Enabling NPU software with Yocto

Follow Us On

Share This Page

Facebook Twitter LinkedIn WhatsApp Email Copy Link

Close

Send Email Using

Gmail (Web) Thunderbird (default mail app)

Close

< >

NXP i.MX95

Table of Contents [Sticky]

• Introduction
  • What is i.MX95?
  • Why choosing RidgeRun for your i.MX95 software?
• Carrier Boards
  • Verdin Development Board
    • Yocto Support
      • Setup Yocto Environment
      • Building an image
      • Flashing the Board
      • Getting into the Board
    • Toradex Easy Installer
      • Hardware Setup
      • Installing a pre-built image using GUI
      • Offline Installation of a pre-built image using GUI
      • Re-Installing Toradex Easy Installer
• Torizon OS
  • Overview
  • Installing Torizon OS Pre-Built Image
  • Containers in Torizon OS
  • Torizon Core Builder
  • Customizing an image with Torizon Core Builder
• GStreamer Support
  • Example Pipelines
• Multimedia Support
  • Audio
  • Streaming
    • RTSP Streaming
    • UDP Streaming
  • Video Capture
    • Capture Subsystem
    • Testing Camera Capture
  • Video Processing Unit-VPU
• Neural Processing Unit-NPU
  • eIQ Neutron NPU overview
  • Machine Learning Software Stack
  • Enabling NPU software with Yocto
  • Testing ML demos
• Contact Us

Introduction

This page shows how to enable software support for hardware accelerated Machine Learning Applications. It will show how integrate the eIQ Toolkit that makes launching ML applications, easier.

Software Prerequisites

• A configured build environment as described in Yocto Support
• Read the Build a Reference Image with Yocto Project article

Setup Instructions

Clone the eIQ Layer

The eIQ stack is distributed through the meta-imx repository, specifically within the meta-imx-ml layer. This layer provides all the required recipes for Machine Learning runtimes, compilers, and supporting tools.

To include this layer in your Yocto setup, clone the repository and copy the Machine Learning layer into your BSP:

cd /tmp
git clone --depth 1 -b scarthgap-6.6.52-2.2.2 https://github.com/nxp-imx/meta-imx.git
sudo cp -a meta-imx/meta-imx-ml/ <toradex_bsp_yocto_dir>/layers/

Warning: Make sure the branch you clone matches the version of the Toradex BSP you are working with. This may change over time. For instance, if you are using Yocto Project 7.5 reference images, the correct branch is scarthgap-6.6.52-2.2.2. Always verify compatibility using the Embedded Linux Release Matrix.

Register the Layer in Yocto

Once the layer has been copied, it must be added to your build configuration. Update your bblayers.conf file by appending the new layer:

BBLAYERS += "                     \
  ${TOPDIR}/../layers/meta-imx-ml \
"

Configure and Build the Image

Next, configure your image to include the required packages for running edge AI workloads. This is typically done on top of the Toradex Reference Multimedia Image.

Add the following to your local.conf:

IMAGE_INSTALL:append = " adwaita-icon-theme opencv python3-pillow tensorflow-lite tensorflow-lite-neutron-delegate vim"

Finally, build the Reference Multimedia Image with this modification. Run the bitbake command:

bitbake tdx-reference-multimedia-image

And then install it following this page: Flashing the Board

< >