Performance Portability Analysis for Real-Time Simulations of Smoke Propagation Using OpenACC

摘要

Real-time simulations of smoke propagation during fires in complex geometries challenge engineers, physicists, mathematicians and computer scientists due to the complexity of fluid dynamics and the large number of involved physical and chemical processes. Recently, several application scenarios emerged that require real-time predictions during an incident to support the rescue teams. Therefore, we develop the CFD-based simulation software JuROr aiming to run in real-time by leveraging parallel computer architectures like CPUs and GPUs. For that, we parallelize the code with OpenACC directives that promise maintenance of a single source base by delegating some architecture-agnostic optimizations to the compiler. We investigate the performance portability of JuROr using PGI's OpenACC implementation across four Intel CPUs and three NVIDIA GPUs. We present the achieved performance shares as part of a roofline model where we focus on traditionally-computed arithmetic code intensities, as well as on a measurement approach based on performance counters.