Skip to content

Flat barebones - #574

Open
dadhi wants to merge 19 commits into
masterfrom
flat-barebones
Open

Flat barebones#574
dadhi wants to merge 19 commits into
masterfrom
flat-barebones

Conversation

@dadhi

@dadhi dadhi commented Nov 2, 2026

Copy link
Copy Markdown
Owner

No description provided.

@github-actions

github-actions Bot commented Nov 2, 2026 •
edited
Loading

Copy link
Copy Markdown

Outdated: superseded by the newer performance report.

Previous measurements

net10.0: Potential regressions detected (reporting only)

Base e82c9500e629 → PR e203ba378540; SDK 10.0.301; harness e82c9500e629 (base).

Backend / phase / workload Base ns PR ns Time change B/op: base → PR Assessment
Light / Compile / Arithmetic 7874.10 7582.09 -3.7% 928.00 → 928.00 Improvement
Light / Compile / CallConditional 4476.62 4323.14 -3.4% 1373.00 → 1372.50 Within variation
Light / Compile / Closure 9210.74 9330.06 +1.6% 3042.50 → 3037.00 Within variation
Light / Compile / Scalar 2987.85 2996.98 +0.4% 888.50 → 890.00 Within variation
Light / CompileAndFirstInvoke / Arithmetic 74463.07 74296.42 -0.2% 928.00 → 928.00 Within variation
Light / CompileAndFirstInvoke / CallConditional 135335.74 134599.65 -0.5% 1360.00 → 1360.00 Within variation
Light / CompileAndFirstInvoke / Closure 333198.69 331956.85 -0.4% 3185.00 → 3185.00 Within variation
Light / CompileAndFirstInvoke / Scalar 25918.79 25822.59 -0.4% 872.00 → 872.00 Within variation
Light / Construct / Arithmetic 1075.31 1072.10 -0.3% 1296.00 → 1296.00 Within variation
Light / Construct / CallConditional 226.80 227.91 +0.5% 360.00 → 360.00 Within variation
Light / Construct / Closure 140.54 138.95 -1.1% 144.00 → 144.00 Within variation
Light / Construct / Scalar 97.19 95.34 -1.9% 120.00 → 120.00 Within variation
Light / Invoke / Arithmetic 2.21 2.52 +13.9% 0.00 → 0.00 Regression (both rounds)
Light / Invoke / CallConditional 2.40 2.52 +5.3% 0.00 → 0.00 Warning / noise
Light / Invoke / Closure 44.63 43.80 -1.8% 184.00 → 184.00 Within variation
Light / Invoke / Scalar 2.21 1.89 -14.3% 0.00 → 0.00 Improvement
Standard / Compile / Arithmetic 7979.93 7732.95 -3.1% 928.00 → 928.00 Within variation
Standard / Compile / CallConditional 4621.20 4382.45 -5.2% 1373.50 → 1373.50 Within variation
Standard / Compile / Closure 10271.79 9673.54 -5.8% 3028.00 → 3040.50 Allocation increase
Standard / Compile / Scalar 2908.00 2851.56 -0.7% 884.50 → 884.50 Within variation
Standard / CompileAndFirstInvoke / Arithmetic 76690.95 75757.72 -1.2% 928.00 → 928.00 Within variation
Standard / CompileAndFirstInvoke / CallConditional 136836.30 137127.87 +0.2% 1360.00 → 1360.00 Within variation
Standard / CompileAndFirstInvoke / Closure 336623.49 336065.26 -0.2% 3185.00 → 3185.00 Within variation
Standard / CompileAndFirstInvoke / Scalar 28630.48 28233.77 -1.4% 872.00 → 872.00 Within variation
Standard / Construct / Arithmetic 1649.82 1636.50 -0.8% 1808.00 → 1808.00 Within variation
Standard / Construct / CallConditional 887.04 874.48 -1.1% 680.00 → 680.00 Within variation
Standard / Construct / Closure 460.75 466.20 +1.2% 440.00 → 440.00 Within variation
Standard / Construct / Scalar 344.85 362.58 +5.2% 416.00 → 416.00 Warning / noise
Standard / Invoke / Arithmetic 2.52 2.52 +0.0% 0.00 → 0.00 Within variation
Standard / Invoke / CallConditional 2.20 2.52 +14.3% 0.00 → 0.00 Regression (both rounds)
Standard / Invoke / Closure 44.32 42.91 -3.1% 184.00 → 184.00 Within variation
Standard / Invoke / Scalar 2.20 1.89 -14.1% 0.00 → 0.00 Improvement

Two opposite-order process rounds on one GitHub-hosted runner; BDN confidence intervals must be separated in both rounds to classify a >5% time regression. A ≥3% change without that evidence is a warning, not proof. Five measurement iterations/round; hosted-runner noise remains.

