University of Vienna
Institute for Software Technology and Parallel Systems
Liechtensteinstrasse 22, A-1090 Vienna, Austria
E-Mail: zima@par.univie.ac.at
Languages such as Vienna Fortran and High Performance Fortran (HPF) have been introduced to allow the programming of scalable parallel machines at a relatively high level of abstraction, based on the data-parallel Single-Program-Multiple-Data (SPMD) model. Their main features are focussed on the facility to express the distribution and alignment of data across the processors of a machine.
At the beginning of this lecture, we outline the basic features of such languages, essentially along the lines of current HPF, together with the related compilation methods and programming tools. In the main part of the talk, we identify some important properties of advanced numerical applications such as multiblock problems, particle-in-cell codes, and sweeps over unstructured meshes. These applications need a more general concept of data and work distribution than HPF currently offers. We propose a set of language extensions and present the compilation and runtime technology needed for the support of these new features. Much of this discussion will be motivated by research in the ESPRIT IV project HPF+, led by the University of Vienna.
The talk concludes with an outlook to future research in languages and compilation systems for scalable parallel machines.