[Bug tree-optimization/106878] New: ICE: verify_gimple failed at -O2 on arm-none-eabi

[Bug tree-optimization/106878] New: ICE: verify_gimple failed at -O2 on arm-none-eabi

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

            Bug ID: 106878
           Summary: ICE: verify_gimple failed at -O2 on arm-none-eabi
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: acoplan at gcc dot gnu.org
  Target Milestone: ---

The following fails for GCC 11 and above (indeed still on the trunk):

$ gcc/xgcc -B gcc -v
Reading specs from gcc/specs
COLLECT_GCC=gcc/xgcc
COLLECT_LTO_WRAPPER=gcc/lto-wrapper
Target: arm-none-eabi
Configured with: /home/alecop01/toolchain/src/gcc/configure
--prefix=/data_sdb/toolchain/cc1s/dbg-arm --enable-languages=c,c++
--disable-bootstrap --target=arm-none-eabi
Thread model: single
Supported LTO compression algorithms: zlib
gcc version 13.0.0 20220907 (experimental) (GCC)

$ cat t.c
int a;
void b(const int *c) {
  unsigned d = ((long)c | (long)&a) & 65535 << 16;
  long e = (long)c;
  if (d != (e & 65535 << 16))
    return;
}

$ gcc/xgcc -B gcc -c t.c -O2 -S -o /dev/null
t.c: In function ‘b’:
t.c:7:1: error: invalid types for ‘bit_not_expr’
    7 | }
      | ^
const int *
const int *
_12 = ~c_7(D);
during GIMPLE pass: forwprop
t.c:7:1: internal compiler error: verify_gimple failed
0xe0c855 verify_gimple_in_cfg(function*, bool)
        /home/alecop01/toolchain/src/gcc/gcc/tree-cfg.cc:5568
0xca78be execute_function_todo
        /home/alecop01/toolchain/src/gcc/gcc/passes.cc:2091
0xca8342 execute_todo
        /home/alecop01/toolchain/src/gcc/gcc/passes.cc:2145
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

$ gcc/xgcc -B gcc -c t.c -O2 -S -o /dev/null -fno-checking
during GIMPLE pass: evrp
t.c: In function ‘b’:
t.c:7:1: internal compiler error: Segmentation fault
    7 | }
      | ^
0xdb06ef crash_signal
        /home/alecop01/toolchain/src/gcc/gcc/toplev.cc:314
0x1a70ca1 range_op_handler::fold_range(vrange&, tree_node*, vrange const&,
vrange const&, relation_kind_t) const
        /home/alecop01/toolchain/src/gcc/gcc/range-op.cc:4217
0x1a755cb operator_bitwise_not::fold_range(irange&, tree_node*, irange const&,
irange const&, relation_kind_t) const
        /home/alecop01/toolchain/src/gcc/gcc/range-op.cc:3507
0x1948cd7 fold_using_range::range_of_range_op(vrange&, gimple*, fur_source&)
        /home/alecop01/toolchain/src/gcc/gcc/gimple-range-fold.cc:611
0x194ae68 fold_using_range::fold_stmt(vrange&, gimple*, fur_source&,
tree_node*)
        /home/alecop01/toolchain/src/gcc/gcc/gimple-range-fold.cc:555
0x19358ec gimple_ranger::fold_range_internal(vrange&, gimple*, tree_node*)
        /home/alecop01/toolchain/src/gcc/gcc/gimple-range.cc:258
0x19371f0 gimple_ranger::range_of_stmt(vrange&, gimple*, tree_node*)
        /home/alecop01/toolchain/src/gcc/gcc/gimple-range.cc:319
0x113ec83 range_query::value_of_stmt(gimple*, tree_node*)
        /home/alecop01/toolchain/src/gcc/gcc/value-query.cc:135
0xfcda70 substitute_and_fold_dom_walker::before_dom_children(basic_block_def*)
        /home/alecop01/toolchain/src/gcc/gcc/tree-ssa-propagate.cc:816
0x18f9e47 dom_walker::walk(basic_block_def*)
        /home/alecop01/toolchain/src/gcc/gcc/domwalk.cc:311
0xfcc8b5 substitute_and_fold_engine::substitute_and_fold(basic_block_def*)
        /home/alecop01/toolchain/src/gcc/gcc/tree-ssa-propagate.cc:987
0x10f3430 execute_ranger_vrp(function*, bool)
        /home/alecop01/toolchain/src/gcc/gcc/tree-vrp.cc:4349
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 on arm-none-eabi

