For our blocking receive(std::vector), we first do a MPI_Probe to get the incoming message size, so that we can resize the data vector to accomodate. This is wonderful behavior.
For our nonblocking equivalent, we instead currently force users to pre-allocate the input data vector. Is there any way to instead make the allocation behavior as user-friendly as in the blocking case? We can't use MPI_Probe, because that would block. We can't use MPI_Iprobe and then return, because (unless the non-blocking probe already found the message) then we'd never end up calling MPI_Irecv. We could use MPI_Iprobe and then hold off on the recv (and additional Iprobe calls as necessary) until Parallel::wait() is called, but that's basically equivalent to (and so redundant with) doing a blocking receive. I'm currently rewriting my app code to switch to all blocking receives, so this isn't critical for me, but it might be a nice feature to add in the future if we can figure out how to work it. --- Roy ------------------------------------------------------------------------------ Live Security Virtual Conference Exclusive live event will cover all the ways today's security and threat landscape has changed and how IT managers can respond. Discussions will include endpoint security, mobile security and the latest in malware threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/ _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
