帮助线程预取质量的实时在线评价方法

针对传统静态枚举设置帮助线程控制参数值的繁杂耗时问题，提出了一种帮助线程预取质量的实时在线评价方法。首先，明确了帮助线程的预取服务质量（QoS）的目标；其次，分析了帮助线程预取性能评价的动态指标，对帮助线程预取QoS进行了建模分析；最后，提出一个帮助线程预取的动态自适应调节算法，算法根据程序的阶段行为变化和动态预取获益变化等信息来判断参数值的适用度以及是否需要进行反馈优化，从而实现对预取控制的自适应调节。实验结果表明，应用自适应预取评价算法之后，Mst热点模块的性能提升加速比为1.496，所提出的自适应预取评价方法能够根据程序的动态阶段行为对帮助线程控制参数值作出自适应控制和调节。

Real-time online evaluation method of helper thread prefetching quality

Focusing on the multifarious and time-consuming optimization process of traditional helper thread parameter value enumeration method, a real-time online helper thread prefetching quality assessment method was proposed. First, the help thread prefetching Quality of Service (QoS) target was defined. Second, the dynamic evaluation index of helper thread prefetching quality was analyzed, as well the helper thread prefetching QoS model. Finally, a dynamic and adaptive helper thread prefetching adjustment algorithm was presented. The algorithm was based on phase behavior and dynamic prefetching benefit information to determine the suitable degree of parameter values, and whether to need feedback optimization, so as to realize the adaptive adjustment and control of helper thread prefetching. By applying the adaptive prefeching algorithm, the speed up of Mst's hotspot module was 1.496. The experimental results show that the proposed adaptive prefetching evaluation method can control parameter values adaptively according to the dynamic phase behavior and prefetching benefit information.