when compiling with fftw3, use the fftw3 provided malloc wrappers instead of the default malloc. those will guarantee proper memory alignment. that will not only help fftw, but all dynamical memory that may be used for vectorized optimizations.