[Bug c++/98216] New: [C++20] std::array<double, N> template parameter error with negative values

[Bug c++/98216] New: [C++20] std::array<double, N> template parameter error with negative values

[emmanuel.le-trong@cnrs-orleans.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

            Bug ID: 98216
           Summary: [C++20] std::array<double, N> template parameter error
                    with negative values
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: emmanuel.le-trong@cnrs-orleans.fr
  Target Milestone: ---

Sorry for the bad title, I can't figure out how to summarize this:

    #include <array>
    #include <cassert>

        using
    an_array = std::array <double, 1>;

        constexpr auto
    add_array (an_array a, an_array const& b)
    {
        a[0] += b[0];
        return a;
    }

        template <an_array A>
        struct
    wrapper
    {
            static constexpr auto
        arr = A;
    };


        template <an_array A1 ,an_array A2 >
        auto
    add (wrapper <A1> const&, wrapper <A2> const&)
    {
        return wrapper <add_array (A1, A2)> {};
    }

    constexpr auto ONE       = an_array {{  1. }};
    constexpr auto MINUS_ONE = an_array {{ -1. }}; 
    constexpr auto MINUS_TWO = an_array {{ -2. }}; 

        int
    main ()
    {
        auto a = wrapper <MINUS_TWO> {};
        auto b = wrapper <ONE> {};
        auto c = add (a, b);
        assert (c.arr == MINUS_TWO); // <- should fail
        assert (c.arr == MINUS_ONE); // <- should pass
    }

Compiled with version 11.0.0 20201209, the first assertion passes and the
second fails. It should be the opposite.

FWIW, if you replace double by float, it works as intended.

[Bug c++/98216] [C++20] std::array<double, N> template parameter error with negative values

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2020-12-09
     Ever confirmed|0                           |1
           Keywords|                            |wrong-code

[Bug c++/98216] [C++20] std::array<double, N> template parameter error with negative values

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I guess the issue has to be with you using 'static constexpr auto arr' and
that not being correctly templated on the expression.

Well, confirmed your observation.

[Bug c++/98216] [C++20] std::array<double, N> template parameter error with negative values

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Reduced:

template<typename T, int N>
struct array
{
  constexpr T& operator[](int n) { return d[n]; }
  constexpr const T& operator[](int n) const { return d[n]; }

  constexpr bool operator==(const array& a) const
  {
    for (int i = 0; i < N; ++i)
      if (d[i] != a.d[i])
        return false;
    return true;
  }

  T d[N];
};

using an_array = array <double, 1>;

constexpr auto
add_array (an_array a, an_array const& b)
{
  a[0] += b[0];
  return a;
}

template <an_array A>
struct wrapper
{
  static constexpr auto
    arr = A;
};


template <an_array A1 , an_array A2>
auto
add (wrapper <A1> const&, wrapper <A2> const&)
{
  return wrapper <add_array (A1, A2)> {};
}

constexpr auto ONE       = an_array {{  1. }};
constexpr auto MINUS_ONE = an_array {{ -1. }}; 
constexpr auto MINUS_TWO = an_array {{ -2. }}; 

  int
main ()
{
  auto a = wrapper <MINUS_TWO> {};
  auto b = wrapper <ONE> {};
  auto c = add (a, b);
  __builtin_printf("%f %f %f\n", a.arr[0], b.arr[0], c.arr[0]);
  if (c.arr != MINUS_ONE)
    __builtin_abort();
}

Prints:
-2.000000 1.000000 -2.000000
Aborted (core dumped)

[Bug c++/98216] [C++20] std::array<double, N> template parameter error with negative values

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

--- Comment #3 from Jonathan Wakely <redi at gcc dot gnu.org> ---
Reduced:

struct an_array
{
  constexpr bool operator==(const an_array& a) const { return d == a.d; }

  double d;
};

constexpr auto
add_array (an_array a, an_array b)
{
  a.d += b.d;
  return a;
}

template <an_array A>
struct wrapper
{
  static constexpr auto arr = A;
};

constexpr auto ONE       = an_array {{  1. }};
constexpr auto MINUS_ONE = an_array {{ -1. }}; 
constexpr auto MINUS_TWO = an_array {{ -2. }}; 

int
main ()
{
#ifndef FIX
  auto a = wrapper <MINUS_TWO> {};
  __builtin_printf("%f\n", a.arr.d);
#endif
  auto c = wrapper <add_array (MINUS_TWO, ONE)> {};
  __builtin_printf("%f\n", c.arr.d);
  if (c.arr != MINUS_ONE)
    __builtin_abort();
}

Prints:

-2.000000
-2.000000
Aborted (core dumped)

If you define FIX (i.e. don't instantiate wrapper<MINUS_TWO> first) it works:

-1.000000

[Bug c++/98216] [C++20] std::array<double, N> template parameter error with negative values

[emmanuel.le-trong@cnrs-orleans.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

--- Comment #4 from Emmanuel Le Trong <emmanuel.le-trong@cnrs-orleans.fr> ---
Looking at the symbols in your snippet:

    $ nm -C toto | grep wrapper
    0000000000402030 u wrapper<an_array{(double)[ffffffff00000000]}>::arr

That looks like a weird NaN to me.

