ddidderr/zstd-rs
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

bench: changed creation/reset function to timedFnState

Browse Source 
for consistency
This commit is contained in:
Yann Collet
2018-08-21 18:19:27 -07:00
parent 1af27a7ed7
commit 77e805e3db
3 changed files with 287 additions and 277 deletions
Download Patch File Download Diff File Expand all files Collapse all files
programs/bench.c
+74 -74
View File
@@ -124,8 +124,8 @@ static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
* Benchmark Parameters
***************************************/
BMK_advancedParams_t BMK_initAdvancedParams(void) {
BMK_advancedParams_t res = {
BMK_advancedParams_t BMK_initAdvancedParams(void) {
BMK_advancedParams_t res = {
BMK_both, /* mode */
BMK_timeMode, /* loopMode */
BMK_TIMETEST_DEFAULT_S, /* nbSeconds */
@@ -133,7 +133,7 @@ BMK_advancedParams_t BMK_initAdvancedParams(void) {
0, /* nbWorkers */
0, /* realTime */
0, /* additionalParam */
0, /* ldmFlag */
0, /* ldmFlag */
0, /* ldmMinMatch */
0, /* ldmHashLog */
0, /* ldmBuckSizeLog */
@@ -168,8 +168,8 @@ struct BMK_timeState_t{
#define MIN(a,b) ((a) < (b) ? (a) : (b))
#define MAX(a,b) ((a) > (b) ? (a) : (b))
static void BMK_initCCtx(ZSTD_CCtx* ctx,
const void* dictBuffer, size_t dictBufferSize, int cLevel,
static void BMK_initCCtx(ZSTD_CCtx* ctx,
const void* dictBuffer, size_t dictBufferSize, int cLevel,
const ZSTD_compressionParameters* comprParams, const BMK_advancedParams_t* adv) {
ZSTD_CCtx_reset(ctx);
ZSTD_CCtx_resetParameters(ctx);
@@ -195,7 +195,7 @@ static void BMK_initCCtx(ZSTD_CCtx* ctx,
}
static void BMK_initDCtx(ZSTD_DCtx* dctx,
static void BMK_initDCtx(ZSTD_DCtx* dctx,
const void* dictBuffer, size_t dictBufferSize) {
ZSTD_DCtx_reset(dctx);
ZSTD_DCtx_loadDictionary(dctx, dictBuffer, dictBufferSize);
@@ -230,8 +230,8 @@ static size_t local_initDCtx(void* payload) {
/* additional argument is just the context */
static size_t local_defaultCompress(
const void* srcBuffer, size_t srcSize,
void* dstBuffer, size_t dstSize,
const void* srcBuffer, size_t srcSize,
void* dstBuffer, size_t dstSize,
void* addArgs) {
size_t moreToFlush = 1;
ZSTD_CCtx* ctx = (ZSTD_CCtx*)addArgs;
@@ -257,8 +257,8 @@ static size_t local_defaultCompress(
/* additional argument is just the context */
static size_t local_defaultDecompress(
const void* srcBuffer, size_t srcSize,
void* dstBuffer, size_t dstSize,
const void* srcBuffer, size_t srcSize,
void* dstBuffer, size_t dstSize,
void* addArgs) {
size_t moreToFlush = 1;
ZSTD_DCtx* dctx = (ZSTD_DCtx*)addArgs;
@@ -294,7 +294,7 @@ BMK_customReturn_t BMK_benchFunction(
BMK_initFn_t initFn, void* initPayload,
size_t blockCount,
const void* const * const srcBlockBuffers, const size_t* srcBlockSizes,
void* const * const dstBlockBuffers, const size_t* dstBlockCapacities, size_t* blockResult,
void* const * const dstBlockBuffers, const size_t* dstBlockCapacities, size_t* blockResult,
unsigned nbLoops) {
size_t dstSize = 0;
U64 totalTime;
@@ -312,15 +312,15 @@ BMK_customReturn_t BMK_benchFunction(
memset(dstBlockBuffers[i], 0xE5, dstBlockCapacities[i]); /* warm up and erase result buffer */
}
#if 0
/* based on testing these seem to lower accuracy of multiple calls of 1 nbLoops vs 1 call of multiple nbLoops
* (Makes former slower)
/* based on testing these seem to lower accuracy of multiple calls of 1 nbLoops vs 1 call of multiple nbLoops
* (Makes former slower)
*/
UTIL_sleepMilli(5); /* give processor time to other processes */
UTIL_waitForNextTick();
#endif
}
{
{
unsigned i, j;
