[Bug optimization/11269] New: [tree-ssa] wrong-code

[Bug optimization/11269] New: [tree-ssa] wrong-code

[reichelt at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269

           Summary: [tree-ssa] wrong-code
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P1
         Component: optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: reichelt@gcc.gnu.org
                CC: dnovillo@gcc.gnu.org,gcc-bugs@gcc.gnu.org
  GCC host triplet: i686-pc-linux-gnu

The following code snippet (inspired by PR6189) should return "1",
but does not, if optimization on the tree-ssa-branch is turned on.

---------------------------------------------------------------
struct A
{
    int  x[2];
    int& GetFirst() { return x[0]; }
};


A foo ()
{
    A a;
    a.GetFirst()=1;
    return a;
}


int main()
{
    A b=foo();
    return b.GetFirst();
}
---------------------------------------------------------------

Without optimization I get the return value "1", with "-O" or "-O2"
I get "192" and with "-O3" I get "0".

This is a regression w.r.t. mainline.

[Bug optimization/11269] [tree-ssa] wrong-code

[bangerth at dealii dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


bangerth at dealii dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4                         |tree-ssa


------- Additional Comments From bangerth at dealii dot org  2003-06-21 01:28 -------
Bulk switch target milestone for all PRs with [tree-ssa] in the summary to
have milestone=tree-ssa.
W.

[Bug optimization/11269] [tree-ssa] wrong-code

[dhazeghi at yahoo dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


dhazeghi at yahoo dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Version|unknown                     |tree-ssa

[Bug optimization/11269] [tree-ssa] wrong-code

[steven at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
            Summary|[tree-ssa] wrong-code       |[tree-ssa] wrong-code


------- Additional Comments From steven at gcc dot gnu dot org  2003-07-06 22:16 -------
I don't see this; with g++-3.5 (GCC) 3.5-tree-ssa 20030704 (merged 20030525) I
always get 1 as the result.  Volkert, do you still have this problem??

[Bug optimization/11269] [tree-ssa] wrong-code

[reichelt at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


reichelt at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|                            |FIXED


------- Additional Comments From reichelt at gcc dot gnu dot org  2003-07-07 11:13 -------
I also don't get the failure any more.
Seems to be fixed.

[Bug optimization/11269] [tree-ssa] wrong-code

[reichelt at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


reichelt at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |UNCONFIRMED
         Resolution|FIXED                       |


------- Additional Comments From reichelt at gcc dot gnu dot org  2003-08-08 09:37 -------
Reappeared in 3.5-tree-ssa 20030808 (merged 20030722). :-(

[Bug optimization/11269] [tree-ssa] wrong-code

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|normal                      |critical
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-08-15 03:10:34
               date|                            |


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-08-15 03:10 -------
I can confirm this on the tree-ssa (20030814).
Looks like the inline is messing this one up:
int main() ()
{
  int & retval.8;
  struct A retval.7;
  struct A * b.3;
  int & T.4;
  int T.5;

  {
    {
      struct A b;

      {
        struct A a;

        {
          {
            struct A a;

            {
              <UL5cb0>:;
            };
            goto <UL5bd0>;;
          }
        };
        <UL5bd0>:;;
        retval.7 = a;
      };
      b = retval.7;
      {
        b.3 = &b;
        {
          struct A * const this;
          int & <UV5ee0>;

          this = (struct A * const)b.3;
          {
            {
              {
                <UV5ee0> = (int &)this;
                goto <UL5e70>;;
              }
            }
          };
          <UL5e70>:;;
          retval.8 = <UV5ee0>;
        };
        T.4 = retval.8;
        T.5 = *T.4;
        if (T.5 != 1)
          {
            {
              abort ()
            }
          }
        else
          {
            (void)0
          };
      };
      return 0;;
    }
  };
  return 0;;
}

There is no setting anything to 1.

Here is the source I used (self checking):
struct A
{
    int  x[2];
    int& GetFirst() { return x[0]; }
};


A foo ()
{
    A a;
    a.GetFirst()=1;
    return a;
}
extern "C" void abort();

int main()
{
    A b=foo();
    if (b.GetFirst()!=1)
        abort();
    return 0;
}

[Bug optimization/11269] [tree-ssa] wrong-code

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269



------- Additional Comments From pinskia at gcc dot gnu dot org  2003-08-20 03:32 -------
Miget get fixed by: <http://gcc.gnu.org/ml/gcc-patches/2003-08/msg01139.html> when it 
gets applied.

[Bug optimization/11269] [tree-ssa] wrong-code

[steven at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269



------- Additional Comments From steven at gcc dot gnu dot org  2003-09-07 13:10 -------
So is this fixed now??  That patch was applied.

