another late topic:
To separate the extended collective discussion, votes and deadlines
from one-sided, I propose that you should include the non-blocking
MPI_WIN_IBarrier(flag,win,request)
into the MPI_WIN_BARRIER section.
I think the reasons for WIN_IBarrier are clear.
Here a short example:
PUT(torank=1)
WIN_IBARRIER
|
compute something
compute something
compute something
compute something
compute something PUT(torank=0)
compute something WIN_IBARRIER
compute something |
compute something compute something <<< in this moment the
| compute something barrier can be
WAIT compute something executed in the
load compute something background
compute something
compute something
|
WAIT
load
In this example the necessary synchronization is done with
MPI_WIN_Ibarrier, but no process had to idle in some wait.
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