diff --git a/NEWS b/NEWS index 5bf184e8d..457105070 100644 --- a/NEWS +++ b/NEWS @@ -1,6 +1,7 @@ v1.3.1 +perf: substantially decreased memory usage in Multi-threading mode, thanks to reports by Tino Reichardt build: fix Visual compilation for non x86/x64 targets, reported by Greg Slazinski (#718) -API exp : breaking change : ZSTD_getframeHeader() +API exp : breaking change : ZSTD_getframeHeader() provides more information v1.3.0 cli : new : `--list` command, by Paul Cruz diff --git a/lib/common/pool.c b/lib/common/pool.c index 749fa4f2f..06d8a5f57 100644 --- a/lib/common/pool.c +++ b/lib/common/pool.c @@ -92,7 +92,7 @@ POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) { * and full queues. */ ctx->queueSize = queueSize + 1; - ctx->queue = (POOL_job *)malloc(ctx->queueSize * sizeof(POOL_job)); + ctx->queue = (POOL_job*) malloc(ctx->queueSize * sizeof(POOL_job)); ctx->queueHead = 0; ctx->queueTail = 0; pthread_mutex_init(&ctx->queueMutex, NULL); @@ -100,7 +100,7 @@ POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) { pthread_cond_init(&ctx->queuePopCond, NULL); ctx->shutdown = 0; /* Allocate space for the thread handles */ - ctx->threads = (pthread_t *)malloc(numThreads * sizeof(pthread_t)); + ctx->threads = (pthread_t*)malloc(numThreads * sizeof(pthread_t)); ctx->numThreads = 0; /* Check for errors */ if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; } @@ -153,8 +153,8 @@ size_t POOL_sizeof(POOL_ctx *ctx) { + ctx->numThreads * sizeof(pthread_t); } -void POOL_add(void *ctxVoid, POOL_function function, void *opaque) { - POOL_ctx *ctx = (POOL_ctx *)ctxVoid; +void POOL_add(void* ctxVoid, POOL_function function, void *opaque) { + POOL_ctx* const ctx = (POOL_ctx*)ctxVoid; if (!ctx) { return; } pthread_mutex_lock(&ctx->queueMutex); @@ -183,22 +183,22 @@ struct POOL_ctx_s { int data; }; -POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) { +POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) { (void)numThreads; (void)queueSize; - return (POOL_ctx *)malloc(sizeof(POOL_ctx)); + return (POOL_ctx*)malloc(sizeof(POOL_ctx)); } -void POOL_free(POOL_ctx *ctx) { - if (ctx) free(ctx); +void POOL_free(POOL_ctx* ctx) { + free(ctx); } -void POOL_add(void *ctx, POOL_function function, void *opaque) { +void POOL_add(void* ctx, POOL_function function, void* opaque) { (void)ctx; function(opaque); } -size_t POOL_sizeof(POOL_ctx *ctx) { +size_t POOL_sizeof(POOL_ctx* ctx) { if (ctx==NULL) return 0; /* supports sizeof NULL */ return sizeof(*ctx); } diff --git a/lib/common/pool.h b/lib/common/pool.h index 386cd674b..957100f46 100644 --- a/lib/common/pool.h +++ b/lib/common/pool.h @@ -19,11 +19,11 @@ extern "C" { typedef struct POOL_ctx_s POOL_ctx; /*! POOL_create() : - Create a thread pool with at most `numThreads` threads. - `numThreads` must be at least 1. - The maximum number of queued jobs before blocking is `queueSize`. - `queueSize` must be at least 1. - @return : The POOL_ctx pointer on success else NULL. + * Create a thread pool with at most `numThreads` threads. + * `numThreads` must be at least 1. + * The maximum number of queued jobs before blocking is `queueSize`. + * `queueSize` must be at least 1. + * @return : POOL_ctx pointer on success, else NULL. */ POOL_ctx *POOL_create(size_t numThreads, size_t queueSize); diff --git a/lib/compress/zstd_compress.c b/lib/compress/zstd_compress.c index 9300357f2..c17645de7 100644 --- a/lib/compress/zstd_compress.c +++ b/lib/compress/zstd_compress.c @@ -377,7 +377,7 @@ size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned v return ERROR(compressionParameter_unsupported); ZSTDMT_freeCCtx(cctx->mtctx); cctx->nbThreads = 1; - cctx->mtctx = ZSTDMT_createCCtx(value); + cctx->mtctx = ZSTDMT_createCCtx_advanced(value, cctx->customMem); if (cctx->mtctx == NULL) return ERROR(memory_allocation); } cctx->nbThreads = value; diff --git a/lib/compress/zstdmt_compress.c b/lib/compress/zstdmt_compress.c index d5607adfd..677e96f08 100644 --- a/lib/compress/zstdmt_compress.c +++ b/lib/compress/zstdmt_compress.c @@ -73,6 +73,7 @@ static unsigned long long GetCurrentClockTimeMicroseconds(void) /* ===== Buffer Pool ===== */ +/* a single Buffer Pool can be invoked from multiple threads in parallel */ typedef struct buffer_s { void* start; @@ -82,6 +83,8 @@ typedef struct buffer_s { static const buffer_t g_nullBuffer = { NULL, 0 }; typedef struct ZSTDMT_bufferPool_s { + pthread_mutex_t poolMutex; + size_t bufferSize; unsigned totalBuffers; unsigned nbBuffers; ZSTD_customMem cMem; @@ -90,10 +93,12 @@ typedef struct ZSTDMT_bufferPool_s { static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads, ZSTD_customMem cMem) { - unsigned const maxNbBuffers = 2*nbThreads + 2; + unsigned const maxNbBuffers = 2*nbThreads + 3; ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc( sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem); if (bufPool==NULL) return NULL; + pthread_mutex_init(&bufPool->poolMutex, NULL); + bufPool->bufferSize = 64 KB; bufPool->totalBuffers = maxNbBuffers; bufPool->nbBuffers = 0; bufPool->cMem = cMem; @@ -106,6 +111,7 @@ static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool) if (!bufPool) return; /* compatibility with free on NULL */ for (u=0; utotalBuffers; u++) ZSTD_free(bufPool->bTable[u].start, bufPool->cMem); + pthread_mutex_destroy(&bufPool->poolMutex); ZSTD_free(bufPool, bufPool->cMem); } @@ -116,65 +122,85 @@ static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool) + (bufPool->totalBuffers - 1) * sizeof(buffer_t); unsigned u; size_t totalBufferSize = 0; + pthread_mutex_lock(&bufPool->poolMutex); for (u=0; utotalBuffers; u++) totalBufferSize += bufPool->bTable[u].size; + pthread_mutex_unlock(&bufPool->poolMutex); return poolSize + totalBufferSize; } -/** ZSTDMT_getBuffer() : - * assumption : invocation from main thread only ! */ -static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* pool, size_t bSize) +static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* bufPool, size_t bSize) { - if (pool->nbBuffers) { /* try to use an existing buffer */ - buffer_t const buf = pool->bTable[--(pool->nbBuffers)]; + bufPool->bufferSize = bSize; +} + +/** ZSTDMT_getBuffer() : + * assumption : bufPool must be valid */ +static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool) +{ + size_t const bSize = bufPool->bufferSize; + DEBUGLOG(5, "ZSTDMT_getBuffer"); + pthread_mutex_lock(&bufPool->poolMutex); + if (bufPool->nbBuffers) { /* try to use an existing buffer */ + buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)]; size_t const availBufferSize = buf.size; - if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize)) + if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize)) { /* large enough, but not too much */ + pthread_mutex_unlock(&bufPool->poolMutex); return buf; + } /* size conditions not respected : scratch this buffer, create new one */ - ZSTD_free(buf.start, pool->cMem); + DEBUGLOG(5, "existing buffer does not meet size conditions => freeing"); + ZSTD_free(buf.start, bufPool->cMem); } + pthread_mutex_unlock(&bufPool->poolMutex); /* create new buffer */ + DEBUGLOG(5, "create a new buffer"); { buffer_t buffer; - void* const start = ZSTD_malloc(bSize, pool->cMem); - if (start==NULL) bSize = 0; + void* const start = ZSTD_malloc(bSize, bufPool->cMem); buffer.start = start; /* note : start can be NULL if malloc fails ! */ - buffer.size = bSize; + buffer.size = (start==NULL) ? 