[Bug target/107242] New: ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284

[Bug target/107242] New: ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284

[asolokha at gmx dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

            Bug ID: 107242
           Summary: ICE: tree check: expected complex_cst, have plus_expr
                    in output_constant, at varasm.cc:5284
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: asolokha at gmx dot com
  Target Milestone: ---
            Target: powerpc-e300c3-linux-gnu

g++ 13.0.0 20221009 snapshot (g:e95e91eccd022a4a3a86da2749809fbad9afd20e) ICEs
when compiling the following testcase, reduced from
libstdc++-v3/testsuite/26_numerics/complex/comparison_operators/more_constexpr.cc,
for a 32-bit BE target w/ -std=c++20 -Og -frounding-math -fno-ipa-pure-const:

#include <complex>

std::complex<float>
test_comparison ()
{
  constexpr std::complex<float> a{1.1, 2.2};
  constexpr std::complex<float> b{3.3, 4.4};

  return a + b;
}

% powerpc-e300c3-linux-gnu-g++-13 -std=c++20 -Og -frounding-math
-fno-ipa-pure-const -c micz0onj.cc
micz0onj.cc:10:1: internal compiler error: tree check: expected complex_cst,
have plus_expr in output_constant, at varasm.cc:5284
   10 | }
      | ^
0x803d1f tree_check_failed(tree_node const*, char const*, int, char const*,
...)
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/tree.cc:8827
0x819d94 tree_check(tree_node*, char const*, int, char const*, tree_code)
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/tree.h:3526
0x819d94 output_constant
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:5284
0x14b1c1a output_constructor_regular_field
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:5562
0x14b1c1a output_constructor
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:5829
0x14b3625 assemble_constant_contents
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:3638
0x14b3d34 output_object_block
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:8070
0x14b3d34 output_object_blocks()
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:8148

powerpc-e300c3-linux-gnu-g++-13 -v
Using built-in specs.
COLLECT_GCC=powerpc-e300c3-linux-gnu-g++-13
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/powerpc-e300c3-linux-gnu/13/lto-wrapper
Target: powerpc-e300c3-linux-gnu
Configured with:
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/configure
--host=x86_64-pc-linux-gnu --target=powerpc-e300c3-linux-gnu
--build=x86_64-pc-linux-gnu --prefix=/usr
--bindir=/usr/x86_64-pc-linux-gnu/powerpc-e300c3-linux-gnu/gcc-bin/13
--includedir=/usr/lib/gcc/powerpc-e300c3-linux-gnu/13/include
--datadir=/usr/share/gcc-data/powerpc-e300c3-linux-gnu/13
--mandir=/usr/share/gcc-data/powerpc-e300c3-linux-gnu/13/man
--infodir=/usr/share/gcc-data/powerpc-e300c3-linux-gnu/13/info
--with-gxx-include-dir=/usr/lib/gcc/powerpc-e300c3-linux-gnu/13/include/g++-v13
--with-python-dir=/share/gcc-data/powerpc-e300c3-linux-gnu/13/python
--enable-languages=c,c++,fortran --enable-obsolete --enable-secureplt
--disable-werror --with-system-zlib --disable-nls
--disable-libunwind-exceptions --enable-checking=yes
--with-gcc-major-version-only --disable-esp --enable-libstdcxx-time
--disable-libstdcxx-pch --enable-poison-system-directories
--with-sysroot=/usr/powerpc-e300c3-linux-gnu --disable-bootstrap
--enable-__cxa_atexit --enable-clocale=gnu --disable-multilib
--disable-fixed-point --enable-targets=all --enable-libgomp --disable-libssp
--disable-libada --disable-cet --disable-systemtap
--enable-valgrind-annotations --disable-vtable-verify --disable-libvtv
--without-zstd --enable-lto --with-isl --disable-isl-version-check
--disable-libsanitizer --enable-default-ssp
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 13.0.0 20221009 (experimental) (GCC)

[Bug c++/107242] ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
It is the gimplification which is causing issues.

