[mpas-developers] COMMIT CHECK BUG FIX: MPI timing issue

[Doug Jacobsen]

Hi all,

Recently I have been noticing some weird behavior from MPI_Wtime and
MPI_Wtick in MPAS, and I think I have a solution for the problem.

After a simple simulation on either my laptop of the IC machines here
(using gfortran and openmpi or mvapich) I was noticing the timers were all
0.0. After that I started writing out what MPI_Wtime and MPI_WTick returned
to see what the timer should be computed with, and I got values like

MPI_WTime = 3.07144143222521536682484349811465461E-4947
MPI_WTick  = 2.92504975975246619939945495478118484E-4947

Which are obviously incorrect. The probably apparently has something to do
with the mpi headers that are being used. So, I made some modifications to
the modules that are used within mpas_timer.F that fix this problem on my
end. I'm going to test it some more on the IC machines here with different
compilers, but I wanted to see if anyone else has problems with my
modifications or if they break anything.

I have attached a modified version of mpas_timer.F that fixes the issue for
me, so please let me know if it causes you any issues.

If I don't hear from anyone, I'll push this change onto the trunk on Friday.

Thanks,
Doug
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20120627/a30ae39f/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: mpas_timer.F
Type: application/octet-stream
Size: 12051 bytes
Desc: not available
Url : http://mailman.ucar.edu/pipermail/mpas-developers/attachments/20120627/a30ae39f/attachment.obj