Compile excludes delegate invocation; CompileAndFirstInvoke includes the first call/JIT of a new delegate. Invoke is warmed; closure invocation includes creating and calling the returned delegate. MemoryDiagnoser reports managed B/op. Loop/checksum overhead is shared by both revisions.

Workflow run and complete raw artifacts

@github-actions

github-actions Bot commented Nov 2, 2026 •
edited
Loading

Copy link
Copy Markdown

Outdated: superseded by the newer performance report.

Previous measurements

net10.0: Potential regressions detected (reporting only)

Base e82c9500e629 → PR 083caf4137a3; SDK 10.0.301; harness e82c9500e629 (base).

Backend / phase / workload Base ns PR ns Time change B/op: base → PR Assessment
Light / Compile / Arithmetic 7828.17 7434.46 -5.0% 928.00 → 928.00 Improvement
Light / Compile / CallConditional 4317.25 4212.07 -2.4% 1373.50 → 1372.50 Within variation
Light / Compile / Closure 9227.39 9183.73 -0.3% 3035.00 → 3041.50 Within variation
Light / Compile / Scalar 2990.31 2950.42 -1.3% 888.50 → 886.50 Within variation
Light / CompileAndFirstInvoke / Arithmetic 75239.27 74269.92 -1.3% 928.00 → 928.00 Within variation
Light / CompileAndFirstInvoke / CallConditional 134369.80 133220.58 -0.9% 1360.00 → 1360.00 Within variation
Light / CompileAndFirstInvoke / Closure 330803.96 332650.79 +0.6% 3185.00 → 3185.00 Within variation
Light / CompileAndFirstInvoke / Scalar 26051.71 26236.61 +0.7% 872.00 → 872.00 Within variation
Light / Construct / Arithmetic 1054.57 1058.82 +0.4% 1296.00 → 1296.00 Within variation
Light / Construct / CallConditional 225.72 227.63 +0.9% 360.00 → 360.00 Within variation
Light / Construct / Closure 140.08 140.12 +0.0% 144.00 → 144.00 Within variation
Light / Construct / Scalar 95.91 98.35 +2.6% 120.00 → 120.00 Within variation
Light / Invoke / Arithmetic 2.21 2.52 +14.0% 0.00 → 0.00 Regression (both rounds)
Light / Invoke / CallConditional 2.40 2.52 +5.4% 0.00 → 0.00 Warning / noise
Light / Invoke / Closure 41.90 42.41 +1.2% 184.00 → 184.00 Within variation
Light / Invoke / Scalar 2.21 1.89 -14.3% 0.00 → 0.00 Improvement
Standard / Compile / Arithmetic 8064.49 7777.08 -3.6% 928.00 → 928.00 Within variation
Standard / Compile / CallConditional 4470.31 4400.23 -1.6% 1373.00 → 1373.00 Within variation
Standard / Compile / Closure 9688.26 9459.92 -2.4% 3034.00 → 3041.00 Allocation increase
Standard / Compile / Scalar 2839.24 2834.12 -0.2% 884.50 → 884.50 Within variation
Standard / CompileAndFirstInvoke / Arithmetic 75311.85 75853.88 +0.7% 928.00 → 928.00 Within variation
Standard / CompileAndFirstInvoke / CallConditional 137313.78 135351.32 -1.4% 1360.00 → 1360.00 Within variation
Standard / CompileAndFirstInvoke / Closure 331816.20 331788.06 -0.0% 3185.00 → 3185.00 Within variation
Standard / CompileAndFirstInvoke / Scalar 27905.31 27165.29 -2.6% 872.00 → 872.00 Within variation
Standard / Construct / Arithmetic 1635.21 1620.84 -0.9% 1808.00 → 1808.00 Within variation
Standard / Construct / CallConditional 849.01 870.85 +2.6% 680.00 → 680.00 Within variation
Standard / Construct / Closure 447.48 451.44 +0.9% 440.00 → 440.00 Within variation
Standard / Construct / Scalar 341.68 341.83 +0.0% 416.00 → 416.00 Within variation
Standard / Invoke / Arithmetic 2.21 2.52 +13.9% 0.00 → 0.00 Regression (both rounds)
Standard / Invoke / CallConditional 2.21 2.52 +14.1% 0.00 → 0.00 Regression (both rounds)
Standard / Invoke / Closure 42.07 42.02 -0.1% 184.00 → 184.00 Within variation
Standard / Invoke / Scalar 2.20 1.89 -14.2% 0.00 → 0.00 Improvement

