[performance] perf(SKRuntimeEffectUniforms): pre-size uniform lookup map to known count

[github-actions]

🤖 Generated by the Fixer - Performance agentic workflow using the performance-fixer skill.

What changed

SKRuntimeEffectUniforms's constructor now builds its internal Dictionary<string, Variable> with
an explicit capacity equal to the known uniform count:

- uniforms = new Dictionary<string, Variable> ();
+ uniforms = new Dictionary<string, Variable> (names.Length);

The constructor already knows the final entry count (names.Length) before the fill loop, so the
default-sized (capacity 0) map needlessly rehashes several times as it is populated, allocating
throwaway bucket/entry arrays. This is on a per-uniform-set (per-frame for animated shaders) path.

Invariant that keeps it correct: Dictionary capacity is a pure sizing hint — it changes only
the initial bucket allocation, never lookup/insert semantics or iteration results. The map still
holds exactly names.Length entries.

Proof it is faster

BenchmarkDotNet, [MemoryDiagnoser], net10.0, AMD EPYC 9V74, 2 runs. Old = default-sized map,
New = pre-sized; both fill N entries:

dotnet run -c Release --project benchmarks/SkiaSharp.Benchmarks -- --filter '*SKRuntimeEffectUniformsDictionaryBenchmark*'

N (uniforms)	Old Mean	New Mean	Ratio	Old Alloc	New Alloc	Alloc Ratio
4	119.7 ns	79.5 ns	0.67	704 B	496 B	0.70
16	372.4 ns	242.8 ns	0.65	1640 B	1016 B	0.62
48	1205.8 ns	689.0 ns	0.57	8296 B	3200 B	0.39

Repeatable across both runs; no allocation regression (fewer allocations from removed rehashes).
Benchmark added at benchmarks/SkiaSharp.Benchmarks/Benchmarks/SKRuntimeEffectUniformsDictionaryBenchmark.cs.

Proof it is identical

New equivalence test SKRuntimeEffectTest.Shaders.UniformsMapIsCompleteForManyUniforms builds a
24-uniform effect (crossing several dictionary rehash boundaries) and asserts:

• every declared uniform is listed, in declaration order, with the right Count;
• the packed uniform buffer is byte-identical whether values are written in forward or reverse
  order (proving each name maps to the correct offset regardless of insertion path);
• each packed float round-trips to its declared value.

Confirmed the test catches a deliberately-wrong result: truncating the constructor loop to
i < names.Length - 1 turns it RED; the shipped fix is GREEN. Full *Uniform* suite: 81 passed /
0 failed against pre-built natives (dotnet cake --target=externals-download) on net10.0.

Scope note

• Empirically measured, not merely statically reasoned.
• Managed C# only (binding/SkiaSharp/SKRuntimeEffect.cs); no *.generated.cs, no externals/skia.
• ABI impact: none — private field, public signatures unchanged.
• Driver of the win: removed rehash allocations → perf/allocations.

Fixes #4511

Generated by Fixer - Performance · ● 39.1M · ◷

[github-actions]

📖 Documentation Preview

The documentation for this PR has been deployed and is available at:

🔗 View Staging Site
🔗 View Staging Docs
🔗 View Staging Gallery (Blazor)
🔗 View Staging Gallery (Uno Platform)
🔗 View Staging SkiaFiddle

This preview will be updated automatically when you push new commits to this PR.

---

This comment is automatically updated by the documentation staging workflow.

[github-actions]

📊 SkiaSharp benchmarks — PR #4512

⭐ this PR (full source build) vs 🌙 nightly · Linux · Windows · macOS

Informational only — this never blocks the PR. 🟢 faster / less allocation · 🔴 slower / more allocation; moves under 5% are hidden as noise.

⏱️ Times are raw BenchmarkDotNet means, and the ⭐ PR and baseline legs run on separate CI runners, so microbenchmarks can swing run-to-run — treat small time deltas as noise. Allocations are deterministic and the reliable signal. The interactive perf-dashboard (linked below) applies smoothing for the trend view.

