论文部分内容阅读
面向按序执行处理器开展预执行机制的设计空间探索,并对预执行机制的优化效果随 Cache 容量和访存延时的变化趋势进行了量化分析。实验结果表明,对于按序执行处理器,保存并复用预执行期间的有效结果和在预执行访存指令之间进行数据传递都能够有效地提升处理器性能,前者还能够有效地降低能耗开销。将两者相结合使用,在平均情况下将基础处理器的性能提升 24. 07% ,而能耗仅增加 4. 93% 。进一步发现,在 Cache 容量较大的情况下,预执行仍然能够带来较大幅度的性能提升。并且,随着访存延时的增加,预执行在提高按序执行处理器性能和能效性方面的优势都将更加显著。
In order to explore the design space of the pre-enforcement mechanism, the optimizer effect of the pre-enforcement mechanism is quantitatively analyzed with the trend of cache capacity and memory latency. Experimental results show that processor performance can be effectively improved by executing processors in sequence, saving and reusing valid results during pre-execution, and transferring data between pre-fetch instructions. The former also can effectively reduce energy consumption Overhead 93%. The combination of the two, under the average performance of basic processor performance 24. 07%, while energy consumption increased by only 4.93%. Further, it is found that pre-execution can still bring about significant performance gains in the case of large cache capacity. And, as memory latency increases, the benefits of pre-execution in order to improve processor performance and energy efficiency are even more pronounced.