Tweak backend selection

[christiangnrd]

This method will be more resilient for if/when we support broadcasted containers and not just AbstractArrays.

[anicusan]

Hmm, I like the idea, but it'd still call get_backend on inputs; we wanted to avoid that when backend is explicitly specified, so that unmaterialised arrays (e.g. UnitRange) would work in GPU kernels.

A bit of a nitpick, but could we rename use_KA_algo to something fully snake_case to be consistent with the rest of the library? Perhaps use_gpu_backend? It can technically work on CPUs via PoCL, but it still goes through the GPUCompiler infrastructure, which is the main thing here.

Also, is there a way to have the buildkite badges in the README point to the latest merged commit? No problem if not.

[christiangnrd]

I misread and accidentally changed it to use_gpu_algo instead of use_gpu_backend. I like it but happy to change it if you want.

It now takes in the provided backend.

Also, is there a way to have the buildkite badges in the README point to the latest merged commit? No problem if not.

I have no idea there.

[christiangnrd]

Also, is there a way to have the buildkite badges in the README point to the latest merged commit? No problem if not.

@anicusan I found the answer while doing some unrelated troubleshooting. Adding ?branch=main to the badge url seems to be what you're looking for.