when lambda insert "__device__" "__host__",report error

[lishi8168797]

Environment

• OS and Version: ubuntu 20.04
• VS Code Version: 1.74.3
• C/C++ Extension Version: 1.14.0
• If using SSH remote, specify OS of remote machine: no

Bug Summary and Steps to Reproduce

please see red line
it can compile,but in vscode it report error and has a red line

Configuration and Logs

{
    "configurations": [
        {
            "name": "Linux",
            "includePath": [
                "${workspaceFolder}/**",
                "/opt/ros/noetic/include",
                "/usr/local/cuda/include"
            ],
            "defines": [],
            "compilerPath": "/usr/bin/gcc",
            "cStandard": "gnu17",
            "cppStandard": "gnu++14",
            "intelliSenseMode": "linux-gcc-x64"
        }
    ],
    "version": 4
}

Other Extensions

No response

Additional context

No response

[browntarik]

Thank you for submitting this issue! We are currently looking into it.

[browntarik]

I am having a hard time reproducing your error, could you provide a simple test case to reproduce the error? Could you also provide the error output?

[Colengms]

Hi @lishi8168797 . It looks like you're using CUDA. Is the code you posted in an .cu (or .cuh) file?

If you're currently using a .cpp file, you might try renaming it to use .cu. Or, you could specify nvcc as the compilerPath to cause all .cpp files to be treated as CUDA. Or, you could use a compile_commands.json, or a custom configuration provider, to cause files to be configured exactly as built. The actual command line used to build the file will be considered in those cases.

Unfortunately, simply changing the language ID of the document to "CUDA C++" in the editor is insufficient to cause the C/C++ Extension to treat the file as CUDA. I've opened #10494 to address that specific scenario.

[lishi8168797]

@browntarik

it reported this error,and make other code after it error,in fact they are all correct.

[lishi8168797]

@Colengms yes,I am using CUDA,and it is a .cu file,but it still report this red line,and make other code after it also have red line,it bother me.

[sean-mcmanus]

Can you provide the output of running the C/C++: Log Diagnostics command and/or look in that logging for an misconfiguration? I'm guessing you may need to set your compilerPath to the nvcc compiler or make sure gcc isn't being used instead.

[lishi8168797]

sorry,I may not understand that "Can you provide the output of running the C/C++",because,this code run correct,it only report this error in vscode. Does C/C++ Extension provied "insert device in a lambda"?
And I set compiler with nvcc in c_cpp_properties.json,but it does not work.

[sean-mcmanus]

The C/C++: Log Diagnostics command is run from the VS Code command palette -- it can show what IntelliSense configuration is being used. It looks like something is causing the CUDA configuration to fail. @Colengms might have more info when he gets back.

[lishi8168797]

@sean-mcmanus
Thank you,this my log:

-------- Diagnostics - 2023/2/10 10:26:20
Version: 1.14.1
Current Configuration:
{
"name": "Linux",
"includePath": [
"/media/lishi/work/图像点云预处理/点云预处理/ros_env/src/pcl_preprocessing/",
"/opt/ros/noetic/include",
"/usr/local/cuda/include"
],
"defines": [],
"compilerPath": "/usr/local/cuda/bin/nvcc",
"cStandard": "gnu17",
"cppStandard": "gnu++14",
"intelliSenseMode": "linux-gcc-x64",
"compilerPathIsExplicit": true,
"cStandardIsExplicit": true,
"cppStandardIsExplicit": true,
"intelliSenseModeIsExplicit": true,
"mergeConfigurations": false,
"browse": {
"path": [
"${workspaceFolder}/",
"/opt/ros/noetic/include",
"/usr/local/cuda/include"
],
"limitSymbolsToIncludedHeaders": true
}
}
Translation Unit Mappings:
[ /media/lishi/work/图像点云预处理/点云预处理/ros_env/src/pcl_preprocessing/src/octree/octree.cu ]:
/media/lishi/work/图像点云预处理/点云预处理/ros_env/src/pcl_preprocessing/src/octree/octree.cu
Translation Unit Configurations:
[ /media/lishi/work/图像点云预处理/点云预处理/ros_env/src/pcl_preprocessing/src/octree/octree.cu ]:
Process ID: 7754
Memory Usage: 553 MB
Compiler Path: /usr/bin/gcc
Includes:
/opt/ros/noetic/include
/usr/local/cuda-11.6/targets/x86_64-linux/include
/usr/include/c++/9
/usr/include/x86_64-linux-gnu/c++/9
/usr/include/c++/9/backward
/usr/lib/gcc/x86_64-linux-gnu/9/include
/usr/local/include
/usr/include/x86_64-linux-gnu
/usr/include
/media/lishi/work/图像点云预处理/点云预处理/ros_env/src/pcl_preprocessing/include
Defines:
CUDA_ARCH=520
CUDA_ARCH_LIST=520
CUDA_DOUBLE_MATH_FUNCTIONS
CUDACC
NVCC
CUDACC_VER_MAJOR=11
CUDACC_VER_MINOR=6
CUDACC_VER_BUILD=55
CUDA_API_VER_MAJOR=11
CUDA_API_VER_MINOR=6
NVCC_DIAG_PRAGMA_SUPPORT=1
Forced Includes:
/usr/local/cuda/include/cuda_runtime.h
Standard Version: c++14
IntelliSense Mode: linux-gcc-x64
Other Flags:
--g++
--gnu_version=90400
--cuda
Total Memory Usage: 553 MB

