Ray Tracing

A Rust reimplementation of a simple ray tracer introduced in Ray Tracing in One Weekend.

This project is heavily influenced by a similar one: ray-tracing-in-one-weekend.

Features

See CHANGELOG for version details.

• BVH (Bounding Volume Hierarchies) from Ray Tracing: The Next Week.
• Parallel computing powered by Rayon.

Example result

The results below are of 1920x1080, sample rate 1000, max depth 50.

Run without feature flags

• scene: lined-up scene (with camera focus)
• rendering time: 201.0s

Run with benchmark feature

• scene: final scene (with camera focus)
• rendering time: 163.8s

Run with course feature

• scene: lined-up scene (without camera focus)
• rendering time: 217.1s

Performance

With benchmark feature (from v0.2.1) enabled (which limits most of the randomness), we get a not-so-serious (but useful enough) benchmark system to measure the performance (rendering time) difference between versions.

Configuration of the benchmark:

• resolution 1200x800, sample rate 100, max depth 50
• run cargo build --release --features benchmark
• run hyperfine --warmup 1 -r 10 'target/release/ray-tracing --ns 100'
• use the min given by hyperfine as the result

Version	Best Rendering Time (s)	Speed (sample/s)	Rel-Speed	Note
v0.1.0	39.179	2.45M	1
v0.2.0	32.212	2.98M	1.22	implemented BVH
v0.2.1	-	-	-	no perf improvement
v0.2.2	19.109	5.02M	2.05	limited BVH depth
v0.3.0	8.256	11.63M	4.75	* see below
v0.4.0	-	-	-	no perf improvement
v0.4.1	-	-	-	no perf improvement
v0.5.0	7.870	12.20M	4.98	removed bench timer

• "Rel-Speed" is the relative speed compared to v0.1.0.
• Note for v0.3.0:
  • Switched to stratified pixel sampling.
  • Optimized AaBb::hit() by removing branches, enabling better SIMD utilization.
  • Refactored Material trait as an enum to reduce runtime overhead.
  • Enabled lto and tuned codegen-units for improved performance.