The Filter2DDispatcher class is the top-level class that provides an end-user API to schedule kernel calls. Every call schedules a kernel enqueue and related data transfers using the Filter2DRequest object, as explained previously. The Filter2DDispatcher is a container class that essentially holds a vector of request objects. The number of Filter2DRequest objects that are instantiated is defined as the max parameter for the dispatcher class at construction time. This parameter’s minimum value can be as small as the number of compute units to allow at least one kernel enqueue call per compute unit to happen in parallel. But a larger value is desired since it will allow overlap between input and output data transfers happening between host and device.