Merge remote-tracking branch 'refs/remotes/origin/dev' into repcodes
# Conflicts: # lib/zstd_compress.c # lib/zstd_decompress.c # lib/zstd_internal.h # lib/zstd_opt.h # programs/bench.c
This commit is contained in:
+177
-143
@@ -44,35 +44,58 @@
|
||||
/* *************************************
|
||||
* Includes
|
||||
***************************************/
|
||||
#define _POSIX_C_SOURCE 199309L /* before time.h */
|
||||
#define _POSIX_C_SOURCE 199309L /* before <time.h> - needed for nanosleep() */
|
||||
#include <stdlib.h> /* malloc, free */
|
||||
#include <string.h> /* memset */
|
||||
#include <stdio.h> /* fprintf, fopen, ftello64 */
|
||||
#include <sys/types.h> /* stat64 */
|
||||
#include <sys/stat.h> /* stat64 */
|
||||
#include <time.h> /* clock_t, clock, nanosleep, CLOCKS_PER_SEC */
|
||||
#include <time.h> /* clock_t, nanosleep, clock, CLOCKS_PER_SEC */
|
||||
|
||||
/* sleep : posix - windows - others */
|
||||
#if !defined(_WIN32) && (defined(__unix__) || defined(__unix) || (defined(__APPLE__) && defined(__MACH__)))
|
||||
# include <unistd.h>
|
||||
# include <sys/resource.h> /* setpriority */
|
||||
# define BMK_sleep(s) sleep(s)
|
||||
# define mili_sleep(mili) { struct timespec t; t.tv_sec=0; t.tv_nsec=mili*1000000L; nanosleep(&t, NULL); }
|
||||
# define mili_sleep(mili) { struct timespec t; t.tv_sec=0; t.tv_nsec=mili*1000000ULL; nanosleep(&t, NULL); }
|
||||
# define SET_HIGH_PRIORITY setpriority(PRIO_PROCESS, 0, -20)
|
||||
#elif defined(_WIN32)
|
||||
# include <windows.h>
|
||||
# define BMK_sleep(s) Sleep(1000*s)
|
||||
# define mili_sleep(mili) Sleep(mili)
|
||||
# define SET_HIGH_PRIORITY SetPriorityClass(GetCurrentProcess(), REALTIME_PRIORITY_CLASS)
|
||||
#else
|
||||
# define BMK_sleep(s) /* disabled */
|
||||
# define mili_sleep(mili) /* disabled */
|
||||
#error "disabled"
|
||||
# define SET_HIGH_PRIORITY /* disabled */
|
||||
#endif
|
||||
|
||||
#if !defined(_WIN32) && (defined(__unix__) || defined(__unix) || (defined(__APPLE__) && defined(__MACH__)))
|
||||
typedef clock_t BMK_time_t;
|
||||
# define BMK_initTimer(ticksPerSecond) ticksPerSecond=0
|
||||
# define BMK_getTime(x) x = clock()
|
||||
# define BMK_getSpanTimeMicro(ticksPerSecond, clockStart, clockEnd) (1000000ULL * (clockEnd - clockStart) / CLOCKS_PER_SEC)
|
||||
# define BMK_getSpanTimeNano(ticksPerSecond, clockStart, clockEnd) (1000000000ULL * (clockEnd - clockStart) / CLOCKS_PER_SEC)
|
||||
#elif defined(_WIN32)
|
||||
typedef LARGE_INTEGER BMK_time_t;
|
||||
# define BMK_initTimer(x) if (!QueryPerformanceFrequency(&x)) { fprintf(stderr, "ERROR: QueryPerformance not present\n"); }
|
||||
# define BMK_getTime(x) QueryPerformanceCounter(&x)
|
||||
# define BMK_getSpanTimeMicro(ticksPerSecond, clockStart, clockEnd) (1000000ULL*(clockEnd.QuadPart - clockStart.QuadPart)/ticksPerSecond.QuadPart)
|
||||
# define BMK_getSpanTimeNano(ticksPerSecond, clockStart, clockEnd) (1000000000ULL*(clockEnd.QuadPart - clockStart.QuadPart)/ticksPerSecond.QuadPart)
|
