[Perf] Linux/arm64: 9 Regressions on 6/24/2026 7:31:32 AM +00:00

[performanceautofiler]

Run Information

Name	Value
Architecture	arm64
OS	azurelinux 3.0
Queue	CobaltAzureLinux
Baseline	0d1a20fb14109f277df06ebee3f83c964f9dcc61
Compare	e3cbc00cd3e0b511ad36dc240010eb2b9eefc3b7
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Text.Json.Tests.Perf_Reader

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ReadSpanEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	2.46 μs	3.40 μs	1.38	0.29	False
ReadReturnBytes - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	103.53 ns	135.76 ns	1.31	0.24	False
ReadSpanEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	95.06 ns	122.44 ns	1.29	0.28	False
ReadSingleSpanSequenceEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	932.79 ns	1121.09 ns	1.20	0.21	False
ReadSpanEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	917.66 ns	1086.47 ns	1.18	0.19	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.Text.Json.Tests.Perf_Reader*'

Details

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: LotsOfNumbers)

ETL Files

Histogram

JIT Disasms

System.Text.Json.Tests.Perf_Reader.ReadReturnBytes(IsDataCompact: False, TestCase: HelloWorld)

ETL Files

Histogram

JIT Disasms

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: HelloWorld)

ETL Files

Histogram

JIT Disasms

System.Text.Json.Tests.Perf_Reader.ReadSingleSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: Json400B)

ETL Files

Histogram

JIT Disasms

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: Json400B)

ETL Files

Histogram

JIT Disasms

Docs

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

---

Run Information

Name	Value
Architecture	arm64
OS	azurelinux 3.0
Queue	CobaltAzureLinux
Baseline	0d1a20fb14109f277df06ebee3f83c964f9dcc61
Compare	e3cbc00cd3e0b511ad36dc240010eb2b9eefc3b7
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Globalization.Tests.StringSearch

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
LastIndexOf_Word_NotFound - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	430.74 ns	691.82 ns	1.61	0.19	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.Globalization.Tests.StringSearch*'

Details

System.Globalization.Tests.StringSearch.LastIndexOf_Word_NotFound(Options: (, IgnoreCase, False))

ETL Files

Histogram

JIT Disasms

Docs

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

---

Run Information

Name	Value
Architecture	arm64
OS	azurelinux 3.0
Queue	CobaltAzureLinux
Baseline	0d1a20fb14109f277df06ebee3f83c964f9dcc61
Compare	e3cbc00cd3e0b511ad36dc240010eb2b9eefc3b7
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.IO.Tests.Perf_RandomAccess

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ReadScatter - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	7.41 ms	16.89 ms	2.28	0.69	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.IO.Tests.Perf_RandomAccess*'

Details

System.IO.Tests.Perf_RandomAccess.ReadScatter(fileSize: 104857600, buffersSize: 65536, options: Asynchronous)

ETL Files

Histogram

JIT Disasms

Docs

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

---

Run Information

Name	Value
Architecture	arm64
OS	azurelinux 3.0
Queue	CobaltAzureLinux
Baseline	0d1a20fb14109f277df06ebee3f83c964f9dcc61
Compare	e3cbc00cd3e0b511ad36dc240010eb2b9eefc3b7
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Collections.CreateAddAndClear<Int32>

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ImmutableDictionary - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	111.54 μs	171.57 μs	1.54	0.11	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.CreateAddAndClear<Int32>*'

Details

System.Collections.CreateAddAndClear<Int32>.ImmutableDictionary(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	arm64
OS	azurelinux 3.0
Queue	CobaltAzureLinux
Baseline	0d1a20fb14109f277df06ebee3f83c964f9dcc61
Compare	e3cbc00cd3e0b511ad36dc240010eb2b9eefc3b7
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Memory.Span<Char>

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
LastIndexOfValue - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	20.12 ns	32.33 ns	1.61	0.14	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.Memory.Span<Char>*'

Details

System.Memory.Span<Char>.LastIndexOfValue(Size: 512)

ETL Files

Histogram

JIT Disasms

Docs

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

[LoopedBard3]

🔍 Automated Triage Analysis

Summary: Verified regression in ReadSpanEmptyLoop(IsDataCompact: False, TestCase: LotsOfNumbers) — caused by commit 65b2508 ("Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"). 5 tests bisected to related culprit commits. ⚠️ Note: One or more tests show bimodal history, but at least one clean test bisected to a related culprit.

Finding Confidence: 5/5 (analysis accuracy)
Regression Confidence: 5/5 (likelihood of true regression)

Likely Cause: 65b2508c9e09b7ab0dad595a06d583df7688a210

🤖 Proposed Next Actions: Known-cause regression — already linked — 0 done, 2 dry-run (rule R3_known_regression_with_prior — see Full Analysis for the table). · 🦆 Re-analysis: ✅ concurs

📊 Full Analysis (click to expand)

Summary: 3 confirmed regressions (System.Text.Json Reader +28–31%, System.IO ReadScatter +128%) and 3 borderline cases attributed to commit 65b2508 ("Use SearchValues in Utf8JsonReader.SkipWhiteSpace"). Finding confidence 4/5 (strong historical data, 283–286 runs), regression confidence 4/5 (direct code path hit, step change pattern with z=2.3–3.2, focused diff targeting exact test area).

