Performance Evaluation for Dense Sparse Matrix Product Algorithms

In optimizing the dense-sparse matrix product (DSMP), we study different versions of the algorithm relative to the compressed formats DNS, CSR and COO. Our aim is (i) first, to choose the best optimizations which improve the algorithm performances, (ii) second, to select the best format which provides the best running time. Our work is validated through a series of experiments achieved on a set of dense and sparse matrices with different sizes and densities.