The Vitis subsystem flow enables the development, testing, and delivery of AI Engine graphs and PL kernels packaged as a platform-independent subsystem. This flow allows for the simulation of the subsystem without the need for a hardware platform, making it faster and more flexible. The VSS component can be configured and rebuilt for testing on multiple devices, making it ideal for feasibility studies, custom library development, and reusable modules.
The subsystem flow uses v++ to link AI Engine graphs and PL
kernels into a .vss deliverable, which can be used as
a component in Vitis or incorporated into another VSS.
The VSS component can be used in either the Vitis Integrated Flow or the Vitis Export to Vivado Flow. A simplified overview of the flow is represented in the following figure:
The primary objective of the VSS flow is to facilitate the development of a combination of AI Engine and PL, independent of the platform. Multiple VSS can contain AI Engine content targeting disjoint partitions. A VSS component can be simulated independently of a platform or running hardware emulation, allowing for AI Engine and PL simulation.
The specific steps for creating and integrating a VSS component are described in Linking a VSS Component and Linking the VSS Component to the Platform.
Key Features of the VSS Flow
The key features of the VSS flow include:
- Platform independent heterogeneous subsystems: Integration of PL kernels and AIE graphs into a VSS component.
- Simulation without needing a platform or running Hardware Emulation: Provides a significant advantage by enabling faster and more flexible testing without platform dependency or hardware emulation.
- Early exploration of AI Engine and PL kernels: Facilitates the early development and testing of key components. It is also very useful when migrating functionality to the AI Engine from the PL.
- Aligns with both Vitis Integrated Flow and Vitis Export to Vivado Flow.