I am intested in getting an estimated improvement ratio for implementing an FFT in Altivec vs not doing so. I looked in AN2115 and saw for a size N FFT Altivec is 6.99 times faster than its scalar equivalent. Is this true? That seems a bit much. Does anyone know how long it would take to perform a complex 2D FFT in Altivec for a 64x64 array?
(Alban updated title & moved to better board)
Message Edited by Alban on 2006-09-20 08:31 AM