ti-sdk-docker

PROCESSOR-SDK-AM67A running on TI Ubuntu

Docker setup

Running the image

docker run -it \
    -v /Volumes/LinuxCS/code:/home/tisdk/shared \
    ghcr.io/goat-hill/ti-sdk-docker:latest /bin/bash

Building image

In same directory as Dockerfile:

docker build -t ghcr.io/goat-hill/ti-sdk-docker:latest .

Push to Github, using classic Github token:

docker login ghcr.io -u your-github-user
docker push ghcr.io/goat-hill/ti-sdk-docker:latest

Running TI Edge AI SDK on BeagleY-AI

I have the TI Edge AI SDK 11.00.00.08 running on BeagleY-AI with a few modifications.

• UBoot - branch ti-u-boot-2024.04-bb
• Linux kernel - branch ti-linux-6.12.y-bb
• TI vision_apps - branch 11.00.00.06-beagley

U-Boot development workflow

Building U-Boot

r5

EVM configure:

make ARCH=arm O=/home/tisdk/uboot-build/r5 j722s_evm_r5_defconfig

BeagleY-AI configure:

make ARCH=arm O=/home/tisdk/uboot-build/r5 beagleyai_r5_defconfig

Compile:

make -j$(nproc) ARCH=arm O=/home/tisdk/uboot-build/r5 \
    CROSS_COMPILE="$CROSS_COMPILE_32" \
    BINMAN_INDIRS=${PREBUILT_IMAGES}

Copy output to shared volume:

cp /home/tisdk/uboot-build/r5/tiboot3-j722s-hs-fs-evm.bin /home/tisdk/shared/ti-uboot-build/tiboot3.bin

Other tasks

Clean:

make ARCH=arm O=/home/tisdk/uboot-build/r5 mrproper

Simplify defconfig:

make ARCH=arm O=/home/tisdk/uboot-build/r5 savedefconfig
cp /home/tisdk/uboot-build/r5/defconfig configs/beagleyai_r5_defconfig

a53

EVM configure:

make ARCH=arm O=/home/tisdk/uboot-build/a53 j722s_evm_a53_defconfig

BeagleY-AI configure:

make ARCH=arm O=/home/tisdk/uboot-build/a53 beagleyai_a53_defconfig

Compile:

make -j$(nproc) ARCH=arm O=/home/tisdk/uboot-build/a53 \
    CROSS_COMPILE="$CROSS_COMPILE_64" \
    CC="$CC_64" \
    BL31=${PREBUILT_IMAGES}/bl31.bin \
    TEE=${PREBUILT_IMAGES}/bl32.bin \
    BINMAN_INDIRS=${PREBUILT_IMAGES}

Copy output to shared volume:

cp /home/tisdk/uboot-build/a53/tispl.bin /home/tisdk/shared/ti-uboot-build/
cp /home/tisdk/uboot-build/a53/u-boot.img /home/tisdk/shared/ti-uboot-build/

Other tasks

Clean:

make ARCH=arm O=/home/tisdk/uboot-build/a53 mrproper

Simplify defconfig:

make ARCH=arm O=/home/tisdk/uboot-build/a53 savedefconfig
cp /home/tisdk/uboot-build/a53/defconfig configs/beagleyai_a53_defconfig

Installing U-Boot

Based on SDK instructions here

sudo cp tiboot3.bin tispl.bin u-boot.img /media/brady/BOOT

Linux kernel development workflow

Building Linux kernel

Image and modules

Must be on same git commit for Image + modules + module_install with no local changes to avoid -dirty flagging.

Clean and configure make:

make -j$(nproc) ARCH=arm64 CROSS_COMPILE="$CROSS_COMPILE_64" distclean
make -j$(nproc) ARCH=arm64 CROSS_COMPILE="$CROSS_COMPILE_64" defconfig ti_arm64_prune.config

Compile linux kernel image and modules:

make -j$(nproc) ARCH=arm64 CROSS_COMPILE="$CROSS_COMPILE_64" Image
make -j$(nproc) ARCH=arm64 CROSS_COMPILE="$CROSS_COMPILE_64" modules

Device tree DTBs

make -j$(nproc) ARCH=arm64 CROSS_COMPILE="$CROSS_COMPILE_64" dtbs

Installing kernel

Instructions based on SDK docs here

sudo cp arch/arm64/boot/Image /media/brady/rootfs/boot/
sudo cp arch/arm64/boot/dts/ti/k3-am67a-beagleyai.dtb /media/brady/rootfs/boot/dtb/
sudo cp arch/arm64/boot/dts/ti/k3-am67a-beagley-ai-edgeai-apps.dtbo /media/brady/rootfs/boot/dtb/ti/
sudo cp arch/arm64/boot/dts/ti/k3-am67a-beagley-ai-csi0-imx219.dtbo /media/brady/rootfs/boot/dtb/ti/
sudo make ARCH=arm64 INSTALL_MOD_PATH=/media/brady/rootfs/ modules_install

