[CIR][CUDA] Register global variables

[AdUhTkJm]

The process is essentially the same as registering global functions.

This currently only works for __device__ and __shared__ variables. For textures and surfaces, another registration function will be needed. (We don't have support for these variables in CIR yet, and I have no idea what they really are.)

[bcardosolopes]

LGTM after minor nit

[bcardosolopes]

Good to land once the conflict is solved

[koparasy]

@AdUhTkJm are you ok with me fixing this and adding HIP support?

@bcardosolopes are you ok with having a single PR about CUDA/HIP. Or do you want these to be separate?

[bcardosolopes]

are you ok with having a single PR about CUDA/HIP. Or do you want these to be separate?

as long as you add tests I'm totally ok with it (it can even share the same file and just use different RUN lines)

[AdUhTkJm]

@AdUhTkJm are you ok with me fixing this and adding HIP support?

@bcardosolopes are you ok with having a single PR about CUDA/HIP. Or do you want these to be separate?

Yes of course. Thanks for your help!

[koparasy]

Pushed on a separate PR (#1980). Added tests for HIP and fixed a minor.

[koparasy]

@bcardosolopes OG has separate tests for HIP/CUDA. It will be really crowded in the LIT expressions. I will keep tests separate for now.

[bcardosolopes]

Thanks folks, closing this one in favor of #1980