[Bug tree-optimization/54713] New: [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[Bug tree-optimization/54713] New: [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[ubizjak at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

             Bug #: 54713
           Summary: [4.8 Regression] error: non-trivial conversion at
                    assignment in gcc.c-torture/compile/pr53410-2.c
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: ubizjak@gmail.com
            Target: alpha-linux-gnu


Following testcase, derived from gcc.c-torture/compile/pr53410-2.c started to
ICE on alpha-linux-gnu recently.

--cut here--
typedef int V __attribute__((vector_size (4 * sizeof (int))));

void
f6 (V *p, V *q, V *r)
{
  *p = (*p & *r) == (*q & *r);
}
--cut here--

ICE can be triggered with a cross from x86_64-pc-linux-gnu:

~/gcc-build-alpha/gcc/cc1 -O2 -quiet tt.c
tt.c: In function ‘f6’:
tt.c:4:1: error: non-trivial conversion at assignment
 f6 (V *p, V *q, V *r)
 ^
int
vector(2) int
_24 = _20;

tt.c:4:1: error: non-trivial conversion at assignment
int
vector(2) int
_26 = _23;

tt.c:4:1: internal compiler error: verify_gimple failed
Please submit a full bug report,
...

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |4.8.0

--- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-26 08:58:49 UTC ---
ISTR alpha has BLKmode vectors.

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-09-26 09:19:55 UTC ---
I think this is a bug in fold_ternary_loc.  The thing is that forwprop tries to
simplify_bitfield_ref, _24 = BIT_FIELD_REF <_9, 32, 0>, where _9 is TImode
4xint
vector, _24 is int, and _9 = { _20, _23 }; where _20 and _23 are V2SI vectors.
Fold then assumes that a vector CONSTRUCTOR will only contain scalars, not
shorter vectors, and returns the first vector as whole.

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2012-09-26
         AssignedTo|unassigned at gcc dot       |jakub at gcc dot gnu.org
                   |gnu.org                     |
     Ever Confirmed|0                           |1

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-09-26 12:51:45 UTC ---
Created attachment 28280
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28280
gcc48-pr54713.patch

Untested fix (with testcases that ICE even on x86_64).

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #28280|0                           |1
        is obsolete|                            |

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-09-27 16:07:29 UTC ---
Created attachment 28293
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28293
gcc48-pr54713.patch

Updated patch.  I had to make the CONSTRUCTOR checking less strict, in
particular there are CONSTRUCTORS with (consecutive) indexes as well as NULL
indexes, there are CONSTRUCTORS with less than nunits elements even when the
elements are scalar.
But even with all this I'm still getting gfortran.dg/loc_2.f90 -O3 ICEs,
apparently SLP creates vector CONSTRUCTOR with type integer(kind=4) vector, but
elements logical(kind=4) (i.e. same TYPE_MODE, but INTEGER_TYPE vs.
BOOLEAN_TYPE.  Shall I make the verification even less strict (check TYPE_MODE
instead of useless_type_conversion?), or shall we fix the vectorizer?
Shouldn't I split the patch into the bugfixes + testcases and submit it before
the verification, so that the latter can be done more slowly?

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-09-28 08:36:28 UTC ---
(In reply to comment #4)
> Created attachment 28293 [details]
> gcc48-pr54713.patch
> 
> Updated patch.  I had to make the CONSTRUCTOR checking less strict, in
> particular there are CONSTRUCTORS with (consecutive) indexes as well as NULL
> indexes, there are CONSTRUCTORS with less than nunits elements even when the
> elements are scalar.

Ugh, less than nunits is probably ok for GENERIC but in GIMPLE we really
want trailing zeros explicit ... :/  Thus, if we need to "gimplify"
vector constructors we can as well clear their indexes?

> But even with all this I'm still getting gfortran.dg/loc_2.f90 -O3 ICEs,
> apparently SLP creates vector CONSTRUCTOR with type integer(kind=4) vector, but
> elements logical(kind=4) (i.e. same TYPE_MODE, but INTEGER_TYPE vs.
> BOOLEAN_TYPE.  Shall I make the verification even less strict (check TYPE_MODE
> instead of useless_type_conversion?), or shall we fix the vectorizer?

