Edge Impulse Example: stand-alone inferencing for CMSIS toolbox

Example standalone inferencing for AVH created for CMSIS toolbox v2.4.0

Update the model to test

Once you have deployed your model as a Open CMSIS pack, unzip the downloaded file, you will find 2 pack files and a model.clayer.yml

To add your model to your component library you need to use the cpackget utility:

cpackget add <your_model>.pack

You probably need to install the Edge Impulse sdk too:

cpackget add EdgeImpulse.EI-SDK.x.yy.zzpack

Then you need to move the model.clayer.yml into the App folder.

Build the firmware

Using command line

The most basic compilation command is:

cbuild inferencing.csolution.yml

This command will build for every target and every configuration available, so probably isn't what you are looking for.

To specify a configuration and target (ie speed for CM55):

cbuild inferencing.csolution.yml --context inferencing.speed+CM55

Other useful arguments are:

--update-rte

Update the RTE directory, which contains target specific header.

--packs

Automatically download any missing software packs with cpackget.

--clean

Remove intermediate and output directories

Usually, the most common way to compiler for a specific target and toolchain is:

cbuild inferencing.csolution.yml --update-rte --packs --context inferencing.speed+CM55 --toolchain GCC

Check the CMSIS-Toolbox User Guide for a complete overview of the cbuild command.

Using CMSIS toolbox extension

In the CMSIS view, click on the hammer icon.

Using docker

This repository contains a docker image with the required dependencies for building with gcc and llvm:

docker build -t standalone-csolution .

Copy the content of the deployment for your project in the pack folder.

Build the firmware with the following command:

docker run --rm -it -v "${PWD}":/app standalone-csolution /bin/bash -c "./build.sh"

Using script

You can use the build.sh script to compile, target and toolchain can be specified as follow:

./build.sh <TARGET> <TOOLCHAIN>

The script will also install any package present in the pack folder.

Note

After building, the bin will be run locally on AVH.

Run on AVH

The basic usage is:

<AVH FVP for the target> -f <fvp config file> <elf file to be loaded>

Some examples:

• Run Test on model for Cortex-M3 (speed) compiled with GCC

FVP_MPS2_Cortex-M3 -f ./Target/CM3/model_config.txt ./build/CM3/GCC/speed/outdir/CM3_inferencing.elf

• Run Test on model for Cortex-M55 with Ethos (none) compiled with Arm Compiler

FVP_Corstone_SSE-300_Ethos-U55 -f ./Target/CM55/model_config.txt ./build/CM55/AC6/speed/outdir/CM55_inferencing.elf

Available target

CM0 CM0plus CM3 CM4 CM4-FP CM7 CM33 CM55 CM55-U128 CM55-U55-128 CM55-U55-128 CM85

Note

Tested using GCC 10.3.1, GCC 12.3.1, GCC 13.2.1, Arm Compiler v6.19 and 6.22 and Arm LLVM v17.1. Doesn't work with GCC 12.2.1 for MCU with Helium (Cortex-M55, Cortex-M85).