GENERATION OF COARSE GRAIN TASKS FROM APPLICATIVE PROGRAMS FOR MULTIPROCESSING

Abstract

The higher speed has always. been a desirable thing for a computing machine. From the days of the -advent of the modern computer systems, scientists and engineers have worked hard to achieve higher and higher speeds. The computing speed of a single processor-system has almost reached the theoretical limits. It has been now established, beyond doubt, that the thirst for higher speed can be quenched only by exploiting the parallelism in the problem using a multiprocessor system. The functional languages, due to their fundamental properties, have been found to be suitable for parallel processing. Some of the issues in implementing functional languages on multiprocessor architecture are partitioning the problem into various tasks and scheduling these tasks onto the various processors. Here in this dissertation, the first problem has been addressed. As the-coarse—grain parallelism has an edge over the fine—grain one, partitioning of a problem into coarse—grain tasks has been considered. The process of generation of' tasks has been enriched by the evaluation of a constant expression using only built in operators. Finally, various suggestions for improvement in this work and for future work have been included.