针对国防科技大学自主研发的异构多核数字信号处理(digital signal processing, DSP)芯片的特征以及卷积算法自身特点,提出了一种面向多核DSP架构的高性能多核并行卷积实现方案。针对1×1卷积提出了特征图级多核并行方案;针对卷积...针对国防科技大学自主研发的异构多核数字信号处理(digital signal processing, DSP)芯片的特征以及卷积算法自身特点,提出了一种面向多核DSP架构的高性能多核并行卷积实现方案。针对1×1卷积提出了特征图级多核并行方案;针对卷积核大于1的卷积提出了窗口级多核并行优化设计,同时提出了逐元素向量化计算的核内并行优化实现。实验结果表明,所提并行优化方法实现单核计算效率最高能达到64.95%,在带宽受限情况下,多核并行扩展效率可达到48.36%~88.52%,在典型网络ResNet50上的执行性能与E5-2640 CPU相比,获得了5.39倍性能加速。展开更多
矩阵乘卷积算法能够为各种卷积配置提供高性能基础实现,是面向给定芯片进行卷积性能优化的首要选择。针对国防科技大学自主研制的飞腾异构多核数字信号处理器(digital signal processor,DSP)芯片的特征以及矩阵乘卷积算法自身的特点,提...矩阵乘卷积算法能够为各种卷积配置提供高性能基础实现,是面向给定芯片进行卷积性能优化的首要选择。针对国防科技大学自主研制的飞腾异构多核数字信号处理器(digital signal processor,DSP)芯片的特征以及矩阵乘卷积算法自身的特点,提出了一种面向多核DSP架构的高性能并行矩阵乘卷积实现算法ftmEConv。该算法由输入特征图转换、卷积核转换、矩阵乘以及输出特征图转换这四个均运行在通用多核DSP上的并行化部分构成,通过有效挖掘通用DSP核中功能单元的潜力来提升各个部分的性能。实验结果表明,ftmEConv实现了高达42.90%的计算效率,与芯片上的其他矩阵乘卷积算法实现相比,获得了高达7.79倍的性能加速。展开更多
矩阵转置是矩阵运算的基本操作,广泛应用于信号处理、科学计算以及深度学习等各种领域。随着国防科技大学自主研制的飞腾异构多核数字信号处理器(digital signal processor, DSP)在各种领域中的推广应用,对高性能矩阵转置实现提出了强...矩阵转置是矩阵运算的基本操作,广泛应用于信号处理、科学计算以及深度学习等各种领域。随着国防科技大学自主研制的飞腾异构多核数字信号处理器(digital signal processor, DSP)在各种领域中的推广应用,对高性能矩阵转置实现提出了强烈需求。针对飞腾异构多核DSP的体系结构特征与矩阵转置操作的特点,提出了一种适配不同数据位宽(8 B、4 B以及2 B)矩阵的并行矩阵转置算法ftmMT。该算法基于DSP中向量处理单元的Load/Store部件实现了向量化,同时基于矩阵分块实现了多个DSP核的并行处理,通过隐式乒乓设计实现了片上向量化转置与片外访存的重叠以及访存性能的大幅提升。实验结果表明,ftmMT能够显著加快矩阵转置操作,与CPU上的开源转置库HPTT相比,可获得高达8.99倍的性能加速。展开更多
The Cs_(2)NaInCl_(6) double perovskite is one of the most promising lead-free perovskites due to its exceptional stability and straightforward synthesis.However,it faces challenges related to inefficient photoluminesc...The Cs_(2)NaInCl_(6) double perovskite is one of the most promising lead-free perovskites due to its exceptional stability and straightforward synthesis.However,it faces challenges related to inefficient photoluminescence.Doping and high pressure are employed to tailor the optical properties of Cs_(2)NaInCl_(6).Herein,Sb^(3+)doped Cs_(2)NaInCl_(6)(Sb^(3+):Cs_(2)NaInCl_(6)) was synthesized and it exhibits blue emission with a photoluminescence quantum yield of up to 37.3%.Further,by employing pressure tuning,a blue stable emission under a very wide range from 2.7 GPa to 9.8 GPa is realized in Sb^(3+):Cs_(2)NaInCl_(6).Subsequently,the emission intensity of Sb^(3+):Cs_(2)NaInCl_(6) experiences a significant increase(3.3 times)at 19.0 GPa.It is revealed that the pressure-induced distinct emissions can be attributed to the carrier self-trapping and detrapping between Cs_(2)NaInCl_(6) and Sb^(3+).Notably,the lattice compression in the cubic phase inevitably modifies the band gap of Sb^(3+):Cs_(2)NaInCl_(6).Our findings provide valuable insights into effects of the high pressure in further boosting unique emission characteristics but also offer promising opportunities for development of doped double perovskites with enhanced optical functionalities.展开更多
