面向ARM64架构多核微处理器的模板计算性能优化研究
发布时间:2018-01-10 19:22
本文关键词:面向ARM64架构多核微处理器的模板计算性能优化研究 出处:《计算机工程与科学》2017年05期 论文类型:期刊论文
更多相关文章: 模板计算 ARM AMCC X-GENE FT-A 并行化 线程绑定
【摘要】:模板计算是一类重要的计算核心,广泛存在于图像和视频处理以及大规模科学和工程计算领域。但是,针对ARM64高性能处理器的模板计算性能的优化研究还很少。为了实现典型模板计算核心在ARM64架构多核微处理器上的并行化和性能优化,基于AMCC X-GENE2和飞腾FT-1500A多核微处理器特点,提出了基于两维度绑定的优化方法,该方法通过线程与CPU绑定以及线程与数据块绑定,减少了线程调度的并行开销,增加了Cache的命中率。实验结果表明,该方法提升了模板计算在ARM64架构多核微处理器上的性能,且在两种ARM64架构多核微处理器平台上都表现出较好的可扩展性。
[Abstract]:Template calculation is an important kind of computing core, widely exists in the image and video processing and large-scale scientific and engineering computing. However, optimization research for ARM64 high performance processor template computing performance is less. In order to realize the typical template calculation core in ARM64 frame parallelization and performance optimization of multi core microprocessor, AMCC X-GENE2 and FT-1500A intelligent multi core microprocessor based on the proposed optimization method based on the two dimensions of the binding, through the method of thread and CPU binding and thread with a block of data binding, reducing the overhead of thread scheduling, increase the hit rate of Cache. The experimental results show that this method improves the performance of the template computation in ARM64 architecture the core microprocessor, and in two ARM64 architecture multi-core microprocessor platform showed good scalability.
【作者单位】: 国防科学技术大学计算机学院;
【基金】:国家自然科学基金(61170046) 国家863计划(2012AA010903)
【分类号】:TP332
【正文快照】: A57[3]是两款主流的64位ARM高性能处理器核1 引言心。ARMv8[1]体系结构的发布把ARM架构推向了大规模企业服务器领域。近年来,ARM架构发展迅速。2011年10月,模板计算是一类科学和工程计算领域的重要ARM公司正式发布了处理器架构ARMv8,计算核心,它具有计算密集和访存密集的特点,
本文编号:1406518
本文链接:https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/1406518.html