[Mlir-commits] [mlir] [mlir][arith] Add `arith.convertf` op (PR #188041)

[Renato Golin]

rengolin wrote:

> I don't necessarily agree that we should be removing extf and truncf at al - extension and truncation (and these sorts of conversions) have meaningfully different properties in a lot of code paths and having them as separate ops makes them easy to match on. They've also go meaningfully different semantics to worry about - extension doesn't _really_ care about rounding mode the way truncation does, for example. Tonally, we have extension and truncation as separate operations for integers, so why not on floats?

That's why I think we should merge this one first (to allow such conversions to exist in the first place), and then decide what to do with the rest. I don't personally see this indecision as a blocker to this PR.

https://github.com/llvm/llvm-project/pull/188041