https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111294
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The difference is jump threading:
old (able to remove foo):
```
Registering killing_def (path_oracle) _4
Registering value_relation (path_oracle) (_4 pe8 _33) (root: bb5)
Checking profitability of path (backwards):
[1] Registering jump thread: (5, 7) incoming edge; (7, 9) nocopy;
path: 5->7->9 SUCCESS
```
vs
new (not able to remove foo):
```
Registering killing_def (path_oracle) _33
Registering value_relation (path_oracle) (_33 pe8 _31) (root: bb5)
Checking profitability of path (backwards): FAIL: Jump-thread path not
considered: duplication of 2 insns is needed and optimizing for size.
path: 5->7->xx REJECTED
```
Old IR before jump threading:
```
<bb 5> [local count: 118111600]:
a.9_28 = a;
if (a.9_28 != 0)
goto <bb 6>; [34.00%]
else
goto <bb 7>; [66.00%]
<bb 6> [local count: 40157944]:
_30 = (char) b.1_2;
<bb 7> [local count: 118111600]:
# _33 = PHI <_30(6), 0(5)>
_4 = (unsigned char) _33;
_6 = (int) a.9_28;
j_22 = (short int) _4;
_20 = (int) _4;
if (_4 > 11)
goto <bb 8>; [50.00%]
else
goto <bb 9>; [50.00%]
```
New:
```
<bb 5> [local count: 118111600]:
a.9_28 = a;
if (a.9_28 != 0)
goto <bb 6>; [34.00%]
else
goto <bb 7>; [66.00%]
<bb 6> [local count: 40157944]:
<bb 7> [local count: 118111600]:
# _31 = PHI <b.1_2(6), 0(5)>
_32 = (char) _31;
_4 = (unsigned char) _31;
_6 = (int) a.9_28;
j_22 = (short int) _4;
_33 = _31 & 255;
if (_33 > 11)
goto <bb 8>; [50.00%]
else
goto <bb 9>; [50.00%]
```
But _32 will be 0, _4 will be 0, j_22 will be 0, _33 will be 0.
But I think jump threading does not recognize j_22 will be 0 on that edge and
considers it a copy ... I think.