why is calculating convolution in frequency domain quicker than conv2()

so i have two programs written that calculate the convolution of a picture with a high-pass filter and i used tic/toc to calculate the time it took to make the calculations. it showed that ifft2(fft2(a)*fft2(b)) was faster than conv2(). could someone explain why these calculations are done faster in the frequency domain?

[Rik]

The direct calculation is the fastest method anyway, so I would image conv2 would need to do some input checking before it concludes that it is the faster method. Or maybe something else is going on. cellfun('length',A) being much faster than cellfun(@length,A) has it's roots in how the function started, although why this is still the case is a mystery to me.

[justin stephens]

thank you very much