.NET等效的_byteswap_ulong/OSSwapInt32 /或bswap32

Question

有没有人知道是否有一个.NET Framework等价于交换uint中的字节顺序？

我正在尝试移植一些使用MSFT _byteswap_ulong的自定义c散列代码(相当于App Nsix中的Apples OSSwapInt32或Bswap32)的到c#.我可以手动编写这个函数,但我怀疑它会利用任何编译器优化(例如c/c ++编译器提供难以超越的内在函数我希望运行时与内置函数相同).如果重要,我不关心保留字节序.

我尝试过基于通用的解决方案,但我不相信这会是最佳的.

BitConverter.ToUInt32(BitConverter.GetBytes(g).Reverse().ToArray<byte>(),0);

编辑:

所以我想出了这个特定函数平均在十分钟内调用了多少次(对于我们的一个哈希消费者).此函数被调用10000000000次.因此,我设置了一些微观分析,以查看c代码与下面提供的解决方案(以及上面提出的解决方案)的性能.

C代码运行许多操作(使用内在)大约.在我可靠的笔记本电脑上1500毫秒.我上面提到的c#代码大约运行2689581 ms.一个巨大的差异.Matthew Watson提供的c#代码运行了将近36000毫秒.Caramiriel提出的第一个解决方案在大约115014毫秒内运行,第二个解决方案提供了近36000个.

虽然这些解决方案都没有接近内在调用的速度,但它们比我原来的解决方案要好得多(对于那么多计算,从44分钟到36秒).这对我的应用程序完全可以接受.虽然.NET编译器提供了一些与本机编译器相同的内在功能,但这样会很好.

为了完整性,这里是我的微基准测试的C代码:

#include "stdafx.h"
#include "windows.h"

unsigned long Swap(unsigned int value)
{
    return _byteswap_uint64(value);
}

int _tmain(int argc, _TCHAR* argv[])
{
    unsigned int value = 0x01020304;
    unsigned long NUMITER = 10000000000;
    unsigned long a=0;
    unsigned long z=0;
    int throwAwayLoopCount = 5;

    for (int k = 0; k < throwAwayLoopCount; ++k)
    {
        a = GetTickCount();
        for (unsigned long  i = 0; i < NUMITER; ++i)
        {
            value = Swap(value);
        }
        z = GetTickCount();
        printf("Baseline, Cached: time is %4lld milliseconds: value%4lld\n", z-a,value);
    }

    printf("Baseline, Cached: time is %4lld milliseconds\n", z-a);

    return 0;
}

以下是用于对提供的解决方案进行基准测试的c#代码:

namespace ByteSwapProfiler
{
    using System.Runtime.InteropServices;
    using System.Diagnostics;

    [StructLayout(LayoutKind.Explicit)]
    internal struct UInt32Union
    {
        [FieldOffset(0)]
        public UInt32 Value;
        [FieldOffset(0)]
        public byte Byte1;
        [FieldOffset(1)]
        public byte Byte2;
        [FieldOffset(2)]
        public byte Byte3;
        [FieldOffset(3)]
        public byte Byte4;
    }


    class Program
    {

        static uint ByteSwapNaive(uint g)
        {
            return BitConverter.ToUInt32(BitConverter.GetBytes(g).Reverse().ToArray<byte>(), 0);
        }

        static uint ByteSwapCaramiriel1(uint value)
        {
            unchecked
            {
                return ((value & 0xff000000) >> 24) |
                        ((value & 0x00ff0000) >> 8) |
                        ((value & 0x0000ff00) << 8) |
                        ((value & 0x000000ff) << 24);
            }
        }

        static uint ByteSwapCaramiriel2(UInt32Union src)
        {
            UInt32Union dest = new UInt32Union
                {
                    Byte1 = src.Byte4,
                    Byte2 = src.Byte3,
                    Byte3 = src.Byte2,
                    Byte4 = src.Byte1
                };

            return dest.Value;
        }

        static uint ByteSwapMatthewWatson(uint word)
        {
            return ((word >> 24) & 0x000000FF) | ((word >> 8) & 0x0000FF00) | ((word << 8) & 0x00FF0000) | ((word << 24) & 0xFF000000);            
        }

        static void Main(string[] args)
        {
            uint value= 0x01020304;
            UInt32Union src = new UInt32Union();
            src.Value = value;

            ulong NUMITER = 10000000000;
            uint throwAwayLoopCount = 5;
            var sw = new Stopwatch();
            string name = "Naive";
            //for (int k = 0; k < throwAwayLoopCount; ++k)
            {
                sw = Stopwatch.StartNew();
                for (ulong i = 0; i < NUMITER; ++i)
                {
                    value = ByteSwapNaive(value);
                }
                sw.Stop();
                Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.  Value:{2} \n", name, (sw.ElapsedMilliseconds).ToString("0"),value);
            }

            Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.\n", name, (sw.ElapsedMilliseconds).ToString("0"));


            name = "MatthewWatson";
            for (int k = 0; k < throwAwayLoopCount; ++k)
            {
                sw = Stopwatch.StartNew();
                for (ulong i = 0; i < NUMITER; ++i)
                {
                    value = ByteSwapMatthewWatson(value);
                }
                sw.Stop();
                Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.  Value:{2} \n", name, (sw.ElapsedMilliseconds).ToString("0"), value);
            }
            Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.\n", name, (sw.ElapsedMilliseconds).ToString("0"));

            name = "Caramiriel2";
            for (int k = 0; k < throwAwayLoopCount; ++k)
            {
                sw = Stopwatch.StartNew();
                for (ulong i = 0; i < NUMITER; ++i)
                {
                    value = ByteSwapCaramiriel2(src);
                }
                sw.Stop();
                Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.  Value:{2} \n", name, (sw.ElapsedMilliseconds).ToString("0"), value);
            }

            Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.\n", name, (sw.ElapsedMilliseconds).ToString("0"));

            name = "Caramiriel1";
            for (int k = 0; k < throwAwayLoopCount; ++k)
            {
                sw = Stopwatch.StartNew();
                for (ulong i = 0; i < NUMITER; ++i)
                {
                    value = ByteSwapCaramiriel1(value);
                }
                sw.Stop();
                Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.  Value:{2} \n", name, (sw.ElapsedMilliseconds).ToString("0"), value);
            }

            Console.Write("{0,-13}, Cached: time is {1,7} milliseconds.\n", name, (sw.ElapsedMilliseconds).ToString("0"));
        }
    }
}

Answer 1

我不认为你会得到像_byteswap_ulong一样快的东西,但如果你使用这个:

public static uint SwapBytes(uint word)
{
    return ((word>>24)&0x000000FF) | ((word>>8)&0x0000FF00) | ((word<<8)&0x00FF0000) | ((word<<24)&0xFF000000);            
}

至少JIT优化器可能会内联它.

Answer 2

从 .NET Core 2.1 开始，BinaryPrimitives.ReverseEndianness 为该功能提供了优化的软件实现。从 .NET Core 3.0 开始，它使用 JIT 内在函数实现，该内部函数被 jitt 到使用 bswap 指令的非常高效的机器代码中。