Two opposite-order process rounds on one GitHub-hosted runner; BDN confidence intervals must be separated in both rounds to classify a >5% time regression. A ≥3% change without that evidence is a warning, not proof. Five measurement iterations/round; hosted-runner noise remains.

Compile excludes delegate invocation; CompileAndFirstInvoke includes the first call/JIT of a new delegate. Invoke is warmed; closure invocation includes creating and calling the returned delegate. MemoryDiagnoser reports managed B/op. Loop/checksum overhead is shared by both revisions.

Workflow run and complete raw artifacts

@github-actions

github-actions Bot commented Nov 2, 2026 •
edited
Loading

Copy link
Copy Markdown

Outdated: superseded by the newer performance report.

Previous measurements

net10.0: Potential regressions detected (reporting only)

Base e82c9500e629 → PR eb98cf36013d; SDK 10.0.301; harness e82c9500e629 (base).

Backend / phase / workload Base ns PR ns Time change B/op: base → PR Assessment
Light / Compile / Arithmetic 7877.16 7465.96 -5.2% 928.00 → 928.00 Improvement
Light / Compile / CallConditional 4407.90 4314.69 -1.9% 1373.00 → 1372.50 Within variation
Light / Compile / Closure 9501.65 9510.13 +0.4% 3037.00 → 3042.00 Within variation
Light / Compile / Scalar 3021.43 3040.39 +0.9% 885.00 → 887.50 Allocation increase
Light / CompileAndFirstInvoke / Arithmetic 74886.04 74674.62 -0.3% 928.00 → 928.00 Within variation
Light / CompileAndFirstInvoke / CallConditional 135946.96 134986.55 -0.7% 1360.00 → 1360.00 Within variation
Light / CompileAndFirstInvoke / Closure 332279.86 333240.95 +0.3% 3185.00 → 3185.00 Within variation
Light / CompileAndFirstInvoke / Scalar 25977.20 26717.44 +2.9% 872.00 → 872.00 Within variation
Light / Construct / Arithmetic 1061.02 1073.25 +1.2% 1296.00 → 1296.00 Within variation
Light / Construct / CallConditional 229.51 231.23 +0.8% 360.00 → 360.00 Within variation
Light / Construct / Closure 141.45 148.08 +4.7% 144.00 → 144.00 Warning / noise
Light / Construct / Scalar 96.94 98.03 +1.2% 120.00 → 120.00 Within variation
Light / Invoke / Arithmetic 2.21 2.52 +14.0% 0.00 → 0.00 Regression (both rounds)
Light / Invoke / CallConditional 2.24 2.36 +5.5% 0.00 → 0.00 Warning / noise
Light / Invoke / Closure 45.81 44.77 -2.2% 184.00 → 184.00 Within variation
Light / Invoke / Scalar 2.21 1.89 -14.2% 0.00 → 0.00 Improvement
Standard / Compile / Arithmetic 8139.10 7667.86 -5.8% 928.00 → 928.00 Improvement
Standard / Compile / CallConditional 4680.09 4423.18 -5.4% 1373.50 → 1373.00 Within variation
Standard / Compile / Closure 9290.23 9882.35 +6.6% 3030.00 → 3028.00 Warning / noise
Standard / Compile / Scalar 2908.01 2719.34 -6.3% 885.00 → 885.00 Within variation
Standard / CompileAndFirstInvoke / Arithmetic 75796.32 74925.19 -1.1% 928.00 → 928.00 Within variation
Standard / CompileAndFirstInvoke / CallConditional 137620.94 138628.07 +0.8% 1360.00 → 1360.00 Within variation
Standard / CompileAndFirstInvoke / Closure 334251.65 334654.49 +0.1% 3185.00 → 3185.00 Within variation
Standard / CompileAndFirstInvoke / Scalar 28010.87 28038.03 +0.1% 872.00 → 872.00 Within variation
Standard / Construct / Arithmetic 1733.22 1716.01 -1.0% 1808.00 → 1808.00 Within variation
Standard / Construct / CallConditional 893.22 1019.84 +14.0% 680.00 → 752.00 Warning / noise
Standard / Construct / Closure 455.96 451.61 -0.9% 440.00 → 440.00 Within variation
Standard / Construct / Scalar 340.58 391.20 +14.7% 416.00 → 416.00 Warning / noise
Standard / Invoke / Arithmetic 2.40 2.52 +5.8% 0.00 → 0.00 Warning / noise
Standard / Invoke / CallConditional 2.21 2.52 +14.2% 0.00 → 0.00 Regression (both rounds)
Standard / Invoke / Closure 44.35 43.76 -1.2% 184.00 → 184.00 Within variation
Standard / Invoke / Scalar 2.21 1.89 -14.3% 0.00 → 0.00 Improvement

