https://hgpu.org/?p=16374
A survey of sparse matrix-vector multiplication performance on large matrices