[Beowulf] Re: Table of best compiler flags for different CPUs?

[Steve Cousins]

On Wed, 11 May 2005, Tom Crick wrote:

> Not sure if this is exactly what you are looking for, but I've
> previously looked at a piece of software called ACOVEA (Analysis of
> Compiler Options via Evolutionary Algorithm) by Scott Robert Ladd
> (http://www.coyotegulch.com/products/acovea/).
> 
> This attempts to find the 'optimal' set of compiler flags for GCC
> (recently updated for GCC 4.0) using genetic algorithms. This seems to
> be particularly relevant for GCC, with the abundance of existing flags
> and options.

This is very interesting.  It seems to show that GCC has the defaults set
more towards the P4/Xeon than the Opteron.  That is, there is less
improvement in fiddling with the more esoteric compiler switches for the
P4/Xeon than for the Opteron.  In turn, this makes the Opteron even more
impressive since it beats the P4/Xeon quite handily even when it is not
completely optimized.

Thanks for the link.

Steve
 
> Cheers,
> 
> Tom
> 
> -- 
> Tom Crick
> Mathematical Logic & Symbolic Computation Group
> Department of Computer Science
> University of Bath
> tc at cs.bath.ac.uk
> http://www.cs.bath.ac.uk/~tc
> 
> On Tue, 2005-05-10 at 20:00, beowulf-request at beowulf.org wrote:
> > ----------------------------------------------------------------------
> > 
> > Message: 1
> > Date: Tue, 10 May 2005 14:39:54 -0400 (EDT)
> > From: Steve Cousins <cousins at limpet.umeoce.maine.edu>
> > Subject: [Beowulf] Table of best compiler flags for different CPUs?
> > To: beowulf at beowulf.org
> > Message-ID:
> > 	<Pine.LNX.4.10.10505101415590.14786-100000 at limpet.umeoce.maine.edu>
> > Content-Type: TEXT/PLAIN; charset=US-ASCII
> > 
> > 
> > The recent thread comparing Opterons to Xeon/Nocona brought up some
> > interesting points about compilers for different CPU's.  Has anyone
> > created a table of CPU type vs Compilers, showing what flags to use to get
> > the best performance for a given CPU/Compiler pair? 
> > 
> > I know "best performance" depends on the application but hopefully there
> > are some basic flags that will work pretty well with any floating point
> > intensive applications (I know, not everyone is doing floating-point
> > applications but that is my domain). Of primary concern to me are finite
> > element and finite difference models.  The compilers that I am mainly
> > interested in are Gnu Fortran, Portland Group Fortran and the Intel
> > Fortran compilers, although a more complete table would of more general
> > interest I'm sure.
> > 
> > I have compared a number of CPU's using one of our ocean models and
> > recently I compared a Dual 3.0 Ghz Xeon Nocona to a Dual 2.6 Ghz Opteron
> > using an old version (3.2.4) of the Portland Group compiler using the
> > -fast switch. The Opteron was 71% faster.  This is presumably a fairly
> > dumb way to test the relative merits of each CPU so (not being a compiler
> > sophisticate) I am interested in what combinations of compilers and
> > switches will give a better comparison, as well as run our code faster.
> > 
> > Thanks,
> > 
> > Steve
> > 
> > ______________________________________________________________________
> >  Steve Cousins, Ocean Modeling Group    Email: cousins at umit.maine.edu
> >  Marine Sciences, 208 Libby Hall        http://rocky.umeoce.maine.edu
> >  Univ. of Maine, Orono, ME 04469        Phone: (207) 581-4302
>