[Bug fortran/38033] New: Bounds of a pointer/allocatable array not stabilized

[Bug fortran/38033] New: Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

Following thread from Thomas Breuel,
http://gcc.gnu.org/ml/fortran/2008-11/msg00044.html

program test
   integer, parameter :: n = 100
   real, pointer :: a(:),temp(:)  ! pointer or allocatable have the same effect
   allocate(a(n), temp(n))
   temp(1:size(a)) = a
end program

produces
test ()
{
  struct array01_real(kind=4) a;
  struct array01_real(kind=4) temp;
  static integer(kind=4) options.0[8] = {68, 255, 0, 0, 0, 1, 0, 1};

  a.data = 0B;
  temp.data = 0B;
  _gfortran_set_options (8, (void *) &options.0);
  {
    void * D.646;

    a.dtype = 281;
    a.dim[0].lbound = 1;
    a.dim[0].ubound = 100;
    a.dim[0].stride = 1;
    D.646 = __builtin_malloc (400);
    if (D.646 == 0B)
      {
        _gfortran_os_error (&"Out of memory"[1]{lb: 1 sz: 1});
      }
    a.data = D.646;
    a.offset = -1;
  }
  {
    void * D.647;

    temp.dtype = 281;
    temp.dim[0].lbound = 1;
    temp.dim[0].ubound = 100;
    temp.dim[0].stride = 1;
    D.647 = __builtin_malloc (400);
    if (D.647 == 0B)
      {
        _gfortran_os_error (&"Out of memory"[1]{lb: 1 sz: 1});
      }
    temp.data = D.647;
    temp.offset = -1;
  }
  {
    integer(kind=4) D.664;
    void * D.662;
    void * D.661;
    integer(kind=4) D.660;
    integer(kind=4) D.659;
    logical(kind=4) D.658;
    struct array01_real(kind=4) atmp.1;
    struct array01_real(kind=4) * D.656;
    integer(kind=4) D.655;
    struct array01_real(kind=4) * D.654;
    integer(kind=4) D.653;
    real(kind=4)[0:] * D.652;
    integer(kind=4) D.651;
    integer(kind=4) D.650;
    integer(kind=4) D.649;
    real(kind=4)[0:] * D.648;

    D.648 = (real(kind=4)[0:] *) a.data;
    D.649 = a.offset;
    D.650 = a.dim[0].lbound;
    D.651 = a.dim[0].ubound;
    D.652 = (real(kind=4)[0:] *) temp.data;
    D.653 = temp.offset;
    D.654 = &a;
    D.655 = _gfortran_size0 (D.654);
    D.656 = &a;
    atmp.1.dtype = 281;
    atmp.1.dim[0].stride = 1;
    atmp.1.dim[0].lbound = 0;
    atmp.1.dim[0].ubound = _gfortran_size0 (D.656) + -1;
    D.658 = _gfortran_size0 (D.656) <= 0;
    D.659 = _gfortran_size0 (D.656);
    D.660 = D.658 ? 0 : D.659 * 4;
    if (D.660 < 0)
      {
        _gfortran_runtime_error (&"Attempt to allocate a negative amount of
memo
ry."[1]{lb: 1 sz: 1});
      }
    D.661 = __builtin_malloc (MAX_EXPR <D.660, 1>);
    if (D.661 == 0B)
      {
        _gfortran_os_error (&"Memory allocation failed"[1]{lb: 1 sz: 1});
      }
    D.662 = D.661;
    atmp.1.data = D.662;
    atmp.1.offset = 0;
    D.664 = NON_LVALUE_EXPR <D.650>;
    {
      integer(kind=4) D.667;
      integer(kind=4) D.666;
      integer(kind=4) S.2;

      D.666 = a.dim[0].stride;
      D.667 = temp.dim[0].stride;
      S.2 = 0;
      while (1)
        {
          if (_gfortran_size0 (D.656) + -1 < S.2) goto L.1;
          (*(real(kind=4)[0:] *) atmp.1.data)[S.2] = (*D.648)[(S.2 + D.664) *
D.
666 + D.649];
          S.2 = S.2 + 1;
        }
      L.1:;
      S.2 = 0;
      while (1)
        {
          if (_gfortran_size0 (D.656) + -1 < S.2) goto L.2;
          (*D.652)[(S.2 + 1) * D.667 + D.653] = (*(real(kind=4)[0:] *)
atmp.1.da
ta)[S.2];
          S.2 = S.2 + 1;
        }
      L.2:;
    }
    {
      void * D.663;

      D.663 = atmp.1.data;
      if (D.663 != 0B)
        {
          __builtin_free (D.663);
        }
    }
  }
}

Calls to gfortran_size0 are appearing withing the scalarized loop because the
expression for the upper bound of atmp is not stabilized.


