An Asynchronous Register Bypass Transformation

Abstract

A register specification typically states that in each cycle there is a possible read followed by a possible write; the sequence is strict. A register core with a separate read and write port is more efficient, because it can read and write to different locations simultaneously, and hence in one cycle. In the Caltech MiniMIPS processor, a control structure was added to such a register core, so that it implements the desired specification.