#include "datagen.h" /* RDG_genBuffer */
#include "xxhash.h"
#include "bench.h"
+#include "zstd_errors.h"
/* *************************************
}
/* error without displaying */
-#define EXM_THROW_ND(errorNum, retType, ...) { \
+#define EXM_THROW_ND(errorNum, retType, ...) { \
retType r; \
memset(&r, 0, sizeof(retType)); \
DEBUGOUTPUT("%s: %i: \n", __FILE__, __LINE__); \
- DEBUGOUTPUT("Error %i : ", errorNum); \
- DEBUGOUTPUT(__VA_ARGS__); \
- DEBUGOUTPUT(" \n"); \
+ DEBUGOUTPUT("Error %i : ", errorNum); \
+ DEBUGOUTPUT(__VA_ARGS__); \
+ DEBUGOUTPUT(" \n"); \
r.error = errorNum; \
return r; \
}
* Benchmark Parameters
***************************************/
-#define BMK_LDM_PARAM_NOTSET 9999
-
BMK_advancedParams_t BMK_initAdvancedParams(void) {
BMK_advancedParams_t res = {
BMK_both, /* mode */
0, /* ldmFlag */
0, /* ldmMinMatch */
0, /* ldmHashLog */
- BMK_LDM_PARAM_NOTSET, /* ldmBuckSizeLog */
- BMK_LDM_PARAM_NOTSET /* ldmHashEveryLog */
+ 0, /* ldmBuckSizeLog */
+ 0 /* ldmHashEveryLog */
};
return res;
}
size_t resSize;
} blockParam_t;
+struct BMK_timeState_t{
+ unsigned nbLoops;
+ U64 timeRemaining;
+ UTIL_time_t coolTime;
+ U64 fastestTime;
+};
+
#undef MIN
#undef MAX
#define MIN(a,b) ((a) < (b) ? (a) : (b))
ZSTD_CCtx_setParameter(ctx, ZSTD_p_enableLongDistanceMatching, adv->ldmFlag);
ZSTD_CCtx_setParameter(ctx, ZSTD_p_ldmMinMatch, adv->ldmMinMatch);
ZSTD_CCtx_setParameter(ctx, ZSTD_p_ldmHashLog, adv->ldmHashLog);
- if (adv->ldmBucketSizeLog != BMK_LDM_PARAM_NOTSET) {
- ZSTD_CCtx_setParameter(ctx, ZSTD_p_ldmBucketSizeLog, adv->ldmBucketSizeLog);
- }
- if (adv->ldmHashEveryLog != BMK_LDM_PARAM_NOTSET) {
- ZSTD_CCtx_setParameter(ctx, ZSTD_p_ldmHashEveryLog, adv->ldmHashEveryLog);
- }
+ ZSTD_CCtx_setParameter(ctx, ZSTD_p_ldmBucketSizeLog, adv->ldmBucketSizeLog);
+ ZSTD_CCtx_setParameter(ctx, ZSTD_p_ldmHashEveryLog, adv->ldmHashEveryLog);
ZSTD_CCtx_setParameter(ctx, ZSTD_p_windowLog, comprParams->windowLog);
ZSTD_CCtx_setParameter(ctx, ZSTD_p_hashLog, comprParams->hashLog);
ZSTD_CCtx_setParameter(ctx, ZSTD_p_chainLog, comprParams->chainLog);
out.size = dstSize;
out.pos = 0;
while (moreToFlush) {
+ if(out.pos == out.size) {
+ return (size_t)-ZSTD_error_dstSize_tooSmall;
+ }
moreToFlush = ZSTD_compress_generic(ctx, &out, &in, ZSTD_e_end);
if (ZSTD_isError(moreToFlush)) {
return moreToFlush;
out.size = dstSize;
out.pos = 0;
while (moreToFlush) {
+ if(out.pos == out.size) {
+ return (size_t)-ZSTD_error_dstSize_tooSmall;
+ }
moreToFlush = ZSTD_decompress_generic(dctx,
&out, &in);
if (ZSTD_isError(moreToFlush)) {
/* initFn will be measured once, bench fn will be measured x times */
