Computer Engineering and Applications ›› 2017, Vol. 53 ›› Issue (1): 172-177.DOI: 10.3778/j.issn.1002-8331.1503-0292

Previous Articles     Next Articles

Research of raster data spatial analysis under CPU/GPU heterogeneous hybrid parallel environment—Take terrain factors analysis as an example

LU Min1,2, WANG Jinyin1,2, LU Gang3, TAO Weidong1,2, WANG Jiechen1,2   

  1. 1.Jiangsu Province Key Laboratory of Geographic Information Science and Technology, Nanjing University, Nanjing 210023, China
    2.Department of Geographic Information Science, Nanjing University, Nanjing 210023, China
    3.Jiangsu Province Surveying and Mapping Engineering Institute, Nanjing 210013, China
  • Online:2017-01-01 Published:2017-01-10

CPU/GPU异构混合并行的栅格数据空间分析研究#br# ——以地形因子计算为例

卢  敏1,2,王金茵1,2,卢  刚3,陶伟东1,2,王结臣1,2   

  1. 1.江苏省地理信息技术重点实验室,南京 210023
    2.南京大学 地理信息科学系,南京 210023
    3.江苏省测绘工程院,南京 210013

Abstract: The calculating efficiency of traditional GIS raster data spatial analysis can no longer meet the demand of fast calculating under the using of mass data. This paper tests on terrain factor computation. Firstly, it analyzes and tests the computational performance of the multi-core CPU based on the shared memory model and the many-core GPU based on the CUDA model, and on the basis of the test, it accomplishes the task partitioning considering load equilibrium between equipment in detail, so as to conduct improvement research about heterogeneous hybrid parallel technology. The results show that the CPU/GPU heterogeneous hybrid parallel model has a better acceleration performance in the analysis of raster data than the multi-core shared-memory model or many-core stream processors model based on the same single computing platform.

Key words: GIS raster data spatial analysis, shared-memory parallel model, stream processors parallel model, CPU/GPU heterogeneous hybrid parallel

摘要: 海量数据背景下传统GIS栅格数据空间分析计算效率已经不能满足快速计算的需求,为此以地形因子计算为例,分析并测试了基于共享内存模型的CPU多核并行模式与基于流处理器模型的GPU众核并行模式的计算性能,在此基础上详细实现了负载均衡的设备间任务划分,进行CPU与GPU异构混合的并行技术改良研究。实验结果表明,基于相同的单机硬件环境,与多核共享内存模型或众核流处理器的单一计算平台并行方案相比,CPU/GPU异构混合并行计算方法对于栅格数据分析具有更好的加速效果。

关键词: GIS栅格数据分析, 共享内存模型, 流处理器模型, CPU/GPU异构混合并行