This eliminates extra extract instructions when loading an i8 vector to a float vector. llvm-svn: 210666