[Perf -13%] Microsoft.Extensions.Primitives.StringSegmentBenchmark.EndsWith

[DrewScoggins]

Run Information

Architecture	x64
OS	Windows 10.0.18362
Changes	diff

Regressions in Microsoft.Extensions.Primitives.StringSegmentBenchmark

Benchmark	Baseline	Test	Test/Base	Modality	Baseline Outlier
EndsWith	9.44 ns	10.78 ns	1.14	Bimodal	True

Historical Data in Reporting System

Repro

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f netcoreapp5.0 --filter 'Microsoft.Extensions.Primitives.StringSegmentBenchmark*'

Details

Histogram

Microsoft.Extensions.Primitives.StringSegmentBenchmark.EndsWith

[ 8.933 ;  9.178) | @@@@@@@@@@@@@@@@@@@@@@@@@@
[ 9.178 ;  9.457) | @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
[ 9.457 ;  9.762) | @
[ 9.762 ; 10.041) | @@@@@@@@@@@@@@@@@@@@@@@
[10.041 ; 10.332) | @
[10.332 ; 10.589) | @@
[10.589 ; 10.868) | @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@

Docs

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

[Dotnet-GitSync-Bot]

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.

[gfoidl]

Just to point out that there is #39140 (correctness-fix is #39853), so maybe it's more interesting to validate the numbers after this change is in.

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

[eerhardt]

Marking as 5.0 to investigate the perf regression here.

[ericstj]

This method hasn't changed since we moved the code. There was a change around the time of this regression in string.Compare which this method calls. @GrabYourPitchforks it looks like 189e1aa#diff-a70aebfa78baa09ba9bb366e971a14ca occurred around the same time. Is it possible that this caused a regression here?

[eerhardt]

I cannot reproduce this regression locally. I've tried from commit ca95479 to 4ec3a25 (which is basically July 25 to Aug 9).

Here are the results I am seeing locally:

Before (commit ca95479)

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Gen 1	Gen 2	Allocated
EndsWith	10.36 ns	0.085 ns	0.075 ns	10.36 ns	10.21 ns	10.50 ns	-	-	-	-
EndsWith	10.31 ns	0.196 ns	0.183 ns	10.33 ns	9.989 ns	10.58 ns	-	-	-	-
EndsWith	10.39 ns	0.104 ns	0.081 ns	10.42 ns	10.21 ns	10.52 ns	-	-	-	-

After (commit 4ec3a25)

Method	Mean	Error	StdDev	Median	Min	Max	Gen 0	Gen 1	Gen 2	Allocated
EndsWith	10.27 ns	0.117 ns	0.110 ns	10.23 ns	10.11 ns	10.47 ns	-	-	-	-
EndsWith	10.31 ns	0.107 ns	0.090 ns	10.29 ns	10.19 ns	10.52 ns	-	-	-	-

Looking at this benchmark, the version of Microsoft.Extensions.Primitives did not change in this time frame. The benchmark was always hard-coded to use 5.0.0-preview.4.20251.6. So if something regressed here, it had to be in the 1 method EndsWith calls - string.Compare(string, int, string, int, int, StringComparison). Note that the benchmark is only using StringComparison.Ordinal.

Looking at the history of string.Compare, @GrabYourPitchforks did change some OrdinalIgnoreCase code around this time, but that change didn't affect the Ordinal code path, so his change should not have affected this.

Given the above information, I'm going to close this issue as "not repro". If we can get a reproducible regression, please re-open and I can take a look.