Skip to content

Accept the 80-bit long double in <xutility> #2073

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
smanna12 wants to merge 9 commits into from
Closed

Accept the 80-bit long double in <xutility> #2073

Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
415e1ce
Accept the 80-bit long double in <xutility> for Clang
smanna12 Jul 22, 2021
9017d1f
fix error
smanna12 Jul 22, 2021
a0cbc0c
update patch
smanna12 Jul 26, 2021
590c6b8
update patch
smanna12 Jul 26, 2021
33a77c2
fix errors
smanna12 Jul 26, 2021
af0b826
update patch with #if defined
smanna12 Jul 26, 2021
224b0d6
fix typo
smanna12 Jul 26, 2021
828f3b4
Add _FINITE macro for _Dtest RETURN
smanna12 Jul 27, 2021
fdb0c5c
remove ^^^ from #endif
smanna12 Jul 28, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions stl/inc/xutility
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5997,8 +5997,12 @@ _NODISCARD _CONSTEXPR_BIT_CAST bool _Is_inf(const _Ty _Xx) { // constexpr isinf(
// FUNCTION TEMPLATE _Is_finite
template <class _Ty, enable_if_t<is_floating_point_v<_Ty>, int> = 0>
_NODISCARD _CONSTEXPR_BIT_CAST bool _Is_finite(const _Ty _Xx) { // constexpr isfinite()
#if defined(__LDBL_DIG__) && __LDBL_DIG__ == 18
return _CSTD _LDtest(&_Xx) == _FINITE;
Copy link
Contributor

@statementreply statementreply Jul 28, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't believe that _LDtest could properly handle 80-bit long double:

STL/stl/src/xldtest.cpp

Lines 10 to 12 in bd7adb4

_CRTIMP2_PURE short __CLRCALL_PURE_OR_CDECL _LDtest(long double* px) { // categorize *px -- 64-bit
return _Dtest(reinterpret_cast<double*>(px));
}

Copy link
Author

@smanna12 smanna12 Jul 28, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

_LDtest is handling long double:

STL/stl/inc/complex

Lines 542 to 557 in bd7adb4

static bool _Isinf(_Ty _Left) { // test for infinity
#if defined(__LDBL_DIG__) && __LDBL_DIG__ == 18
return _CSTD _LDtest(&_Left) == _INFCODE;
#else // ^^^ 80-bit long double (not supported by MSVC in general, see GH-1316) / 64-bit long double vvv
const auto _Uint = _Bit_cast<uint64_t>(_Left);
return (_Uint & 0x7fffffffffffffffU) == 0x7ff0000000000000U;
#endif // ^^^ 64-bit long double ^^^
}
static _CONSTEXPR20 bool _Isnan(_Ty _Left) {
#if defined(__LDBL_DIG__) && __LDBL_DIG__ == 18
return _CSTD _LDtest(&_Left) == _NANCODE;
#else // ^^^ 80-bit long double (not supported by MSVC in general, see GH-1316) / 64-bit long double vvv
const auto _Uint = _Bit_cast<uint64_t>(_Left);
return (_Uint & 0x7fffffffffffffffU) > 0x7ff0000000000000U;
#endif // ^^^ 64-bit long double ^^^

Copy link
Author

@smanna12 smanna12 Jul 28, 2021 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I work with Intel C++ compiler team. The Intel C++ compiler uses /Qlong-double option to set the long double type to 80 bits wide. The _Bit_cast function in <xutility> header file does not allow 80 bits wide long double.

Intel C++ Compiler with /Qlong-double enabled, produces the following compilation error messages when we compile <complex> header file with /std:c++latest option:

