This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch 2/2] PR27313 Transform conditional stores
Hi,
On Tue, 15 May 2007, Andrew Pinski wrote:
> On 5/15/07, Richard Guenther <richard.guenther@gmail.com> wrote:
> > and so can be possibly vectorized? Otherwise I would have expected rtl
> > level ifcvt to catch the cmov possibility (which isn't profitable in
> > general).
>
> It is very profitable for SPU, I mean very. Even if you add 10
> selects (cmovs), it is still profitable because a mispredicted branch
> is 40 cycles.
Hmm, wait, you say that making conditional stores unconditional is bad,
but doing the same for conditional loads is good? Wasn't the basis for
the former the fear for load-store hazards in one constructed example? A
similar example can be constructed for the moved load. So, what has
changed?
Ciao,
Michael.