https://hgpu.org/?p=13756
Fast Sparse Matrix Multiplication on GPU