1 /**

     2  * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.

     3  * All rights reserved.

     4  *

     5  * This source code is licensed under the BSD-style license found in the

     6  * LICENSE file in the root directory of this source tree. An additional grant

     7  * of patent rights can be found in the PATENTS file in the same directory.

     8  */

     9 

    10 

    11 /* ======   Tuning parameters   ====== */

    12 #define ZSTDMT_NBTHREADS_MAX 128

    13 

    14 

    15 /* ======   Compiler specifics   ====== */

    16 #if defined(_MSC_VER)

    17 #  pragma warning(disable : 4204)        /* disable: C4204: non-constant aggregate initializer */

    18 #endif

    19 

    20 

    21 /* ======   Dependencies   ====== */

    22 #include <stdlib.h>   /* malloc */

    23 #include <string.h>   /* memcpy */

    24 #include "pool.h"     /* threadpool */

    25 #include "threading.h"  /* mutex */

    26 #include "zstd_internal.h"   /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */

    27 #include "zstdmt_compress.h"

    28 #define XXH_STATIC_LINKING_ONLY   /* XXH64_state_t */

    29 #include "xxhash.h"

    30 

    31 

    32 /* ======   Debug   ====== */

    33 #if 0

    34 

    35 #  include <stdio.h>

    36 #  include <unistd.h>

    37 #  include <sys/times.h>

    38    static unsigned g_debugLevel = 3;

    39 #  define DEBUGLOGRAW(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __VA_ARGS__); }

    40 #  define DEBUGLOG(l, ...) if (l<=g_debugLevel) { fprintf(stderr, __FILE__ ": "); fprintf(stderr, __VA_ARGS__); fprintf(stderr, " \n"); }

    41 

    42 #  define DEBUG_PRINTHEX(l,p,n) { \

    43     unsigned debug_u;                   \

    44     for (debug_u=0; debug_u<(n); debug_u++)           \

    45         DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \

    46     DEBUGLOGRAW(l, " \n");       \

    47 }

    48 

    49 static unsigned long long GetCurrentClockTimeMicroseconds()

    50 {

    51    static clock_t _ticksPerSecond = 0;

    52    if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);

    53 

    54    struct tms junk; clock_t newTicks = (clock_t) times(&junk);

    55    return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond);

    56 }

    57 

    58 #define MUTEX_WAIT_TIME_DLEVEL 5

    59 #define PTHREAD_MUTEX_LOCK(mutex) \

    60 if (g_debugLevel>=MUTEX_WAIT_TIME_DLEVEL) { \

    61    unsigned long long beforeTime = GetCurrentClockTimeMicroseconds(); \

    62    pthread_mutex_lock(mutex); \

    63    unsigned long long afterTime = GetCurrentClockTimeMicroseconds(); \

    64    unsigned long long elapsedTime = (afterTime-beforeTime); \

    65    if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \

    66       DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \

    67                elapsedTime, #mutex); \

    68   } \

    69 } else pthread_mutex_lock(mutex);

    70 

    71 #else

    72 

    73 #  define DEBUGLOG(l, ...)      {}    /* disabled */

    74 #  define PTHREAD_MUTEX_LOCK(m) pthread_mutex_lock(m)

    75 #  define DEBUG_PRINTHEX(l,p,n) {}

    76 

    77 #endif

    78 

    79 

    80 /* =====   Buffer Pool   ===== */

    81 

    82 typedef struct buffer_s {

    83     void* start;

    84     size_t size;

    85 } buffer_t;

    86 

    87 static const buffer_t g_nullBuffer = { NULL, 0 };

    88 

    89 typedef struct ZSTDMT_bufferPool_s {

    90     unsigned totalBuffers;

    91     unsigned nbBuffers;

    92     buffer_t bTable[1];   /* variable size */

    93 } ZSTDMT_bufferPool;

    94 

    95 static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads)

    96 {

    97     unsigned const maxNbBuffers = 2*nbThreads + 2;

    98     ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)calloc(1, sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t));

    99     if (bufPool==NULL) return NULL;

   100     bufPool->totalBuffers = maxNbBuffers;

   101     bufPool->nbBuffers = 0;

   102     return bufPool;

   103 }

   104 

   105 static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)

   106 {

   107     unsigned u;

   108     if (!bufPool) return;   /* compatibility with free on NULL */

   109     for (u=0; u<bufPool->totalBuffers; u++)

   110         free(bufPool->bTable[u].start);

   111     free(bufPool);

   112 }

   113 

   114 /* assumption : invocation from main thread only ! */

   115 static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* pool, size_t bSize)

