This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] Avoid nondeterminism in tree-ssanames
- From: law at redhat dot com
- To: Jan Hubicka <jh at suse dot cz>
- Cc: Geoff Keating <geoffk at geoffk dot org>, gcc-patches at gcc dot gnu dot org
- Date: Mon, 01 Dec 2003 10:49:38 -0700
- Subject: Re: [tree-ssa] Avoid nondeterminism in tree-ssanames
- Reply-to: law at redhat dot com
In message <20031201163929.GX4715@kam.mff.cuni.cz>, Jan Hubicka writes:
>static GTY ((deletable (""))) tree free_ssanames;
>
>To:
>
>static GTY tree free_ssanames;
>
>So we don't get the nondeterminism?
Sure.
>What are yours plans concerning the GGC? (given the fact that I need to
>face similar problems while GTYizing the CFG, I would like to be in
>sync)
I'm not sure what you're asking. But some general thoughts:
1. I think that everyone would agree that ultimately we need to be GC safe
between SSA optimization passes, much like we are GC safe between RTL
optimization passes.
2. We desparately need a way to tell the GC system that we are done with
a node. SSA_NAMEs, PHI_NODEs are common examples, as are certain temporary
nodes optimizers create. Another would be virtual arrays which are resized.
This allows us to remove the SSA_NAME and PHI_NODE managers since the only
reason those exist is to re-use nodes we know are available, but which
hasn't been made available for re-use at the GC level.
jeff