xref: /freebsd/sys/contrib/zstd/programs/benchzstd.c (revision a90b9d0159070121c221b966469c3e36d912bf82)
1 /*
2  * Copyright (c) Yann Collet, Facebook, Inc.
3  * All rights reserved.
4  *
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.
9  */
10 
11 
12 /* **************************************
13 *  Tuning parameters
14 ****************************************/
15 #ifndef BMK_TIMETEST_DEFAULT_S   /* default minimum time per test */
16 #define BMK_TIMETEST_DEFAULT_S 3
17 #endif
18 
19 
20 /* *************************************
21 *  Includes
22 ***************************************/
23 #include "platform.h"    /* Large Files support */
24 #include "util.h"        /* UTIL_getFileSize, UTIL_sleep */
25 #include <stdlib.h>      /* malloc, free */
26 #include <string.h>      /* memset, strerror */
27 #include <stdio.h>       /* fprintf, fopen */
28 #include <errno.h>
29 #include <assert.h>      /* assert */
30 
31 #include "timefn.h"      /* UTIL_time_t */
32 #include "benchfn.h"
33 #include "../lib/common/mem.h"
34 #ifndef ZSTD_STATIC_LINKING_ONLY
35 #define ZSTD_STATIC_LINKING_ONLY
36 #endif
37 #include "../lib/zstd.h"
38 #include "datagen.h"     /* RDG_genBuffer */
39 #ifndef XXH_INLINE_ALL
40 #define XXH_INLINE_ALL
41 #endif
42 #include "../lib/common/xxhash.h"
43 #include "benchzstd.h"
44 #include "../lib/zstd_errors.h"
45 
46 
47 /* *************************************
48 *  Constants
49 ***************************************/
50 #ifndef ZSTD_GIT_COMMIT
51 #  define ZSTD_GIT_COMMIT_STRING ""
52 #else
53 #  define ZSTD_GIT_COMMIT_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_GIT_COMMIT)
54 #endif
55 
56 #define TIMELOOP_MICROSEC     (1*1000000ULL) /* 1 second */
57 #define TIMELOOP_NANOSEC      (1*1000000000ULL) /* 1 second */
58 #define ACTIVEPERIOD_MICROSEC (70*TIMELOOP_MICROSEC) /* 70 seconds */
59 #define COOLPERIOD_SEC        10
60 
61 #define KB *(1 <<10)
62 #define MB *(1 <<20)
63 #define GB *(1U<<30)
64 
65 #define BMK_RUNTEST_DEFAULT_MS 1000
66 
67 static const size_t maxMemory = (sizeof(size_t)==4)  ?
68                     /* 32-bit */ (2 GB - 64 MB) :
69                     /* 64-bit */ (size_t)(1ULL << ((sizeof(size_t)*8)-31));
70 
71 
72 /* *************************************
73 *  console display
74 ***************************************/
75 #define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush(NULL); }
76 #define DISPLAYLEVEL(l, ...) if (displayLevel>=l) { DISPLAY(__VA_ARGS__); }
77 /* 0 : no display;   1: errors;   2 : + result + interaction + warnings;   3 : + progression;   4 : + information */
78 #define OUTPUT(...)          { fprintf(stdout, __VA_ARGS__); fflush(NULL); }
79 #define OUTPUTLEVEL(l, ...)  if (displayLevel>=l) { OUTPUT(__VA_ARGS__); }
80 
81 
82 /* *************************************
83 *  Exceptions
84 ***************************************/
85 #ifndef DEBUG
86 #  define DEBUG 0
87 #endif
88 #define DEBUGOUTPUT(...) { if (DEBUG) DISPLAY(__VA_ARGS__); }
89 
90 #define RETURN_ERROR_INT(errorNum, ...)  {               \
91     DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__);    \
92     DISPLAYLEVEL(1, "Error %i : ", errorNum);         \
93     DISPLAYLEVEL(1, __VA_ARGS__);                     \
94     DISPLAYLEVEL(1, " \n");                           \
95     return errorNum;                                  \
96 }
97 
98 #define CHECK_Z(zf) {              \
99     size_t const zerr = zf;        \
100     if (ZSTD_isError(zerr)) {      \
101         DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__);  \
102         DISPLAY("Error : ");       \
103         DISPLAY("%s failed : %s",  \
104                 #zf, ZSTD_getErrorName(zerr));   \
105         DISPLAY(" \n");            \
106         exit(1);                   \
107     }                              \
108 }
109 
110 #define RETURN_ERROR(errorNum, retType, ...)  {       \
111     retType r;                                        \
112     memset(&r, 0, sizeof(retType));                   \
113     DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__);    \
114     DISPLAYLEVEL(1, "Error %i : ", errorNum);         \
115     DISPLAYLEVEL(1, __VA_ARGS__);                     \
116     DISPLAYLEVEL(1, " \n");                           \
117     r.tag = errorNum;                                 \
118     return r;                                         \
119 }
120 
121 
122 /* *************************************
123 *  Benchmark Parameters
124 ***************************************/
125 
126 BMK_advancedParams_t BMK_initAdvancedParams(void) {
127     BMK_advancedParams_t const res = {
128         BMK_both, /* mode */
129         BMK_TIMETEST_DEFAULT_S, /* nbSeconds */
130         0, /* blockSize */
131         0, /* nbWorkers */
132         0, /* realTime */
133         0, /* additionalParam */
134         0, /* ldmFlag */
135         0, /* ldmMinMatch */
136         0, /* ldmHashLog */
