I was running into issues where even when a CUDA compiler was found, compilation would still fail because of missing include directories, and we didn't have a way to turn off CUDA compilation. This PR updates the build infrastructure to only compile the CUDA extensions when they are explicitly requested using -DJAX_FINUFFT_USE_CUDA=ON. It should be sufficient to set a CMAKE_ARGS environment variable when running pip install.
I was running into issues where even when a CUDA compiler was found, compilation would still fail because of missing include directories, and we didn't have a way to turn off CUDA compilation. This PR updates the build infrastructure to only compile the CUDA extensions when they are explicitly requested using
-DJAX_FINUFFT_USE_CUDA=ON. It should be sufficient to set aCMAKE_ARGSenvironment variable when runningpip install.Still to do: Update the README.