-- 
           Summary: Bounds of  a pointer/allocatable array not stabilized
           Product: gcc
           Version: 4.4.0
            Status: UNCONFIRMED
          Keywords: wrong-code, missed-optimization
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: pault at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

-- 

pault at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |pault at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2008-11-06 14:24:59
               date|                            |
            Summary|Bounds of  a                |Bounds of  a
                   |pointer/allocatable array   |pointer/allocatable array
                   |not stabilized              |not stabilized


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #1 from pault at gcc dot gnu dot org  2008-11-06 21:09 -------
(In reply to comment #0)
Index: gcc/fortran/trans-array.c
===================================================================
--- gcc/fortran/trans-array.c   (revision 141655)
+++ gcc/fortran/trans-array.c   (working copy)
@@ -645,13 +645,16 @@
     {
       n = loop->order[dim];
       if (n < loop->temp_dim)
-      gcc_assert (integer_zerop (loop->from[n]));
+       gcc_assert (integer_zerop (loop->from[n]));
       else
        {
          /* Callee allocated arrays may not have a known bound yet.  */
          if (loop->to[n])
-             loop->to[n] = fold_build2 (MINUS_EXPR, gfc_array_index_type,
-                                        loop->to[n], loop->from[n]);
+           {
+             tmp = fold_build2 (MINUS_EXPR, gfc_array_index_type,
+                                loop->to[n], loop->from[n]);
+             loop->to[n] = gfc_evaluate_now (tmp, pre);
+           }
          loop->from[n] = gfc_index_zero_node;
        }

Removes the _gfortran_size0 calls from the loop but I have not yet located the
doubling up of calls here:

    D.1551 = temp.offset;
    D.1552 = &a;
    D.1553 = (integer(kind=8)) (integer(kind=4)) _gfortran_size0 (D.1552);
    D.1554 = &a;
    D.1555 = (integer(kind=8)) (integer(kind=4)) _gfortran_size0 (D.1554) + -1;
    atmp.1.dtype = 281;
    atmp.1.dim[0].stride = 1;
    atmp.1.dim[0].lbound = 0;
    atmp.1.dim[0].ubound = D.1555;
    D.1557 = D.1555 + 1 <= 0;
    D.1558 = D.1555 + 1;
    D.1559 = D.1557 ? 0 : D.1558 * 4;
    if (D.1559 < 0)
      {
        _gfortran_runtime_error (&"Attempt to allocate a negative amount of
memory."[1]{lb: 1 sz: 1});
      }
    D.1560 = __builtin_malloc (MAX_EXPR <D.1559, 1>);
    if (D.1560 == 0B)
      {
        _gfortran_os_error (&"Memory allocation failed"[1]{lb: 1 sz: 1});
      }
    D.1561 = D.1560;
    atmp.1.data = D.1561;
    atmp.1.offset = 0;
    D.1563 = NON_LVALUE_EXPR <D.1548>;
    {
      integer(kind=8) D.1566;
      integer(kind=8) D.1565;
      integer(kind=8) S.2;

      D.1565 = a.dim[0].stride;
      D.1566 = temp.dim[0].stride;
      S.2 = 0;
      while (1)
        {
          if (S.2 > D.1555) goto L.1;
          (*(real(kind=4)[0:] *) atmp.1.data)[S.2] = (*D.1546)[(S.2 + D.1563) *
D.1565 + D.1547];
          S.2 = S.2 + 1;
        }
      L.1:;
      S.2 = 0;
      while (1)
        {
          if (S.2 > D.1555) goto L.2;
          (*D.1550)[(S.2 + 1) * D.1566 + D.1551] = (*(real(kind=4)[0:] *)
atmp.1.data)[S.2];
          S.2 = S.2 + 1;
        }
      L.2:;
    }


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #2 from pault at gcc dot gnu dot org  2008-11-07 13:59 -------
(In reply to comment #1)
> (In reply to comment #0)

