哇，我简直不敢相信我在这里读到的一些答案。实际上汇编中有一条指令比其他任何程序都快。bswap。你可以简单地写一个这样的函数…

__declspec(naked) uint32_t EndianSwap(uint32 value)
{
    __asm
    {
        mov eax, dword ptr[esp + 4]
        bswap eax
        ret
    }
}

它比之前提到的内在函数要快得多。我把它们拆开看了看。上面的函数没有序言/尾声，因此实际上没有任何开销。

unsigned long _byteswap_ulong(unsigned long value);

做16位同样容易，除了你会使用xchg al，啊。Bswap仅适用于32位寄存器。

64位有点棘手，但也不过分。比上面所有带有循环和模板的例子都要好得多。

这里有一些注意事项……首先，bswap只能在80x486以上的CPU上使用。有人打算在386上运行吗?!?如果是这样，你仍然可以用…替换bswap。

mov ebx, eax
shr ebx, 16
xchg al, ah
xchg bl, bh
shl eax, 16
or eax, ebx

内联汇编也只能在Visual Studio的x86代码中使用。裸函数不能内衬，而且在x64版本中也不可用。对于那个实例，你必须使用编译器的内在函数。

我从这篇文章中得到了一些建议，并把它们放在一起形成了这个:

#include <boost/type_traits.hpp>
#include <boost/static_assert.hpp>
#include <boost/detail/endian.hpp>
#include <stdexcept>
#include <cstdint>

enum endianness
{
    little_endian,
    big_endian,
    network_endian = big_endian,
    
    #if defined(BOOST_LITTLE_ENDIAN)
        host_endian = little_endian
    #elif defined(BOOST_BIG_ENDIAN)
        host_endian = big_endian
    #else
        #error "unable to determine system endianness"
    #endif
};

namespace detail {

template<typename T, size_t sz>
struct swap_bytes
{
    inline T operator()(T val)
    {
        throw std::out_of_range("data size");
    }
};

template<typename T>
struct swap_bytes<T, 1>
{
    inline T operator()(T val)
    {
        return val;
    }
};

template<typename T>
struct swap_bytes<T, 2>
{
    inline T operator()(T val)
    {
        return ((((val) >> 8) & 0xff) | (((val) & 0xff) << 8));
    }
};

template<typename T>
struct swap_bytes<T, 4>
{
    inline T operator()(T val)
    {
        return ((((val) & 0xff000000) >> 24) |
                (((val) & 0x00ff0000) >>  8) |
                (((val) & 0x0000ff00) <<  8) |
                (((val) & 0x000000ff) << 24));
    }
};

template<>
struct swap_bytes<float, 4>
{
    inline float operator()(float val)
    {
        uint32_t mem =swap_bytes<uint32_t, sizeof(uint32_t)>()(*(uint32_t*)&val);
        return *(float*)&mem;
    }
};

template<typename T>
struct swap_bytes<T, 8>
{
    inline T operator()(T val)
    {
        return ((((val) & 0xff00000000000000ull) >> 56) |
                (((val) & 0x00ff000000000000ull) >> 40) |
                (((val) & 0x0000ff0000000000ull) >> 24) |
                (((val) & 0x000000ff00000000ull) >> 8 ) |
                (((val) & 0x00000000ff000000ull) << 8 ) |
                (((val) & 0x0000000000ff0000ull) << 24) |
                (((val) & 0x000000000000ff00ull) << 40) |
                (((val) & 0x00000000000000ffull) << 56));
    }
};

template<>
struct swap_bytes<double, 8>
{
    inline double operator()(double val)
    {
        uint64_t mem =swap_bytes<uint64_t, sizeof(uint64_t)>()(*(uint64_t*)&val);
        return *(double*)&mem;
    }
};

template<endianness from, endianness to, class T>
struct do_byte_swap
{
    inline T operator()(T value)
    {
        return swap_bytes<T, sizeof(T)>()(value);
    }
};
// specialisations when attempting to swap to the same endianess
template<class T> struct do_byte_swap<little_endian, little_endian, T> { inline T operator()(T value) { return value; } };
template<class T> struct do_byte_swap<big_endian,    big_endian,    T> { inline T operator()(T value) { return value; } };

