Charm++ parallel programming system

[Phil Miller <mille121 AT illinois.edu>]

As a quick expedient, you could use CkCallbackResumeThread either directly with the all-to-all routine (possibly a reduction from that array after they're all done) or with quiescence detection (QD).

On Thu, Mar 20, 2014 at 4:07 AM, Mustafa Abdul Jabbar <musbar AT gmail.com> wrote:

Hello,

I am porting an MPI based code to a Charm++ based one without major code reforms. 

Problem is from within the array node I want to do blocking all to all communication without using call backs to trigger completion, because those will make the code a big mess given its current structure. I have the following SDAG entry

entry void allToAllBodyCount()

{

atomic

{

transportBodyCountToAll(); // will broadcast to all nodes

}

for(count = 0; count < numChares; ++count){

            when transportBodyCount(int bodyCount, int sender) atomic 

{

                processBodyCount(bodyCount,sender);

            }

        }

}

I am calling allToAllBodyCount() locally from the array node, and I want to wait until I have processed incoming messages.

I have read about the [sync] methods, but there's no evidence that an SDAG method can be made sync since the SDAG translates the code above to a sort of Future Completion as I noticed in the generated header file

Any workaround?

_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm