[Bug tree-optimization/48098] New: internal compiler error: in build_vector_from_val, at tree.c:1380

[Bug tree-optimization/48098] New: internal compiler error: in build_vector_from_val, at tree.c:1380

[schnetter at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

           Summary: internal compiler error: in build_vector_from_val, at
                    tree.c:1380
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: schnetter@gmail.com


Created attachment 23638
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=23638
Failing preprocessed source coden

I am using

$ /Users/eschnett/gcc/bin/gcc --version
gcc (GCC) 4.6.0 20110311 (experimental)

$ svn info
Path: .
URL: svn://gcc.gnu.org/svn/gcc/trunk
Repository Root: svn://gcc.gnu.org/svn/gcc
Repository UUID: 138bc75d-0d04-0410-961f-82ee72b054a4
Revision: 170879
Node Kind: directory
Schedule: normal
Last Changed Author: jsm28
Last Changed Rev: 170879
Last Changed Date: 2011-03-11 11:51:29 -0500 (Fri, 11 Mar 2011)

I issue the command

$ /Users/eschnett/gcc/bin/gcc -std=gnu99 -O3 -c rotatingsymmetry180.i

and I receive the error message

/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:
In function ‘BndRot180VI’:
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:495:11:
warning: passing argument 1 of ‘free’ discards ‘restrict’ qualifier from
pointer target type [enabled by default]
/usr/include/stdlib.h:160:6: note: expected ‘void *’ but argument is of type
‘struct slabsetup * restrict* restrict’
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:514:12:
warning: passing argument 5 of ‘Slab_MultiTransfer_Apply’ from incompatible
pointer type [enabled by default]
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/Slab/src/slab.h:113:1:
note: expected ‘const void * const restrict* const restrict’ but argument is of
type ‘void * restrict* restrict’
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:521:12:
warning: passing argument 7 of ‘Slab_MultiTransfer’ from incompatible pointer
type [enabled by default]
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/Slab/src/slab.h:128:1:
note: expected ‘const void * const restrict* const restrict’ but argument is of
type ‘void * restrict* restrict’
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:544:10:
warning: passing argument 10 of ‘Hyperslab_GlobalMappingByIndex’ from
incompatible pointer type [enabled by default]
/Users/eschnett/EinsteinToolkit-hg/configs/sim-gcc46/bindings/include/RotatingSymmetry180_Prototypes.h:61:5:
note: expected ‘int *’ but argument is of type ‘int (*)[3]’
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:716:3:
warning: passing argument 1 of ‘free’ discards ‘restrict’ qualifier from
pointer target type [enabled by default]
/usr/include/stdlib.h:160:6: note: expected ‘void *’ but argument is of type
‘void * restrict* restrict’
/Users/eschnett/EinsteinToolkit-hg/arrangements/CactusNumerical/RotatingSymmetry180/src/rotatingsymmetry180.c:19:5:
internal compiler error: in build_vector_from_val, at tree.c:1380
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

The same code compiles fine with gcc 4.5.2. It also compiles fine if I use -O2
instead of -O3.

[Bug tree-optimization/48098] internal compiler error: in build_vector_from_val, at tree.c:1380

[schnetter at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

--- Comment #1 from Erik Schnetter <schnetter at gmail dot com> 2011-03-12 23:30:41 UTC ---
Here is a reduced test case:

void BndRot180VI ()
{
  static char *restrict * slab_setups;
  static int num_slab_setups = 0;
  num_slab_setups = GetRefinementLevels (0);
  slab_setups = malloc (num_slab_setups);
  for (int rl=0; rl<num_slab_setups; ++rl) {
    slab_setups[rl] = 0;
  }
}

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
             Status|UNCONFIRMED                 |NEW
      Known to work|                            |4.5.2
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|                            |2011.03.14 09:21:19
                 CC|                            |jakub at gcc dot gnu.org
     Ever Confirmed|0                           |1
            Summary|internal compiler error: in |[4.6 Regression] internal
                   |build_vector_from_val, at   |compiler error: in
                   |tree.c:1380                 |build_vector_from_val, at
                   |                            |tree.c:1380
   Target Milestone|---                         |4.6.0
      Known to fail|                            |4.6.0

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-14 09:21:19 UTC ---
Caused by http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=166136
together with the earlier
http://gcc.gnu.org/viewcvs?view=revision&revision=166102
change.

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-14 09:29:02 UTC ---
Smaller testcase:

void
foo (int n)
{
  static char *__restrict *p;
  int i;
  p = __builtin_malloc (n);
  for (i = 0; i < n; i++)
    p[i] = 0;
}

The reason why the conversion between char * types isn't useless is that one of
the pointers is TYPE_RESTRICT, while the other one is not.

