------- Comment #5 from law at redhat dot com 2005-11-29 00:22 -------
Subject: Re: [4.1/4.2 Regression] ICE in
coalesce_abnormal_edges, at tree-outof-ssa.c:646
On Mon, 2005-11-28 at 23:26 +0000, janis at gcc dot gnu dot org wrote:
>
> ------- Comment #4 from janis at gcc dot gnu dot org 2005-11-28 23:26 -------
> A regression hunt using the reduced testcase from comment #3 identified the
> following patch:
>
> http://gcc.gnu.org/viewcvs?view=rev&rev=98066
>
> r98066 | law | 2005-04-13 04:29:40 +0000 (Wed, 13 Apr 2005)
Thanks. This is clearly a problem in in the uncprop code as it
totally ignores abnormal PHIs.
We replace a constant (0) with an SSA_NAME which occurs in an abnormal
PHI. That causes the lifetime of the SSA_NAME to be lengthened, which
is never safe if the SSA_NAME occurs in an abnormal PHI. Eventually
this leads to the abort.
I've got a potential fix, but it'll take until tomorrow before
testing is finished.
jeff
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25000
