Nsight version deprecation? NVTX3 errors.

[ha-ves]

Describe the bug Nsight breaking changes.

The implementation of NVTX from Cuda 11.x to Cuda 12.x seem to have been changed. Compilation on Cuda 12.x (seem like >12.2) failed with NVTX errors.

---

To Reproduce Steps to reproduce the behavior. If the code is not attached and cannot be reproduced easily, the bug report will be closed without any comments.

Run py setup.py install

---

Expected behavior A clear and concise description of what you expected to happen.

Successful installation after suggested changes in #607

---

Desktop (please complete the following information):

• OS: Ubuntu 22.04 on WSL2
• Python version: 3.10.11
• Pytorch version: 2.5.1+cu124
• CUDA version: 12.6
• NVIDIA Driver version: 561.17
• Minkowski Engine version 02fc608
• Output of the following command. (If you installed the latest MinkowskiEngine, paste the output of python -c "import MinkowskiEngine as ME; ME.print_diagnostics()". Otherwise, paste the output of the following command.)

wget -q https://raw.githubusercontent.com/NVIDIA/MinkowskiEngine/master/MinkowskiEngine/diagnostics.py ; python diagnostics.py

==========System========== Linux-5.15.167.4-microsoft-standard-WSL2-x86_64-with-glibc2.35 DISTRIB_ID=Ubuntu DISTRIB_RELEASE=22.04 DISTRIB_CODENAME=jammy DISTRIB_DESCRIPTION="Ubuntu 22.04.5 LTS" 3.10.12 (main, Nov 6 2024, 20:22:13) [GCC 11.4.0] ==========Pytorch========== 2.5.1+cu124 torch.cuda.is_available(): True ==========NVIDIA-SMI========== /usr/lib/wsl/lib/nvidia-smi Driver Version 561.17 CUDA Version 12.6 VBIOS Version 90.17.31.00.23 Image Version G001.0000.02.04 GSP Firmware Version N/A ==========NVCC========== /usr/local/cuda-12.6/bin/nvcc nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2024 NVIDIA Corporation Built on Tue_Oct_29_23:50:19_PDT_2024 Cuda compilation tools, release 12.6, V12.6.85 Build cuda_12.6.r12.6/compiler.35059454_0 ==========CC========== /usr/bin/c++ c++ (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 Copyright (C) 2021 Free Software Foundation, Inc. This is free software; see the source for copying conditions. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

==========MinkowskiEngine========== MinkowskiEngine not installed

---

Additional context Add any other context about the problem here.

Seem this repository is not updated.

For now you need to manually reimplement the NVTX events/markings on this library, or just disable it with -DNVTX_DISABLE

[RivettiLuciano]

sorry how did you handle the NVTX errors? I followed https://github.com/NVIDIA/MinkowskiEngine/issues/601 but I still have the nvtx3 errors.

[ha-ves]

@RivettiLuciano, sorry, I decided not to pursue this path and dropped the project.

Haven't looked into the code yet, but assuming Visual Studio--and by extension, IntelliSense-- can point out the errors, you may be able to reimplement the NVTX events/markings.

NVTX is telemetry for profiling and such, if you don't need it, use the -DNVTX_DISABLE variable.

[RivettiLuciano]

Hi @ha-ves I tried adding to the setup.py

extra_compile_args = { "cxx": [ARGS, "-DNVTX_DISABLE"] }

but it doesnt work i still have the errors with nvtx3

[ha-ves]

@RivettiLuciano I believe the flag/variable is for nvcc, try put it in NVCC_FLAGS

[RivettiLuciano]

@ha-ves that was correct. I could install Monkoski thank you very much for your answer

[Xin200203]

Thank you all for the pointers!

I was finally able to work around the NVTX3 header collisions by completely disabling the cuDF NVTX code paths at compile time. In my setup.py, I added the NVTX_DISABLE macro both as a define and in the compiler flags:

ext_modules = [
    Extension(
        name="MinkowskiEngineBackend._C",
        sources=[…],
+       define_macros=[('NVTX_DISABLE', None)],
        extra_compile_args={
            'cxx': CC_FLAGS + ['-DNVTX_DISABLE'],
            'nvcc': NVCC_FLAGS + ['-DNVTX_DISABLE'],
        },
        libraries=libraries,
    ),
]

After cleaning and reinstalling:

python setup.py clean --all
python setup.py install

the build succeeds without any more ambiguous domain/category errors.

Hope this helps anyone else hitting the same CUDA 12.x + NVTX3 clash!

[linQian99]

Thank you all for the pointers!

I was finally able to work around the NVTX3 header collisions by completely disabling the cuDF NVTX code paths at compile time. In my setup.py, I added the NVTX_DISABLE macro both as a define and in the compiler flags:

ext_modules = [ Extension( name="MinkowskiEngineBackend._C", sources=[…],

•   define_macros=[('NVTX_DISABLE', None)],
    extra_compile_args={
        'cxx': CC_FLAGS + ['-DNVTX_DISABLE'],
        'nvcc': NVCC_FLAGS + ['-DNVTX_DISABLE'],
    },
    libraries=libraries,
  

  ), ] After cleaning and reinstalling:

python setup.py clean --all python setup.py install the build succeeds without any more ambiguous domain/category errors.

Hope this helps anyone else hitting the same CUDA 12.x + NVTX3 clash!

Thank u mate! Really helpful!