A Generalized Software Toolkit for Portable GPU-Enabled Chemistry Acceleration in CFD Applications

Current combustor design simulations aimed at reducing greenhouse gas emissions and improving fuel-lean combustion have entailed using large amounts of dedicated CPU resources for extended time periods due to the expense of solving detailed, strongly-coupled, chemical kinetic models. Such models are inherently data parallel, and much faster solutions can be obtained using low-cost graphics processing unit (GPU) hardware without loss of accuracy. This proposal describes development of a user-friendly software toolkit that facilitates implementing detailed or reduced fuel chemistry solvers directly onto GPUs to substantially accelerate CFD simulation runtimes. The approach is significant because it provides a cost-effective path to substantially reduce the wall-clock times currently bottlenecking high-fidelity combustion simulations. It accommodates the incorporation of self-contained, real fuel kinetic mechanisms and validated chemistry solvers, written using standard GPU-recognized program language extensions such as CUDA and OpenCL, for use in CFD analyses with minimal end-user code modifications. Using inputs that are Chemkin-format compatible, the proposed software toolkit will generate portable, GPU-enabled kernels that can be directly compiled into existing CFD codes, such as the National Combustion Code (NCC), to accelerate detailed combustion simulations for improved design support.