[Perf] Linux/arm64: 6 Improvements 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

Improvements in System.Text.Json.Tests.Perf_Reader

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ReadSingleSpanSequenceEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	13.19 μs	3.98 μs	0.30	0.18	False
ReadMultiSpanSequenceEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	14.69 μs	4.71 μs	0.32	0.15	False
ReadSpanEmptyLoop - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	13.27 μs	6.30 μs	0.47	0.18	False
ReadReturnBytes - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	13.58 μs	6.70 μs	0.49	0.29	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.ReadSingleSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: DeepTree)

ETL Files

Histogram

JIT Disasms

System.Text.Json.Tests.Perf_Reader.ReadMultiSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: DeepTree)

ETL Files

Histogram

JIT Disasms

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

ETL Files

Histogram

JIT Disasms

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

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

Improvements in System.Text.Json.Document.Tests.Perf_ParseThenWrite

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ParseThenWrite - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	23.39 μs	16.53 μs	0.71	0.16	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.Document.Tests.Perf_ParseThenWrite*'

Details

System.Text.Json.Document.Tests.Perf_ParseThenWrite.ParseThenWrite(IsDataIndented: True, TestCase: DeepTree)

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

Improvements in System.Text.Json.Node.Tests.Perf_ParseThenWrite

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ParseThenWrite - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	37.18 μs	30.62 μs	0.82	0.18	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.Node.Tests.Perf_ParseThenWrite*'

Details

System.Text.Json.Node.Tests.Perf_ParseThenWrite.ParseThenWrite(IsDataIndented: True, TestCase: DeepTree)

ETL Files

Histogram

JIT Disasms

Docs

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

[LoopedBard3]

🔍 Automated Triage Analysis

Summary: Verified improvement in ReadSingleSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: DeepTree) — caused by commit 65b2508c9e09 ("Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"). 6 tests bisected to related culprit commits.

Finding Confidence: 5/5 (analysis accuracy)
Improvement Confidence: 5/5 (likelihood of true improvement)

Likely Cause: 65b2508c9e09b7ab0dad595a06d583df7688a210

🤖 Proposed Next Actions: Known-cause improvement — link & close — 0 done, 3 dry-run (rule R2_known_improvement — see Full Analysis for the table). · 🦆 Re-analysis: ✅ concurs

📊 Full Analysis (click to expand)

Summary: Clear performance improvement across 6 System.Text.Json DeepTree benchmarks (-18% to -70%) attributed to commit 65b2508c9e09 which vectorized whitespace-skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept. Finding confidence 4/5 (excellent historical data, 283-287 runs), improvement confidence 5/5 (surgical change directly targeting the hot path, extreme z-scores up to -8.2, step-change pattern confirmed).

Issue Overview

• Issue: [Perf] Linux/arm64: 6 Improvements on 6/24/2026 7:31:32 AM +00:00 #75011 - [Perf] Linux/arm64: 6 Improvements on 6/24/2026 7:31:32 AM +00:00
• Type: Performance Improvement
• Date: 2026-06-24 | Queue: AzureLinux.3.Cobalt.Arm64.Perf | OS/Arch: linux/arm64
• Commits: 0d1a20fb1410 → e3cbc00cd3e0 (16 commits)
• Labels: untriaged, perf-improvement, os-linux, branch-refs/heads/main, arch-arm64, runtime-coreclr, kind-micro, compilationmode-tiered, runkind-micro
• Historical Data: Available (283-287 runs per test)
• Runtime Stack: Desktop + CoreCLR (implicit)

