1 /*
2 * Copyright (c) 2015-present, 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 * Includes
14 **************************************/
15 #include "util.h" /* Compiler options, UTIL_GetFileSize */
16 #include <stdlib.h> /* malloc */
17 #include <stdio.h> /* fprintf, fopen, ftello64 */
18 #include <assert.h>
19
20 #include "timefn.h" /* UTIL_clockSpanNano, UTIL_getTime */
21 #include "mem.h" /* U32 */
22 #ifndef ZSTD_DLL_IMPORT
23 #include "zstd_internal.h" /* ZSTD_decodeSeqHeaders, ZSTD_blockHeaderSize, blockType_e, KB, MB */
24 #else
25 #define KB *(1 <<10)
26 #define MB *(1 <<20)
27 #define GB *(1U<<30)
28 typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
29 #endif
30 #define ZSTD_STATIC_LINKING_ONLY /* ZSTD_compressBegin, ZSTD_compressContinue, etc. */
31 #include "zstd.h" /* ZSTD_versionString */
32 #include "util.h" /* time functions */
33 #include "datagen.h"
34 #include "benchfn.h" /* CustomBench */
35 #include "benchzstd.h" /* MB_UNIT */
36
37
38 /*_************************************
39 * Constants
40 **************************************/
41 #define PROGRAM_DESCRIPTION "Zstandard speed analyzer"
42 #define AUTHOR "Yann Collet"
43 #define WELCOME_MESSAGE "*** %s %s %i-bits, by %s (%s) ***\n", PROGRAM_DESCRIPTION, ZSTD_versionString(), (int)(sizeof(void*)*8), AUTHOR, __DATE__
44
45 #define NBLOOPS 6
46 #define TIMELOOP_S 2
47
48 #define MAX_MEM (1984 MB)
49
50 #define DEFAULT_CLEVEL 1
51
52 #define COMPRESSIBILITY_DEFAULT 0.50
53 static const size_t kSampleSizeDefault = 10000000;
54
55 #define TIMELOOP_NANOSEC (1*1000000000ULL) /* 1 second */
56
57
58 /*_************************************
59 * Macros
60 **************************************/
61 #define DISPLAY(...) fprintf(stderr, __VA_ARGS__)
62
63 #define CONTROL(c) { if (!(c)) { abort(); } } /* like assert(), but cannot be disabled */
64
65 /*_************************************
66 * Benchmark Parameters
67 **************************************/
68 static unsigned g_nbIterations = NBLOOPS;
69
70
71 /*_*******************************************************
72 * Private functions
73 *********************************************************/
BMK_findMaxMem(U64 requiredMem)74 static size_t BMK_findMaxMem(U64 requiredMem)
75 {
76 size_t const step = 64 MB;
77 void* testmem = NULL;
78
79 requiredMem = (((requiredMem >> 26) + 1) << 26);
80 if (requiredMem > MAX_MEM) requiredMem = MAX_MEM;
81
82 requiredMem += step;
83 do {
84 testmem = malloc ((size_t)requiredMem);
85 requiredMem -= step;
86 } while (!testmem);
87
88 free (testmem);
89 return (size_t) requiredMem;
90 }
91
92
93 /*_*******************************************************
94 * Benchmark wrappers
95 *********************************************************/
96
97 static ZSTD_CCtx* g_zcc = NULL;
98
99 static size_t
local_ZSTD_compress(const void * src,size_t srcSize,void * dst,size_t dstSize,void * payload)100 local_ZSTD_compress(const void* src, size_t srcSize,
101 void* dst, size_t dstSize,
102 void* payload)
103 {
104 ZSTD_parameters p;
105 ZSTD_frameParameters f = { 1 /* contentSizeHeader*/, 0, 0 };
106 p.fParams = f;
107 p.cParams = *(ZSTD_compressionParameters*)payload;
108 return ZSTD_compress_advanced (g_zcc, dst, dstSize, src, srcSize, NULL ,0, p);
109 //return ZSTD_compress(dst, dstSize, src, srcSize, cLevel);
110 }
111
112 static size_t g_cSize = 0;
local_ZSTD_decompress(const void * src,size_t srcSize,void * dst,size_t dstSize,void * buff2)113 static size_t local_ZSTD_decompress(const void* src, size_t srcSize,
114 void* dst, size_t dstSize,
115 void* buff2)
116 {
117 (void)src; (void)srcSize;
118 return ZSTD_decompress(dst, dstSize, buff2, g_cSize);
119 }
120
121 static ZSTD_DCtx* g_zdc = NULL;
122
123 #ifndef ZSTD_DLL_IMPORT
