C# 过滤数据的三种姿势：yield return、LINQ Where、ToList 全景性能对决

测试代码

[MemoryDiagnoser]
public class YieldBenchmark
{
    private readonly Consumer _consumer = new Consumer();

    [Params(100, 10000, 100000)]
    public int N;

    private int[] _intArray;
    private MyClass[] _classArray;
    private MyStruct[] _structArray;

    [GlobalSetup]
    public void Setup()
    {
        _intArray = Enumerable.Range(0, N).ToArray();
        _classArray = Enumerable.Range(0, N).Select(i => new MyClass { Value = i }).ToArray();
        _structArray = Enumerable.Range(0, N).Select(i => new MyStruct { Value = i }).ToArray();
    }


    // ==================== int ====================
    [Benchmark] public void Int_Yield() => YieldInt(_intArray).Consume(_consumer);
    [Benchmark] public void Int_IfYield() => IfYieldInt(_intArray).Consume(_consumer);
    [Benchmark] public List<int> Int_ToList() => YieldInt(_intArray).ToList();
    [Benchmark] public int[] Int_ToArray() => YieldInt(_intArray).ToArray();

    private static IEnumerable<int> YieldInt(int[] source)
    {
        return source.Where(x => x % 2 == 0);
    }

    private static IEnumerable<int> IfYieldInt(int[] source)
    {
        foreach (var item in source)
            if (item % 2 == 0)
                yield return item;
    }

    // ==================== class ====================
    [Benchmark] public void Class_Yield() => YieldClass(_classArray).Consume(_consumer);
    [Benchmark] public void Class_IfYield() => IfYieldClass(_classArray).Consume(_consumer);
    [Benchmark] public List<MyClass> Class_ToList() => YieldClass(_classArray).ToList();
    [Benchmark] public MyClass[] Class_ToArray() => YieldClass(_classArray).ToArray();

    private static IEnumerable<MyClass> YieldClass(MyClass[] source)
    {
        return source.Where(x => x.Value % 2 == 0);
    }

    private static IEnumerable<MyClass> IfYieldClass(MyClass[] source)
    {
        foreach (var item in source)
            if (item.Value % 2 == 0)
                yield return item;
    }

    // ==================== struct ====================
    [Benchmark] public void Struct_Yield() => YieldStruct(_structArray).Consume(_consumer);
    [Benchmark] public void Struct_IfYield() => IfYieldStruct(_structArray).Consume(_consumer);
    [Benchmark] public List<MyStruct> Struct_ToList() => YieldStruct(_structArray).ToList();
    [Benchmark] public MyStruct[] Struct_ToArray() => YieldStruct(_structArray).ToArray();

    private static IEnumerable<MyStruct> YieldStruct(MyStruct[] source)
    {
        return source.Where(x => x.Value % 2 == 0);
    }

    private static IEnumerable<MyStruct> IfYieldStruct(MyStruct[] source)
    {
        foreach (var item in source)
            if (item.Value % 2 == 0)
                yield return item;
    }
}

public class MyClass
{
    public int Value;
}

public struct MyStruct
{
    public int Value;
}

测试结果

总结

1. list通常比array更占内存
2. 使用LINQ Where，会生成迭代器对象迭代器对象（WhereArrayIterator），产生48B内存
3. 手写循环yield return返回，会生成一个状态机，产生56B/64B(引用类型占8B)内存
4. 大数据量用IEnumerable可以减少内存分配