How to include torch/extension.h correctly?

C++
1

Hi, I’m following this tutorial to extend torch to cpp/cuda.
My setup is Ubuntu 22, Visual Code, Cuda toolkit 12.1, torch install by conda.

>>> torch.version.cuda
'12.1'
>>> torch.__version__
'2.1.0.dev20230614'

I create my cpp file as follows:

#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>


// Forward pass 

// Backward pass

In .vscode/c_cpp_properties.json, I added include path:

{
    "configurations": [
        {
            "name": "Linux",
            "includePath": [
                "${workspaceFolder}/**",
                "/usr/local/cuda/include",
                "/home/tien/anaconda3/lib/python3.9/site-packages/torch/include",
                "/home/tien/anaconda3/lib/python3.9/site-packages/torch/include/torch/csrc/api/include"
            ],
            "defines": [],
            "compilerPath": "/usr/bin/gcc",
            "cStandard": "c17",
            "cppStandard": "gnu++17",
            "intelliSenseMode": "linux-gcc-x64"
        }
    ],
    "version": 4
}

But visual code still yields an error:

image
image1923×716 142 KB

How can I fix this problem?

Thanks in advance!

2

I found that you should include “/torch/include/torch/csrc/api/include” before “/torch/include”