摘要
为克服交叉相关外推算法时间复杂度高、运算时间过长的缺点,提出一种基于GPU的快速并行化算法,应用于地闪落点的外推预测。首先分析串行的算法流程,然后对算法进行并行化分析设计,再针对AMD系列GPU硬件架构特点,运用OpenCL技术从主存与设备内存之间的数据传输、显存访问模式等方面对算法进一步优化。最后将地闪监测实况数据与本算法外推计算结果进行比对,分析不同精度下串行与并行算法的计算效率。实验结果表明,该算法充分利用GPU强大的并行计算能力,计算速度提高了近17倍。
To overcome the shortcomings of high time complexity and long computing time about cross-correlation extrapolation al- gorithm, an improved GPU-based fast parallel algorithm is presented, which is applied to extrapolate the Cloud-to-Ground flash development trend. First of all, we analyze the serial algorithm flow, and then design the parallel algorithm flow, optimize the al- gorithm by way of improving the data transfer between device memory and main memory, and optimize the memory access pat- terns. These optimization measures are based on OpenCL technology and aimed at the hardware architecture of AMD series GPU. Finally, we compare the Cloud-to-Ground flash monitoring data against the extrapolation results computed by this algorithm, and analyze the efficiency of the serial and the parallel algorithms under different precisions. The experimental result indicates that, the algorithm takes advantage of the powerful GPU parallel computing capability, and the calculation speed increases by nearly 17 times.
出处
《计算机与现代化》
2014年第2期213-218,共6页
Computer and Modernization
基金
江苏省2012年度普通高校研究生科研创新计划项目(CXZZ12_0513)
国家科技支撑计划项目(2012BAH05B01)
