Added X86 SSE2 intrinsics which can be represented as vector_shuffles. This is

a temporary workaround for the 2-wide vector_shuffle problem (i.e. its mask
would have type v2i32 which is not legal).

llvm-svn: 27964