| 1 | /* |
| 2 | * Copyright (c) Meta Platforms, Inc. and affiliates. |
| 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 | /* largeNbDicts |
| 12 | * This is a benchmark test tool |
| 13 | * dedicated to the specific case of dictionary decompression |
| 14 | * using a very large nb of dictionaries |
| 15 | * thus suffering latency from lots of cache misses. |
| 16 | * It's created in a bid to investigate performance and find optimizations. */ |
| 17 | |
| 18 | |
| 19 | /*--- Dependencies ---*/ |
| 20 | |
| 21 | #include <stddef.h> /* size_t */ |
| 22 | #include <stdlib.h> /* malloc, free, abort, qsort*/ |
| 23 | #include <stdio.h> /* fprintf */ |
| 24 | #include <limits.h> /* UINT_MAX */ |
| 25 | #include <assert.h> /* assert */ |
| 26 | |
| 27 | #include "util.h" |
| 28 | #include "benchfn.h" |
| 29 | #define ZSTD_STATIC_LINKING_ONLY |
| 30 | #include "zstd.h" |
| 31 | #include "zdict.h" |
| 32 | |
| 33 | |
| 34 | /*--- Constants --- */ |
| 35 | |
| 36 | #define KB *(1<<10) |
| 37 | #define MB *(1<<20) |
| 38 | |
| 39 | #define BLOCKSIZE_DEFAULT 0 /* no slicing into blocks */ |
| 40 | #define DICTSIZE (4 KB) |
| 41 | #define CLEVEL_DEFAULT 3 |
| 42 | #define DICT_LOAD_METHOD ZSTD_dlm_byCopy |
| 43 | |
| 44 | #define BENCH_TIME_DEFAULT_S 6 |
| 45 | #define RUN_TIME_DEFAULT_MS 1000 |
| 46 | #define BENCH_TIME_DEFAULT_MS (BENCH_TIME_DEFAULT_S * RUN_TIME_DEFAULT_MS) |
| 47 | |
| 48 | #define DISPLAY_LEVEL_DEFAULT 3 |
| 49 | |
| 50 | #define BENCH_SIZE_MAX (1200 MB) |
| 51 | |
| 52 | |
| 53 | /*--- Macros ---*/ |
| 54 | |
| 55 | #define CONTROL(c) { if (!(c)) abort(); } |
| 56 | #undef MIN |
| 57 | #define MIN(a,b) ((a) < (b) ? (a) : (b)) |
| 58 | |
| 59 | |
| 60 | /*--- Display Macros ---*/ |
| 61 | |
| 62 | #define DISPLAY(...) fprintf(stdout, __VA_ARGS__) |
| 63 | #define DISPLAYLEVEL(l, ...) { if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); } } |
| 64 | static int g_displayLevel = DISPLAY_LEVEL_DEFAULT; /* 0 : no display, 1: errors, 2 : + result + interaction + warnings, 3 : + progression, 4 : + information */ |
| 65 | |
| 66 | |
| 67 | /*--- buffer_t ---*/ |
| 68 | |
| 69 | typedef struct { |
| 70 | void* ptr; |
| 71 | size_t size; |
| 72 | size_t capacity; |
| 73 | } buffer_t; |
| 74 | |
| 75 | static const buffer_t kBuffNull = { NULL, 0, 0 }; |
| 76 | |
| 77 | /* @return : kBuffNull if any error */ |
| 78 | static buffer_t createBuffer(size_t capacity) |
| 79 | { |
| 80 | assert(capacity > 0); |
| 81 | void* const ptr = malloc(capacity); |
| 82 | if (ptr==NULL) return kBuffNull; |
| 83 | |
| 84 | buffer_t buffer; |
| 85 | buffer.ptr = ptr; |
| 86 | buffer.capacity = capacity; |
| 87 | buffer.size = 0; |
| 88 | return buffer; |
| 89 | } |
| 90 | |
| 91 | static void freeBuffer(buffer_t buff) |
| 92 | { |
| 93 | free(buff.ptr); |
| 94 | } |
| 95 | |
| 96 | |
| 97 | static void fillBuffer_fromHandle(buffer_t* buff, FILE* f) |
| 98 | { |
| 99 | size_t const readSize = fread(buff->ptr, 1, buff->capacity, f); |
| 100 | buff->size = readSize; |
| 101 | } |
| 102 | |
| 103 | |
| 104 | /* @return : kBuffNull if any error */ |
| 105 | static buffer_t createBuffer_fromFile(const char* fileName) |
| 106 | { |
| 107 | U64 const fileSize = UTIL_getFileSize(fileName); |
| 108 | size_t const bufferSize = (size_t) fileSize; |
| 109 | |
| 110 | if (fileSize == UTIL_FILESIZE_UNKNOWN) return kBuffNull; |
| 111 | assert((U64)bufferSize == fileSize); /* check overflow */ |
| 112 | |
| 113 | { FILE* const f = fopen(fileName, "rb"); |
| 114 | if (f == NULL) return kBuffNull; |
| 115 | |
| 116 | buffer_t buff = createBuffer(bufferSize); |
| 117 | CONTROL(buff.ptr != NULL); |
| 118 | |
| 119 | fillBuffer_fromHandle(&buff, f); |
| 120 | CONTROL(buff.size == buff.capacity); |
| 121 | |
| 122 | fclose(f); /* do nothing specific if fclose() fails */ |
| 123 | return buff; |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | |
| 128 | /* @return : kBuffNull if any error */ |
| 129 | static buffer_t |
| 130 | createDictionaryBuffer(const char* dictionaryName, |
| 131 | const void* srcBuffer, |
| 132 | const size_t* srcBlockSizes, size_t nbBlocks, |
| 133 | size_t requestedDictSize) |
| 134 | { |
| 135 | if (dictionaryName) { |
| 136 | DISPLAYLEVEL(3, "loading dictionary %s \n", dictionaryName); |
| 137 | return createBuffer_fromFile(dictionaryName); /* note : result might be kBuffNull */ |
| 138 | |
| 139 | } else { |
| 140 | |
| 141 | DISPLAYLEVEL(3, "creating dictionary, of target size %u bytes \n", |
| 142 | (unsigned)requestedDictSize); |
| 143 | void* const dictBuffer = malloc(requestedDictSize); |
| 144 | CONTROL(dictBuffer != NULL); |
| 145 | |
| 146 | assert(nbBlocks <= UINT_MAX); |
| 147 | size_t const dictSize = ZDICT_trainFromBuffer(dictBuffer, requestedDictSize, |
| 148 | srcBuffer, |
| 149 | srcBlockSizes, (unsigned)nbBlocks); |
| 150 | CONTROL(!ZSTD_isError(dictSize)); |
| 151 | |
| 152 | buffer_t result; |
| 153 | result.ptr = dictBuffer; |
| 154 | result.capacity = requestedDictSize; |
| 155 | result.size = dictSize; |
| 156 | return result; |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | /*! BMK_loadFiles() : |
| 161 | * Loads `buffer`, with content from files listed within `fileNamesTable`. |
| 162 | * Fills `buffer` entirely. |
| 163 | * @return : 0 on success, !=0 on error */ |
| 164 | static int loadFiles(void* buffer, size_t bufferSize, |