/* benchFn should return error value or out Size */
/* takes # of blocks and list of size & stuff for each. */
-/* only does iterations*/
-/* note time/iter could be zero if interval too short */
+/* only does looping */
+/* note time per loop could be zero if interval too short */
BMK_customReturn_t BMK_benchFunction(
- size_t (*benchFn)(const void*, size_t, void*, size_t, void*), void* benchPayload,
- size_t (*initFn)(void*), void* initPayload,
+ 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,
- unsigned iter) {
+ unsigned nbLoops) {
size_t srcSize = 0, dstSize = 0, ind = 0;
- unsigned toAdd = 1;
U64 totalTime;
BMK_customReturn_t retval;
UTIL_waitForNextTick();
}
- if(!iter) {
+ if(!nbLoops) {
EXM_THROW_ND(1, BMK_customReturn_t, "nbLoops must be nonzero \n");
}
srcSize += srcBlockSizes[ind];
}
- {
- unsigned i, j;
+ {
+ unsigned i, j, firstIter = 1;
clockStart = UTIL_getTime();
- if(initFn != NULL) { (*initFn)(initPayload); }
- for(i = 0; i < iter; i++) {
+ if(initFn != NULL) { initFn(initPayload); }
+ for(i = 0; i < nbLoops; i++) {
for(j = 0; j < blockCount; j++) {
- size_t res = (*benchFn)(srcBlockBuffers[j], srcBlockSizes[j], dstBlockBuffers[j], dstBlockCapacities[j], benchPayload);
+ size_t res = benchFn(srcBlockBuffers[j], srcBlockSizes[j], dstBlockBuffers[j], dstBlockCapacities[j], benchPayload);
if(ZSTD_isError(res)) {
EXM_THROW_ND(2, BMK_customReturn_t, "Function benchmarking failed on block %u of size %u : %s \n",
j, (U32)dstBlockCapacities[j], ZSTD_getErrorName(res));
- } else if(toAdd) {
+ } else if(firstIter) {
dstSize += res;
}
}
- toAdd = 0;
+ firstIter = 0;
}
totalTime = UTIL_clockSpanNano(clockStart);
}
retval.error = 0;
- retval.result.nanoSecPerRun = totalTime / iter;
+ retval.result.nanoSecPerRun = totalTime / nbLoops;
retval.result.sumOfReturn = dstSize;
return retval;
}
-BMK_customResultContinuation_t BMK_init_customResultContinuation(unsigned iter) {
- BMK_customResultContinuation_t c;
- c.completed = 0;
- c.state.nbLoops = 1;
- c.state.coolTime = UTIL_getTime();
- c.state.timeRemaining = (U64)iter * TIMELOOP_NANOSEC;
- c.intermediateResult.error = 0;
- c.intermediateResult.result.nanoSecPerRun = (U64)(-1LL);
- c.intermediateResult.result.sumOfReturn = 0;
- return c;
+#define MINUSABLETIME 500000000ULL /* 0.5 seconds in ns */
+
+void BMK_resetTimeState(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* r = (BMK_timedFnState_t*)malloc(sizeof(struct BMK_timeState_t));
+ BMK_resetTimeState(r, nbSeconds);
+ return r;
}
-#define MINUSABLETIME 500000000ULL
+void BMK_freeTimeState(BMK_timedFnState_t* state) {
+ free(state);
+}
-//how to use minusabletime?
