Abstract: We present optimizations for sparse matrix-vector multiply SpMV and its generalization to multiple vectors, SpMM, when the matrix is symmetric: (1) symmetric storage, (2) register blocking, ...