https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103037
--- Comment #6 from rguenther at suse dot de <rguenther at suse dot de> --- On Tue, 2 Nov 2021, jakub at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103037 > > --- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> --- > We have: > var_3.2_4 = var_3; > iftmp.1_14 = (int) var_3.2_4; > var_11_lsm.14_8 = _7(D); > > <bb 3> [local count: 955630225]: > # RANGE [0, 24] NONZERO 31 > # a_20 = PHI <a_18(10), 0(9)> > _19 = MEM <short unsigned int[23]> [(short unsigned int &)&arr_4][a_20]; > if (_19 > 1) > goto <bb 5>; [50.00%] > else > goto <bb 4>; [50.00%] > > <bb 4> [local count: 477815112]: > # RANGE [0, 1] NONZERO 1 > _3 = (int) _19; > if (_19 != 0) > goto <bb 5>; [20.00%] > else > goto <bb 6>; [80.00%] > > <bb 5> [local count: 477815112]: > # RANGE [0, 1] NONZERO 1 > # _24 = PHI <_3(4), 1(3)> > # RANGE [1, 65535] NONZERO 65535 > iftmp.1_15 = (int) _19; > goto <bb 7>; [100.00%] > > <bb 6> [local count: 477815112]: > > <bb 7> [local count: 955630225]: > # RANGE [-32768, 65535] > # iftmp.1_10 = PHI <iftmp.1_15(5), iftmp.1_14(6)> > # RANGE [0, 1] NONZERO 1 > # _25 = PHI <_24(5), 0(6)> > # RANGE [-1, 1] > _5 = _25 / iftmp.1_10; > and seems PRE is trying top simplify 1 / _3 and _3 / _3. _3 has correctly > range of [0, 1] - in that bb _19 can't be anything but 0 or 1 given the > condition, but > iftmp.1_15 = (int) _19; isn't equivalent to that, because it is reachable from > other bbs, even when it is also (int) _19. > So, does PRE need to temporarily reset_flow_sensitive_info in such cases and > restore if it didn't succeed? No, we have means to avoid this situation but somehow it doesn't work. Give me some time to look into this.