Test Analysis

Test Name	Baseline	Compare	Δ%	μ	σ	n	z	Thresh%	Noise?	Assessment
Perf_Reader.ReadSingleSpanSequenceEmptyLoop (DeepTree)	13.19 μs	3.98 μs	-69.9%	12946	1094	286	-8.20	16.9%	N	Step change, clear improvement
Perf_Reader.ReadMultiSpanSequenceEmptyLoop (DeepTree)	14.69 μs	4.71 μs	-67.9%	14248	1390	287	-6.86	19.5%	N	Step change, clear improvement
Perf_Reader.ReadSpanEmptyLoop (DeepTree)	13.27 μs	6.30 μs	-52.5%	12786	1299	285	-4.99	20.3%	N	Step change, clear improvement
Perf_Reader.ReadReturnBytes (DeepTree)	13.58 μs	6.70 μs	-50.7%	12925	1599	286	-3.90	24.7%	N	Step change, clear improvement
Perf_ParseThenWrite.ParseThenWrite (Document, DeepTree)	23.39 μs	16.53 μs	-29.3%	22674	2252	283	-2.73	19.9%	N	Step change, clear improvement
Perf_ParseThenWrite.ParseThenWrite (Node, DeepTree)	37.18 μs	30.62 μs	-17.7%	36994	2989	283	-2.13	16.2%	N	Borderline, pre-existing drift

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

Finding Confidence: 4/5 - Excellent historical data (283-287 prior runs per test), sufficient after-runs (6 per test), clear step-change pattern visible. Not 5/5 only because no independent verification via RunWithBuildAtHash was performed.

Candidate Stack Relevance

Commit	Touched Paths (summary)	Relevance	Notes
65b2508c9e09	src/libraries/System.Text.Json/src/System/Text/Json/Reader/	Shared	BCL library change; benefits all runtimes via vectorized whitespace scan
371731cc1a49	CoreCLR thread statics alignment	Relevant	CoreCLR-specific, but unrelated to JSON
a79c6fbd8514	src/mono/wasm/	LessLikely	Mono/WASM only

Confirmed Improvements

System.Text.Json.Tests.Perf_Reader.ReadSingleSpanSequenceEmptyLoop (-69.9%)

• Delta: 13.19 μs → 3.98 μs (-69.9%)
• Historical Context: μ=12946ns (σ=1094ns, n=286); z-score=-8.20; adaptive threshold=16.9%
• Assessment: Step change. Clear, extreme improvement with z=-8.2 far exceeding all thresholds. After-values cluster ~5900-6500ns; compare run (3976ns) is a low outlier even within the improved group.
• Finding Confidence: 4/5 - Excellent historical baseline (286 runs), clear step-change visible in data
• Improvement Confidence: 5/5 - Extreme z-score (-8.2), surgical commit directly targeting this code path, improvement magnitude matches PR benchmarks
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"
  • Affected Areas: System.Text.Json / Utf8JsonReader / whitespace scanning
  • Stack Relevance: Shared (BCL library; affects all runtimes using System.Text.Json)
  • Rationale: This commit vectorizes SkipWhiteSpace in Utf8JsonReader using SearchValues<byte> and IndexOfAnyExcept. DeepTree JSON documents have extensive indentation (whitespace), making this the exact hot path optimized. The PR's own benchmarks show "deeply nested" documents achieving 0.42× ratio (~2.4× faster), perfectly matching the ~70% improvement seen here. The commit was authored on 2026-06-24T06:25:29Z, landing within the baseline→compare window.
  • Candidates Considered:
    • 65b2508c9e09 - Direct hit: vectorized whitespace skip in Utf8JsonReader
    • 371731cc1a49 - Thread statics alignment fix; unrelated to JSON parsing
• Verification Status: Not verified via RunWithBuildAtHash. However, the commit's own PR benchmarks demonstrate the exact improvement pattern on deeply nested documents.

System.Text.Json.Tests.Perf_Reader.ReadMultiSpanSequenceEmptyLoop (-67.9%)

• Delta: 14.69 μs → 4.71 μs (-67.9%)
• Historical Context: μ=14248ns (σ=1390ns, n=287); z-score=-6.86; adaptive threshold=19.5%
• Assessment: Step change. The multi-span path also benefits because SkipWhiteSpaceMultiSegment delegates to SkipWhiteSpace. After-values cluster ~7575-7767ns; compare run (4709ns) is an outlier within the after group.
• Finding Confidence: 4/5 - Excellent historical baseline (287 runs), clear step-change
• Improvement Confidence: 5/5 - Extreme z-score (-6.86), same commit directly benefits this path per the PR description
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"
  • Affected Areas: System.Text.Json / Utf8JsonReader / multi-span whitespace scanning
  • Stack Relevance: Shared
  • Rationale: The commit explicitly notes "The multi-segment reader benefits for free, since SkipWhiteSpaceMultiSegment delegates to SkipWhiteSpace." DeepTree documents exercise heavy whitespace, making this a direct hit.