clockStart = UTIL_getTime();
if(initFn != NULL) { initFn(initPayload); }
@@ -335,7 +335,7 @@ BMK_customReturn_t BMK_benchFunction(
if(blockResult != NULL) {
blockResult[j] = res;
}
}
}
}
}
totalTime = UTIL_clockSpanNano(clockStart);
@@ -345,27 +345,27 @@ BMK_customReturn_t BMK_benchFunction(
retval.result.nanoSecPerRun = totalTime / nbLoops;
retval.result.sumOfReturn = dstSize;
return retval;
}
}
#define MINUSABLETIME 500000000ULL /* 0.5 seconds in ns */
void BMK_resetTimeState(BMK_timedFnState_t* r, unsigned nbSeconds) {
void BMK_resetTimedFnState(BMK_timedFnState_t* r, unsigned nbSeconds) {
r->nbLoops = 1;
r->timeRemaining = (U64)nbSeconds * TIMELOOP_NANOSEC;
r->coolTime = UTIL_getTime();
r->fastestTime = (U64)(-1LL);
}
BMK_timedFnState_t* BMK_createTimeState(unsigned nbSeconds) {
BMK_timedFnState_t* BMK_createTimedFnState(unsigned nbSeconds) {
BMK_timedFnState_t* r = (BMK_timedFnState_t*)malloc(sizeof(struct BMK_timeState_t));
if(r == NULL) {
return r;
}
BMK_resetTimeState(r, nbSeconds);
BMK_resetTimedFnState(r, nbSeconds);
return r;
}
void BMK_freeTimeState(BMK_timedFnState_t* state) {
void BMK_freeTimedFnState(BMK_timedFnState_t* state) {
free(state);
}
@@ -376,7 +376,7 @@ BMK_customTimedReturn_t BMK_benchFunctionTimed(
BMK_initFn_t initFn, void* initPayload,
size_t blockCount,
const void* const* const srcBlockBuffers, const size_t* srcBlockSizes,
void * const * const dstBlockBuffers, const size_t * dstBlockCapacities, size_t* blockResults)
void * const * const dstBlockBuffers, const size_t * dstBlockCapacities, size_t* blockResults)
{
U64 fastest = cont->fastestTime;
int completed = 0;
@@ -402,7 +402,7 @@ BMK_customTimedReturn_t BMK_benchFunctionTimed(
{ U64 const loopDuration = r.result.result.nanoSecPerRun * cont->nbLoops;
r.completed = (cont->timeRemaining <= loopDuration);
cont->timeRemaining -= loopDuration;
if (loopDuration > (TIMELOOP_NANOSEC / 100)) {
if (loopDuration > (TIMELOOP_NANOSEC / 100)) {
fastest = MIN(fastest, r.result.result.nanoSecPerRun);
if(loopDuration >= MINUSABLETIME) {
r.result.result.nanoSecPerRun = fastest;
@@ -427,7 +427,7 @@ BMK_customTimedReturn_t BMK_benchFunctionTimed(
/* benchMem with no allocation */
static BMK_return_t BMK_benchMemAdvancedNoAlloc(
const void ** const srcPtrs, size_t* const srcSizes,
void** const cPtrs, size_t* const cCapacities, size_t* const cSizes,
void** const cPtrs, size_t* const cCapacities, size_t* const cSizes,
void** const resPtrs, size_t* const resSizes,
void** resultBufferPtr, void* compressedBuffer,
const size_t maxCompressedSize,
@@ -438,7 +438,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
const int cLevel, const ZSTD_compressionParameters* comprParams,
const void* dictBuffer, size_t dictBufferSize,
ZSTD_CCtx* ctx, ZSTD_DCtx* dctx,
int displayLevel, const char* displayName, const BMK_advancedParams_t* adv)
int displayLevel, const char* displayName, const BMK_advancedParams_t* adv)
{
size_t const blockSize = ((adv->blockSize>=32 && (adv->mode != BMK_decodeOnly)) ? adv->blockSize : srcSize) + (!srcSize); /* avoid div by 0 */
BMK_return_t results = { { 0, 0, 0, 0 }, 0 } ;
@@ -447,7 +447,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
double ratio = 0.;
U32 nbBlocks;
if(!ctx || !dctx)
if(!ctx || !dctx)
EXM_THROW(31, BMK_return_t, "error: passed in null context");
/* init */
@@ -466,16 +466,16 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
free(*resultBufferPtr);
*resultBufferPtr = malloc(decodedSize);
if (!(*resultBufferPtr)) {
EXM_THROW(33, BMK_return_t, "not enough memory");
