Algorithmic aspects of balancing techniques for pipelined data flow code generation

Abstract

This paper outlines the algorithmic aspects of balancing techniques for data flow code produced by a pipelined program transformation scheme for static data flow computers. The goal of optimal balancing is to introduce a minimum amount of buffering into data flow graphs such that their execution can be fully pipelined. The balancing procedure is based on the global structure of the code and the configurations of pipelined data flow. Based on a graph-theoretic formulation of the balancing procedures, a critical review of previous related work is given. Our discussions are primarily centered on the pipelining of acyclic connected data flow graphs generated from a data flow language. We show that the optimal balancing for such data flow graphs can be formulated into certain linear programming problems which have efficient algorithmic solutions.