An Implementation of Parallel 2-D FFT Using Intel AVX Instructions on Multi-core Processors

摘要

In this paper, we propose an implementation of a parallel two-dimensional fast Fourier transform (FFT) using Intel Advanced Vector Extensions (AVX) instructions on multi-core processors. The combination of vectorization and a block two-dimensional FFT algorithm is shown to effectively improve performance. We vectorized FFT kernels using the AVX instructions. Performance results of two-dimensional FFTs on multi-core processors are reported. We successfully achieved a performance of over 61 GFlops on an Intel Xeon E5-2670 (2.6 GHz, two CPUs, 16 cores) and over 24 GFlops on an Intel Core i7-3930K (3.2 GHz, one CPU, six cores) for a 2~(12) × 2~(12)-point FFT.