Condensed Graphs: A Multi-level, Parallel, Abstract Machine

Abstract

This paper outlines a programme of research based on a recently developed model of computing[ 1, 2, 3] which uses the concept of a Condensed Graph as an intermediate representation of functional and imperative languages. To efficiently execute Condensed Graphs, a multi-level abstract machine is used. This machine incorporates characteristics and feedback information of underlying architectures and guides dynamic program transformations through many levels of abstraction. These dynamic transformations can add parallelism and change evaluation orders. The CG abstract machine dynamically constrains the parallelism, preventing it getting out of control. Simulations of the intermediate representation which utilize varying evaluation orders on an idealized Condensed Graph machine are presented.