张量转置(tensor transposition)作为基础张量运算原语,广泛应用于信号处理、科学计算以及深度学习等各种领域,在张量数据密集型应用及高性能计算中具有重要作用。随着能效指标在高性能计算系统中的重要性日益凸显,基于数字信号处理器(d...张量转置(tensor transposition)作为基础张量运算原语,广泛应用于信号处理、科学计算以及深度学习等各种领域,在张量数据密集型应用及高性能计算中具有重要作用。随着能效指标在高性能计算系统中的重要性日益凸显,基于数字信号处理器(digital signal processors,DSPs)的加速器已被集成至通用计算系统。然而,传统面向多核CPU和GPU的张量转置库因架构差异无法充分适配DSP架构。一方面,DSP架构的向量化计算潜力尚未得到充分挖掘;另一方面,其复杂的片上存储体系与多层次共享内存结构为张量并行程序设计带来了显著挑战。针对国产多核DSP的架构特点,提出ftmTT算法,并设计实现了一个面向多核DSP架构的通用张量转置库。ftmTT算法通过设计适配DSP架构的高效内存访问模式充分挖掘其并行化和向量化潜力,其核心创新包括:1)采用分块策略将高维张量转置转化为多核DSP平台所提供的矩阵转置内核操作;2)提出基于DMA点对点传输的张量数据块访存合并方案来降低数据搬运开销;3)通过双缓冲设计异步重叠转置计算与DMA传输实现计算通信隐藏,最终面向多核DSP实现高性能并行张量转置。在国产多核DSP平台FT-M7032的实验表明,ftmTT张量转置算法取得了最高达理论带宽75.96%的性能,达到FT-M7032平台STREAM带宽99.23%的性能。展开更多
The human brain is a complex intelligent system composed of tens of billions of neurons interconnected through synapses,and its intricate network structure has consistently attracted numerous scientists to explore the...The human brain is a complex intelligent system composed of tens of billions of neurons interconnected through synapses,and its intricate network structure has consistently attracted numerous scientists to explore the mysteries of brain functions.However,most existing studies have only verified the biological mimicry characteristics of memristors at the single neuron-synapse level,and there is still a lack of research on memristors simulating synaptic coupling between neurons in multi-neuron networks.Based on this,this paper uses discrete memristors to couple dual discrete Rulkov neurons,and adds synaptic crosstalk between the two discrete memristors to form a neuronal network.A memristor-coupled dual-neuron map,called the Rulkov-memristor-Rulkov(R-M-R)map,is constructed to simulate synaptic connections between neurons in biological tissues.Then,the equilibrium points of the R-M-R map are studied.Subsequently,the effect of parameter variations on the dynamic performance of the R-M-R map is comprehensively analyzed using bifurcation diagram,phase diagram,Lyapunov exponent spectrum(LEs),firing diagram,and spectral entropy(SE)complexity algorithms.In the RM-R map,diverse categories of periodic,chaotic,and hyperchaotic attractors,as well as different states of firing patterns,can be observed.Additionally,different types of state transitions and coexisting attractors are discovered.Finally,the feasibility of the model in digital circuits is verified using a DSP hardware platform.In this study,the coupling principle of biological neurons is simulated,the chaotic dynamic behavior of the R-M-R map is analyzed,and a foundation is laid for deciphering the complex working mechanisms of the brain.展开更多
Si P微系统是一种高度集成化的系统,其内部可能集成1个或多个DSP、NOR Flash和DDR存储器、AI加速芯片等,有些复杂的微系统还集成了FPGA芯片。由于内部集成了多个微组件,芯片之间相互连接,传统的测试单一微组件的方法并不适用于微系统的...Si P微系统是一种高度集成化的系统,其内部可能集成1个或多个DSP、NOR Flash和DDR存储器、AI加速芯片等,有些复杂的微系统还集成了FPGA芯片。由于内部集成了多个微组件,芯片之间相互连接,传统的测试单一微组件的方法并不适用于微系统的测试。提出了一套DSP微组件测试方法,该系统包括1块专门的测试板、可调试的电脑测试环境和JTAG通信。与单一的DSP裸芯测试相比,它可以快速稳定地实现DSP微组件的性能测试,满足大批量生产测试的需求。展开更多
就电动机变频调速自动控制系统的设计与实现展开研究。首先,给出了电动机变频调速自动控制系统整体框架,并详细阐述了数字信号处理器(Digital Signal Processor,DSP)技术。在软件设计部分,针对DSP技术存在控制复杂度高的问题,引入比例-...就电动机变频调速自动控制系统的设计与实现展开研究。首先,给出了电动机变频调速自动控制系统整体框架,并详细阐述了数字信号处理器(Digital Signal Processor,DSP)技术。在软件设计部分,针对DSP技术存在控制复杂度高的问题,引入比例-积分-微分(Proportional-Integral-Derivative,PID)控制策略简化控制流程,并提出了基于模糊控制的电动机变频调速自动控制方法,以实现更精确的变频调速控制。系统应用测试结果表明,新系统控制误差低至1.2%,波动范围小且故障率低至0.01次/h,使得系统稳定性指数(System Stability Index,SSI)仅为0.95,说明本研究为电动机变频调速领域的自动化和智能化提供了有价值的参考。展开更多
