Regressions in System.Buffers.Text.Tests.Utf8FormatterTests

[performanceautofiler]

Run Information

Architecture	x64
OS	ubuntu 18.04
Baseline	ce041c81b689499cc02b0a4d6914f53d7c44b5c7
Compare	09c81d8bef22ee8350b86911b20715a19ffc05ab
Diff	Diff

Regressions in System.Buffers.Text.Tests.Utf8FormatterTests

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
FormatterInt64 - Duration of single invocation	23.03 ns	29.21 ns	1.27	0.02	False
FormatterUInt32 - Duration of single invocation	9.91 ns	12.82 ns	1.29	0.01	False

Test Report

Repro

git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f net6.0 --filter 'System.Buffers.Text.Tests.Utf8FormatterTests*'

Related Issues

Regressions

Improvements

Payloads

Baseline
Compare

Histogram

Edge Detector Info

Collection Data

System.Buffers.Text.Tests.Utf8FormatterTests.FormatterInt64(value: 9223372036854775807)

Description of detection logic

IsRegressionChecked: Marked as regression because the three check build points were 0.05 greater than the baseline.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsRegressionBase: Marked as regression because the compare was 5% greater than the baseline, and the value was not too small.
IsRegressionWindowed: Marked as regression because 29.212626133931302 > 24.328699978006973.
IsChangePoint: Marked as a change because one of 9/3/2022 11:44:47 AM, 10/11/2022 2:16:14 AM, 11/4/2022 9:06:14 PM, 11/8/2022 2:29:58 AM falls between 10/30/2022 12:44:54 AM and 11/8/2022 2:29:58 AM.
IsRegressionStdDev: Marked as regression because -75.5984167211087 (T) = (0 -29.211252705184588) / Math.Sqrt((0.09798937478021806 / (41)) + (0.032984641590120704 / (9))) is less than -2.010634757623041 = MathNet.Numerics.Distributions.StudentT.InvCDF(0, 1, (41) + (9) - 2, .025) and -0.25216090685751974 = (23.32867329207273 - 29.211252705184588) / 23.32867329207273 is less than -0.05.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.

```#### System.Buffers.Text.Tests.Utf8FormatterTests.FormatterUInt32(value: 4294967295)

```log

Description of detection logic

IsRegressionChecked: Marked as regression because the three check build points were 0.05 greater than the baseline.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsRegressionBase: Marked as regression because the compare was 5% greater than the baseline, and the value was not too small.
IsRegressionWindowed: Marked as regression because 12.823637156237847 > 10.427979517056931.
IsChangePoint: Marked as a change because one of 9/3/2022 11:44:47 AM, 10/11/2022 2:16:14 AM, 11/4/2022 9:06:14 PM, 11/8/2022 2:29:58 AM falls between 10/30/2022 12:44:54 AM and 11/8/2022 2:29:58 AM.
IsRegressionStdDev: Marked as regression because -76.51693901950912 (T) = (0 -13.032252353193137) / Math.Sqrt((0.0016959854340174783 / (41)) + (0.014481213515299272 / (9))) is less than -2.010634757623041 = MathNet.Numerics.Distributions.StudentT.InvCDF(0, 1, (41) + (9) - 2, .025) and -0.3132382797505535 = (9.92375302650223 - 13.032252353193137) / 9.92375302650223 is less than -0.05.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[EgorBo]

Likely regressed from #77137 (e.g. changed alignment?) cc @TIHan

[JulieLeeMSFT]

Ping @TIHan .

[TIHan]

It looks like the optimization I did with x << 3 and x << 2 may have caused it. Specifically looking at the diffs of the FormatterInt64 test:

Now, I manually tested, on my machine, FormatterInt64 with the correct input value, 9223372036854775807 showing before and after regarding #77137. It either showed an improvement or no difference; so it may be machine specific? Maybe newer hardware it is an improvement? The results below:

Benchmark 1

    public class Utf8FormatterTests
    {
        private readonly byte[] _destination = new byte[1000];

        // This is using TryFormat(long value...).
        [Benchmark]
        public bool FormatterInt64() => Utf8Formatter.TryFormat(9223372036854775807, _destination, out _);
    }

Before #77137

BenchmarkDotNet=v0.13.3, OS=Windows 11 (10.0.22621.963)
AMD Ryzen 9 7950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=7.0.101
  [Host]     : .NET 7.0.1 (7.0.122.56804), X64 RyuJIT AVX2
  Job-TYOOUA : .NET 8.0.0 (42.42.42.42424), X64 RyuJIT AVX2

Toolchain=CoreRun

|         Method |     Mean |    Error |   StdDev |
|--------------- |---------:|---------:|---------:|
| FormatterInt64 | 19.25 ns | 0.033 ns | 0.030 ns |

After #77137

BenchmarkDotNet=v0.13.3, OS=Windows 11 (10.0.22621.963)
AMD Ryzen 9 7950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=7.0.101
  [Host]     : .NET 7.0.1 (7.0.122.56804), X64 RyuJIT AVX2
  Job-OYHQOX : .NET 8.0.0 (42.42.42.42424), X64 RyuJIT AVX2

Toolchain=CoreRun

|         Method |     Mean |    Error |   StdDev |
|--------------- |---------:|---------:|---------:|
| FormatterInt64 | 18.35 ns | 0.092 ns | 0.086 ns |

The example above shows that #77137 is improvement, at least on my machine.

Benchmark 2

    public class Utf8FormatterTests
    {
        private readonly byte[] _destination = new byte[1000];

        [MethodImpl(MethodImplOptions.NoInlining)]
        public bool FormatterInt64(long value) => Utf8Formatter.TryFormat(value, _destination, out _);

        [Benchmark]
        public bool FormatterInt64() => FormatterInt64(9223372036854775807);
    }

Before #77137

BenchmarkDotNet=v0.13.3, OS=Windows 11 (10.0.22621.963)
AMD Ryzen 9 7950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=7.0.101
  [Host]     : .NET 7.0.1 (7.0.122.56804), X64 RyuJIT AVX2
  Job-VVQGGN : .NET 8.0.0 (42.42.42.42424), X64 RyuJIT AVX2

Toolchain=CoreRun

|         Method |     Mean |    Error |   StdDev |
|--------------- |---------:|---------:|---------:|
| FormatterInt64 | 19.61 ns | 0.080 ns | 0.075 ns |

After #77137

BenchmarkDotNet=v0.13.3, OS=Windows 11 (10.0.22621.963)
AMD Ryzen 9 7950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=7.0.101
  [Host]     : .NET 7.0.1 (7.0.122.56804), X64 RyuJIT AVX2
  Job-GQAVBD : .NET 8.0.0 (42.42.42.42424), X64 RyuJIT AVX2

Toolchain=CoreRun

|         Method |     Mean |    Error |   StdDev |
|--------------- |---------:|---------:|---------:|
| FormatterInt64 | 19.62 ns | 0.047 ns | 0.039 ns |

The example above shows that there is basically no difference.

---

@dotnet/jit-contrib how should I move forward with this? Maybe test more hardware?

[TIHan]

cc @EgorBo

[EgorBo]

cc @EgorBo

I would not bother since there is no clear suspect, on the global scale it looks like a bimodal behavior so I'd close this