"Garth N. Wells" <[email protected]> writes:
> A 'fix' for this needs to go deeper. With global dofs, building the 
> sparsity pattern is only one of the difficulties. The other is insertion 
> into a sparse matrix. PETSc, in particular, doesn't perform well for 
> dense rows.

If you know which rows will be dense, you can balance the ownership.
Alternatively, you can use MatNest and MatCreateTranspose to distribute
the dense rows.  (You would then use a fieldsplit preconditioner.)

Attachment: pgpwzBolBWMDL.pgp
Description: PGP signature

_______________________________________________
fenics mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics

Reply via email to