On Thu, Nov 29, 2018 at 10:22:15AM +0100, Richard Biener wrote: > OK. I didn't spent a lot of time trying to see if we can merge the > scalar and vector variants but I trust you did ;)
Yeah, sadly it is too different, even if we were to introduce a predicate whether a tree is either INTEGER_CST, or VECTOR_CST where uniform_vector_p is non-NULL and INTEGER_CST, e.g. because we need to use build_vector_from_val and take care of the sometimes vector, sometimes element types. Jakub