Mathematical functions

It seems the techniques for generating math functions that are correctly rounded (to within 0.5 ulps) for all valid inputs, and are decently fast, is a relatively new thing. And then there are all the new floating point functions (new functions, as well as old functions for new floating point formats), handling rounding modes, and whatnot. Then add in SIMD implementations (libmvec) etc etc., and it's just a huge amount of work.