EXM_THROW(33, BMK_return_t, "not enough memory");
}
if (totalDSize64 > decodedSize) {
free(*resultBufferPtr);
free(*resultBufferPtr);
EXM_THROW(32, BMK_return_t, "original size is too large"); /* size_t overflow */
}
cSize = srcSize;
srcSize = decodedSize;
ratio = (double)srcSize / (double)cSize;
}
}
}
/* Init data blocks */
@@ -499,8 +499,8 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
cPtr += cCapacities[nbBlocks];
resPtr += thisBlockSize;
remaining -= thisBlockSize;
}
}
}
}
}
/* warmimg up memory */
@@ -511,7 +511,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
}
/* Bench */
{
{
U64 const crcOrig = (adv->mode == BMK_decodeOnly) ? 0 : XXH64(srcBuffer, srcSize, 0);
# define NB_MARKS 4
const char* const marks[NB_MARKS] = { " |", " /", " =", "\\" };
@@ -545,7 +545,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
intermediateResultDecompress.completed = 0;
}
while(!(intermediateResultCompress.completed && intermediateResultDecompress.completed)) {
if(!intermediateResultCompress.completed) {
if(!intermediateResultCompress.completed) {
intermediateResultCompress = BMK_benchFunctionTimed(timeStateCompress, &local_defaultCompress, (void*)ctx, &local_initCCtx, (void*)&cctxprep,
nbBlocks, srcPtrs, srcSizes, cPtrs, cCapacities, cSizes);
if(intermediateResultCompress.result.error) {
@@ -553,7 +553,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
return results;
}
ratio = (double)(srcSize / intermediateResultCompress.result.result.sumOfReturn);
{
{
int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
results.result.cSpeed = (srcSize * TIMELOOP_NANOSEC / intermediateResultCompress.result.result.nanoSecPerRun);
cSize = intermediateResultCompress.result.result.sumOfReturn;
@@ -574,8 +574,8 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
results.error = intermediateResultDecompress.result.error;
return results;
}
{
{
int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
results.result.dSpeed = (srcSize * TIMELOOP_NANOSEC/ intermediateResultDecompress.result.result.nanoSecPerRun);
markNb = (markNb+1) % NB_MARKS;
@@ -592,7 +592,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
if(adv->mode != BMK_decodeOnly) {
BMK_customReturn_t compressionResults = BMK_benchFunction(&local_defaultCompress, (void*)ctx, &local_initCCtx, (void*)&cctxprep,
nbBlocks, srcPtrs, srcSizes, cPtrs, cCapacities, cSizes, adv->nbSeconds);
nbBlocks, srcPtrs, srcSizes, cPtrs, cCapacities, cSizes, adv->nbSeconds);
if(compressionResults.error) {
results.error = compressionResults.error;
return results;
@@ -605,7 +605,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
}
results.result.cSize = compressionResults.result.sumOfReturn;
{
{
int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
cSize = compressionResults.result.sumOfReturn;
results.result.cSize = cSize;
@@ -621,7 +621,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
BMK_customReturn_t decompressionResults = BMK_benchFunction(
&local_defaultDecompress, (void*)(dctx),
&local_initDCtx, (void*)&dctxprep, nbBlocks,
(const void* const*)cPtrs, cSizes, resPtrs, resSizes, NULL,
(const void* const*)cPtrs, cSizes, resPtrs, resSizes, NULL,
adv->nbSeconds);
if(decompressionResults.error) {
results.error = decompressionResults.error;
@@ -634,7 +634,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
results.result.dSpeed = srcSize * TIMELOOP_NANOSEC / decompressionResults.result.nanoSecPerRun;
}
{
{
int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
markNb = (markNb+1) % NB_MARKS;
DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s ,%6.1f MB/s \r",
@@ -683,9 +683,9 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
}
if (u==srcSize-1) { /* should never happen */
DISPLAY("no difference detected\n");
}
}
}
}
}
} /* CRC Checking */
if (displayLevel == 1) { /* hidden display mode -q, used by python speed benchmark */
@@ -705,7 +705,7 @@ static BMK_return_t BMK_benchMemAdvancedNoAlloc(
}
BMK_return_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
void* dstBuffer, size_t dstCapacity,
void* dstBuffer, size_t dstCapacity,
const size_t* fileSizes, unsigned nbFiles,
const int cLevel, const ZSTD_compressionParameters* comprParams,
const void* dictBuffer, size_t dictBufferSize,
@@ -727,8 +727,8 @@ BMK_return_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
void ** const resPtrs = (void**)malloc(maxNbBlocks * sizeof(void*));
size_t* const resSizes = (size_t*)malloc(maxNbBlocks * sizeof(size_t));
BMK_timedFnState_t* timeStateCompress = BMK_createTimeState(adv->nbSeconds);
BMK_timedFnState_t* timeStateDecompress = BMK_createTimeState(adv->nbSeconds);
BMK_timedFnState_t* timeStateCompress = BMK_createTimedFnState(adv->nbSeconds);
BMK_timedFnState_t* timeStateDecompress = BMK_createTimedFnState(adv->nbSeconds);
ZSTD_CCtx* ctx = ZSTD_createCCtx();
ZSTD_DCtx* dctx = ZSTD_createDCtx();
@@ -744,8 +744,8 @@ BMK_return_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
void* resultBuffer = srcSize ? malloc(srcSize) : NULL;
int allocationincomplete = !srcPtrs || !srcSizes || !cPtrs ||
!cSizes || !cCapacities || !resPtrs || !resSizes ||
int allocationincomplete = !srcPtrs || !srcSizes || !cPtrs ||
!cSizes || !cCapacities || !resPtrs || !resSizes ||
!timeStateCompress || !timeStateDecompress || !compressedBuffer || !resultBuffer;
@@ -756,20 +756,20 @@ BMK_return_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
srcBuffer, srcSize, fileSizes, nbFiles, cLevel, comprParams,
dictBuffer, dictBufferSize, ctx, dctx, displayLevel, displayName, adv);
}
/* clean up */
BMK_freeTimeState(timeStateCompress);
BMK_freeTimeState(timeStateDecompress);
BMK_freeTimedFnState(timeStateCompress);
BMK_freeTimedFnState(timeStateDecompress);
ZSTD_freeCCtx(ctx);
ZSTD_freeDCtx(dctx);
free(internalDstBuffer);
free(resultBuffer);
free((void*)srcPtrs);
free(srcSizes);
free(cPtrs);
free((void*)srcPtrs);
free(srcSizes);
free(cPtrs);
free(cSizes);
free(cCapacities);
free(resPtrs);
@@ -778,9 +778,9 @@ BMK_return_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
if(allocationincomplete) {
EXM_THROW(31, BMK_return_t, "allocation error : not enough memory");
}
if(parametersConflict) {
EXM_THROW(32, BMK_return_t, "Conflicting input results");
EXM_THROW(32, BMK_return_t, "Conflicting input results");
}
return results;
}
@@ -840,11 +840,11 @@ static BMK_return_t BMK_benchCLevel(const void* srcBuffer, size_t benchedSize,
if (displayLevel == 1 && !adv->additionalParam)
DISPLAY("bench %s %s: input %u bytes, %u seconds, %u KB blocks\n", ZSTD_VERSION_STRING, ZSTD_GIT_COMMIT_STRING, (U32)benchedSize, adv->nbSeconds, (U32)(adv->blockSize>>10));
res = BMK_benchMemAdvanced(srcBuffer, benchedSize,
NULL, 0,
fileSizes, nbFiles,
cLevel, comprParams,
dictBuffer, dictBufferSize,
res = BMK_benchMemAdvanced(srcBuffer, benchedSize,
NULL, 0,
fileSizes, nbFiles,
cLevel, comprParams,
dictBuffer, dictBufferSize,
displayLevel, displayName, adv);
return res;
@@ -855,7 +855,7 @@ static BMK_return_t BMK_benchCLevel(const void* srcBuffer, size_t benchedSize,
* Loads `buffer` with content of files listed within `fileNamesTable`.
