At 07:21 AM 1/22/2004, you wrote:
>Would you ever expect something like a 30% improvement for
>Opteron, based on special compilation?
>
>-P.
Opteron-specific optimizations gain in some applications from use of the
larger register sets. The floating point units are designed to run as fast
with scalar instructions as with parallel, in some cases. Still, PGI
Fortran offers the option to "cache-align" arrays in order to gain
advantage from parallel instructions. While there is an advantage in these
optimizations, it's not as great as you appear to be looking for.
Tim Prince
|