Implementation Of Run Time Techniques In The Polaris Fortran Restructurer

INTRODUCTION 1.1 Motivation Parallel computing can provide very high levels of performance for scientific applications. However, these applications must be specifically structured to take advantage of the power of parallel computers. There are two means of achieving this. The programmer can write explicitly parallel code, or the programmer can use a tool to automatically transform programs written in conventional sequential programming languages for parallel execution. Automatic parallelization is appealing because it is less tedious than writing explicitly parallel programs since the programmer is freed from worrying about the details of interprocessor communication, synchronization, and data distribution. In addition, it is easier to achieve good performance on a wide range of machines, because the task of tuning a program for a specific machine can be left to the restructurer. Unfortunately, automatic parallelization has had limited success because compilers in som