137         0, /* ldmBuckSizeLog */
138         0,  /* ldmHashRateLog */
139         ZSTD_ps_auto, /* literalCompressionMode */
140         0 /* useRowMatchFinder */
141     };
142     return res;
143 }
144 
145 
146 /* ********************************************************
147 *  Bench functions
148 **********************************************************/
149 typedef struct {
150     const void* srcPtr;
151     size_t srcSize;
152     void*  cPtr;
153     size_t cRoom;
154     size_t cSize;
155     void*  resPtr;
156     size_t resSize;
157 } blockParam_t;
158 
159 #undef MIN
160 #undef MAX
161 #define MIN(a,b)    ((a) < (b) ? (a) : (b))
162 #define MAX(a,b)    ((a) > (b) ? (a) : (b))
163 
164 static void
165 BMK_initCCtx(ZSTD_CCtx* ctx,
166             const void* dictBuffer, size_t dictBufferSize,
167             int cLevel,
168             const ZSTD_compressionParameters* comprParams,
169             const BMK_advancedParams_t* adv)
170 {
171     ZSTD_CCtx_reset(ctx, ZSTD_reset_session_and_parameters);
172     if (adv->nbWorkers==1) {
173         CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_nbWorkers, 0));
174     } else {
175         CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_nbWorkers, adv->nbWorkers));
176     }
177     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_compressionLevel, cLevel));
178     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_useRowMatchFinder, adv->useRowMatchFinder));
179     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_enableLongDistanceMatching, adv->ldmFlag));
180     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_ldmMinMatch, adv->ldmMinMatch));
181     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_ldmHashLog, adv->ldmHashLog));
182     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_ldmBucketSizeLog, adv->ldmBucketSizeLog));
183     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_ldmHashRateLog, adv->ldmHashRateLog));
184     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_windowLog, (int)comprParams->windowLog));
185     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_hashLog, (int)comprParams->hashLog));
186     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_chainLog, (int)comprParams->chainLog));
187     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_searchLog, (int)comprParams->searchLog));
188     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_minMatch, (int)comprParams->minMatch));
189     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_targetLength, (int)comprParams->targetLength));
190     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_literalCompressionMode, (int)adv->literalCompressionMode));
191     CHECK_Z(ZSTD_CCtx_setParameter(ctx, ZSTD_c_strategy, (int)comprParams->strategy));
192     CHECK_Z(ZSTD_CCtx_loadDictionary(ctx, dictBuffer, dictBufferSize));
193 }
194 
195 static void BMK_initDCtx(ZSTD_DCtx* dctx,
196     const void* dictBuffer, size_t dictBufferSize) {
197     CHECK_Z(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_and_parameters));
198     CHECK_Z(ZSTD_DCtx_loadDictionary(dctx, dictBuffer, dictBufferSize));
199 }
200 
201 
202 typedef struct {
203     ZSTD_CCtx* cctx;
204     const void* dictBuffer;
205     size_t dictBufferSize;
206     int cLevel;
207     const ZSTD_compressionParameters* comprParams;
208     const BMK_advancedParams_t* adv;
209 } BMK_initCCtxArgs;
210 
211 static size_t local_initCCtx(void* payload) {
212     BMK_initCCtxArgs* ag = (BMK_initCCtxArgs*)payload;
213     BMK_initCCtx(ag->cctx, ag->dictBuffer, ag->dictBufferSize, ag->cLevel, ag->comprParams, ag->adv);
214     return 0;
215 }
216 
217 typedef struct {
218     ZSTD_DCtx* dctx;
219     const void* dictBuffer;
220     size_t dictBufferSize;
221 } BMK_initDCtxArgs;
222 
223 static size_t local_initDCtx(void* payload) {
224     BMK_initDCtxArgs* ag = (BMK_initDCtxArgs*)payload;
225     BMK_initDCtx(ag->dctx, ag->dictBuffer, ag->dictBufferSize);
226     return 0;
227 }
228 
229 
230 /* `addArgs` is the context */
231 static size_t local_defaultCompress(
232                     const void* srcBuffer, size_t srcSize,
233                     void* dstBuffer, size_t dstSize,
234                     void* addArgs)
235 {
236     ZSTD_CCtx* const cctx = (ZSTD_CCtx*)addArgs;
237     return ZSTD_compress2(cctx, dstBuffer, dstSize, srcBuffer, srcSize);
238 }
239 
240 /* `addArgs` is the context */
241 static size_t local_defaultDecompress(
242                     const void* srcBuffer, size_t srcSize,
243                     void* dstBuffer, size_t dstCapacity,
244                     void* addArgs)
245 {
246     size_t moreToFlush = 1;
247     ZSTD_DCtx* const dctx = (ZSTD_DCtx*)addArgs;
248     ZSTD_inBuffer in;
249     ZSTD_outBuffer out;
250     in.src = srcBuffer; in.size = srcSize; in.pos = 0;
251     out.dst = dstBuffer; out.size = dstCapacity; out.pos = 0;
252     while (moreToFlush) {
