| 引用本文: | 吴建平,王正华,李晓梅.利用混合编程改善SMP机群上并行矩阵乘法的性能.[J].国防科技大学学报,2006,28(4):68-72.[点击复制] |
| WU Jianping,WANG Zhenghua,LI Xiaomei.Improve the Performance of Parallel Matrix Multiplication on Clustered SMP Systems Through Hybrid Programming[J].Journal of National University of Defense Technology,2006,28(4):68-72[点击复制] |
|
| |
|
|
| 本文已被：浏览 6564次 下载 5508次 |
| 利用混合编程改善SMP机群上并行矩阵乘法的性能 |
| 吴建平1, 王正华1, 李晓梅2 |
|
(1.国防科技大学 计算机学院，湖南 长沙 410073;2.装备指挥技术学院，北京 101416)
|
| 摘要: |
| 针对SMP机群，探讨了分别利用单机优化、OpenMP与MPI从指令级、共享存储级与分布主存级三个层次上改善矩阵并行乘Fox算法性能的方法。并通过调用数学函数库与混合编程的方式，在深腾6800上进行了实验，取得了相当满意的数值效果。 |
| 关键词: SMP机群 OpenMP MPI 混合编程 Fox算法 |
| DOI： |
| 投稿日期：2006-03-17 |
| 基金项目:国家自然科学重点基金资助项目(69933030) |
|
| Improve the Performance of Parallel Matrix Multiplication on Clustered SMP Systems Through Hybrid Programming |
| WU Jianping1, WANG Zhenghua1, LI Xiaomei2 |
|
(1.College of Computer, National Univ. of Defense Technology，Changsha 410073，China;2.Academy of Command and Technology of Equipment, Beijing 101416, China)
|
| Abstract: |
| In order to give a full play of SMP clusters, this research studied the method to improve the performance of Fox algorithm for parallel matrix multiplication exploiting optimization on single processor, OpenMP and MPI, which involves the levels of instruction, shared memory and distributed memory respectively. Through invoking mathematic library and hybrid programming, the numerical results derived on DeepComp 6800 are satisfactory. |
| Keywords: SMP cluster OpenMP MPI hybrid programming Fox algorithm |
|
|