-//only report times which are > minUsable
-void BMK_benchFunctionTimed(
- size_t (*benchFn)(const void*, size_t, void*, size_t, void*), void* benchPayload,
- size_t (*initFn)(void*), void* initPayload,
+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,
- BMK_customResultContinuation_t* cont)
+ const void* const* const srcBlockBuffers, const size_t* srcBlockSizes,
+ void* const* const dstBlockBuffers, const size_t* dstBlockCapacities)
{
- U64 fastest = cont->intermediateResult.result.nanoSecPerRun;
+ U64 fastest = cont->fastestTime;
int completed = 0;
+ BMK_customTimedReturn_t r;
+ r.completed = 0;
- while(!cont->completed && !completed)
+ while(!r.completed && !completed)
{
/* Overheat protection */
- if (UTIL_clockSpanMicro(cont->state.coolTime) > ACTIVEPERIOD_MICROSEC) {
+ if (UTIL_clockSpanMicro(cont->coolTime) > ACTIVEPERIOD_MICROSEC) {
DEBUGOUTPUT("\rcooling down ... \r");
UTIL_sleep(COOLPERIOD_SEC);
- cont->state.coolTime = UTIL_getTime();
+ cont->coolTime = UTIL_getTime();
}
- cont->intermediateResult = BMK_benchFunction(benchFn, benchPayload, initFn, initPayload,
- blockCount, srcBlockBuffers, srcBlockSizes, dstBlockBuffers, dstBlockCapacities, cont->state.nbLoops);
- if(cont->intermediateResult.error) { /* completed w/ error */
- cont->completed = 1;
- return;
+ r.result = BMK_benchFunction(benchFn, benchPayload, initFn, initPayload,
+ blockCount, srcBlockBuffers, srcBlockSizes, dstBlockBuffers, dstBlockCapacities, cont->nbLoops);
+ if(r.result.error) { /* completed w/ error */
+ r.completed = 1;
+ return r;
}
- { U64 const loopDuration = cont->intermediateResult.result.nanoSecPerRun * cont->state.nbLoops;
- cont->completed = (cont->state.timeRemaining <= loopDuration);
- cont->state.timeRemaining -= loopDuration;
+ { U64 const loopDuration = r.result.result.nanoSecPerRun * cont->nbLoops;
+ r.completed = (cont->timeRemaining <= loopDuration);
+ cont->timeRemaining -= loopDuration;
if (loopDuration > 0) {
- fastest = MIN(fastest, cont->intermediateResult.result.nanoSecPerRun);
- cont->intermediateResult.result.nanoSecPerRun = fastest;
- cont->state.nbLoops = (U32)(TIMELOOP_NANOSEC / fastest) + 1;
+ if(loopDuration >= MINUSABLETIME) { /* don't report results which have time too low */
+ fastest = MIN(fastest, r.result.result.nanoSecPerRun);
+ }
+ r.result.result.nanoSecPerRun = fastest;
+ cont->fastestTime = fastest;
+ cont->nbLoops = (U32)(TIMELOOP_NANOSEC / fastest) + 1;
} else {
const unsigned multiplier = 2;
- assert(cont->state.nbLoops < ((unsigned)-1) / multiplier); /* avoid overflow */
- cont->state.nbLoops *= multiplier;
+ assert(cont->nbLoops < ((unsigned)-1) / multiplier); /* avoid overflow */
+ cont->nbLoops *= multiplier;
}
if(loopDuration < MINUSABLETIME) { /* don't report results which have time too low */
continue;
}
completed = 1;
}
+ return r;
}
/* benchMem with no allocation */
void** const resPtrs, size_t* const resSizes,
void* resultBuffer, void* compressedBuffer,
const size_t maxCompressedSize,
+ BMK_timedFnState_t* timeState,
const void* srcBuffer, size_t srcSize,
const size_t* fileSizes, unsigned nbFiles,
/* Compression */
DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->\r", marks[markNb], displayName, (U32)srcSize);
if(adv->loopMode == BMK_timeMode) {