Issue Overview

• Issue: [Perf] Linux/arm64: 9 Regressions on 6/24/2026 7:31:32 AM +00:00 #129860 - [Perf] Linux/arm64: 9 Regressions on 6/24/2026 7:31:32 AM +00:00
• Date: 2026-06-24 | Queue: AzureLinux.3.Cobalt.Arm64.Perf | OS/Arch: linux/arm64
• Commits: 0d1a20fb1410 → e3cbc00cd3e0 (16 commits)
• Labels: untriaged, branch-refs/heads/main, perf-regression, arch-arm64, os-linux, runtime-coreclr, kind-micro, compilationmode-tiered, runkind-micro
• Runtime Stack: Desktop + CoreCLR (implicit)
• Historical Data: Available (283–286 runs per test)

Test Analysis

Test Name	Baseline	Compare	Δ%	μ	σ	n	z	Thresh%	Noise?	Assessment
Perf_Reader.ReadReturnBytes(HelloWorld)	103.53	135.76	+31.1%	95.91	12.57	286	3.17	26.2%	N	Step change, clean break
Perf_Reader.ReadSpanEmptyLoop(HelloWorld)	95.06	122.44	+28.8%	92.95	12.90	285	2.29	27.8%	N	Step change, no overlap
Perf_RandomAccess.ReadScatter(Async)	7.41μs	16.89μs	+128.0%	12.34μs	1.54μs	286	2.96	24.9%	N	Step change, elevated cluster
Perf_Reader.ReadSingleSpanSequenceEmptyLoop(Json400B)	932.79	1121.09	+20.2%	921.38	94.30	286	2.12	20.5%	N	Borderline, step pattern
Perf_Reader.ReadSpanEmptyLoop(Json400B)	917.66	1086.47	+18.4%	890.21	91.35	286	2.15	20.5%	N	Borderline, drift history
Perf_Reader.ReadSpanEmptyLoop(LotsOfNumbers)	2.46ns	3.40ns	+38.0%	2.79ns	0.31ns	283	1.95	22.5%	N	Borderline, drift reversion
StringSearch.LastIndexOf_Word_NotFound	430.74	691.82	+60.6%	675.50	64.78	286	0.25	19.2%	Y	Bimodal baseline outlier
CreateAddAndClear.ImmutableDictionary	111540	171569	+53.8%	177997	19062	286	-0.34	21.4%	Y	Bimodal baseline outlier
Span.LastIndexOfValue	20.12	32.34	+60.8%	31.51	2.96	286	0.28	18.8%	Y	Bimodal baseline outlier

Legend: μ=mean, σ=stddev, n=runs, z=z-score, Thresh%=adaptive threshold

Finding Confidence: 4/5 - Strong historical data (283–286 runs per test), adequate variance characterization, complete commit range information.

Candidate Stack Relevance

Commit	Touched Paths (summary)	Relevance	Notes
65b2508c9e09	src/libraries/System.Text.Json/src/.../Reader/	Shared	BCL library; directly exercises the Json Reader path under test
46b72c7c8159	src/coreclr/System.Private.CoreLib/...AsyncProfiler, src/libraries/.../CompilerServices/	Relevant	CoreCLR async profiler instrumentation; potential I/O overhead
371731cc1a49	src/coreclr/vm/threadstatics.cpp	Relevant	Thread-local alignment fix; unlikely perf impact
20703f6bb2b7	src/libraries/System.Memory/, src/libraries/System.IO/ (Stream wrappers)	Shared	New Stream types; unlikely to affect existing paths
a79c6fbd8514	src/mono/	LessLikely	Mono/WASM only; not exercised on CoreCLR

Confirmed Regressions

System.Text.Json.Tests.Perf_Reader.ReadReturnBytes(IsDataCompact: False, TestCase: HelloWorld) (+31.1%)

• Delta: 103.53 → 135.76 (+31.1%)
• Historical Context: μ=95.91 (σ=12.57, n=286); z-score=3.17; adaptive threshold=26.2%
• Assessment: Step change — all 5 after values (129.7–135.8) exceed the historical max (112.8), confirming a clean distributional break with no overlap.
• Finding Confidence: 4/5 - Strong historical data, clear statistical signal
• Regression Confidence: 4/5 - Direct code path hit (z=3.17), focused diff targets exact test area, step change pattern
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace"
  • Affected Areas: System.Text.Json / Utf8JsonReader / whitespace scanning
  • Stack Relevance: Shared (BCL library used by CoreCLR)
  • Rationale: This commit replaces the scalar whitespace-skipping loop in Utf8JsonReader.SkipWhiteSpace() with a vectorized SearchValues<byte>-based IndexOfAnyExcept approach. The test uses IsDataCompact: False (pretty-printed JSON), which exercises the whitespace-skipping path directly. While benchmarks showed improvements on x64, the arm64 SearchValues implementation may have different performance characteristics — specifically, the additional CountNewLines bookkeeping pass over the skipped whitespace run introduces a second traversal that on arm64 may cost more than the scalar loop's single-pass approach. The commit timestamp (06:25:29 UTC) immediately precedes the regression date (07:31:32 UTC).
  • Candidates Considered:
    • 65b2508c9e09 - Direct modification of Utf8JsonReader whitespace skipping
    • 46b72c7c8159 - Async profiler V1 instrumentation (timing coincidence, but no direct Json path)
• Verification Status: Not verified

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: HelloWorld) (+28.8%)

• Delta: 95.06 → 122.44 (+28.8%)
• Historical Context: μ=92.95 (σ=12.90, n=285); z-score=2.29; adaptive threshold=27.75%
• Assessment: Step change — all 5 after values (122.4–124.6) exceed the historical max (109.7), complete separation.
• Finding Confidence: 4/5 - Strong historical data, clear separation
• Regression Confidence: 4/5 - Same root cause as ReadReturnBytes, corroborating pattern, step change
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace"
  • Affected Areas: System.Text.Json / Utf8JsonReader / whitespace scanning
  • Stack Relevance: Shared
  • Rationale: Same mechanism as ReadReturnBytes — this test also uses IsDataCompact: False and exercises the SkipWhiteSpace path. The corroboration across multiple Json Reader tests with the same IsDataCompact: False parameter strongly implicates the whitespace-skipping change.
  • Candidates Considered:
    • 65b2508c9e09 - Direct modification of whitespace skipping
• Verification Status: Not verified

System.IO.Tests.Perf_RandomAccess.ReadScatter(fileSize: 104857600, buffersSize: 65536, options: Asynchronous) (+128%)

