Software Design of the HyperSolve CFD Tool

Recent efforts at NASA have focused on developing advanced unstructured mesh adaptation techniques and discretization technologies that reduce the burden of mesh generation on the CFD practitioner. The HyperSolve CFD code focuses specifically on leveraging advanced numerical methods to enable accurate aerothermal predictions for NASA missions with high-fidelity geometry requirements. HyperSolve uses economical discretization
methods to achieve higher-order accuracy on adapted simplex element meshes. Notably, this is achieved without requiring mixed-element meshes with prisms in the boundary layer. A JFNK nonlinear solver with nonlinear preconditioning is employed to solve the HyperSolve discretization efficiently. Automatic differentiation, using expression templates and SIMD instructions, is highlighted as an essential tool for the correct and efficient implementation of the Jacobian matrix required for preconditioning the linear system. A focus on software architecture has enabled the rapid development of the HyperSolve code. Adherence to the SOLID principles promotes code that is both readable and main-tainable. Specifically, the dependency inversion principle encourages the composability of features without creating highly fragile and rigid software. The dependency inversion principle is also used to enable the self-registration of subcommands, a technique that simplifies the organization and documentation of the software’s capabilities.