EVON: AN EXTENDED VON NEUMANN MODEL FOR PARALLEL PROCESSING.

Abstract

The author proposes an extended von Neumann model called Evon for parallel processing. It is centrally controlled, and is capable of exploiting instruction-level as well as expression-level parallelism inherent in high-level language programs. Its effectiveness in exploiting parallelism depends crucially on powerful primitives, and the author illustrates this point by several short programming examples written in APL. A report is then presented on work on the design of an instruction set embodying the Evon model, and a portable compiler for that instruction set aimed at extracting parallelism automatically. Finally, a comparison is made of Evon with other computational models and architectures proposed for parallel processing with pragmatic considerations and programming concerns.