• Delta: 7.41μs → 16.89μs (+128%)
• Historical Context: μ=12.34μs (σ=1.54μs, n=286); z-score=2.96; adaptive threshold=24.9%
• Assessment: Step change — all 5 after values (13.2–19.2μs) exceed the historical mean. The baseline (7.41μs) was a low outlier inflating the raw delta, but the recentAfterMean (~16.4μs) is genuinely ~29% above recentBeforeMean (~12.7μs), confirming an elevated post-regression cluster.
• Finding Confidence: 3/5 - Strong historical data but baseline was anomalous low outlier; true delta is closer to +29% (recentAfter vs recentBefore) rather than +128%
• Regression Confidence: 3/5 - Multiple plausible commits, I/O test with high environmental sensitivity, baseline outlier complicates interpretation
• Likely Cause: 46b72c7c8159 - "Async profiler: V1 (TaskAsync) instrumentation + tests"
  • Affected Areas: CoreCLR async profiler / TaskAsync / async I/O paths
  • Stack Relevance: Relevant (CoreCLR async infrastructure)
  • Rationale: This commit adds V1 async profiler instrumentation to AsyncTaskMethodBuilder, ConfiguredValueTaskAwaitable, and other async infrastructure. The ReadScatter test uses options: Asynchronous, exercising the async I/O path. The instrumentation adds InstrumentCheckPoint guards at await-completion sites that may add overhead to every async continuation. The commit timestamp (07:31:32 UTC) exactly matches the regression date. However, the baseline was a low outlier and the true elevation vs recent history is ~29%, not 128%.
  • Candidates Considered:
    • 46b72c7c8159 - Async profiler V1 instrumentation affecting async paths
    • 20703f6bb2b7 - Stream wrappers (adds new stream types but shouldn't affect existing RandomAccess)
• Verification Status: Not verified

Borderline Cases

System.Text.Json.Tests.Perf_Reader.ReadSingleSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: Json400B) (+20.2%)

• Delta: 932.79 → 1121.09 (+20.2%)
• Historical Context: μ=921.38 (σ=94.30, n=286); z=2.12; threshold=20.47%
• Assessment: Borderline — delta falls 0.28pp below threshold, but all 5 after values exceed beforeMax (1086.2). Structurally identical pattern to the confirmed regressions. Likely same root cause (65b2508).

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: Json400B) (+18.4%)

• Delta: 917.66 → 1086.47 (+18.4%)
• Historical Context: μ=890.21 (σ=91.35, n=286); z=2.15; threshold=20.52%
• Assessment: Borderline — delta below threshold, marginal z. Historical drift from early ~1018 down to recent ~850 complicates interpretation. Same likely cause (65b2508).

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: LotsOfNumbers) (+38.0%)

• Delta: 2.46ns → 3.40ns (+38.0%)
• Historical Context: μ=2.79ns (σ=0.31ns, n=283); z=1.95; threshold=22.49%
• Assessment: Borderline — z-score just under 2.0. Historical series shows sustained downward drift (early ~3.2ns → recent ~2.4ns). Compare value (~3.4ns) sits near the early historical mean, suggesting reversion from drift rather than a new regression.

Noise (Dismissed)

• StringSearch.LastIndexOf_Word_NotFound (+60.6%): z=0.25; compare (691.8) at historical mean. Bimodal baseline landed on low mode.
• CreateAddAndClear.ImmutableDictionary (+53.8%): z=-0.34; compare is 3.6% below historical mean. Bimodal baseline outlier.
• Span.LastIndexOfValue (+60.8%): z=0.28; compare (32.3) matches historical mean (31.5). Baseline landed in rare low cluster (~20).

Open Questions

• Historical Data: Available and strong (283–286 runs), but high CV (10–14%) on the Json tests raises adaptive thresholds.
• Confidence Limitations: arm64-specific behavior of SearchValues<byte> not benchmarked in the PR. The commit author noted improvements on x64 — arm64 may not benefit equally.
• Verification Needed: RunWithBuildAtHash on commits 65b2508 (before/after) to confirm the Json Reader regression on arm64, and on 46b72c7 for the ReadScatter regression.
• Bisection Status: Not performed. Range is 16 commits — manageable but verification of the top candidate would be more efficient.

Recommended Actions

1. Verify Json Reader regression: Run bisection/verification with RunWithBuildAtHash on 5bd107431d14 (before) vs 65b2508c9e09 (after) using the Perf_Reader.ReadReturnBytes test on arm64
2. Investigate arm64 SearchValues behavior: The IndexOfAnyExcept + CountNewLines double-pass may regress on arm64 where the scalar single-pass loop had better branch prediction / smaller working set
3. Verify ReadScatter regression: Run verification on 65b2508c9e09 (before) vs 46b72c7c8159 (after) for the async I/O path

Bisection Results

Test	Status	Culprit Commit	Relevance	Reported Diff (issue)	Bisected Diff (at culprit)	Duration
ReadSpanEmptyLoop(IsDataCompact: Fals...	✅ Success	65b2508c9e09	✅ RELATED	2.5→3.4 μs (+38%)	2.4→3.4 μs (+39%)	1672s
ReadReturnBytes(IsDataCompact: False,...	✅ Success	65b2508c9e09	✅ RELATED	103.5→135.8 ns (+31%)	98.6→132.4 ns (+34%)	1384s
ReadSpanEmptyLoop(IsDataCompact: Fals...	✅ Success	65b2508c9e09	✅ RELATED	95.1→122.4 ns (+29%)	96.3→124.7 ns (+30%)	1307s
ReadSingleSpanSequenceEmptyLoop(IsDat...	✅ Success	65b2508c9e09	✅ RELATED	0.9→1.1 μs (+20%)	0.9→1.1 μs (+20%)	791s
ReadSpanEmptyLoop(IsDataCompact: Fals...	✅ Success	65b2508c9e09	✅ RELATED	0.9→1.1 μs (+18%)	0.9→1.1 μs (+20%)	1602s
LastIndexOf_Word_NotFound(Options: (,...	❌ timeout	-	-	430.7→691.8 ns (+61%)	— (not measured)	9000s
ReadScatter(fileSize: 104857600, buff...	❌ failed	-	-	7.4→16.9 μs (+128%)	— (not measured)	2187s
ImmutableDictionary(Size: 512)	❌ failed	-	-	111.5→171.6 ns (+54%)	— (not measured)	1475s
LastIndexOfValue(Size: 512)	❌ failed	-	-	20.1→32.3 ns (+61%)	— (not measured)	1682s

