Improve performance of `Enumerable.ToDictionary()` #96574

xin9le · 2024-01-06T09:43:01Z

Summary

In this Pull-Request, I've improved the performance of the Enumerable.ToDictionary() method in the following two points.

Set the initial capacity when the source implements ICollection or IIListProvider<T>
- The implementation patterns where the initial capacity is set are increasing
Combine the specializations for T[] and List<T> into a specialization for ReadOnlySpan<T>
- Reduction of code volume
- Particularly, speed up of enumeration for List<T>

Benchmark

BenchmarkDotNet v0.13.12, Windows 11 (10.0.22621.2861/22H2/2022Update/SunValley2)
13th Gen Intel Core i7-1360P, 1 CPU, 16 logical and 12 physical cores
.NET SDK 8.0.100
  [Host]   : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2
  ShortRun : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1
WarmupCount=3

| Method                    | Mean     | Error     | StdDev   | Ratio | RatioSD | Gen0     | Gen1     | Gen2     | Allocated | Alloc Ratio |
|-------------------------- |---------:|----------:|---------:|------:|--------:|---------:|---------:|---------:|----------:|------------:|
| EnumerableToDictionary    | 236.9 us |  29.52 us |  1.62 us |  1.00 |    0.00 | 153.8086 | 153.8086 | 153.8086 | 788.75 KB |        1.00 |
| ArrayToDictionary         | 110.9 us |  16.13 us |  0.88 us |  0.47 |    0.01 |  62.3779 |  62.3779 |  62.3779 | 236.94 KB |        0.30 |
| ListToDictionary          | 136.4 us | 660.76 us | 36.22 us |  0.58 |    0.15 |  62.3779 |  62.3779 |  62.3779 | 236.94 KB |        0.30 |
| PR_EnumerableToDictionary | 170.8 us |  19.04 us |  1.04 us |  0.72 |    0.01 |  62.2559 |  62.2559 |  62.2559 | 236.99 KB |        0.30 |
| PR_ArrayToDictionary      | 114.8 us |  20.21 us |  1.11 us |  0.48 |    0.00 |  62.3779 |  62.3779 |  62.3779 | 236.94 KB |        0.30 |
| PR_ListToDictionary       | 115.1 us |  20.52 us |  1.12 us |  0.49 |    0.01 |  62.3779 |  62.3779 |  62.3779 | 236.94 KB |        0.30 |

public class ToDictionaryBenchmark
{
    private const int Length = 10000;

    private static readonly IEnumerable<Kvs> s_enumerable = Enumerable.Range(0, Length).Select(x => new Kvs(x, x));
    private static readonly IEnumerable<Kvs> s_array = s_enumerable.ToArray();
    private static readonly IEnumerable<Kvs> s_list = s_enumerable.ToList();

    [Benchmark(Baseline = true)]
    public Dictionary<int, Kvs> EnumerableToDictionary()
        => s_enumerable.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> ArrayToDictionary()
        => s_array.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> ListToDictionary()
        => s_list.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_EnumerableToDictionary()
        => s_enumerable.ToDictionary_PR(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_ArrayToDictionary()
        => s_array.ToDictionary_PR(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_ListToDictionary()
        => s_list.ToDictionary_PR(static x => x.Key);
}

public record struct Kvs(int Key, int Value);

ghost · 2024-01-06T09:43:14Z

Tagging subscribers to this area: @dotnet/area-system-linq
See info in area-owners.md if you want to be subscribed.

Issue Details

Summary

In this Pull-Request, I've improved the performance of the Enumerable.ToDictionary() method in the following two points.

