[patch] enabling vectorization by default at -O3

Dorit Nuzman DORIT@il.ibm.com
Tue Sep 18 12:19:00 GMT 2007


committed.

dorit

> >    I think we should be go to go now as Dorit seems to have fixed all
> > of the ICEs and we now have PR26449 fixed in gcc trunk as well (which
> > I believe were the only holdups to enabling vectorization as the
> > default for -O3).
>
> yes, I started a bootstrap + make check cycle and hope to be able to
commit
> the patch tomorrow.
>
> > I also thought there was some discussion of enabling
> > a more restricted subset of vectorization for -O2 as well.
>
> I think this can be done as a separate step
>
> dorit
>
> >                   Jack
> >
> > On Mon, Sep 17, 2007 at 03:14:55PM -0400, Michael Meissner wrote:
> > > On Thu, Sep 13, 2007 at 11:49:51AM -0700, Mark Mitchell wrote:
> > > > Dorit Nuzman wrote:
> > > >
> > > > > SpecCPU2006 on powerpc970 shows 1.5% on specINT (mostly due to a
> 14%
> > > > > improvement on hmmer), and a -0.4% on specFP. This was
> > comparing '-O3' with
> > > > > '-O3 -ftree-vectorize -maltivec'. The costmodel wasn't on, cause
> it is
> > > > > still undergoing some changes, and also it is not tuned yet
> > for powerpc970,
> > > > > so I hope that with the costmodel on we'll be able to get rid
> > of the small
> > > > > degradations we have here and there later on.
> > > >
> > > > I was hoping for a bit more of an overall win.  The negative result
> on
> > > > specFP worries me; although the overall scores are higher, it would
> be
> > > > nice if we could eliminate that negative.
> > > >
> > > > Do any of the other optimization maintainers have a feeling about
> this?
> > > >  Should we turn on vectorization at -O3, or are we not quite there
> yet?
> > >
> > > IMHO, the best way to getting the bugs fixed is enabling it to come
on
> by
> > > default.  My sense is for x86 platforms, it is more of a win, but
> > I don't have
> > > the numbers at hand.
> > >
> > > --
> > > Michael Meissner, AMD
> > > 90 Central Street, MS 83-29, Boxborough, MA, 01719, USA
> > > michael.meissner@amd.com
> > >
>



More information about the Gcc-patches mailing list