[FLASH-USERS] OpenMP Scheduling

[Gian-Andrea Wetten]

Dear all,


I am currently looking at performance implications of OpenMP scheduling (including scheduling techniques not present in the standard OpenMP implementation).


I read the section about multithreaded FLASH on the documentation and executed the examples listed under performance results. I saw that the openmp loops are executed a fair amount of times with these examples but the number of iterations seems to be rather small, which is not very beneficial when wanting to analyze the impact of the scheduling methods. A run with 8 MPI ranks and 20 threads per rank had ~20 iterations per loop execution on most openmp loops with the Sedov 2D example (42.5.4 in the pdf of the latest documentation).

The only interesting one I've found so far seems to be the laser simulation unit (loop in ed_traceblockrays2dcyl3d).


Could you point me to a simulation (a setup command or par file) which might have a high number of iterations in its OpenMP loop(s)? I am only interested in the performance aspect and not in the results from the simulation.


BR,

Gian-Andrea Wetten
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://flash.rochester.edu/pipermail/flash-users/attachments/20210830/259baf9d/attachment.htm>