| 165 | size_t* fileSizes, |
| 166 | const char* const * fileNamesTable, unsigned nbFiles) |
| 167 | { |
| 168 | size_t pos = 0, totalSize = 0; |
| 169 | |
| 170 | for (unsigned n=0; n<nbFiles; n++) { |
| 171 | U64 fileSize = UTIL_getFileSize(fileNamesTable[n]); |
| 172 | if (UTIL_isDirectory(fileNamesTable[n])) { |
| 173 | fileSizes[n] = 0; |
| 174 | continue; |
| 175 | } |
| 176 | if (fileSize == UTIL_FILESIZE_UNKNOWN) { |
| 177 | fileSizes[n] = 0; |
| 178 | continue; |
| 179 | } |
| 180 | |
| 181 | FILE* const f = fopen(fileNamesTable[n], "rb"); |
| 182 | assert(f!=NULL); |
| 183 | |
| 184 | assert(pos <= bufferSize); |
| 185 | assert(fileSize <= bufferSize - pos); |
| 186 | |
| 187 | { size_t const readSize = fread(((char*)buffer)+pos, 1, (size_t)fileSize, f); |
| 188 | assert(readSize == fileSize); |
| 189 | pos += readSize; |
| 190 | } |
| 191 | fileSizes[n] = (size_t)fileSize; |
| 192 | totalSize += (size_t)fileSize; |
| 193 | fclose(f); |
| 194 | } |
| 195 | |
| 196 | assert(totalSize == bufferSize); |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | |
| 201 | |
| 202 | /*--- slice_collection_t ---*/ |
| 203 | |
| 204 | typedef struct { |
| 205 | void** slicePtrs; |
| 206 | size_t* capacities; |
| 207 | size_t nbSlices; |
| 208 | } slice_collection_t; |
| 209 | |
| 210 | static const slice_collection_t kNullCollection = { NULL, NULL, 0 }; |
| 211 | |
| 212 | static void freeSliceCollection(slice_collection_t collection) |
| 213 | { |
| 214 | free(collection.slicePtrs); |
| 215 | free(collection.capacities); |
| 216 | } |
| 217 | |
| 218 | /* shrinkSizes() : |
| 219 | * downsizes sizes of slices within collection, according to `newSizes`. |
| 220 | * every `newSizes` entry must be <= than its corresponding collection size */ |
| 221 | void shrinkSizes(slice_collection_t collection, |
| 222 | const size_t* newSizes) /* presumed same size as collection */ |
| 223 | { |
| 224 | size_t const nbSlices = collection.nbSlices; |
| 225 | for (size_t blockNb = 0; blockNb < nbSlices; blockNb++) { |
| 226 | assert(newSizes[blockNb] <= collection.capacities[blockNb]); |
| 227 | collection.capacities[blockNb] = newSizes[blockNb]; |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | |
| 232 | /* splitSlices() : |
| 233 | * nbSlices : if == 0, nbSlices is automatically determined from srcSlices and blockSize. |
| 234 | * otherwise, creates exactly nbSlices slices, |
| 235 | * by either truncating input (when smaller) |
| 236 | * or repeating input from beginning */ |
| 237 | static slice_collection_t |
| 238 | splitSlices(slice_collection_t srcSlices, size_t blockSize, size_t nbSlices) |
| 239 | { |
| 240 | if (blockSize==0) blockSize = (size_t)(-1); /* means "do not cut" */ |
| 241 | size_t nbSrcBlocks = 0; |
| 242 | for (size_t ssnb=0; ssnb < srcSlices.nbSlices; ssnb++) { |
| 243 | size_t pos = 0; |
| 244 | while (pos <= srcSlices.capacities[ssnb]) { |
| 245 | nbSrcBlocks++; |
| 246 | pos += blockSize; |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | if (nbSlices == 0) nbSlices = nbSrcBlocks; |
| 251 | |
| 252 | void** const sliceTable = (void**)malloc(nbSlices * sizeof(*sliceTable)); |
| 253 | size_t* const capacities = (size_t*)malloc(nbSlices * sizeof(*capacities)); |
| 254 | if (sliceTable == NULL || capacities == NULL) { |
| 255 | free(sliceTable); |
| 256 | free(capacities); |
| 257 | return kNullCollection; |
| 258 | } |
| 259 | |
| 260 | size_t ssnb = 0; |
| 261 | for (size_t sliceNb=0; sliceNb < nbSlices; ) { |
| 262 | ssnb = (ssnb + 1) % srcSlices.nbSlices; |
| 263 | size_t pos = 0; |
| 264 | char* const ptr = (char*)srcSlices.slicePtrs[ssnb]; |
| 265 | while (pos < srcSlices.capacities[ssnb] && sliceNb < nbSlices) { |
| 266 | size_t const size = MIN(blockSize, srcSlices.capacities[ssnb] - pos); |
| 267 | sliceTable[sliceNb] = ptr + pos; |
| 268 | capacities[sliceNb] = size; |
| 269 | sliceNb++; |
| 270 | pos += blockSize; |
| 271 | } |
| 272 | } |
| 273 | |
| 274 | slice_collection_t result; |
| 275 | result.nbSlices = nbSlices; |
| 276 | result.slicePtrs = sliceTable; |
| 277 | result.capacities = capacities; |
| 278 | return result; |
| 279 | } |
| 280 | |
| 281 | |
| 282 | static size_t sliceCollection_totalCapacity(slice_collection_t sc) |
| 283 | { |
| 284 | size_t totalSize = 0; |
| 285 | for (size_t n=0; n<sc.nbSlices; n++) |
| 286 | totalSize += sc.capacities[n]; |
| 287 | return totalSize; |
| 288 | } |
| 289 | |
| 290 | |
| 291 | /* --- buffer collection --- */ |
| 292 | |
| 293 | typedef struct { |
| 294 | buffer_t buffer; |
| 295 | slice_collection_t slices; |
| 296 | } buffer_collection_t; |
| 297 | |
| 298 | |
| 299 | static void freeBufferCollection(buffer_collection_t bc) |
| 300 | { |
| 301 | freeBuffer(bc.buffer); |
| 302 | freeSliceCollection(bc.slices); |
| 303 | } |
| 304 | |
| 305 | |
| 306 | static buffer_collection_t |
| 307 | createBufferCollection_fromSliceCollectionSizes(slice_collection_t sc) |
| 308 | { |
| 309 | size_t const bufferSize = sliceCollection_totalCapacity(sc); |
| 310 | |
| 311 | buffer_t buffer = createBuffer(bufferSize); |
| 312 | CONTROL(buffer.ptr != NULL); |
| 313 | |
| 314 | size_t const nbSlices = sc.nbSlices; |
| 315 | void** const slices = (void**)malloc(nbSlices * sizeof(*slices)); |
| 316 | CONTROL(slices != NULL); |
| 317 | |
| 318 | size_t* const capacities = (size_t*)malloc(nbSlices * sizeof(*capacities)); |