124 extern size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* ctx, const void* src, size_t srcSize);
local_ZSTD_decodeLiteralsBlock(const void * src,size_t srcSize,void * dst,size_t dstSize,void * buff2)125 static size_t local_ZSTD_decodeLiteralsBlock(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2)
126 {
127 (void)src; (void)srcSize; (void)dst; (void)dstSize;
128 return ZSTD_decodeLiteralsBlock(g_zdc, buff2, g_cSize);
129 }
130
local_ZSTD_decodeSeqHeaders(const void * src,size_t srcSize,void * dst,size_t dstSize,void * buff2)131 static size_t local_ZSTD_decodeSeqHeaders(const void* src, size_t srcSize, void* dst, size_t dstSize, void* buff2)
132 {
133 int nbSeq;
134 (void)src; (void)srcSize; (void)dst; (void)dstSize;
135 return ZSTD_decodeSeqHeaders(g_zdc, &nbSeq, buff2, g_cSize);
136 }
137 #endif
138
139 static ZSTD_CStream* g_cstream= NULL;
140 static size_t
local_ZSTD_compressStream(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)141 local_ZSTD_compressStream(const void* src, size_t srcSize,
142 void* dst, size_t dstCapacity,
143 void* payload)
144 {
145 ZSTD_outBuffer buffOut;
146 ZSTD_inBuffer buffIn;
147 ZSTD_parameters p;
148 ZSTD_frameParameters f = {1 /* contentSizeHeader*/, 0, 0};
149 p.fParams = f;
150 p.cParams = *(ZSTD_compressionParameters*)payload;
151 ZSTD_initCStream_advanced(g_cstream, NULL, 0, p, ZSTD_CONTENTSIZE_UNKNOWN);
152 buffOut.dst = dst;
153 buffOut.size = dstCapacity;
154 buffOut.pos = 0;
155 buffIn.src = src;
156 buffIn.size = srcSize;
157 buffIn.pos = 0;
158 ZSTD_compressStream(g_cstream, &buffOut, &buffIn);
159 ZSTD_endStream(g_cstream, &buffOut);
160 return buffOut.pos;
161 }
162
163 static size_t
local_ZSTD_compressStream_freshCCtx(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)164 local_ZSTD_compressStream_freshCCtx(const void* src, size_t srcSize,
165 void* dst, size_t dstCapacity,
166 void* payload)
167 {
168 ZSTD_CCtx* const cctx = ZSTD_createCCtx();
169 size_t r;
170 assert(cctx != NULL);
171
172 r = local_ZSTD_compressStream(src, srcSize, dst, dstCapacity, payload);
173
174 ZSTD_freeCCtx(cctx);
175
176 return r;
177 }
178
179 static size_t
local_ZSTD_compress_generic_end(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)180 local_ZSTD_compress_generic_end(const void* src, size_t srcSize,
181 void* dst, size_t dstCapacity,
182 void* payload)
183 {
184 (void)payload;
185 return ZSTD_compress2(g_cstream, dst, dstCapacity, src, srcSize);
186 }
187
188 static size_t
local_ZSTD_compress_generic_continue(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)189 local_ZSTD_compress_generic_continue(const void* src, size_t srcSize,
190 void* dst, size_t dstCapacity,
191 void* payload)
192 {
193 ZSTD_outBuffer buffOut;
194 ZSTD_inBuffer buffIn;
195 (void)payload;
196 buffOut.dst = dst;
197 buffOut.size = dstCapacity;
198 buffOut.pos = 0;
199 buffIn.src = src;
200 buffIn.size = srcSize;
201 buffIn.pos = 0;
202 ZSTD_compressStream2(g_cstream, &buffOut, &buffIn, ZSTD_e_continue);
203 ZSTD_compressStream2(g_cstream, &buffOut, &buffIn, ZSTD_e_end);
204 return buffOut.pos;
205 }
206
207 static size_t
local_ZSTD_compress_generic_T2_end(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)208 local_ZSTD_compress_generic_T2_end(const void* src, size_t srcSize,
209 void* dst, size_t dstCapacity,
210 void* payload)
211 {
212 (void)payload;
213 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_nbWorkers, 2);
214 return ZSTD_compress2(g_cstream, dst, dstCapacity, src, srcSize);
215 }
216
217 static size_t
local_ZSTD_compress_generic_T2_continue(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)218 local_ZSTD_compress_generic_T2_continue(const void* src, size_t srcSize,
219 void* dst, size_t dstCapacity,
220 void* payload)
221 {
222 ZSTD_outBuffer buffOut;
223 ZSTD_inBuffer buffIn;
224 (void)payload;
