[Bug tree-optimization/109945] Escape analysis hates copy elision: different result with -O1 vs -O2

["rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109945

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c++                         |tree-optimization
   Last reconfirmed|                            |2023-05-23
                 CC|                            |jakub at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
     Ever confirmed|0                           |1
            Version|unknown                     |14.0

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
ESCAPED, points-to vars: { }

intD.9 mainD.2827 ()
{
  intD.9 w$iD.2845;
  intD.9 iD.2831;
  struct WidgetD.2773 wD.2829;
  intD.9 _1; 
  boolD.2740 _2;
  intD.9 _7; 

  <bb 2> :
  # USE = nonlocal  
  # CLB = nonlocal escaped
  wD.2829 = makeD.2813 (); [return slot optimization]
  w$i_9 = wD.2829.iD.2778;
  i_5 = w$i_9;
  # USE = nonlocal escaped
  # CLB = nonlocal escaped
  gD.2825 ();
  _1 = w$i_9;
  _2 = _1 == i_5;
  _7 = (intD.9) _2;
  wD.2829 ={v} {CLOBBER(eol)};
  return _7;

}

so indeed GCC points-to doesn't consider

  wD.2829 = makeD.2813 (); [return slot optimization]

as escape point for the return.  We have

  /* And if we applied NRV the address of the return slot escapes as well.  */
  if (gimple_call_return_slot_opt_p (stmt)
      && gimple_call_lhs (stmt) != NULL_TREE
      && TREE_ADDRESSABLE (TREE_TYPE (gimple_call_lhs (stmt))))
    {

but in this case the type is not TREE_ADDRESSABLE.  I'm not sure why this
check is here.  Removing it fixes the issue.  A !is_gimple_reg (lhs)
check may be a substitute in case we can get stale return-slot-opt flags?