Enable cuTENSOR preload support for CUDA 13.x binary packages #462
Conversation
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
Co-authored-by: kmaehashi <[email protected]>
Co-authored-by: kmaehashi <[email protected]>
Copilot
AI
changed the title
[WIP] Fix build scripts to preload cuTENSOR for cupy-cuda13x
Enable cuTENSOR preload support for CUDA 13.x binary packages
Oct 17, 2025
Copilot finished work on behalf of
kmaehashi
October 17, 2025 08:39
kmaehashi
approved these changes
Oct 17, 2025
Member
|
xref #9429 |
Member
|
/test |
7 tasks
Member
|
Will retrigger CI after merging cupy/cupy#9492 |
Member
|
/test mini |
Member
|
/test mini |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
Problem
The CUDA 13.x configurations in
dist_config.pyhad TODO comments indicating that cuTENSOR support needed to be re-added. This meant that CuPy wheels built for CUDA 13 (cupy-cuda13x) could not preload cuTENSOR, unlike CUDA 11.x and 12.x packages which already had this functionality.Solution
This PR enables cuTENSOR preloading for CUDA 13.x packages by:
Adding cuTENSOR to Linux x86_64 preloads: Updated the CUDA 13.x configuration to include
'cutensor'in the preloads list alongside'nccl', matching the pattern used in CUDA 11.x and 12.x.Adding cuTENSOR to Windows preloads: Updated the CUDA 13.x Windows configuration to include
'cutensor'in the preloads list.Maintaining aarch64 consistency: Kept the CUDA 13.x aarch64 configuration with only
'nccl'in preloads, following the established pattern for aarch64 builds across all CUDA versions (11.x, 12.x, and 13.x).Updating tests: Removed the skip condition in
test_preload.pythat prevented cuTENSOR testing on CUDA 13+.Changes
dist_config.py:
'preloads': ['nccl']→'preloads': ['cutensor', 'nccl'](CUDA 13.x x86_64)'preloads': []→'preloads': ['cutensor'](CUDA 13.x Windows)release-tests/pkg_wheel/test_preload.py:
Verification
All linters pass (ruff, autopep8, mypy) and the configurations now follow the established pattern from CUDA 11.x and 12.x releases.
Original prompt
💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.