AI Engine Graphs and Arrays - 2021.2 English

Vitis Unified Software Platform Documentation: Application Acceleration Development (UG1393)

Document ID
UG1393
Release Date
2022-03-29
Version
2021.2 English

The AI Engine Graph and Array diagrams provide a quick overview of the structure of the ADF graph application implemented in the AI Engine tiles.

Figure 1. AI Engine Graph Application

The AI Engine Graph view shows the connectivity of the ADF graph as seen by the AI Engine compiler. The canvas shows nodes representing kernels, kernel arguments, memory buffers, and primary inputs and outputs, with edges drawn to represent the connectivity between these elements.

To the right of the canvas, a Settings panel that can be used to customize the view of the graph, by letting you show or hide elements of the displayed graph.

Below the graphical view is a set of tables containing all of the objects drawn in the graphical view: kernels, connections, paths, memories, etc. Selecting an element in one of these tables will cross-probe to the graphical view and vice versa.

Figure 2. AI Engine Array Diagram

The Array view shows how the ADF graph is spatially placed on the AI Engine tile array. The array canvas contains all the core and memory components of the array. Kernels are drawn in the core where they are programmed, and connectivity between cores and/or memory are shown with connectivity lines. There is an abstract representation of the PL area as needed to model PL cores and interfaces to the programmable logic region.

To the right of the canvas, a Settings panel that can be used to customize the view of the array, by letting you show or hide elements of the resources and/or tile array.