2 * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
5 * This source code is licensed under both the BSD-style license (found in the
6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7 * in the COPYING file in the root directory of this source tree).
8 * You may select, at your option, one of the above-listed licenses.
12 /* ====== Tuning parameters ====== */
13 #define ZSTDMT_NBTHREADS_MAX 200
14 #define ZSTDMT_OVERLAPLOG_DEFAULT 6
17 /* ====== Compiler specifics ====== */
19 # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */
23 /* ====== Dependencies ====== */
24 #include <string.h> /* memcpy, memset */
25 #include "pool.h" /* threadpool */
26 #include "threading.h" /* mutex */
27 #include "zstd_internal.h" /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
28 #include "zstdmt_compress.h"
31 /* ====== Debug ====== */
32 #if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
36 # include <sys/times.h>
37 # define DEBUGLOGRAW(l, ...) if (l<=ZSTD_DEBUG) { fprintf(stderr, __VA_ARGS__); }
39 # define DEBUG_PRINTHEX(l,p,n) { \
41 for (debug_u=0; debug_u<(n); debug_u++) \
42 DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
43 DEBUGLOGRAW(l, " \n"); \
46 static unsigned long long GetCurrentClockTimeMicroseconds(void)
48 static clock_t _ticksPerSecond = 0;
49 if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
51 { struct tms junk; clock_t newTicks = (clock_t) times(&junk);
52 return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond); }
55 #define MUTEX_WAIT_TIME_DLEVEL 6
56 #define ZSTD_PTHREAD_MUTEX_LOCK(mutex) { \
57 if (ZSTD_DEBUG >= MUTEX_WAIT_TIME_DLEVEL) { \
58 unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
59 ZSTD_pthread_mutex_lock(mutex); \
60 { unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
61 unsigned long long const elapsedTime = (afterTime-beforeTime); \
62 if (elapsedTime > 1000) { /* or whatever threshold you like; I'm using 1 millisecond here */ \
63 DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
64 elapsedTime, #mutex); \
67 ZSTD_pthread_mutex_lock(mutex); \
73 # define ZSTD_PTHREAD_MUTEX_LOCK(m) ZSTD_pthread_mutex_lock(m)
74 # define DEBUG_PRINTHEX(l,p,n) {}
79 /* ===== Buffer Pool ===== */
80 /* a single Buffer Pool can be invoked from multiple threads in parallel */
82 typedef struct buffer_s {
87 static const buffer_t g_nullBuffer = { NULL, 0 };
89 typedef struct ZSTDMT_bufferPool_s {
90 ZSTD_pthread_mutex_t poolMutex;
92 unsigned totalBuffers;
95 buffer_t bTable[1]; /* variable size */
98 static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads, ZSTD_customMem cMem)
100 unsigned const maxNbBuffers = 2*nbThreads + 3;
101 ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
102 sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
103 if (bufPool==NULL) return NULL;
104 if (ZSTD_pthread_mutex_init(&bufPool->poolMutex, NULL)) {
105 ZSTD_free(bufPool, cMem);
108 bufPool->bufferSize = 64 KB;
109 bufPool->totalBuffers = maxNbBuffers;
110 bufPool->nbBuffers = 0;
111 bufPool->cMem = cMem;
115 static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
118 DEBUGLOG(3, "ZSTDMT_freeBufferPool (address:%08X)", (U32)(size_t)bufPool);
119 if (!bufPool) return; /* compatibility with free on NULL */
120 for (u=0; u<bufPool->totalBuffers; u++) {
121 DEBUGLOG(4, "free buffer %2u (address:%08X)", u, (U32)(size_t)bufPool->bTable[u].start);
122 ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
124 ZSTD_pthread_mutex_destroy(&bufPool->poolMutex);
125 ZSTD_free(bufPool, bufPool->cMem);
128 /* only works at initialization, not during compression */
129 static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
131 size_t const poolSize = sizeof(*bufPool)
132 + (bufPool->totalBuffers - 1) * sizeof(buffer_t);
134 size_t totalBufferSize = 0;
135 ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
136 for (u=0; u<bufPool->totalBuffers; u++)
137 totalBufferSize += bufPool->bTable[u].size;
138 ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
140 return poolSize + totalBufferSize;
143 static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* bufPool, size_t bSize)
145 bufPool->bufferSize = bSize;
148 /** ZSTDMT_getBuffer() :
149 * assumption : bufPool must be valid */
150 static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
152 size_t const bSize = bufPool->bufferSize;
153 DEBUGLOG(5, "ZSTDMT_getBuffer");
154 ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
155 if (bufPool->nbBuffers) { /* try to use an existing buffer */
156 buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)];
157 size_t const availBufferSize = buf.size;
158 bufPool->bTable[bufPool->nbBuffers] = g_nullBuffer;
159 if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize)) {
160 /* large enough, but not too much */
161 ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
164 /* size conditions not respected : scratch this buffer, create new one */
165 DEBUGLOG(5, "existing buffer does not meet size conditions => freeing");
166 ZSTD_free(buf.start, bufPool->cMem);
168 ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
169 /* create new buffer */
170 DEBUGLOG(5, "create a new buffer");
172 void* const start = ZSTD_malloc(bSize, bufPool->cMem);
173 buffer.start = start; /* note : start can be NULL if malloc fails ! */
174 buffer.size = (start==NULL) ? 0 : bSize;
179 /* store buffer for later re-use, up to pool capacity */
180 static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
182 if (buf.start == NULL) return; /* compatible with release on NULL */
183 DEBUGLOG(5, "ZSTDMT_releaseBuffer");
184 ZSTD_pthread_mutex_lock(&bufPool->poolMutex);
185 if (bufPool->nbBuffers < bufPool->totalBuffers) {
186 bufPool->bTable[bufPool->nbBuffers++] = buf; /* stored for later use */
187 ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
190 ZSTD_pthread_mutex_unlock(&bufPool->poolMutex);
