[Bug tree-optimization/62021] New: ICE in verify_gimple_assign_single

[Bug tree-optimization/62021] New: ICE in verify_gimple_assign_single

[ysrumyan at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

            Bug ID: 62021
           Summary: ICE in verify_gimple_assign_single
           Product: gcc
           Version: 4.10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ysrumyan at gmail dot com

For attached simple test-case if we omit 'uniform' specification compiler
produces ICE:
 error: incorrect type of vector CONSTRUCTOR elements
Note that for stmt
_38 = {vect_cst_.62_39, vect_cst_.62_39};
we have type mismatch - type of rhs constructor element is vector(2) long
unsigned int whereas type of lhs is vector(4) float*.
If we add 'uniform' specification (through -DUNIFORM option) test compiles
successfully. I assume that ICE must be fixed.

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[ysrumyan at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

--- Comment #1 from Yuri Rumyantsev <ysrumyan at gmail dot com> ---
Created attachment 33247
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=33247&action=edit
test-case to reprroduce

Test should be compiled with
-O2 -fopenmp -march=core-avx2
options. If we add -DUNIFORM ICE will disappeared.

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

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

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

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  The constructor is built in vectorizable_simd_clone_call

2958                              vec_oprnd0 = build_constructor (atype,
ctor_elts);
2959                              new_stmt
2960                                = gimple_build_assign (make_ssa_name
(atype, NULL),
2961                                                       vec_oprnd0);

(gdb) p debug_generic_expr (atype)
vector(4) float *

huh.  The actual elements in the constructor are of type

vector(2) long unsigned int


Seems bestn->simdclone->args[i].vector_type seems to miss that we substitute
integer types for all pointer types in vectors (pointer element vectors are
disallowed)?

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The problem is that get_vectype_for_scalar_type_and_size for pointers just
returns vectors of pointer sized integers instead of vectors of pointers.
So, either we need to VCE it, or change the simd_clones to instead use vectors
of
pointer sized integers instead of vectors of pointers for the arguments.

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #3)
> The problem is that get_vectype_for_scalar_type_and_size for pointers just
> returns vectors of pointer sized integers instead of vectors of pointers.
> So, either we need to VCE it, or change the simd_clones to instead use
> vectors of
> pointer sized integers instead of vectors of pointers for the arguments.

The latter.  There are no vectors of pointers.

Richard.

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-12-09
                 CC|                            |mpolacek at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #5 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
So confirmed.

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |jakub at gcc dot gnu.org

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 34240
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34240&action=edit
gcc5-pr62021.patch

Untested fix.

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Wed Dec 10 20:01:12 2014
New Revision: 218603

URL: https://gcc.gnu.org/viewcvs?rev=218603&root=gcc&view=rev
Log:
    PR tree-optimization/62021
    * omp-low.c (simd_clone_adjust_return_type): Use
    vector of pointer_sized_int_node types instead vector of pointer
    types.
    (simd_clone_adjust_argument_types): Likewise.

    * gcc.dg/vect/pr62021.c: New test.

Added:
    trunk/gcc/testsuite/gcc.dg/vect/pr62021.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/omp-low.c
    trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Author: jakub
Date: Wed Dec 10 20:02:19 2014
New Revision: 218604

URL: https://gcc.gnu.org/viewcvs?rev=218604&root=gcc&view=rev
Log:
    PR tree-optimization/62021
    * omp-low.c (simd_clone_adjust_return_type): Use
    vector of pointer_sized_int_node types instead vector of pointer
    types.
    (simd_clone_adjust_argument_types): Likewise.

    * gcc.dg/vect/pr62021.c: New test.

Added:
    branches/gcc-4_9-branch/gcc/testsuite/gcc.dg/vect/pr62021.c
Modified:
    branches/gcc-4_9-branch/gcc/ChangeLog
    branches/gcc-4_9-branch/gcc/omp-low.c
    branches/gcc-4_9-branch/gcc/testsuite/ChangeLog

[Bug tree-optimization/62021] ICE in verify_gimple_assign_single

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=62021

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED
   Target Milestone|---                         |4.9.3

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed.