Having seen the article " Simplifying CUDA Upgrades for NVIDIA Jetson Users" from last October, I thought I would try upgrading my Orin Developer Toolkit to CUDA 12.2. I used the instructions on the CUDA Toolkit 12.2 Downloads page; that seemed to work fine.
I now need to do the same for my Ubuntu 20.04 host environment (which is where I do compilations using the Eclipse plug-in). The instructions on the same downloads page end up with installation of the cuda-cross-aarch64 package. However, that does not seem to provide quite a lot of the stuff in /usr/local/cuda-12.2 that I think is needed (and is present from when I originally set the host up for CUDA 11.4).
I’m thinking that what I need is the CUDA 12.2 equivalent of section 1.5 (“JetPack Debian Packages on Host”) in the How to Install JetPack document, which ends up with the command
sudo apt-get install cuda-cross-aarch64-11-4 cuda-cupti-cross-aarch64-11-7 cuda-sanitizer-11-7 cuda-toolkit-11-4 libnvvpi2 nsight-compute-2022.2.1 nsight-compute-addon-l4t-2022.2.1 nsight-graphics-for-embeddedlinux-2022.3.0.0 nsight-systems-2022.3.3 nvsci python3.8-vpi2 vpi2-demos vpi2-cross-aarch64-l4t vpi2-dev vpi2-samples
In other words, I think I probably need the CUDA 12.2 equivalent of that, but I don’t know the precise names of the packages to specify for cross-compiling to CUDA 12.2 (and it probably isn’t sensible to try to guess for all of them …).
So if somebody could suggest how I should go about doing this – upgrading the host cross-compile environment to match the CUDA 12.2 upgrade that I have done on my Orin box – I’d be most grateful.
Thanks … Andrew