Parallaxis-III: A Language for Structured

Data-Parallel Programming

Thomas Bräunl
Computer Science Dept., IPVR
Computer Vision Group
Univ. Stuttgart, Germany

Abstract:

Parallaxis is a machine-independent language for data-parallel programming. Sequential Modula-2 is used as the base language. Programming in Parallaxis is done on a level of abstraction with virtual processors and virtual connections, which may be defined by the application programmer. This paper describes Parallaxis-III, the current version of the Parallaxis language together with a number of parallel sample algorithms.

1.Parallel Language Constructs

Parallaxis [Bräunl 89], [Bräunl 91], [Bräunl 93] is based on Modula-2 [Wirth 83], extended by data-parallel concepts. The language is fully machine-independent across SIMD architectures; therefore programs written in Parallaxis run on different SIMD parallel computer systems. For a large number of (single-processor) workstations and personal computers, there is also a Parallaxis simulation system with source level debugging and tools for visualization and timing. A compiler for data-parallel programming of MIMD computer systems in SPMD style (same program, multiple data) is being developed. Parallel programs with small data sets can be developed, tested and debugged with this simulation system. Then, Parallaxis compilers can be used to generate parallel code for the MasPar MP-1 / MP-2 or the Connection Machine CM-2. The simulation environment allows both the study of data parallel fundamentals on simple computer systems and the development of parallel programs which can later be executed on expensive parallel computer systems. The programming environment for Parallaxis is available as public domain software [Barth, Bräunl, Engelhardt, Sembach 92].

We had two major goals in developing a new parallel programming language:

  1. We believe that `structured programming' has a number of advantages in developing sequential and parallel software, as well as for learning programming concepts in general. Therefore, we chose Modula-2 as base language and not C like many other approaches.
  2. Almost all commercial SIMD programming languages are machine-dependent, since they have been specifically designed for a single hardware platform to achieve maximum performance. Therefore, one may not easily port a program from one SIMD system to another. To avoid this problem, we designed Parallaxis to be completely machine-independent. This refers to SIMD architectures only, for switching from SIMD to MIMD will almost always require to change the algorithm of your program, in order to keep it efficient.
The central point of Parallaxis is programming on a level of abstraction with virtual PEs and virtual connections. In addition to the algorithmic description, every program includes a semi-dynamic connection declaration in functional form. This means that the desired PE network topology is specified in advance for each program (or for each procedure) and can be addressed in the algorithmic section with symbolic names (instead of complicated arithmetic index or pointer expressions). However, for completeness, also full-dynamic data exchange operations may be performed.

The following describes the latest version of the language, Parallaxis-III, which is not fully compatible to older versions of the language.

[ Table of Contents | Next Page ]