[Mpi3-rma] RMA op compatibility matrix
jhammond at alcf.anl.gov
Sat Oct 15 07:19:35 CDT 2011
Given the complex nature of MPI RMA semantics, I think you want a
Hermitian matrix rather than a symmetry one :-)
I am awful. I'm sorry. I could not resist.
On Fri, Oct 14, 2011 at 10:11 PM, Torsten Hoefler <htor at illinois.edu> wrote:
> On Fri, 2011-10-14 at 17:52 -0500, James Dinan wrote:
>> Ahh, right. Thanks for your help. Updated tables are attached.
> What about rules 2+3 in the unified model? They explicitly allow
> overlapping put/acc + load and get + store but the behavior depends on
> the implementation. I think we should have another color for those.
> I agree to the separate matrix.
> Since all matrices are symmetric by definition, we could draw them into
> one, making the upper triangle unified and the lower triangle
> separate :-).
> All the Best,
> mpi3-rma mailing list
> mpi3-rma at lists.mpi-forum.org
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
More information about the mpiwg-rma