[Perf] Windows/x64: 4 Improvements on 8/7/2024 8:05:53 AM

[performanceautofiler]

Run Information

Name	Value
Architecture	x64
OS	Mac OS X 10.18
Queue	IPhone
Baseline	dee8a8be71d755a1d27537ea8b3c59ee2a3d49c5
Compare	47ebcf3831a4a9b43d9abb19c21e4d66752e1d7c
Diff	Diff
Configs	CompilationMode:tiered, HybridGlobalization:true, iOSStripSymbols:false, RunKind:ios_scenarios, RuntimeType:nativeaot

Improvements in SOD - iOS HelloWorld NativeAOT .app Size nollvm symbols

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
app Size nollvm symbols - pub ADX - Test Multi Config Graph	4.32 MB	4.31 MB	1.00	0.00	True
app Size nollvm symbols - pub/iosHelloWorld/HelloiOS.app/HelloiOS ADX - Test Multi Config Graph	4.32 MB	4.31 MB	1.00	0.00	True
app Size nollvm symbols - Aggregate - No Extension ADX - Test Multi Config Graph	4.32 MB	4.31 MB	1.00	0.00	True
app Size nollvm symbols - SOD - iOS HelloWorld NativeAOT .app Size nollvm symbols ADX - Test Multi Config Graph	4.32 MB	4.31 MB	1.00	0.00	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 'SOD - iOS HelloWorld NativeAOT .app Size nollvm symbols*'

SOD - iOS HelloWorld NativeAOT .app Size nollvm symbols

ETL Files

Histogram

JIT Disasms

Docs

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

[matouskozak]

Small improvement, likely caused by dotnet/runtime#105945 @MichalStrehovsky !

[MichalStrehovsky]

Small improvement, likely caused by dotnet/runtime#105945 @MichalStrehovsky !

Nice. My guess would be dotnet/runtime#104148, but 45 also probably contributed.

[matouskozak]

I tried extracting the shortest range and only 45 pop-out to me as likely caused dotnet/runtime@af5a23c...c94cacb. In the range containing 48, no changes in measurements were recorded. Did it show improvements in local measurements? If so, perhaps something is off in the testing infra...

[MichalStrehovsky]

Hm, it might be 45 after all, even though 45 wasn't a size fix, just deleting some dead code that happened to save size. I ran measurements on both.

• 45: 105945 MichalStrehovsky/rt-sz#55
• 48: 104148 MichalStrehovsky/rt-sz#42