Charm++ parallel programming system

["Kale, Laxmikant V" <kale AT illinois.edu>]

Joseph, I am posting this to the Charm++ mailing list (please post such bug reports there in future). -- Sanjay

----------------------

Dear Professor Kale,

    While working on my project, I noticed a neat little bug with the SDAG 'forall' construct.

    Let's say we want to execute a 'forall' inside a 'when' clause:

when entryMethodA(int totalIterations) {

   forall [i] (1 : totalIterations, 1) {

       //do work here

   }

}

    The variable 'totalIterations' should be in scope for the 'forall' construct. But when we compile, the .def.h code gets a little bit wrong. It'll throw an error like "totalIterations" not in scope.

The offending line looks like: int __first = (1), __last = (totalIterations), __stride = (1);

This is the 'forall''s iteration construct. It appears that there is direct textual substitution from the SDAG code to the generated .def.h code.

Yet, the closures (such as entryMethodA) are passed to this method as parameters. Thus, the 'forall' generated code has access to the parameter called totalIterations. Instead of "totalIterations", though, it would be something like "gen0->getP0()"; (gen0 is the closure object representing the "when entryMethodA" scope)

Were 'forall's intended to have dynamic iteration? Given that it has access to its parent closures' parameters, I think it is in the realm of possibility.

--- 

Laxmikant (Sanjay) Kale         http://charm.cs.uiuc.edu

Professor, Computer Science     kale AT illinois.edu

201 N. Goodwin Avenue           Ph:  (217) 244-0094

Urbana, IL  61801-2302          FAX: (217) 265-6582