The optimization a + (-0.0f) -> a was being misapplied to a + (+0.0f) in the vector...

 The optimization a + (-0.0f) -> a was being misapplied to a + (+0.0f) in the vector case (because
we weren't differntiating floating-point zeroinitializers from other zero-initializers)
which was causing problems for code relying upon a + (+0.0f) to, eg, flush denormals to
0. Make the scalar and vector cases have the same behaviour.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177279 91177308-0d34-0410-b5e6-96231b3b80d8