Delegate instance allocation with method group compared to

馋奶兔 提交于 2021-01-21 07:31:34

问题


I started to use the method group syntax a couple of years ago based on some suggestion from ReSharper and recently I gave a try to ClrHeapAllocationAnalyzer and it flagged every location where I was using a method group in a lambda with the issue HAA0603 - This will allocate a delegate instance.

As I was curious to see if this suggestion was actually useful, I wrote a simple console app for the 2 cases.

Code1:

class Program
{
    static void Main(string[] args)
    {
        var temp = args.AsEnumerable();

        for (int i = 0; i < 10_000_000; i++)
        {
            temp = temp.Select(x => Foo(x));
        }

        Console.ReadKey();
    }

    private static string Foo(string x)
    {
        return x;
    }
}

Code2:

class Program
{
    static void Main(string[] args)
    {
        var temp = args.AsEnumerable();

        for (int i = 0; i < 10_000_000; i++)
        {
            temp = temp.Select(Foo);
        }

        Console.ReadKey();
    }

    private static string Foo(string x)
    {
        return x;
    }
}

Putting a break point on the Console.ReadKey(); of Code1 shows a memory consumption of ~500MB and on Code2 a consumption of ~800MB. Even if we can argue on whether this test case is good enough to explain something it actually shows a difference.

So I decided to have a look at the IL code produced to try to understand the difference between the 2 code.

IL Code1:

.method private hidebysig static 
    void Main (
        string[] args
    ) cil managed 
{
    // Method begins at RVA 0x2050
    // Code size 75 (0x4b)
    .maxstack 3
    .entrypoint
    .locals init (
        [0] class [mscorlib]System.Collections.Generic.IEnumerable`1<string>,
        [1] int32,
        [2] bool
    )

    //      temp = from x in temp
    //      select Foo(x);
    IL_0000: nop
    // IEnumerable<string> temp = args.AsEnumerable();
    IL_0001: ldarg.0
    IL_0002: call class [mscorlib]System.Collections.Generic.IEnumerable`1<!!0> [System.Core]System.Linq.Enumerable::AsEnumerable<string>(class [mscorlib]System.Collections.Generic.IEnumerable`1<!!0>)
    IL_0007: stloc.0
    // for (int i = 0; i < 10000000; i++)
    IL_0008: ldc.i4.0
    IL_0009: stloc.1
    // (no C# code)
    IL_000a: br.s IL_0038
    // loop start (head: IL_0038)
        IL_000c: nop
        IL_000d: ldloc.0
        IL_000e: ldsfld class [mscorlib]System.Func`2<string, string> ConsoleApp1.Program/'<>c'::'<>9__0_0'
        IL_0013: dup
        IL_0014: brtrue.s IL_002d

        IL_0016: pop
        IL_0017: ldsfld class ConsoleApp1.Program/'<>c' ConsoleApp1.Program/'<>c'::'<>9'
        IL_001c: ldftn instance string ConsoleApp1.Program/'<>c'::'<Main>b__0_0'(string)
        IL_0022: newobj instance void class [mscorlib]System.Func`2<string, string>::.ctor(object, native int)
        IL_0027: dup
        IL_0028: stsfld class [mscorlib]System.Func`2<string, string> ConsoleApp1.Program/'<>c'::'<>9__0_0'

        IL_002d: call class [mscorlib]System.Collections.Generic.IEnumerable`1<!!1> [System.Core]System.Linq.Enumerable::Select<string, string>(class [mscorlib]System.Collections.Generic.IEnumerable`1<!!0>, class [mscorlib]System.Func`2<!!0, !!1>)
        IL_0032: stloc.0
        IL_0033: nop
        // for (int i = 0; i < 10000000; i++)
        IL_0034: ldloc.1
        IL_0035: ldc.i4.1
        IL_0036: add
        IL_0037: stloc.1

        // for (int i = 0; i < 10000000; i++)
        IL_0038: ldloc.1
        IL_0039: ldc.i4 10000000
        IL_003e: clt
        IL_0040: stloc.2
        // (no C# code)
        IL_0041: ldloc.2
        IL_0042: brtrue.s IL_000c
    // end loop

