Enable AVX support with Intel compilers #8372
Conversation
icc does not define the __AVX*__ macros if the corresponding -m architecture flag was not provided. Thus, make sure we always provide it for icc (not not necessarily for gcc). Signed-off-by: George Bosilca <[email protected]>
ggouaillardet
left a comment
ggouaillardet
left a comment
There was a problem hiding this comment.
minor typo in the first commit message ("not not"), otherwise looks good to me
|
@rajachan Can you try this patch with your app performance test? |
|
Corresponding 4.1.x PR: #8373 |
rajachan
left a comment
rajachan
left a comment
There was a problem hiding this comment.
Tested the PR, looks good to me. I no longer observe the performance degradation when op/avx component is used.
|
@rajachan Just curious -- is AVX2 being used? And if so, do you see any performance benefit? |
|
This PR disables AVX2 as well. Assuming you were asking about AVX, I see a marginal (~1%) improvement in the loop time for the specific test case I was running. I have not evaluated microbenchmarks yet. |
bosilca
force-pushed
the
topic/icc_avx
branch
from
db1a271 to
a2041be
Compare
|
This PR only fixes the support for icc. The restrictions in AVX support are now in #8376. |
|
The icc fix is good to go, thanks for splitting out the restrictions change to a separate PR. |
4 participants
Refs #8334