SDAA422 June   2026 TDA4VE-Q1

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1Introduction
  5. 2Understanding OpenVX Kernel Architecture
    1. 2.1 OpenVX Node Execution Flow
    2. 2.2 Host and Target Kernels
    3. 2.3 A72 Target Execution Model
  6. 3Creating a Custom Kernel Package
    1. 3.1 Package Structure
    2. 3.2 Defining the Kernel Interface
  7. 4Developing a Custom OpenVX Node for the A72 Core: A Threshold Example
    1. 4.1 Implementing the Host Kernel
    2. 4.2 Implementing the A72 Target Kernel
    3. 4.3 Accessing Image Buffers on the A72 Core
    4. 4.4 Implementing the Threshold Algorithm
    5. 4.5 Registering the Kernel Package
  8. 5Integrating the Custom Node into vision_apps
    1. 5.1 Implementing the app_custom_threshold Example
    2. 5.2 Example Execution Log
  9. 6Summary
  10. 7References

6 Summary

This application note presented a practical workflow for developing a custom OpenVX node that executes on the Arm Cortex-A72 core using the TIOVX framework provided by Texas Instruments Processor SDK RTOS. A threshold node was used as a reference implementation to illustrate the complete development process, including kernel interface definition, Host Kernel implementation, A72 Target Kernel development, image buffer access, kernel package registration, and integration into the vision_apps framework.

The example demonstrated how application-specific processing can be incorporated into an OpenVX graph while preserving the standard TIOVX execution model. Although the threshold operation was intentionally chosen for its simplicity, the overall development methodology is applicable to a wide range of custom image processing, sensor processing, and AI pre-processing or post-processing algorithms. By modifying only the processing logic within the Process() callback, developers can reuse the same Host Kernel, Target Kernel, memory access, and graph integration framework to accelerate the development of custom A72-based OpenVX nodes.

As vision and edge AI applications continue to evolve, integrating application-specific processing into heterogeneous computing platforms becomes increasingly important. The workflow presented in this application note provides a practical foundation for extending the TIOVX framework with custom functionality while maintaining compatibility with the existing Processor SDK RTOS software architecture. By leveraging the flexibility of the Arm Cortex-A72 core together with the graph-based execution model of OpenVX, developers can rapidly prototype, validate, and deploy custom processing nodes for a broad range of embedded vision and AI applications.