endian.hpp

// Copyright Take Vos 2019-2022.
// Distributed under the Boost Software License, Version 1.0.
// (See accompanying file LICENSE_1_0.txt or copy at https://www.boost.org/LICENSE_1_0.txt)
 
#pragma once
 
#include "memory.hpp"
 
#ifdef HI_HAS_SSE
#include <immintrin.h>
#endif
#ifdef HI_HAS_SSE2
#include <emmintrin.h>
#endif
#ifdef HI_HAS_SSE4_1
#include <smmintrin.h>
#endif
#if HI_COMPILER == HI_CC_MSVC
#include <stdlib.h>
#endif
#include <bit>
#include <concepts>
 
hi_warning_push();
// C26472: Don't use a static_cast for arithmetic conversions. Use brace initialization, gsl::narrow_cast or gsl::narrow
// (type.1).
// static_cast are used to cheaply cast integers to unsigned and back, for byteswapping.
hi_warning_ignore_msvc(26472);
 
namespace hi::inline v1 {
 
template<std::unsigned_integral T>
[[nodiscard]] constexpr T byte_swap(T x) noexcept
{
    if (not std::is_constant_evaluated()) {
#if HI_COMPILER == HI_CC_CLANG || HI_COMPILER == HI_CC_GCC
        if constexpr (sizeof(T) == sizeof(uint64_t)) {
            return static_cast<T>(__builtin_bswap64(static_cast<uint64_t>(x)));
        } else if constexpr (sizeof(T) == sizeof(uint32_t)) {
            return static_cast<T>(__builtin_bswap32(static_cast<uint32_t>(x)));
        } else if constexpr (sizeof(T) == sizeof(uint16_t)) {
            return static_cast<T>(__builtin_bswap16(static_cast<uint16_t>(x)));
        }
#elif HI_COMPILER == HI_CC_MSVC
        if constexpr (sizeof(T) == sizeof(uint64_t)) {
            return static_cast<T>(_byteswap_uint64(static_cast<uint64_t>(x)));
        } else if constexpr (sizeof(T) == sizeof(unsigned long)) {
            return static_cast<T>(_byteswap_ulong(static_cast<unsigned long>(x)));
        } else if constexpr (sizeof(T) == sizeof(unsigned short)) {
            return static_cast<T>(_byteswap_ushort(static_cast<unsigned short>(x)));
        }
#endif
    }
 
    if constexpr (sizeof(T) == 1) {
        return x;
    } else {
        auto r = T{};
        for (auto i = 0_uz; i != sizeof(T); ++i) {
            r <<= 8;
            r |= static_cast<uint8_t>(x);
            x >>= 8;
        }
        return r;
    }
}
 
template<std::signed_integral T>
[[nodiscard]] constexpr T byte_swap(T x) noexcept
{
    return static_cast<T>(byte_swap(static_cast<std::make_unsigned_t<T>>(x)));
}
 
template<std::floating_point T>
[[nodiscard]] constexpr T byte_swap(T x) noexcept
{
    if constexpr (std::is_same_v<T, float>) {
        auto utmp = std::bit_cast<uint32_t>(x);
        utmp = byte_swap(utmp);
        return std::bit_cast<float>(x);
    } else if constexpr (std::is_same_v<T, double>) {
        auto utmp = std::bit_cast<uint64_t>(x);
        utmp = byte_swap(utmp);
        return std::bit_cast<double>(x);
    } else {
        hi_static_no_default();
    }
}
 
template<std::integral T>
[[nodiscard]] constexpr T little_to_native(T x)
{
    if constexpr (std::endian::native == std::endian::little) {
        return x;
    } else {
        return byte_swap(x);
    }
}
 
template<std::integral T>
[[nodiscard]] constexpr T big_to_native(T x)
{
    if constexpr (std::endian::native == std::endian::big) {
        return x;
    } else {
        return byte_swap(x);
    }
}
 
template<std::integral T>
[[nodiscard]] constexpr T native_to_little(T x)
{
    if constexpr (std::endian::native == std::endian::little) {
        return x;
    } else {
        return byte_swap(x);
    }
}
 
template<std::integral T>
[[nodiscard]] constexpr T native_to_big(T x)
{
    if constexpr (std::endian::native == std::endian::big) {
        return x;
    } else {
        return byte_swap(x);
    }
}
 
