dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Perhaps it's worth focusing on getting DofMap (with re-working of the 
mapping) and SparsityPattern sorted out for parallel assembly before 
worrying too much about the PETSc wrappers. Once you have the 
SparsityPattern (for each process) it should give you what you need to 
initialise parallel matrices.

Garth

Gustav Magnus Vikstrøm wrote:
> The PETSc function MatCreateMPIAIJ  
> ( http://www.caspur.it/risorse/softappl/doc/petsc_docs/manualpages/Mat/MatCreateMPIAIJ.html  
> ) takes a number of arguments. Two of the arguments are:
>
> d_nnz: Array containing the number of nonzeros in the various rows of the  
> DIAGONAL portion of the local submatrix
> o_nnz: Array containing the number of nonzeros in the various rows of the  
> OFF-DIAGONAL portion of the local submatrix
>
> Currently we are using PETSC_NULL for these to values and a guess  
> (currently set to 50) for the number of non-zeroes on all rows. According  
> to the PETSc manual performance can be increased by more than a factor of  
> 50 by setting these parameters accurately.
>
> My question is how to create d_nnz and o_nnz?
>
> For sequential matrices it is done by:
> SparsityPattern::numNonZeroPerRow(uint nzrow[])
>
> Maybe we could add methods to SparsityPattern?
> SparsityPattern::numNonZeroPerDiagonalRow(uint nzdiagrow[])
> SparsityPattern::numNonZeroPerOffDiagonalRow(uint nzoffdiagrow[])
>
> - Magnus
> _______________________________________________
> DOLFIN-dev mailing list
> DOLFIN-dev@xxxxxxxxxx
> http://www.fenics.org/mailman/listinfo/dolfin-dev