[llvm] [NVPTX] support packed f32 instructions for sm_100+ (PR #126337)

[Alex MacLean via llvm-commits]

https://github.com/AlexMaclean commented:

Left some comments but will need to review more thoroughly. This change is quite large and complex, would it be possible to break it up in any way? Such as adding some of the DAG combine rules in a separate change? 

https://github.com/llvm/llvm-project/pull/126337