int main(void) {
const int size[NSTEPS] = {500, 1000, 1500, 2000, 2500, 3000, 3500, 4000, 4500, 5000}; // specification of array sizes
static const double epsilon = 0.00001; // maximum floating point error allowed between doubles to be considered equal
const double flops_per_iteration = 2.0;
int i; // index variables for looping
double time[2]; // elapsed time. 0 is test case, 1 is base case
int mflops[2]; // calculated mflops. 0 is test case, 1 is base case
double* a;
double* b;
double* ctest;
double* cbase;
stopwatch* sw = stopwatch_new();
for(i = 0; i < NSTEPS; i++) {
int n = size[i];
int n2 = n * n;
a = (double*) malloc(n2*sizeof(double));
b = (double*) malloc(n2*sizeof(double));
ctest = (double*) malloc(n2*sizeof(double));
cbase = (double*) malloc(n2*sizeof(double));
rand_square_double_matrix(i+10, n, a);
rand_square_double_matrix(i+11, n, b);
zero_square_double_matrix(n, ctest);
zero_square_double_matrix(n, cbase);
stopwatch_restart(sw);
my_dgemm(n, a, b, ctest);
stopwatch_stop(sw);
time[0] = stopwatch_time(sw);
stopwatch_restart(sw);
cblas_dgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, n, n, n, 1.0, a, n, b, n, 1.0, cbase, n);
stopwatch_stop(sw);
//printf("A\n");
//print_square_matrix(n, a);
//printf("B\n");
//print_square_matrix(n, b);
//printf("my_dgemm\n");
//print_square_matrix(n, ctest);
//printf("cdgemm\n");
//print_square_matrix(n, cbase);
assert_equal(n, ctest, cbase, epsilon);
time[1] = stopwatch_time(sw);
mflops[0] = calc_mflops(flops_per_iteration, n, time[0]);
mflops[1] = calc_mflops(flops_per_iteration, n, time[1]);
printf("%d, %5.2f, %d, %5.2f, %d\n", n, time[0], mflops[0], time[1], mflops[1]);
free(a);
free(b);
free(ctest);
free(cbase);
}
stopwatch_delete(sw);
return 0;
}
void stopwatch_stop(struct timeval *start_time) {
struct timeval *end_time = stopwatch_time();
unsigned long msec;
msec = (end_time->tv_sec - start_time->tv_sec) * 1000;
msec += (end_time->tv_usec - start_time->tv_usec) / 1000;
info("request completed in %lu ms", msec);
free(start_time);
free(end_time);
}
