/* * Copyright (c) Meta Platforms, Inc. and affiliates. * All rights reserved. * * This source code is licensed under both the BSD-style license (found in the * LICENSE file in the root directory of this source tree) and the GPLv2 (found * in the COPYING file in the root directory of this source tree). */ /* ************************************* * Includes ***************************************/ #include "util.h" /* Compiler options, UTIL_GetFileSize, UTIL_sleep */ #include /* malloc, free */ #include /* memset */ #include /* fprintf, fopen, ftello64 */ #include /* clock_t, clock, CLOCKS_PER_SEC */ #include /* toupper */ #include /* errno */ #include "timefn.h" /* UTIL_time_t, UTIL_getTime, UTIL_clockSpanMicro, UTIL_waitForNextTick */ #include "mem.h" #define ZSTD_STATIC_LINKING_ONLY #include "zstd.h" #include "datagen.h" /* RDG_genBuffer */ #include "xxhash.h" #include "../zstd_zlibwrapper.h" /*-************************************ * Tuning parameters **************************************/ #ifndef ZSTDCLI_CLEVEL_DEFAULT # define ZSTDCLI_CLEVEL_DEFAULT 3 #endif /*-************************************ * Constants **************************************/ #define COMPRESSOR_NAME "Zstandard wrapper for zlib command line interface" #ifndef ZSTD_VERSION # define ZSTD_VERSION "v" ZSTD_VERSION_STRING #endif #define AUTHOR "Yann Collet" #define WELCOME_MESSAGE "*** %s %i-bits %s, by %s ***\n", COMPRESSOR_NAME, (int)(sizeof(size_t)*8), ZSTD_VERSION, AUTHOR #ifndef ZSTD_GIT_COMMIT # define ZSTD_GIT_COMMIT_STRING "" #else # define ZSTD_GIT_COMMIT_STRING ZSTD_EXPAND_AND_QUOTE(ZSTD_GIT_COMMIT) #endif #define NBLOOPS 3 #define TIMELOOP_MICROSEC 1*1000000ULL /* 1 second */ #define ACTIVEPERIOD_MICROSEC 70*1000000ULL /* 70 seconds */ #define COOLPERIOD_SEC 10 #define KB *(1 <<10) #define MB *(1 <<20) #define GB *(1U<<30) static const size_t maxMemory = (sizeof(size_t)==4) ? (2 GB - 64 MB) : (size_t)(1ULL << ((sizeof(size_t)*8)-31)); static U32 g_compressibilityDefault = 50; /* ************************************* * console display ***************************************/ #define DEFAULT_DISPLAY_LEVEL 2 #define DISPLAY(...) fprintf(displayOut, __VA_ARGS__) #define DISPLAYLEVEL(l, ...) if (g_displayLevel>=l) { DISPLAY(__VA_ARGS__); } static unsigned g_displayLevel = DEFAULT_DISPLAY_LEVEL; /* 0 : no display; 1: errors; 2 : + result + interaction + warnings; 3 : + progression; 4 : + information */ static FILE* displayOut; #define DISPLAYUPDATE(l, ...) if (g_displayLevel>=l) { \ if ((clock() - g_time > refreshRate) || (g_displayLevel>=4)) \ { g_time = clock(); DISPLAY(__VA_ARGS__); \ if (g_displayLevel>=4) fflush(displayOut); } } static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100; static clock_t g_time = 0; /* ************************************* * Exceptions ***************************************/ #ifndef DEBUG # define DEBUG 0 #endif #define DEBUGOUTPUT(...) { if (DEBUG) DISPLAY(__VA_ARGS__); } #define EXM_THROW(error, ...) \ { \ DEBUGOUTPUT("Error defined at %s, line %i : \n", __FILE__, __LINE__); \ DISPLAYLEVEL(1, "Error %i : ", error); \ DISPLAYLEVEL(1, __VA_ARGS__); \ DISPLAYLEVEL(1, "\n"); \ exit(error); \ } /* ************************************* * Benchmark Parameters ***************************************/ static unsigned g_nbIterations = NBLOOPS; static size_t g_blockSize = 0; int g_additionalParam = 0; static void BMK_setNotificationLevel(unsigned level) { g_displayLevel=level; } static void BMK_setAdditionalParam(int additionalParam) { g_additionalParam=additionalParam; } static void BMK_SetNbIterations(unsigned nbLoops) { g_nbIterations = nbLoops; DISPLAYLEVEL(3, "- test >= %u seconds per compression / decompression -\n", g_nbIterations); } static void BMK_SetBlockSize(size_t blockSize) { g_blockSize = blockSize; DISPLAYLEVEL(2, "using blocks of size %u KB \n", (unsigned)(blockSize>>10)); } /* ******************************************************** * Bench functions **********************************************************/ #undef MIN #undef MAX #define MIN(a,b) ((a)<(b) ? (a) : (b)) #define MAX(a,b) ((a)>(b) ? (a) : (b)) typedef struct { z_const char* srcPtr; size_t srcSize; char* cPtr; size_t cRoom; size_t cSize; char* resPtr; size_t resSize; } blockParam_t; typedef enum { BMK_ZSTD, BMK_ZSTD_STREAM, BMK_ZLIB, BMK_ZWRAP_ZLIB, BMK_ZWRAP_ZSTD, BMK_ZLIB_REUSE, BMK_ZWRAP_ZLIB_REUSE, BMK_ZWRAP_ZSTD_REUSE } BMK_compressor; static int BMK_benchMem(z_const void* srcBuffer, size_t srcSize, const char* displayName, int cLevel, const size_t* fileSizes, U32 nbFiles, const void* dictBuffer, size_t dictBufferSize, BMK_compressor compressor) { size_t const blockSize = (g_blockSize>=32 ? g_blockSize : srcSize) + (!srcSize) /* avoid div by 0 */ ; size_t const avgSize = MIN(g_blockSize, (srcSize / nbFiles)); U32 const maxNbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize) + nbFiles; blockParam_t* const blockTable = (blockParam_t*) malloc(maxNbBlocks * sizeof(blockParam_t)); size_t const maxCompressedSize = ZSTD_compressBound(srcSize) + (maxNbBlocks * 1024); /* add some room for safety */ void* const compressedBuffer = malloc(maxCompressedSize); void* const resultBuffer = malloc(srcSize); ZSTD_CCtx* const ctx = ZSTD_createCCtx(); ZSTD_DCtx* const dctx = ZSTD_createDCtx(); U32 nbBlocks; /* checks */ if (!compressedBuffer || !resultBuffer || !blockTable || !ctx || !dctx) EXM_THROW(31, "allocation error : not enough memory"); /* init */ if (strlen(displayName)>17) displayName += strlen(displayName)-17; /* can only display 17 characters */ /* Init blockTable data */ { z_const char* srcPtr = (z_const char*)srcBuffer; char* cPtr = (char*)compressedBuffer; char* resPtr = (char*)resultBuffer; U32 fileNb; for (nbBlocks=0, fileNb=0; fileNb ACTIVEPERIOD_MICROSEC) { DISPLAYLEVEL(2, "\rcooling down ... \r"); UTIL_sleep(COOLPERIOD_SEC); coolTime = UTIL_getTime(); } /* Compression */ DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->\r", marks[markNb], displayName, (unsigned)srcSize); if (!cCompleted) memset(compressedBuffer, 0xE5, maxCompressedSize); /* warm up and erase result buffer */ UTIL_sleepMilli(1); /* give processor time to other processes */ UTIL_waitForNextTick(); clockStart = UTIL_getTime(); if (!cCompleted) { /* still some time to do compression tests */ U32 nbLoops = 0; if (compressor == BMK_ZSTD) { ZSTD_parameters const zparams = ZSTD_getParams(cLevel, avgSize, dictBufferSize); ZSTD_customMem const cmem = { NULL, NULL, NULL }; ZSTD_CDict* const cdict = ZSTD_createCDict_advanced(dictBuffer, dictBufferSize, ZSTD_dlm_byRef, ZSTD_dct_auto, zparams.cParams, cmem); if (cdict==NULL) EXM_THROW(1, "ZSTD_createCDict_advanced() allocation failure"); do { U32 blockNb; size_t rSize; for (blockNb=0; blockNbmaxTime; } } cSize = 0; { U32 blockNb; for (blockNb=0; blockNb%10u (%5.3f),%6.1f MB/s\r", marks[markNb], displayName, (unsigned)srcSize, (unsigned)cSize, ratio, (double)srcSize / (double)fastestC ); (void)fastestD; (void)crcOrig; /* unused when decompression disabled */ #if 1 /* Decompression */ if (!dCompleted) memset(resultBuffer, 0xD6, srcSize); /* warm result buffer */ UTIL_sleepMilli(1); /* give processor time to other processes */ UTIL_waitForNextTick(); clockStart = UTIL_getTime(); if (!dCompleted) { U32 nbLoops = 0; if (compressor == BMK_ZSTD) { ZSTD_DDict* ddict = ZSTD_createDDict(dictBuffer, dictBufferSize); if (!ddict) EXM_THROW(2, "ZSTD_createDDict() allocation failure"); do { unsigned blockNb; for (blockNb=0; blockNbmaxTime; } } markNb = (markNb+1) % NB_MARKS; DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.3f),%6.1f MB/s ,%6.1f MB/s\r", marks[markNb], displayName, (unsigned)srcSize, (unsigned)cSize, ratio, (double)srcSize / (double)fastestC, (double)srcSize / (double)fastestD ); /* CRC Checking */ { U64 const crcCheck = XXH64(resultBuffer, srcSize, 0); if (crcOrig!