https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89223
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- The error is clearly in the dataref code, first that it ends up asserting instead of failing analysis and second that it, dependent on context(!) keeps to-be interpreted as "signed" values sizetype constants in (unsigned) sizetype. Papering over this in the gimplifier shouldn't be done. Fixing it "properly" is on my longer TODO list but with low priority give it triggers only in these kind of weird circumstances (and the issue is so old).