Highlights

⏱️ Time — 🔴 16 slower · 🟢 42 faster

• 🟢 LargeImageScaleBenchmark.UpscaleOpaque(Size: 1024) · macOS · 827.21 µs → 507.99 µs (-39%)
• 🟢 BitmapDrawBenchmark.DrawScaledTiles(Tiles: 256) · macOS · 1.07 ms → 669.01 µs (-37%)
• 🔴 RasterImageLifecycleBenchmark.CreateDataWithReleaseProc(Count: 256) · Windows · 95.00 µs → 129.66 µs (+36%)
• 🟢 CanvasDrawBenchmark.Draw(Shapes: 64) · macOS · 2.12 ms → 1.35 ms (-36%)
• 🟢 LargeImageScaleBenchmark.UpscaleCrossfade(Size: 1024) · macOS · 1.79 ms → 1.24 ms (-31%)
• …and 53 more (see details below)

Full per-OS benchmark deltas

Linux

⏱️ Time (vs 🌙 nightly 4.151.0-nightly.83)

Benchmark	baseline	this PR	Δ
RasterImageLifecycleBenchmark.CreateDataWithReleaseProc(Count: 256)	100.78 µs	120.98 µs	🔴 +20%
ColorMathBenchmark.ToColor(Colors: 4096)	16.21 µs	14.34 µs	🟢 -12%
MatrixMapPointsBenchmark.MapRect(Points: 256)	11.93 µs	11.12 µs	🟢 -7%
RasterImageLifecycleBenchmark.CreateRasterImage(Count: 256)	257.50 µs	273.19 µs	🔴 +6%
MatrixMapPointsBenchmark.MapPoints(Points: 256)	208.1 ns	220.2 ns	🔴 +6%

Windows

⏱️ Time (vs 🌙 nightly 4.151.0-nightly.83)

Benchmark	baseline	this PR	Δ
RasterImageLifecycleBenchmark.CreateDataWithReleaseProc(Count: 256)	95.00 µs	129.66 µs	🔴 +36%
ColorMathBenchmark.ToColor(Colors: 4096)	13.48 µs	17.06 µs	🔴 +27%
PathBoundsBenchmark.TightBounds(Points: 64)	1.53 µs	1.86 µs	🔴 +21%
PathBoundsBenchmark.TightBounds(Points: 1024)	16.41 µs	19.87 µs	🔴 +21%
RuntimeEffectShaderBenchmark.DrawFrame	1.22 ms	1.45 ms	🔴 +18%
RasterImageLifecycleBenchmark.CreateRasterImage(Count: 256)	253.92 µs	284.61 µs	🔴 +12%
MatrixMapPointsBenchmark.MapPoint(Points: 256)	5.51 µs	4.90 µs	🟢 -11%
MatrixMapPointsBenchmark.MapPoints(Points: 4096)	2.02 µs	2.24 µs	🔴 +11%
CanvasDrawBenchmark.Draw(Shapes: 64)	2.58 ms	2.86 ms	🔴 +11%
CanvasDrawBenchmark.Draw(Shapes: 512)	20.68 ms	22.89 ms	🔴 +11%
MatrixOpsBenchmark.Invert(Count: 4096)	149.85 µs	134.08 µs	🟢 -11%
LargeImageScaleBenchmark.UpscaleOpaque(Size: 2048)	3.31 ms	2.97 ms	🟢 -10%
LargeImageScaleBenchmark.UpscaleCrossfade(Size: 2048)	11.35 ms	10.25 ms	🟢 -10%
ColorMathBenchmark.UnPreMultiply(Colors: 4096)	12.80 µs	11.61 µs	🟢 -9%
ColorMathBenchmark.PreMultiply(Colors: 4096)	14.70 µs	13.37 µs	🟢 -9%
BitmapDrawBenchmark.DrawScaledTiles(Tiles: 256)	1.32 ms	1.42 ms	🔴 +8%
ColorParseBenchmark.Parse(Iterations: 1000)	78.09 µs	83.90 µs	🔴 +7%
LargeImageScaleBenchmark.UpscaleOpaque(Size: 1024)	819.07 µs	760.08 µs	🟢 -7%
MatrixMapPointsBenchmark.MapRadius(Points: 4096)	155.35 µs	165.65 µs	🔴 +7%
LargeImageScaleBenchmark.UpscaleCrossfade(Size: 1024)	2.82 ms	2.63 ms	🟢 -7%
MatrixMapPointsBenchmark.MapPoint(Points: 4096)	85.91 µs	80.45 µs	🟢 -6%
MatrixMapPointsBenchmark.MapRect(Points: 256)	8.12 µs	8.54 µs	🔴 +5%

