* Avoid generating vzeroupper instruction
@ 2012-05-02 13:44 Gunther Piez
2012-05-02 14:50 ` Ian Lance Taylor
0 siblings, 1 reply; 2+ messages in thread
From: Gunther Piez @ 2012-05-02 13:44 UTC (permalink / raw)
To: gcc-help
Hello,
is there a way to tell gcc to not emit the `vzeroupper` instruction
while using -march=corei7avx? I do realize that this instruction is
neccessary if legacy 128 bit SSE instructions are intermixed with AVX
128/256 bit code, like calling a libc function which makes use of SSE in
a non-AVX glibc or the other way around.
But in a pure AVX code environment I noticed it actually does hurt
performance. So how can I tell gcc to not emit vzeroupper while
generating AVX instructions?
- Gunther
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: Avoid generating vzeroupper instruction
2012-05-02 13:44 Avoid generating vzeroupper instruction Gunther Piez
@ 2012-05-02 14:50 ` Ian Lance Taylor
0 siblings, 0 replies; 2+ messages in thread
From: Ian Lance Taylor @ 2012-05-02 14:50 UTC (permalink / raw)
To: Gunther Piez; +Cc: gcc-help
Gunther Piez <gupiez@googlemail.com> writes:
> is there a way to tell gcc to not emit the `vzeroupper` instruction
> while using -march=corei7avx? I do realize that this instruction is
> neccessary if legacy 128 bit SSE instructions are intermixed with AVX
> 128/256 bit code, like calling a libc function which makes use of SSE in
> a non-AVX glibc or the other way around.
>
> But in a pure AVX code environment I noticed it actually does hurt
> performance. So how can I tell gcc to not emit vzeroupper while
> generating AVX instructions?
Try -mno-vzeroupper.
Ian
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2012-05-02 14:50 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-05-02 13:44 Avoid generating vzeroupper instruction Gunther Piez
2012-05-02 14:50 ` Ian Lance Taylor
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).