191 /* Reached bufferPool capacity (should not happen) */
192 DEBUGLOG(5, "buffer pool capacity reached => freeing ");
193 ZSTD_free(buf.start, bufPool->cMem);
196 /* Sets parameters relevant to the compression job, initializing others to
197 * default values. Notably, nbThreads should probably be zero. */
198 static ZSTD_CCtx_params ZSTDMT_makeJobCCtxParams(ZSTD_CCtx_params const params)
200 ZSTD_CCtx_params jobParams;
201 memset(&jobParams, 0, sizeof(jobParams));
203 jobParams.cParams = params.cParams;
204 jobParams.fParams = params.fParams;
205 jobParams.compressionLevel = params.compressionLevel;
207 jobParams.ldmParams = params.ldmParams;
211 /* ===== CCtx Pool ===== */
212 /* a single CCtx Pool can be invoked from multiple threads in parallel */
215 ZSTD_pthread_mutex_t poolMutex;
219 ZSTD_CCtx* cctx[1]; /* variable size */
222 /* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
223 static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
226 for (u=0; u<pool->totalCCtx; u++)
227 ZSTD_freeCCtx(pool->cctx[u]); /* note : compatible with free on NULL */
228 ZSTD_pthread_mutex_destroy(&pool->poolMutex);
229 ZSTD_free(pool, pool->cMem);
232 /* ZSTDMT_createCCtxPool() :
233 * implies nbThreads >= 1 , checked by caller ZSTDMT_createCCtx() */
234 static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads,
237 ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
238 sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*), cMem);
239 if (!cctxPool) return NULL;
240 if (ZSTD_pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
241 ZSTD_free(cctxPool, cMem);
244 cctxPool->cMem = cMem;
245 cctxPool->totalCCtx = nbThreads;
246 cctxPool->availCCtx = 1; /* at least one cctx for single-thread mode */
247 cctxPool->cctx[0] = ZSTD_createCCtx_advanced(cMem);
248 if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
249 DEBUGLOG(3, "cctxPool created, with %u threads", nbThreads);
253 /* only works during initialization phase, not during compression */
254 static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
256 ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
257 { unsigned const nbThreads = cctxPool->totalCCtx;
258 size_t const poolSize = sizeof(*cctxPool)
259 + (nbThreads-1)*sizeof(ZSTD_CCtx*);
261 size_t totalCCtxSize = 0;
262 for (u=0; u<nbThreads; u++) {
263 totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctx[u]);
265 ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
266 return poolSize + totalCCtxSize;
270 static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
272 DEBUGLOG(5, "ZSTDMT_getCCtx");
273 ZSTD_pthread_mutex_lock(&cctxPool->poolMutex);
274 if (cctxPool->availCCtx) {
275 cctxPool->availCCtx--;
276 { ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx];
277 ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
280 ZSTD_pthread_mutex_unlock(&cctxPool->poolMutex);
281 DEBUGLOG(5, "create one more CCtx");
282 return ZSTD_createCCtx_advanced(cctxPool->cMem); /* note : can be NULL, when creation fails ! */
285 static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
287 if (cctx==NULL) return; /* compatibility with release on NULL */
288 ZSTD_pthread_mutex_lock(&pool->poolMutex);
289 if (pool->availCCtx < pool->totalCCtx)
290 pool->cctx[pool->availCCtx++] = cctx;
292 /* pool overflow : should not happen, since totalCCtx==nbThreads */
293 DEBUGLOG(5, "CCtx pool overflow : free cctx");
296 ZSTD_pthread_mutex_unlock(&pool->poolMutex);
300 /* ===== Thread worker ===== */
304 const void* srcStart;
312 unsigned jobCompleted;
314 ZSTD_pthread_mutex_t* jobCompleted_mutex;
315 ZSTD_pthread_cond_t* jobCompleted_cond;
316 ZSTD_CCtx_params params;
317 const ZSTD_CDict* cdict;
318 ZSTDMT_CCtxPool* cctxPool;
319 ZSTDMT_bufferPool* bufPool;
320 unsigned long long fullFrameSize;
321 } ZSTDMT_jobDescription;
323 /* ZSTDMT_compressChunk() : POOL_function type */
324 void ZSTDMT_compressChunk(void* jobDescription)
326 ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
327 ZSTD_CCtx* cctx = ZSTDMT_getCCtx(job->cctxPool);
328 const void* const src = (const char*)job->srcStart + job->dictSize;
329 buffer_t dstBuff = job->dstBuff;
330 DEBUGLOG(5, "job (first:%u) (last:%u) : dictSize %u, srcSize %u",
331 job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize);
334 job->cSize = ERROR(memory_allocation);
338 if (dstBuff.start == NULL) {
339 dstBuff = ZSTDMT_getBuffer(job->bufPool);
340 if (dstBuff.start==NULL) {
341 job->cSize = ERROR(memory_allocation);
344 job->dstBuff = dstBuff;
347 if (job->cdict) { /* should only happen for first segment */
348 size_t const initError = ZSTD_compressBegin_usingCDict_advanced(cctx, job->cdict, job->params.fParams, job->fullFrameSize);
349 DEBUGLOG(5, "using CDict");
350 if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
351 } else { /* srcStart points at reloaded section */
352 if (!job->firstChunk) job->params.fParams.contentSizeFlag = 0; /* ensure no srcSize control */
353 { ZSTD_CCtx_params jobParams = job->params;
354 size_t const forceWindowError =
355 ZSTD_CCtxParam_setParameter(&jobParams, ZSTD_p_forceMaxWindow, !job->firstChunk);
356 /* Force loading dictionary in "content-only" mode (no header analysis) */
357 size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, job->srcStart, job->dictSize, ZSTD_dm_rawContent, jobParams, job->fullFrameSize);
358 if (ZSTD_isError(initError) || ZSTD_isError(forceWindowError)) {
359 job->cSize = initError;
363 if (!job->firstChunk) { /* flush and overwrite frame header when it's not first segment */
364 size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, 0);
365 if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; }
366 ZSTD_invalidateRepCodes(cctx);
369 DEBUGLOG(5, "Compressing : ");
370 DEBUG_PRINTHEX(4, job->srcStart, 12);
371 job->cSize = (job->lastChunk) ?
