基准

Question

有问题的行在这里：

memcpy(v[0], b_y1[0], 160U * sizeof(double));

其中v和b_y1构成双精度数组（double[]）。

这条线到底在做什么？它是通过将MatLab从C ++转换为C＃生成的。如果需要，我可以在下面提供完整的功能。

Answer 1

您可以通过多种方式做到这一点。如果您选择并使用memcpy

，则导入Dll

但是您也可以使用

Buffer.BlockCopy

从源数组中复制指定数量的字节，起始于a 从特定位置开始到目标数组的特定偏移量偏移量。

Array.Copy

将一个数组中的一系列元素复制到另一个数组中并执行根据需要键入投放和拳击内容。

按建议的方式逐项手动复制项目
使用fixed和unsafe以及指针

但是，如果您确实想要速度，Array.Copy和Buffer.BlockCopy可能是您最快和最接近的比赛

基准

因为我很烦，所以我打开了基准测试器。赛马！

----------------------------------------------------------------------------
Operating System : Microsoft Windows 10 Pro
Version          : 10.0.17134
----------------------------------------------------------------------------
CPU Name         : Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz
Description      : Intel64 Family 6 Model 58 Stepping 9
Cores (Threads)  : 4 (8)      : Architecture  : x64
Clock Speed      : 3901 MHz   : Bus Speed     : 100 MHz
L2Cache          : 1 MB       : L3Cache       : 8 MB
----------------------------------------------------------------------------

Total Benchmarks : Inputs (1) * Scales (7) * Benchmarks (4) * Runs (500) = 14,000

结果1

----------------------------------------------------------------------------
Mode             : Release (64Bit)
Test Framework   : .NET Framework 4.7.1 (CLR 4.0.30319.42000)
----------------------------------------------------------------------------

Total Benchmarks : Inputs (1) * Scales (7) * Benchmarks (4) * Runs (500) = 14,000

NET Framework 4.7.1

--- Standard input --------------------------------------------------------------------
| Value           |   Average |   Fastest |    Cycles |    Garbage | Test |      Gain |
--- Scale 10 ----------------------------------------------------------- Time 0.994 ---
| ArrayCopy       |  0.005 ms |  0.004 ms |  23.016 K |   7.927 KB | N/A  |   14.98 % |
| ElemCopy        |  0.006 ms |  0.004 ms |  23.591 K |   8.000 KB | N/A  |   12.08 % |
| ElemCopy Unsafe |  0.006 ms |  0.004 ms |  26.426 K |   7.915 KB | N/A  |    0.77 % |
| BlockCopy       |  0.006 ms |  0.004 ms |  26.295 K |   7.836 KB | Base |    0.00 % |
--- Scale 100 ---------------------------------------------------------- Time 1.124 ---
| ArrayCopy       |  0.005 ms |  0.004 ms |  20.581 K |   7.937 KB | N/A  |    1.60 % |
| BlockCopy       |  0.005 ms |  0.004 ms |  20.915 K |   8.000 KB | Base |    0.00 % |
| ElemCopy        |  0.005 ms |  0.004 ms |  21.836 K |   8.000 KB | N/A  |   -4.15 % |
| ElemCopy Unsafe |  0.005 ms |  0.004 ms |  22.357 K |   7.970 KB | N/A  |  -10.85 % |
--- Scale 1,000 -------------------------------------------------------- Time 1.322 ---
| ArrayCopy       |  0.005 ms |  0.005 ms |  23.106 K |   8.000 KB | N/A  |    6.31 % |
| ElemCopy Unsafe |  0.006 ms |  0.005 ms |  24.075 K |   8.000 KB | N/A  |    1.41 % |
| ElemCopy        |  0.006 ms |  0.005 ms |  24.392 K |   8.000 KB | N/A  |    0.46 % |
| BlockCopy       |  0.006 ms |  0.004 ms |  24.766 K |   8.015 KB | Base |    0.00 % |
--- Scale 10,000 ------------------------------------------------------- Time 1.727 ---
| BlockCopy       |  0.013 ms |  0.009 ms |  51.749 K |  86.172 KB | Base |    0.00 % |
| ArrayCopy       |  0.016 ms |  0.014 ms |  61.467 K |  86.172 KB | N/A  |  -23.61 % |
| ElemCopy Unsafe |  0.017 ms |  0.015 ms |  63.659 K |  86.172 KB | N/A  |  -28.22 % |
| ElemCopy        |  0.019 ms |  0.016 ms |  70.479 K |  86.172 KB | N/A  |  -41.93 % |
--- Scale 100,000 ------------------------------------------------------ Time 1.825 ---
| BlockCopy       |  0.050 ms |  0.045 ms | 178.829 K | 789.273 KB | Base |    0.00 % |
| ArrayCopy       |  0.101 ms |  0.089 ms | 357.518 K | 789.273 KB | N/A  | -102.24 % |
| ElemCopy Unsafe |  0.121 ms |  0.108 ms | 428.179 K | 789.273 KB | N/A  | -143.35 % |
| ElemCopy        |  0.133 ms |  0.118 ms | 469.168 K | 789.273 KB | N/A  | -166.41 % |
--- Scale 1,000,000 ---------------------------------------------------- Time 4.946 ---
| BlockCopy       |  0.494 ms |  0.409 ms |   1.730 M |   7.637 MB | Base |    0.00 % |
| ElemCopy Unsafe |  1.336 ms |  1.164 ms |   4.674 M |   7.637 MB | N/A  | -170.59 % |
| ArrayCopy       |  1.478 ms |  1.298 ms |   5.169 M |   7.637 MB | N/A  | -199.40 % |
| ElemCopy        |  1.910 ms |  1.607 ms |   6.675 M |   7.637 MB | N/A  | -286.95 % |
--- Scale 10,000,000 -------------------------------------------------- Time 31.376 ---
| BlockCopy       |  5.408 ms |  4.589 ms |  18.896 M |  76.302 MB | Base |    0.00 % |
| ElemCopy Unsafe | 43.981 ms | 35.344 ms | 153.137 M |  76.302 MB | N/A  | -713.21 % |
| ElemCopy        | 46.318 ms | 37.623 ms | 161.225 M |  76.302 MB | N/A  | -756.43 % |
| ArrayCopy       | 48.171 ms | 38.471 ms | 167.548 M |  76.302 MB | N/A  | -790.69 % |
---------------------------------------------------------------------------------------

