How to across compile the tensorrt sample code in x86_64 for aarch64

[FengtuWang]

Hi,all
I want to across compile the tensorrt sample code for aarch64 in a x86_64 machine.
I used the SDK manager 1.5.0.7774 software to install CUDA in the host machine. It can not find the related TensorRT and cuDNN softwares. When I build the demo trtexec, I got some errors about that can not found some lib files. I find that the same name lib in the x86_64 target, but it is not in the aarch64 target.
I use a PC without nvidia platform GPU.
I want to install the related software about Tensorrt and cudnn.
some one would give me some guide?
Thanks for your time.

[FengtuWang]

There are some information about the CUDA libs with SDK Manager installed

• In the x86_64
  

  

• in the aarch64
  

  

There are different library between x86_64 and aarch64 platform. When I build the TensorRT sample trtexec code, I got the error.

There are some libs need by the code exist in the x86_64 but don’t exist in the aarch64 platform.
So should I reinstall CUDA aarch64 version manual in the x86_64 platform?
The sample code also need the TensorRT libs, should I install Tensorrt aarch64 in the x86_64 platform ? How I get it ?

Thanks for your times.

[AastaLLL]

Hi,

Since there are lots of dependencies to cross-compile a TensorRT software. ​
Please use our docker container for the environment setup.

You can find the detailed steps below:

GitHub - NVIDIA/TensorRT: TensorRT is a C++ library for high performance...

Thanks.

[FengtuWang]

HI, AastaLLL

thanks for your reply.

I also have some questions about the README.md in the project NVIDIA/TensorRT.

There are some contents.

(Optional) JetPack SDK for Jetson builds
Using the JetPack SDK manager, download the host components. Steps:

Download and launch the SDK manager. Login with your developer account.
Select the platform and target OS (example: Jetson AGX Xavier, Linux Jetpack 4.4), and click Continue.
Under Download & Install Options change the download folder and select Download now, Install later. Agree to the license terms and click Continue.
Move the extracted files into the <TensorRT-OSS>/docker/jetpack_files folder.

I got some .deb ubuntu install package through sdkmanager. I would like to know which files should to be extracted and then move to the releated folder.

[FengtuWang]

I got some .deb package as below.

[AastaLLL]

Hi,

You can find the required file below:

https://github.com/NVIDIA/TensorRT/blob/master/docker/ubuntu-cross-aarch64.Dockerfile#L72

cuda-repo-cross-aarch64\*.deb
cuda-repo-ubuntu\*_amd64.deb
libcudnn[7-8]_\*-1+cuda10.[0-9]_arm64.deb
...

Thanks.