[Bug tree-optimization/30252] [4.2 regression] miscompilation of sigc++-2.0 based code with -fstrict-aliasing

[rguenth at gcc dot gnu dot org]

------- Comment #17 from rguenth at gcc dot gnu dot org  2007-05-01 14:00 -------
Without -fstrict-aliasing we mark this_20->functor_ = D.2946; obviously
necessary
because it is is_hidden_global_store ():

 ;; Function sigc::slot0 A::bar() (_ZN1A3barEv)

-Marking useful stmt: this_20->functor_ = D.2946;
-
-Marking useful stmt: D.3047_33 = operator new (20);
+Marking useful stmt: D.3054_33 = operator new (20);

 Marking useful stmt: this_35->functor_ = D.2514;

-Marking useful stmt: D.3053_46->call_ = call_it;
+Marking useful stmt: D.3060_46->call_ = call_it;

 Marking useful stmt: return <retval>;

without that we DCE too much as

  #   SFT.80_6 = V_MUST_DEF <SFT.80_5>;
  D.2378.func_ptr_ = foo;
...
  #   D.2951_27 = V_MUST_DEF <D.2951_26>;
  #   VUSE <SFT.80_6>;
  D.2951 = D.2378;
  #   SFT.77_50 = V_MAY_DEF <SFT.77_49>;
  #   VUSE <D.2951_27>;
  this_20->functor_ = D.2951;
  #   SFT.77_63 = V_MUST_DEF <SFT.77_50>;
  D.2514.bound1_.visited_ = D.2953_31;

the last store kills the assignment to this_20->functor_.  Note that while
we have uses for D.2951_27 and SFT.80_6 the structure assignment to
this_20->functor_ for some reason shares the SFT with the assignment
to D.2514.bound1_.visited_.  So this may be as well a frontend bug.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |NEW
     Ever Confirmed|0                           |1
           Keywords|                            |alias
   Last reconfirmed|0000-00-00 00:00:00         |2007-05-01 14:00:47
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30252