Merge 99c457dc2ef395872d7448c85609f6cb73a7f89b into 18.x #82903
Description
Unbreak *tf builtins for hexfloat (#82208)
This re-lands cc0065a in a way that
keeps existing targets working.
Original commit message:
#68132 ended up removing
__multc3 & __divtc3 from compiler-rt library builds that have
QUAD_PRECISION but not TF_MODE due to missing int128 support.
I added support for QUAD_PRECISION to use the native hex float long double representation.