Mesh Copy constructor uses incorrect MPI communicator
Hi,
When copying a dolfin::mesh using the copy constructor, the new dolfin::mesh is initiated with the MPI_COMM_WORLD communicator, rather than the "comm" communicator associated with the mesh. This causes stalls when using split MPI communicators. (i,e when using MPI_Comm_split to break the world communicator into chunks)
On a similar note, dolfin::Scalar can only be constructed on MPI_COMM_WORLD. A dolfin::Scalar(MPI_Comm comm) type constructor would be useful. Without that, assembling a functional on a split communicator is not possible.
Comments (8)
-
-
-
assigned issue to
-
assigned issue to
-
- changed status to resolved
Merge branch 'garth/fix-issue-617-mesh-mpi-comm' into next
Fixes Issue
#617(MPI communicators)→ <<cset 33016f820015>>
-
Merge branch 'garth/fix-issue-617-mesh-mpi-comm' into next
Fixes Issue
#617(MPI communicators)→ <<cset 33016f820015>>
-
Merge branch 'garth/fix-issue-617-mesh-mpi-comm' into next
Fixes Issue
#617(MPI communicators)→ <<cset 33016f820015>>
-
Why has this happened? I just pulled next locally, and merged my branch. I haven't touched my local 'next' for a long time, I guess.
-
@chris_richardson
next
have been reset since you last touched it. -
- removed milestone
Removing milestone: 1.7 (automated comment)
- Log in to comment
Thanks. The transition to using communicators other than
MPI_COMM_WORLD
was started a while back, but wasn't completely finished. Reports on what needs fixing are appreciated.