|
Optimizing IRS Benchmark for IBM Cell BE Processors
Goal
The goal of this project is to run the IRS (Implicit Radiation Solver) parallel benchmark [1][2][3] on multiple IBM cell processor [5] nodes using MPI and accelerate the runtime of the benchmark on each node using the multiprocessing capability available within the IBM Cell BE processor (PPE and SPEs).
Purpose of IRS Benchmark The name IRS stands for Implicit Radiation Solver [1][2][3]. IRS solves a diffusion equation on a three-dimensional, block structured mesh. The single CPU instruction mix for this benchmark is roughly 40%load/store, 18% floating point, 31% fixed-point, and 11% branch. More Resources |