=crcCheck) { size_t u; DISPLAY("!!! WARNING !!! %14s : Invalid Checksum : %x != %x \n", displayName, (unsigned)crcOrig, (unsigned)crcCheck); for (u=0; u u) break; bacc += blockTable[segNb].srcSize; } pos = (U32)(u - bacc); bNb = pos / (128 KB); DISPLAY("(block %u, sub %u, pos %u) \n", segNb, bNb, pos); break; } if (u==srcSize-1) { /* should never happen */ DISPLAY("no difference detected\n"); } } break; } } /* CRC Checking */ #endif } /* for (testNb = 1; testNb <= (g_nbIterations + !g_nbIterations); testNb++) */ if (g_displayLevel == 1) { double cSpeed = (double)srcSize / (double)fastestC; double dSpeed = (double)srcSize / (double)fastestD; if (g_additionalParam) DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s %s (param=%d)\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName, g_additionalParam); else DISPLAY("-%-3i%11i (%5.3f) %6.2f MB/s %6.1f MB/s %s\n", cLevel, (int)cSize, ratio, cSpeed, dSpeed, displayName); } DISPLAYLEVEL(2, "%2i#\n", cLevel); } /* Bench */ /* clean up */ free(blockTable); free(compressedBuffer); free(resultBuffer); ZSTD_freeCCtx(ctx); ZSTD_freeDCtx(dctx); return 0; } static size_t BMK_findMaxMem(U64 requiredMem) { size_t const step = 64 MB; BYTE* testmem = NULL; requiredMem = (((requiredMem >> 26) + 1) << 26); requiredMem += step; if (requiredMem > maxMemory) requiredMem = maxMemory; do { testmem = (BYTE*)malloc((size_t)requiredMem); requiredMem -= step; } while (!testmem && requiredMem); /* do not allocate zero bytes */ free(testmem); return (size_t)(requiredMem+1); /* avoid zero */ } static void BMK_benchCLevel(void* srcBuffer, size_t benchedSize, const char* displayName, int cLevel, int cLevelLast, const size_t* fileSizes, unsigned nbFiles, const void* dictBuffer, size_t dictBufferSize) { int l; const char* pch = strrchr(displayName, '\\'); /* Windows */ if (!pch) pch = strrchr(displayName, '/'); /* Linux */ if (pch) displayName = pch+1; SET_REALTIME_PRIORITY; if (g_displayLevel == 1 && !g_additionalParam) DISPLAY("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n", ZSTD_VERSION_STRING, ZSTD_GIT_COMMIT_STRING, (unsigned)benchedSize, g_nbIterations, (unsigned)(g_blockSize>>10)); if (cLevelLast < cLevel) cLevelLast = cLevel; DISPLAY("benchmarking zstd %s (using ZSTD_CStream)\n", ZSTD_VERSION_STRING); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZSTD_STREAM); } DISPLAY("benchmarking zstd %s (using ZSTD_CCtx)\n", ZSTD_VERSION_STRING); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZSTD); } DISPLAY("benchmarking zstd %s (using zlibWrapper)\n", ZSTD_VERSION_STRING); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZWRAP_ZSTD_REUSE); } DISPLAY("benchmarking zstd %s (zlibWrapper not reusing a context)\n", ZSTD_VERSION_STRING); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZWRAP_ZSTD); } if (cLevelLast > Z_BEST_COMPRESSION) cLevelLast = Z_BEST_COMPRESSION; DISPLAY("\n"); DISPLAY("benchmarking zlib %s\n", ZLIB_VERSION); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZLIB_REUSE); } DISPLAY("benchmarking zlib %s (zlib not reusing a context)\n", ZLIB_VERSION); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZLIB); } DISPLAY("benchmarking zlib %s (using