This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: SAVE_EXPRs still there at RTL-expansion time?
- From: Richard Henderson <rth at redhat dot com>
- To: Paolo Bonzini <bonzini at gnu dot org>
- Cc: gcc at gcc dot gnu dot org
- Date: Fri, 2 Jul 2004 09:34:52 -0700
- Subject: Re: SAVE_EXPRs still there at RTL-expansion time?
- References: <cc3kcb$uai$1@sea.gmane.org>
On Fri, Jul 02, 2004 at 02:31:08PM +0200, Paolo Bonzini wrote:
> I see that even though SAVE_EXPRs are not GIMPLE, they are still found
> by the RTL expander. I'd think that they are not in the subset of
> GENERIC that is produced at the end of tree optimizations...
They are produced by expand_builtin_foo. :-/
> Only a few of them really need SAVE_EXPRs semantics: if I abort in
> SAVE_EXPR's expansion, stage2 dies almost instantly, while if I abort on
> SAVE_EXPR_RTL (x) != 0 it takes a while before I hit it.
I have a patch in testing that expunges most of the ugly bits
of SAVE_EXPRs. Removes the 2nd and 3rd arguments, gets rid of
a lot of the machinery to handle them at the rtl level.
r~