This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
RE: [tree-ssa] SSA bug or feature?
- From: "S. Bosscher" <S dot Bosscher at student dot tudelft dot nl>
- To: 'Pop Sébastian ' <pop at gauvain dot u-strasbg dot fr>,"'gcc at gcc dot gnu dot org '" <gcc at gcc dot gnu dot org>
- Date: Thu, 4 Sep 2003 14:21:27 +0200
- Subject: RE: [tree-ssa] SSA bug or feature?
Sebastian Pop wrote:
> Recently I've ran into a case where the arguments of two phi nodes
> reference each other, creating an unusual loop in the SSA representation:
>
> normal_call_insns_8 = PHI <0B(380), normal_call_insns_9(782)>;
> ...
> normal_call_insns_9 = PHI <normal_call_insns_8(773), T.1529_2087(777),
> T.1529_2087(778)>;
>
> This case showed up during the analysis of the expand_call function in
> gcc/gcc/calls.c.
>
> Is this a valid SSA code, or is the SSA screwed up by one of the
> optimizers?
This means that one of them is used before it is defined, so how
could it possibly be valid? Is this just after rewriting into SSA,
or after one or more ssa passes?
Gr.
Steven