Simultaneous way-footprint prediction and branch prediction for ener gy savings in set-associative instruction caches

Caches are partitioned intosubarrays for optimal timing. In a set-associative cache, if the way holding the data is known before anaccess, only subarrays for that way need tobe accessed. Reduction in cache switching activities results in energy savings. In this paper, we propose to extend the branch prediction framework to enable wayfootprint prediction. The next fetch address and its way-footprint are predicted simultaneously for one-way instruction cache access. Because the way-footprint prediction shares some prediction hardware with the branch prediction, additional hardware cost is small. To enlarge the number of one-way cache accesses, we have made modi cations to the branch prediction. Speci cally, we have investigated three BTBallocation policies. Each policy results in average 29%, 33 % and 62 % energy savings with normalized execution time 1, 1, and 1.001 respectively