在编写加密代码时，通常避免依赖数据的分支和表查找，以确保运行时不依赖于数据，因为依赖数据的计时可能会导致侧通道攻击。

它也很快。

static string ByteToHexBitFiddle(byte[] bytes)
{
    char[] c = new char[bytes.Length * 2];
    int b;
    for (int i = 0; i < bytes.Length; i++) {
        b = bytes[i] >> 4;
        c[i * 2] = (char)(55 + b + (((b-10)>>31)&-7));
        b = bytes[i] & 0xF;
        c[i * 2 + 1] = (char)(55 + b + (((b-10)>>31)&-7));
    }
    return new string(c);
}

Ph'nglui mglw'naph Cthulhu R'lieh wgah'nagl fhtagn公司

---

你们这些进入这里的人，放弃一切希望吧

一个奇怪的拨弄解释：

bytes[i]>>4提取字节的高位半字节bytes[i]&0xF提取字节的低位半字节b-10对于值b<10，为<0，将变为十进制数字对于值b>10，为>=0，这将成为从a到F的字母。在有符号32位整数上使用i>>31可以提取符号，这得益于符号扩展。当i<0时为-1，当i>=0时为0。结合2）和3），表明（b-10）>>31将是字母0，数字-1。看看字母的大小写，最后一个被加数变为0，b在10到15的范围内。我们希望将其映射到A（65）到F（70），这意味着添加55（'A'-10）。看看数字的情况，我们希望调整最后一个被加数，使其将b从范围0到9映射到范围0（48）到9（57）。这意味着它需要变为-7（'0'-55）。现在我们可以乘以7。但由于-1由所有位表示为1，因此我们可以改用&-7，因为（0&-7）==0和（-1&-7）==-7。

进一步考虑：

我没有使用第二个循环变量来索引c，因为测量表明从I计算它更便宜。正好使用i<bytes.Length作为循环的上限允许JITter消除对bytes[i]的边界检查，所以我选择了这个变量。将b设为int允许不必要的从和到字节的转换。

如果性能很重要，这里有一个优化的解决方案：

    static readonly char[] _hexDigits = "0123456789abcdef".ToCharArray();
    public static string ToHexString(this byte[] bytes)
    {
        char[] digits = new char[bytes.Length * 2];
        for (int i = 0; i < bytes.Length; i++)
        {
            int d1, d2;
            d1 = Math.DivRem(bytes[i], 16, out d2);
            digits[2 * i] = _hexDigits[d1];
            digits[2 * i + 1] = _hexDigits[d2];
        }
        return new string(digits);
    }

它比BitConverter.ToString快2.5倍，比BitConverter.ToString+删除“-”字符快7倍。

如果你想得到wcoenen报告的“4倍速度增长”，那么如果不明显：用hex[i]+hex[i+1]替换hex.Substring（i，2）

您还可以再进一步，通过在两个地方使用i++来消除i+=2。

另一种基于查找表的方法。该方法只为每个字节使用一个查找表，而不是为每个半字节使用查找表。

private static readonly uint[] _lookup32 = CreateLookup32();

private static uint[] CreateLookup32()
{
    var result = new uint[256];
    for (int i = 0; i < 256; i++)
    {
        string s=i.ToString("X2");
        result[i] = ((uint)s[0]) + ((uint)s[1] << 16);
    }
    return result;
}

private static string ByteArrayToHexViaLookup32(byte[] bytes)
{
    var lookup32 = _lookup32;
    var result = new char[bytes.Length * 2];
    for (int i = 0; i < bytes.Length; i++)
    {
        var val = lookup32[bytes[i]];
        result[2*i] = (char)val;
        result[2*i + 1] = (char) (val >> 16);
    }
    return new string(result);
}

我还使用查找表中的ushort、struct｛char X1，X2｝、struct{byte X1，X2}测试了这个变体。

根据编译目标（x86、X64）的不同，它们要么具有大致相同的性能，要么稍慢于此变体。

---

为了获得更高的性能，其不安全的兄弟：

private static readonly uint[] _lookup32Unsafe = CreateLookup32Unsafe();
private static readonly uint* _lookup32UnsafeP = (uint*)GCHandle.Alloc(_lookup32Unsafe,GCHandleType.Pinned).AddrOfPinnedObject();

private static uint[] CreateLookup32Unsafe()
{
    var result = new uint[256];
    for (int i = 0; i < 256; i++)
    {
        string s=i.ToString("X2");
        if(BitConverter.IsLittleEndian)
            result[i] = ((uint)s[0]) + ((uint)s[1] << 16);
        else
            result[i] = ((uint)s[1]) + ((uint)s[0] << 16);
    }
    return result;
}

public static string ByteArrayToHexViaLookup32Unsafe(byte[] bytes)
{
    var lookupP = _lookup32UnsafeP;
    var result = new char[bytes.Length * 2];
    fixed(byte* bytesP = bytes)
    fixed (char* resultP = result)
    {
        uint* resultP2 = (uint*)resultP;
        for (int i = 0; i < bytes.Length; i++)
        {
            resultP2[i] = lookupP[bytesP[i]];
        }
    }
    return new string(result);
}