225 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_nbWorkers, 2);
226 buffOut.dst = dst;
227 buffOut.size = dstCapacity;
228 buffOut.pos = 0;
229 buffIn.src = src;
230 buffIn.size = srcSize;
231 buffIn.pos = 0;
232 ZSTD_compressStream2(g_cstream, &buffOut, &buffIn, ZSTD_e_continue);
233 while(ZSTD_compressStream2(g_cstream, &buffOut, &buffIn, ZSTD_e_end)) {}
234 return buffOut.pos;
235 }
236
237 static ZSTD_DStream* g_dstream= NULL;
238 static size_t
local_ZSTD_decompressStream(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * buff2)239 local_ZSTD_decompressStream(const void* src, size_t srcSize,
240 void* dst, size_t dstCapacity,
241 void* buff2)
242 {
243 ZSTD_outBuffer buffOut;
244 ZSTD_inBuffer buffIn;
245 (void)src; (void)srcSize;
246 ZSTD_initDStream(g_dstream);
247 buffOut.dst = dst;
248 buffOut.size = dstCapacity;
249 buffOut.pos = 0;
250 buffIn.src = buff2;
251 buffIn.size = g_cSize;
252 buffIn.pos = 0;
253 ZSTD_decompressStream(g_dstream, &buffOut, &buffIn);
254 return buffOut.pos;
255 }
256
257 #ifndef ZSTD_DLL_IMPORT
local_ZSTD_compressContinue(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)258 static size_t local_ZSTD_compressContinue(const void* src, size_t srcSize,
259 void* dst, size_t dstCapacity,
260 void* payload)
261 {
262 ZSTD_parameters p;
263 ZSTD_frameParameters f = { 1 /* contentSizeHeader*/, 0, 0 };
264 p.fParams = f;
265 p.cParams = *(ZSTD_compressionParameters*)payload;
266 ZSTD_compressBegin_advanced(g_zcc, NULL, 0, p, srcSize);
267 return ZSTD_compressEnd(g_zcc, dst, dstCapacity, src, srcSize);
268 }
269
270 #define FIRST_BLOCK_SIZE 8
271 static size_t
local_ZSTD_compressContinue_extDict(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * payload)272 local_ZSTD_compressContinue_extDict(const void* src, size_t srcSize,
273 void* dst, size_t dstCapacity,
274 void* payload)
275 {
276 BYTE firstBlockBuf[FIRST_BLOCK_SIZE];
277
278 ZSTD_parameters p;
279 ZSTD_frameParameters const f = { 1, 0, 0 };
280 p.fParams = f;
281 p.cParams = *(ZSTD_compressionParameters*)payload;
282 ZSTD_compressBegin_advanced(g_zcc, NULL, 0, p, srcSize);
283 memcpy(firstBlockBuf, src, FIRST_BLOCK_SIZE);
284
285 { size_t const compressResult = ZSTD_compressContinue(g_zcc,
286 dst, dstCapacity,
287 firstBlockBuf, FIRST_BLOCK_SIZE);
288 if (ZSTD_isError(compressResult)) {
289 DISPLAY("local_ZSTD_compressContinue_extDict error : %s\n",
290 ZSTD_getErrorName(compressResult));
291 return compressResult;
292 }
293 dst = (BYTE*)dst + compressResult;
294 dstCapacity -= compressResult;
295 }
296 return ZSTD_compressEnd(g_zcc, dst, dstCapacity,
297 (const BYTE*)src + FIRST_BLOCK_SIZE,
298 srcSize - FIRST_BLOCK_SIZE);
299 }
300
local_ZSTD_decompressContinue(const void * src,size_t srcSize,void * dst,size_t dstCapacity,void * buff2)301 static size_t local_ZSTD_decompressContinue(const void* src, size_t srcSize,
302 void* dst, size_t dstCapacity,
303 void* buff2)
304 {
305 size_t regeneratedSize = 0;
306 const BYTE* ip = (const BYTE*)buff2;
307 const BYTE* const iend = ip + g_cSize;
308 BYTE* op = (BYTE*)dst;
309 size_t remainingCapacity = dstCapacity;
310
311 (void)src; (void)srcSize; /* unused */
312 ZSTD_decompressBegin(g_zdc);
313 while (ip < iend) {
314 size_t const iSize = ZSTD_nextSrcSizeToDecompress(g_zdc);
315 size_t const decodedSize = ZSTD_decompressContinue(g_zdc, op, remainingCapacity, ip, iSize);
316 ip += iSize;
317 regeneratedSize += decodedSize;
318 op += decodedSize;
319 remainingCapacity -= decodedSize;
320 }
321
322 return regeneratedSize;
323 }
324 #endif
325
326
327 /*_*******************************************************
328 * Bench functions
329 *********************************************************/
benchMem(unsigned benchNb,const void * src,size_t srcSize,int cLevel,ZSTD_compressionParameters cparams)330 static int benchMem(unsigned benchNb,
