A System for Performance Porting of Iterative Structured Grid Applications in HPC Environments

Abstract

This paper presents a system for the efficient implementation of cellular automata problems on heterogeneous HPC platforms, with separation of the numerical models developed by domain scientists from the low-level details that are typically handled by experienced programmers familiar with MPI+X programming models. We describe a framework designed to reduce development and maintenance time required for new applications, which can also be used to port existing applications to a new HPC environment. We demonstrate how to use the framework to implement cellular automata problems while highlighting the various differences in performance between framework-assisted and native implementations. The evaluation of the framework shows that the performance of application developed using the frame work is comparable to the same application optimized for a specific architecture.