https://hgpu.org/?p=11291
Hybrid strategy for stencil computations on the APU