[Mpi3-rma] Memory barriers in MPI_WIN_LOCK_ALL mode

[Torsten Hoefler]

On Tue, Oct 30, 2012 at 02:21:12PM +0000, Underwood, Keith D wrote:
> > > Ok, but, the kind of thing that delays that visibility (e.g. the
> > > uncertainty between when a posted write starts over PCIExpress and
> > > when it is known to be globally visible) is completely unaffected by a
> > > memory barrier.  Memory barrier until your heart is content and that
> > > thing could still be getting retried across a PCIExpress link and the
> > > process (and the NIC and other side of the link and just about
> > > anything else) would have no way to know...
> > 
> > Hmm.  Yes, you are right.  That's even worse though :-(.
> 
> Not usually.  If you start synchronizing outside the network you
> better know exactly what you are doing.  Most networks have some
> ordering primitives to help here.
Yes, so I assumed that this was all happening in shared memory (thus the
mfence). If we're on a network, we need to ensure that we have ordering
between the fence and the hand waiving :-).

As for the retry, I think the flush may not return before the value is
successfully committed into the target memory (either public or private
copy).

Best,
  Torsten

-- 
### qreharg rug ebs fv crryF ------------- http://www.unixer.de/ -----
Torsten Hoefler           | Assistant Professor
Dept. of Computer Science | ETH Zürich
Universitätsstrasse 6     | Zurich-8092, Switzerland
CAB E 64.1                | Phone: +41 76 309 79 29