[llvm] [NVPTX] Lower bfloat16 add/mul/sub as fma on SM80 (PR #121065)

[Justin Fargnoli via llvm-commits]

https://github.com/justinfargnoli commented:

The purpose of this PR is to eliminate the conversions to/from `fp32`, correct? If so, can you add that to the description? 

https://github.com/llvm/llvm-project/pull/121065