It’s a collection of Fast Fourier Transform algorithms, featured in Joel Yliluoma’s master’s thesis (“pro gradu”) in University of Helsinki.

Supported algorithms:

• Discrete Fourier Transform (naive DFT)
• Cooley-Tukey FFT (radix-2)
• Cooley-Tukey FFT (generic)
• Rader’s FFT
• Bluestein’s FFT
• Combined (heuristic) algorithm
• FFTW (for reference)

• defs.hh: Definitions for complex types and complex vectors.
• exp.hh and exp.cc: A module for handling exp-vectors containing $e^{-i2\pi x/N}$ values.
• factor.hh and factor.cc: A module for integer factorization utilities.
• dft.hh and dft.cc: The naive DFT. Also works as a fallback for any algorithm. Contains optimized FFTs for sizes ≤ 4.
• fft_fftw.hh and fft_fftw.cc: FFT through libFFTW3.
• fft_radix2.hh and fft_radix2.cc: Fast FFT for powers-of-two lengths (Cooley-Tukey radix-2).
• fft_tukey.hh and fft_tukey.cc: Generic Cooley-Tukey algorithm. Requires that the input length is a composite number.
• fft_rader.hh and fft_rader.cc: Rader’s FFT. Requires that the input length is a prime number. Performs sub-FFTs using FFT_any for length $N-1$.
• fft_bluestein.hh and fft_bluestein.cc: Bluestein’s FFT. Performs sub-FFTs using FFT_any for length $m$. Provided are three versions:
• • FFT_bluestein selects smallest convolution length $m ≥ 2N-1$ such that $m$ consists of only these factors: 2, 3, 5 or 7.
• • FFT_bluestein_pow2 selects smallest convolution length $m ≥ 2N-1$ such that $m$ consists of only these factors: 2.
• • FFT_bluestein_fac2_3 selects smallest convolution length $m ≥ 2N-1$ such that $m$ consists of only these factors: 2 or 3.
• fft_any.hh and fft_any.cc: Combined FFT. Uses heuristics to decide the fastest method for any input size (excluding FFTW).

Also

• render.cc: See below
• planner.cc: A tool for generating FFTW wisdom files

The program render.cc is used to render speed comparison graphs for the thesis. This tool was created because LibreOffice Calc was too sluggish to operate with CSV files containing tens of thousands of rows and multiple columns.

The program main.cc in the speech/ branch is used for analyzing vowel formants in speech by FFT.