372 ZSTD_compressEnd (cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :
373 ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, job->srcSize);
374 DEBUGLOG(5, "compressed %u bytes into %u bytes (first:%u) (last:%u)",
375 (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk);
376 DEBUGLOG(5, "dstBuff.size : %u ; => %s", (U32)dstBuff.size, ZSTD_getErrorName(job->cSize));
379 ZSTDMT_releaseCCtx(job->cctxPool, cctx);
380 ZSTDMT_releaseBuffer(job->bufPool, job->src);
381 job->src = g_nullBuffer; job->srcStart = NULL;
382 ZSTD_PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex);
383 job->jobCompleted = 1;
385 ZSTD_pthread_cond_signal(job->jobCompleted_cond);
386 ZSTD_pthread_mutex_unlock(job->jobCompleted_mutex);
390 /* ------------------------------------------ */
391 /* ===== Multi-threaded compression ===== */
392 /* ------------------------------------------ */
399 struct ZSTDMT_CCtx_s {
401 ZSTDMT_jobDescription* jobs;
402 ZSTDMT_bufferPool* bufPool;
403 ZSTDMT_CCtxPool* cctxPool;
404 ZSTD_pthread_mutex_t jobCompleted_mutex;
405 ZSTD_pthread_cond_t jobCompleted_cond;
406 size_t targetSectionSize;
409 size_t targetDictSize;
411 ZSTD_CCtx_params params;
412 XXH64_state_t xxhState;
417 unsigned allJobsCompleted;
418 unsigned long long frameContentSize;
420 ZSTD_CDict* cdictLocal;
421 const ZSTD_CDict* cdict;
424 static ZSTDMT_jobDescription* ZSTDMT_allocJobsTable(U32* nbJobsPtr, ZSTD_customMem cMem)
426 U32 const nbJobsLog2 = ZSTD_highbit32(*nbJobsPtr) + 1;
427 U32 const nbJobs = 1 << nbJobsLog2;
429 return (ZSTDMT_jobDescription*) ZSTD_calloc(
430 nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
434 size_t ZSTDMT_initializeCCtxParameters(ZSTD_CCtx_params* params, unsigned nbThreads)
436 params->nbThreads = nbThreads;
437 params->overlapSizeLog = ZSTDMT_OVERLAPLOG_DEFAULT;
442 ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbThreads, ZSTD_customMem cMem)
445 U32 nbJobs = nbThreads + 2;
446 DEBUGLOG(3, "ZSTDMT_createCCtx_advanced");
448 if (nbThreads < 1) return NULL;
449 nbThreads = MIN(nbThreads , ZSTDMT_NBTHREADS_MAX);
450 if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
451 /* invalid custom allocator */
454 mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
455 if (!mtctx) return NULL;
456 ZSTDMT_initializeCCtxParameters(&mtctx->params, nbThreads);
458 mtctx->allJobsCompleted = 1;
459 mtctx->factory = POOL_create_advanced(nbThreads, 0, cMem);
460 mtctx->jobs = ZSTDMT_allocJobsTable(&nbJobs, cMem);
461 mtctx->jobIDMask = nbJobs - 1;
462 mtctx->bufPool = ZSTDMT_createBufferPool(nbThreads, cMem);
463 mtctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads, cMem);
464 if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool) {
465 ZSTDMT_freeCCtx(mtctx);
468 if (ZSTD_pthread_mutex_init(&mtctx->jobCompleted_mutex, NULL)) {
469 ZSTDMT_freeCCtx(mtctx);
472 if (ZSTD_pthread_cond_init(&mtctx->jobCompleted_cond, NULL)) {
473 ZSTDMT_freeCCtx(mtctx);
476 DEBUGLOG(3, "mt_cctx created, for %u threads", nbThreads);
480 ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbThreads)
482 return ZSTDMT_createCCtx_advanced(nbThreads, ZSTD_defaultCMem);
485 /* ZSTDMT_releaseAllJobResources() :
486 * note : ensure all workers are killed first ! */
487 static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
490 DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
491 for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
492 DEBUGLOG(4, "job%02u: release dst address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].dstBuff.start);
493 ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
494 mtctx->jobs[jobID].dstBuff = g_nullBuffer;
495 DEBUGLOG(4, "job%02u: release src address %08X", jobID, (U32)(size_t)mtctx->jobs[jobID].src.start);
496 ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].src);
497 mtctx->jobs[jobID].src = g_nullBuffer;
499 memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
500 DEBUGLOG(4, "input: release address %08X", (U32)(size_t)mtctx->inBuff.buffer.start);
501 ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer);
502 mtctx->inBuff.buffer = g_nullBuffer;
503 mtctx->allJobsCompleted = 1;
506 static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* zcs)
508 DEBUGLOG(4, "ZSTDMT_waitForAllJobsCompleted");
509 while (zcs->doneJobID < zcs->nextJobID) {
510 unsigned const jobID = zcs->doneJobID & zcs->jobIDMask;
511 ZSTD_PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
512 while (zcs->jobs[jobID].jobCompleted==0) {
513 DEBUGLOG(5, "waiting for jobCompleted signal from chunk %u", zcs->doneJobID); /* we want to block when waiting for data to flush */
514 ZSTD_pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);
