I think I’m going to take this into a target-specific optimization, now that I understand the assumption of “setCC has one output type” is pretty heavily baked into the current DAG/IR. Fiona