   116 {

   117     if (pool->nbBuffers) {   /* try to use an existing buffer */

   118         buffer_t const buf = pool->bTable[--(pool->nbBuffers)];

   119         size_t const availBufferSize = buf.size;

   120         if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize))   /* large enough, but not too much */

   121             return buf;

   122         free(buf.start);   /* size conditions not respected : scratch this buffer and create a new one */

   123     }

   124     /* create new buffer */

   125     {   buffer_t buffer;

   126         void* const start = malloc(bSize);

   127         if (start==NULL) bSize = 0;

   128         buffer.start = start;   /* note : start can be NULL if malloc fails ! */

   129         buffer.size = bSize;

   130         return buffer;

   131     }

   132 }

   133 

   134 /* store buffer for later re-use, up to pool capacity */

   135 static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* pool, buffer_t buf)

   136 {

   137     if (buf.start == NULL) return;   /* release on NULL */

   138     if (pool->nbBuffers < pool->totalBuffers) {

   139         pool->bTable[pool->nbBuffers++] = buf;   /* store for later re-use */

   140         return;

   141     }

   142     /* Reached bufferPool capacity (should not happen) */

   143     free(buf.start);

   144 }

   145 

   146 

   147 /* =====   CCtx Pool   ===== */

   148 

   149 typedef struct {

   150     unsigned totalCCtx;

   151     unsigned availCCtx;

   152     ZSTD_CCtx* cctx[1];   /* variable size */

   153 } ZSTDMT_CCtxPool;

   154 

   155 /* assumption : CCtxPool invocation only from main thread */

   156 

   157 /* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */

   158 static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)

   159 {

   160     unsigned u;

   161     for (u=0; u<pool->totalCCtx; u++)

   162         ZSTD_freeCCtx(pool->cctx[u]);  /* note : compatible with free on NULL */

   163     free(pool);

   164 }

   165 

   166 /* ZSTDMT_createCCtxPool() :

   167  * implies nbThreads >= 1 , checked by caller ZSTDMT_createCCtx() */

   168 static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads)

   169 {

   170     ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) calloc(1, sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*));

   171     if (!cctxPool) return NULL;

   172     cctxPool->totalCCtx = nbThreads;

   173     cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */

   174     cctxPool->cctx[0] = ZSTD_createCCtx();

   175     if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }

   176     DEBUGLOG(1, "cctxPool created, with %u threads", nbThreads);

   177     return cctxPool;

   178 }

   179 

   180 static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* pool)

   181 {

   182     if (pool->availCCtx) {

   183         pool->availCCtx--;

   184         return pool->cctx[pool->availCCtx];

   185     }

   186     return ZSTD_createCCtx();   /* note : can be NULL, when creation fails ! */

   187 }

   188 

   189 static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)

   190 {

   191     if (cctx==NULL) return;   /* compatibility with release on NULL */

   192     if (pool->availCCtx < pool->totalCCtx)

   193         pool->cctx[pool->availCCtx++] = cctx;

   194     else

   195         /* pool overflow : should not happen, since totalCCtx==nbThreads */

   196         ZSTD_freeCCtx(cctx);

   197 }

   198 

   199 

   200 /* =====   Thread worker   ===== */

   201 

   202 typedef struct {

   203     buffer_t buffer;

   204     size_t filled;

   205 } inBuff_t;

   206 

   207 typedef struct {

   208     ZSTD_CCtx* cctx;

   209     buffer_t src;

   210     const void* srcStart;

   211     size_t   srcSize;

   212     size_t   dictSize;

   213     buffer_t dstBuff;

   214     size_t   cSize;

   215     size_t   dstFlushed;

   216     unsigned firstChunk;

   217     unsigned lastChunk;

   218     unsigned jobCompleted;

   219     unsigned jobScanned;

   220     pthread_mutex_t* jobCompleted_mutex;

   221     pthread_cond_t* jobCompleted_cond;

   222     ZSTD_parameters params;

   223     ZSTD_CDict* cdict;

   224     unsigned long long fullFrameSize;

   225 } ZSTDMT_jobDescription;

   226 

   227 /* ZSTDMT_compressChunk() : POOL_function type */

   228 void ZSTDMT_compressChunk(void* jobDescription)

   229 {

   230     ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;

   231     const void* const src = (const char*)job->srcStart + job->dictSize;

   232     buffer_t const dstBuff = job->dstBuff;

   233     DEBUGLOG(3, "job (first:%u) (last:%u) : dictSize %u, srcSize %u", job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize);

   234     if (job->cdict) {

   235         size_t const initError = ZSTD_compressBegin_usingCDict(job->cctx, job->cdict, job->fullFrameSize);

