[patch] Merge loop-invariant changes from killloop branch
Richard Guenther
richard.guenther@gmail.com
Mon Nov 28 17:06:00 GMT 2005
On 11/28/05, Mark Mitchell <mark@codesourcery.com> wrote:
> Steven Bosscher wrote:
>
> > I can't comment on the state of the dataflow-branch, but removing
> > the old RTL loop optimizer is an independent project that shouldn't
> > have to wait for the dataflow-branch merge.
>
> I don't see any inherent reason that should be true. Sometimes one
> thing depends on another. I think that if the dataflow work was not
> going to be ready for a long time, that would be an issue; but, since we
> do expect it to be ready, we can wait for it, if that's best.
I don't think it is best at all to wait for the dataflow branch.
Instead merging
killloop will allow a lot of cleanups and will get loop-invariant wider testing
so that it may be backported/fixed on the 4.1 branch, too. The earlier we
can do that, the better.
One temporary workaround for a df.c defect should not block the branch
merge at this point.
Please re-consider your position,
Richard.
More information about the Gcc-patches
mailing list