Design and implementation of parallel programming language COIS based on UNITY

Abstract

According as scientific, technique as well as business users need to a high-performance computing power and parallel computers are emerging to satisfy their demands, a development of an efficient parallel programming environment is required. Today, programming parallelism is still very difficult for most programmers due to the fact that existing languages were originally developed for sequential computers. Programmers are often forced to program hardware-dependent features instead of programming parallelism using high-level abstraction. In this thesis, a new parallel programming language, COIS(concurrent iterative statements) is developed. It is a high-level, implicitly parallel language. It makes the abstract and architecture-independant parallel programming possible. The COIS is based on UNITY theory, which is proposed by Chandy and Misra. UNITY is a computational model based on unbouned iterative nondeterministic transition. Nondeterministic transition systems are natural method for specifying parallel systems. A UNITY program has one global state and transitions operating on it, possibly concurrently, it does not have an explict notion of a procase; it terminate implicitly, by reaching a fixed point state. In UNITY, statements are constructs to represent transitions. They should be executed atomically. COIS is designed with a computational model of UNITY with some modificationas; addition of new sections (macro and terminate), introduction of explicit termination scheme, definition of declarations and expressions, and some changes for programming notations. We implement a COIS-to-C translator in order to execute a COIS program on a hypercube multicomputer. The translator generates host and node programs with SPMD (single program and multiple data) style. The translator consists of two modules of front-end and back-end. The front-end module includes lexical analyzer, parser, and semantic processor. It transfers AST (abstract syntax tree) and DAL (data access...