This is the payoff for D31156 - if a target has efficient comparison instructions for vector-sized equality,
we can replace memcmp calls with inline code that is both smaller and faster.
Differential Revision: https://reviews.llvm.org/D31290
llvm-svn: 298775