论文部分内容阅读
研究目的:性能估计已成为异构MPSo C设计中一个非常重要且具有挑战性的任务。在设计早期进行准确快速估计性能对于设计空间探索十分必要。本文采用GCC剖析技术和代码标注技术,结合MPSo C分层抽象概念,探讨逐层次完善的性能评估技术在MPSo C体系结构探索中的应用。创新要点:为面向多媒体应用的MPSo C性能估计提出一个从VA层到TA层的剖析和标注相结合流程,使性能估计可以被有效逐层完善。研究方法:基于GNU gcov工具,在本机模拟过程中剖析给定应用程序代码执行的统计信息,并且支持实时性能分析,快速、准确估计VA模型的计算负载。基于VA模型得到的计算负载性能的结果标注,利用TA模型的System C时序精确级仿真得到通信延时结果,使TA模型性能估计更高效,完善整个MPSo C性能估计。重要结论:研究一个剖析和标注技术相结合的MPSo C性能评估方法和流程。在VA层得到准确计算负载性能并标注给TA层;在TA层利用基于标注的仿真方法完善通信延时,使得性能估计更高效。通过M-JPEG和MEPG2两个典型视频多媒体应用实验,展示本文方法的高效、快速与准确。
Research Purpose: Performance estimation has become a very important and challenging task in the design of heterogeneous MPSoCs. Accurate and fast estimation of performance early in the design is essential for exploring the design space. In this paper, GCC profiling technology and code labeling technology, combined with MPSo C layered abstract concept, to explore the level of perfection of the performance evaluation technology in MPSo C architecture exploration. Innovative points: For the MPSo C performance evaluation for multimedia applications, a combination of profiling and tagging from the VA layer to the TA layer is proposed so that the performance estimation can be effectively improved layer by layer. Research Methods: Based on the GNU gcov tool, we parse the statistics of a given application code execution in the native simulation process and support real-time performance analysis to quickly and accurately estimate the computational load of the VA model. Based on the result of calculating the load performance obtained by the VA model, the communication delay result is obtained by using the simulation precision of the System C timing TA model, which makes the TA model performance estimation more efficient and improves the estimation performance of the entire MPSo C. Important Conclusions: Investigate an MPSo C performance assessment methodology and process that combines profiling and labeling techniques. Accurately calculate the load performance in the VA layer and mark it to the TA layer. Improve the communication delay by using the annotation-based simulation method in the TA layer, which makes the performance estimation more efficient. By M-JPEG and MEPG2 two typical video multimedia applications experiment, demonstrate the method of this paper efficient, fast and accurate.