This is the mail archive of the
mailing list for the GCC project.
Re: Speedup into SSA (not committed) for PR 8361, PR 15855, PR13766
- From: Jeffrey A Law <law at redhat dot com>
- To: Diego Novillo <dnovillo at redhat dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, Mark Mitchell <mark at codesourcery dot com>
- Date: Fri, 10 Dec 2004 11:13:08 -0700
- Subject: Re: Speedup into SSA (not committed) for PR 8361, PR 15855, PR13766
- Organization: Red Hat, Inc
- References: <41B8DA6D.firstname.lastname@example.org>
- Reply-to: law at redhat dot com
On Thu, 2004-12-09 at 18:06 -0500, Diego Novillo wrote:
> This patch adds a shortcut to avoid processing statements that don't
> need renaming when the renamer is called multiple times.
> The compile time benefits of this patch are measurable, and in the
> process it uncovered a couple of bugs in SRA and the vectorizer. These
> are the compile time speedups I got with this patch:
> Into SSA Overall
> cc-i-files +18.3% +0.2%
> DLV (PR 8361) +15.9% +3.3%
> MICO (PR 13766) +21.9% +2.3%
> tramp3d (PR ?) +20.9% +7.8%
> PR 15855 +26.0% +1.4%
> Mark, I'm deferring to you. The patch is only a compile time
> improvement and fixes no regressions. If you prefer, I can hold on to
> it until mainline is out of the regression-only state.
I'd say go for it. It's clearly attacking compile-time regressions
and should not affect the generated code. But if you'd prefer to wait
for Mark to chime in, then that's OK with me.