The code you showed was incorrect -- you were waiting on an uninitialized
variable. Perhaps that code was only a snipit...?
On Oct 12, 2010, at 8:00 AM, Ed Peddycoart wrote:
> Actually, that wasn't the problem. My code is working now with no changes to
> it. Not sure what the problem was but it wasn't the called to MPI_Send
> blocking.
> Ed
>
>
> From: users-boun...@open-mpi.org on behalf of Jeff Squyres
> Sent: Tue 10/12/2010 6:52 AM
> To: Open MPI Users
> Subject: Re: [OMPI users] nonblocking send/receive question
>
> On Oct 11, 2010, at 1:29 PM, Bowen Zhou wrote:
>
> > Try MPI_Isend?
>
> 'zactly correct.
>
> You currently have an MPI_Wait on the sender side for no reason -- the
> request is only filled in on the receiver. So you're waiting on an
> uninitialized variable on the sender.
>
> MPI_Send is a "blocking" send. MPI_Isend is a non-blocking send.
> MPI_Recv is a blocking receiver. MPI_Irecv is a non-blocking receiver.
>
> MPI_Send is more-or-less equivalent to MPI_Isend immediately followed by an
> MPI_Wait on the corresponding request. Ditto with MPI_Recv / MPI_Irecv.
>
>
> >
> >> I have a glut application I am trying to add MPI to. In the display
> >> callback, for rank >= 1, I want to send data to the rank =0 process. I am
> >> not concerned at this point about sending data from the rank 0 process
> >> back to the rank >= 1 process, so my data is one direction. I would like
> >> to do this with non-blocking send/receive but I am not having much success.
> >> Within my display callback I do the following:
> >> if( myrank == 0 ) {
> >> MPI_Irecv( receiveData, DATA_SIZE, MPI_DOUBLE, 1, 19, MPI_COMM_WORLD,
> >> &request );
> >> MPI_Wait( &request, &status );
> >> }
> >> else if( myrank == 1 ) {
> >> /* Post a receive, send a message, then wait */
> >> MPI_Send( sendData, DATA_SIZE, MPI_DOUBLE, 0, 19, MPI_COMM_WORLD );
> >> MPI_Wait( &request, &status );
> >> }
> >> But it appears that the app is still blocking after the MPI_Send.... (I
> >> have various debug prints in the actual code, this is stripped down for
> >> ease of reading). A sample app that i have that does this works... Is
> >> doing this from the glut display call back causing the problem? Any
> >> suggestions would be greatly appreciated.
> >> Thanks,
> >> Ed
> >> ------------------------------------------------------------------------
> >> _______________________________________________
> >> users mailing list
> >> us...@open-mpi.org
> >> http://www.open-mpi.org/mailman/listinfo.cgi/users
> >
> > _______________________________________________
> > users mailing list
> > us...@open-mpi.org
> > http://www.open-mpi.org/mailman/listinfo.cgi/users
>
>
> --
> Jeff Squyres
> jsquy...@cisco.com
> For corporate legal information go to:
> http://www.cisco.com/web/about/doing_business/legal/cri/
>
>
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users
>
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> http://www.open-mpi.org/mailman/listinfo.cgi/users
--
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/
