[Issue]: nvshmem.h:__device__ function 'nvshmem_team_my_pe' cannot overload __host__ __device__ function 'nvshmem_team_my_pe'

[Qizhi697]

After compiling normally with nvcc, clangd reports errors. How can I resolve this issue?

NVSHMEM 3.4.5-0 & CUDA 13.0

[benhg]

What version of clang/llvm are you using? When we build our bitcode libraries for release, we build with clang 19.

[benhg]

I don't think your static analysis tool is expanding the macros correctly then. When we build the llvm bitcode library, we define NVSHMEM_HOSTLIB_ONLY, which results in skipping the branch on line 22 in your screenshot.

[benhg]

This is because we build the bitcode library as a device library only. So there's no host device implementation of those functions available through the bitcode library.

[Qizhi697]

Thanks，I set NVSHMEM_HOSTLIB_ONLY, and then the problem was solved!