trueform

Benchmarks | C++

Overview

Performance comparisons against VTK, CGAL, libigl, Coal, FCL, and nanoflann.

TrueForm enables real-time geometric operations on million-polygon meshes. On a pair of 1M-polygon meshes, boolean union completes in 28ms and intersection curve extraction in 7ms. Try it live in your browser →

These benchmarks quantify performance against established libraries across spatial indexing, topology analysis, intersection computation, and mesh cutting.

Experimental Setup

Property	Value
CPU	Apple M4 Max (12P + 4E cores)
Memory	64 GB
Compiler	Clang with `-O3 -march=native`
Allocator	mimalloc
Test Data	Stanford Dragon, 58K–1M polygons (contains non-manifold flaps)
Sampling	10–1000 evaluations per configuration

Construction benchmarks use 10 runs; query benchmarks use 1000. All libraries use parallel execution where available. For sub-millisecond queries, mean time is reported; for longer operations, minimum time is reported to reduce noise from OS scheduling. Source code →

Summary

Sample comparisons against VTK, CGAL, libigl, Coal, FCL, and nanoflann:

Operation	Input	Time	Speedup	Baseline	TrueForm
Boolean Union	2 × 1M	28 ms	84×	CGAL `Simple_cartesian<double>`	reduction diagrams, double
Mesh–Mesh Curves	2 × 1M	7 ms	233×	CGAL `Simple_cartesian<double>`	reduction diagrams, double
Self-Intersection	1M	78 ms	37×	libigl EPECK (GMP/MPFR)	reduction diagrams, double
Isocontours	1M, 16 cuts	3.8 ms	38×	VTK `vtkContourFilter`	reduction diagrams, float
Connected Components	1M	15 ms	10×	CGAL	parallel union-find
Boundary Paths	1M	12 ms	11×	CGAL	Hierholzer's algorithm
k-NN Query	500K	1.7 µs	3×	nanoflann k-d tree	AABB tree
Mesh–Mesh Distance	2 × 1M	0.2 ms	2×	Coal (FCL) `OBBRSS`	OBBRSS tree
Principal Curvatures	1M	25 ms	55×	libigl	parallel k-ring quadric fitting

To learn more about the reduction diagrams and the theoretical foundations of trueform, see our Research.

Detailed Results

Explore full benchmark charts and scaling analysis for each module:

Spatial

AABB tree construction, k-NN queries, and mesh–mesh distance.

Topology

Connected component extraction and boundary path detection.

Geometry

Curvature estimation and geometric analysis.

Intersect

Isocontour extraction and mesh–mesh intersection curves.

Cut

Boolean operations, self-intersection detection, and curve embedding.

I/O

File reading and writing performance.

Running Benchmarks

Dependencies

brew install boost gmp mpfr libccd tbb assimp libomp

sudo apt-get install libboost-dev libgmp-dev libmpfr-dev libccd-dev libtbb-dev libassimp-dev libomp-dev

vcpkg install boost gmp mpfr libccd tbb assimp llvm-openmp

Comparison Libraries

CGAL, libigl, FCL, Coal, and nanoflann are fetched automatically via CMake FetchContent.

VTK is optional due to its size. Install it separately to include VTK benchmarks; otherwise they are skipped.

brew install vtk

sudo apt-get install libvtk9-dev

vcpkg install vtk

Build and Run

cmake -B build -DTF_BUILD_BENCHMARKS=ON -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel --target benchmarks
./build/run-benchmarks

Results are written to build/results/. Use consolidate_results.py to merge per-library CSVs.

I/O

File I/O operations for reading and writing mesh data.

Spatial

Tree construction and proximity queries.