Evaluation of high performance fortran on some real applications

Abstract

Due to the existence of a global name space, the data parallel programming model is very comfortable to use. All communication requirements are implicit and therefore left to the compiler. To support this model even to general classes of parallel architectures, i.e. MIMD machines with distributed memory, High Performance Fortran was developed. Beside a data parallel point of view the programmer has the ability to specify data distributions explicitly via directives.

Some real application programs have been investigated to show the usefulness of High Performance Fortran. Therefore these programs have been transformed into the data parallel style and a prototype compilation system has been used to translate them to message-passing programs running on different platforms.

The results verify that the generated message-passing programs scale and can achieve nearly the same performance as their hand-written counterparts. The experiences have also shown where the data parallel programming model has difficulties and where additional optimizations are required.

This work was funded by the Esprit-6643 project PPPE (Portable Parallel Programming Environment)