Two opposite-order process rounds on one GitHub-hosted runner; BDN confidence intervals must be separated in both rounds to classify a >5% time regression. A ≥3% change without that evidence is a warning, not proof. Five measurement iterations/round; hosted-runner noise remains.

Compile excludes delegate invocation; CompileAndFirstInvoke includes the first call/JIT of a new delegate. Invoke is warmed; closure invocation includes creating and calling the returned delegate. MemoryDiagnoser reports managed B/op. Loop/checksum overhead is shared by both revisions.

Workflow run and complete raw artifacts

dadhi and others added 16 commits November 2, 2026 22:36
@dadhi
FlatExpression: drop node-collection lists and closure-usage tracking…
cdeee2c 
…, require non-empty leading children

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat expr: explicit presence flags for Switch default/Try finally, Cs…
1adf731 
…Check fuzzer, fix SwitchCase, Block Eq/Hash and WithOneChild stale ref

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Experimental flat-native collect and emit path with comparison test
1e87f0c 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native emit: pop block scope, use body type as the return type, …
3972284 
…test bool and fallback

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Direct Flat -> Light conversion without System; ideas notes in .scratch
f79d569 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Direct Light -> Flat; fix Light loop dropping the continue label with…
e7754c6 
…out a break label

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat: README expressions test and timing, notes
38ef227 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native compile: New, Call, members, reference types, closure con…
a9ac2f1 
…stants

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native collect: drop per-node argument type checks, match Light
1af09c2 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native compile: nested lambdas and captured variables
9fa8fc4 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native compile: construction-time lambda count and assigned para…
94e9a50 
…ms, by-value captures, delegate info cache

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native: MemberInit, NewArrayInit, ArrayIndex, Convert; drop via-…
78a982c 
…System from prod; README and Complex tests with alloc numbers

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native: per-thread compile context, cached StrongBox members (-1…
fb96544 
…76 B per compile)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native: reuse FEC pooled ILGenerator (-240 B per compile)
f1fbf6e 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Flat native: fold constant Func<bool>, FEC local var indexes, create+…
b0f4d14 
…compile benchmark rows

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Fix the test for the TryCollectInfo scopeIdx API change
03276f6 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
dadhi and others added 2 commits November 2, 2026 22:40
@dadhi
Flat review: reject value-type instances natively, short constant IL,…
ac5873d 
… perf todos

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@dadhi
Fuzz the native flat compile: closures and objects/members
c11856d 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@github-actions

github-actions Bot commented Nov 2, 2026 •
edited
Loading

Copy link
Copy Markdown

Outdated: superseded by the newer performance report.

Previous measurements

net10.0: Potential regressions detected (reporting only)

Base 76c9f3f18e1b → PR c11856d3b6ad; SDK 10.0.301; harness 76c9f3f18e1b (base).

