[Bug target/111268] New: internal compiler error: in to_constant, at poly-int.h:504

[Bug target/111268] New: internal compiler error: in to_constant, at poly-int.h:504

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

            Bug ID: 111268
           Summary: internal compiler error: in to_constant, at
                    poly-int.h:504
           Product: gcc
           Version: 13.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: malat at debian dot org
  Target Milestone: ---

I am well aware of PR/109176 but it looks somewhat different. Reporting it
here:

ref:
*
https://buildd.debian.org/status/fetch.php?pkg=highway&arch=arm64&ver=1.0.8%7Egit20230830.fed142a-1&stamp=1693567540&raw=0

In file included from /<<PKGBUILDDIR>>/hwy/foreach_target.h:207,
                 from /<<PKGBUILDDIR>>/hwy/tests/shift_test.cc:18:
/<<PKGBUILDDIR>>/hwy/tests/shift_test.cc: In function 'void
hwy::N_SVE2_128::TestRotateRight::operator()(T, D) [with T = long unsigned int;
D = hwy::N_SVE2_128::Simd<long unsigned int, 2, 0>]':
/<<PKGBUILDDIR>>/hwy/tests/shift_test.cc:157:34: internal compiler error: in
to_constant, at poly-int.h:504
0xe96ec3 internal_error(char const*, ...)
        ???:0
0xe81f63 fancy_abort(char const*, int, char const*)
        ???:0
0x19f174f vect_slp_analyze_operations(vec_info*)
        ???:0
0x17b9f63 vect_slp_function(function*)
        ???:0
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <file:///usr/share/doc/gcc-snapshot/README.Bugs> for instructions.

[Bug target/111268] internal compiler error: in to_constant, at poly-int.h:504

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #1 from Mathieu Malaterre <malat at debian dot org> ---
Created attachment 55830
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=55830&action=edit
preprocessed source

% /usr/lib/gcc-snapshot/bin/g++ -O2 -g -DNDEBUG -c testcase.i

[Bug target/111268] internal compiler error: in to_constant, at poly-int.h:504

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #2 from Mathieu Malaterre <malat at debian dot org> ---
% /usr/lib/gcc-snapshot/bin/g++ --version
g++ (Debian 20230811-1) 14.0.0 20230811 (experimental) [master
r14-3139-g68783211f66]
Copyright (C) 2023 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

[Bug target/111268] internal compiler error: in to_constant, at poly-int.h:504

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #3 from Mathieu Malaterre <malat at debian dot org> ---
For reference

% /usr/lib/gcc-snapshot/bin/gcc -march=armv8.2-a+sve -O2 -c m.c && echo
"success"
success

with

% cat m.c
#pragma GCC aarch64 "arm_sve.h"

svbool_t
foo (svint8_t a, svint8_t b, svbool_t c)
{
  svbool_t d = svcmplt_s8 (svptrue_pat_b8 (SV_ALL), a, b);
  return svsel_b (d, c, d);
}

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code,
                   |                            |needs-bisection
          Component|target                      |tree-optimization
   Target Milestone|---                         |14.0
            Summary|internal compiler error: in |[14 Regression] internal
                   |to_constant, at             |compiler error: in
                   |poly-int.h:504              |to_constant, at
                   |                            |poly-int.h:504

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Looks like this one is already fixed.

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #5 from Mathieu Malaterre <malat at debian dot org> ---
(In reply to Andrew Pinski from comment #4)
> Looks like this one is already fixed.

Same symptoms with 20230902 snapshot:

*
https://buildd.debian.org/status/fetch.php?pkg=highway&arch=arm64&ver=1.0.8%7Egit20230830.fed142a-1&stamp=1693905326&raw=0

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
reducing ...

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Interesting:
/* We checked above that the vectors are constant-length.  */
unsigned vnunits = TYPE_VECTOR_SUBPARTS (vtype).to_constant ();

But it is not constant ...

