Statically Scheduling Communication Resources in Multiprocessor DSP Architectures

In statically scheduled multiprocessors inter-processor communication resources can be scheduled by determining, at compile time, the order in which processors require access to shared resources and enforcing this order at run time. We show how to choose an access order such that, under certain assumptions, imposing that order incurs no performance penalty. 1.0 Introduction In this paper we address multiprocessor implementation of applications that are specified as Synchronous Data Flow (SDF) graphs. We recall that in a dataflow representation, an algorithm is represented as a graph where nodes (actors) are individual computations and directed arcs between them represent flow of data (tokens). SDF refers to a subclass of dataflow graphs where the actors lack data dependency in their firing patterns. I.e. the number of tokens produced and consumed in each of the output and input arcs of each actor is constant and fixed at compile time. In this paper we assume that the application is a..