From Newsgroup: comp.lang.c++

I've developed fpclassify() and similar parts of the glibc.
Mostly faster than the glibc, especially fpclassify() itself.

#pragma once
#include <limits>
#include <concepts>
#include <type_traits>
#include <cstdint>
#include <bit>
#include <cmath>

template<typename Fp>
concept fp_classify_types =
(sizeof( Fp ) == 8 && std::same_as<Fp, double> && std::numeric_limits<double>::is_iec559)
|| (sizeof( Fp ) == 4 && std::same_as<Fp, float> && std::numeric_limits<float>::is_iec559);

template<fp_classify_types Fp>
struct fp_classify
{
static constexpr int type( Fp v ) noexcept;
static constexpr bool normal( Fp v ) noexcept;
static constexpr bool subnormal( Fp v ) noexcept;
static constexpr bool finite( Fp v ) noexcept;
static constexpr bool inf( Fp v ) noexcept;
static constexpr bool nan( Fp v ) noexcept;
private:
static constexpr bool Fp8 = sizeof( Fp ) == 8;
using word = std::conditional_t<Fp8, uint64_t, uint32_t>;
static constexpr unsigned
MantBits = Fp8 ? 52 : 23,
MaxExp = Fp8 ? 0x7FFu : 0xFFu;
static constexpr word
Sign = 1ull << (Fp8 ? 63 : 31),
Exp = (word)MaxExp << MantBits,
MinNormal = 1ull << MantBits,
Mant = MinNormal - 1;
};

template<fp_classify_types Fp>
constexpr int fp_classify<Fp>::type( Fp v ) noexcept
{
word bin = std::bit_cast<word>( v ) & ~Sign;
if( bin - MinNormal < Exp - MinNormal ) [[likely]]
return FP_NORMAL;
unsigned exp = bin >> MantBits;
bool nzMant = bin & Mant;
int
zroExpType = nzMant ? FP_SUBNORMAL : FP_ZERO,
maxExpType = nzMant ? FP_NAN : FP_INFINITE;
return exp ? maxExpType : zroExpType;
}

template<fp_classify_types Fp>
constexpr bool fp_classify<Fp>::normal( Fp v ) noexcept
{
word bin = std::bit_cast<word>( v ) & ~Sign;
return bin - MinNormal < Exp - MinNormal;
}

template<fp_classify_types Fp>
constexpr bool fp_classify<Fp>::subnormal( Fp v ) noexcept
{
word bin = std::bit_cast<word>( v ) & ~Sign;
return bin - 1u < MinNormal - 1u;
}

template<fp_classify_types Fp>
constexpr bool fp_classify<Fp>::finite( Fp v ) noexcept
{
return (std::bit_cast<word>( v ) & ~Sign) < Exp;
}

template<fp_classify_types Fp>
constexpr bool fp_classify<Fp>::inf( Fp v ) noexcept
{
return (std::bit_cast<word>( v ) & ~Sign) == Exp;
}

template<fp_classify_types Fp>
constexpr bool fp_classify<Fp>::nan( Fp v ) noexcept
{
return (std::bit_cast<word>( v ) & ~Sign) > Exp;
}
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.c++

On Fri, 24 Jul 2026 13:49:48 +0200
Bonita Montero <Bonita.Montero@gmail.com> wrote:

I've developed fpclassify() and similar parts of the glibc.
Mostly faster than the glibc, especially fpclassify() itself.

Any alternative to glibc has to be linkable with C object code so
re-write the below in C or assembler and then get back to us.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.c++

Am 24.07.2026 um 19:22 schrieb boltar@capricaworld.com:

On Fri, 24 Jul 2026 13:49:48 +0200

Bonita Montero <Bonita.Montero@gmail.com> wrote:

I've developed fpclassify() and similar parts of the glibc.
Mostly faster than the glibc, especially fpclassify() itself.

Any alternative to glibc has to be linkable with C object code so
re-write the below in C or assembler and then get back to us.

Idiot, this is a C++ group.

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.c++

On Fri, 24 Jul 2026 19:28:49 +0200
Bonita Montero <Bonita.Montero@gmail.com> gabbled:

Am 24.07.2026 um 19:22 schrieb boltar@capricaworld.com:

On Fri, 24 Jul 2026 13:49:48 +0200

Bonita Montero <Bonita.Montero@gmail.com> wrote:

I've developed fpclassify() and similar parts of the glibc.
Mostly faster than the glibc, especially fpclassify() itself.

Any alternative to glibc has to be linkable with C object code so
re-write the below in C or assembler and then get back to us.

Idiot, this is a C++ group.

No shit, thanks for the heads up. Whats the point of re-writing glibc in C++?

--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.c++

Am 26.07.2026 um 10:44 schrieb boltar@caprica.universe:

No shit, thanks for the heads up.
Whats the point of re-writing glibc in C++?

Just for fun and because it's faster.


--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.lang.c++

On 7/26/2026 3:59 AM, Bonita Montero wrote:

Am 26.07.2026 um 10:44 schrieb boltar@caprica.universe:

No shit, thanks for the heads up.
Whats the point of re-writing glibc in C++?

Just for fun and because it's faster.

;^)
--- Synchronet 3.22a-Linux NewsLink 1.2