* At most, fills `buffer` entirely. */
static int BMK_loadFiles(void* buffer, size_t bufferSize,
size_t* fileSizes, const char* const * const fileNamesTable,
size_t* fileSizes, const char* const * const fileNamesTable,
unsigned nbFiles, int displayLevel)
{
size_t pos = 0, totalSize = 0;
@@ -890,8 +890,8 @@ static int BMK_loadFiles(void* buffer, size_t bufferSize,
}
BMK_return_t BMK_benchFilesAdvanced(const char* const * const fileNamesTable, unsigned const nbFiles,
const char* const dictFileName, int const cLevel,
const ZSTD_compressionParameters* const compressionParams,
const char* const dictFileName, int const cLevel,
const ZSTD_compressionParameters* const compressionParams,
int displayLevel, const BMK_advancedParams_t * const adv)
{
void* srcBuffer = NULL;
@@ -960,13 +960,13 @@ BMK_return_t BMK_benchFilesAdvanced(const char* const * const fileNamesTable, un
{
char mfName[20] = {0};
snprintf (mfName, sizeof(mfName), " %u files", nbFiles);
{
{
const char* const displayName = (nbFiles > 1) ? mfName : fileNamesTable[0];
res = BMK_benchCLevel(srcBuffer, benchedSize,
fileSizes, nbFiles,
res = BMK_benchCLevel(srcBuffer, benchedSize,
fileSizes, nbFiles,
cLevel, compressionParams,
dictBuffer, dictBufferSize,
displayLevel, displayName,
dictBuffer, dictBufferSize,
displayLevel, displayName,
adv);
} }
@@ -1000,10 +1000,10 @@ BMK_return_t BMK_syntheticTest(int cLevel, double compressibility,
/* Bench */
snprintf (name, sizeof(name), "Synthetic %2u%%", (unsigned)(compressibility*100));
res = BMK_benchCLevel(srcBuffer, benchedSize,
&benchedSize, 1,
cLevel, compressionParams,
NULL, 0,
res = BMK_benchCLevel(srcBuffer, benchedSize,
&benchedSize, 1,
cLevel, compressionParams,
NULL, 0,
displayLevel, name, adv);
/* clean up */
@@ -1013,8 +1013,8 @@ BMK_return_t BMK_syntheticTest(int cLevel, double compressibility,
}
BMK_return_t BMK_benchFiles(const char* const * const fileNamesTable, unsigned const nbFiles,
const char* const dictFileName,
int const cLevel, const ZSTD_compressionParameters* const compressionParams,
const char* const dictFileName,
int const cLevel, const ZSTD_compressionParameters* const compressionParams,
int displayLevel) {
const BMK_advancedParams_t adv = BMK_initAdvancedParams();
return BMK_benchFilesAdvanced(fileNamesTable, nbFiles, dictFileName, cLevel, compressionParams, displayLevel, &adv);
programs/bench.h
+96 -86
View File
@@ -19,11 +19,12 @@ extern "C" {
#define ZSTD_STATIC_LINKING_ONLY /* ZSTD_compressionParameters */
#include "zstd.h" /* ZSTD_compressionParameters */
/* Creates a struct of type typeName with an int type .error field
* and a .result field of some baseType. Functions with return
* typeName pass a successful result with .error = 0 and .result
* with the intended result, while returning an error will result
* in .error != 0.
/* Creates a struct type typeName, featuring:
* - an .error field of type int
* - a .result field of some baseType.
* Functions with return type typeName
* will either be successful, with .error = 0, providing a valid .result,
* or return an error, with .error != 0, in which case .result is invalid.
