[Bug lto/45586] New: ICE non-trivial conversion at assignment

[Bug lto/45586] New: ICE non-trivial conversion at assignment

[jv244 at cam dot ac dot uk]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain, Size: 1734 bytes --]

I'm trying to do an lto build of CP2K, but this fails with:

/data03/vondele/clean/cp2k/makefiles/../src/qs_linres_current.F: In function
‘calculate_jrho_resp’:
/data03/vondele/clean/cp2k/makefiles/../src/qs_linres_current.F:614:0: error:
non-trivial conversion at assignment
struct array3_real(kind=8)
struct array3_real(kind=8)
# .MEM_3681 = VDEF <.MEM_3286>
my_rho = D.13693_938->r;

/data03/vondele/clean/cp2k/makefiles/../src/qs_linres_current.F:614:0: error:
non-trivial conversion at assignment
struct array3_real(kind=8)
struct array3_real(kind=8)
# .MEM_3682 = VDEF <.MEM_3681>
my_current = D.13701_946->r;

/data03/vondele/clean/cp2k/makefiles/../src/qs_linres_current.F:614:0: error:
non-trivial conversion at assignment
struct array3_real(kind=8)
struct array3_real(kind=8)
# .MEM_3750 = VDEF <.MEM_3287>
my_gauge = D.14057_1326->r;

/data03/vondele/clean/cp2k/makefiles/../src/qs_linres_current.F:614:0: internal
compiler error: verify_stmts failed
Please submit a full bug report.

Not sure how to get a testcase for this one....

It appears to be for example this pointer assignment

my_gauge=>rs_gauge(1)%rs(igrid_level)%rs_grid%r

where 

REAL(dp), DIMENSION(:, :, :), POINTER    ::  my_gauge

and r is:

  TYPE realspace_grid_type

     REAL(KIND=dp), DIMENSION ( :, :, : ), ALLOCATABLE :: r   ! the grid

  END TYPE realspace_grid_type


-- 
           Summary: ICE non-trivial conversion at assignment
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: lto
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: jv244 at cam dot ac dot uk


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

[Bug lto/45586] ICE non-trivial conversion at assignment

[jv244 at cam dot ac dot uk]

------- Comment #1 from jv244 at cam dot ac dot uk  2010-09-07 19:28 -------
Simple testcase (gfortran -flto test.f90):

MODULE M1
  INTEGER, PARAMETER :: dp=8
  TYPE realspace_grid_type

     REAL(KIND=dp), DIMENSION ( :, :, : ), ALLOCATABLE :: r

  END TYPE realspace_grid_type
END MODULE

MODULE M2
 USE m1
CONTAINS
 SUBROUTINE S1(x)
  TYPE(realspace_grid_type), POINTER :: x
  REAL(dp), DIMENSION(:, :, :), POINTER    :: y
  y=>x%r
  y=0

 END SUBROUTINE
END MODULE

USE M2
  TYPE(realspace_grid_type), POINTER :: x
  ALLOCATE(x)
  ALLOCATE(x%r(10,10,10))
  CALL S1(x)
  write(6,*) x%r
END


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[jv244 at cam dot ac dot uk]

------- Comment #2 from jv244 at cam dot ac dot uk  2010-09-07 19:30 -------
Actually works with 4.5 but fails with trunk


-- 

jv244 at cam dot ac dot uk changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
            Summary|ICE non-trivial conversion  |[4.6 Regression] ICE non-
                   |at assignment               |trivial conversion at
                   |                            |assignment
   Target Milestone|---                         |4.6.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[rguenth at gcc dot gnu dot org]

------- Comment #3 from rguenth at gcc dot gnu dot org  2010-09-08 09:36 -------
Confirmed.  We have two different array3_real(kind=8) record types that are
not considered compatible.  One data pointer member is restrict qualified
while the other one is not.

Why do we have an aggregate assignment of an array descriptor anyway?

  struct array3_real(kind=8) y;
  struct realspace_grid_type * D.2157;

<bb 2>:
  y.data = 0B;
  D.2157_5 = *x_4(D);
  y = D.2157_5->r;
  D.2172_6 = y.data;

The issue here is of course that LTO re-computes TYPE_CANONICAL and the FE
sets it in a way that the above situation is not detected as non-trivial
conversion.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
           Keywords|                            |lto
   Last reconfirmed|0000-00-00 00:00:00         |2010-09-08 09:36:18
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[Joost dot VandeVondele at pci dot uzh dot ch]

------- Comment #4 from Joost dot VandeVondele at pci dot uzh dot ch  2010-09-15 18:25 -------
If realspace_grid_type%r is declared as pointer instead of allocatable, the
testcase can be compiled, and a similar workaround in CP2K allows an LTO build
of CP2K to finish (after updating binutils to its CVS version).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-10-21 15:28:01 UTC ---
Created attachment 22107
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=22107
patch

Bootstrapped, tested and SPEC CPU 2006 tested.

I don't like it too much (it's really a FE hack we put in place that haunts
us now ...).

Deferred installing to stage3 or later (--enable-checking=release will
get you an equivalent solution).

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-checking
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-09-30 11:49:44 UTC ---
Mine.

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[Joost.VandeVondele at pci dot uzh.ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

--- Comment #6 from Joost VandeVondele <Joost.VandeVondele at pci dot uzh.ch> 2010-09-24 10:46:08 UTC ---
(In reply to comment #5)
> Actually, looks like there might be some vaguely related issue here in the FE,
> which I'll open in another PR.

See PR45777

-- 
Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

[Bug lto/45586] [4.6 Regression] ICE non-trivial conversion at assignment

[Joost.VandeVondele at pci dot uzh.ch]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45586

--- Comment #5 from Joost VandeVondele <Joost.VandeVondele at pci dot uzh.ch> 2010-09-24 10:33:13 UTC ---
(In reply to comment #3)
> The issue here is of course that LTO re-computes TYPE_CANONICAL and the FE
> sets it in a way that the above situation is not detected as non-trivial
> conversion.

Does the fix need to be done in the FE, or will this be addressed by changes to
LTO?

Actually, looks like there might be some vaguely related issue here in the FE,
which I'll open in another PR.

-- 
Configure bugmail: http://gcc.gnu.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.