[Bug libgomp/31974] New: [4.3 regression]: Many libgomp failures

[Bug libgomp/31974] New: [4.3 regression]: Many libgomp failures

[hjl at lucon dot org]

In

http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg00804.html

there are

FAIL: libgomp.fortran/vla1.f90  -O2  execution test
FAIL: libgomp.fortran/vla1.f90  -O3 -fomit-frame-pointer  execution test
FAIL: libgomp.fortran/vla1.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: libgomp.fortran/vla1.f90  -O3 -fomit-frame-pointer -funroll-all-loops
-finline-functions  execution test
FAIL: libgomp.fortran/vla1.f90  -O3 -g  execution test
FAIL: libgomp.fortran/vla1.f90  -Os  execution test
FAIL: libgomp.fortran/vla2.f90  -O2  execution test
FAIL: libgomp.fortran/vla2.f90  -O3 -fomit-frame-pointer  execution test
FAIL: libgomp.fortran/vla2.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: libgomp.fortran/vla2.f90  -O3 -fomit-frame-pointer -funroll-all-loops
-finline-functions  execution test
FAIL: libgomp.fortran/vla2.f90  -O3 -g  execution test
FAIL: libgomp.fortran/vla2.f90  -Os  execution test
FAIL: libgomp.fortran/vla3.f90  -O2  execution test
FAIL: libgomp.fortran/vla3.f90  -O3 -fomit-frame-pointer  execution test
FAIL: libgomp.fortran/vla3.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: libgomp.fortran/vla3.f90  -O3 -fomit-frame-pointer -funroll-all-loops
-finline-functions  execution test
FAIL: libgomp.fortran/vla3.f90  -O3 -g  execution test
FAIL: libgomp.fortran/vla3.f90  -Os  execution test
FAIL: libgomp.fortran/vla4.f90  -O2  execution test
FAIL: libgomp.fortran/vla4.f90  -O3 -fomit-frame-pointer  execution test
FAIL: libgomp.fortran/vla4.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: libgomp.fortran/vla4.f90  -O3 -fomit-frame-pointer -funroll-all-loops
-finline-functions  execution test
FAIL: libgomp.fortran/vla4.f90  -O3 -g  execution test
FAIL: libgomp.fortran/vla4.f90  -Os  execution test
FAIL: libgomp.fortran/vla5.f90  -O2  execution test
FAIL: libgomp.fortran/vla5.f90  -O3 -fomit-frame-pointer  execution test
FAIL: libgomp.fortran/vla5.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: libgomp.fortran/vla5.f90  -O3 -fomit-frame-pointer -funroll-all-loops
-finline-functions  execution test
FAIL: libgomp.fortran/vla5.f90  -O3 -g  execution test
FAIL: libgomp.fortran/vla5.f90  -Os  execution test
FAIL: libgomp.fortran/vla6.f90  -O2  execution test
FAIL: libgomp.fortran/vla6.f90  -O3 -fomit-frame-pointer  execution test
FAIL: libgomp.fortran/vla6.f90  -O3 -fomit-frame-pointer -funroll-loops 
execution test
FAIL: libgomp.fortran/vla6.f90  -O3 -fomit-frame-pointer -funroll-all-loops
-finline-functions  execution test
FAIL: libgomp.fortran/vla6.f90  -O3 -g  execution test
FAIL: libgomp.fortran/vla6.f90  -Os  execution test


-- 
           Summary: [4.3 regression]: Many libgomp failures
           Product: gcc
           Version: 4.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: libgomp
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: hjl at lucon dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug libgomp/31974] [4.3 regression]: Many libgomp failures

[hjl at lucon dot org]

------- Comment #1 from hjl at lucon dot org  2007-05-17 15:06 -------
Failures look like

Fortran runtime error: Attempt to allocate a negative amount of memory.
FAIL: libgomp.fortran/vla1.f90  -O2  execution test

Is this related to

http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00148.html


-- 

hjl at lucon dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug libgomp/31974] [4.3 regression]: Many libgomp failures

[pinskia at gcc dot gnu dot org]

-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
   Target Milestone|---                         |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug libgomp/31974] [4.3 regression]: Many libgomp failures

[hjl at lucon dot org]

------- Comment #2 from hjl at lucon dot org  2007-05-19 00:46 -------
I have verified that this patch

http://gcc.gnu.org/ml/fortran/2007-05/msg00123.html

causes the regression. FX, can you take a look? Thanks.


