On Sep 16, 2014, at 6:37 AM, Florian Lindner <[email protected]> wrote:
> Hello,
>
> I'm currently replacing an RBF implementation with petsc linear algebra. The
> program itself runs parallel using MPI but the piece of code I work on runs
> strictly sequentially without making any use of MPI, just the same code on
> every node. Right now we're more interessted in patsc sparse matrix abilities
> then in its parallelization. Though parallelization is certainly interesting
> later....
>
> What is the best way to run petsc sequentially?
>
> 1) MatSetType the matrix to MATSEQSBAIJ e.g. -> expects MPI communicator of
> size 1.
> 2) MatSetSizes(matrix, n, n, n, n) does not work.
This should certainly work on one process
> 2) MatCreate not with PETSC_COMM_WORLD but with the communicator of size 1.
> Where do I get it from? (probably MPI_Comm_create and friends)
Just use PETSC_COMM_SELF
>
> Is there another more petsc like way?
>
> Thanks,
> Florian
