不同IComparer对数组排序Array.Sort,Linq orderby的性能的影响

      今天很热闹啊。老赵的数组排序方法的性能比较（中）：Array.Sort<T>实现分析，Ivony的 数组排序LINQ的性能优势初步分析 —— 不起眼的属性等有关linq性能文章让我忍不住写下此文章。

      Ivony的提供的代码中，对数组的排序时采用自定义的PersonComparer，而linq排序使用的确是Comparer<int>.Default，代码分析如下：

      数组排序     

private   static   void  SortWithCustomComparer(Person[] array)
 {
    Array.Sort(array,  new  PersonComparer());
}

    LINQ排序    

private   static   void  SortWithLinq( Person[] array )
{
    var sorted  =
        ( from person  in  array
         orderby person.ID
         select person ).ToList();
}

     上述lambda表达式对应为varsorted = Enumerable.OrderBy<Person,int>(array,person =>person.ID).ToList()。通过reflector反汇编可以得到以下代码

代码

// Enumerable类中OrderBy的代码
public   static  IOrderedEnumerable < TSource >  OrderBy < TSource, TKey > ( this  IEnumerable < TSource >  source, Func < TSource, TKey >  keySelector)
{
     return   new  OrderedEnumerable < TSource, TKey > (source, keySelector,  null ,  false );
}

// 内部类OrderedEnumerable的代码
internal   class  OrderedEnumerable < TElement, TKey >  : OrderedEnumerable < TElement >
{
     //  Fields
     internal  IComparer < TKey >  comparer;
     internal   bool  descending;
     internal  Func < TElement, TKey >  keySelector;
     internal  OrderedEnumerable < TElement >  parent;

     //  Methods
     internal  OrderedEnumerable(IEnumerable < TElement >  source, Func < TElement, TKey >  keySelector, IComparer < TKey >  comparer,  bool  descending)
    {
         if  (source  ==   null )
        {
             throw  Error.ArgumentNull( " source " );
        }
         if  (keySelector  ==   null )
        {
             throw  Error.ArgumentNull( " keySelector " );
        }
         base .source  =  source;
         this .parent  =   null ;
         this .keySelector  =  keySelector;
         this .comparer  =  (comparer  !=   null )  ?  comparer : ((IComparer < TKey > ) Comparer < TKey > .Default);
         this .descending  =  descending;
    }

     internal   override  EnumerableSorter < TElement >  GetEnumerableSorter(EnumerableSorter < TElement >  next)
    {
        EnumerableSorter < TElement >  enumerableSorter  =   new  EnumerableSorter < TElement, TKey > ( this .keySelector,  this .comparer,  this .descending, next);
         if  ( this .parent  !=   null )
        {
            enumerableSorter  =   this .parent.GetEnumerableSorter(enumerableSorter);
        }
         return  enumerableSorter;
    }
}

 

      在文章数组排序方法的性能比较（上）：注意事项及试验，老赵测试得出的结果可以得知PersonComparer比Comparer<int>.Default排序慢。

      在增加一个新的SortWithLinq2方法之后，release运行测试，我得到新的结果。

      新代码： 

代码

using  System;
using  System.Collections.Generic;
using  System.Diagnostics;
using  System.Linq;
using  System.Runtime.InteropServices;
using  System.Threading;

namespace  Exam11
{
     public   static   class  CodeTimer
    {
         public   static   void  Initialize()
        {
            Process.GetCurrentProcess().PriorityClass  =  ProcessPriorityClass.High;
            Thread.CurrentThread.Priority  =  ThreadPriority.Highest;
            Time( "" ,  1 , ()  =>  { });
        }


