SAR成像并行仿真的访存局部性优化技术

摘要

Synthetic aperture radar (SAR)has critical performance requirements in terms of the computing speed and memory space. Thus,the locality optimisation technique based on parallel computing platform has become one of the key techniques for reducing the memory access overhead and increasing the imaging speed.Represented by a typical SAR imaging simulation program,namely R-D,we study the locality optimising technique in SAR imaging parallel simulation.Especially,according to the locality optimisation theories of loop and data layout transformations,we propose the locality optimising technique of transposing and blocking for large-scale data of SAR echo signals. Experimental results show that the SAR imaging program with the memory access locality optimising technique applied can achieve obvious performance improvement.%合成孔径雷达SAR（Synthetic Aperture Radar）成像仿真对处理速度和存储空间都有很高的要求，因此基于并行计算平台的局部性优化技术成为了降低访存开销、提高SAR成像速度的关键技术之一。以典型SAR成像仿真程序---R-D程序为代表，研究SAR成像并行仿真中的访存局部性优化技术，重点根据循环变换和数据布局变换的局部性优化理论，提出对大规模SAR回波数据进行转置和分块的局部性优化技术。实验结果证明，经过访存局部性优化的SAR成像程序能够获得显著的性能提升。