||||
#else
|
||||
typedef int BMK_time_t;
|
||||
# define BMK_initTimer(ticksPerSecond) ticksPerSecond=0
|
||||
# define BMK_getTimeMicro(clockStart) clockStart=1
|
||||
# define BMK_getSpanTimeMicro(ticksPerSecond, clockStart, clockEnd) (TIMELOOP_S*1000000ULL+clockEnd-clockStart)
|
||||
# define BMK_getSpanTimeNano(ticksPerSecond, clockStart, clockEnd) (TIMELOOP_S*1000000000ULL+clockEnd-clockStart)
|
||||
#endif
|
||||
|
||||
#include "mem.h"
|
||||
#include "zstd_static.h"
|
||||
#include "zstd_internal.h" /* ZSTD_setAdditionalParam */
|
||||
#include "zstd_internal.h" /* ZSTD_compressBegin_targetSrcSize */
|
||||
#include "datagen.h" /* RDG_genBuffer */
|
||||
#include "xxhash.h"
|
||||
#include "datagen.h" /* RDG_genBuffer */
|
||||
|
||||
|
||||
|
||||
/* *************************************
|
||||
@@ -138,9 +161,12 @@ static U32 g_displayLevel = 2; /* 0 : no display; 1: errors; 2 : + result
|
||||
***************************************/
|
||||
static U32 g_nbIterations = NBLOOPS;
|
||||
static size_t g_blockSize = 0;
|
||||
int g_additionalParam = 0;
|
||||
|
||||
void BMK_setNotificationLevel(unsigned level) { g_displayLevel=level; }
|
||||
|
||||
void BMK_setAdditionalParam(int additionalParam) { g_additionalParam=additionalParam; }
|
||||
|
||||
void BMK_SetNbIterations(unsigned nbLoops)
|
||||
{
|
||||
g_nbIterations = nbLoops;
|
||||
@@ -157,12 +183,16 @@ void BMK_SetBlockSize(size_t blockSize)
|
||||
/* ********************************************************
|
||||
* Private functions
|
||||
**********************************************************/
|
||||
static clock_t BMK_clockSpan( clock_t clockStart )
|
||||
/* returns time span in microseconds */
|
||||
static U64 BMK_clockSpan( BMK_time_t clockStart, BMK_time_t ticksPerSecond )
|
||||
{
|
||||
return clock() - clockStart; /* works even if overflow, span limited to <= ~30mn */
|
||||
BMK_time_t clockEnd;
|
||||
|
||||
(void)ticksPerSecond;
|
||||
BMK_getTime(clockEnd);
|
||||
return BMK_getSpanTimeMicro(ticksPerSecond, clockStart, clockEnd);
|
||||
}
|
||||
|
||||
|
||||
static U64 BMK_getFileSize(const char* infilename)
|
||||
{
|
||||
int r;
|
||||
@@ -207,44 +237,42 @@ typedef struct
|
||||
int kSlotNew = 0;
|
||||
|
||||
static int BMK_benchMem(const void* srcBuffer, size_t srcSize,
|
||||
const char* displayName, int cLevel, int additionalParam,
|
||||
const char* displayName, int cLevel,
|
||||
const size_t* fileSizes, U32 nbFiles,
|
||||
const void* dictBuffer, size_t dictBufferSize, benchResult_t *result)
|
||||
{
|
||||
const size_t blockSize = (g_blockSize ? g_blockSize : srcSize) + (!srcSize); /* avoid div by 0 */
|
||||
const U32 maxNbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize) + nbFiles;
|
||||
size_t largestBlockSize = 0;
|
||||
size_t const blockSize = (g_blockSize ? g_blockSize : srcSize) + (!srcSize); /* avoid div by 0 */
|
||||
U32 const maxNbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize) + nbFiles;
|
||||
