Intel mkl FFT

[DD42]

Is there an efficient way to get the complete frequency domain incl. negative frequencies (yes - I know it's redundant) when doing a real- complex forward FFT?

Background: I want to perform a Hilbert transform by forward FFT of a signal, setting the obtained negative frequencies to zero and then backward FFTing. The only possibility, I currently see, is to make the original signal complex, but that requires of lot of data copying.

[CpejiuAndrewparker]

For a real input, the negative frequencies are implied by conjugate symmetry, so you normally don't need to copy the signal into a complex array. You could reconstruct the negative half from the positive-frequency output when needed. For a Hilbert transform, though, checking whether MKL's real FFT layout already gives you the bins you need might be simpler and avoid extra copying.