conv does not use fft, or at least not when I last checked, as I wrote my own convolution function to use the fft in place of it.
If you get the padding and all that right then within computational error the results of conv should be equal to those of doing an ifft on the product of two ffts.