Thanks Thomas! ----- Mail original -----
De: "Thomas Ropars" <[email protected]> À: [email protected] Envoyé: Vendredi 20 Septembre 2013 09:57:27 Objet: Re: [mpich-discuss] Order of completion in MPI_Waitany
Hi Matthieu,
MPI does not provide any guarantee on the completion order of the reception requests in this case. More generally, regarding waitAny, the MPI-3.0 specification says that "If more than one operation is enabled and can terminate, one is arbitrarily chosen".
Thomas
On 09/19/2013 04:31 PM, Matthieu Dorier wrote:
Hi,
I would like a precision on the semantics of MPI_Waitany in this situation:
- Process 1 does 2 MPI_Irecv with different tags A and B (the order doesn't matter); it then waits on the two requests with MPI_Waitany.
- In the mean time, process 2 does an MPI_Send with tag A, then MPI_Send with tag B.
Can I ensure that MPI_Waitany will always give me the index of the message corresponding to the first MPI_Send, i.e. the one with the tag A, even in situations where process 1 called MPI_Irecv on a tag B first?
Thanks
Matthieu Dorier
PhD student at ENS Cachan Brittany and IRISA
_______________________________________________
discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss
_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss