基于以太网的多FPGA矩阵乘法并行计算系统设计(英文)

被引:5
作者
田翔 [1 ]
周凡 [1 ]
陈耀武 [1 ]
刘莉 [2 ]
陈耀 [2 ]
机构
[1] 浙江大学数字技术及仪器研究所
[2] 通用电气全球研发中心
关键词
矩阵乘法; 多FPGA; 以太网; 并行计算; 稀疏矩阵;
D O I
10.19650/j.cnki.cjsi.2007.08.005
中图分类号
TN791 [];
学科分类号
摘要
在过程控制、图像处理等应用领域中需要用到大量的矩阵乘法操作,并且矩阵乘法的计算性能是系统性能的关键因素。本文设计了一个基于以太网的双精度浮点矩阵乘法并行计算系统,并在Xilinx XUP Virtex-II Pro开发平台上进行了原型验证。系统中主机负责将计算任务分配及将计算数据发送给计算单元。当多个计算单元需要相同的数据进行计算时,主机采用广播方式将数据发送所有单元,有效降低了系统的通信开销。计算单元中采用的矩阵乘法器针对稀疏矩阵乘法进行了优化,能够避免零元素块参与计算而提高系统性能。通过理论分析和实验验证,该系统达到了较高的计算性能。
引用
收藏
页码:1373 / 1377
页数:5
相关论文
共 8 条
  • [1] An FPGA based co-pro-cessor design for embedded real-time model predictive control. TIAN X,,LIU H,CHEN Y W. Control and Instruments in Chemical Indus-try . 2007
  • [2] Introduction to Parallel Computing. Grama A,,Gupta A,Karypis K,Kumar V. . 2003
  • [3] H-SIMD machine:config-urable parallel computing for matrix multiplication. XU XI-ZHEN,ZIAVRAS S G. Proceedings of the2005International Conference on Com-puter Design . 2005
  • [4] Resource and delay effi-cient matrix multiplication using newer FPGA devices. CAMPBELL S J,KHATRI S P. Proceedings of the16th ACM Great Lakes Symposi-um on VLSI . 2006
  • [5] 64-bit floating-point FPGA matrix multiplication. DOU Y,VASSILIADIS S,KUZMANOV G K,et al. Pro-ceedings of the International Symposium on Field Program-mable Gate Arrays . 2005
  • [6] Scalable and modular algo-rithms for floating-point matrix multiplication on FPGAs. ZHUO L,PRASANNA V K. Proceedings of the18th International Parallel and Distributed Processing Symposium(IPDPS′04) . 2004
  • [7] Closing the gap:CPU and FPGA trends in sustainable floating-point BLAS per-formance. UNDERWOOD K,HEMMERTK. Proceedings of the12th Annual IEEE Sym-posium on Field-Programmable Custom Computing Ma-chines(FCCM′04) . 2004
  • [8] FPGAs vs CPUs:trends in peak floating-point performance. UNDERWOOD K. Proceedings of the Inter-national Symposium on Field Programmable Gate Arrays . 2004