Repository to build wheels for openmm.
We are using github actions to build wheels. Wheels are found in the github actions artifacts section and when a release is made in this github repo, wheels are uploaded to the release as well.
Use ./rerender.sh
to rerender this feedstock
There are two wheels for each platform. Eg: for linux
OpenMM-8.1.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
openmm_cuda-8.1.1.12-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl
The first wheel contains all files except the CUDA plugins and the second wheel contains all the CUDA plugins.
When uploaded to PyPI, these wheels can be installed with
pip install openmm[cuda12]
or
pip install openmm[cuda11.2]
or
pip install openmm[cuda11.8]
To install without the CUDA plugins,
pip install openmm
version
and git_rev
in recipe/meta.yaml
../rerender.sh
.v8.1.1-0
../download_wheels.sh v8.1.1-0
.twine upload dist
../rerender.sh
.v8.1.1-1
.To get downstream packages as wheels, we need to do the following,
conda-forge
feedstock up and running.recipe/build_openmm*
, conda-forge.yml
, recipe/conda_build_config.yaml
, rerender.sh
, download_wheels.sh
from this repo.recipe/meta.yaml
's build
indicated by # START WHEEL CHANGES
.host
in recipe/meta.yaml
has no C++ shared dependencies on Linux.
To be compatible with manylinux
spec, we need to build C++ shared libraries
in recipe/build_openmm.sh
.openmm-cuda
. See recipe/0001-wheels.patch
for how to do that.recipe/build_openmm.sh
to update the lists of headers, libraries to include in each wheel.