331 const void* src, size_t srcSize,
332 int cLevel, ZSTD_compressionParameters cparams)
333 {
334 size_t dstBuffSize = ZSTD_compressBound(srcSize);
335 BYTE* dstBuff;
336 void* dstBuff2;
337 void* payload;
338 const char* benchName;
339 BMK_benchFn_t benchFunction;
340 int errorcode = 0;
341
342 /* Selection */
343 switch(benchNb)
344 {
345 case 1:
346 benchFunction = local_ZSTD_compress; benchName = "compress";
347 break;
348 case 2:
349 benchFunction = local_ZSTD_decompress; benchName = "decompress";
350 break;
351 #ifndef ZSTD_DLL_IMPORT
352 case 11:
353 benchFunction = local_ZSTD_compressContinue; benchName = "compressContinue";
354 break;
355 case 12:
356 benchFunction = local_ZSTD_compressContinue_extDict; benchName = "compressContinue_extDict";
357 break;
358 case 13:
359 benchFunction = local_ZSTD_decompressContinue; benchName = "decompressContinue";
360 break;
361 case 31:
362 benchFunction = local_ZSTD_decodeLiteralsBlock; benchName = "decodeLiteralsBlock";
363 break;
364 case 32:
365 benchFunction = local_ZSTD_decodeSeqHeaders; benchName = "decodeSeqHeaders";
366 break;
367 #endif
368 case 41:
369 benchFunction = local_ZSTD_compressStream; benchName = "compressStream";
370 break;
371 case 42:
372 benchFunction = local_ZSTD_decompressStream; benchName = "decompressStream";
373 break;
374 case 43:
375 benchFunction = local_ZSTD_compressStream_freshCCtx; benchName = "compressStream_freshCCtx";
376 break;
377 case 51:
378 benchFunction = local_ZSTD_compress_generic_continue; benchName = "compress_generic, continue";
379 break;
380 case 52:
381 benchFunction = local_ZSTD_compress_generic_end; benchName = "compress_generic, end";
382 break;
383 case 61:
384 benchFunction = local_ZSTD_compress_generic_T2_continue; benchName = "compress_generic, -T2, continue";
385 break;
386 case 62:
387 benchFunction = local_ZSTD_compress_generic_T2_end; benchName = "compress_generic, -T2, end";
388 break;
389 default :
390 return 0;
391 }
392
393 /* Allocation */
394 dstBuff = (BYTE*)malloc(dstBuffSize);
395 dstBuff2 = malloc(dstBuffSize);
396 if ((!dstBuff) || (!dstBuff2)) {
397 DISPLAY("\nError: not enough memory!\n");
398 free(dstBuff); free(dstBuff2);
399 return 12;
400 }
401 payload = dstBuff2;
402 if (g_zcc==NULL) g_zcc = ZSTD_createCCtx();
403 if (g_zdc==NULL) g_zdc = ZSTD_createDCtx();
404 if (g_cstream==NULL) g_cstream = ZSTD_createCStream();
405 if (g_dstream==NULL) g_dstream = ZSTD_createDStream();
406
407 /* DISPLAY("params: cLevel %d, wlog %d hlog %d clog %d slog %d mml %d tlen %d strat %d \n",
408 cLevel, cparams->windowLog, cparams->hashLog, cparams->chainLog, cparams->searchLog,
409 cparams->minMatch, cparams->targetLength, cparams->strategy); */
410
411 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_compressionLevel, cLevel);
412 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_windowLog, (int)cparams.windowLog);
413 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_hashLog, (int)cparams.hashLog);
414 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_chainLog, (int)cparams.chainLog);
415 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_searchLog, (int)cparams.searchLog);
416 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_minMatch, (int)cparams.minMatch);
417 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_targetLength, (int)cparams.targetLength);
418 ZSTD_CCtx_setParameter(g_zcc, ZSTD_c_strategy, cparams.strategy);
419
420
421 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_compressionLevel, cLevel);
422 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_windowLog, (int)cparams.windowLog);
423 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_hashLog, (int)cparams.hashLog);
