[llvm] [NVPTX] Add support for atomic add for f16 type (PR #84295)

[Adrian Kuegel via llvm-commits]

akuegel wrote:

Tried to disable the instruction variant with an immediate argument, and that seems to work:

https://github.com/llvm/llvm-project/pull/85197

https://github.com/llvm/llvm-project/pull/84295