If you have the Intel C compiler, you also have MKL, which supports FFTW3 interfaces. https://software.intel.com/en-us/mkl-developer-reference-c-fftw3-interface-to-intel-math-kernel-library
MKL and FFTW3 have slightly different performance properties for the sizes that I have studied, but MKL has the substantial advantage of not requiring an extremely expensive "planning" step. (Planning is not particularly expensive for small FFTs (e.g., 2^10), but for large FFTs (e.g., 2^20) using multiple threads, I often see planning times of hours.)
