机构地区: 广东金融学院计算机科学与技术系
出 处: 《微电子学与计算机》 2015年第3期63-66,72,共5页
摘 要: 2维数据以1维的形式进行处理和存储,2维空间相邻的节点被不同的映射算法映射到1维空间的不同位置.不同位置导致了不同的访问距离和访问延迟.提出了度量访问距离的指标,计算了4种主要映射算法的指标值,发现并分析了不同算法指标值之间的差异.通过在缓存中增加用于预取的空间,利用了Z-ordering和Hilbert映射算法在一定范围内能更好地簇集映射位置的特性,提高了缓存命中率,从而改善了2维数据的访问速度,优化了系统性能. Two-dimensional data is processed and stored in the form of one-dimensional.Different algorithms map 2-dimensional data nodes onto different linear addresses.These diverse addresses lead to various access distances and latencies.Several indicators are proposed in this paper to evaluate the distances made by the 4 major mapping algorithms.The value of these indicators are calculated and analyzed.In many cases,the Z-ordering and Hilbert mapping algorithms can aggregate the neighboring data nodes in a small linear scope.With this feature,increasing prefetch space in cache will enhance the cache hit rate,thereby improve the access speed of the two-dimensional data and optimize system performance.
领 域: [自动化与计算机技术] [自动化与计算机技术]