异构多核系统中协处理器优化

发布时间：2018-01-01 03:33

本文关键词：异构多核系统中协处理器优化　出处：《合肥工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：随着集成电路工艺技术的进步和对芯片更高性能的需求,传统的通过走不断小型化的路线来提升主频从而获得性能的提升的方法已经不能满足设计需求,片上多核系统(MPSoC)技术应运而生。相对于单处理器系统而言,多核系统具有低功耗、高并行度的优势。多核处理器的出现将微处理器的发展从追求单核心更高主频变为追求更多的处理器核。层次化、可拓展的大阵列将成为多核系统的发展方向。本文的研究对象就是一种基于NoC网络的小核大阵列SoC系统。在多核系统平台中,浮点协处理器的加入能够提高特殊运算的效率,提供更大的发挥空间给主处理器。同时给予软件编程足够的自由度,使多核编程简化,能够更好的处理子任务的控制,利于核间的协同合作,提高多核系统的整体性能。论文的主要工作如下：首先,分析异构多核系统下协处理器的设计需求,对已有协处理器的不足分析,并寻找改进方法。其次,通过研究乱序多发射计算机体系,设计了一种能够应用到面向高密度计算,能够进行批量运算及杂散运算的协处理器上的指令乱序多发射方案。最后,根据协处理优化方案完成RTL设计并实现,通过实验对系统性能进行分析及评估,证明了该优化设计的性能优势和实用性。
[Abstract]:With the progress of the integrated circuit technology and the demand for higher performance of the chip, the traditional method of improving the main frequency by continuously miniaturizing has been unable to meet the design requirements. Multi-core system on chip (MPSoC) technology emerges as the times require. Compared with single-processor system, multi-core system has low power consumption. The advantage of high parallelism. The emergence of multi-core processors will change the development of microprocessors from the pursuit of single-core higher main frequency to the pursuit of more processor cores. Extensible large array will become the development direction of multi-core system. The research object of this paper is a small core large array SoC system based on NoC network. The addition of floating-point coprocessor can improve the efficiency of special operation, and provide more space for the main processor. At the same time, it gives the software programming enough freedom to simplify the multi-core programming. It can better control the sub-tasks, facilitate the cooperation between cores, improve the overall performance of multi-core systems. The main work of this paper is as follows: firstly, the design requirements of coprocessor in heterogeneous multi-core systems are analyzed. The shortcomings of the existing coprocessor are analyzed, and the improved method is found. Secondly, by studying the scrambled multi-transmission computer system, a new method which can be applied to high-density computing is designed. The instruction scrambling multi-transmission scheme on the coprocessor which can carry out batch operation and spurious operation. Finally, according to the coprocessing optimization scheme, the RTL design and implementation are completed, and the system performance is analyzed and evaluated through experiments. The performance advantages and practicability of the optimized design are proved.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP332

【相似文献】