    // Console.ReadKey();
    IL_0044: call valuetype [mscorlib]System.ConsoleKeyInfo [mscorlib]System.Console::ReadKey()
    IL_0049: pop
    // (no C# code)
    IL_004a: ret
} // end of method Program::Main

IL Code2:

.method private hidebysig static 
    void Main (
        string[] args
    ) cil managed 
{
    // Method begins at RVA 0x2050
    // Code size 56 (0x38)
    .maxstack 3
    .entrypoint
    .locals init (
        [0] class [mscorlib]System.Collections.Generic.IEnumerable`1<string>,
        [1] int32,
        [2] bool
    )

    // (no C# code)
    IL_0000: nop
    // IEnumerable<string> temp = args.AsEnumerable();
    IL_0001: ldarg.0
    IL_0002: call class [mscorlib]System.Collections.Generic.IEnumerable`1<!!0> [System.Core]System.Linq.Enumerable::AsEnumerable<string>(class [mscorlib]System.Collections.Generic.IEnumerable`1<!!0>)
    IL_0007: stloc.0
    // for (int i = 0; i < 10000000; i++)
    IL_0008: ldc.i4.0
    IL_0009: stloc.1
    // (no C# code)
    IL_000a: br.s IL_0025
    // loop start (head: IL_0025)
        IL_000c: nop
        // temp = temp.Select(Foo);
        IL_000d: ldloc.0
        IL_000e: ldnull
        IL_000f: ldftn string ConsoleApp1.Program::Foo(string)
        IL_0015: newobj instance void class [mscorlib]System.Func`2<string, string>::.ctor(object, native int)
        IL_001a: call class [mscorlib]System.Collections.Generic.IEnumerable`1<!!1> [System.Core]System.Linq.Enumerable::Select<string, string>(class [mscorlib]System.Collections.Generic.IEnumerable`1<!!0>, class [mscorlib]System.Func`2<!!0, !!1>)
        IL_001f: stloc.0
        // (no C# code)
        IL_0020: nop
        // for (int i = 0; i < 10000000; i++)
        IL_0021: ldloc.1
        IL_0022: ldc.i4.1
        IL_0023: add
        IL_0024: stloc.1

        // for (int i = 0; i < 10000000; i++)
        IL_0025: ldloc.1
        IL_0026: ldc.i4 10000000
        IL_002b: clt
        IL_002d: stloc.2
        // (no C# code)
        IL_002e: ldloc.2
        IL_002f: brtrue.s IL_000c
    // end loop

    // Console.ReadKey();
    IL_0031: call valuetype [mscorlib]System.ConsoleKeyInfo [mscorlib]System.Console::ReadKey()
    IL_0036: pop
    // (no C# code)
    IL_0037: ret
} // end of method Program::Main

I have to admit I am not enough expert in IL code to actually fully understand the difference and that's why I am raising this thread.

As far as I have understood, the actual Select seems to generate more instructions when not done through a method group (Code1) BUT is using some pointer to native functions. Is it reusing the method through the pointer compared to the other case which is always generating a new delegate?

Also I have noticed that the method group IL (Code2) is generating 3 comments linked to the for loop compared to the IL code of Code1.

Any help in understanding the difference of allocation would be appreciated.


回答1:


Spending some more time understanding why ReSharper is recommending to use method group instead of lambdas and reading the articles quoted in the rule page description, I am now able to answer my own question.

For cases where the number of iterations is small enough, around 1M with the code snippet I provided (so probably the majority of cases), the difference in memory allocation is small enough so that the 2 implementations are equivalent. Besides, and as we can see in the 2 generated IL Codes the compilation is faster as there is less instructions to generate. Note this was clearly stated by ReSharper :

to achieve more compact syntax and prevent compile-time overhead caused by using lambdas.

Which explain ReSharper recommendation.

But if you know that the delegate is going to be heavily used then the lambda is a better choice.



来源:https://stackoverflow.com/questions/53208516/delegate-instance-allocation-with-method-group-compared-to

易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!