blockParam_t* const blockTable = (blockParam_t*) malloc(maxNbBlocks * sizeof(blockParam_t));
|
||||
const size_t maxCompressedSize = ZSTD_compressBound(srcSize) + (maxNbBlocks * 1024); /* add some room for safety */
|
||||
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* refCtx = ZSTD_createCCtx();
|
||||
ZSTD_CCtx* ctx = ZSTD_createCCtx();
|
||||
ZSTD_DCtx* refDCtx = ZSTD_createDCtx();
|
||||
ZSTD_DCtx* dctx = ZSTD_createDCtx();
|
||||
U64 crcOrig = XXH64(srcBuffer, srcSize, 0);
|
||||
U32 nbBlocks = 0;
|
||||
size_t cSize = 0;
|
||||
|
||||
/* init */
|
||||
if (strlen(displayName)>17) displayName += strlen(displayName)-17; /* can only display 17 characters */
|
||||
U32 nbBlocks;
|
||||
BMK_time_t ticksPerSecond;
|
||||
|
||||
/* Memory allocation & restrictions */
|
||||
/* checks */
|
||||
if (!compressedBuffer || !resultBuffer || !blockTable || !refCtx || !ctx || !refDCtx || !dctx)
|
||||
EXM_THROW(31, "not enough memory");
|
||||
|
||||
/* init */
|
||||
if (strlen(displayName)>17) displayName += strlen(displayName)-17; /* can only display 17 characters */
|
||||
BMK_initTimer(ticksPerSecond);
|
||||
|
||||
/* Init blockTable data */
|
||||
{
|
||||
const char* srcPtr = (const char*)srcBuffer;
|
||||
{ const char* srcPtr = (const char*)srcBuffer;
|
||||
char* cPtr = (char*)compressedBuffer;
|
||||
char* resPtr = (char*)resultBuffer;
|
||||
U32 fileNb;
|
||||
for (fileNb=0; fileNb<nbFiles; fileNb++) {
|
||||
for (nbBlocks=0, fileNb=0; fileNb<nbFiles; fileNb++) {
|
||||
size_t remaining = fileSizes[fileNb];
|
||||
U32 const nbBlocksforThisFile = (U32)((remaining + (blockSize-1)) / blockSize);
|
||||
U32 const blockEnd = nbBlocks + nbBlocksforThisFile;
|
||||
for ( ; nbBlocks<blockEnd; nbBlocks++) {
|
||||
size_t thisBlockSize = MIN(remaining, blockSize);
|
||||
size_t const thisBlockSize = MIN(remaining, blockSize);
|
||||
blockTable[nbBlocks].srcPtr = srcPtr;
|
||||
blockTable[nbBlocks].cPtr = cPtr;
|
||||
blockTable[nbBlocks].resPtr = resPtr;
|
||||
@@ -254,137 +282,137 @@ static int BMK_benchMem(const void* srcBuffer, size_t srcSize,
|
||||
cPtr += blockTable[nbBlocks].cRoom;
|
||||
resPtr += thisBlockSize;
|
||||
remaining -= thisBlockSize;
|
||||
if (thisBlockSize > largestBlockSize) largestBlockSize = thisBlockSize;
|
||||
} } }
|
||||
|
||||
/* warmimg up memory */
|
||||
// int timeloop = additionalParam ? additionalParam : 2500;
|
||||
kSlotNew = additionalParam;
|
||||
RDG_genBuffer(compressedBuffer, maxCompressedSize, 0.10, 0.50, 1);
|
||||
|
||||
/* Bench */
|
||||
{
|
||||
U32 loopNb;
|
||||
double fastestC = 100000000., fastestD = 100000000.;
|
||||
double ratio = 0.;
|
||||
{ double fastestC = 100000000., fastestD = 100000000.;
|
||||
U64 const crcOrig = XXH64(srcBuffer, srcSize, 0);
|
||||
U64 crcCheck = 0;
|
||||
clock_t coolTime = clock();
|
||||
BMK_time_t coolTime;
|
||||
U32 testNb;
|
||||
size_t cSize = 0;
|
||||
double ratio = 0.;
|
||||
|
||||
BMK_getTime(coolTime);
|
||||
DISPLAYLEVEL(2, "\r%79s\r", "");
|
||||
