This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran 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: Performance testing

From: Scott Robert Ladd <coyote at coyotegulch dot com>
To: "Ole W. Saastad" <ole at scali dot com>
Cc: fortran at gcc dot gnu dot org
Date: Wed, 10 Nov 2004 14:07:36 -0500
Subject: Re: Performance testing
References: <1100067350.9865.19.camel@pc-2.office.scali.no>

Ole W. Saastad wrote:

Over time I have seen the gnuforan change from a state where none of my test would compile and run to a state today where it is an alternative to Portland and Intel compilers.

I'm glad to hear it!

However, the performance is not optimal. A simple test
of a molecular dynamics code (MPI code) showed numbers
like 20.06s and 20.72s for gfortran while 13.8s for Intel
ifc 7.1. It must be said that Intel fortran 8.0 and 8.1
failed to run, segv after a few seconds run time.

Your experience has been much like mine. In general, code generation is sub-optimal at best; gfortran (actually, GCC in general) lacks good vectorization support, thought it is being worked on.

My biggest problem has been broken code generation, especially on x86_64 (Opteron) and when using SSE2. I'm not a code generation guru by any stretch of the imagination, but I'm digging into some of the more egregious problems as time permits.

Which version of MPI have you used with gfortran? I've found MPICH to be most uncooperative.

..Scott

--
Scott Robert Ladd
site: http://www.coyotegulch.com
blog: http://chaoticcoyote.blogspot.com

References:
- Performance testing
  - From: Ole W. Saastad

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