sunshaoce wrote: > Isn't the code in LowerOperation only used for vectors? >From what I saw, `lowerAbsDiffToMinMax` works for both scalars and vectors. Should we keep it generic? https://github.com/llvm/llvm-project/pull/155888