Signed overflow and vectors?

Looking at , I realized that
we are setting nsw flags on vector instructions. To fix that bug, at
the very least, we need to fix *mmintrin.h to not use instructions
with undefined overflow behavior. But generally, do we want to be
setting nsw on vector addition/subtraction/etc. generated by clang?


My thought would be "No?", but I may be suffering from a lack of
imagination where it would be helpful to do that.


I think “no”, we shouldn’t be setting these on vector operations.


Done in r131000.