[dataflow] [RFC] Remove many almost useless DCE passes
Richard Guenther
richard.guenther@gmail.com
Tue Feb 27 17:44:00 GMT 2007
On 2/27/07, Vladimir Makarov <vmakarov@redhat.com> wrote:
> Paolo Bonzini wrote:
>
> >
> >> Then you are probably moving in wrong direction worsening the
> >> generated code. It is not only performance. The generated code size
> >> for x86_64 (SPECInt2000) was 0.5% bigger on the branch before the
> >> patch and I am sure it will be even bigger after the patch.
> >
> >
> > The patch will never be applied. It was not meant to be.
> >
> Good because SPECFP2000 for ppc64 is 0.5% better (SPECInt2000 is the
> same) on the branch than on the mainline at the last merge point. And
> this patch could remove this achievement.
There's now daily SPEC2k and C++bench (tramp3d, DLV and others) numbers from
dataflow branch available from
http://www.suse.de/~gcctest/SPEC/C{INT|FP}/sb-haydn-df-64/
and
http://www.suse.de/~gcctest/c++bench-haydn-df/
(or follow the links from gcctest's home page)
Richard.
More information about the Gcc-patches
mailing list