Implementation and Optimization of the FFT Using OpenCL on Heterogeneous Platforms
Zhao Mei-chao · 2011
Fourier methods have revolutionized fields of science and engineering,from astronomy to medical imaging,from seismology to spectroscopy.A fast Fourier transform(FFT) is an efficient algorithm to compute the discrete Fourier transform(DFT) and its inverse.OpenCL(Open Computing Language) is a new framework for writing programs that execute across heterogeneous platforms consisting of CPUs,GPUs,and other processors,and it provides parallel computing using task-based and data-based parallelism.In this paper,we first implemented FFT with OpenCL,then tested and analyzed the performance of it on heterogeneous multi-core platforms like Cell,NVIDIA GPU.The perfor-mance we achieved is about 65% of Cell SDK,and 75% of CUDA CUFFT,and it needs to improve in the near further.Furthermore,we acquire unprecedented performance results that nearly 140% of CUFFT on Fermi GPU by exploiting hardware features when the size of FFT is small.