System.Text.Json.Tests.Perf_Reader.ReadSpanEmptyLoop (-52.5%)

• Delta: 13.27 μs → 6.30 μs (-52.5%)
• Historical Context: μ=12786ns (σ=1299ns, n=285); z-score=-4.99; adaptive threshold=20.3%
• Assessment: Step change. After-values tightly clustered at ~6298-6437ns with no outliers.
• Finding Confidence: 4/5 - Excellent historical baseline (285 runs)
• Improvement Confidence: 5/5 - Strong z-score (-5.0), direct code path hit
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"
  • Affected Areas: System.Text.Json / Utf8JsonReader / span whitespace scanning
  • Stack Relevance: Shared
  • Rationale: ReadSpanEmptyLoop reads through the entire JSON document calling Read() in a loop; the vectorized whitespace skip accelerates every inter-token transition in deeply indented JSON.

System.Text.Json.Tests.Perf_Reader.ReadReturnBytes (-50.7%)

• Delta: 13.58 μs → 6.70 μs (-50.7%)
• Historical Context: μ=12925ns (σ=1599ns, n=286); z-score=-3.90; adaptive threshold=24.7%
• Assessment: Step change. After-values mostly cluster ~6356-6798ns; one after outlier (4129ns) does not affect the overall pattern.
• Finding Confidence: 4/5 - Excellent historical baseline (286 runs), slightly higher CV (12.4%) but delta far exceeds noise
• Improvement Confidence: 5/5 - z=-3.9, direct code path, consistent with other reader tests
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"
  • Affected Areas: System.Text.Json / Utf8JsonReader / whitespace scanning
  • Stack Relevance: Shared
  • Rationale: ReadReturnBytes reads tokens and extracts ValueSpan bytes; the whitespace-heavy DeepTree test case makes SkipWhiteSpace the dominant cost, now vectorized.

System.Text.Json.Document.Tests.Perf_ParseThenWrite.ParseThenWrite (-29.3%)

• Delta: 23.39 μs → 16.53 μs (-29.3%)
• Historical Context: μ=22674ns (σ=2252ns, n=283); z-score=-2.73; adaptive threshold=19.9%
• Assessment: Step change. After-values stable at ~16550ns vs recent baseline ~23430ns. The smaller improvement magnitude (vs reader tests) is expected because ParseThenWrite includes parsing + writing, diluting the SkipWhiteSpace speedup.
• Finding Confidence: 4/5 - Excellent historical baseline (283 runs)
• Improvement Confidence: 4/5 - Good z-score (-2.73), logically consistent with reader optimization, but writing portion dilutes the signal
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"
  • Affected Areas: System.Text.Json / JsonDocument parsing (uses Utf8JsonReader internally)
  • Stack Relevance: Shared
  • Rationale: JsonDocument.Parse uses Utf8JsonReader internally for tokenization. The DeepTree+Indented test case exercises heavy whitespace, so the vectorized SkipWhiteSpace accelerates the parsing phase. The ~29% improvement (vs ~50-70% for pure reader tests) reflects that writing is unaffected.

System.Text.Json.Node.Tests.Perf_ParseThenWrite.ParseThenWrite (-17.7%)

• Delta: 37.18 μs → 30.62 μs (-17.7%)
• Historical Context: μ=36994ns (σ=2989ns, n=283); z-score=-2.13; adaptive threshold=16.2%
• Assessment: Borderline. Delta (-17.7%) barely exceeds threshold (16.2%), z=-2.13 just clears the z=2 bar. Pre-existing downward drift complicates attribution (earlyBeforeMean=38640 vs recentBeforeMean=37233). After-values are consistent (~30400ns) suggesting a real change but with marginal confidence.
• Finding Confidence: 4/5 - Good historical baseline (283 runs), but pre-existing drift reduces certainty
• Improvement Confidence: 3/5 - Borderline z-score, pre-existing drift, and Node.ParseThenWrite has significant overhead beyond parsing (DOM construction + serialization), diluting the signal further
• Likely Cause: 65b2508c9e09 - "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)"
  • Affected Areas: System.Text.Json / JsonNode parsing (uses Utf8JsonReader internally)
  • Stack Relevance: Shared
  • Rationale: JsonNode parsing also uses Utf8JsonReader. The smaller magnitude (-18% vs -50-70% for pure reader) is expected given DOM construction and serialization overhead. Borderline classification due to pre-existing drift and marginal z-score.
  • Candidates Considered:
    • 65b2508c9e09 - Vectorized whitespace skip
    • Pre-existing drift — part of the improvement may be continuation of existing trend

