Take a look at this PR for a few examples of usage of mesh_loop:
https://github.com/dealii/dealii/pull/7806
The function WorkStream::run
takes (a minimum of) 5 arguments:
WorkStream::run(cell, endc, cell_worker, copier, scratch, copy);
initial and final iterator, a worker function, a copier function, a scratch
object, and a copy object.
Logically, WorkStream::run does the following:
for(it = cell; it != endc; ++it) {
cell_worker(cell, scratch, copy);
copier(copy);
}
and is agnostic about what type of iterator you pass to it.
MeshWorker::mesh_loop is a specific implementation of the above in which the
iterator is a cell iterator, and the loop is thought to run also on each face
of each cell as well, in order to assemble DG type contributions, i.e., it
allows you to specify two additional functions, a boundary_worker and a
face_worker:
WorkStream::mesh_loop(cell, endc, cell_worker, copier, scratch, copy, flags,
boundary_worker, face_worker);
This is logically identical to
for(it = cell; it != endc; ++it) {
cell_worker(cell, scratch, copy);
for(unsigned int f=0; f<NFaces; ++f) {
if(cell->face(f)->at_boundary() {
boundary_worker(cell, f, scratch, copy);
} else {
// Get subface, neighbor face index, and neighbor subface index
const auto [sf, nf, nsf] = get_neighbor_info(cell, f);
// Run the face worker
face_worker(cell, f, sf, cell->neighbor(f), nf, nsf);
}
}
copier(copy);
}
What mesh_loop does is controlled in a finer way by the assembly flags. For
example it is possible to visit each face only once, or to assemble first faces
and then cells, etc.
MeshWorker::integration_loop
is a specific implementation of MeshWorker::mesh_loop where you use the DoFInfo
and DoFInfoBox objects, that contain scratch and copy data with a different
structure. The two behave in a substantially identical way, but with different
data structures.
In particular, you could implement MeshWorker::integration_loop using
MeshWorker::mesh_loop, and some wrapping around DoFInfo and DoFInfoBox objects.
I hope this clarifies a little the scope of run (agnostic about iterators) and
mesh_loop (actually expecting cell iterators, and taking care of
subface/face/neighbor matching when looping over faces).
Best,
Luca.
> On 18 Mar 2019, at 17:46, 'Maxi Miller' via deal.II User Group
> <[email protected]> wrote:
>
> Similar question: How are the different loop-functions differentiated, i.e.
> MeshWorker::integration_loop and MeshWorker::mesh_loop? Both are able to loop
> over faces, boundaries and cells, but what are the differences here?
> Thanks!
>
> Am Dienstag, 22. Januar 2019 16:56:28 UTC+1 schrieb Bruno Turcksin:
> Le mar. 22 janv. 2019 à 10:48, 'Maxi Miller' via deal.II User Group
> <[email protected]> a écrit :
> > I. e. if I would like to calculate f. ex. the L2-norm of a vector (while
> > neglecting that there already is a function for that), I can use WorkStream
> > for parallelization of that, but not MeshWorker, is that correct?
> That's right. You can use WorkStream using the iterators for the
> vector but MeshWorker won't work because it expects cell iterators.
>
> Best,
>
> Bruno
>
> --
> The deal.II project is located at http://www.dealii.org/
> For mailing list/forum options, see
> https://groups.google.com/d/forum/dealii?hl=en
> ---
> You received this message because you are subscribed to the Google Groups
> "deal.II User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
