Algorithm Design for High-Performance Computers

摘要

In this paper we look at two ways of constructing algorithms for parallel architectures. The first is based on taking well-understood algorithms and restructuring them to isolate simple, computationally intense modules. We then parallelize the well-defined modules for the specific architecture. In the second approach we develop new algorithms for parallel architectures. Our main emphasis is on algorithms for solving linear algebra problems where the software we develop would reside in a library on high-performance computers. We also discuss a software environment for developing parallel algorithms in Fortran. This environment provides for ease of program development, debugging, and algorithm understanding. (ERA citation 13:038998)