projects
/
platform
/
upstream
/
openblas.git
/ commitdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
| commitdiff |
tree
raw
|
patch
| inline |
side by side
(parent:
b37e5fa
)
Disable FMA intrinsics in the srot kernel when the compiler is PGI/NVIDIA
author
Martin Kroeker
<martin@ruby.chemie.uni-freiburg.de>
Sat, 19 Dec 2020 21:15:58 +0000
(22:15 +0100)
committer
GitHub
<noreply@github.com>
Sat, 19 Dec 2020 21:15:58 +0000
(22:15 +0100)
kernel/x86_64/srot.c
patch
|
blob
|
history
diff --git
a/kernel/x86_64/srot.c
b/kernel/x86_64/srot.c
index 3de586cb8e8ab38c613fb98ddcda2f7d1540cd6c..3264d251aad3f7ea93e3bb201b65b365f5bbe464 100644
(file)
--- a/
kernel/x86_64/srot.c
+++ b/
kernel/x86_64/srot.c
@@
-13,7
+13,7
@@
static void srot_kernel(BLASLONG n, FLOAT *x, FLOAT *y, FLOAT c, FLOAT s)
{
BLASLONG i = 0;
-#if V_SIMD && (defined(HAVE_FMA3) || V_SIMD > 128)
+#if V_SIMD &&
!defined(C_PGI) &&
(defined(HAVE_FMA3) || V_SIMD > 128)
const int vstep = v_nlanes_f32;
const int unrollx4 = n & (-vstep * 4);
const int unrollx = n & -vstep;