253         if(out.pos == out.size) {
254             return (size_t)-ZSTD_error_dstSize_tooSmall;
255         }
256         moreToFlush = ZSTD_decompressStream(dctx, &out, &in);
257         if (ZSTD_isError(moreToFlush)) {
258             return moreToFlush;
259         }
260     }
261     return out.pos;
262 
263 }
264 
265 
266 /* ================================================================= */
267 /*      Benchmark Zstandard, mem-to-mem scenarios                    */
268 /* ================================================================= */
269 
270 int BMK_isSuccessful_benchOutcome(BMK_benchOutcome_t outcome)
271 {
272     return outcome.tag == 0;
273 }
274 
275 BMK_benchResult_t BMK_extract_benchResult(BMK_benchOutcome_t outcome)
276 {
277     assert(outcome.tag == 0);
278     return outcome.internal_never_use_directly;
279 }
280 
281 static BMK_benchOutcome_t BMK_benchOutcome_error(void)
282 {
283     BMK_benchOutcome_t b;
284     memset(&b, 0, sizeof(b));
285     b.tag = 1;
286     return b;
287 }
288 
289 static BMK_benchOutcome_t BMK_benchOutcome_setValidResult(BMK_benchResult_t result)
290 {
291     BMK_benchOutcome_t b;
292     b.tag = 0;
293     b.internal_never_use_directly = result;
294     return b;
295 }
296 
297 
298 /* benchMem with no allocation */
299 static BMK_benchOutcome_t
300 BMK_benchMemAdvancedNoAlloc(
301                     const void** srcPtrs, size_t* srcSizes,
302                     void** cPtrs, size_t* cCapacities, size_t* cSizes,
303                     void** resPtrs, size_t* resSizes,
304                     void** resultBufferPtr, void* compressedBuffer,
305                     size_t maxCompressedSize,
306                     BMK_timedFnState_t* timeStateCompress,
307                     BMK_timedFnState_t* timeStateDecompress,
308 
309                     const void* srcBuffer, size_t srcSize,
310                     const size_t* fileSizes, unsigned nbFiles,
311                     const int cLevel,
312                     const ZSTD_compressionParameters* comprParams,
313                     const void* dictBuffer, size_t dictBufferSize,
314                     ZSTD_CCtx* cctx, ZSTD_DCtx* dctx,
315                     int displayLevel, const char* displayName,
316                     const BMK_advancedParams_t* adv)
317 {
318     size_t const blockSize = ((adv->blockSize>=32 && (adv->mode != BMK_decodeOnly)) ? adv->blockSize : srcSize) + (!srcSize);  /* avoid div by 0 */
319     BMK_benchResult_t benchResult;
320     size_t const loadedCompressedSize = srcSize;
321     size_t cSize = 0;
322     double ratio = 0.;
323     U32 nbBlocks;
324 
325     assert(cctx != NULL); assert(dctx != NULL);
326 
327     /* init */
328     memset(&benchResult, 0, sizeof(benchResult));
329     if (strlen(displayName)>17) displayName += strlen(displayName) - 17;   /* display last 17 characters */
330     if (adv->mode == BMK_decodeOnly) {  /* benchmark only decompression : source must be already compressed */
331         const char* srcPtr = (const char*)srcBuffer;
332         U64 totalDSize64 = 0;
333         U32 fileNb;
334         for (fileNb=0; fileNb<nbFiles; fileNb++) {
335             U64 const fSize64 = ZSTD_findDecompressedSize(srcPtr, fileSizes[fileNb]);
336             if (fSize64==0) RETURN_ERROR(32, BMK_benchOutcome_t, "Impossible to determine original size ");
337             totalDSize64 += fSize64;
338             srcPtr += fileSizes[fileNb];
339         }
340         {   size_t const decodedSize = (size_t)totalDSize64;
341             assert((U64)decodedSize == totalDSize64);   /* check overflow */
342             free(*resultBufferPtr);
343             *resultBufferPtr = malloc(decodedSize);
344             if (!(*resultBufferPtr)) {
345                 RETURN_ERROR(33, BMK_benchOutcome_t, "not enough memory");
346             }
347             if (totalDSize64 > decodedSize) {  /* size_t overflow */
348                 free(*resultBufferPtr);
349                 RETURN_ERROR(32, BMK_benchOutcome_t, "original size is too large");
350             }
351             cSize = srcSize;
352             srcSize = decodedSize;
353             ratio = (double)srcSize / (double)cSize;
354         }
355     }
356 
357     /* Init data blocks  */
358     {   const char* srcPtr = (const char*)srcBuffer;
359         char* cPtr = (char*)compressedBuffer;
360         char* resPtr = (char*)(*resultBufferPtr);
361         U32 fileNb;
362         for (nbBlocks=0, fileNb=0; fileNb<nbFiles; fileNb++) {
363             size_t remaining = fileSizes[fileNb];
364             U32 const nbBlocksforThisFile = (adv->mode == BMK_decodeOnly) ? 1 : (U32)((remaining + (blockSize-1)) / blockSize);
365             U32 const blockEnd = nbBlocks + nbBlocksforThisFile;
366             for ( ; nbBlocks<blockEnd; nbBlocks++) {
367                 size_t const thisBlockSize = MIN(remaining, blockSize);
