https://hgpu.org/?p=2468
Speeding up Mutual Information Computation Using NVIDIA CUDA Hardware