摘要
快速福利叶变换在图像处理领域,尤其是在图像复原算法中作为常用的计算工具,将时域计算转变为频域计算,在工程应用中有着非常重要的意义。采取多线程分块以及并行的映射方法,可以使FFT算法最大程度并行。针对OpenCL的存储层次特点和算法层次的优化,在AMD GPU平台上取得了明显的加速效果。优化后的算法性能比具有相同处理能力的CPU平台提高了7倍,比具有相同处理能力的CUDA提高了4倍。
Fast Fourier transform, as a commonly used computational tool in the field of image processing, especially in image restoration algorithm, transforms the time-domain computation into frequency-domain computation and has great significance for engineering applications. By adopting thread blocks and parallel mapping method,, we can make FFT algorithm reach the maximum degree of parallelism. In view of the storage features of OpenCL and the optimisation of the algorithm, the AMD GPU platform has been significantly accelerated. Compared with CPU platform and CUDA with the same processing capability, the performance of the optimised algorithm has increased by 7 times and 4 times respectively.
作者
贾格
彭先蓉
左颢睿
Jia Ge Peng Xianrong Zuo Haorui(Institute of Optics and Electronics ,Chinese Academy of Sciences, Chengdu 610209, Sichuan, China Graduate University of Chinese Academy of Sciences, Beijing 100039, China)
出处
《计算机应用与软件》
2017年第3期233-237,283,共6页
Computer Applications and Software