面向CUDA程序的性能预测框架

doi:10.3969/j.issn.0372-2112.2020.04.006

电子学报 ›› 2020, Vol. 48 ›› Issue (4): 654-661.DOI: 10.3969/j.issn.0372-2112.2020.04.006

面向CUDA程序的性能预测框架

曲海成, 于思淼, 刘万军, 王鑫源

辽宁工程技术大学软件学院, 辽宁葫芦岛 125105

收稿日期:2018-11-26 修回日期:2019-11-12 出版日期:2020-04-25
- 通讯作者:
- 于思淼
- 作者简介:
- 曲海成 男,1981年9月出生,山东烟台人.副教授、硕士生导师、CCF会员、IEEE会员.2005年、2008年和2016年分别在青岛理工大学、辽宁工程技术大学和哈尔滨工业大学获学士学位、硕士学位和博士学位.现为辽宁工程技术大学软件学院软件工程系主任,主要研究方向为遥感影像高性能计算,数字图像处理等.E-mail:quhaicheng@lntu.edu.cn;刘万军 男,1959年10月出生,辽宁北镇人.教授、博士生导师、CCF高级会员.1991年在辽宁工程技术大学获硕士学位,主要从事数字图像处理,运动目标检测与跟踪等研究工作.Email:liuwanjun@lntu.edu.cn
- 基金资助:
- 国家自然科学基金青年基金 (No.41701479）; 辽宁省自然科学基金 (No.20180550529）; 辽宁省教育厅科学研究基础研究 (No.LJ2019JL010）

Performance Prediction Framework for CUDA Programs

QU Hai-cheng, YU Si-miao, LIU Wan-jun, WANG Xin-yuan

College of Software, Liaoning Technical University, Huludao, Liaoning 125105, China

Received:2018-11-26 Revised:2019-11-12 Online:2020-04-25 Published:2020-04-25

摘要/Abstract

摘要： 为对CUDA并行程序内核性能进行分析和预测，从而指导并行程序设计及性能优化，提出一种性能预测框架.1）从GPU编程模型和设备架构细节入手，以线程束为研究单位，通过整合与GPU程序用时密切相关的软硬件基本特征，定义了并行空间闲置度、流处理器线程束负载、并行效应因子等高层次性能相关特征.2）基于上述特征，框架针对线程负载均衡型GPU程序，评估内核函数在不同问题规模以及执行配置下的执行时间.3）依据性能评估原理提出了内核函数执行配置参数的优化策略.验证实验结果表明，该框架在两种典型情境下对现有程序性能的平均预测准确率分别达到89%和94%，客观归纳了高层次特征与程序性能间的相关关系，且能定性分析并行算法性能水平.

关键词: 性能预测, 线程束, 设备并行空间, 并行效应, 性能特征, 执行配置参数优化

Abstract: In order to analyze and predict the performance of CUDA program kernel and guide parallel program design and performance optimization,a performance prediction framework is proposed.This paper starts with the GPU programming model and hardware architecture details,with warp as the research unit.By integrating hardware and software factors closely related to GPU program time,high-level performance-related features such as device parallel space idle degree (DPSID),number of streaming multiprocessor warp (NSMW) are defined.Based on the above features,a framework for evaluating the execution time of kernel functions under different problem sizes and execution configurations is built for thread load balancing GPU programs.The principle of optimizing configuration parameters of kernel function execution is put forward to guide optimizing program performance.The experimental results show that the average prediction accuracy of the framework is 89% and 94% in the two scenarios,respectively.

Key words: performance prediction, thread warp, device active warps, parallel effect, performance features, execution configuration optimization

中图分类号:

TP302.7

曲海成, 于思淼, 刘万军, 王鑫源. 面向CUDA程序的性能预测框架[J]. 电子学报, 2020, 48(4): 654-661.

QU Hai-cheng, YU Si-miao, LIU Wan-jun, WANG Xin-yuan. Performance Prediction Framework for CUDA Programs[J]. Acta Electronica Sinica, 2020, 48(4): 654-661.

