This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/56309] conditional moves instead of compare and branch result in almost 2x slower code
- From: "rguenther at suse dot de" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Fri, 18 Oct 2013 11:47:11 +0000
- Subject: [Bug target/56309] conditional moves instead of compare and branch result in almost 2x slower code
- Auto-submitted: auto-generated
- References: <bug-56309-4 at http dot gcc dot gnu dot org/bugzilla/>
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56309
--- Comment #31 from rguenther at suse dot de <rguenther at suse dot de> ---
On Fri, 18 Oct 2013, ubizjak at gmail dot com wrote:
> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56309
>
> --- Comment #30 from Uro? Bizjak <ubizjak at gmail dot com> ---
> (In reply to Jakub Jelinek from comment #29)
>
> > On the by-ref-O3.ii testcase there is no change, but by-val-O3.ii seems to
> > be consistently faster, around 20% on AMD 8354 and 35% on Intel 2600.
>
> According to detailed analysis in Comment 26, perhaps a new tuning flag or
> target hook could be used for if-conversion of scalars. Also, I think that -Os
> can still use cmoves due to smaller code size.
But RTL if-conversion still will generate cmov, right?