Multicore Optimization of an Astrophysical Simulation Code using Performance Annotations

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

Intellectual Merit: This project will demonstrate a performance annotations approach to tune a production simulation code for use with computers containing more than O(1000) multicore processors. The code to be optimized is FLASH, a publicly available astrophysical simulation code. The work will focus on three code units of broad interest to the astrophysical simulation community: hydrodynamics (Euler equations), particle transport, and gravitation (Poisson?s equation). PI will focus on two machines, the TACC Ranger machine and the Argonne BlueGene/P, both of which are new and relatively untested architectures. Broader Impact: The annotation based approach to tunig codes for petascale can have tremendous impact on a number of other "legacy codes" if successful