Adaptive tuning of sparse matrix-vector multiplication on Cell architecture

摘要: Sparse matrix-vector multiplication is a kernel which widely used in scientific applications. The sparse-data usually stored the compressed row storage format, introduces irregular reference pattern. It's problem for software cache on Cell architecture because line always set to specific size traditional strategies, limits utilization and increases memory bandwidth overhead. In this paper, we propose an adaptive strategy continuously adjusts during SpMV execution. Therefore, transferred data execution time are significantly decreased. Moreover, prefetching scheme proposed further improve performance. evaluation indicates that our achieves speedup factor from 2.11 3.57 compared approach. And approach translates into up 3.2 multiplications.