   236         if (job->cdict) DEBUGLOG(3, "using CDict ");

   237         if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }

   238     } else {

   239         size_t const initError = ZSTD_compressBegin_advanced(job->cctx, job->srcStart, job->dictSize, job->params, job->fullFrameSize);

   240         if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }

   241         ZSTD_setCCtxParameter(job->cctx, ZSTD_p_forceWindow, 1);

   242     }

   243     if (!job->firstChunk) {  /* flush frame header */

   244         size_t const hSize = ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, 0);

   245         if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; }

   246         ZSTD_invalidateRepCodes(job->cctx);

   247     }

   248 

   249     DEBUGLOG(4, "Compressing : ");

   250     DEBUG_PRINTHEX(4, job->srcStart, 12);

   251     job->cSize = (job->lastChunk) ?   /* last chunk signal */

   252                  ZSTD_compressEnd     (job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :

   253                  ZSTD_compressContinue(job->cctx, dstBuff.start, dstBuff.size, src, job->srcSize);

   254     DEBUGLOG(3, "compressed %u bytes into %u bytes   (first:%u) (last:%u)", (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk);

   255 

   256 _endJob:

   257     PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex);

   258     job->jobCompleted = 1;

   259     job->jobScanned = 0;

   260     pthread_cond_signal(job->jobCompleted_cond);

   261     pthread_mutex_unlock(job->jobCompleted_mutex);

   262 }

   263 

   264 

   265 /* ------------------------------------------ */

   266 /* =====   Multi-threaded compression   ===== */

   267 /* ------------------------------------------ */

   268 

   269 struct ZSTDMT_CCtx_s {

   270     POOL_ctx* factory;

   271     ZSTDMT_bufferPool* buffPool;

   272     ZSTDMT_CCtxPool* cctxPool;

   273     pthread_mutex_t jobCompleted_mutex;

   274     pthread_cond_t jobCompleted_cond;

   275     size_t targetSectionSize;

   276     size_t marginSize;

   277     size_t inBuffSize;

   278     size_t dictSize;

   279     size_t targetDictSize;

   280     inBuff_t inBuff;

   281     ZSTD_parameters params;

   282     XXH64_state_t xxhState;

   283     unsigned nbThreads;

   284     unsigned jobIDMask;

   285     unsigned doneJobID;

   286     unsigned nextJobID;

   287     unsigned frameEnded;

   288     unsigned allJobsCompleted;

   289     unsigned overlapRLog;

   290     unsigned long long frameContentSize;

   291     size_t sectionSize;

   292     ZSTD_CDict* cdict;

   293     ZSTD_CStream* cstream;

   294     ZSTDMT_jobDescription jobs[1];   /* variable size (must lies at the end) */

   295 };

   296 

   297 ZSTDMT_CCtx *ZSTDMT_createCCtx(unsigned nbThreads)

   298 {

   299     ZSTDMT_CCtx* cctx;

   300     U32 const minNbJobs = nbThreads + 2;

   301     U32 const nbJobsLog2 = ZSTD_highbit32(minNbJobs) + 1;

   302     U32 const nbJobs = 1 << nbJobsLog2;

   303     DEBUGLOG(5, "nbThreads : %u  ; minNbJobs : %u ;  nbJobsLog2 : %u ;  nbJobs : %u  \n",

   304             nbThreads, minNbJobs, nbJobsLog2, nbJobs);

   305     if ((nbThreads < 1) | (nbThreads > ZSTDMT_NBTHREADS_MAX)) return NULL;

   306     cctx = (ZSTDMT_CCtx*) calloc(1, sizeof(ZSTDMT_CCtx) + nbJobs*sizeof(ZSTDMT_jobDescription));

   307     if (!cctx) return NULL;

   308     cctx->nbThreads = nbThreads;

   309     cctx->jobIDMask = nbJobs - 1;

   310     cctx->allJobsCompleted = 1;

   311     cctx->sectionSize = 0;

   312     cctx->overlapRLog = 3;

   313     cctx->factory = POOL_create(nbThreads, 1);

   314     cctx->buffPool = ZSTDMT_createBufferPool(nbThreads);

   315     cctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads);

   316     if (!cctx->factory | !cctx->buffPool | !cctx->cctxPool) {  /* one object was not created */

   317         ZSTDMT_freeCCtx(cctx);

   318         return NULL;

   319     }

   320     if (nbThreads==1) {

   321         cctx->cstream = ZSTD_createCStream();

   322         if (!cctx->cstream) {

   323             ZSTDMT_freeCCtx(cctx); return NULL;

   324     }   }

   325     pthread_mutex_init(&cctx->jobCompleted_mutex, NULL);   /* Todo : check init function return */

