Tuning the ANSYS kernel LSOLVE for a parallel computer

Abstract

A solver kernel, LSOLVE, has been added to ANSYS, a finite-element analysis code. LSOLVE is a rank-n update routine that can be implemented efficiently on parallel and vector machines with a hierarchical memory structure. An implementation of LSOLVE on a parallel computer is described by examining the Fortran code used on the Alliant parallel computer. The utility of LSOLVE is demonstrated by performance on standard ANSYS benchmarks.<>