This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] Run copyrename one extra time before out-of-ssa


On Friday 21 January 2005 18:59, Mark Mitchell wrote:
> Jeffrey A Law wrote:
> >>May I interpret your comments as "OK" after bootstrapping/testing? ;-)
> >
> > I can't speak for Andrew, but it would have my support.
>
> In theory, this makes sense, but we need more analysis than an isolated
> testcase.  Please obtain benchmark numbers on at least one, and
> hopefully two or three, platforms.  If this is not a demonstrable win,
> then let's not do it.

OK, so that went a lot smoother than I expected...

Bootstrapped and tested on i686, ia64, x86_64, ppc64, and ppc.  There
are no new regressions, of course.

Benchmark results are not all in yet, but for AMD64 I have SPECint and
SPECfp numbers, at -O2.  Here is a summary of the results:

- Not one binary was bigger, and a few were smaller with the patch.
  Overall the binaries are ~0.5% smaller.
- Compile time was 2s faster on 194 seconds for the unpatched compiler.
  Let's call it compile time neutral.
- For SPECint there is a 5% win for twolf, all other benchmarks are
  within 0.5%.
- For SPECfp there is a 30% improvement for mgrid (!), and 6% for mesa.
  The rest is in the noise (within 0.4%).

Good enough?

Gr.
Steven


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]