| 319 | CONTROL(capacities != NULL); |
| 320 | |
| 321 | char* const ptr = (char*)buffer.ptr; |
| 322 | size_t pos = 0; |
| 323 | for (size_t n=0; n < nbSlices; n++) { |
| 324 | capacities[n] = sc.capacities[n]; |
| 325 | slices[n] = ptr + pos; |
| 326 | pos += capacities[n]; |
| 327 | } |
| 328 | |
| 329 | buffer_collection_t result; |
| 330 | result.buffer = buffer; |
| 331 | result.slices.nbSlices = nbSlices; |
| 332 | result.slices.capacities = capacities; |
| 333 | result.slices.slicePtrs = slices; |
| 334 | return result; |
| 335 | } |
| 336 | |
| 337 | static buffer_collection_t |
| 338 | createBufferCollection_fromSliceCollection(slice_collection_t sc) |
| 339 | { |
| 340 | size_t const bufferSize = sliceCollection_totalCapacity(sc); |
| 341 | |
| 342 | buffer_t buffer = createBuffer(bufferSize); |
| 343 | CONTROL(buffer.ptr != NULL); |
| 344 | |
| 345 | size_t const nbSlices = sc.nbSlices; |
| 346 | void** const slices = (void**)malloc(nbSlices * sizeof(*slices)); |
| 347 | CONTROL(slices != NULL); |
| 348 | |
| 349 | size_t* const capacities = (size_t*)malloc(nbSlices * sizeof(*capacities)); |
| 350 | CONTROL(capacities != NULL); |
| 351 | |
| 352 | char* const ptr = (char*)buffer.ptr; |
| 353 | size_t pos = 0; |
| 354 | for (size_t n=0; n < nbSlices; n++) { |
| 355 | capacities[n] = sc.capacities[n]; |
| 356 | slices[n] = ptr + pos; |
| 357 | pos += capacities[n]; |
| 358 | } |
| 359 | |
| 360 | for (size_t i = 0; i < nbSlices; i++) { |
| 361 | memcpy(slices[i], sc.slicePtrs[i], sc.capacities[i]); |
| 362 | capacities[i] = sc.capacities[i]; |
| 363 | } |
| 364 | |
| 365 | buffer_collection_t result; |
| 366 | result.buffer = buffer; |
| 367 | result.slices.nbSlices = nbSlices; |
| 368 | result.slices.capacities = capacities; |
| 369 | result.slices.slicePtrs = slices; |
| 370 | |
| 371 | return result; |
| 372 | } |
| 373 | |
| 374 | /* @return : kBuffNull if any error */ |
| 375 | static buffer_collection_t |
| 376 | createBufferCollection_fromFiles(const char* const * fileNamesTable, unsigned nbFiles) |
| 377 | { |
| 378 | U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles); |
| 379 | assert(totalSizeToLoad != UTIL_FILESIZE_UNKNOWN); |
| 380 | assert(totalSizeToLoad <= BENCH_SIZE_MAX); |
| 381 | size_t const loadedSize = (size_t)totalSizeToLoad; |
| 382 | assert(loadedSize > 0); |
| 383 | void* const srcBuffer = malloc(loadedSize); |
| 384 | assert(srcBuffer != NULL); |
| 385 | |
| 386 | assert(nbFiles > 0); |
| 387 | size_t* const fileSizes = (size_t*)calloc(nbFiles, sizeof(*fileSizes)); |
| 388 | assert(fileSizes != NULL); |
| 389 | |
| 390 | /* Load input buffer */ |
| 391 | int const errorCode = loadFiles(srcBuffer, loadedSize, |
| 392 | fileSizes, |
| 393 | fileNamesTable, nbFiles); |
| 394 | assert(errorCode == 0); |
| 395 | |
| 396 | void** sliceTable = (void**)malloc(nbFiles * sizeof(*sliceTable)); |
| 397 | assert(sliceTable != NULL); |
| 398 | |
| 399 | char* const ptr = (char*)srcBuffer; |
| 400 | size_t pos = 0; |
| 401 | unsigned fileNb = 0; |
| 402 | for ( ; (pos < loadedSize) && (fileNb < nbFiles); fileNb++) { |
| 403 | sliceTable[fileNb] = ptr + pos; |
| 404 | pos += fileSizes[fileNb]; |
| 405 | } |
| 406 | assert(pos == loadedSize); |
| 407 | assert(fileNb == nbFiles); |
| 408 | |
| 409 | |
| 410 | buffer_t buffer; |
| 411 | buffer.ptr = srcBuffer; |
| 412 | buffer.capacity = loadedSize; |
| 413 | buffer.size = loadedSize; |
| 414 | |
| 415 | slice_collection_t slices; |
| 416 | slices.slicePtrs = sliceTable; |
| 417 | slices.capacities = fileSizes; |
| 418 | slices.nbSlices = nbFiles; |
| 419 | |
| 420 | buffer_collection_t bc; |
| 421 | bc.buffer = buffer; |
| 422 | bc.slices = slices; |
| 423 | return bc; |
| 424 | } |
| 425 | |
| 426 | |
| 427 | |
| 428 | |
| 429 | /*--- ddict_collection_t ---*/ |
| 430 | |
| 431 | typedef struct { |
| 432 | ZSTD_DDict** ddicts; |
| 433 | size_t nbDDict; |
| 434 | } ddict_collection_t; |
| 435 | |
| 436 | typedef struct { |
| 437 | ZSTD_CDict** cdicts; |
| 438 | size_t nbCDict; |
| 439 | } cdict_collection_t; |
| 440 | |
| 441 | static const cdict_collection_t kNullCDictCollection = { NULL, 0 }; |
| 442 | |
| 443 | static void freeCDictCollection(cdict_collection_t cdictc) |
| 444 | { |
| 445 | for (size_t dictNb=0; dictNb < cdictc.nbCDict; dictNb++) { |
| 446 | ZSTD_freeCDict(cdictc.cdicts[dictNb]); |
| 447 | } |
| 448 | free(cdictc.cdicts); |
| 449 | } |
| 450 | |
| 451 | /* returns .buffers=NULL if operation fails */ |
| 452 | static cdict_collection_t createCDictCollection(const void* dictBuffer, size_t dictSize, size_t nbCDict, ZSTD_dictContentType_e dictContentType, ZSTD_CCtx_params* cctxParams) |
| 453 | { |
| 454 | ZSTD_CDict** const cdicts = malloc(nbCDict * sizeof(ZSTD_CDict*)); |
| 455 | if (cdicts==NULL) return kNullCDictCollection; |
| 456 | for (size_t dictNb=0; dictNb < nbCDict; dictNb++) { |
| 457 | cdicts[dictNb] = ZSTD_createCDict_advanced2(dictBuffer, dictSize, DICT_LOAD_METHOD, dictContentType, cctxParams, ZSTD_defaultCMem); |
| 458 | CONTROL(cdicts[dictNb] != NULL); |
| 459 | } |
| 460 | cdict_collection_t cdictc; |
| 461 | cdictc.cdicts = cdicts; |
| 462 | cdictc.nbCDict = nbCDict; |
| 463 | return cdictc; |
| 464 | } |
| 465 | |
| 466 | static const ddict_collection_t kNullDDictCollection = { NULL, 0 }; |
| 467 | |
