Optimization Method for Reusing GPU Data Based on Dynamic Spanning Tree

doi:10.7652/xjtuxb201310008

您当前的位置：

首页 >

文章列表页 >

Optimization Method for Reusing GPU Data Based on Dynamic Spanning Tree

更新时间：2025-07-09

- Optimization Method for Reusing GPU Data Based on Dynamic Spanning Tree
- Vol. 47, Issue 10, Pages: 44-50(2013)
- 作者机构：
  
  西安交通大学电子与信息工程学院,西安,710049
- 作者简介：
- 基金信息：
- DOI：10.7652/xjtuxb201310008
  CLC： TP399
- Online First：10 October 2013，
  
  Published：2013
- 稿件说明：
移动端阅览
Optimization Method for Reusing GPU Data Based on Dynamic Spanning Tree[J]. 2013, 47(10): 44-50.
DOI：

Optimization Method for Reusing GPU Data Based on Dynamic Spanning Tree[J]. 2013, 47(10): 44-50. DOI： 10.7652/xjtuxb201310008.

摘要

针对手工优化GPU(Graphic Processing Unit)显存级数据复用过程复杂和编译时优化数据复用开销过大的问题

提出了一种基于动态生成树在运行时进行数据复用的优化方法

可为程序员提供一种透明且高效简单的优化方式。该方法将已经执行的GPU计算任务的数据访问抽象为生成树的叶子节点

利用动态生成树管理GPU数据访问信息

实现了运行时GPU显存级数据的复用优化

并通过运行时对生成树的搜索和维护

动态地发掘和优化GPU显存级数据复用

因此

在编程时不需要进行复杂且困难的数据复用分析

直接调用文中提出的运行时库就能有效减少程序执行过程中CPU内存和GPU显存之间的冗余数据传输次数

从而提升应用的运行性能。实验结果表明

使用文中提出的优化方法可有效消除未进行数据复用优化的CPU-GPU应用程序中的冗余数据传输

最大加速比达原始执行的3～10倍

额外开销不到优化后程序总执行时间的5%。

Abstract

An optimization method for reusing graphic processing unit(GPU)data is proposed based on the dynamic spanning tree to solve the problem that the reusing process with manual optimization is complicated and compilation overhead in optimization is expensive. The proposed method is transparent to programmers with simplicity and effectiveness. The approach abstracts the executed accesses of GPU tasks as leaf nodes of a spanning tree

and uses the tree to dynamically manage the access information. Then the identification and optimization for the data reusing is realized by searching and managing the tree. The complex and challenging data reuse analysis is not required since invoking the run-time library is sufficient to reduce the CPU-GPU data transfers. Experiments show that the proposed optimization method can eliminate redundant CPU-GPU data transfers in non-reuse CPU-GPU application programs and can achieve a speedup as large as 3 to 10 times over the original execution. Moreover

the additional cost is just less than 5% of the execution time of the program adopting the proposed optimization.

关键词

Keywords

references

YANG Xuejun, LIAO Xiangke, LU Kai, et al. The TianHe-1A supercomputer: its hardware and software [J]. Journal of Computer Science and Technology, 2011, 26(3): 344-351.

ZHU Xiaoqian, LIU Xin, MENG Xiangfei, et al. Performance analysis and optimization of gyrokinetic torodial code on TH-1A supercomputer [C]∥Proceedings of 2nd International Conference on Electrical and Control Engineering. Piscataway, NJ, USA: IEEE, 2011: 6027-6031.

FENG Xiaowen, JIN Hai, ZHENG Ran, et al. Optimization of sparse matrix-vector multiplication with variant CSR on GPUs [C]∥Proceedings of 17th IEEE International Conference on Parallel and Distributed Systems(ICPADS). Piscataway, NJ, USA: IEEE, 2011: 165-172.

WU Haicheng, DIAMOS G, Wang Jin, et al. Optimizing data warehousing applications for GPUs using kernel fusion/fission [C]∥Proceedings of IEEE 26th International Parallel and Distributed Processing Symposium, Workshops PhD Forum(IPDPSW). Piscataway, NJ, USA: IEEE, 2011: 2433-2442.

WOLF M E, LAM M S. A loop transformation theory and an algorithm to maximize parallelism [J]. IEEE Trans on Parallel Distrib Syst, 1991, 2(4): 452-471.

WOLF M E, LAM M S. A data locality optimizing algorithm [C]∥Proceedings of the ACM SIGPLAN'91 Conference on Programming Language Design and Implementation(PLDI). Washington, DC, USA: ACM, 1991: 30-44.

SMITH M D, RAMSEY N, HOLLOWAY G H. A generalized algorithm for graph-coloring register allocation [C]∥Proceedings of the ACM SIGPLAN 2004 Conference on Programming Language Design and Implementation(PLDI). Washington, DC, USA: ACM, 2004: 277-288.

WOLFE M. Implementing the PGI accelerator model [C]∥Proceedings of the 3rd Workshop on General-Purpose Computation on Graphics Processing Units(GPGPU). Washington, DC, USA: ACM, 2010: 43-50.

HAN T D, ABDELRAHMAN T S. hiCUDA: high-level GPGPU programming [J]. IEEE Trans on Parallel Distrib Syst, 2011, 22(1): 78-90.

WOLFE M. Optimizing data movement in the PGI accelerator programming model [EB/OL]. [2012-06-09]. http:∥www.pgroup.com/lit/articles/insider/v3n1a1.htm.

冯国富, 董小社, 胡冰, 等. 一种支持多种访存技术的CBEA片上多核MPI并行编程模型 [J]. 计算机学报, 2008, 11(31): 1965-1974.

FENG Guofu, DONG Xiaoshe, HU Bing, et al. A MPI parallel programming model for CBEA based on hybrid memory access technology [J]. Chinese Journal of Computers, 2008, 31(11): 1965-1974.

IOSEVICH V, SCHUSTER A. A comparison of sequential consistency with home-based lazy release consistency for software distributed shared memory [C]∥Proceedings of 18th Annual ACM International Conference on Supercomputing(ICS). Washington, DC, USA: ACM, 2004: 306-315.

NAS Parallel Benchmarks Team. Problem sizes and parameters in NAS Parallel Benchmarks [EB/OL]. [2012-06-11]. http:∥www.nas.nasa.gov/publications/npb_problem_sizes.html.

YARROW M, KUSZMAUL C. NPB2.3-FT Benchmark-C+CUDA [EB/OL]. [2012-06-11]. http:∥hpc gpu.codeplex.com/releases/view/34770.

张宝, 董小社, 白秀秀, 等. CPU-GPU系统中基于剖分的全局性能优化方法 [J]. 西安交通大学学报, 2012, 46(2): 17-23.

ZHANG Bao, DONG Xiaoshe, BAI Xiuxiu et al. Profiling based optimization method for CPU-GPU heterogeneous parallel processing systems [J]. Journal of Xi'an Jiaotong University, 2012, 46(2): 17-23.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Non-Uniform Random Membership Management to Construct Overlays for Transferring Scalable Video Coding

Research on the Electrohydrodynamic Effects of Ionic Liquid-Modified Fuels

A Fast Calculation Method for Thermal Flow Fields in Transformer Windings Based on PI-DeepONet

Energy Efficiency Optimization of the Cryogenic Air Separation Process for Electronic-Grade Ultra Pure Nitrogen

Effects of Expansion Angle Crater Film Hole on the Film Cooling Characteristics of a Double-Wall Structure

Related Author

黄泳翔 1

钱德沛 1

伍卫国 1

刘涛 1

Related Institution

No data

⁰