Sparse Matrix-Vector Multiplication on GPGPUs

摘要: The multiplication of a sparse matrix by dense vector (SpMV) is centerpiece scientific computing applications: it the essential kernel for solution linear systems and eigenvalue problems iterative methods. efficient implementation matrix-vector therefore crucial has been subject an immense amount research, with interest renewed every major new trend in high-performance architectures. introduction General-Purpose Graphics Processing Units (GPGPUs) no exception, many articles have devoted to this problem.With article, we provide review techniques implementing SpMV on GPGPUs that appeared literature last few years. We discuss issues tradeoffs encountered various researchers, list solutions, organized categories according common features. also performance comparison across different GPGPU models set test matrices coming from application domains.