Now that we have vector support for [US](ADD|SUB)O we no longer need to scalarize when expanding [US](ADD|SUB)SAT. This matches what the cost model already does. Differential Revision: https://reviews.llvm.org/D57348 llvm-svn: 353651