================ ---------------- joker-eph wrote: That change seems fishy to me. I understand you seem to be using tensor splat in the arith-expand pass, but why here? https://github.com/llvm/llvm-project/pull/206249