Skip to content

Performance changes related to vzeroupper emit changes #98705

New issue
New issue
Closed
Closed
Performance changes related to vzeroupper emit changes#98705
Labels
arch-x64area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMIos-windowsruntime-coreclrspecific to the CoreCLR runtime

Description

@performanceautofiler
performanceautofiler
bot
opened on Feb 20, 2024

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in System.Tests.Perf_Char

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
17.21 ns 18.88 ns 1.10 0.03 False

graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Tests.Perf_Char*'
Details

Payloads

Baseline
Compare

System.Tests.Perf_Char.Char_ToUpper(c: 'İ', cultureName: en-US)

ETL Files

Histogram

JIT Disasms

Docs

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

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in System.Text.Perf_Ascii

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
5.97 ns 7.36 ns 1.23 0.03 False
5.87 ns 7.15 ns 1.22 0.03 False

graph
graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Text.Perf_Ascii*'
Details

Payloads

Baseline
Compare

System.Text.Perf_Ascii.EqualsIgnoreCase_ExactlyTheSame_Chars(Size: 6)

ETL Files

Histogram

JIT Disasms

System.Text.Perf_Ascii.EqualsIgnoreCase_ExactlyTheSame_Bytes(Size: 6)

ETL Files

Histogram

JIT Disasms

Docs

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

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in System.Collections.TryGetValueFalse<Int32, Int32>

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
20.47 μs 23.18 μs 1.13 0.09 False
2.80 μs 3.77 μs 1.34 0.10 True

graph
graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Collections.TryGetValueFalse&lt;Int32, Int32&gt;*'
Details

Payloads

Baseline
Compare

System.Collections.TryGetValueFalse<Int32, Int32>.ImmutableDictionary(Size: 512)

ETL Files

Histogram

JIT Disasms

System.Collections.TryGetValueFalse<Int32, Int32>.Dictionary(Size: 512)

ETL Files

Histogram

JIT Disasms

Docs

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

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in System.Collections.Sort<BigStruct>

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
7.21 μs 11.43 μs 1.58 0.61 False
7.51 μs 11.76 μs 1.56 0.65 False

graph
graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Collections.Sort&lt;BigStruct&gt;*'
Details

Payloads

Baseline
Compare

System.Collections.Sort<BigStruct>.List(Size: 512)

ETL Files

Histogram

JIT Disasms

System.Collections.Sort<BigStruct>.Array(Size: 512)

ETL Files

Histogram

JIT Disasms

Docs

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

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in System.Buffers.Tests.SearchValuesByteTests

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
9.34 ns 11.14 ns 1.19 0.06 False

graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Buffers.Tests.SearchValuesByteTests*'
Details

Payloads

Baseline
Compare

System.Buffers.Tests.SearchValuesByteTests.IndexOfAnyExcept(Values: "abcdefABCDEF0123456789")

ETL Files

Histogram

JIT Disasms

Docs

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

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in PerfLabTests.CastingPerf

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
311.83 μs 362.26 μs 1.16 0.00 True
311.83 μs 363.82 μs 1.17 0.00 True
364.28 μs 420.81 μs 1.16 0.03 False

graph
graph
graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'PerfLabTests.CastingPerf*'
Details

Payloads

Baseline
Compare

PerfLabTests.CastingPerf.ObjFooIsObj2

ETL Files

Histogram

JIT Disasms

PerfLabTests.CastingPerf.IFooFooIsIFoo

ETL Files

Histogram

JIT Disasms

PerfLabTests.CastingPerf.FooObjCastIfIsa

ETL Files

Histogram

JIT Disasms

Docs

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

Run Information

Name Value
Architecture x64
OS Windows 10.0.18362
Queue TigerWindows
Baseline e3af00bf62f3280b4db46e2a01caf68a4dd5c991
Compare 49fe3b06cdfce737ea1c8964e01f2dd2a4e77d44
Diff Diff
Configs CompilationMode:tiered, RunKind:micro

