> Hmm, so the "idea" was that we'd move towards overloading > vector_costs::add_slp_cost (x86 does, handling one specific case) > so as to get all "components" of the overall operation that you could > then classify on. That the components use scalar_load/store would > be of secondary concern. > > Would that achieve a similar result and maybe improve handling of > gather/scatter costing for you?
Yeah, I think we can use add_slp_cost as well (I tried it for the in-flight permute-cost stuff). I think all information we need is available there but it felt more work than just passing the "right" thing here. If there's no way forward here, we'll try reconstructing what we need in add_slp_cost. -- Regards Robin
