https://hgpu.org/?p=11606
Efficient Preconditioned Conjugate Gradient Parallelization on GPU