Cross-Compilation for AGX Orin

I need to perform cross-compilation on LINUX PC for Jetson AGX Orin
I need to do for my applications that use CUDA,Tensorrt,Deepstream.

How to do it?
Jetpack - 5.1.2
Jetson LINUX - 35.4.1

Please refer to Jetson Linux Toolchain — Jetson Linux Developer Guide documentation (

This involves compiling for code that doesn’t involve CUDA/GPU/deepstream right?

how do i compile for CUDA and tensorrt, as the versions on PC will be 12+ and on the device i have11.4?

Please install the cross-compile toolkit from the SDK manager.

The package is also available in the below link: cuda-cross-aarch64*

Once i install these i have export the toolchain path and the CUDA library path and compile right?


Yes, please give it a try.
Please note that CUDA cross-compile is supported but TensorRT is not.


for deepstream?

hi any updates?

i tried to compile a sample tcp program with this
$ export CROSS_COMPILE=$HOME/l4t-gcc/bin/aarch64-buildroot-linux-gnu-

and compilation command
g++ serverkey.cpp -o server
on the same terminal
but the files was not compiled for aarch64

This enviroment variable only works for kernel build system.

You need to directly call the cross compiler binary here instead of g++.

is there any example that i can refer to
the one given in the previous chat was referring to this toolchain export path

i was able to compile for the tcp program
using aarch64-linux-gnu-g++
but when i tried for the gstreamer program
i was not able to
what to do?


We don’t have cross compile tool for Deepstream.

Would you mind sharing more info on why you need it?
Do you want to narrow down the compiling time or align the environment between the host/the Jetson app?


We are building our DS pipeline using c++ sources. It involve detection, distance, image processing and streaming all integrated to the pipeline.
For that we’re using CUDA, tensorrt, image framework and gst.
So we need to build the sources- cross compile them on the Host PC and then add executable to the Linux_for_Tegra source folder and flash it on the board as a single image.

So i need to know how to cross compile for CUDA,tensorrt,deepstream,opencv and gstreamer involved in a program.

Srivathsan P


We don’t have the cross-compile tool for TensorRT and Deepstream.

For your usecase, maybe you can try to compile it on the Jetson locally.
And copy the application back to host to create the image?


atleast for gstreamer and opencv and cuda based programs can you give the solution excluding deepstream and tensorrt