[petsc-users] MatMatSolve in sequential call

Hong Zhang hzhang at mcs.anl.gov
Mon Aug 13 10:56:53 CDT 2012

Alexander :
petsc-mumps interface does not support MatMatSolve()
see ~petsc/src/mat/impls/aij/mpi/mumps/mumps.c
Likely MatMatSolve_Basic() is called which calls MatSolve()
repeatedly, thus slow.

In addition, MUMPS
only supports centralized rhs. Petsc interface must scatter
distributed rhs into a seqential rhs vector for MUMPS.
If num of rhs is large, would be slow here as well.

The error in your report likely is a bug in petsc. I'm investigating it
and will let you know once it is fixed.


On 13.08.2012 17:23, Hong Zhang wrote:
>> Alexander :
>> Hmm, isrow is a matrix ordering, likely a bug :-(
>> Can you send us a stand-alone code that reproduces this error?
> Hong,
> I can try to pull it out, but this may be tricky.
>  Do you use an external package for MatMatSolve()?
> Yes, MUMPS.
> I use ksp with preonly.
> Originally I thought it will be faster to solve multiple problems using
> MatMatSolve.
> This turned out to be not true however, but the code is still like that.
> --
> Regards,
> Alexander
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20120813/4fb4ee96/attachment.html>

More information about the petsc-users mailing list