Close CUDA 12.0 migration#5613
Conversation
Now that a large majority of feedstocks have been rebuilt or have an open migration PR, let's go ahead and close out this PR and add CUDA 12 to the matrix. This adds CUDA 12.0 & CUDA 12.4. Where CUDA 12.0 represents the oldest CUDA 12 version built against and supported in conda-forge. Similarly CUDA 12.4 represents the latest CUDA version in conda-forge. This way if users are still have packages that depend on an older CUDA 12 version, packages built with CUDA 12.0 will satisfy that need. Also if users are trying to take advantage of packages built with the latest optimizations, they can install packages built with the latest CUDA Toolkit in conda-forge. We keep the latest version pinned to avoid pulling in partially rolled out CUDA Toolkit updates. Meaning we will need to go back through and bump this number when a new CUDA Toolkit becomes available.
|
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 ( |
|
cc @conda-forge/core |
| - 11.2 # [(linux or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] | ||
| - 11.8 # [(linux or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] | ||
| - 12.0 # [(linux or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] | ||
| - 12.4 # [(linux or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] |
There was a problem hiding this comment.
Are packages built for 12.0 expected to always successfully build with 12.4? Shouldn't we separate the closing of the 12.0 migration from adding 12.4...?
There was a problem hiding this comment.
I remember that in some feedstocks I had to add some code to explicitly use 12.0 to compile. This code should work with 12.4 but as it is not tested, I would feel better if we had a migrator first.
h-vetinari
left a comment
There was a problem hiding this comment.
I don't think we should add 12.4 here; I'd be open to dropping 11.2 in the same PR though.
| # case: native compilation (build == target) | ||
| - quay.io/condaforge/linux-anvil-ppc64le # [ppc64le and os.environ.get("BUILD_PLATFORM") == "linux-ppc64le"] | ||
| - quay.io/condaforge/linux-anvil-aarch64 # [aarch64 and os.environ.get("BUILD_PLATFORM") == "linux-aarch64"] | ||
| # case: cross-compilation (build != target) | ||
| - quay.io/condaforge/linux-anvil-cos7-x86_64 # [ppc64le and os.environ.get("BUILD_PLATFORM") == "linux-64"] | ||
| - quay.io/condaforge/linux-anvil-cos7-x86_64 # [aarch64 and os.environ.get("BUILD_PLATFORM") == "linux-64"] |
There was a problem hiding this comment.
I'm also pretty sure that the selectors here need os.environ.get("CF_CUDA_ENABLED", "False") == "True" (which was active in the migrator through the selector on docker_images).
|
I think we should move forward here with dropping 11.2 and adding 12.0. Let's split off 12.4. |
h-vetinari
left a comment
There was a problem hiding this comment.
The announced date for dropping 11.2 has now passed; given that and the various unanswered review requests here, I've opened #5799 which builds on this PR and incorporates them.
| - 12 # [os.environ.get("cf_cuda_enabled", "false") == "true" and linux] | ||
| - 12 # [os.environ.get("cf_cuda_enabled", "false") == "true" and linux] |
There was a problem hiding this comment.
This is incorrectly lower-cased here (that environment variable is not set by smithy)
| - 12.4 # [(linux or win64) and os.environ.get("CF_CUDA_ENABLED", "False") == "True"] | ||
| cuda_compiler_version_min: | ||
| - None # [osx] | ||
| - 11.2 # [linux or win64] |
There was a problem hiding this comment.
This also needs to be bumped (assuming we drop CUDA 11.2 here).
|
@jakirkham, what's the status here? What still needs to be done from your POV? (since you asked not to close this migration as part of #5799) |
Gentle ping on this |
|
I've opened a new PR for this in #6263 |
|
Made obsolete by #6263 |
Now that a large majority of feedstocks have been rebuilt or have an open migration PR, let's go ahead and close out this PR and add CUDA 12 to the matrix.
This adds CUDA 12.0 & CUDA 12.4. Where CUDA 12.0 represents the oldest CUDA 12 version built against and supported in conda-forge. Similarly CUDA 12.4 represents the latest CUDA version in conda-forge. This way if users are still have packages that depend on an older CUDA 12 version, packages built with CUDA 12.0 will satisfy that need. Also if users are trying to take advantage of packages built with the latest optimizations, they can install packages built with the latest CUDA Toolkit in conda-forge. We keep the latest version pinned to avoid pulling in partially rolled out CUDA Toolkit updates. Meaning we will need to go back through and bump this number when a new CUDA Toolkit becomes available.
xref: #5390
Checklist
0(if the version changed)conda-smithy(Use the phrase@conda-forge-admin, please rerenderin a comment in this PR for automated rerendering)