Massively Parallel Programming
with Parallaxis

Thomas Bräunl
Universität Stuttgart, Institut für Informatik
Azenbergstr. 12, D-7000 Stuttgart 1, FRG

e-mail: braunl@informatik.uni-stuttgart.de

In the Parallaxis programming language, the model of a parallel architecture is included as an integral part of the problem solution. That is, the combination of algorithm and machine model accounts for an entire specification. Since the choice of the computer architecture largely determines the structure of the algorithm applied, a structured parallel programming language should be given the expressive power of stating, or even better, selecting the structure of the parallel architecture, for which an algorithm is bound. Our language model Parallaxis operates on the class of SIMD array processors. In addition to the concurrent algorithm, Parallaxis allows the specification of arbitrary network topologies by means of a functional description with configuration and connection specifications. Simple concepts for concurrent execution of statements and message passing are based on these definitions. Variable declaration takes the parallel machine model into account and therefore splits into variables for the control unit (scalars) and variables for each of the parallel processing units (vectors).