[petsc-users] Does MUMPS check the nonzero pattern for each iteration?

[Jed Brown]

Santos Teixeira Frederico <fsantost at student.ethz.ch> writes:

> Hi,
>
> I am using PETSc with MUMPS and based on the time required for each
> iteraction (it is almost the same amount of time), I suspect MUMPS is
> not using the information about the nonzero pattern.  My question is:
> how do I make sure it is correctly configured? I am already setting
> "KSPSetOperators(ksp, A, A, SAME_NONZERO_PATTERN)".

Profile, don't speculate.  If you see something like this, then symbolic
factorization is indeed reused.

MatLUFactorSym         1 1.0 5.0068e-06 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
MatLUFactorNum         3 1.0 1.2076e-03 1.0 0.00e+00 0.0 0.0e+00 0.0e+00 0.0e+00  4  0  0  0  0   4  0  0  0  0     0


The vast majority of the work is in numeric factorization.  MUMPS has a
sequential stage in symbolic factorization so symbolic can be
significant at high core count, but otherwise, it's negligible.