参考文献

[1] 巨涛,朱正东,董小社.异构众核系统及其编程模型与性能优化技术研究综述[J].电子学报,2015,43(1):111-119. JU Tao,ZHU Zheng-dong,DONG Xiao-she.The feature,programming model and performance optimization strategy of heterogeneous many-core system:a review[J].Acta Electronica Sinica,2015,43(1):111-119.(in Chinese)
[2] NVIDIA Corporation.CUDA C programming guide[OL].https://docs.nvidia.com/cuda/archive/8.0/,2018.
[3] 张珩,张立波,武延军.基于Multi-GPU平台的大规模图数据处理[J].计算机研究与发展,2018,55(2):273-288. ZHANG Heng,ZHANG Li-bo,WU Yan-jun.Large-scale graph processing on multi-GPU platforms[J].Journal of Computer Research and Development,2018,55(2):273-288.(in Chinese)
[4] ZHU Z,XU S,TANG J,et al.GraphVite:A high-performanceCPU-GPU hybrid system for node embedding[A].Proceedings of the The World Wide Web Conference[C].New York:ACM,2019.2494-2504.
[5] GONG L,ZHANG C,DUAN L,et al.Nonrigid image registration using spatially region-weighted correlation ratio and GPU-acceleration[J].IEEE Journal of Biomedical and Health Informatics,2018,23(2):766-778.
[6] CHAPUIS G,EIDENBENZ S,SANTHI N.GPU performance prediction through parallel discrete event simulation and common sense[A].Proceedings of the 9th EAI International Conference on Performance Evaluation Methodologies and Tools[C].Brussels:ICST (Institute for Computer Sciences,Social-Informatics and Telecommunications Engineering),2016.204-211.
[7] 冯晓,戴紫彬,蔡路亭,等.基于Amdahl定律扩展的多核处理器性能模型研究[J].电子学报,2017,45(6):1424-1430. FENG Xiao,DAI Zi-bin,CAI Lu-ting,et al.Performance model for multicore processor based on extended Amdahl’s law[J].Acta Electronica Sinica,2017,45(6):1424-1430.(in Chinese)
[8] 郑祯,翟季冬,李焱,等.基于CUPTI接口的典型GPU程序负载特征分析[J].计算机研究与发展,2016,53(6):1249-1262. ZHENG Zhen,ZHAI Ji-dong,LI yan,et al.Workload analysis for typical GPU programs using CUPTI interface[J].Journal of Computer Research and Development,2016,53(6):1249-1262.(in Chinese)
[9] PARK I K,SINGHAL N,LEE M H,et al.Design and performance evaluation of image processing algorithms on GPUs[J].IEEE Transactions on Parallel and Distributed Systems,2010,22(1):91-104.
[10] CUI Z,LIANG Y,RUPNOW K,et al.An accurate GPU performance model for effective control flow divergence optimization[A].Proceedings of the 2012 IEEE 26th International Parallel and Distributed Processing Symposium[C].Piscataway:IEEE,2012.83-94.
[11] ZHOU K,TAN G,ZHANG X,et al.A performance analysis framework for exploiting GPU microarchitectural capability[A].Proceedings of the International Conference on Supercomputing[C].New York:ACM,2017.15.
[12] BALDINI I,FINK S J,ALTMAN E.Predicting GPU performance from CPU runs using machine learning[A].Proceedings of the 2014 IEEE 26th International Symposium on Computer Architecture and High Performance Computing[C].Piscataway:IEEE,2014.254-261.
[13] ARDALANI N,LESTOURGEON C,SANKARALINGAM K,et al.Cross-architecture performance prediction (XAPP) using CPU code to predict GPU performance[A].Proceedings of the 48th International Symposium on Microarchitecture[C].New York:ACM,2015.725-737.
[14] O’NEAL K,BRISK P,ABOUSAMRA A,et al.GPU performance estimation using software rasterization and machine learning[J].ACM Transactions on Embedded Computing Systems (TECS),2017,16(5s):148.
[15] LOUBOUTIN M,LANGE M,HERRMANN F J,et al.Performance prediction of finite-difference solvers for different computer architectures[J].Computers & Geosciences,2017,105:148-157.
[16] LYM S,LEE D,O’CONNOR M,et al.DeLTA:GPU performance model for deep learning applications with in-depth memory system traffic analysis[A].Proceedings of the 2019 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)[C].Piscataway:IEEE,2019.293-303.
[17] GU J,LIU H,ZHOU Y,et al.DeepProf:performance analysis for deep learning applications via mining GPU execution patterns[J].arXiv preprint,2017,arXiv:1707.03750.
[18] O’NEAL K,BRISK P.Predictive modeling for CPU,GPU,and FPGA performance and power consumption:a survey[A].Proceedings of the 2018 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)[C].Piscataway:IEEE,2018.763-768.
[19] KONSTANTINIDIS E,COTRONIS Y.A quantitative roofline model for GPU kernel performance estimation using micro-benchmarks and hardware metric profiling[J].Journal of Parallel and Distributed Computing,2017,107:37-56.
[20] 黄品丰,赵荣彩,姚远,等.面向异构多核处理器的并行代价模型[J].计算机应用,2013,33(6):1544-1547. HUANG Pin-feng,ZHAO Rong-cai,YAO Yuan,et al.Parallel cost model for heterogeneous multi-core processors[J].Journal of Computer Applications,2013,33(6):1544-1547.(in Chinese)
[21] NVIDIA Corporation.CUDA occupancy calculato[OL].http://developer.download.nvidia.com/compute/cuda/CUDA_Occupancy_calculator.xls,2018.
[22] LUEBKE D.CUDA并行编程入门[OL].https://cn.udacity.com/course/intro-to-parallel-programming-cs344,2018.
[23] NVIDIA Corporation.NVIDIA visual profiler[OL].https://developer.nvidia.com/nvidia-visual-profiler,2018.

面向CUDA程序的性能预测框架

Performance Prediction Framework for CUDA Programs

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 2

编辑推荐

Metrics

[1]	车向北, 康文倩, 邓彬, 杨柯涵, 李剑. 一种基于图神经网络的SDN路由性能预测模型[J]. 电子学报, 2021, 49(3): 484-491.
[2]	乔亚男;齐勇. 查询语义图辅助的信息检索性能预测模型[J]. 电子学报, 2011, 39(3A): 158-162.