424 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_chainLog, (int)cparams.chainLog);
425 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_searchLog, (int)cparams.searchLog);
426 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_minMatch, (int)cparams.minMatch);
427 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_targetLength, (int)cparams.targetLength);
428 ZSTD_CCtx_setParameter(g_cstream, ZSTD_c_strategy, cparams.strategy);
429
430 /* Preparation */
431 switch(benchNb)
432 {
433 case 1:
434 payload = &cparams;
435 break;
436 case 2:
437 g_cSize = ZSTD_compress(dstBuff2, dstBuffSize, src, srcSize, cLevel);
438 break;
439 #ifndef ZSTD_DLL_IMPORT
440 case 11:
441 payload = &cparams;
442 break;
443 case 12:
444 payload = &cparams;
445 break;
446 case 13 :
447 g_cSize = ZSTD_compress(dstBuff2, dstBuffSize, src, srcSize, cLevel);
448 break;
449 case 31: /* ZSTD_decodeLiteralsBlock : starts literals block in dstBuff2 */
450 { size_t frameHeaderSize;
451 g_cSize = ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, cLevel);
452 frameHeaderSize = ZSTD_frameHeaderSize(dstBuff, ZSTD_FRAMEHEADERSIZE_PREFIX(ZSTD_f_zstd1));
453 CONTROL(!ZSTD_isError(frameHeaderSize));
454 /* check block is compressible, hence contains a literals section */
455 { blockProperties_t bp;
456 ZSTD_getcBlockSize(dstBuff+frameHeaderSize, dstBuffSize, &bp); /* Get 1st block type */
457 if (bp.blockType != bt_compressed) {
458 DISPLAY("ZSTD_decodeLiteralsBlock : impossible to test on this sample (not compressible)\n");
459 goto _cleanOut;
460 } }
461 { size_t const skippedSize = frameHeaderSize + ZSTD_blockHeaderSize;
462 memcpy(dstBuff2, dstBuff+skippedSize, g_cSize-skippedSize);
463 }
464 srcSize = srcSize > 128 KB ? 128 KB : srcSize; /* speed relative to block */
465 ZSTD_decompressBegin(g_zdc);
466 break;
467 }
468 case 32: /* ZSTD_decodeSeqHeaders */
469 { blockProperties_t bp;
470 const BYTE* ip = dstBuff;
471 const BYTE* iend;
472 { size_t const cSize = ZSTD_compress(dstBuff, dstBuffSize, src, srcSize, cLevel);
473 CONTROL(cSize > ZSTD_FRAMEHEADERSIZE_PREFIX(ZSTD_f_zstd1));
474 }
475 /* Skip frame Header */
476 { size_t const frameHeaderSize = ZSTD_frameHeaderSize(dstBuff, ZSTD_FRAMEHEADERSIZE_PREFIX(ZSTD_f_zstd1));
477 CONTROL(!ZSTD_isError(frameHeaderSize));
478 ip += frameHeaderSize;
479 }
480 /* Find end of block */
481 { size_t const cBlockSize = ZSTD_getcBlockSize(ip, dstBuffSize, &bp); /* Get 1st block type */
482 if (bp.blockType != bt_compressed) {
483 DISPLAY("ZSTD_decodeSeqHeaders : impossible to test on this sample (not compressible)\n");
484 goto _cleanOut;
485 }
486 iend = ip + ZSTD_blockHeaderSize + cBlockSize; /* End of first block */
487 }
488 ip += ZSTD_blockHeaderSize; /* skip block header */
489 ZSTD_decompressBegin(g_zdc);
490 CONTROL(iend > ip);
491 ip += ZSTD_decodeLiteralsBlock(g_zdc, ip, (size_t)(iend-ip)); /* skip literal segment */
492 g_cSize = (size_t)(iend-ip);
493 memcpy(dstBuff2, ip, g_cSize); /* copy rest of block (it starts by SeqHeader) */
494 srcSize = srcSize > 128 KB ? 128 KB : srcSize; /* speed relative to block */
495 break;
496 }
497 #else
498 case 31:
499 goto _cleanOut;
500 #endif
501 case 41 :
502 payload = &cparams;
503 break;
504 case 42 :
505 g_cSize = ZSTD_compress(payload, dstBuffSize, src, srcSize, cLevel);
506 break;
507 case 43 :
508 payload = &cparams;
509 break;
510
511 /* test functions */
512 /* convention: test functions have ID > 100 */
513
514 default : ;
515 }
516
517 /* warming up dstBuff */
518 { size_t i; for (i=0; i<dstBuffSize; i++) dstBuff[i]=(BYTE)i; }
519
520 /* benchmark loop */
521 { BMK_timedFnState_t* const tfs = BMK_createTimedFnState(g_nbIterations * 1000, 1000);
522 void* const avoidStrictAliasingPtr = &dstBuff;
523 BMK_benchParams_t bp;
524 BMK_runTime_t bestResult;