516 ZSTD_pthread_mutex_unlock(&zcs->jobCompleted_mutex);
521 size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
523 if (mtctx==NULL) return 0; /* compatible with free on NULL */
524 POOL_free(mtctx->factory); /* stop and free worker threads */
525 ZSTDMT_releaseAllJobResources(mtctx); /* release job resources into pools first */
526 ZSTD_free(mtctx->jobs, mtctx->cMem);
527 ZSTDMT_freeBufferPool(mtctx->bufPool);
528 ZSTDMT_freeCCtxPool(mtctx->cctxPool);
529 ZSTD_freeCDict(mtctx->cdictLocal);
530 ZSTD_pthread_mutex_destroy(&mtctx->jobCompleted_mutex);
531 ZSTD_pthread_cond_destroy(&mtctx->jobCompleted_cond);
532 ZSTD_free(mtctx, mtctx->cMem);
536 size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
538 if (mtctx == NULL) return 0; /* supports sizeof NULL */
539 return sizeof(*mtctx)
540 + POOL_sizeof(mtctx->factory)
541 + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
542 + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
543 + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
544 + ZSTD_sizeof_CDict(mtctx->cdictLocal);
548 size_t ZSTDMT_CCtxParam_setMTCtxParameter(
549 ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, unsigned value) {
552 case ZSTDMT_p_sectionSize :
553 params->jobSize = value;
555 case ZSTDMT_p_overlapSectionLog :
556 DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);
557 params->overlapSizeLog = (value >= 9) ? 9 : value;
560 return ERROR(parameter_unsupported);
564 size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value)
568 case ZSTDMT_p_sectionSize :
569 return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
570 case ZSTDMT_p_overlapSectionLog :
571 return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
573 return ERROR(parameter_unsupported);
577 /* ------------------------------------------ */
578 /* ===== Multi-threaded compression ===== */
579 /* ------------------------------------------ */
581 static unsigned computeNbChunks(size_t srcSize, unsigned windowLog, unsigned nbThreads) {
582 size_t const chunkSizeTarget = (size_t)1 << (windowLog + 2);
583 size_t const chunkMaxSize = chunkSizeTarget << 2;
584 size_t const passSizeMax = chunkMaxSize * nbThreads;
585 unsigned const multiplier = (unsigned)(srcSize / passSizeMax) + 1;
586 unsigned const nbChunksLarge = multiplier * nbThreads;
587 unsigned const nbChunksMax = (unsigned)(srcSize / chunkSizeTarget) + 1;
588 unsigned const nbChunksSmall = MIN(nbChunksMax, nbThreads);
589 return (multiplier>1) ? nbChunksLarge : nbChunksSmall;
592 static size_t ZSTDMT_compress_advanced_internal(
594 void* dst, size_t dstCapacity,
595 const void* src, size_t srcSize,
596 const ZSTD_CDict* cdict,
597 ZSTD_CCtx_params const params)
599 ZSTD_CCtx_params const jobParams = ZSTDMT_makeJobCCtxParams(params);
600 unsigned const overlapRLog = (params.overlapSizeLog>9) ? 0 : 9-params.overlapSizeLog;
601 size_t const overlapSize = (overlapRLog>=9) ? 0 : (size_t)1 << (params.cParams.windowLog - overlapRLog);
602 unsigned nbChunks = computeNbChunks(srcSize, params.cParams.windowLog, params.nbThreads);
603 size_t const proposedChunkSize = (srcSize + (nbChunks-1)) / nbChunks;
604 size_t const avgChunkSize = ((proposedChunkSize & 0x1FFFF) < 0x7FFF) ? proposedChunkSize + 0xFFFF : proposedChunkSize; /* avoid too small last block */
605 const char* const srcStart = (const char*)src;
606 size_t remainingSrcSize = srcSize;
607 unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbChunks : (unsigned)(dstCapacity / ZSTD_compressBound(avgChunkSize)); /* presumes avgChunkSize >= 256 KB, which should be the case */
608 size_t frameStartPos = 0, dstBufferPos = 0;
610 assert(jobParams.nbThreads == 0);
611 assert(mtctx->cctxPool->totalCCtx == params.nbThreads);
613 DEBUGLOG(4, "nbChunks : %2u (chunkSize : %u bytes) ", nbChunks, (U32)avgChunkSize);
614 if (nbChunks==1) { /* fallback to single-thread mode */
615 ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
616 if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
617 return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, jobParams);
619 assert(avgChunkSize >= 256 KB); /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), which is required for compressWithinDst */
620 ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgChunkSize) );
621 XXH64_reset(&xxh64, 0);
623 if (nbChunks > mtctx->jobIDMask+1) { /* enlarge job table */
624 U32 nbJobs = nbChunks;
625 ZSTD_free(mtctx->jobs, mtctx->cMem);
626 mtctx->jobIDMask = 0;
627 mtctx->jobs = ZSTDMT_allocJobsTable(&nbJobs, mtctx->cMem);
628 if (mtctx->jobs==NULL) return ERROR(memory_allocation);
629 mtctx->jobIDMask = nbJobs - 1;