结果2

----------------------------------------------------------------------------
Mode             : Release (64Bit)
Test Framework   : .Net Core 2.0 (CLR 4.0.30319.42000)
----------------------------------------------------------------------------

Total Benchmarks : Inputs (1) * Scales (7) * Benchmarks (4) * Runs (500) = 14,000

.Net Core 2.0

测试1

--- Standard input ----------------------------------------------------------------------
| Value           |   Average |   Fastest |    Cycles |    Garbage | Test |        Gain |
--- Scale 10 ------------------------------------------------------------- Time 1.221 ---
| ElemCopy Unsafe |  0.006 ms |  0.003 ms |  23.940 K |   8.000 KB | N/A  |     13.53 % |
| BlockCopy       |  0.007 ms |  0.004 ms |  27.573 K |   7.923 KB | Base |      0.00 % |
| ArrayCopy       |  0.007 ms |  0.004 ms |  28.341 K |   7.914 KB | N/A  |     -5.98 % |
| ElemCopy        |  0.007 ms |  0.003 ms |  28.939 K |   7.914 KB | N/A  |     -6.12 % |
--- Scale 100 ------------------------------------------------------------ Time 1.333 ---
| BlockCopy       |  0.005 ms |  0.004 ms |  19.855 K |   7.970 KB | Base |      0.00 % |
| ElemCopy        |  0.005 ms |  0.004 ms |  22.061 K |   7.950 KB | N/A  |    -11.44 % |
| ArrayCopy       |  0.005 ms |  0.004 ms |  22.793 K |   8.000 KB | N/A  |    -15.94 % |
| ElemCopy Unsafe |  0.006 ms |  0.004 ms |  23.715 K |   7.999 KB | N/A  |    -21.65 % |
--- Scale 1,000 ---------------------------------------------------------- Time 1.464 ---
| BlockCopy       |  0.005 ms |  0.004 ms |  21.045 K |   8.001 KB | Base |      0.00 % |
| ElemCopy Unsafe |  0.005 ms |  0.004 ms |  21.731 K |   8.016 KB | N/A  |     -5.53 % |
| ElemCopy        |  0.006 ms |  0.004 ms |  24.120 K |   8.000 KB | N/A  |    -17.49 % |
| ArrayCopy       |  0.006 ms |  0.004 ms |  27.113 K |   8.013 KB | N/A  |    -31.62 % |
--- Scale 10,000 --------------------------------------------------------- Time 1.846 ---
| BlockCopy       |  0.010 ms |  0.008 ms |  37.962 K |  86.172 KB | Base |      0.00 % |
| ArrayCopy       |  0.018 ms |  0.014 ms |  67.134 K |  86.172 KB | N/A  |    -83.72 % |
| ElemCopy Unsafe |  0.019 ms |  0.014 ms |  72.097 K |  86.172 KB | N/A  |    -99.35 % |
| ElemCopy        |  0.021 ms |  0.016 ms |  77.657 K |  86.172 KB | N/A  |   -113.22 % |
--- Scale 100,000 -------------------------------------------------------- Time 2.880 ---
| BlockCopy       |  0.027 ms |  0.020 ms | 100.355 K | 789.305 KB | Base |      0.00 % |
| ArrayCopy       |  0.385 ms |  0.289 ms |   1.346 M | 789.305 KB | N/A  | -1,305.09 % |