   326     pthread_cond_init(&cctx->jobCompleted_cond, NULL);

   327     DEBUGLOG(4, "mt_cctx created, for %u threads \n", nbThreads);

   328     return cctx;

   329 }

   330 

   331 /* ZSTDMT_releaseAllJobResources() :

   332  * Ensure all workers are killed first. */

   333 static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)

   334 {

   335     unsigned jobID;

   336     for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {

   337         ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].dstBuff);

   338         mtctx->jobs[jobID].dstBuff = g_nullBuffer;

   339         ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[jobID].src);

   340         mtctx->jobs[jobID].src = g_nullBuffer;

   341         ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[jobID].cctx);

   342         mtctx->jobs[jobID].cctx = NULL;

   343     }

   344     memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));

   345     ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->inBuff.buffer);

   346     mtctx->inBuff.buffer = g_nullBuffer;

   347     mtctx->allJobsCompleted = 1;

   348 }

   349 

   350 size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)

   351 {

   352     if (mtctx==NULL) return 0;   /* compatible with free on NULL */

   353     POOL_free(mtctx->factory);

   354     if (!mtctx->allJobsCompleted) ZSTDMT_releaseAllJobResources(mtctx); /* stop workers first */

   355     ZSTDMT_freeBufferPool(mtctx->buffPool);  /* release job resources into pools first */

   356     ZSTDMT_freeCCtxPool(mtctx->cctxPool);

   357     ZSTD_freeCDict(mtctx->cdict);

   358     ZSTD_freeCStream(mtctx->cstream);

   359     pthread_mutex_destroy(&mtctx->jobCompleted_mutex);

   360     pthread_cond_destroy(&mtctx->jobCompleted_cond);

   361     free(mtctx);

   362     return 0;

   363 }

   364 

   365 size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSDTMT_parameter parameter, unsigned value)

   366 {

   367     switch(parameter)

   368     {

   369     case ZSTDMT_p_sectionSize :

   370         mtctx->sectionSize = value;

   371         return 0;

   372     case ZSTDMT_p_overlapSectionLog :

   373     DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);

   374         mtctx->overlapRLog = (value >= 9) ? 0 : 9 - value;

   375         return 0;

   376     default :

   377         return ERROR(compressionParameter_unsupported);

   378     }

   379 }

   380 

   381 

   382 /* ------------------------------------------ */

   383 /* =====   Multi-threaded compression   ===== */

   384 /* ------------------------------------------ */

   385 

   386 size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,

   387                            void* dst, size_t dstCapacity,

   388                      const void* src, size_t srcSize,

   389                            int compressionLevel)

   390 {

   391     ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);

   392     size_t const chunkTargetSize = (size_t)1 << (params.cParams.windowLog + 2);

   393     unsigned const nbChunksMax = (unsigned)(srcSize / chunkTargetSize) + (srcSize < chunkTargetSize) /* min 1 */;

   394     unsigned nbChunks = MIN(nbChunksMax, mtctx->nbThreads);

   395     size_t const proposedChunkSize = (srcSize + (nbChunks-1)) / nbChunks;

   396     size_t const avgChunkSize = ((proposedChunkSize & 0x1FFFF) < 0xFFFF) ? proposedChunkSize + 0xFFFF : proposedChunkSize;   /* avoid too small last block */

   397     size_t remainingSrcSize = srcSize;

   398     const char* const srcStart = (const char*)src;

   399     size_t frameStartPos = 0;

   400 

   401     DEBUGLOG(3, "windowLog : %2u => chunkTargetSize : %u bytes  ", params.cParams.windowLog, (U32)chunkTargetSize);

   402     DEBUGLOG(2, "nbChunks  : %2u   (chunkSize : %u bytes)   ", nbChunks, (U32)avgChunkSize);

   403     params.fParams.contentSizeFlag = 1;

   404 

   405     if (nbChunks==1) {   /* fallback to single-thread mode */

   406         ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];

   407         return ZSTD_compressCCtx(cctx, dst, dstCapacity, src, srcSize, compressionLevel);

   408     }

   409 

   410     {   unsigned u;

   411         for (u=0; u<nbChunks; u++) {

   412             size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize);

   413             size_t const dstBufferCapacity = u ? ZSTD_compressBound(chunkSize) : dstCapacity;

   414             buffer_t const dstAsBuffer = { dst, dstCapacity };

   415             buffer_t const dstBuffer = u ? ZSTDMT_getBuffer(mtctx->buffPool, dstBufferCapacity) : dstAsBuffer;

