSTD_pthread_mutex_unlock(&serialState->mutex);
 
@@ -388,6 +578,7 @@ typedef struct {
     ZSTD_pthread_cond_t job_cond;        /* Thread-safe - used by mtctx and worker */
     ZSTDMT_CCtxPool* cctxPool;           /* Thread-safe - used by mtctx and (all) workers */
     ZSTDMT_bufferPool* bufPool;          /* Thread-safe - used by mtctx and (all) workers */
+    ZSTDMT_seqPool* seqPool;             /* Thread-safe - used by mtctx and (all) workers */
     serialState_t* serial;               /* Thread-safe - used by mtctx and (all) workers */
     buffer_t dstBuff;                    /* set by worker (or mtctx), then read by worker & mtctx, then modified by mtctx => no barrier */
     range_t prefix;                      /* set by mtctx, then read by worker & mtctx => no barrier */
@@ -408,10 +599,15 @@ void ZSTDMT_compressionJob(void* jobDescription)
     ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
     ZSTD_CCtx_params jobParams = job->params;   /* do not modify job->params ! copy it, modify the copy */
     ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(job->cctxPool);
+    rawSeqStore_t rawSeqStore = ZSTDMT_getSeq(job->seqPool);
     buffer_t dstBuff = job->dstBuff;
 
-    /* Don't compute the checksum for chunks, but write it in the header */
+    /* Don't compute the checksum for chunks, since we compute it externally,
+     * but write it in the header.
+     */
     if (job->jobID != 0) jobParams.fParams.checksumFlag = 0;
+    /* Don't run LDM for the chunks, since we handle it externally */
+    jobParams.ldmParams.enableLdm = 0;
 
     /* ressources */
     if (cctx==NULL) {
@@ -448,8 +644,8 @@ void ZSTDMT_compressionJob(void* jobDescription)
                 goto _endJob;
     }   }   }
 
-    /* Perform serial step as early as possible */
+    /* Perform serial step as early as possible, but after CCtx initialization */
-    ZSTDMT_serialState_update(job->serial, job->src, job->jobID);
+    ZSTDMT_serialState_update(job->serial, cctx, rawSeqStore, job->src, job->jobID);
 
     if (!job->firstJob) {  /* flush and overwrite frame header when it's not first job */
         size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.capacity, job->src.start, 0);
@@ -504,6 +700,7 @@ _endJob:
         DEBUGLOG(5, "Finished with prefix: %zx", (size_t)job->prefix.start);
     DEBUGLOG(5, "Finished with source: %zx", (size_t)job->src.start);
     /* release resources */
+    ZSTDMT_releaseSeq(job->seqPool, rawSeqStore);
     ZSTDMT_releaseCCtx(job->cctxPool, cctx);
     /* report */
     ZSTD_PTHREAD_MUTEX_LOCK(&job->job_mutex);
@@ -544,6 +741,7 @@ struct ZSTDMT_CCtx_s {
     ZSTDMT_jobDescription* jobs;
     ZSTDMT_bufferPool* bufPool;
     ZSTDMT_CCtxPool* cctxPool;
+    ZSTDMT_seqPool* seqPool;
     ZSTD_CCtx_params params;
     size_t targetSectionSize;
     size_t targetPrefixSize;
@@ -635,9 +833,10 @@ ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbWorkers, ZSTD_customMem cMem)
     mtctx->jobIDMask = nbJobs - 1;
     mtctx->bufPool = ZSTDMT_createBufferPool(nbWorkers, cMem);
     mtctx->cctxPool = ZSTDMT_createCCtxPool(nbWorkers, cMem);
+    mtctx->seqPool = ZSTDMT_createSeqPool(nbWorkers, cMem);
     initError = ZSTDMT_serialState_init(&mtctx->serial);
     mtctx->roundBuff = kNullRoundBuff;
-    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool | initError) {
+    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool | !mtctx->seqPool | initError) {
         ZSTDMT_freeCCtx(mtctx);
         return NULL;
     }
@@ -692,6 +891,7 @@ size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
     ZSTDMT_freeJobsTable(mtctx->jobs, mtctx->jobIDMask+1, mtctx->cMem);
     ZSTDMT_freeBufferPool(mtctx->bufPool);
     ZSTDMT_freeCCtxPool(mtctx->cctxPool);
+    ZSTDMT_freeSeqPool(mtctx->seqPool);
     ZSTDMT_serialState_free(&mtctx->serial);
     ZSTD_freeCDict(mtctx->cdictLocal);
     if (mtctx->roundBuff.buffer)
@@ -708,6 +908,7 @@ size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
             + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
             + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
             + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
+            + ZSTDMT_sizeof_seqPool(mtctx->seqPool)
             + ZSTD_sizeof_CDict(mtctx->cdictLocal)
             + mtctx->roundBuff.capacity;
 }
@@ -761,7 +962,6 @@ static ZSTD_CCtx_params ZSTDMT_initJobCCtxParams(ZSTD_CCtx_params const params)
     jobParams.compressionLevel = params.compressionLevel;
     jobParams.disableLiteralCompression = params.disableLiteralCompression;
 
