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

[Srinivasa Ravi via llvm-commits]

Wolfram70 wrote:

> I wonder if https://github.com/llvm/llvm-project/issues/213332 has any relevance here? Do you think we should restrict these patterns to single use cases?

I wasn't aware of this, thanks!
Yes, it looks like it would be better to restrict the upconverting patterns here to single-use cases.

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