Parallel matrix transposition and vector multiplication using openMP

Abstract

In this chapter, we propose two parallel algorithms for sparse matrix transposition and vector multiplication using CSR format: with and without actual matrix transposition. Both algorithms are parallelized using OpenMP. Experimentations are run on a quad-core Intel Xeon64 CPU E5507. We measure and compare the performance of our algorithms with that of using CSB scheme. Our experimental results show that actual matrix transposition algorithm is comparable to the CSB-based algorithm; on the other hand, direct sparse matrix-transpose-vector multiplication using CSR significantly outperforms CSB-based algorithm. © 2013 Springer Science+Business Media New York.