[python-torch-scatter] doesn't build

[4rozenwolves]

I got this error when building.

gcc -Wno-unused-result -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -march=x86-64 -mtune=generic -O3 -pipe -fno-plt -fexceptions -Wp,-D_FORTIFY_SOURCE=2 -Wformat -Werror=format-security -fstack-clash-protection -fcf-protection -g -ffile-prefix-map=/build/python/src=/usr/src/debug/python -flto=auto -ffat-lto-objects -march=x86-64 -mtune=generic -O3 -pipe -fno-plt -fexceptions -Wp,-D_FORTIFY_SOURCE=2 -Wformat -Werror=format-security -fstack-clash-protection -fcf-protection -g -ffile-prefix-map=/build/python/src=/usr/src/debug/python -flto=auto -march=x86-64 -mtune=generic -O3 -pipe -fno-plt -fexceptions -Wp,-D_FORTIFY_SOURCE=2 -Wformat -Werror=format-security -fstack-clash-protection -fcf-protection -g -ffile-prefix-map=/build/python/src=/usr/src/debug/python -flto=auto -march=native -O2 -pipe -fno-plt -fexceptions -Wp,-D_FORTIFY_SOURCE=2 -Wformat -Werror=format-security -fstack-clash-protection -fcf-protection -fPIC -DWITH_PYTHON -DWITH_CUDA -Icsrc -I/usr/lib/python3.10/site-packages/torch/include -I/usr/lib/python3.10/site-packages/torch/include/torch/csrc/api/include -I/usr/lib/python3.10/site-packages/torch/include/TH -I/usr/lib/python3.10/site-packages/torch/include/THC -I/opt/cuda/include -I/usr/include/python3.10 -c csrc/cpu/scatter_cpu.cpp -o build/temp.linux-x86_64-cpython-310/csrc/cpu/scatter_cpu.o -O3 -Wno-sign-compare -DAT_PARALLEL_OPENMP -fopenmp -DTORCH_API_INCLUDE_EXTENSION_H -DPYBIND11_COMPILER_TYPE=\"_gcc\" -DPYBIND11_STDLIB=\"_libstdcpp\" -DPYBIND11_BUILD_ABI=\"_cxxabi1017\" -DTORCH_EXTENSION_NAME=_scatter_cuda -D_GLIBCXX_USE_CXX11_ABI=1 -std=c++17
/opt/cuda/bin/nvcc -DWITH_PYTHON -DWITH_CUDA -Icsrc -I/usr/lib/python3.10/site-packages/torch/include -I/usr/lib/python3.10/site-packages/torch/include/torch/csrc/api/include -I/usr/lib/python3.10/site-packages/torch/include/TH -I/usr/lib/python3.10/site-packages/torch/include/THC -I/opt/cuda/include -I/usr/include/python3.10 -c csrc/cuda/scatter_cuda.cu -o build/temp.linux-x86_64-cpython-310/csrc/cuda/scatter_cuda.o -D__CUDA_NO_HALF_OPERATORS__ -D__CUDA_NO_HALF_CONVERSIONS__ -D__CUDA_NO_BFLOAT16_CONVERSIONS__ -D__CUDA_NO_HALF2_OPERATORS__ --expt-relaxed-constexpr --compiler-options '-fPIC' -O3 --expt-relaxed-constexpr -DTORCH_API_INCLUDE_EXTENSION_H -DPYBIND11_COMPILER_TYPE=\"_gcc\" -DPYBIND11_STDLIB=\"_libstdcpp\" -DPYBIND11_BUILD_ABI=\"_cxxabi1017\" -DTORCH_EXTENSION_NAME=_scatter_cuda -D_GLIBCXX_USE_CXX11_ABI=1 -gencode=arch=compute_75,code=compute_75 -gencode=arch=compute_75,code=sm_75 -std=c++17
/usr/include/pybind11/detail/../cast.h: In function ‘typename pybind11::detail::type_caster<typename pybind11::detail::intrinsic_type<T>::type>::cast_op_type<T> pybind11::detail::cast_op(make_caster<T>&)’:
/usr/include/pybind11/detail/../cast.h:45:120: error: expected template-name before ‘<’ token
   45 |     return caster.operator typename make_caster<T>::template cast_op_type<T>();
      |                                                                                                                        ^
/usr/include/pybind11/detail/../cast.h:45:120: error: expected identifier before ‘<’ token
/usr/include/pybind11/detail/../cast.h:45:123: error: expected primary-expression before ‘>’ token
   45 |     return caster.operator typename make_caster<T>::template cast_op_type<T>();
      |                                                                                                                           ^
/usr/include/pybind11/detail/../cast.h:45:126: error: expected primary-expression before ‘)’ token
   45 |     return caster.operator typename make_caster<T>::template cast_op_type<T>();
      |                                                                                                                              ^
error: command '/opt/cuda/bin/nvcc' failed with exit code 1

[4rozenwolves]

I managed to compile it when I downgrade cuda to 11.8.0 instead of 12.1.0 with cudnn to 8.6 instead of 8.8. The same error also happen with python-torch-sparse and is solved similarly.

[acxz]

Interesting. Clever find. Let's keep this issue open so that it has more visibility or until it gets fixed upstream to work with later cuda versions.

[4rozenwolves]

Another solution to build python-torch-scatter when using cuda 12.1 is to change to gcc-11 and g++-11 (doesn't work with gcc-12). It can be done by using sudo ln -sf /usr/bin/gcc-11 /opt/cuda/bin/gcc and sudo ln -sf /usr/bin/gcc-11 /opt/cuda/bin/gcc

[acxz]

Looks like updates to the dependency stack has resolved any build issues, as I can build just fine.