https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126997
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> --- So we again have an invariant conversion which was pattern matched to a widen-sum reduction. I believe we should work towards avoiding this situation. It's again unswitching that exposes this. So I'm fixing this one by ensuring to hoist invariants after unswitching.
