Extending a 3D Parallel Particle-In-Cell Code For Heterogeneous Hardware
Grischa Jacobs · The 7th International Conference on Computational Methods (ICCM2016) · 2016
An evaluation for a parallel Particle-In-Cell code leveraging heterogeneous hardware is presented. Hybrid parallelization is implemented to support optional workload offloading to 40 Intel Xeon Phi coprocessors. A performance model is applied to load balance the particle data for this heterogeneous setup. Performance measurements of a benchmark show the speedups for the balanced and unbalanced cases and the execution without the coprocessor.