|
|
Going back to this thread from earlier this calendar year...
Ganesh wrote:
Hi Dick,
Jeff paraphrased an unnamed source as suggesting that: "any
MPI program that relies on a barrier for correctness is an incorrect
MPI application." . That is probably too strong.
How about this assertion?
If there are no wildcard receives - every MPI_Barrier call is
semantically irrelevant.
This depends on what 'semantically irrelevant' means. It is clear that
one can write a wildcard-free program that will deadlock if you insert
a barrier incorrectly, but that removing the barrier will avoid the
deadlock. (Imagine P1 doing a Send; Barrier and P2 doing a Barrier;
Receive(nonwildcard)).
So a wildcard-free program may still deadlock (semantically noticeable
effect) by having barriers. I'm sure you did not mean to include this
degenerate nit-pick - but yes otherwise you are right!
And similarly (to disprove Jeff's original assertion), if there are
wildcards, then an MPI_Barrier call could be semantically relevent.
E.g., let's say P0 and P1 each send a message to P2, both using the
same tag and communicator. Let's say P2 does two receives on that
communicator and tag, using a wildcard source. So, the messages could
be received in either order. One could introduce barriers to order the
messages. E.g.,
P0:
Send
Barrier
P1:
Barrier
Send
P2:
Recv
Barrier
Recv
|
|
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}