- Boost - lists.boost.org

multiprecision issues
by Gero Peterhoff 22 Apr '23

22 Apr '23

Hi John, i stumbled over some more issues https://godbolt.org/z/x5srG9eof : - the unary operator- does not work with boost::multiprecision::cpp_bin_float_N on nan - operator<<(boost::multiprecision::float128) does not support std::showpos with std::hexfloat - operator<<(boost::multiprecision::cpp_bin_float_N) does not support std::hexfloat and in the case of nan does not support std::showpos I also miss C++17 from_chars/to_chars. I realize that this is extremely tedious work, but multiprecision should be as compatible to the standard as possible. If you want, I'll be happy to look for various issues and report them to you, but you'll have to fix them yourself. cu Gero

1 0

Re: [boost] ccmath::copysign/signbit
by Matt Borland 22 Apr '23

22 Apr '23

Gero, > 1) copysign uses abs, which is wrong. copysign/signbit may *only* change/check the sign (https://en.cppreference.com/w/cpp/numeric/math/copysign https://en.cppreference.com/w/cpp/numeric/math/signbit) I think you are conflating signbit and copysign here. Copysign is to return the magnitude of the first argument (which by definition is the absolute value) with the sign of the second argument. We use signbit in the implementation of this because copysign can be used to manipulate the signs of NaNs. > 2) In signbit.hpp you implement struct IEEEfN2bits. But these are incomplete: > - long double can be 16/12/10 bytes depending on platform/compiler option. You do not take this into account. > - Correct types are already available in boost/math/cstdfloat/cstdfloat_types.hpp. You will find implementations of IEEEl2bits here https://github.com/boostorg/math/blob/develop/include/boost/math/ccmath/sig…. They account for 64, 80, and 128 bit long doubles based on LDBL_MANT_DIG and LDBL_MAX_EXP. If for instance you use compiler flags on GCC to alter the size of a long double those macros are changed accordingly. > PS > gcc13 is about to be released (https://gcc.gnu.org) so it is high time to support std::floatN_t as well. Yes, the release of GCC 13 is scheduled for next week. Once it’s available for download I will open issues/PRs to fix any breakages that come from the new types. In the interim I believe, but cannot confirm that if you use a policy that disables promotion you should be fine. Matt

2 1

Re: [boost] ccmath implementations
by John Maddock 21 Apr '23

21 Apr '23

On 21/04/2023 15:05, Gero Peterhoff via Boost wrote: > Hi Matt, > the ccmath functions must/could only work with std types. Is this > assumption correct? > The constexpr code path is almost certainly for std types only. However, they can be used in generic functions declared constexpr, which are not necessarily used in a constexpr context: in that case the argument could be any suitable type. John.

1 0

ccmath::copysign/signbit
by Gero Peterhoff 20 Apr '23

20 Apr '23

Hi Matt, I see the following problems in your implementations for ccmath::copysign/signbit: 1) copysign uses abs, which is wrong. copysign/signbit may *only* change/check the sign (https://en.cppreference.com/w/cpp/numeric/math/copysign https://en.cppreference.com/w/cpp/numeric/math/signbit) 2) In signbit.hpp you implement struct IEEEfN2bits. But these are incomplete: - long double can be 16/12/10 bytes depending on platform/compiler option. You do not take this into account. - Correct types are already available in boost/math/cstdfloat/cstdfloat_types.hpp. Therefore, I provide the complete structures in ccmath/bits.hpp. This also contains structures for std::(b)float16_t and can be extended if needed. Extended type-traits for floats To minimize completely sprawling manual checks (#if + std::is_same) I provide more advanced TTs in type_traits/is_floating_point_more.hpp. This way ccmath::copysign/signbit can be implemented correctly. Possibly other functions are affected. thx Gero PS gcc13 is about to be released (https://gcc.gnu.org) so it is high time to support std::floatN_t as well.

1 0

ccmath implementations
by Gero Peterhoff 20 Apr '23

20 Apr '23

Hi Matt, the ccmath functions must/could only work with std types. Is this assumption correct? If true: Wouldn't it make sense to do such a check for CT? This way errors can be detected early and it generally simplifies error handling. namespace boost::math::ccmath { namespace detail { template <typename type> inline constexpr auto foo_impl(const Type x) noexcept { return x; } } // detail template <typename Type> constexpr auto foo(const Type x) noexcept { BOOST_IF_CONSTEXPR (std::is_arithmetic_v<Type>) { if (BOOST_MATH_IS_CONSTANT_EVALUATED(x)) { BOOST_IF_CONSTEXPR (std::is_integral_v<Type>) { return detail::foo_impl(double(x)); } else { return detail::foo_impl(x); } } else { using ::std::foo; return foo(x); } } else { static_assert(boost::dependent_false_v<Type>, "foo is not supported with this type or platform"); return 0; } } } // boost::math::ccmath 1) In particular, this approach is useful for functions that require BOOST_MATH_BIT_CAST in their implementation. To be honest, I don't like your implementations of signbit/copysign (buggy) - I'll write you something more on that. 2) The idiom "always false" (dependent_false) guarantees that static_assert is only used for Type!=is_arithmetic. cu Gero

