gpuroofline: a model for guiding performance optimizations on gpus

ISCAS OpenIR

	gpuroofline: a model for guiding performance optimizations on gpus
	Jia Haipeng; Zhang Yunquan; Long Guoping; Xu Jianliang; Yan Shengen; Li Yan
	2012
会议名称	18th International Conference on Parallel Processing, Euro-Par 2012
会议录名称	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
页码	920-932
会议日期	August 27, 2012 - August 31, 2012
会议地点	Rhodes Island, Greece
收录类别	EI
ISSN	0302-9743
ISBN	9783642328190
部门归属	(1) Lab. of Parallel Software and Computational Science Institute of Software Chinese Academy of Sciences China; (2) College of Information Science and Engineering Ocean University of China China; (3) State Key Laboratory of Computing Science Chinese Academy of Sciences China; (4) Graduate University of Chinese Academy of Sciences China
摘要	Performance optimization on GPUs requires deep technical knowledge of the underlying hardware. Modern GPU architectures are becoming more and more diversified, which further exacerbates the already difficult problem. This paper presents GPURoofline, an empirical model for guiding optimizations on GPUs. The goal is to help non-expert programmers with limited knowledge of GPU architectures implement high performance GPU kernels. The model addresses this problem by exploring potential performance bottlenecks and evaluating whether specific optimization techniques bring any performance improvement. To demonstrate the usage of the model, we optimize four representative kernels with different computation densities, namely matrix transpose, Laplace transform, integral and face-dection, on both NVIDIA and AMD GPUs. Experimental results show that under the guidance of GPURoofline, performance of those kernels achieves 3.74~14.8 times speedup compared to their nai¨ve implementations on both NVIDIA and AMD GPU platforms. © 2012 Springer-Verlag.; Performance optimization on GPUs requires deep technical knowledge of the underlying hardware. Modern GPU architectures are becoming more and more diversified, which further exacerbates the already difficult problem. This paper presents GPURoofline, an empirical model for guiding optimizations on GPUs. The goal is to help non-expert programmers with limited knowledge of GPU architectures implement high performance GPU kernels. The model addresses this problem by exploring potential performance bottlenecks and evaluating whether specific optimization techniques bring any performance improvement. To demonstrate the usage of the model, we optimize four representative kernels with different computation densities, namely matrix transpose, Laplace transform, integral and face-dection, on both NVIDIA and AMD GPUs. Experimental results show that under the guidance of GPURoofline, performance of those kernels achieves 3.74~14.8 times speedup compared to their nai¨ve implementations on both NVIDIA and AMD GPU platforms. © 2012 Springer-Verlag.
关键词	Laplace Transforms Optimization
语种	英语
内容类型	会议论文
URI标识	http://ir.iscas.ac.cn/handle/311060/15892
专题	中国科学院软件研究所
推荐引用方式 GB/T 7714	Jia Haipeng,Zhang Yunquan,Long Guoping,et al. gpuroofline: a model for guiding performance optimizations on gpus[C],2012:920-932.