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

Fix thinko in skylake beta handling 

casting ints is cheaper but it has a rounding, not memory casing effect, resulting in
invalid outcome
This commit is contained in:
Arjan van de Ven 2018-12-24 18:49:50 +00:00
parent e1eab96502
commit 795285c587
2 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
kernel/x86_64/dgemm_beta_skylakex.c
View File

@ -50,7 +50,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT beta,
FLOAT ctemp5, ctemp6, ctemp7, ctemp8;
/* fast path.. just zero the whole matrix */
if (m == ldc && (unsigned long)beta == (unsigned long)ZERO) {
if (m == ldc && beta == ZERO) {
memset(c, 0, m * n * sizeof(FLOAT));
return 0;
}

2
kernel/x86_64/sgemm_beta_skylakex.c
View File

@ -50,7 +50,7 @@ int CNAME(BLASLONG m, BLASLONG n, BLASLONG dummy1, FLOAT beta,
FLOAT ctemp5, ctemp6, ctemp7, ctemp8;
/* fast path.. just zero the whole matrix */
if (m == ldc && (unsigned long)beta == (unsigned long)ZERO) {
if (m == ldc && beta == ZERO) {
memset(c, 0, m * n * sizeof(FLOAT));
return 0;
}