Energy Efficient Branch Prediction on the Cell SPE

We propose novel power efficient branch predictors for the Cell SPU, which normally depends on compiler inserted hint instructions to predict taken branches. Several prediction schemes were designed all using a Branch Target Buffer (BTB) to store the branch target address and the prediction, which is computed using a bimodal counter. One prediction scheme pre-decodes instructions once they are fetched from the local store and accesses the BTB only for a branch instruction. Several ways to combine the previous with the existing hint instructions are studied. We also introduce branch warning instructions which initiates a branch prediction before the branch instruction is even fetched. It allows to fetch the instructions starting at the branch target address and completely removes the branch penalty for correctly predicted branches. Assuming a 256 entry BTB a speedup of up to 18.8% was achieved. The power consumption of the branch prediction schemes is estimated to be 1% or less and the average energy delay product is reduced by up to 6.2%. Keywords-branch prediction, cell spu, energy efficient