视觉阵列处理器超越函数加速单元设计

doi:10.3969/j.issn.1001-2400.2018.04.029

西安电子科技大学学报

视觉阵列处理器超越函数加速单元设计

山蕊1;李涛1;蒋林1;邓军勇1;杨博文1;沈绪榜2

(1. 西安邮电大学电子工程学院，陕西西安 710121；
2. 西安电子科技大学微电子学院，陕西西安 710071)

收稿日期:2017-06-20 发布日期:2018-09-25
作者简介:山蕊(1986-)，女，西安邮电大学博士研究生，E-mail: shanrui0112@163.com
基金资助:
国家自然科学基金资助项目(61772417，61634004，61602377)；陕西省科技统筹计划资助项目(2016KTZDGY02-04-02)；陕西省教育厅专项科研计划资助项目(17JK0689)

Design of the transcendental function computing unit of the computer vision array processor

SHAN Rui1;LI Tao1;JIANG Lin1;DENG Junyong1;YANG Bowen1;SHEN Xubang2

(1. School of Electronic Engineering, Xian Univ. of Posts and Telecommunication, Xian 710121, China;
2. School of Microelectronics, Xidian Univ., Xian 710071, China)

Received:2017-06-20 Published:2018-09-25

摘要/Abstract

摘要：

由于在计算机视觉算法中涉及大量超越函数的运算，为了加快算法在视觉阵列处理器上的运行速度，满足视觉应用的实时性需求，提出了一种调整截距分段线性逼近算法，并基于此设计了一种定点运算系统和对数运算系统混合的超越函数硬件流水线结构．新算法实现了正余弦函数的近似计算，并通过对数/指数计算转换、采用对数运算系统实现了反正切、开平方根、指数运算、对数运算等超越函数的计算．通过数百个随机数据的仿真验证，表明该结构可有效地降低计算误差，提高吞吐率．基于SMIC 0.13μm CMOS集成电路工艺对混合计算系统进行了综合，电路工作频率达200MHz．

关键词: 视觉阵列处理器, 超越函数, 定点数运算, 对数运算系统, 流水线

Abstract:

Computer vision algorithms involve a large number of transcendental function calculations. For accelerating the processing speed of the computer vision algorithm on a computer vision array processor and meeting the real time requirement of vision application, an algorithm of piece-wise linear approximation based on adjusting intercept is proposed, and a transcendental function calculator pipeline structure is designed based on the proposed algorithm, which employs a mixture operation system of the fixed-point number and the logarithmic number system. The function of cos/sin is computed approximately. Meanwhile, based on the logarithm/exponent convertor, the other functions are computed in the logarithmic number system. Simulation is done based on hundreds of random data and the architecture is synthesized using SMIC 0.13μm technology. The proposed structure improves the system throughput and reduces the computational error with the clock frequency reaching 200MHz.

Key words: computer vision array processor, transcendental function, fixed-point number system, logarithmic number system, pipeline

山蕊;李涛;蒋林;邓军勇;杨博文;沈绪榜. 视觉阵列处理器超越函数加速单元设计[J]. 西安电子科技大学学报, 2018, 45(4): 166-173.

SHAN Rui;LI Tao;JIANG Lin;DENG Junyong;YANG Bowen;SHEN Xubang. Design of the transcendental function computing unit of the computer vision array processor[J]. Journal of Xidian University, 2018, 45(4): 166-173.

