[Perf] Windows/x64: 4 Regressions on 4/11/2024 11:29:08 PM

[performanceautofiler]

Run Information

Name	Value
Architecture	x64
OS	Windows 10.0.22621
Queue	TigerWindows
Baseline	a8158c170b694f8c1dbae114c63c346b38244901
Compare	324bd561b9e9d01ded8d2cb086cf55d23fc37eab
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Collections.AddGivenSize<String>

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ICollection - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	2.85 μs	3.16 μs	1.11	0.11	False
List - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	2.03 μs	2.18 μs	1.08	0.12	False

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.AddGivenSize<String>*'

System.Collections.AddGivenSize<String>.ICollection(Size: 512)

ETL Files

Histogram

JIT Disasms

System.Collections.AddGivenSize<String>.List(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 10.0.22621
Queue	TigerWindows
Baseline	a8158c170b694f8c1dbae114c63c346b38244901
Compare	324bd561b9e9d01ded8d2cb086cf55d23fc37eab
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Reflection.Invoke

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
StaticMethod4_ByRefParams_int_string_struct_class - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	217.96 ns	240.77 ns	1.10	0.06	False
StaticMethod5_ByRefParams_int_string_struct_class_bool - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	303.04 ns	350.44 ns	1.16	0.05	True

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.Reflection.Invoke*'

System.Reflection.Invoke.StaticMethod4_ByRefParams_int_string_struct_class

ETL Files

Histogram

JIT Disasms

System.Reflection.Invoke.StaticMethod5_ByRefParams_int_string_struct_class_bool

ETL Files

Histogram

JIT Disasms

Docs

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

[dotnet-policy-service]

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

[jkotas]

We are leaving the performance on the table in JIT_Box helper. It should be possible to structure this helper to have fast and slow paths, similar to how other allocation helpers are structured.

JIT_Box is inlined by the JIT for situations that matter, so the perf did not matter much. We are calling it directly now that is changing the equation.

[jkoritzinsky]

I'm still investigating, but I think one of the culprits (based on the related worse Mono regression in the exact same two benchmarks) is the introduction of the type.IsNullableOfT check in AllocateValueType from the MethodBaseInvoker.TryByRefFastPath path (for which that will never be true).

I'm looking now to see if inlining that branch into the callers where it fixes the regression.

[jkoritzinsky]

Fixed by #101137