This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] for PR23855
On 1/21/06, Zdenek Dvorak <rakdver@atrey.karlin.mff.cuni.cz> wrote:
> Hello,
>
> > >For these reasons, I believe having this common special case handled
> > >separately makes sense. Also, this makes it easy to have this
> > >transformation performed very early in the compilation process, thus
> > >making it possible for other optimizations to benefit from it.
> > >
> > >Bootstrapped & regtested on ppc and ia64.
> >
> > Do you have performance numbers because PR 23855 blocks PR 23970 which
> > is
> > a reduced testcase from SPEC?
> > Yes this should help SPEC but knowing how much would be helpful.
>
> I have checked the testcase from PR23970, it seems that the patch has
> the desired effect (loads of stride* are moved completely out of loops,
> multiplication by k to the outermost loop, etc.). For the effects on
> specs, see below (base without the patch, peak with, both -O2
> -march=athlon).
>
> Btw. which benchmark is testcase from PR 23970 supposed to be extracted
> from? I would expect more significant improvement in that benchmark, so
> I wanted to have a look at it, but I could not find anything resembling
> the code.
It's of course extracted from tramp3d ;) I'll try the patch on it in a second.
Thanks for the work!
Richard.