Our paper “RAOP: Recurrent Neural Network Augmented Offset Prefetcher” has been accepted at the International Symposium on Memory Systems (MEMSYS ’20). This paper proposes an RNN Augmented Offset prefetcher (RAOP) that makes use of both the RNN predicting accuracy and the principle of spatial locality. By performing offset prefetching for both the current and the predicted memory addresses, the prefetching performance is significantly improved.
