多維陣列初始化器的性能下降

Question

我有一些奇怪的性能結果，我無法完全解釋。 看來，這個線

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

比這個

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（這甚至沒有考慮到一個事實，在這個例子中，我可以離開了所有那些= 0分配）

我知道在c＃中的多維數組上循環會因邊界檢查而變慢。但是這裏沒有循環，不需要邊界檢查，整個數組初始化行可以在編譯時解析。

然而，第二個代碼塊必須首先將數組初始化爲零，然後分別覆蓋每個值。
那麼這裏有什麼問題？

如果性能是一個問題，那麼初始化這個數組的最佳方法是什麼？

---

我用下面的代碼來衡量性能：

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); 
    } 
} 

---

結果：

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

Answer 1

這裏是數組的初始化的很好的解釋，爲什麼你看到了這些不同的結果：http://bartdesmet.net/blogs/bart/archive/2008/08/21/how-c-array-initializers-work.aspx

基本上，數組初始值設定項涉及創建自定義結構，而直接分配每個項目只是在堆棧中進行直接分配，儘管速度更快。