This is the mail archive of the
mailing list for the GCC project.
Re: Possible store motion tweak
- From: law at redhat dot com
- To: Roger Sayle <roger at eyesopen dot com>
- Cc: Richard Sandiford <rsandifo at redhat dot com>, gcc-patches at gcc dot gnu dot org
- Date: Thu, 18 Mar 2004 11:01:09 -0700
- Subject: Re: Possible store motion tweak
- Reply-to: law at redhat dot com
In message <Pine.LNX.email@example.com>, Roger Sa
>I was wondering if you could try bootstraps and regression tests on
>one or two other platforms in addition to mips64-linux-gnu, perhaps
>x86 and powerpc? It might also help if you described your benchmark
>results in more detail, or presented a more comprehensive set of
If the proprietary benchmark license is anything like proprietary testsuite
licenses, then we really can't give lots of detail -- those licenses are
What we have done in the past (and what Richard has done in this case) is
to generate some synthesized code from scratch which shows a simple example
of what the optimization is supposed to do.
>My apologies for my "tingling maintainer senses" but it isn't clear
>why we weren't performing these load/store motions previously. It
>looks like a fabulous improvement, but I worry about breaking bootstraps,
>generating incorrect code or serious performance regression.
The original code was simply trying to be conservative about the kinds of
memory references it tried to optimize. It was always meant to be extended
to handle more non-aliased memory references.
>This patch is Ok for mainline if bootstrapping and regression testing
>on two more platforms is Ok. Once committed, if the automated SPEC
>testers don't show any serious regressions, it can stay. I think its
>unreasonable to ask all contributors to run SPEC, but if you have
>access to it, those numbers would provide confidence before the commit.
How about testing + SPEC on i686-pc-linux-gnu and call it good? Richard has
already tested mips64-pc-linux-gnu and seen good benchmarks on that
platform as well.