for (loopNb = 1; loopNb <= (g_nbIterations + !g_nbIterations); loopNb++) {
|
||||
int nbLoops;
|
||||
U32 blockNb;
|
||||
clock_t clockStart, clockSpan;
|
||||
clock_t const clockLoop = g_nbIterations ? TIMELOOP_S * CLOCKS_PER_SEC : 10;
|
||||
for (testNb = 1; testNb <= (g_nbIterations + !g_nbIterations); testNb++) {
|
||||
BMK_time_t clockStart, clockEnd;
|
||||
U64 clockLoop = g_nbIterations ? TIMELOOP_S*1000000ULL : 10;
|
||||
|
||||
/* overheat protection */
|
||||
if (BMK_clockSpan(coolTime) > ACTIVEPERIOD_S * CLOCKS_PER_SEC) {
|
||||
if (BMK_clockSpan(coolTime, ticksPerSecond) > ACTIVEPERIOD_S*1000000ULL) {
|
||||
DISPLAY("\rcooling down ... \r");
|
||||
BMK_sleep(COOLPERIOD_S);
|
||||
coolTime = clock();
|
||||
BMK_getTime(coolTime);
|
||||
}
|
||||
|
||||
/* Compression */
|
||||
DISPLAYLEVEL(2, "%2i-%-17.17s :%10u ->\r", loopNb, displayName, (U32)srcSize);
|
||||
DISPLAYLEVEL(2, "%2i-%-17.17s :%10u ->\r", testNb, displayName, (U32)srcSize);
|
||||
memset(compressedBuffer, 0xE5, maxCompressedSize); /* warm up and erase result buffer */
|
||||
|
||||
nbLoops = 0;
|
||||
mili_sleep(1); // give processor to other processes
|
||||
clockStart = clock();
|
||||
while (clock() == clockStart);
|
||||
clockStart = clock();
|
||||
while (BMK_clockSpan(clockStart) < clockLoop) {
|
||||
ZSTD_compressBegin_advanced(refCtx, dictBuffer, dictBufferSize, ZSTD_getParams(cLevel, MAX(dictBufferSize, largestBlockSize)));
|
||||
for (blockNb=0; blockNb<nbBlocks; blockNb++) {
|
||||
size_t rSize = ZSTD_compress_usingPreparedCCtx(ctx, refCtx,
|
||||
blockTable[blockNb].cPtr, blockTable[blockNb].cRoom,
|
||||
blockTable[blockNb].srcPtr,blockTable[blockNb].srcSize);
|
||||
if (ZSTD_isError(rSize)) EXM_THROW(1, "ZSTD_compress_usingPreparedCCtx() failed : %s", ZSTD_getErrorName(rSize));
|
||||
blockTable[blockNb].cSize = rSize;
|
||||
}
|
||||
nbLoops++;
|
||||
}
|
||||
clockSpan = BMK_clockSpan(clockStart);
|
||||
mili_sleep(1); /* give processor time to other processes */
|
||||
BMK_getTime(clockStart);
|
||||
do { BMK_getTime(clockEnd); }
|
||||
while (BMK_getSpanTimeNano(ticksPerSecond, clockStart, clockEnd) == 0);
|
||||
BMK_getTime(clockStart);
|
||||
|
||||
{ U32 nbLoops;
|
||||
for (nbLoops = 0 ; BMK_clockSpan(clockStart, ticksPerSecond) < clockLoop ; nbLoops++) {
|
||||
U32 blockNb;
|
||||
ZSTD_compressBegin_targetSrcSize(refCtx, dictBuffer, dictBufferSize, blockSize, cLevel);
|
||||
// ZSTD_compressBegin_usingDict(refCtx, dictBuffer, dictBufferSize, cLevel);
|
||||
for (blockNb=0; blockNb<nbBlocks; blockNb++) {
|
||||
size_t const rSize = ZSTD_compress_usingPreparedCCtx(ctx, refCtx,
|
||||
blockTable[blockNb].cPtr, blockTable[blockNb].cRoom,
|
||||
blockTable[blockNb].srcPtr,blockTable[blockNb].srcSize);
|
||||
if (ZSTD_isError(rSize)) EXM_THROW(1, "ZSTD_compress_usingPreparedCCtx() failed : %s", ZSTD_getErrorName(rSize));
|
||||
blockTable[blockNb].cSize = rSize;
|
||||
} }
|
||||
{ U64 const clockSpan = BMK_clockSpan(clockStart, ticksPerSecond);
|
||||
