Test mdpsan for handling memory

alpaka-group / vikunja

Vikunja is a performance portable algorithm library that defines functions operating on ranges of elements for a variety of purposes . It supports the execution on multi-core CPUs and various GPUs. Vikunja uses alpaka to implement platform-independent primitives such as reduce or transform.

https://vikunja.readthedocs.io/en/latest/

Mozilla Public License 2.0

14 stars 5 forks source link

Test mdpsan for handling memory #91

Open SimeonEhrig opened 1 year ago

SimeonEhrig commented 1 year ago

The PR tests mdspan as alternative for raw pointers to handle the input and output memory.

Benefits of mdspan:

bundling of memory and size in one object
native support for multidimensional memory
handling of objects instead pointer makes the code easier
slicing of memory

The PR requires the mdspan support, which is not merged yet: https://github.com/alpaka-group/alpaka/pull/1788

Is solving the same problem like in PR #88 with another approach.

SimeonEhrig commented 1 year ago

@bernhardmgruber ping

bussmann commented 1 year ago

I think any tensor example would fit extremely well for mdspan. Tensors are THE thing for machine learning.

bernhardmgruber commented 1 year ago

I think any tensor example would fit extremely well for mdspan. Tensors are THE thing for machine learning.

I fully agree that mdspan is a good fit to represent tensors. I am unsure though whether tensors fit into vikunja. We can definitely have a tensor example in alpaka though. Can you please open an issue at alpaka and also suggest a concrete algorithm? And ideally assign a professional support to implement it! Thx!