368                 srcPtrs[nbBlocks] = srcPtr;
369                 srcSizes[nbBlocks] = thisBlockSize;
370                 cPtrs[nbBlocks] = cPtr;
371                 cCapacities[nbBlocks] = (adv->mode == BMK_decodeOnly) ? thisBlockSize : ZSTD_compressBound(thisBlockSize);
372                 resPtrs[nbBlocks] = resPtr;
373                 resSizes[nbBlocks] = (adv->mode == BMK_decodeOnly) ? (size_t) ZSTD_findDecompressedSize(srcPtr, thisBlockSize) : thisBlockSize;
374                 srcPtr += thisBlockSize;
375                 cPtr += cCapacities[nbBlocks];
376                 resPtr += thisBlockSize;
377                 remaining -= thisBlockSize;
378                 if (adv->mode == BMK_decodeOnly) {
379                     cSizes[nbBlocks] = thisBlockSize;
380                     benchResult.cSize = thisBlockSize;
381     }   }   }   }
382 
383     /* warming up `compressedBuffer` */
384     if (adv->mode == BMK_decodeOnly) {
385         memcpy(compressedBuffer, srcBuffer, loadedCompressedSize);
386     } else {
387         RDG_genBuffer(compressedBuffer, maxCompressedSize, 0.10, 0.50, 1);
388     }
389 
390     /* Bench */
391     {   U64 const crcOrig = (adv->mode == BMK_decodeOnly) ? 0 : XXH64(srcBuffer, srcSize, 0);
392 #       define NB_MARKS 4
393         const char* marks[NB_MARKS] = { " |", " /", " =", " \\" };
394         U32 markNb = 0;
395         int compressionCompleted = (adv->mode == BMK_decodeOnly);
396         int decompressionCompleted = (adv->mode == BMK_compressOnly);
397         BMK_benchParams_t cbp, dbp;
398         BMK_initCCtxArgs cctxprep;
399         BMK_initDCtxArgs dctxprep;
400 
401         cbp.benchFn = local_defaultCompress;   /* ZSTD_compress2 */
402         cbp.benchPayload = cctx;
403         cbp.initFn = local_initCCtx;   /* BMK_initCCtx */
404         cbp.initPayload = &cctxprep;
405         cbp.errorFn = ZSTD_isError;
406         cbp.blockCount = nbBlocks;
407         cbp.srcBuffers = srcPtrs;
408         cbp.srcSizes = srcSizes;
409         cbp.dstBuffers = cPtrs;
410         cbp.dstCapacities = cCapacities;
411         cbp.blockResults = cSizes;
412 
413         cctxprep.cctx = cctx;
414         cctxprep.dictBuffer = dictBuffer;
415         cctxprep.dictBufferSize = dictBufferSize;
416         cctxprep.cLevel = cLevel;
417         cctxprep.comprParams = comprParams;
418         cctxprep.adv = adv;
419 
420         dbp.benchFn = local_defaultDecompress;
421         dbp.benchPayload = dctx;
422         dbp.initFn = local_initDCtx;
423         dbp.initPayload = &dctxprep;
424         dbp.errorFn = ZSTD_isError;
425         dbp.blockCount = nbBlocks;
426         dbp.srcBuffers = (const void* const *) cPtrs;
427         dbp.srcSizes = cSizes;
428         dbp.dstBuffers = resPtrs;
429         dbp.dstCapacities = resSizes;
430         dbp.blockResults = NULL;
431 
432         dctxprep.dctx = dctx;
433         dctxprep.dictBuffer = dictBuffer;
434         dctxprep.dictBufferSize = dictBufferSize;
435 
436         OUTPUTLEVEL(2, "\r%70s\r", "");   /* blank line */
437         assert(srcSize < UINT_MAX);
438         OUTPUTLEVEL(2, "%2s-%-17.17s :%10u -> \r", marks[markNb], displayName, (unsigned)srcSize);
439 
440         while (!(compressionCompleted && decompressionCompleted)) {
441             if (!compressionCompleted) {
442                 BMK_runOutcome_t const cOutcome = BMK_benchTimedFn( timeStateCompress, cbp);
443 
444                 if (!BMK_isSuccessful_runOutcome(cOutcome)) {
445                     return BMK_benchOutcome_error();
446                 }
447 
448                 {   BMK_runTime_t const cResult = BMK_extract_runTime(cOutcome);
449                     cSize = cResult.sumOfReturn;
450                     ratio = (double)srcSize / (double)cSize;
451                     {   BMK_benchResult_t newResult;
452                         newResult.cSpeed = (U64)((double)srcSize * TIMELOOP_NANOSEC / cResult.nanoSecPerRun);
453                         benchResult.cSize = cSize;
454                         if (newResult.cSpeed > benchResult.cSpeed)
455                             benchResult.cSpeed = newResult.cSpeed;
456                 }   }
457 
458                 {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
459                     assert(cSize < UINT_MAX);
460                     OUTPUTLEVEL(2, "%2s-%-17.17s :%10u ->%10u (x%5.*f), %6.*f MB/s \r",
461                             marks[markNb], displayName,
462                             (unsigned)srcSize, (unsigned)cSize,
463                             ratioAccuracy, ratio,
464                             benchResult.cSpeed < (10 * MB_UNIT) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT);
465                 }
466                 compressionCompleted = BMK_isCompleted_TimedFn(timeStateCompress);
467             }
468 
469             if(!decompressionCompleted) {
470                 BMK_runOutcome_t const dOutcome = BMK_benchTimedFn(timeStateDecompress, dbp);
