How to update my project with a new version of ST Edge AI Core

Context

This document presents an easy scheme to migrate a project that was written using an old version of STEdgeAI (and its outputs) to a newer version of the tool. This page is specific to projects written for using the ST Neural-ART runtime stack and using the NPU.

The issues faced when re-generating a network.c file with a newer version of STEdgeAI are leading to compilation errors/bad behaviours:

• The generated code is expecting with a newer version of the ST Neural-ART runtime stack (i.e. it might use new functions, or expect new behaviours from functions)

  • The generated code raises a compiler error if an old version of ST Neural-ART runtime stack is used.

• The generated code is aligned with a newer version of the NetworkRuntime library (i.e. it might expect new behaviours, new API, etc…)

How to migrate

Update the optimized AI network runtime library and related headers

For epochs that are executed on the CPU (i.e. not handled directly by the NPU), the generated .c file needs to access functions that are shipped in the AI network runtime library.

When migrating to a new version of STEdgeAI, the runtime library expected is newer than the one that was used in previous versions of STEdgeAI. As such,

• the library must be updated (to access “newer” versions of the functions provided).

• the library headers must be updated (to allow compilation to be done properly).

The AI network runtime library is shipped with the installation of STEdgeAI and is located in Middlewares/ST/AI/Lib/<toolchain>/ARMCortexM55, the old library used for linking the project should be replaced by the new one.

The AI network runtime headers are shipped with the installation of STEdgeAI and are located in Middlewares/ST/AI/Inc/, the old header files used in the project should be replaced by the new ones.

Update the ST Neural-Art runtime software stack

In a similar way, the generated .c file must be linked with correct version of ST Neural-ART runtime software stack (or the compilation will eventually fail with an error).

The software stack is shipped with STEdgeAI installation in Middlewares/ST/AI/Npu/ll_aton. When migrating a project written with an old version of STEdgeAI, those files should be replaced by the newer ones in the project.

Summary

• Update the library with the new one in Middlewares/ST/AI/Lib/<toolchain>/ARMCortexM55

• Update the library headers with the new ones in Middlewares/ST/AI/Inc/

• Update the Neural-Art runtime software stack with the new one in Middlewares/ST/AI/Npu/ll_aton