         public   static   void  Time( string  name,  int  iteration, Action action)
        {
             if  (String.IsNullOrEmpty(name))  return ;
             //  warm up            
            action();
             //  1.           
            ConsoleColor currentForeColor  =  Console.ForegroundColor;
            Console.ForegroundColor  =  ConsoleColor.Yellow;
            Console.WriteLine(name);
             //  2.          
            GC.Collect(GC.MaxGeneration, GCCollectionMode.Forced);
             int [] gcCounts  =   new   int [GC.MaxGeneration  +   1 ];
             for  ( int  i  =   0 ; i  <=  GC.MaxGeneration; i ++ )
            {
                gcCounts[i]  =  GC.CollectionCount(i);
            }
             //  3.         
            Stopwatch watch  =   new  Stopwatch();
            watch.Start();
             ulong  cycleCount  =  GetCycleCount();
             for  ( int  i  =   0 ; i  <  iteration; i ++ ) action();
             ulong  cpuCycles  =  GetCycleCount()  -  cycleCount;
            watch.Stop();
             //  4.           
            Console.ForegroundColor  =  currentForeColor;
            Console.WriteLine( " \tTime Elapsed:\t "   +  watch.ElapsedMilliseconds.ToString( " N0 " )  +   " ms " );
            Console.WriteLine( " \tCPU Cycles:\t "   +  cpuCycles.ToString( " N0 " ));
             //  5.        
             for  ( int  i  =   0 ; i  <=  GC.MaxGeneration; i ++ )
            {
                 int  count  =  GC.CollectionCount(i)  -  gcCounts[i];
                Console.WriteLine( " \tGen  "   +  i  +   " : \t\t "   +  count);
            }
            Console.WriteLine();
        }
         private   static   ulong  GetCycleCount()
        {
             ulong  cycleCount  =   0 ;
            QueryThreadCycleTime(GetCurrentThread(),  ref  cycleCount);
             return  cycleCount;
        }
        [DllImport( " kernel32.dll " )]
        [ return : MarshalAs(UnmanagedType.Bool)]
         static   extern   bool  QueryThreadCycleTime(IntPtr threadHandle,  ref   ulong  cycleTime);
        [DllImport( " kernel32.dll " )]
         static   extern  IntPtr GetCurrentThread();
    }



     class  Program
    {
         static   void  Main( string [] args)
        {
            var random  =   new  Random(DateTime.Now.Millisecond);
            var array  =  Enumerable.Repeat( 0 ,  50000 ).Select(_  =>   new  Person { ID  =  random.Next() }).ToArray();

            JetBrains.dotTrace.Api.CPUProfiler.Start();

             // 老赵程序
            CodeTimer.Initialize();
            CodeTimer.Time( " SortWithCustomComparer " ,  500 , ()  =>  SortWithCustomComparer(CloneArray(array)));
            CodeTimer.Time( " SortWithLinq2 " ,  500 , ()  =>  SortWithLinq(CloneArray(array)));
            CodeTimer.Time( " SortWithLinq " ,  500 , ()  =>  SortWithLinq2(CloneArray(array)));

            JetBrains.dotTrace.Api.CPUProfiler.StopAndSaveSnapShot();                   
        }

         public   static   void  Time( string  name,  int  iteration, Action action)
        {
            action();

            Console.WriteLine(name);

             for  ( int  i  =   0 ; i  <  iteration; i ++ ) action();
        }

         private   static   readonly  PersonComparer comparer  =   new  PersonComparer();

         private   static   void  SortWithCustomComparer(Person[] array)
        {
            Array.Sort(array, comparer);
        }

         private   static   void  SortWithLinq(Person[] array)
        {
             // array = (from person in array
             //          orderby person.ID
             //          select person).ToList();
             Enumerable.OrderBy < Person, int > (array,person  => person.ID).ToList();
        }

         private   static   void  SortWithLinq2(Person[] array)
        {
             // array = (from person in array
             //          orderby person.ID
             //          select person).ToList();
            Enumerable.OrderBy < Person, Person > (array, person  =>  person, comparer).ToList();
        }

         private   static  T[] CloneArray < T > (T[] source)
        {
            var dest  =   new  T[source.Length];
            Array.Copy(source, dest, source.Length);
             return  dest;
        }

    }
     public   class  Person
    {
         public   string  FirstName
        {
             get ;
             set ;
        }

         public   string  LastName
        {
             get ;
             set ;
        }

         public   int  ID
        {
             get ;
             set ;
        }

         public   override   int  GetHashCode()
        {
             return   this .ID.GetHashCode();
        }
    }

     public   class  PersonComparer : IComparer < Person >
    {
         public   int  Compare(Person x, Person y)
        {
             return  x.ID  -  y.ID;
        }

    }
}

 

 

    不使用dottrace探测得到测试结果（使用dottrace之后运行态慢了，而且不好得到控制台的显示结果，除非使用辅助软件）：

SortWithCustomComparer
        Time Elapsed:   11,446ms
        CPU Cycles:     26,201,674,700
        Gen 0:          31
        Gen 1:          31
        Gen 2:          31

SortWithLinq
        Time Elapsed:   14,970ms
        CPU Cycles:     34,259,890,929
        Gen 0:          311
        Gen 1:          310
        Gen 2:          187

SortWithLinq2
        Time Elapsed:   18,573ms
        CPU Cycles:     42,491,287,773
        Gen 0:          311
        Gen 1:          311
        Gen 2:          187 

       使用dottrace的分析图如如下（启动dottrace后程序运行的时间比之前的慢多了）：

 

      上述结果测试表明，在使用PersonComparer之后SortWithLinq2比SortWithLinq慢了差不多一倍。在此也印证老赵的相关分析。