基于超标量处理器的高效FFT映射方法

An Efficient FFT-Mapping Method Based on Superscalar Processor

  • 摘要: 针对超标量处理器的结构特点,研究新的映射方法,实现高效FFT运算.对现代超标量结构处理器进行建模,分析FFT算法在其上执行情况,得出内存访问是FFT算法执行的关键点.并进一步对FFT的内访问过程进行建模分析,最终实现了一种基于cache优化的高效FFT映射方法,该方法将FFT进行拆分实现,充分发挥了cache的作用,进而提高了处理性能.最后在ADI公司的TS201数字信号处理器上,以该映射方法为指导实现了基2FFT算法,实验结果显示在处理点数超出cache容量时,本映射方法可以大幅度提高处理性能.

     

    Abstract: Considering the technology background and the structure characteristic of superscalar processor, novel mapping method was studied to implement the FFT effectively. Firstly, the structure of modern superscalar processor was modeled, and then the effect of FFT implementation on the processor was analyzed. The analysis results show that the EMS memory accessing is the key point of FFT algorithm implementation. Further more, the accessing process of FFT processing was modeled to implement FFT mapping effectively based on cache optimization. The new mapping method splits sequence of FFT into short sequence to exert the function of cache sufficiently. Finally, a radix-2 FFT was implemented on the ADI's TS201 digital signal processor based on the new mapping method. The result shows that the implementation time of FFT is improved greatly.

     

/

返回文章
返回