-    jobParams.ldmParams = params.ldmParams;
     return jobParams;
 }
 
@@ -827,12 +1027,16 @@ ZSTD_frameProgression ZSTDMT_getFrameProgression(ZSTDMT_CCtx* mtctx)
 
 static size_t ZSTDMT_computeTargetJobLog(ZSTD_CCtx_params const params)
 {
+    if (params.ldmParams.enableLdm)
+        return MAX(21, params.cParams.chainLog + 4);
     return params.cParams.windowLog + 2;
 }
 
 static size_t ZSTDMT_computeOverlapLog(ZSTD_CCtx_params const params)
 {
     unsigned const overlapRLog = (params.overlapSizeLog>9) ? 0 : 9-params.overlapSizeLog;
+    if (params.ldmParams.enableLdm)
+        return (MIN(params.cParams.windowLog, ZSTDMT_computeTargetJobLog(params) - 2) - overlapRLog);
     return overlapRLog >= 9 ? 0 : (params.cParams.windowLog - overlapRLog);
 }
 
@@ -856,7 +1060,7 @@ static size_t ZSTDMT_compress_advanced_internal(
                 void* dst, size_t dstCapacity,
           const void* src, size_t srcSize,
           const ZSTD_CDict* cdict,
-                ZSTD_CCtx_params const params)
+                ZSTD_CCtx_params params)
 {
     ZSTD_CCtx_params const jobParams = ZSTDMT_initJobCCtxParams(params);
     size_t const overlapSize = (size_t)1 << ZSTDMT_computeOverlapLog(params);
@@ -870,6 +1074,7 @@ static size_t ZSTDMT_compress_advanced_internal(
     assert(jobParams.nbWorkers == 0);
     assert(mtctx->cctxPool->totalCCtx == params.nbWorkers);
 
+    params.jobSize = (U32)avgJobSize;
     DEBUGLOG(4, "ZSTDMT_compress_advanced_internal: nbJobs=%2u (rawSize=%u bytes; fixedSize=%u) ",
                 nbJobs, (U32)proposedJobSize, (U32)avgJobSize);
 
@@ -882,7 +1087,8 @@ static size_t ZSTDMT_compress_advanced_internal(
 
     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) */
     ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgJobSize) );
-    ZSTDMT_serialState_reset(&mtctx->serial, params);
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+        return ERROR(memory_allocation);
 
     if (nbJobs > mtctx->jobIDMask+1) {  /* enlarge job table */
         U32 jobsTableSize = nbJobs;
@@ -915,6 +1121,7 @@ static size_t ZSTDMT_compress_advanced_internal(
             mtctx->jobs[u].dstBuff = dstBuffer;
             mtctx->jobs[u].cctxPool = mtctx->cctxPool;
             mtctx->jobs[u].bufPool = mtctx->bufPool;
+            mtctx->jobs[u].seqPool = mtctx->seqPool;
             mtctx->jobs[u].serial = &mtctx->serial;
             mtctx->jobs[u].jobID = u;
             mtctx->jobs[u].firstJob = (u==0);
@@ -1023,9 +1230,6 @@ size_t ZSTDMT_initCStream_internal(
 
     /* init */
     if (params.jobSize == 0) {
-        if (params.cParams.windowLog >= 29)
-            params.jobSize = ZSTDMT_JOBSIZE_MAX;
-        else
         params.jobSize = 1U << ZSTDMT_computeTargetJobLog(params);
     }
     if (params.jobSize > ZSTDMT_JOBSIZE_MAX) params.jobSize = ZSTDMT_JOBSIZE_MAX;
@@ -1072,11 +1276,20 @@ size_t ZSTDMT_initCStream_internal(
     DEBUGLOG(4, "inBuff Size : %u KB", (U32)(mtctx->targetSectionSize>>10));
     ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(mtctx->targetSectionSize));
     {
-        /* Two buffers of slack, plus extra space for the overlap */
+        /* If ldm is enabled we need windowSize space. */
+        size_t const windowSize = mtctx->params.ldmParams.enableLdm ? (1U << mtctx->params.cParams.windowLog) : 0;
+        /* Two buffers of slack, plus extra space for the overlap
+         * This is the minimum slack that LDM works with. One extra because
+         * flush might waste up to targetSectionSize-1 bytes. Another extra
+         * for the overlap (if > 0), then one to fill which doesn't overlap
+         * with the LDM window.
+         */
+        size_t const nbSlackBuffers = 2 + (mtctx->targetPrefixSize > 0);
+        size_t const slackSize = mtctx->targetSectionSize * nbSlackBuffers;
+        /* Compute the total size, and always have enough slack */
         size_t const nbWorkers = MAX(mtctx->params.nbWorkers, 1);
-        size_t const nbSlackBuffers = MIN(nbWorkers, 2) + (mtctx->targetPrefixSize > 0);
+        size_t const sectionsSize = mtctx->targetSectionSize * nbWorkers;
-        size_t const nbSections = nbWorkers + nbSlackBuffers;
+        size_t const capacity = MAX(windowSize, sectionsSize) + slackSize;
-        size_t const capacity = mtctx->targetSectionSize * nbSections;
         if (mtctx->roundBuff.capacity < capacity) {
             if (mtctx->roundBuff.buffer)
                 ZSTD_free(mtctx->roundBuff.buffer, mtctx->cMem);
@@ -1099,7 +1312,8 @@ size_t ZSTDMT_initCStream_internal(
     mtctx->allJobsCompleted = 0;
     mtctx->consumed = 0;
     mtctx->produced = 0;
-    ZSTDMT_serialState_reset(&mtctx->serial, params);
+    if (ZSTDMT_serialState_reset(&mtctx->serial, mtctx->seqPool, params))
+        return ERROR(memory_allocation);
     return 0;
 }
 