   416             ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(mtctx->cctxPool);

   417 

   418             if ((cctx==NULL) || (dstBuffer.start==NULL)) {

   419                 mtctx->jobs[u].cSize = ERROR(memory_allocation);   /* job result */

   420                 mtctx->jobs[u].jobCompleted = 1;

   421                 nbChunks = u+1;

   422                 break;   /* let's wait for previous jobs to complete, but don't start new ones */

   423             }

   424 

   425             mtctx->jobs[u].srcStart = srcStart + frameStartPos;

   426             mtctx->jobs[u].srcSize = chunkSize;

   427             mtctx->jobs[u].fullFrameSize = srcSize;

   428             mtctx->jobs[u].params = params;

   429             mtctx->jobs[u].dstBuff = dstBuffer;

   430             mtctx->jobs[u].cctx = cctx;

   431             mtctx->jobs[u].firstChunk = (u==0);

   432             mtctx->jobs[u].lastChunk = (u==nbChunks-1);

   433             mtctx->jobs[u].jobCompleted = 0;

   434             mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex;

   435             mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond;

   436 

   437             DEBUGLOG(3, "posting job %u   (%u bytes)", u, (U32)chunkSize);

   438             DEBUG_PRINTHEX(3, mtctx->jobs[u].srcStart, 12);

   439             POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]);

   440 

   441             frameStartPos += chunkSize;

   442             remainingSrcSize -= chunkSize;

   443     }   }

   444     /* note : since nbChunks <= nbThreads, all jobs should be running immediately in parallel */

   445 

   446     {   unsigned chunkID;

   447         size_t error = 0, dstPos = 0;

   448         for (chunkID=0; chunkID<nbChunks; chunkID++) {

   449             DEBUGLOG(3, "waiting for chunk %u ", chunkID);

   450             PTHREAD_MUTEX_LOCK(&mtctx->jobCompleted_mutex);

   451             while (mtctx->jobs[chunkID].jobCompleted==0) {

   452                 DEBUGLOG(4, "waiting for jobCompleted signal from chunk %u", chunkID);

   453                 pthread_cond_wait(&mtctx->jobCompleted_cond, &mtctx->jobCompleted_mutex);

   454             }

   455             pthread_mutex_unlock(&mtctx->jobCompleted_mutex);

   456             DEBUGLOG(3, "ready to write chunk %u ", chunkID);

   457 

   458             ZSTDMT_releaseCCtx(mtctx->cctxPool, mtctx->jobs[chunkID].cctx);

   459             mtctx->jobs[chunkID].cctx = NULL;

   460             mtctx->jobs[chunkID].srcStart = NULL;

   461             {   size_t const cSize = mtctx->jobs[chunkID].cSize;

   462                 if (ZSTD_isError(cSize)) error = cSize;

   463                 if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);

   464                 if (chunkID) {   /* note : chunk 0 is already written directly into dst */

   465                     if (!error) memcpy((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize);

   466                     ZSTDMT_releaseBuffer(mtctx->buffPool, mtctx->jobs[chunkID].dstBuff);

   467                     mtctx->jobs[chunkID].dstBuff = g_nullBuffer;

   468                 }

   469                 dstPos += cSize ;

   470             }

   471         }

   472         if (!error) DEBUGLOG(3, "compressed size : %u  ", (U32)dstPos);

   473         return error ? error : dstPos;

   474     }

   475 

   476 }

   477 

   478 

   479 /* ====================================== */

   480 /* =======      Streaming API     ======= */

   481 /* ====================================== */

   482 

   483 static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* zcs) {

   484     while (zcs->doneJobID < zcs->nextJobID) {

   485         unsigned const jobID = zcs->doneJobID & zcs->jobIDMask;

   486         PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);

   487         while (zcs->jobs[jobID].jobCompleted==0) {

   488             DEBUGLOG(4, "waiting for jobCompleted signal from chunk %u", zcs->doneJobID);   /* we want to block when waiting for data to flush */

   489             pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);

   490         }

   491         pthread_mutex_unlock(&zcs->jobCompleted_mutex);

   492         zcs->doneJobID++;

   493     }

   494 }

   495 

   496 

   497 static size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,

   498                                     const void* dict, size_t dictSize, unsigned updateDict,

   499                                     ZSTD_parameters params, unsigned long long pledgedSrcSize)

   500 {

   501     ZSTD_customMem const cmem = { NULL, NULL, NULL };

   502     DEBUGLOG(3, "Started new compression, with windowLog : %u", params.cParams.windowLog);

   503     if (zcs->nbThreads==1) return ZSTD_initCStream_advanced(zcs->cstream, dict, dictSize, params, pledgedSrcSize);

