[Mlir-commits] [mlir] [mlir] [arith] Fold trunc(extremum(ext(lhs), ext(rhs))) (PR #214658)

[Jakub Kuderski]

================

----------------
kuhar wrote:

I'd rather see this implemented as an additional parameter passed to `isLosslesslyConvertibleTo`. That is safe for today’s types, but not fully future-proof: a future wider format that converts source NaN or infinity into an ordinary finite value could make the original satisfy nnan while the rewritten operation becomes poison or returns a different value.

https://github.com/llvm/llvm-project/pull/214658