https://hgpu.org/?p=6093
Matrix inversion speed up with CUDA