|
| Parallelization of 3D explicit material point method using OpenMP |
| Received:March 28, 2008 |
| View Full Text View/Add Comment Download reader |
| DOI：10.7511/jslx20101004 |
| KeyWord:OpenMP Explicit MPM domain decomposition method loop splitting method speedup |
| Author | Institution |
| 黄鹏 |
清华大学 航天航空学院,北京 ;中国工程物理研究院 总体工程研究所,绵阳 |
| 张雄 |
清华大学 航天航空学院,北京 ;大连理工大学工业装备结构分析国家重点实验室,大连 |
| 马上 |
清华大学 航天航空学院,北京 |
| 王汉奎 |
清华大学 航天航空学院,北京 |
|
| Hits: 2169 |
| Download times: 1445 |
| Abstract: |
| Based on the OpenMP technique, a parallel 3D explicit material point method (MPM) code, MPM3DMP, is developed in this paper. The domain decomposition method is presented for avoiding data races in updating nodal variables. The background grid is decomposed into some uniform patches, and each thread deal with a patch in domain decomposition method. After updating nodes in all patches, their nodal variables are assembled into the global one. The code for updating particle variables can be parallelized using the loop splitting method directly. To test the performance of the developed code, the Taylor bar impact is simulated by using three models with different number of particles on a computer with dual quad-core Intel Woodcrest processors. A speedup of 3.82 is achieved for the coarse model in 4 cores, and 6.23 in 8 cores, while a speedup of 3.79 is achieved for the medium-sized model in 4 cores, and 6.23 in 8 cores. A speedup of 3.75 is achieved for the fine model in 4 cores, and 6.26 in 8 cores. Thus, MPM3DMP has good parallel efficiency and extension ability. |
|
|
|