| |
[摘要]：Caches are underutilized because not all of the excess data brought into the cache, to exploit spatial locality, is used. We define cache utilization as the percentage of data brought into the cache that is actually used. In this paper, we focus on prediction mechanisms to predict the useless data in a cache block ( cache noise). The prediction mechanisms consider the word usage history of cache blocks for predicting the useful data so that only the useful data is fetched into the cache on a cache miss. We found that the cache noise can have a predictability of about 95 percent with a simple last word usage predictor. When applying cache noise prediction to the L1 data cache, we observed about 37 percent improvement in cache utilization and about 23 percent and 28 percent reduction in cache energy consumption and bandwidth requirement, respectively. Cache noise mispredictions increased the miss rate by 0.1 percent and had almost no impact on instructions per cycle (IPC) count. When compared to a subblocked cache, fetching the to-be-referenced data resulted in 97 percent and 10 percent improvement in miss rate and cache utilization, respectively. However, the subblocked cache had a bandwidth requirement of about 35 percent of the cache noise prediction-based approach and saved 10 percent more power. We also observed that cache noise prediction significantly improves the utilization and reduces the bandwidth requirement for prefetching. |
|
