When calculating xxhash, net7.0 is slower than net6.0 #90090

Cricle · 2023-08-07T07:34:12Z

Description

    internal unsafe class Program
    {
        public static void Main(string[] args)
        {
            var buffer = Encoding.UTF8.GetBytes("aadwejkadjgb8c27tr874c3/./[}|P{OP&^&$%^^TGERfgea");
            var sw = Stopwatch.GetTimestamp();
            for (int i = 0; i < 1_000_000; i++)
            {
                XxHash64.Hash(buffer);
            }
            Console.WriteLine(new TimeSpan(Stopwatch.GetTimestamp() - sw));

        }
    }

The csproj

<Project Sdk="Microsoft.NET.Sdk">

	<PropertyGroup>
		<OutputType>Exe</OutputType>
		<TargetFrameworks>net6.0;net7.0</TargetFrameworks>
	</PropertyGroup>

	<ItemGroup>
	  <PackageReference Include="System.IO.Hashing" Version="7.0.0" />
	</ItemGroup>

</Project>

dotnet run -c Release -f net6.0

Result 00:00:00.1590156

dotnet run -c Release -f net7.0

Result 00:00:00.2138358

Configuration

OS Name: Windows
OS Version: 10.0.19045
OS Platform: Windows
RID: win10-x64

Regression?

I don't know.

Data

None.

ghost · 2023-08-07T07:34:21Z

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

Issue Details

Description

    internal unsafe class Program
    {
        public static void Main(string[] args)
        {
            var buffer = Encoding.UTF8.GetBytes("aadwejkadjgb8c27tr874c3/./[}|P{OP&^&$%^^TGERfgea");
            var sw = Stopwatch.GetTimestamp();
            for (int i = 0; i < 1_000_000; i++)
            {
                XxHash64.Hash(buffer);
            }
            Console.WriteLine(new TimeSpan(Stopwatch.GetTimestamp() - sw));

        }
    }

The csproj

<Project Sdk="Microsoft.NET.Sdk">

	<PropertyGroup>
		<OutputType>Exe</OutputType>
		<TargetFrameworks>net6.0;net7.0</TargetFrameworks>
	</PropertyGroup>

	<ItemGroup>
	  <PackageReference Include="System.IO.Hashing" Version="7.0.0" />
	</ItemGroup>

</Project>

dotnet run -c Release -f net6.0

Result 00:00:00.1590156

dotnet run -c Release -f net7.0

Result 00:00:00.2138358

Configuration

OS Name: Windows
OS Version: 10.0.19045
OS Platform: Windows
RID: win10-x64

Regression?

I don't know.

Data

None.

Author:	Cricle
Assignees:	-
Labels:	`tenet-performance`, `area-CodeGen-coreclr`, `untriaged`
Milestone:	-

KeterSCP · 2023-08-07T08:24:07Z

The difference between net6.0 and net7.0 is not that big, but net8.0 shows 2x+ regression:

Benchmark code

using System.IO.Hashing;
using System.Text;
using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Jobs;
using BenchmarkDotNet.Running;

BenchmarkRunner.Run<Benchs>();

[SimpleJob(RuntimeMoniker.Net60)]
[SimpleJob(RuntimeMoniker.Net70)]
[SimpleJob(RuntimeMoniker.Net80)]
public class Benchs
{
    private readonly byte[] buffer = Encoding.UTF8.GetBytes("aadwejkadjgb8c27tr874c3/./[}|P{OP&^&$%^^TGERfgea");

    [Benchmark]
    public byte[] BenchXxhash()
    {
        return XxHash64.Hash(buffer);
    }
}

BenchmarkDotNet v0.13.7, Windows 11 (10.0.22621.1992/22H2/2022Update/SunValley2)
AMD Ryzen 9 5950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK 8.0.100-preview.6.23330.14
  [Host]   : .NET 7.0.9 (7.0.923.32018), X64 RyuJIT AVX2
  .NET 6.0 : .NET 6.0.20 (6.0.2023.32017), X64 RyuJIT AVX2
  .NET 7.0 : .NET 7.0.9 (7.0.923.32018), X64 RyuJIT AVX2
  .NET 8.0 : .NET 8.0.0 (8.0.23.32907), X64 RyuJIT AVX2

