* Update Python and CUDA
* Update GLIBC
* Debug 1
* Add sysroot 2.17
* Clean up Linux dependencies
* Fix OpenCL tests
* Update CUDA installation
* Fix CUDA version
* Try to fix CUDA installation
* Debug
* Rename env file
* Don't set NVCC version
* Set NVCC version
* Clean up
* Fix GCC version
* Clean up
* Update tests
* Exclude the CUDA tests
* Fix
* Set names
* Set PyTorch version
* Fix
* Add the latest versions
* Remove an obsolete script
* Try to fix MacOS
* Downgrade to CUDA 11.2
* Reduce verbosity
* Debug MacOS
* Show test results on the failure
* Clean up MacOS dependencies
* Clean up
* Use MacOS 11
* Set the PyTorch version on MacOS
* Set PyTorch to 1.9 on MacOS
* Clean up
* Properly set the PyTorch version for MacOS
* Remove unused flags
* Remove more unused flags
* Remove more unused flags
* add CI
* point to pytorch install
* fix vars
* fix vars again
* create symlink
* sudo
* add mesa
* Retry
* make sure we only use conda-forge
* pin pytorch
* try other env config for ubuntu
* try other way for cuda
* try cmake_args
* patch envs with versions
* export CUDA_PATH too?
* add nvcc metapkg
* restraint nvcc too
* link to stubs
* make CUDA required (temporarily)
* debug libcuda.so
* symlink libcuda stub
* fix symlink
* try luck with library paths
* try cmake_prefix_path
* fix python paths
* try different env vars
* try like this instead
* use extra_compile_flags
* summarize skipped tests too
* Clean up a bit
* update readme
* try pocl for tests
* debug opencl platforms
* more debugging
* pytorch >= 1.7
* fix typo
* explicit pytorch gpu
* rename jobs
* workaround with custom flag
* alternative attempt
* do not use cmake_args
* is it opencl's fault?
* is it openmm?
* openmm _on linux_
* remove nvidia's opencl
* try pocl again
* force pytorch gpu
* build opencl again
* fix pytorch req
* plugin info
* try ld_library_path
* add python tests too
* fix ld path
* fix pytest call
* export ld for python tests too
