Many implementation of the library goes deep down to FPATAN instuction for all arc-functions. How is FPATAN implemented? Assuming that we have 1 bit sign, M bits mantissa and N bits exponent, what is the algorithm to get the arctangent of this number? There should be such algorithm, since the FPU does it.

# Best How To :

Trigonometric functions do have pretty ugly implementations that are hacky and do lots of bit fiddling. I think it will be pretty hard to find someone here that is able to explain an algorithm that is actually used.

Here is an atan2 implementation: https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/ieee754/dbl-64/e_atan2.c;h=a287ca6656b210c77367eec3c46d72f18476d61d;hb=HEAD

Edit: Actually I found this one: http://www.netlib.org/fdlibm/e_atan2.c which is a lot easier to follow, but probably slower because of that (?).

The FPU does all this in some circuits so the CPU doesn't have to do all this work.