I am a novice in CUDA. I want to debug and inspect a project with Python (PyTorch + Pybind) & CUDA integrated on an RTX 3090 Windows 11 device. I tried to use Visual Studio and finally managed to debug the Python and C++ parts. However, for the CUDA part, although I specified the executable (python.exe in the conda environment folder) and arguments in the GUI, it gets stuck because the conda environment is not activated. I managed to use the command line (with the environment activated following Nsight Systems CLI (nsys.exe) Program not found:) to generate the report. However, I still don’t know how to use Nsight Systems/VSE (Nsight User Properties) to profile the Python project. Can somebody teach me how to do that? Thanks!
Hi, @xrr233333
For Nsight System, you can use the CLI tool to get the report, and then open in Nsight System UI.
For Nsight VSE, it is not profile tool, it is for debug.
Yes, I hope to both view the performance result and debug, and previously I was using CLI for generating report. As for debugging, I found a way to debug by using Nsight VSCE and attaching cppvsdbg on the running python program to enable debugging C++ code inside *.cu file. However, it seems that on Windows system VSCE is not able to do CUDA debugging by seeing inside the kernels.
I was here just wanting to find a set of solution that enabling python/CUDA intergrated project debugging on Windows 11 OS, preferably with GUI operations. But current solution is already enough for me to check the project. Just write my solution here to benifit the later comers.
Hi, @xrr233333
Thanks for the reply and good to know our develop tools work for you now.
This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.