The vectorizer needs to be fixed.  See the comment in
get_vectype_for_scalar_type_and_size:

  /* For vector types of elements whose mode precision doesn't
     match their types precision we use a element type of mode
     precision.  The vectorization routines will have to make sure
     they support the proper result truncation/extension.
     We also make sure to build vector types with INTEGER_TYPE
     component type only.  */

> Shouldn't I split the patch into the bugfixes + testcases and submit it before
> the verification, so that the latter can be done more slowly?

Yes, I think separate patches would be prefered.

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-09-28 12:19:07 UTC ---
Author: jakub
Date: Fri Sep 28 12:18:57 2012
New Revision: 191826

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=191826
Log:
    PR tree-optimization/54713
    * fold-const.c (vec_cst_ctor_to_array): Give up if vector CONSTRUCTOR
    has vector elements.
    (fold_ternary_loc) <case BIT_FIELD_REF>: Likewise.
    * tree-vect-generic.c (vector_element): Don't rely on CONSTRUCTOR elts
    indexes.  Use BIT_FIELD_REF if CONSTRUCTOR has vector elements.
    (lower_vec_perm): Use NULL_TREE CONSTRUCTOR indexes.

    * gcc.c-torture/compile/pr54713-1.c: New test.
    * gcc.c-torture/compile/pr54713-2.c: New test.
    * gcc.c-torture/compile/pr54713-3.c: New test.

Added:
    trunk/gcc/testsuite/gcc.c-torture/compile/pr54713-1.c
    trunk/gcc/testsuite/gcc.c-torture/compile/pr54713-2.c
    trunk/gcc/testsuite/gcc.c-torture/compile/pr54713-3.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/fold-const.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/tree-vect-generic.c

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-02 09:18:38 UTC ---
Created attachment 28322
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28322
gcc48-pr54713.patch

Updated version of the verifier.  This one doesn't pass regtest, e.g.
g++.dg/torture/vshuf*.C fail, because non-gimplified CONSTRUCTORs with non-NULL
indexes get into the IL from DECL_INITIAL during fold_stmt.

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-10-02 13:43:13 UTC ---
Author: jakub
Date: Tue Oct  2 13:43:09 2012
New Revision: 191983

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=191983
Log:
    PR tree-optimization/54713
    * expr.c (categorize_ctor_elements_1): Don't assume purpose is
    non-NULL.
    * tree-cfg.c (verify_gimple_assign_single): Add verification of
    vector CONSTRUCTORs.
    * tree-ssa-sccvn.c (vn_reference_lookup_3): For VECTOR_TYPE
    CONSTRUCTORs, don't do anything if element type is VECTOR_TYPE,
    and don't check index.
    * tree-vect-slp.c (vect_get_constant_vectors): VIEW_CONVERT_EXPR
    ctor elements first if their type isn't compatible with vector
    element type.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/expr.c
    trunk/gcc/tree-cfg.c
    trunk/gcc/tree-ssa-sccvn.c
    trunk/gcc/tree-vect-slp.c

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[danglin at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

John David Anglin <danglin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |danglin at gcc dot gnu.org

--- Comment #9 from John David Anglin <danglin at gcc dot gnu.org> 2012-10-15 23:25:52 UTC ---
This test has recently started to fail on 32-bit hppa targets.

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |WAITING

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-07 13:29:52 UTC ---
Which test started failing?  I don't see either pr53410 or pr54713 in
http://gcc.gnu.org/ml/gcc-testresults/2012-11/msg00082.html
nor
http://gcc.gnu.org/ml/gcc-testresults/2012-10/msg02866.html

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[dave.anglin at bell dot net]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

--- Comment #11 from dave.anglin at bell dot net 2012-11-07 17:17:47 UTC ---
On 11/7/2012 8:29 AM, jakub at gcc dot gnu.org wrote:
> Which test started failing?
pr53410 was failing here
http://gcc.gnu.org/ml/gcc-testresults/2012-10/msg02016.html
but the problem seems to have gone away in later runs.

Dave

[Bug tree-optimization/54713] [4.8 Regression] error: non-trivial conversion at assignment in gcc.c-torture/compile/pr53410-2.c

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54713

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

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

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-07 17:19:37 UTC ---
Fixed then.