[Bug target/111260] New: arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[Bug target/111260] New: arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

            Bug ID: 111260
           Summary: arm: ice in maybe_legitimize_operand, at
                    optabs.cc:8054
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: dcb314 at hotmail dot com
  Target Milestone: ---

Created attachment 55827
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=55827&action=edit
C source code

For the attached C code, on a raspberry Pi 3, with
compiler flag -O2 and recent gcc, does this:

bug952.c:809:17: internal compiler error: in maybe_legitimize_operand, at
optabs.cc:8054
  809 | static uint32_t  func_16(int8_t * p_17, int64_t  p_18)
      |                 ^~~~~~~
0x8811b3 maybe_legitimize_operand
        ../../trunk/gcc/optabs.cc:8054
0x8811b3 maybe_legitimize_operands(insn_code, unsigned int, unsigned int,
expand_operand*)
        ../../trunk/gcc/optabs.cc:8191
0x87dfa7 maybe_gen_insn(insn_code, unsigned int, expand_operand*)
        ../../trunk/gcc/optabs.cc:8210
0x87ecff maybe_expand_insn(insn_code, unsigned int, expand_operand*)
        ../../trunk/gcc/optabs.cc:8265

The bug first seems to appear sometime between g:bb0da4542df887a5
and g:7c04da768c1fc22e

[Bug target/111260] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[dcb314 at hotmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

David Binderman <dcb314 at hotmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |needs-bisection

--- Comment #1 from David Binderman <dcb314 at hotmail dot com> ---
Reduced C code seems to be:

long long g_263;
func_16_l_1358;
func_1() {
  int p_18 = 357942, __trans_tmp_5 = p_18 - (p_18 != g_263);
  if (__trans_tmp_5 <= func_16_l_1358)
    func_16_l_1358 = 0;
}

[Bug target/111260] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[zsojka at seznam dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

Zdenek Sojka <zsojka at seznam dot cz> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |zsojka at seznam dot cz

--- Comment #2 from Zdenek Sojka <zsojka at seznam dot cz> ---
Different testcase:
$ cat testcase.c
long
bar (long x)
{
  x *= x == 0 / 0;
  return (x);
}

long
foo (void)
{
  return bar (822920);
}

$ aarch64-unknown-linux-gnu-gcc -O testcase.c 
testcase.c: In function 'bar':
testcase.c:4:15: warning: division by zero [-Wdiv-by-zero]
    4 |   x *= x == 0 / 0;
      |               ^
during RTL pass: expand
testcase.c: In function 'foo':
testcase.c:11:10: internal compiler error: in maybe_legitimize_operand, at
optabs.cc:8046
   11 |   return bar (822920);
      |          ^~~~~~~~~~~~
0x7ff794 maybe_legitimize_operand
        /repo/gcc-trunk/gcc/optabs.cc:8046
0x7ff794 maybe_legitimize_operands(insn_code, unsigned int, unsigned int,
expand_operand*)
        /repo/gcc-trunk/gcc/optabs.cc:8194
...

[Bug target/111260] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Zdenek Sojka from comment #2)
> Different testcase:
> $ cat testcase.c
> long
> bar (long x)
> {
>   x *= x == 0 / 0;
>   return (x);
> }
> 
> long
> foo (void)
> {
>   return bar (822920);
> }
> 
> $ aarch64-unknown-linux-gnu-gcc -O testcase.c 
> testcase.c: In function 'bar':
> testcase.c:4:15: warning: division by zero [-Wdiv-by-zero]
>     4 |   x *= x == 0 / 0;
>       |               ^
> during RTL pass: expand
> testcase.c: In function 'foo':
> testcase.c:11:10: internal compiler error: in maybe_legitimize_operand, at
> optabs.cc:8046
>    11 |   return bar (822920);
>       |          ^~~~~~~~~~~~
> 0x7ff794 maybe_legitimize_operand
>         /repo/gcc-trunk/gcc/optabs.cc:8046
> 0x7ff794 maybe_legitimize_operands(insn_code, unsigned int, unsigned int,
> expand_operand*)
>         /repo/gcc-trunk/gcc/optabs.cc:8194
> ...

That is 100% a different bug and should be filed separately.

[Bug target/111260] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[zsojka at seznam dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

--- Comment #4 from Zdenek Sojka <zsojka at seznam dot cz> ---
(In reply to Andrew Pinski from comment #3)
> (In reply to Zdenek Sojka from comment #2)

...

> > ...
> 
> That is 100% a different bug and should be filed separately.

