Dear Experts,
I am am trying to parallelise the computation and inversion of the influence matrix in neBEM.
I’m using “nebem.SetNumberOf Threads()” to do so, but it seems that increasing the number of requested cores is slowing down significantly the matrix inversion and increasing the solution error.
For example, for a 1100 x 1100 matrix, i got the following results:
- 1 core: setup time of 11 sec, inversion time of 13 sec, max error 10^-8
- 2 cores: setup time of 12 sec, inversion time of 20 sec, max error 15k.
- 8 cores: setup time of 20 sec, inversion time of 400 sec, max error 600k.
I think openMP is correctly installed (the CMakeList.txt contains the following lines: find_package(OpenMP), <BOOL:${OpenMP_CXX_FOUND}:OpenMP::OpenMP_CXX>).
Is there a way to correctly set up and run neBEM with OpenMP?
I did not manage to find any instruction about enabling OpenMP while building/making the project.
Thank you in advance!