- BMK_customResultContinuation_t cont = BMK_init_customResultContinuation(adv->nbSeconds);
- while(!cont.completed) {
- BMK_benchFunctionTimed(&local_defaultCompress, (void*)ctx, &local_initCCtx, (void*)&cctxprep,
- nbBlocks, srcPtrs, srcSizes, cPtrs, cSizes, &cont);
- if(cont.intermediateResult.error) {
- results.error = cont.intermediateResult.error;
+ BMK_customTimedReturn_t intermediateResult;
+ intermediateResult.completed = 0;
+ while(!intermediateResult.completed) {
+ intermediateResult = BMK_benchFunctionTimed(timeState, &local_defaultCompress, (void*)ctx, &local_initCCtx, (void*)&cctxprep,
+ nbBlocks, srcPtrs, srcSizes, cPtrs, cSizes);
+ if(intermediateResult.result.error) {
+ results.error = intermediateResult.result.error;
return results;
}
- ratio = (double)(srcSize / cont.intermediateResult.result.sumOfReturn);
+ ratio = (double)(srcSize / intermediateResult.result.result.sumOfReturn);
{
int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
- double const compressionSpeed = ((double)srcSize / cont.intermediateResult.result.nanoSecPerRun) * 1000;
+ double const compressionSpeed = ((double)srcSize / intermediateResult.result.result.nanoSecPerRun) * 1000;
int const cSpeedAccuracy = (compressionSpeed < 10.) ? 2 : 1;
results.result.cSpeed = compressionSpeed * 1000000;
- results.result.cSize = cont.intermediateResult.result.sumOfReturn;
+ results.result.cSize = intermediateResult.result.result.sumOfReturn;
ratio = (double)srcSize / results.result.cSize;
markNb = (markNb+1) % NB_MARKS;
DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s\r",
dctxprep.dictBuffer = dictBuffer;
dctxprep.dictBufferSize = dictBufferSize;
if(adv->loopMode == BMK_timeMode) {
- BMK_customResultContinuation_t cont = BMK_init_customResultContinuation(adv->nbSeconds);
- while(!cont.completed) {
- BMK_benchFunctionTimed(&local_defaultDecompress, (void*)(dctx), &local_initDCtx, (void*)&dctxprep,
- nbBlocks, (const void * const *)cPtrs, cSizes, resPtrs, resSizes, &cont);
- if(cont.intermediateResult.error) {
- results.error = cont.intermediateResult.error;
+ BMK_customTimedReturn_t intermediateResult;
+ intermediateResult.completed = 0;
+ BMK_resetTimeState(timeState, adv->nbSeconds);
+ while(!intermediateResult.completed) {
+ intermediateResult = BMK_benchFunctionTimed(timeState, &local_defaultDecompress, (void*)(dctx), &local_initDCtx, (void*)&dctxprep,
+ nbBlocks, (const void* const*)cPtrs, cSizes, resPtrs, resSizes);
+ if(intermediateResult.result.error) {
+ results.error = intermediateResult.result.error;
return results;
}
{ int const ratioAccuracy = (ratio < 10.) ? 3 : 2;
double const compressionSpeed = results.result.cSpeed / 1000000;
int const cSpeedAccuracy = (compressionSpeed < 10.) ? 2 : 1;
- double const decompressionSpeed = ((double)srcSize / cont.intermediateResult.result.nanoSecPerRun) * 1000;
+ double const decompressionSpeed = ((double)srcSize / intermediateResult.result.result.nanoSecPerRun) * 1000;
results.result.dSpeed = decompressionSpeed * 1000000;
markNb = (markNb+1) % NB_MARKS;
DISPLAYLEVEL(2, "%2s-%-17.17s :%10u ->%10u (%5.*f),%6.*f MB/s ,%6.1f MB/s \r",
decompressionResults = BMK_benchFunction(
&local_defaultDecompress, (void*)(dctx),
&local_initDCtx, (void*)&dctxprep, nbBlocks,
- (const void * const *)cPtrs, cSizes, resPtrs, resSizes,
