Sparse matrix-vector multiplication on GPGPUs