Towards High-Performance Optimizations of the Unstructured Open-Source

This article presents a snapshot of ongoing efforts to optimize an open-source CFD anal-ysis and design suite, SU2, for high-performance, scalable Reynolds-averaged Navier-Stokes calculations using implicit time integration. We focus on performance optimizations with a particular emphasis on code profiling, the opportunities for parallelism of the software components, and finding highly-scalable algorithms. Consequently, the resulting code mod-ifications are geared toward achieving coarse- and fine-grained parallelism for edge-based, finite volume CFD solvers, making efficient use of memory within a heavily object-oriented solver, and choosing appropriate algorithms for maximizing parallelism, especially when solving the linear systems arising from implicit time integration of the governing equa-tions. All lessons learned are reported for the benefit of both the users and developers of SU2, as well as the larger CFD community for pursuing performance improvements in similar analysis and design software packages. I