Rebuild for CUDA 12 w/arch support

[regro-cf-autotick-bot]

This PR has been triggered in an effort to update cuda120.

Notes and instructions for merging this PR:

1. Please merge the PR only after the tests have passed.
2. Feel free to push to the bot's branch to update this PR if needed.

Please note that if you close this PR we presume that the feedstock has been rebuilt, so if you are going to perform the rebuild yourself don't close this PR until the your rebuild has been merged.

---

Here are some more details about this specific migrator:

The transition to CUDA 12 SDK includes new packages for all CUDA libraries and build tools. Notably, the cudatoolkit package no longer exists, and packages should depend directly on the specific CUDA libraries (libcublas, libcusolver, etc) as needed. For an in-depth overview of the changes and to report problems see this issue. Please feel free to raise any issues encountered there. Thank you! :pray:

---

If this PR was opened in error or needs to be updated please add the bot-rerun label to this PR. The bot will close this PR and schedule another one. If you do not have permissions to add this label, you can use the phrase code>@<space/conda-forge-admin, please rerun bot in a PR comment to have the conda-forge-admin add it for you.

_{This PR was created by the regro-cf-autotick-bot. The regro-cf-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. Feel free to drop us a line if there are any issues! This PR was generated by https://github.com/regro/cf-scripts/actions/runs/6638413162, please use this URL for debugging.}

[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.

[github-actions[bot]]

Hi! This is the friendly conda-forge automerge bot!

I considered the following status checks when analyzing this PR:

• linter: passed
• azure: failed

Thus the PR was not passing and not merged.

[dhirschfeld]

The job running on agent Azure Pipelines 31 ran longer than the maximum time of 360 minutes.

😭

[github-actions[bot]]

Hi! This is the friendly conda-forge automerge bot!

Commits were made to this PR after the automerge label was added. For security reasons, I have disabled automerge by removing the automerge label. Please add the automerge label again (or ask a maintainer to do so) if you'd like to enable automerge again!

[xhochy]

Seems like even reducing capabilities doesn't help anymore 😢 I have to revisit my idea of pre-building the LLVM part.

[dhirschfeld]

Only 4 out of 21 failed. Maybe they suffered from noisy-neighbours so took longer than usual to compile?

Is it worthwhile to hit the re-run failed checks button a couple of times to see if they can scrape in?

I'd give it a go myself, but don't have permissions here.

[carterbox]

I'd be interested in y'alls opinions on choosing what archs to target / what peeps are thinking the best way to prune archs to reduce build times. Since y'all seem to be running into the same issue as pytorch and magma when it comes to builds that are too long.

https://github.com/conda-forge/conda-forge.github.io/issues/1901

[dhirschfeld]

I'd be interested in y'alls opinions on choosing what archs to target / what peeps are thinking the best way to prune archs to reduce build times

If pruning versions built could help I'd recommend first getting rid of numpy 1.22 and then python 3.9 builds

[xhochy]

I'd be interested in y'alls opinions on choosing what archs to target / what peeps are thinking the best way to prune archs to reduce build times

If pruning versions built could help I'd recommend first getting rid of numpy 1.22 and then python 3.9 builds

That doesn't help. Pruning CUDA architectures is meant here.

[jakirkham]

Would it make sense to do something similar to what was done for libmagma?

[ngam]

From experience playing around with tf and pt in the past, one would likely need go down to one or two cuda arches. That's too narrow to be a viable solution unless there is a push to adopt differentiating microarches in __cuda or elsewhere