Total bisection time: 21101s
Success rate: 5/9 tests

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: LotsOfNumbers)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (Use SearchValues in Utf8JsonReader.SkipWhiteSpace #129701)
• Relevance: RELATED
• Explanation: This commit directly modifies Utf8JsonReader.SkipWhiteSpace(), replacing the scalar whitespace-skipping loop with a vectorized SearchValues-based IndexOfAnyExcept path. The ReadSpanEmptyLoop(IsDataCompact: False, TestCase: LotsOfNumbers) benchmark exercises exactly this code path — it reads non-compact JSON token-by-token, hitting SkipWhiteSpace at every token boundary. The added CountNewLines bookkeeping overhead for short whitespace runs between numbers likely explains the +38% regression.

Helix jobs

Job ID	Submitted by bisection	Last status
ce6be6bb-1f26-481e-b8c2-63cffadb4319	✅	submitted
27f3649e-9c75-4c07-8758-e1c0cb5d9ac8	✅	submitted

Per-commit measurements

Commit	Position	Measurements
46b72c7c8159	3/4, regressed	3.400 μs
5efb24e1e9f9	index 9, good	2.448 μs
5bd107431d14	index 10, good	2.449 μs
65b2508c9e09	index 11, regressed	3.405 μs
aedbbd2b17ca	mid, good	2.452 μs

Bisection details

…[earlier output omitted]

The regression makes sense: the ReadSpanEmptyLoop benchmark reads with IsDataCompact: False (pretty-printed JSON with whitespace), and while the PR claims improvements for deeply-indented JSON, the LotsOfNumbers test case likely has short whitespace runs between tokens where the vectorized SearchValues + CountNewLines overhead is costlier than the original simple scalar loop.

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 5
- Confidence: 4/5
- Commit Relevance: RELATED
- Relevance Explanation: This commit directly modifies `Utf8JsonReader.SkipWhiteSpace()`, replacing the scalar whitespace-skipping loop with a vectorized `SearchValues`-based `IndexOfAnyExcept` path. The `ReadSpanEmptyLoop(IsDataCompact: False, TestCase: LotsOfNumbers)` benchmark exercises exactly this code path — it reads non-compact JSON token-by-token, hitting `SkipWhiteSpace` at every token boundary. The added `CountNewLines` bookkeeping overhead for short whitespace runs between numbers likely explains the +38% regression.
- Baseline Value: 2.449 us
- Culprit Value: 3.405 us
- Summary: Bisected a 16-commit range using double bisection. First job narrowed to Q3 (commits 9-12), second job pinpointed commit 65b2508c (index 11) as the exact point where performance degraded from ~2.45 μs to ~3.40 μs.

HELIX JOBS:
- ce6be6bb-1f26-481e-b8c2-63cffadb4319
- 27f3649e-9c75-4c07-8758-e1c0cb5d9ac8

MEASUREMENTS:
- aedbbd2b17ca (mid, good): 2.452 us
- 46b72c7c8159 (3/4, regressed): 3.400 us
- 5efb24e1e9f9 (index 9, good): 2.448 us
- 5bd107431d14 (index 10, good): 2.449 us
- 65b2508c9e09 (index 11, regressed): 3.405 us

System.Text.Json.Tests.Perf_Reader.ReadReturnBytes(IsDataCompact: False, TestCase: HelloWorld)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (Use SearchValues in Utf8JsonReader.SkipWhiteSpace #129701)
• Relevance: RELATED
• Explanation: This commit directly modifies Utf8JsonReader.SkipWhiteSpace() in System.Text.Json, replacing the scalar whitespace-skipping loop with a vectorized SearchValues-based IndexOfAnyExcept path. The Perf_Reader.ReadReturnBytes benchmark exercises Utf8JsonReader token-by-token reading, which invokes SkipWhiteSpace between tokens. The "HelloWorld" test case with IsDataCompact: False (pretty-printed JSON) specifically stresses whitespace handling, making this commit the direct and obvious cause of the performance change.

Helix jobs

Job ID	Submitted by bisection	Last status
78c5a6ee-321b-48be-90cd-5b644ce21b85	✅	submitted
f73c122f-7757-4dc6-be8e-12daf5591f8e	✅	submitted
cff05619-50e8-4868-849e-1f5fee4febd6	✅	submitted

Per-commit measurements

Commit	Position	Measurements
0d1a20fb1410	baseline	98.330 ns
5bd107431d14	confirm baseline	98.620, 103.240 ns
46b72c7c8159	3/4	133.020 ns
5efb24e1e9f9	idx 9	99.050 ns
65b2508c9e09	idx 11 - culprit	132.220, 132.440 ns
aedbbd2b17ca	mid	100.980 ns
e3cbc00cd3e0	head	128.630 ns

Bisection details

…[earlier output omitted]

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 10
- Confidence: 4/5
- Commit Relevance: RELATED
- Relevance Explanation: This commit directly modifies `Utf8JsonReader.SkipWhiteSpace()` in `System.Text.Json`, replacing the scalar whitespace-skipping loop with a vectorized `SearchValues`-based `IndexOfAnyExcept` path. The `Perf_Reader.ReadReturnBytes` benchmark exercises `Utf8JsonReader` token-by-token reading, which invokes `SkipWhiteSpace` between tokens. The "HelloWorld" test case with `IsDataCompact: False` (pretty-printed JSON) specifically stresses whitespace handling, making this commit the direct and obvious cause of the performance change.
- Baseline Value: 98.62 ns
- Culprit Value: 132.44 ns
- Summary: Bisected 16 commits between baseline (0d1a20fb) and compare (e3cbc00cd) using double-bisection, narrowing to the exact commit in 3 Helix jobs. The regression is confirmed reproducible with a ~34 ns step change at commit 65b2508c, which rewrites the whitespace-skipping hot path in Utf8JsonReader. Note: while the PR claims improvements for whitespace-heavy documents, the ReadReturnBytes(HelloWorld) benchmark with compact=False appears to hit a case where the new vectorized path is slower, possibly due to short whitespace runs between tokens not amortizing the SearchValues setup cost on ARM64.