Open Questions

• Historical Data: Available and sufficient (283-287 runs). CV values elevated (8-12%) due to scattered outliers in the before series, but absolute delta magnitudes far exceed noise.
• Confidence Limitations: Compare-run values for ReadSingleSpan (3976ns) and ReadMultiSpan (4709ns) are outliers even within their own after-value clusters (~6400ns and ~7600ns respectively). The true steady-state improvement may be ~50% rather than ~70% for those tests. Monitor subsequent runs.
• Metric Direction: The autofile system reports these as improvements (lower duration = faster). The commit PR benchmarks confirm "lower is better" semantics for these duration-based microbenchmarks.
• Verification Needed: Not strictly necessary — the commit is a near-perfect match (same library, same function, same test pattern, magnitude matches PR benchmarks). Optional verification could confirm the steady-state improvement level.
• Bisection Status: Not performed (unnecessary — single obvious candidate commit with direct code path match)

Recommended Actions

1. Close as attributed - The improvement is clearly caused by 65b2508c9e09 ("Use SearchValues in Utf8JsonReader.SkipWhiteSpace")
2. Monitor - Track subsequent runs to confirm steady-state improvement levels, particularly for ReadSingleSpan/ReadMultiSpan where compare-run values appear to be outliers within the after group
3. Cross-platform check - Verify similar improvements on x64 platforms, as the SearchValues optimization is architecture-portable

Bisection Results

Test	Status	Culprit Commit	Relevance	Reported Diff (issue)	Bisected Diff (at culprit)	Duration
ReadSingleSpanSequenceEmptyLoop(IsDat...	✅ Success	65b2508c9e09	✅ RELATED	13.2→4.0 μs (-70%)	13.2→6.3 μs (-52%)	-
ReadMultiSpanSequenceEmptyLoop(IsData...	✅ Success	65b2508c9e09	✅ RELATED	14.7→4.7 μs (-68%)	14.7→7.8 μs (-47%)	-
ReadSpanEmptyLoop(IsDataCompact: Fals...	✅ Success	65b2508c9e09	✅ RELATED	13.3→6.3 μs (-53%)	13.2→6.3 μs (-52%)	-
ReadReturnBytes(IsDataCompact: False,...	✅ Success	65b2508c9e09	✅ RELATED	13.6→6.7 μs (-51%)	13.6→6.7 μs (-51%)	-
ParseThenWrite(IsDataIndented: True, ...	✅ Success	65b2508c9e09	✅ RELATED	23.4→16.5 μs (-29%)	23.4→16.6 μs (-29%)	1278s
ParseThenWrite(IsDataIndented: True, ...	✅ Success	65b2508c9e09	✅ RELATED	37.2→30.6 μs (-18%)	37.2→30.2 μs (-19%)	1166s

Total bisection time: 2444s
Success rate: 6/6 tests

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

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
• Relevance: RELATED
• Explanation: The commit vectorizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which is the exact hot path exercised by the Perf_Reader.ReadSingleSpanSequenceEmptyLoop test with DeepTree (deeply nested, whitespace-heavy JSON). The PR explicitly benchmarks this scenario and reports ~2.4× speedup for deeply nested documents.

Helix jobs

Job ID	Submitted by bisection	Last status
9a7d1e1b-3066-405e-8506-1c21e643d42e	✅	submitted
af6db0f9-0772-4f82-9b2f-e04151b6abf9	✅	submitted

Per-commit measurements

Commit	Position	Measurements
46b72c7c8159	3/4 point	6.320 μs
5efb24e1e9f9	index 9	13.239 μs
5bd107431d14	index 10	13.338 μs
65b2508c9e09	index 11	6.309 μs
aedbbd2b17ca	midpoint	13.195 μs

Bisection details

…[earlier output omitted]

The PR description even explicitly mentions:

• "pretty, deeply nested | 1,248 µs | 519 µs | 0.42× — ~2.4× faster"

This is clearly the culprit commit. Let me note that the measured improvement (~52% at 6308 ns vs 13238 ns baseline) is somewhat less than the reported 69.9%, but that's reasonable given we're measuring a slightly different metric (the full reader loop vs just whitespace skipping).

