Changes the priority for MPI tasks to thread_priority::boost, to give higher priority to the submission of MPI communication, but then fall back to normal priority for polling. Also changes to the static_priority scheduler for the MPI pool since it only has a single thread. I haven't benchmarked this in isolation and doubt that it makes much of a difference, but it's a bit more appropriate than the default local_priority_fifo since no work stealing needs to happen.
Changes the priority for MPI tasks to
thread_priority::boost
, to give higher priority to the submission of MPI communication, but then fall back to normal priority for polling. Also changes to thestatic_priority
scheduler for the MPI pool since it only has a single thread. I haven't benchmarked this in isolation and doubt that it makes much of a difference, but it's a bit more appropriate than the defaultlocal_priority_fifo
since no work stealing needs to happen.