Memory Dependence Predictors

This paper details the implementation of a novel Memory Dependence Predictor based on the relative distance between loads and their dependent stores in the dynamic program order. In presence of unknown store addresses blind speculation never stalls loads which are ready. This results in load violations since the loads may depend on those stores with unknown addresses. To prevent the violation penalty we employ the use of memory dependence predictors that train a predictor table to detect load-store dependency. In this project we will be implementing two kinds of memory dependence predictors namely, Sticky Bit and Store Vectors.