Architecture-Independent Locality Analysis and Efficient PRAM Simulations

We introduced an approach to parallel computing which unites the automatic and direct programming paradigms within the core BSPlib environment. Our PRAM simulator and companion C++ macro-based language[LS96] is scalable and portable; it has been tested on the IBM SP2, Cray T3D, SGI Power challenge, and a cluster of Sun Workstations. Directly-programmed solutions to regular problems inevitably can obtain greater performance than our simulator. Our approach wins for irregular problems with poor locality.