Marko Kabić

Hi Leopold, I think this is a great idea, would be certainly nice to have it. Although I don't have much experience with writing conda packages. Would you maybe like...

Would we have to make a conda package for each COSMA dependency (?), which are basically the following: - [COSTA](https://github.com/eth-cscs/COSTA): the algorithm for transposing and transforming distributed matrices between different...

Hi @ltalirz , sorry for keeping you wait -- I cannot find the PR for a conda package, can you please send me this again?

Hi @ltalirz, do you maybe know what's the problem with: https://github.com/conda-forge/staged-recipes/pull/15073 ?

Since @gsitaraman mentioned [here](https://github.com/cp2k/cp2k/pull/1557#issuecomment-860751137) the potential performance benefit of using rocBLAS, I would still hesitate to switch to hipBLAS, unless you believe it's absolutely necessary.

This should be possible. @AdhocMan, do you think this would be a simple change, or would it require adding a lot of boilerplate code to the [gpu-backend](https://github.com/kabicm/Tiled-MM)?

Would we still have to add hipBLAS as a dependency of COSMA in that case?

As written [here](https://sep5.readthedocs.io/en/latest/ROCm_Tools/rocblaswiki.html) it seems the cmake for the CUDA backend of rocBLAS is broken at the moment: > rocBLAS is written with HiP kernels, so it should build and...

Hi @yaoyi92! I am not sure if we already discussed this by email? COSMA does not have any wrapper around `MPI_Finalize` and in general, COSMA does not have a "initialize/finalize"...

Is the problem you mentioned still present with the latest version: [COSMA-v2.6.1](https://github.com/eth-cscs/COSMA/releases/tag/v2.6.1)?