This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Polyhedron benchmark on Opteron
- From: Tim Prince <timothyprince at sbcglobal dot net>
- To: François-Xavier Coudert <fxcoudert at gmail dot com>
- Cc: Fortran List <fortran at gcc dot gnu dot org>, franke dot daniel at gmail dot com, burnus at net-b dot de
- Date: Sat, 30 Sep 2006 08:18:31 -0700
- Subject: Re: Polyhedron benchmark on Opteron
- Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=s1024; d=sbcglobal.net; h=Received:Message-ID:Date:From:Reply-To:User-Agent:MIME-Version:To:CC:Subject:References:In-Reply-To:Content-Type:Content-Transfer-Encoding; b=m/Y35fH7ie2GksoYJLpkG8yZnGE6XhnvxBzBxtiQe02JvxFUDFi8LNcyw+ALaYCW5cs9+XyLlFtPMj91cFJpN6RJcl/n9cHDghFSbLHS6jjqQU/QcSZDskILmT6+TaG0BJNsG7qjRF/ehBY2QbAOe+s159fKWr10Iro1CCTOnE0= ;
- References: <19c433eb0609290713x64f74089m45e5ea291343e1d7@mail.gmail.com>
- Reply-to: tprince at myrealbox dot com
François-Xavier Coudert wrote:
I wanted to report some results for the Polyhedron benchmark** on
Opteron (hardware details at the bottom of this mail).
Unfortunately, there are also tests for which Intel is a clear winner:
-- aermod, by 44%
-- air, by 30%
-- nf, by 29%
The differences aren't so large on the Core 2 Duo.
gprof profiling doesn't capture the deficits in gfortran performance.
In aermod, with either ifort or gfortran, less than 50% of the time is
spent in the code covered by gprof profiling. Most of the time spent by
the ifort version of air is captured in gprof profile, but the gfortran
version spends significant unaccounted time. The profile for the
gfortran version of nf allocates by far the largest time segment to a
mysterious frame_dummy symbol.
Some day, I'll check aermod and air with VTune, to look for the
mysterious extra time.