High-performance Fortran (HPF) encompasses substantial language and compiler constructs for parallel machine architectures. HPF introduced an important software design scheme that augmented portability and programmer productivity. However, its compiler failed to generate the efficient machine codes required by various real-time software applications. HPF has perhaps not won over a dominant user community because of a lack of support by the persuasive parallel programmers in industry and academia. Will a widely acceptable model of parallel programming emerge in the future?
The authors call on the high-performance community to adopt innovative parallel programming models similar to the HPF model. They chronicle HPF’s contributions to concurrent programming and different parallel-machine categories. They creatively use a diagram and Fortran code segments to present the major HPF features augmenting parallelism, portability, and interoperability on multiple-instruction multiple-data machines. The authors insightfully discuss the unsuccessful efforts to standardize HPF, and the notable influences of HPF on the evolution of high-level parallel languages. Over the years, HPF progressed to support asymmetrical data distributions, task parallelism, and general block distributions suitable for parallel scripting languages.
Will Fortran--the earliest high-level programming language [1]--and HPF perish as long as there are mathematicians and engineers? This is a debatable question, given the variety of existing and emerging applications of parallel algorithms. All doomsayers and HPF advocates ought to read this appealing paper and partake in this debate.