Performance Enhancement of GPU Parallel Computing Using Memory Allocation Optimization

摘要

The Fourier transform converts a signal from its original domain to a representation in the frequency domain. Applications of the Fourier Transform are far-reaching, spanning fields such as intelligent information processing, machine vision, physics, mathematics, medical science, and telecommunications; hence, its applications have become an indispensable part in our daily life. Therefore, it is essential to construct efficient and high-reliability schemes to guarantee smooth performance of the systems using Fourier Transforms. This study compares performances of Fast Fourier Transforms on a host CPU, GPU parallel computing, and GPU parallel computing with memory allocation optimization. From the experimental results, GPU parallel computing is proven to be effective in enhancing computation speed of the FFT; the speedup ratio of GPU parallel computing over the CPU can reach 48 when operating on 32678 8-byte complex input data. In addition, by optimizing GPU memory allocation, the computation speed of the FFT can be further enhanced; the speedup ratio of GPU parallel computing with memory allocation optimization over the CPU can reach 114.7 when operating on 32678 8-byte complex input data.