} // namespace detail

template<endianness from, endianness to, class T>
inline T byte_swap(T value)
{
    // ensure the data is only 1, 2, 4 or 8 bytes
    BOOST_STATIC_ASSERT(sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4 || sizeof(T) == 8);
    // ensure we're only swapping arithmetic types
    BOOST_STATIC_ASSERT(boost::is_arithmetic<T>::value);

    return detail::do_byte_swap<from, to, T>()(value);
}

然后你可以这样使用它:

// swaps val from host-byte-order to network-byte-order
auto swapped = byte_swap<host_endian, network_endian>(val);

反之亦然

// swap a value received from the network into host-byte-order
auto val = byte_swap<network_endian, host_endian>(val_from_network);

这是我想到的一个通用版本，用于在适当的位置交换值。如果性能存在问题，其他建议会更好。

 template<typename T>
    void ByteSwap(T * p)
    {
        for (int i = 0;  i < sizeof(T)/2;  ++i)
            std::swap(((char *)p)[i], ((char *)p)[sizeof(T)-1-i]);
    }

免责声明:我还没有尝试编译或测试它。

下面介绍如何读取以IEEE 754 64位格式存储的double，即使您的主机使用不同的系统。

/*
* read a double from a stream in ieee754 format regardless of host
*  encoding.
*  fp - the stream
*  bigendian - set to if big bytes first, clear for little bytes
*              first
*
*/
double freadieee754(FILE *fp, int bigendian)
{
    unsigned char buff[8];
    int i;
    double fnorm = 0.0;
    unsigned char temp;
    int sign;
    int exponent;
    double bitval;
    int maski, mask;
    int expbits = 11;
    int significandbits = 52;
    int shift;
    double answer;

    /* read the data */
    for (i = 0; i < 8; i++)
        buff[i] = fgetc(fp);
    /* just reverse if not big-endian*/
    if (!bigendian)
    {
        for (i = 0; i < 4; i++)
        {
            temp = buff[i];
            buff[i] = buff[8 - i - 1];
            buff[8 - i - 1] = temp;
        }
    }
    sign = buff[0] & 0x80 ? -1 : 1;
    /* exponet in raw format*/
    exponent = ((buff[0] & 0x7F) << 4) | ((buff[1] & 0xF0) >> 4);

    /* read inthe mantissa. Top bit is 0.5, the successive bits half*/
    bitval = 0.5;
    maski = 1;
    mask = 0x08;
    for (i = 0; i < significandbits; i++)
    {
        if (buff[maski] & mask)
            fnorm += bitval;

        bitval /= 2.0;
        mask >>= 1;
        if (mask == 0)
        {
            mask = 0x80;
            maski++;
        }
    }
    /* handle zero specially */
    if (exponent == 0 && fnorm == 0)
        return 0.0;

    shift = exponent - ((1 << (expbits - 1)) - 1); /* exponent = shift + bias */
    /* nans have exp 1024 and non-zero mantissa */
    if (shift == 1024 && fnorm != 0)
        return sqrt(-1.0);
    /*infinity*/
    if (shift == 1024 && fnorm == 0)
    {

#ifdef INFINITY
        return sign == 1 ? INFINITY : -INFINITY;
#endif
        return  (sign * 1.0) / 0.0;
    }
    if (shift > -1023)
    {
        answer = ldexp(fnorm + 1.0, shift);
        return answer * sign;
    }
    else
    {
        /* denormalised numbers */
        if (fnorm == 0.0)
            return 0.0;
        shift = -1022;
        while (fnorm < 1.0)
        {
            fnorm *= 2;
            shift--;
        }
        answer = ldexp(fnorm, shift);
        return answer * sign;
    }
}

