This is clangd extended to give feedback on host and device parts of cuda, including things like syntax errors within inline ptx. Actual cuda as it is written or the clang dialect as you prefer.
I like the tarball release as that's a set of x86 binaries that run as your normal user. There's other good stuff in the package too - it's a complete cuda to amdgpu or nvptx compiler toolchain.
Questions / feedback welcome here or to jon@spectralcompute.com or via https://discord.com/invite/KNpgGbTc38 etc