525 bestResult.sumOfReturn = 0;
526 bestResult.nanoSecPerRun = (double)TIMELOOP_NANOSEC * 2000000000; /* hopefully large enough : must be larger than any potential measurement */
527 CONTROL(tfs != NULL);
528
529 bp.benchFn = benchFunction;
530 bp.benchPayload = payload;
531 bp.initFn = NULL;
532 bp.initPayload = NULL;
533 bp.errorFn = ZSTD_isError;
534 bp.blockCount = 1;
535 bp.srcBuffers = &src;
536 bp.srcSizes = &srcSize;
537 bp.dstBuffers = (void* const*) avoidStrictAliasingPtr; /* circumvent strict aliasing warning on gcc-8,
538 * because gcc considers that `void* const *` and `void**` are 2 different types */
539 bp.dstCapacities = &dstBuffSize;
540 bp.blockResults = NULL;
541
542 for (;;) {
543 BMK_runOutcome_t const bOutcome = BMK_benchTimedFn(tfs, bp);
544
545 if (!BMK_isSuccessful_runOutcome(bOutcome)) {
546 DISPLAY("ERROR benchmarking function ! ! \n");
547 errorcode = 1;
548 goto _cleanOut;
549 }
550
551 { BMK_runTime_t const newResult = BMK_extract_runTime(bOutcome);
552 if (newResult.nanoSecPerRun < bestResult.nanoSecPerRun )
553 bestResult.nanoSecPerRun = newResult.nanoSecPerRun;
554 DISPLAY("\r%2u#%-29.29s:%8.1f MB/s (%8u) ",
555 benchNb, benchName,
556 (double)srcSize * TIMELOOP_NANOSEC / bestResult.nanoSecPerRun / MB_UNIT,
557 (unsigned)newResult.sumOfReturn );
558 }
559
560 if ( BMK_isCompleted_TimedFn(tfs) ) break;
561 }
562 BMK_freeTimedFnState(tfs);
563 }
564 DISPLAY("\n");
565
566 _cleanOut:
567 free(dstBuff);
568 free(dstBuff2);
569 ZSTD_freeCCtx(g_zcc); g_zcc=NULL;
570 ZSTD_freeDCtx(g_zdc); g_zdc=NULL;
571 ZSTD_freeCStream(g_cstream); g_cstream=NULL;
572 ZSTD_freeDStream(g_dstream); g_dstream=NULL;
573 return errorcode;
574 }
575
576
benchSample(U32 benchNb,size_t benchedSize,double compressibility,int cLevel,ZSTD_compressionParameters cparams)577 static int benchSample(U32 benchNb,
578 size_t benchedSize, double compressibility,
579 int cLevel, ZSTD_compressionParameters cparams)
580 {
581 /* Allocation */
582 void* const origBuff = malloc(benchedSize);
583 if (!origBuff) { DISPLAY("\nError: not enough memory!\n"); return 12; }
584
585 /* Fill buffer */
586 RDG_genBuffer(origBuff, benchedSize, compressibility, 0.0, 0);
587
588 /* bench */
589 DISPLAY("\r%70s\r", "");
590 DISPLAY(" Sample %u bytes : \n", (unsigned)benchedSize);
591 if (benchNb) {
592 benchMem(benchNb, origBuff, benchedSize, cLevel, cparams);
593 } else { /* 0 == run all tests */
594 for (benchNb=0; benchNb<100; benchNb++) {
595 benchMem(benchNb, origBuff, benchedSize, cLevel, cparams);
596 } }
597
598 free(origBuff);
599 return 0;
600 }
601
602
benchFiles(U32 benchNb,const char ** fileNamesTable,const int nbFiles,int cLevel,ZSTD_compressionParameters cparams)603 static int benchFiles(U32 benchNb,
604 const char** fileNamesTable, const int nbFiles,
605 int cLevel, ZSTD_compressionParameters cparams)
606 {
607 /* Loop for each file */
608 int fileIdx;
609 for (fileIdx=0; fileIdx<nbFiles; fileIdx++) {
610 const char* const inFileName = fileNamesTable[fileIdx];
611 FILE* const inFile = fopen( inFileName, "rb" );
612 size_t benchedSize;
613
614 /* Check file existence */
615 if (inFile==NULL) { DISPLAY( "Pb opening %s\n", inFileName); return 11; }
616
617 /* Memory allocation & restrictions */
618 { U64 const inFileSize = UTIL_getFileSize(inFileName);
619 if (inFileSize == UTIL_FILESIZE_UNKNOWN) {
620 DISPLAY( "Cannot measure size of %s\n", inFileName);
621 fclose(inFile);
622 return 11;
623 }
624 benchedSize = BMK_findMaxMem(inFileSize*3) / 3;
625 if ((U64)benchedSize > inFileSize)
626 benchedSize = (size_t)inFileSize;
627 if ((U64)benchedSize < inFileSize) {
628 DISPLAY("Not enough memory for '%s' full size; testing %u MB only... \n",
629 inFileName, (unsigned)(benchedSize>>20));
630 } }
631
632 /* Alloc */
633 { void* const origBuff = malloc(benchedSize);
