Author(s)

Youcai Luo, Hongsheng Wang and Lin Han

Corresponding Author

Youcai Luo

ABSTRACT

The Fast Fourier Transform is one of the most fundamental algorithms used in digital signal processing. It is of great significance to implement an efficient Fast Fourier Transform algorithm on Sunway embedded platform. In this paper, based on the radix 2-radix 4 mixed-radix algorithm implemented in DSPLIB Library of Texas Instruments, it takes the lookup table to replace the reversal computation function, which improves the performance by 1.69 times. In addition, the algorithm is manually optimized for the Sunway embedded processor by using loop unrolling optimization, SIMD vectorization, and data prefetch optimization with the studying of Sunway processor chip architecture. Comparing the initial state, the optimal average performance of the algorithm has been improved by about three times on Sunway.

KEYWORDS

Sunway processor; fast fourier transform; lookup table; loop unrolling; vectorization; data prefetching