| 468 | static void freeDDictCollection(ddict_collection_t ddictc) |
| 469 | { |
| 470 | for (size_t dictNb=0; dictNb < ddictc.nbDDict; dictNb++) { |
| 471 | ZSTD_freeDDict(ddictc.ddicts[dictNb]); |
| 472 | } |
| 473 | free(ddictc.ddicts); |
| 474 | } |
| 475 | |
| 476 | /* returns .buffers=NULL if operation fails */ |
| 477 | static ddict_collection_t createDDictCollection(const void* dictBuffer, size_t dictSize, size_t nbDDict) |
| 478 | { |
| 479 | ZSTD_DDict** const ddicts = malloc(nbDDict * sizeof(ZSTD_DDict*)); |
| 480 | assert(ddicts != NULL); |
| 481 | if (ddicts==NULL) return kNullDDictCollection; |
| 482 | for (size_t dictNb=0; dictNb < nbDDict; dictNb++) { |
| 483 | ddicts[dictNb] = ZSTD_createDDict(dictBuffer, dictSize); |
| 484 | assert(ddicts[dictNb] != NULL); |
| 485 | } |
| 486 | ddict_collection_t ddictc; |
| 487 | ddictc.ddicts = ddicts; |
| 488 | ddictc.nbDDict = nbDDict; |
| 489 | return ddictc; |
| 490 | } |
| 491 | |
| 492 | |
| 493 | /* mess with addresses, so that linear scanning dictionaries != linear address scanning */ |
| 494 | void shuffleCDictionaries(cdict_collection_t dicts) |
| 495 | { |
| 496 | size_t const nbDicts = dicts.nbCDict; |
| 497 | for (size_t r=0; r<nbDicts; r++) { |
| 498 | size_t const d = (size_t)rand() % nbDicts; |
| 499 | ZSTD_CDict* tmpd = dicts.cdicts[d]; |
| 500 | dicts.cdicts[d] = dicts.cdicts[r]; |
| 501 | dicts.cdicts[r] = tmpd; |
| 502 | } |
| 503 | for (size_t r=0; r<nbDicts; r++) { |
| 504 | size_t const d1 = (size_t)rand() % nbDicts; |
| 505 | size_t const d2 = (size_t)rand() % nbDicts; |
| 506 | ZSTD_CDict* tmpd = dicts.cdicts[d1]; |
| 507 | dicts.cdicts[d1] = dicts.cdicts[d2]; |
| 508 | dicts.cdicts[d2] = tmpd; |
| 509 | } |
| 510 | } |
| 511 | |
| 512 | /* mess with addresses, so that linear scanning dictionaries != linear address scanning */ |
| 513 | void shuffleDDictionaries(ddict_collection_t dicts) |
| 514 | { |
| 515 | size_t const nbDicts = dicts.nbDDict; |
| 516 | for (size_t r=0; r<nbDicts; r++) { |
| 517 | size_t const d = (size_t)rand() % nbDicts; |
| 518 | ZSTD_DDict* tmpd = dicts.ddicts[d]; |
| 519 | dicts.ddicts[d] = dicts.ddicts[r]; |
| 520 | dicts.ddicts[r] = tmpd; |
| 521 | } |
| 522 | for (size_t r=0; r<nbDicts; r++) { |
| 523 | size_t const d1 = (size_t)rand() % nbDicts; |
| 524 | size_t const d2 = (size_t)rand() % nbDicts; |
| 525 | ZSTD_DDict* tmpd = dicts.ddicts[d1]; |
| 526 | dicts.ddicts[d1] = dicts.ddicts[d2]; |
| 527 | dicts.ddicts[d2] = tmpd; |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | |
| 532 | /* --- Compression --- */ |
| 533 | |
| 534 | /* compressBlocks() : |
| 535 | * @return : total compressed size of all blocks, |
| 536 | * or 0 if error. |
| 537 | */ |
| 538 | static size_t compressBlocks(size_t* cSizes, /* optional (can be NULL). If present, must contain at least nbBlocks fields */ |
| 539 | slice_collection_t dstBlockBuffers, |
| 540 | slice_collection_t srcBlockBuffers, |
| 541 | ZSTD_CDict* cdict, int cLevel) |
| 542 | { |
| 543 | size_t const nbBlocks = srcBlockBuffers.nbSlices; |
| 544 | assert(dstBlockBuffers.nbSlices == srcBlockBuffers.nbSlices); |
| 545 | |
| 546 | ZSTD_CCtx* const cctx = ZSTD_createCCtx(); |
| 547 | assert(cctx != NULL); |
| 548 | |
| 549 | size_t totalCSize = 0; |
| 550 | for (size_t blockNb=0; blockNb < nbBlocks; blockNb++) { |
| 551 | size_t cBlockSize; |
| 552 | if (cdict == NULL) { |
| 553 | cBlockSize = ZSTD_compressCCtx(cctx, |
| 554 | dstBlockBuffers.slicePtrs[blockNb], dstBlockBuffers.capacities[blockNb], |
| 555 | srcBlockBuffers.slicePtrs[blockNb], srcBlockBuffers.capacities[blockNb], |
| 556 | cLevel); |
| 557 | } else { |
| 558 | cBlockSize = ZSTD_compress_usingCDict(cctx, |
| 559 | dstBlockBuffers.slicePtrs[blockNb], dstBlockBuffers.capacities[blockNb], |
| 560 | srcBlockBuffers.slicePtrs[blockNb], srcBlockBuffers.capacities[blockNb], |
| 561 | cdict); |
| 562 | } |
| 563 | CONTROL(!ZSTD_isError(cBlockSize)); |
| 564 | if (cSizes) cSizes[blockNb] = cBlockSize; |
| 565 | totalCSize += cBlockSize; |
| 566 | } |
| 567 | return totalCSize; |
| 568 | } |
| 569 | |
| 570 | |
| 571 | /* --- Benchmark --- */ |
| 572 | |
| 573 | typedef struct { |
| 574 | ZSTD_CCtx* cctx; |
| 575 | size_t nbDicts; |
| 576 | size_t dictNb; |
| 577 | cdict_collection_t dictionaries; |
| 578 | } compressInstructions; |
| 579 | |
| 580 | compressInstructions createCompressInstructions(cdict_collection_t dictionaries, ZSTD_CCtx_params* cctxParams) |
| 581 | { |
| 582 | compressInstructions ci; |
| 583 | ci.cctx = ZSTD_createCCtx(); |
| 584 | CONTROL(ci.cctx != NULL); |
| 585 | if (cctxParams) |
| 586 | ZSTD_CCtx_setParametersUsingCCtxParams(ci.cctx, cctxParams); |
| 587 | ci.nbDicts = dictionaries.nbCDict; |
| 588 | ci.dictNb = 0; |
| 589 | ci.dictionaries = dictionaries; |
| 590 | return ci; |
| 591 | } |
| 592 | |
| 593 | void freeCompressInstructions(compressInstructions ci) |
| 594 | { |
| 595 | ZSTD_freeCCtx(ci.cctx); |
| 596 | } |
| 597 | |
| 598 | typedef struct { |
| 599 | ZSTD_DCtx* dctx; |
| 600 | size_t nbDicts; |
| 601 | size_t dictNb; |
| 602 | ddict_collection_t dictionaries; |
| 603 | } decompressInstructions; |
| 604 | |
| 605 | decompressInstructions createDecompressInstructions(ddict_collection_t dictionaries) |
| 606 | { |
| 607 | decompressInstructions di; |
| 608 | di.dctx = ZSTD_createDCtx(); |
| 609 | assert(di.dctx != NULL); |
| 610 | di.nbDicts = dictionaries.nbDDict; |