Now let me compile the final results:

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 5
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: The commit vectorizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which is the exact hot path exercised by the Perf_Reader.ReadSingleSpanSequenceEmptyLoop test with DeepTree (deeply nested, whitespace-heavy JSON). The PR explicitly benchmarks this scenario and reports ~2.4× speedup for deeply nested documents.
- Baseline Value: 13238.91 ns   (the measured perf at the last-good commit before the culprit, e.g. "1245.0 ns")
- Culprit Value: 6308.55 ns    (the measured perf at the culprit commit, same units as Baseline Value)
- Summary: Bisected 16 commits in 2 Helix jobs. The improvement was clearly introduced at commit 65b2508c which vectorizes JSON whitespace skipping using SearchValues, directly benefiting the deeply-nested JSON reader benchmark.

HELIX JOBS:
- 9a7d1e1b-3066-405e-8506-1c21e643d42e
- af6db0f9-0772-4f82-9b2f-e04151b6abf9

MEASUREMENTS:
- aedbbd2b17ca (midpoint): 13195.48 ns
- 46b72c7c8159 (3/4 point): 6320.12 ns
- 5efb24e1e9f9 (index 9): 13238.91 ns
- 5bd107431d14 (index 10): 13337.89 ns
- 65b2508c9e09 (index 11): 6308.55 ns

System.Text.Json.Tests.Perf_Reader.ReadMultiSpanSequenceEmptyLoop(IsDataCompact: False, TestCase: DeepTree)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
• Relevance: RELATED
• Explanation: The commit vectorizes whitespace-skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, directly optimizing the hot path exercised by ReadMultiSpanSequenceEmptyLoop with DeepTree (heavily indented JSON). The PR's own benchmarks show ~2.4× speedup for deeply nested documents, matching the observed improvement.

Helix jobs

Job ID	Submitted by bisection	Last status
d229ff4c-45ea-46b7-9f46-24cf397ea9f6	✅	submitted
f79d12f6-aacd-49ea-9b15-751da5588e66	✅	submitted

Per-commit measurements

Commit	Position	Measurements
0d1a20fb1410	baseline	14.720 μs
46b72c7c8159	3/4	7.600 μs
5efb24e1e9f9	q3-start	14.620 μs
5bd107431d14	pre-culprit	14.670 μs
65b2508c9e09	culprit	7.750 μs
aedbbd2b17ca	mid	14.630 μs
e3cbc00cd3e0	head	7.690 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 clearly the culprit. The commit "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)" vectorizes the whitespace-skipping in Utf8JsonReader, which is directly exercised by the ReadMultiSpanSequenceEmptyLoop(DeepTree) benchmark. The improvement from ~14.6 us to ~7.7 us (~47% faster) matches the PR's own benchmarks showing deeply nested documents getting ~2.4× faster.

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 7
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: The commit vectorizes whitespace-skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, directly optimizing the hot path exercised by ReadMultiSpanSequenceEmptyLoop with DeepTree (heavily indented JSON). The PR's own benchmarks show ~2.4× speedup for deeply nested documents, matching the observed improvement.
- Baseline Value: 14.67 us (5bd107431d14, last commit before improvement)
- Culprit Value: 7.75 us (65b2508c9e09, the improvement commit)
- Summary: Double bisection of 16 commits identified 65b2508c9e09 in two Helix jobs. The improvement is a clear step-change from ~14.6 us to ~7.7 us occurring exactly at this commit.

