[Bug tree-optimization/58570] New: wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[Bug tree-optimization/58570] New: wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[su at cs dot ucdavis.edu]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

            Bug ID: 58570
           Summary: wrong code at -Os and above on x86_64-linux-gnu (both
                    32-bit and 64-bit modes)
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: su at cs dot ucdavis.edu

The current gcc trunk mis-compiles the following code on x86_64-linux at -Os
and above in both 32-bit and 64-bit modes.

This is a regression from 4.8.x.

$ gcc-trunk -v
Using built-in specs.
COLLECT_GCC=gcc-trunk
COLLECT_LTO_WRAPPER=/usr/local/gcc-trunk/libexec/gcc/x86_64-unknown-linux-gnu/4.9.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-trunk/configure
--enable-languages=c,c++,objc,obj-c++,fortran,lto --disable-werror
--enable-checking=release --with-gmp=/usr/local/gcc-trunk
--with-mpfr=/usr/local/gcc-trunk --with-mpc=/usr/local/gcc-trunk
--with-cloog=/usr/local/gcc-trunk --prefix=/usr/local/gcc-trunk
Thread model: posix
gcc version 4.9.0 20130928 (experimental) [trunk revision 203003] (GCC) 
$ 
$ gcc-trunk -O1 small.c
$ a.out
1
$ gcc-4.8 -O3 small.c
$ a.out
1
$ gcc-trunk -O3 small.c
$ a.out
0
$ gcc-trunk -O2 small.c
$ a.out
0
$ gcc-trunk -Os small.c
$ a.out
0
$



------------------------------------


int printf (const char *, ...);

#pragma pack(1)
struct S
{
  int f0:15;
  int f1:29;
};

int a, b, c, e = 1, g;
static struct S d[6];
long long f;

static short
foo (int p)
{
  for (b = 0; b < 1; b++)
    {
      g |= 1;
      for (; a < 1; a++)
    {
      int i;
      if (p < c - 1)
        return 4;
      if (p)
        for (i = 0; i < 6; i++)
          {
        struct S j = {1, 1};
        d[i] = j;
          }
      continue;
    }
    }
  return 0;
}

int
main ()
{
  if (e)
    foo (!f);
  printf ("%d\n", d[0].f1);
  return 0;
}

[Bug tree-optimization/58570] [4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.9.0
            Summary|wrong code at -Os and above |[4.9 Regression] wrong code
                   |on x86_64-linux-gnu (both   |at -Os and above on
                   |32-bit and 64-bit modes)    |x86_64-linux-gnu (both
                   |                            |32-bit and 64-bit modes)

[Bug tree-optimization/58570] [4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

Bernd Edlinger <bernd.edlinger at hotmail dot de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bernd.edlinger at hotmail dot de

--- Comment #1 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
Created attachment 30961
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30961&action=edit
reduced test case

the code generation is not aware that d[i].f0 and d[i].f1 overlap one byte.

[Bug tree-optimization/58570] [4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #2 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
Created attachment 30962
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30962&action=edit
for a possible fix

comments?

[Bug tree-optimization/58570] [4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[mikpelinux at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

Mikael Pettersson <mikpelinux at gmail dot com> changed:

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

--- Comment #3 from Mikael Pettersson <mikpelinux at gmail dot com> ---
Started with r198024.  Author CC:d.

[Bug tree-optimization/58570] [4.9 Regression] wrong code at -Os and above on x86_64-linux-gnu (both 32-bit and 64-bit modes)

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-10-07
     Ever confirmed|0                           |1

--- Comment #4 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Ouch.  The tree-ssa-alias.c changes were not intended to affect alias.c, but
they of course do nowadays...

[Bug tree-optimization/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|bernd.edlinger at hotmail dot de   |
           Assignee|unassigned at gcc dot gnu.org      |ebotcazou at gcc dot gnu.org
            Summary|[4.9 Regression] wrong code |[4.9 Regression] wrong code
                   |at -Os and above on         |for bitfields at -O2 and
                   |x86_64-linux-gnu (both      |above
                   |32-bit and 64-bit modes)    |

--- Comment #5 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
I think we just want to copy the following from
nonoverlapping_component_refs_p:

      /* If we're left with accessing different fields of a structure, then no
     possible overlap, unless they are both bitfields.  */
      if (TREE_CODE (typex) == RECORD_TYPE && fieldx != fieldy)
    return !(DECL_BIT_FIELD (fieldx) && DECL_BIT_FIELD (fieldy));

over to nonoverlapping_component_refs_of_decl_p.

[Bug tree-optimization/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[bernd.edlinger at hotmail dot de]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="UTF-8", Size: 3039 bytes --]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #6 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
(In reply to Eric Botcazou from comment #5)
> I think we just want to copy the following from
> nonoverlapping_component_refs_p:
> 
>       /* If we're left with accessing different fields of a structure, then
> no
> 	 possible overlap, unless they are both bitfields.  */
>       if (TREE_CODE (typex) == RECORD_TYPE && fieldx != fieldy)
> 	return !(DECL_BIT_FIELD (fieldx) && DECL_BIT_FIELD (fieldy));
> 
> over to nonoverlapping_component_refs_of_decl_p.

