Parallel Computing 3 (1986) 305-326 305
North-Holland
Restructuring SIMPLE for the
CHiP architecture
Dennis GANNON * and Jairo PANETTA
Department of Computer Sciences, Purdue University, West Lafayette. IN 47907, U.S.A.
Received May 1985
Abstract. The SIMPLE program is a commonly used benchmark for testing new architectures designed for high
speed scientific computatt,~n. As the name implies, the code is a simple example of a Lagrangian hydrodynamics
application. In this paper we describe the SIMPLE benchmark in detail and discuss the way in which parallelism
can be used to speed up execution. The focus of the work is a mapping of the algorithms to a configurable
highly parallel (CHIP) computer being designed at the University of Washington.
Keywords. Parallel numerical algorithms, Lagrangian hydrodynamics, parallelism, architecture, configurable
networks, performance analysis, SIMPLE, CHIP.
1. Introduction
Research in the area of solving partial differential equations on parallel computers has been
going on for fifteen years. While a wide variety of algorithms and software has been developed
for vector and SIMD array processors, only recently has there been much research or
experimentation with large scale MIMD parallelism. For the most part, new research for these
highly parallel systems has concentrated on solving simple scalar elliptic PDEs on regular
domains. In terms of algorithm design and analysis there still is much research that remains to
be done for these problems. On the other hand, there are aspects of certain applications, such
as computational fluid dynamics and structural mechanics, that pose important problems for
the designers of new computer architectures. Among the important issues that deserve more
attention by the computer architecture community are:
(1) Parallel computation algorithms for coupled non-linear systems of PDEs such as the
Navier-Stokes equations;
(2) Problems where the geometry is complex and the structural data base is very large;
(3) The problems associated with the parallel processing of locally refined or dynamically
self adaptive grids.
Various studies have touched on each of these issues. Zave and Cole [17], and Gannon and
Van Rosendale [7] have considered items (2) and (3) for scalar problems; Adams has
considered systems of linear PDEs [1]; and Johnson [10], and Greenberg and Stevens [8] have
done a substantial amount with solutions of computational fluids problems on MIMD systems
with a small number of processors. The Denelcor HEP has stimulated other good work.
* Present address: Department of Computer Sciences, Indiana University, Bloomington, IN 47405, U.S.A.
016%8191/86/$3.50 © 1986, Elsevier Science Publishers B.V. (North-Holland)