[Bug middle-end/20030] New: Broken arguments access

[Bug middle-end/20030] New: Broken arguments access

[grigory dot zagorodnev at intel dot com]

Broken arguments access

Starting February 13th, gfrotran fails on the following test case on 
ia32/ia64/x86_64 platforms with all kinds of optimizations: -O0, -O1, -O2 etc. 
The issue is critical because it affects SPEC cpu2k tests wupwise and galgel on 
all platforms. 

$ cat foo.f
      program foo
      character*1 a1, a2, b
      a1='A'
      a2='A'
      b='B'
      x = LSAME(a1,a2)
      if ( x.eq.1 ) then
        write(*,*) 'passsed'
      else
        write(*,*) 'failed'
      endif
      end
 
      logical function LSAME( CA, CB )
      character CA, CB
      integer   INTA, INTB
      
      INTA = ICHAR( CA )
      INTB = ICHAR( CB )
      
      write(*,*) INTA, INTB
      
      LSAME = INTA.EQ.INTB
      end


The reason of failure is the broken argument access for the function. Here is 
the ia32 Linux code obtained with -O0 optimization.
        .type   lsame_, @function
lsame_:
        pushl   %ebp
        movl    %esp, %ebp
        subl    $16, %esp
        movl    8(%ebp), %eax
        movb    -1(%eax), %al		<-- Wrong. Must be "movb (%eax), %al"
        movzbl  %al, %eax
        movl    %eax, -8(%ebp)
        movl    12(%ebp), %eax
        movb    -1(%eax), %al		<-- Wrong. Must be "movb (%eax), %al"
        movzbl  %al, %eax

This issue has been discussed here http://gcc.gnu.org/ml/gcc/2005-
02/msg00685.html but not resolved yet and bug search gave me nothing. It 
appears that breakage introduced by revision 2.110 (and later 2.112) of 
gcc/gimplify.c. The exact place is:

$ cvs diff -r 2.111 -r 2.112 
gcc/gimplify.c                                                              
3797a3799,3803
>       case INDIRECT_REF:
>         *expr_p = fold_indirect_ref (*expr_p);
>         if (*expr_p != save_expr)
>           break;
>         /* else fall through.  */
3800d3805
<       case INDIRECT_REF:

-- 
           Summary: Broken arguments access
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: middle-end
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: grigory dot zagorodnev at intel dot com
                CC: gcc-bugs at gcc dot gnu dot org
  GCC host triplet: i686-redhat-linux


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] Broken arguments access

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-17 17:03 -------
I think this is a bug in fold_indirect_ref and not in the fortran front-end as Jason as has said, the front-
end provides a cast to char* and then dereferences the pointer.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] Broken arguments access

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
           Keywords|                            |wrong-code
            Summary|Broken arguments access     |[4.0 Regression] Broken
                   |                            |arguments access


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-17 17:25 -------
This is a bug in fold_indirect_ref.

We get now:
  D.485 = (*ca)[0]{lb: 1 sz: 1};

See how the lower bound is 1, that is wrong.

A related C testcase is:
typedef char a[1];
int f(a * b)
{
  char a1 = *(char*)(b);
  return a1;
}

But for C we don't have lower bounds.  In fact in fortran we lower already all the array access so they 
start at 0.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2005-02-17 17:25:38
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-17 17:40 -------
I have a fix which I am testing.

Yes this was a bug in fold_indirect_ref and not in the fortran front-end (it could also effect Ada too but I 
have no test case for Ada).

Basically what happens is that the fortran front-end creates an array which has a lower bound of 1 and 
then casts the reference to one of those to the pointer type of the type (of what is in the array) and the 
dereferences it.  fold_indirect_ref would get the lower bound wrong which causes us to create wrong 
code.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-17 17:45 -------
Patch here: <http://gcc.gnu.org/ml/gcc-patches/2005-02/msg01002.html>.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[tobi at gcc dot gnu dot org]

------- Additional Comments From tobi at gcc dot gnu dot org  2005-02-17 18:12 -------
Adding Steve Kargl to CC, as this is most likely the BLAS bug he's been looking
into.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |sgk at troutmask dot apl dot
                   |                            |washington dot edu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[sgk at troutmask dot apl dot washington dot edu]

------- Additional Comments From sgk at troutmask dot apl dot washington dot edu  2005-02-17 19:17 -------
Tobi, thanks for adding me to the CC list.  Andrew's patch
fixes the problems with BLAS.  I extemely happy someone else
was able to find a small testcase, because I was starting
to trim 10K LOC.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-02-18 14:32 -------
Subject: Bug 20030

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	pinskia@gcc.gnu.org	2005-02-18 14:31:54

Modified files:
	gcc            : ChangeLog fold-const.c 

Log message:
	2005-02-18  Andrew Pinski  <pinskia@physics.uc.edu>
	
	PR middle-end/20030
	* fold-const.c (fold_indirect_ref_1): Use the correct index for zero access,
	the lower bound of the array type if it exists.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.7523&r2=2.7524
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/fold-const.c.diff?cvsroot=gcc&r1=1.514&r2=1.515



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-18 14:37 -------
Fixed.
------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-02-18 14:37 -------
Subject: Bug 20030

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	pinskia@gcc.gnu.org	2005-02-18 14:36:55

Modified files:
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gfortran.fortran-torture/execute: 
	                                                character_passing.f90 

Log message:
	2004-02-18  Andrew Pinski  <pinskia@physics.uc.edu>
	
	PR middle-end/20030
	* gfortran.fortran-torture/execute/character_passing.f90: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.5047&r2=1.5048
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.fortran-torture/execute/character_passing.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030

[Bug middle-end/20030] [4.0 Regression] Broken arguments access

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-02-18 14:37 -------
Fixed.
------- Additional Comments From cvs-commit at gcc dot gnu dot org  2005-02-18 14:37 -------
Subject: Bug 20030

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	pinskia@gcc.gnu.org	2005-02-18 14:36:55

Modified files:
	gcc/testsuite  : ChangeLog 
Added files:
	gcc/testsuite/gfortran.fortran-torture/execute: 
	                                                character_passing.f90 

Log message:
	2004-02-18  Andrew Pinski  <pinskia@physics.uc.edu>
	
	PR middle-end/20030
	* gfortran.fortran-torture/execute/character_passing.f90: New test.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/ChangeLog.diff?cvsroot=gcc&r1=1.5047&r2=1.5048
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/testsuite/gfortran.fortran-torture/execute/character_passing.f90.diff?cvsroot=gcc&r1=NONE&r2=1.1



-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20030