This section briefly discusses how the host application is written to orchestrate the execution of a convolution kernel. It was estimated in the previous lab that multiple compute units will be needed to meet the 60 FPS performance target for processing 1080p HD Video. The host application is designed to be agnostic to the number of compute units. More specifically, if the compute units are symmetric ( instance of the same kernel and memory connectivity to device DDR banks is identical), the host application can deal with any number of compute units.
TIP: Additional tutorials and examples about host programming and Vitis tools are available.