GPU实现的高速FIR数字滤波算法

High Speed FIR Digital Filtering on GPU

针对目前基于GPU的FIR算法速度低、扩展性差的缺点,提出一种高速的多通道FIR数字滤波的并行算法,并利用平衡并行运算负载的技术以及降低内存访问密度的方法进行加速.该算法采用矩阵乘法的并行运算技术在GPU上建立并行滤波模型,通过每个线程在单个指令周期内执行2个信号运算,实现了多通道信号的高速滤波.实验结果表明,在GTX260+平台上,采用文中算法的平均加速比达到了203,效率超过40%,并且具有更好的扩展性.

This paper proposes a massively parallel FIR filter algorithm with its GPU implementation to improve the efficiency and scalability of GPU based FIR.By the algorithm proposed,the problem is formulated as a matrix multiplication operation that offers sufficient data level parallelism for parallel filtering on modern GPUs.In addition,the GPU implementation guarantees that each thread could complete a two-signal operation within a single instruction cycle.Efficient and effective strategies for load balancing and memory mapping are also introduced to further improve the performance.The proposed algorithm and the corresponding GPU implementation could achieve an efficient multi-channel signal filtering.Experimental results on a GTX260＋ graphics card prove that the FIR filter algorithm could be used to deliver on average a speed-up of 203X and an efficiency increase over 40%.