Profiler-driven performance optimization: CPU profiling, hotspot fixes, and new benchmarks

[Copilot]

Adds CPU profiling to benchmarks, runs two full profile→fix iterations, and extends benchmark coverage to previously untested APIs. All 997 tests pass; CodeQL clean.

CPU Profiling Setup

• Added [EventPipeProfiler(EventPipeProfile.CpuSampling)] to both ConstrainedSwedenBenchmarks and SmallDatasetBenchmarks — produces .nettrace/.speedscope.json alongside BDN results

Optimization Candidates A–E (initial pass)

• A (Types.cs): Edge.GetHashCode — kept as HashCode.Combine(V1, V2) for collision-resistance in HashSet<Edge> / Dictionary<Edge,...>
• B (Triangulation.cs): Lift polyL/polyR/outerTris/intersected out of InsertEdgeIteration — allocate once in InsertEdges, pass as parameters and Clear() per call
• C (Triangulation.cs): InsertVertexInsideTriangle / InsertVertexOnEdge changed to accept a caller-owned Stack<int> (cleared at entry). New InsertVertex(int, int, Stack<int>) and InsertVertex(int, Stack<int>) overloads let all three bulk insertion loops (KDTreeBFS, Randomized, AsProvided) allocate one stack per batch and thread it through every vertex — eliminates O(N) per-vertex stack allocations
• D (KdTree.cs): Cache T.One + T.One as private readonly T _two in both constructors; GetMid changed to instance method
• E (Triangulation.cs): NthElement made generic over TComparer : struct, IComparer<int>; two readonly struct comparers replace the capturing lambdas in InsertVertices_KDTreeBFS — zero delegate allocations

Hotspot Fixes F–G (first profile round)

Profile showed List<Triangle>.AddWithResize at 4.9–11.6% and IsFlipNeeded at 9–20%.

• F (Triangulation.cs): Pre-allocate _triangles, _vertices, _vertTris in InsertVertices on first insertion using Euler's formula (~2N triangles for N points):

  _triangles.EnsureCapacity(2 * n + 4);
  _vertices.EnsureCapacity(n + Indices.SuperTriangleVertexCount);
  _vertTris.EnsureCapacity(n + Indices.SuperTriangleVertexCount);

• G (Triangulation.cs): Fast path in IsFlipNeeded: skip _fixedEdges.Contains entirely when _fixedEdges.Count == 0 (pure vertex-insertion path). Original branch structure and orientation predicates are preserved exactly.

Hotspot Fixes H–I (second profile round, all APIs)

• H (Triangulation.cs): _two = T.One + T.One cached as readonly field on Triangulation<T>; removes recomputation in AddSuperTriangle and ConformToEdgeIteration
• I (Triangulation.cs): Thread a single reusable Stack<int> + List<Edge> from ConformToEdges through ConformToEdgeIteration → InsertVertex_FlipFixedEdges — eliminates O(N) per-midpoint-insertion allocations in the conforming path

New Benchmarks

Added to cover previously untested APIs:

Class	New benchmarks
ConstrainedSwedenBenchmarks	Conforming–AsProvided, EraseSuperTriangle–Auto, EraseOuterTriangles–Auto
SmallDatasetBenchmarks	Small–ConformingAuto, Small–EraseSuperTriangle, Small–EraseOuterTrianglesAndHoles

Benchmark Delta (ConstrainedSweden, ~2 600 verts / ~2 600 edges)

Benchmark	Before	After	Δ Time	Δ Alloc
Vertices only – Auto	2.20 ms	1.89 ms	−14%	863 KB → 492 KB −43%
Vertices only – AsProvided	4.12 ms	3.66 ms	−11%	995 KB → 625 KB −37%
Constrained – Auto	4.94 ms	4.71 ms	−5%	1015 KB → 644 KB −37%
Constrained – AsProvided	6.79 ms	6.45 ms	−5%	1147 KB → 777 KB −32%
Full pipeline – Auto	14.32 ms	13.84 ms	−3%	1399 KB → 1029 KB −26%

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[Copilot]

Pull request overview

This PR adds profiler-enabled benchmarks and applies several targeted allocation/CPU optimizations to the constrained Delaunay triangulation implementation, with the goal of improving hot-path performance while expanding benchmark coverage for previously unmeasured APIs.

Changes:

• Enable EventPipe CPU sampling profiling on benchmark suites and add new benchmark cases for conforming and finalization APIs.
• Reduce allocations in constrained/conforming insertion paths (workspace reuse, stack/list reuse) and pre-size core collections on first insertion.
• Micro-optimize hot paths (custom Edge.GetHashCode, cached two constant, delegate-free quickselect comparers, IsFlipNeeded fast paths).

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.

