Setting Up the Yocto Environment Using Docker Containers
Note: The steps described below are applicable for NVONLINE Users and
NVIDIA Developer Users working with the NGC Docker image.
Note: NVONLINE Users and NVIDIA Developer Users working with the
SDK need to download the additional resources from NGC
(https://registry.ngc.nvidia.com/orgs/${MY_NGC_ORG}/teams/driveos-sdk/resources/driveos_sdk/files)
prior to following the "Install and Set Up the Toolkits" section.
Note: ${MY_NGC_ORG} is the NGC organization allocated for your
team. NVIDIA Developer Users should replace ${MY_NGC_ORG} with "drive". NVONLINE
Users should use their allocated team org.
Install Linux and Foundation Debian Packages
Linux and Foundation packages are pre-installed in the Docker image for NVONLINE Users and NVIDIA Developer Users.
Install and Set Up the Toolkits
Note: Follow these steps to install and set up the
toolkits:
Install CUDA Toolkit on the Host
Note: This step is mandatory if CUDA is enabled with Yocto. CUDA
toolkit must be installed on the host before Yocto build initialization. Skip this
section if CUDA toolkit is already installed.
Cuda host packages are pre-installed in Docker image. Copy the CUDA arm64 Debian package into a specific path :
mkdir <top>/drive-linux/toolkits/cuda/drive-os-av/
cp cuda-tegra-repo-ubuntu*_arm64.deb <top>/drive-linux/toolkits/cuda/drive-os-av/
Install CuDNN Packages if Including CuDNN from Yocto
CuDNN host packages are pre-installed in Docker image. Copy the CuDNN deb file into a
specific path
:
mkdir <top>/drive-linux/toolkits/cudnn/
cp cudnn-local-tegra-repo-ubuntu2004-<CUDNN_MAJOR_VER>-<CUDNN_MINOR_VER>-local_1.0-1_arm64.deb <top>/drive-linux/toolkits/cudnn
Install TensorRT Packages if Including TensorRT from Yocto
TensorRT host packages are pre-installed in Docker image. Copy the TensorRT deb file
into a specific path
:
mkdir -p <top>/drive-linux/toolkits/tensorRT
cp nv-tensorrt-repo-ubuntu2004-cuda${CUDA_VER}trt${TRT_VER}-d6l-target-ga<BUILD_ID>-1_arm64.deb <top>/drive-linux/toolkits/tensorRT
Install NVIDIA DriveWorks if Including NVIDIA DriveWorks from Yocto
Before including NVIDIA DriveWorks in Yocto builds, users must follow the installation steps for CUDA, CuDNN, and TensorRT toolkits as described in the preceding sections.
DriveWorks host packages are pre-installed in Docker image
[OPTIONAL ] Only
execute
the following steps if enabling STM and CGF with
Yocto:
mkdir <top>/drive-linux/toolkits/driveworks
cp driveworks-cgf_<dw-ver>~linux<release-ver>-<GCID>_arm64.deb <top>/drive-linux/toolkits/driveworks
cp driveworks-stm_<dw-ver>~linux<release-ver>-<GCID>_arm64.deb <top>/drive-linux/toolkits/driveworks