floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

added conf option for number of loops

This commit is contained in:
wernsaar 2014-07-12 11:54:39 +02:00
parent b0e7810a6b
commit 1d4ffddf69
1 changed files with 32 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
benchmark/gemm.c
View File

@ -123,17 +123,18 @@ int MAIN__(int argc, char *argv[]){
FLOAT *a, *b, *c; FLOAT *a, *b, *c;
FLOAT alpha[] = {1.0, 1.0}; FLOAT alpha[] = {1.0, 1.0};
FLOAT beta [] = {1.0, 1.0}; FLOAT beta [] = {1.0, 1.0};
char trans='N'; char trans='N';
blasint m, i, j; blasint m, i, j;
int loops = 1;
int l;
char *p;
int from = 1; int from = 1;
int to = 200; int to = 200;
int step = 1; int step = 1;
struct timeval start, stop; struct timeval start, stop;
double time1; double time1,timeg;
argc--;argv++; argc--;argv++;
@ -155,6 +156,9 @@ int MAIN__(int argc, char *argv[]){
fprintf(stderr,"Out of Memory!!\n");exit(1); fprintf(stderr,"Out of Memory!!\n");exit(1);
} }
p = getenv("OPENBLAS_LOOPS");
if ( p != NULL )
loops = atoi(p);
#ifdef linux #ifdef linux
@ -166,8 +170,13 @@ int MAIN__(int argc, char *argv[]){
for(m = from; m <= to; m += step) for(m = from; m <= to; m += step)
{ {
timeg=0;
fprintf(stderr, " %6d : ", (int)m); fprintf(stderr, " %6d : ", (int)m);
for (l=0; l<loops; l++)
{
for(j = 0; j < m; j++){ for(j = 0; j < m; j++){
for(i = 0; i < m * COMPSIZE; i++){ for(i = 0; i < m * COMPSIZE; i++){
a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5; a[i + j * m * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
@ -184,11 +193,14 @@ int MAIN__(int argc, char *argv[]){
time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6; time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
gettimeofday( &start, (struct timezone *)0); timeg += time1;
}
timeg /= loops;
fprintf(stderr, fprintf(stderr,
" %10.2f MFlops\n", " %10.2f MFlops\n",
COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / time1 * 1.e-6); COMPSIZE * COMPSIZE * 2. * (double)m * (double)m * (double)m / timeg * 1.e-6);
} }