摘要
GPU具有很高的显存带宽和大量计算单元,随着其可编程性的不断提高,GPU越来越多地用于图像渲染以外的其他通用计算。研究了利用GPU丰富的运算资源并行加速实现脉冲压缩雷达的点迹凝聚问题。首先研究了每个目标的点迹采用一个CUDA线程的粗粒度并行方式,结果发现处理时间反而有所增加,分析了处理时间加长的原因。然后增加了并行的尺度,对单个目标的凝聚过程进行并行分解。结果表明大尺度细粒度的并行方式可以有效利用GPU强大的运算能力,加快脉冲压缩雷达的点迹凝聚速度。
GPU has the rather high video memory width and a mass of parallel calculating units.With the advance of GPU＇s programmability,GPU is used in common computing other than image manipulation.The acceleration of pulse compression Radar plots＇ agglomerating by GPU＇s powerful compute performance is studied.At first,coarse grain parallelization that each target agglomerating ran on one CUDA threads is introduced,but the result showed that plots＇ processing time increases.Then the causation of the result is analyzed,and parallel scale is enlarged by parallelization of agglomerating process inside each target.New result shows that large scale fine grain parallelization could make full use of GPU＇s powerful compute capacity and accelerate pulse compression Radar＇s plots＇ agglomerating.
出处
《火力与指挥控制》
CSCD
北大核心
2013年第3期81-85,共5页
Fire Control & Command Control
基金
海军预先研究基金资助项目(4010604020103)
