国防科技大学自主研制的高性能加速器采用中央处理器(CPU)+通用数字信号处理器(GPDSP)的片上异构融合架构,使用超长指令集(VLIW)+单指令多数据流(SIMD)的向量化结构的GPDSP是峰值性能主要支撑的加速核。主流编译器在密集的数据计算指令...国防科技大学自主研制的高性能加速器采用中央处理器(CPU)+通用数字信号处理器(GPDSP)的片上异构融合架构,使用超长指令集(VLIW)+单指令多数据流(SIMD)的向量化结构的GPDSP是峰值性能主要支撑的加速核。主流编译器在密集的数据计算指令排布、为指令静态分配硬件执行单元、GPDSP特有的向量指令等方面不能很好地支持高性能加速器。基于低级虚拟器(LLVM)编译框架,在前寄存器分配调度阶段,结合峰值寄存器压力感知方法(PERP)、蚁群优化(ACO)算法与GPDSP结构特点,优化代价模型,设计支持寄存器压力感知的指令调度模块;在后寄存器分配阶段提出支持静态功能单元分配的指令调度策略,通过冲突检测机制保证功能单元分配的正确性,为指令并行执行提供软件基础;在后端封装一系列丰富且规整的向量指令接口,实现对GPDSP向量指令的支持。实验结果表明,所提出的LLVM编译架构优化方法从功能和性能上实现了对GPDSP的良好支撑,GCC testsuite测试整体性能平均加速比为4.539,SPEC CPU 2017浮点测试整体性能平均加速比为4.49,SPEC CPU 2017整型测试整体性能平均加速比为3.24,使用向量接口的向量程序实现了平均97.1%的性能提升率。展开更多
The fast acceptance of cloud technology to industry explains increasing energy conservation needs and adoption of energy aware scheduling methods to cloud. Power consumption is one of the top of mind issues in cloud, ...The fast acceptance of cloud technology to industry explains increasing energy conservation needs and adoption of energy aware scheduling methods to cloud. Power consumption is one of the top of mind issues in cloud, because the usage of cloud storage by the individuals or organization grows rapidly. Developing an efficient power management processor architecture has gained considerable attention. However, the conventional power management mechanism fails to consider task scheduling policies. Therefore, this work presents a novel energy aware framework for power management. The proposed system leads to the development of Inclusive Power-Cognizant Processor Controller (IPCPC) for efficient power utilization. To evaluate the performance of the proposed method, simulation experiments inputting random tasks as well as tasks collected from Google Trace Logs were conducted to validate the supremacy of IPCPC. The research based on Real world Google Trace Logs gives results that proposed framework leads to less than 9% of total power consumption per task of server which proves reduction in the overall power needed.展开更多
文摘国防科技大学自主研制的高性能加速器采用中央处理器(CPU)+通用数字信号处理器(GPDSP)的片上异构融合架构,使用超长指令集(VLIW)+单指令多数据流(SIMD)的向量化结构的GPDSP是峰值性能主要支撑的加速核。主流编译器在密集的数据计算指令排布、为指令静态分配硬件执行单元、GPDSP特有的向量指令等方面不能很好地支持高性能加速器。基于低级虚拟器(LLVM)编译框架,在前寄存器分配调度阶段,结合峰值寄存器压力感知方法(PERP)、蚁群优化(ACO)算法与GPDSP结构特点,优化代价模型,设计支持寄存器压力感知的指令调度模块;在后寄存器分配阶段提出支持静态功能单元分配的指令调度策略,通过冲突检测机制保证功能单元分配的正确性,为指令并行执行提供软件基础;在后端封装一系列丰富且规整的向量指令接口,实现对GPDSP向量指令的支持。实验结果表明,所提出的LLVM编译架构优化方法从功能和性能上实现了对GPDSP的良好支撑,GCC testsuite测试整体性能平均加速比为4.539,SPEC CPU 2017浮点测试整体性能平均加速比为4.49,SPEC CPU 2017整型测试整体性能平均加速比为3.24,使用向量接口的向量程序实现了平均97.1%的性能提升率。
文摘The fast acceptance of cloud technology to industry explains increasing energy conservation needs and adoption of energy aware scheduling methods to cloud. Power consumption is one of the top of mind issues in cloud, because the usage of cloud storage by the individuals or organization grows rapidly. Developing an efficient power management processor architecture has gained considerable attention. However, the conventional power management mechanism fails to consider task scheduling policies. Therefore, this work presents a novel energy aware framework for power management. The proposed system leads to the development of Inclusive Power-Cognizant Processor Controller (IPCPC) for efficient power utilization. To evaluate the performance of the proposed method, simulation experiments inputting random tasks as well as tasks collected from Google Trace Logs were conducted to validate the supremacy of IPCPC. The research based on Real world Google Trace Logs gives results that proposed framework leads to less than 9% of total power consumption per task of server which proves reduction in the overall power needed.
