https://hgpu.org/?p=7276
Asynchronous Parallel Computing Model of Global Motion Estimation with CUDA