if ((double)clockSpan < fastestC*nbLoops) fastestC = (double)clockSpan / nbLoops;
|
||||
} }
|
||||
|
||||
cSize = 0;
|
||||
for (blockNb=0; blockNb<nbBlocks; blockNb++)
|
||||
cSize += blockTable[blockNb].cSize;
|
||||
if ((double)clockSpan < fastestC*nbLoops) fastestC = (double)clockSpan / nbLoops;
|
||||
{ U32 blockNb; for (blockNb=0; blockNb<nbBlocks; blockNb++) cSize += blockTable[blockNb].cSize; }
|
||||
ratio = (double)srcSize / (double)cSize;
|
||||
DISPLAYLEVEL(2, "%2i-%-17.17s :%10u ->%10u (%5.3f),%6.1f MB/s\r",
|
||||
loopNb, displayName, (U32)srcSize, (U32)cSize, ratio,
|
||||
(double)srcSize / 1000000. / (fastestC / CLOCKS_PER_SEC) );
|
||||
testNb, displayName, (U32)srcSize, (U32)cSize, ratio,
|
||||
(double)srcSize / fastestC );
|
||||
|
||||
(void)fastestD; (void)crcOrig; /* unused when decompression disabled */
|
||||
#if 1
|
||||
/* Decompression */
|
||||
memset(resultBuffer, 0xD6, srcSize); /* warm result buffer */
|
||||
|
||||
nbLoops = 0;
|
||||
mili_sleep(1); // give processor to other processes
|
||||
clockStart = clock();
|
||||
while (clock() == clockStart);
|
||||
clockStart = clock();
|
||||
mili_sleep(1); /* give processor time to other processes */
|
||||
BMK_getTime(clockStart);
|
||||
do { BMK_getTime(clockEnd); }
|
||||
while (BMK_getSpanTimeNano(ticksPerSecond, clockStart, clockEnd) == 0);
|
||||
BMK_getTime(clockStart);
|
||||
|
||||
for ( ; BMK_clockSpan(clockStart) < clockLoop; nbLoops++) {
|
||||
ZSTD_decompressBegin_usingDict(refDCtx, dictBuffer, dictBufferSize);
|
||||
for (blockNb=0; blockNb<nbBlocks; blockNb++) {
|
||||
size_t regenSize = ZSTD_decompress_usingPreparedDCtx(dctx, refDCtx,
|
||||
blockTable[blockNb].resPtr, blockTable[blockNb].srcSize,
|
||||
blockTable[blockNb].cPtr, blockTable[blockNb].cSize);
|
||||
if (ZSTD_isError(regenSize)) {
|
||||
DISPLAY("ZSTD_decompress_usingPreparedDCtx() failed on block %u : %s",
|
||||
blockNb, ZSTD_getErrorName(regenSize));
|
||||
goto _findError;
|
||||
}
|
||||
blockTable[blockNb].resSize = regenSize;
|
||||
{ U32 nbLoops;
|
||||
for (nbLoops = 0 ; BMK_clockSpan(clockStart, ticksPerSecond) < clockLoop ; nbLoops++) {
|
||||
U32 blockNb;
|
||||
ZSTD_decompressBegin_usingDict(refDCtx, dictBuffer, dictBufferSize);
|
||||
for (blockNb=0; blockNb<nbBlocks; blockNb++) {
|
||||
size_t const regenSize = ZSTD_decompress_usingPreparedDCtx(dctx, refDCtx,
|
||||
blockTable[blockNb].resPtr, blockTable[blockNb].srcSize,
|
||||
blockTable[blockNb].cPtr, blockTable[blockNb].cSize);
|
||||
if (ZSTD_isError(regenSize)) {
|
||||
DISPLAY("ZSTD_decompress_usingPreparedDCtx() failed on block %u : %s \n",
|
||||
blockNb, ZSTD_getErrorName(regenSize));
|
||||
clockLoop = 0; /* force immediate test end */
|
||||
break;
|
||||
}
|
||||
blockTable[blockNb].resSize = regenSize;
|
||||
} }
|
||||
{ U64 const clockSpan = BMK_clockSpan(clockStart, ticksPerSecond);
|
||||
if ((double)clockSpan < fastestD*nbLoops) fastestD = (double)clockSpan / nbLoops;
|
||||
} }
|
||||
|
||||
