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

Small Matrix: skylakex: sgemm nn: fix n6 conflicts with n4

This commit is contained in:
Wangyang Guo 2021-05-20 11:24:31 +00:00
parent 0ecaa99fc2
commit 5f91668904
1 changed files with 0 additions and 62 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
kernel/x86_64/sgemm_small_kernel_nn_skylakex.c
View File

@ -191,26 +191,6 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alp
STORE_512(0, 4); STORE_512(1, 4); STORE_512(0, 4); STORE_512(1, 4);
STORE_512(0, 5); STORE_512(1, 5); STORE_512(0, 5); STORE_512(1, 5);
} }
for (;j < n4; j += 4) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
DECLARE_RESULT_512(0, 2); DECLARE_RESULT_512(1, 2);
DECLARE_RESULT_512(0, 3); DECLARE_RESULT_512(1, 3);
for (k = 0; k < K; k++) {
LOAD_A_512(0, x); LOAD_A_512(1, x);
BROADCAST_LOAD_B_512(x, 0); BROADCAST_LOAD_B_512(x, 1);
BROADCAST_LOAD_B_512(x, 2); BROADCAST_LOAD_B_512(x, 3);
MATMUL_512(0, 0); MATMUL_512(1, 0);
MATMUL_512(0, 1); MATMUL_512(1, 1);
MATMUL_512(0, 2); MATMUL_512(1, 2);
MATMUL_512(0, 3); MATMUL_512(1, 3);
}
STORE_512(0, 0); STORE_512(1, 0);
STORE_512(0, 1); STORE_512(1, 1);
STORE_512(0, 2); STORE_512(1, 2);
STORE_512(0, 3); STORE_512(1, 3);
}
for (; j < n2; j += 2) { for (; j < n2; j += 2) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0); DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(1, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1); DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(1, 1);
@ -261,27 +241,6 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alp
STORE_512(0, 4); STORE_512(0, 4);
STORE_512(0, 5); STORE_512(0, 5);
} }
for (; j < n4; j += 4) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
DECLARE_RESULT_512(0, 2);
DECLARE_RESULT_512(0, 3);
for (k = 0; k < K; k++) {
LOAD_A_512(0, x);
BROADCAST_LOAD_B_512(x, 0); BROADCAST_LOAD_B_512(x, 1);
BROADCAST_LOAD_B_512(x, 2); BROADCAST_LOAD_B_512(x, 3);
MATMUL_512(0, 0);
MATMUL_512(0, 1);
MATMUL_512(0, 2);
MATMUL_512(0, 3);
}
STORE_512(0, 0);
STORE_512(0, 1);
STORE_512(0, 2);
STORE_512(0, 3);
}
for (; j < n2; j += 2) { for (; j < n2; j += 2) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(0, 1);
@ -335,27 +294,6 @@ int CNAME(BLASLONG M, BLASLONG N, BLASLONG K, FLOAT * A, BLASLONG lda, FLOAT alp
MASK_STORE_512(0, 4); MASK_STORE_512(0, 4);
MASK_STORE_512(0, 5); MASK_STORE_512(0, 5);
} }
for (; j < n4; j += 4) {
DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1);
DECLARE_RESULT_512(0, 2);
DECLARE_RESULT_512(0, 3);
for (k = 0; k < K; k++) {
MASK_LOAD_A_512(0, x);
BROADCAST_LOAD_B_512(x, 0); BROADCAST_LOAD_B_512(x, 1);
BROADCAST_LOAD_B_512(x, 2); BROADCAST_LOAD_B_512(x, 3);
MATMUL_512(0, 0);
MATMUL_512(0, 1);
MATMUL_512(0, 2);
MATMUL_512(0, 3);
}
MASK_STORE_512(0, 0);
MASK_STORE_512(0, 1);
MASK_STORE_512(0, 2);
MASK_STORE_512(0, 3);
}
for (; j < n2; j += 2) { for (; j < n2; j += 2) {
DECLARE_RESULT_512(0, 0); DECLARE_RESULT_512(0, 0);
DECLARE_RESULT_512(0, 1); DECLARE_RESULT_512(0, 1);