基于CUDA平台的FIR滤波算法的设计与优化

针对目前基于普通DSP的FIR算法速度低、扩展性差的缺点,提出并实现基于CUDA平台实现的FIR滤波算法。由于在CUDA中程序可以直接操作数据而无需借助于图形系统的API,使开发者能够在GPU 强大计算能力的基础上建立起一种效率更高的密集数据计算解决方案。该算法将CUDA用于FIR滤波器输入输出关系计算,采用矩阵乘法的并行运算技术,在GPU上建立并行滤波模型,并对算法进行了优化。实验结果表明,在Tesla C1060平台上,和传统的基于DSP的FIR滤波算法计算速度相比,基于CUDA平台计算FIR滤波算法时,其加速比可接近30,解决了传统基于DSP计算FIR滤波算法速度较慢、扩展性差的问题。

Design and Optimization of FIR Filtering Algorithm Based on CUDA Platform

It is well known that FIR algorithm based on normal DSP has low computing speed and extensive capabilities. In order to over-come these,present a new FIR filter algorithm based on CUDA platform. Since in CUDA program can directly manipulate data without graphics API of the system,enables developers on the basis of the powerful GPU computing power to set up a efficient dense data compu-ting solutions. The algorithm adopts CUDA for FIR filter calculation of input and output relationship,using the parallel computing tech-nology of matrix multiplication,on the GPU the parallel filtering model is established,and the algorithm is optimized. Experiment on Tes-la C1060 shows that,compared with traditional FIR filter algorithm＆#39;s speed based on DSP,it can accelerate its computation speed up to 30 times,solving conventional FIR filter＆#39;s defect based on DSP of low speed and bad extending capabilities.