Closed fdrmrc closed 10 months ago
In a DG implementation, in order not to compute jumps and averages twice, one needs to do
//inside assembly routine
if(current_cell_index < neighboring_cell_index){
//compute jumps, averages on faces
}
With a p::d::T
active cells indices are not unique among processors, so one should acutally use the unique identifier provided by cell->id()
. Hence, an AgglomerationAccessor::id()
has been added.
@luca-heltai Do you foresee any problems? (I've not thought about FECouplingValues/FEInterfaceValues approaches yet.) This would allow to have a distributed version of the classical DG methods on polygons just by replacing the dof_handler.active_cell_iterators()
with ah.polytope_iterators()
.
This draft PR introduces the concept of locally_owned polygon. In particular, the accessor class provides a
is_locally_owned()
methods that forwards the request to a deal.II cell.A test with two MPI ranks has been added.