633 for (u=0; u<nbChunks; u++) {
634 size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize);
635 size_t const dstBufferCapacity = ZSTD_compressBound(chunkSize);
636 buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
637 buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
638 size_t dictSize = u ? overlapSize : 0;
640 mtctx->jobs[u].src = g_nullBuffer;
641 mtctx->jobs[u].srcStart = srcStart + frameStartPos - dictSize;
642 mtctx->jobs[u].dictSize = dictSize;
643 mtctx->jobs[u].srcSize = chunkSize;
644 mtctx->jobs[u].cdict = mtctx->nextJobID==0 ? cdict : NULL;
645 mtctx->jobs[u].fullFrameSize = srcSize;
646 mtctx->jobs[u].params = jobParams;
647 /* do not calculate checksum within sections, but write it in header for first section */
648 if (u!=0) mtctx->jobs[u].params.fParams.checksumFlag = 0;
649 mtctx->jobs[u].dstBuff = dstBuffer;
650 mtctx->jobs[u].cctxPool = mtctx->cctxPool;
651 mtctx->jobs[u].bufPool = mtctx->bufPool;
652 mtctx->jobs[u].firstChunk = (u==0);
653 mtctx->jobs[u].lastChunk = (u==nbChunks-1);
654 mtctx->jobs[u].jobCompleted = 0;
655 mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex;
656 mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond;
658 if (params.fParams.checksumFlag) {
659 XXH64_update(&xxh64, srcStart + frameStartPos, chunkSize);
662 DEBUGLOG(5, "posting job %u (%u bytes)", u, (U32)chunkSize);
663 DEBUG_PRINTHEX(6, mtctx->jobs[u].srcStart, 12);
664 POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]);
666 frameStartPos += chunkSize;
667 dstBufferPos += dstBufferCapacity;
668 remainingSrcSize -= chunkSize;
672 { size_t error = 0, dstPos = 0;
674 for (chunkID=0; chunkID<nbChunks; chunkID++) {
675 DEBUGLOG(5, "waiting for chunk %u ", chunkID);
676 ZSTD_PTHREAD_MUTEX_LOCK(&mtctx->jobCompleted_mutex);
677 while (mtctx->jobs[chunkID].jobCompleted==0) {
678 DEBUGLOG(5, "waiting for jobCompleted signal from chunk %u", chunkID);
679 ZSTD_pthread_cond_wait(&mtctx->jobCompleted_cond, &mtctx->jobCompleted_mutex);
681 ZSTD_pthread_mutex_unlock(&mtctx->jobCompleted_mutex);
682 DEBUGLOG(5, "ready to write chunk %u ", chunkID);
684 mtctx->jobs[chunkID].srcStart = NULL;
685 { size_t const cSize = mtctx->jobs[chunkID].cSize;
686 if (ZSTD_isError(cSize)) error = cSize;
687 if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
688 if (chunkID) { /* note : chunk 0 is written directly at dst, which is correct position */
690 memmove((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize); /* may overlap when chunk compressed within dst */
691 if (chunkID >= compressWithinDst) { /* chunk compressed into its own buffer, which must be released */
692 DEBUGLOG(5, "releasing buffer %u>=%u", chunkID, compressWithinDst);
693 ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[chunkID].dstBuff);
695 mtctx->jobs[chunkID].dstBuff = g_nullBuffer;
698 } /* for (chunkID=0; chunkID<nbChunks; chunkID++) */
700 DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
701 if (params.fParams.checksumFlag) {
702 U32 const checksum = (U32)XXH64_digest(&xxh64);
703 if (dstPos + 4 > dstCapacity) {
704 error = ERROR(dstSize_tooSmall);
706 DEBUGLOG(4, "writing checksum : %08X \n", checksum);
707 MEM_writeLE32((char*)dst + dstPos, checksum);
711 if (!error) DEBUGLOG(4, "compressed size : %u ", (U32)dstPos);
712 return error ? error : dstPos;
716 size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
717 void* dst, size_t dstCapacity,
718 const void* src, size_t srcSize,
719 const ZSTD_CDict* cdict,
720 ZSTD_parameters const params,
723 ZSTD_CCtx_params cctxParams = mtctx->params;
724 cctxParams.cParams = params.cParams;
725 cctxParams.fParams = params.fParams;
726 cctxParams.overlapSizeLog = overlapLog;
727 return ZSTDMT_compress_advanced_internal(mtctx,
734 size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
735 void* dst, size_t dstCapacity,
736 const void* src, size_t srcSize,
737 int compressionLevel)
739 U32 const overlapLog = (compressionLevel >= ZSTD_maxCLevel()) ? 9 : ZSTDMT_OVERLAPLOG_DEFAULT;
740 ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
741 params.fParams.contentSizeFlag = 1;
742 return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
746 /* ====================================== */
747 /* ======= Streaming API ======= */
748 /* ====================================== */
750 size_t ZSTDMT_initCStream_internal(
752 const void* dict, size_t dictSize, ZSTD_dictMode_e dictMode,
753 const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
754 unsigned long long pledgedSrcSize)
756 DEBUGLOG(4, "ZSTDMT_initCStream_internal");
757 /* params are supposed to be fully validated at this point */
758 assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
