[Mlir-commits] [mlir] [mlir] [arith] Fix ceildivsi lowering in arith-expand (PR #133774)

[Jakub Kuderski]

kuhar wrote:

> > What I can easily do is add integration test for some values that do not trigger poison/UB if you want, but I'm just not sure how to proceed with poison/UB.
> 
> Yeah this sounds good. I know we can't test for poison/UB, but let's at least make sure that the non-ub results are correct.

My worry would be that something that attempts to fix UB/poison propagation could just as well break the correctness for well-defined inputs and it's hard to tell if the math checks out or not.

https://github.com/llvm/llvm-project/pull/133774