[Bug c++/107242] ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
From the original dump:
{
  const struct complex a = {._M_value=__complex__
(1.10000002384185791015625e+0, 2.2000000476837158203125e+0)};
  const struct complex b = {._M_value=__complex__ (3.2999999523162841796875e+0,
4.400000095367431640625e+0)};

  <<cleanup_point   const struct complex a = {._M_value=__complex__
(1.10000002384185791015625e+0, 2.2000000476837158203125e+0)};>>;
  <<cleanup_point   const struct complex b = {._M_value=__complex__
(3.2999999523162841796875e+0, 4.400000095367431640625e+0)};>>;
  <<cleanup_point return <retval> = TARGET_EXPR <D.68654, TARGET_EXPR <D.68676,
{._M_value=__complex__ (1.10000002384185791015625e+0,
2.2000000476837158203125e+0) + __complex__ (3.2999999523162841796875e+0,
4.400000095367431640625e+0)}>>>>;
}
Which looks fine as we have COMPLEX_CST+COMPLEX_CST

But then in the gimple dump we get:
      <retval> = *.LC2D.78102;

Which is broken.

[Bug c++/107242] ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284 with frounding-math

[asolokha at gmx dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

--- Comment #3 from Arseny Solokha <asolokha at gmx dot com> ---
Testcase w/o dependency on libstdc++:

template <typename>
class complex;

template <>
struct complex<float> {
  template <typename _Tp>
  constexpr void
  operator+= (_Tp __z)
  {
    _M_value += __z._M_value;
  }

  _Complex float _M_value;
};

constexpr complex<float>
operator+ (complex<float> __x, complex<float> __y)
{
  complex __r = __x;

  __r += __y;

  return __r;
}

complex<float>
test_comparison ()
{
  constexpr complex<float> a{2.2}, b{4.4};

  return a + b;
}

% powerpc-e300c3-linux-gnu-g++-13 -Og -frounding-math -c micz0onj.cc
micz0onj.cc:32:1: internal compiler error: tree check: expected complex_cst,
have plus_expr in output_constant, at varasm.cc:5284
   32 | }
      | ^
0x803d1f tree_check_failed(tree_node const*, char const*, int, char const*,
...)
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/tree.cc:8827
0x819d94 tree_check(tree_node*, char const*, int, char const*, tree_code)
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/tree.h:3526
0x819d94 output_constant
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:5284
0x14b1c1a output_constructor_regular_field
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:5562
0x14b1c1a output_constructor
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:5829
0x14b3625 assemble_constant_contents
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:3638
0x14b3d34 output_object_block
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:8070
0x14b3d34 output_object_blocks()
       
/var/tmp/portage/cross-powerpc-e300c3-linux-gnu/gcc-13.0.0_p20221009/work/gcc-13-20221009/gcc/varasm.cc:8148

[Bug c++/107242] ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284 with frounding-math

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2022-10-13

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Arseny Solokha from comment #3)
> Testcase w/o dependency on libstdc++:

And reduced all the way:
struct cf {
  _Complex float _M_value;
};

cf
test_comparison ()
{
  constexpr _Complex float  a{2.2}, b{4.4};
  return {a + b};
}

[Bug c++/107242] ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284 with frounding-math

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think that either the C++ frontend needs to reject this code or we need to
ignore -frounding-math in constant folding this (I think we do that elsewhere).

x86 is fine, presumambly because of ... !?

I get

struct cf test_comparison ()
{
  struct cf D.2394;
  const complex float a;
  const complex float b;

  a = __complex__ (2.2000000476837158203125e+0, 0.0);
  b = __complex__ (4.400000095367431640625e+0, 0.0);
  _1 = __complex__ (2.2000000476837158203125e+0, 0.0) + __complex__
(4.400000095367431640625e+0, 0.0);
  D.2394._M_value = _1;
  return D.2394;
}

[Bug c++/107242] ICE: tree check: expected complex_cst, have plus_expr in output_constant, at varasm.cc:5284 with frounding-math

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107242

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #5)
> I think that either the C++ frontend needs to reject this code or we need to
> ignore -frounding-math in constant folding this (I think we do that
> elsewhere).
> 
> x86 is fine, presumambly because of ... !?

Because of the gimplifier deciding it should do the complex float move struct
by peices rather than a block.