Article directory
foreword
On July 7, 2023, Canaan Technology released the software and hardware development kit of the latest generation K230 chip on the Internet. I tried to click on the released GitHub repository, and was pleasantly surprised to find that this open source project is very well done. The entire sdk project has been released, the project directory is concise and clear, and the technical documentation is also done with great care.
It happened that I encountered a closed-source problem when I used Aixin's ax620a to make a mobile AI sensor product. For me, custom hardware is not flexible and the cost is high, which is a headache. Now I just came across a seemingly suitable replacement plan, so I immediately contacted Jianan's young lady. The young lady is very enthusiastic and nice. After knowing my small needs, she also professionally explained and introduced the situation of this chip to me, and tried to provide me with this latest development board. I am so touched. >_<~+
1. Unpacking
- I unpacked it immediately after receiving it, and I was a little excited. First of all, there are small cards and stickers for precautions, small details with care
- Then there is the main body, which is bigger than expected. Then there is a small complaint, that is, the decorative panel on the LCD is too thin and fragile, and there is no protective edging on the four sides. Perhaps this screen was directly used in those smart door lock products.
- Accessories are two typec cables and a typec to ethernet cable
- The protagonist is the big and small core controller with k230 printed on it.
- High-definition camera and LCD, the speed of the preview screen is still good for the first impression
2. Understanding of operating system concepts
Compared with the previous generation of Internet celebrity products K210, k230 uses a more complex Linux&RT-smart dual-core heterogeneous system. The development environment needs to use the container environment built by docker under ubuntu for cross-compilation, and the development language is currently mainly c/c++.
Among them, the small core runs the linux system and realizes services such as linux driver and network control. The big core runs the RT-smart system to realize the control of the camera sensor and audio and video hardware. The large and small cores need to use their own cross-compilation tool chains, and the communication between them needs to use a set of inter-core communication APIs . File access between large and small cores uses a shared file system called Sharefs . 
In actual experience, the linux side is a standard linux buildroot, and the CLI on the rt-smart side is similar to linux, so both are relatively simple to operate.
3. Use vscode to build a development environment
To quickly learn how to use a new solution, of course, directly look at the application demo/example inside, so it is very necessary to use a good IDE to read and track the source code
By viewing the directory structure of the k230 sdk source code, the sdk itself builds the project through Makefile, and the reference routines in the directory https://gitee.com/kendryte/k230_sdk/tree/main/src/reference/ use CMake to constructed. VSCODEAt this time , it is very convenient to manage and use this sdk project by using Microsoft .
Because the document mentions that the docker container environment is preferred for compilation, it will be a little troublesome when configuring vscode.
The following is my process of configuring the vscode development environment under windows11 :
1. Use WSL2 to install ubuntu
You can refer to my previous article AX620A to run the yolov5s self-training model whole process record (windows) for the operation steps of using WSL2 to install ubuntu, and now you can install the newer 22.04. Note that there is no need to install the Docker-Desktop client in Windows, just install docker under ubuntu.
After installing ubuntu, enter install docker in the terminal of ubuntu
sudo apt update
sudo apt install docker-ce docker-ce-cli containerd.io
2. Pull the k230 sdk warehouse and download the cross-compilation tool
Under ubuntu (root user)
git clone https://github.com/kendryte/k230_sdk
cd k230_sdk
#下载toolchain
source tools/get_download_url.sh && make prepare_sourcecode
- If there is an error
, install itbzip2
apt install bzip2
3. Build docker image
docker build -f tools/docker/Dockerfile -t k230_docker tools/docker
The above process is very smooth, there is no troublesome problem of ladders, and the domestic production is good.
- Create and enter the container
docker run -u root -it -v $(pwd):$(pwd) -v $(pwd)/toolchain:/opt/toolchain -w $(pwd) k230_docker /bin/bash
This step will configure the toolchain location in the container environment to be in the /opt/toolchain directory
- At this point, in fact, you can directly enter make in the container to compile the sdk and generate the final burning image. The whole process is a bit long. My cpu is 7840H 16 threads. After compiling, the entire output directory is just 10 G in size.
make CONF=k230_evb_defconfig
4. vscode installs wsl, Dev Containers and docker plugins
-
Open vscode, search for WSL and Dev Containers in the extension to install.
-
Enter the wsl environment
-
After entering the vscode of the wsl environment, install the docker plugin
5. Start the container in vscode
- After starting the container, click the right-click menu, click "Add VIsual Studio Code", select the k230 container, and a new vscode window will open. This window has pointed to the container development environment of k230 in docker.
- Find the k230_sdk directory in the resource manager and open it.
If it is correct, the lower left of vscode will show that it is currently a container development environment.
6. Configure cmake and cross compile
- Open the command panel: Ctrl+Shift+P, enter: cmake:q, select
GCC9.4.0this compiler first, then manually editcmake-tools-kits.jsonthis file later, and add the cross compiler used by k230
- Open the command panel, enter: cmake configure, select a CMakeLists. Here I chose src/reference/ai_poc/CMakeLists.txt to compile the sample program here
- Then open the command panel, enter: cmake, find
编辑用户本地Cmake工具包. After clicking,cmake-tools-kits.jsonthe configuration file will be opened and a Risv64 option will be added:
[
{
"name": "GCC 9.4.0 x86_64-linux-gnu",
"compilers": {
"C": "/usr/bin/gcc",
"CXX": "/usr/bin/g++"
},
"isTrusted": true
},
{
"name": "Risv64",
"toolchainFile": "/home/k230_sdk/src/reference/ai_poc/cmake/Riscv64.cmake"
}
]
- Open the command panel, enter: cmake, find
选择工具包, select the newly added Risv64
- At this time, if you build directly, you will be prompted that the two paths of riscv64-unknown-linux-musl-gcc and g++ cannot be found. One way is to edit the file in the container and add a
.bashrcline at the end of the file
export PATH=$PATH:/opt/toolchain/riscv64-linux-musleabi_for_x86_64-pc-linux-gnu/bin/
# 保存文件后再执行
source .bashrc
- After closing vscode and reopening, you should be able to find the compiler and configure cmake correctly. Click the button next to the build button at the bottom of vscode
[all], and the compilation target options included in CMakeLists.txt will be listed. Here I choose a routine of yolov8 as the compilation target
- Finally, click the build button to cross-compile the specified program in vscdoe. Finally, the executable file
ob_det.elfis generated in the build directory of the sdk root directory.
Summarize
Thanks to Canaan's detailed and friendly technical documentation, developers can set up a k230 development environment in a short period of time. Of course, the above is only the first step in developing an application, and there are still many related principles, concepts, and module functions that need to be learned and used. The next step should be to obtain video from the camera on the board, send it to the self-trained yolo model for inference, and output the result to the LCD in real time for display.