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-09-07
     Ever confirmed|0                           |1

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Reduced testcase:
```
#include <arm_sve.h>

void f(unsigned long *expected, svuint64_t values) {
  svst1_u64(svptrue_b64(), expected, values);
  expected[0] = (expected[0]+1);
  expected[1] = (expected[1]+1);
}
```

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
   Last reconfirmed|2023-09-07 00:00:00         |2023-10-17

--- Comment #9 from Richard Biener <rguenth at gcc dot gnu.org> ---
Re-confirmed.

during GIMPLE pass: slp
t.c: In function 'f':
t.c:3:6: internal compiler error: in to_constant, at poly-int.h:588
    3 | void f(unsigned long *expected, svuint64_t values) {
      |      ^
0xe915f8 poly_int<2u, unsigned long>::to_constant() const
        /space/rguenther/src/gcc/gcc/poly-int.h:588
0x1aedf84 vectorizable_slp_permutation_1
        /space/rguenther/src/gcc/gcc/tree-vect-slp.cc:8766
0x1aeee54 vectorizable_slp_permutation
        /space/rguenther/src/gcc/gcc/tree-vect-slp.cc:8931
0x1ae4321 vect_slp_analyze_node_operations_1
        /space/rguenther/src/gcc/gcc/tree-vect-slp.cc:6049

the comment is outdated, we check

  unsigned olanes = ncopies * SLP_TREE_LANES (node);
  gcc_assert (repeating_p || multiple_p (olanes, nunits));

but then we have to adjust the following to poly arith somehow:

          if (repeating_p)
            vperm.quick_push ({{p.first, 0}, p.second + active_lane[p.first]});
          else
            {
              /* We checked above that the vectors are constant-length.  */
              unsigned vnunits = TYPE_VECTOR_SUBPARTS (vtype).to_constant ();
              unsigned vi = (active_lane[p.first] + p.second) / vnunits;
              unsigned vl = (active_lane[p.first] + p.second) % vnunits;
              vperm.quick_push ({{p.first, vi}, vl});

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #10 from Richard Biener <rguenth at gcc dot gnu.org> ---
Btw, we have

(gdb) p debug (node)
t.c:3:6: note: node 0x3f341d0 (max_nunits=1, refcnt=1) vector(2) long unsigned
int
t.c:3:6: note: op: VEC_PERM_EXPR
t.c:3:6: note:  stmt 0 _1 = BIT_FIELD_REF <values_7(D), 64, 0>;
t.c:3:6: note:  stmt 1 _4 = BIT_FIELD_REF <values_7(D), 64, 64>;
t.c:3:6: note:  lane permutation { 0[0] 0[1] }
t.c:3:6: note:  children 0x3f34260
$1 = void
(gdb) p vtype
$2 = <vector_type 0x7ffff6ca6bd0 svuint64_t>

it looks like "compatibility" somehow requires/allows us to V_C_E the
permuted argument to a fixed length vector?  Otherwise vect_build_slp_tree_1
of course could check the BIT_FIELD_REF args are fixed length.

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |testsuite-fail

--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
testsuite/gcc.dg/torture/neon-sve-bridge.c ICEs at -O2 and above because of
this bug.

[Bug tree-optimization/111268] [14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

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

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

--- Comment #12 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The #c8 testcase started to ICE with
r13-926-g08afab6f8642f58f702010ec196dce3b00955627
The #c3 one started to ICE with
r11-1445-g502d63b6d6141597bb18fd23c87736a1b384cf8f
So, not really sure why is this marked as [14 Regression] only.

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to work|12.3.0                      |
            Summary|[14 Regression] internal    |[11/12/13/14 Regression]
                   |compiler error: in          |internal compiler error: in
                   |to_constant, at             |to_constant, at
                   |poly-int.h:504              |poly-int.h:504
   Target Milestone|14.0                        |11.5
           Keywords|                            |ice-checking

--- Comment #13 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #12)
> The #c8 testcase started to ICE with
> r13-926-g08afab6f8642f58f702010ec196dce3b00955627
> The #c3 one started to ICE with
> r11-1445-g502d63b6d6141597bb18fd23c87736a1b384cf8f
> So, not really sure why is this marked as [14 Regression] only.

