COREFL: An open-source GPU-accelerated high-fidelity solver for compressible reactive flows on generalized curvilinear coordinates
Description
COmpressible REactive Flow soLver (COREFL) is an open-source computational fluid dynamics solver for high-fidelity simulations of compressible reactive flows. COREFL is written in C++/CUDA, and parallelized by message passing interface to achieve large scale computations on modern high-performance computing architectures represented by multi-CPU/GPU clusters. The solver solves the compressible Navier-Stokes equations coupled with species conservation equations, where a finite difference framework in structured curvilinear coordinates is adopted. For scale-resolving simulations of compressible reactive flows, a hybrid of the seventh-order Weighted Essentially Non-Oscillatory scheme and the linear upwind scheme is used to discretize convective terms. The transport properties are evaluated by the mixture-averaged model to treat the viscous terms accurately, while time integration of detailed chemical kinetics is handled via a balanced splitting method to ensure both efficiency and robustness. The static polymorphism based on template class/function of C++ is used to provide a flexible way of extending the codes to additional physical models without losing runtime performance and without changing code structures. Data structures are designed, and computational logics are optimized to fully exploit the power of modern GPUs. COREFL is validated with benchmark cases covering multiple scenarios. The solver is proved to be capable of simulating compressible reactive flows efficiently and acquires reliable results. Finally, a speedup over 800 is achieved on an Nvidia A100 GPU compared to a CPU code developed by the authors in reactive cases with a kinetic mechanism of 9 species, 19 reactions.