Simon Frasch

hipBLAS is just an additional layer on top of rocBLAS and you can compile rocBLAS with cuda. So it should work already, although I haven't tested it.

No, if rocBLAS works correctly with cuda. There would just be few changes to the CMake config required.

That's unfortunate, it would have made things easier. I think, adding hipBLAS alongside rocBLAS does not make much sense, since adding a code path, that's only used for testing somewhat...

I already have a version of Tiled-MM with hipblas instead of rocblas: https://github.com/AdhocMan/Tiled-MM/tree/hipblas It compiles fine, I just can't run it at the moment because of an issue with our...

The only reason I can think of is NVCC, since switching to C++17 will require CUDA 11 or later.

I just came across desribing a model using a devlop branch: [https://nvie.com/posts/a-successful-git-branching-model/](https://nvie.com/posts/a-successful-git-branching-model/) This might be a good workflow to follow.

The idea is to provide the building blocks for such cases, without having to extent arbor to use otherwise unnecessary information like a global cell position. So assuming the user...

> 1. I think the docs need some love, especially the motivation and a more prominent first example in that section. > How about contrasting the 'old' way and the...

> @AdhocMan How's the feature going? If you're stuck, please let me know, so we can get you going again :) I don't have as much time as I was...

I've updated the description to reflect to current state. It's not quite ready for a full review, but some feedback on the new general concept would be welcome.