Stopped supported -0 as a float output because (a) it is non-portable, only existing on some architectures and (b) relies on the undefined behavior of 1.0/0.0