Jan Blechta <[email protected]> writes:

> I'm not sure now with MatDestroy/VecDestroy. Following snippet should
> deadlock at two places but it does not. Possible explanations are at
> least:
>  - Mat/VecDestroy do not communicate at all
>  - Mat/VecDestroy communicate using communnication which is
>    implemented as non-blocking; MPI-2 does not guarantee whether
>    collective communication (except barrier) is blocking or not
>    http://www.mpi-forum.org/docs/mpi-2.2/mpi22-report/node87.htm
>  - Mat/VecDestroy communicate using point-to-point non-blocking
>    communcation; (PETSc collective may not imply MPI collective
>    communication)

The documentation (which states that MatDestroy() is collective) should
be followed on this.  Whether a specific instance deadlocks or not may
depend on debug mode (sometimes consistency checks will run in debug
mode) and implementation.  So please call them collectively.

Attachment: pgpRDAPgdxXIZ.pgp
Description: PGP signature

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

Reply via email to