Xilinx / Vitis-AI

Vitis AI is Xilinx’s development stack for AI inference on Xilinx hardware platforms, including both edge devices and Alveo cards.
https://www.xilinx.com/ai
Apache License 2.0
1.44k stars 626 forks source link

Please add a (cross-) compile guide for custom Ultrascale+ MPSoC with any Linux #1260

Open RapidEdgeAI opened 1 year ago

RapidEdgeAI commented 1 year ago

Hi,

We have a custom Ultrascale+ MPSoC device with a Debian derived OS. Unfortunately, the VAI 3.0 User Guide and older versions of it only contain a build guide for Petalinux (link). The same applies to the Setting up a Zynq UltraScale+ MPSoC Target guide link in the github.io documentation. We have built XRT (Xilinx Runtime), ZOCL-Driver, VART (Vitis AI Runtime) and the Vitis-AI-Library on the device itself. But finding out the dependencies and how to do it was a little bit challenging. Is somewhere maybe a custom build guide, which includes the installation of the necessary dependencies, compilation of XRT and ZOCL and compilation of the VART und Vitis-AI-Library? Or could you create a custom build guide? This would be extremely helpful. AFAIK there is an Ubuntu apt-package, developed by Canonical, which installs Vitis-AI (link). So, there might be already somewhere a build guide.

Thank you very much for your help!

Mickeyyyang commented 1 year ago

Very good question, looking forward to an official reply from Xilinx...

yangyecs commented 1 year ago

Unfortunately, Xilinx currently maintains the Petalinux tool exclusively. Regarding the VitisAI framework, it does not provide support for any other systems. However, Xilinx has a project called PYNQ, which is based on the Ubuntu system and can support the specified version of VitisAI and DPU. Below are some links to the project that you can refer to: https://github.com/Xilinx/DPU-PYNQ https://github.com/Xilinx/Kria-PYNQ/blob/main/README.md

RapidEdgeAI commented 1 year ago

Ok, what a pity. But thank you very much for your response. My request has been clarified.

quentonh commented 1 year ago

@RapidEdgeAI have you also been discussing this with our ML Specialist in Europe, Sophie? I just want to understand if this is the same request that I am aware of.

RapidEdgeAI commented 1 year ago

Hi @quentonh. Yes, but not directly. In February we asked the Xilinx support if there are prebuilt Debian packages for Vitis-AI because Xilinx offers the certified Ubuntu for the Kria boards. But the answer was no because the Ubuntu image that includes those packages is distributed by Canonical. Because of this it would be super helpful if a build guide, including dependencies, cross compilation, etc. for custom Linux systems would be included in a future Vitis AI release. Petalinux works very well for us with our prototype. But unfortunately, we cannot use Petalinux in the product because our whole product ecosystem is based on our own Linux derivate due to maintainability reasons.

RapidEdgeAI commented 1 year ago

I'm sorry. I accidentally skipped "Sophie". No, we didn't talk to Sophie. We talked to another Xilinx employee via the Xilinx support page.

tyler274 commented 1 year ago

This is also an issue for our usecase. Having detailed guides for bringing up the runtime environment is crucial.