HELIX JOBS:
- d229ff4c-45ea-46b7-9f46-24cf397ea9f6
- f79d12f6-aacd-49ea-9b15-751da5588e66

MEASUREMENTS:
- 0d1a20fb1410 (baseline): 14.72 us
- aedbbd2b17ca (mid): 14.63 us
- 46b72c7c8159 (3/4): 7.60 us
- e3cbc00cd3e0 (head): 7.69 us
- 5efb24e1e9f9 (Q3-start): 14.62 us
- 5bd107431d14 (pre-culprit): 14.67 us
- 65b2508c9e09 (culprit): 7.75 us

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

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
• Relevance: RELATED
• Explanation: The commit vectorizes the SkipWhiteSpace hot path in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which is precisely the code path exercised by the ReadSpanEmptyLoop(DeepTree) benchmark. The PR's own benchmarks show ~2.4× speedup on deeply nested JSON, matching the observed 52% improvement.

Helix jobs

Job ID	Submitted by bisection	Last status
d409478a-fffe-4732-b806-d8c93e1b0471	✅	submitted
7b238d1b-9fc0-4ae2-bab9-a40ff7b77251	✅	submitted

Per-commit measurements

Commit	Position	Measurements
5efb24e1e9f9	index 9	13.180 μs
5bd107431d14	index 10	13.160 μs
65b2508c9e09	index 11	6.300 μs
46b72c7c8159	three-quarter	6.320 μs
aedbbd2b17ca	mid	13.190 μs

Bisection details

…[earlier output omitted]

2% improvement, matching the reported delta perfectly. This is well beyond the 2σ noise floor of 2.598.

Now let me examine this commit.

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

This is a clear match. The commit 65b2508c9e09 titled "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)" directly optimizes the Utf8JsonReader.SkipWhiteSpace method using vectorized SearchValues/IndexOfAnyExcept. The test Perf_Reader.ReadSpanEmptyLoop(IsDataCompact: False, TestCase: DeepTree) exercises the JSON reader's whitespace-skipping path on deeply nested (pretty-printed) JSON — exactly the scenario this PR optimizes by ~2.4× as noted in the commit message.

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 5
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: The commit vectorizes the SkipWhiteSpace hot path in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which is precisely the code path exercised by the ReadSpanEmptyLoop(DeepTree) benchmark. The PR's own benchmarks show ~2.4× speedup on deeply nested JSON, matching the observed 52% improvement.
- Baseline Value: 13.16 us
- Culprit Value: 6.30 us
- Summary: Double bisection narrowed 16 commits to a 4-commit window (Q3), then testing all 3 candidates pinpointed the improvement at commit 65b2508c which vectorizes Utf8JsonReader.SkipWhiteSpace using SearchValues. The step change from 13.16→6.30 μs matches the reported ~52% improvement exactly.

HELIX JOBS:
- d409478a-fffe-4732-b806-d8c93e1b0471
- 7b238d1b-9fc0-4ae2-bab9-a40ff7b77251

MEASUREMENTS:
- aedbbd2b17ca (mid): 13.19 us
- 46b72c7c8159 (three-quarter): 6.32 us
- 5efb24e1e9f9 (index 9): 13.18 us
- 5bd107431d14 (index 10): 13.16 us
- 65b2508c9e09 (index 11): 6.30 us

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

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
• Relevance: RELATED
• Explanation: This commit vectorizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept. The test reads a deeply-nested JSON tree (DeepTree with IsDataCompact=False), which has extensive whitespace between tokens — exactly the hot path this optimization targets. The PR's own benchmarks show a ~2.4× speedup for deeply nested documents, matching the observed ~50% improvement.

Helix jobs

Job ID	Submitted by bisection	Last status
a5d28700-fdb3-4805-ae53-0c525f4fdc3e	✅	submitted
1721e3a0-185a-4352-bc76-9035f3dd084f	✅	submitted

Per-commit measurements

Commit	Position	Measurements
5efb24e1e9f9	index 9	13.534 μs
5bd107431d14	index 10	13.561 μs
65b2508c9e09	index 11	6.710 μs
46b72c7c8159	three-quarter	6.767 μs
aedbbd2b17ca	mid	13.525 μs

Bisection details

…[earlier output omitted]

The PR itself shows benchmarks with a deeply nested document improving by ~2.4× (0.42× ratio), which closely matches the ~50% improvement observed in this issue.