759 assert(!((dict) && (cdict))); /* either dict or cdict, not both */
760 assert(zcs->cctxPool->totalCCtx == params.nbThreads);
762 if (params.nbThreads==1) {
763 ZSTD_CCtx_params const singleThreadParams = ZSTDMT_makeJobCCtxParams(params);
764 DEBUGLOG(4, "single thread mode");
765 assert(singleThreadParams.nbThreads == 0);
766 return ZSTD_initCStream_internal(zcs->cctxPool->cctx[0],
767 dict, dictSize, cdict,
768 singleThreadParams, pledgedSrcSize);
771 if (zcs->allJobsCompleted == 0) { /* previous compression not correctly finished */
772 ZSTDMT_waitForAllJobsCompleted(zcs);
773 ZSTDMT_releaseAllJobResources(zcs);
774 zcs->allJobsCompleted = 1;
777 zcs->params = params;
778 zcs->frameContentSize = pledgedSrcSize;
780 DEBUGLOG(4,"cdictLocal: %08X", (U32)(size_t)zcs->cdictLocal);
781 ZSTD_freeCDict(zcs->cdictLocal);
782 zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
783 ZSTD_dlm_byCopy, dictMode, /* note : a loadPrefix becomes an internal CDict */
784 params.cParams, zcs->cMem);
785 zcs->cdict = zcs->cdictLocal;
786 if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
788 DEBUGLOG(4,"cdictLocal: %08X", (U32)(size_t)zcs->cdictLocal);
789 ZSTD_freeCDict(zcs->cdictLocal);
790 zcs->cdictLocal = NULL;
794 zcs->targetDictSize = (params.overlapSizeLog==0) ? 0 : (size_t)1 << (params.cParams.windowLog - (9 - params.overlapSizeLog));
795 DEBUGLOG(4, "overlapLog : %u ", params.overlapSizeLog);
796 DEBUGLOG(4, "overlap Size : %u KB", (U32)(zcs->targetDictSize>>10));
797 zcs->targetSectionSize = params.jobSize ? params.jobSize : (size_t)1 << (params.cParams.windowLog + 2);
798 zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize);
799 zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize);
800 DEBUGLOG(4, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10));
801 zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize;
802 ZSTDMT_setBufferSize(zcs->bufPool, MAX(zcs->inBuffSize, ZSTD_compressBound(zcs->targetSectionSize)) );
803 zcs->inBuff.buffer = g_nullBuffer;
808 zcs->allJobsCompleted = 0;
809 if (params.fParams.checksumFlag) XXH64_reset(&zcs->xxhState, 0);
813 size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
814 const void* dict, size_t dictSize,
815 ZSTD_parameters params,
816 unsigned long long pledgedSrcSize)
818 ZSTD_CCtx_params cctxParams = mtctx->params;
819 DEBUGLOG(5, "ZSTDMT_initCStream_advanced");
820 cctxParams.cParams = params.cParams;
821 cctxParams.fParams = params.fParams;
822 return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, ZSTD_dm_auto, NULL,
823 cctxParams, pledgedSrcSize);
826 size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
827 const ZSTD_CDict* cdict,
828 ZSTD_frameParameters fParams,
829 unsigned long long pledgedSrcSize)
831 ZSTD_CCtx_params cctxParams = mtctx->params;
832 cctxParams.cParams = ZSTD_getCParamsFromCDict(cdict);
833 cctxParams.fParams = fParams;
834 if (cdict==NULL) return ERROR(dictionary_wrong); /* method incompatible with NULL cdict */
835 return ZSTDMT_initCStream_internal(mtctx, NULL, 0 /*dictSize*/, ZSTD_dm_auto, cdict,
836 cctxParams, pledgedSrcSize);
840 /* ZSTDMT_resetCStream() :
841 * pledgedSrcSize is optional and can be zero == unknown */
842 size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* zcs, unsigned long long pledgedSrcSize)
844 if (zcs->params.nbThreads==1)
845 return ZSTD_resetCStream(zcs->cctxPool->cctx[0], pledgedSrcSize);
846 return ZSTDMT_initCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, 0, zcs->params,
850 size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) {
851 ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, 0);
852 ZSTD_CCtx_params cctxParams = zcs->params;
853 cctxParams.cParams = params.cParams;
854 cctxParams.fParams = params.fParams;
855 return ZSTDMT_initCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, NULL, cctxParams, 0);
859 static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame)
861 unsigned const jobID = zcs->nextJobID & zcs->jobIDMask;
863 DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ",
864 zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize);
865 zcs->jobs[jobID].src = zcs->inBuff.buffer;
866 zcs->jobs[jobID].srcStart = zcs->inBuff.buffer.start;
867 zcs->jobs[jobID].srcSize = srcSize;
868 zcs->jobs[jobID].dictSize = zcs->dictSize;
869 assert(zcs->inBuff.filled >= srcSize + zcs->dictSize);
870 zcs->jobs[jobID].params = zcs->params;
871 /* do not calculate checksum within sections, but write it in header for first section */
872 if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0;
873 zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL;