0 : bSize; return buffer; } } /* store buffer for later re-use, up to pool capacity */ -static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* pool, buffer_t buf) +static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf) { - if (buf.start == NULL) return; /* release on NULL */ - if (pool->nbBuffers < pool->totalBuffers) { - pool->bTable[pool->nbBuffers++] = buf; /* store for later re-use */ + if (buf.start == NULL) return; /* compatible with release on NULL */ + DEBUGLOG(5, "ZSTDMT_releaseBuffer"); + pthread_mutex_lock(&bufPool->poolMutex); + if (bufPool->nbBuffers < bufPool->totalBuffers) { + bufPool->bTable[bufPool->nbBuffers++] = buf; /* stored for later use */ + pthread_mutex_unlock(&bufPool->poolMutex); return; } + pthread_mutex_unlock(&bufPool->poolMutex); /* Reached bufferPool capacity (should not happen) */ - ZSTD_free(buf.start, pool->cMem); + DEBUGLOG(5, "buffer pool capacity reached => freeing "); + ZSTD_free(buf.start, bufPool->cMem); } /* ===== CCtx Pool ===== */ +/* a single CCtx Pool can be invoked from multiple threads in parallel */ typedef struct { + pthread_mutex_t poolMutex; unsigned totalCCtx; unsigned availCCtx; ZSTD_customMem cMem; ZSTD_CCtx* cctx[1]; /* variable size */ } ZSTDMT_CCtxPool; -/* assumption : CCtxPool invocation only from main thread */ - /* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */ static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool) { unsigned u; for (u=0; utotalCCtx; u++) ZSTD_freeCCtx(pool->cctx[u]); /* note : compatible with free on NULL */ + pthread_mutex_destroy(&pool->poolMutex); ZSTD_free(pool, pool->cMem); } @@ -186,6 +212,7 @@ static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads, ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc( sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*), cMem); if (!cctxPool) return NULL; + pthread_mutex_init(&cctxPool->poolMutex, NULL); cctxPool->cMem = cMem; cctxPool->totalCCtx = nbThreads; cctxPool->availCCtx = 1; /* at least one cctx for single-thread mode */ @@ -198,50 +225,57 @@ static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads, /* only works during initialization phase, not during compression */ static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool) { - unsigned const nbThreads = cctxPool->totalCCtx; - size_t const poolSize = sizeof(*cctxPool) - + (nbThreads-1)*sizeof(ZSTD_CCtx*); - unsigned u; - size_t totalCCtxSize = 0; - for (u=0; ucctx[u]); - - return poolSize + totalCCtxSize; + pthread_mutex_lock(&cctxPool->poolMutex); + { unsigned const nbThreads = cctxPool->totalCCtx; + size_t const poolSize = sizeof(*cctxPool) + + (nbThreads-1)*sizeof(ZSTD_CCtx*); + unsigned u; + size_t totalCCtxSize = 0; + for (u=0; ucctx[u]); + } + pthread_mutex_unlock(&cctxPool->poolMutex); + return poolSize + totalCCtxSize; + } } -static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* pool) +static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool) { - if (pool->availCCtx) { - pool->availCCtx--; - return pool->cctx[pool->availCCtx]; - } - return ZSTD_createCCtx(); /* note : can be NULL, when creation fails ! */ + DEBUGLOG(5, "ZSTDMT_getCCtx"); + pthread_mutex_lock(&cctxPool->poolMutex); + if (cctxPool->availCCtx) { + cctxPool->availCCtx--; + { ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx]; + pthread_mutex_unlock(&cctxPool->poolMutex); + return cctx; + } } + pthread_mutex_unlock(&cctxPool->poolMutex); + DEBUGLOG(5, "create one more CCtx"); + return ZSTD_createCCtx_advanced(cctxPool->cMem); /* note : can be NULL, when creation fails ! */ } static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx) { if (cctx==NULL) return; /* compatibility with release on NULL */ + pthread_mutex_lock(&pool->poolMutex); if (pool->availCCtx < pool->totalCCtx) pool->cctx[pool->availCCtx++] = cctx; - else + else { /* pool overflow : should not happen, since totalCCtx==nbThreads */ + DEBUGLOG(5, "CCtx pool overflow : free cctx"); ZSTD_freeCCtx(cctx); + } + pthread_mutex_unlock(&pool->poolMutex); } /* ===== Thread worker ===== */ typedef struct { - buffer_t buffer; - size_t filled; -} inBuff_t; - -typedef struct { - ZSTD_CCtx* cctx; buffer_t src; const void* srcStart; - size_t srcSize; size_t dictSize; + size_t srcSize; buffer_t dstBuff; size_t cSize; size_t dstFlushed; @@ -253,6 +287,8 @@ typedef struct { pthread_cond_t* jobCompleted_cond; ZSTD_parameters params; const ZSTD_CDict* cdict; + ZSTDMT_CCtxPool* cctxPool; + ZSTDMT_bufferPool* bufPool; unsigned long long fullFrameSize; } ZSTDMT_jobDescription; @@ -260,37 +296,56 @@ typedef struct { void ZSTDMT_compressChunk(void* jobDescription) { ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription; + ZSTD_CCtx* cctx = ZSTDMT_getCCtx(job->cctxPool); const void* const src = (const char*)job->srcStart + job->dictSize; - buffer_t const dstBuff = job->dstBuff; + buffer_t dstBuff = job->dstBuff; DEBUGLOG(5, "job (first:%u) (last:%u) : dictSize %u, srcSize %u", job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize); + + if (cctx==NULL) { + job->cSize = ERROR(memory_allocation); + goto _endJob; + } + + if (dstBuff.start == NULL) { + dstBuff = ZSTDMT_getBuffer(job->bufPool); + if (dstBuff.start==NULL) { + job->cSize = ERROR(memory_allocation); + goto _endJob; + } + job->dstBuff = dstBuff; + } + if (job->cdict) { /* should only happen for first segment */ - size_t const initError = ZSTD_compressBegin_usingCDict_advanced(job->cctx, job->cdict, job->params.fParams, job->fullFrameSize); + size_t const initError = ZSTD_compressBegin_usingCDict_advanced(cctx, job->cdict, job->params.fParams, job->fullFrameSize); DEBUGLOG(5, "using CDict"); if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; } } else { /* srcStart points at reloaded section */ if (!job->firstChunk) job->params.fParams.contentSizeFlag = 0; /* ensure no srcSize control */ - { size_t const dictModeError = ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceRawDict, 1); /* Force loading dictionary in "content-only" mode (no header analysis) */ - size_t const initError = ZSTD_compressBegin_advanced(job->cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize); + { size_t const dictModeError = ZSTD_setCCtxParameter(cctx, ZSTD_p_forceRawDict, 1); /* Force loading dictionary in "content-only" mode (no header analysis) */ + size_t const initError = ZSTD_compressBegin_advanced(cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize); if (ZSTD_isError(initError) || ZSTD_isError(dictModeError)) { job->cSize = initError; goto _endJob; } - ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceWindow, 1); + ZSTD_setCCtxParameter(cctx, ZSTD_p_forceWindow, 1); } } if (!job->firstChunk) { /* flush and overwrite frame header when it's not first segment */ - size_t const hSize = ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, 0); + size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, 0); if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; } - ZSTD_invalidateRepCodes(job->cctx); + ZSTD_invalidateRepCodes(cctx); } DEBUGLOG(5, "Compressing : "); DEBUG_PRINTHEX(4, job->srcStart, 12); job->cSize = (job->lastChunk) ? - ZSTD_compressEnd (job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize) : - ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize); + ZSTD_compressEnd (cctx, dstBuff.start, dstBuff.size, src, job->srcSize) : + ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, job->srcSize); DEBUGLOG(5, "compressed %u bytes into %u bytes (first:%u) (last:%u)", (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk); DEBUGLOG(5, "dstBuff.size : %u ; => %s", (U32)dstBuff.size, ZSTD_getErrorName(job->cSize)); _endJob: + ZSTDMT_releaseCCtx(job->cctxPool, cctx); + ZSTDMT_releaseBuffer(job->bufPool, job->src); + job->src = g_nullBuffer; job->srcStart = NULL; PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex); job->jobCompleted = 1; job->jobScanned = 0; @@ -303,15 +358,19 @@ _endJob: /* ===== Multi-threaded compression ===== */ /* ------------------------------------------ */ +typedef struct { + buffer_t buffer; + size_t filled; +} inBuff_t; + struct ZSTDMT_CCtx_s { POOL_ctx* factory; ZSTDMT_jobDescription* jobs; - ZSTDMT_bufferPool* buffPool; + ZSTDMT_bufferPool* bufPool; ZSTDMT_CCtxPool* cctxPool; pthread_mutex_t jobCompleted_mutex; pthread_cond_t jobCompleted_cond; size_t targetSectionSize; - size_t marginSize; size_t inBuffSize; size_t dictSize; size_t targetDictSize; @@ -362,9 +421,9 @@ ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbThreads, ZSTD_customMem cMem) mtctx->factory = POOL_create(nbThreads, 1); mtctx->jobs = ZSTDMT_allocJobsTable(&nbJobs, cMem); mtctx->jobIDMask = nbJobs - 1; - mtctx->buffPool = ZSTDMT_createBufferPool(nbThreads, cMem); + mtctx->bufPool = ZSTDMT_createBufferPool(nbThreads, cMem); mtctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads, cMem); - if (!mtctx->factory | !mtctx->jobs | !mtctx->buffPool | !mtctx->cctxPool) { + if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool) { ZSTDMT_freeCCtx(mtctx); return NULL; } @@ -386,15 +445,13 @@ static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx) unsigned jobID; DEBUGLOG(3, "ZSTDMT_releaseAllJobResources"); for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) { - ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].dstBuff); + ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff); mtctx->jobs[jobID].dstBuff = g_nullBuffer; - ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].src); + ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].src); mtctx->jobs[jobID].src = g_nullBuffer; - ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[jobID].cctx); - mtctx->jobs[jobID].cctx = NULL; } memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription)); - ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer); + ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer); mtctx->inBuff.buffer = g_nullBuffer; mtctx->allJobsCompleted = 1; } @@ -404,7 +461,7 @@ size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx) if (mtctx==NULL) return 0; /* compatible with free on NULL */ POOL_free(mtctx->factory); if (!mtctx->allJobsCompleted) ZSTDMT_releaseAllJobResources(mtctx); /* stop workers first */ - ZSTDMT_freeBufferPool(mtctx->buffPool); /* release job resources into pools first */ + ZSTDMT_freeBufferPool(mtctx->bufPool); /* release job resources into pools first */ ZSTD_free(mtctx->jobs, mtctx->cMem); ZSTDMT_freeCCtxPool(mtctx->cctxPool); ZSTD_freeCDict(mtctx->cdictLocal); @@ -418,11 +475,11 @@ size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx) { if (mtctx == NULL) return 0; /* supports sizeof NULL */ return sizeof(*mtctx) - + POOL_sizeof(mtctx->factory) - + ZSTDMT_sizeof_bufferPool(mtctx->buffPool) - + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription) - + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool) - + ZSTD_sizeof_CDict(mtctx->cdictLocal); + + POOL_sizeof(mtctx->factory) + + ZSTDMT_sizeof_bufferPool(mtctx->bufPool) + + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription) + + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool) + + ZSTD_sizeof_CDict(mtctx->cdictLocal); } size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter, unsigned value) @@ -459,11 +516,11 @@ static unsigned computeNbChunks(size_t srcSize, unsigned windowLog, unsigned nbT size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, - void* dst, size_t dstCapacity, - const void* src, size_t srcSize, - const ZSTD_CDict* cdict, - ZSTD_parameters const params, - unsigned overlapRLog) + void* dst, size_t dstCapacity, + const void* src, size_t srcSize, + const ZSTD_CDict* cdict, + ZSTD_parameters const params, + unsigned overlapRLog) { size_t const overlapSize = (overlapRLog>=9) ? 