@@ -1201,6 +1415,7 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* mtctx, size_t srcSize, ZS
         mtctx->jobs[jobID].dstBuff = g_nullBuffer;
         mtctx->jobs[jobID].cctxPool = mtctx->cctxPool;
         mtctx->jobs[jobID].bufPool = mtctx->bufPool;
+        mtctx->jobs[jobID].seqPool = mtctx->seqPool;
         mtctx->jobs[jobID].serial = &mtctx->serial;
         mtctx->jobs[jobID].jobID = mtctx->nextJobID;
         mtctx->jobs[jobID].firstJob = (mtctx->nextJobID==0);
@@ -1383,10 +1598,51 @@ static int ZSTDMT_isOverlapped(buffer_t buffer, range_t range)
 
     if (rangeStart == NULL || bufferStart == NULL)
         return 0;
+    /* Empty ranges cannot overlap */
+    if (bufferStart == bufferEnd || rangeStart == rangeEnd)
+        return 0;
 
     return bufferStart < rangeEnd && rangeStart < bufferEnd;
 }
 
+static int ZSTDMT_doesOverlapWindow(buffer_t buffer, ZSTD_window_t window)
+{
+    range_t extDict;
+    range_t prefix;
+
+    extDict.start = window.dictBase + window.lowLimit;
+    extDict.size = window.dictLimit - window.lowLimit;
+
+    prefix.start = window.base + window.dictLimit;
+    prefix.size = window.nextSrc - (window.base + window.dictLimit);
+    DEBUGLOG(5, "extDict [0x%zx, 0x%zx)",
+                (size_t)extDict.start,
+                (size_t)extDict.start + extDict.size);
+    DEBUGLOG(5, "prefix  [0x%zx, 0x%zx)",
+                (size_t)prefix.start,
+                (size_t)prefix.start + prefix.size);
+
+    return ZSTDMT_isOverlapped(buffer, extDict)
+        || ZSTDMT_isOverlapped(buffer, prefix);
+}
+
+static void ZSTDMT_waitForLdmComplete(ZSTDMT_CCtx* mtctx, buffer_t buffer)
+{
+    if (mtctx->params.ldmParams.enableLdm) {
+        ZSTD_pthread_mutex_t* mutex = &mtctx->serial.ldmWindowMutex;
+        DEBUGLOG(5, "source  [0x%zx, 0x%zx)",
+                    (size_t)buffer.start,
+                    (size_t)buffer.start + buffer.capacity);
+        ZSTD_PTHREAD_MUTEX_LOCK(mutex);
+        while (ZSTDMT_doesOverlapWindow(buffer, mtctx->serial.ldmWindow)) {
+            DEBUGLOG(6, "Waiting for LDM to finish...");
+            ZSTD_pthread_cond_wait(&mtctx->serial.ldmWindowCond, mutex);
+        }
+        DEBUGLOG(6, "Done waiting for LDM to finish");
+        ZSTD_pthread_mutex_unlock(mutex);
+    }
+}
+
 /**
  * Attempts to set the inBuff to the next section to fill.
  * If any part of the new section is still in use we give up.
@@ -1415,6 +1671,7 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
             DEBUGLOG(6, "Waiting for buffer...");
             return 0;
         }
+        ZSTDMT_waitForLdmComplete(mtctx, buffer);
         memmove(start, mtctx->inBuff.prefix.start, prefixSize);
         mtctx->inBuff.prefix.start = start;
         mtctx->roundBuff.pos = prefixSize;
@@ -1427,6 +1684,9 @@ static int ZSTDMT_tryGetInputRange(ZSTDMT_CCtx* mtctx)
         return 0;
     }
     assert(!ZSTDMT_isOverlapped(buffer, mtctx->inBuff.prefix));
+
+    ZSTDMT_waitForLdmComplete(mtctx, buffer);
+
     DEBUGLOG(5, "Using prefix range [%zx, %zx)",
                 (size_t)mtctx->inBuff.prefix.start,
                 (size_t)mtctx->inBuff.prefix.start + mtctx->inBuff.prefix.size);