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 mesoscale ensemble prediction system based on the Tropical Regional Atmosphere Model for the South China Sea(CMA-TRAMS(EPS))has been pre-operational since April 2020 at South China Regional Meteorological Center(S...The mesoscale ensemble prediction system based on the Tropical Regional Atmosphere Model for the South China Sea(CMA-TRAMS(EPS))has been pre-operational since April 2020 at South China Regional Meteorological Center(SCRMC),which was developed by the Guangzhou Institute of Tropical and Marine Meteorology(GITMM).To better understand the performance of the CMA-TRAMS(EPS)and provide guidance to forecasters,we assess the performance of this system on both deterministic and probabilistic forecasts from April to September 2020 in this study through objective verification.Compared with the control(deterministic)forecasts,the ensemble mean of the CMATRAMS(EPS)shows advantages in most non-precipitation variables.In addition,the threat score indicates that the CMA-TRAMS(EPS)obviously improves light and heavy rainfall forecasts in terms of the probability-matched mean.Compared with the European Center for Medium-range Weather Forecasts operational ensemble prediction system(ECMWF-EPS),the CMA-TRAMS(EPS)improves the probabilistic forecasts of light rainfall in terms of accuracy,reliability and discrimination,and this system also improves the heavy rainfall forecasts in terms of discrimination.Moreover,two typical heavy rainfall cases in south China during the pre-summer rainy season are investigated to visually demonstrate the deterministic and probabilistic forecasts,and the results of these two cases indicate the differences and advantages(deficiencies)of the two ensemble systems.展开更多
针对集合预报存在的偏差和集合离散度通常偏小的问题,在卡尔曼滤波递减平均的一阶矩和二阶矩偏差订正方案的基础上发展了综合偏差订正方案,并利用B08RDP WWRP(The WWRP Beijing 2008 Olympics Research and Development Project)项目中...针对集合预报存在的偏差和集合离散度通常偏小的问题,在卡尔曼滤波递减平均的一阶矩和二阶矩偏差订正方案的基础上发展了综合偏差订正方案,并利用B08RDP WWRP(The WWRP Beijing 2008 Olympics Research and Development Project)项目中日本气象厅(JMA)区域集合预报的850 hPa温度资料,将敏感性试验得到的一阶矩和二阶矩订正的最优权重系数应用于综合偏差订正方案,并对其订正效果进行多方面检验分析。试验结果表明,一阶矩订正可以有效减小集合平均偏差,集合平均预报质量得到了明显改善;二阶矩订正对集合离散度具有较强的调整能力,订正后的集合预报可靠性、区分不同天气事件的能力总体上得到了提高;综合偏差订正方案有效融合了一阶矩和二阶矩订正的优势,其各自的最优权重系数适用于综合偏差订正方案,对集合平均偏差和离散度具有良好的订正效果,能够改善集合预报的整体质量。但一阶矩与二阶矩订正对综合偏差订正的贡献程度随评分指标而异,一阶矩订正对等级概率(RPS)评分和异常值百分比评分的贡献分别为83.75%和18.83%,可信度的改善约83.98%源于二阶矩订正,而相对作用特征(ROC)评分中二者的贡献基本相当。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。
基金National Key Research and Development Project(2019YFEO110100)National Natural Science Foundation of China(41975136)+5 种基金the Intelligent Gridded Forecasting Team of Guangdong Meteorological Bureau(GRMCTD202004)Guangdong Basic and Applied Basic Research Foundation(2019A1515011118)Science and Technology Planning Project of Guangzhou(202103000030)the Innovation and Development Project of the China Meteorological Administration(CXF2021Z009)the Science and Technology Research Project of Guangdong Meteorological Bureau(GMRC2020M06)the Open Fund of Guangdong Provincial Key Laboratory of Regional Numerical Weather Prediction(J202006)。
文摘The mesoscale ensemble prediction system based on the Tropical Regional Atmosphere Model for the South China Sea(CMA-TRAMS(EPS))has been pre-operational since April 2020 at South China Regional Meteorological Center(SCRMC),which was developed by the Guangzhou Institute of Tropical and Marine Meteorology(GITMM).To better understand the performance of the CMA-TRAMS(EPS)and provide guidance to forecasters,we assess the performance of this system on both deterministic and probabilistic forecasts from April to September 2020 in this study through objective verification.Compared with the control(deterministic)forecasts,the ensemble mean of the CMATRAMS(EPS)shows advantages in most non-precipitation variables.In addition,the threat score indicates that the CMA-TRAMS(EPS)obviously improves light and heavy rainfall forecasts in terms of the probability-matched mean.Compared with the European Center for Medium-range Weather Forecasts operational ensemble prediction system(ECMWF-EPS),the CMA-TRAMS(EPS)improves the probabilistic forecasts of light rainfall in terms of accuracy,reliability and discrimination,and this system also improves the heavy rainfall forecasts in terms of discrimination.Moreover,two typical heavy rainfall cases in south China during the pre-summer rainy season are investigated to visually demonstrate the deterministic and probabilistic forecasts,and the results of these two cases indicate the differences and advantages(deficiencies)of the two ensemble systems.
文摘针对集合预报存在的偏差和集合离散度通常偏小的问题,在卡尔曼滤波递减平均的一阶矩和二阶矩偏差订正方案的基础上发展了综合偏差订正方案,并利用B08RDP WWRP(The WWRP Beijing 2008 Olympics Research and Development Project)项目中日本气象厅(JMA)区域集合预报的850 hPa温度资料,将敏感性试验得到的一阶矩和二阶矩订正的最优权重系数应用于综合偏差订正方案,并对其订正效果进行多方面检验分析。试验结果表明,一阶矩订正可以有效减小集合平均偏差,集合平均预报质量得到了明显改善;二阶矩订正对集合离散度具有较强的调整能力,订正后的集合预报可靠性、区分不同天气事件的能力总体上得到了提高;综合偏差订正方案有效融合了一阶矩和二阶矩订正的优势,其各自的最优权重系数适用于综合偏差订正方案,对集合平均偏差和离散度具有良好的订正效果,能够改善集合预报的整体质量。但一阶矩与二阶矩订正对综合偏差订正的贡献程度随评分指标而异,一阶矩订正对等级概率(RPS)评分和异常值百分比评分的贡献分别为83.75%和18.83%,可信度的改善约83.98%源于二阶矩订正,而相对作用特征(ROC)评分中二者的贡献基本相当。
