int main (int argc, char *argv[]) {
float *A, *B, *C;
int N, M, i, j;
#ifdef CHECK
FILE *fd;
#endif
if (argc != 3) {
printf("Usage: mm-test matrix_size iter\n");
exit(0);
}
N = atoi(argv[1]);
M = atoi(argv[2]);
/* Alokacja i inicjowanie macierzy */
A = (float *)malloc(N * N * sizeof(float));
B = (float *)malloc(N * N * sizeof(float));
C = (float *)malloc(N * N * sizeof(float));
#ifdef CHECK
if ((fd = fopen("tmp111", "r")) == NULL) {
printf("Cannot open tmp111\n");
exit(0);
}
for (i = 0; i < N; i++)
for (j = 0; j < N; j++)
fscanf(fd, "%f", (A + i * N + j));
fclose(fd);
if ((fd = fopen("tmp222", "r")) == NULL) {
printf("Cannot open tmp222\n");
exit(0);
}
for (i = 0; i < N; i++)
for (j = 0; j < N; j++)
fscanf(fd, " %f", (B + i * N + j));
fclose(fd);
#else
srand48(100);
for (i = 0; i < N * N; ++i) {
A[i] = drand48();
B[i] = drand48();
}
#endif
/* Wielokrotne mnożenie */
gettimeofday(&t1, NULL);
for (i = 0; i < M; i++)
naive_mm(N, B, A, C);
gettimeofday(&t2, NULL);
printf("Time for the matrix multiplication is %d milliseconds\n",
(t2.tv_sec - t1.tv_sec) * 1000 +
(t2.tv_usec - t1.tv_usec) / 1000);
#ifdef CHECK
if ((fd = fopen("tmp333", "w")) == NULL) {
printf("Cannot open tmp333\n");
exit(0);
}
for (i = 0; i < N; i++) {
for (j = 0; j < N; j++)
fprintf(fd, "%6.2lf ", *(C + i * N + j));
fprintf(fd, "\n");
}
fclose(fd);
#endif
return 0;
}
void
checkCorrect ()
{
double *A,*B,*C,*cA,*cB,*cC;
int minDim = 1;
int maxDim = 256;
int maxnbytes = sizeof(double) * SQR (maxDim);
int i, j;
A = (double*) malloc (maxnbytes);
B = (double*) malloc (maxnbytes);
C = (double*) malloc (maxnbytes);
cA = (double*) malloc (maxnbytes);
cB = (double*) malloc (maxnbytes);
cC = (double*) malloc (maxnbytes);
fprintf (stderr, "Checking for correctness on sizes:");
#if !RANDOM_TESTS
for (i = 0; i < 2; i++) for (j = 0; j < num_tests[i]; j++)
{
int matdim = test_sizes[i][j];
#else
for (i = 0; i < NUM_CORRECTNESS_CHECKS; i++)
{
int matdim = rrand (minDim, maxDim);
#endif
double err;
int nbytes = sizeof(double) * SQR(matdim);
fprintf (stderr, " %d", matdim);
mat_init (A, matdim, matdim);
mat_init (B, matdim, matdim);
mat_init (C, matdim, matdim);
bcopy ((void*)A, (void*)cA, nbytes);
bcopy ((void*)B, (void*)cB, nbytes);
bcopy ((void*)C, (void*)cC, nbytes);
naive_mm (matdim, matdim, matdim, cA, cB, cC);
MUL_MFMF_MF (matdim, A, B, C);
if (bcmp ((void*)A, (void*)cA, nbytes) != 0 ||
bcmp ((void*)B, (void*)cB, nbytes) != 0)
{
fprintf (stderr, "Source matrices were modified. DISQUALIFIED!!!\n")
;
//exit (0);
}
if ((err = error (C, cC, matdim, matdim)) > MAX_ERROR)
{
fprintf (stderr, "Error for test case %dx%d is %f > %f. DISQUALIFIED!
!!\n",
matdim, matdim, err, MAX_ERROR);
//exit (0);
}
}
fprintf (stderr,"\n");
free (A); free (B); free (C);
free (cA); free (cB); free (cC);
}
void
timeIt ()
{
double *A, *B, *C;
double *oA[TEST_RUNS], *oB[TEST_RUNS], *oC[TEST_RUNS];
int i, j, k;
int test;
for (k = 0; k < 2; k++)
{
if (k > 0) printf ("\n");
for (test = 0; test < num_tests[k]; test++)
{
int matdim = test_sizes[k][test];
const int num_iters = CALC_ITERS (matdim);
double max_mflops = 0.0;
int run;
/* make sure these are quad-word (i.e., 16-byte) aligned */
#if 0
A = oA = (double*) malloc ((SQR(matdim)+1) * sizeof(double));
B = oB = (double*) malloc ((SQR(matdim)+1) * sizeof(double));
C = oC = (double*) malloc ((SQR(matdim)+1) * sizeof(double));
#endif
for (run = 0; run < TEST_RUNS; run++)
{
int iter;
double mflops;
double utime;
/* use different matricies for each trial so that the OS page map
ping */
/* won't affect the results... */
A = oA[run] = (double*) malloc ((SQR(matdim)+rrand(1,10)) * sizeof(double));
B = oB[run] = (double*) malloc ((SQR(matdim)+rrand(1,10)) * sizeof(double));
C = oC[run] = (double*) malloc ((SQR(matdim)+rrand(1,10)) * sizeof(double));
if (((unsigned)A) & 0x8) A = (double*)(((unsigned)A)+0x8);
if (((unsigned)B) & 0x8) B = (double*)(((unsigned)B)+0x8);
if (((unsigned)C) & 0x8) C = (double*)(((unsigned)C)+0x8);
mat_init (A, matdim, matdim);
mat_init (B, matdim, matdim);
mat_init (C, matdim, matdim);
START_TIMING;
for (iter=0;iter<num_iters;iter++)
{
// iteratively accumulate into C
MUL_MFMF_MF (matdim, A, B, C);
}
STOP_TIMING;
utime = reportTiming();
// (2 * n^3) FLOPs (n^3 mul-adds)
mflops = 2.0 * CUBE(matdim) * num_iters * 1e-6 / utime;
max_mflops = MAX (max_mflops, mflops);
}
printf("%d %.0f\n", matdim, max_mflops);
fflush(stdout);
for (run = 0; run < TEST_RUNS; run++)
{
free (oA[run]);
free (oB[run]);
free (oC[run]);
}
}
}
}
