[petsc-users] Columnwise and 2D partitioning for matrices that have irregular sparsity patterns

[Jed Brown]

"R. Oğuz Selvitopi" <roguzsel at gmail.com> writes:

> Hello,
>
> Is it possible to partition a parallel matrix in PETSc as follows:
>
> A B C
> D E F
> G H I
>
> The blocks A-D-G belong to processor 0 (A is the diagonal block, D and
> G are off-diagonal blocks)
>
> The blocks B-E-H belong to processor 1 (E is the diagonal block, B and
> H are off-diagonal blocks)
>
> The blocks C-F-I belong to processor 2 (I is the diagonal block, C and
> F are off-diagonal blocks)

Assemble the transpose, then either use MatMultTranspose or
MatCreateTranspose.

> Or, is it possible to have nine processors and each has a block of the
> matrix above? Block A belongs to processor 0, block B belongs to
> processor 1, and so on...

Not with sparse matrices (except using MatShell, in which case you are
responsible for the implementation).  You can do it for dense matrices
(see MatElemental).
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20140102/90c8f308/attachment.pgp>