多维阵列初始化器的性能下降

Question

我有一些奇怪的性能结果，我无法完全解释。 看来，这个线

d = new double[4, 4]{{1, 0, 0, 0}, 
        {0, 1, 0, 0}, 
        {0, 0, 1, 0}, 
        {0, 0, 0, 1},}; 

比这个

d = new double[4, 4]; 
d[0, 0] = 1; d[0, 1] = 0; d[0, 2] = 0; d[0, 3] = 0; 
d[1, 0] = 0; d[1, 1] = 1; d[1, 2] = 0; d[1, 3] = 0; 
d[2, 0] = 0; d[2, 1] = 0; d[2, 2] = 1; d[2, 3] = 0; 
d[3, 0] = 0; d[3, 1] = 0; d[3, 2] = 0; d[3, 3] = 1; 

慢4倍（这甚至没有考虑到一个事实，在这个例子中，我可以离开了所有那些= 0分配）

我知道在c＃中的多维数组上循环会因边界检查而变慢。但是这里没有循环，不需要边界检查，整个数组初始化行可以在编译时解析。

然而，第二个代码块必须首先将数组初始化为零，然后分别覆盖每个值。
那么这里有什么问题？

如果性能是一个问题，那么初始化这个数组的最佳方法是什么？

---

我用下面的代码来衡量性能：

using System; 
using System.Diagnostics; 
class Program 
{ 
    public static double[,] d; // global static variable to prevent the JIT optimizing it away 

    static void Main(string[] args) 
    { 
     Stopwatch watch; 
     int numIter = 10000000; // repeat all tests this often 

     double[,] d2 = new double[4, 4]{{1, 0, 0, 0}, 
             {0, 1, 0, 0}, 
             {0, 0, 1, 0}, 
             {0, 0, 0, 1},}; 

     // ================================================================ 
     // use arrayInitializer: slowest 
     watch = Stopwatch.StartNew(); 
     for (int i = 0; i < numIter; i++) 
     { 
      d = new double[4, 4]{{1, 0, 0, 0}, 
           {0, 1, 0, 0}, 
           {0, 0, 1, 0}, 
           {0, 0, 0, 1},}; 
     } 
     Console.WriteLine("ArrayInitializer: \t{0:0.##########}ms", watch.ElapsedMilliseconds * 1.0/numIter); 

     // ================================================================ 
     // use Array.Copy: faster 
     watch = Stopwatch.StartNew(); 
     for (int i = 0; i < numIter; i++) 
     { 
      d = new double[4, 4]; 
      Array.Copy(d2, d, d2.Length); 
     } 
     Console.WriteLine("new + Array.Copy: \t{0:0.##########}ms", watch.ElapsedMilliseconds * 1.0/numIter); 

     // ================================================================ 
     // direct assignment: fastest 
     watch = Stopwatch.StartNew(); 
     for (int i = 0; i < numIter; i++) 
     { 
      d = new double[4, 4]; 
      d[0, 0] = 1; d[0, 1] = 0; d[0, 2] = 0; d[0, 3] = 0; 
      d[1, 0] = 0; d[1, 1] = 1; d[1, 2] = 0; d[1, 3] = 0; 
      d[2, 0] = 0; d[2, 1] = 0; d[2, 2] = 1; d[2, 3] = 0; 
      d[3, 0] = 0; d[3, 1] = 0; d[3, 2] = 0; d[3, 3] = 1; 
     } 
     Console.WriteLine("direct assignment: \t{0:0.##########}ms", watch.ElapsedMilliseconds * 1.0/numIter); 
    } 
} 

---

结果：

ArrayInitializer:  0,0007917ms 
new + Array.Copy:  0,0002739ms 
direct assignment:  0,0002281ms 

Answer 1

这里是数组的初始化的很好的解释，为什么你看到了这些不同的结果：http://bartdesmet.net/blogs/bart/archive/2008/08/21/how-c-array-initializers-work.aspx

基本上，数组初始值设定项涉及创建自定义结构，而直接分配每个项目只是在堆栈中进行直接分配，尽管速度更快。