[Bug tree-optimization/125250] [15/16/17 Regression] wrong code at -O1/2/3

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125250

--- Comment #5 from Drea Pinski <pinskia at gcc dot gnu.org> ---
So what happens in GCC 15, is jump threading we have:
```
  a5__lsm.14_18 = MEM[(_Bool *)&a5];
...
  <bb 4> [local count: 114863531]:
  # g9_lsm.12_29 = PHI <_7(3), g9_lsm.12_10(2)>
  # g9_lsm_flag.13_14 = PHI <1(3), 0(2)>
  # a5__lsm.14_27 = PHI <0(3), a5__lsm.14_18(2)>
  MEM[(_Bool *)&a5] = a5__lsm.14_27;
  if (g9_lsm_flag.13_14 != 0)
...
```

And then jump threading does:
```
  a5__lsm.14_18 = MEM[(_Bool *)&a5];
...
  <bb 4> [local count: 6317494]:
  # g9_lsm.12_20 = PHI <g9_lsm.12_10(2)>
  # g9_lsm_flag.13_12 = PHI <0(2)>
  # a5__lsm.14_22 = PHI <a5__lsm.14_18(2)>
  MEM[(_Bool *)&a5] = a5__lsm.14_22;
  goto <bb 7>; [100.00%]
```

And then DOM notices the store to MEM[(_Bool *)&a5] can be removed as it is the
same value.

But that is undefined behavior even back in GCC 15; just it was happened to
remove then.

GCC 16 changes the first statement to:
```
  _11 = MEM[(_Bool *)&a5];
  a5__lsm.14_21 = (_Bool) _11;
```

Which does not detect that they are the same value.

What happened in GCC 14 and before was instead:

  a5__lsm.14_24 = _23(D);
  a5__lsm_flag.15_25 = 0;

And then conditional store to a5.