[mpich-discuss] MPI_Send() may or may not wait for an MPI_Recv()

Dave Goodell goodell at mcs.anl.gov
Tue Aug 24 11:12:27 CDT 2010


MPI_Send is allowed to block or not, based on whatever the implementation thinks will provide the best performance.  MPI_Bsend will always buffer the send.  MPI_Isend initiate a non-blocking send, but then you must make sure to complete the request with MPI_Wait/MPI_Test/etc.  MPI_Ssend will block the sender until the receiver posts a corresponding receive operation.

-Dave

On Aug 24, 2010, at 10:49 AM CDT, michael bane wrote:

> Sounds like it does what the standard does! If you want buffered then use buffered send, don't presume!
> 
> Michael
> http://www.twitter.com/mkbane_mcr
> 
> 
> On 24 Aug 2010, at 16:45, Alexandru Blidaru wrote:
> 
>> Hi guys,
>> 
>> I discovered that MPI_Send sometimes waits for the MPI_Recv, while at other times it doesn't. Does anyone know the reason why this is happening or any way I can make sure it either always doesn't or does wait?
>> 
>> Cheers,
>> 
>> -- 
>> Alexandru Blidaru
>> 
>> University of Waterloo - Electrical Engineering '15
>> University email:              asblidar at uwaterloo.ca
>> LinkedIn:                         http://ca.linkedin.com/in/alexandrublidaru
>> Blog:                              http://alexblidaru.wordpress.com/
>> Twitter handle:                @G_raph
>> 
>> _______________________________________________
>> mpich-discuss mailing list
>> mpich-discuss at mcs.anl.gov
>> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 
> _______________________________________________
> mpich-discuss mailing list
> mpich-discuss at mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss



More information about the mpich-discuss mailing list