   504     if (zcs->allJobsCompleted == 0) {   /* previous job not correctly finished */

   505         ZSTDMT_waitForAllJobsCompleted(zcs);

   506         ZSTDMT_releaseAllJobResources(zcs);

   507         zcs->allJobsCompleted = 1;

   508     }

   509     zcs->params = params;

   510     if (updateDict) {

   511         ZSTD_freeCDict(zcs->cdict); zcs->cdict = NULL;

   512         if (dict && dictSize) {

   513             zcs->cdict = ZSTD_createCDict_advanced(dict, dictSize, 0, params, cmem);

   514             if (zcs->cdict == NULL) return ERROR(memory_allocation);

   515     }   }

   516     zcs->frameContentSize = pledgedSrcSize;

   517     zcs->targetDictSize = (zcs->overlapRLog>=9) ? 0 : (size_t)1 << (zcs->params.cParams.windowLog - zcs->overlapRLog);

   518     DEBUGLOG(4, "overlapRLog : %u ", zcs->overlapRLog);

   519     DEBUGLOG(3, "overlap Size : %u KB", (U32)(zcs->targetDictSize>>10));

   520     zcs->targetSectionSize = zcs->sectionSize ? zcs->sectionSize : (size_t)1 << (zcs->params.cParams.windowLog + 2);

   521     zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize);

   522     zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize);

   523     DEBUGLOG(3, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10));

   524     zcs->marginSize = zcs->targetSectionSize >> 2;

   525     zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize + zcs->marginSize;

   526     zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);

   527     if (zcs->inBuff.buffer.start == NULL) return ERROR(memory_allocation);

   528     zcs->inBuff.filled = 0;

   529     zcs->dictSize = 0;

   530     zcs->doneJobID = 0;

   531     zcs->nextJobID = 0;

   532     zcs->frameEnded = 0;

   533     zcs->allJobsCompleted = 0;

   534     if (params.fParams.checksumFlag) XXH64_reset(&zcs->xxhState, 0);

   535     return 0;

   536 }

   537 

   538 size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* zcs,

   539                                 const void* dict, size_t dictSize,

   540                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)

   541 {

   542     return ZSTDMT_initCStream_internal(zcs, dict, dictSize, 1, params, pledgedSrcSize);

   543 }

   544 

   545 /* ZSTDMT_resetCStream() :

   546  * pledgedSrcSize is optional and can be zero == unknown */

   547 size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* zcs, unsigned long long pledgedSrcSize)

   548 {

   549     if (zcs->nbThreads==1) return ZSTD_resetCStream(zcs->cstream, pledgedSrcSize);

   550     return ZSTDMT_initCStream_internal(zcs, NULL, 0, 0, zcs->params, pledgedSrcSize);

   551 }

   552 

   553 size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) {

   554     ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, 0);

   555     return ZSTDMT_initCStream_internal(zcs, NULL, 0, 1, params, 0);

   556 }

   557 

   558 

   559 static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame)

   560 {

   561     size_t const dstBufferCapacity = ZSTD_compressBound(srcSize);

   562     buffer_t const dstBuffer = ZSTDMT_getBuffer(zcs->buffPool, dstBufferCapacity);

   563     ZSTD_CCtx* const cctx = ZSTDMT_getCCtx(zcs->cctxPool);

   564     unsigned const jobID = zcs->nextJobID & zcs->jobIDMask;

   565 

   566     if ((cctx==NULL) || (dstBuffer.start==NULL)) {

   567         zcs->jobs[jobID].jobCompleted = 1;

   568         zcs->nextJobID++;

   569         ZSTDMT_waitForAllJobsCompleted(zcs);

   570         ZSTDMT_releaseAllJobResources(zcs);

   571         return ERROR(memory_allocation);

   572     }

   573 

   574     DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ", zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize);

   575     zcs->jobs[jobID].src = zcs->inBuff.buffer;

   576     zcs->jobs[jobID].srcStart = zcs->inBuff.buffer.start;

   577     zcs->jobs[jobID].srcSize = srcSize;

   578     zcs->jobs[jobID].dictSize = zcs->dictSize;   /* note : zcs->inBuff.filled is presumed >= srcSize + dictSize */

   579     zcs->jobs[jobID].params = zcs->params;

   580     if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0;  /* do not calculate checksum within sections, just keep it in header for first section */

   581     zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL;

   582     zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize;

   583     zcs->jobs[jobID].dstBuff = dstBuffer;

   584     zcs->jobs[jobID].cctx = cctx;

   585     zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0);

