API Reference lacks logical organization and is hardly readable

[leofang]

I am doing some internal tests using CUDA Python, and I find the Python API Reference is hardly readable. Currently all APIs (functions, classes (=C structs), attributes, etc) are all dumped in the same page. Furthermore, the docs for all modules (cuda, cudart, nvrtc) are also coalesced in the same page, making the situation even worse. This is a screenshot of the gargantua page:

What I'd expect:

• Under the entry "CUDA Python API Reference" of the ToC on the left, we list the 3 modules as sub-entries
• Under each module, we further list 3 (or more) sub-sub-entries: "Functions", "Classes", "Attributes" (maybe defines/typedefs/enums can all be combined as Attributes)
• In each of the "Functions" sub-sub-entry, we organize the contents into sub-sub-sub-entries based on their purpose. For example, for the CUDA Runtime APIs we can follow how the parent page does it: https://docs.nvidia.com/cuda/cuda-runtime-api/modules.html#modules

What we'll achieve by doing so:

• Match the way CUDA programmers search the CUDA documentations
• Provide a better user-friendly, more logical organization for the docs

[leofang]

Closing this issue as it's resolved since CUDA Python 11.6.0 👍