LocOp with Dynamic Heterogenous Task Scheduling Performance Model

[vlkale]

Can there be a way to enable local operations (LocOp) that are determined through a performance model of work-stealing or a dynamic heterogeneous processor loop/task scheduling along with the cache/memory hierarchy? Such a model would also adapt to the MPI 'slack' observed in the previous invocation of the MPI communication function. Note that this performance model includes heterogeneous processors.

I work on a draft PR # to do this at GitHub.com/vlkale/logopsim .