Those are mostly overloads of already existing CUDA core functions for more vector types.
Those are all for convenience and there is little need to document an fminf() function or the like.
You should be able to see what they are doing just from looking into that header.
The root functions called in there are documented inside the CUDA programming guide: https://docs.nvidia.com/cuda/cuda-c-programming-guide/
The ones which are added for convenience (reflect, refract, fresnel_schlick, etc.) are common graphics programming knowledge you can find in any 3D graphics programming book or other 3D graphics APIs and some are documented additionally inside the optixu_math_namespace.h header.
When starting to learn OptiX, please follow the recommendations I gave plenty of times on this forum.
There is a search field in the top right. Search for anything, then click on the “Show” link in the results page and refine your search to the OptiX forum by most recent date to drill down on relevant answers.