[Perf] Windows/x64: 3 Regressions on 7/27/2026 11:29:43 AM +00:00

[performanceautofiler]

Run Information

Name	Value
Architecture	x64
OS	Windows 10.0.22631
Queue	ViperWindows
Baseline	ca073c0866e1504313b43781c209f7f680d0ecb3
Compare	da6081391e5d3227e462c082b345af8720db05c1
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Numerics.Tests.Perf_BigInteger

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
Equals - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	5.17 ns	6.36 ns	1.23	0.02	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.Numerics.Tests.Perf_BigInteger*'

Details

System.Numerics.Tests.Perf_BigInteger.Equals(arguments: 259 bytes, Same)

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 10.0.22631
Queue	ViperWindows
Baseline	2aa2ce008a33a25ac77f4f340b628e135b94087f
Compare	da6081391e5d3227e462c082b345af8720db05c1
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.IO.Tests.Perf_StreamWriter

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
WriteString - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	78.95 ms	83.90 ms	1.06	0.01	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_StreamWriter*'

Details

System.IO.Tests.Perf_StreamWriter.WriteString(writeLength: 2)

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 10.0.22631
Queue	ViperWindows
Baseline	2aa2ce008a33a25ac77f4f340b628e135b94087f
Compare	da6081391e5d3227e462c082b345af8720db05c1
Diff	Diff
Configs	CompilationMode:tiered, RunKind:micro

Regressions in System.Buffers.Tests.RentReturnArrayPoolTests<Byte>

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio
ProducerConsumer - Duration of single invocation 📝 - Benchmark Source ADX - Test Multi Config Graph	1.32 μs	1.54 μs	1.17	0.45	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.Buffers.Tests.RentReturnArrayPoolTests<Byte>*'

Details

System.Buffers.Tests.RentReturnArrayPoolTests<Byte>.ProducerConsumer(RentalSize: 4096, ManipulateArray: True, Async: False, UseSharedPool: False)

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 WriteString(writeLength: 2) — caused by commit 65b0190 ("Avoid materializing Task/ValueTask in StreamReader/StreamWriter (#130689"). 1 other test(s) found unrelated commits (possible alignment artifacts).

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

Likely Cause: 65b0190679e7afaea3c09bcfc474c55414403972

🤖 Proposed Next Actions: Known-cause regression — transfer & notify — 0 done, 3 dry-run (rule R4_known_regression_new — see Full Analysis for the table). · 🦆 Re-analysis: ⚠️ disputes (overconfident, overconfident-change, mixed-bisection-evidence)

📊 Full Analysis (click to expand)

Summary: One clear regression confirmed — System.IO.Tests.Perf_StreamWriter.WriteString(writeLength: 2) slowed +6.3% (z=27.4 on a historically rock-stable test, CV=0.22%), attributed with strong confidence to commit 65b0190 (PR #130689, "Avoid materializing Task/ValueTask in StreamReader/StreamWriter"), which is a direct code hit on the exact API and landed on the regression date. BigInteger.Equals (+23.2%) is borderline (multi-modal history depresses z to 1.5) and the RentReturnArrayPool ProducerConsumer (+17.3%) move is measurement noise (CV=11.5%). Finding confidence 4/5, regression confidence 4/5 for the StreamWriter finding.

Issue Overview

• Issue: [Perf] Windows/x64: 3 Regressions on 7/27/2026 11:29:43 AM +00:00 #131812 - [Perf] Windows/x64: 3 Regressions on 7/27/2026 11:29:43 AM +00:00
• Date: 2026-07-27 | Queue: Windows.11.Amd64.Viper.Perf (ViperWindows) | OS/Arch: Windows 10.0.22631/x64
• Commits: ca073c08 → da608139 (20 commits)
• Labels: untriaged, perf-regression, os-windows, arch-x64, branch-refs/heads/main, runtime-coreclr, kind-micro, compilationmode-tiered, runkind-micro
• Runtime stack: Desktop + CoreCLR (implicit; runtime-coreclr, CompilationMode=tiered)
• Historical Data: Available (298–300 prior runs per test; 333 examined total)
• Correlation prior: 2 co-occurring sibling issues ([mono][s390x] Wrong implementation of OP_CHECK_THIS #76915, [Perf] Windows/x64: 5 Regressions on 7/20/2026 2:22:01 PM +00:00 perf-autofiling-issues#76914) on the same platform/date — raises prior that borderline movements are real signal, not pure noise.