Than you for checking, I've created PR111711 for that

[Bug middle-end/111260] [14 Regression] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end
            Summary|arm: ice in                 |[14 Regression] arm: ice in
                   |maybe_legitimize_operand,   |maybe_legitimize_operand,
                   |at optabs.cc:8054           |at optabs.cc:8054
           Keywords|needs-bisection             |
   Target Milestone|---                         |14.0

[Bug middle-end/111260] [14 Regression] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|arm-linux-gnueabihf         |arm-*-* aarch64-*-*
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-10-06
     Ever confirmed|0                           |1
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=111711

--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Zdenek Sojka from comment #4)
> (In reply to Andrew Pinski from comment #3)
> > (In reply to Zdenek Sojka from comment #2)
> 
> ...
> 
> > > ...
> > 
> > That is 100% a different bug and should be filed separately.
> 
> Than you for checking, I've created PR111711 for that

Actually it looks like I might be wrong in the end.

For the reduced testcase in comment #1 we have:
  int __trans_tmp_5;
  long long int g_263.0_1;
  int func_16_l_1358.1_2;
  _Bool _4;

  <bb 2> [local count: 1073741824]:
  g_263.0_1 = g_263;
  _4 = g_263.0_1 == 357942;
  __trans_tmp_5_6 = _4 ? 357942 : 357941;


In this case, 357942 are 2 different modes (SImode and DImode).

Still causes by r14-2667-gceae1400cf24f329393e96dd9720 .

[Bug middle-end/111260] [14 Regression] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 111711 has been marked as a duplicate of this bug. ***

[Bug middle-end/111260] [14 Regression] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Better reduced testcase:
```
long f(int a)
{
  int b = 822920;
  int t = a == b;
  return t * (long)b;
}
```

[Bug middle-end/111260] [14 Regression] arm: ice in maybe_legitimize_operand, at optabs.cc:8054

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

--- Comment #8 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #7)
> Better reduced testcase:
> ```
> long f(int a)
> {
>   int b = 822920;
>   int t = a == b;
>   return t * (long)b;
> }
> ```

Here is one that ICEs on both arm and aarch64:
```
int f1(long long a)
{
  int b = 822920;
  int t = a == b;
  return t * (int)b;
}
```

[Bug target/111260] [14 Regression] arm/aarch64: ice in maybe_legitimize_operand, at optabs.cc:8054

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|middle-end                  |target
           Priority|P3                          |P1
           Keywords|                            |needs-bisection

[Bug middle-end/111260] [14 Regression] arm/aarch64: ice in maybe_legitimize_operand with ?: and constants and different types since r14-2667-gceae1400cf24f329393e96dd9720

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |middle-end

--- Comment #9 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The issue is inside emit_conditional_move where we are trying to use the same
register for constants which are of 2 different types (modes).

[Bug middle-end/111260] [14 Regression] arm/aarch64: ice in maybe_legitimize_operand with ?: and constants and different types since r14-2667-gceae1400cf24f329393e96dd9720

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

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

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
I have a fix which I think is correct.
Testing right now.

[Bug middle-end/111260] [14 Regression] arm/aarch64: ice in maybe_legitimize_operand with ?: and constants and different types since r14-2667-gceae1400cf24f329393e96dd9720

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                URL|                            |https://gcc.gnu.org/piperma
                   |                            |il/gcc-patches/2023-Decembe
                   |                            |r/640520.html
           Keywords|                            |patch

--- Comment #11 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Patch posted:
https://gcc.gnu.org/pipermail/gcc-patches/2023-December/640520.html

[Bug middle-end/111260] [14 Regression] arm/aarch64: ice in maybe_legitimize_operand with ?: and constants and different types since r14-2667-gceae1400cf24f329393e96dd9720

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

--- Comment #12 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The trunk branch has been updated by Andrew Pinski <pinskia@gcc.gnu.org>:

https://gcc.gnu.org/g:5fa27d9f8c4bec65887654e374146926d76690b0

commit r14-6562-g5fa27d9f8c4bec65887654e374146926d76690b0
Author: Andrew Pinski <quic_apinski@quicinc.com>
Date:   Tue Dec 12 21:55:50 2023 -0800

    middle-end: Fix up constant handling in emit_conditional_move [PR111260]

    After r14-2667-gceae1400cf24f329393e96dd9720, we force a constant to a
register
    if it is shared with one of the other operands. The problem is used the
comparison
    mode for the register but that could be different from the operand mode.
This
    causes some issues on some targets.
    To fix it, we need to make sure the mode of the comparison matches the mode
    of the other operands, before we can compare the constants (CONST_INT has
no
    modes so compare_rtx returns true if they have the same value even if the
usage
    is in a different mode).

    Bootstrapped and tested on both aarch64-linux-gnu and x86_64-linux.

            PR middle-end/111260

    gcc/ChangeLog:

            * optabs.cc (emit_conditional_move): Change the modes to be
            equal before forcing the constant to a register.

    gcc/testsuite/ChangeLog:

            * gcc.c-torture/compile/condmove-1.c: New test.

    Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>

[Bug middle-end/111260] [14 Regression] arm/aarch64: ice in maybe_legitimize_operand with ?: and constants and different types since r14-2667-gceae1400cf24f329393e96dd9720

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111260

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

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

--- Comment #13 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Fixed.