https://hgpu.org/?p=9459
A Tuned, Concurrent-Kernel Approach to Speed Up the APSP Problem