template<numeric T, std::endian Endian = std::endian::native>
[[nodiscard]] constexpr T load(T const *src) noexcept
{
    auto value = *src;
    if constexpr (Endian != std::endian::native) {
        value = byte_swap(value);
    }
    return value;
}
 
template<numeric T, std::endian Endian = std::endian::native, byte_like B>
[[nodiscard]] constexpr T load(B const *src) noexcept
{
    auto value = unaligned_load<T>(src);
    if constexpr (Endian != std::endian::native) {
        value = byte_swap(value);
    }
    return value;
}
 
template<numeric T, std::endian Endian = std::endian::native>
[[nodiscard]] inline T load(void const *src) noexcept
{
    auto value = unaligned_load<T>(src);
    if constexpr (Endian != std::endian::native) {
        value = byte_swap(value);
    }
    return value;
}
 
template<numeric T>
[[nodiscard]] constexpr T load_le(T const *src) noexcept
{
    return load<T, std::endian::little>(src);
}
 
template<numeric T, byte_like B>
[[nodiscard]] constexpr T load_le(B const *src) noexcept
{
    return load<T, std::endian::little>(src);
}
 
template<numeric T>
[[nodiscard]] inline T load_le(void const *src) noexcept
{
    return load<T, std::endian::little>(src);
}
 
template<numeric T>
[[nodiscard]] constexpr T load_be(T const *src) noexcept
{
    return load<T, std::endian::big>(src);
}
 
template<numeric T, byte_like B>
[[nodiscard]] constexpr T load_be(B const *src) noexcept
{
    return load<T, std::endian::big>(src);
}
 
template<numeric T>
[[nodiscard]] inline T load_be(void const *src) noexcept
{
    return load<T, std::endian::big>(src);
}
 
template<unsigned int NumBits, byte_like B>
[[nodiscard]] constexpr auto load_bits_be(B const *src, size_t bit_index) noexcept
{
    static_assert(NumBits <= sizeof(unsigned long long) * CHAR_BIT);
 
    constexpr auto num_bits = NumBits;
    constexpr auto num_bytes = (num_bits + CHAR_BIT - 1) / CHAR_BIT;
 
    // Determine an unsigned type that can be used to read NumBits in a single `load_be()` on every bit offset.
    // clang-format off
    using value_type =
        std::conditional_t<num_bytes < sizeof(unsigned short), unsigned short,
        std::conditional_t<num_bytes < sizeof(unsigned int), unsigned int,
        std::conditional_t<num_bytes < sizeof(unsigned long), unsigned long, unsigned long long>>>;
    // clang-format on
 
    constexpr auto value_bits = sizeof(value_type) * CHAR_BIT;
 
    hilet byte_offset = bit_index / CHAR_BIT;
    hilet bit_offset = bit_index % CHAR_BIT;
 
    // Optimization of reading a byte, aligned to a byte.
    if constexpr (num_bits == CHAR_BIT) {
        if (bit_offset == 0) {
            return char_cast<value_type>(src[byte_offset]);
        }
    }
 
    // load_be allows unaligned reads.
    auto r = load_be<value_type>(std::addressof(src[byte_offset]));
 
    // Align to most significant bit. In preparation for loading
    // one more byte.
    r <<= bit_offset;
 
    if constexpr (num_bytes == sizeof(value_type)) {
        // In this case it is possible we could not read the whole value in one go.
        // We may need to read one more byte.
 
        auto bits_done = value_bits - bit_offset;
        if (bits_done < num_bits) {
            auto rest = char_cast<value_type>(src[byte_offset + sizeof(value_type)]);
            rest >>= CHAR_BIT - bit_offset;
            r |= rest;
        }
    }
 
    // Align number to least significant bit.
    r >>= value_bits - num_bits;
    return r;
}
 
