Accel uses the LLVM PTX backend, which is unusable for serious projects and doesn't work on windows, i have more about it here https://github.com/RDambrosio016/Rust-CUDA/blob/master/guide...
What is missing in your opinion aside from working debug info? I've worked on experimental CUDA support in GNAT (the Ada frontend of GCC), using LLVM's ptx backend to target GPUs and it was mostly working.