This is the mail archive of the
mailing list for the GCC project.
Re: [RFA:] Fix PR optimization/15296, delayed-branch-slot bug.
- From: Hans-Peter Nilsson <hans-peter dot nilsson at axis dot com>
- To: law at redhat dot com
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Wed, 5 May 2004 18:54:17 +0200
- Subject: Re: [RFA:] Fix PR optimization/15296, delayed-branch-slot bug.
> From: email@example.com
> Date: Wed, 05 May 2004 10:20:46 -0600
Thanks for the extremely quick review!
> I agree that reorg needs to be rewritten and I strongly feel any rewrite
> ought be centered around building a real data dependency graph (or re-using
> one from a previous pass such as sched2). Given a real data dependency graph,
> most of the stuff reorg does becomes simpler and faster (consider something
> like 20001226-1.c on most delay slot targets).
Yeah, you've mentioned that once or twice. ;-)
(I have this futile hope that I'll get to fix it some time.
Maybe it'll actually happen this year.)
> The testcase at least at first glance looks like it may not work for our
> 16bit targets.... You might want to restrict it to working on 32/64 bit
Something in particular that looks nonworking? I believe the
assumption that "void *" is assignment-compatible with
"int __attribute__ ((mode (__pointer__)))" is safe, but is that all?
I've started a h8300-elf run to see if it fails there.
Is that a sufficient test for approval?