Because I didn't notice the ICE is due to an assert which is only enabled with
checking:
  gcc_checking_assert (is_constant ());

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P2

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |adhemerval.zanella at linaro dot o
                   |                            |rg

--- Comment #14 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 113489 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[ricbal02 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Richard Ball <ricbal02 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ricbal02 at gcc dot gnu.org

--- Comment #15 from Richard Ball <ricbal02 at gcc dot gnu.org> ---
The following patch fixes the neon-sve-bridge.c regression around this bug.
https://sourceware.org/pipermail/gcc-patches/2024-January/644432.html

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[ricbal02 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #16 from Richard Ball <ricbal02 at gcc dot gnu.org> ---
Confirmed that approved patch:
https://sourceware.org/pipermail/gcc-patches/2024-February/644711.html

Fixes neon-sve-bridge.c error and the error in #c8.
The error in #c3 appears to already be fixed.

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[ricbal02 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #17 from Richard Ball <ricbal02 at gcc dot gnu.org> ---
(In reply to Richard Ball from comment #16)
> Confirmed that approved patch:
> https://sourceware.org/pipermail/gcc-patches/2024-February/644711.html
> 
> Fixes neon-sve-bridge.c error and the error in #c8.
> The error in #c3 appears to already be fixed.

Apologies this approved patch:
https://sourceware.org/pipermail/gcc-patches/2024-February/644714.html

[Bug tree-optimization/111268] [11/12/13/14 Regression] internal compiler error: in to_constant, at poly-int.h:504

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #18 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Ball <ricbal02@gcc.gnu.org>:

https://gcc.gnu.org/g:4571b4d413a4ba5f1e2d429a2623180ad1c73c0f

commit r14-8703-g4571b4d413a4ba5f1e2d429a2623180ad1c73c0f
Author: Richard Ball <richard.ball@arm.com>
Date:   Thu Feb 1 17:18:28 2024 +0000

    middle-end: Fix ICE in poly-int.h due to SLP.

    Adds a check to ensure that the input vector arguments
    to a function are not variable length. Previously, only the
    output vector of a function was checked.

    The ICE in question is within the neon-sve-bridge.c test,
    and is related to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

    gcc/ChangeLog:
            PR tree-optimization/111268
            * tree-vect-slp.cc (vectorizable_slp_permutation_1):
            Add variable-length check for vector input arguments
            to a function.

[Bug tree-optimization/111268] [11/12/13 Regression] internal compiler error: in to_constant, at poly-int.h:504

[ricbal02 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

Richard Ball <ricbal02 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[11/12/13/14 Regression]    |[11/12/13 Regression]
                   |internal compiler error: in |internal compiler error: in
                   |to_constant, at             |to_constant, at
                   |poly-int.h:504              |poly-int.h:504

--- Comment #19 from Richard Ball <ricbal02 at gcc dot gnu.org> ---
Fixed on trunk so far.

[Bug tree-optimization/111268] [11/12/13 Regression] internal compiler error: in to_constant, at poly-int.h:504

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

--- Comment #20 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:b0632c06a7f61d3b42170d51aa5b88237a722ae0

commit r13-8725-gb0632c06a7f61d3b42170d51aa5b88237a722ae0
Author: Richard Ball <richard.ball@arm.com>
Date:   Thu Feb 1 17:18:28 2024 +0000

    middle-end: Fix ICE in poly-int.h due to SLP.

    Adds a check to ensure that the input vector arguments
    to a function are not variable length. Previously, only the
    output vector of a function was checked.

    The ICE in question is within the neon-sve-bridge.c test,
    and is related to https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

    gcc/ChangeLog:
            PR tree-optimization/111268
            * tree-vect-slp.cc (vectorizable_slp_permutation_1):
            Add variable-length check for vector input arguments
            to a function.

    (cherry picked from commit 4571b4d413a4ba5f1e2d429a2623180ad1c73c0f)

[Bug tree-optimization/111268] [11/12 Regression] internal compiler error: in to_constant, at poly-int.h:504

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111268

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED