https://hgpu.org/?p=19304
Sparse matrix partitioning for optimizing SpMV on CPU-GPU heterogeneous platforms