Architecture of the cu$Q$-RTM code package

In this section, we outline the architecture of cu$Q$-RTM code package and some program optimization schemes. From an overall perspective, this package can be roughly separated into four components: memory manipulation, modules, kernels, and multi-level parallelism. As shown in Figure 1, each component plays an indispensable role in GPU-CPU cooperative computing. The following is a brief description of each component and how it interacts with the others.