https://hgpu.org/?p=10441
A Scalable, Efficient Scheme for Evaluation of Stencil Computations over Unstructured Meshes