Real DFT storage

[xnoreq]

Hello,

why does an N-point real DFT need a storage size of sizeof(Meow_FFT_Complex) * N in your example? The point of a real DFT is that N real samples are converted into N/2 complex samples due to the symmetry around Nyquist. And for the inverse only N/2 complex samples can be used anyway since any asymmetry (different values for negative frequencies) would result in a complex time-domain output .. which is not possible in a real DFT.

Also, since this is an out-of-place implementation why does it need an additional temp with a sizeof(Meow_FFT_Complex) * N? With N=16384 the workset already requires a lot of memory, 556840 bytes to be precise.

[JodiTheTigger]

Oooh, thanks for asking those questions - I'll check what I did, and get back to you with an answer.

[JodiTheTigger]

A bit late - but you're correct. I'll see to making an update to the example and readme.

[JodiTheTigger]

Also, since this is an out-of-place implementation why does it need an additional temp

Due to the way it's implemented, the mixer stage of the inverse transform needs a temporary location before running the non-inplace fft stage. This is because the input array is treated as read only.