-- 

hjl at lucon dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug libgomp/31974] [4.3 regression]: Many libgomp failures

[hjl at lucon dot org]

------- Comment #3 from hjl at lucon dot org  2007-05-19 02:52 -------
malloc is

void *malloc(size_t size);

and size_t is unsigned. I am not sure if checking size < 0 in Fortran is needed
or done properly.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug libgomp/31974] [4.3 regression]: Many libgomp failures

[hjl at lucon dot org]

------- Comment #4 from hjl at lucon dot org  2007-05-19 16:12 -------
Here is a simple testcase. It fails on Linux/x86-64 and Linux/ia64. I think
it may fail on all 64bit systems:

[hjl@gnu-26 31974]$ cat foo.f90
! { dg-do run }

  call test
contains
  subroutine foo (n)
    integer :: n
    integer :: p, q
    character (len = n) :: s
    character (len = n), dimension (5, 3:n) :: v
    s = 'P'
    forall (p = 1:5, q = 3:7, p + q .gt. 8) v(p, q) = '!'
  end subroutine foo

  subroutine test
    call foo (7)
  end subroutine test
end
[hjl@gnu-26 31974]$ make
/export/build/gnu/gcc-last/build-x86_64-linux/gcc/gfortran
-B/export/build/gnu/gcc-last/build-x86_64-linux/gcc/
-L/export/build/gnu/gcc-last/build-x86_64-linux/gcc/../x86_64-unknown-linux-gnu/libgfortran/.libs
-static -g -o foo foo.o
./foo
Fortran runtime error: Attempt to allocate a negative amount of memory.
make: *** [all] Error 2
[hjl@gnu-26 31974]$


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug libgomp/31974] [4.3 regression]: Many libgomp failures

[hjl at lucon dot org]

------- Comment #5 from hjl at lucon dot org  2007-05-19 17:04 -------
VRP says

D.1387_39: [-10737418225, 0]

Simplified relational D.1387_39 < 0 into D.1387_39 != 0

  if (D.1387_39 != 0)
    goto <bb 4>;
  else
    goto <bb 5> (<L37>);

<bb 4>:
  _gfortran_runtime_error ("Attempt to allocate a negative amount of memory.");

Is this a VRP bug?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug tree-optimization/31974] [4.3 regression]: Many libgomp failures

[fxcoudert at gcc dot gnu dot org]

------- Comment #6 from fxcoudert at gcc dot gnu dot org  2007-05-19 18:16 -------
Indeed...

$ cat u.f90               
  subroutine foo (n)
    integer :: n
    character (len = n) :: v(n)
    v = ''
  end subroutine foo

  call foo(7)
  end
$ gfortran u.f90 -O && ./a.out
$ gfortran u.f90 -O -ftree-vrp && ./a.out
Fortran runtime error: Attempt to allocate a negative amount of memory.


The tree-original output is a bit convoluted, but I think it's doing the right
thing. I'm attaching it here.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
          Component|libgomp                     |tree-optimization
     Ever Confirmed|0                           |1
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|0000-00-00 00:00:00         |2007-05-19 18:16:20
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug tree-optimization/31974] [4.3 regression]: Many libgomp failures

[fxcoudert at gcc dot gnu dot org]

------- Comment #7 from fxcoudert at gcc dot gnu dot org  2007-05-19 18:19 -------
Created an attachment (id=13584)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=13584&action=view)
Output of -fdump-tree-original for the testcase u.f90


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.3 regression]: Many libgomp failures

[fxcoudert at gcc dot gnu dot org]

------- Comment #8 from fxcoudert at gcc dot gnu dot org  2007-05-19 18:34 -------
Andrew Pinski noticed on IRC a type mismatch in the generated code, that leads
to this wrong-code. This ptach fixes it, I'm regtesting it right now.

