Cannot cmake Open3D-ML with PyTorch & CUDA 11.3

[raynehe]

Checklist

• [X] I have searched for similar issues.
• [X] I have tested with the latest development wheel.
• [X] I have checked the release documentation and the latest documentation (for master branch).

Steps to reproduce the issue

I followed the instructions on http://www.open3d.org/docs/release/compilation.html#ubuntu-macos and http://www.open3d.org/docs/release/compilation.html#ml-module to build Open3D from source:

git clone https://github.com/isl-org/Open3D
cd Open3D
util/install_deps_ubuntu.sh
mkdir build
cd build
cmake -DBUILD_CUDA_MODULE=ON \
      -DGLIBCXX_USE_CXX11_ABI=OFF \
      -DBUILD_PYTORCH_OPS=ON \
      -DBUNDLE_OPEN3D_ML=ON \
      -DOPEN3D_ML_ROOT=https://github.com/isl-org/Open3D-ML.git \
      -DPython3_ROOT=/home/rayne/anaconda3/envs/nf2/bin/python \
      ..

And the last sentence (cmake) introduce some errors.

Error message

The content of CMakeError.log is shown below:

Performing C SOURCE FILE Test CMAKE_HAVE_LIBC_PTHREAD failed with the following output:
Change Dir: /home/rayne/code/NeuroFluid/Open3D/build/CMakeFiles/CMakeTmp

Run Build Command(s):/usr/bin/make -f Makefile cmTC_ebb67/fast && /usr/bin/make  -f CMakeFiles/cmTC_ebb67.dir/build.make CMakeFiles/cmTC_ebb67.dir/build
make[1]: 进入目录“/home/rayne/code/NeuroFluid/Open3D/build/CMakeFiles/CMakeTmp”
Building C object CMakeFiles/cmTC_ebb67.dir/src.c.o
/usr/bin/cc -DCMAKE_HAVE_LIBC_PTHREAD  -fPIE -o CMakeFiles/cmTC_ebb67.dir/src.c.o -c /home/rayne/code/NeuroFluid/Open3D/build/CMakeFiles/CMakeTmp/src.c
Linking C executable cmTC_ebb67
/usr/local/bin/cmake -E cmake_link_script CMakeFiles/cmTC_ebb67.dir/link.txt --verbose=1
/usr/bin/cc CMakeFiles/cmTC_ebb67.dir/src.c.o -o cmTC_ebb67 
/usr/bin/ld: CMakeFiles/cmTC_ebb67.dir/src.c.o: in function `main':
src.c:(.text+0x3e): undefined reference to `pthread_create'
/usr/bin/ld: src.c:(.text+0x4a): undefined reference to `pthread_detach'
/usr/bin/ld: src.c:(.text+0x56): undefined reference to `pthread_cancel'
/usr/bin/ld: src.c:(.text+0x67): undefined reference to `pthread_join'
collect2: error: ld returned 1 exit status
make[1]: *** [CMakeFiles/cmTC_ebb67.dir/build.make:99：cmTC_ebb67] 错误 1
make[1]: 离开目录“/home/rayne/code/NeuroFluid/Open3D/build/CMakeFiles/CMakeTmp”
make: *** [Makefile:127：cmTC_ebb67/fast] 错误 2


Source file was:
#include <pthread.h>

static void* test_func(void* data)
{
  return data;
}

int main(void)
{
  pthread_t thread;
  pthread_create(&thread, NULL, test_func, NULL);
  pthread_detach(thread);
  pthread_cancel(thread);
  pthread_join(thread, NULL);
  pthread_atfork(NULL, NULL, NULL);
  pthread_exit(NULL);

  return 0;
}

Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.
Compiler: /usr/bin/nvcc 
Build flags: 
Id flags: --keep;--keep-dir;tmp;-gencode=arch=compute_86,code=sm_86 -v

The output was:
1
nvcc fatal   : Unsupported gpu architecture 'compute_86'


Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.
Compiler: /usr/bin/nvcc 
Build flags: 
Id flags: --keep;--keep-dir;tmp;-gencode=arch=compute_86,code=sm_86 -v

The output was:
1
nvcc fatal   : Unsupported gpu architecture 'compute_86'


Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.
Compiler: /usr/bin/nvcc 
Build flags: 
Id flags: --keep;--keep-dir;tmp;-gencode=arch=compute_86,code=sm_86 -v

The output was:
1
nvcc fatal   : Unsupported gpu architecture 'compute_86'


Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.
Compiler: /usr/bin/nvcc 
Build flags: 
Id flags: --keep;--keep-dir;tmp;-gencode=arch=compute_86,code=sm_86 -v

The output was:
1
nvcc fatal   : Unsupported gpu architecture 'compute_86'


Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.
Compiler: /usr/bin/nvcc 
Build flags: 
Id flags: --keep;--keep-dir;tmp;-gencode=arch=compute_86,code=sm_86 -v

The output was:
1
nvcc fatal   : Unsupported gpu architecture 'compute_86'


Compiling the CUDA compiler identification source file "CMakeCUDACompilerId.cu" failed.
Compiler: /usr/bin/nvcc 
Build flags: 
Id flags: --keep;--keep-dir;tmp;-gencode=arch=compute_86,code=sm_86 -v

The output was:
1
nvcc fatal   : Unsupported gpu architecture 'compute_86'

Open3D, Python and System information

- Operating system: Ubuntu 20.04
- Python version: Python 3.8.15
- Open3D version: (output from python: `print(open3d.__version__)`)
- System type: x86_64
- Is this remote workstation?: no
- How did you install Open3D?: build from source
- Compiler version (if built from source): GCC 11.2.0, cmake 3.20.0

Additional information

This is my conda list if helpful:

# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main  
_openmp_mutex             5.1                       1_gnu  
ca-certificates           2022.10.11           h06a4308_0  
certifi                   2022.9.24        py38h06a4308_0  
ld_impl_linux-64          2.38                 h1181459_1  
libffi                    3.4.2                h295c915_4  
libgcc-ng                 11.2.0               h1234567_1  
libgomp                   11.2.0               h1234567_1  
libstdcxx-ng              11.2.0               h1234567_1  
ncurses                   6.3                  h5eee18b_3  
openssl                   1.1.1s               h7f8727e_0  
pip                       22.2.2           py38h06a4308_0  
python                    3.8.15               h3fd9d12_0  
readline                  8.2                  h5eee18b_0  
setuptools                65.5.0           py38h06a4308_0  
sqlite                    3.39.3               h5082296_0  
tk                        8.6.12               h1ccaba5_0  
wheel                     0.37.1             pyhd3eb1b0_0  
xz                        5.2.6                h5eee18b_0  
zlib                      1.2.13               h5eee18b_0

Thank you for your help!

[ssheorey]

Hi @raynehe you can either switch to a more recent CUDA or specify a specific architecture with:

cmake -DCMAKE_CUDA_ARCHITECTURE="70-real" <other_cmake_options> ..

(if you have a Turing card, for example).