https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107176
--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> --- Hmm, and interestingly this fix causes PR66375 to regress again: FAIL: gcc.dg/torture/pr66375.c -O2 execution test ... OK, so that's because my suggested fix is exactly an (incomplete) reversal of the fix for PR66375 (r6-1346-gb9b79ba4264cf6) :/