Test Analysis

Test Name	Baseline	Compare	Δ%	μ	σ	n	z	Thresh%	Noise?	Assessment
Perf_StreamWriter.WriteString(writeLength: 2)	78.95 ms	83.90 ms	+6.27%	79.06 ms	0.177 ms	300	27.42	5%	N	Step change, extremely stable history
Perf_BigInteger.Equals(259 bytes, Same)	5.17 ns	6.36 ns	+23.15%	5.70 ns	0.44 ns	298	1.51	15.56%	N (Borderline)	Step-up but multi-modal history
RentReturnArrayPoolTests<Byte>.ProducerConsumer(4096, True, False, False)	1.32 μs	1.54 μs	+17.26%	1.47 μs	0.17 μs	300	0.41	23%	Y	Within natural variance (CV=11.5%)

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

Finding Confidence: 4/5 - Comprehensive historical data (298–300 runs/test), low variance on the confirmed test, complete commit range (20 commits). Not raised to 5 because no independent rerun/bisection was performed in this pass.

Candidate Stack Relevance

Detected stack: Desktop + CoreCLR. BCL (src/libraries/System.Private.CoreLib/) is directly exercised by these managed micro-benchmarks on CoreCLR; JIT/VM changes are Relevant to CoreCLR codegen.

Commit	Touched Paths (summary)	Relevance	Notes
65b0190679e7	System.Private.CoreLib/.../IO/StreamWriter.cs, StreamReader.cs	Shared (BCL)	Direct implementation change to the exact API under test; landed on regression date. Primary suspect for StreamWriter.
c40bf1205afb (#130871)	src/coreclr/jit/ (importercalls, gentree, compiler.h, corinfo)	Relevant (JIT)	"Remove boxing from value type instance calls" — could alter codegen of struct BigInteger.Equals; candidate for the BigInteger borderline.
2177d80507aa (#131362)	src/coreclr/jit/optimizer.cpp	Relevant (JIT)	Loop-inversion exit-weight change; broad codegen effect, weak specific link.
55810a448212 (#130436)	src/coreclr/pal/, crashreporter	Relevant (VM/PAL)	Signal/process handling; unlikely to affect these managed micro-benchmarks.
51c924dc634e (#131359)	src/coreclr/vm/peassembly.*	Relevant (VM)	Assembly path handling; no plausible link to hot paths here.

Confirmed Regressions

System.IO.Tests.Perf_StreamWriter.WriteString(writeLength: 2) (+6.3%)