634 if (!origBuff) { DISPLAY("\nError: not enough memory!\n"); fclose(inFile); return 12; }
635
636 /* Fill input buffer */
637 DISPLAY("Loading %s... \r", inFileName);
638 { size_t const readSize = fread(origBuff, 1, benchedSize, inFile);
639 fclose(inFile);
640 if (readSize != benchedSize) {
641 DISPLAY("\nError: problem reading file '%s' !! \n", inFileName);
642 free(origBuff);
643 return 13;
644 } }
645
646 /* bench */
647 DISPLAY("\r%70s\r", ""); /* blank line */
648 DISPLAY(" %s : \n", inFileName);
649 if (benchNb) {
650 benchMem(benchNb, origBuff, benchedSize, cLevel, cparams);
651 } else {
652 for (benchNb=0; benchNb<100; benchNb++) {
653 benchMem(benchNb, origBuff, benchedSize, cLevel, cparams);
654 } }
655
656 free(origBuff);
657 } }
658
659 return 0;
660 }
661
662
663
664 /*_*******************************************************
665 * Argument Parsing
666 *********************************************************/
667
668 #define ERROR_OUT(msg) { DISPLAY("%s \n", msg); exit(1); }
669
readU32FromChar(const char ** stringPtr)670 static unsigned readU32FromChar(const char** stringPtr)
671 {
672 const char errorMsg[] = "error: numeric value too large";
673 unsigned result = 0;
674 while ((**stringPtr >='0') && (**stringPtr <='9')) {
675 unsigned const max = (((unsigned)(-1)) / 10) - 1;
676 if (result > max) ERROR_OUT(errorMsg);
677 result *= 10;
678 result += (unsigned)(**stringPtr - '0');
679 (*stringPtr)++ ;
680 }
681 if ((**stringPtr=='K') || (**stringPtr=='M')) {
682 unsigned const maxK = ((unsigned)(-1)) >> 10;
683 if (result > maxK) ERROR_OUT(errorMsg);
684 result <<= 10;
685 if (**stringPtr=='M') {
686 if (result > maxK) ERROR_OUT(errorMsg);
687 result <<= 10;
688 }
689 (*stringPtr)++; /* skip `K` or `M` */
690 if (**stringPtr=='i') (*stringPtr)++;
691 if (**stringPtr=='B') (*stringPtr)++;
692 }
693 return result;
694 }
695
longCommandWArg(const char ** stringPtr,const char * longCommand)696 static int longCommandWArg(const char** stringPtr, const char* longCommand)
697 {
698 size_t const comSize = strlen(longCommand);
699 int const result = !strncmp(*stringPtr, longCommand, comSize);
700 if (result) *stringPtr += comSize;
701 return result;
702 }
703
704
705 /*_*******************************************************
706 * Command line
707 *********************************************************/
708
usage(const char * exename)709 static int usage(const char* exename)
710 {
711 DISPLAY( "Usage :\n");
712 DISPLAY( " %s [arg] file1 file2 ... fileX\n", exename);
713 DISPLAY( "Arguments :\n");
714 DISPLAY( " -H/-h : Help (this text + advanced options)\n");
715 return 0;
716 }
717
usage_advanced(const char * exename)718 static int usage_advanced(const char* exename)
719 {
720 usage(exename);
721 DISPLAY( "\nAdvanced options :\n");
722 DISPLAY( " -b# : test only function # \n");
723 DISPLAY( " -l# : benchmark functions at that compression level (default : %i)\n", DEFAULT_CLEVEL);
724 DISPLAY( "--zstd= : custom parameter selection. Format same as zstdcli \n");
725 DISPLAY( " -P# : sample compressibility (default : %.1f%%)\n", COMPRESSIBILITY_DEFAULT * 100);
726 DISPLAY( " -B# : sample size (default : %u)\n", (unsigned)kSampleSizeDefault);
727 DISPLAY( " -i# : iteration loops [1-9](default : %i)\n", NBLOOPS);
728 return 0;
729 }
730
badusage(const char * exename)731 static int badusage(const char* exename)
732 {
733 DISPLAY("Wrong parameters\n");
734 usage(exename);
735 return 1;
736 }
737
main(int argc,const char ** argv)738 int main(int argc, const char** argv)
739 {
740 int argNb, filenamesStart=0, result;
741 const char* const exename = argv[0];
742 const char* input_filename = NULL;
743 U32 benchNb = 0, main_pause = 0;
744 int cLevel = DEFAULT_CLEVEL;
745 ZSTD_compressionParameters cparams = ZSTD_getCParams(cLevel, 0, 0);