Set the initial capacity when the source implements ICollection or IIListProvider<T>
- The implementation patterns where the initial capacity is set are increasing
Combine the specializations for T[] and List<T> into a specialization for ReadOnlySpan<T>
- Reduction of code volume
- Particularly, speed up of enumeration for List<T>

Benchmark

BenchmarkDotNet v0.13.12, Windows 11 (10.0.22621.2861/22H2/2022Update/SunValley2)
13th Gen Intel Core i7-1360P, 1 CPU, 16 logical and 12 physical cores
.NET SDK 8.0.100
  [Host]   : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2
  ShortRun : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1
WarmupCount=3

Method	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
EnumerableToDictionary	236.9 us	29.52 us	1.62 us	1.00	0.00	153.8086	153.8086	153.8086	788.75 KB	1.00
ArrayToDictionary	110.9 us	16.13 us	0.88 us	0.47	0.01	62.3779	62.3779	62.3779	236.94 KB	0.30
ListToDictionary	136.4 us	660.76 us	36.22 us	0.58	0.15	62.3779	62.3779	62.3779	236.94 KB	0.30
PR_EnumerableToDictionary	170.8 us	19.04 us	1.04 us	0.72	0.01	62.2559	62.2559	62.2559	236.99 KB	0.30
PR_ArrayToDictionary	114.8 us	20.21 us	1.11 us	0.48	0.00	62.3779	62.3779	62.3779	236.94 KB	0.30
PR_ListToDictionary	115.1 us	20.52 us	1.12 us	0.49	0.01	62.3779	62.3779	62.3779	236.94 KB	0.30

public class ToDictionaryBenchmark
{
    private const int Length = 10000;

    private static readonly IEnumerable<Kvs> s_enumerable = Enumerable.Range(0, Length).Select(x => new Kvs(x, x));
    private static readonly IEnumerable<Kvs> s_array = s_enumerable.ToArray();
    private static readonly IEnumerable<Kvs> s_list = s_enumerable.ToList();

    [Benchmark(Baseline = true)]
    public Dictionary<int, Kvs> EnumerableToDictionary()
        => s_enumerable.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> ArrayToDictionary()
        => s_array.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> ListToDictionary()
        => s_list.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_EnumerableToDictionary()
        => s_enumerable.ToDictionary_PR(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_ArrayToDictionary()
        => s_array.ToDictionary_PR(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_ListToDictionary()
        => s_list.ToDictionary_PR(static x => x.Key);
}

public record struct Kvs(int Key, int Value);

Author:	xin9le
Assignees:	-
Labels:	`area-System.Linq`, `community-contribution`
Milestone:	-

xin9le · 2024-01-06T09:46:00Z

@dotnet-policy-service agree

stephentoub · 2024-01-06T15:45:22Z

Thanks. What does the comparison look like, in particular for an iterator that conveys no size information, with a small length, like 3?

stephentoub

LGTM. Thanks.

xin9le · 2024-01-08T05:44:48Z

@stephentoub
Thank you for your approval. As you already understand, I believe there may be slight overhead in the following cases:

When size information cannot be obtained
When the internal buffer of the collection type is not expanded

※ This is due to the increased type determination in .TryGetNonEnumeratedCount()

Whether or not this overhead is considered negligible compared to the benefits gained from setting the initial capacity, I believe, is the determining factor in whether or not to incorporate this Pull-Request. Although you have already approved, I will document the measurement results just in case.

Benchmark

BenchmarkDotNet v0.13.12, Windows 11 (10.0.22621.2861/22H2/2022Update/SunValley2)
13th Gen Intel Core i7-1360P, 1 CPU, 16 logical and 12 physical cores
.NET SDK 8.0.100
  [Host]     : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2
  DefaultJob : .NET 8.0.0 (8.0.23.53103), X64 RyuJIT AVX2

| Method                    | Mean     | Error    | StdDev   | Ratio | RatioSD | Gen0   | Allocated | Alloc Ratio |
|-------------------------- |---------:|---------:|---------:|------:|--------:|-------:|----------:|------------:|
| EnumerableToDictionary    | 39.45 ns | 0.817 ns | 1.431 ns |  1.00 |    0.00 | 0.0263 |     248 B |        1.00 |
| ArrayToDictionary         | 34.26 ns | 0.698 ns | 0.776 ns |  0.87 |    0.05 | 0.0221 |     208 B |        0.84 |
| ListToDictionary          | 34.20 ns | 0.570 ns | 0.701 ns |  0.87 |    0.04 | 0.0221 |     208 B |        0.84 |
| PR_EnumerableToDictionary | 41.02 ns | 0.814 ns | 0.762 ns |  1.05 |    0.07 | 0.0263 |     248 B |        1.00 |
| PR_ArrayToDictionary      | 34.18 ns | 0.695 ns | 0.543 ns |  0.88 |    0.05 | 0.0221 |     208 B |        0.84 |
| PR_ListToDictionary       | 32.78 ns | 0.683 ns | 0.670 ns |  0.84 |    0.05 | 0.0221 |     208 B |        0.84 |

public class ToDictionaryBenchmark
{
    private static readonly IEnumerable<Kvs> s_enumerable = EnumerateSample();
    private static readonly IEnumerable<Kvs> s_array = s_enumerable.ToArray();
    private static readonly IEnumerable<Kvs> s_list = s_enumerable.ToList();

    private static IEnumerable<Kvs> EnumerateSample()
    {
        for (var i = 0; i < 3; i++)
            yield return new Kvs(i, i);
    }

    [Benchmark(Baseline = true)]
    public Dictionary<int, Kvs> EnumerableToDictionary()
        => s_enumerable.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> ArrayToDictionary()
        => s_array.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> ListToDictionary()
        => s_list.ToDictionary(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_EnumerableToDictionary()
        => s_enumerable.ToDictionary2(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_ArrayToDictionary()
        => s_array.ToDictionary2(static x => x.Key);

    [Benchmark]
    public Dictionary<int, Kvs> PR_ListToDictionary()
        => s_list.ToDictionary2(static x => x.Key);
}

public record struct Kvs(int Key, int Value);

stephentoub · 2024-01-08T14:36:50Z

Thanks for the numbers. A ns on an extreme case is reasonable.

improve Enumerable.ToDictionary performance

8683446

ghost added the community-contribution Indicates that the PR has been added by a community member label Jan 6, 2024

dotnet-issue-labeler bot added the area-System.Linq label Jan 6, 2024

stephentoub approved these changes Jan 8, 2024

View reviewed changes

eiriktsarpalis requested a review from stephentoub January 8, 2024 13:17

stephentoub merged commit 1063fc2 into dotnet:main Jan 8, 2024

xin9le deleted the improve-enumerable-todictionary branch January 12, 2024 15:54

github-actions bot locked and limited conversation to collaborators Feb 12, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Improve performance of `Enumerable.ToDictionary()` #96574

Improve performance of `Enumerable.ToDictionary()` #96574

xin9le commented Jan 6, 2024 •

edited

Loading

ghost commented Jan 6, 2024

Summary

Benchmark

xin9le commented Jan 6, 2024

stephentoub commented Jan 6, 2024 •

edited

Loading

stephentoub left a comment

xin9le commented Jan 8, 2024

stephentoub commented Jan 8, 2024

Improve performance of Enumerable.ToDictionary() #96574

Improve performance of Enumerable.ToDictionary() #96574

Conversation

xin9le commented Jan 6, 2024 • edited Loading

Summary

Benchmark

ghost commented Jan 6, 2024

Summary

Benchmark

xin9le commented Jan 6, 2024

stephentoub commented Jan 6, 2024 • edited Loading

stephentoub left a comment

Choose a reason for hiding this comment

xin9le commented Jan 8, 2024

Benchmark

stephentoub commented Jan 8, 2024

Improve performance of `Enumerable.ToDictionary()` #96574

Improve performance of `Enumerable.ToDictionary()` #96574

xin9le commented Jan 6, 2024 •

edited

Loading

stephentoub commented Jan 6, 2024 •

edited

Loading