[Bug middle-end/99782] [13/14/15/16 Regression] Compile time and memory hog w/ __int128 on aarch64

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99782

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the problem is we start with the outer expression and work our way down and
that becomes a O(n^2) because we fail for each side but we don't save around
the expansion of the operands.

This is similar to PR 69619 which was was "fixed" and there was supposed to be
a bug filed by Wilco on improving the ccmp interface but that never happened.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69619#c7