https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127793
--- Comment #2 from Trevor Gross <tg at trevorgross dot com> --- This one has some specific motivation because, as I understand it, it's gotten in the way of distros migrating from the AIX long double to IEEE long double. Since altivec is typically in the BE baseline but VSX is not. It's slower out of the box but I believe the libgcc routines are already using ifunc, so it will wind up accelerated if you're on hardware that would support manual -mcpu=power9 anyway.
