Jim Dinan <[email protected]> writes:

Another question that was raised outside of this email discussion is whether we can rely on the architecture (in the absence of MPI calls) to make the results of operations visible in the order in which the origin process expected (e.g. through ordered accumulates, flushes, etc).

I'm not sure exactly what the question was, but you certainly don't get (portable) read memory ordering without memory barriers on the read side. Recall that explicit data or control dependencies are not enough (Alpha and ARMv7, respectively).