As discussed in the last developer's meeting, and as highlighted in #1360, we need some form of CI of the GPU. The 2 options are:
-
Setup some crawler program that regularly pulls and tests master on a local machine (e.g. at EPFL), and warn us in case of failure
-
Use the CSCS infrastructure (AMD and NVIDIA devices available)
Option 2 has the advantage of using an existing pipeline (also used in some Julia projects, e.g. here), but ties DFTK to an external organization.
The details of how to implement option 1 are not as clear, but it insures independence in the long term.
I do not have a strong preference. Any comment is welcome.
As discussed in the last developer's meeting, and as highlighted in #1360, we need some form of CI of the GPU. The 2 options are:
Setup some crawler program that regularly pulls and tests master on a local machine (e.g. at EPFL), and warn us in case of failure
Use the CSCS infrastructure (AMD and NVIDIA devices available)
Option 2 has the advantage of using an existing pipeline (also used in some Julia projects, e.g. here), but ties DFTK to an external organization.
The details of how to implement option 1 are not as clear, but it insures independence in the long term.
I do not have a strong preference. Any comment is welcome.