   586     zcs->jobs[jobID].lastChunk = endFrame;

   587     zcs->jobs[jobID].jobCompleted = 0;

   588     zcs->jobs[jobID].dstFlushed = 0;

   589     zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex;

   590     zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond;

   591 

   592     /* get a new buffer for next input */

   593     if (!endFrame) {

   594         size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize);

   595         zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->buffPool, zcs->inBuffSize);

   596         if (zcs->inBuff.buffer.start == NULL) {   /* not enough memory to allocate next input buffer */

   597             zcs->jobs[jobID].jobCompleted = 1;

   598             zcs->nextJobID++;

   599             ZSTDMT_waitForAllJobsCompleted(zcs);

   600             ZSTDMT_releaseAllJobResources(zcs);

   601             return ERROR(memory_allocation);

   602         }

   603         DEBUGLOG(5, "inBuff filled to %u", (U32)zcs->inBuff.filled);

   604         zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize;

   605         DEBUGLOG(5, "new job : filled to %u, with %u dict and %u src", (U32)zcs->inBuff.filled, (U32)newDictSize, (U32)(zcs->inBuff.filled - newDictSize));

   606         memmove(zcs->inBuff.buffer.start, (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize, zcs->inBuff.filled);

   607         DEBUGLOG(5, "new inBuff pre-filled");

   608         zcs->dictSize = newDictSize;

   609     } else {

   610         zcs->inBuff.buffer = g_nullBuffer;

   611         zcs->inBuff.filled = 0;

   612         zcs->dictSize = 0;

   613         zcs->frameEnded = 1;

   614         if (zcs->nextJobID == 0)

   615             zcs->params.fParams.checksumFlag = 0;   /* single chunk : checksum is calculated directly within worker thread */

   616     }

   617 

   618     DEBUGLOG(3, "posting job %u : %u bytes  (end:%u) (note : doneJob = %u=>%u)", zcs->nextJobID, (U32)zcs->jobs[jobID].srcSize, zcs->jobs[jobID].lastChunk, zcs->doneJobID, zcs->doneJobID & zcs->jobIDMask);

   619     POOL_add(zcs->factory, ZSTDMT_compressChunk, &zcs->jobs[jobID]);   /* this call is blocking when thread worker pool is exhausted */

   620     zcs->nextJobID++;

   621     return 0;

   622 }

   623 

   624 

   625 /* ZSTDMT_flushNextJob() :

   626  * output : will be updated with amount of data flushed .

   627  * blockToFlush : if >0, the function will block and wait if there is no data available to flush .

   628  * @return : amount of data remaining within internal buffer, 1 if unknown but > 0, 0 if no more, or an error code */

   629 static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned blockToFlush)

   630 {

   631     unsigned const wJobID = zcs->doneJobID & zcs->jobIDMask;

   632     if (zcs->doneJobID == zcs->nextJobID) return 0;   /* all flushed ! */

   633     PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);

   634     while (zcs->jobs[wJobID].jobCompleted==0) {

   635         DEBUGLOG(5, "waiting for jobCompleted signal from job %u", zcs->doneJobID);

   636         if (!blockToFlush) { pthread_mutex_unlock(&zcs->jobCompleted_mutex); return 0; }  /* nothing ready to be flushed => skip */

   637         pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);  /* block when nothing available to flush */

   638     }

   639     pthread_mutex_unlock(&zcs->jobCompleted_mutex);

   640     /* compression job completed : output can be flushed */

   641     {   ZSTDMT_jobDescription job = zcs->jobs[wJobID];

   642         if (!job.jobScanned) {

   643             if (ZSTD_isError(job.cSize)) {

   644                 DEBUGLOG(5, "compression error detected ");

   645                 ZSTDMT_waitForAllJobsCompleted(zcs);

   646                 ZSTDMT_releaseAllJobResources(zcs);

   647                 return job.cSize;

   648             }

   649             ZSTDMT_releaseCCtx(zcs->cctxPool, job.cctx);

   650             zcs->jobs[wJobID].cctx = NULL;

   651             DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag);

   652             if (zcs->params.fParams.checksumFlag) {

   653                 XXH64_update(&zcs->xxhState, (const char*)job.srcStart + job.dictSize, job.srcSize);

   654                 if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) {  /* write checksum at end of last section */

   655                     U32 const checksum = (U32)XXH64_digest(&zcs->xxhState);

   656                     DEBUGLOG(4, "writing checksum : %08X \n", checksum);

   657                     MEM_writeLE32((char*)job.dstBuff.start + job.cSize, checksum);

   658                     job.cSize += 4;

