[Mpi3-hybridpm] First cut at slides
jhammond at alcf.anl.gov
Tue Jun 18 10:25:24 CDT 2013
> > I don't see the issue with connected dynamic processes and finalize.
> > Doesn't each process just wait until count=0 to actually finalize? In this
> > case, is it not sufficient for each process to use its local ref count?
> > Your slide looks fine to me. What's the example where borkification
> > happens?
> The issue is that finalize might actually need to do something
> synchronizing between all connected processes. I think it therefore might
> actually cause connected processes to block until everyone actually
> finalizes (vs. calling FINALIZE).
> But maybe that isn't disastrous. If you don't want that (potential)
> behavior, you should COMM_DISCONNECT before finalize.
Does FINALIZE imply DISCONNECT? Could we add that behavior without
breaking existing semantics?
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
ALCF docs: http://www.alcf.anl.gov/user-guides
More information about the mpiwg-hybridpm