When can we expect reductions to work with the CUDA backend?

[mfbalin]

When I use sycl::ONEAPI::reduction with the CUDA backend, I get the following error:
ptxas fatal : Unresolved extern function '_Z17__spirv_GroupIAddjjm'

[bader]

Tagging @Pennycook.

[Pennycook]

_Z17__spirv_GroupIAddjjm is one of the functions from the GroupAlgorithms extension, which the Reduction extension builds upon. As soon as the GroupAlgorithms extension is supported, the Reduction extension should (hopefully) start working automatically.

GroupAlgorithms is next on my list of things to look at after I get SubGroups working. We're still sketching out what the implementation should look like, so I'm not sure how long it will take to get working. But I'll post an update here when I have a better idea.

[Pennycook]

Ok, I'm hopeful that we can get this working within a few weeks. I have a prototype passing the tests for 32-bit types, and just need to spend a little longer working out the details for 64-bit types.

[Pennycook]

I'm apparently really bad at predicting how long things will take.

#2647 enables reductions with the CUDA backend. Please take a look and let me know if you encounter any issues with it.