https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122083
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Assignee|unassigned at gcc dot gnu.org |pinskia at gcc dot
gnu.org
CC| |pinskia at gcc dot gnu.org
Last reconfirmed| |2025-10-03
Status|UNCONFIRMED |ASSIGNED
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the difference between GCC 14 and 15 is that GCC 14 would early inline:
D.8563 = min_cmp<long int> (_2, _1);
But GCC 15 does not.
The difference is:
MEM <char[8]> [(struct Res *)&D.8662 + 8B] = {};
D.8662.M_min = y_5(D);
D.8662.M_cmp._M_value = 1;
vs
D.8854.M_min = y_5(D);
D.8854.M_cmp._M_value = 1;
So zeroing the padding.
And the warning happens after inlining but before we can remove the (undefined)
memcmp call from being dead.
I wonder if we could optimize this:
```
if (c$_M_value_21 == 1)
goto <bb 6>; [INV]
else
goto <bb 7>; [INV]
<bb 6> :
MEM <char[8]> [(struct Res *)&D.8662 + 8B] = {};
D.8662.M_min = y_5(D);
D.8662.M_cmp._M_value = 1;
goto <bb 8>; [INV]
<bb 7> :
MEM <char[8]> [(struct Res *)&D.8662 + 8B] = {};
D.8662.M_min = x_3(D);
D.8662.M_cmp._M_value = c$_M_value_21;
<bb 8> :
```
Into:
if (c$_M_value_21 == 1)
goto <bb 6>; [INV]
else
goto <bb 7>; [INV]
<bb 6> :
MEM <char[8]> [(struct Res *)&D.8662 + 8B] = {};
goto <bb 8>; [INV]
<bb 7> :
MEM <char[8]> [(struct Res *)&D.8662 + 8B] = {};
<bb 8> :
_9 = PHI<x_3(D)(7), y_5(D)(6)>
D.8662.M_min = _9;
D.8662.M_cmp._M_value = c$_M_value_21;
In phiopt1, it might be better.
Sink is able to do it too:
/app/example.cpp:13:17: optimized: sinking common stores to
D.8662.M_cmp._M_value
/app/example.cpp:13:17: optimized: sinking common stores to D.8662.M_min
/app/example.cpp:13:17: optimized: sinking common stores with same value to MEM
<char[7]> [(struct Res *)&D.8662 + 9B]
But cselim does not know how to handle the {} I think.
In this case cselim should be able to handle it even without DR. Let me see if
I can improve cselim/cselim light in phiopt1 to handle it.
Because I think that will fixup the early inlining issue.
