https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85116
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> --- So loop-header copying now DTRT but then PRE is faced with <bb 4> [local count: 955630223]: # __first_20 = PHI <&testArray._M_elems(2), __first_21(7)> # __first_21 = PHI <&MEM[(void *)&testArray + 8B](2), __first_7(7)> # __result_22 = PHI <&testArray._M_elems(2), __result_15(7)> _8 = MEM[(const double *)__first_20 + 8B]; _10 = *__result_22; _11 = _8 - offset_2(D); _12 = ABS_EXPR <_11>; _13 = _10 - offset_2(D); _14 = ABS_EXPR <_13>; if (_12 < _14) goto <bb 5>; [50.00%] else goto <bb 8>; [50.00%] <bb 5> [local count: 477815111]: goto <bb 3>; [100.00%] <bb 8> [local count: 477815112]: <bb 3> [local count: 955630224]: # __result_15 = PHI <__first_21(5), __result_22(8)> __first_7 = __first_21 + 8; if (__first_7 != &MEM[(void *)&testArray + 8000B]) goto <bb 7>; [89.00%] else goto <bb 6>; [11.00%] <bb 7> [local count: 850510900]: goto <bb 4>; where the situation is too confusing to be recognized. As said, loop-header copying doing a SSA update introduces the extra IV somehow. That needs to be investigated.