Complexity and Correctness of a Super-Pipelined Processor

Abstract

This thesis introduces the DLXπ+, a super-pipelined processor with variable cycle time. The cycle time of the DLXπ+ may be as low as 9 gate delays (including 5 gate delays for registers), which is assumed to be a lower bound for the cycle time. For the parts of the DLXπ+ that significantly differ form previous implementations correctness proofs are provided. Formulas are developed which compute restrictions to the parameters of the DLXπ+, e.g., the maximum number of reservation station entries for a given cycle time. The formulas also compute what modifications to the base design have to be made in order to realize a certain cycle time and what the impact is on the number of pipeline stages. This lays the foundation for computing the time per instruction of the DLXπ+ for a given benchmark and different cycle times in future work in order to determine the "optimum" cycle time.