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

[Ivan Butygin]

Hardcode84 wrote:

The fact it duplicates lowering with the existing extf/truncf ops is unfortunate. Maybe add a separate legalization pass which transforms `arith.fptofp -> arith.extf/truncf` and only lower `f16<->bf16`? Alternatively we can deprecate `extf/truncf` and switch all lowering to `fptofp` eventually.

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