How to install NVIDIA packages (CUDA, tensorrt... )

[CHNtentes]

Hello,

I came to this github repo from Ridgerun guide https://developer.ridgerun.com/wiki/index.php/Yocto_Support_for_NVIDIA_Jetson_Platforms/ . Although I'm able to build the image, adding NV packages brings some issues for me:

1. In the guide, it tells me to use IMAGE_INSTALL_append = " cuda-samples" to add CUDA package. However, under /usr/local/cuda-11.4/bin, there's nothing but only cuda-install-samples-11.4.sh. Obviously, I expect some binary files in this directory.
2. When I add IMAGE_INSTALL_append = " tensorrt" in local.conf, bitbake returns ERROR: Nothing RPROVIDES 'tensorrt'. I tried IMAGE_INSTALL_append = " tensorrt-samples" and no error occurs, but I doubt this is the solution here.

I'm using kirkstone branch for all layers and Orin AGX devkit. Thanks for support.

[madisongh]

For cuda-samples, by reading the recipe you'll see that it builds only a couple of the sample programs, and they get installed in /usr/bin/cuda-samples/.

For TensorRT, I suspect that the wiki page is just a bit out of date. To provide more control over exactly what packages get installed, and to maximize the number of packages that get built from sources, there are different recipes for the different bits of TensorRT that you might need. Take a look at the recipes for more information.

Unlike JetPack, which dumps a huge blob containing as much of NVIDIA's software on your device as possible so you can play with it, the goal here is to let you pick and choose exactly what you need for your product/application. Because of that basic difference, you won't find a direct 1:1 mapping between JetPack packages and those generated by the recipes here.

[ichergui]

Hey @leadercxn
Any updates ?

[leadercxn]

Hey @leadercxn Any updates ?

@ichergui Yes, I copied the bb files and directories from the latest master branch and used them.

When I used jtop again, it was better than before

Thank you ~

[leadercxn]

Hey @leadercxn Any updates ?

Will you sync to the kirkstone-l4t-r32.7.x branch of the meta-tegra-community repository ?

[ichergui]

Yes, I will do it over the weekend

[ichergui]

Hey @leadercxn

The branch kirkstone-l4t-r32.7.x from meta-tegra-community repo is updated

[dwalkes]

@CHNtentes if you haven't noticed it already you may want to look at the repo at https://github.com/OE4T/tegra-demo-distro which is a fully working build setup with some customized images. The demo images there include a packagegroup-demo-basetests which includes cuda-samples. Regarding how these images are tested you can see previous release tests and test results in this spreadsheet. See the CUDA tests there for the tests we currently have for cuda-samples. Contributions in terms of running or adding tests are welcome!