GVigne

Thanks for the examples! Here is what I came up with to solve them: ``` basetype(T::Type) = T.name.wrapper basetype(T::UnionAll) = basetype(T.body) basetype(x::Union) = Union{basetype(x.a), basetype(x.b)} ``` Can you think of...

I understand what you are saying. Encouraging package developers to use a fragile function can only lead to bad surprises. Here is the issue I'm facing and trying to solve...

I am aware of `similar` and am currently using it. However, it only works if the way the array is built is GPU-compatible. Sometimes, we just want to build an...

We figured out a much simpler way to do what we wanted. We simply added the array type in one of our structure as a parametric type, so it's very...

Unfortunately, it doesn't seem to work, but for a different reason. When running the script in the REPL, the result is computed accordingly, but the warning "Performing scalar indexing" is...

Ok, I have a different error now, but I still can't really understand it. The 5-arg `mul!` works correctly when using only CuArrays.  At first I thought it came...

Sorry, I wasn't clear. I am using the same example as above, that is `X = mortar((CUDA.randn(5,1),CUDA.randn(5,1)))` `Y = mortar((CUDA.randn(5,1),CUDA.randn(5,1)))`

Alright, I think I figured out the overall issue. Replacing `muladd!` by the 5-argument `mul!` isn't enough because the array holding the result is not Cuda compatible. Modifying `similar` at...

Well, we could always remove [the check for the threadid](https://github.com/JuliaMolSim/DFTK.jl/blob/d3a2963b34f16491eb6aa213d59b8f0d0929f1f5/src/terms/Hamiltonian.jl#L148-L149), and we would be merging all timers into the main one. But I think @mfherbst said there were a few...

I just stumbled upon this PR: it looks great! It implements a feature I was talking about in issue #1572, which is a support for LinearAlgebra's `eigen` function. Just a...