OpenCL Path Tracer
Monte Carlo path tracer solving the rendering equation on the GPU, applied to 3D fractal geometry.
- Method
- Monte Carlo
- Compute
- OpenCL kernels
A physically based path tracer that runs the sampling loop as OpenCL kernels on the GPU. Written to work through the math directly rather than to wrap an existing renderer.
Approach
The renderer solves the rendering equation by Monte Carlo integration, tracing paths from the camera and accumulating radiance across samples. Importance sampling concentrates paths where they contribute most, which is what makes the estimator converge in a workable number of samples instead of an intractable one.
Ray marching against distance fields provides the geometry, which is what makes 3D fractals tractable as subjects — they have no mesh to intersect against, only a distance estimator.
Why OpenCL
Path tracing is close to the ideal shape for data-parallel hardware: each sample is independent, and the work is arithmetic-bound rather than branch-bound. The kernel work here is the most direct evidence behind the GPU compute experience on my resume — offset surface computation at E4D was the same class of problem applied to a production CAD pipeline.