Method	Job	Runtime	Mean	Error	StdDev
BenchXxhash	.NET 6.0	.NET 6.0	14.09 ns	0.060 ns	0.056 ns
BenchXxhash	.NET 7.0	.NET 7.0	15.54 ns	0.062 ns	0.055 ns
BenchXxhash	.NET 8.0	.NET 8.0	36.63 ns	0.090 ns	0.084 ns

EgorBo · 2023-08-07T10:49:06Z

I bet it's #87113 (comment)

EgorBo · 2023-08-07T13:26:15Z

Or it can be PGO actually... cc @AndyAyersMS

using System.Diagnostics;
using System.IO.Hashing;

internal class Program
{
    public static void Main()
    {
        byte[] buffer = "aadwejkadjgb8c27tr874c3/./[}|P{OP&^&$%^^TGERfgea"u8.ToArray();
        Stopwatch sw = Stopwatch.StartNew();
        while (true)
        {
            sw.Restart();
            for (int i = 0; i < 1_000_000; i++)
                XxHash64.Hash(buffer);
            sw.Stop();
            Console.WriteLine(sw.ElapsedMilliseconds);
        }
    }
}

smells like the inliner budget problem - i see BinaryPrimitives.Read* apis non-inlined on hot path

EgorBo · 2023-08-07T13:34:43Z

public static void Main()
{
    byte[] buffer = "aadwejkadjgb8c27tr874c3/./[}|P{OP&^&$%^^TGERfgea"u8.ToArray();
    for (int i = 0; i < 1_000_000; i++)
        XxHash64.Hash(buffer);
}

Inlines into Program:Main():
  [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:.ctor(ulong,int):this
    [INLINED: below ALWAYS_INLINE size] System.Runtime.CompilerServices.RuntimeHelpers:IsReferenceOrContainsReferences[ubyte]():bool
    [FAILED: unprofitable inline] System.ThrowHelper:ThrowInvalidTypeWithPointersNotSupported(System.Type)
  [FAILED: unprofitable inline] System.ReadOnlySpan`1[ubyte]:ToArray():ubyte[]:this
  [INLINED: profitable inline] System.IO.Hashing.XxHash64:Hash(ubyte[]):ubyte[]
    [FAILED: too many il bytes] System.ArgumentNullException:.ctor(System.String):this
    [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:.ctor(ubyte[]):this
    [INLINED: profitable inline] System.IO.Hashing.XxHash64:Hash(System.ReadOnlySpan`1[ubyte],long):ubyte[]
      [INLINED: profitable inline] System.IO.Hashing.XxHash64:HashToUInt64(System.ReadOnlySpan`1[ubyte],long):ulong
        [INLINED: profitable inline] System.IO.Hashing.XxHash64+State:.ctor(ulong):this
        [INLINED: profitable inline] System.IO.Hashing.XxHash64+State:ProcessStripe(System.ReadOnlySpan`1[ubyte]):this
          [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:Slice(int,int):System.ReadOnlySpan`1[ubyte]:this
            [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException()
            [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:.ctor(byref,int):this
              [INLINED: profitable inline] System.Diagnostics.Debug:Assert(bool)
                [INLINED: below ALWAYS_INLINE size] System.Diagnostics.Debug:Assert(bool,System.String,System.String)
          [INLINED: below ALWAYS_INLINE size] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,System.ReadOnlySpan`1[ubyte]):ulong
            [INLINED: aggressive inline attribute] System.Buffers.Binary.BinaryPrimitives:ReadUInt64LittleEndian(System.ReadOnlySpan`1[ubyte]):ulong
              [INLINED: aggressive inline attribute] System.Runtime.InteropServices.MemoryMarshal:Read[ulong](System.ReadOnlySpan`1[ubyte]):ulong
                [INLINED: below ALWAYS_INLINE size] System.Runtime.CompilerServices.RuntimeHelpers:IsReferenceOrContainsReferences[ulong]():bool
                [FAILED: unprofitable inline] System.ThrowHelper:ThrowInvalidTypeWithPointersNotSupported(System.Type)
                [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException(int)
                [INLINED: below ALWAYS_INLINE size] System.Runtime.InteropServices.MemoryMarshal:GetReference[ubyte](System.ReadOnlySpan`1[ubyte]):byref
            [INLINED: profitable inline] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,ulong):ulong
          [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:Slice(int):System.ReadOnlySpan`1[ubyte]:this
            [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException()
            [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:.ctor(byref,int):this
              [INLINED: profitable inline] System.Diagnostics.Debug:Assert(bool)
                [INLINED: below ALWAYS_INLINE size] System.Diagnostics.Debug:Assert(bool,System.String,System.String)
                  [FAILED: noinline per IL/cached result] System.Diagnostics.Debug:Fail(System.String,System.String)
          [INLINED: below ALWAYS_INLINE size] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,System.ReadOnlySpan`1[ubyte]):ulong
            [INLINED: aggressive inline attribute] System.Buffers.Binary.BinaryPrimitives:ReadUInt64LittleEndian(System.ReadOnlySpan`1[ubyte]):ulong
              [INLINED: aggressive inline attribute] System.Runtime.InteropServices.MemoryMarshal:Read[ulong](System.ReadOnlySpan`1[ubyte]):ulong
                [INLINED: below ALWAYS_INLINE size] System.Runtime.CompilerServices.RuntimeHelpers:IsReferenceOrContainsReferences[ulong]():bool
                [FAILED: unprofitable inline] System.ThrowHelper:ThrowInvalidTypeWithPointersNotSupported(System.Type)
                [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException(int)
                [INLINED: below ALWAYS_INLINE size] System.Runtime.InteropServices.MemoryMarshal:GetReference[ubyte](System.ReadOnlySpan`1[ubyte]):byref
            [INLINED: profitable inline] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,ulong):ulong
          [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:Slice(int):System.ReadOnlySpan`1[ubyte]:this
            [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException()
            [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:.ctor(byref,int):this
              [INLINED: profitable inline] System.Diagnostics.Debug:Assert(bool)
                [INLINED: below ALWAYS_INLINE size] System.Diagnostics.Debug:Assert(bool,System.String,System.String)
                  [FAILED: noinline per IL/cached result] System.Diagnostics.Debug:Fail(System.String,System.String)
          [INLINED: below ALWAYS_INLINE size] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,System.ReadOnlySpan`1[ubyte]):ulong
            [INLINED: aggressive inline attribute] System.Buffers.Binary.BinaryPrimitives:ReadUInt64LittleEndian(System.ReadOnlySpan`1[ubyte]):ulong
              [INLINED: aggressive inline attribute] System.Runtime.InteropServices.MemoryMarshal:Read[ulong](System.ReadOnlySpan`1[ubyte]):ulong
                [INLINED: below ALWAYS_INLINE size] System.Runtime.CompilerServices.RuntimeHelpers:IsReferenceOrContainsReferences[ulong]():bool
                [FAILED: unprofitable inline] System.ThrowHelper:ThrowInvalidTypeWithPointersNotSupported(System.Type)
                [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException(int)
                [INLINED: below ALWAYS_INLINE size] System.Runtime.InteropServices.MemoryMarshal:GetReference[ubyte](System.ReadOnlySpan`1[ubyte]):byref
            [INLINED: profitable inline] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,ulong):ulong
          [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:Slice(int):System.ReadOnlySpan`1[ubyte]:this
            [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException()
            [INLINED: aggressive inline attribute] System.ReadOnlySpan`1[ubyte]:.ctor(byref,int):this
              [INLINED: profitable inline] System.Diagnostics.Debug:Assert(bool)
                [INLINED: below ALWAYS_INLINE size] System.Diagnostics.Debug:Assert(bool,System.String,System.String)
                  [FAILED: noinline per IL/cached result] System.Diagnostics.Debug:Fail(System.String,System.String)
          [INLINED: below ALWAYS_INLINE size] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,System.ReadOnlySpan`1[ubyte]):ulong
            [INLINED: aggressive inline attribute] System.Buffers.Binary.BinaryPrimitives:ReadUInt64LittleEndian(System.ReadOnlySpan`1[ubyte]):ulong
              [INLINED: aggressive inline attribute] System.Runtime.InteropServices.MemoryMarshal:Read[ulong](System.ReadOnlySpan`1[ubyte]):ulong
                [INLINED: below ALWAYS_INLINE size] System.Runtime.CompilerServices.RuntimeHelpers:IsReferenceOrContainsReferences[ulong]():bool
                [FAILED: unprofitable inline] System.ThrowHelper:ThrowInvalidTypeWithPointersNotSupported(System.Type)
                [FAILED: does not return] System.ThrowHelper:ThrowArgumentOutOfRangeException(int)
                [INLINED: below ALWAYS_INLINE size] System.Runtime.InteropServices.MemoryMarshal:GetReference[ubyte](System.ReadOnlySpan`1[ubyte]):byref
            [INLINED: profitable inline] System.IO.Hashing.XxHash64+State:ApplyRound(ulong,ulong):ulong
        [FAILED: inline exceeds budget] System.ReadOnlySpan`1[ubyte]:Slice(int):System.ReadOnlySpan`1[ubyte]:this
        [FAILED: inline exceeds budget] System.IO.Hashing.XxHash64+State:Complete(long,System.ReadOnlySpan`1[ubyte]):ulong:this
      [INLINED: below ALWAYS_INLINE size] System.Span`1[ubyte]:op_Implicit(ubyte[]):System.Span`1[ubyte]
        [FAILED: inline exceeds budget] System.Span`1[ubyte]:.ctor(ubyte[]):this
      [FAILED: inline exceeds budget] System.Buffers.Binary.BinaryPrimitives:WriteUInt64BigEndian(System.Span`1[ubyte],ulong)

Note a few FAILED: inline exceeds budget

Cricle added the tenet-performance Performance related issue label Aug 7, 2023

dotnet-issue-labeler bot added the area-CodeGen-coreclr CLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI label Aug 7, 2023

ghost added the untriaged New issue has not been triaged by the area owner label Aug 7, 2023

EgorBo mentioned this issue Aug 7, 2023

Add support for optimized AVX2/SSE2 multiply for VectorXXX<ulong/long> #87113

Closed

EgorBo added this to the 8.0.0 milestone Aug 7, 2023

ghost removed the untriaged New issue has not been triaged by the area owner label Aug 7, 2023

EgorBo self-assigned this Aug 7, 2023

EgorBo mentioned this issue Aug 8, 2023

Mark XxHash64.Complete as noinline #90142

Merged

ghost added the in-pr There is an active PR which will close this issue when it is merged label Aug 8, 2023

EgorBo mentioned this issue Aug 5, 2023

Microbenchmarks where inliner runs out of budget with TieredPGO #85531

Open

EgorBo closed this as completed in #90142 Aug 8, 2023

ghost removed the in-pr There is an active PR which will close this issue when it is merged label Aug 8, 2023

ghost locked as resolved and limited conversation to collaborators Sep 7, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

When calculating xxhash, net7.0 is slower than net6.0 #90090

When calculating xxhash, net7.0 is slower than net6.0 #90090

Cricle commented Aug 7, 2023

ghost commented Aug 7, 2023

Description

Configuration

Regression?

Data

KeterSCP commented Aug 7, 2023

EgorBo commented Aug 7, 2023

EgorBo commented Aug 7, 2023 •

edited

Loading

EgorBo commented Aug 7, 2023

When calculating xxhash, net7.0 is slower than net6.0 #90090

When calculating xxhash, net7.0 is slower than net6.0 #90090

Comments

Cricle commented Aug 7, 2023

Description

Configuration

Regression?

Data

ghost commented Aug 7, 2023

Description

Configuration

Regression?

Data

KeterSCP commented Aug 7, 2023

EgorBo commented Aug 7, 2023

EgorBo commented Aug 7, 2023 • edited Loading

EgorBo commented Aug 7, 2023

EgorBo commented Aug 7, 2023 •

edited

Loading