https://hgpu.org/?p=5579
Performing with CUDA