874 zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize;
875 zcs->jobs[jobID].dstBuff = g_nullBuffer;
876 zcs->jobs[jobID].cctxPool = zcs->cctxPool;
877 zcs->jobs[jobID].bufPool = zcs->bufPool;
878 zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0);
879 zcs->jobs[jobID].lastChunk = endFrame;
880 zcs->jobs[jobID].jobCompleted = 0;
881 zcs->jobs[jobID].dstFlushed = 0;
882 zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex;
883 zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond;
885 if (zcs->params.fParams.checksumFlag)
886 XXH64_update(&zcs->xxhState, (const char*)zcs->inBuff.buffer.start + zcs->dictSize, srcSize);
888 /* get a new buffer for next input */
890 size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize);
891 zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->bufPool);
892 if (zcs->inBuff.buffer.start == NULL) { /* not enough memory to allocate next input buffer */
893 zcs->jobs[jobID].jobCompleted = 1;
895 ZSTDMT_waitForAllJobsCompleted(zcs);
896 ZSTDMT_releaseAllJobResources(zcs);
897 return ERROR(memory_allocation);
899 zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize;
900 memmove(zcs->inBuff.buffer.start,
901 (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize,
903 zcs->dictSize = newDictSize;
904 } else { /* if (endFrame==1) */
905 zcs->inBuff.buffer = g_nullBuffer;
906 zcs->inBuff.filled = 0;
909 if (zcs->nextJobID == 0) {
910 /* single chunk exception : checksum is calculated directly within worker thread */
911 zcs->params.fParams.checksumFlag = 0;
914 DEBUGLOG(4, "posting job %u : %u bytes (end:%u) (note : doneJob = %u=>%u)",
916 (U32)zcs->jobs[jobID].srcSize,
917 zcs->jobs[jobID].lastChunk,
919 zcs->doneJobID & zcs->jobIDMask);
920 POOL_add(zcs->factory, ZSTDMT_compressChunk, &zcs->jobs[jobID]); /* this call is blocking when thread worker pool is exhausted */
926 /* ZSTDMT_flushNextJob() :
927 * output : will be updated with amount of data flushed .
928 * blockToFlush : if >0, the function will block and wait if there is no data available to flush .
929 * @return : amount of data remaining within internal buffer, 1 if unknown but > 0, 0 if no more, or an error code */
930 static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned blockToFlush)
932 unsigned const wJobID = zcs->doneJobID & zcs->jobIDMask;
933 if (zcs->doneJobID == zcs->nextJobID) return 0; /* all flushed ! */
934 ZSTD_PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
935 while (zcs->jobs[wJobID].jobCompleted==0) {
936 DEBUGLOG(5, "waiting for jobCompleted signal from job %u", zcs->doneJobID);
937 if (!blockToFlush) { ZSTD_pthread_mutex_unlock(&zcs->jobCompleted_mutex); return 0; } /* nothing ready to be flushed => skip */
938 ZSTD_pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex); /* block when nothing available to flush */
940 ZSTD_pthread_mutex_unlock(&zcs->jobCompleted_mutex);
941 /* compression job completed : output can be flushed */
942 { ZSTDMT_jobDescription job = zcs->jobs[wJobID];
943 if (!job.jobScanned) {
944 if (ZSTD_isError(job.cSize)) {
945 DEBUGLOG(5, "compression error detected ");
946 ZSTDMT_waitForAllJobsCompleted(zcs);
947 ZSTDMT_releaseAllJobResources(zcs);
950 DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag);
951 if (zcs->params.fParams.checksumFlag) {
952 if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) { /* write checksum at end of last section */
953 U32 const checksum = (U32)XXH64_digest(&zcs->xxhState);
954 DEBUGLOG(5, "writing checksum : %08X \n", checksum);
955 MEM_writeLE32((char*)job.dstBuff.start + job.cSize, checksum);
957 zcs->jobs[wJobID].cSize += 4;
959 zcs->jobs[wJobID].jobScanned = 1;
961 { size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos);
962 DEBUGLOG(5, "Flushing %u bytes from job %u ", (U32)toWrite, zcs->doneJobID);
963 memcpy((char*)output->dst + output->pos, (const char*)job.dstBuff.start + job.dstFlushed, toWrite);
964 output->pos += toWrite;
965 job.dstFlushed += toWrite;
967 if (job.dstFlushed == job.cSize) { /* output buffer fully flushed => move to next one */
968 ZSTDMT_releaseBuffer(zcs->bufPool, job.dstBuff);
969 zcs->jobs[wJobID].dstBuff = g_nullBuffer;
970 zcs->jobs[wJobID].jobCompleted = 0;
973 zcs->jobs[wJobID].dstFlushed = job.dstFlushed;
975 /* return value : how many bytes left in buffer ; fake it to 1 if unknown but >0 */
976 if (job.cSize > job.dstFlushed) return (job.cSize - job.dstFlushed);
977 if (zcs->doneJobID < zcs->nextJobID) return 1; /* still some buffer to flush */
978 zcs->allJobsCompleted = zcs->frameEnded; /* frame completed and entirely flushed */
979 return 0; /* everything flushed */
983 /** ZSTDMT_compressStream_generic() :