[Bug optimization/11269] [tree-ssa] wrong-code

[dnovillo at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269


dnovillo at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jason at gcc dot gnu dot org
                   |dot org                     |
             Status|NEW                         |ASSIGNED


------- Additional Comments From dnovillo at gcc dot gnu dot org  2003-09-14 15:01 -------

This seems to be an FE problem.  The GENERIC and GIMPLE versions of function
foo() show it not returning 'a'.  This causes all pointer dereferences inside
the body to be considered local modifications.  This causes DCE to kill just
about everything in the body.

A foo() ()
{
  int & retval.3;
  struct A * a.1;
  int & T.2;

  {
    struct A a;

    a.1 = &a;
    {
      struct A * const this;
      int & <U8930>;

      this = (struct A * const)a.1;
      {
        {
          {
            <U8930> = (int *)this;
            goto <U88c0>;;
          }
        }
      };
      <U88c0>:;;
      retval.3 = <U8930>;
    };
    T.2 = retval.3;
    *T.2 = 1;
    return;;     <-- This should be 'return *T.2' or something similar
  }
}

[Bug optimization/11269] [tree-ssa] wrong-code

[steven at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269



------- Additional Comments From steven at gcc dot gnu dot org  2003-09-14 20:22 -------
Diego,

Funny you looked at this just today.  I did so as well and also came to the
conclusion that this has to be a front end problem.  I compiled the test case
with -fdump-tree-{original,inlined,gimple}.  The "original" dump already shows
the problem you also identified:


;; Function int& A::GetFirst() (_ZN1A8GetFirstEv)
;; enabled by -tree-original


{
   {
      return <return-value> = (this);
   }
}



;; Function A foo() (_Z3foov)
;; enabled by -tree-original


{
   {
      struct A a;
      (void)(*GetFirst((&a))= 1);
      return;  // return *WHAT* ??? This was "return a;" in the test case.
   }
}



;; Function int main() (main)
;; enabled by -tree-original


{
   {
      struct A b;
      ({
         (void)(b = (__builtin_memcpy(&<Udee0>, & foo()));
      });
      if (*GetFirst((&b)) != 1)
         {
            abort();
         }
      return <return-value> = 0;
   }
}

[Bug optimization/11269] [tree-ssa] wrong-code

[jason at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269



------- Additional Comments From jason at gcc dot gnu dot org  2003-10-22 02:35 -------
Hmm, the front end should be calling set_has_hidden_use on 'a', which should
prevent it from being deleted.  Perhaps it doesn't work for BLKmode types?

[Bug optimization/11269] [tree-ssa] wrong-code

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2003-11-17 00:31 -------
Subject: Bug 11269

CVSROOT:	/cvs/gcc
Module name:	gcc
Branch: 	tree-ssa-20020619-branch
Changes by:	jason@gcc.gnu.org	2003-11-17 00:31:03

Modified files:
	gcc            : ChangeLog.tree-ssa dwarf2out.c tree-inline.c 
	                 c-common.c gimplify.c 
	gcc/cp         : ChangeLog.tree-ssa semantics.c decl.c tree.c 
	                 cp-tree.h 

Log message:
	PR optimization/11269
	* semantics.c (finalize_nrv_r): Rename from nullify_returns_r.
	Also replace uses of the nrv with our RESULT_DECL.
	(cxx_expand_function_start): Don't mess with the nrv.
	(finalize_nrv): New fn.
	* cp-tree.h: Declare it.
	* decl.c (finish_function): Call it.
	* tree.c (cp_copy_res_decl_for_inlining): Don't mess with the nrv.
	* dwarf2out.c (gen_subprogram_die): Generate a DIE for a named
	return value.
	(loc_descriptor_from_tree): Treat RESULT_DECL like VAR_DECL.
	(add_location_or_const_value_attribute): Likewise.
	(add_bound_info): Likewise.
	(gen_decl_die): Likewise.
	
	* tree-inline.c (walk_tree): Don't walk into the BIND_EXPR_VARS
	of a BIND_EXPR.
	* c-common.c (c_walk_subtrees): Don't walk into the decl of a
	DECL_STMT.
	
	* gimplify.c (mostly_copy_tree_r): Don't walk into a BLOCK.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.tree-ssa.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.1.2.874&r2=1.1.2.875
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/dwarf2out.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.379.2.30&r2=1.379.2.31
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-inline.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.26.2.61&r2=1.26.2.62
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/c-common.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.344.2.50&r2=1.344.2.51
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/gimplify.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.1.2.115&r2=1.1.2.116
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/ChangeLog.tree-ssa.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.1.2.51&r2=1.1.2.52
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/semantics.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.264.2.39&r2=1.264.2.40
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/decl.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.911.2.55&r2=1.911.2.56
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/tree.c.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.286.2.37&r2=1.286.2.38
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/cp/cp-tree.h.diff?cvsroot=gcc&only_with_tag=tree-ssa-20020619-branch&r1=1.719.2.41&r2=1.719.2.42



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269

[Bug optimization/11269] [tree-ssa] wrong-code

[jason at gcc dot gnu dot org]

------- Additional Comments From jason at gcc dot gnu dot org  2003-11-17 00:34 -------
.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11269