https://hgpu.org/?p=4173
Efficent multiple pass, multiple output algorithms on the GPU