Re: [PATCH] Extend dse

["Richard Guenther" <richard dot guenther at gmail dot com>]

On 4/29/07, Jeffrey Law <law@redhat.com> wrote:
> On Sun, 2007-04-29 at 09:54 +0300, Revital1 Eres wrote:
> >
> > gcc-patches-owner@gcc.gnu.org wrote on 26/04/2007 21:44:51:
> >
> > > Revital1 Eres wrote on 03/23/07 07:34:
> > >
> > > > I would like to propose the following patch to mainline:
> > > >
> > > > http://gcc.gnu.org/ml/gcc-patches/2007-03/msg01257.html
> > >
> > > After you re-work the store-ccp extension that we discussed earlier
> > > today, this will probably not be needed.  Regular DCE should catch these
> > > cases, right?
> >
> > AFAICT regular DCE nor DSE eliminate cases like the following which this
> > patch aims to fix:
> >
> > S[5].x = 0;
> > S[5].y = 0;
> >
> > S[5].x = x;
> > S[5].y = y;
> Umm, I would think the existing DSE code can/should eliminate the
> dead stores in this case if the aliasing information is accurate
> enough to allow disambiguation.  What makes you think it can't/won't?

Because 5 > 4, so we don't split array elements (and the element structures)
into sub-variables for aliasing.  So aliasing cannot tell S[5].x apart from
S[5].y.

Richard.