[Bug rtl-optimization/125683] New: possible wrong code due to ce1 combining 2 loads of different incompatible aliasing sets

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125683

            Bug ID: 125683
           Summary: possible wrong code due to ce1 combining 2 loads of
                    different incompatible aliasing sets
           Product: gcc
           Version: 16.0
            Status: UNCONFIRMED
          Keywords: alias, wrong-code
          Severity: normal
          Priority: P3
         Component: rtl-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: pinskia at gcc dot gnu.org
  Target Milestone: ---

Take (for LP64 targets, ILP32 has a similar thing for long and int):
```
long f(int a, void *cc)
{
  long long c;
  if (a)
    c = *(long*)cc;
  else
    c = *(long long*)cc;
  return c;
}
```

Before ce1 we have:
```
(note 9 8 10 3 [bb 3] NOTE_INSN_BASIC_BLOCK)
(insn 10 9 13 3 (set (reg/v:DI 101 [ <retval> ])
        (mem:DI (reg/v/f:DI 103 [ cc ]) [2 MEM[(long int *)cc_4(D)]+0 S8 A64]))
"/app/example.cpp":5:7 106 {*movdi_aarch64}
     (expr_list:REG_DEAD (reg/v/f:DI 103 [ cc ])
        (nil)))

(code_label 13 10 14 4 2 (nil) [1 uses])
(note 14 13 15 4 [bb 4] NOTE_INSN_BASIC_BLOCK)
(insn 15 14 20 4 (set (reg/v:DI 101 [ <retval> ])
        (mem:DI (reg/v/f:DI 103 [ cc ]) [1 MEM[(long long int *)cc_4(D)]+0 S8
A64])) "/app/example.cpp":7:7 106 {*movdi_aarch64}
     (expr_list:REG_DEAD (reg/v/f:DI 103 [ cc ])
        (nil)))

(code_label 20 15 23 5 1 (nil) [0 uses])
```

But afterwards the we get just have:
```
(insn 35 4 21 2 (set (reg/v:DI 101 [ <retval> ])
        (mem:DI (reg/v/f:DI 103 [ cc ]) [1 MEM[(long long int *)cc_4(D)]+0 S8
A64])) "/app/example.cpp":5:7 106 {*movdi_aarch64}
     (nil))
```

Note the two aliasing sets are 1 and 2 both are incompatible and only 1 is
chosen here.

I have not found a testcase which this could cause wrong code but I suspect
there could be one somewhere.

I found this while reviewing
https://gcc.gnu.org/pipermail/gcc-patches/2026-June/719780.html .