Tweak *mmintrin.h so that they don't make any bad assumptions about alignment (which probably has little effect in practice, but better to get it right).  Make the load in _mm_loadh_pi and _mm_loadl_pi a single LLVM IR instruction to make optimizing easier for CodeGen.

rdar://10054986



git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@139874 91177308-0d34-0410-b5e6-96231b3b80d8