HELIX JOBS:
- 78c5a6ee-321b-48be-90cd-5b644ce21b85
- f73c122f-7757-4dc6-be8e-12daf5591f8e
- cff05619-50e8-4868-849e-1f5fee4febd6

MEASUREMENTS:
- 0d1a20fb1410 (baseline): 98.33 ns
- aedbbd2b17ca (mid): 100.98 ns
- 46b72c7c8159 (3/4): 133.02 ns
- e3cbc00cd3e0 (head): 128.63 ns
- 5efb24e1e9f9 (idx 9): 99.05 ns
- 5bd107431d14 (idx 10): 103.24 ns
- 65b2508c9e09 (idx 11 - culprit): 132.22 ns
- 5bd107431d14 (confirm baseline): 98.62 ns
- 65b2508c9e09 (confirm culprit): 132.44 ns

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: HelloWorld)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (Use SearchValues in Utf8JsonReader.SkipWhiteSpace #129701)
• Relevance: RELATED
• Explanation: This commit directly modifies the Utf8JsonReader.SkipWhiteSpace() method to use vectorized SearchValues/IndexOfAnyExcept scanning. The benchmark ReadSpanEmptyLoop(IsDataCompact: False) reads non-compact (pretty-printed) JSON in a tight loop, exercising SkipWhiteSpace on every token. For the small "HelloWorld" document, the vectorized path's setup overhead likely exceeds the cost of the previous scalar loop on short whitespace runs.

Helix jobs

Job ID	Submitted by bisection	Last status
174d85b0-943b-4a19-8081-89890b04ec37	✅	submitted
6c85d7e0-302e-47a4-85c0-14bae74f33cc	✅	submitted

Per-commit measurements

Commit	Position	Measurements
5efb24e1e9f9	index 9	96.120 ns
5bd107431d14	index 10	96.280 ns
65b2508c9e09	index 11	124.710 ns
46b72c7c8159	three-quarter	124.000 ns
aedbbd2b17ca	mid	95.430 ns

Bisection details

…[earlier output omitted]

The regression makes sense: the ReadSpanEmptyLoop test with IsDataCompact: False (pretty-printed JSON with whitespace) and HelloWorld (a very small document) may be encountering overhead from the vectorized path on very short whitespace runs in a small document, where the old scalar loop was faster due to lower setup cost.

Now let me compile the final results:

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 5
- Confidence: 4/5
- Commit Relevance: RELATED
- Relevance Explanation: This commit directly modifies the `Utf8JsonReader.SkipWhiteSpace()` method to use vectorized `SearchValues`/`IndexOfAnyExcept` scanning. The benchmark `ReadSpanEmptyLoop(IsDataCompact: False)` reads non-compact (pretty-printed) JSON in a tight loop, exercising `SkipWhiteSpace` on every token. For the small "HelloWorld" document, the vectorized path's setup overhead likely exceeds the cost of the previous scalar loop on short whitespace runs.
- Baseline Value: 96.28 ns
- Culprit Value: 124.71 ns
- Summary: Bisected 16 commits in 2 Helix jobs. First job identified regression between index 8 (aedbbd2b, 95.43 ns) and index 12 (46b72c7c, 124.00 ns). Second job narrowed to the exact commit: 5bd10743 (96.28 ns, good) → 65b2508c (124.71 ns, regressed). The step change of ~28 ns aligns with the reported +28.8% regression.

HELIX JOBS:
- 174d85b0-943b-4a19-8081-89890b04ec37
- 6c85d7e0-302e-47a4-85c0-14bae74f33cc

MEASUREMENTS:
- aedbbd2b17ca (mid): 95.43 ns
- 46b72c7c8159 (three-quarter): 124.00 ns
- 5efb24e1e9f9 (index 9): 96.12 ns
- 5bd107431d14 (index 10): 96.28 ns
- 65b2508c9e09 (index 11): 124.71 ns

System.Text.Json.Tests.Perf_Reader.ReadSingleSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: Json400B)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (Use SearchValues in Utf8JsonReader.SkipWhiteSpace #129701)
• Relevance: RELATED
• Explanation: This commit directly modifies Utf8JsonReader.SkipWhiteSpace() to use vectorized SearchValues-based scanning. The regression test Perf_Reader.ReadSingleSpanSequenceEmptyLoop exercises exactly this code path by reading JSON tokens through the Utf8JsonReader, which calls SkipWhiteSpace on every token advance. The +20% regression on the Json400B (IsDataCompact=False) case is plausibly caused by the overhead of the new vectorized path on relatively short whitespace runs in moderately-sized JSON.

Helix jobs

Job ID	Submitted by bisection	Last status
1b38d144-797d-4cde-8bb5-9a0340c0f7a9	✅	submitted
e77c853d-38ef-448e-b072-0f9a30be15c2	✅	submitted

Per-commit measurements

Commit	Position	Measurements
46b72c7c8159	3/4 point	1.119 μs
5efb24e1e9f9	index 9	0.951 μs
5bd107431d14	index 10	0.934 μs
65b2508c9e09	index 11	1.116 μs
aedbbd2b17ca	midpoint	0.935 μs

Bisection details

…[earlier output omitted]

● get_single_commit_summary (MCP: PerfTriageMcp) · owner: "dotnet", repo: "runtime", sha: "65b2508c…
└ {"index":null,"sha":"65b2508c9e09b7ab0dad595a06d583df7688a210","abbrevSha":"6...

This is very clearly the culprit. The commit "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)" directly modifies the Utf8JsonReader code path that the test Perf_Reader.ReadSingleSpanSequenceEmptyLoop exercises. The PR optimizes whitespace skipping with vectorized operations - while it shows improvements for pretty-printed JSON, it appears to regress the Json400B compact=False test case by ~20%.

Now let me compile my findings:

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 5
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: This commit directly modifies `Utf8JsonReader.SkipWhiteSpace()` to use vectorized `SearchValues`-based scanning. The regression test `Perf_Reader.ReadSingleSpanSequenceEmptyLoop` exercises exactly this code path by reading JSON tokens through the Utf8JsonReader, which calls SkipWhiteSpace on every token advance. The +20% regression on the `Json400B` (IsDataCompact=False) case is plausibly caused by the overhead of the new vectorized path on relatively short whitespace runs in moderately-sized JSON.
- Baseline Value: 934.07 ns
- Culprit Value: 1116.33 ns
- Summary: Bisected 16 commits in 2 Helix jobs. The regression is cleanly isolated to commit 65b2508c which replaced the scalar whitespace-skipping loop in Utf8JsonReader with a vectorized SearchValues-based approach.

HELIX JOBS:
- 1b38d144-797d-4cde-8bb5-9a0340c0f7a9
- e77c853d-38ef-448e-b072-0f9a30be15c2

MEASUREMENTS:
- aedbbd2b17ca (midpoint): 935.39 ns
- 46b72c7c8159 (3/4 point): 1119.50 ns
- 5efb24e1e9f9 (index 9): 950.97 ns
- 5bd107431d14 (index 10): 934.07 ns
- 65b2508c9e09 (index 11): 1116.33 ns

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: Json400B)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (Use SearchValues in Utf8JsonReader.SkipWhiteSpace #129701)
• Relevance: RELATED
• Explanation: The commit directly modifies Utf8JsonReader.SkipWhiteSpace() — the exact hot path exercised by the Perf_Reader.ReadSpanEmptyLoop benchmark. It replaces the scalar whitespace-skipping loop with a vectorized SearchValues/IndexOfAnyExcept approach. For the small 400B non-compact JSON payload, the additional overhead of CountNewLines bookkeeping in the new path causes the ~18% regression.

Helix jobs

Job ID	Submitted by bisection	Last status
79821f6e-2303-41aa-8026-61f557e0cfa4	✅	submitted
0f775f7d-ddf6-4bb7-bfcf-973f44f706fa	✅	submitted

Per-commit measurements

Commit	Position	Measurements
aedbbd2b17ca	mid, index 8	0.902 μs
5efb24e1e9f9	index 9	0.920 μs
5bd107431d14	index 10	0.906 μs
65b2508c9e09	index 11	1.083 μs
46b72c7c8159	three-quarter, index 12	1.084 μs

Bisection details

…[earlier output omitted]

688a210","abbrevSha":"6...

This is clearly the culprit. The commit "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)" directly modifies the Utf8JsonReader's SkipWhiteSpace method — which is the exact code path exercised by the ReadSpanEmptyLoop benchmark. The test reads a compact=False (pretty-printed) JSON document, so whitespace skipping is heavily used.

While the PR claims improvements for pretty-printed JSON, the ReadSpanEmptyLoop test with Json400B and IsDataCompact: False shows a regression. This makes sense — for a small 400-byte document, the overhead of the vectorized path (SearchValues setup, CountNewLines) may outweigh the scalar loop which was fast for short whitespace runs.

Let me now compile the final results:

BISECTION RESULT:

• Culprit Commit: 65b2508
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (Use SearchValues in Utf8JsonReader.SkipWhiteSpace #129701)
• Commits Tested: 5
• Confidence: 5
• Commit Relevance: RELATED
• Relevance Explanation: The commit directly modifies Utf8JsonReader.SkipWhiteSpace() — the exact hot path exercised by the Perf_Reader.ReadSpanEmptyLoop benchmark. It replaces the scalar whitespace-skipping loop with a vectorized SearchValues/IndexOfAnyExcept approach. For the small 400B non-compact JSON payload, the additional overhead of CountNewLines bookkeeping in the new path causes the ~18% regression.
• Baseline Value: 905.74 ns
• Culprit Value: 1083.26 ns
• Summary: Double bisection narrowed 16 commits to Q3 (indices 9-12), then a second Helix job tested all 3 intermediate commits and identified a clear step change at commit 65b2508 where performance jumped from ~906ns to ~1083ns.

HELIX JOBS:

• 79821f6e-2303-41aa-8026-61f557e0cfa4
• 0f775f7d-ddf6-4bb7-bfcf-973f44f706fa

MEASUREMENTS:

• aedbbd2 (mid, index 8): 901.61 ns
• 46b72c7 (three-quarter, index 12): 1084.38 ns
• 5efb24e (index 9): 919.80 ns
• 5bd1074 (index 10): 905.74 ns
• 65b2508 (index 11): 1083.26 ns

System.Globalization.Tests.StringSearch.LastIndexOf_Word_NotFound(Options: (, IgnoreCase, False))

• Status: timeout

System.IO.Tests.Perf_RandomAccess.ReadScatter(fileSize: 104857600, buffersSize: 65536, options: Asynchronous)

• Status: failed

Helix jobs

Job ID	Submitted by bisection	Last status
3e9ccf5d-6470-446a-bb92-08ec8976c56c	✅	submitted
569fb187-142d-4566-ad85-e172985eb46e	✅	completed
16657901-6926-4b54-9e4c-37e66e1c82d0	✅	submitted
94df3b5c-d691-48df-a3b6-cb929e0d675d	✅	submitted

Per-commit measurements

Commit	Position	Measurements
0d1a20fb1410	baseline	13.870 ms
a79c6fbd8514	index 0 (run 1, 2)	13.480, 13.850 ms
df7a83ee3fa9	index 1	14.930 ms
b92fa572f7d0	index 2	13.600 ms
5fcd603a0a22	index 3	18.840 ms
371731cc1a49	index 5	12.880 ms
20703f6bb2b7	index 7	17.770 ms
aedbbd2b17ca	index 8	15.180 ms
46b72c7c8159	index 12	14.420 ms

Bisection details

…[earlier output omitted]

• Index 3 (5fcd603): 18.84 ms
• Index 5 (371731c): 12.88 ms
• Index 7 (20703f6): 17.77 ms
• Index 8 (aedbbd2): 15.18 ms
• Index 12 (46b72c7): 14.42 ms

With 2σ noise floor of ~3.4 ms (CV 12.47%), all measurements are within noise of each other. The high values (18.84, 17.77) are interspersed with low values (12.88) — this is noise, not a step change.

