Performance Optimization of D3Q19 Lattice Boltzmann Kernels on Intel KNL

摘要

This work discusses and assesses the impact of fundamental code optimization steps performed to maximize computing performances and memory throughput on Intel? Knights Landing (KNL) processor for Lattice Boltzmann (LB) applications. The benefits of using different memory data layouts is presented in regards to the most computationally intensive kernels of such applications, reporting performance results measured for the LBE3D code developed at the Applied Physics Department of the Eindhoven University of Technology, and run on a single KNL node for a common flow simulation case. We finally analyze and discuss the impact of different memory layouts on energy efficiency.