A small CPU path tracer that renders a Cornell Box–style scene to a PPM image.

This project is intentionally minimal: everything lives in a single file, pathtracer.cpp.

• Unidirectional path tracing with throughput accumulation
• Russian Roulette termination (starts after a few bounces)
• BVH acceleration (templated BVH, SAH-like split search)
• Multithreaded tile renderer using std::thread + atomics
• Materials:
  • Lambertian diffuse
  • Diffuse area light (DiffuseLight)
  • Simple PBR-style material (PBRMaterial)
  • GGX rough metal (GGXMetal) with importance sampling
• Output:
  • Linear HDR accumulation per pixel
  • Optional Reinhard tonemapping
  • Gamma 2.2 correction

By default, the program writes:

• cornell_box_ggx_pt.ppm

PPM is a simple text image format. Many viewers can open it directly; otherwise you can convert it.

clang++ -std=c++17 -O3 -DNDEBUG pathtracer.cpp -o pathtracer.exe

This produces pathtracer.exe.

Run this inside a Developer Command Prompt (so MSVC libraries are available):

Notes:

• You can add -march=native for extra performance on your CPU.

pathtracer.exe

You’ll see progress in the terminal (tile-based percentage + elapsed time). When finished, the PPM is written to the working directory.

Open pathtracer.cpp and modify the constants in main():

• image_width / image_height: output resolution
• samples_per_pixel: main quality knob (noise decreases ~ $1/\sqrt{\text{spp}}$)
• max_depth: max bounces

The render call at the end of main() is:

render_to_ppm(world, cam,
              image_width, image_height,
              samples_per_pixel, max_depth,
              "cornell_box_ggx_pt.ppm",
              1.0,  // exposure
              true  // enable_reinhard_tonemap
);

• Exposure scales the final radiance before tonemapping/gamma.
• Reinhard tonemapping helps prevent highlight blowout for bright lights.

• Cornell Box walls (red/green/white)
• A ceiling area light (rectangle)
• Three spheres inside the box:
  • a metallic PBR sphere
  • a diffuse PBR sphere
  • a GGX rough metal sphere

• The environment is treated as black (no sky); rays that miss the scene contribute nothing.
• This is a learning-focused renderer; it does not aim to match a reference implementation pixel-for-pixel.