对于这套函数的其余部分，包括写和整数例程，请参阅我的github项目

https://github.com/MalcolmMcLean/ieee754

实现优化器友好的未对齐非就地末端访问器的可移植技术。它们处理每个编译器、每个边界对齐和每个字节排序。这些未对齐的例程被补充或讨论，取决于本机的端序和对齐方式。部分列出，但你懂的。BO*是基于本机字节排序的常数值。

uint32_t sw_get_uint32_1234(pu32)
uint32_1234 *pu32;
{
  union {
    uint32_1234 u32_1234;
    uint32_t u32;
  } bou32;
  bou32.u32_1234[0] = (*pu32)[BO32_0];
  bou32.u32_1234[1] = (*pu32)[BO32_1];
  bou32.u32_1234[2] = (*pu32)[BO32_2];
  bou32.u32_1234[3] = (*pu32)[BO32_3];
  return(bou32.u32);
}

void sw_set_uint32_1234(pu32, u32)
uint32_1234 *pu32;
uint32_t u32;
{
  union {
    uint32_1234 u32_1234;
    uint32_t u32;
  } bou32;
  bou32.u32 = u32;
  (*pu32)[BO32_0] = bou32.u32_1234[0];
  (*pu32)[BO32_1] = bou32.u32_1234[1];
  (*pu32)[BO32_2] = bou32.u32_1234[2];
  (*pu32)[BO32_3] = bou32.u32_1234[3];
}

#if HAS_SW_INT64
int64 sw_get_int64_12345678(pi64)
int64_12345678 *pi64;
{
  union {
    int64_12345678 i64_12345678;
    int64 i64;
  } boi64;
  boi64.i64_12345678[0] = (*pi64)[BO64_0];
  boi64.i64_12345678[1] = (*pi64)[BO64_1];
  boi64.i64_12345678[2] = (*pi64)[BO64_2];
  boi64.i64_12345678[3] = (*pi64)[BO64_3];
  boi64.i64_12345678[4] = (*pi64)[BO64_4];
  boi64.i64_12345678[5] = (*pi64)[BO64_5];
  boi64.i64_12345678[6] = (*pi64)[BO64_6];
  boi64.i64_12345678[7] = (*pi64)[BO64_7];
  return(boi64.i64);
}
#endif

int32_t sw_get_int32_3412(pi32)
int32_3412 *pi32;
{
  union {
    int32_3412 i32_3412;
    int32_t i32;
  } boi32;
  boi32.i32_3412[2] = (*pi32)[BO32_0];
  boi32.i32_3412[3] = (*pi32)[BO32_1];
  boi32.i32_3412[0] = (*pi32)[BO32_2];
  boi32.i32_3412[1] = (*pi32)[BO32_3];
  return(boi32.i32);
}

void sw_set_int32_3412(pi32, i32)
int32_3412 *pi32;
int32_t i32;
{
  union {
    int32_3412 i32_3412;
    int32_t i32;
  } boi32;
  boi32.i32 = i32;
  (*pi32)[BO32_0] = boi32.i32_3412[2];
  (*pi32)[BO32_1] = boi32.i32_3412[3];
  (*pi32)[BO32_2] = boi32.i32_3412[0];
  (*pi32)[BO32_3] = boi32.i32_3412[1];
}

uint32_t sw_get_uint32_3412(pu32)
uint32_3412 *pu32;
{
  union {
    uint32_3412 u32_3412;
    uint32_t u32;
  } bou32;
  bou32.u32_3412[2] = (*pu32)[BO32_0];
  bou32.u32_3412[3] = (*pu32)[BO32_1];
  bou32.u32_3412[0] = (*pu32)[BO32_2];
  bou32.u32_3412[1] = (*pu32)[BO32_3];
  return(bou32.u32);
}

void sw_set_uint32_3412(pu32, u32)
uint32_3412 *pu32;
uint32_t u32;
{
  union {
    uint32_3412 u32_3412;
    uint32_t u32;
  } bou32;
  bou32.u32 = u32;
  (*pu32)[BO32_0] = bou32.u32_3412[2];
  (*pu32)[BO32_1] = bou32.u32_3412[3];
  (*pu32)[BO32_2] = bou32.u32_3412[0];
  (*pu32)[BO32_3] = bou32.u32_3412[1];
}

float sw_get_float_1234(pf)
float_1234 *pf;
{
  union {
    float_1234 f_1234;
    float f;
  } bof;
  bof.f_1234[0] = (*pf)[BO32_0];
  bof.f_1234[1] = (*pf)[BO32_1];
  bof.f_1234[2] = (*pf)[BO32_2];
  bof.f_1234[3] = (*pf)[BO32_3];
  return(bof.f);
}

