> From: Fortran 90 List [mailto:[log in to unmask]] On Behalf Of
> > Drew McCormack
> > Sent: Wednesday, July 14, 2004 12:16 AM
> > [...] So if you want it to go as fast as it can, without writing it in
> > assembler ... you generally are better off in Fortran. [...]
On Wed, 14 Jul 2004, 6:06am -0700, [log in to unmask] wrote:
> - And even so, I think it would take a lot of expertise to write code in
> assembler that runs as fast as what's produced by modern optimizers. = Loren
Hmmmm... how well does a simply coded Fortran matrix-matrix multiply
do, compared with a tuned DGEMM produced, by, say, Atlas, when used
to perform the same operation?
Maybe matrix-matrix multiply isn't the best example (or maybe it
is), but if these sorts of simple operations are generally well
optimized by code optimizers, there would not be a need for ATLAS
(though there might still be a good rationale for BLAS itself).
(Yes, of course it depends on size of the data and on the platform
and compiler used.)
-P.
--
**************** "The web is a scary place." - J. Gunn ****************
Peter S. Shenkin Schrodinger, Inc.
VP, Software Development 120 W. 45th St., 32nd Floor
646 366 9555 x111 Tel New York, NY 10036
646 366 9550 FAX USERID: shenkin
http://www.schrodinger.com DOMAIN: schrodinger DOT com
Pre-arranged conf. calls: 702-759-8420 or 888-867-7084; passcode 646-366
|