746 size_t sampleSize = kSampleSizeDefault;
747 double compressibility = COMPRESSIBILITY_DEFAULT;
748
749 DISPLAY(WELCOME_MESSAGE);
750 if (argc<1) return badusage(exename);
751
752 for (argNb=1; argNb<argc; argNb++) {
753 const char* argument = argv[argNb];
754 CONTROL(argument != NULL);
755
756 if (longCommandWArg(&argument, "--zstd=")) {
757 for ( ; ;) {
758 if (longCommandWArg(&argument, "windowLog=") || longCommandWArg(&argument, "wlog=")) { cparams.windowLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; }
759 if (longCommandWArg(&argument, "chainLog=") || longCommandWArg(&argument, "clog=")) { cparams.chainLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; }
760 if (longCommandWArg(&argument, "hashLog=") || longCommandWArg(&argument, "hlog=")) { cparams.hashLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; }
761 if (longCommandWArg(&argument, "searchLog=") || longCommandWArg(&argument, "slog=")) { cparams.searchLog = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; }
762 if (longCommandWArg(&argument, "minMatch=") || longCommandWArg(&argument, "mml=")) { cparams.minMatch = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; }
763 if (longCommandWArg(&argument, "targetLength=") || longCommandWArg(&argument, "tlen=")) { cparams.targetLength = readU32FromChar(&argument); if (argument[0]==',') { argument++; continue; } else break; }
764 if (longCommandWArg(&argument, "strategy=") || longCommandWArg(&argument, "strat=")) { cparams.strategy = (ZSTD_strategy)(readU32FromChar(&argument)); if (argument[0]==',') { argument++; continue; } else break; }
765 if (longCommandWArg(&argument, "level=") || longCommandWArg(&argument, "lvl=")) { cLevel = (int)readU32FromChar(&argument); cparams = ZSTD_getCParams(cLevel, 0, 0); if (argument[0]==',') { argument++; continue; } else break; }
766 DISPLAY("invalid compression parameter \n");
767 return 1;
768 }
769
770 /* check end of string */
771 if (argument[0] != 0) {
772 DISPLAY("invalid --zstd= format \n");
773 return 1;
774 } else {
775 continue;
776 }
777
778 } else if (argument[0]=='-') { /* Commands (note : aggregated commands are allowed) */
779 argument++;
780 while (argument[0]!=0) {
781
782 switch(argument[0])
783 {
784 /* Display help on usage */
785 case 'h':
786 case 'H': return usage_advanced(exename);
787
788 /* Pause at the end (hidden option) */
789 case 'p': main_pause = 1; break;
790
791 /* Select specific algorithm to bench */
792 case 'b':
793 argument++;
794 benchNb = readU32FromChar(&argument);
795 break;
796
797 /* Select compression level to use */
798 case 'l':
799 argument++;
800 cLevel = (int)readU32FromChar(&argument);
801 cparams = ZSTD_getCParams(cLevel, 0, 0);
802 break;
803
804 /* Select compressibility of synthetic sample */
805 case 'P':
806 argument++;
807 compressibility = (double)readU32FromChar(&argument) / 100.;
808 break;
809
810 /* Select size of synthetic sample */
811 case 'B':
812 argument++;
813 sampleSize = (size_t)readU32FromChar(&argument);
814 break;
815
816 /* Modify Nb Iterations */
817 case 'i':
818 argument++;
819 g_nbIterations = readU32FromChar(&argument);
820 break;
821
822 /* Unknown command */
823 default : return badusage(exename);
824 }
825 }
826 continue;
827 }
828
829 /* first provided filename is input */
830 if (!input_filename) { input_filename=argument; filenamesStart=argNb; continue; }
831 }
832
833
834
835 if (filenamesStart==0) /* no input file */
836 result = benchSample(benchNb, sampleSize, compressibility, cLevel, cparams);
837 else
838 result = benchFiles(benchNb, argv+filenamesStart, argc-filenamesStart, cLevel, cparams);
839
840 if (main_pause) { int unused; printf("press enter...\n"); unused = getchar(); (void)unused; }
841
842 return result;
843 }
844