This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] Fix properties for pass manager
- From: Richard Guenther <richard dot guenther at gmail dot com>
- To: Steven Bosscher <stevenb dot gcc at gmail dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Tue, 10 Mar 2009 11:31:49 +0100
- Subject: Re: [patch] Fix properties for pass manager
- References: <571f6b510903091537n1d8090c9ied25a7b27e3d2767@mail.gmail.com>
On Mon, Mar 9, 2009 at 11:37 PM, Steven Bosscher <stevenb.gcc@gmail.com> wrote:
> Hi,
>
> There are lies, damn lies, and GCC passes.
>
> GCC has passes that falsely claim to provide properties, just to avoid
> crashing in the properties checkers. The problem is the pass_lower_omp
> pass, which has a gate function. ?The result is that PROP_gimple_lomp
> is never set unless -fopenmp is given. ?But some passes require
> PASS_gimple_lomp, via the fake property PROP_trees which is actually a
> union of properties that are available after fully lowering GIMPLE.
> So we had IPA passes cheating, claiming to provide "PROP_trees".
>
> I believe there should be a general rule for the pass properties that
> GCC currently does not honor: There should be exactly one pass that
> provides a property. ?With the attached patch applied, this is
> actually the case, but it is not checked (nor enforced via the
> existing properties checking). ?I intend to submit a patch for 4.5
> that adds this checking. ?Comments/thoughts on this idea?
It certainly makes sense.
> Patch is still bootstrapping on ia64. ?OK for GCC 4.5 if
> bootstrap/testing finishes successfully?
Looks good. Ok.
Thanks,
Richard.
> Gr.
> Steven
>