This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] autopar - prepartion step for integration with Graphite
- From: Razya Ladelsky <RAZYA at il dot ibm dot com>
- To: Li Feng <nemokingdom at gmail dot com>
- Cc: gcc-patches at gcc dot gnu dot org, rakdver <rakdver%kam dot mff dot cuni dot _cz at il dot ibm dot com>
- Date: Tue, 21 Jul 2009 17:37:32 +0300
- Subject: Re: [PATCH] autopar - prepartion step for integration with Graphite
Li Feng <nemokingdom@gmail.com> wrote on 21/07/2009 17:05:23:
> Hi Razya,
>
> Thanks for your work.
> Just a little comment.
>
> > + /* FIXME: Only consider innermost loops with just one exit. */
> > + if (loop->inner || !single_dom_exit (loop))
> > + continue;
>
> Since here have judged !single_dom_exit(loop), then I think
> it's not necessary to judge again in:
> try_get_loop_niter (loop, &niter_desc)
> and
> try_create_reduction_list (loop, reduction_list)
Yes.
We could probably just assert it there.
Thanks for reviewing.
Razya
>
> Li
> On Mon, Jul 20, 2009 at 6:58 PM, Razya Ladelsky<RAZYA@il.ibm.com> wrote:
> > Hi,
> >
> > This patch is a preparation patch for the integration of autopar with
> > Graphite framework.
> > It is mainly reorganizing the code in a way that autopar's code
generation
> > could be called
> > even if the analysis wasn't, i.e. building up all needed information
> > independently of the analysis.
> > This will enable performing the dependency analysis in Graphite
framework,
> > and later calling
> > the code generation on the resulting parallel loops.
> > Another change is that canonicalize_loop_ivs was moved to
> > tree-ssa-loop-manip.c, so it can be
> > used externally (Graphite is using it as well).
> >
> > Bootstrap and regressions tests passed successfully (for
c,c++,fortran) on
> > linux powerpc.
> > Ok for mainline?
> >
> > Thanks,
> > Razya
> >
> >
> >
> >