c++20无分支版本，现在std::endian已经存在，但在c++23之前增加了std::byteswap

#include <bit>
#include <type_traits>
#include <concepts>
#include <array>
#include <cstring>
#include <iostream>
#include <bitset>

template <int LEN, int OFF=LEN/2>
class do_swap
{
    // FOR 8 bytes:
    // LEN=8 (LEN/2==4)       <H><G><F><E><D><C><B><A>
    // OFF=4: FROM=0, TO=7 => [A]<G><F><E><D><C><B>[H]
    // OFF=3: FROM=1, TO=6 => [A][B]<F><E><D><C>[G][H]
    // OFF=2: FROM=2, TO=5 => [A][B][C]<E><D>[F][G][H]
    // OFF=1: FROM=3, TO=4 => [A][B][C][D][E][F][G][H]
    // OFF=0: FROM=4, TO=3 => DONE
public:
    enum consts {FROM=LEN/2-OFF, TO=(LEN-1)-FROM};
    using NXT=do_swap<LEN, OFF-1>;
// flip the first and last for the current iteration's range
    static void flip(std::array<std::byte, LEN>& b)
    {
        std::byte tmp=b[FROM];
        b[FROM]=b[TO];
        b[TO]=tmp;
        NXT::flip(b);
    }
};
template <int LEN>
class do_swap<LEN, 0> // STOP the template recursion
{
public:
    static void flip(std::array<std::byte, LEN>&)
    {
    }
};

template<std::integral T, std::endian TO, std::endian FROM=std::endian::native>
        requires ((TO==std::endian::big) || (TO==std::endian::little))
              && ((FROM==std::endian::big) || (FROM==std::endian::little))
class endian_swap
{
public:
    enum consts {BYTE_COUNT=sizeof(T)};
    static T cvt(const T integral)
    {
    // if FROM and TO are the same -- nothing to do
        if (TO==FROM)
        {
                return integral;
        }

    // endian::big --> endian::little is the same as endian::little --> endian::big
    // the bytes have to be reversed
    // memcpy seems to be the most supported way to do byte swaps in a defined way
        std::array<std::byte, BYTE_COUNT> bytes;
        std::memcpy(&bytes, &integral, BYTE_COUNT);
        do_swap<BYTE_COUNT>::flip(bytes);
        T ret;
        std::memcpy(&ret, &bytes, BYTE_COUNT);
        return ret;
    }
};

std::endian big()
{
    return std::endian::big;
}

std::endian little()
{
    return std::endian::little;
}

std::endian native()
{
    return std::endian::native;
}

long long swap_to_big(long long x)
{
    return endian_swap<long long, std::endian::big>::cvt(x);
}

long long swap_to_little(long long x)
{
    return endian_swap<long long, std::endian::little>::cvt(x);
}

void show(std::string label, long long x)
{
    std::cout << label << "\t: " << std::bitset<64>(x) << " (" << x << ")" << std::endl;
}

int main(int argv, char ** argc)
{
    long long init=0xF8FCFEFF7F3F1F0;
    long long to_big=swap_to_big(init);
    long long to_little=swap_to_little(init);
    show("Init", init);
    show(">big", to_big);
    show(">little", to_little);
}

如果您采用反转单词中位序的常见模式，并剔除每个字节中反转位的部分，那么您将只剩下反转单词中的字节的部分。对于64位:

x = ((x & 0x00000000ffffffff) << 32) ^ ((x >> 32) & 0x00000000ffffffff);
x = ((x & 0x0000ffff0000ffff) << 16) ^ ((x >> 16) & 0x0000ffff0000ffff);
x = ((x & 0x00ff00ff00ff00ff) <<  8) ^ ((x >>  8) & 0x00ff00ff00ff00ff);

编译器应该清除多余的位屏蔽操作(我把它们留在了突出显示模式)，但如果它没有，你可以这样重写第一行:

x = ( x                       << 32) ^  (x >> 32);

在大多数架构上，这通常应该简化为一条旋转指令(忽略整个操作可能是一条指令)。

在RISC处理器上，大而复杂的常量可能会导致编译困难。不过，您可以简单地计算前一个的每个常数。像这样:

uint64_t k = 0x00000000ffffffff; /* compiler should know a trick for this */
x = ((x & k) << 32) ^ ((x >> 32) & k);
k ^= k << 16;
x = ((x & k) << 16) ^ ((x >> 16) & k);
k ^= k << 8;
x = ((x & k) <<  8) ^ ((x >>  8) & k);

如果你愿意，你可以把它写成一个循环。这样做效率不高，只是为了好玩:

int i = sizeof(x) * CHAR_BIT / 2;
uintmax_t k = (1 << i) - 1;
while (i >= 8)
{
    x = ((x & k) << i) ^ ((x >> i) & k);
    i >>= 1;
    k ^= k << i;
}

为了完整起见，这里是第一种形式的简化32位版本:

x = ( x               << 16) ^  (x >> 16);
x = ((x & 0x00ff00ff) <<  8) ^ ((x >>  8) & 0x00ff00ff);

摘自Rob Pike的《字节顺序谬误》:

假设数据流有一个小端编码的32位整数。下面是如何提取它(假设无符号字节):

i = (data[0]<<0) | (data[1]<<8) | (data[2]<<16) | ((unsigned)data[3]<<24);

如果它是big-endian，下面是如何提取它:

i = (data[3]<<0) | (data[2]<<8) | (data[1]<<16) | ((unsigned)data[0]<<24);

TL;DR:不要担心你的平台原生顺序，重要的是你从中读取的流的字节顺序，你最好希望它是定义良好的。

注1:这里int和unsigned int是32位，否则类型可能需要调整。

注2:最后一个字节必须在移位前显式转换为unsigned，因为默认情况下它被提升为int，移位24位意味着操作符号位，这是未定义行为。

我喜欢这个，只是为了风格:-)

long swap(long i) {
    char *c = (char *) &i;
    return * (long *) (char[]) {c[3], c[2], c[1], c[0] };
}

void writeLittleEndianToBigEndian(void* ptrLittleEndian, void* ptrBigEndian , size_t bufLen )
{
    char *pchLittleEndian = (char*)ptrLittleEndian;

    char *pchBigEndian = (char*)ptrBigEndian;

    for ( size_t i = 0 ; i < bufLen ; i++ )    
        pchBigEndian[bufLen-1-i] = pchLittleEndian[i];
}

std::uint32_t row = 0x12345678;

char buf[4]; 

writeLittleEndianToBigEndian( &row, &buf, sizeof(row) );

简单地说:

#include <climits>

template <typename T>
T swap_endian(T u)
{
    static_assert (CHAR_BIT == 8, "CHAR_BIT != 8");

    union
    {
        T u;
        unsigned char u8[sizeof(T)];
    } source, dest;

    source.u = u;

    for (size_t k = 0; k < sizeof(T); k++)
        dest.u8[k] = source.u8[sizeof(T) - k - 1];

    return dest.u;
}

用法:swap_endian < uint32_t >(42)。