https://hgpu.org/?p=1197
A Task Parallel Algorithm for Computing the Costs of All-Pairs Shortest Paths on the CUDA-Compatible GPU