| ElemCopy Unsafe |  0.404 ms |  0.280 ms |   1.414 M | 789.305 KB | N/A  | -1,374.97 % |
| ElemCopy        |  0.408 ms |  0.322 ms |   1.424 M | 789.305 KB | N/A  | -1,389.61 % |
--- Scale 1,000,000 ----------------------------------------------------- Time 11.892 ---
| BlockCopy       |  0.632 ms |  0.415 ms |   2.198 M |   7.637 MB | Base |      0.00 % |
| ElemCopy        |  4.645 ms |  3.347 ms |  16.159 M |   7.637 MB | N/A  |   -635.36 % |
| ArrayCopy       |  4.706 ms |  3.684 ms |  16.376 M |   7.637 MB | N/A  |   -645.01 % |
| ElemCopy Unsafe |  4.774 ms |  3.467 ms |  16.568 M |   7.637 MB | N/A  |   -655.66 % |
--- Scale 10,000,000 ---------------------------------------------------- Time 35.806 ---
| BlockCopy       |  6.116 ms |  4.635 ms |  21.294 M |  76.302 MB | Base |      0.00 % |
| ElemCopy Unsafe | 44.132 ms | 35.039 ms | 153.807 M |  76.302 MB | N/A  |   -621.63 % |
| ArrayCopy       | 49.990 ms | 40.360 ms | 173.860 M |  76.302 MB | N/A  |   -717.41 % |
| ElemCopy        | 50.552 ms | 38.044 ms | 175.432 M |  76.302 MB | N/A  |   -726.60 % |
-----------------------------------------------------------------------------------------

输入

public static double[] ArrayOfDouble(int scale)
{
   return Enumerable.Range(1, scale)
                    .Select(x =>_rand.NextDouble())
                    .ToArray();
}

代码

[Test("BlockCopy", "", true)]
public double[] Test1(double[] input, int scale)
{
   var result = new double[scale];
   Buffer.BlockCopy(input, 0, result, 0, input.Length);
   return result;
}

[Test("ArrayCopy", "", false)]
public  double[] Test2(double[] input, int scale)
{
   var result = new double[scale];
   Array.Copy(input, 0, result, 0, input.Length);
   return result;
}

[Test("ElemCopy", "", false)]
public  double[] Test3(double[] input, int scale)
{
   var result = new double[scale];
   for (var i=0; i <input.Length; i++)
      result[i] = input[i];
   return result;
}

[Test("ElemCopy Unsafe", "", false)]
unsafe public  double[] Test4(double[] input, int scale)
{
   var result = new double[scale];
   fixed(double* pInput = input, pResult = result)
   for (var i=0; i <input.Length; i++)
      *(pResult+i) = *(pResult+i);
   return result;
}

摘要

这应该与一粒盐一起服用，任何真实的测试都应该在自己的系统上进行。但这是对我缩放后得到的东西的很好估计

在这种情况下，BlockCopy的表现确实不错，ArrayCopy的表现却很差，我不确定为什么，Unsafe在这类事情上总是会带来性能上的好处，但无法与之相提并论。专用的BCL方法。

Answer 2

这只是将数据从b_y1复制到v（160倍）。

最明显的替代品是这样的：

for (int i=0; i<160; i++)
    v[i] = b_y1[i];

在从C ++转换为C＃的代码中，应该使用什么代替memcpy？

2 个答案:

基准

结果1

结果2

摘要