[2024] github/dogefromage/pathtracer/README.md
pathtracer from scratch
Started out as a small project over the break and has spiraled out of control over time. Runs on cuda-12.5.
Usage:
- Build the program using
make(requires cuda compiler) - The output will be placed in the
./binfolder - Test by writing
./bin/raytracer -c pathtracer.yaml gltf/bust.gltf - The output will be rendered to
./output.png
# usage:
Usage: ./bin/raytracer [options] <path_to_gltf>
Expects a gltf 2.0 model. Further settings can be applied in pathtracer.yaml.
-c <pathtracer.yaml> Pathtracer render settings file.
-o <output.png> Path to output image.
-v Enable verbose printing.
# pathtracer.yaml layout:
TODO
# example output:
Parsing assets/spheres.gltf... Done
Building bvh_t... [Done]
Copying scene to device... Done [105kB]
Copying bvh_t to device... Done [86kB]
Launching kernel...
Rendering 300 samples in batches of 10, img size (1600, 1600)
Kernel params <<<(100,100), (16,16)>>>
Rendered 10 / 300 samples in 0.3s - 35.59 samples/s - 12.81 MPS/s
Rendered 20 / 300 samples in 0.4s - 45.25 samples/s - 16.29 MPS/s
Rendered 30 / 300 samples in 0.6s - 52.17 samples/s - 18.78 MPS/s
Rendered 40 / 300 samples in 0.7s - 56.58 samples/s - 20.37 MPS/s
Rendered 50 / 300 samples in 0.8s - 59.74 samples/s - 21.51 MPS/s
Rendered 60 / 300 samples in 1.0s - 61.98 samples/s - 22.31 MPS/s
Rendered 70 / 300 samples in 1.1s - 63.64 samples/s - 22.91 MPS/s
Rendered 80 / 300 samples in 1.2s - 64.99 samples/s - 23.40 MPS/s
...
Currently implements:
- basic path tracing on gpu using CUDA or optionally using CPU
- handles large scenes thanks to BVH spacial acceleration structure
- BSDF global illumination and transmission
- rendering to .png image
- BVH construction on CPU with surface area heuristic
TODO:
- more materials / principled bsdf
- light source sampling