I am trying to solve a symmetric generalized eigenproblem with slepc (n
= 40500) using 8 processors. The command I use is:
mpiexec -n 8 ./main -eps_nev 10 -eps_target 0 -st_type sinvert
The program assembles the two matrices (A,B), sets up the eigensolver
and then calls EPSSolve. The issue is that the matrix assembly runs fine
in parallel. And for about 1 minute or so, the EPSSolve call uses all 8
processors (according to the top program). But then after around 1
minute or so, the program collapses to just 1 thread and just runs that
way for a long time (still in the EPSSolve call).
Is there a way to diagnose what is going on, and how to make the program
use all 8 processors for the eigensolver?