clockSpan = BMK_clockSpan(clockStart);
|
||||
if ((double)clockSpan < fastestD*nbLoops) fastestD = (double)clockSpan / nbLoops;
|
||||
DISPLAYLEVEL(2, "%2i-%-17.17s :%10u ->%10u (%5.3f),%6.1f MB/s ,%6.1f MB/s\r",
|
||||
loopNb, displayName, (U32)srcSize, (U32)cSize, ratio,
|
||||
(double)srcSize / 1000000. / (fastestC / CLOCKS_PER_SEC),
|
||||
(double)srcSize / 1000000. / (fastestD / CLOCKS_PER_SEC) );
|
||||
testNb, displayName, (U32)srcSize, (U32)cSize, ratio,
|
||||
(double)srcSize / fastestC,
|
||||
(double)srcSize / fastestD );
|
||||
|
||||
/* CRC Checking */
|
||||
_findError:
|
||||
crcCheck = XXH64(resultBuffer, srcSize, 0);
|
||||
if (crcOrig!=crcCheck) {
|
||||
size_t u;
|
||||
DISPLAY("\n!!! WARNING !!! %14s : Invalid Checksum : %x != %x\n", displayName, (unsigned)crcOrig, (unsigned)crcCheck);
|
||||
for (u=0; u<srcSize; u++) {
|
||||
if (((const BYTE*)srcBuffer)[u] != ((const BYTE*)resultBuffer)[u]) {
|
||||
U32 segNb, bNb, pos;
|
||||
size_t bacc = 0;
|
||||
printf("Decoding error at pos %u ", (U32)u);
|
||||
for (segNb = 0; segNb < nbBlocks; segNb++) {
|
||||
if (bacc + blockTable[segNb].srcSize > u) break;
|
||||
bacc += blockTable[segNb].srcSize;
|
||||
{ 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<srcSize; u++) {
|
||||
if (((const BYTE*)srcBuffer)[u] != ((const BYTE*)resultBuffer)[u]) {
|
||||
U32 segNb, bNb, pos;
|
||||
size_t bacc = 0;
|
||||
DISPLAY("Decoding error at pos %u ", (U32)u);
|
||||
for (segNb = 0; segNb < nbBlocks; segNb++) {
|
||||
if (bacc + blockTable[segNb].srcSize > 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;
|
||||
}
|
||||
pos = (U32)(u - bacc);
|
||||
bNb = pos / (128 KB);
|
||||
printf("(block %u, sub %u, pos %u) \n", segNb, bNb, pos);
|
||||
break;
|
||||
}
|
||||
if (u==srcSize-1) { /* should never happen */
|
||||
printf("no difference detected\n");
|
||||
} }
|
||||
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 (crcOrig == crcCheck)
|
||||
{
|
||||
if (crcOrig == crcCheck) {
|
||||
result->ratio = ratio;
|
||||
result->cSize = cSize;
|
||||
result->cSpeed = (double)srcSize / 1000000. / (fastestC / CLOCKS_PER_SEC);
|
||||
result->dSpeed = (double)srcSize / 1000000. / (fastestD / CLOCKS_PER_SEC);
|
||||
DISPLAYLEVEL(2, "%2i-%-17.17s :%10i ->%10i (%5.3f),%6.1f MB/s ,%6.1f MB/s \n", cLevel, displayName, (int)srcSize, (int)cSize, ratio, result->cSpeed, result->dSpeed);
|
||||
result->cSpeed = (double)srcSize / fastestC;
|
||||
result->dSpeed = (double)srcSize / fastestD;
|
||||
}
|
||||
else
|
||||
DISPLAYLEVEL(2, "%2i-\n", cLevel);
|
||||
}
|
||||
DISPLAYLEVEL(2, "%2i#\n", cLevel);
|
||||
} /* Bench */
|
||||
|
||||
/* clean up */
|
||||
free(compressedBuffer);
|
||||
@@ -399,23 +427,24 @@ _findError:
|
||||
|
||||
static size_t BMK_findMaxMem(U64 requiredMem)
|
||||
{
|
||||
size_t step = 64 MB;
|
||||
size_t const step = 64 MB;
|
||||
BYTE* testmem = NULL;
|
||||
|
||||
requiredMem = (((requiredMem >> 26) + 1) << 26);
|
||||
requiredMem += 2 * step;
|
||||
requiredMem += step;
|
||||
