Notice, this is just a proof of concept, it is not intended to be a real Julia package!
This package demonstrates how GPUCompiler.jl/CUDA.jl can be used to create OptiX "programs" written in Julia.
Potential solutions to several problems in wrapping and abstracting OptiX are demonstrated, for example:
- The OptiX library is accessed by function pointers obtained in runtime;
- Parameters are passed to OptiX "programs" as constant global GPU memory;
- Device side functions must be called using custom assembly code (some have multiple return variables);
- OptiX consumes PTX containing functions with very specific signatures.
A usage sample is available in samples folder, it produces this:
