Development Guide - Development Guide - 6.3 English

Development Guide - Development Guide - 6.3 English - XD056

Vitis Video Analytics SDK (VVAS) 6.3 User Guide (XD056)

Document_ID

XD056

Release_Date

2026-09-24

Version

6.3 English

This section covers detailed description and steps to achieve for advanced developers to develop a kernel or VVAS acceleration software library for the kernel. The acceleration software library implements the logic to control the kernel. Each acceleration software library must implement four APIs that are then called by VVAS Infrastructure plug-ins to interact with the kernel.

Table 41 Table 34: Sections for Advanced Developers
Title	Description
Acceleration s/w Library Development Guide	This section covers the intefaces exposed by the VVAS framework to develop the acceleration software library. It also covers the various types of kernels supported and how to develop an acceleration software library for each type of kernel.
Adding new Postprocessing Implementation to VvasPostprocess	This section covers details about how to add a new postprocess implementation with VvasPostprocess.
vvas_xinfer Buffer Allocation Strategies	This section documents how `vvas_xinfer` negotiates the input buffer pool and allocator with upstream, prepares the input frame in the PPE and no-PPE paths, and handles DMA-BUF cache synchronization, for each combination of upstream memory class (system memory, VVAS memory, generic DMA-BUF) and preprocessing mode (internal PPE, external preprocessing).
Custom AI Tensor Formats	This section covers the custom GStreamer tensor format names used in caps, logs, and VVAS image-processing configuration.
VVAS Pipeline Performance Considerations	This section lists practical checks for GStreamer pipelines that miss throughput or latency targets, including measurement, profiler use, batching, zero-copy, and sink choices.