Ricardo, yes.. that's going to be your best bet. There's a bit of error in the bands because of the use of Integer math, but it's at the cost of speed to go to floating point of any sort.
Unlike my above project, you are going to want to actually tune the sample code to actually get timed samples, as my approximation (or even "fast as you can" code) is NOT lab accurate by any means.
I've been reading up and am starting to get a grip on the math. By the way, to the other poster.. the output is N/2 where N is the number of samples, this gives 64 bins due to 128 byte sample space.
I might tweak this code at some point and stick it somewhere as a downloadable lib-- I haven't seen the author back to the thread, but this code shouldn't disappear.. far too useful, if even just for a light show..