0 : (size_t)1 << (params.cParams.windowLog - overlapRLog); unsigned nbChunks = computeNbChunks(srcSize, params.cParams.windowLog, mtctx->nbThreads); @@ -473,6 +530,7 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, size_t remainingSrcSize = srcSize; unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbChunks : (unsigned)(dstCapacity / ZSTD_compressBound(avgChunkSize)); /* presumes avgChunkSize >= 256 KB, which should be the case */ size_t frameStartPos = 0, dstBufferPos = 0; + XXH64_state_t xxh64; DEBUGLOG(4, "nbChunks : %2u (chunkSize : %u bytes) ", nbChunks, (U32)avgChunkSize); if (nbChunks==1) { /* fallback to single-thread mode */ @@ -481,6 +539,8 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, return ZSTD_compress_advanced(cctx, dst, dstCapacity, src, srcSize, NULL, 0, params); } assert(avgChunkSize >= 256 KB); /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), which is useful to avoid allocating extra buffers */ + ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgChunkSize) ); + XXH64_reset(&xxh64, 0); if (nbChunks > mtctx->jobIDMask+1) { /* enlarge job table */ U32 nbJobs = nbChunks; @@ -496,17 +556,10 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize); size_t const dstBufferCapacity = ZSTD_compressBound(chunkSize); buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity }; - buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : ZSTDMT_getBuffer(mtctx->buffPool, dstBufferCapacity); - ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(mtctx->cctxPool); + buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer; size_t dictSize = u ? overlapSize : 0; - if ((cctx==NULL) || (dstBuffer.start==NULL)) { - mtctx->jobs[u].cSize = ERROR(memory_allocation); /* job result */ - mtctx->jobs[u].jobCompleted = 1; - nbChunks = u+1; /* only wait and free u jobs, instead of initially expected nbChunks ones */ - break; /* let's wait for previous jobs to complete, but don't start new ones */ - } - + mtctx->jobs[u].src = g_nullBuffer; mtctx->jobs[u].srcStart = srcStart + frameStartPos - dictSize; mtctx->jobs[u].dictSize = dictSize; mtctx->jobs[u].srcSize = chunkSize; @@ -516,13 +569,18 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, /* do not calculate checksum within sections, but write it in header for first section */ if (u!=0) mtctx->jobs[u].params.fParams.checksumFlag = 0; mtctx->jobs[u].dstBuff = dstBuffer; - mtctx->jobs[u].cctx = cctx; + mtctx->jobs[u].cctxPool = mtctx->cctxPool; + mtctx->jobs[u].bufPool = mtctx->bufPool; mtctx->jobs[u].firstChunk = (u==0); mtctx->jobs[u].lastChunk = (u==nbChunks-1); mtctx->jobs[u].jobCompleted = 0; mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex; mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond; + if (params.fParams.checksumFlag) { + XXH64_update(&xxh64, srcStart + frameStartPos, chunkSize); + } + DEBUGLOG(5, "posting job %u (%u bytes)", u, (U32)chunkSize); DEBUG_PRINTHEX(6, mtctx->jobs[u].srcStart, 12); POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]); @@ -533,8 +591,8 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, } } /* collect result */ - { unsigned chunkID; - size_t error = 0, dstPos = 0; + { size_t error = 0, dstPos = 0; + unsigned chunkID; for (chunkID=0; chunkIDjobCompleted_mutex); @@ -545,8 +603,6 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, pthread_mutex_unlock(&mtctx->jobCompleted_mutex); DEBUGLOG(5, "ready to write chunk %u ", chunkID); - ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[chunkID].cctx); - mtctx->jobs[chunkID].cctx = NULL; mtctx->jobs[chunkID].srcStart = NULL; { size_t const cSize = mtctx->jobs[chunkID].cSize; if (ZSTD_isError(cSize)) error = cSize; @@ -556,13 +612,25 @@ size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, memmove((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize); /* may overlap when chunk compressed within dst */ if (chunkID >= compressWithinDst) { /* chunk compressed into its own buffer, which must be released */ DEBUGLOG(5, "releasing buffer %u>=%u", chunkID, compressWithinDst); - ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[chunkID].dstBuff); + ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[chunkID].dstBuff); } mtctx->jobs[chunkID].dstBuff = g_nullBuffer; } dstPos += cSize ; } - } + } /* for (chunkID=0; chunkID dstCapacity) { + error = ERROR(dstSize_tooSmall); + } else { + DEBUGLOG(4, "writing checksum : %08X \n", checksum); + MEM_writeLE32((char*)dst + dstPos, checksum); + dstPos += 4; + } } + if (!error) DEBUGLOG(4, "compressed size : %u ", (U32)dstPos); return error ? error : dstPos; } @@ -615,8 +683,8 @@ size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs, if (zcs->nbThreads==1) { DEBUGLOG(4, "single thread mode"); return ZSTD_initCStream_internal(zcs->cctxPool->cctx[0], - dict, dictSize, cdict, - params, pledgedSrcSize); + dict, dictSize, cdict, + params, pledgedSrcSize); } if (zcs->allJobsCompleted == 0) { /* previous compression not correctly finished */ @@ -649,11 +717,9 @@ size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs, zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize); zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize); DEBUGLOG(4, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10)); - zcs->marginSize = zcs->targetSectionSize >> 2; - zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize + zcs->marginSize; - zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize); - if (zcs->inBuff.buffer.start == NULL) return ERROR(memory_allocation); - zcs->inBuff.filled = 0; + zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize; + ZSTDMT_setBufferSize(zcs->bufPool, MAX(zcs->inBuffSize, ZSTD_compressBound(zcs->targetSectionSize)) ); + zcs->inBuff.buffer = g_nullBuffer; zcs->dictSize = 0; zcs->doneJobID = 0; zcs->nextJobID = 0; @@ -664,8 +730,9 @@ size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs, } size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx, - const void* dict, size_t dictSize, - ZSTD_parameters params, unsigned long long pledgedSrcSize) + const void* dict, size_t dictSize, + ZSTD_parameters params, + unsigned long long pledgedSrcSize) { DEBUGLOG(5, "ZSTDMT_initCStream_advanced"); return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, NULL, params, pledgedSrcSize); @@ -701,19 +768,8 @@ size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) { static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame) { - size_t const dstBufferCapacity = ZSTD_compressBound(srcSize); - buffer_t const dstBuffer = ZSTDMT_getBuffer(zcs->buffPool, dstBufferCapacity); - ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(zcs->cctxPool); unsigned const jobID = zcs->nextJobID & zcs->jobIDMask; - if ((cctx==NULL) || (dstBuffer.start==NULL)) { - zcs->jobs[jobID].jobCompleted = 1; - zcs->nextJobID++; - ZSTDMT_waitForAllJobsCompleted(zcs); - ZSTDMT_releaseAllJobResources(zcs); - return ERROR(memory_allocation); - } - DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ", zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize); zcs->jobs[jobID].src = zcs->inBuff.buffer; @@ -726,8 +782,9 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsi if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0; zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL; zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize; - zcs->jobs[jobID].dstBuff = dstBuffer; - zcs->jobs[jobID].cctx = cctx; + zcs->jobs[jobID].dstBuff = g_nullBuffer; + zcs->jobs[jobID].cctxPool = zcs->cctxPool; + zcs->jobs[jobID].bufPool = zcs->bufPool; zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0); zcs->jobs[jobID].lastChunk = endFrame; zcs->jobs[jobID].jobCompleted = 0; @@ -735,11 +792,13 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsi zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex; zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond; + if (zcs->params.fParams.checksumFlag) + XXH64_update(&zcs->xxhState, (const char*)zcs->inBuff.buffer.start + zcs->dictSize, srcSize); + /* get a new buffer for next input */ if (!endFrame) { size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize); - DEBUGLOG(5, "ZSTDMT_createCompressionJob::endFrame = %u", endFrame); - zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize); + zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->bufPool); if (zcs->inBuff.buffer.start == NULL) { /* not enough memory to allocate next input buffer */ zcs->jobs[jobID].jobCompleted = 1; zcs->nextJobID++; @@ -747,18 +806,12 @@ static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsi ZSTDMT_releaseAllJobResources(zcs); return ERROR(memory_allocation); } - DEBUGLOG(5, "inBuff currently filled to %u", (U32)zcs->inBuff.filled); zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize; - DEBUGLOG(5, "new job : inBuff filled to %u, with %u dict and %u src", - (U32)zcs->inBuff.filled, (U32)newDictSize, - (U32)(zcs->inBuff.filled - newDictSize)); memmove(zcs->inBuff.buffer.start, (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize, zcs->inBuff.filled); - DEBUGLOG(5, "new inBuff pre-filled"); zcs->dictSize = newDictSize; } else { /* if (endFrame==1) */ - DEBUGLOG(5, "ZSTDMT_createCompressionJob::endFrame = %u", endFrame); zcs->inBuff.buffer = g_nullBuffer; zcs->inBuff.filled = 0; zcs->dictSize = 0; @@ -804,11 +857,8 @@ static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsi ZSTDMT_releaseAllJobResources(zcs); return job.cSize; } - ZSTDMT_releaseCCtx(zcs->cctxPool, job.cctx); - zcs->jobs[wJobID].cctx = NULL; DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag); if (zcs->params.fParams.checksumFlag) { - XXH64_update(&zcs->xxhState, (const char*)job.srcStart + job.dictSize, job.srcSize); if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) { /* write checksum at end of last section */ U32 const checksum = (U32)XXH64_digest(&zcs->xxhState); DEBUGLOG(5, "writing checksum : %08X \n", checksum); @@ -816,9 +866,6 @@ static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsi job.cSize += 4; zcs->jobs[wJobID].cSize += 4; } } - ZSTDMT_releaseBuffer(zcs->buffPool, job.src); - zcs->jobs[wJobID].srcStart = NULL; - zcs->jobs[wJobID].src = g_nullBuffer; zcs->jobs[wJobID].jobScanned = 1; } { size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos); @@ -828,7 +875,7 @@ static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsi job.dstFlushed += toWrite; } if (job.dstFlushed == job.cSize) { /* output buffer fully flushed => move to next one */ - ZSTDMT_releaseBuffer(zcs->buffPool, job.dstBuff); + ZSTDMT_releaseBuffer(zcs->bufPool, job.dstBuff); zcs->jobs[wJobID].dstBuff = g_nullBuffer; zcs->jobs[wJobID].jobCompleted = 0; zcs->doneJobID++; @@ -852,18 +899,18 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx, ZSTD_inBuffer* input, ZSTD_EndDirective endOp) { - size_t const newJobThreshold = mtctx->dictSize + mtctx->targetSectionSize + mtctx->marginSize; + size_t const newJobThreshold = mtctx->dictSize + mtctx->targetSectionSize; assert(output->pos <= output->size); assert(input->pos <= input->size); if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) { /* current frame being ended. Only flush/end are allowed. Or start new frame with init */ return ERROR(stage_wrong); } - if (mtctx->nbThreads==1) { + if (mtctx->nbThreads==1) { /* delegate to single-thread (synchronous) */ return ZSTD_compressStream_generic(mtctx->cctxPool->cctx[0], output, input, endOp); } - /* single-pass shortcut (note : this is blocking-mode) */ + /* single-pass shortcut (note : this is synchronous-mode) */ if ( (mtctx->nextJobID==0) /* just started */ && (mtctx->inBuff.filled==0) /* nothing buffered */ && (endOp==ZSTD_e_end) /* end order */ @@ -875,20 +922,25 @@ size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx, if (ZSTD_isError(cSize)) return cSize; input->pos = input->size; output->pos += cSize; - ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer); /* was allocated in initStream */ + ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer); /* was allocated in initStream */ mtctx->allJobsCompleted = 1; mtctx->frameEnded = 1; return 0; } /* fill input buffer */ - if ((input->src) && (mtctx->inBuff.buffer.start)) { /* support NULL input */ - size_t const toLoad = MIN(input->size - input->pos, mtctx->inBuffSize - mtctx->inBuff.filled); - DEBUGLOG(2, "inBuff:%08X; inBuffSize=%u; ToCopy=%u", (U32)(size_t)mtctx->inBuff.buffer.start, (U32)mtctx->inBuffSize, (U32)toLoad); - memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad); - input->pos += toLoad; - mtctx->inBuff.filled += toLoad; - } + if (input->size > input->pos) { /* support NULL input */ + if (mtctx->inBuff.buffer.start == NULL) { + mtctx->inBuff.buffer = ZSTDMT_getBuffer(mtctx->bufPool); + if (mtctx->inBuff.buffer.start == NULL) return ERROR(memory_allocation); + mtctx->inBuff.filled = 0; + } + { size_t const toLoad = MIN(input->size - input->pos, mtctx->inBuffSize - mtctx->inBuff.filled); + DEBUGLOG(5, "inBuff:%08X; inBuffSize=%u; ToCopy=%u", (U32)(size_t)mtctx->inBuff.buffer.start, (U32)mtctx->inBuffSize, (U32)toLoad); + memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad); + input->pos += toLoad; + mtctx->inBuff.filled += toLoad; + } } if ( (mtctx->inBuff.filled >= newJobThreshold) /* filled enough : let's compress */ && (mtctx->nextJobID <= mtctx->doneJobID + mtctx->jobIDMask) ) { /* avoid overwriting job round buffer */ diff --git a/lib/compress/zstdmt_compress.h b/lib/compress/zstdmt_compress.h index a6e1759b9..7584007f1 100644 --- a/lib/compress/zstdmt_compress.h +++ b/lib/compress/zstdmt_compress.h @@ -16,8 +16,8 @@ /* Note : This is an internal