Institutional Repository
| gpuroofline: a model for guiding performance optimizations on gpus | |
| Jia Haipeng; Zhang Yunquan; Long Guoping; Xu Jianliang; Yan Shengen; Li Yan | |
| 2012 | |
| Conference Name | 18th International Conference on Parallel Processing, Euro-Par 2012 |
| Source | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
| Pages | 920-932 |
| Conference Date | August 27, 2012 - August 31, 2012 |
| Conference Place | Rhodes Island, Greece |
| Indexed Type | EI |
| ISSN | 0302-9743 |
| ISBN | 9783642328190 |
| Department | (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 |
| English Abstract | 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. |
| Keyword | Laplace Transforms Optimization |
| Language | 英语 |
| Content Type | 会议论文 |
| URI | http://ir.iscas.ac.cn/handle/311060/15892 |
| Collection | 中国科学院软件研究所 |
| Recommended Citation GB/T 7714 | Jia Haipeng,Zhang Yunquan,Long Guoping,et al. gpuroofline: a model for guiding performance optimizations on gpus[C],2012:920-932. |
| Files in This Item: | There are no files associated with this item. | |||||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment