On Sep 27, 2011, at 5:43 AM, Xin He wrote: > sorry that it takes so long for paper works of TIPC BTL. Hopefully I can get > some feedback this week.
Great. > However, I would like to do some work while waiting. For future work, I would > like to add TIPC_broadcast > to support MPI_Bcast. But as I looked into TCP BTL code, I could not find any > code doing broadcast. Does > that mean MPI_Bcast is actually done by sending one by one? In that case, in > I want to add actual broadcast > support, which part should I change, I guess not just adding a BTL but also > modify PML as well? MPI collectives are handled from another framework -- not the BTL, but the "coll" (collective) framework. Look at ompi/mca/coll/coll.h. Most coll modules use the PML for transporting their messages. That is, most of our current coll modules concentrate on algorithms (e.g., different bcast/reduce/gather/etc. algorithms) and use whatever transport is provided underneath. So MPI_Bcast is likely already using the TIPC BTL for the individual point-to-point messages used by whatever algorithm is being used for the broadcast. Do you have some kind of TIPC native primitive for broadcast that you want to use? If so, you could write a tipc coll component that has just the bcast method (and NULL for everything else). -- Jeff Squyres jsquy...@cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