Regressions in System.Numerics.Tests.Perf_VectorOf<Int16>

Benchmark Baseline Test Test/Base Test Quality Edge Detector Baseline IR Compare IR IR Ratio
39.54 ns 44.49 ns 1.13 0.01 True
39.66 ns 44.49 ns 1.12 0.02 True

graph
graph
Test Report

Repro

General Docs link: https://github.com/dotnet/performance/blob/main/docs/benchmarking-workflow-dotnet-runtime.md

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net8.0 --filter 'System.Numerics.Tests.Perf_VectorOf&lt;Int16&gt;*'
Details

Payloads

Baseline
Compare

System.Numerics.Tests.Perf_VectorOf<Int16>.DivisionOperatorBenchmark

ETL Files

Histogram

JIT Disasms

System.Numerics.Tests.Perf_VectorOf<Int16>.DivideBenchmark

ETL Files

Histogram

JIT Disasms

Docs

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

Activity

  1. performanceautofiler
    added
    arch-x64
    os-windows
    runtime-coreclrspecific to the CoreCLR runtime
    untriagedNew issue has not been triaged by the area owner
    on Feb 20, 2024

  2. DrewScoggins commented on Feb 20, 2024

    @DrewScoggins
    DrewScoggins
    on Feb 20, 2024
    Member

    #98261

  3. DrewScoggins
    transferred this issue fromdotnet/perf-autofiling-issueson Feb 20, 2024
  4. ghost added
    needs-area-labelAn area label is needed to ensure this gets routed to the appropriate area owners
    on Feb 20, 2024
  5. DrewScoggins
    changed the title [-][Perf] Windows/x64: 13 Regressions on 2/13/2024 8:56:49 PM[/-] [+]Performance changes related to vzeroupper emit changes[/+] on Feb 20, 2024

  6. DrewScoggins commented on Feb 20, 2024

    @DrewScoggins
    DrewScoggins
    on Feb 20, 2024
    Member

    cc @tannergooding

  7. DrewScoggins commented on Feb 20, 2024

    @DrewScoggins
    DrewScoggins
    on Feb 20, 2024
    Member

    Regressions:

    Windows AMD x64: dotnet/perf-autofiling-issues#29540, dotnet/perf-autofiling-issues#29544
    Linux x64: dotnet/perf-autofiling-issues#29441, dotnet/perf-autofiling-issues#29444, dotnet/perf-autofiling-issues#29445

    Improvements:
    Windows x64: dotnet/perf-autofiling-issues#29530
    Linux x64: dotnet/perf-autofiling-issues#29455

  8. danmoseley
    added
    area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMI
    and removed
    needs-area-labelAn area label is needed to ensure this gets routed to the appropriate area owners
    on Feb 20, 2024

  10. tannergooding commented on Feb 21, 2024

    @tannergooding
    tannergooding
    on Feb 21, 2024
    Member

    This is effectively "by design" and we are now simply following the Architecture Optimization Manual guidance on where to emit vzeroupper.

    Previously we were failing to emit it in some places where it was required, this could lead to a significant performance hit on some hardware (which multiple customers reported hitting). However, if we aren't in a scenario where the native code actually needs vzeroupper, then some hardware will itself see a perf hit from the additional instructions needing to execute (newer hardware is typically not impacted).

    Correspondingly, we were emitting it in too many unnecessary places as well. Removing these unnecessary places can result in a perf increase on older hardware.

    It's ultimately a set of tradeoffs, but since we're now matching the official guidance on where to emit vzeroupper, we are in a better overall place.

  11. tannergooding
    closed this as completedon Feb 21, 2024
  12. ghost removed
    untriagedNew issue has not been triaged by the area owner
    on Feb 21, 2024
  13. github-actions
    locked and limited conversation to collaborators on Mar 23, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    arch-x64area-CodeGen-coreclrCLR JIT compiler in src/coreclr/src/jit and related components such as SuperPMIos-windowsruntime-coreclrspecific to the CoreCLR runtime

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions