https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118684
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to Andreas Schwab from comment #1) > Does that also happen if you remove the undefined behaviour? The code generation still happens when I make it conditionally executed. It's the "easiest" case I currently have that triggers a stack temp dropping alignment, but given how assign_temp works it might not be the only one. Note the vectorizer does create aligned out-of-bound accesses if the first lane is known to be inside an object, so while the source is technically UB on the GIMPLE side we should not consider this UB. This was split out from PR117424 where it's the vectorizers doing.
