I am trying to set up my local development environment following the documentation at IDE Setup, but I have encountered a few issues regarding the configuration correctness and IDE features (VSCode + Nsight Visual Studio Code Edition + C/C++ extension).

1. Potential documentation error regarding nvcc flags
The documentation mentions configuration options that seem to include -xcuda with Compiler: /usr/local/cuda/bin/nvcc. However, nvcc does not appear to support the -xcuda flag.
Could you please verify if the documentation is outdated or if this flag is intended for a specific compiler setup that isn't clearly distinguished?

2. "Go to Definition" fails inside __device__ functions
I am using VSCode with the Nsight Visual Studio Code Edition plugin and have configured c_cpp_properties.json as suggested.

• Working: I can successfully use Ctrl + Left Click (Go to Definition) on CuTe structures like TiledCopy and ThrCopy when they are used inside the main function (host code).
• Not Working: The same "Go to Definition" feature fails when clicking on TiledCopy or ThrCopy inside __device__ functions or kernels. The IDE does not seem to resolve the definition correctly in the device context.

Request
Could you please share a working .vscode configuration (specifically c_cpp_properties.json or settings.json) that is verified to support full code navigation (Go to Definition) for CuTe/Cutlass structures within both host and device code? Here is my configuration:

// .vscode/c_cpp_properties.json
{
    "configurations": [
        {
            "name": "Linux",
            "includePath": [
                "${workspaceFolder}/src/cutlass/include",
                "${workspaceFolder}/src/cutlass/tools/util/include",
                "${workspaceFolder}/src/cutlass/examples/common"
            ],
            "defines": [
                "__CUDACC__",
                "__NVCC__",
                "__clang__",
                "__CUDACC_VER_MAJOR__=12",
                "__CUDACC_VER_MINOR__=8",
                "__CUDA_ARCH__=890",
                "__CUDA_ARCH_FEAT_SM89_ALL"
            ],
            "compilerPath": "/usr/local/cuda/bin/nvcc",
            "cStandard": "gnu17",
            "cppStandard": "gnu++17",
            "intelliSenseMode": "linux-gcc-x64"
        }
    ],
    "version": 4
}

Thanks!