*/
#define ERROR_STRUCT(baseType, typeName) typedef struct { \
baseType result; \
@@ -32,36 +33,37 @@ extern "C" {
typedef struct {
size_t cSize;
U64 cSpeed; /* bytes / sec */
U64 dSpeed;
size_t cMem;
unsigned long long cSpeed; /* bytes / sec */
unsigned long long dSpeed;
size_t cMem; /* ? what does it reports ? */
} BMK_result_t;
ERROR_STRUCT(BMK_result_t, BMK_return_t);
/* called in cli */
/* Loads files in fileNamesTable into memory, as well as a dictionary
* from dictFileName, and then uses benchMem */
/* fileNamesTable - name of files to benchmark
* nbFiles - number of files (size of fileNamesTable), must be > 0
* dictFileName - name of dictionary file to load
* cLevel - compression level to benchmark, errors if invalid
* compressionParams - basic compression Parameters
* displayLevel - what gets printed
* 0 : no display;
* 1 : errors;
* 2 : + result + interaction + warnings;
* 3 : + progression;
/*! BMK_benchFiles() -- called by zstdcli */
/* Loads files from fileNamesTable into memory,
* loads dictionary from dictFileName,
* then uses benchMem().
* fileNamesTable - name of files to benchmark
* nbFiles - number of files (size of fileNamesTable), must be > 0 (what happens if not ?)
* dictFileName - name of dictionary file to load
* cLevel - compression level to benchmark, errors if invalid
* compressionParams - advanced compression Parameters
* displayLevel - what gets printed
* 0 : no display;
* 1 : errors;
* 2 : + result + interaction + warnings;
* 3 : + progression;
* 4 : + information
* return
* @return
* .error will give a nonzero error value if an error has occured
* .result - if .error = 0, .result will return the time taken to compression speed
* (.cSpeed), decompression speed (.dSpeed), and compressed size (.cSize) of the original
* file
* .result - only valid if .error = 0,
* .result will return compression speed (.cSpeed),
* decompression speed (.dSpeed), and compressed size (.cSize).
*/
BMK_return_t BMK_benchFiles(const char* const * const fileNamesTable, unsigned const nbFiles,
const char* const dictFileName,
int const cLevel, const ZSTD_compressionParameters* const compressionParams,
BMK_return_t BMK_benchFiles(const char* const * fileNamesTable, unsigned nbFiles,
const char* dictFileName,
int cLevel, const ZSTD_compressionParameters* compressionParams,
int displayLevel);
typedef enum {
@@ -85,7 +87,7 @@ typedef struct {
int additionalParam; /* used by python speed benchmark */
unsigned ldmFlag; /* enables long distance matching */
unsigned ldmMinMatch; /* below: parameters for long distance matching, see zstd.1.md for meaning */
unsigned ldmHashLog;
unsigned ldmHashLog;
unsigned ldmBucketSizeLog;
unsigned ldmHashEveryLog;
} BMK_advancedParams_t;
@@ -93,67 +95,75 @@ typedef struct {
/* returns default parameters used by nonAdvanced functions */
BMK_advancedParams_t BMK_initAdvancedParams(void);
/* See benchFiles for normal parameter uses and return, see advancedParams_t for adv */
BMK_return_t BMK_benchFilesAdvanced(const char* const * const fileNamesTable, unsigned const nbFiles,
const char* const dictFileName,
int const cLevel, const ZSTD_compressionParameters* const compressionParams,
int displayLevel, const BMK_advancedParams_t* const adv);
/*! BMK_benchFilesAdvanced():
* Same as BMK_benchFiles(),
* with more controls, provided through advancedParams_t structure */
BMK_return_t BMK_benchFilesAdvanced(const char* const * fileNamesTable, unsigned nbFiles,
const char* dictFileName,
int cLevel, const ZSTD_compressionParameters* compressionParams,
int displayLevel, const BMK_advancedParams_t* adv);
/* called in cli */
/* Generates a sample with datagen with the compressibility argument*/
/* cLevel - compression level to benchmark, errors if invalid
* compressibility - determines compressibility of sample
* compressionParams - basic compression Parameters
* displayLevel - see benchFiles
* adv - see advanced_Params_t
* return
/*! BMK_syntheticTest() -- called from zstdcli */
/* Generates a sample with datagen, using compressibility argument */
/* cLevel - compression level to benchmark, errors if invalid
* compressibility - determines compressibility of sample
* compressionParams - basic compression Parameters
* displayLevel - see benchFiles
* adv - see advanced_Params_t
* @return:
* .error will give a nonzero error value if an error has occured
* .result - if .error = 0, .result will return the time taken to compression speed
* (.cSpeed), decompression speed (.dSpeed), and compressed size (.cSize) of the original
* file
* .result - only valid if .error = 0,
* .result will return the compression speed (.cSpeed),
* decompression speed (.dSpeed), and compressed size (.cSize).
