On Wed, Sep 14, 2016 at 09:35:50AM +0200, Richard Biener wrote:
> On Tue, Sep 13, 2016 at 5:29 PM, Segher Boessenkool
> <seg...@kernel.crashing.org> wrote:
> > On Tue, Sep 13, 2016 at 10:14:57AM -0500, Pat Haugen wrote:
> >> PRs 77536 and 68212 document cases where the estimated frequencies become
> >> unreliable to make appropriate decisions based on them. This patch removes
> >> the optimize_insn_for_speed_p check on the splitter so that the
> >> transformation will take place when -ffast-math is specified.
> >>
> >> Bootstrap/regtest on powerpc64le, ok for trunk?
> >
> > Okay for trunk. Please mention the PRs in the changelog, i.e.:
>
> Maybe you instead want !optimize_insn_for_size then? (if such thing
> exists?) Otherwise you're
> triggering this for -Os.
That is the same as was there before, see predict.c:
bool
optimize_insn_for_speed_p (void)
{
return !optimize_insn_for_size_p ();
}
The best solution of course is to have sane profiles always, but that
doesn't always happen currently (notably with the unrollers). Pat left
a comment with the code so we can put that check back when the profile
is usable again.
Segher
