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

merging error fixed

Browse Source
This commit is contained in:
Euna Kim
2019-11-25 20:53:41 -05:00
parent a9970ae62f
commit 72361b3afe
2 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.c
View File

@@ -53,8 +53,6 @@ int main()
for (int i = 0; i < w; i=+4) { for (int i = 0; i < w; i=+4) {
for (int m = 0; m < d; m++) { for (int m = 0; m < d; m++) {
vx_vec_sgemm_nn(i, m, n, a1, b1, c1, ldc, vsize); vx_vec_sgemm_nn(i, m, n, a1, b1, c1, ldc, vsize);
//d1[i+n*ldc] += a1[m+n*ldc]*b1[m*ldc+i];
vx_vec_sgemm_nn(i, r, c, a1, b1, c1, ldc, vsize);
i = i + vsize; i = i + vsize;
} }
} }
@@ -77,10 +75,10 @@ int main()
} }
#if 1 #if 1
printf("\n\nc[%d]:\n", m*n); printf("\n\nc[%d]:\n", w*d);
for(int i = 0; i < m; ++i) { for(int i = 0; i < w; ++i) {
for(int j = 0; j < n; ++j) { for(int j = 0; j < d; ++j) {
printf("%d ", d1[i*m+j]); printf("%d ", d1[i*w+j]);
} }
printf("\n"); printf("\n");
} }

12
benchmarks/vector/sgemm_nn/vx_vec_sgemm_nn.s
View File

@@ -16,17 +16,19 @@ vx_vec_sgemm_nn:
vsetvli t0, a7, e32 # <--- vsize vsetvli t0, a7, e32 # <--- vsize
mul x11, a6, a2 # n*ldc mul x11, a6, a2 # n*ldc
add x12, x11, a1 # i + (n*ldc) add x12, x11, a1 # i + (n*ldc)
slli x12, x12, 2
add a3, x12, a3 # a[i+ n*ldc] add a3, x12, a3 # a[i+ n*ldc]
lw x13, (a3) lw x13, (a3)
mul x14, a1, a6 # m*ldc mul x14, a1, a6 # m*ldc
add x15, a0, x14 # i + m*ldc add x15, a0, x14 # i + m*ldc
slli x15, x15, 2
add a4, x15, a4 # b[i + m*ldc] add a4, x15, a4 # b[i + m*ldc]
vlw.v v0, (a4) vlw.v v0, (a4)
vmul.vx v2, v1, x13 vmul.vx v2, v1, x13
## lw x6, (a4) ## lw x6, (a4)
# lw x10, (a4) # b ## lw x10, (a4) # b
# mul x11, x3, x10 ## mul x11, x3, x10
mul x6, a2, a6 # n*ldc mul x6, a2, a6 # n*ldc
add x7, a0, x6 # i + n*ldc add x7, a0, x6 # i + n*ldc
@@ -34,7 +36,7 @@ vx_vec_sgemm_nn:
vlw.v v3, (a5) # c vlw.v v3, (a5) # c
vadd.vv v3, v3, v2 vadd.vv v3, v3, v2
vsw.v v3, (a5) vsw.v v3, (a5)
# lw x12, (a5) ## lw x12, (a5)
# add x12, x12, x11 ## add x12, x12, x11
# sw x12, (a5) ## sw x12, (a5)
ret ret