Skip to Main Content
Scientific modeling and simulations are popularly used in science and engineering communities to explain complicate phenomena or to extract knowledge from structured or unstructured data along with theoretical analysis and physical experiments. Generally, these models are represented as partial differential equations (PDEs) which can be solved numerically using meshes and sparse matrices. Typically, matrix vector multiplication is the most dominating module in the solution of PDEs. Therefore, efficient matrix vector multiplication algorithm is a critical component in scientific computing simulations. In this paper, we proposed a sparse matrix vector multiplication using compressed graph. Our experiments show that the proposed algorithm reduces cache misses by 65% at best with a little bit of memory overhead.
Date of Conference: 18-21 March 2010