471 
472                 if(!BMK_isSuccessful_runOutcome(dOutcome)) {
473                     return BMK_benchOutcome_error();
474                 }
475 
476                 {   BMK_runTime_t const dResult = BMK_extract_runTime(dOutcome);
477                     U64 const newDSpeed = (U64)((double)srcSize * TIMELOOP_NANOSEC / dResult.nanoSecPerRun);
478                     if (newDSpeed > benchResult.dSpeed)
479                         benchResult.dSpeed = newDSpeed;
480                 }
481 
482                 {   int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
483                     OUTPUTLEVEL(2, "%2s-%-17.17s :%10u ->%10u (x%5.*f), %6.*f MB/s, %6.1f MB/s\r",
484                             marks[markNb], displayName,
485                             (unsigned)srcSize, (unsigned)cSize,
486                             ratioAccuracy, ratio,
487                             benchResult.cSpeed < (10 * MB_UNIT) ? 2 : 1, (double)benchResult.cSpeed / MB_UNIT,
488                             (double)benchResult.dSpeed / MB_UNIT);
489                 }
490                 decompressionCompleted = BMK_isCompleted_TimedFn(timeStateDecompress);
491             }
492             markNb = (markNb+1) % NB_MARKS;
493         }   /* while (!(compressionCompleted && decompressionCompleted)) */
494 
495         /* CRC Checking */
496         {   const BYTE* resultBuffer = (const BYTE*)(*resultBufferPtr);
497             U64 const crcCheck = XXH64(resultBuffer, srcSize, 0);
498             if ((adv->mode == BMK_both) && (crcOrig!=crcCheck)) {
499                 size_t u;
500                 DISPLAY("!!! WARNING !!! %14s : Invalid Checksum : %x != %x   \n",
501                         displayName, (unsigned)crcOrig, (unsigned)crcCheck);
502                 for (u=0; u<srcSize; u++) {
503                     if (((const BYTE*)srcBuffer)[u] != resultBuffer[u]) {
504                         unsigned segNb, bNb, pos;
505                         size_t bacc = 0;
506                         DISPLAY("Decoding error at pos %u ", (unsigned)u);
507                         for (segNb = 0; segNb < nbBlocks; segNb++) {
508                             if (bacc + srcSizes[segNb] > u) break;
509                             bacc += srcSizes[segNb];
510                         }
511                         pos = (U32)(u - bacc);
512                         bNb = pos / (128 KB);
513                         DISPLAY("(sample %u, block %u, pos %u) \n", segNb, bNb, pos);
514                         {   size_t const lowest = (u>5) ? 5 : u;
515                             size_t n;
516                             DISPLAY("origin: ");
517                             for (n=lowest; n>0; n--)
518                                 DISPLAY("%02X ", ((const BYTE*)srcBuffer)[u-n]);
519                             DISPLAY(" :%02X:  ", ((const BYTE*)srcBuffer)[u]);
520                             for (n=1; n<3; n++)
521                                 DISPLAY("%02X ", ((const BYTE*)srcBuffer)[u+n]);
522                             DISPLAY(" \n");
523                             DISPLAY("decode: ");
524                             for (n=lowest; n>0; n--)
525                                 DISPLAY("%02X ", resultBuffer[u-n]);
526                             DISPLAY(" :%02X:  ", resultBuffer[u]);
527                             for (n=1; n<3; n++)
528                                 DISPLAY("%02X ", resultBuffer[u+n]);
529                             DISPLAY(" \n");
530                         }
531                         break;
532                     }
533                     if (u==srcSize-1) {  /* should never happen */
534                         DISPLAY("no difference detected\n");
535                     }
536                 }   /* for (u=0; u<srcSize; u++) */
537             }   /* if ((adv->mode == BMK_both) && (crcOrig!=crcCheck)) */
538         }   /* CRC Checking */
539 
540         if (displayLevel == 1) {   /* hidden display mode -q, used by python speed benchmark */
541             double const cSpeed = (double)benchResult.cSpeed / MB_UNIT;
542             double const dSpeed = (double)benchResult.dSpeed / MB_UNIT;
543             if (adv->additionalParam) {
544                 OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, adv->additionalParam);
545             } else {
546                 OUTPUT("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s  %s\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName);
547             }
548         }
549 
550         OUTPUTLEVEL(2, "%2i#\n", cLevel);
551     }   /* Bench */
552 
553     benchResult.cMem = (1ULL << (comprParams->windowLog)) + ZSTD_sizeof_CCtx(cctx);
554     return BMK_benchOutcome_setValidResult(benchResult);
555 }
556 
557 BMK_benchOutcome_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
558                         void* dstBuffer, size_t dstCapacity,
