<span style="font-family: sans-serif;">Optimization of Vector Function Library Based on Domestic PuDianNao Chip</span>

NAVIGATE

Table of Contents

STATISTICS

Viewed1145

Downloads894

Optimization of Vector Function Library Based on Domestic PuDianNao Chip

[HTML] PDF下载 (894)

[1]Yang Zhizheng,Du Zidong,Wen Yuanbo..Optimization of Vector Function Library Based on Domestic PuDianNao Chip[J].Journal of Zhengzhou University (Engineering Science),2023,44(01):31-37.[doi:10.13705/j.issn.1671-6833.2023.01.013]

Copy

Journal of Zhengzhou University (Engineering Science)[ISSN 1671-6833/CN 41-1339/T] Volume: 44卷 Number of periods: 2023 01 Page number: 31-37 Column: Public date: 2022-12-06

Title:: Optimization of Vector Function Library Based on Domestic PuDianNao Chip

Author(s):: Yang Zhizheng¹; Du Zidong²; Wen Yuanbo³.; Zhengzhou University Henan Advanced Technical Research Institute, Henan Zhengzhou 450001, National Key Laboratory of Computing Technology Structure Structure of the Institute of Computing Technology of the Chinese Academy of Sciences, Beijing 100190, Chinese University of Science and Technology of Science and Technology

Keywords:: Direction function; Pudiannao-VECMATH; domestic artificial intelligence processor; exposed branch scope and branch flat

CLC:: TP311

DOI:: 10.13705/j.issn.1671-6833.2023.01.013

Abstract:: At present, the vector math functions on the PuDianNao chip of the domestic artificial intelligence processor can only be implemented by calling scalar functions cyclically, and the performance of this method is relatively low. Based on the PuDianNao chip, three optimization methods were proposed. The first two were interpolation method and SIMD masking method. Thirdly, based on a hardware array structure on PuDianNao, VLIW instructions were used to operate each processing unit in the array, and the SIMT programming model was encapsulated programmatically. The accuracy and performance of the above three methods were tested, and the experimental results showed that the third method had the best accuracy and performance. The third method was used to implement the vector mathematical function library PuDianNao-VecMath based on the domestic PuDianNao chip, which solved the problem that the multi-branch structure of mathematical functions was difficult to vectorize. The function library had good precision performance, stable functions and correct operation. The provided interfaces included rounding functions, transcendental functions, comparison functions, activation functions and other common basic math library functions. In terms of precision, the entire data of the function definition domain interval was used as input, and the operation result was compared with the result of the scalar function running on the CPU i7. The results showed that the maximum ULP value was 2, and the maximum ULP value of the half-precision version was 1. Compared with the use of scalar loop, the performance was greatly improved. Compared with the scalar loop, the single-precision version had an average speed-up ratio of 18. 26 and a maximum speed-up ratio of 35. 90. The halfprecision version had an average speed-up ratio of 15. 65 and a maximum speed-up ratio of 30. 11.

References:: [1] Intel Corporation. Intel ® oneAPI math kernel library [EB / OL] . ( 2019 - 11 - 01) [ 2022 - 02 - 10] . https: / / software. intel. com / en-us/ mkl.
[2] Intel Corporation. Intel short vector math library [ EB / OL] . ( 2021 - 06 - 20) [ 2022 - 02 - 10 ] . https: / / software. intel. com / en-us/ node / 523613.
[3] Advanced Micro Devices, Inc. . AMD core math library [EB / OL] . (2013-07-24) [2022- 02- 10] . http: / / developer. amd. com / tools-and-sdks/ archive / acml-productfeatures/ .
[4] ANAND C K, KAHL W. An optimized cell BE special function library generated by coconut[ J] . IEEE transactions on computers, 2009, 58(8) : 1126-1138.
[5] LAUTER C. A new open-source SIMD vector libm fully implemented with high-level scalar C [ C ] / / 2016 50th Asilomar Conference on Signals, Systems and Computers. Piscataway: IEEE, 2016: 407-411.
[6] PIPARO D, INNOCENTE V, HAUTH T. Speeding up HEP experiment software with a library of fast and autovectorisable mathematical functions[ J] . Journal of physics: conference series, 2014, 513(5) : 052027.
[7] 刘聃, 郭绍忠, 郝江伟, 等. 基于 SIMD 扩展部件的长向量超越函数实现方法[ J] . 计算机科学, 2021, 48 (6) : 26-33.
LIU D, GUO S Z, HAO J W, et al. Implementation of transcendental functions on vectors based on SIMD exten- 第 1 期杨指政,等:基于国产 PuDianNao 芯片的向量函数库优化 37 sions[J]. Computer science, 2021, 48(6): 26-33.
[8] LIU D F, CHEN T S, LIU S L, et al. PuDianNao[ J] . ACM SIGPLAN notices, 2015, 50(4) : 369-381.
[9] HUCK J, MORRIS D, ROSS J, et al. Introducing the IA64 architecture[J]. IEEE micro, 2000, 20(5): 12-23.
[10] ZHANG Y, HU Y, LI B, et al. Performance and power analysis of ATI GPU: a statistical approach [ C] / / 2011 IEEE Sixth International Conference on Networking, Architecture, and Storage. Piscataway: IEEE, 2011: 149- 158.
[11] KUMURA T, IKEKAWA M, YOSBIDA M, et al. VLIW DSP for mobile applications[ J] . IEEE signal processing magazine, 2002, 19(4) : 10-21.
[12] KYUNG G, JUNG C M, LEE K. An implementation of a SIMT architecture-based stream processor[C] / / TENCON 2014-2014 IEEE Region 10 Conference. Piscataway: IEEE, 2014: 1-5.
[13] XIONG Y Q. A unified programming model for heterogeneous computing with CPU and accelerator technologies [C] / / 12th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) . Piscataway: IEEE, 2019: 1-4.
[14] MULLER J K. On the definition of ulp ( x ) [ R / OL ] . (2005-02-01) [ 2022- 02- 10] . https: / / www. researchgate. net / publication / 236944278_On _ the _ definition _ of _ ulpx.
[15] Free Software Foundation. The GNU C Library ( glibc ) [EB / OL] . ( 2019 - 11 - 01) [ 2022 - 02 - 10] . https: / / www. gnu. org / software / libc / .
[16] NVIDIA. CUDA Math API [ EB / OL] . ( 2022 - 01 - 12) [2022- 02 - 10 ] . https: / / docs. nvidia. com / cuda / cudamath-api / index. html.

Similar References:

Memo

Last Update: 2022-12-07