[Bug tree-optimization/71258] Missed optimizations: dynamic allocation, virtual calls, empty destructors

[Bug tree-optimization/71258] Missed optimizations: dynamic allocation, virtual calls, empty destructors

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71258

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57722
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57722&action=edit
testcase

[Bug tree-optimization/71258] Missed optimizations: dynamic allocation, virtual calls, empty destructors

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71258

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |enhancement

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
  _5 = operator new (16);
  MEM[(struct __as_base  &)_5] ={v} {CLOBBER(bob)};
  MEM[(struct model *)_5].D.38294._vptr.concept = &MEM <int (*) ()[5]> [(void
*)&_ZTVN4poly5modelIiEE + 16B];
  MEM[(struct model *)_5]._data = 40;
  _6 = operator new (16);

  <bb 3> [local count: 1073741824]:
  MEM[(struct __as_base  &)_6] ={v} {CLOBBER(bob)};
  MEM[(struct model *)_6].D.41483._vptr.concept = &MEM <int (*) ()[5]> [(void
*)&_ZTVN4poly5modelIfEE + 16B];
  MEM[(struct model *)_6]._data = 2.0e+0;
  _7 = MEM[(struct concept *)_5]._vptr.concept;

Basically the second `operator new` is treated as being able to touch the
pointer that is returned the first `operator new`.

[Bug tree-optimization/71258] Missed optimizations: dynamic allocation, virtual calls, empty destructors

[phosit at autistici dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71258

Phosit <phosit at autistici dot org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |phosit at autistici dot org

--- Comment #5 from Phosit <phosit at autistici dot org> ---
struct Base{
    virtual ~Base() = default;
};

int main(){
    delete new Base;
    return 0;
}

In this partial test case the destructor is not inlined.
Curiously when changing the function name the destructor is inlined.
I compiled with -O3

[Bug tree-optimization/71258] Missed optimizations: dynamic allocation, virtual calls, empty destructors

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71258

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Phosit from comment #5)
> struct Base{
>     virtual ~Base() = default;
> };
> 
> int main(){
>     delete new Base;
>     return 0;
> }
> 
> In this partial test case the destructor is not inlined.
> Curiously when changing the function name the destructor is inlined.
> I compiled with -O3

Yes main is known to be only called once so heuristics decide inlining is not
profitable. Gcc does this because things called directly from main (without a
loop) is usually not on the hot path of applications.