File	Description
src/CDT.Core/Types.cs	Replaces Edge.GetHashCode implementation to reduce hashing overhead.
src/CDT.Core/Triangulation.cs	Multiple performance refactors: capacity preallocation, workspace reuse, delegate-free NthElement, cached constants, and IsFlipNeeded fast paths.
src/CDT.Core/KdTree.cs	Caches constant two and updates midpoint computation to use cached value.
benchmark/CDT.Benchmarks/Benchmarks.cs	Adds EventPipe profiler attributes and new benchmarks for conforming/finalization APIs.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[MichaConrad]

looks good 👍
performance and memory allocation actually improved quite a bit

Start:

Method	Categories	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
'Conforming – Auto'	Conforming	2.028 ms	0.2340 ms	0.0128 ms	89.8438	89.8438	89.8438	1417.03 KB
'Constrained – AsProvided'	Constrained	3.021 ms	0.1272 ms	0.0070 ms	89.8438	89.8438	89.8438	1220.82 KB
'Constrained – Auto'	Constrained	1.752 ms	0.2863 ms	0.0157 ms	89.8438	89.8438	89.8438	1196.81 KB
'Full pipeline – Auto'	FullPipeline	2.325 ms	0.1574 ms	0.0086 ms	179.6875	89.8438	89.8438	1581.55 KB
'Vertices only – AsProvided'	VerticesOnly	2.745 ms	0.2556 ms	0.0140 ms	89.8438	89.8438	89.8438	1040.3 KB
'Vertices only – Auto'	VerticesOnly	1.701 ms	0.5033 ms	0.0276 ms	89.8438	89.8438	89.8438	1016.29 KB

Final (with some new benchmarks)

Method	Categories	Mean	Error	StdDev	Gen0	Gen1	Gen2	Allocated
'Conforming – Auto'	Conforming	1.589 ms	0.1569 ms	0.0086 ms	117.1875	117.1875	117.1875	1023.33 KB
'Conforming – AsProvided'	Conforming	2.315 ms	0.4291 ms	0.0235 ms	117.1875	117.1875	117.1875	1008.67 KB
'Constrained – AsProvided'	Constrained	2.196 ms	0.1036 ms	0.0057 ms	39.0625	39.0625	39.0625	417.46 KB
'Constrained – Auto'	Constrained	1.339 ms	0.0211 ms	0.0012 ms	39.0625	39.0625	39.0625	433.15 KB
'EraseSuperTriangle – Auto'	Finalization	1.585 ms	0.0344 ms	0.0019 ms	39.0625	39.0625	39.0625	509.9 KB
'EraseOuterTriangles – Auto'	Finalization	1.740 ms	0.1420 ms	0.0078 ms	39.0625	39.0625	39.0625	629.86 KB
'Full pipeline – Auto'	FullPipeline	1.944 ms	0.0956 ms	0.0052 ms	78.1250	39.0625	39.0625	817.9 KB
'Vertices only – AsProvided'	VerticesOnly	2.075 ms	0.0723 ms	0.0040 ms	39.0625	39.0625	39.0625	265.48 KB
'Vertices only – Auto'	VerticesOnly	1.164 ms	0.0873 ms	0.0048 ms	39.0625	39.0625	39.0625	281.22 KB

[github-actions]

Summary

Summary

Generated on:	2/22/2026 - 6:43:27 PM
Coverage date:	2/22/2026 - 6:43:23 PM
Parser:	Cobertura
Assemblies:	1
Classes:	17
Files:	7
Line coverage:	85% (1257 of 1478)
Covered lines:	1257
Uncovered lines:	221
Coverable lines:	1478
Total lines:	3694
Branch coverage:	79.9% (606 of 758)
Covered branches:	606
Total branches:	758
Method coverage:	Feature is only available for sponsors
Tag:	46_22282827581

Coverage

CDT.Core - 85%

Name	Line	Branch
CDT.Core	85%	79.9%
CDT.Box2d`1	68.7%	100%
CDT.CdtUtils	83.1%	72.8%
CDT.CovariantReadOnlyDictionary`3	18.1%	50%
CDT.DictionaryExtensions	100%	100%
CDT.DuplicatesInfo	100%
CDT.DuplicateVertexException	66.6%
CDT.Edge	75%	66.6%
CDT.IntersectingConstraintsException	0%
CDT.KdTree`1	72.7%	65%
CDT.LayerDepth	0%
CDT.Predicates.PredicatesAdaptive	97.6%	91.5%
CDT.Predicates.PredicatesExact	33.7%	100%
CDT.TopologyVerifier	77.2%	63.8%
CDT.Triangle	66.6%	71.4%
CDT.Triangulation`1	93.7%	83.9%
CDT.TriangulationException	100%
CDT.V2d`1	44.4%	25%