if (requiredMem > maxMemory) requiredMem = maxMemory;
|
||||
|
||||
while (!testmem) {
|
||||
requiredMem -= step;
|
||||
do {
|
||||
testmem = (BYTE*)malloc((size_t)requiredMem);
|
||||
}
|
||||
requiredMem -= step;
|
||||
} while (!testmem);
|
||||
|
||||
free(testmem);
|
||||
return (size_t)(requiredMem - step);
|
||||
return (size_t)(requiredMem);
|
||||
}
|
||||
|
||||
static void BMK_benchCLevel(void* srcBuffer, size_t benchedSize,
|
||||
const char* displayName, int cLevel, int cLevelLast, int additionalParam,
|
||||
const char* displayName, int cLevel, int cLevelLast,
|
||||
const size_t* fileSizes, unsigned nbFiles,
|
||||
const void* dictBuffer, size_t dictBufferSize)
|
||||
{
|
||||
@@ -427,6 +456,8 @@ static void BMK_benchCLevel(void* srcBuffer, size_t benchedSize,
|
||||
setpriority(PRIO_PROCESS, 0, -20);
|
||||
#endif
|
||||
|
||||
SET_HIGH_PRIORITY;
|
||||
|
||||
const char* pch = strrchr(displayName, '\\'); /* Windows */
|
||||
if (!pch) pch = strrchr(displayName, '/'); /* Linux */
|
||||
if (pch) displayName = pch+1;
|
||||
@@ -434,19 +465,20 @@ static void BMK_benchCLevel(void* srcBuffer, size_t benchedSize,
|
||||
memset(&result, 0, sizeof(result));
|
||||
memset(&total, 0, sizeof(total));
|
||||
|
||||
// if (g_displayLevel == 1 && !additionalParam)
|
||||
// DISPLAY("bench %s: input %u bytes, %i iterations, %u KB blocks\n", ZSTD_VERSION, (U32)benchedSize, g_nbIterations, (U32)(g_blockSize>>10));
|
||||
kSlotNew = g_additionalParam;
|
||||
if (g_displayLevel == 1 && !g_additionalParam)
|
||||
DISPLAY("bench %s: input %u bytes, %i iterations, %u KB blocks\n", ZSTD_VERSION, (U32)benchedSize, g_nbIterations, (U32)(g_blockSize>>10));
|
||||
|
||||
if (cLevelLast < cLevel) cLevelLast = cLevel;
|
||||
|
||||
for (l=cLevel; l <= cLevelLast; l++) {
|
||||
BMK_benchMem(srcBuffer, benchedSize,
|
||||
displayName, l, additionalParam,
|
||||
displayName, l,
|
||||
fileSizes, nbFiles,
|
||||
dictBuffer, dictBufferSize, &result);
|
||||
if (g_displayLevel == 1) {
|
||||
if (1)// && additionalParam)
|
||||
DISPLAY("%-3i%11i (%5.3f) %6.1f MB/s %6.1f MB/s %s (kSlotNew=%d)\n", -l, (int)result.cSize, result.ratio, result.cSpeed, result.dSpeed, displayName, additionalParam);
|
||||
if (g_additionalParam)
|
||||
DISPLAY("%-3i%11i (%5.3f) %6.1f MB/s %6.1f MB/s %s (param=%d)\n", -l, (int)result.cSize, result.ratio, result.cSpeed, result.dSpeed, displayName, g_additionalParam);
|
||||
else
|
||||
DISPLAY("%-3i%11i (%5.3f) %6.1f MB/s %6.1f MB/s %s\n", -l, (int)result.cSize, result.ratio, result.cSpeed, result.dSpeed, displayName);
|
||||
total.cSize += result.cSize;
|
||||
@@ -474,30 +506,32 @@ static U64 BMK_getTotalFileSize(const char** fileNamesTable, unsigned nbFiles)
|
||||
return total;
|
||||
}
|
||||
|
||||
/*! 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 const nbFiles)
|
||||
const char** fileNamesTable, unsigned nbFiles)
|
||||
{
|
||||
size_t pos = 0;
|
||||
|
||||
unsigned n;
|
||||
for (n=0; n<nbFiles; n++) {
|
||||
size_t readSize;
