High Performance Fortran Comes of Age: Guest Editors′ Introduction

Abstract

The High Performance Fortran (HPF) language was defined in 1992-1993 to address the need for high-level, portable, parallel programming for data parallel algorithms. The language provides directives which allow programmers to control the distribution of data across the processors of the underlying parallel machine. Computation is expressed in a distributionindependent manner using the Fortran 90 array syntax and the language extensions and directives provided by HPF. An HPF programmer expresses the parallel computation in a global index space without any explicit communication. It is the compiler's responsibility to analyze the code and generate the complex, low-level details of communication required for sharing data on the target machine. The specification of HPF version 1. 0 was published in 1993, in Scientific Programming, volume 2, numbers 1-2. Information on the current effort to develop a revision, version 2.0, is available at http: I /www. crpc. rice.edu/HPFF/home.html. The first commercial compilers for High Performance Fortran (HPF) are now available, and more are expected shortly. In addition, res!'arch groups have developed prototype implementations including advanced optimization and user interface capabilities. It is therefore an opportune time to consider the technologies required to make HPF usable and efficient. This special issue of Scient(fic Programming includes articles by researchers and commercial organizations describing their implementations of compilers and other HPF tools. The first three papers describe important HPF implementations. In their paper A Linear Algebra Framework for Static HPF Code Distribution, C. Ancourt,