This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Status of new graph-coloring register allocator?
- From: Denis Chertykov <denisc at overta dot ru>
- To: Michael Matz <matz at suse dot de>
- Cc: Nathanael Nerode <neroden at doctormoo dot dyndns dot org>, Andrew Pinski <pinskia at physics dot uc dot edu>, <gcc at gcc dot gnu dot org>
- Date: 29 Oct 2002 21:00:24 +0300
- Subject: Re: Status of new graph-coloring register allocator?
- References: <Pine.LNX.4.33.0210291300021.20702-100000@wotan.suse.de>
Michael Matz <matz@suse.de> writes:
> Hi,
>
> On Mon, 28 Oct 2002, Andrew Pinski wrote:
>
> > > It looks (unsurprisingly) like a lot of the nastiest and hardest-to-fix
> > > bugs are showing up in reload.
> > >
> > > The new register allocator has the potential to get rid of reload.
> > >
> > > Hence, I'm curious about its current status; if it's close to ready
> > > (i.e. very likely to be in 3.4), we may not want to work too hard on
> > > reload bugs in the meantime.
> >
> > It will be in 3.3 but not turned on by default, use -fnew-ra to turn it
> > one.
>
> That, and reload will still be used in 3.3. On the new-regalloc-branch
> reload is also used, but it hasn't much to do anymore, and only is there
> for particularly nasty situations. (Well, the routines in reload do also
> some more, like register elimination and caller-save handling and _those_
> thing will anyway stay in the one or other form). In my local version
> reload has even less to do, but I've not yet begun to delete hunks
> of it.
Are you support register elimination in your internal version ?
Are you support web splitting for right regclass choosing ?
Denis.