[Mlir-commits] [mlir] [mlir][math] Fix casting mistake in tanh expansion and test (PR #85364)

[Maksim Levental]

makslevental wrote:

Is the original/current actually wrong? `arith::CmpFOp` returns `i1` i.e., signless, which means roughly (in my understanding, which might be mistaken) "[up to you](https://mlir.llvm.org/docs/Rationale/Rationale/#integer-signedness-semantics)", i.e., target (dialect or arch or whatever) dependent. So how can you pick `SIToFPOp` or `UIToFPOp` at all here? I mean, maybe the real issue is that there's no `arith::IToFP` for cases like this.


https://github.com/llvm/llvm-project/pull/85364