Experiences of Performance Optimization for Large Eddy Simulation on Intel MIC Platforms

Abstract

Large Eddy Simulation (LES) is a mathematical model for turbulence used in Computational Fluid Dynamics (CFD). We have implemented LES on multi-core CPUs and General Purpose Graphics Processing Units (GPGPUs). In this work, we port and optimize LES on Intel Many Integrated Core (MIC) platforms. On Intel MIC co-processor (KNC), we implement LES using the main execution modes, including native, offload and symmetric execution modes. The newly emerging second generation of Intel MIC processor (Knights Landing, i.e. KNL) acts as an independent multi-core computing node, it is more convenient to port the application. On both of the MIC platforms, some important performance optimization techniques are implemented and evaluated, such as parallelization with OpenMP threads and MPI processes, single-instruction-multiple-data (SIMD) vectorization, memory access optimization, threads scheduling, etc. The experimental results demonstrate that performance optimization techniques are very important when porting applications on MIC platforms.