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.4
            Summary|ICE: verify_gimple failed   |[11/12/13 Regression] ICE:
                   |at -O2 on arm-none-eabi     |verify_gimple failed at -O2
                   |                            |on arm-none-eabi
      Known to work|                            |10.1.0, 10.4.0
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-09-07
             Target|arm-*-*                     |arm-*-* x86_64-linux-gnu
                   |                            |i?86-linux-gnu
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fails on x86_64 with -m32.
Also it worked in GCC 10.4.0 (I checked the IR to see there was no BIT_NOT_EXPR
too).

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

Alex Coplan <acoplan at gcc dot gnu.org> changed:

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

--- Comment #2 from Alex Coplan <acoplan at gcc dot gnu.org> ---
Started with r11-35-g496f4f884716ae061f771a62e44868a32dbd502f

commit 496f4f884716ae061f771a62e44868a32dbd502f
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Mon May 4 11:01:08 2020 +0200

    match.pd: Decrease number of nop conversions around bitwise ops [PR94718]

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Adjusted testcase that ICEs also on x86_64:
typedef __INTPTR_TYPE__ intptr_t;
typedef __UINTPTR_TYPE__ uintptr_t;
int a;

int
foo (const int *c)
{
  uintptr_t d = ((intptr_t) c | (intptr_t) &a) & 65535 << 16;
  intptr_t e = (intptr_t) c;
  if (d != (e & 65535 << 16))
    return 1;
  return 0;
}

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 53550
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53550&action=edit
gcc13-pr106878-1.patch

Ugh, this is a mess.  The verifiers do allow BIT_AND_EXPR (intentionally for
realignment) and BIT_{I,X}OR_EXPR (just because it is similar) on pointer
types,
but since r12-1608-g2f1686ff70b25f BIT_NOT_EXPR is rejected on pointers.

I'm afraid allowing just BIT_AND_EXPR and disallowing BIT_{I,X}OR_EXPR or
BIT_NOT_EXPR would be a lot of work, this patch is what I spotted during
skimming of match.pd to make BIT_{AND,IOR,XOR}_EXPR work on pointer types and
not BIT_NOT_EXPR.

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 53551
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53551&action=edit
gcc13-pr106878-2.patch

Or we can instead just tweak my r11-35-g496f4f884716ae06 change and avoid
creating pointer BIT_*_EXPR in that case.  That fixes the testcase, but makes
all this stuff latent...

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 53552
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53552&action=edit
gcc13-pr106878-3.patch

Or we can revert parts of Andrew's change and allow BIT_NOT_EXPR on pointers
and OFFSET_TYPE, because it is just weird if the 3 BIT_*_EXPR are inconsitent
with it.
Except this ICEs in the ranger, which wants to treat BIT_NOT_EXPR as
~X is simply -1 - X except that MINUS_EXPR isn't supported on pointers.

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 53569
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53569&action=edit
gcc13-pr106878-4.patch

So, I've tried to restrict the verifiers so that they allow pointer arguments
only for BIT_AND_EXPR with constant second argument and not for BIT_IOR_EXPR
nor BIT_XOR_EXPR.  Ran into reassoc issue which was converting (ptr1 == 0 &&
ptr2 == 0) into (ptr1 | ptr2) == 0 with BIT_IOR_EXPR in pointer type.
This untested patch changes that, will see tonight how far I get with this
adjusted patch.

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #8 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jakub Jelinek <jakub@gcc.gnu.org>:

https://gcc.gnu.org/g:645ef01a463f15fc230e2155719c7a12cec89acf

commit r13-2658-g645ef01a463f15fc230e2155719c7a12cec89acf
Author: Jakub Jelinek <jakub@redhat.com>
Date:   Wed Sep 14 12:36:36 2022 +0200

    Disallow pointer operands for |, ^ and partly & [PR106878]

    My change to match.pd (that added the two simplifications this patch
    touches) results in more |/^/& assignments with pointer arguments,
    but since r12-1608 we reject pointer operands for BIT_NOT_EXPR.

