gpuroofline: a model for guiding performance optimizations on gpus

CORC > 软件研究所 > 软件所图书馆 > 会议论文

	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
会议日期	August 27, 2012 - August 31, 2012
会议地点	Rhodes Island, Greece
关键词	Laplace transforms Optimization
页码	920-932
中文摘要	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.
英文摘要	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.
收录类别	EI
会议录	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
语种	英语
ISSN号	0302-9743
ISBN号	9783642328190
内容类型	会议论文
源URL	[http://ir.iscas.ac.cn/handle/311060/15892]
专题	软件研究所_软件所图书馆_会议论文
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]. 见:18th International Conference on Parallel Processing, Euro-Par 2012. Rhodes Island, Greece. August 27, 2012 - August 31, 2012.