API. - * Some methods are still exposed (ZSTDLIB_API), because for some time, - * it used to be the only way to invoke MT compression. + * Some methods are still exposed (ZSTDLIB_API), + * because it used to be the only way to invoke MT compression. * Now, it's recommended to use ZSTD_compress_generic() instead. * These methods will stop being exposed in a future version */ @@ -68,7 +68,7 @@ ZSTDLIB_API size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx, const void* src, size_t srcSize, const ZSTD_CDict* cdict, ZSTD_parameters const params, - unsigned overlapRLog); + unsigned overlapRLog); /* overlapRLog = 9 - overlapLog */ ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx, const void* dict, size_t dictSize, /* dict can be released after init, a local copy is preserved within zcs */ diff --git a/tests/Makefile b/tests/Makefile index 345e0e8eb..4784136e1 100644 --- a/tests/Makefile +++ b/tests/Makefile @@ -79,7 +79,7 @@ all32: fullbench32 fuzzer32 zstreamtest32 allnothread: fullbench fuzzer paramgrill datagen decodecorpus -dll: fuzzer-dll zstreamtest-dll +dll: fuzzer-dll zstreamtest-dll zstd: $(MAKE) -C $(PRGDIR) $@ @@ -108,11 +108,11 @@ fullbench-dll: $(PRGDIR)/datagen.c fullbench.c $(MAKE) -C $(ZSTDDIR) libzstd $(CC) $(FLAGS) $^ -o $@$(EXT) -DZSTD_DLL_IMPORT=1 $(ZSTDDIR)/dll/libzstd.dll -fuzzer : $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c fuzzer.c - $(CC) $(FLAGS) $^ -o $@$(EXT) - -fuzzer32 : $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c fuzzer.c - $(CC) -m32 $(FLAGS) $^ -o $@$(EXT) +fuzzer : CPPFLAGS += $(MULTITHREAD_CPP) +fuzzer : LDFLAGS += $(MULTITHREAD_LD) +fuzzer32: CFLAGS += -m32 +fuzzer fuzzer32 : $(ZSTD_FILES) $(ZDICT_FILES) $(PRGDIR)/datagen.c fuzzer.c + $(CC) $(FLAGS) $^ -o $@$(EXT) fuzzer-dll : LDFLAGS+= -L$(ZSTDDIR) -lzstd fuzzer-dll : $(ZSTDDIR)/common/xxhash.c $(PRGDIR)/datagen.c fuzzer.c diff --git a/tests/fuzzer.c b/tests/fuzzer.c index c9461ee90..904ce6fd8 100644 --- a/tests/fuzzer.c +++ b/tests/fuzzer.c @@ -51,14 +51,14 @@ static const U32 nbTestsDefault = 30000; /*-************************************ * Display Macros **************************************/ -#define DISPLAY(...) fprintf(stderr, __VA_ARGS__) +#define DISPLAY(...) fprintf(stdout, __VA_ARGS__) #define DISPLAYLEVEL(l, ...) if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); } static U32 g_displayLevel = 2; #define DISPLAYUPDATE(l, ...) if (g_displayLevel>=l) { \ if ((FUZ_clockSpan(g_displayClock) > g_refreshRate) || (g_displayLevel>=4)) \ { g_displayClock = clock(); DISPLAY(__VA_ARGS__); \ - if (g_displayLevel>=4) fflush(stderr); } } + if (g_displayLevel>=4) fflush(stdout); } } static const clock_t g_refreshRate = CLOCKS_PER_SEC / 6; static clock_t g_displayClock = 0; @@ -97,7 +97,161 @@ static unsigned FUZ_highbit32(U32 v32) /*============================================= -* Basic Unit tests +* Memory Tests +=============================================*/ +#if defined(__APPLE__) && defined(__MACH__) + +#include /* malloc_size */ + +typedef struct { + unsigned long long totalMalloc; + size_t currentMalloc; + size_t peakMalloc; + unsigned nbMalloc; + unsigned nbFree; +} mallocCounter_t; + +static const mallocCounter_t INIT_MALLOC_COUNTER = { 0, 0, 0, 0, 0 }; + +static void* FUZ_mallocDebug(void* counter, size_t size) +{ + mallocCounter_t* const mcPtr = (mallocCounter_t*)counter; + void* const ptr = malloc(size); + if (ptr==NULL) return NULL; + DISPLAYLEVEL(4, "allocating %u KB => effectively %u KB \n", + (U32)(size >> 10), (U32)(malloc_size(ptr) >> 10)); /* OS-X specific */ + mcPtr->totalMalloc += size; + mcPtr->currentMalloc += size; + if (mcPtr->currentMalloc > mcPtr->peakMalloc) + mcPtr->peakMalloc = mcPtr->currentMalloc; + mcPtr->nbMalloc += 1; + return ptr; +} + +static void FUZ_freeDebug(void* counter, void* address) +{ + mallocCounter_t* const mcPtr = (mallocCounter_t*)counter; + DISPLAYLEVEL(4, "freeing %u KB \n", (U32)(malloc_size(address) >> 10)); + mcPtr->nbFree += 1; + mcPtr->currentMalloc -= malloc_size(address); /* OS-X specific */ + free(address); +} + +static void FUZ_displayMallocStats(mallocCounter_t count) +{ + DISPLAYLEVEL(3, "peak:%u KB, nbMallocs:%u, total:%u KB \n", + (U32)(count.peakMalloc >> 10), + count.nbMalloc, + (U32)(count.totalMalloc >> 10)); +} + +#define CHECK_Z(f) { \ + size_t const err = f; \ + if (ZSTD_isError(err)) { \ + DISPLAY("Error => %s : %s ", \ + #f, ZSTD_getErrorName(err)); \ + exit(1); \ +} } + +static int FUZ_mallocTests(unsigned seed, double compressibility) +{ + size_t const inSize = 64 MB + 16 MB + 4 MB + 1 MB + 256 KB + 64 KB; /* 85.3 MB */ + size_t const outSize = ZSTD_compressBound(inSize); + void* const inBuffer = malloc(inSize); + void* const outBuffer = malloc(outSize); + + /* test only played in verbose mode, as they are long */ + if (g_displayLevel<3) return 0; + + /* Create compressible noise */ + if (!inBuffer || !outBuffer) { + DISPLAY("Not enough memory, aborting\n"); + exit(1); + } + RDG_genBuffer(inBuffer, inSize, compressibility, 0. /*auto*/, seed); + + /* simple compression tests */ + { int compressionLevel; + for (compressionLevel=1; compressionLevel<=6; compressionLevel++) { + mallocCounter_t malcount = INIT_MALLOC_COUNTER; + ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount }; + ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(cMem); + CHECK_Z( ZSTD_compressCCtx(cctx, outBuffer, outSize, inBuffer, inSize, compressionLevel) ); + ZSTD_freeCCtx(cctx); + DISPLAYLEVEL(3, "compressCCtx level %i : ", compressionLevel); + FUZ_displayMallocStats(malcount); + } } + + /* streaming compression tests */ + { int compressionLevel; + for (compressionLevel=1; compressionLevel<=6; compressionLevel++) { + mallocCounter_t malcount = INIT_MALLOC_COUNTER; + ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount }; + ZSTD_CCtx* const cstream = ZSTD_createCStream_advanced(cMem); + ZSTD_outBuffer out = { outBuffer, outSize, 0 }; + ZSTD_inBuffer in = { inBuffer, inSize, 0 }; + CHECK_Z( ZSTD_initCStream(cstream, compressionLevel) ); + CHECK_Z( ZSTD_compressStream(cstream, &out, &in) ); + CHECK_Z( ZSTD_endStream(cstream, &out) ); + ZSTD_freeCStream(cstream); + DISPLAYLEVEL(3, "compressStream level %i : ", compressionLevel); + FUZ_displayMallocStats(malcount); + } } + + /* advanced MT API test */ + { U32 nbThreads; + for (nbThreads=1; nbThreads<=4; nbThreads++) { + int compressionLevel; + for (compressionLevel=1; compressionLevel<=6; compressionLevel++) { + mallocCounter_t malcount = INIT_MALLOC_COUNTER; + ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount }; + ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(cMem); + ZSTD_outBuffer out = { outBuffer, outSize, 0 }; + ZSTD_inBuffer in = { inBuffer, inSize, 0 }; + CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, (U32)compressionLevel) ); + CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_nbThreads, nbThreads) ); + while ( ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_end) ) {} + ZSTD_freeCCtx(cctx); + DISPLAYLEVEL(3, "compress_generic,-T%u,end level %i : ", + nbThreads, compressionLevel); + FUZ_displayMallocStats(malcount); + } } } + + /* advanced MT streaming API test */ + { U32 nbThreads; + for (nbThreads=1; nbThreads<=4; nbThreads++) { + int compressionLevel; + for (compressionLevel=1; compressionLevel<=6; compressionLevel++) { + mallocCounter_t malcount = INIT_MALLOC_COUNTER; + ZSTD_customMem const cMem = { FUZ_mallocDebug, FUZ_freeDebug, &malcount }; + ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(cMem); + ZSTD_outBuffer out = { outBuffer, outSize, 0 }; + ZSTD_inBuffer in = { inBuffer, inSize, 0 }; + CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_compressionLevel, (U32)compressionLevel) ); + CHECK_Z( ZSTD_CCtx_setParameter(cctx, ZSTD_p_nbThreads, nbThreads) ); + CHECK_Z( ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_continue) ); + while ( ZSTD_compress_generic(cctx, &out, &in, ZSTD_e_end) ) {} + ZSTD_freeCCtx(cctx); + DISPLAYLEVEL(3, "compress_generic,-T%u,continue level %i : ", + nbThreads, compressionLevel); + FUZ_displayMallocStats(malcount); + } } } + + return 0; +} + +#else + +static int FUZ_mallocTests(unsigned seed, double compressibility) +{ + (void)seed; (void)compressibility; + return 0; +} + +#endif + +/*============================================= +* Unit tests =============================================*/ #define CHECK_V(var, fn) size_t const var = fn; if (ZSTD_isError(var)) goto _output_error @@ -108,7 +262,8 @@ static int basicUnitTests(U32 seed, double compressibility) { size_t const CNBuffSize = 5 MB; void* const CNBuffer = malloc(CNBuffSize); - void* const compressedBuffer = malloc(ZSTD_compressBound(CNBuffSize)); + size_t const compressedBufferSize = ZSTD_compressBound(CNBuffSize); + void* const compressedBuffer = malloc(compressedBufferSize); void* const decodedBuffer = malloc(CNBuffSize); ZSTD_DCtx* dctx = ZSTD_createDCtx(); int testResult = 0; @@ -139,7 +294,7 @@ static int basicUnitTests(U32 seed, double compressibility) { ZSTD_CCtx* cctx = ZSTD_createCCtx(); if (cctx==NULL) goto _output_error; CHECKPLUS(r, ZSTD_compressCCtx(cctx, - compressedBuffer, ZSTD_compressBound(CNBuffSize), + compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, 1), cSize=r ); DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", (U32)cSize, (double)cSize/CNBuffSize*100); @@ -226,7 +381,7 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(4, "test%3i : simple compression test with static CCtx : ", testNb++); CHECKPLUS(r, ZSTD_compressCCtx(staticCCtx, - compressedBuffer, ZSTD_compressBound(CNBuffSize), + compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, STATIC_CCTX_LEVEL), cSize=r ); DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", @@ -295,7 +450,7 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(4, "test%3i : compress %u bytes with 2 threads : ", testNb++, (U32)CNBuffSize); CHECKPLUS(r, ZSTDMT_compressCCtx(mtctx, - compressedBuffer, ZSTD_compressBound(CNBuffSize), + compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, 1), cSize=r ); @@ -321,6 +476,23 @@ static int basicUnitTests(U32 seed, double compressibility) } } DISPLAYLEVEL(4, "OK \n"); + DISPLAYLEVEL(4, "test%3i : compress -T2 with checksum : ", testNb++); + { ZSTD_parameters params = ZSTD_getParams(1, CNBuffSize, 0); + params.fParams.checksumFlag = 1; + params.fParams.contentSizeFlag = 1; + CHECKPLUS(r, ZSTDMT_compress_advanced(mtctx, + compressedBuffer, compressedBufferSize, + CNBuffer, CNBuffSize, + NULL, params, 3 /*overlapRLog*/), + cSize=r ); + } + DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", (U32)cSize, (double)cSize/CNBuffSize*100); + + DISPLAYLEVEL(4, "test%3i : decompress %u bytes : ", testNb++, (U32)CNBuffSize); + { size_t const r = ZSTD_decompress(decodedBuffer, CNBuffSize, compressedBuffer, cSize); + if (r != CNBuffSize) goto _output_error; } + DISPLAYLEVEL(4, "OK \n"); + ZSTDMT_freeCCtx(mtctx); } @@ -382,7 +554,7 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(4, "test%3i : compress with flat dictionary : ", testNb++); cSize = 0; - CHECKPLUS(r, ZSTD_compressEnd(ctxOrig, compressedBuffer, ZSTD_compressBound(CNBuffSize), + CHECKPLUS(r, ZSTD_compressEnd(ctxOrig, compressedBuffer, compressedBufferSize, (const char*)CNBuffer + dictSize, CNBuffSize - dictSize), cSize += r); DISPLAYLEVEL(4, "OK (%u bytes : %.2f%%)\n", (U32)cSize, (double)cSize/CNBuffSize*100); @@ -398,7 +570,7 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(4, "test%3i : compress with duplicated context : ", testNb++); { size_t const cSizeOrig = cSize; cSize = 0; - CHECKPLUS(r, ZSTD_compressEnd(ctxDuplicated, compressedBuffer, ZSTD_compressBound(CNBuffSize), + CHECKPLUS(r, ZSTD_compressEnd(ctxDuplicated, compressedBuffer, compressedBufferSize, (const char*)CNBuffer + dictSize, CNBuffSize - dictSize), cSize += r); if (cSize != cSizeOrig) goto _output_error; /* should be identical ==> same size */ @@ -483,7 +655,7 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(4, "OK : %u \n", dictID); DISPLAYLEVEL(4, "test%3i : compress with dictionary : ", testNb++); - cSize = ZSTD_compress_usingDict(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize), + cSize = ZSTD_compress_usingDict(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, dictBuffer, dictSize, 4); if (ZSTD_isError(cSize)) goto _output_error; @@ -521,7 +693,7 @@ static int basicUnitTests(U32 seed, double compressibility) 1 /* byReference */, ZSTD_dm_auto, cParams, ZSTD_defaultCMem); DISPLAYLEVEL(4, "(size : %u) : ", (U32)ZSTD_sizeof_CDict(cdict)); - cSize = ZSTD_compress_usingCDict(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize), + cSize = ZSTD_compress_usingCDict(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, cdict); ZSTD_freeCDict(cdict); if (ZSTD_isError(cSize)) goto _output_error; @@ -556,7 +728,7 @@ static int basicUnitTests(U32 seed, double compressibility) goto _output_error; } cSize = ZSTD_compress_usingCDict(cctx, - compressedBuffer, ZSTD_compressBound(CNBuffSize), + compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, cdict); if (ZSTD_isError(cSize)) { DISPLAY("ZSTD_compress_usingCDict failed "); @@ -570,7 +742,7 @@ static int basicUnitTests(U32 seed, double compressibility) { ZSTD_frameParameters const fParams = { 0 /* frameSize */, 1 /* checksum */, 1 /* noDictID*/ }; ZSTD_compressionParameters const cParams = ZSTD_getCParams(1, CNBuffSize, dictSize); ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dictBuffer, dictSize, 1 /*byRef*/, ZSTD_dm_auto, cParams, ZSTD_defaultCMem); - cSize = ZSTD_compress_usingCDict_advanced(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize), + cSize = ZSTD_compress_usingCDict_advanced(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, cdict, fParams); ZSTD_freeCDict(cdict); if (ZSTD_isError(cSize)) goto _output_error; @@ -594,7 +766,7 @@ static int basicUnitTests(U32 seed, double compressibility) DISPLAYLEVEL(4, "test%3i : ZSTD_compress_advanced, no dictID : ", testNb++); { ZSTD_parameters p = ZSTD_getParams(3, CNBuffSize, dictSize); p.fParams.noDictIDFlag = 1; - cSize = ZSTD_compress_advanced(cctx, compressedBuffer, ZSTD_compressBound(CNBuffSize), + cSize = ZSTD_compress_advanced(cctx, compressedBuffer, compressedBufferSize, CNBuffer, CNBuffSize, dictBuffer, dictSize, p); if (ZSTD_isError(cSize)) goto _output_error; @@ -902,6 +1074,7 @@ static size_t FUZ_randomLength(U32* seed, U32 maxLog) goto _output_error; \ } } +#undef CHECK_Z #define CHECK_Z(f) { \ size_t const err = f; \ if (ZSTD_isError(err)) { \ @@ -1245,6 +1418,7 @@ int main(int argc, const char** argv) U32 mainPause = 0; U32 maxDuration = 0; int bigTests = 1; + U32 memTestsOnly = 0; const char* const programName = argv[0]; /* Check command line */ @@ -1255,6 +1429,7 @@ int main(int argc, const char** argv) /* Handle commands. Aggregated commands are allowed */ if (argument[0]=='-') { + if (!strcmp(argument, "--memtest")) { memTestsOnly=1; continue; } if (!strcmp(argument, "--no-big-tests")) { bigTests=0; continue; } argument++; @@ -1326,6 +1501,11 @@ int main(int argc, const char** argv) DISPLAY("Seed = %u\n", seed); if (proba!=FUZ_compressibility_default) DISPLAY("Compressibility : %u%%\n", proba); + if (memTestsOnly) { + g_displayLevel = MAX(3, g_displayLevel); + return FUZ_mallocTests(seed, ((double)proba) / 100); + } + if (nbTests < testNb) nbTests = testNb; if (testNb==0) diff --git a/tests/playTests.sh b/tests/playTests.sh index 2e1cc6826..88a1c2ab4 100755 --- a/tests/playTests.sh +++ b/tests/playTests.sh @@ -7,17 +7,17 @@ die() { roundTripTest() { if [ -n "$3" ]; then - local_c="$3" - local_p="$2" + cLevel="$3" + proba="$2" else - local_c="$2" - local_p="" + cLevel="$2" + proba="" fi rm -f tmp1 tmp2 - $ECHO "roundTripTest: ./datagen $1 $local_p | $ZSTD -v$local_c | $ZSTD -d" - ./datagen $1 $local_p | $MD5SUM > tmp1 - ./datagen $1 $local_p | $ZSTD --ultra -v$local_c | $ZSTD -d | $MD5SUM > tmp2 + $ECHO "roundTripTest: ./datagen $1 $proba | $ZSTD -v$cLevel | $ZSTD -d" + ./datagen $1 $proba | $MD5SUM > tmp1 + ./datagen $1 $proba | $ZSTD --ultra -v$cLevel | $ZSTD -d | $MD5SUM > tmp2 $DIFF -q tmp1 tmp2 } @@ -625,16 +625,15 @@ roundTripTest -g35000000 -P75 10 roundTripTest -g35000000 -P75 11 roundTripTest -g35000000 -P75 12 -roundTripTest -g18000000 -P80 13 -roundTripTest -g18000000 -P80 14 -roundTripTest -g18000000 -P80 15 -roundTripTest -g18000000 -P80 16 -roundTripTest -g18000000 -P80 17 +roundTripTest -g18000013 -P80 13 +roundTripTest -g18000014 -P80 14 +roundTripTest -g18000015 -P80 15 +roundTripTest -g18000016 -P80 16 +roundTripTest -g18000017 -P80 17 +roundTripTest -g18000018 -P94 18 +roundTripTest -g18000019 -P94 19 -roundTripTest -g50000000 -P94 18 -roundTripTest -g50000000 -P94 19 - -roundTripTest -g99000000 -P99 20 +roundTripTest -g68000020 -P99 20 roundTripTest -g6000000000 -P99 1 fileRoundTripTest -g4193M -P99 1 diff --git a/tests/zstreamtest.c b/tests/zstreamtest.c index 9b2b8eaf8..3e551e331 100644 --- a/tests/zstreamtest.c +++ b/tests/zstreamtest.c @@ -95,19 +95,6 @@ unsigned int FUZ_rand(unsigned int* seedPtr) return rand32 >> 5; } -static void* allocFunction(void* opaque, size_t size) -{ - void* address = malloc(size); - (void)opaque; - return address; -} - -static void freeFunction(void* opaque, void* address) -{ - (void)opaque; - free(address); -} - /*====================================================== * Basic Unit tests @@ -1390,13 +1377,12 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double /* multi-segments compression test */ XXH64_reset(&xxhState, 0); { ZSTD_outBuffer outBuff = { cBuffer, cBufferSize, 0 } ; - U32 n; - for (n=0, cSize=0, totalTestSize=0 ; totalTestSize < maxTestSize ; n++) { + for (cSize=0, totalTestSize=0 ; (totalTestSize < maxTestSize) ; ) { /* compress random chunks into randomly sized dst buffers */ size_t const randomSrcSize = FUZ_randomLength(&lseed, maxSampleLog); size_t const srcSize = MIN(maxTestSize-totalTestSize, randomSrcSize); size_t const srcStart = FUZ_rand(&lseed) % (srcBufferSize - srcSize); - size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog); + size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog+1); size_t const dstBuffSize = MIN(cBufferSize - cSize, randomDstSize); ZSTD_EndDirective const flush = (FUZ_rand(&lseed) & 15) ? ZSTD_e_continue : ZSTD_e_flush; ZSTD_inBuffer inBuff = { srcBuffer+srcStart, srcSize, 0 }; @@ -1415,7 +1401,7 @@ static int fuzzerTests_newAPI(U32 seed, U32 nbTests, unsigned startTest, double { size_t remainingToFlush = (size_t)(-1); while (remainingToFlush) { ZSTD_inBuffer inBuff = { NULL, 0, 0 }; - size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog); + size_t const randomDstSize = FUZ_randomLength(&lseed, maxSampleLog+1); size_t const adjustedDstSize = MIN(cBufferSize - cSize, randomDstSize); outBuff.size = outBuff.pos + adjustedDstSize; DISPLAYLEVEL(5, "End-flush into dst buffer of size %u \n", (U32)adjustedDstSize); @@ -1543,7 +1529,6 @@ int main(int argc, const char** argv) int bigTests = (sizeof(size_t) == 8); e_api selected_api = simple_api; const char* const programName = argv[0]; - ZSTD_customMem const customMem = { allocFunction, freeFunction, NULL }; ZSTD_customMem const customNULL = ZSTD_defaultCMem; /* Check command line */ @@ -1657,10 +1642,7 @@ int main(int argc, const char** argv) if (testNb==0) { result = basicUnitTests(0, ((double)proba) / 100, customNULL); /* constant seed for predictability */ - if (!result) { - DISPLAYLEVEL(3, "Unit tests using customMem :\n") - result = basicUnitTests(0, ((double)proba) / 100, customMem); /* use custom memory allocation functions */ - } } + } if (!result) { switch(selected_api)