In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\iostream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\istream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\ostream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\ios:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xlocnum:12:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath:26:
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(5918,24): error: no matching function for call to '_Bit_cast'
    const auto _Bits = _Bit_cast<_Uint_type>(_Xx);
                       ^~~~~~~~~~~~~~~~~~~~~
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(5967,12): note: in instantiation of function template specialization 'std::_Float_abs_bits<long double, 0>'
      requested here
    return _Float_abs_bits(_Xx) < _Traits::_Shifted_exponent_mask;
           ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1307,31): note: in instantiation of function template specialization 'std::_Is_finite<long double, 0>' requested here
    const bool _T_is_finite = _Is_finite(_ArgT);
                              ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1392,12): note: in instantiation of function template specialization 'std::_Common_lerp<long double>' requested here
    return _Common_lerp(_ArgA, _ArgB, _ArgT);
           ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(67,36): note: candidate template ignored: requirement 'conjunction_v<std::integral_constant<bool, false>,
      std::is_trivially_copyable<unsigned long long>, std::is_trivially_copyable<long double>>' was not satisfied [with _To = unsigned long long, _From = long double]
_NODISCARD _CONSTEXPR_BIT_CAST _To _Bit_cast(const _From& _Val) noexcept {
                                   ^

In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\iostream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\istream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\ostream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\ios:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xlocnum:12:
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1308,25): error: no matching function for call to '_Is_finite'
    if (_T_is_finite && _Is_finite(_ArgA) && _Is_finite(_ArgB)) {
                        ^~~~~~~~~~
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1392,12): note: in instantiation of function template specialization 'std::_Common_lerp<long double>' requested here
    return _Common_lerp(_ArgA, _ArgB, _ArgT);
           ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(5965,37): note: candidate template ignored: substitution failure [with _Ty = long double, $1 = 0]
_NODISCARD _CONSTEXPR_BIT_CAST bool _Is_finite(const _Ty _Xx) { // constexpr isfinite()
                                    ^

In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\iostream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\istream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\ostream:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\ios:11:
In file included from c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xlocnum:12:
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1308,46): error: no matching function for call to '_Is_finite'
    if (_T_is_finite && _Is_finite(_ArgA) && _Is_finite(_ArgB)) {
                                             ^~~~~~~~~~
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(5965,37): note: candidate template ignored: substitution failure [with _Ty = long double, $1 = 0]
_NODISCARD _CONSTEXPR_BIT_CAST bool _Is_finite(const _Ty _Xx) { // constexpr isfinite()
                                    ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(5941,12): error: no matching function for call to '_Float_abs_bits'
    return _Float_abs_bits(_Xx) > _Traits::_Shifted_exponent_mask;
           ^~~~~~~~~~~~~~~
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1337,13): note: in instantiation of function template specialization 'std::_Is_nan<long double, 0>' requested here
        if (_Is_nan(_ArgA)) {
            ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\cmath(1392,12): note: in instantiation of function template specialization 'std::_Common_lerp<long double>' requested here
    return _Common_lerp(_ArgA, _ArgB, _ArgT);
           ^
c:/Program files/Microsoft Visual Studio/2022/Preview/VC/Tools/MSVC/14.29.30130/include\xutility(5915,37): note: candidate template ignored: substitution failure [with _Ty = long double, $1 = 0]
_NODISCARD _CONSTEXPR_BIT_CAST auto _Float_abs_bits(const _Ty& _Xx) {
                                  ^
4 errors generated

I think the problem is that _Bit_cast is unable to work on a long double (probably because sizeof(long double) > sizeof(unsigned long long)?).

I am not sure whether _Bit_cast needs to be fixed, or whether we should be avoiding trying to call it in the first place (generalizes the 80 bit long double check in xutility header file using same logic #1316)

It would be very helpful if we could get any input or assistance in figuring out the right path forward to fix the problem.
Tagging @CaseyCarter, @statementreply, @StephanTLavavej, @sam20908, @simmse for any suggestion. Thank you.

Adding @AaronBallman, @erichkeane for any more information.

Copy link
Member

@StephanTLavavej StephanTLavavej Jul 28, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I believe there's a separate correctness issue here, because this codepath was being used for 32-bit float as well, and that won't be properly handled by _LDtest (regardless of whether it's the original 64-bit implementation or a replaced 80-bit implementation).

#else // ^^^ 80-bit long double (not supported by MSVC in general, see GH-1316) / 64-bit long double vvv
using _Traits = _Floating_type_traits<_Ty>;
return _Float_abs_bits(_Xx) < _Traits::_Shifted_exponent_mask;
#endif // 64-bit long double
}

// STRUCT _Nontrivial_dummy_type
Expand Down