This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PING]^2 New if-combining pass
On Sun, 10 Jun 2007, Richard Guenther wrote:
> On 6/10/07, Mark Mitchell <mark@codesourcery.com> wrote:
> > Richard Guenther wrote:
> > > On 5/17/07, Richard Guenther <rguenther@suse.de> wrote:
> > > >
> > > > [PATCH] New if-combining pass, addresses PRs 15353 and 31657
> > > > http://gcc.gnu.org/ml/gcc-patches/2007-04/msg01703.html
> > > >
> > > > (with pass-oder swapped)
> >
> > This patch is OK.
> >
> > What does this patch do for code size on (say) CSiBE when compiling with
> > -Os? I'm not concerned about the particular benchmark, but I would like
> > to understand if there's any size impact. (I can well imagine it would
> > be helpful for -Os: eliminating control flow instructions in lieu of a
> > logical instruction. However, on an architecture with predication, it
> > might be harmful if we introduce a logical instruction where we could
> > otherwise use predication.
>
> I'll do a CSiBE compare tomorrow.
Results are (x86_64, -Os) 3473918 bytes with the patch and 3473968 without
(that's a 0.0015% improvement). Compile and runtimes are slightly better
with the patch but I consider that noise (the machine was not completely
idle).
I'll re-post what I commit (I had to do some adjustments due to the cfg
changes and improved commentary while at it) after the df-branch merge
settled.
Richard.