代表与方法-为什么我得到奇怪的结果

Question

当我实现xxHash时，我尝试将内联汇编用作“ ROL”操作。这是我的测试代码：

class Program
{
    private delegate uint RolFunc(uint x, int y);
    private static readonly byte[] _AsmCodeROL64 = new byte[] { 0x8B, 0xC1, 0x8B, 0xCA, 0xD3, 0xC0, 0xC3 };
    /*X64 ASM
    8B C1                - mov eax,ecx
    8B CA                - mov ecx,edx
    D3 C0                - rol eax,cl
    C3                   - ret 
    */
    private static readonly byte[] _AsmCodeROL32 = new byte[] { 0x8B, 0x44, 0x24, 0x08, 0x8B, 0x4C, 0x24, 0x04, 0xD3, 0xC0, 0xC2, 0x08, 0x00 };
    /*X86 ASM
    8B 44 24 08           - mov eax,[esp+08]
    8B 4C 24 04           - mov ecx,[esp+04]
    D3 C0                 - rol eax,cl
    C2 0800               - ret 0008
    */
    private static readonly RolFunc AsmRol;

    [DllImport("kernel32.dll", SetLastError = false, CallingConvention = CallingConvention.Winapi)]
    private static extern IntPtr GetCurrentProcess();

    [DllImport("kernel32.dll", SetLastError = false, CallingConvention = CallingConvention.Winapi)]
    [return: MarshalAs(UnmanagedType.Bool)]
    private unsafe static extern bool VirtualProtectEx([In] IntPtr hProcess, [In] void* lpAddress, [In, MarshalAs(UnmanagedType.SysUInt)] UIntPtr dwSize, [In] uint flNewProtect, [Out] out uint lpflOldProtect);

    private const uint PAGE_EXECUTE_READWRITE = 0x40;

    private static uint RotateLeft32(uint value, int count)
    {
        return (value << count) | (value >> (32 - count));
    }


    static unsafe Program()
    {
        var asmCodeROL = IntPtr.Size == 4 ? _AsmCodeROL32 : _AsmCodeROL64;
        fixed (void* funcAddr = asmCodeROL)
        {
            VirtualProtectEx(GetCurrentProcess(), funcAddr, (UIntPtr)asmCodeROL.Length, PAGE_EXECUTE_READWRITE, out var lastProtect);
            AsmRol = Marshal.GetDelegateForFunctionPointer<RolFunc>(new IntPtr(funcAddr));
        }
    }

    static void Main(string[] args)
    {
        var sw = new Stopwatch();
        sw.Start();
        var asmResult = AsmRol(10, 2);
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        var mtdResult = RotateLeft32(10, 2);
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        asmResult = AsmRol(11, 2);
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        mtdResult = RotateLeft32(11, 2);
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        asmResult = AsmRol(12, 2);
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        mtdResult = RotateLeft32(12, 2);
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        for (uint i = 0; i < 0x10000; i++)
        {
            asmResult = AsmRol(i, 2);
        }
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
        sw.Reset();
        sw.Start();
        for (uint i = 0; i < 0x10000; i++)
        {
            mtdResult = RotateLeft32(i, 2);
        }
        sw.Stop();
        Console.WriteLine(sw.Elapsed.TotalMilliseconds);
    }
}

结果如下：

  

0.0108   0.4562   0.002   0.0008   0.0014   0.0008   8.892   0.9648

我发现内联汇编的代码长度比该方法短得多，但是为什么除了第一次之外，AsmRol却比RotateLeft32慢得多？