https://hgpu.org/?p=18106
Design Principles for Sparse Matrix Multiplication on the GPU