https://hgpu.org/?p=9739
Computing Strongly Connected Components with CUDA