The AI Engine compiler, Vitis linker, and embedded XRT support the independent compilation and linking of AI Engine graphs into partitions. These AI Engine partitions, along with the PL region, can be loaded at boot time. XRT enables runtime control of AI Engine partitions and PL kernels. XRT also supports dynamic reloading of an AI Engine partition while keeping other AI Engine partitions running. The reloaded graph for a partition can be the same or different, provided that the AI Engine-PL and AI Engine-NoC interfaces remain unchanged.
Important: The XRT-based Reset and Reload
feature of the AI Engine
array is deprecated. AMD recommends using the Reset
and Reload of AI Engine
Partitions feature described in this section.
This section details the process of building AI Engine partitions for reloading and the
host code required for runtime control and partition reloading.
Note: Only the hardware flow supports the runtime
control and reloading of AI Engine partitions. Hardware emulation flow is not
supported.