https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120894
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|middle-end |ipa
Status|UNCONFIRMED |NEW
Ever confirmed|0 |1
Last reconfirmed| |2025-06-30
Severity|normal |enhancement
--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This seems related to how std::function optimizing. Basically we don't
devirutalize things until late .
Before inlining we have:
```
_1 = std::pmr::get_default_resource ();
res ={v} {CLOBBER(bob)};
res.D.43232._vptr.memory_resource = &MEM <int (*) ()[7]> [(void
*)&_ZTVN12_GLOBAL__N_114StaticResourceE + 16B];
res.buffer = &buf[0];
res.buffer_left = 16;
res.upstream = _1;
MEM[(struct polymorphic_allocator *)&D.44315] ={v} {CLOBBER(bob)};
MEM[(struct polymorphic_allocator *)&D.44315]._M_resource = &res.D.43232;
std::vector<std::byte, std::pmr::polymorphic_allocator<std::byte> >::vector
(&vec, 16, &D.44315);
```
And that function was:
```
MEM[(struct _Vector_impl *)this_2(D)].D.43600 = MEM[(const struct
polymorphic_allocator &)__a_5(D)];
MEM[(struct _Vector_impl_data *)this_2(D) + 8B] ={v} {CLOBBER(bob)};
MEM[(struct _Vector_impl_data *)this_2(D) + 8B]._M_start = 0B;
MEM[(struct _Vector_impl_data *)this_2(D) + 8B]._M_finish = 0B;
MEM[(struct _Vector_impl_data *)this_2(D) + 8B]._M_end_of_storage = 0B;
if (__n_4(D) != 0)
goto <bb 5>; [71.00%]
else
goto <bb 6>; [29.00%]
<bb 5> [local count: 83859236]:
_10 = MEM[(struct polymorphic_allocator *)this_2(D)]._M_resource;
_11 = *_10._vptr.memory_resource;
_12 = MEM[(int (*) () *)_11 + 16B];
_13 = OBJ_TYPE_REF(_12;(struct memory_resource)_10->2B) (_10, __n_4(D), 1);
```
MEM[(struct polymorphic_allocator *)this_2(D)]._M_resource
could be optimize to just :
MEM[(const struct polymorphic_allocator &)__a_5(D)]._M_resource I think.
That might improve things slightly but not enough though.
