GCC on SPEC.org

GCC on SPEC.org

[Dara Hazeghi]

Hello,

in light of the recent discussion of gcc performance
(with particular emphasis on comparison with Intel's
compiler), the 3rd and 4th entries at
<http://www.spec.org/cpu2000/results/res2003q4/> may
be interesting. Perhaps we aren't so far off from
parity, at least in some areas? Cheers,

Dara

__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus

Re: GCC on SPEC.org

[Andreas Jaeger]

[-- Attachment #1: Type: text/plain, Size: 913 bytes --]

Dara Hazeghi <dhazeghi@yahoo.com> writes:

> Hello,
>
> in light of the recent discussion of gcc performance
> (with particular emphasis on comparison with Intel's
> compiler), the 3rd and 4th entries at
> <http://www.spec.org/cpu2000/results/res2003q4/> may
> be interesting. Perhaps we aren't so far off from
> parity, at least in some areas? Cheers,

Those are AMD64 platforms and GCC was the first compiler for that
platform, we had some advantage over other vendors - and the GCC
developers optimizing for it did a great job!

I suggest to compare the following platforms instead to see how good
we are:
- Intel x86: icc vs. gcc
- Intel IPF: icc vs. gcc
- Alpha: A system compiler vs. gcc

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Maxfeldstr. 5, 90409 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: GCC on SPEC.org

[Vladimir Makarov]

Dara Hazeghi wrote:
> 
> Hello,
> 
> in light of the recent discussion of gcc performance
> (with particular emphasis on comparison with Intel's
> compiler), the 3rd and 4th entries at
> <http://www.spec.org/cpu2000/results/res2003q4/> may
> be interesting. Perhaps we aren't so far off from
> parity, at least in some areas? Cheers,
> 

I'd take the comparison with Intel's compiler with a grain of salt.  The
testing was made for AMD64 not for Intel P4.  Intel is focused to
improve code mainly for their processors.

There is one area where we could try to achieve a better results.  It is
a peak performance which permits to use a specific set of options for
each SPEC test and characterizes variety and quality of compiler
optimizations.  Gcc has much more optimization options than Intel's
one.  So there is a possibility to improve SPEC peak performance.

Actually it is another argument to add tree-ssa for mainline.  It could
aid to improve peak performance too.

Vlad

Re: GCC on SPEC.org

[Dara Hazeghi]

--- Vladimir Makarov <vmakarov@redhat.com> wrote:
> Dara Hazeghi wrote:
> > Perhaps we aren't so far off from
> > parity, at least in some areas? Cheers,
> > 
> I'd take the comparison with Intel's compiler with a
> grain of salt.  The
> testing was made for AMD64 not for Intel P4.  Intel
> is focused to
> improve code mainly for their processors.

This is true. But on the other hand, Intel also tunes
their compiler for SPEC a lot more than we've done
(do?). Anyhow, I thought it was a nice sign that we
are a preferred SPEC compiler on at least one arch...

> 
> There is one area where we could try to achieve a
> better results.  It is
> a peak performance which permits to use a specific
> set of options for
> each SPEC test and characterizes variety and quality
> of compiler
> optimizations.  Gcc has much more optimization
> options than Intel's
> one.  So there is a possibility to improve SPEC peak
> performance.

While true, the other side is that unfortunately, to
make something run optimally with gcc takes a lot more
tweaking. It'd be awfully nice if we had a -fast that
did what their -O2 did for SPEC, but that's an
argument for another day...

Dara

__________________________________
Do you Yahoo!?
Yahoo! Hotjobs: Enter the "Signing Bonus" Sweepstakes
http://hotjobs.sweepstakes.yahoo.com/signingbonus