As requested, Artem and I started some documentation on the CUDA support in LLVM. The CUDA support is not fully ready yet (some patches pending as the doc mentioned), but in a pretty reasonable shape for early adopters.
We will document more content down the road such as CUDA-specific compilation flags, performance numbers, and performance tuning instructions. At the same time, feel free to improve it and contribute more documentation as you see appropriate. Thanks!
Jingyue