https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93674
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |amker at gcc dot gnu.org
--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Anyway, it seems the bug is somewhere in ivopts, before ivopts we have:
# j_18 = PHI <3(3), j_9(8)>
# ivtmp_14 = PHI <4(3), ivtmp_13(8)>
_1 = i_17 + j_18;
_2 = (some_enum) _1;
sink (_2);
j_9 = j_18 + -1;
ivtmp_13 = ivtmp_14 - 1;
where _2 is the only SSA_NAME with some_enum type.
ivopts starts using that type everywhere:
some_enum ivtmp.19;
some_enum ivtmp.9;
int j;
int i;
some_enum _2;
some_enum _23;
unsigned int _24;
int _25;
<bb 2> [local count: 71582789]:
<bb 3> [local count: 214748369]:
# ivtmp.19_21 = PHI <4294967295(2), ivtmp.19_22(7)>
_24 = ivtmp.19_21 + 1;
_25 = (int) _24;
i_17 = _25;
_23 = ivtmp.19_21 + 4;
ivtmp.9_3 = _23;
<bb 4> [local count: 858993457]:
# ivtmp.9_5 = PHI <ivtmp.9_3(3), ivtmp.9_4(8)>
_2 = ivtmp.9_5;
sink (_2);
ivtmp.9_4 = ivtmp.9_5 - 1;
if (ivtmp.9_4 == ivtmp.19_21)
goto <bb 5>; [25.00%]
else
goto <bb 8>; [75.00%]
<bb 8> [local count: 644245086]:
goto <bb 4>; [100.00%]
<bb 5> [local count: 214748369]:
ivtmp.19_22 = ivtmp.19_21 + 1;
if (ivtmp.19_22 == 2)
goto <bb 6>; [33.33%]
else
goto <bb 7>; [66.67%]
