Jeroen Ruigrok/asmodai wrote me to mention that his recent libm changes “included optimised assembly routines for certain mathematical functions for x86 and amd64.”