559                         const size_t* fileSizes, unsigned nbFiles,
560                         int cLevel, const ZSTD_compressionParameters* comprParams,
561                         const void* dictBuffer, size_t dictBufferSize,
562                         int displayLevel, const char* displayName, const BMK_advancedParams_t* adv)
563 
564 {
565     int const dstParamsError = !dstBuffer ^ !dstCapacity;  /* must be both NULL or none */
566 
567     size_t const blockSize = ((adv->blockSize>=32 && (adv->mode != BMK_decodeOnly)) ? adv->blockSize : srcSize) + (!srcSize) /* avoid div by 0 */ ;
568     U32 const maxNbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize) + nbFiles;
569 
570     /* these are the blockTable parameters, just split up */
571     const void ** const srcPtrs = (const void**)malloc(maxNbBlocks * sizeof(void*));
572     size_t* const srcSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t));
573 
574 
575     void ** const cPtrs = (void**)malloc(maxNbBlocks * sizeof(void*));
576     size_t* const cSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t));
577     size_t* const cCapacities = (size_t*)malloc(maxNbBlocks * sizeof(size_t));
578 
579     void ** const resPtrs = (void**)malloc(maxNbBlocks * sizeof(void*));
580     size_t* const resSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t));
581 
582     BMK_timedFnState_t* timeStateCompress = BMK_createTimedFnState(adv->nbSeconds * 1000, BMK_RUNTEST_DEFAULT_MS);
583     BMK_timedFnState_t* timeStateDecompress = BMK_createTimedFnState(adv->nbSeconds * 1000, BMK_RUNTEST_DEFAULT_MS);
584 
585     ZSTD_CCtx* const cctx = ZSTD_createCCtx();
586     ZSTD_DCtx* const dctx = ZSTD_createDCtx();
587 
588     const size_t maxCompressedSize = dstCapacity ? dstCapacity : ZSTD_compressBound(srcSize) + (maxNbBlocks * 1024);
589 
590     void* const internalDstBuffer = dstBuffer ? NULL : malloc(maxCompressedSize);
591     void* const compressedBuffer = dstBuffer ? dstBuffer : internalDstBuffer;
592 
593     BMK_benchOutcome_t outcome = BMK_benchOutcome_error();  /* error by default */
594 
595     void* resultBuffer = srcSize ? malloc(srcSize) : NULL;
596 
597     int allocationincomplete = !srcPtrs || !srcSizes || !cPtrs ||
598         !cSizes || !cCapacities || !resPtrs || !resSizes ||
599         !timeStateCompress || !timeStateDecompress ||
600         !cctx || !dctx ||
601         !compressedBuffer || !resultBuffer;
602 
603 
604     if (!allocationincomplete && !dstParamsError) {
605         outcome = BMK_benchMemAdvancedNoAlloc(srcPtrs, srcSizes,
606                                             cPtrs, cCapacities, cSizes,
607                                             resPtrs, resSizes,
608                                             &resultBuffer,
609                                             compressedBuffer, maxCompressedSize,
610                                             timeStateCompress, timeStateDecompress,
611                                             srcBuffer, srcSize,
612                                             fileSizes, nbFiles,
613                                             cLevel, comprParams,
614                                             dictBuffer, dictBufferSize,
615                                             cctx, dctx,
616                                             displayLevel, displayName, adv);
617     }
618 
619     /* clean up */
620     BMK_freeTimedFnState(timeStateCompress);
621     BMK_freeTimedFnState(timeStateDecompress);
622 
623     ZSTD_freeCCtx(cctx);
624     ZSTD_freeDCtx(dctx);
625 
626     free(internalDstBuffer);
627     free(resultBuffer);
628 
629     free((void*)srcPtrs);
630     free(srcSizes);
631     free(cPtrs);
632     free(cSizes);
633     free(cCapacities);
634     free(resPtrs);
635     free(resSizes);
636 
637     if(allocationincomplete) {
638         RETURN_ERROR(31, BMK_benchOutcome_t, "allocation error : not enough memory");
639     }
640 
641     if(dstParamsError) {
642         RETURN_ERROR(32, BMK_benchOutcome_t, "Dst parameters not coherent");
643     }
644     return outcome;
645 }
646 
647 BMK_benchOutcome_t BMK_benchMem(const void* srcBuffer, size_t srcSize,
648                         const size_t* fileSizes, unsigned nbFiles,
649                         int cLevel, const ZSTD_compressionParameters* comprParams,
650                         const void* dictBuffer, size_t dictBufferSize,
651                         int displayLevel, const char* displayName) {
652 
653     BMK_advancedParams_t const adv = BMK_initAdvancedParams();
654     return BMK_benchMemAdvanced(srcBuffer, srcSize,
655                                 NULL, 0,
656                                 fileSizes, nbFiles,
657                                 cLevel, comprParams,
658                                 dictBuffer, dictBufferSize,
659                                 displayLevel, displayName, &adv);
660 }
661 
