wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Vector evaluations

Browse Source
This commit is contained in:
proshan3
2019-11-25 22:18:12 -05:00
parent 72361b3afe
commit 374d991a20
16 changed files with 104163 additions and 104289 deletions
Download Patch File Download Diff File Expand all files Collapse all files

87
benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.c
View File

@@ -16,76 +16,89 @@ int main()
{
vx_tmc(1);
int w = 4;
int h = 4;
int d = 4;
int m = 64;
int k = 64;
int n = 64;
int* a1 = (int*)malloc(sizeof(int) * w * h);
int* b1 = (int*)malloc(sizeof(int) * h * d);
int* c1 = (int*)malloc(sizeof(int) * w * d);
int* d1 = (int*)malloc(sizeof(int) * w * d); //verfication
int* a1 = (int*)malloc(sizeof(int) * m * k);
int* b1 = (int*)malloc(sizeof(int) * k * n);
int* c1 = (int*)malloc(sizeof(int) * m * n);
int* d1 = (int*)malloc(sizeof(int) * m * n); //verfication
for (int i = 0; i < (w * h); ++i) a1[i] = i;
for (int i = 0; i < (h * d); ++i) b1[i] = 1;
for (int i = 0; i < (w * d); ++i) c1[i] = 0;
for (int i = 0; i < (w * d); ++i) d1[i] = 0;
for (int i = 0; i < (m * k); ++i) a1[i] = i;
for (int i = 0; i < (k * n); ++i) b1[i] = 1;
for (int i = 0; i < (m * n); ++i) c1[i] = 0;
for (int i = 0; i < (m * n); ++i) d1[i] = 0;
#if 0
printf("sgemm_nn\na[%d]:", w*h);
for (int i = 0; i < w*h; ++i) {
if(!(i % h)) printf("\n");
printf("sgemm_nn\na[%d]:", m*k);
for (int i = 0; i < m*k; ++i) {
if(!(i % k)) printf("\n");
printf("%d ", a1[i]);
}
printf("\n\nb[%d]:", h*d);
for (int i = 0; i < h*d; ++i) {
if (!(i % d)) printf("\n");
printf("\n\nb[%d]:", k*n);
for (int i = 0; i < k*n; ++i) {
if (!(i % n)) printf("\n");
printf("%d ", b1[i]);
}
#endif
int lda = 4;
int ldb = 4;
int ldc = 4; //64;
int vsize = 4;
int ldc = 64; //64;
int vsize = 32;
for (int n = 0; n < h; n++) {
for (int i = 0; i < w; i=+4) {
for (int m = 0; m < d; m++) {
vx_vec_sgemm_nn(i, m, n, a1, b1, c1, ldc, vsize);
int startCycles = vx_getCycles();
int startInst = vx_getInst();
for (int r = 0; r < m; r++) {
for (int c = 0; c < n; c++) {
for (int i = 0; i < k;) {
// d1[r*k+i] += a1[r*k+c]*b1[i*n+c];
vx_vec_sgemm_nn(i, r, c, a1, b1, c1, ldc, vsize);
i = i + vsize;
}
}
}
int endCycles = vx_getCycles();
int endInst = vx_getInst();
#if 1
printf("\n\nc[%d]:", d*h);
for (int i = 0; i < d*h; ++i) {
if (!(i % h)) printf("\n");
int totalInst = (endInst - startInst);
int totalCycles = (endCycles - startCycles);
printf("\nCycles = %d, Instructions = %d", totalCycles, totalInst);
// vx_vec_sgemm_nn(n, a1, b1, c1);
#if 0
printf("\n\nc[%d]:", m*n);
for (int i = 0; i < m*n; ++i) {
if (!(i % n)) printf("\n");
printf("%d ", c1[i]);
}
#endif
for (int r = 0; r < h; r++) {
for (int c = 0; c < w; c++) {
for (int i = 0; i < d; i++) {
d1[r*h+i] += a1[r*h+c]*b1[i*d+c];
for (int r = 0; r < m; r++) {
for (int c = 0; c < n; c++) {
for (int i = 0; i < k; i++) {
d1[c*ldc+i] += a1[c*ldc+r]*b1[i + (r*ldc)];
//printf("d[%d] += a[%d]*b[%d]\n", c*ldc+i, c*ldc+r , i + (r*ldc));
//printf("%d %d %d\n", d1[c*ldc+i] , a1[c*ldc+r] , b1[i + (r*ldc)]);
}
}
}
#if 1
printf("\n\nc[%d]:\n", w*d);
for(int i = 0; i < w; ++i) {
for(int j = 0; j < d; ++j) {
printf("%d ", d1[i*w+j]);
#if 0
printf("\n\nc[%d]:\n", m*n);
for(int i = 0; i < m; ++i) {
for(int j = 0; j < n; ++j) {
printf("%d ", d1[i*m+j]);
}
printf("\n");
}
#endif
for(int i = 0; i < w*d; ++i)
for(int i = 0; i < m*n; ++i)
{
if(c1[i] != d1[i])
{