1 0

Re: [boost] Master is open for merges
by Vinnie Falco 19 Apr '23

19 Apr '23

On Sat, Apr 15, 2023 at 7:37 AM Marshall Clow via Boost <boost(a)lists.boost.org> wrote: > Next milestone: June 21 Boost 1.83.0 closed for new libraries and breaking changes Are we going to do a bugfix update for the overly aggressive inline function issue? Thanks

1 0

Re: [boost] using boost::multiprecision functions
by Seth 19 Apr '23

19 Apr '23

This is what ADL (argument dependent lookup) is for: https://en.cppreference.com/w/cpp/language/adl See e.g. https://compiler-explorer.com/z/oej1ehhnY template <typename T> T foo(T x) { using std::sin; return sin(x) * 42; } int main() { using A = double; using B = boost::multiprecision::cpp_bin_float_oct; A a = foo(A(23)); // ok B b = foo(B(23)); // ok std::cout << std::setprecision(32); std::cout << "a: " << a << "\n"; std::cout << "b: " << b << "\n"; using std::abs; std::cout << "|b-a|: " << abs(b-a) << "\n"; } As you'd expect, b is more precise than a: a: -35.541256975357164549222943605855 b: -35.541256975357166680135703420035 |b-a|: 2.1309127598141799216575314028644e-15 Seth On Sun, Apr 16, 2023, at 9:02 PM, Gero Peterhoff via Boost wrote: > Hello John, > * problem description > The functions of boost::multiprecision are not available in std, e.g. > template <typename type> > Type foo(const Type x) noexcept > { > return std::sin(x) * 42; > } > > using A = double; > A a = foo(A(23)); // ok > > using B = boost::multiprecision::cpp_bin_float_oct; > B b = foo(B(23)); // error > > > > As a workaround you currently have to write > template <typename Type> > Type bar(const Type x) noexcept > { > using ::std::sin; > using ::boost::multiprecision::sin; > return sin(x) * 42; > } > > using A = double; > A a = bar(A(23)); // ok > > using B = boost::multiprecision::cpp_bin_float_oct; > B b = bar(B(23)); // ok > > > This contradicts the template idea and is time-consuming/error-prone. > > > thx > Gero > > > _______________________________________________ > Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost > > Attachments: > * OpenPGP_signature

1 0

Re: [boost] Typo in Mysql URL on release page
by Ruben Perez 18 Apr '23

18 Apr '23

Hi Tim, Thanks for reporting! That's now fixed. Regards, Ruben. On Fri, 14 Apr 2023 at 21:27, Tim Haines via Boost <boost(a)lists.boost.org> wrote: > > The link to Mysql on the release page ( > https://www.boost.org/users/history/version_1_82_0.html) points to > > https://www.boost.org/doc/libs/1_82_0/libs/msyql/ > ^^ > > Thanks. > - Tim > > _______________________________________________ > Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

1 0

Re: [boost] using boost::multiprecision functions
by John Maddock 17 Apr '23

17 Apr '23

On 16/04/2023 19:52, Gero Peterhoff via Boost wrote: > Hello John, > * problem description > The functions of boost::multiprecision are not available in std, e.g. > template <typename type> > Type foo(const Type x) noexcept > { > return std::sin(x) * 42; > } > > using A = double; > A a = foo(A(23)); // ok > > using B = boost::multiprecision::cpp_bin_float_oct; > B b = foo(B(23)); // error > > > > As a workaround you currently have to write > template <typename Type> > Type bar(const Type x) noexcept > { > using ::std::sin; > using ::boost::multiprecision::sin; > return sin(x) * 42; > } > > using A = double; > A a = bar(A(23)); // ok > > using B = boost::multiprecision::cpp_bin_float_oct; > B b = bar(B(23)); // ok > > > This contradicts the template idea and is time-consuming/error-prone. > We are not allowed to overload anything in namespace std, so the correct incantation for any generic code is: template <typename Type> Type bar(const Type x) // cannot be noexcept if used in multiprecision case { using ::std::sin; return sin(x) * 42; // potentially found via ADL } John.

1 0

using boost::multiprecision functions
by Gero Peterhoff 16 Apr '23

16 Apr '23

Hello John, * problem description The functions of boost::multiprecision are not available in std, e.g. template <typename type> Type foo(const Type x) noexcept { return std::sin(x) * 42; } using A = double; A a = foo(A(23)); // ok using B = boost::multiprecision::cpp_bin_float_oct; B b = foo(B(23)); // error As a workaround you currently have to write template <typename Type> Type bar(const Type x) noexcept { using ::std::sin; using ::boost::multiprecision::sin; return sin(x) * 42; } using A = double; A a = bar(A(23)); // ok using B = boost::multiprecision::cpp_bin_float_oct; B b = bar(B(23)); // ok This contradicts the template idea and is time-consuming/error-prone. thx Gero

1 0