文摘张量转置(tensor transposition)作为基础张量运算原语,广泛应用于信号处理、科学计算以及深度学习等各种领域,在张量数据密集型应用及高性能计算中具有重要作用。随着能效指标在高性能计算系统中的重要性日益凸显,基于数字信号处理器(digital signal processors,DSPs)的加速器已被集成至通用计算系统。然而,传统面向多核CPU和GPU的张量转置库因架构差异无法充分适配DSP架构。一方面,DSP架构的向量化计算潜力尚未得到充分挖掘;另一方面,其复杂的片上存储体系与多层次共享内存结构为张量并行程序设计带来了显著挑战。针对国产多核DSP的架构特点,提出ftmTT算法,并设计实现了一个面向多核DSP架构的通用张量转置库。ftmTT算法通过设计适配DSP架构的高效内存访问模式充分挖掘其并行化和向量化潜力,其核心创新包括:1)采用分块策略将高维张量转置转化为多核DSP平台所提供的矩阵转置内核操作;2)提出基于DMA点对点传输的张量数据块访存合并方案来降低数据搬运开销;3)通过双缓冲设计异步重叠转置计算与DMA传输实现计算通信隐藏,最终面向多核DSP实现高性能并行张量转置。在国产多核DSP平台FT-M7032的实验表明,ftmTT张量转置算法取得了最高达理论带宽75.96%的性能,达到FT-M7032平台STREAM带宽99.23%的性能。
基金supported by the National Natural Science Foundation of China(Grant No.62571079)the Technological Innovation Projects in the Field of Artificial Intelligence in Liaoning Province(Grant No.2023JH26/10300011)+1 种基金the Basic Scientific Research Projects in the Department of Education of Liaoning Province(Grant No.LJ212410152049)the Liaoning Provincial Science and Technology Plan Joint Project(Grant No.2025-BSLH-041)。
文摘The human brain is a complex intelligent system composed of tens of billions of neurons interconnected through synapses,and its intricate network structure has consistently attracted numerous scientists to explore the mysteries of brain functions.However,most existing studies have only verified the biological mimicry characteristics of memristors at the single neuron-synapse level,and there is still a lack of research on memristors simulating synaptic coupling between neurons in multi-neuron networks.Based on this,this paper uses discrete memristors to couple dual discrete Rulkov neurons,and adds synaptic crosstalk between the two discrete memristors to form a neuronal network.A memristor-coupled dual-neuron map,called the Rulkov-memristor-Rulkov(R-M-R)map,is constructed to simulate synaptic connections between neurons in biological tissues.Then,the equilibrium points of the R-M-R map are studied.Subsequently,the effect of parameter variations on the dynamic performance of the R-M-R map is comprehensively analyzed using bifurcation diagram,phase diagram,Lyapunov exponent spectrum(LEs),firing diagram,and spectral entropy(SE)complexity algorithms.In the RM-R map,diverse categories of periodic,chaotic,and hyperchaotic attractors,as well as different states of firing patterns,can be observed.Additionally,different types of state transitions and coexisting attractors are discovered.Finally,the feasibility of the model in digital circuits is verified using a DSP hardware platform.In this study,the coupling principle of biological neurons is simulated,the chaotic dynamic behavior of the R-M-R map is analyzed,and a foundation is laid for deciphering the complex working mechanisms of the brain.
文摘就电动机变频调速自动控制系统的设计与实现展开研究。首先,给出了电动机变频调速自动控制系统整体框架,并详细阐述了数字信号处理器(Digital Signal Processor,DSP)技术。在软件设计部分,针对DSP技术存在控制复杂度高的问题,引入比例-积分-微分(Proportional-Integral-Derivative,PID)控制策略简化控制流程,并提出了基于模糊控制的电动机变频调速自动控制方法,以实现更精确的变频调速控制。系统应用测试结果表明,新系统控制误差低至1.2%,波动范围小且故障率低至0.01次/h,使得系统稳定性指数(System Stability Index,SSI)仅为0.95,说明本研究为电动机变频调速领域的自动化和智能化提供了有价值的参考。