• Delta: 78.95 ms → 83.90 ms (+6.27%)
• Historical Context: μ=79.06 ms (σ=0.177 ms, n=300); z-score=27.42; adaptive threshold=5%; CV=0.22%
• Assessment: Step change. 300 before-values tightly clustered ~78.7–79.8 ms; all 33 after-values shift uniformly to ~83.3–84.2 ms with no overlap.
• Finding Confidence: 4/5 - Large, low-variance historical baseline; unambiguous step. Not verified via rerun/bisection in this pass.
• Regression Confidence: 4/5 - Direct code path hit, extreme outlier (z=27.4), focused 2-file diff. Not 5 because attribution is not yet reproduced via RunWithBuildAtHash.
• Likely Cause: 65b0190679e7 - "Avoid materializing Task/ValueTask in StreamReader/StreamWriter (Avoid materializing Task/ValueTask in StreamReader/StreamWriter #130689)"
  • Affected Areas: BCL I/O (StreamWriter)
  • Stack Relevance: Shared (BCL) — directly exercised by this CoreCLR managed benchmark; no cross-stack ambiguity.
  • Rationale: The commit modifies the exact type under test (StreamWriter.cs, +60/-38) and was authored/committed at 2026-07-27T11:29:43Z — the precise regression timestamp. It replaces the Task-based concurrent-write diagnostic with a try/finally inside the callees to avoid allocating Task objects, but retains the Task path only in a subset of cases. For the tiny synchronous WriteString(writeLength: 2) hot path the added try/finally / restructured diagnostic bookkeeping is a plausible source of the uniform +6.3% step. Corroborated by exact temporal alignment and the test's rock-stable history.
  • Candidates Considered:
    • 65b0190679e7 - direct StreamWriter change on the regression date (chosen).
    • c40bf1205afb - JIT boxing-removal; broad codegen but no StreamWriter-specific path.
    • 2177d80507aa - JIT loop inversion; generic, no specific link.
• Verification Status: Not verified - rerun/bisection deferred (handled separately downstream). Recommend RunWithBuildAtHash on 55810a4 (parent) vs 65b0190 to confirm.

Borderline (Manual Review)

System.Numerics.Tests.Perf_BigInteger.Equals(arguments: 259 bytes, Same) (+23.2%)

• Delta: 5.17 ns → 6.36 ns (+23.15%)
• Historical Context: μ=5.70 ns (σ=0.44 ns, n=298); z-score=1.51; adaptive threshold=15.56%; CV=7.78%
• Assessment: Delta exceeds the adaptive threshold (23.15% > 15.56%) but z=1.51 < 2. History is multi-modal (~5.15/5.9/6.0/6.4 ns regimes), inflating σ and depressing z. Recent before-values were stable ~5.15 ns; after-values cluster ~6.3–6.5 ns, suggesting a genuine step-up — but the per-rule conflict warrants Borderline.
• Finding Confidence: 3/5 - Ample data but multi-modal series makes z interpretation unreliable.
• Regression Confidence: 3/5 - Threshold exceeded and step-up visible; sibling co-occurrence raises prior, but no clean statistical signal and no confirmed culprit.
• Possible Cause: c40bf1205afb (JIT: Remove boxing from value type instance calls #130871) "JIT: Remove boxing from value type instance calls" — BigInteger.Equals is a value-type instance call, so codegen changes here could plausibly shift its cost. Unconfirmed; verification recommended.

Noise (Deprioritized)

System.Buffers.Tests.RentReturnArrayPoolTests<Byte>.ProducerConsumer(RentalSize: 4096, ManipulateArray: True, Async: False, UseSharedPool: False) (+17.3%)

• Delta: 1.32 μs → 1.54 μs (+17.26%)
• Historical Context: μ=1.47 μs (σ=0.17 μs, n=300); z-score=0.41; adaptive threshold=23%; CV=11.5%
• Assessment: Inherently noisy (before-values 1264–2303 ns, after-values 1351–2660 ns). Delta below the 23% adaptive threshold and z≈0.41 negligible; compare value sits comfortably inside the historical distribution. Classified as measurement noise.

Open Questions

• Historical Data: Sufficient (298–300 runs/test). BigInteger series is multi-modal, weakening z-score reliability.
• Confidence Limitations: BigInteger attribution to the JIT boxing change is speculative; StreamWriter attribution is strong but not yet reproduced.
• Verification Needed: RunWithBuildAtHash on 65b0190 vs its parent 55810a4 for StreamWriter; separately test c40bf12 for BigInteger. Bisection likely unnecessary given the direct StreamWriter hit.
• Bisection Status: Not performed (deferred to downstream step).

