Cross compilation from x86 servers to Orin

[hmedicharla]

Hi,
Can I build code on host tensor rt and execute the built binary on Orin device? Is such a cross compilation and execution possible?

Thanks,
Hema

[SivaRamaKrishnaNV]

Dear @hmedicharla,
Yes. we recommend to develop code, cross compile on host and test on target. Please see "Cross compile samples " section in TensorRT development guide

[hmedicharla]

Hi @SivaRamaKrishnaNV,

I am facing the following error while launching the Tensor RT OSS build container as per the steps given in the link GitHub - NVIDIA/TensorRT: NVIDIA® TensorRT™, an SDK for high-performance deep learning inference, includes a deep learning inference optimizer and runtime that delivers low latency and high throughput for inference applications.

sudo ./docker/launch.sh --tag tensorrt-ubuntu20.04-cuda11.6 --gpus all
Launching container:

docker run --gpus all -v /home/deltlo22/Downloads/nvidia/TensorRT/TensorRT:/workspace/TensorRT --rm -it tensorrt-ubuntu20.04-cuda11.6:latest
docker: Error response from daemon: could not select device driver “” with capabilities: [[gpu]].

Upon search for a fix to this error, I followed steps in the link (20.04 - Docker only works with Nvidia drivers upon reinstall - Ask Ubuntu) suggested in could not select device driver "" with capabilities: [[gpu]]. - #7 by a.adeyemi13

This also did not help resolve the error. Please help on further steps to resolve this error.

Thanks,
Hema

[hmedicharla]

I get the following error while setting up the build environment for the cross-compile for Jetson (step: Example: Ubuntu 20.04 cross-compile for Jetson (aarch64) with cuda-11.4.2 (JetPack SDK) as per the link GitHub - NVIDIA/TensorRT: TensorRT is a C++ library for high performance inference on NVIDIA GPUs and deep learning accelerators.)

The command ‘/bin/sh -c dpkg -i /pdk_files/cuda-repo-cross-aarch64*.deb /pdk_files/cuda-repo-ubuntu*_amd64.deb && apt-get update && apt-get install -y cuda-cross-aarch64 && rm -rf /var/lib/apt/lists/*’ returned a non-zero code: 100

[hmedicharla]

Can some one from Nvidia support team help here? I am kind of stuck

Thanks,
Hema

[AastaLLL]

Hi,

Sorry for the late update.

Please noted that the TensorRT engine has dependencies on the hardware.
So you will need to convert the engine on Orin directly.

But it is possible to cross-compile a TensorRT source code.
We will double-confirm the workflow and share more details with you.

Thanks.

[AastaLLL]

Hi,

Sorry that we have confirmed that the cross-compiling toolchain for Jetson doesn’t be included in the JetPack.
Currently, please build the binary directly on the device for simplicity.

We will update the TensorRT document to reduce the confusing.

Thanks.