This set of instructions was only supported by AMD chips starting in
the K6-2 (introduced 1998), and before the "Bulldozer" family
(2011). They were never much used, as they were effectively superseded
by the more-widely-implemented SSE (first implemented on the AMD side
in Athlon XP in 2001).
As part of the general removal of MMX register usage
(https://reviews.llvm.org/D86855), something also needs to be done
with the 3dnow intrinsics. Since there is almost no usage of these,
and no hardware implements them, simple removal of both the intrinsics
and the corresponding __3dNOW__ feature define seems like the best
option.
Support for the underlying LLVM intrinsics remains unchanged, at least
for the moment, although it quite likely makes sense to remove support
on the LLVM side as well.