void sw_set_float_1234(pf, f)
float_1234 *pf;
float f;
{
  union {
    float_1234 f_1234;
    float f;
  } bof;
  bof.f = (float)f;
  (*pf)[BO32_0] = bof.f_1234[0];
  (*pf)[BO32_1] = bof.f_1234[1];
  (*pf)[BO32_2] = bof.f_1234[2];
  (*pf)[BO32_3] = bof.f_1234[3];
}

double sw_get_double_12345678(pd)
double_12345678 *pd;
{
  union {
    double_12345678 d_12345678;
    double d;
  } bod;
  bod.d_12345678[0] = (*pd)[BO64_0];
  bod.d_12345678[1] = (*pd)[BO64_1];
  bod.d_12345678[2] = (*pd)[BO64_2];
  bod.d_12345678[3] = (*pd)[BO64_3];
  bod.d_12345678[4] = (*pd)[BO64_4];
  bod.d_12345678[5] = (*pd)[BO64_5];
  bod.d_12345678[6] = (*pd)[BO64_6];
  bod.d_12345678[7] = (*pd)[BO64_7];
  return(bod.d);
}

void sw_set_double_12345678(pd, d)
double_12345678 *pd;
double d;
{
  union {
    double_12345678 d_12345678;
    double d;
  } bod;
  bod.d = d;
  (*pd)[BO64_0] = bod.d_12345678[0];
  (*pd)[BO64_1] = bod.d_12345678[1];
  (*pd)[BO64_2] = bod.d_12345678[2];
  (*pd)[BO64_3] = bod.d_12345678[3];
  (*pd)[BO64_4] = bod.d_12345678[4];
  (*pd)[BO64_5] = bod.d_12345678[5];
  (*pd)[BO64_6] = bod.d_12345678[6];
  (*pd)[BO64_7] = bod.d_12345678[7];
}

如果不与访问器一起使用，这些类型def的好处是会引发编译器错误，从而减少被遗忘的访问器错误。

typedef char int8_1[1], uint8_1[1];

typedef char int16_12[2], uint16_12[2]; /* little endian */
typedef char int16_21[2], uint16_21[2]; /* big endian */

typedef char int24_321[3], uint24_321[3]; /* Alpha Micro, PDP-11 */

typedef char int32_1234[4], uint32_1234[4]; /* little endian */
typedef char int32_3412[4], uint32_3412[4]; /* Alpha Micro, PDP-11 */
typedef char int32_4321[4], uint32_4321[4]; /* big endian */

typedef char int64_12345678[8], uint64_12345678[8]; /* little endian */
typedef char int64_34128756[8], uint64_34128756[8]; /* Alpha Micro, PDP-11 */
typedef char int64_87654321[8], uint64_87654321[8]; /* big endian */

typedef char float_1234[4]; /* little endian */
typedef char float_3412[4]; /* Alpha Micro, PDP-11 */
typedef char float_4321[4]; /* big endian */

typedef char double_12345678[8]; /* little endian */
typedef char double_78563412[8]; /* Alpha Micro? */
typedef char double_87654321[8]; /* big endian */

虽然没有使用固有函数有效，但肯定是可移植的。我的回答:

#include <cstdint>
#include <type_traits>

/**
 * Perform an endian swap of bytes against a templatized unsigned word.
 *
 * @tparam value_type The data type to perform the endian swap against.
 * @param value       The data value to swap.
 *
 * @return value_type The resulting swapped word.
 */
template <typename value_type>
constexpr inline auto endian_swap(value_type value) -> value_type
{
    using half_type = typename std::conditional<
        sizeof(value_type) == 8u,
        uint32_t,
        typename std::conditional<sizeof(value_type) == 4u, uint16_t, uint8_t>::
            type>::type;

    size_t const    half_bits  = sizeof(value_type) * 8u / 2u;
    half_type const upper_half = static_cast<half_type>(value >> half_bits);
    half_type const lower_half = static_cast<half_type>(value);

    if (sizeof(value_type) == 2u)
    {
        return (static_cast<value_type>(lower_half) << half_bits) | upper_half;
    }

    return ((static_cast<value_type>(endian_swap(lower_half)) << half_bits) |
            endian_swap(upper_half));
}