Application-Based Tuning of Mathematical Kernels on Petascale Systems

Petascale supercomputers offer the computational power to make significant progress toward the DoE¿s most complex electromagnetic problems. Existing simulation tools have to be carefully tuned to take advantage of the high degree of concurrency offered by these systems. This project will optimize a plasma physics modeling tool that is widely used within DoE's scientific community for leadershipclass supercomputing systems. Good performance on the largest-scale systems will be achieved by addressing algorithm scalability, single processor performance, and code portability. The physics algorithms will be separated from the numerical kernel routines, thereby taking advantage of highly tuned numerical libraries, obtaining good performance, and maintaining a high level of portability. Phase I will concentrate on the scalability of the sparse matrix/vector product and compare the performance of different implementations. Phase II will focus on single processor performance, as well as on the scaling of linear solvers. Commercial Applications and other Benefits as described by the awardee: This project should lead to a highly optimized commercial plasma physics code that fully utilizes leadership class supercomputing systems. In addition, the libraries and models developed will be made available to the scientific computing community.