Thomas Grützmacher

format-rebase!

Personally, I prefer the suffix `_kernel`, so it is clear at every call and by looking at the function directly (namespace would require to look at the neighborhood of the...

@tcojean True, I guess my search was not thorough enough. Updated the issue description.

@pratikvn Really, where? Every part is inside a namespace `kernels`, after which a namespace `omp`, `reference` or `cuda` comes. But I did not find any function encapsulated inside a namespace...

The question is how we should encapsulate actual CUDA kernels (that have to be called with ``) in order to associate them with the global function. For example if you...

We might actually need the namespace, so we can also put the `__device__` functions in there.

While investigating a bit more, we need to change multiple parts of our implementation if we want to properly support complex: 1. `Dense::compute_norm2` should put the result into a `Dense`...

I agree, making everything a `signed` value will definitively remove the need for a lot of casts. For sizes, I think you are correct that we can give up the...

rebase!

I am not 100% happy with the way I deal with the complex `next_krylov` norm computation (currently, I need additional storage for that). I am looking into alternative ways, but...