摘要
FFT算法是高度并行的分治算法,因此适合在GPU(Graphics Processing Unit,图形处理器)的CUDA(Compute Unified Device Architecture,计算统一设备体系结构)构架上实现.阐述了GPU用于通用计算的原理和方法,并在Geforce8800 GT平台上完成了二维卷积FFT的运算实验.实验结果表明,随着图像尺寸的增加,CPU和GPU上的运算量和运算时间大幅度增加,GPU上运算的速度提高倍数也随之增加,平均提升20倍左右.
The fact that FFT is a highly paralleled divide-and-conquer algorithm determines that it can be applied to compute unified dovice architecture （CUDA） of graphics processing unit （GPU）. This paper deals with the principle and method of applying GPU to general purpose computation. And the algorithm of 2D FFT was simulated on the platform of Geforce8800 GT. The results indicate that with the increase of image size, the calculation and calculating time of CPU and GPU increase significantly and the calculating speed increases by 20 times on the average.
出处
《南京工程学院学报（自然科学版）》
2009年第2期41-45,共5页
Journal of Nanjing Institute of Technology(Natural Science Edition)
基金
南京工程学院科研基金项目(KXJ07014)
关键词
图形处理器
计算统一设备体系结构
通用计算
二维FFT
graphics processing unit
computer unified device architecture
general purpose computation
2D FFT