*/
BMK_return_t BMK_syntheticTest(int cLevel, double compressibility,
const ZSTD_compressionParameters* compressionParams,
int displayLevel, const BMK_advancedParams_t * const adv);
/* basic benchmarking function, called in paramgrill
* applies ZSTD_compress_generic() and ZSTD_decompress_generic() on data in srcBuffer
* with specific compression parameters specified by other arguments using benchFunction
* (cLevel, comprParams + adv in advanced Mode) */
/* srcBuffer - data source, expected to be valid compressed data if in Decode Only Mode
* srcSize - size of data in srcBuffer
* cLevel - compression level
* comprParams - basic compression parameters
* dictBuffer - a dictionary if used, null otherwise
* dictBufferSize - size of dictBuffer, 0 otherwise
* diplayLevel - see BMK_benchFiles
* displayName - name used by display
* return
/** BMK_benchMem() -- core benchmarking function, called in paramgrill
* applies ZSTD_compress_generic() and ZSTD_decompress_generic() on data in srcBuffer
* with specific compression parameters provided by other arguments using benchFunction
* (cLevel, comprParams + adv in advanced Mode) */
/* srcBuffer - data source, expected to be valid compressed data if in Decode Only Mode
* srcSize - size of data in srcBuffer
* fileSizes - srcBuffer is considered cut into 1+ segments, to compress separately.
* note : sum(fileSizes) must be == srcSize. (<== ensure it's properly checked)
* nbFiles - nb of segments
* cLevel - compression level
* comprParams - basic compression parameters
* dictBuffer - a dictionary if used, null otherwise
* dictBufferSize - size of dictBuffer, 0 otherwise
* diplayLevel - see BMK_benchFiles
* displayName - name used by display
* @return
* .error will give a nonzero value if an error has occured
* .result - if .error = 0, will give the same results as benchFiles
* .result - only valid if .error = 0,
* provide the same results as benchFiles()
* but for the data stored in srcBuffer
*/
BMK_return_t BMK_benchMem(const void* srcBuffer, size_t srcSize,
const size_t* fileSizes, unsigned nbFiles,
const int cLevel, const ZSTD_compressionParameters* comprParams,
int cLevel, const ZSTD_compressionParameters* comprParams,
const void* dictBuffer, size_t dictBufferSize,
int displayLevel, const char* displayName);
/* See benchMem for normal parameter uses and return, see advancedParams_t for adv
/* BMK_benchMemAdvanced() : same as BMK_benchMem()
* with following additional options :
* dstBuffer - destination buffer to write compressed output in, NULL if none provided.
* dstCapacity - capacity of destination buffer, give 0 if dstBuffer = NULL
* adv = see advancedParams_t
*/
BMK_return_t BMK_benchMemAdvanced(const void* srcBuffer, size_t srcSize,
void* dstBuffer, size_t dstCapacity,
void* dstBuffer, size_t dstCapacity,
const size_t* fileSizes, unsigned nbFiles,
const int cLevel, const ZSTD_compressionParameters* comprParams,
int cLevel, const ZSTD_compressionParameters* comprParams,
const void* dictBuffer, size_t dictBufferSize,
int displayLevel, const char* displayName,
const BMK_advancedParams_t* adv);
typedef struct {
size_t sumOfReturn; /* sum of return values */
U64 nanoSecPerRun; /* time per iteration */
unsigned long long nanoSecPerRun; /* time per iteration */
} BMK_customResult_t;
ERROR_STRUCT(BMK_customResult_t, BMK_customReturn_t);
@@ -165,60 +175,60 @@ typedef size_t (*BMK_initFn_t)(void*);
/* benchFn - (*benchFn)(srcBuffers[i], srcSizes[i], dstBuffers[i], dstCapacities[i], benchPayload)
* is run nbLoops times
* initFn - (*initFn)(initPayload) is run once per benchmark at the beginning. This argument can
* initFn - (*initFn)(initPayload) is run once per benchmark at the beginning. This argument can
* be NULL, in which case nothing is run.
