For those of us in signal processing research, the built-in fft function in Matlab (or Octave) is what we use almost all the time. It is adaptive in that it will choose the best algorithm available for the desired transform size.
For C or C++ applications, there are several highly optimized FFT variants in the FFTW package (``Fastest Fourier Transform in the West'') . FFTW is free for non-commercial or free-software applications under the terms of the GNU General Public License.
For embedded DSP applications (software running on special purpose DSP chips), consult your vendor's software libraries and support website for FFT algorithms written in optimized assembly language for your DSP hardware platform. Nearly all DSP chip vendors supply free FFT software (and other signal processing utilities) as a way of promoting their hardware.