Context streams a theoretical basis for a generic form of MIMD pipelining

Abstract

Context flow (CF) is a canonical form of MIMD pipelining in which process contexts rather than either instructions or data are manipulated. The paper presents a formal definition of context flow. It defines the state of a CF computation as the stream of contexts which pass a particular point in a context flow graph during a given period of time. It shows that for a class of context flow graphs corresponding to dynamic multifunctional pipelines, the number of contexts which may exist at any point in a graph is bounded, and independent of the computation being performed. It presents methods for determining the numerical value of the upper bound or queue length for a variety of configurations of a acyclic graph.<>