* blockCount - number of blocks (size of srcBuffers, srcSizes, dstBuffers, dstCapacities)
* blockCount - number of blocks. Size of all array parameters : srcBuffers, srcSizes, dstBuffers, dstCapacities, blockResults
* srcBuffers - an array of buffers to be operated on by benchFn
* srcSizes - an array of the sizes of above buffers
* dstBuffers - an array of buffers to be written into by benchFn
* dstCapacities - an array of the capacities of above buffers
* blockResults - the return value of benchFn called on each block.
* nbLoops - defines number of times benchFn is run.
* assumed array of size blockCount, will have compressed size of each block written to it.
* return
* @return:
* .error will give a nonzero value if ZSTD_isError() is nonzero for any of the return
* of the calls to initFn and benchFn, or if benchFunction errors internally
* .result - if .error = 0, then .result will contain the sum of all return values of
* benchFn on the first iteration through all of the blocks (.sumOfReturn) and also
* the time per run of benchFn (.nanoSecPerRun). For the former, this
* is generally intended to be used on functions which return the # of bytes written
* into dstBuffer, hence this value will be the total amount of bytes written to
* dstBuffer.
* .result - if .error = 0, then .result will contain
* the sum of all return values of benchFn on the first iteration through all of the blocks (.sumOfReturn)
* and also the time per run of benchFn (.nanoSecPerRun).
* For the former, this is generally intended to be used on functions which return the # of bytes written into dstBuffer,
* hence this value will be the total amount of bytes written into dstBuffer.
*/
BMK_customReturn_t BMK_benchFunction(BMK_benchFn_t benchFn, void* benchPayload,
BMK_initFn_t initFn, void* initPayload,
size_t blockCount,
const void* const * const srcBuffers, const size_t* srcSizes,
void * const * const dstBuffers, const size_t* dstCapacities, size_t* blockResults,
const void *const * srcBuffers, const size_t* srcSizes,
void *const * dstBuffers, const size_t* dstCapacities, size_t* blockResults,
unsigned nbLoops);
/* state information needed to advance computation for benchFunctionTimed */
typedef struct BMK_timeState_t BMK_timedFnState_t;
/* initializes timeState object with desired number of seconds */
BMK_timedFnState_t* BMK_createTimeState(unsigned nbSeconds);
BMK_timedFnState_t* BMK_createTimedFnState(unsigned nbSeconds);
/* resets existing timeState object */
void BMK_resetTimeState(BMK_timedFnState_t*, unsigned nbSeconds);
void BMK_resetTimedFnState(BMK_timedFnState_t*, unsigned nbSeconds);
/* deletes timeState object */
void BMK_freeTimeState(BMK_timedFnState_t* state);
void BMK_freeTimedFnState(BMK_timedFnState_t* state);
typedef struct {
BMK_customReturn_t result;
int completed;
} BMK_customTimedReturn_t;
/*
* Benchmarks custom functions like BMK_benchFunction(), but runs for nbSeconds seconds rather than a fixed number of loops
* arguments mostly the same other than BMK_benchFunction()
* Usage - benchFunctionTimed will return in approximately one second. Keep calling BMK_benchFunctionTimed() until the return's completed field = 1.
* to continue updating intermediate result. Intermediate return values are returned by the function.
/* BMK_benchFunctionTimed() :
* Same as BMK_benchFunction(), but runs for nbSeconds seconds rather than a fixed number of loops.
* Arguments are mostly the same other as BMK_benchFunction()
* Usage - benchFunctionTimed will return in approximately one second.
* Keep calling BMK_benchFunctionTimed() until @return.completed == 1,
* to continue updating intermediate result.
* Intermediate return values are returned by the function.
*/
BMK_customTimedReturn_t BMK_benchFunctionTimed(BMK_timedFnState_t* cont,
BMK_benchFn_t benchFn, void* benchPayload,
BMK_initFn_t initFn, void* initPayload,
size_t blockCount,
const void* const * const srcBlockBuffers, const size_t* srcBlockSizes,
void* const * const dstBlockBuffers, const size_t* dstBlockCapacities, size_t* blockResults);
const void *const * srcBlockBuffers, const size_t* srcBlockSizes,
void *const * dstBlockBuffers, const size_t* dstBlockCapacities, size_t* blockResults);
#endif /* BENCH_H_121279284357 */