problems with parallel fortran programs (openmp)

problems with parallel fortran programs (openmp)

wurzel's picture

I have problems with fortran programs (weather forecast models) parallized with openmp. When the programs are run for some period the threads hangs. The threads exists, but are using no cpu time. I am using fortran version 7.0. It is 2cpu 2.4 ghz Xeon on a 7500 chipset with hyperthreading. the problem is both when hyperthreading are enabled or not. the linux kernel is 2.4.18-3smp from the Red Hat Linux 7.3 2.96-110. When I using the portland compiler there no problem (but the programs are slower, up to 33%), or on other platforms as True64unix (alpha) or SGI. The programs does not hang at the same place. Sometimes they can run for long period before they hangs.

6 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.
one_glen's picture

i'd suggest go out and get bert, if the code is less that 8000 lines there is a free version for linux that analyses codz and find out where the threads are hanging. sounds like the openmp implementation is hosed, or what version of openmp does pgcc use 1.1? 2? ifc v7 uses 2?

wurzel's picture

thanks for then replies: pgf77 is using Openmp version 1.1
ifc is using version 2. I will try to use bert and see the results.

It is the workstation version of ETA from NCEP and MM5 model, I have problems with. I have at MM5's homepage seen compiler options for ifc, but only for version 6, and without openMP parallelization. I have not seen it for version 7 and with openMP.

Martyn Corden (Intel)'s picture

This sounds like an issue which should be reported to Intel Premier Support at https://premier.intel.com for follow up.

I do know of weather codes such as MM5 that have been run successfully with OpenMP using the Intel compiler.

Martyn

madstynx's picture


Hello Wurzel, and others.

I really want to know if you are succesfull in finding what causes this behavior.

Running into the same problem, Redhat 7.3/Suse8.0) kernel 2.4.18-SMP, Intel Xeon 2cpu'smachines, not hyperthreaded, Intel compiler 7.0 (patched).
Solver program, programmed with Fortran and C, with OpenMP.

Exactly the same behavior Wurzel has, parallel runs kan go for days (30/40/50 hours), but they alway's stop(not at the same place). The processes are still there, but no cpu use or whatever.

Are here people with more idea's and/or clues?

Thanks in advance

misty12@yandex.ru's picture

Hello,
Have anyone found out what causes this problem, because I ran into the same problem but under Windows on 2Xeon processors 2.4GHz. My program can ran for 20 or more hours and than stops calculating though the process is still here. This error occurs often but each time on the different place. Can anyone help me with it, please?
Olga

Login to leave a comment.