“Implementing out-of-order execution with multiple physical registers for each architectural register becomes prohibitively expensive for very wide vector registers, such as 1K or 2K bits.”