Register synthesis for speculative computation

Speculative computation and branch prediction have been used in high-performance processor design for many years. Recently it has also been applied to high-level synthesis where a priori knowledge of possible control paths provides an even higher performance potential. One problem of speculative techniques is the circuit overhead necessary for correctness preservation. While in processors, overhead is high due to the required generality, high-level synthesis can, again, employ a priori knowledge. The paper presents a register synthesis and allocation technique for speculative computation with branch prediction which is based on life time trees. It creates shift register structures with little register and control overhead.