Iterative sparse matrix vector multiplication (SpMV) over GF(2) with CUDA

Abstract

Solving linear systems of equations (LSEs) is a very common computational problem appearing in numerous research disciplines. From a complexity theoretical point of view, the solution of an LSE is efficiently computable, e.g. by using for example the well known Gaussian elimination algorithm any LSE can be solved in at most cubic time. However, for some areas current algorithms and their sequential implementations are too slow. This is often due to the large dimension or number of LSEs that must be solved in order to accomplish a specific task.

To fulfil such purposes Iterative Sparse Matrix Vector Multiplication (SpMV)

algorithms need to be designed using useful languages like CUDA that enable

seamless communication with GPUs and this is the primary intent of this

project.