https://hgpu.org/?p=8345
Programming NVIDIA cards by means of transitive closure based parallelization algorithms