Hello,
I've questions about performances and parallelism with openMP run on a single thread (mono-thread).
My team and I are developing a fortran source code which is compiled with iFort12.
For some reason, I have to compile it in parallel mode and run it on a single thread. Unfortunatly I can not do otherwise.
The execution time (elaps time, not user time) seems to have a strange behaviour when I change the server where I compil and run it.
On a HP Z600, my code has the same run time (elaps) if it's compiled in parallel (with OMP_NUM_THREADS=1) or if it's compiled sequentially.
If I run the code n times simultaneousely (n could be 8, 16, 64, etc ...) the elaps time of parallel version is the same as the elaps time of sequential version.
On a HPDL580, my code is very longer in the parallel (but mono-thread) version than in the sequential version.
The parallel time is average twice the time of the sequential version.
Why is there so much difference in the behavior of the parallel code between two servers ?
Is there a compiler option that can reduce my execution time in parallel mode ?
I have made the same tests with my code compiled with gfortran and no problems appears.
Thanks a lot for your answer.
Patrick BODET