基于交织预取率的帮助线程预取质量调节算法

Helper thread prefetching quality adjust algorithm based on prefetching rate

预执行帮助线程在预取过程中需要进行动态预取调节,而传统静态枚举控制参数值的控制方法在预取执行过程中保持固定不变,从而使得该方法不能够有效地为主线程提供预取质量保证(quality of service,Qo S)。针对该问题,提出了一种基于交织预取率的帮助线程预取质量参数调节方法。首先,对帮助线程的预取Qo S优化进行了建模分析;其次,在前期交织预取工作的基础上,提出了基于交织预取率的帮助线程参数值调节算法;最后,在真实的商用多核平台上对所提出帮助线程预取调节算法进行了评测和分析。实验结果是所提出的帮助线程预取调节算法使得基准测试程序的几何平均性能加速比为1. 114,而传统静态枚举方法的几何平均性能加速比为1. 135。实验结果表明,所提出的帮助线程预取质量调节算法解决了帮助线程预取过程中的参数值自动调节问题,算法无须静态枚举参数值便可以快速获得与之相近似的预取性能提升。

Pre-execution helper threads were needed to perform dynamic prefetching in the prefetch process. However,the traditional static enumeration parameter value was keep constant during in the execution of prefetching,so that the method can not effectively provide prefetching quality. To solve this problem,this paper proposed a method of prefetching quality control based on prefetching rate. Firstly,it analyzed the model of helper thread prefetching QoS optimization. Secondly,based on the previous work,this paper put forward the algorithm of parameter value adjustment based on prefetching rate. In the end,it evaluated and analyzed the algorithm on the real commercial multi-core platform. Experimental results were as follows. With the proposed prefetching adjusting algorithm,the geometric average performance speed up of the benchmarks was 1. 114. It was similar to the traditional static enumeration method’s geometric average performance speed up which was 1. 135. The experimental results show that the proposed algorithm can solve the helper thread prefetching quality adjust problem,as well can quickly obtain the similar prefetching performance compared with the static enumeration method.