Make sure boot partition uEnv.txt indicates the overlays:

name_overlays=ti/k3-am67a-beagley-ai-edgeai-apps.dtbo ti/k3-am67a-beagley-ai-csi0-imx219.dtbo

TI vision_apps development workflow

Install PROCESSOR-SDK-RTOS-J722S on TI Ubuntu docker image. Download SDK here: https://www.ti.com/tool/PROCESSOR-SDK-J722S

I'm mostly following AM67A firmware builder guide

To setup firmware builder run the following below. Exclude --firmware-only flag as we need Linux changes. Execute this in the root directory of PSDK RTOS.

./sdk_builder/scripts/setup_psdk_rtos.sh

Swap to modified BeagleY-AI modified vision_apps

We need to use a modified verison of vision_apps to support memory map for BeagleY-AI with 4 GB DDR vs 8 GB EVM. Modifications are based on instructions here

mv vision_apps/ vision_apps_bak/
git clone https://github.com/goat-hill/ti-vision-apps.git vision_apps
cd vision_apps/
git checkout 11.00.00.06-beagley

Compilation

Then:

cd sdk_builder/
TISDK_IMAGE=edgeai ./make_firmware.sh

Gather build outputs

cp -r /tmp/tivision_apps_targetfs_stage /home/tisdk/shared/tivision_apps_targetfs_stage

Installing vision_apps

export LINUX_FS_PATH=/media/brady/rootfs
export LINUX_FS_STAGE_PATH=/home/brady/host-shared/code/tivision_apps_targetfs_stage

# remove old remote files from filesystem
sudo rm -f $LINUX_FS_PATH/usr/lib/firmware/j722s-*-fw
sudo rm -f $LINUX_FS_PATH/usr/lib/firmware/j722s-*-fw-sec
sudo rm -rf $LINUX_FS_PATH/usr/lib/firmware/vision_apps_eaik
sudo rm -rf $LINUX_FS_PATH/opt/tidl_test/*
sudo rm -rf $LINUX_FS_PATH/opt/notebooks/*
sudo rm -rf $LINUX_FS_PATH/usr/include/processor_sdk/*

# create new directories
sudo mkdir -p $LINUX_FS_PATH/usr/include/processor_sdk

# copy full vision apps linux fs stage directory into linux fs
sudo cp -r $LINUX_FS_STAGE_PATH/* $LINUX_FS_PATH/.

Edge AI development workflow

Obtaining Edge AI image rootfs .tar.gz

Download same TI SDK Edge AI .wic.gz used for flashing SD card. Now we want to get the rootfs of the .wic available in the same directory. A bit tricky becuase it's a .wic image and not .tar.gz like TI SDK adas edition.

Steps if in a docker image:

sudo losetup -P $(losetup -f) ~/host-shared/code/tisdk-edgeai-image-j722s-evm.wic 
losetup -a
sudo mount /dev/loop13p2 /mnt/rootfs-edgeai
sudo tar -czpf targetfs.tar.gz -C /mnt/rootfs-edgeai/ .

Hold on to this .tar.gz. Now cleanup the mess:

sudo umount /mnt/rootfs-edgeai 
sudo losetup -d /dev/loop13

Creating build environment

On TI Ubuntu docker image with Edge AI SDK installed:

git clone https://github.com/TexasInstruments/edgeai-app-stack
git submodule init
git submodule update

Extract EdgeAI rootfs to targetfs/ local directory.

tar -zxf targetfs.tar.gz -C targetfs

Compilation

Note: Required modifications to Makefile (point to local compile toolchain, targetfs/, and new targetfs-install/ dir)

Now compile:

ARCH=arm64 make -j$(nproc)

I had to compile it a few times...

Copy to SD card

sudo cp -r targetfs-install/* /media/brady/rootfs/

Testing TI Edge AI SDK

Kill the default app started at launch

killall edgeai-gui-app

Pipe OpenVX logs to shell:

source /opt/vision_apps/vision_apps_init.sh

Run IMX219 demo:

cd /opt/edgeai-gst-apps/apps_cpp/
./bin/Release/app_edgeai ../configs/imx219_cam_example.yaml

Issues

• Need to add libncurses-dev apt package for menuconfig
• TI Ubuntu Docker build is broken (there was a gnutls package missing as a result)