I've written a bit of CUDA before. If I want to go pretty bare-bones, what's the equivalent setup for writing code for my AMD card?
How turn-key / happy an experience that is depends on how closely your system correlates with one of the documented/tested distro versions and what GPU you have. If it's one that doesn't have binary versions of rocblas etc in the binary blob, either build rocm from source or don't bother with rocblas.