Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Remove unnecessary MPI barriers #13812

Merged
merged 1 commit into from
May 25, 2022
Merged

Remove unnecessary MPI barriers #13812

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
45 changes: 0 additions & 45 deletions source/dofs/dof_handler_policy.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3789,35 +3789,6 @@ namespace internal
GridTools::exchange_cell_data_to_ghosts<
std::vector<types::global_dof_index>,
DoFHandler<dim, spacedim>>(dof_handler, pack, unpack, filter);

// have a barrier so that sends between two calls to this
// function are not mixed up.
//
// this is necessary because above we just see if there are
// messages and then receive them, without discriminating
// where they come from and whether they were sent in phase
// 1 or 2 (the function is called twice in a row). the need
// for a global communication step like this barrier could
// be avoided by receiving messages specifically from those
// processors from which we expect messages, and by using
// different tags for phase 1 and 2, but the cost of a
// barrier is negligible compared to everything else we do
// here
if (const auto *triangulation =
dynamic_cast<const dealii::parallel::
DistributedTriangulationBase<dim, spacedim> *>(
&dof_handler.get_triangulation()))
{
const int ierr = MPI_Barrier(triangulation->get_communicator());
AssertThrowMPI(ierr);
}
else
{
Assert(false,
ExcMessage(
"The function communicate_dof_indices_on_marked_cells() "
"only works with parallel distributed triangulations."));
}
# endif
}

Expand Down Expand Up @@ -4213,22 +4184,6 @@ namespace internal
// will request them again in the step below.
communicate_mg_ghost_cells(*triangulation, *dof_handler);

// have a barrier so that sends from above and below this
// place are not mixed up.
//
// this is necessary because above we just see if there are
// messages and then receive them, without discriminating
// where they come from and whether they were sent in phase
// 1 or 2 in communicate_mg_ghost_cells() on another
// processor. the need for a global communication step like
// this barrier could be avoided by receiving messages
// specifically from those processors from which we expect
// messages, and by using different tags for phase 1 and 2,
// but the cost of a barrier is negligible compared to
// everything else we do here
const int ierr = MPI_Barrier(triangulation->get_communicator());
AssertThrowMPI(ierr);

// Phase 2, only request the cells that were not completed
// in Phase 1.
communicate_mg_ghost_cells(*triangulation, *dof_handler);
Expand Down