Computer architectures for functional and logic languages

Abstract

In recent years interest in functional and logic languages has grown considerably. Both classes of language offer advantages for programming and have an influential group of people promoting them. As yet no consensus has formed as to which class is best, and such a consensus may never form. Future general-purpose computer architectures may well be required to support both classes of language efficiently. Novel architectures designed to support both classes of languages could even add impetus to the area of hybrid functional/logic languages. Treleaven et al[68] have proposed a classification of computational mechanisms which they believe underly several types of novel computer architecture (i.e. control flow, data flow and reduction). The classification partitions novel general-purpose architectures into the following classes: control driven - where a statement is executed when it is selected by flow(s) of control, data driven - where a statement is executed when some combination of its arguments are available, and demand driven - where a statement is executed when the result it produces is needed by another, already active instruction. This thesis investigates the efficient support of both functional and logic languages using an architecture that attempts to be general purpose by embodying all the mechanisms that underly the above classification. A novel packet communication architecture is presented which intergrates the control driven, data driven and demand driven computational mechanisms. A software emulator for the machine was used as the basis for separate implementations of functional and logic languages, which were in turn used to evaluate the effectiveness of the computational mechanisms described in the classification. These mechanisms allowed functional languages to be implemented wi th ease, but caused severe problems when used to support logic languages. The difficulties with these mechanisms are taken as signifying that they do not provide adequate support for logic languages. The problems encountered led to the development of a novel implementation technique for logic languages, which also proved to be a good basis for a combined functional and logic model. This model is believed to provide a sound foundation for a parallel computer system that would support functional and logic languages with equal elegance and efficiency, and would therefore also support hybrid languages. The design for such a computer is described at the end of this thesis.