<span class="var-sub_title">Refactoring and Optimizing Multiphysics Combustion Models for Data Parallelism</span> SC18 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Refactoring and Optimizing Multiphysics Combustion Models for Data Parallelism

Authors: Christopher Stone (US Department of Defense HPC Modernization Program, Engility Corporation), Alexei Poludnenko (Texas A&M University), Brian Taylor (US Air Force Research Laboratory)

Abstract: High-fidelity combustion simulations combine high-resolution computational fluid dynamics numerical methods with multi-physics models to capture chemical kinetics and transport processes. These multi-physics models can dominate the computation cost of the simulation. Due to the high cost of combustion simulations and the important role simulations play in propulsion and power research, acceleration methods are needed to reduce the computational time and cost. Multi-physics models within each mesh cell are often independent leading to significant parallelism. However, the iterative algorithms often impede efficient SIMD data parallelism, a key performance feature on modern HPC systems. Refactoring methods for multi-physics models (e.g., kinetics, equation-of-state, diffusion) with nonuniform workloads are demonstrated and benchmarked on a range of platforms (AVX2, KNL, AVX-512). Realized speed-ups over 6x were achieved on KNL and 4x on Skylake (SKX) for complex chemical kinetics models and over 3x on SKX for iterative EOS computations.

Best Poster Finalist (BP): no

Poster: pdf
Poster summary: PDF

Back to Poster Archive Listing