或者如果您认为可以直接写入字符串：

public static string ByteArrayToHexViaLookup32UnsafeDirect(byte[] bytes)
{
    var lookupP = _lookup32UnsafeP;
    var result = new string((char)0, bytes.Length * 2);
    fixed (byte* bytesP = bytes)
    fixed (char* resultP = result)
    {
        uint* resultP2 = (uint*)resultP;
        for (int i = 0; i < bytes.Length; i++)
        {
            resultP2[i] = lookupP[bytesP[i]];
        }
    }
    return result;
}

测试：十六进制字符串到字节数组

我注意到，大多数测试都是在将Bytes数组转换为十六进制字符串的函数上执行的。因此，在这篇文章中，我将关注另一方面：将十六进制字符串转换为字节数组的函数。若您只对结果感兴趣，可以跳到“摘要”部分。测试代码文件在文章末尾提供。

标签

我想根据接受的答案（Tomalak）将函数命名为StringToByteArrayV1，或将其快捷到V1。其余函数将以相同的方式命名：V2、V3、V4、…、。。。，等

参与功能索引

Tomalak的StringToByteArrayV1（公认答案）Mykroft的StringToByteArrayV2（使用SoapHexBinary）drphrozen的StringToByteArrayV3（查找表）CoperNick的StringToByteArrayV4（字节操作）Chris F编写的StringToByteArrayV5_1（字节操作）Chris F的StringToByteArrayV5_2（V5_1+根据Amir Rezaei的评论修改）Chris F的StringToByteArrayV5_3（V5_2+根据Ben Voigt的评论对其进行了修改）（您可以在发布后的测试代码中看到它的最终形状）Ben Mosher编写的StringToByteArrayV6（字节操作）Maratius的StringToByteArrayV7（字节操作-安全版本）Maratius的StringToByteArrayV8（字节操作-不安全版本）StringToByteArrayV9（按Geograph）AlejandroAlis编写的StringToByteArrayV10Fredrik Hu编写的StringToByteArrayV11Maarten Bodewes编写的StringToByteArrayV12ClausAndersen编写的StringToByteArrayV13Stas Makutin编写的StringToByteArrayV14JJJ的StringToByteArrayV15JamieSee的StringToByteArrayV16spacepille的StringToByteArrayV17Gregory Morse编写的StringToByteArrayV18Rick编写的StringToByteArrayV19SandRock的StringToByteArrayV20Paul编写的StringToByteArrayV21

正确性测试

我通过传递1字节的所有256个可能值来测试正确性，然后检查输出是否正确。结果：

V18中以“00”开头的字符串有问题（请参阅Roger Stewart对此的评论）。除了通过所有测试。如果十六进制字符串字母是大写的：所有函数都成功传递如果十六进制字符串字母是小写的，则以下函数失败：V5_1、V5_2、v7、V8、V15、V19

注：V5_3解决了这个问题（V5_1和V5_2）

性能测试

我已经使用Stopwatch类进行了性能测试。

长字符串的性能

input length: 10,000,000 bytes
runs: 100
average elapsed time per run:
V1 = 136.4ms
V2 = 104.5ms
V3 = 22.0ms
V4 = 9.9ms
V5_1 = 10.2ms
V5_2 = 9.0ms
V5_3 = 9.3ms
V6 = 18.3ms
V7 = 9.8ms
V8 = 8.8ms
V9 = 10.2ms
V10 = 19.0ms
V11 = 12.2ms
V12 = 27.4ms
V13 = 21.8ms
V14 = 12.0ms
V15 = 14.9ms
V16 = 15.3ms
V17 = 9.5ms
V18 got excluded from this test, because it was very slow when using very long string
V19 = 222.8ms
V20 = 66.0ms
V21 = 15.4ms