文摘针对国防科技大学自主研发的异构多核数字信号处理(digital signal processing, DSP)芯片的特征以及卷积算法自身特点,提出了一种面向多核DSP架构的高性能多核并行卷积实现方案。针对1×1卷积提出了特征图级多核并行方案;针对卷积核大于1的卷积提出了窗口级多核并行优化设计,同时提出了逐元素向量化计算的核内并行优化实现。实验结果表明,所提并行优化方法实现单核计算效率最高能达到64.95%,在带宽受限情况下,多核并行扩展效率可达到48.36%~88.52%,在典型网络ResNet50上的执行性能与E5-2640 CPU相比,获得了5.39倍性能加速。
文摘矩阵乘卷积算法能够为各种卷积配置提供高性能基础实现,是面向给定芯片进行卷积性能优化的首要选择。针对国防科技大学自主研制的飞腾异构多核数字信号处理器(digital signal processor,DSP)芯片的特征以及矩阵乘卷积算法自身的特点,提出了一种面向多核DSP架构的高性能并行矩阵乘卷积实现算法ftmEConv。该算法由输入特征图转换、卷积核转换、矩阵乘以及输出特征图转换这四个均运行在通用多核DSP上的并行化部分构成,通过有效挖掘通用DSP核中功能单元的潜力来提升各个部分的性能。实验结果表明,ftmEConv实现了高达42.90%的计算效率,与芯片上的其他矩阵乘卷积算法实现相比,获得了高达7.79倍的性能加速。
基金supported by the National Key Research and Development Program of China(Grant Nos.2021YFA1400200 and2021YFA0718701)the National Natural Science Foundation of China(Grant Nos.U2032127,11904322,12104411,12174347)+4 种基金the Natural Science Foundation of Henan province of China(Grant No.202300410356)the China Postdoctoral Science Foundation(Grant Nos.2019M652560 and 2020M682326)the CAS Interdisciplinary Innovation Team(Grant No.JCTD-2019-01)the Postdoctoral Research Grant in Henan Province(Grant No.1902013)the Science Foundation for Highlevel Talents of Wuyi University(Grant No.2021AL019)。
文摘The Cs_(2)NaInCl_(6) double perovskite is one of the most promising lead-free perovskites due to its exceptional stability and straightforward synthesis.However,it faces challenges related to inefficient photoluminescence.Doping and high pressure are employed to tailor the optical properties of Cs_(2)NaInCl_(6).Herein,Sb^(3+)doped Cs_(2)NaInCl_(6)(Sb^(3+):Cs_(2)NaInCl_(6)) was synthesized and it exhibits blue emission with a photoluminescence quantum yield of up to 37.3%.Further,by employing pressure tuning,a blue stable emission under a very wide range from 2.7 GPa to 9.8 GPa is realized in Sb^(3+):Cs_(2)NaInCl_(6).Subsequently,the emission intensity of Sb^(3+):Cs_(2)NaInCl_(6) experiences a significant increase(3.3 times)at 19.0 GPa.It is revealed that the pressure-induced distinct emissions can be attributed to the carrier self-trapping and detrapping between Cs_(2)NaInCl_(6) and Sb^(3+).Notably,the lattice compression in the cubic phase inevitably modifies the band gap of Sb^(3+):Cs_(2)NaInCl_(6).Our findings provide valuable insights into effects of the high pressure in further boosting unique emission characteristics but also offer promising opportunities for development of doped double perovskites with enhanced optical functionalities.
