https://hgpu.org/?p=18768
Using Compiler Directives for Performance Portability in Scientific Computing: Kernels from Molecular Simulation