BISECTION FAILED:
- Reason: No observable step change in measurements across the commit range; all values cluster around 13-15 ms (with noise spikes to 17-19 ms) including the baseline commit itself, which measures 13.87 ms rather than the reported 7.41 ms
- Helix Errors: None — all jobs completed successfully
- What was attempted: Submitted 4 Helix jobs testing 10 commits (including the baseline) across the full commit range. Measured baseline commit (0d1a20fb) at 13.87 ms vs reported 7.41 ms (1.87x magnitude mismatch). No consistent step change observed between any adjacent commits — values fluctuate between 12.88-18.84 ms with no monotonic transition.
- What is needed: The reported regression (7.41→16.89 ms) may not be reproducible on the AzureLinux.3.Cobalt.Arm64.Perf queue with default tiered-JIT configuration, or requires multiple samples per commit to overcome the extremely high noise floor (CV 12.47%, 2σ ≈ 41.56% of baseline). Re-running with the exact original run configuration or with many more iterations per commit may be needed.

HELIX JOBS:
- 3e9ccf5d-6470-446a-bb92-08ec8976c56c
- 569fb187-142d-4566-ad85-e172985eb46e
- 16657901-6926-4b54-9e4c-37e66e1c82d0
- 94df3b5c-d691-48df-a3b6-cb929e0d675d

MEASUREMENTS:
- 0d1a20fb1410 (baseline): 13.87 ms
- a79c6fbd8514 (index 0, run 1): 13.48 ms
- a79c6fbd8514 (index 0, run 2): 13.85 ms
- df7a83ee3fa9 (index 1): 14.93 ms
- b92fa572f7d0 (index 2): 13.60 ms
- 5fcd603a0a22 (index 3): 18.84 ms
- 371731cc1a49 (index 5): 12.88 ms
- 20703f6bb2b7 (index 7): 17.77 ms
- aedbbd2b17ca (index 8): 15.18 ms
- 46b72c7c8159 (index 12): 14.42 ms

System.Collections.CreateAddAndClear.ImmutableDictionary(Size: 512)

• Status: failed

Helix jobs

Job ID	Submitted by bisection	Last status
cd283563-cc16-4130-9c5c-68d56c298930	✅	submitted
18e380d9-aa61-4648-b6e8-61ac33084103	✅	submitted
7040dd31-f628-4b80-a9d0-fc577e4724fc	✅	submitted
8f8933f2-c934-4ba5-9257-295289967ae4	✅	submitted

Per-commit measurements

Commit	Position	Measurements
0d1a20fb1410	baseline commit	171.225 μs
a79c6fbd8514	index 0	166.680 μs
df7a83ee3fa9	index 1	167.585 μs
5fcd603a0a22	index 3	165.074 μs
371731cc1a49	index 5	168.338 μs
feb1fe23e41b	index 6	167.843 μs
20703f6bb2b7	index 7	164.941 μs
aedbbd2b17ca	index 8	165.284 μs
46b72c7c8159	index 12	168.673 μs

Bisection details

…[earlier output omitted]

Since there is no observable step change in my measurements (all commits show the same performance level), and the historical analysis classifies this as noise:

BISECTION FAILED:
- Reason: No observable step change in measurements across the entire commit range. Historical analysis confirms this is noise — the reported 111.54 μs baseline is a recurring low-value outlier spike (at least 15 similar occurrences in 286 runs), while the compare value of 171.57 μs is within the normal historical range (mean ~178 μs, CV=10.7%).
- Helix Errors: None — all 3 Helix jobs completed successfully
- What was attempted: Submitted 3 Helix jobs testing 8 unique commits plus the baseline commit across the full range. All measurements (165-171 μs) cluster around the "regressed" value with no step change. AnalyzeHistoricalData confirms classification as "Noise" with z-score of -0.337.
- What is needed: This issue should be closed as noise. The reported regression is an artifact of the baseline landing on a recurring anomalous low-value outlier spike (~109k-125k ns) rather than a true performance change.

HELIX JOBS:

• cd283563-cc16-4130-9c5c-68d56c298930
• 18e380d9-aa61-4648-b6e8-61ac33084103
• 7040dd31-f628-4b80-a9d0-fc577e4724fc
• 8f8933f2-c934-4ba5-9257-295289967ae4

MEASUREMENTS:

• 0d1a20f (baseline commit): 171225.07 ns
• a79c6fb (index 0): 166679.87 ns, 169356.73 ns
• df7a83e (index 1): 167584.57 ns
• 5fcd603 (index 3): 165073.78 ns
• 371731c (index 5): 168338.36 ns
• feb1fe2 (index 6): 167843.09 ns
• 20703f6 (index 7): 164940.91 ns
• aedbbd2 (index 8): 165284.45 ns
• 46b72c7 (index 12): 168673.44 ns

System.Memory.Span.LastIndexOfValue(Size: 512)

• Status: failed

Helix jobs

Job ID	Submitted by bisection	Last status
9980f603-8f88-40c3-b0d3-eea0ac123364	✅	submitted
f6308925-e6a8-4785-ad30-a144a42fc9e1	✅	submitted
202fa3f3-1dec-461a-bbc0-723c45a952bf	✅	submitted
4344aefc-7976-4eb5-85de-4d21c0c4d0c3	✅	submitted

Per-commit measurements

Commit	Position	Measurements
0d1a20fb1410	baseline	32.290 ns
a79c6fbd8514	index 0	32.370 ns
df7a83ee3fa9	index 1	32.430 ns
5fcd603a0a22	index 3	32.450 ns
feb1fe23e41b	index 6	32.350 ns
20703f6bb2b7	index 7	32.430 ns
aedbbd2b17ca	midpoint, index 8	32.340 ns
46b72c7c8159	3/4 point, index 12	32.270 ns

Bisection details

…[earlier output omitted]

.read()
│ parts = content.split('\n\n')…
└ 14 lines…

As expected, feb1fe23e41b also measures ~32 ns in my rerun, confirming that the historical ~20 ns was a transient anomaly. All commits in this range measure ~32 ns consistently.

