Treffer: Compiler Assisted Distributed Memory Parallelization of an Iterative Solver for Irregular Sparse Linear Systems

Distributed memory parallel processors (DMPPs) can deliver high peak performance comparable or higher than vector supercomputers while promising a better cost-performance ratio. Programming comfort, however, is significantly lower than on traditional vector systems, especially when problems necessitating irregularly structured solution methods are considered. A class of such applications, with large resource requirements, is the numerical solution of highly nonlinear partial differential equations on irregularly refined three-dimensional finite-element discretizations. Porting such an application from vector (and shared memory parallel) machines to DMPPs involves some fundamental algorithm changes (for instance, grid decomposition, mapping, and coloring strategies). In addition, no standardized language interface is available to ease the efficient parallelization and porting among DMPPs and between vector computers and DMPPs. We have ported PILS---an existing package for the iterativ.