○单周期、多周期、流水线控制性能比较 Single-Cycle EXecution Clock=350+250+350+350+250=1550 ps IFReg ALU MEM Reg CPI=1, but long clock cycle Each instruction =1550 ps IF Reg AL MEM Reg Each instruction 1550 ps IF Reg ALU MEM Reg Multi-Cycle EXecution: Clock =350 ps Each instruction= 1550 ps -+ F|Reg| ALUMEM| Reg Average CP=5×02+4×01+4×04+3×0.3=39 350ps 350ps 350ps 350ps 350ps IFReg| ALU Reg -Load 5 cycles 350ps350ps350ps350ps Reg ALU ALU =4 cycles Pipelined Execution 350ps350s350ps Branch=3 cycles→ IFReg ALU MEMReg Tcok=350ps=ma(350,250) 350ps Reg ALU MEM Reg One instruction completes each cycle 350ps IF Reg aLU mEm reg Average CPI=1 350ps350ps350ps350ps350ps Ignore time to fill pipeline 2/112021 chapters 10单周期、多周期、流水线控制性能比较 2/11/2021 chapter03.10 • 假设5段指令执行流水线 • 某一程序段假设： – 20% load, 10% store, 40% ALU, and 30% branch • 比较三种执行模式的性能