984 * internal use only - exposed to be invoked from zstd_compress.c
985 * assumption : output and input are valid (pos <= size)
986 * @return : minimum amount of data remaining to flush, 0 if none */
987 size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
988 ZSTD_outBuffer* output,
989 ZSTD_inBuffer* input,
990 ZSTD_EndDirective endOp)
992 size_t const newJobThreshold = mtctx->dictSize + mtctx->targetSectionSize;
993 unsigned forwardInputProgress = 0;
994 assert(output->pos <= output->size);
995 assert(input->pos <= input->size);
996 if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
997 /* current frame being ended. Only flush/end are allowed */
998 return ERROR(stage_wrong);
1000 if (mtctx->params.nbThreads==1) { /* delegate to single-thread (synchronous) */
1001 return ZSTD_compressStream_generic(mtctx->cctxPool->cctx[0], output, input, endOp);
1004 /* single-pass shortcut (note : synchronous-mode) */
1005 if ( (mtctx->nextJobID == 0) /* just started */
1006 && (mtctx->inBuff.filled == 0) /* nothing buffered */
1007 && (endOp == ZSTD_e_end) /* end order */
1008 && (output->size - output->pos >= ZSTD_compressBound(input->size - input->pos)) ) { /* enough room */
1009 size_t const cSize = ZSTDMT_compress_advanced_internal(mtctx,
1010 (char*)output->dst + output->pos, output->size - output->pos,
1011 (const char*)input->src + input->pos, input->size - input->pos,
1012 mtctx->cdict, mtctx->params);
1013 if (ZSTD_isError(cSize)) return cSize;
1014 input->pos = input->size;
1015 output->pos += cSize;
1016 ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer); /* was allocated in initStream */
1017 mtctx->allJobsCompleted = 1;
1018 mtctx->frameEnded = 1;
1022 /* fill input buffer */
1023 if (input->size > input->pos) { /* support NULL input */
1024 if (mtctx->inBuff.buffer.start == NULL) {
1025 mtctx->inBuff.buffer = ZSTDMT_getBuffer(mtctx->bufPool); /* note : may fail, in which case, no forward input progress */
1026 mtctx->inBuff.filled = 0;
1028 if (mtctx->inBuff.buffer.start) {
1029 size_t const toLoad = MIN(input->size - input->pos, mtctx->inBuffSize - mtctx->inBuff.filled);
1030 DEBUGLOG(5, "inBuff:%08X; inBuffSize=%u; ToCopy=%u", (U32)(size_t)mtctx->inBuff.buffer.start, (U32)mtctx->inBuffSize, (U32)toLoad);
1031 memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad);
1032 input->pos += toLoad;
1033 mtctx->inBuff.filled += toLoad;
1034 forwardInputProgress = toLoad>0;
1037 if ( (mtctx->inBuff.filled >= newJobThreshold) /* filled enough : let's compress */
1038 && (mtctx->nextJobID <= mtctx->doneJobID + mtctx->jobIDMask) ) { /* avoid overwriting job round buffer */
1039 CHECK_F( ZSTDMT_createCompressionJob(mtctx, mtctx->targetSectionSize, 0 /* endFrame */) );
1042 /* check for potential compressed data ready to be flushed */
1043 CHECK_F( ZSTDMT_flushNextJob(mtctx, output, !forwardInputProgress /* blockToFlush */) ); /* block if there was no forward input progress */
1045 if (input->pos < input->size) /* input not consumed : do not flush yet */
1046 endOp = ZSTD_e_continue;
1051 return ZSTDMT_flushStream(mtctx, output);
1053 return ZSTDMT_endStream(mtctx, output);
1054 case ZSTD_e_continue:
1057 return ERROR(GENERIC); /* invalid endDirective */
1062 size_t ZSTDMT_compressStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
1064 CHECK_F( ZSTDMT_compressStream_generic(zcs, output, input, ZSTD_e_continue) );
1066 /* recommended next input size : fill current input buffer */
1067 return zcs->inBuffSize - zcs->inBuff.filled; /* note : could be zero when input buffer is fully filled and no more availability to create new job */
1071 static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned endFrame)
1073 size_t const srcSize = zcs->inBuff.filled - zcs->dictSize;
1075 if ( ((srcSize > 0) || (endFrame && !zcs->frameEnded))
1076 && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {
1077 CHECK_F( ZSTDMT_createCompressionJob(zcs, srcSize, endFrame) );
1080 /* check if there is any data available to flush */
1081 return ZSTDMT_flushNextJob(zcs, output, 1 /* blockToFlush */);
1085 size_t ZSTDMT_flushStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
1087 DEBUGLOG(5, "ZSTDMT_flushStream");
1088 if (zcs->params.nbThreads==1)
1089 return ZSTD_flushStream(zcs->cctxPool->cctx[0], output);
1090 return ZSTDMT_flushStream_internal(zcs, output, 0 /* endFrame */);
1093 size_t ZSTDMT_endStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
1095 DEBUGLOG(4, "ZSTDMT_endStream");
1096 if (zcs->params.nbThreads==1)
1097 return ZSTD_endStream(zcs->cctxPool->cctx[0], output);
1098 return ZSTDMT_flushStream_internal(zcs, output, 1 /* endFrame */);