Recommended Actions

1. Verify StreamWriter culprit - Run RunWithBuildAtHash comparing parent 55810a4482 vs 65b0190679e7 on System.IO.Tests.Perf_StreamWriter.WriteString(writeLength: 2) to confirm the +6.3% step.
2. Investigate BigInteger borderline - Test c40bf1205afb (JIT: Remove boxing from value type instance calls #130871) against its parent for Perf_BigInteger.Equals(259 bytes, Same); inspect struct-call codegen.
3. Close ArrayPool as noise - No action; ProducerConsumer delta is within natural variance (CV=11.5%).

Bisection Results

Test	Status	Culprit Commit	Relevance	Reported Diff (issue)	Bisected Diff (at culprit)
Equals(arguments: 259 bytes, Same)	✅ Culprit found	65b0190679e7	⚠️ LIKELY_UNRELATED	5.2→6.4 ns (+23%)	5.2→6.3 ns (+21%)
WriteString(writeLength: 2)	✅ Culprit found	65b0190679e7	✅ RELATED	78.9→83.9 ns (+6%)	79.1→83.4 ns (+5%)
ProducerConsumer(RentalSize: 4096, Ma...	🔍 No change found	-	-	1.3→1.5 μs (+17%)	— (no culprit)

Total bisection time: 7597s
Tests measured: 3/3
Culprit found: 2/3

⚠️ Alignment Warning

Bisection found commits that are unrelated to the regressed tests. This strongly suggests the performance changes are caused by code alignment, binary layout, or other non-functional factors rather than actual regressions. Consider marking as noise.

System.Numerics.Tests.Perf_BigInteger.Equals(arguments: 259 bytes, Same)

• Culprit: 65b0190679e7
• Commit Message: Avoid materializing Task/ValueTask in StreamReader/StreamWriter (Avoid materializing Task/ValueTask in StreamReader/StreamWriter #130689)
• Relevance: LIKELY_UNRELATED
• Explanation: The commit only modifies StreamReader.cs and StreamWriter.cs async-IO diagnostic paths in System.Private.CoreLib. The System.Numerics BigInteger.Equals benchmark never touches StreamReader/StreamWriter, so there is no functional mechanism. Because the change lands in CoreLib (loaded by everything), it can shift binary/code layout, which is the most plausible cause of this reproducible +21% step.
• ⚠️ This commit is UNRELATED to the test — likely a code alignment artifact

Helix jobs

Job ID	Submitted by bisection	Last status
f94926d0-6fde-4c07-a1d2-5d3c483ce2ec	✅	submitted
b2a9d598-20b8-49c6-8b20-1d460390ad25	✅	submitted
ce3d7ac3-a096-4a55-bcbb-5ce24f789ba3	✅	completed

Per-commit measurements

Commit	Position	Measurements
ca073c0866e1	baseline/good	5.164 ns
39aaf28f0916	idx2	5.162 ns
3d59a088c3bb	idx5, noisy	5.883 ns
55810a448212	idx7, last good	5.167 ns
0e04f95f1c39	idx9	6.412 ns
65b0190679e7	idx8, culprit	6.262 ns
da6081391e5d	head/compare	6.290 ns

Bisection details

…[earlier output omitted]

Bisection complete. The culprit commit modifies StreamReader.cs/StreamWriter.cs (async IO diagnostics) — no functional connection to BigInteger.Equals. The clean, reproducible step change (5.17→6.26 ns) at this commit points to a code-layout/alignment artifact in System.Private.CoreLib.

BISECTION RESULT:
- Culprit Commit: 65b0190679e7afaea3c09bcfc474c55414403972
- Commit Message: Avoid materializing `Task`/`ValueTask` in `StreamReader`/`StreamWriter` (#130689)
- Commits Tested: 8
- Confidence: 4
- Commit Relevance: LIKELY_UNRELATED
- Relevance Explanation: The commit only modifies StreamReader.cs and StreamWriter.cs async-IO diagnostic paths in System.Private.CoreLib. The System.Numerics BigInteger.Equals benchmark never touches StreamReader/StreamWriter, so there is no functional mechanism. Because the change lands in CoreLib (loaded by everything), it can shift binary/code layout, which is the most plausible cause of this reproducible +21% step.
- Baseline Value: 5.167 ns
- Culprit Value: 6.262 ns
- Summary: I confirmed the regression at the head (6.29 ns) vs baseline ca073c08 (5.16 ns), then bisected the 20-commit range. 55810a44 was still good (5.167 ns) and its immediate child 65b01906 jumped to 6.262 ns, isolating the single culprit commit.