    Disallowing them for BIT_NOT_EXPR and allowing for BIT_{IOR,XOR,AND}_EXPR
    leads to a match.pd maintainance nightmare (see one of the patches in the
    PR), so either we want to allow pointer operand on BIT_NOT_EXPR (but then
    we run into issues e.g. with the ranger which expects it can emulate
    BIT_NOT_EXPR ~X as - 1 - X which doesn't work for pointers which don't
    support MINUS_EXPR), or the following patch disallows pointer arguments
    for all of BIT_{IOR,XOR,AND}_EXPR with the exception of BIT_AND_EXPR
    with INTEGER_CST last operand (for simpler pointer realignment).
    I had to tweak one reassoc optimization and the two match.pd
    simplifications.

    2022-09-14  Jakub Jelinek  <jakub@redhat.com>

            PR tree-optimization/106878
            * tree-cfg.cc (verify_gimple_assign_binary): Disallow pointer,
            reference or OFFSET_TYPE BIT_IOR_EXPR, BIT_XOR_EXPR or, unless
            the second argument is INTEGER_CST, BIT_AND_EXPR.
            * match.pd ((type) X op CST -> (type) (X op ((type-x) CST)),
            (type) (((type2) X) op Y) -> (X op (type) Y)): Punt for
            POINTER_TYPE_P or OFFSET_TYPE.
            * tree-ssa-reassoc.cc (optimize_range_tests_cmp_bitwise): For
            pointers cast them to pointer sized integers first.

            * gcc.c-torture/compile/pr106878.c: New test.

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jwhitakera at gmail dot com

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 107478 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/106878] [11/12/13 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise caluclation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |gcc-hbfypl at hansdejong dot eu

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 105663 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[11/12/13 Regression] ICE:  |[11/12 Regression] ICE:
                   |verify_gimple failed at -O2 |verify_gimple failed at -O2
                   |with pointers and bitwise   |with pointers and bitwise
                   |caluclation                 |calculation

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Fixed on the trunk I believe.

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |julien.staub at se dot com

--- Comment #12 from Martin Liška <marxin at gcc dot gnu.org> ---
*** Bug 108340 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[vvinayag at arm dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

vvinayag at arm dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vvinayag at arm dot com

--- Comment #13 from vvinayag at arm dot com ---
Will this fix be backported to GCC 12 and GCC 11 ?

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-checking

--- Comment #14 from Richard Biener <rguenth at gcc dot gnu.org> ---
Iff then I think we want to backport all but the verifier changes.

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #15 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Just note this had various follow-ups.
r13-2658
r13-2709
r13-2891
at least.

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

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

--- Comment #16 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 109841 has been marked as a duplicate of this bug. ***

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|11.4                        |11.5

--- Comment #17 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
GCC 11.4 is being released, retargeting bugs to GCC 11.5.

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #18 from Alex Coplan <acoplan at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #15)
> Just note this had various follow-ups.
> r13-2658
> r13-2709
> r13-2891
> at least.

So for backports, it sounds like we want r13-2658 without the verify_gimple
changes, and the other two patches as is. Is that right? Would it make sense to
squash these if we were to backport them or should they be kept as separate
patches?

[Bug tree-optimization/106878] [11/12 Regression] ICE: verify_gimple failed at -O2 with pointers and bitwise calculation

[acoplan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106878

--- Comment #19 from Alex Coplan <acoplan at gcc dot gnu.org> ---
(In reply to Alex Coplan from comment #18)

> So for backports, it sounds like we want r13-2658 without the verify_gimple
> changes, and the other two patches as is. Is that right? Would it make sense
> to squash these if we were to backport them or should they be kept as
> separate patches?

FWIW I tested the backports as described above on the GCC 12 branch,
bootstrap/regtest passed on both x86_64-linux-gnu and aarch64-linux-gnu. Is it
OK to backport those patches to GCC 12?