[llvm] [NVPTX] Add support for f32x2 mixed-precision add/sub (PR #221957)

[Srinivasa Ravi via llvm-commits]

Wolfram70 wrote:

> One question, another nit: do we need a negative tests when sm_107f isn't targeted, or does this pattern occur in other tests?

Sorry, I missed this bit. That makes sense, added another test to make sure the pattern is folded only on `sm_107f`.

https://github.com/llvm/llvm-project/pull/221957