+ (const void* const*)cPtrs, cSizes, resPtrs, resSizes,
adv->nbSeconds);
if(decompressionResults.error) {
results.error = decompressionResults.error;
U32 const maxNbBlocks = (U32) ((srcSize + (blockSize-1)) / blockSize) + nbFiles;
/* these are the blockTable parameters, just split up */
- const void ** const srcPtrs = (const void ** const)malloc(maxNbBlocks * sizeof(void*));
+ const void ** const srcPtrs = (const void** const)malloc(maxNbBlocks * sizeof(void*));
size_t* const srcSizes = (size_t* const)malloc(maxNbBlocks * sizeof(size_t));
void ** const cPtrs = (void** const)malloc(maxNbBlocks * sizeof(void*));
const size_t maxCompressedSize = ZSTD_compressBound(srcSize) + (maxNbBlocks * 1024); /* add some room for safety */
void* compressedBuffer = malloc(maxCompressedSize);
void* resultBuffer = malloc(srcSize);
-
+ BMK_timedFnState_t* timeState = BMK_createTimeState(adv->nbSeconds);
+
+
BMK_return_t results;
int allocationincomplete = !compressedBuffer || !resultBuffer ||
!srcPtrs || !srcSizes || !cPtrs || !cSizes || !resPtrs || !resSizes;
if (!allocationincomplete) {
results = BMK_benchMemAdvancedNoAlloc(srcPtrs, srcSizes, cPtrs, cSizes,
- resPtrs, resSizes, resultBuffer, compressedBuffer, maxCompressedSize,
+ resPtrs, resSizes, resultBuffer, compressedBuffer, maxCompressedSize, timeState,
srcBuffer, srcSize, fileSizes, nbFiles, cLevel, comprParams,
dictBuffer, dictBufferSize, ctx, dctx, displayLevel, displayName, adv);
}
/* clean up */
+ BMK_freeTimeState(timeState);
free(compressedBuffer);
free(resultBuffer);
int cLevel, const ZSTD_compressionParameters* const comprParams,
const void* dictBuffer, size_t dictBufferSize,
int displayLevel, const char* displayName,
- const BMK_advancedParams_t * const adv)
+ const BMK_advancedParams_t* const adv)
{
BMK_return_t res;
ZSTD_CCtx* ctx = ZSTD_createCCtx();
const ZSTD_compressionParameters* const compressionParams,
int displayLevel, const BMK_advancedParams_t * const adv)
{
- void* srcBuffer;
+ void* srcBuffer = NULL;
size_t benchedSize;
void* dictBuffer = NULL;
size_t dictBufferSize = 0;
- size_t* fileSizes;
+ size_t* fileSizes = NULL;
BMK_return_t res;
U64 const totalSizeToLoad = UTIL_getTotalFileSize(fileNamesTable, nbFiles);
int errorCode = BMK_loadFiles(dictBuffer, dictBufferSize, fileSizes, &dictFileName, 1, displayLevel);
if(errorCode) {
res.error = errorCode;
- return res;
+ goto _cleanUp;
}
}
}
int errorCode = BMK_loadFiles(srcBuffer, benchedSize, fileSizes, fileNamesTable, nbFiles, displayLevel);
if(errorCode) {
res.error = errorCode;
- return res;
+ goto _cleanUp;
}
}
/* Bench */
adv);
} }
- /* clean up */
+_cleanUp:
free(srcBuffer);
free(dictBuffer);
free(fileSizes);
displayLevel, name, adv);
/* clean up */
- free((void*)srcBuffer);
+ free(srcBuffer);
return res;
}
* in .error != 0.
*/
#define ERROR_STRUCT(baseType, typeName) typedef struct { \
- int error; \
baseType result; \
+ int error; \
} typeName
typedef struct {
double dSpeed;
} BMK_result_t;
-typedef struct {
- size_t sumOfReturn; /* sum of return values */
- U64 nanoSecPerRun; /* time per iteration */
-} BMK_customResult_t;
-//we might need a nbRuns or nbSecs if we're keeping timeMode / iterMode respectively.
-//give benchMem responsibility to incrementally update display.