> Removes the _gfortran_size0 calls from the loop but I have not yet located the
> doubling up of calls here:
> 
>     D.1551 = temp.offset;
>     D.1552 = &a;
>     D.1553 = (integer(kind=8)) (integer(kind=4)) _gfortran_size0 (D.1552);
>     D.1554 = &a;
>     D.1555 = (integer(kind=8)) (integer(kind=4)) _gfortran_size0 (D.1554) + -1

The second call and stabilization of &a comes from the patch in #1.  The first
is from gfc_conv_section_startstride, after:

  /* Similarly calculate the end.  Although this is not used in the
     scalarizer, it is needed when checking bounds and where the end
     is an expression with side-effects.  */

Thus, rather than using loop->to[n], we will probably have to detect
GFC_SS_SECTION && ss->data.info.end[n] and the use ss->data.info.start[n] and
ss->data.info.end[n].

I'm working on it!

Paul


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #3 from pault at gcc dot gnu dot org  2008-11-13 15:01 -------
Created an attachment (id=16662)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=16662&action=view)
Patch for PR

This is not regtested yet but it fixes the testcase completely.  Watch this
space!

Paul


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[mikael at gcc dot gnu dot org]

------- Comment #4 from mikael at gcc dot gnu dot org  2008-11-13 19:05 -------
Maybe we can drop gfc_conv_section_upper_bound completely. 
It looks redundant with how info->end[n] is calculated in
gfc_conv_section_startstride. 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #5 from pault at gcc dot gnu dot org  2008-11-14 11:16 -------
Mikael,

(In reply to comment #4)
> Maybe we can drop gfc_conv_section_upper_bound completely. 
> It looks redundant with how info->end[n] is calculated in
> gfc_conv_section_startstride. 
> 

I tried that and generated a load of regressions.  I didn't analyse this but
stuck with the patch as is.  I maybe will think about it some more but I am
inclined to put a TODO in there.

Thanks

Paul 


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[mikael at gcc dot gnu dot org]

------- Comment #6 from mikael at gcc dot gnu dot org  2008-11-14 12:54 -------
(In reply to comment #5)
> I tried that and generated a load of regressions.  
Fine. Let's keep it as is then. 
> 
> Thanks
Thanks to you.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #7 from pault at gcc dot gnu dot org  2008-11-14 18:04 -------
Subject: Bug 38033

Author: pault
Date: Fri Nov 14 18:03:05 2008
New Revision: 141861

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=141861
Log:
2008-10-14  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/38033
        * trans-array.c (gfc_trans_create_temp_array): Stabilize the
        'to' expression.
        (gfc_conv_loop_setup): Use the end expression for the loop 'to'
        if it is available.

2008-10-14  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/38033
        * gfortran.dg/array_section_2.f90: New test.



Added:
    trunk/gcc/testsuite/gfortran.dg/array_section_2.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/trans-array.c
    trunk/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #8 from pault at gcc dot gnu dot org  2008-11-14 18:39 -------
Fixed on trunk

Paul


-- 

pault at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #9 from pault at gcc dot gnu dot org  2008-11-24 21:37 -------
Subject: Bug 38033

Author: pault
Date: Mon Nov 24 21:36:05 2008
New Revision: 142174

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=142174
Log:
2008-11-24  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/38033
        * trans-array.c (gfc_trans_create_temp_array): Stabilize the
        'to' expression.
        (gfc_conv_loop_setup): Use the end expression for the loop 'to'
        if it is available.

2008-11-24  Paul Thomas  <pault@gcc.gnu.org>

        PR fortran/38033
        * gfortran.dg/array_section_2.f90: New test.



Added:
    branches/gcc-4_3-branch/gcc/testsuite/gfortran.dg/array_section_2.f90
Modified:
    branches/gcc-4_3-branch/gcc/fortran/ChangeLog
    branches/gcc-4_3-branch/gcc/fortran/trans-array.c
    branches/gcc-4_3-branch/gcc/testsuite/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033

[Bug fortran/38033] Bounds of a pointer/allocatable array not stabilized

[pault at gcc dot gnu dot org]

------- Comment #10 from pault at gcc dot gnu dot org  2008-11-24 21:38 -------
Fixed on trunk and 4.3.

Thanks for the report.

Paul


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38033