[Mpi3-rma] MPI-3 UNIFIED model clarification

[Underwood, Keith D]

At one point, them using their own memory fences was accepted as part of it, because it provides " the performance and
"safety" of threads, localized to a window".  Those same people have to use memory fences for threads to work right too.

> -----Original Message-----
> From: Jed Brown [mailto:five9a2 at gmail.com] On Behalf Of Jed Brown
> Sent: Friday, August 02, 2013 3:02 PM
> To: Underwood, Keith D; MPI 3.0 Remote Memory Access working group;
> Pavan Balaji
> Subject: Re: [Mpi3-rma] MPI-3 UNIFIED model clarification
> 
> "Underwood, Keith D" <keith.d.underwood at intel.com> writes:
> 
> > Oh, I think we want it to say that they are identical.  I believe that
> > is the only way to let the user actually use it.
> 
> Given that there is nothing the source rank can do to provide read ordering
> at the target, in what scenario can this be useful?  If users are specializing for
> UNIFIED and intend to elide the somewhat heavyweight MPI_Win_sync,
> how can they avoid using their own memory fences?  (I thought this was the
> use case that UNIFIED specialization was intended for: the performance and
> "safety" of threads, localized to a window.)