HELIX JOBS:
- f94926d0-6fde-4c07-a1d2-5d3c483ce2ec
- b2a9d598-20b8-49c6-8b20-1d460390ad25
- ce3d7ac3-a096-4a55-bcbb-5ce24f789ba3

MEASUREMENTS:
- ca073c0866e1 (baseline/good): 5.164 ns
- 39aaf28f0916 (idx2): 5.162 ns
- 3d59a088c3bb (idx5, noisy): 5.883 ns
- 55810a448212 (idx7, last good): 5.167 ns
- 65b0190679e7 (idx8, culprit): 6.262 ns
- 0e04f95f1c39 (idx9): 6.412 ns
- da6081391e5d (head/compare): 6.290 ns

ALIGNMENT WARNING:
This bisection found a commit that does not appear related to the regression test. This suggests the performance change may be due to code alignment, binary layout, or other non-functional factors rather than a regression in the tested code path.

System.IO.Tests.Perf_StreamWriter.WriteString(writeLength: 2)

• Culprit: 65b0190679e7
• Commit Message: Avoid materializing Task/ValueTask in StreamReader/StreamWriter (Avoid materializing Task/ValueTask in StreamReader/StreamWriter #130689)
• Relevance: RELATED
• Explanation: The commit directly modifies System/IO/StreamWriter.cs (and StreamReader.cs), the exact class exercised by Perf_StreamWriter.WriteString. It reworks the async write path and the CheckAsyncTaskInProgress diagnostic, adding an _asyncIOInProgress bool and a try-finally IDisposable scope (GuardAgainstOtherWrites) inside WriteAsyncInternal, which is on the hot path the benchmark measures. There is a clear, plausible mechanism for the +6.3% write-path regression.

Helix jobs

Job ID	Submitted by bisection	Last status
e5fe8117-e60b-42ac-837a-bcc3f7635476	✅	submitted
1b466706-bc6f-4869-81e4-933e0cfbd519	✅	submitted

Per-commit measurements

Commit	Position	Measurements
ca073c0866e1	baseline, idx-1	78.890 ns
3d59a088c3bb	idx5, quarter	78.996 ns
fefaaf43570b	idx6	78.975 ns
0e04f95f1c39	idx9	83.614 ns
6b6ef2eae64e	idx10, mid	83.437 ns
1103cabc8eff	idx15, three-quarter	83.350 ns
55810a448212	idx7, last-good/culprit parent	79.124 ns
65b0190679e7	idx8, culprit	83.379 ns
da6081391e5d	idx19, head	83.341 ns

Bisection details

…[earlier output omitted]

└ {"index":null,"sha":"65b0190679e7afaea3c09bcfc474c55414403972","abbrevSha":"6...

Bisection complete. The culprit 65b0190 directly modifies StreamWriter.cs — the exact class the Perf_StreamWriter.WriteString test exercises. Measurements matched reported values closely and default config reproduced the regression.

