https://github.com/hilbertcube/simd-pathtracer
A path tracer with 3D model importing, written in modern C++ and accelerated using SIMD, OpenMP, and tile-based rendering
https://github.com/hilbertcube/simd-pathtracer
3d-graphics 3d-models cpp openmp-parallelization path-tracing raytracing simd
Last synced: 4 months ago
JSON representation
A path tracer with 3D model importing, written in modern C++ and accelerated using SIMD, OpenMP, and tile-based rendering
- Host: GitHub
- URL: https://github.com/hilbertcube/simd-pathtracer
- Owner: hilbertcube
- License: mit
- Created: 2025-09-15T07:32:03.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2025-09-25T08:56:01.000Z (5 months ago)
- Last Synced: 2025-09-25T10:30:24.587Z (5 months ago)
- Topics: 3d-graphics, 3d-models, cpp, openmp-parallelization, path-tracing, raytracing, simd
- Language: C++
- Homepage:
- Size: 26.7 MB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Ray tracing in one weekend, with SIMD
An optimized implementation of Peter Shirley’s classic ray tracing (actually path tracing) series: [*Ray Tracing in One Weekend*](https://raytracing.github.io/books/RayTracingInOneWeekend.html) and [*Ray Tracing: The Next Week*](https://raytracing.github.io/books/RayTracingTheNextWeek.html). This version is written in C++17, enhanced with SIMD intrinsics and OpenMP multithreading for better performance.
### Preview:
## Features Added
Peter Shirley’s originals were intentionally C-style. This project reimagines them with modern C++ practices and performance optimizations.
Some features added:
- A 3D model importer.
- Generic vector class: implemented a generic N-dimensional vector class.
- Image Writer: Instead of relying on a PPM writer, which is very slow and produces very large files, I implemented an image writer using stb_image_write. It supports multiple output formats, including PPM, PNG, JPEG, BMP, and TGA.
- Multithreading: Initially, I implemented a custom thread pool, but later transitioned to OpenMP for simplicity and better performance (skill issue on my part).
- Tile-based rendering: The image is divided into tiles for more efficient parallel rendering.
- CPU SIMD intrinsics: Used for optimized vector operations and improved performance.
## Building with CMake
### External (header-only) dependencies
- [stb_image_write.h](https://github.com/nothings/stb/blob/master/stb_image_write.h)
- [stb_image.h](https://github.com/nothings/stb/blob/master/stb_image.h)
- [tiny_obj_loader.h](https://github.com/tinyobjloader/tinyobjloader)
### Requirements
- C++17 (GCC >= 9, Clang >= 10, or MSVC >= 2019)
- [CMake](https://cmake.org/)
- OpenMP support
### Build Instructions
```bash
# Clone the repository
git clone https://github.com/hilbertcube/SIMD-Pathtracer.git
cd SIMD-Pathtracer
# Create a build directory
mkdir build && cd build
# Configure the project
cmake .. -DCMAKE_BUILD_TYPE=Release
# Build the project
cmake --build .
```
### Running
After building, you’ll find the executable inside the build/ directory:
```bash
./SIMD-Pathtracer
```
By default, it renders the final scene and saves it to `output.png` (or another format if specified).
### Usage
To enable SIMD, simply add `#define _USE_SIMD_` before `#include "vector.hpp"`, but remember to do this only once, or else just use `#ifndef`.
## Future plans
- CUDA acceleration
- Command line arguments