------- Additional Comments From law at redhat dot com 2004-12-29 21:06 -------
Subject: Re: [4.0 Regression] out-of ssa
causing loops to have more than one BB
On Wed, 2004-12-29 at 20:50 +0000, pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org 2004-12-29
> 20:50 -------
> The reference patch does not fix the problem all the time, see PR 14741 for a
> testcase where the patch
> fails, and it fails still for sixtrack but I should note that this would be
> now caused by IV-OPTs.
I don't expect to fix it all the time. There's a fundamental problem in that
copy propagation can lengthen lifetimes and thus require a copy at
out-of-ssa time. The patch is meant to fix those cases which can be
trivially detected and for which we can generate better code.
Changing the out-of-ssa code to use Sreedhar's algorithm (which appears
to do a lot better at avoiding copies) isn't something I'd advocate
for 4.1. But even with a better out-of-ssa algorithm there are still
going to be cases where out-of-ssa is going to have to generate a
copy. No way around it.
Now if you think that PR is a trivial case that should be caught, then,
show me why and I'll take a closer look.
jeff
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19038