BISECTION RESULT:
- Culprit Commit: 65b0190679e7afaea3c09bcfc474c55414403972
- Commit Message: Avoid materializing `Task`/`ValueTask` in `StreamReader`/`StreamWriter` (#130689)
- Commits Tested: 9
- Confidence: 5
- Commit Relevance: RELATED
- Relevance Explanation: The commit directly modifies System/IO/StreamWriter.cs (and StreamReader.cs), the exact class exercised by Perf_StreamWriter.WriteString. It reworks the async write path and the CheckAsyncTaskInProgress diagnostic, adding an `_asyncIOInProgress` bool and a try-finally IDisposable scope (GuardAgainstOtherWrites) inside WriteAsyncInternal, which is on the hot path the benchmark measures. There is a clear, plausible mechanism for the +6.3% write-path regression.
- Baseline Value: 79.124 ns
- Culprit Value: 83.379 ns
- Summary: A double-bisect over the 20-commit range located a clean step change: idx5 (3d59a08)=79.0 ns and idx7 (55810a4)=79.1 ns were good, while idx8 (65b0190)=83.4 ns and beyond were regressed. Narrowing commits idx6–10 pinned the single transition at 65b0190, whose good parent 55810a4 measured 79.1 ns.

HELIX JOBS:
- e5fe8117-e60b-42ac-837a-bcc3f7635476
- 1b466706-bc6f-4869-81e4-933e0cfbd519

MEASUREMENTS:
- ca073c0866e1 (baseline, idx-1): 78.890 ns
- 3d59a088c3bb (idx5, quarter): 78.996 ns
- fefaaf43570b (idx6): 78.975 ns
- 55810a448212 (idx7, last-good/culprit parent): 79.124 ns
- 65b0190679e7 (idx8, culprit): 83.379 ns
- 0e04f95f1c39 (idx9): 83.614 ns
- 6b6ef2eae64e (idx10, mid): 83.437 ns
- 1103cabc8eff (idx15, three-quarter): 83.350 ns
- da6081391e5d (idx19, head): 83.341 ns

System.Buffers.Tests.RentReturnArrayPoolTests.ProducerConsumer(RentalSize: 4096, ManipulateArray: True, Async: False, UseSharedPool: False)

• Status: inconclusive

Helix jobs

Job ID	Submitted by bisection	Last status
7cc48832-b060-41a2-94b8-8e03ab211257	✅	completed
4054c89b-8d4b-4ab8-a83a-5dba4aec7c49	✅	completed
5d41a01b-513c-47ec-a15c-e23475e8ed63	✅	completed
dd080515-18ec-4831-b97b-3c1097d59c59	✅	completed
b5b24dd2-f835-41bc-b60e-574df927e036	✅	completed
92c7441e-4090-48d6-8620-0015b6f73f21	✅	completed
7c6e80be-b169-4ddf-961e-de2b693578b1	✅	completed
03d29c71-c4d4-47a3-ac4b-e115e5896eba	✅	completed
7d278466-59ad-440b-90ff-3943919cbc51	✅	completed
ca4b6c3f-5c78-483e-a80e-773e41a1ee14	✅	completed
928d8a8d-f6ca-4ec3-8564-af2abb63e708	✅	completed
441c502e-dea6-4c07-bf10-ba66a397234f	✅	submitted
9a367283-e2b1-4f3b-adf0-880083cfc062	✅	submitted
b2ab5232-7833-4826-9b5f-89790be14f11	✅	submitted
82a3e2f4-1ccd-41f1-bcb9-bbe5e530b44c	✅	submitted
fba9e80c-5b2c-4965-9e66-2b87ba63fc3f	✅	submitted

Bisection details

…[earlier output omitted]

Notes for human triage (NOT a bisection result — unconfirmed by signal): Of the 20 commits, only four touch coreclr code that ships on x64/Windows and could plausibly affect this write-heavy ArrayPool path; the rest are wasm/mono/macOS/XML/JSON/config/test/tooling changes with no x64 impact. The plausible candidates, if a real regression exists, are:

• cea5f60b "Fix g_gc_lowest_address updating" (src/coreclr/gc/card_table.cpp) — GC write-barrier/card-table area, directly relevant since ManipulateArray=True dirties the rented array.
• c40bf120 "JIT: Remove boxing from value type instance calls" — broad JIT codegen change.
• 1103cabc "Fix GC.GetTotalMemory returning negative under regions GC" (gc/interface.cpp).
• 2177d805 "JIT: recompute exit weight in loop inversion" (optimizer.cpp).

