[Bug ipa/114703] New: Missed devirtualization in rather simple case

[Bug ipa/114703] New: Missed devirtualization in rather simple case

[konstantin.vladimirov at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114703

            Bug ID: 114703
           Summary: Missed devirtualization in rather simple case
           Product: gcc
           Version: 13.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: ipa
          Assignee: unassigned at gcc dot gnu.org
          Reporter: konstantin.vladimirov at gmail dot com
  Target Milestone: ---

See code here:

https://godbolt.org/z/T9Ma7qe3E

struct A {
  virtual int foo() { return 1; }
};

struct B : A {
  int foo() override { return 2; }
  virtual int foo(int i) { return i + 1; }
};

int use(int, int, int);

int main() {
  B *p = new B;
#if 0
  A* q = p;
#else
  A* q = new B;
#endif  
  int result1 = p->foo(1);
  int result2 = p->foo();
  int result3 = q->foo();
  use(result1, result2, result3);
}

clang devirtualizes just fine, gcc lacks this one.

Interesting that if you flip preprocessor switch, everything works with gcc
either.

[Bug ipa/114703] Missed devirtualization in rather simple case

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114703

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-04-12
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The problem I think is we don't realize that `operator new` can't touch the
other memory that was just allocated partly because the escape analysis is not
flow sensitive .

[Bug ipa/114703] Missed devirtualization in rather simple case

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114703

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

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

[Bug ipa/114703] Missed devirtualization in rather simple case

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114703

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu.org

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Yep, 'new' memory escapes.

[Bug ipa/114703] Missed devirtualization in rather simple case

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114703

--- Comment #3 from Jan Hubicka <hubicka at ucw dot cz> ---
> Yep, 'new' memory escapes.
Yep, this is blocking a lot of propagation in common C++ code.
Here it may help to do speculative devirtualization during IPA stage
that will let the late optimization to get rid of the speculation (since
after inlning we will know that the virtual call does not overwrite
virtual table pointer).  This is technically not too hard to add.  We
can optimistically rule out (some) may aliases while walking the alias
oracle.   I will take a look next stage1.