Unable to step into torch.nn.functional.linear using VS Code debugging

Question

I want to step into the linear function using VS Code's step-in , but it skips automatically when I click "step into". Could anyone help me with this? I used DEBUG=1 when compiling PyTorch.

I can't step into the linear function using either python launch or gdb attach.

However, strangely, if I set a breakpoint on the line const KernelFunction& kernel = op.operatorDef_->op.lookup(dispatchKeySet); in the directory /home/sss/WORK/PYTORCH/pytorch/aten/src/ATen/core/dispatch/Dispatcher.h, I can step into or enter that line.

It doesn't work without the breakpoint. Is this normal? Is there any way to solve this?

my test code:

import torch
import torch.nn.functional as F
x = torch.tensor([[1.0, 2.0]]).to('cuda')
weight = torch.tensor([[0.5, 0.5], [0.5, 0.5]]).to('cuda')
y = F.linear(x, weight)

My launch.json:

{
    "version": "0.2.0",
    "configurations": [
        {
            "name": "Python: Current File",
            "type": "python",
            "request": "launch",
            "program": "/home/sss/WORK/PYTORCH/test.py",
            "console": "integratedTerminal",
            "justMyCode": false ,
        },
        {
            "name": "(gdb) Attach to Process",
            "type": "cppdbg",
            "request": "attach",
            "program": "/home/sss/WORK/VLLM/vllm/.venv/bin/python",
            "processId": "${command:pickProcess}",
            "MIMode": "gdb",
            "setupCommands": [
                {
                    "description": "Enable pretty-printing for gdb",
                    "text": "-enable-pretty-printing",
                    "ignoreFailures": true 
                }
            ]
        }
    ],
    "compounds": [
        {
            "name": "Python + C++ (Attach)",
            "configurations": ["Python: Current File", "(gdb) Attach to Process"]
        }
    ]
}

MinhNH · Accepted Answer · 2025-11-12 10:33:38Z

0

You can't be cause under the hood, the calculation is now out of Python scope (it's dispatched to C++ or CUDA code for actual execution), therefore the Python Debugger cannot catch what the function is doing.

On the other hand, if you set a breakpoint in C++ code, the C++ debugger now catch it and you can enter the code. This is because GDB (C++ debugger) does not actively catch all native call events. It's like in Python if you use step over it will run all the lines after and don't listen to what they do until it hit another breakpoint or an exception.

edited Nov 12 at 10:33

answered Nov 12 at 10:24

MinhNH

5744 silver badges11 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Unable to step into torch.nn.functional.linear using VS Code debugging

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related