In this paper, we study the straggler mitigation via coded computing in distributed computations. In particular, we consider the coded matrix-vector multiplication where the matrix is sparse and coding may break the sparsity of the matrix. We construct a class of sparse universal decodable matrices (UDMs) for coded computing. In simulations, it shows that the proposed code possesses better sparsity than other schemes with randomly generated sparse matrices. Besides, the proposed code performs well in numerical stability.