Our counter model adds a synchronization overhead (the addition to
the counter) in each call although the target only wants to know
that the latest PUT is ready on the target.
The first IPUTs with increment=0 and the last PUT with
increment=1 is no alternative because there is no chance
to guarantee that the first IPUTs are ready on the target
when the last IPUT is ready.
FENCE is no alternative because there is no IFENCE :-) !?!
In the moment I think the application must put all data with
one PUT with a complex origin and target datatype addressing
all data locations.
Rolf Rabenseifner (Computer Center )
Rechenzentrum Universitaet Stuttgart (University of Stuttgart)
Allmandring 30 Phone: ++49 711 6855530
D-70550 Stuttgart 80 FAX: ++49 711 6787626
Germany rabenseifner@rus.uni-stuttgart.de