WIP: Revamp CUDA support

[white238]

This removes the old deprecated FindCUDA and mimic's what @davidbeckingsale did for HIP support.

Breaking changes:

• cuda -> blt::cuda
• cuda_runtime -> blt::cuda_runtime
• CMAKE_LINK_WITH_NVCC -> BLT_CMAKE_LINK_WITH_NVCC (warning about former being deprecated)
• Removes Clang CUDA support (does anyone still use this @davidbeckingsale ?)

Thanks to @kennyweiss for sitting on a webex while I drastically rip support out from under myself!

[davidbeckingsale]

Does this need to be CACHE FORCE?

[white238]

I was thinking that too as I did it =) I'll test it out

[white238]

I reworked it to just populate the option()..

[rhornung67]

@white238 will this require a syntax change for BLT users like the recent change in HIP support; e.g., 'cuda_runtime' --> 'cuda::runtime'?

[white238]

@white238 will this require a syntax change for BLT users like the recent change in HIP support; e.g., 'cuda_runtime' --> 'cuda::runtime'?

Yes this is a breaking change in the following ways:

cuda -> blt::cuda
cuda_runtime -> blt::cuda_runtime
CMAKE_LINK_WITH_NVCC -> BLT_CMAKE_LINK_WITH_NVCC

[white238]

We should probably add this to the blt::cuda target:

https://cmake.org/cmake/help/latest/module/FindCUDAToolkit.html#cuda-driver-library

[victorapm]

Hi Chris, I'm having issues compiling hypre+umpire in shared mode. It seems this PR solves the issue

cannot find -lcuda_runtime: No such file or directory
  >> 1199    clang++: error: linker command failed with exit code 1 (use -v to see invocation)

[white238]

Hi Chris, I'm having issues compiling hypre+umpire in shared mode. It seems this PR solves the issue

cannot find -lcuda_runtime: No such file or directory
  >> 1199    clang++: error: linker command failed with exit code 1 (use -v to see invocation)

That's great to know. I was hoping to have time to revive this PR very soon. (honestly was hoping last year, it pains me how long this sat). The blueos machines were just retired and it makes it a good time to revamp the CUDA support w/o having to support that machine.

I'll bump it up my list. Hopefully its just fixing conflicts and testing it on a few codes.

[white238]

Relevent info:

https://nvidia.github.io/cccl/cccl/3.0_migration_guide.html