Charm++ parallel programming system

[Phil Miller <mille121 AT illinois.edu>]

On Sat, Dec 29, 2012 at 2:48 AM, Kale, Laxmikant V 
<kale AT illinois.edu>
 wrote:
> Then, are you (the programmer) required to make sure only one of the sdag
> methods is active at a time? Otherwise, there is an ambiguity as to which
> of them will advance if both flows are waiting  on (i.e.e expecting) the
> same entry method.

Yes, the programmer is responsible for making sure that there's no
ambiguity, or that it's not a problem. This could come by
1. limiting the control flow to just one of those methods executing at a time
2. using distinct tags in the two methods so that each matches unambiguously
3. writing the methods so that either one can process whichever
message it happens to get

AMR uses the first of these. CharmLU uses the first and the second -
pivoting can happen in a few different places, but it shares the same
methods.

> I'd prefer even a restrictive semantics, as long as it is unambiguous.

I'm not sure what you mean by this.