662 static BMK_benchOutcome_t BMK_benchCLevel(const void* srcBuffer, size_t benchedSize,
663                             const size_t* fileSizes, unsigned nbFiles,
664                             int cLevel, const ZSTD_compressionParameters* comprParams,
665                             const void* dictBuffer, size_t dictBufferSize,
666                             int displayLevel, const char* displayName,
667                             BMK_advancedParams_t const * const adv)
668 {
669     const char* pch = strrchr(displayName, '\\'); /* Windows */
670     if (!pch) pch = strrchr(displayName, '/');    /* Linux */
671     if (pch) displayName = pch+1;
672 
673     if (adv->realTime) {
674         DISPLAYLEVEL(2, "Note : switching to real-time priority \n");
675         SET_REALTIME_PRIORITY;
676     }
677 
678     if (displayLevel == 1 && !adv->additionalParam)   /* --quiet mode */
679         OUTPUT("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n",
680                 ZSTD_VERSION_STRING, ZSTD_GIT_COMMIT_STRING,
681                 (unsigned)benchedSize, adv->nbSeconds, (unsigned)(adv->blockSize>>10));
682 
683     return BMK_benchMemAdvanced(srcBuffer, benchedSize,
684                                 NULL, 0,
685                                 fileSizes, nbFiles,
686                                 cLevel, comprParams,
687                                 dictBuffer, dictBufferSize,
688                                 displayLevel, displayName, adv);
689 }
690 
691 BMK_benchOutcome_t BMK_syntheticTest(int cLevel, double compressibility,
692                           const ZSTD_compressionParameters* compressionParams,
693                           int displayLevel, const BMK_advancedParams_t* adv)
694 {
695     char name[20] = {0};
696     size_t const benchedSize = 10000000;
697     void* srcBuffer;
698     BMK_benchOutcome_t res;
699 
700     if (cLevel > ZSTD_maxCLevel()) {
701         RETURN_ERROR(15, BMK_benchOutcome_t, "Invalid Compression Level");
702     }
703 
704     /* Memory allocation */
705     srcBuffer = malloc(benchedSize);
706     if (!srcBuffer) RETURN_ERROR(21, BMK_benchOutcome_t, "not enough memory");
707 
708     /* Fill input buffer */
709     RDG_genBuffer(srcBuffer, benchedSize, compressibility, 0.0, 0);
710 
711     /* Bench */
712     snprintf (name, sizeof(name), "Synthetic %2u%%", (unsigned)(compressibility*100));
713     res = BMK_benchCLevel(srcBuffer, benchedSize,
714                     &benchedSize /* ? */, 1 /* ? */,
715                     cLevel, compressionParams,
716                     NULL, 0,  /* dictionary */
717                     displayLevel, name, adv);
718 
719     /* clean up */
720     free(srcBuffer);
721 
722     return res;
723 }
724 
725 
726 
727 static size_t BMK_findMaxMem(U64 requiredMem)
728 {
729     size_t const step = 64 MB;
730     BYTE* testmem = NULL;
731 
732     requiredMem = (((requiredMem >> 26) + 1) << 26);
733     requiredMem += step;
734     if (requiredMem > maxMemory) requiredMem = maxMemory;
735 
736     do {
737         testmem = (BYTE*)malloc((size_t)requiredMem);
738         requiredMem -= step;
739     } while (!testmem && requiredMem > 0);
740 
741     free(testmem);
742     return (size_t)(requiredMem);
743 }
744 
745 /*! BMK_loadFiles() :
746  *  Loads `buffer` with content of files listed within `fileNamesTable`.
747  *  At most, fills `buffer` entirely. */
748 static int BMK_loadFiles(void* buffer, size_t bufferSize,
749                          size_t* fileSizes,
750                          const char* const * fileNamesTable, unsigned nbFiles,
751                          int displayLevel)
752 {
753     size_t pos = 0, totalSize = 0;
754     unsigned n;
755     for (n=0; n<nbFiles; n++) {
756         U64 fileSize = UTIL_getFileSize(fileNamesTable[n]);  /* last file may be shortened */
757         if (UTIL_isDirectory(fileNamesTable[n])) {
758             DISPLAYLEVEL(2, "Ignoring %s directory...       \n", fileNamesTable[n]);
759             fileSizes[n] = 0;
760             continue;
761         }
762         if (fileSize == UTIL_FILESIZE_UNKNOWN) {
763             DISPLAYLEVEL(2, "Cannot evaluate size of %s, ignoring ... \n", fileNamesTable[n]);
764             fileSizes[n] = 0;
765             continue;
766         }
767         {   FILE* const f = fopen(fileNamesTable[n], "rb");
768             if (f==NULL) RETURN_ERROR_INT(10, "impossible to open file %s", fileNamesTable[n]);
769             OUTPUTLEVEL(2, "Loading %s...       \r", fileNamesTable[n]);
770             if (fileSize > bufferSize-pos) fileSize = bufferSize-pos, nbFiles=n;   /* buffer too small - stop after this file */
771             {   size_t const readSize = fread(((char*)buffer)+pos, 1, (size_t)fileSize, f);
772                 if (readSize != (size_t)fileSize) RETURN_ERROR_INT(11, "could not read %s", fileNamesTable[n]);
773                 pos += readSize;
774             }
