| 摘要: |
| 灰度弥散模型被广泛应用于模拟星模拟器的成像过程.在实际问题域中,该模型需要巨大的计算能力以完成繁重的数值计算,而目前图形处理单元(GPU)已经发展成为一种有效的数值处理平台,对于计算密集型模拟具有很好的加速能力.设计并实现了GPU 平台下,基于统一计算架构(CUDA)的并行灰度模型,可应用于大规模星模拟器的快速灰度模拟.首先分析了该模型具有的双重并行特性,并采用CUDA 模型模拟其良好的数据并行特征.为了便于对比研究,设计了两类模拟器:一类是串行模拟器作为基准模拟器;另一类是基于CUDA 的并行模拟器.同时,在并行策略、模型以及GPU 实现层面分别给出不同的优化方法以有效提高并行效率.最后,设计对应于双重并行粒度的两类测试基准,以评估并行模拟器的性能.数据分析表明,CUDA 并行模拟器取得良好的性能提升,同时也给出了该模拟器中存在的一些限制. |
| 关键词: GPU 计算 CUDA 星模拟器 弥散效果 灰度模型 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金(60303020); 国家高技术研究发展计划(863)(2009AA01Z303) |
|
| Intensity Model with Blur Effect on GPUs Applied to Large-Scale Star Simulators |
|
LI Chao1,2, ZHANG Yun-Quan3, ZHENG Chang-Wen1, HU Xiao-Hui1
|
|
1.National Key Laboratory of Integrated Information System Technology, Institute of Software, The Chinese Academy of Sciences, Beijing 100190, China;2.Graduate University, The Chinese Academy of Sciences, Beijing 100049, China;3.Laboratoy of Parallel Software and Computational Science, Institute of Software, The Chinese Academy of Sciences, Beijing 100190 China
|
| Abstract: |
| Intensity model with blur effect is widely employed to accurately simulate the imaging process of star simulator used for attitude determination and guiding feedback. It imposes great demands of computing power for realistic domains, and modern Graphics Processing Units (GPUs) have demonstrated to be a powerful accelerator for this kind of computationally intensive simulations. This paper presents a parallel design and implementation of the intensity model applied to large-scale star simulators on GPUs using the compute unified device architecture (CUDA) programming model. The study analyzes the double parallel nature inherent in this model and use the CUDA framework to efficiently exploit the potential fine-grain data parallelism. Two versions of simulator are designed and studied: One is sequential simulator used as the baseline simulator, and another is parallel simulator using CUDA. In parallel strategy, model, and GPU implementation level, the study employs specific optimized strategies to efficiently improve the parallel performance. Finally, two benchmarks corresponding with the double parallelism are developed to fully evaluate the performance behavior of our simulators. The result analysis demonstrates the efficiency of the CUDA simulators and also illustrates the restriction and bottlenecks presented in this simulator. |
| Key words: GPU computing CUDA star simulator blur effect intensity model |