My measurements show no observable step change across the entire commit range. Every commit tested (including the baseline) measures ~32 ns. The reported baseline of ~20 ns was an anomalous, non-reproducible measurement.

Based on all the evidence, this is a case where bisection cannot identify a culprit because the regression is not real - the baseline was an outlier.

BISECTION FAILED:
- Reason: No observable step change in measurements across the commit range. All commits (including the reported baseline 0d1a20fb1410) measure ~32 ns when re-run on Helix. The reported baseline of 20.12 ns appears to be a transient, non-reproducible outlier measurement.
- Helix Errors: None - all jobs completed successfully
- What was attempted: Submitted 4 Helix jobs testing 9 commits across the range (including the baseline commit). All measurements returned ~32.3 ns consistently with no step change. Historical data confirms the ~20 ns measurements at the baseline and feb1fe23e41b were anomalous dips in an otherwise stable ~32 ns signal (commits before and after the baseline also measured ~32 ns historically).
- What is needed: This regression is likely noise — the baseline measurement was an outlier. The issue should be classified as noise/false-positive.

HELIX JOBS:
- 9980f603-8f88-40c3-b0d3-eea0ac123364
- f6308925-e6a8-4785-ad30-a144a42fc9e1
- 202fa3f3-1dec-461a-bbc0-723c45a952bf
- 4344aefc-7976-4eb5-85de-4d21c0c4d0c3

MEASUREMENTS:
- 0d1a20fb1410 (baseline): 32.29 ns
- a79c6fbd8514 (index 0): 32.37 ns
- df7a83ee3fa9 (index 1): 32.43 ns
- 5fcd603a0a22 (index 3): 32.45 ns
- feb1fe23e41b (index 6): 32.35 ns
- 20703f6bb2b7 (index 7): 32.43 ns
- aedbbd2b17ca (midpoint, index 8): 32.34 ns
- 46b72c7c8159 (3/4 point, index 12): 32.27 ns

🤖 Proposed Next Actions

Rule: R3_known_regression_with_prior — Known-cause regression — already linked.

Verified regression attributed to commit 65b2508; an autofile issue is already linked: dotnet/perf-autofiling-issues#75007.

#	Action	Status	Details
1	Append related regression	⏭️ Skipped (dry-run)	on dotnet/perf-autofiling-issues#75007
2	Remove label	⏭️ Skipped (dry-run)	untriaged

🦆 Pre-bisection Re-analysis (advisory)

⚠️ Disputes the pre-bisection plan — advisory only, not applied

Flags: wasted-bisection-slot, possible-noise

Rationale: Three candidate tests (StringSearch.LastIndexOf_Word_NotFound, ImmutableDictionary CreateAddAndClear, and Span.LastIndexOfValue) have no corresponding per-finding signal entry to justify bisection and, given the bimodal history flag, are likely noise burning expensive slots. The six JSON Reader variants sharing a single likely commit (65b2508) could also be consolidated to fewer representatives, but are individually defensible given the 4/5 change confidence.

Suggested drop from bisection: System.Globalization.Tests.StringSearch.LastIndexOf_Word_NotFound(Options: (, IgnoreCase, False)), System.Collections.CreateAddAndClear<Int32>.ImmutableDictionary(Size: 512), System.Memory.Span<Char>.LastIndexOfValue(Size: 512) (advisory — not applied)

This critique is advisory in the current configuration and did not change which tests were bisected.

🦆 Re-analysis (advisory)

✅ Agrees with the analysis — advisory only, not applied

Flags: bimodal-history-caution

Rationale: Five independent bisections converging on the same commit (65b2508) provide strong culprit evidence and justify high confidence, but bimodal history flagged=True introduces inherent ambiguity about whether the shift is a permanent regression or a mode-switch, warranting a slight reduction in change confidence from 5 to 4.

Suggested confidence: Change Confidence → 4/5 (advisory — not applied)

This critique is advisory in the current configuration and did not change the confidences, culprit, or proposed actions above.

---

Generated by PerfTriageAgent on 2026-06-25 15:08 UTC

[LoopedBard3]

Likely due to #129701 (65b2508c9e) from range: feb1fe2...46b72c7.

FYI @eiriktsarpalis.

Wasm paths may already have fix proposed: #129838.

Related Regressions:

• [Perf] Linux/x64: 80 Regressions on 6/24/2026 7:31:32 AM +00:00 perf-autofiling-issues#75007
• [Perf] Linux/x64: 55 Regressions on 6/24/2026 7:31:32 AM +00:00 perf-autofiling-issues#74995
• [Perf] Windows/x64: 49 Regressions on 6/24/2026 7:31:32 AM +00:00 perf-autofiling-issues#75180
• [Perf] Linux/x64: 63 Regressions on 6/24/2026 7:31:32 AM +00:00 perf-autofiling-issues#75226
• [Perf] Linux/arm64: 18 Regressions on 6/24/2026 7:31:32 AM +00:00 perf-autofiling-issues#75477
• [Perf] Linux/arm64: 55 Regressions on 6/24/2026 6:25:29 AM +00:00 perf-autofiling-issues#75490