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

[Princeton Ferro via llvm-commits]

Prince781 wrote:

@npanchen it looks like there's a similar thing at the LLVM IR level ([`@llvm.arithmetic.fence`](https://llvm.org/docs/LangRef.html#llvm-arithmetic-fence-intrinsic)) you could also look into, although I haven't tried it myself.

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