zlibWrapper)\n", ZLIB_VERSION); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZWRAP_ZLIB_REUSE); } DISPLAY("benchmarking zlib %s (zlibWrapper not reusing a context)\n", ZLIB_VERSION); for (l=cLevel; l <= cLevelLast; l++) { BMK_benchMem(srcBuffer, benchedSize, displayName, l, fileSizes, nbFiles, dictBuffer, dictBufferSize, BMK_ZWRAP_ZLIB); } } /*! BMK_loadFiles() : Loads `buffer` with content of files listed within `fileNamesTable`. At most, fills `buffer` entirely */ static void BMK_loadFiles(void* buffer, size_t bufferSize, size_t* fileSizes, const char** fileNamesTable, unsigned nbFiles) { size_t pos = 0, totalSize = 0; unsigned n; for (n=0; n bufferSize-pos) fileSize = bufferSize-pos, nbFiles=n; /* buffer too small - stop after this file */ { size_t const readSize = fread(((char*)buffer)+pos, 1, (size_t)fileSize, f); if (readSize != (size_t)fileSize) EXM_THROW(11, "could not read %s", fileNamesTable[n]); pos += readSize; } fileSizes[n] = (size_t)fileSize; totalSize += (size_t)fileSize; fclose(f); } if (totalSize == 0) EXM_THROW(12, "no data to bench"); } static void BMK_benchFileTable(const char** fileNamesTable, unsigned nbFiles, const char* dictFileName, int cLevel, int cLevelLast) { void* srcBuffer; size_t benchedSize; void* dictBuffer = NULL; size_t dictBufferSize = 0; size_t* fileSizes = (size_t*)malloc(nbFiles * sizeof(size_t)); U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles); char mfName[20] = {0}; if (!fileSizes) EXM_THROW(12, "not enough memory for fileSizes"); /* Load dictionary */ if (dictFileName != NULL) { U64 const dictFileSize = UTIL_getFileSize(dictFileName); if (dictFileSize > 64 MB) EXM_THROW(10, "dictionary file %s too large", dictFileName); dictBufferSize = (size_t)dictFileSize; dictBuffer = malloc(dictBufferSize); if (dictBuffer==NULL) EXM_THROW(11, "not enough memory for dictionary (%u bytes)", (unsigned)dictBufferSize); BMK_loadFiles(dictBuffer, dictBufferSize, fileSizes, &dictFileName, 1); } /* Memory allocation & restrictions */ benchedSize = BMK_findMaxMem(totalSizeToLoad * 3) / 3; if ((U64)benchedSize > totalSizeToLoad) benchedSize = (size_t)totalSizeToLoad; if (benchedSize < totalSizeToLoad) DISPLAY("Not enough memory; testing %u MB only...\n", (unsigned)(benchedSize >> 20)); srcBuffer = malloc(benchedSize + !benchedSize); if (!srcBuffer) EXM_THROW(12, "not enough memory"); /* Load input buffer */ BMK_loadFiles(srcBuffer, benchedSize, fileSizes, fileNamesTable, nbFiles); /* Bench */ snprintf (mfName, sizeof(mfName), " %u files", nbFiles); { const char* displayName = (nbFiles > 1) ? mfName : fileNamesTable[0]; BMK_benchCLevel(srcBuffer, benchedSize, displayName, cLevel, cLevelLast, fileSizes, nbFiles, dictBuffer, dictBufferSize); } /* clean up */ free(srcBuffer); free(dictBuffer); free(fileSizes); } static void BMK_syntheticTest(int cLevel, int cLevelLast, double compressibility) { char name[20] = {0}; size_t benchedSize = 10000000; void* const srcBuffer = malloc(benchedSize); /* Memory allocation */ if (!srcBuffer) EXM_THROW(21, "not enough memory"); /* Fill input buffer */ RDG_genBuffer(srcBuffer, benchedSize, compressibility, 0.0, 0); /* Bench */ snprintf (name, sizeof(name), "Synthetic %2u%%", (unsigned)(compressibility*100)); BMK_benchCLevel(srcBuffer, benchedSize, name, cLevel, cLevelLast, &benchedSize, 1, NULL, 0); /* clean up */ free(srcBuffer); } static int BMK_benchFiles(const char** fileNamesTable, unsigned nbFiles, const char* dictFileName, int cLevel, int cLevelLast) { double const compressibility = (double)g_compressibilityDefault / 100; if (nbFiles == 0) BMK_syntheticTest(cLevel, cLevelLast, compressibility); else BMK_benchFileTable(fileNamesTable, nbFiles, dictFileName, cLevel, cLevelLast); return 