Alexis Montoison

I propose the names `GESYRK` and `GEHERK`. GEMMT is like a generalized SYRK / HERK.

@abelsiqueira It seems that we have too many functions in `reference.md`. Do you have an idea how we can solve this issue?

It's more generic if you add an argument with the storage type `S`. It will also work with other GPU backends whereas your current implementation is restricted to Nvidia GPUs.

> > And is it possible to remove CUDA from the deps? I can only see 3 lines where you use CUDA functions, maybe you could add something for the...

> > Then I would advice not adding CUDA in the dependencies. > > We used the following multiple times in this package: > > > https://github.com/JuliaSmoothOptimizers/LinearOperators.jl/blob/a2043722a09efdf64ed89a6d4912606a9c419ab9/src/constructors.jl#L28 > > It...

Paul, could you test with the branch master of CUDA.jl? You just need to do a modification similar to https://github.com/JuliaSmoothOptimizers/Krylov.jl/blob/main/.buildkite/pipeline.yml#L33

> > I still think that removing CUDA entirely is a better solution. As mentionned by @amontoison you would also be able to use other GPU backends. You could add...

The issue is that I didn't understand what you want to achieve. Your previous message helped me to understand. Your example has a big drawback, you will have a different...

Can you try with this precompiled [metis](https://github.com/JuliaBinaryWrappers/METIS_jll.jl/releases/download/METIS-v5.1.2%2B0/METIS.v5.1.2.x86_64-linux-gnu.tar.gz)?

If you compile SPRAL with OpenBLAS, you need different link flags for generating the shared library: ```shell gfortran -fPIC -shared -Wl,--whole-archive libspral.a -Wl,--no-whole-archive -lgomp -lopenblas -lhwloc -lmetis -lstdc++ -o libspral.so...