| 611 | di.dictNb = 0; |
| 612 | di.dictionaries = dictionaries; |
| 613 | return di; |
| 614 | } |
| 615 | |
| 616 | void freeDecompressInstructions(decompressInstructions di) |
| 617 | { |
| 618 | ZSTD_freeDCtx(di.dctx); |
| 619 | } |
| 620 | |
| 621 | /* benched function */ |
| 622 | size_t compress(const void* src, size_t srcSize, void* dst, size_t dstCapacity, void* payload) |
| 623 | { |
| 624 | compressInstructions* const ci = (compressInstructions*) payload; |
| 625 | (void)dstCapacity; |
| 626 | |
| 627 | ZSTD_CCtx_refCDict(ci->cctx, ci->dictionaries.cdicts[ci->dictNb]); |
| 628 | ZSTD_compress2(ci->cctx, |
| 629 | dst, srcSize, |
| 630 | src, srcSize); |
| 631 | |
| 632 | ci->dictNb = ci->dictNb + 1; |
| 633 | if (ci->dictNb >= ci->nbDicts) ci->dictNb = 0; |
| 634 | |
| 635 | return srcSize; |
| 636 | } |
| 637 | |
| 638 | /* benched function */ |
| 639 | size_t decompress(const void* src, size_t srcSize, void* dst, size_t dstCapacity, void* payload) |
| 640 | { |
| 641 | decompressInstructions* const di = (decompressInstructions*) payload; |
| 642 | |
| 643 | size_t const result = ZSTD_decompress_usingDDict(di->dctx, |
| 644 | dst, dstCapacity, |
| 645 | src, srcSize, |
| 646 | di->dictionaries.ddicts[di->dictNb]); |
| 647 | |
| 648 | di->dictNb = di->dictNb + 1; |
| 649 | if (di->dictNb >= di->nbDicts) di->dictNb = 0; |
| 650 | |
| 651 | return result; |
| 652 | } |
| 653 | |
| 654 | typedef enum { |
| 655 | fastest = 0, |
| 656 | median = 1, |
| 657 | } metricAggregatePref_e; |
| 658 | |
| 659 | /* compareFunction() : |
| 660 | * Sort input in decreasing order when used with qsort() */ |
| 661 | int compareFunction(const void *a, const void *b) |
| 662 | { |
| 663 | double x = *(const double *)a; |
| 664 | double y = *(const double *)b; |
| 665 | if (x < y) |
| 666 | return 1; |
| 667 | else if (x > y) |
| 668 | return -1; |
| 669 | return 0; |
| 670 | } |
| 671 | |
| 672 | double aggregateData(double *data, size_t size, |
| 673 | metricAggregatePref_e metricAggregatePref) |
| 674 | { |
| 675 | qsort(data, size, sizeof(*data), compareFunction); |
| 676 | if (metricAggregatePref == fastest) |
| 677 | return data[0]; |
| 678 | else /* median */ |
| 679 | return (data[(size - 1) / 2] + data[size / 2]) / 2; |
| 680 | } |
| 681 | |
| 682 | static int benchMem(slice_collection_t dstBlocks, slice_collection_t srcBlocks, |
| 683 | ddict_collection_t ddictionaries, |
| 684 | cdict_collection_t cdictionaries, unsigned nbRounds, |
| 685 | int benchCompression, const char *exeName, |
| 686 | ZSTD_CCtx_params *cctxParams, |
| 687 | metricAggregatePref_e metricAggregatePref) |
| 688 | { |
| 689 | assert(dstBlocks.nbSlices == srcBlocks.nbSlices); |
| 690 | if (benchCompression) assert(cctxParams); |
| 691 | |
| 692 | unsigned const ms_per_round = RUN_TIME_DEFAULT_MS; |
| 693 | unsigned const total_time_ms = nbRounds * ms_per_round; |
| 694 | |
| 695 | double *const speedPerRound = (double *)malloc(nbRounds * sizeof(double)); |
| 696 | |
| 697 | BMK_timedFnState_t* const benchState = |
| 698 | BMK_createTimedFnState(total_time_ms, ms_per_round); |
| 699 | |
| 700 | decompressInstructions di = createDecompressInstructions(ddictionaries); |
| 701 | compressInstructions ci = |
| 702 | createCompressInstructions(cdictionaries, cctxParams); |
| 703 | void* payload = benchCompression ? (void*)&ci : (void*)&di; |
| 704 | BMK_benchParams_t const bp = { |
| 705 | .benchFn = benchCompression ? compress : decompress, |
| 706 | .benchPayload = payload, |
| 707 | .initFn = NULL, |
| 708 | .initPayload = NULL, |
| 709 | .errorFn = ZSTD_isError, |
| 710 | .blockCount = dstBlocks.nbSlices, |
| 711 | .srcBuffers = (const void* const*) srcBlocks.slicePtrs, |
| 712 | .srcSizes = srcBlocks.capacities, |
| 713 | .dstBuffers = dstBlocks.slicePtrs, |
| 714 | .dstCapacities = dstBlocks.capacities, |
| 715 | .blockResults = NULL |
| 716 | }; |
| 717 | |
| 718 | size_t roundNb = 0; |
| 719 | for (;;) { |
| 720 | BMK_runOutcome_t const outcome = BMK_benchTimedFn(benchState, bp); |
| 721 | CONTROL(BMK_isSuccessful_runOutcome(outcome)); |
| 722 | |
| 723 | BMK_runTime_t const result = BMK_extract_runTime(outcome); |
| 724 | double const dTime_ns = result.nanoSecPerRun; |
| 725 | double const dTime_sec = (double)dTime_ns / 1000000000; |
| 726 | size_t const srcSize = result.sumOfReturn; |
| 727 | double const speed_MBps = (double)srcSize / dTime_sec / (1 MB); |
| 728 | speedPerRound[roundNb] = speed_MBps; |
| 729 | if (benchCompression) |
| 730 | DISPLAY("Compression Speed : %.1f MB/s \r", speed_MBps); |
| 731 | else |
| 732 | DISPLAY("Decompression Speed : %.1f MB/s \r", speed_MBps); |
| 733 | |
| 734 | fflush(stdout); |
| 735 | if (BMK_isCompleted_TimedFn(benchState)) break; |
| 736 | roundNb++; |
| 737 | } |
| 738 | DISPLAY("\n"); |
| 739 | /* BMK_benchTimedFn may not run exactly nbRounds iterations */ |
| 740 | double speedAggregated = |
| 741 | aggregateData(speedPerRound, roundNb + 1, metricAggregatePref); |
| 742 | if (metricAggregatePref == fastest) |
| 743 | DISPLAY("Fastest Speed : %.1f MB/s \n", speedAggregated); |
| 744 | else |
| 745 | DISPLAY("Median Speed : %.1f MB/s \n", speedAggregated); |
| 746 | |
| 747 | char* csvFileName = malloc(strlen(exeName) + 5); |
| 748 | strcpy(csvFileName, exeName); |
| 749 | strcat(csvFileName, ".csv"); |
