Add CUDA 11.8 migrator #4834
Add CUDA 11.8 migrator #4834
Conversation
|
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 ( |
|
Started a couple test PRs using this migration (updated below). These both work:
This PR was created and builds part way ok. Then runs into a tooling issue outside the scope of this change ( some bug in cross-compiling linking: conda-forge/pycolmap-feedstock#7 (comment) ) |
|
It appears that native builds work ok. However CUDA 11.8 cross-compilation support is absent atm. Started PR ( conda-forge/conda-forge-ci-setup-feedstock#261 ) to add CUDA 11.8 cross-compilation support Edit: Marking as draft until CUDA 11.8 cross-compilation support is added |
|
Thoughts on this @conda-forge/core ? |
| - quay.io/condaforge/linux-anvil-cuda:9.2 # [linux64 and os.environ.get("BUILD_PLATFORM") == "linux-64"] | ||
| - quay.io/condaforge/linux-anvil-cuda:10.0 # [linux64 and os.environ.get("BUILD_PLATFORM") == "linux-64"] | ||
| - quay.io/condaforge/linux-anvil-cuda:10.1 # [linux64 and os.environ.get("BUILD_PLATFORM") == "linux-64"] |
There was a problem hiding this comment.
Q: Do we still need these? I thought they are not zippable because cuda_compiler_version does not list them?
There was a problem hiding this comment.
They didn't seem to cause any issues when applying the migration to different feedstocks
Zipping isn't relevant here as this is the ordering section and is only used to preserve the order of these when they do appear
recipe/migrations/cuda118.yaml
Outdated
| # non-CUDA arch: cross-compilation (build != target) | ||
| - quay.io/condaforge/linux-anvil-cos7-x86_64 # [linux64 and os.environ.get("BUILD_PLATFORM") == "linux-64"] |
There was a problem hiding this comment.
Q: Wouldn't this last item be zipped with cuda_compiler_version == 12.0? (I guess I never understood the meaning of ordering in the migrator, even after the CUDA 12 migration. Would be nice if someone could shed some light 🙂)
There was a problem hiding this comment.
(My very same question from the old days: #4400 (comment))
There was a problem hiding this comment.
Applied this to feedstocks that completed the CUDA 12 migration (so they had both migrators like NCCL). AFAICT it worked without issues
This is due to conda-build reordering values of keys (so losing the original ordering). As a result we have to reorder things after the fact in conda-smithy
In any event as the long term goal is to eliminate the need for different Docker images (as is already happening with CUDA 12.0). Think the docker_image ordering can eventually go away. So don't think it is worth thinking to hard about
|
Thanks all! 🙏 Going to go ahead and get this started then. We can follow up on anything else separately |
|
I noticed that and the bot can't see any diff before/after applying the migrator. This would likely happen to other feedstocks |
|
Are there differences in the |
Adds a migrator to include CUDA 11.8 as discussed in issue ( conda-forge/conda-forge.github.io#1981 ).
Checklist
0(if the version changed)conda-smithy(Use the phrase@conda-forge-admin, please rerenderin a comment in this PR for automated rerendering)Fixes conda-forge/conda-forge.github.io#1981
Closes conda-forge/ucx-split-feedstock#133
Closes conda-forge/nccl-feedstock#96
Closes conda-forge/pycolmap-feedstock#7