Cache-optimized implemention of long sequences FFT on TS201

This paper proposes an improved method for Winograd algorithm, aiming to solve the problem that the existing methods of long sequences Fast Fourier Transform (FFT) on the TS201 processor doesn´t take full account of the Cache´s miss influence on efficiency. The new method makes maximum use of the Cache´s advantages in reading and writing by optimizing the access method of rows and columns to avoid three explicitly matrix transposition, and hiding the twiddle factor multiplication by reconfiguration butterfly computation. Test results show that the performance of Cache-Optimized Implementation of FFT has been significantly improved.