面向应用领域的处理器软硬件协同设计关键技术研究

The increasingly diversified computer applications have brought the growing demand of performance and the power consumption constraints. How to meet the applications' demand for performance and power consumption limit under a variety of constraints, and shorten the product design cycle to fit the rapid changes in the market, is one of the most important problems in modern processor design. Hardware/software co-design from high-level language programs can efficiently improve the performance and energy of specific applications while maintaining the flexible programming model of general-purpose processors, bridging the gap between the design methodology development and the semiconductor manufacturing technology development, and thus is being paid more and more attention by both academics and industry, which becoming one of the research focuses in the recent years. This research work will explore the practical problems in application-specific hardware/software co-design technology, bring compiler optimization technologies into the automatic processor design workflow. We will build a unified compilation support platform for hardware/software co-design, and study the hardware/software co-design optimization technologies for application-specific processors based on the platform. This research work will reduce the design and development cost of application-specific computer systems, improve design and deveopment efficiency. It will contribute to accelerate the application of self-innovated CPUs and relevant supporting system softwares.

日益丰富和多样化的应用带来了不断增长的对计算能力的需求和功耗的约束，如何在多种约束条件下满足应用的性能、功耗等需求，同时缩短产品的设计周期，以适应市场的快速变化，是现代处理器设计的重要问题之一。从高级语言源程序开始的处理器协同设计技术可以在有效改善特定应用的性能和功耗的情况下保持通用处理器灵活的编程模式，拉近设计方法学的发展与生产工艺发展之间的差距，因此日渐得到学术界和工业界的重视，是近年来的研究热点之一。 本基金项目申请的研究工作将深入探索面向应用领域的处理器协同设计技术在实际应用中存在的不足，有针对性地将编译优化技术引入自动设计流程，建立和完善面向处理器软硬件协同设计的统一的编译支撑平台，并基于该平台开展面向特定应用的协同设计优化技术研究。这对于降低计算机系统的设计开发成本，改善设计效果，提高研发效率，具有重要理论意义；对加快国产自主CPU及配套系统软件的推广应用也具有实际推动作用。

日益丰富和多样化的应用带来了不断增长的对计算能力的需求和功耗的约束，如何在多种约束条件下满足应用的性能、功耗等需求，同时缩短产品的设计周期，是现代处理器设计的重要问题之一。从高级语言开始的处理器协同设计技术可以在有效改善特定应用的性能和功耗的情况下保持通用处理器灵活的编程模式，拉近设计方法学的发展与生产工艺发展之间的差距，日渐得到学术界和工业界的重视，是近年来的研究热点。.本项目立足于当前面向特定应用的处理器软硬件协同设计的相关研究，基于北大众志自主CPU架构和软硬件系统平台，面向虚拟机、嵌入式多媒体运算和系统安全等特定应用领域开展处理器的软硬件协同设计中的编译优化问题研究。项目探索了面向应用领域的处理器协同设计技术在实际应用中存在的不足，有针对性地将编译优化技术引入设计流程，建立和完善了面向自主处理器软硬件协同设计的统一的编译支撑平台；并基于该平台开展了面向特定应用的协同设计优化技术研究。在虚拟机设计和优化、嵌入式多媒体运算优化和系统安全改进策略方面形成了一些成果。这对于降低计算机系统的设计开发成本，改善设计效果，提高研发效率，具有理论意义；对加快国产自主CPU及配套系统软件的推广应用也具有实际推动作用。