faster FFT for dsPIC
Here (attached) for everyone's use is a speedier version of the FFTComplexIP function from the Microchip DSP library. I've tested it only on the dsPIC30F6014 and 6014A, and the code is safe from the YMEM errata of the early 6014. The code is longer, some of the speedup is achieved by computing the last two stages of the decimation-in-frequency FFT without using twiddle factors since there are only trivial multiplications in these stages. Most of the speedup comes from reducing the 16-cycle inner loop of the original code to 13 cycles. This function does not give the _exact_ result the DSP library one does only because most of the trivial multiplications by +1 do not suffer from a rounding error in the twiddle factor in this version.
My apologies for the sparse comments in the code. Shoot me an email if you can make use of it.