https://hgpu.org/?p=3271
Implementing a GPU Programming Model on a non-GPU Accelerator Architecture