| 750 | FILE* csvFile = fopen(csvFileName, "r"); |
| 751 | if (!csvFile) { |
| 752 | csvFile = fopen(csvFileName, "wt"); |
| 753 | assert(csvFile); |
| 754 | fprintf(csvFile, "%s\n", exeName); |
| 755 | /* Print table headers */ |
| 756 | fprintf( |
| 757 | csvFile, |
| 758 | "Compression/Decompression,Level,nbDicts,dictAttachPref,metricAggregatePref,Speed\n"); |
| 759 | } else { |
| 760 | fclose(csvFile); |
| 761 | csvFile = fopen(csvFileName, "at"); |
| 762 | assert(csvFile); |
| 763 | } |
| 764 | |
| 765 | int cLevel = -1; |
| 766 | int dictAttachPref = -1; |
| 767 | if (benchCompression) { |
| 768 | ZSTD_CCtxParams_getParameter(cctxParams, ZSTD_c_compressionLevel, |
| 769 | &cLevel); |
| 770 | ZSTD_CCtxParams_getParameter(cctxParams, ZSTD_c_forceAttachDict, |
| 771 | &dictAttachPref); |
| 772 | } |
| 773 | fprintf(csvFile, "%s,%d,%ld,%d,%d,%.1f\n", |
| 774 | benchCompression ? "Compression" : "Decompression", cLevel, |
| 775 | benchCompression ? ci.nbDicts : di.nbDicts, dictAttachPref, |
| 776 | metricAggregatePref, speedAggregated); |
| 777 | fclose(csvFile); |
| 778 | free(csvFileName); |
| 779 | |
| 780 | freeDecompressInstructions(di); |
| 781 | freeCompressInstructions(ci); |
| 782 | BMK_freeTimedFnState(benchState); |
| 783 | |
| 784 | return 0; /* success */ |
| 785 | } |
| 786 | |
| 787 | |
| 788 | /*! bench() : |
| 789 | * fileName : file to load for benchmarking purpose |
| 790 | * dictionary : optional (can be NULL), file to load as dictionary, |
| 791 | * if none provided : will be calculated on the fly by the program. |
| 792 | * @return : 0 is success, 1+ otherwise */ |
| 793 | int bench(const char **fileNameTable, unsigned nbFiles, const char *dictionary, |
| 794 | size_t blockSize, int clevel, unsigned nbDictMax, unsigned nbBlocks, |
| 795 | unsigned nbRounds, int benchCompression, |
| 796 | ZSTD_dictContentType_e dictContentType, ZSTD_CCtx_params *cctxParams, |
| 797 | const char *exeName, metricAggregatePref_e metricAggregatePref) |
| 798 | { |
| 799 | int result = 0; |
| 800 | |
| 801 | DISPLAYLEVEL(3, "loading %u files... \n", nbFiles); |
| 802 | buffer_collection_t const srcs = createBufferCollection_fromFiles(fileNameTable, nbFiles); |
| 803 | CONTROL(srcs.buffer.ptr != NULL); |
| 804 | buffer_t srcBuffer = srcs.buffer; |
| 805 | size_t const srcSize = srcBuffer.size; |
| 806 | DISPLAYLEVEL(3, "created src buffer of size %.1f MB \n", |
| 807 | (double)srcSize / (1 MB)); |
| 808 | |
| 809 | slice_collection_t const srcSlices = splitSlices(srcs.slices, blockSize, nbBlocks); |
| 810 | nbBlocks = (unsigned)(srcSlices.nbSlices); |
| 811 | DISPLAYLEVEL(3, "split input into %u blocks ", nbBlocks); |
| 812 | if (blockSize) |
| 813 | DISPLAYLEVEL(3, "of max size %u bytes ", (unsigned)blockSize); |
| 814 | DISPLAYLEVEL(3, "\n"); |
| 815 | size_t const totalSrcSlicesSize = sliceCollection_totalCapacity(srcSlices); |
| 816 | |
| 817 | |
| 818 | size_t* const dstCapacities = malloc(nbBlocks * sizeof(*dstCapacities)); |
| 819 | CONTROL(dstCapacities != NULL); |
| 820 | size_t dstBufferCapacity = 0; |
| 821 | for (size_t bnb=0; bnb<nbBlocks; bnb++) { |
| 822 | dstCapacities[bnb] = ZSTD_compressBound(srcSlices.capacities[bnb]); |
| 823 | dstBufferCapacity += dstCapacities[bnb]; |
| 824 | } |
| 825 | |
| 826 | buffer_t dstBuffer = createBuffer(dstBufferCapacity); |
| 827 | CONTROL(dstBuffer.ptr != NULL); |
| 828 | |
| 829 | void** const sliceTable = malloc(nbBlocks * sizeof(*sliceTable)); |
| 830 | CONTROL(sliceTable != NULL); |
| 831 | |
| 832 | { char* const ptr = dstBuffer.ptr; |
| 833 | size_t pos = 0; |
| 834 | for (size_t snb=0; snb < nbBlocks; snb++) { |
| 835 | sliceTable[snb] = ptr + pos; |
| 836 | pos += dstCapacities[snb]; |
| 837 | } } |
| 838 | |
| 839 | slice_collection_t dstSlices; |
| 840 | dstSlices.capacities = dstCapacities; |
| 841 | dstSlices.slicePtrs = sliceTable; |
| 842 | dstSlices.nbSlices = nbBlocks; |
| 843 | |
| 844 | |
| 845 | /* dictionary determination */ |
| 846 | buffer_t const dictBuffer = createDictionaryBuffer(dictionary, |
| 847 | srcs.buffer.ptr, |
| 848 | srcSlices.capacities, srcSlices.nbSlices, |
| 849 | DICTSIZE); |
| 850 | CONTROL(dictBuffer.ptr != NULL); |
| 851 | |
| 852 | ZSTD_CDict* const cdict = ZSTD_createCDict_advanced2(dictBuffer.ptr, dictBuffer.size, DICT_LOAD_METHOD, dictContentType, cctxParams, ZSTD_defaultCMem); |
| 853 | CONTROL(cdict != NULL); |
| 854 | |
| 855 | size_t const cTotalSizeNoDict = compressBlocks(NULL, dstSlices, srcSlices, NULL, clevel); |
| 856 | CONTROL(cTotalSizeNoDict != 0); |
| 857 | DISPLAYLEVEL(3, "compressing at level %u without dictionary : Ratio=%.2f (%u bytes) \n", |
| 858 | clevel, |
| 859 | (double)totalSrcSlicesSize / (double)cTotalSizeNoDict, (unsigned)cTotalSizeNoDict); |
| 860 | |
| 861 | size_t* const cSizes = malloc(nbBlocks * sizeof(size_t)); |
| 862 | CONTROL(cSizes != NULL); |
| 863 | |
| 864 | size_t const cTotalSize = compressBlocks(cSizes, dstSlices, srcSlices, cdict, clevel); |
| 865 | CONTROL(cTotalSize != 0); |
| 866 | DISPLAYLEVEL(3, "compressed using a %u bytes dictionary : Ratio=%.2f (%u bytes) \n", |
| 867 | (unsigned)dictBuffer.size, |
| 868 | (double)totalSrcSlicesSize / (double)cTotalSize, (unsigned)cTotalSize); |
| 869 | |
| 870 | /* now dstSlices contain the real compressed size of each block, instead of the maximum capacity */ |
| 871 | shrinkSizes(dstSlices, cSizes); |
