Publish the newest releases on conda-forge (for Python)

[ogrisel]

This would make it easy to use Intel GPUs via OpenCL in Python via dpnp / dpctl /numba-dpex.

The last few release have had failed builds and nobody fixed them:

• https://github.com/conda-forge/intel-compute-runtime-feedstock/pulls

I think it would be great if at least one maintainer of intel/compute-runtime would request to be registered as a recipe maintainer on conda-forge/intel-compute-runtime-feedstock in addition to @isuruf to adjust the build recipe on the conda-forge CI when needed.

[ogrisel]

Related issue for intel-level-zero on conda-forge: https://github.com/oneapi-src/level-zero/issues/98

[AdamCetnerowski]

There are no plans for proactive enabling and distro adoption except for Ubuntu. We are open to accepting contributions (pull requests) that unblock support.

[ogrisel]

I think not providing a way to install all the runtime dependencies for packages such as dpctl, dpnp and numba_dpex to users without sudo rights is a very strong impediment for adoption of the Intel xPU technology stack in the Python datascience ecosystem.

Publishing such packages on conda channels such as intel would be an improvement but it would even be better to directly contribute to maintain packages on the main conda-forge channels or on pypi.org to ease dependency management of third party library developers that would like to ship software on this stack.

There are no plans for proactive enabling and distro adoption except for Ubuntu.

BTW: #588 was also closed and was specifically about installation/packaging problems on Ubuntu.

We are open to accepting contributions (pull requests) that unblock support.

I am already maintaining way too many open source packages myself, I do not have the bandwidth to handle the packaging of a system level libraries I only interact with via higher level components such as the aforementioned packages.