【作者】 翟文正；

【导师】 胡越黎；

【作者基本信息】 上海大学 ， 控制理论与控制工程， 2013， 博士

【摘要】 伴随集成电路工艺的不断进步以及处理器体系的发展，异构多核处理器在嵌入式领域广泛应用并成为设计的主流体系结构，与单核平台上软件开发不同，应用任务需要合理划分和并行性调度到恰当的核上执行，实现资源的最佳化配置。异构多核软件面临巨大挑战，迫切需要相应开发方法的指导和开发工具的支持。传统的以编程为中心的方法难以满足市场对开发效率和质量的要求，基于模型的开发体现了嵌入式领域的现代设计思想和方法，是当前在异构环境中设计复杂软件的有效途径，对其深入研究具有重要的理论和实际意义。论文针对领域应用的异构多核处理器，紧密围绕基于模型的嵌入式系统开发这一主题进行各阶段的探索、研究和开发，同时完成必要的创新性工作。所开展的工作和取得的成果主要体现在以下四个方面：1提出基于模型的异构多核嵌入式系统开发方法。结合异构多核特性和嵌入式软件开发模式的特点，提出从任务划分、目标平台与应用建模、任务映射、模型编译到最终代码生成的开发方法。克服了多核并行开发的复杂性，提高了软件开发效率和软件可靠性。2构建基于模型的异构多核开发环境。设计实现了模型化、可视化集成开发环境MV12-ModelIDE，其中模型编辑器MV_Modeling实现应用需求基于嵌入式实时操作系统的抽象建模和架构建模描述，任务分配器MV-Mapping实现DAG任务模型到异构多核平台的最优分配和调度仿真，模型编译器MV-Compilation则根据既定转换规则实现模型到代码的自动生成，汇编器MV-Assembler针对MV10从核实现了包括定制指令的汇编源程序准确和快速编译。3提出一种求解异构多核DAG任务分配与调度问题的改进离散粒子群优化算法。以任务高度和粒子的运动位置作为优先级构造调度队列，采用基于插入的原则选择调度列表中首个任务具有最小累计EFT的处理器核作为任务分配对象，使任务集有序并行运行达到最小完成时间和最大加速比，充分发挥异构多核高性能特性。4将上述方法和技术在MV12异构多核处理器及汽车车身控制网络实验平台实施和验证后，结果证明所研究的方法与技术正确有效。课题研究所取得的成果可推广应用于其它嵌入式多核处理器上。更多还原

【Abstract】 Along with the continuous advancement in the semiconductor manufacturingtechnology and processor architecture, a heterogeneous multi-core processor hasbeen widely used in the embedded system and becomes the mainstream. Unlike theconventional single-core software development, the embedded application isrequired to be divided and scheduled reasonably on the appropriate core to achievethe optimal performance. Software development on heterogeneous multi-core isfacing great challenges, requiring the appropriate development methods and toolsurgently.The traditional embedded software development method which is programmingcentered is hard to meet the market demand for development efficiency and quality.The model-based development presents modern software design idea and methodand is an effective solution to the complex software development in heterogeneousenvironment. The in-depth study will have great theoretical and practicalsignificance.Addressing the domain-specific heterogeneous multi-core processor, the thesisfocuses on the model-based embedded software design method and relevanttechnologies. The research results and main contributions of this dissertation are asfollows:First, it proposed a model-based embedded software development method forheterogeneous multi-core system by combining the MDA (Model DrivenArchitecture) techniques with heterogeneous multi-core properties and embeddedsoftware programming framework. It covered the entire development process ofembedded software, such as task partitioning, architecture and application modeling,task mapping, model compilation and code generation. It overcomes the complexityof multi-core parallel development and increases the efficiency of the softwaredevelopment.Second, it built a model-based integrated development environmentMV12-ModelIDE for the embedded heterogeneous multi-core system. The modeleditor MV-Modeling was used to implement the application requirements andarchitecture modeling. The task distributer MV-Mapping maps DAG (DirectedAcyclic Graph) upon processors and the model compiler MV-Compiliation implements code generation automatically according to the model transformationrules. The assembler MV-Assembler for MV10processor implements accurate andfast instruction compilation.Third, it proposed a novel DAG task scheduling algorithm named IDPSO(Improved Discrete Particle Swarm Optimization) for the heterogeneous multi-coreprocessor. It takes the task height and the particle position as the task priority tobuild scheduling list, and applies the insertion technique to obtain a feasible schedulethat a processor with minimum cumulative EFT (Earliest Finish Time) serves as theobject of the first task assignment, so that the task precedence relationships aresatisfied, the makespan is minimum and the speedup is maximum.Finally, it applied the above proposed methods and techniques to MV12andvehicle body control network experimental platform, the results of which illustratedthe feasibility and validity of the research in the thesis.The proposed methods and tools developed in the thesis are also suitable forother embedded multi-core platform.更多还原