This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Loop optimizer issues
- From: Jan Hubicka <jh at suse dot cz>
- To: David Edelsohn <dje at watson dot ibm dot com>
- Cc: Jan Hubicka <jh at suse dot cz>,"S. Bosscher" <S dot Bosscher at student dot tudelft dot nl>,'Zdenek Dvorak ' <rakdver at atrey dot karlin dot mff dot cuni dot cz>,'Richard Henderson ' <rth at redhat dot com>,'Geoff Keating ' <geoffk at geoffk dot org>,'Diego Novillo ' <dnovillo at redhat dot com>,'Jason Merrill ' <jason at redhat dot com>,'Daniel Berlin ' <dberlin at dberlin dot org>,"'pop at gauvain dot u-strasbg dot fr '" <pop at gauvain dot u-strasbg dot fr>,"'gcc at gcc dot gnu dot org '" <gcc at gcc dot gnu dot org>
- Date: Thu, 3 Jul 2003 10:48:59 +0200
- Subject: Re: Loop optimizer issues
- References: <20030702223328.GF9134@kam.mff.cuni.cz> <200307022243.SAA33174@makai.watson.ibm.com>
> >>>>> Jan Hubicka writes:
>
> Jan> Zdenek is speaking about the invariant motion. Clearly there are many
> Jan> cases where RTL expansion will bring oppurtunities for invariant motion,
> Jan> so these needs to be re-done on the RTL.
>
> Jan> Concerning libcalls, they should not exist in post-tree-SSA world as
> Jan> they are just papering over the problem that such complex things should
> Jan> have been moved earlier.
>
> How about starting with a Tree-SSA loop optimization
> infrastructure and then back-filling RTL loop optimizations where they are
> profitable instead of guessing?
Majority of loop optimization is actually about thinigs that can be done
in a way independent on IL. I am not sure whether Tree-SSA is quite
ready for playing with things like, say, loop unrolling when there are
still no ways to duplicate basic blocks and redirect edges.
So I believe that testing some of code on current RTL infrastructure
helps to get things done in parallel and tree-SSA will benefit from
tested algorithms.
Also once tree-SSA is in (with the syntactic control flow reduction pass
I plan to implement), we will likely RTL loop notes, so old loop
optimizer must be performance NIL at that point in case we don't want to
see performance rgressions in 3.5.
I hope to get into tree-SSA CFG manipulation soonish, but in reality not too
soonish as my recent days were entirely consumed by unit-at-a-time work
and I am leaving for roughly one month.
Honza
>
> David