Improving efficiency of parallel computations using parallel arrays

Abstract

One of the reasons why C/C++ programming language is considered less suitable for scientific computations is the belief that C++ programs produce less optimized code. C programs, on the other hand, may produce very optimal code, but this language is not as convenient as FORTRAN for the purposes of scientific programming. In this article, two complementing approaches are analysed: • The methods to improve efficiency of C++ programs while still using advanced C++ features such as templates, OOP, and others. These methods may help to improve both sequential and parallel programs. • For parallel programs, different ways to reduce communication costs are analysed. The methods described in this article are implemented in ParSol parallel array package. In this article, implementation specifics are described and performance test results are presented.