Intel Fortran 11, MPICH2, 32bit Windows-based system : Bad Performance

[tnvquynh]

Hello,

I have written 2 codes. The first one is a non-parallel code and the second one uses MPICH2. They are compiled with Intel Fortran 11 (Trial Version) on 32 bits Windows system.

The options for the compilations are /fast . So, the first code is compiled by the following command:

ifort /fast -o a1.exe mycode.f90

The second code, implementing MPICH2, is compiled by:

ifort /fast -o a2.exe mycode.f90 /link fmpich2.lib


The compilations were successfully finished and gave 2 executables.

When I launch a2.exe using only 1 CPU, I find it runs almost 2 times slower than the a1.exe. Note that two executables treat the same problem and give the same results (Sure!).

I seem the Intel Fortran optimization (here /fast) fails when I compile a mpich code.


Anyone has experience on this problem? Please give me some advice!

Thank you,

tnvquynh.

[TimP]

Run with a profiler such as PTU, preferably with static link, and high threshold for yielding CPU during spin waits. Or, use an MPI profiler, such as Intel Trace Analyzer/Collector, which should be compatible with MPICH2. You are getting more into the territory of the HPC forum.