Heterogeneous computing on tightly-integrated CPU-GPU systems is ubiquitous, and to increase programmability, many of these systems support virtual address accesses from GPU hardware. However, there is no free lunch. ...