CUDA configuration with Kepler and >=cuda-11 fails in vtkm
Hello,
when configuring the VTK project, version 9.0.1, with a Kepler CUDA compute arch using >=cuda-11 cmake isn't able to finish this step properly. In ThirdParty/vtkm/vtkvtkm/vtk-m/CMake/VTKmDeviceAdapters.cmake
both, compute_30 and compute_35 arches are set and passed to the --generate-code
flag for nvcc.
However, starting with CUDA-11, the compute_30 arches are no longer supported, so the configuration fails. For Gentoo, I found the attached patch to solve this issue (see also https://github.com/gentoo/gentoo/blob/b4515ef3e96a9e3cdfe5723141fe0a0ca5fa4d56/sci-libs/vtk/files/vtk-9.0.1-0001-fix-kepler-compute_arch-if-CUDA-toolkit-11-is-used.patch).
I haven't tested with the master branch, but the code hasn't changed, so I assume, the issue will be present for current HEAD too.
vtk-9.0.1-0001-fix-kepler-compute_arch-if-CUDA-toolkit-11-is-used.patch