AI Engine APIs - 2025.1 English - XD100

Vitis Tutorials: AI Engine Development (XD100)

Document ID
XD100
Release Date
2025-08-25
Version
2025.1 English

The AI engine APIs are provided as header files allowing a higher level of abstraction than intrinsics. They are also architecture-agnostic, that is, the generated intrinsics match the selected target device. Browse through the AI Engine API User Guide to view the available functions (see Fig. 3).

Fig. 3: AI Engine API User Guide Fig. 3: AI Engine API User Guide

The following figure shows how an intrinsic and an API with the same basic operation, such as multiply-accumulate, may differ.

Fig. 4: Intrinsic vs. API Fig. 4: Intrinsic vs. API

Note: Intrinsics have more parameters, allowing more flexibility, but require more detailed knowledge of the architecture. APIs assume a specific use-case, allowing the use of fewer parameters.

The use of APIs is highly recommended, and only utilize intrinsics to optimize critical portions of the code. Note that intrinsics and APIs may be mixed together in the same kernel code.