Open jan-janssen opened 1 year ago
You'll need the rocm-smi-lib packaged first. Since it's an open source library, it's fine to link it here.
I started a corresponding pull request in https://github.com/conda-forge/staged-recipes/pull/23288
@isuruf Unfortunately, I am still not sure how to move on. Can we have an rcom* build just like we have cuda and cpu_ builds? Can this be achieved with a conda_build_config.yaml
file?
No need to have a third variant. We can add rocm support for the existing builds. It's a small, open source library, so I'm fine with linking to it. It would be better if there was a rocm-smi-libs output that is just the library without the cli tool.
Comment:
As the first Exascale computer Frontier uses AMD GPUs I would like to compile a libhwloc version with RMSI support. But I am not sure how to set the priorities for this one. I feel for Nvidia Cuda there is already a lot of support on conda-forge but I am not aware of a similar selector mechanism for AMD GPUs. @isuruf Can you help me out here?