Hi All - Likely a simple configuration issue I'm struggling with.
I have a very simple simulation (almost identical to the GaussianBeamSource example), but when I run with MPI (local node only, multi-core) it hangs before returning from 'sim.run(until=60)'. The simulation appears to proceed correctly until the sim should be complete: ' ... Meep progress: 45.56/60.0 = 75.9% done in 16.0s, 5.1s to go on time step 4556 (time=45.56), 0.0034461 s/step Meep progress: 57.18/60.0 = 95.3% done in 20.0s, 1.0s to go on time step 5718 (time=57.18), 0.00344415 s/step ' [Hang forever]. It hangs with 2 threads or 32 threads (Ryzen 3990x CPU). Performance scales apparently correctly through the simulation otherwise (2 threads is 2x serial; 32 again much faster). Meep setup: Ubuntu 20.04, miniconda3 From: https://meep.readthedocs.io/en/latest/Installation/#conda-packages conda create -n pmp -c conda-forge pymeep=*=mpi_mpich_* conda activate pmp I've also tried pmp_test, to same result. I have a 'print("after run")' right after sim.run(until=60), which does not appear. When I attach to the individual python processes with 'gdb --pid [pid]', they generally appear to be hung in some MPI all wait or reduce. But I'm not certain; they seem to be polling. Any advice appreciated; can provide any additional useful information. Very much need parallel meep for performance. Thanks! Luke
_______________________________________________ meep-discuss mailing list meep-discuss@ab-initio.mit.edu http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/meep-discuss