|
||||
U64 fileSize = BMK_getFileSize(fileNamesTable[n]);
|
||||
FILE* f = fopen(fileNamesTable[n], "rb");
|
||||
FILE* const f = fopen(fileNamesTable[n], "rb");
|
||||
if (f==NULL) EXM_THROW(10, "impossible to open file %s", fileNamesTable[n]);
|
||||
DISPLAYLEVEL(2, "Loading %s... \r", fileNamesTable[n]);
|
||||
if (fileSize > bufferSize-pos) fileSize = bufferSize-pos;
|
||||
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;
|
||||
if (fileSize > 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;
|
||||
fclose(f);
|
||||
}
|
||||
}
|
||||
|
||||
static void BMK_benchFileTable(const char** fileNamesTable, unsigned nbFiles,
|
||||
const char* dictFileName, int cLevel, int cLevelLast, int additionalParam)
|
||||
const char* dictFileName, int cLevel, int cLevelLast)
|
||||
{
|
||||
void* srcBuffer;
|
||||
size_t benchedSize;
|
||||
@@ -537,7 +571,7 @@ static void BMK_benchFileTable(const char** fileNamesTable, unsigned nbFiles,
|
||||
else displayName = fileNamesTable[0];
|
||||
|
||||
BMK_benchCLevel(srcBuffer, benchedSize,
|
||||
displayName, cLevel, cLevelLast, additionalParam,
|
||||
displayName, cLevel, cLevelLast,
|
||||
fileSizes, nbFiles,
|
||||
dictBuffer, dictBufferSize);
|
||||
|
||||
@@ -548,7 +582,7 @@ static void BMK_benchFileTable(const char** fileNamesTable, unsigned nbFiles,
|
||||
}
|
||||
|
||||
|
||||
static void BMK_syntheticTest(int cLevel, int cLevelLast, int additionalParam, double compressibility)
|
||||
static void BMK_syntheticTest(int cLevel, int cLevelLast, double compressibility)
|
||||
{
|
||||
char name[20] = {0};
|
||||
size_t benchedSize = 10000000;
|
||||
@@ -562,7 +596,7 @@ static void BMK_syntheticTest(int cLevel, int cLevelLast, int additionalParam, d
|
||||
|
||||
/* Bench */
|
||||
snprintf (name, sizeof(name), "Synthetic %2u%%", (unsigned)(compressibility*100));
|
||||
BMK_benchCLevel(srcBuffer, benchedSize, name, cLevel, cLevelLast, additionalParam, &benchedSize, 1, NULL, 0);
|
||||
BMK_benchCLevel(srcBuffer, benchedSize, name, cLevel, cLevelLast, &benchedSize, 1, NULL, 0);
|
||||
|
||||
/* clean up */
|
||||
free(srcBuffer);
|
||||
@@ -570,14 +604,14 @@ static void BMK_syntheticTest(int cLevel, int cLevelLast, int additionalParam, d
|
||||
|
||||
|
||||
int BMK_benchFiles(const char** fileNamesTable, unsigned nbFiles,
|
||||
const char* dictFileName, int cLevel, int cLevelLast, int additionalParam)
|
||||
const char* dictFileName, int cLevel, int cLevelLast)
|
||||
{
|
||||
double const compressibility = (double)g_compressibilityDefault / 100;
|
||||
|
||||
if (nbFiles == 0)
|
||||
BMK_syntheticTest(cLevel, cLevelLast, additionalParam, compressibility);
|
||||
BMK_syntheticTest(cLevel, cLevelLast, compressibility);
|
||||
else
|
||||
BMK_benchFileTable(fileNamesTable, nbFiles, dictFileName, cLevel, cLevelLast, additionalParam);
|
||||
BMK_benchFileTable(fileNamesTable, nbFiles, dictFileName, cLevel, cLevelLast);
|
||||
return 0;
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user