Alpaka container and algorithm library

[erikzenker]

A container somehow similar to the thrust::device vectors would reduce are code base again and should make algorithms on its memory more safe. For HASEonGPU we only need a container equal to std::array. Therefore no dynamic size increase is necessary.

For the algorithm side, we need a reduce and exclusive scan/prefix sum algorithm. Based on alpaka buffers and a wrapper for the container would be perfect :smile_cat:

I would suggest to create a separate repository for this containers and algorithms. Name suggestions ? :tiger:

[slizzered]

Is it possible to use thrust directly as the implementation for the CUDA backend? Or would that possibly interfere with the use of alpaka buffers?

As for naming suggestions: Since it emulates the "look and feel" of the STL, what about "Alpaka Standard Template Library [ASTL]", with a namespace alp::

[erikzenker]

I am still not sure about using thrust together with alpaka:

• Thrust itself does only support a single accelerator at a time
• Thrust could use an alpaka backend, but how much work is that ?
• Using thrust would be only a temporary solution in my eyes

we could rewrite thrust containers and algorithms we need to alp.

[erikzenker]

Alp sounds create, what do you think about creating a gist and collecting the interface.