-
ERROR_STRUCT(BMK_result_t, BMK_return_t);
-ERROR_STRUCT(BMK_customResult_t, BMK_customReturn_t);
-
-typedef enum {
- BMK_timeMode = 0,
- BMK_iterMode = 1
-} BMK_loopMode_t;
-
-typedef enum {
- BMK_both = 0,
- BMK_decodeOnly = 1,
- BMK_compressOnly = 2
-} BMK_mode_t;
-
-typedef struct {
- BMK_mode_t mode; /* 0: all, 1: compress only 2: decode only */
- BMK_loopMode_t loopMode; /* if loopmode, then nbSeconds = nbLoops */
- unsigned nbSeconds; /* default timing is in nbSeconds */
- size_t blockSize; /* Maximum allowable size of a block*/
- unsigned nbWorkers; /* multithreading */
- unsigned realTime; /* real time priority */
- 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 ldmBucketSizeLog;
- unsigned ldmHashEveryLog;
-} BMK_advancedParams_t;
-
-/* returns default parameters used by nonAdvanced functions */
-BMK_advancedParams_t BMK_initAdvancedParams(void);
/* called in cli */
/* Loads files in fileNamesTable into memory, as well as a dictionary
int const cLevel, const ZSTD_compressionParameters* const compressionParams,
int displayLevel);
+typedef enum {
+ BMK_timeMode = 0,
+ BMK_iterMode = 1
+} BMK_loopMode_t;
+
+typedef enum {
+ BMK_both = 0,
+ BMK_decodeOnly = 1,
+ BMK_compressOnly = 2
+} BMK_mode_t;
+
+typedef struct {
+ BMK_mode_t mode; /* 0: all, 1: compress only 2: decode only */
+ BMK_loopMode_t loopMode; /* if loopmode, then nbSeconds = nbLoops */
+ unsigned nbSeconds; /* default timing is in nbSeconds */
+ size_t blockSize; /* Maximum allowable size of a block*/
+ unsigned nbWorkers; /* multithreading */
+ unsigned realTime; /* real time priority */
+ 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 ldmBucketSizeLog;
+ unsigned ldmHashEveryLog;
+} BMK_advancedParams_t;
+
+/* 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 displayLevel, const char* displayName,
const BMK_advancedParams_t* adv);
+typedef struct {
+ size_t sumOfReturn; /* sum of return values */
+ U64 nanoSecPerRun; /* time per iteration */
+} BMK_customResult_t;
+
+ERROR_STRUCT(BMK_customResult_t, BMK_customReturn_t);
+
+typedef size_t (*BMK_benchFn_t)(const void*, size_t, void*, size_t, void*);
+typedef size_t (*BMK_initFn_t)(void*);
+
/* This function times the execution of 2 argument functions, benchFn and initFn */
/* benchFn - (*benchFn)(srcBuffers[i], srcSizes[i], dstBuffers[i], dstCapacities[i], benchPayload)
- * is run iter times
+ * is run nbLoops times
* 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)
* 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.
- * iter - defines number of times benchFn is run.
+ * nbLoops - defines number of times benchFn is run.
* 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
* dstBuffer.
*/
BMK_customReturn_t BMK_benchFunction(
- size_t (*benchFn)(const void*, size_t, void*, size_t, void*), void* benchPayload,
- size_t (*initFn)(void*), void* initPayload,
+ 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,
- unsigned sec);
+ unsigned nbLoops);
-typedef struct {
- unsigned nbLoops;
- U64 timeRemaining;
- UTIL_time_t coolTime;
-} BMK_timeState_t;
+
+/* 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);
+/* resets existing timeState object */
+void BMK_resetTimeState(BMK_timedFnState_t*, unsigned nbSeconds);
+/* deletes timeState object */
+void BMK_freeTimeState(BMK_timedFnState_t* state);
typedef struct {
+ BMK_customReturn_t result;
int completed;
- BMK_customReturn_t intermediateResult; /* since the wrapper can't err, don't need ERROR_STRUCT(cRC, just check here) */
- BMK_timeState_t state;
-} BMK_customResultContinuation_t;
-
-/*
- * initializes the last argument of benchFunctionTimed, with iter being the number of seconds to bench (see below)
- */
-BMK_customResultContinuation_t BMK_init_customResultContinuation(unsigned iter);
+} BMK_customTimedReturn_t;
/*
- * Benchmarks custom functions like BMK_benchFunction(), but runs for iter seconds rather than a fixed number of iterations
+ * 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, where the intermediate results can be found in
- * the *cont passed in and be displayed/used as wanted. Keep calling BMK_benchFunctionTimed() until cont->completed = 1
- * to continue updating intermediate result.
+ * 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.
*/
-void BMK_benchFunctionTimed(
- size_t (*benchFn)(const void*, size_t, void*, size_t, void*), void* benchPayload,
- size_t (*initFn)(void*), void* initPayload,
+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,
- BMK_customResultContinuation_t* cont);
+ void* const * const dstBlockBuffers, const size_t* dstBlockCapacities);
#endif /* BENCH_H_121279284357 */