Backend / phase / workload Base ns PR ns Time change B/op: base → PR Assessment
Light / Compile / Arithmetic 7788.70 7761.45 -0.4% 928.00 → 928.00 Within variation
Light / Compile / CallConditional 4225.46 4152.26 -1.7% 1372.00 → 1372.00 Within variation
Light / Compile / Closure 9129.37 9192.20 +0.7% 3041.00 → 3043.00 Within variation
Light / Compile / Scalar 3011.31 2908.41 -3.4% 889.50 → 890.00 Within variation
Light / CompileAndFirstInvoke / Arithmetic 73273.17 73661.13 +0.5% 928.00 → 928.00 Within variation
Light / CompileAndFirstInvoke / CallConditional 133477.57 133400.33 -0.1% 1360.00 → 1360.00 Within variation
Light / CompileAndFirstInvoke / Closure 328417.78 330121.05 +0.5% 3185.00 → 3185.00 Within variation
Light / CompileAndFirstInvoke / Scalar 25569.61 25450.14 -0.5% 872.00 → 872.00 Within variation
Light / Construct / Arithmetic 1049.76 1034.53 -1.4% 1296.00 → 1296.00 Within variation
Light / Construct / CallConditional 218.75 219.90 +0.5% 360.00 → 360.00 Within variation
Light / Construct / Closure 137.74 135.32 -1.8% 144.00 → 144.00 Within variation
Light / Construct / Scalar 93.25 93.63 +0.4% 120.00 → 120.00 Within variation
Light / Invoke / Arithmetic 2.21 2.21 +0.0% 0.00 → 0.00 Within variation
Light / Invoke / CallConditional 2.24 2.51 +12.4% 0.00 → 0.00 Regression (both rounds)
Light / Invoke / Closure 39.88 40.63 +1.9% 184.00 → 184.00 Within variation
Light / Invoke / Scalar 1.89 1.89 +0.0% 0.00 → 0.00 Within variation
Standard / Compile / Arithmetic 8116.70 7993.35 -1.5% 928.00 → 928.00 Within variation
Standard / Compile / CallConditional 4364.16 4506.87 +3.3% 1373.50 → 1373.00 Warning / noise
Standard / Compile / Closure 9542.42 9615.20 +0.8% 3040.50 → 3041.00 Within variation
Standard / Compile / Scalar 2808.59 2839.67 +1.2% 884.50 → 884.00 Within variation
Standard / CompileAndFirstInvoke / Arithmetic 74693.53 74932.92 +0.3% 928.00 → 928.00 Within variation
Standard / CompileAndFirstInvoke / CallConditional 134272.29 134493.13 +0.2% 1360.00 → 1360.00 Within variation
Standard / CompileAndFirstInvoke / Closure 331643.10 330540.39 -0.3% 3185.00 → 3185.00 Within variation
Standard / CompileAndFirstInvoke / Scalar 27556.29 27267.39 -1.0% 872.00 → 872.00 Within variation
Standard / Construct / Arithmetic 1591.79 1635.88 +2.8% 1808.00 → 1808.00 Within variation
Standard / Construct / CallConditional 856.52 863.01 +0.9% 680.00 → 680.00 Within variation
Standard / Construct / Closure 452.84 448.81 -0.9% 440.00 → 440.00 Within variation
Standard / Construct / Scalar 343.65 342.13 -0.4% 416.00 → 416.00 Within variation
Standard / Invoke / Arithmetic 2.52 2.52 +0.1% 0.00 → 0.00 Within variation
Standard / Invoke / CallConditional 2.20 2.21 +0.0% 0.00 → 0.00 Within variation
Standard / Invoke / Closure 40.24 40.56 +0.8% 184.00 → 184.00 Within variation
Standard / Invoke / Scalar 2.21 2.21 +0.0% 0.00 → 0.00 Within variation

Two opposite-order process rounds on one GitHub-hosted runner; BDN confidence intervals must be separated in both rounds to classify a >5% time regression. A ≥3% change without that evidence is a warning, not proof. Five measurement iterations/round; hosted-runner noise remains.

Compile excludes delegate invocation; CompileAndFirstInvoke includes the first call/JIT of a new delegate. Invoke is warmed; closure invocation includes creating and calling the returned delegate. MemoryDiagnoser reports managed B/op. Loop/checksum overhead is shared by both revisions.

Workflow run and complete raw artifacts

@dadhi
FlatCtx: stack struct passed by ref instead of the thread-static class
39648cc 
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
@github-actions

github-actions Bot commented Nov 3, 2026

Copy link
Copy Markdown

net10.0: Comparison complete (reporting only)

Base 76c9f3f18e1b -> PR 39648cc37499; SDK 10.0.301; harness 76c9f3f18e1b (base).

