CUDA Accelerated Ray Tracer

A CUDA implementation of the Ray Tracing in One Weekend Book Series with per-pixel RNG, thin-lens depth of field, metal/dielectric materials, procedural textures,
quads, instancing, light sources, Cornell box, texture mapping, and the randomized final scene. The code mirrors the books’ progression, adapted for GPU kernels and
CUDA memory/launch patterns.

References
• Peter Shirley, Ray Tracing in One Weekend Book Series — https://raytracing.github.io/
• NVIDIA Developer Blog, “Accelerated Ray Tracing in One Weekend in CUDA” — https://developer.nvidia.com/blog/accelerated-ray-tracing-cuda/

---

Features

Rendering pipeline

• Antialiasing via stochastic supersampling (per-pixel jitter using cuRAND)

  

• Gamma correction with configurable values

  

Materials

• Metal with fuzz parameter for rough/microfacet reflections
• Lambertian (diffuse) with random cosine-ish scattering
• Dielectric (glass): refraction, total internal reflection, and Schlick reflectance

  

Camera

• Thin-lens defocus blur (depth of field) (aperture + focus distance)
• Proper camera basis (u/v/w) with configurable FOV and aspect ratio

  

Procedural Textures

• Checkered planes and spheres
• Perlin noise textures with turbulence & marble patterns

Motion Blur

• Shutter interval sampling: each ray carries a randomized time in [time0, time1]
• Animated primitives: object positions are interpolated across the shutter window
• Produces natural blur trails when geometry moves during exposure

Light Sources

• Emissive materials for area lights
• Example: Cornell box ceiling quad + glowing sphere

Quads & Rectangular Geometry

• General quad primitive
• Used for walls, floors, ceilings, and light sources

Instancing & Object Transforms

• Translate / rotate geometry without duplicating vertex data
• Used to place rotated blocks in the Cornell Box

Texture Mapping

• Spherical coordinate texture mapping
• Example: Earth texture on a sphere

---

Requirements

• Windows with Visual Studio 2022 (MSVC toolset)
• CUDA Toolkit 12+ (13.x tested)
• CMake 3.24+
• NVIDIA GPU with supported compute capability (current flags target Ada 8.9; adjust for your GPU)

If you target a different architecture, set CMAKE_CUDA_ARCHITECTURES (or -gencode in NVCC flags) appropriately in CMakeLists.txt.

---

Build & Run

Quick start (Windows)

A helper script automates clean, configure, build, and run. It will prompt for an output name and append .ppm:

setup.bat

Manual CMake build

rmdir /s /q build
cmake -S . -B build
cmake --build build --config Debug
Build\\bin\\Debug
RayTracer.exe > output.ppm

Running & output

• The renderer writes a PPM (P3) image to stdout.

• Run after building (Debug example):

  Build\\bin\\Debug
  rayTracer.exe > output.ppm

• Or use the helper script (prompts for a filename and appends .ppm):

  setup.bat

• View/convert the PPM:

  • ImageMagick: magick convert output.ppm output.png
  • GIMP/Photoshop: open output.ppm directly

Scene parameters (resolution nx/ny, samples ns, camera, aperture, focus distance, and scene generator) live in src/main.cu.

---

How it works (GPU notes)

• Bounded depth instead of recursion: the book’s recursive color() is turned into a loop (default max depth = 50) to avoid device stack overflows.
• Per-pixel RNG: Each thread has a curandState. We copy the state to a local variable, sample multiple times, then write it back.
• Unified memory for the framebuffer (cudaMallocManaged) to simplify host readout (stdout → PPM).
• Device-side scene build: A small kernel constructs the world and camera once, then the main render kernel traces rays.
• Thin-lens DOF: random_in_unit_disk samples the aperture; lower_left_corner, horizontal, and vertical are scaled by the focus distance; lens_radius = aperture/2.

---

Tuning

• Samples per pixel (spp): higher ns → cleaner images (time ∝ spp).
• Resolution: increase nx/ny for detail.
• Max depth: 50 is a good default; raising it gives diminishing returns.
• Aperture: small (0.1) = subtle blur; large (2.0) = strong DOF (needs more spp).
• Build config: use Release for speed; set CMAKE_CUDA_ARCHITECTURES to match your GPU.

---

Acknowledgements

• Peter Shirley et al. for the Ray Tracing in One Weekend series.
• NVIDIA Developer Blog for the CUDA adaptation guidance.

---