So do we want to prohibit diff op atomicity in all cases or is that the default? I think many implementations won't care but the RMA-in-hardware ones really want same_op_no_op_replace or less (as in more restrictive). Jeff On Thu, Mar 13, 2014 at 3:28 PM, Jim Dinan <[email protected]> wrote:
Because it's an info key, we can't make the relax the semantic, we can only restrict it. So, the most permissive semantic needs to be the default.
On Thu, Mar 13, 2014 at 3:54 PM, Dave Goodell (dgoodell) <[email protected]> wrote:
On Mar 13, 2014, at 2:47 PM, Jeff Hammond <[email protected]> wrote:
same_op_no_op_replace might not be completely SHMEM-compliant but it almost certainly meets the needs of reasonable SHMEM programs. Right now, the incompatibility of REPLACE and <any reduce op> puts me in a very bad place for essentially all PGAS models.
Wouldn't you be in a perfectly fine place if we just add "same_op_no_op_replace" to the standard and then you modify your codes to use it?
I really don't understand your insistence on changing the default behavior...
-Dave
_______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
_______________________________________________ mpiwg-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-rma
-- Jeff Hammond [email protected]