[Bug c++/98216] [C++20] std::array<double, N> template parameter error with negative values

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

Patrick Palka <ppalka at gcc dot gnu.org> changed:

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

--- Comment #5 from Patrick Palka <ppalka at gcc dot gnu.org> ---
Related testcase which may or may not be caused by the same underlying bug:

$ cat testcase.C
template <double s> void foo() {}
template void foo<-1.0>();
template void foo<-2.0>();

$ g++ -std=c++20 testcase.C
testcase.C:3:26: error: Two symbols with same comdat_group are not linked by
the same_comdat_group list.
    3 | template void foo<-2.0>();                                              
      |                          ^  
_Z3fooILdffffffff00000000EEvv/1 (void foo() [with double s = -2.0e+0])
@0x7fdb12c4aee0         
  Type: function definition analyzed
  Visibility: force_output forced_by_abi no_reorder public weak
comdat_group:_Z3fooILdffffffff00000000EEvv one_only
  previous sharing asm name: 0
  References: 
  Referring:          
  Function flags: body
  Called by: 
  Calls:                                                                        
_Z3fooILdffffffff00000000EEvv/0 (void foo() [with double s = -1.0e+0])
@0x7fdb12c4add0
  Type: function definition analyzed                                            
  Visibility: force_output forced_by_abi no_reorder public weak
comdat_group:_Z3fooILdffffffff00000000EEvv one_only
  next sharing asm name: 1
  References: 
  Referring: 
  Function flags: body
  Called by: 
  Calls: 
testcase.C:3:26: internal compiler error: symtab_node::verify failed
0xbebe81 symtab_node::verify_symtab_nodes()
        /home/patrick/code/gcc/gcc/symtab.c:1406
0xc00586 symtab_node::checking_verify_symtab_nodes()
        /home/patrick/code/gcc/gcc/cgraph.h:675 
0xc00586 symbol_table::compile()
        /home/patrick/code/gcc/gcc/cgraphunit.c:2277
0xc02f67 symbol_table::compile()
        /home/patrick/code/gcc/gcc/cgraphunit.c:2274
0xc02f67 symbol_table::finalize_compilation_unit()
        /home/patrick/code/gcc/gcc/cgraphunit.c:2542

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |bobmiller at nvidia dot com

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
*** Bug 100279 has been marked as a duplicate of this bug. ***

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

Patrick Palka <ppalka at gcc dot gnu.org> changed:

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

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

Patrick Palka <ppalka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |andrei.popa105 at yahoo dot com

--- Comment #7 from Patrick Palka <ppalka at gcc dot gnu.org> ---
*** Bug 102092 has been marked as a duplicate of this bug. ***

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

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

https://gcc.gnu.org/g:34947d4e97ee72b26491cfe5ff4fa8258fadbe95

commit r12-3882-g34947d4e97ee72b26491cfe5ff4fa8258fadbe95
Author: Patrick Palka <ppalka@redhat.com>
Date:   Fri Sep 24 12:36:26 2021 -0400

    real: fix encoding of negative IEEE double/quad values [PR98216]

    In encode_ieee_double/quad, the assignment

      unsigned long WORD = r->sign << 31;

    is intended to set the 31st bit of WORD whenever the sign bit is set.
    But on LP64 hosts it also unintentionally sets the upper 32 bits of WORD,
    because r->sign gets promoted from unsigned:1 to int and then the result
    of the shift (equal to INT_MIN) gets sign extended from int to long.

    In the C++ frontend, this bug causes incorrect mangling of negative
    floating point values because the output of real_to_target called from
    write_real_cst unexpectedly has the upper 32 bits of this word set,
    which the caller doesn't mask out.

    This patch fixes this by avoiding the unwanted sign extension.  Note
    that r0-53976 fixed the same bug in encode_ieee_single long ago.

            PR c++/98216
            PR c++/91292

    gcc/ChangeLog:

            * real.c (encode_ieee_double): Avoid unwanted sign extension.
            (encode_ieee_quad): Likewise.

    gcc/testsuite/ChangeLog:

            * g++.dg/cpp2a/nontype-float2.C: New test.

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[emmanuel.le-trong@cnrs-orleans.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

Emmanuel Le Trong <emmanuel.le-trong@cnrs-orleans.fr> changed:

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

--- Comment #9 from Emmanuel Le Trong <emmanuel.le-trong@cnrs-orleans.fr> ---
Fixed, thank you.

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-11 branch has been updated by Patrick Palka
<ppalka@gcc.gnu.org>:

https://gcc.gnu.org/g:1682576e62d41cd761472943372b83aee514254a

commit r11-9083-g1682576e62d41cd761472943372b83aee514254a
Author: Patrick Palka <ppalka@redhat.com>
Date:   Fri Sep 24 12:36:26 2021 -0400

    real: fix encoding of negative IEEE double/quad values [PR98216]

    In encode_ieee_double/quad, the assignment

      unsigned long WORD = r->sign << 31;

    is intended to set the 31st bit of WORD whenever the sign bit is set.
    But on LP64 hosts it also unintentionally sets the upper 32 bits of WORD,
    because r->sign gets promoted from unsigned:1 to int and then the result
    of the shift (equal to INT_MIN) gets sign extended from int to long.

    In the C++ frontend, this bug causes incorrect mangling of negative
    floating point values because the output of real_to_target called from
    write_real_cst unexpectedly has the upper 32 bits of this word set,
    which the caller doesn't mask out.

    This patch fixes this by avoiding the unwanted sign extension.  Note
    that r0-53976 fixed the same bug in encode_ieee_single long ago.

            PR c++/98216
            PR c++/91292

    gcc/ChangeLog:

            * real.c (encode_ieee_double): Avoid unwanted sign extension.
            (encode_ieee_quad): Likewise.

    gcc/testsuite/ChangeLog:

            * g++.dg/cpp2a/nontype-float2.C: New test.

    (cherry picked from commit 34947d4e97ee72b26491cfe5ff4fa8258fadbe95)

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[hq.ks at web dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

hq.ks at web dot de changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hq.ks at web dot de

--- Comment #11 from hq.ks at web dot de ---
*** Bug 102754 has been marked as a duplicate of this bug. ***

[Bug c++/98216] [C++20] template mangling for double template argument is wrong

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98216

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |11.3