775             fileSizes[n] = (size_t)fileSize;
776             totalSize += (size_t)fileSize;
777             fclose(f);
778     }   }
779 
780     if (totalSize == 0) RETURN_ERROR_INT(12, "no data to bench");
781     return 0;
782 }
783 
784 BMK_benchOutcome_t BMK_benchFilesAdvanced(
785                         const char* const * fileNamesTable, unsigned nbFiles,
786                         const char* dictFileName, int cLevel,
787                         const ZSTD_compressionParameters* compressionParams,
788                         int displayLevel, const BMK_advancedParams_t* adv)
789 {
790     void* srcBuffer = NULL;
791     size_t benchedSize;
792     void* dictBuffer = NULL;
793     size_t dictBufferSize = 0;
794     size_t* fileSizes = NULL;
795     BMK_benchOutcome_t res;
796     U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles);
797 
798     if (!nbFiles) {
799         RETURN_ERROR(14, BMK_benchOutcome_t, "No Files to Benchmark");
800     }
801 
802     if (cLevel > ZSTD_maxCLevel()) {
803         RETURN_ERROR(15, BMK_benchOutcome_t, "Invalid Compression Level");
804     }
805 
806     if (totalSizeToLoad == UTIL_FILESIZE_UNKNOWN) {
807         RETURN_ERROR(9, BMK_benchOutcome_t, "Error loading files");
808     }
809 
810     fileSizes = (size_t*)calloc(nbFiles, sizeof(size_t));
811     if (!fileSizes) RETURN_ERROR(12, BMK_benchOutcome_t, "not enough memory for fileSizes");
812 
813     /* Load dictionary */
814     if (dictFileName != NULL) {
815         U64 const dictFileSize = UTIL_getFileSize(dictFileName);
816         if (dictFileSize == UTIL_FILESIZE_UNKNOWN) {
817             DISPLAYLEVEL(1, "error loading %s : %s \n", dictFileName, strerror(errno));
818             free(fileSizes);
819             RETURN_ERROR(9, BMK_benchOutcome_t, "benchmark aborted");
820         }
821         if (dictFileSize > 64 MB) {
822             free(fileSizes);
823             RETURN_ERROR(10, BMK_benchOutcome_t, "dictionary file %s too large", dictFileName);
824         }
825         dictBufferSize = (size_t)dictFileSize;
826         dictBuffer = malloc(dictBufferSize);
827         if (dictBuffer==NULL) {
828             free(fileSizes);
829             RETURN_ERROR(11, BMK_benchOutcome_t, "not enough memory for dictionary (%u bytes)",
830                             (unsigned)dictBufferSize);
831         }
832 
833         {   int const errorCode = BMK_loadFiles(dictBuffer, dictBufferSize,
834                                                 fileSizes, &dictFileName /*?*/,
835                                                 1 /*?*/, displayLevel);
836             if (errorCode) {
837                 res = BMK_benchOutcome_error();
838                 goto _cleanUp;
839         }   }
840     }
841 
842     /* Memory allocation & restrictions */
843     benchedSize = BMK_findMaxMem(totalSizeToLoad * 3) / 3;
844     if ((U64)benchedSize > totalSizeToLoad) benchedSize = (size_t)totalSizeToLoad;
845     if (benchedSize < totalSizeToLoad)
846         DISPLAY("Not enough memory; testing %u MB only...\n", (unsigned)(benchedSize >> 20));
847 
848     srcBuffer = benchedSize ? malloc(benchedSize) : NULL;
849     if (!srcBuffer) {
850         free(dictBuffer);
851         free(fileSizes);
852         RETURN_ERROR(12, BMK_benchOutcome_t, "not enough memory");
853     }
854 
855     /* Load input buffer */
856     {   int const errorCode = BMK_loadFiles(srcBuffer, benchedSize,
857                                         fileSizes, fileNamesTable, nbFiles,
858                                         displayLevel);
859         if (errorCode) {
860             res = BMK_benchOutcome_error();
861             goto _cleanUp;
862     }   }
863 
864     /* Bench */
865     {   char mfName[20] = {0};
866         snprintf (mfName, sizeof(mfName), " %u files", nbFiles);
867         {   const char* const displayName = (nbFiles > 1) ? mfName : fileNamesTable[0];
868             res = BMK_benchCLevel(srcBuffer, benchedSize,
869                                 fileSizes, nbFiles,
870                                 cLevel, compressionParams,
871                                 dictBuffer, dictBufferSize,
872                                 displayLevel, displayName,
873                                 adv);
874     }   }
875 
876 _cleanUp:
877     free(srcBuffer);
878     free(dictBuffer);
879     free(fileSizes);
880     return res;
881 }
882 
883 
884 BMK_benchOutcome_t BMK_benchFiles(
885                     const char* const * fileNamesTable, unsigned nbFiles,
886                     const char* dictFileName,
887                     int cLevel, const ZSTD_compressionParameters* compressionParams,
888                     int displayLevel)
889 {
890     BMK_advancedParams_t const adv = BMK_initAdvancedParams();
891     return BMK_benchFilesAdvanced(fileNamesTable, nbFiles, dictFileName, cLevel, compressionParams, displayLevel, &adv);
892 }
893