| 872 | |
| 873 | unsigned const nbDicts = nbDictMax ? nbDictMax : nbBlocks; |
| 874 | |
| 875 | cdict_collection_t const cdictionaries = createCDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts, dictContentType, cctxParams); |
| 876 | CONTROL(cdictionaries.cdicts != NULL); |
| 877 | |
| 878 | ddict_collection_t const ddictionaries = createDDictCollection(dictBuffer.ptr, dictBuffer.size, nbDicts); |
| 879 | CONTROL(ddictionaries.ddicts != NULL); |
| 880 | |
| 881 | if (benchCompression) { |
| 882 | size_t const dictMem = ZSTD_sizeof_CDict(cdictionaries.cdicts[0]); |
| 883 | size_t const allDictMem = dictMem * nbDicts; |
| 884 | DISPLAYLEVEL(3, "generating %u dictionaries, using %.1f MB of memory \n", |
| 885 | nbDicts, (double)allDictMem / (1 MB)); |
| 886 | |
| 887 | shuffleCDictionaries(cdictionaries); |
| 888 | |
| 889 | buffer_collection_t resultCollection = createBufferCollection_fromSliceCollection(srcSlices); |
| 890 | CONTROL(resultCollection.buffer.ptr != NULL); |
| 891 | |
| 892 | result = benchMem(dstSlices, resultCollection.slices, ddictionaries, |
| 893 | cdictionaries, nbRounds, benchCompression, exeName, |
| 894 | cctxParams, metricAggregatePref); |
| 895 | |
| 896 | freeBufferCollection(resultCollection); |
| 897 | } else { |
| 898 | size_t const dictMem = ZSTD_estimateDDictSize(dictBuffer.size, DICT_LOAD_METHOD); |
| 899 | size_t const allDictMem = dictMem * nbDicts; |
| 900 | DISPLAYLEVEL(3, "generating %u dictionaries, using %.1f MB of memory \n", |
| 901 | nbDicts, (double)allDictMem / (1 MB)); |
| 902 | |
| 903 | shuffleDDictionaries(ddictionaries); |
| 904 | |
| 905 | buffer_collection_t resultCollection = createBufferCollection_fromSliceCollectionSizes(srcSlices); |
| 906 | CONTROL(resultCollection.buffer.ptr != NULL); |
| 907 | |
| 908 | result = benchMem(resultCollection.slices, dstSlices, ddictionaries, |
| 909 | cdictionaries, nbRounds, benchCompression, exeName, |
| 910 | NULL, metricAggregatePref); |
| 911 | |
| 912 | freeBufferCollection(resultCollection); |
| 913 | } |
| 914 | |
| 915 | /* free all heap objects in reverse order */ |
| 916 | freeCDictCollection(cdictionaries); |
| 917 | freeDDictCollection(ddictionaries); |
| 918 | free(cSizes); |
| 919 | ZSTD_freeCDict(cdict); |
| 920 | freeBuffer(dictBuffer); |
| 921 | freeSliceCollection(dstSlices); |
| 922 | freeBuffer(dstBuffer); |
| 923 | freeSliceCollection(srcSlices); |
| 924 | freeBufferCollection(srcs); |
| 925 | |
| 926 | return result; |
| 927 | } |
| 928 | |
| 929 | |
| 930 | |
| 931 | /* --- Command Line --- */ |
| 932 | |
| 933 | /*! readU32FromChar() : |
| 934 | * @return : unsigned integer value read from input in `char` format. |
| 935 | * allows and interprets K, KB, KiB, M, MB and MiB suffix. |
| 936 | * Will also modify `*stringPtr`, advancing it to position where it stopped reading. |
| 937 | * Note : function will exit() program if digit sequence overflows */ |
| 938 | static unsigned readU32FromChar(const char** stringPtr) |
| 939 | { |
| 940 | unsigned result = 0; |
| 941 | while ((**stringPtr >='0') && (**stringPtr <='9')) { |
| 942 | unsigned const max = (((unsigned)(-1)) / 10) - 1; |
| 943 | assert(result <= max); /* check overflow */ |
| 944 | result *= 10, result += (unsigned)**stringPtr - '0', (*stringPtr)++ ; |
| 945 | } |
| 946 | if ((**stringPtr=='K') || (**stringPtr=='M')) { |
| 947 | unsigned const maxK = ((unsigned)(-1)) >> 10; |
| 948 | assert(result <= maxK); /* check overflow */ |
| 949 | result <<= 10; |
| 950 | if (**stringPtr=='M') { |
| 951 | assert(result <= maxK); /* check overflow */ |
| 952 | result <<= 10; |
| 953 | } |
| 954 | (*stringPtr)++; /* skip `K` or `M` */ |
| 955 | if (**stringPtr=='i') (*stringPtr)++; |
| 956 | if (**stringPtr=='B') (*stringPtr)++; |
| 957 | } |
| 958 | return result; |
| 959 | } |
| 960 | |
| 961 | /** longCommandWArg() : |
| 962 | * check if *stringPtr is the same as longCommand. |
| 963 | * If yes, @return 1 and advances *stringPtr to the position which immediately follows longCommand. |
| 964 | * @return 0 and doesn't modify *stringPtr otherwise. |
| 965 | */ |
| 966 | static int longCommandWArg(const char** stringPtr, const char* longCommand) |
| 967 | { |
| 968 | size_t const comSize = strlen(longCommand); |
| 969 | int const result = !strncmp(*stringPtr, longCommand, comSize); |
| 970 | if (result) *stringPtr += comSize; |
| 971 | return result; |
| 972 | } |
| 973 | |
| 974 | |
| 975 | int usage(const char* exeName) |
| 976 | { |
| 977 | DISPLAY (" \n"); |
| 978 | DISPLAY (" %s [Options] filename(s) \n", exeName); |
| 979 | DISPLAY (" \n"); |
| 980 | DISPLAY ("Options : \n"); |
| 981 | DISPLAY ("-z : benchmark compression (default) \n"); |
| 982 | DISPLAY ("-d : benchmark decompression \n"); |
| 983 | DISPLAY ("-r : recursively load all files in subdirectories (default: off) \n"); |
| 984 | DISPLAY ("-B# : split input into blocks of size # (default: no split) \n"); |
| 985 | DISPLAY ("-# : use compression level # (default: %u) \n", CLEVEL_DEFAULT); |
| 986 | DISPLAY ("-D # : use # as a dictionary (default: create one) \n"); |
| 987 | DISPLAY ("-i# : nb benchmark rounds (default: %u) \n", BENCH_TIME_DEFAULT_S); |
| 988 | DISPLAY ("-p# : print speed for all rounds 0=fastest 1=median (default: 0) \n"); |
| 989 | DISPLAY ("--nbBlocks=#: use # blocks for bench (default: one per file) \n"); |
