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

[Nikolay Panchenko via llvm-commits]

npanchen wrote:

> Looks like we need to add the following:
> 
> ```
>   setTruncStoreAction(MVT::v2f32, MVT::v2f16, Expand);
>   setTruncStoreAction(MVT::v2f32, MVT::v2bf16, Expand);
> ```
> 
> Otherwise the fp_round will get folded into the store which we can't handle.

Yes, I finally got a chance to debug it and realized there's incorrect handling of truncating stores for a long period of time. Will you(@AlexMaclean) or @Artem-B or @Prince781 create a fix ?

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