https://hgpu.org/?p=8233
Parallelization of a Block-Matching Algorithm