The Parallel Pre-processor: A Compiler For Distributed and Shared Memory Computers

Abstract

The goal in developing a robust parallelization tool is that it is easy to use, it requires minimal modifications to the original serial code, it is extensible to a wide variety of applications, and that it provide good portable performance. A directive-based parallelization tool is described called the Parallel Pre-processor (PPP) that meets most of these goals. The user inserts directives, in the form of comments, into existing Fortran code. PPP translates the code and directives into a parallel version that runs efficiently on shared and distributed memory high-performance computing platforms including: SGI Origin, IBM SP2, Cray T3E, SUN, Alpha and Intel Clusters. Twenty directives are available to support operations including array re-declarations, inter-process communications, loop transformations, and parallel I/O operations. PPP also provides support for incremental parallelization and parallel debugging.