Index: trans-array.c
===================================================================
--- trans-array.c       (revision 124856)
+++ trans-array.c       (working copy)
@@ -3846,7 +3846,8 @@ gfc_trans_auto_array_allocation (tree de
   /* The size is the number of elements in the array, so multiply by the
      size of an element to get the total size.  */
   tmp = TYPE_SIZE_UNIT (gfc_get_element_type (type));
-  size = fold_build2 (MULT_EXPR, gfc_array_index_type, size, tmp);
+  size = fold_build2 (MULT_EXPR, gfc_array_index_type, size,
+                     fold_convert (gfc_array_index_type, tmp));

   /* Allocate memory to hold the data.  */
   tmp = gfc_call_malloc (&block, TREE_TYPE (decl), size);


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |fxcoudert at gcc dot gnu dot
                   |dot org                     |org
             Status|NEW                         |ASSIGNED
          Component|tree-optimization           |fortran
           Keywords|ice-on-valid-code           |patch, wrong-code
   Last reconfirmed|2007-05-19 18:16:20         |2007-05-19 18:34:53
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.3 regression]: Many libgomp failures

[fxcoudert at gcc dot gnu dot org]

------- Comment #9 from fxcoudert at gcc dot gnu dot org  2007-05-19 20:51 -------
Subject: Bug 31974

Author: fxcoudert
Date: Sat May 19 19:51:21 2007
New Revision: 124858

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=124858
Log:
        PR fortran/31974

        * trans-array.c (gfc_trans_auto_array_allocation): Avoid
        multiplication of mismatched types.

        * gfortran.dg/char_allocation_1.f90: New test.

Added:
    trunk/gcc/testsuite/gfortran.dg/char_allocation_1.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=31974

[Bug fortran/31974] [4.3 regression]: Many libgomp failures

[fxcoudert at gcc dot gnu dot org]

------- Comment #10 from fxcoudert at gcc dot gnu dot org  2007-05-19 20:52 -------
Fixed.


-- 

fxcoudert at gcc dot gnu dot org changed:

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.2]: array allocation multiplication of mismatched types

[hjl at lucon dot org]

------- Comment #11 from hjl at lucon dot org  2007-05-19 21:01 -------
Gcc 4.2 has similar issue.


-- 

hjl at lucon dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
         Resolution|FIXED                       |
            Summary|[4.3 regression]: Many      |[4.2]: array allocation
                   |libgomp failures            |multiplication of mismatched
                   |                            |types
   Target Milestone|4.3.0                       |4.2.1
            Version|4.3.0                       |4.2.1


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.2]: array allocation multiplication of mismatched types

[fxcoudert at gcc dot gnu dot org]

------- Comment #12 from fxcoudert at gcc dot gnu dot org  2007-05-19 21:12 -------
(In reply to comment #11)
> Gcc 4.2 has similar issue.

It's not showing up in the ia64 testresults
(http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg00918.html). Is it triggered
by the testcase I committed?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.2]: array allocation multiplication of mismatched types

[hjl at lucon dot org]

------- Comment #13 from hjl at lucon dot org  2007-05-19 21:16 -------
> (In reply to comment #11)
> > Gcc 4.2 has similar issue.
> 
> It's not showing up in the ia64 testresults
> (http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg00918.html). Is it triggered
> by the testcase I committed?
>

It is a latent bug in gcc 4.2, the same way as in gcc 4.3. We didn't notice it
before your malloc change.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.2]: array allocation multiplication of mismatched types

[fxcoudert at gcc dot gnu dot org]

------- Comment #14 from fxcoudert at gcc dot gnu dot org  2007-05-19 21:21 -------
(In reply to comment #13)
> It is a latent bug in gcc 4.2, the same way as in gcc 4.3. We didn't notice it
> before your malloc change.

I'm not sure a latent bug can be considered for backport.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.2]: array allocation multiplication of mismatched types

[fxcoudert at gcc dot gnu dot org]

------- Comment #15 from fxcoudert at gcc dot gnu dot org  2007-05-28 16:17 -------
Removing the wrong-code keyword until we manage to find if it actually happens
on 4.2, or if it's just not exposed. I'm reluctant to backport the patch unless
we have a proof of it happening on 4.2 (and if it happens, it's probably not a
regression).


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |SUSPENDED
           Keywords|wrong-code                  |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974

[Bug fortran/31974] [4.2]: array allocation multiplication of mismatched types

[fxcoudert at gcc dot gnu dot org]

------- Comment #16 from fxcoudert at gcc dot gnu dot org  2007-06-05 22:58 -------
Fixed on mainline, no evidence of a regression on 4.2.


-- 

fxcoudert at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|SUSPENDED                   |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|4.2.1                       |4.3.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31974