Conversation
Since 1.2.0, Buffer.from_handle binds a default-stream deallocation token to the current context for every owning memory resource. Memory the device cannot access has no stream ordering to preserve, so skip the binding when mr.is_device_accessible is False. Such buffers record no deallocation stream and never call the driver. Fixes NVIDIA#2769 Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com> (cherry picked from commit 3fef50b)
(cherry picked from commit 9d762dd)
(cherry picked from commit 7cfdad7)
(cherry picked from commit a6f6adc)
(cherry picked from commit c280d47)
(cherry picked from commit f59229e)
Release branch off the cuda-core-v1.2.0 tag. Windows on ARM wheels need the CUDA 13.4 toolkit, so the build moves to CUDA 13.4.2 for every platform (ci/versions.yml and the pixi cuda-version pins). The WoA build jobs themselves are cherry-picked from main in the preceding commits. From the CUDA 13.4 integration on main (NVIDIA#2437, NVIDIA#2788, merged via NVIDIA#2789): - Map the ClocksEventReasons BOARD_LIMIT and RELIABILITY values that CUDA 13.4 added. Device.current_clock_event_reasons and Device.supported_clock_event_reasons raised ValueError for an unknown reason bit on devices that report them. - Map the NVIDIA DLA, vGameDev and NPU brand types instead of "Unknown". - Test updates: accept the "DLA-" UUID prefix, exclude the new 13.4 memory-location enum value from the coverage check, and cover Device.get_all_devices. The 1.2.1 release notes and the install page describe the new wheels: CUDA 13 only, Python 3.11 and newer, cuda-bindings 13.4.1 or newer. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Scheduled pixi update --no-install for this workspace only. Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> (cherry picked from commit 04d4753) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Scheduled pixi update --no-install for this workspace only. Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> (cherry picked from commit 7aa0648) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Andy-Jost
added
P0
github-actions
Bot
added
CI/CD
Contributor
|
This branch builds with CUDA 13.4.2 but tests on 13.3.0, so the test jobs install published cuda-bindings 13.3.x to cover the real-world combination. No 13.3.x wheel exists for Python 3.15, so the three 3.15 and 3.15t test jobs fail at pip install before any test runs. 1.2.1 ships no 3.15 wheels, so nothing is lost; the 3.15 build jobs remain. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Member
|
Q: Is it possible to avoid touching anything under |
Contributor
Author
I don't think so because WoA requires we build with CUDA 13.4. |
isVoid
approved these changes
Sep 23, 2026
Andy-Jost
merged commit Sep 23, 2026
bb136a3
into
NVIDIA:release/cuda-core-v1.2.1
430 of 434 checks passed
Andy-Jost
added a commit
that referenced
this pull request
Sep 30, 2026
…itcher (#2962) The "Choose Version" dropdown on every cuda.core docs page is driven by one file at the docs root, cuda-core/nv-versions.json, which each page fetches by absolute URL. Every docs deploy copies its own checkout's copy over that root file. Deploys from main keep it current; a release built from a release branch overwrote it with that branch's older list, which hid every version released from main after the branch point until the next main deploy restored it (seen with cuda-core-v1.2.1, #2935/#2944). Delete the switcher files from the release build output next to the existing removal of latest/. The deploy action runs with clean: false, so the root copy on gh-pages is left untouched and main alone owns it. The new version's entry is added on main as part of release prep. Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Patch release of cuda.core 1.2.0 that adds Windows on ARM (
win_arm64) wheels. The branch starts at thecuda-core-v1.2.0tag and carries a minimal set of changes; everything else onmainwaits for 1.3.0.Cherry-picked from
main:build-wheel.yml; this branch keeps v4.1.1.main(chore: refresh pixi.lock (cuda_core) #2830, chore: refresh pixi.lock (cuda_bindings) #2828).From the CUDA 13.4 integration (#2437, #2788 via #2789):
ClocksEventReasonsvaluesBOARD_LIMITandRELIABILITYadded in CUDA 13.4. Reading a device's clock-event reasons raisedValueErrorfor an unknown reason bit on hardware that reports them."Unknown".DLA-UUID prefix, exclude the new memory-location enum value from the coverage check, and coverDevice.get_all_devices.Build and docs:
ci/versions.ymland the pixi pins). Windows on ARM toolkits start at 13.4, and one toolkit for all platforms avoids a mixed release.Validation
pre-commit run --all-filespasses.cuda-core-python3XX-win-arm64-*wheel from this PR's CI run on a Windows on ARM system with cuda-bindings 13.4.2 and run thecuda_coretest suite. The clock-event and brand fixes address the failures seen there with 1.2.0.Release plan
Merge here, tag
cuda-core-v1.2.1on the merge commit, wait for the tag-triggered CI run, run the release workflow in dry-run mode, then release.🤖 Generated with Claude Code