Tobias Ribizel

I see. Yes, I meant launching a single SpMV per column instead of one large SpMM for multiple columns. Whether you can get some benefit/concurrent execution out of it somewhat...

I believe we didn't talk about the actual error reporting so far, I could imagine multiple approaches: 1. `validate_data()` doesn't return anything and throws an exception with information on which...

This is also slightly related to the "double-destroy" segfaults we had with #418, since the `automatical` strategy is default-constructed with a `CudaExecutor`

also, what you are reporting is not the residual norm, but the square of the energy norm $||x||_A$ (since it has the suffix `_sq`)

Thanks for the details. On your first question: You are using a relative residual norm stopping criterion, so we are looking at $||b - Ax||_2 / ||b||_2$ by default. If...

This is really impressive work. I think there are probably better ways to make the interactions with factories more smooth, but in general, this sounds really useful. Before I continue...

I think you may have misunderstood my example a bit. This is only meant to be a single configuration for a specific LinOpFactory. An application of course needs to be...

Coming back to this, since @thoasm and I discussed some related issues or questions about code style, I want to propose a few changes that might clean up our code...

Yes, that would be a long-term goal. Maybe we can collect other changes/issues that this switch would cause. Until then, I'll try to use static_cast where possible

After working on #464, I now have a slightly clearer picture of what the impact of the proposed changes would be, so my updated proposal would be: 1. Make `size_type`...