Upgrade CUDA to 11.6 Update 2? about autoware

[yukkysaito]

To be sure, I am currently checking tit works properly with cuda 11.4.
After that, I will check it works with cuda11.6.

[esteve]

@kenji-miyake is it possible to mix the rest of CUDA 11.6 with libnvinfer-plugin 11.4? If not, I'd wait until the new version is released.

[kenji-miyake]

is it possible to mix the rest of CUDA 11.6 with libnvinfer-plugin 11.4? If not, I'd wait until the new version is released.

Actually, I believe it's impossible. We have to wait for the release in any case.

Also, I'm concerned about the dependencies in https://github.com/autowarefoundation/autoware.universe.
@yukkysaito knows well about this.

[kenji-miyake]

@mitsudome-r @xmfcx @yukkysaito @esteve What do you think about this?

[kenji-miyake]

It seems there is no 11.6 for libnvinfer yet.
https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/

[kenji-miyake]

The new version should be available in both x86_64(amd64) and sbsa(arm64).
#154 (comment)

https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/
https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/sbsa/

[kenji-miyake]

I've created a helper tool to find CUDA versions.
https://github.com/kenji-miyake/find-cuda-versions

Now we can find if there are any corresponding versions to a certain CUDA version.

$ pip install git+https://github.com/kenji-miyake/find-cuda-versions.git
$ find-cuda-versions --cuda 11.6
[cudnn x86_64]
8.4.0.27-1+cuda11.6

[cudnn sbsa]
8.4.0.27-1+cuda11.6

[tensorrt x86_64]

[tensorrt sbsa]

[stale]

This pull request has been automatically marked as stale because it has not had recent activity.

[kenji-miyake]

Updated to 11.6 in the humble branch.
https://github.com/autowarefoundation/autoware.ai/blob/d2feaef806f93ceb60ae50d49ae98d904fef3e09/amd64.env#L5

But we should also consider updating to 11.7 because 11.6 isn't released to ubuntu 22.04 repo.
https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/

Currently, we use 20.04 repo with ubuntu 22.04 because 11.7 doesn't have corresponding cuDNN and TensorRT yet.