------- Workspace parsing diagnostics -------
Number of files discovered (not excluded): 38229

[sean-mcmanus]

Your logging does indicate that it's compiling in CUDA mode, so that looks like __device__ __host__ should be recognized -- I'll wait for @Colengms to get back to see if he has any ideas what could be going wrong. (I'm guessing the /usr/bin/gcc compiler path is just showing what nvcc uses to compile the C++ code with).

[Colengms]

Hi @lishi8168797 . If your file is being correctly interpreted as CUDA, and your code compiles properly, perhaps there is an issue with IntelliSense resulting in __host__ and __device__ (aka __attribute__((host)) and __attribute__((device))) not being recognized in certain contexts.

Can you provide a simple isolated example that we might pass along to our IntelliSense team, to further investigate the issue?

[lishi8168797]

Hi @Colengms .This is my example,thanks for your help.

    float dim_x,dim_y,dim_z;
    float x_min,y_min_,z_min; 
    int depth_mult=1024;
    auto morton_func=[dim_x,dim_y,dim_z,depth_mult,x_min_,y_min_,z_min_]  
        __host__ __device__ 
        (const float& x,const float& y,const float& z,int& code){
        int cellx,celly,cellz;
        cellx = min((int)std::floor(depth_mult * (x-x_min_)/dim_x), depth_mult - 1);
        celly = min((int)std::floor(depth_mult * (y-y_min_)/dim_y), depth_mult - 1);
        cellz = min((int)std::floor(depth_mult * (z-z_min_)/dim_z), depth_mult - 1);
        //......................9876543210
        cellx = (cellx | (cellx << 10)) & 0x000f801f; //............98765..........43210
        cellx = (cellx | (cellx <<  4)) & 0x00e181c3; //........987....56......432....10
        cellx = (cellx | (cellx <<  2)) & 0x03248649; //......98..7..5..6....43..2..1..0
        cellx = (cellx | (cellx <<  2)) & 0x09249249; //....9..8..7..5..6..4..3..2..1..0
        //......................9876543210
        celly = (celly | (celly << 10)) & 0x000f801f; //............98765..........43210
        celly = (celly | (celly <<  4)) & 0x00e181c3; //........987....56......432....10
        celly = (celly | (celly <<  2)) & 0x03248649; //......98..7..5..6....43..2..1..0
        celly = (celly | (celly <<  2)) & 0x09249249; //....9..8..7..5..6..4..3..2..1..0
        celly=celly << 1;
        //......................9876543210
        cellz = (cellz | (cellz << 10)) & 0x000f801f; //............98765..........43210
        cellz = (cellz | (cellz <<  4)) & 0x00e181c3; //........987....56......432....10
        cellz = (cellz | (cellz <<  2)) & 0x03248649; //......98..7..5..6....43..2..1..0
        cellz = (cellz | (cellz <<  2)) & 0x09249249; //....9..8..7..5..6..4..3..2..1..0
        cellz=cellz << 2;
        code=cellx | celly | cellz;
    };

    thrust::for_each(thrust::device,thrust::make_zip_iterator(x_ptr,y_ptr,z_ptr,temp_code.begin()),
    thrust::make_zip_iterator(x_ptr+size,y_ptr+size,z_ptr+size,temp_code.begin()+size),
    thrust::make_zip_function(morton_func));

[Colengms]

Hi @lishi8168797 . I've opened an internal issue (1746549) against VS, as this issue is with the IntelliSense parser which is shared with VS.

[sean-mcmanus]

The fix is available with 1.15.0 (pre-release): https://github.com/microsoft/vscode-cpptools/releases/tag/v1.15.0

[lishi8168797]

The fix is available with 1.15.0 (pre-release): https://github.com/microsoft/vscode-cpptools/releases/tag/v1.15.0

It works well ! Thank you very much !