面向流计算的能量自适应多核处理器设计与实现

发布时间：2018-04-14 07:41

本文选题：多核处理器 + 能量自适应　；参考：《电子科技大学》2013年硕士论文

【摘要】：近年来兴起的能量获取方式和能量输送方式给集成电路带来了新的机遇和挑战。在构建新型的电子系统能量消耗与供应模型问题上，人们提出了能量自适应的概念，并开始研究能量自适应处理系统架构及其功耗模型。通过例如基于片上网络的研究使电子系统具备能量自适应的能力，使之能适应不同的能量供应方式以及能量状态，可以使电子系统具备更强的可靠性和适应性。目前主流的嵌入式微处理器都为多核处理器且大部分都移动化、便携化，随之而来的功耗问题成为目前移动设备最需关注的问题之一。若为自适应能量采集系统的能量输入，在输入能量支持最小系统工作的基础上，如何通过实时检测控制以实现系统的最大资源利用成为能量自适应系统的主要关注点之一。本文首先概述了能量自适应技术的发展，能量自适应片上系统和多核处理器的发展和现状，以及研究能量自适应下多核处理器技术的意义。然后介绍了流计算模型的基本原理，及面向流计算在多核处理器中的重要地位。流计算模型通过并行计算若干个数据通道，使得应用的多个数据可以同时进行交换。流应用具有计算密集性、并行性和局域性等特征，使得流体系结构能较好地解决现有高性能处理器体系结构所面临的难题。论文重点研究了流处理器的流体系结构。介绍了面向流计算的多核处理器的系统架构和具体结构，包括RISC-CPU处理器内核，控制和计算子系统结构、存储层次结构及对外接口。研究了能量自适应下基于动态可变能量的任务调度，通过任务调度控制信息完成对多核的任务重新分配及频率调节。并在此基础上，针对流计算的特性，通过基于FFT流数据计算的多核任务调度控制设计验证面向流计算的能量自适应多核处理器的正确性。文章设计了多核处理器在Altera公司提供的DE3FPGA平台上的实现，并在相应的验证环境及参数下进行验证。完成在动态能量下对多核处理器的任务和频率的可调可控功能，通过在FPGA上实现分析所得结果提出改进方案，以此实现有限能量下的多核系统最大性能。
[Abstract]:In recent years, energy acquisition and energy transmission have brought new opportunities and challenges to integrated circuits.In order to construct a new energy consumption and supply model for electronic systems, the concept of energy adaptation is put forward, and the architecture of energy adaptive processing system and its power consumption model are studied.For example, the research based on on-chip network can make the electronic system have the ability of energy adaptation, which can adapt to different energy supply modes and energy states, which can make the electronic system more reliable and adaptable.At present, the mainstream embedded microprocessors are multi-core processors and most of them are mobile and portable. The following power consumption problem has become one of the most important issues for mobile devices.For the energy input of adaptive energy acquisition system, on the basis of input energy supporting the minimum system work, how to realize the maximum resource utilization of the system through real-time detection control becomes one of the main concerns of the energy adaptive system.In this paper, the development of energy adaptive technology, the development and status of energy adaptive on-chip system and multi-core processor, and the significance of studying multi-core processor technology under energy adaptation are summarized in this paper.Then the basic principle of stream computing model and the importance of flow oriented computing in multi-core processors are introduced.The stream computing model computes several data channels in parallel, which enables multiple data to be exchanged at the same time.Stream applications are characterized by computational denseness, parallelism and locality, which enable the flow architecture to solve the problems faced by the existing high-performance processor architectures.This paper focuses on the stream architecture of stream processor.This paper introduces the architecture and structure of multi-core processor for stream computing, including RISC-CPU processor kernel, control and computing subsystem architecture, storage hierarchy and external interface.The task scheduling based on dynamic variable energy is studied in this paper. The multi-core task redistribution and frequency adjustment are accomplished by the task scheduling control information.On this basis, according to the characteristics of flow computation, the validity of the energy adaptive multi-core processor for flow computing is verified by the design of multi-core task scheduling control based on FFT stream data computing.This paper designs the implementation of multi-core processor on the DE3FPGA platform provided by Altera, and verifies it under the corresponding verification environment and parameters.The task and frequency of the multi-core processor can be adjusted and controlled under the dynamic energy. By implementing the analysis results on FPGA, an improved scheme is proposed to achieve the maximum performance of the multi-core system with limited energy.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP332

【参考文献】