With GCC and clang, __m128 is not a struct, so there are no struct members
to access the individual items in the vector.
Use standard load/store function and some hackery to workaround this
limitation.
Lots of bugs/rough edges/etc - issues will be filed.
Old-style debugging still works (just use --emit_source_annotations to get
the helpful movs back and --break_on_instruction will still fire).