Backend / phase / workload Base ns PR ns Time change B/op: base -> PR Assessment
Light / Compile / Arithmetic 7230.03 7075.42 -2.1% 928.00 -> 928.00 Within variation
Light / Compile / CallConditional 2797.07 2809.03 +0.4% 1360.00 -> 1360.00 Within variation
Light / Compile / Closure 5267.84 5206.29 -1.2% 3000.00 -> 3000.00 Within variation
Light / Compile / Scalar 1630.15 1614.98 -0.9% 872.00 -> 872.00 Within variation
Light / CompileAndFirstInvoke / Arithmetic 51974.94 51792.40 -0.3% 928.00 -> 928.00 Within variation
Light / CompileAndFirstInvoke / CallConditional 54842.84 56402.47 +2.8% 1360.00 -> 1360.00 Within variation
Light / CompileAndFirstInvoke / Closure 248398.30 244320.45 -1.6% 3185.00 -> 3185.00 Within variation
Light / CompileAndFirstInvoke / Scalar 12104.82 12364.90 +2.1% 872.00 -> 872.00 Within variation
Light / Construct / Arithmetic 1052.16 1054.10 +0.2% 1296.00 -> 1296.00 Within variation
Light / Construct / CallConditional 240.83 228.77 -4.8% 360.00 -> 360.00 Within variation
Light / Construct / Closure 144.58 143.74 -0.6% 144.00 -> 144.00 Within variation
Light / Construct / Scalar 104.06 103.25 -0.8% 120.00 -> 120.00 Within variation
Light / Invoke / Arithmetic 2.20 2.29 +4.2% 0.00 -> 0.00 Warning / noise
Light / Invoke / CallConditional 2.71 3.01 +10.8% 0.00 -> 0.00 Warning / noise
Light / Invoke / Closure 54.71 53.65 -1.9% 184.00 -> 184.00 Within variation
Light / Invoke / Scalar 2.38 2.04 -14.4% 0.00 -> 0.00 Improvement
Standard / Compile / Arithmetic 7431.11 7475.18 +0.6% 928.00 -> 928.00 Within variation
Standard / Compile / CallConditional 3043.91 3038.31 -0.2% 1360.00 -> 1360.00 Within variation
Standard / Compile / Closure 5538.68 5588.06 +0.9% 3000.00 -> 3000.00 Within variation
Standard / Compile / Scalar 1783.47 1804.90 +1.2% 872.00 -> 872.00 Within variation
Standard / CompileAndFirstInvoke / Arithmetic 51898.92 51605.65 -0.6% 928.00 -> 928.00 Within variation
Standard / CompileAndFirstInvoke / CallConditional 54937.61 54528.62 -0.7% 1360.00 -> 1360.00 Within variation
Standard / CompileAndFirstInvoke / Closure 252634.42 249869.02 -1.1% 3184.00 -> 3184.00 Within variation
Standard / CompileAndFirstInvoke / Scalar 12535.36 12854.27 +2.6% 872.00 -> 872.00 Within variation
Standard / Construct / Arithmetic 1683.23 1667.21 -1.0% 1808.00 -> 1808.00 Within variation
Standard / Construct / CallConditional 780.03 780.66 +0.1% 680.00 -> 680.00 Within variation
Standard / Construct / Closure 483.06 474.21 -1.8% 440.00 -> 440.00 Within variation
Standard / Construct / Scalar 363.88 367.15 +0.9% 416.00 -> 416.00 Within variation
Standard / Invoke / Arithmetic 2.08 2.08 -0.0% 0.00 -> 0.00 Within variation
Standard / Invoke / CallConditional 3.03 3.08 +1.6% 0.00 -> 0.00 Within variation
Standard / Invoke / Closure 52.80 60.53 +14.6% 184.00 -> 184.00 Warning / noise
Standard / Invoke / Scalar 2.38 2.37 -0.2% 0.00 -> 0.00 Within variation

Two opposite-order process rounds on one GitHub-hosted runner; BDN confidence intervals must be separated in both rounds to classify a >5% time regression. A >=3% change without that evidence is a warning, not proof. Five measurement iterations/round; hosted-runner noise remains.

Compile excludes delegate invocation; CompileAndFirstInvoke includes the first call/JIT of a new delegate. Invoke is warmed; closure invocation includes creating and calling the returned delegate. MemoryDiagnoser reports managed B/op. Loop/checksum overhead is shared by both revisions.

Workflow run and complete raw artifacts

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@dadhi