   659                     zcs->jobs[wJobID].cSize += 4;

   660             }   }

   661             ZSTDMT_releaseBuffer(zcs->buffPool, job.src);

   662             zcs->jobs[wJobID].srcStart = NULL;

   663             zcs->jobs[wJobID].src = g_nullBuffer;

   664             zcs->jobs[wJobID].jobScanned = 1;

   665         }

   666         {   size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos);

   667             DEBUGLOG(4, "Flushing %u bytes from job %u ", (U32)toWrite, zcs->doneJobID);

   668             memcpy((char*)output->dst + output->pos, (const char*)job.dstBuff.start + job.dstFlushed, toWrite);

   669             output->pos += toWrite;

   670             job.dstFlushed += toWrite;

   671         }

   672         if (job.dstFlushed == job.cSize) {   /* output buffer fully flushed => move to next one */

   673             ZSTDMT_releaseBuffer(zcs->buffPool, job.dstBuff);

   674             zcs->jobs[wJobID].dstBuff = g_nullBuffer;

   675             zcs->jobs[wJobID].jobCompleted = 0;

   676             zcs->doneJobID++;

   677         } else {

   678             zcs->jobs[wJobID].dstFlushed = job.dstFlushed;

   679         }

   680         /* return value : how many bytes left in buffer ; fake it to 1 if unknown but >0 */

   681         if (job.cSize > job.dstFlushed) return (job.cSize - job.dstFlushed);

   682         if (zcs->doneJobID < zcs->nextJobID) return 1;   /* still some buffer to flush */

   683         zcs->allJobsCompleted = zcs->frameEnded;   /* frame completed and entirely flushed */

   684         return 0;   /* everything flushed */

   685 }   }

   686 

   687 

   688 size_t ZSTDMT_compressStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)

   689 {

   690     size_t const newJobThreshold = zcs->dictSize + zcs->targetSectionSize + zcs->marginSize;

   691     if (zcs->frameEnded) return ERROR(stage_wrong);   /* current frame being ended. Only flush is allowed. Restart with init */

   692     if (zcs->nbThreads==1) return ZSTD_compressStream(zcs->cstream, output, input);

   693 

   694     /* fill input buffer */

   695     {   size_t const toLoad = MIN(input->size - input->pos, zcs->inBuffSize - zcs->inBuff.filled);

   696         memcpy((char*)zcs->inBuff.buffer.start + zcs->inBuff.filled, input->src, toLoad);

   697         input->pos += toLoad;

   698         zcs->inBuff.filled += toLoad;

   699     }

   700 

   701     if ( (zcs->inBuff.filled >= newJobThreshold)  /* filled enough : let's compress */

   702         && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {   /* avoid overwriting job round buffer */

   703         CHECK_F( ZSTDMT_createCompressionJob(zcs, zcs->targetSectionSize, 0) );

   704     }

   705 

   706     /* check for data to flush */

   707     CHECK_F( ZSTDMT_flushNextJob(zcs, output, (zcs->inBuff.filled == zcs->inBuffSize)) ); /* block if it wasn't possible to create new job due to saturation */

   708 

   709     /* recommended next input size : fill current input buffer */

   710     return zcs->inBuffSize - zcs->inBuff.filled;   /* note : could be zero when input buffer is fully filled and no more availability to create new job */

   711 }

   712 

   713 

   714 static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned endFrame)

   715 {

   716     size_t const srcSize = zcs->inBuff.filled - zcs->dictSize;

   717 

   718     if (srcSize) DEBUGLOG(4, "flushing : %u bytes left to compress", (U32)srcSize);

   719     if ( ((srcSize > 0) || (endFrame && !zcs->frameEnded))

   720        && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {

   721         CHECK_F( ZSTDMT_createCompressionJob(zcs, srcSize, endFrame) );

   722     }

   723 

   724     /* check if there is any data available to flush */

   725     DEBUGLOG(5, "zcs->doneJobID : %u  ; zcs->nextJobID : %u ", zcs->doneJobID, zcs->nextJobID);

   726     return ZSTDMT_flushNextJob(zcs, output, 1);

   727 }

   728 

   729 

   730 size_t ZSTDMT_flushStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)

   731 {

   732     if (zcs->nbThreads==1) return ZSTD_flushStream(zcs->cstream, output);

   733     return ZSTDMT_flushStream_internal(zcs, output, 0);

   734 }

   735 

   736 size_t ZSTDMT_endStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)

   737 {

   738     if (zcs->nbThreads==1) return ZSTD_endStream(zcs->cstream, output);

   739     return ZSTDMT_flushStream_internal(zcs, output, 1);

   740 }