Yes.
Just one idea:

What if both bit fields have different DECL_BIT_FIELD_REPRESENTATIVE?

Then they can't possibly overlap?
>From gcc-bugs-return-431289-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Tue Oct 08 07:15:32 2013
Return-Path: <gcc-bugs-return-431289-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 28442 invoked by alias); 8 Oct 2013 07:15:31 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 28388 invoked by uid 48); 8 Oct 2013 07:15:27 -0000
From: "jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug libgomp/58642] gomp regression: not "honoring" anymore task set and numactl
Date: Tue, 08 Oct 2013 07:15:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: libgomp
X-Bugzilla-Version: 4.9.0
X-Bugzilla-Keywords:
X-Bugzilla-Severity: normal
X-Bugzilla-Who: jakub at gcc dot gnu.org
X-Bugzilla-Status: ASSIGNED
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: jakub at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: attachments.created
Message-ID: <bug-58642-4-OWByswFP8A@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-58642-4@http.gcc.gnu.org/bugzilla/>
References: <bug-58642-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2013-10/txt/msg00433.txt.bz2
Content-length: 510

http://gcc.gnu.org/bugzilla/show_bug.cgi?idX642

--- Comment #29 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 30967
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id0967&actioníit
Y

Ah, thanks, I can see where the failing sched_getaffinity calls are coming
from, hopefully this patch should fix that.

That doesn't explain the libgomp errors you were getting though (if you can
still reproduce them).
To track that perhaps we could instrument libgomp to tell us more details.

[Bug tree-optimization/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #7 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> What if both bit fields have different DECL_BIT_FIELD_REPRESENTATIVE?
> 
> Then they can't possibly overlap?

Probably, yes, that could be a nice enhancement.

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|tree-optimization           |middle-end

--- Comment #8 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Recategorizing.

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #9 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
Eric,

there is one more thing to consider for your proposed patch,
that is the damned -fstrict-volatile-bitfields:

if strict_volatile_bitfields>0 and the BIT_FIELD access
is _volatile_ it does not respect the BIT_FIELD_REPRESENTATIVE at all.

that means for instance:
struct
{
  int a:24;
  char b;
} s;

s.a=1;
s.b=2;

we have a INT32 read-modify-write over b.

even worse:
struct
{
  int a:24;
} __attribute__((packed)) s;

s.a=1;

uses an (unaligned) INT32 read-modify-write and may overwrite
one member of an adjacent structure.

Note that at least the second example should no longer write beyond
the structure if this patch is applied:
 http://gcc.gnu.org/ml/gcc-patches/2013-09/msg02058.html

So you should expect an alias if either field is a VOLATILE
bit-field access and flag_strict_volatile_bitfields > 0.

Bernd.

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #10 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
> there is one more thing to consider for your proposed patch,
> that is the damned -fstrict-volatile-bitfields:
> 
> if strict_volatile_bitfields>0 and the BIT_FIELD access
> is _volatile_ it does not respect the BIT_FIELD_REPRESENTATIVE at all.

My patch as written doesn't use BIT_FIELD_REPRESENTATIVE so it isn't affected.

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[bernd.edlinger at hotmail dot de]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #11 from Bernd Edlinger <bernd.edlinger at hotmail dot de> ---
(In reply to Eric Botcazou from comment #10)
> > there is one more thing to consider for your proposed patch,
> > that is the damned -fstrict-volatile-bitfields:
> > 
> > if strict_volatile_bitfields>0 and the BIT_FIELD access
> > is _volatile_ it does not respect the BIT_FIELD_REPRESENTATIVE at all.
>
> My patch as written doesn't use BIT_FIELD_REPRESENTATIVE so it isn't affected.

No. You only assume an alias if _both_ fields are bit fields.
But in my example only one "a" is a volatile bit field the other
is a normal member "b".

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #13 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Author: ebotcazou
Date: Wed Oct  9 12:59:02 2013
New Revision: 203315

URL: http://gcc.gnu.org/viewcvs?rev=203315&root=gcc&view=rev
Log:
    PR middle-end/58570
    * tree-ssa-alias.c (nonoverlapping_component_refs_of_decl_p): Return
    false if both components are bitfields.

Added:
    trunk/gcc/testsuite/gcc.c-torture/execute/pr58570.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-ssa-alias.c

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[ebotcazou at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

Eric Botcazou <ebotcazou at gcc dot gnu.org> changed:

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

--- Comment #14 from Eric Botcazou <ebotcazou at gcc dot gnu.org> ---
Thanks for reporting the problem.

[Bug middle-end/58570] [4.9 Regression] wrong code for bitfields at -O2 and above

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58570

--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
*** Bug 58677 has been marked as a duplicate of this bug. ***