HELIX JOBS:

• dd080515-18ec-4831-b97b-3c1097d59c59
• 4054c89b-8d4b-4ab8-a83a-5dba4aec7c49
• 5d41a01b-513c-47ec-a15c-e23475e8ed63
• b5b24dd2-f835-41bc-b60e-574df927e036
• 7cc48832-b060-41a2-94b8-8e03ab211257
• 03d29c71-c4d4-47a3-ac4b-e115e5896eba
• 7d278466-59ad-440b-90ff-3943919cbc51
• 92c7441e-4090-48d6-8620-0015b6f73f21
• 928d8a8d-f6ca-4ec3-8564-af2abb63e708
• 7c6e80be-b169-4ddf-961e-de2b693578b1
• ca4b6c3f-5c78-483e-a80e-773e41a1ee14
• fba9e80c-5b2c-4965-9e66-2b87ba63fc3f
• 441c502e-dea6-4c07-bf10-ba66a397234f
• 9a367283-e2b1-4f3b-adf0-880083cfc062
• 82a3e2f4-1ccd-41f1-bcb9-bbe5e530b44c
• b2ab523-7833-4826-9b5f-89790be14f11

MEASUREMENTS: (median ns; note endpoint overlap = no resolvable signal)

• ca073c0 (base/last-good, idx0): 1380 ; repeats 1452, 1782
• f561b0b (idx5): 1242
• fefaaf4 (idx7): 1455
• 65b0190 (idx9): 1630
• 0e04f95 (idx10): 1432
• 6b6ef2e (idx11): 1489
• 3aa40d2 (idx12): 1409 ; repeats 1824, 1356
• cea5f60 (idx13): 1791
• c40bf12 (idx14): 1419 ; repeats 1435, 1350
• 2177d80 (idx17): repeats 2316, 1262
• c6e9091 (idx15): 1542
• da60813 (head/regressed, idx20): 1583 ; repeats 1977, 1406

🤖 Proposed Next Actions

Rule: R4_known_regression_new — Known-cause regression — transfer & notify.

Verified regression attributed to commit 65b0190 (finding 5/5, regression 5/5, bisection RELATED). No prior linked autofile issue found.

#	Action	Status	Details
1	Back-reference cause PR	⏭️ Skipped (dry-run)	on culprit PR for 65b0190679e7
2	Transfer issue	⏭️ Skipped (dry-run)	→ dotnet/runtime
3	Ping owners	⏭️ Skipped (dry-run)	ping jakobbotsch

🦆 Pre-bisection Re-analysis (advisory)

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

Flags: unsupported-candidate

Rationale: The lone per-finding signal (Perf_StreamWriter.WriteString noise=False, changeConf=4/5) is a well-supported actionable regression and rightly queued for bisection, so the noise call is reasonable. The other two candidates (Perf_BigInteger.Equals and RentReturnArrayPoolTests) carry no per-finding signal in the packet, but there is likewise no evidence they are noise, so I do not push to drop them without their z-score/history.

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

🦆 Re-analysis (advisory)

⚠️ Disputes the analysis — advisory only, not applied

Flags: overconfident, overconfident-change, mixed-bisection-evidence

Rationale: The culprit is well-supported by the one RELATED bisection that converged cleanly (no alignment warning), so the SHA stands, but the packet's 5/5 finding and 5/5 regression scores overstate evidence whose own per-finding signal is only 4/5 and which includes one LIKELY_UNRELATED result with an alignment warning and one inconclusive bisection. Both scores should be lowered to 4.

Suggested confidence: Finding Confidence → 4/5; 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 2026-08-04 10:59 UTC

[DrewScoggins]

The only actual regression here is the StreamWriter test. The rest are just noise.

[dotnet-policy-service]

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

[jakobbotsch]

I still plan to take a look at this one after snap, so will leave it in .NET 11.