help about fft code.

I'm not familiar with this specific piece of code, but it looks like a common technique for fixed-point arithmetic.

Rather than using floating point, which is very expensive in gate count, values between zero and one (like the sine and cosine) are expressed as fractions. For convenience, all fractions have powers of two in the denominator. Also, since the denominator is the same for all fractions, adding and subtracting can just ignore it and use only the numerator.

In floating point, that code would be:

tmp = cos * data + sinreal * data;

Using fractions instead of floating point:

tmp = (cosNumer/denom) * data + (sinnumer/denom) * data;

or

tmp = (cosNumer * data + sinnumer * data ) / denom;

Since denom is a power of two of the form (1 << SCALE),

tmp = (cosNumer * data + sinnumer * data ) >> SCALE;