.NET 9.0.6中FrozenDictionary迭代速度慢于普通Dictionary的原因排查求助

最近我了解到FrozenDictionary<TKey, TValue>有着专门优化的实现，目标是让读操作（包括查找和枚举）尽可能快，不像普通ImmutableDictionary那样只是简单包装Dictionary。刚好我手头有个计算密集型应用，里面用到的集合加载后就不会再变更，所以想实际测试下它的性能表现。

我用BenchmarkDotNet编写了如下测试代码：

using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Running;
using System.Collections.Frozen;
using System.Collections.Generic;
using System.Linq;

[MemoryDiagnoser]
public class DictionaryVsFrozenDictionaryBenchmarks
{
    private KeyValuePair<string, string>[] _source;
    private Dictionary<string, string> _dictionary;
    private FrozenDictionary<string, string> _frozen;
    private string _middleKey;
    private string _middleKeyPlusOne;

    [Params(10, 100, 10_000)]
    public int Size;

    [GlobalSetup]
    public void Setup()
    {
        _source = Enumerable.Range(0, Size)
             .Select(i => new KeyValuePair<string, string>($"Key{i}", $"Value{i}"))
             .ToArray();
        _middleKey = $"Key{Size / 2}";
        _middleKeyPlusOne = $"Key{Size / 2 + 1}";
        _dictionary = new Dictionary<string, string>(_source);
        _frozen = _dictionary.ToFrozenDictionary();
    }

    [Benchmark]
    public Dictionary<string, string> Dictionary_Create() => new Dictionary<string, string>(_source);

    [Benchmark]
    public FrozenDictionary<string, string> FrozenDictionary_Create()
    {
        return new Dictionary<string, string>(_source).ToFrozenDictionary();
    }

    [Benchmark]
    public string Dictionary_Read1() => _dictionary[_middleKey];

    [Benchmark]
    public string FrozenDictionary_Read1() => _frozen[_middleKey];

    [Benchmark]
    public string Dictionary_Read2() => _dictionary[_middleKeyPlusOne];

    [Benchmark]
    public string FrozenDictionary_Read2() => _frozen[_middleKeyPlusOne];

    [Benchmark]
    public int Dictionary_Iteration()
    {
        int count = 0;
        foreach (var kvp in _dictionary)
        {
            count += kvp.Value.Length;
        }
        return count;
    }

    [Benchmark]
    public int FrozenDictionary_Iteration()
    {
        int count = 0;
        foreach (var kvp in _frozen)
        {
            count += kvp.Value.Length;
        }
        return count;
    }
}

public class Program
{
    public static void Main(string[] args)
    {
        BenchmarkRunner.Run<DictionaryVsFrozenDictionaryBenchmarks>();
    }
}

测试环境如下：

• BenchmarkDotNet v0.14.0
• Windows 11 (10.0.26100.4351)
• AMD Ryzen 9 7900，1 CPU，24逻辑核心/12物理核心
• .NET SDK 9.0.301，运行时.NET 9.0.6，X64 RyuJIT AVX-512F+CD+BW+DQ+VL+VBMI

得到的基准测试结果如下：

从结果里能看到，读操作确实比普通Dictionary快，但迭代速度却明显落后——Size=10的时候甚至慢了50%左右。我有点搞不懂，是不是我的测试代码哪里写得有问题？还是这个情况和CPU型号有关？有没有大佬能帮忙分析下原因？

内容来源于stack exchange