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 vectorizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept. The test reads a deeply-nested JSON tree (DeepTree with IsDataCompact=False), which has extensive whitespace between tokens — exactly the hot path this optimization targets. The PR's own benchmarks show a ~2.4× speedup for deeply nested documents, matching the observed ~50% improvement.
- Baseline Value: 13560.92 ns   (the measured perf at the last-good commit before the culprit, e.g. "1245.0 ns")
- Culprit Value: 6709.66 ns    (the measured perf at the culprit commit, same units as Baseline Value)
- Summary: Double bisection narrowed 16 commits to a 4-commit window, then tested 3 intermediate commits to pinpoint the exact commit. The improvement cleanly appears at commit 65b2508c9e09 which vectorizes Utf8JsonReader.SkipWhiteSpace.

HELIX JOBS:
- a5d28700-fdb3-4805-ae53-0c525f4fdc3e
- 1721e3a0-185a-4352-bc76-9035f3dd084f

MEASUREMENTS:
- aedbbd2b17ca (mid): 13525.38 ns
- 46b72c7c8159 (three-quarter): 6767.41 ns
- 5efb24e1e9f9 (index 9): 13534.21 ns
- 5bd107431d14 (index 10): 13560.92 ns
- 65b2508c9e09 (index 11): 6709.66 ns

System.Text.Json.Document.Tests.Perf_ParseThenWrite.ParseThenWrite(IsDataIndented: True, TestCase: DeepTree)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
• Relevance: RELATED
• Explanation: The commit vectorizes whitespace-skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept. The test ParseThenWrite with DeepTree and IsDataIndented=True exercises exactly this hot path—deeply indented JSON has extensive whitespace runs that this optimization targets. The PR's own benchmarks show ~2.4× speedup for deeply nested documents, matching the observed improvement.

Helix jobs

Job ID	Submitted by bisection	Last status
0c46174a-e425-4879-9b4e-35e064972d0f	✅	submitted
18f2a61c-000c-48d1-a984-8c43c555e588	✅	submitted

Per-commit measurements

Commit	Position	Measurements
0d1a20fb1410	baseline	23.427 μs
46b72c7c8159	3/4	16.591 μs
5efb24e1e9f9	index 9	23.397 μs
5bd107431d14	index 10	23.438 μs
65b2508c9e09	index 11	16.613 μs
aedbbd2b17ca	mid	23.488 μs
e3cbc00cd3e0	head	16.598 μs

Bisection details

…[earlier output omitted]

The commit 65b2508c9e09 titled "Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)" directly optimizes the JSON reader's whitespace-skipping path using vectorized SearchValues/IndexOfAnyExcept. The benchmark test ParseThenWrite(IsDataIndented: True, TestCase: DeepTree) would heavily exercise whitespace skipping in deeply indented JSON, and the commit's own PR shows a ~2.4× improvement for deeply nested documents — matching the ~29% improvement observed.

Here are my findings:

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 7
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: The commit vectorizes whitespace-skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept. The test ParseThenWrite with DeepTree and IsDataIndented=True exercises exactly this hot path—deeply indented JSON has extensive whitespace runs that this optimization targets. The PR's own benchmarks show ~2.4× speedup for deeply nested documents, matching the observed improvement.
- Baseline Value: 23438.40 ns   (the measured perf at the last-good commit before the culprit, e.g. "1245.0 ns")
- Culprit Value: 16612.86 ns    (the measured perf at the culprit commit, same units as Baseline Value)
- Summary: Double bisection narrowed 16 commits to Q3 (indices 9-12), then testing all 3 remaining commits pinpointed the improvement at commit 65b2508c9e09 which vectorizes JSON whitespace skipping. The step change from ~23.4μs to ~16.6μs (-29%) is consistent and well above noise.

HELIX JOBS:
- 0c46174a-e425-4879-9b4e-35e064972d0f
- 18f2a61c-000c-48d1-a984-8c43c555e588

MEASUREMENTS:
- 0d1a20fb1410 (baseline): 23427.12 ns
- aedbbd2b17ca (mid): 23487.67 ns
- 46b72c7c8159 (3/4): 16591.11 ns
- e3cbc00cd3e0 (head): 16598.11 ns
- 5efb24e1e9f9 (index 9): 23396.92 ns
- 5bd107431d14 (index 10): 23438.40 ns
- 65b2508c9e09 (index 11): 16612.86 ns

System.Text.Json.Node.Tests.Perf_ParseThenWrite.ParseThenWrite(IsDataIndented: True, TestCase: DeepTree)

• Culprit: 65b2508c9e09
• Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
• Relevance: RELATED
• Explanation: The commit vectorizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which directly accelerates parsing of indented/deeply-nested JSON. The PR benchmarks explicitly show ~2.4× speedup for deeply nested documents, matching the observed improvement in this test.