 
 
template<std::endian Endian = std::endian::native, numeric T, byte_like B>
constexpr void store(T value, B const *dst) noexcept
{
    if constexpr (Endian != std::endian::native) {
        value = byte_swap(value);
    }
    unaligned_store<T>(value, dst);
}
 
template<std::endian Endian = std::endian::native, numeric T>
constexpr void store(T value, void const *dst) noexcept
{
    if constexpr (Endian != std::endian::native) {
        value = byte_swap(value);
    }
    unaligned_store<T>(value, dst);
}
 
template<numeric T, byte_like B>
constexpr void store_le(T value, B const *dst) noexcept
{
    store<std::endian::little>(value, dst);
}
 
template<numeric T>
inline void store_le(T value, void const *dst) noexcept
{
    store<std::endian::little>(value, dst);
}
 
template<numeric T, byte_like B>
constexpr void store_be(T value, B const *dst) noexcept
{
    store<std::endian::big>(value, dst);
}
 
template<numeric T>
inline void store_be(T value, void const *dst) noexcept
{
    store<std::endian::big>(value, dst);
}
 
 
template<typename T, std::endian E, std::size_t A = alignof(T)>
struct endian_buf_t {
    using value_type = T;
    constexpr static std::endian endian = E;
    constexpr static std::size_t alignment = A;
 
    alignas(A) std::byte _value[sizeof(T)];
 
    [[nodiscard]] constexpr value_type operator*() const noexcept
    {
        return load<value_type, endian>(_value);
    }
 
    constexpr endian_buf_t& operator=(value_type x) noexcept
    {
        store<endian>(x, _value);
        return *this;
    }
};
 
using big_uint64_buf_t = endian_buf_t<uint64_t, std::endian::big, 1>;
using big_uint32_buf_t = endian_buf_t<uint32_t, std::endian::big, 1>;
using big_uint16_buf_t = endian_buf_t<uint16_t, std::endian::big, 1>;
using big_int64_buf_t = endian_buf_t<int64_t, std::endian::big, 1>;
using big_int32_buf_t = endian_buf_t<int32_t, std::endian::big, 1>;
using big_int16_buf_t = endian_buf_t<int16_t, std::endian::big, 1>;
using little_uint64_buf_t = endian_buf_t<uint64_t, std::endian::little, 1>;
using little_uint32_buf_t = endian_buf_t<uint32_t, std::endian::little, 1>;
using little_uint16_buf_t = endian_buf_t<uint16_t, std::endian::little, 1>;
using little_int64_buf_t = endian_buf_t<int64_t, std::endian::little, 1>;
using little_int32_buf_t = endian_buf_t<int32_t, std::endian::little, 1>;
using little_int16_buf_t = endian_buf_t<int16_t, std::endian::little, 1>;
using native_uint64_buf_t = endian_buf_t<uint64_t, std::endian::native, 1>;
using native_uint32_buf_t = endian_buf_t<uint32_t, std::endian::native, 1>;
using native_uint16_buf_t = endian_buf_t<uint16_t, std::endian::native, 1>;
using native_int64_buf_t = endian_buf_t<int64_t, std::endian::native, 1>;
using native_int32_buf_t = endian_buf_t<int32_t, std::endian::native, 1>;
using native_int16_buf_t = endian_buf_t<int16_t, std::endian::native, 1>;
 
using big_uint64_buf_at = endian_buf_t<uint64_t, std::endian::big>;
using big_uint32_buf_at = endian_buf_t<uint32_t, std::endian::big>;
using big_uint16_buf_at = endian_buf_t<uint16_t, std::endian::big>;
using big_int64_buf_at = endian_buf_t<int64_t, std::endian::big>;
using big_int32_buf_at = endian_buf_t<int32_t, std::endian::big>;
using big_int16_buf_at = endian_buf_t<int16_t, std::endian::big>;
using little_uint64_buf_at = endian_buf_t<uint64_t, std::endian::little>;
using little_uint32_buf_at = endian_buf_t<uint32_t, std::endian::little>;
using little_uint16_buf_at = endian_buf_t<uint16_t, std::endian::little>;
using little_int64_buf_at = endian_buf_t<int64_t, std::endian::little>;
using little_int32_buf_at = endian_buf_t<int32_t, std::endian::little>;
using little_int16_buf_at = endian_buf_t<int16_t, std::endian::little>;
using native_uint64_buf_at = endian_buf_t<uint64_t, std::endian::native>;
using native_uint32_buf_at = endian_buf_t<uint32_t, std::endian::native>;
using native_uint16_buf_at = endian_buf_t<uint16_t, std::endian::native>;
using native_int64_buf_at = endian_buf_t<int64_t, std::endian::native>;
using native_int32_buf_at = endian_buf_t<int32_t, std::endian::native>;
using native_int16_buf_at = endian_buf_t<int16_t, std::endian::native>;
 
} // namespace hi::inline v1
 
hi_warning_pop();