CUDA + OpenGL + BVH Path Tracing Renderer

An interactive path tracing sample on Windows using CUDA–OpenGL interop (PBO). It supports:

• CUDA device path tracing (multi-bounce, GGX specular, cosine / uniform hemisphere sampling, Russian Roulette)

• Unified scene triangle consolidation and single BVH acceleration structure

• Direct / Indirect / MIS (Multiple Importance Sampling) lighting modes

• Runtime ImGui controls (samples per pixel, max depth, lighting mode, importance sampling, Russian Roulette)

• Procedural scattering of many OBJ models (example: 100 randomized dragons with random color + emission probability)

• GPU accumulation buffer (progressive refinement / noise reduction)

  

Core Pipeline

1. Application init: Window → OpenGL → ImGui → GPU PBO → Scene (scatter models)
2. Scene collects triangles from all Model objects, builds a single BVH, gathers materials & light triangle indices.
3. Render loop:
   • Detect camera/scene dirty → reset accumulation → rebuild + upload
   • CUDA kernel performs per-pixel multi-sample path tracing (Spp)
   • Kernel writes to registered PBO; OpenGL updates screen texture
   • ImGui overlays control panel

Path Tracing Highlights

• BRDF:
  • Diffuse: Lambert (albedo / π)
  • Metallic: GGX microfacet + Smith geometry term + Fresnel Schlick
• Importance Sampling: optional cosine-weighted hemisphere for diffuse
• MIS: combines light sampling and BSDF sampling using power heuristic
• Light handling: all emissive model triangles become light sampling set (area sampling)
• Russian Roulette: after threshold depth, terminate based on max throughput component, energy-conserving
• Accumulation: accumulated_radiance[pixel] = mix(previous, current, 1/(frame+1))

Models & Emission

• RapidOBJ parses arbitrary polygon faces → triangulated by fan method
• Each Model can receive randomized Material + emission on CPU; emission stored in MaterialGpu during upload
• Emissive test: glm::dot(emission, emission) > 0 marks triangle as light

ImGui Adjustable Parameters

• Samples Per Pixel (Spp)
• Max Path Depth
• Lighting Mode: Direct / Indirect / MIS
• Diffuse Importance Sampling toggle
• Russian Roulette toggle + start depth
• Camera: FOV / position / yaw / pitch / movement
• Per-model editing: position, rotation, scale, Albedo, Metallic, Emission, dynamic OBJ loading

Build Requirements (Windows)

• Visual Studio 2022 (MSVC x64 toolchain)
• CUDA Toolkit 12.x (compatible with VS)
• CMake >= 3.18

Build Steps (PowerShell)

mkdir build
cd build
cmake .. -G "Visual Studio 17 2022" -A x64
cmake --build . --config Release

Run after build (inside build/Release/):

./raytracer.exe

Runtime Interaction

• Keyboard: W/A/S/D move, Arrow keys rotate, Space up, Shift down, ESC exit
• Any change to camera or scene resets accumulation (re-converge)
• Higher Spp → slower but cleaner; lower Spp → faster but noisier

Code Landmarks

• CUDA kernel entry: kernel_path_tracer_impl (in path_tracer.cu)
• Ray tracing loop: traceRay
• Direct lighting sample: sampleDirectLightingContribution
• BVH intersection: intersectScene + shadowIntersectScene
• Importance sampling & GGX: sampleGGX, cosineSampleHemisphere (in pt_utils.cu)
• Scene rebuild/upload: Scene::buildAndUploadScene

Extension Ideas

• Texture support (UV + sampler)
• Refraction / transparent dielectric materials
• Advanced MIS (many-light + BSDF hybrid strategies)
• Environment map / HDR sky sampling
• GPU-side BVH building / dynamic updates
• Denoising integration: OIDN / OptiX

Third-Party Dependencies

• GLFW (Zlib)
• GLM (MIT)
• ImGui (MIT)
• rapidobj (MIT)
• glad (MIT)