参考文献

［1］ ANIRUDH B K, VENKATRAMAN V, KUMAR A R, et al. Accelerating Real-time Computer Vision Applications Using HW/SW Co-design［C］//Proceedings of the 2017 International Conference on Computer, Communications and Electronics. Piscataway: IEEE, 2017: 458-463.
［2］ LEE Y, JANG C, KIM H. Accelerating a Computer Vision Algorithm on a Mobile SoC Using CPU-GPU Co-processing—a Case Study on Face Detection［C］//Proceedings of the 2016 International Conference on Mobile Software Engineering and Systems. Piscataway: IEEE, 2016: 70-76.
［3］ TAGLIAVINI G, HAUGOU G, MARONGIU A, et al. ADRENALINE: an OpenVX Environment to Optimize Embedded Vision Applications on Many-core Accelerators［C］//Proceedings of the 2015 IEEE 9th International Symposium on Embedded Multicore/Many SoCs. Piscataway: IEEE, 2015: 289-296
［4］陈云霁. 从人工智能到神经网络处理器［J］. 领导科学论坛, 2017(2): 38-55.
CHEN Yunji. From artificial Intelligence to Neural Network Processors［J］. The Forum of Leadership Science, 2017(2): 38-55.
［5］ HAMEED R, QADEER W, WACHS M, et al. Understanding Sources of Ineffciency in General-purpose Chips［J］. Communications of the ACM, 2011, 54(10): 85-93.
［6］ CHANG C H, CHEN S H, CHEN B W, et al. Fixed-point Computing Element Design for Transcendental Functions and Primary Operations in Speech Processing［J］. IEEE Transactions on Very Large Scale Integration Systems, 2016, 24(5): 1993-1997.
［7］ GUPTA S D, THORNTON M A. A Fixed-point Squaring Algorithm Using an Implicit Arbitrary Radix Number System［J］. IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 2016, 6(1): 34-43.
［8］ CHANG C H, CHEN S H, CHEN B W, et al. A Division-free Algorithm for Fixed-point Power Exponential Function in Embedded System［C］//Proceedings of the 2013 1st International Conference on Orange Technologies. Washington: IEEE Computer Society, 2013: 223-226.
［9］ RENARDY A P, AHMADI N, FADILA A A, et al. FPGA Implementation of CORDIC Algorithms for Sine and Cosine Generator［C］//Proceedings of the 2015 5th International Conference on Electrical Engineering and Informatics. Piscataway: IEEE, 2015: 1-6.
［10］ SINGH A K, SINGH M K, RAY K C. Design and Implementation of Quadruple Floating-point CORDIC［C］//Proceedings of the 2015 IEEE International Symposium on Nanoelectronic and Information Systems. Piscataway: IEEE, 2015: 286-290.
［11］ KAPGATE P S, GUGULOTHU S P. Design and Implementation of Complex Arithmetic Operations Using Binary Logarithmic Number System［C］//Proceedings of the 2015 International Conference on Communications and Signal Processing. Piscataway: IEEE, 2015: 1374-1377.
［12］ NAM B G, KIM H, YOO H J. Power and Area-efficient Unified Computation of Vector and Elementary Functions for Handheld 3D Graphics Systems［J］. IEEE Transactions on Computers, 2008, 57(4): 490-504.
［13］ NAM B G, YOO H J. An Embedded Stream Processor Core Based on Logarithmic Arithmetic for a Low-power 3-D Graphics SoC［J］. IEEE Journal of Solid-State Circuits, 2009, 44(5): 1554-1570.

[1]	陈珍海, 魏敬和, 于宗光, 苏小波, 薛颜, 张鸿. 高精度电荷域ADC共模电荷误差前台校准电路[J]. 西安电子科技大学学报, 2018, 45(6): 137-143+149.
[2]	陈珍海;魏敬和;苏小波;邹家轩;张鸿;于宗光. 低功耗时间交织12位500MS/s电荷域ADC[J]. 西安电子科技大学学报, 2017, 44(6): 109-115+137.
[3]	陈珍海;于宗光;李现坤;魏敬和;黄嵩人;苏小波. 用于16bit 100MS/s ADC的高精度参考电压产生电路[J]. 西安电子科技大学学报, 2017, 44(3): 127-132+180.
[4]	黄嵩人;陈珍海;张鸿;李雪;钱宏文;于宗光. 用于电荷域流水线ADC的1.5位子级电路[J]. 西安电子科技大学学报, 2016, 43(6): 170-175.
[5]	王晓飞;郝跃. 一种双采样1.2V 7位125MS/s流水线ADC的设计[J]. 西安电子科技大学学报, 2016, 43(4): 23-28.
[6]	董嗣万;朱樟明;刘敏杰;杨银堂. 用于8位80MS/s模数转换器的增益数模单元电路[J]. J4, 2016, 43(1): 162-166+172.
[7]	张茂森;邱智亮;高雅;黎军. 星上Clos交换网络的逆向调度算法[J]. J4, 2013, 40(4): 96-101.
[8]	张静波1;戴显英2;张鹤鸣2;胡辉勇1;贾大中1. 一个面积优化的高速RS(255，239)译码器VLSI设计 [J]. J4, 2008, 35(1): 116-120.
[9]	赵波;杜建超;颜尧平. 一种易于硬件实现的运动估计算法及其VLSI实现[J]. J4, 2003, 30(2): 160-164.
[10]	梁政;杨银堂;沈绪榜. LSC87嵌入式协处理器中超越函数的实现方法[J]. J4, 2001, 28(2): 258-261.

视觉阵列处理器超越函数加速单元设计

Design of the transcendental function computing unit of the computer vision array processor

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 10

Metrics

本文评价

推荐阅读 10