On 10/25/12 10:55 AM, "Jeff Squyres" <jsquy...@cisco.com> wrote:
>Something that might not be clear from my initial writeup: > >1. I had to go change C code to disable libnbc. Since non-blocking >collectives are part of MPI-3: > a) we have no convenient configure argument to not build the libnbc >coll component (there is a way, but it's laborious), and > b) even if we did, OMPI's coll selection will fail at run time because >it didn't find modules for the non-blocking collective operations. With r27498, the progress function in libnbc is only registered when the first non-blocking collective is performed. It will be unregistered whenever there are no longer any live communicators which have participated in a non-blocking collective. Does this help with your benchmarking? At some point, we really need to move from a progress function to using the progress provided by the completion callbacks available for point to point requests. But it's unlikely I'll have time to do that, particularly given how long its taken me to get the onesided stuff going. Brian