macOS

⏱️ Time (vs 🌙 nightly 4.151.0-nightly.83)

Benchmark	baseline	this PR	Δ
LargeImageScaleBenchmark.UpscaleOpaque(Size: 1024)	827.21 µs	507.99 µs	🟢 -39%
BitmapDrawBenchmark.DrawScaledTiles(Tiles: 256)	1.07 ms	669.01 µs	🟢 -37%
CanvasDrawBenchmark.Draw(Shapes: 64)	2.12 ms	1.35 ms	🟢 -36%
LargeImageScaleBenchmark.UpscaleCrossfade(Size: 1024)	1.79 ms	1.24 ms	🟢 -31%
ColorMathBenchmark.ToColor(Colors: 4096)	23.03 µs	16.07 µs	🟢 -30%
BitmapDrawBenchmark.DrawUnscaledTiles(Tiles: 64)	221.47 µs	157.78 µs	🟢 -29%
BitmapDrawBenchmark.DrawScaledTiles(Tiles: 64)	288.50 µs	207.07 µs	🟢 -28%
BitmapDrawBenchmark.DrawUnscaledTiles(Tiles: 256)	738.94 µs	541.26 µs	🟢 -27%
LargeImageScaleBenchmark.UpscaleOpaque(Size: 2048)	3.27 ms	2.46 ms	🟢 -25%
SceneRenderBenchmark.RenderFrame(Complexity: 1)	1.52 ms	1.18 ms	🟢 -23%
MatrixOpsBenchmark.Invert(Count: 4096)	79.15 µs	62.09 µs	🟢 -22%
LargeImageScaleBenchmark.UpscaleCrossfade(Size: 2048)	6.97 ms	5.53 ms	🟢 -21%
ColorMathBenchmark.ToColorF(Colors: 4096)	6.16 µs	4.90 µs	🟢 -21%
RuntimeEffectShaderBenchmark.DrawFrame	660.79 µs	529.40 µs	🟢 -20%
CanvasDrawBenchmark.Draw(Shapes: 512)	13.49 ms	10.90 ms	🟢 -19%
SceneRenderBenchmark.RenderFrame(Complexity: 4)	6.76 ms	5.58 ms	🟢 -18%
ColorMathBenchmark.UnPreMultiply(Colors: 4096)	8.91 µs	7.43 µs	🟢 -17%
ColorMathBenchmark.PreMultiply(Colors: 4096)	9.83 µs	8.22 µs	🟢 -16%
RasterImageLifecycleBenchmark.CreateDataWithReleaseProc(Count: 256)	90.56 µs	76.69 µs	🟢 -15%
RasterImageLifecycleBenchmark.CreateRasterImage(Count: 256)	184.84 µs	157.40 µs	🟢 -15%
PathBoundsBenchmark.TightBounds(Points: 1024)	10.21 µs	8.80 µs	🟢 -14%
MatrixOpsBenchmark.Concat(Count: 4096)	141.36 µs	122.08 µs	🟢 -14%
MatrixMapPointsBenchmark.MapRect(Points: 4096)	58.15 µs	51.37 µs	🟢 -12%
MatrixMapPointsBenchmark.MapRadius(Points: 256)	7.39 µs	6.54 µs	🟢 -11%
ColorParseBenchmark.Parse(Iterations: 1000)	74.27 µs	66.03 µs	🟢 -11%
…and 6 more

📈 Full interactive perf-dashboard & run details →