0; } /*-************************************ * Command Line **************************************/ static int usage(const char* programName) { DISPLAY(WELCOME_MESSAGE); DISPLAY( "Usage :\n"); DISPLAY( " %s [args] [FILE(s)] [-o file]\n", programName); DISPLAY( "\n"); DISPLAY( "FILE : a filename\n"); DISPLAY( " with no FILE, or when FILE is - , read standard input\n"); DISPLAY( "Arguments :\n"); DISPLAY( " -D file: use `file` as Dictionary \n"); DISPLAY( " -h/-H : display help/long help and exit\n"); DISPLAY( " -V : display Version number and exit\n"); DISPLAY( " -v : verbose mode; specify multiple times to increase log level (default:%d)\n", DEFAULT_DISPLAY_LEVEL); DISPLAY( " -q : suppress warnings; specify twice to suppress errors too\n"); #ifdef UTIL_HAS_CREATEFILELIST DISPLAY( " -r : operate recursively on directories\n"); #endif DISPLAY( "\n"); DISPLAY( "Benchmark arguments :\n"); DISPLAY( " -b# : benchmark file(s), using # compression level (default : %d) \n", ZSTDCLI_CLEVEL_DEFAULT); DISPLAY( " -e# : test all compression levels from -bX to # (default: %d)\n", ZSTDCLI_CLEVEL_DEFAULT); DISPLAY( " -i# : minimum evaluation time in seconds (default : 3s)\n"); DISPLAY( " -B# : cut file into independent chunks of size # (default: no chunking)\n"); return 0; } static int badusage(const char* programName) { DISPLAYLEVEL(1, "Incorrect parameters\n"); if (g_displayLevel >= 1) usage(programName); return 1; } static void waitEnter(void) { int unused; DISPLAY("Press enter to continue...\n"); unused = getchar(); (void)unused; } /*! readU32FromChar() : @return : unsigned integer value reach from input in `char` format Will also modify `*stringPtr`, advancing it to position where it stopped reading. Note : this function can overflow if digit string > MAX_UINT */ static unsigned readU32FromChar(const char** stringPtr) { unsigned result = 0; while ((**stringPtr >='0') && (**stringPtr <='9')) result *= 10, result += (unsigned)(**stringPtr - '0'), (*stringPtr)++ ; return result; } #define CLEAN_RETURN(i) { operationResult = (i); goto _end; } int main(int argCount, char** argv) { int argNb, main_pause=0, nextEntryIsDictionary=0, operationResult=0, nextArgumentIsFile=0; int cLevel = ZSTDCLI_CLEVEL_DEFAULT; int cLevelLast = 1; unsigned recursive = 0; FileNamesTable* filenames = UTIL_allocateFileNamesTable((size_t)argCount); const char* programName = argv[0]; const char* dictFileName = NULL; char* dynNameSpace = NULL; /* init */ if (filenames==NULL) { DISPLAY("zstd: %s \n", strerror(errno)); exit(1); } displayOut = stderr; /* Pick out program name from path. Don't rely on stdlib because of conflicting behavior */ { size_t pos; for (pos = strlen(programName); pos > 0; pos--) { if (programName[pos] == '/') { pos++; break; } } programName += pos; } /* command switches */ for(argNb=1; argNb='0') && (*argument<='9')) { BMK_setAdditionalParam((int)readU32FromChar(&argument)); } else main_pause=1; break; /* unknown command */ default : CLEAN_RETURN(badusage(programName)); } } continue; } /* if (argument[0]=='-') */ } /* if (nextArgumentIsAFile==0) */ if (nextEntryIsDictionary) { nextEntryIsDictionary = 0; dictFileName = argument; continue; } /* add filename to list */ UTIL_refFilename(filenames, argument); } /* Welcome message (if verbose) */ DISPLAYLEVEL(3, WELCOME_MESSAGE); #ifdef UTIL_HAS_CREATEFILELIST if (recursive) { UTIL_expandFNT(&filenames, 1); } #endif BMK_setNotificationLevel(g_displayLevel); BMK_benchFiles(filenames->fileNames, (unsigned)filenames->tableSize, dictFileName, cLevel, cLevelLast); _end: if (main_pause) waitEnter(); free(dynNameSpace); UTIL_freeFileNamesTable(filenames); return operationResult; }