同一个程序运行在不同微体系结构配置的处理器上时,不同处理器的软错误率通常会有非常大的差别.所以,为了设计高可靠处理器,对处理器微体系结构设计空间进行搜索,选择出高可靠的配置,是至关重要的.然而,目前的处理器软错误率预测方法忽视了执行程序的影响,导致微体系结构设计空间搜索仍然需要大量周期精确的模拟.因此,本文提出了基于同时考虑程序固有特征(与微体系结构无关的程序特征)和处理器微体系结构配置而构建的统一的软错误率预测模型的设计空间快速搜索方案.该方案对新程序或新处理器配置都不需要额外的周期精确的模拟就可以准确预测软错误率,从而极大地缩短了高可靠处理器微体系结构设计空间的搜索时间.对SPEC CPU2000基准测试程序的实验结果表明,与基于反应构建软错误率预测模型的方案相比较,该方案搜索高可靠处理器微体系结构设计空间所需的模拟时间减少了99.8%.
Since the soft error rate of the same program varies widely across the different microarchitecture configurations, it is crucial to do the microarchitecture design space exploration to design highly reliable processors. However, the current methods of predicting soft error rate do not consider program characteristics, so they need excessive timeconsuming simulations for microarchitecture program pairs. In this paper, the authors propose a novel approach which utilizes inherent program characteristics coupled with microarchitectural configurations to construct a universal predictive model of soft error rate so as to accelerate the microarchitectural design space exploration. For a new program or a new microarchitectural configuration, the proposed approach could predict the soft error rate of processors accurately, without additional cycle-accurate simulations. Therefore, it significantly reduces the overall simulation time for design space exploration of the highly reliable processors. The results of the experiment on SPEC CPU2000 benchmarks show that the proposed approach can reduce 99.8% of the simulation time compared with the previous reaction-based approaches.