V1 average ticks per run: 1363529.4
V2 is more fast than V1 by: 1.3 times (ticks ratio)
V3 is more fast than V1 by: 6.2 times (ticks ratio)
V4 is more fast than V1 by: 13.8 times (ticks ratio)
V5_1 is more fast than V1 by: 13.3 times (ticks ratio)
V5_2 is more fast than V1 by: 15.2 times (ticks ratio)
V5_3 is more fast than V1 by: 14.8 times (ticks ratio)
V6 is more fast than V1 by: 7.4 times (ticks ratio)
V7 is more fast than V1 by: 13.9 times (ticks ratio)
V8 is more fast than V1 by: 15.4 times (ticks ratio)
V9 is more fast than V1 by: 13.4 times (ticks ratio)
V10 is more fast than V1 by: 7.2 times (ticks ratio)
V11 is more fast than V1 by: 11.1 times (ticks ratio)
V12 is more fast than V1 by: 5.0 times (ticks ratio)
V13 is more fast than V1 by: 6.3 times (ticks ratio)
V14 is more fast than V1 by: 11.4 times (ticks ratio)
V15 is more fast than V1 by: 9.2 times (ticks ratio)
V16 is more fast than V1 by: 8.9 times (ticks ratio)
V17 is more fast than V1 by: 14.4 times (ticks ratio)
V19 is more SLOW than V1 by: 1.6 times (ticks ratio)
V20 is more fast than V1 by: 2.1 times (ticks ratio)
V21 is more fast than V1 by: 8.9 times (ticks ratio)

V18的长串性能

V18 took long time at the previous test, 
so let's decrease length for it:  
input length: 1,000,000 bytes
runs: 100
average elapsed time per run: V1 = 14.1ms , V18 = 146.7ms
V1 average ticks per run: 140630.3
V18 is more SLOW than V1 by: 10.4 times (ticks ratio)

短字符串的性能

input length: 100 byte
runs: 1,000,000
V1 average ticks per run: 14.6
V2 is more fast than V1 by: 1.4 times (ticks ratio)
V3 is more fast than V1 by: 5.9 times (ticks ratio)
V4 is more fast than V1 by: 15.7 times (ticks ratio)
V5_1 is more fast than V1 by: 15.1 times (ticks ratio)
V5_2 is more fast than V1 by: 18.4 times (ticks ratio)
V5_3 is more fast than V1 by: 16.3 times (ticks ratio)
V6 is more fast than V1 by: 5.3 times (ticks ratio)
V7 is more fast than V1 by: 15.7 times (ticks ratio)
V8 is more fast than V1 by: 18.0 times (ticks ratio)
V9 is more fast than V1 by: 15.5 times (ticks ratio)
V10 is more fast than V1 by: 7.8 times (ticks ratio)
V11 is more fast than V1 by: 12.4 times (ticks ratio)
V12 is more fast than V1 by: 5.3 times (ticks ratio)
V13 is more fast than V1 by: 5.2 times (ticks ratio)
V14 is more fast than V1 by: 13.4 times (ticks ratio)
V15 is more fast than V1 by: 9.9 times (ticks ratio)
V16 is more fast than V1 by: 9.2 times (ticks ratio)
V17 is more fast than V1 by: 16.2 times (ticks ratio)
V18 is more fast than V1 by: 1.1 times (ticks ratio)
V19 is more SLOW than V1 by: 1.6 times (ticks ratio)
V20 is more fast than V1 by: 1.9 times (ticks ratio)
V21 is more fast than V1 by: 11.4 times (ticks ratio)

测试代码

在使用以下代码之前，最好先阅读本文下面的免责声明部分https://github.com/Ghosticollis/performance-tests/blob/main/MTestPerformance.cs

总结

由于性能良好，我建议使用以下函数之一，并支持大写和小写：

CoperNick的StringToByteArrayV4StringToByteArrayV9（按Geograph）spacepille的StringToByteArrayV17StringToByteArrayV5_3基本上由Chris F开发（它基于V5_1，但我根据Amir Rezaei和Ben Voigt的评论对其进行了增强）。

以下是V5_3的最终形状：

static byte[] HexStringToByteArrayV5_3(string hexString) {
    int hexStringLength = hexString.Length;
    byte[] b = new byte[hexStringLength / 2];
    for (int i = 0; i < hexStringLength; i += 2) {
        int topChar = hexString[i];
        topChar = (topChar > 0x40 ? (topChar & ~0x20) - 0x37 : topChar - 0x30) << 4;
        int bottomChar = hexString[i + 1];
        bottomChar = bottomChar > 0x40 ? (bottomChar & ~0x20) - 0x37 : bottomChar - 0x30;
        b[i / 2] = (byte)(topChar + bottomChar);
    }
    return b;
}

免责声明

警告：我没有适当的测试知识。这些原始测试的主要目的是快速概述所有发布的函数的优点。如果您需要准确的结果，请使用适当的测试工具。

最后，我想说，我是新来的，在斯塔科弗洛活跃，如果我的职位空缺，我很抱歉。如果您能发表评论，我们将不胜感激。

如果您希望比BitConverter更灵活，但不希望使用那些笨重的90年代风格的显式循环，那么您可以这样做：

String.Join(String.Empty, Array.ConvertAll(bytes, x => x.ToString("X2")));

或者，如果您使用的是.NET 4.0：

String.Concat(Array.ConvertAll(bytes, x => x.ToString("X2")));

（后者来自对原帖子的评论。）