https://hgpu.org/?p=2026
New Row-grouped CSR format for storing the sparse matrices on GPU with implementation in CUDA