The Tradeoffs of Fused Memory Hierarchies in Heterogeneous Computing Architectures