Toy Path Tracer

Toy path tracer I did in 2018 for my own learning purposes. Somewhat based on Peter Shirley's Ray Tracing in One Weekend minibook (highly recommended!), and on Kevin Beason's smallpt.

I decided to write blog posts about things I discover as I do this:

• Part 0: Intro
• Part 1: Initial C++ and walkthrough
• Part 2: Fix stupid performance issue
• Part 3: C#, Unity and Burst
• Part 4: Correctness fixes and Mitsuba
• Part 5: simple GPU version via Metal
• Part 6: simple GPU version via D3D11
• Part 7: initial C++ SIMD & SoA
• Part 8: SSE SIMD for HitSpheres
• Part 9: ryg optimizes my code
• Part 10: Update all implementations to match
• Part 11: Buffer-oriented approach on CPU
• Part 12: Buffer-oriented approach on GPU D3D11
• Part 13: GPU thread group data optimization
• Part 14: Make it run on iOS
• Part 15: A bunch of path tracing links
• Part 16: Unity C# Burst optimization
• Part 17: WebAssembly

Note: it can only do spheres, no bounding volume hierachy of any sorts, a lot of stuff hardcoded.

Performance numbers in Mray/s on a scene with ~50 spheres and two light sources, running on the CPU:

| Language | Approach | Ryzen 5950 | AMD TR1950 | MBP 2021 | MBP 2018 | MBA 2020 | iPhone 11 | iPhone X | iPhone SE | |:--- |:--- | ---:| ---:| ---:| ---:| ---:| ---:| ---:| ---:| | C++ | SIMD Intrinsics | 281.0 | 187.0 | 105.4 | 74.0 | 32.3 | 26.4 | 12.9 | 8.5 | | | Scalar | 141.2 | 100.0 | 84.8 | 35.7 | | 15.9 | | | WebAssembly (no threads, no SIMD) | 8.4 | 5.0 | 8.1 | | | 5.6 | | C# | Unity Burst "manual" SIMD | 227.2 | 133.0 | 103.7 | 60.0 | | 29.7 | | | Unity Burst | | 82.0 | | 36.0 | | | | | Unity (Editor) | 6.5 | | 3.4 | | | | | | Unity (player Mono) | 6.7 | | 3.5 | | | | | | Unity (player IL2CPP) | 39.1 | | 63.8 | | | 17.2 | | | .NET 6.0 | 91.5 | 53.0 | 40.9 | | | | .NET Core 2.0 | 86.1 | 53.0 | | 23.6 | | | Mono --llvm | | | 35.1 | 22.0 | | | Mono | 23.6 | | 3.6 | 6.1 |

More detailed specs of the machines above are:

• Ryzen 5950: AMD Ryzen 5950X (3.4GHz, 16c/32t), Visual Studio 2022.
• AMD TR1950: AMD ThreadRipper 1950X (3.4GHz, SMT disabled - 16c/16t), Visual Studio 2017.
• MBP 2021: Apple MacBook Pro M1 Max (8+2 cores), Xcode 13.2.
• MBP 2018: Apple MacBook Pro mid-2018 (Core i9 2.9GHz, 6c/12t).
• MBA 2020: Apple MacBook Air 2020 (Core i7 1.2GHz, 4c/8t).
• iPhone 11: A13 chip.
• iPhone X: A11 chip.
• iPhone SE: A9 chip.

Software versions:

• Unity 2021.3.16. Burst 1.6.6 (safety checks off). C# testing in editor, Release mode.
• Mono 6.12.

And on the GPU, via a compute shader in D3D11 or Metal depending on the platform:

| GPU | Perf | |:--- | ---:| | D3D11 | | | GeForce RTX 3080Ti | 3920 | | GeForce GTX 1080Ti | 1854 | | Metal | | | MBP 2024 (M4 Max) | 1680 | | MBP 2021 (M1 Max) | 1065 | | MBP 2018 (Radeon Pro 560X) | 246 | | MBA 2020 (Iris Plus) | 201 | | iPhone 11 Pro (A13) | 80 | | iPhone X (A11) | 46 | | iPhone SE (A9) | 20 |

A lot of stuff in the implementation is totally suboptimal or using the tech in a "wrong" way. I know it's just a simple toy, ok :)

Building

• C++ projects:
  • Windows (Visual Studio 2017) in Cpp/Windows/ToyPathTracer.sln. DX11 Win32 app that displays result as a fullscreen CPU-updated or GPU-rendered texture. Pressing G toggles between GPU and CPU tracing, A toggles animation, P toggles progressive accumulation.
  • Mac/iOS (Xcode 10) in Cpp/Apple/ToyPathTracer.xcodeproj. Metal app that displays result as a fullscreen CPU-updated or GPU-rendered texture. Pressing G toggles between GPU and CPU tracing, A toggles animation, P toggles progressive accumulation. Should work on both Mac (Test Mac target) and iOS (Test iOS target).
  • WebAssembly in Cpp/Emscripten/build.sh. CPU, single threaded, no SIMD.
• C# project in Cs/TestCs.sln. A command line app that renders some frames and dumps out final TGA screenshot at the end.
• Unity project in Unity. I used Unity 2021.3.16.