Sorry everyone - this is not a "core" lam question so may be slightly off
topic but does relate to Lam 7 and I have a feeling someone on this list
can make a suggestion to resolve the problem.
Using Luke Palmer's ctype.c (thanks Luke!!) I've successfully had the
Intel C compiler to compile lam7 & the high performance linpack (HPL)
benchmark suite. I used the Intel math kernel library as well. Now here
is the interesting part - I actually get worse performance by about 25%
with this combination than I get when I do the entire thing with gcc &
ATLAS the free software math library. This is all occuring on pentium IV
systems running Red Hat Linux 8 with all the glibc updates and 2.4.20
kernel.
Are there any specific compiler optimization flags people are using when
compiling Lam with the intel compiler? Any/all suggestions appreciated.
(If someone who has also built HPL with the intel compiler and gotten
better results this way can comment that would be appreciated as well.)
Thanks everyone (and especially to the LAM developers for a very robust
MPI environment),
Peter McLachlan
Office: (905) 316-8646
|
