Memory models: a case for rethinking parallel languages and hardware