When optimzing a mul by immediate into two, the resulting mul's should get a...

When optimzing a mul by immediate into two, the resulting mul's should get a x86 specific node to avoid dag combiner from hacking on them further.

llvm-svn: 68066