Helix jobs

Job ID	Submitted by bisection	Last status
8c88ccd5-7941-4749-b083-37184fdd93c7	✅	submitted
5a0a206b-d247-41aa-8381-3260032b660a	✅	submitted

Per-commit measurements

Commit	Position	Measurements
5efb24e1e9f9	index 9	37.109 μs
5bd107431d14	index 10	37.230 μs
65b2508c9e09	index 11	30.249 μs
46b72c7c8159	three-quarter	30.182 μs
aedbbd2b17ca	mid	37.084 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 clearly the culprit commit. It's directly related — it optimizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which directly benefits the ParseThenWrite(DeepTree) benchmark (deeply indented JSON has lots of whitespace to skip).

The commit message explicitly mentions:

| pretty, deeply nested | 1,248 µs | 519 µs | 0.42× — ~2.4× faster |

This is a clear, strong match. The improvement in the ParseThenWrite(IsDataIndented: True, TestCase: DeepTree) test is directly caused by the vectorized whitespace-skipping optimization.

BISECTION RESULT:
- Culprit Commit: 65b2508c9e09b7ab0dad595a06d583df7688a210
- Commit Message: Use SearchValues in Utf8JsonReader.SkipWhiteSpace (#129701)
- Commits Tested: 5
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: The commit vectorizes whitespace skipping in Utf8JsonReader using SearchValues/IndexOfAnyExcept, which directly accelerates parsing of indented/deeply-nested JSON. The PR benchmarks explicitly show ~2.4× speedup for deeply nested documents, matching the observed improvement in this test.
- Baseline Value: 37230.01 ns
- Culprit Value: 30249.34 ns
- Summary: Double bisection narrowed 16 commits to 4, then a second job pinpointed commit 65b2508c as the exact introduction point. The improvement is a clear step change from ~37k ns to ~30k ns, consistent with the reported -17.7% improvement.

HELIX JOBS:
- 8c88ccd5-7941-4749-b083-37184fdd93c7
- 5a0a206b-d247-41aa-8381-3260032b660a

MEASUREMENTS:
- aedbbd2b17ca (mid): 37083.73 ns
- 46b72c7c8159 (three-quarter): 30182.47 ns
- 5efb24e1e9f9 (index 9): 37109.09 ns
- 5bd107431d14 (index 10): 37230.01 ns
- 65b2508c9e09 (index 11): 30249.34 ns

🤖 Proposed Next Actions

Rule: R2_known_improvement — Known-cause improvement — link & close.

Verified improvement attributed to commit 65b2508c9e09 (finding 5/5, improvement 5/5, bisection RELATED).

#	Action	Status	Details
1	Back-reference cause PR	⏭️ Skipped (dry-run)	on culprit PR for 65b2508c9e09
2	Remove label	⏭️ Skipped (dry-run)	untriaged
3	Close issue	⏭️ Skipped (dry-run)	reason completed

🦆 Pre-bisection Re-analysis (advisory)

✅ Agrees with the pre-bisection plan — advisory only, not applied

Rationale: All six findings share the same likely commit (65b2508c9e09), the same DeepTree test case, and high change-confidence scores (3–5/5) consistent with a real improvement in the System.Text.Json reader/writer path. Since they all point to a single culprit, bisecting even 2–3 representative tests would likely suffice, but none are obviously noisy enough to drop given the strong signal.

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

Rationale: All six bisections converged successfully on the same culprit commit with no alignment warnings, all benchmarks are in the same System.Text.Json DeepTree family with shared stack relevance, and no bimodal history is flagged. The 5/5 change confidence is well-supported by consistent, convergent bisection evidence across multiple related benchmarks.

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 21:14 UTC

[AndyAyersMS]

dotnet/runtime#129701