| 990 | DISPLAY ("--nbDicts=# : create # dictionaries for bench (default: one per block) \n"); |
| 991 | DISPLAY ("-h : help (this text) \n"); |
| 992 | DISPLAY (" \n"); |
| 993 | DISPLAY ("Advanced Options (see zstd.h for documentation) : \n"); |
| 994 | DISPLAY ("--dedicated-dict-search\n"); |
| 995 | DISPLAY ("--dict-content-type=#\n"); |
| 996 | DISPLAY ("--dict-attach-pref=#\n"); |
| 997 | return 0; |
| 998 | } |
| 999 | |
| 1000 | int bad_usage(const char* exeName) |
| 1001 | { |
| 1002 | DISPLAY (" bad usage : \n"); |
| 1003 | usage(exeName); |
| 1004 | return 1; |
| 1005 | } |
| 1006 | |
| 1007 | int main (int argc, const char** argv) |
| 1008 | { |
| 1009 | int recursiveMode = 0; |
| 1010 | int benchCompression = 1; |
| 1011 | int dedicatedDictSearch = 0; |
| 1012 | unsigned nbRounds = BENCH_TIME_DEFAULT_S; |
| 1013 | const char* const exeName = argv[0]; |
| 1014 | |
| 1015 | if (argc < 2) return bad_usage(exeName); |
| 1016 | |
| 1017 | const char** nameTable = (const char**)malloc((size_t)argc * sizeof(const char*)); |
| 1018 | assert(nameTable != NULL); |
| 1019 | unsigned nameIdx = 0; |
| 1020 | |
| 1021 | const char* dictionary = NULL; |
| 1022 | int cLevel = CLEVEL_DEFAULT; |
| 1023 | size_t blockSize = BLOCKSIZE_DEFAULT; |
| 1024 | unsigned nbDicts = 0; /* determine nbDicts automatically: 1 dictionary per block */ |
| 1025 | unsigned nbBlocks = 0; /* determine nbBlocks automatically, from source and blockSize */ |
| 1026 | ZSTD_dictContentType_e dictContentType = ZSTD_dct_auto; |
| 1027 | ZSTD_dictAttachPref_e dictAttachPref = ZSTD_dictDefaultAttach; |
| 1028 | ZSTD_paramSwitch_e prefetchCDictTables = ZSTD_ps_auto; |
| 1029 | metricAggregatePref_e metricAggregatePref = fastest; |
| 1030 | |
| 1031 | for (int argNb = 1; argNb < argc ; argNb++) { |
| 1032 | const char* argument = argv[argNb]; |
| 1033 | if (!strcmp(argument, "-h")) { free(nameTable); return usage(exeName); } |
| 1034 | if (!strcmp(argument, "-d")) { benchCompression = 0; continue; } |
| 1035 | if (!strcmp(argument, "-z")) { benchCompression = 1; continue; } |
| 1036 | if (!strcmp(argument, "-r")) { recursiveMode = 1; continue; } |
| 1037 | if (!strcmp(argument, "-D")) { argNb++; assert(argNb < argc); dictionary = argv[argNb]; continue; } |
| 1038 | if (longCommandWArg(&argument, "-i")) { nbRounds = readU32FromChar(&argument); continue; } |
| 1039 | if (longCommandWArg(&argument, "-p")) { metricAggregatePref = (int)readU32FromChar(&argument); continue;} |
| 1040 | if (longCommandWArg(&argument, "--dictionary=")) { dictionary = argument; continue; } |
| 1041 | if (longCommandWArg(&argument, "-B")) { blockSize = readU32FromChar(&argument); continue; } |
| 1042 | if (longCommandWArg(&argument, "--blockSize=")) { blockSize = readU32FromChar(&argument); continue; } |
| 1043 | if (longCommandWArg(&argument, "--nbDicts=")) { nbDicts = readU32FromChar(&argument); continue; } |
| 1044 | if (longCommandWArg(&argument, "--nbBlocks=")) { nbBlocks = readU32FromChar(&argument); continue; } |
| 1045 | if (longCommandWArg(&argument, "--clevel=")) { cLevel = (int)readU32FromChar(&argument); continue; } |
| 1046 | if (longCommandWArg(&argument, "--dedicated-dict-search")) { dedicatedDictSearch = 1; continue; } |
| 1047 | if (longCommandWArg(&argument, "--dict-content-type=")) { dictContentType = (int)readU32FromChar(&argument); continue; } |
| 1048 | if (longCommandWArg(&argument, "--dict-attach-pref=")) { dictAttachPref = (int)readU32FromChar(&argument); continue; } |
| 1049 | if (longCommandWArg(&argument, "--prefetch-cdict-tables=")) { prefetchCDictTables = (int)readU32FromChar(&argument); continue; } |
| 1050 | if (longCommandWArg(&argument, "-")) { cLevel = (int)readU32FromChar(&argument); continue; } |
| 1051 | /* anything that's not a command is a filename */ |
| 1052 | nameTable[nameIdx++] = argument; |
| 1053 | } |
| 1054 | |
| 1055 | FileNamesTable* filenameTable; |
| 1056 | |
| 1057 | if (recursiveMode) { |
| 1058 | #ifndef UTIL_HAS_CREATEFILELIST |
| 1059 | assert(0); /* missing capability, do not run */ |
| 1060 | #endif |
| 1061 | filenameTable = UTIL_createExpandedFNT(nameTable, nameIdx, 1 /* follow_links */); |
| 1062 | } else { |
| 1063 | filenameTable = UTIL_assembleFileNamesTable(nameTable, nameIdx, NULL); |
| 1064 | nameTable = NULL; /* UTIL_createFileNamesTable() takes ownership of nameTable */ |
| 1065 | } |
| 1066 | |
| 1067 | ZSTD_CCtx_params* cctxParams = ZSTD_createCCtxParams(); |
| 1068 | ZSTD_CCtxParams_init(cctxParams, cLevel); |
| 1069 | ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_enableDedicatedDictSearch, dedicatedDictSearch); |
| 1070 | ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_nbWorkers, 0); |
| 1071 | ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_forceAttachDict, dictAttachPref); |
| 1072 | ZSTD_CCtxParams_setParameter(cctxParams, ZSTD_c_prefetchCDictTables, prefetchCDictTables); |
| 1073 | |
| 1074 | int result = |
| 1075 | bench(filenameTable->fileNames, (unsigned)filenameTable->tableSize, |
| 1076 | dictionary, blockSize, cLevel, nbDicts, nbBlocks, nbRounds, |
| 1077 | benchCompression, dictContentType, cctxParams, exeName, |
| 1078 | metricAggregatePref); |
| 1079 | |
| 1080 | UTIL_freeFileNamesTable(filenameTable); |
| 1081 | free(nameTable); |
| 1082 | ZSTD_freeCCtxParams(cctxParams); |
| 1083 | |
| 1084 | return result; |
| 1085 | } |