Program Partitioning for a Control/Data Driven Computer

Abstract

The paper examines the problem of dataflow graph partitioning aiming to improve the efficiency of macro-dataflow computing on a hybrid control/data driven architecture. The partitioning consists of dataflow graph synchronization and scheduling of the synchronous graph. A new scheduling algorithm, called Global Arc Minimization (GAM), is introduced. The performance of the GAM algorithm is evaluated relative to some other known heuristic methods for static scheduling. When interprocessor communication delays are taken into account, the GAM algorithm achieves better performance on the simulated hybrid architecture.