--- gcc/tree.c.jj    2011-03-11 12:16:39.000000000 +0100
+++ gcc/tree.c    2011-03-14 10:27:25.000000000 +0100
@@ -1376,8 +1376,8 @@ build_vector_from_val (tree vectype, tre
   if (sc == error_mark_node)
     return sc;

-  gcc_assert (useless_type_conversion_p (TREE_TYPE (sc),
-                     TREE_TYPE (vectype)));
+  gcc_assert (useless_type_conversion_p (TYPE_MAIN_VARIANT (TREE_TYPE (sc)),
+                     TYPE_MAIN_VARIANT (TREE_TYPE (vectype))));

   v = VEC_alloc (constructor_elt, gc, nunits);
   for (i = 0; i < nunits; ++i)

fixes the ICE, but I wonder whether it isn't instead a bug in the vectorizer.

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P3
   Target Milestone|4.6.0                       |---

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-14 09:38:12 UTC ---
Though, apparently make_vector_type intentionally uses TYPE_MAIN_VARIANT of
innertype as VECTOR_TYPE's type, and puts the quals on the vector type itself.

Richard, what do you prefer?

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.6.0

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-03-14 09:59:32 UTC ---
Just swapping the order of arguments to useless_type_conversion_p works too.
If we consider build_vector_from_val as conversion from sc's type to the inner
type of the vector, then vectype's inner type as outer_type needs to go first:

--- gcc/tree.c.jj    2011-03-11 12:16:39.000000000 +0100
+++ gcc/tree.c    2011-03-14 10:57:21.000000000 +0100
@@ -1376,8 +1376,8 @@ build_vector_from_val (tree vectype, tre
   if (sc == error_mark_node)
     return sc;

-  gcc_assert (useless_type_conversion_p (TREE_TYPE (sc),
-                     TREE_TYPE (vectype)));
+  gcc_assert (useless_type_conversion_p (TREE_TYPE (vectype),
+                     TREE_TYPE (sc)));

   v = VEC_alloc (constructor_elt, gc, nunits);
   for (i = 0; i < nunits; ++i)

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

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

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

--- Comment #6 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-14 10:43:09 UTC ---
(In reply to comment #5)
> Just swapping the order of arguments to useless_type_conversion_p works too.
> If we consider build_vector_from_val as conversion from sc's type to the inner
> type of the vector, then vectype's inner type as outer_type needs to go first:
> 
> --- gcc/tree.c.jj    2011-03-11 12:16:39.000000000 +0100
> +++ gcc/tree.c    2011-03-14 10:57:21.000000000 +0100
> @@ -1376,8 +1376,8 @@ build_vector_from_val (tree vectype, tre
>    if (sc == error_mark_node)
>      return sc;
> 
> -  gcc_assert (useless_type_conversion_p (TREE_TYPE (sc),
> -                     TREE_TYPE (vectype)));
> +  gcc_assert (useless_type_conversion_p (TREE_TYPE (vectype),
> +                     TREE_TYPE (sc)));
> 
>    v = VEC_alloc (constructor_elt, gc, nunits);
>    for (i = 0; i < nunits; ++i)

The assert is supposed to make sure that in vectorized code vector extracts
use the correct type for assignments to scalars.  Thus if we have before
vectorization

  scalar = X;

and we vectorized the code that produced X we should use a vector type
for the vector result X that has an element type that is trivially
convertible to that of the above scalar.  Thus, before and after
vectorization

  useless_type_conversion_p (type-of-scalar, type-of-vector-element-type)

should be true.  The scalar X is sc above which is trivially convertible
to scalar, so if type-of-vector-element-type is trivially convertible to
X then it's trivially convertible to scalar.

Thus the assert is correct.

I think treating restrict as ordinary qualifier in make_vector_type is
bogus, similarly not properly maintaining a ref-all "qualification".
I suppose nobody thought of pointer vector element types before.

I'll cook up a patch.

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

--- Comment #7 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-14 10:50:57 UTC ---
It doesn't seem to strip ref-all, that isn't a variant kind.

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-14 10:58:06 UTC ---
Hm, we also want the elements to be trivially convertible to the vector
element type.  So we really want to assert types_compatible_p.

I can't see any real problems with the restrict case in question, apart from
maybe ICEing in a different place.  Any "fix" at this point seems to be
more risky than just disabling the assert.

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

--- Comment #9 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-14 12:22:32 UTC ---
Author: rguenth
Date: Mon Mar 14 12:20:48 2011
New Revision: 170934

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=170934
Log:
2011-03-14  Richard Guenther  <rguenther@suse.de>

    PR middle-end/48098
    * tree.c (build_vector_from_val): Adjust assert to requirements
    and reality.

    * gcc.dg/torture/pr48098.c: New testcase.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr48098.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree.c

[Bug tree-optimization/48098] [4.6 Regression] internal compiler error: in build_vector_from_val, at tree.c:1380

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48098

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

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

--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-14 12:35:29 UTC ---
Fixed.