Jeff Squyres wrote:
> You might want to check and ensure that you don't have any pending send or
> receive requests when you call MPI_Finalize. LAM will try to finalize all
> outstanding requests before finishing, which is probably what you're
> seeing.
>
This sounds like the problem. Does this apply to all types of calls,
ie., if you have an Irecv posted that doesn't complete MPI_Finalize will
hang? Is there any way to force completion or cancel all pending send
and receives?
Thanks
-Mike
>
> On Mon, 18 Aug 2003, Michael Lees wrote:
>
>
>>I'm having a problem where my program seems to reach the end of its
>>execution and then hang.
>>
>>The program finishes with two steps, MPI_Barrier followed by
>>MPI_Finalize?
>>
>>All the nodes within the program complete the barrier synchronisation
>>and then call MPI_Finalize. The MPI_Finalize never completes, it just
>>seems to hang?
>>
>>Each node has two seperate threads, with the master thread calling
>>MPI_Finalize. The child threads do terminate via pthread_exit and seem
>>to be okay? To ensure this each of the master threads on the nodes makes
>>a pthread_join call prior to the final MPI_Barrier call.
>>
>>Could anyone offer a suggestions as to what I'm doing wrong or what the
>>problem might be?
>
>
|
