nvidia containers support

[edgarriba]

Is there any example or anyone tried cross with any of the NVIDIA L4T containers ? **https://catalog.ngc.nvidia.com/orgs/nvidia/containers/l4t-base

I made a couple of attempts to use as base nvidia but seems that it requires to much configuration happening in this repo.

Any help will pretty much appreciated as rust and cuda builds are starting to get common :)

[Emilgardis]

not sure I understand this issue. What's the end goal?

[Emilgardis]

does these containers contain a cross-compiler (x86_64 -> cuda?) for rust to use?

[edgarriba]

the end goal is being able to cross compile libraries that have to be linked against cuda dependencies (which usually it's a pain to install and that's why L4T or related containers are very useful as they come with all the goodies for cuda). One example would be this crate: https://github.com/mstallmo/tensorrt-rs

As I mentioned above, I tried to take a l4t-base image and install myself cargo as by default those containers do not have any cargo installed.

[Emilgardis]

for cross, there is no need to have cargo in the container. Cross provides cargo from the host in to the running container via mounting. Alright if it's just libraries maybe just copying them into a suitable place could work?

[edgarriba]

when I try to cross build from an nvidia image nvcr.io/nvidia/l4t-base:r35.2.1

I get the following error:

WARNING: The requested image's platform (linux/arm64) does not match the detected host platform (linux/amd64/v4) and no specific platform was requested
/usr/bin/sh: 1: cargo: not found
==> ERROR: Build failed

[Emilgardis]

That just means that something funky is going on, not neccesarily that cargo wasn't found but probably a lib dependency was not found, most likely libc.

I'm a bit confused about the ==> ERROR: Build failed in your snippet. Can you share the entire snippet together with running cross with -v for erbose output