enable cuda build

[xiki-tempula]

Checklist

• [x] Used a personal fork of the feedstock to propose changes
• [x] Bumped the build number (if the version is unchanged)
• [ ] Reset the build number to 0 (if the version changed)
• [x] Re-rendered with the latest conda-smithy (Use the phrase code>@<space/conda-forge-admin, please rerender in a comment in this PR for automated rerendering)
• [ ] Ensured the license file is being packaged.

[conda-forge-webservices[bot]]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[xiki-tempula]

@conda-forge-admin, please rerender

[mattwthompson]

119

[xiki-tempula]

The CUDA builds but the MPI+cuda failed due to No space left on device

[xiki-tempula]

@conda-forge-admin, please rerender

[xiki-tempula]

@mattwthompson This working fine when tested locally. Shall we merge this PR?

[xiki-tempula]

@mattwthompson I think we could potentially test cpptraj.cuda. However, I don't know how to test this only when there is cuda. I wonder if I can have some of your help? Thank you.

[mattwthompson]

I think it can be tucked away under a conditional like already happens for MPI: https://github.com/conda-forge/ambertools-feedstock/blob/bc93c747c80f8592d002ea0aa5ac321a82e9fe6f/recipe/run_test.sh#L50-L56

[xiki-tempula]

@conda-forge-admin, please rerender

[xiki-tempula]

@mattwthompson Ok, I have added a which and a lld test for cpptraj.cuda, which seems to pass.

[xiki-tempula]

@conda-forge-admin, please restart ci​

[xiki-tempula]

@conda-forge-admin, please rerender

[xiki-tempula]

@conda-forge-admin, please restart ci​

[xiki-tempula]

@mattwthompson Shall we merge this before working on 24? All the builds are passing and I have test for cpptraj.cuda.

[mattwthompson]

It's probably fine but I'd like somebody from the Amber community to chime in

[xiki-tempula]

@njzjz Do you mind give this PR a review? Thanks.

[dacase]

Is there some way I could provisionally do a conda install of this? It doesn't have to be before the PR is accepted.

Maybe I should just be building this myself locally(?) Are there instructions for doing that? Apologies for being so clueless -- I prepared conda (not conda-forge) packages years ago, but my grey cells are mostly gone in that (and other) areas.....

[mattwthompson]

You can re-configure the build process to store artifacts if they're not already there (I haven't checked), re-render, and then on the next build they're be available to download somewhere on the Azure pages that are linked in "CI" from this PR. Then you can install those as local conda packages

https://conda-forge.org/docs/maintainer/conda_forge_yml/#azure

Everything can be done locally (there's a build_locally.py file somewhere and/or information in the docs https://conda-forge.org/docs/maintainer/) but given the complexity of this feedstock I would worry there's a risk of false positives compared to what happens in automation

[xiki-tempula]

@dacase Building the linux version is quite easy, you just find a linux machine. clone the repo, run build_locally.py and select the relevant option. The linux build is run with docker so it is quite independent of your local environment. Then upload the .conda file to your local conda channel and you can then install it with conda. In our company, we use this branch to build amber, where I change the source to build from local tar.gz file. Then upload the .conda to company conda channel and deploy in our production environment where pmemd.cuda.mpi works as expected.

[mattwthompson]

@xiki-tempula could you merge the main branch into this once more? I don't want to accidentally revert #147

After that I'll merge tomorrow morning if nobody else got to it by then (or objects)