[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |NEW

--- Comment #13 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-06-29 13:23:15 UTC ---
(In reply to comment #5)
> Grrr:
> Visiting PHI node: i_3 = PHI <i_17(4), i_13(13)>;
>     Argument #0 (4 -> 12 executable)
>         i_17
>         Value: [1, 1]  EQUIVALENCES: { } (0 elements)
> 
>     Argument #1 (13 -> 12 executable)
>         i_13
>         Value: [2, +INF]  EQUIVALENCES: { } (0 elements)
> 
> so we have [1,1] UNION [2, +INF] and we just get ~[0,0] bogus
> and it also means this is PR 23744.

This is now fixed but the rest still applies.

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

--- Comment #14 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
So the difference now is:

```
void f0(int n, float *v)
{
  int i;
  if (n <= 0)
    return;
  else
    {
      i = 1;
      do {
        if (i > n)  __builtin_abort ();
        v[i] = 0.0;
        i++;
      } while (i != n);
    }
}

void f1(int n, float *v)
{
  int i;
  if (n <= 0)
    return;
  else
    {
      i = 1;
      do {
        if (i > n)  __builtin_abort ();
        v[i] = 0.0;
        i++;
      } while (i <= n);
    }
}
```

Dom2 can remove the if for f1 while not for f0 due to the relationship in the
f0 is `!=` and `>` while in the second case it is `<=`/`>` for both.

Also the original fortran testcase still does not optimize either for LP64
targets. where fortran uses basically long to do the math
 something like:
```
float v[1000];
void f1(int n)
{
  int i;
  if (n <= 0)
    return;
  else
    {
      i = 1;
      do {
        long t = i;
        if (t > (long)n)  __builtin_abort ();
        v[i] = 0.0;
        t++;
        i = t;
      } while (i <= n);
    }
}

```

[Bug tree-optimization/25643] New: VRP does not remove -fbounds-checking for Fortran

[pinskia at gcc dot gnu dot org]

Testcase:
subroutine mysub(n,v)
  integer :: n
  real    :: v(n)

  if (n<=0) return
  do i=1, n
    v(i) = i*i
  end do
  return
end subroutine mysub
------
Compile with -fbounds-checking -O2, and notice that there is still
_gfortran_runtime_error left in the asm.


-- 
           Summary: VRP does not remove -fbounds-checking for Fortran
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: enhancement
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pinskia at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #1 from pinskia at gcc dot gnu dot org  2006-01-02 20:23 -------
C testcase:
int v[10000]={0};
void f(int n)
{
  int i;
  if (n <= 0)
    return;
  if (n > 0)
    {
      i = 1;
      do {
        _Bool t = i <= 0;
        _Bool t1 = i > n;
        _Bool t2 = t || t1;
        if (t2)  __builtin_abort ();

        v[i] = i*i;

        i++;
      } while (i != n);
    }
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #2 from pinskia at gcc dot gnu dot org  2006-01-03 20:37 -------
This is related to PR 18527 but not fixed with -funsafe-loop-optimizations.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  BugsThisDependsOn|                            |18527


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2006-01-07 16:20 -------
This was NOT fixed by the patch which fixed PR 18527.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #4 from pinskia at gcc dot gnu dot org  2006-01-29 21:05 -------
Confirmed.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2006-01-29 21:05:41
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #5 from pinskia at gcc dot gnu dot org  2006-05-01 06:18 -------
Grrr:
Visiting PHI node: i_3 = PHI <i_17(4), i_13(13)>;
    Argument #0 (4 -> 12 executable)
        i_17
        Value: [1, 1]  EQUIVALENCES: { } (0 elements)

    Argument #1 (13 -> 12 executable)
        i_13
        Value: [2, +INF]  EQUIVALENCES: { } (0 elements)

so we have [1,1] UNION [2, +INF] and we just get ~[0,0] bogus
and it also means this is PR 23744.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  BugsThisDependsOn|                            |23744


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[baldrick at free dot fr]

------- Comment #6 from baldrick at free dot fr  2006-05-01 10:09 -------
Re comment #5:

> so we have [1,1] UNION [2, +INF] and we just get ~[0,0] bogus
> and it also means this is PR 23744.

This is more than PR 23744: with the fix for PR 23744 applied,
__builtin_abort () is still not eliminated due to VRP failing
to eliminate the i > n comparison (symbolic range).


-- 

baldrick at free dot fr changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |baldrick at free dot fr


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[rguenth at gcc dot gnu dot org]

------- Comment #7 from rguenth at gcc dot gnu dot org  2008-01-12 12:38 -------
ifcombine fixes this.


-- 

rguenth at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2008-01-12 12:42 -------
Actually VRP1 catches both the fortran and the C testcase.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #9 from pinskia at gcc dot gnu dot org  2008-01-12 12:44 -------
This is not fixed for me with 
gcc version 4.3.0 20071110 (experimental) [trunk revision 130075] (GCC) 

or with Sun Jan  6 02:53:41 UTC 2008 (revision 131347)


-- 

pinskia at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #10 from pinskia at gcc dot gnu dot org  2008-01-12 12:48 -------
There should be no call to _gfortran_runtime_error_at and I still get one:
  # i_2 = PHI <1(3), i_27(6)>
  if (i_2 > size.1_4)
    goto <bb 5>;
  else
    goto <bb 6>;

<bb 5>:
  # i_32 = PHI <i_2(4)>  _gfortran_runtime_error_at (&"At line 7 of file
t.f90"[1]{lb: 1 sz: 1}, &"Array reference out of bounds for array \'v\', upper
bound of dimension 1 exceeded (%ld > %ld)"[1]{lb: 1 sz: 1}, i_32, size.1_4);


Yes one call to _gfortran_runtime_error_at is removed but that is the lower
bound one.

Even the C testcase has the same issue.
we still get a call to abort:
L4:
        call    _abort


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[pinskia at gcc dot gnu dot org]

------- Comment #11 from pinskia at gcc dot gnu dot org  2008-01-12 12:49 -------
>Compile with -fbounds-checking -O2, and notice that there is still

That should be just -fbounds-check .


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[rguenth at gcc dot gnu dot org]

------- Comment #12 from rguenth at gcc dot gnu dot org  2008-01-12 14:23 -------
Simplified C testcase (omitting the parts that are optimized).  We cannot
figure out the number of iterations of this loop:

void f(int n, float *v)
{
  int i;
  if (n <= 0)
    return;
  else
    {
      i = 1;
      do {
        if (i > n)  __builtin_abort ();

        v[i] = 0.;

        i++;
      } while (i <= n);
    }
}

Note that for this simplified testcase DOM figures out the redundant test,
but not if you use i != n as the exit test, as for n == 1 the assert
will trigger.  If you look at the fortran IL, it does instead

   } while (i++ != n);

which would be ok again, but is still not optimized.

Testcase:

extern void link_error (void);
int i, n;
int main()
{
  if (i > n)
    return;
  if (i != n)
    if (i >= n)
        link_error ();
}


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643

[Bug tree-optimization/25643] VRP does not remove -fbounds-check for Fortran

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.0                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25643