Skip to content

Fix #962: Don't perform unnecessary version checks - #969

Merged
mdboom merged 3 commits into
NVIDIA:mainfrom
mdboom:issue926
Sep 16, 2025
Merged

Fix #962: Don't perform unnecessary version checks#969
mdboom merged 3 commits into
NVIDIA:mainfrom
mdboom:issue926

Conversation

@mdboom

@mdboommdboom commented Sep 15, 2025

Copy link
Copy Markdown
Contributor

Description

closes#962

Checklist

  • New or existing tests cover these changes.
  • The documentation is up to date with these changes.

@copy-pr-bot

copy-pr-botBot commented Sep 15, 2025

Copy link
Copy Markdown
Contributor

Auto-sync is disabled for ready for review pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

@mdboom

mdboom commented Sep 15, 2025

Copy link
Copy Markdown
ContributorAuthor

/ok to test

cpcloud
cpcloud reviewed Sep 15, 2025
View reviewed changes
Comment threadcuda_bindings/cuda/bindings/_internal/cufile_linux.pyx
cdef void* handle = NULL

with gil, __symbol_lock:
driver_ver = get_cuda_version()

@cpcloudcpcloudSep 15, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks like this is being used below and causing at least some of the compilation failures. Does loading the library actually require computing the version?

@mdboommdboomSep 15, 2025

Copy link
Copy Markdown
ContributorAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, it doesn't actually require it. Just need to change what's generated...

@mdboom

mdboom commented Sep 15, 2025

Copy link
Copy Markdown
ContributorAuthor

/ok to test

@leofangleofang added bug Something isn't working cuda.bindings Everything related to the cuda.bindings module P1 Medium priority - Should do labels Sep 15, 2025
@github-actions

This comment has been minimized.

Sign in to view
@mdboom

mdboom commented Sep 15, 2025

Copy link
Copy Markdown
ContributorAuthor

/ok to test

rwgk
rwgk approved these changes Sep 15, 2025
View reviewed changes

@rwgkrwgk left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice cleanup!

Is there an easy/obvious way to specify libcuda_so = 'libcuda.so.1' in one central place? cuda/bindings/_internal/utils.pyx maybe? Maybe even better, move the entire get_cuda_version() function there? — We could also do that later.

Comment threadcuda_bindings/cuda/bindings/_internal/cufile_linux.pyx
Comment threadcuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx
Comment threadcuda_bindings/cuda/bindings/_internal/nvvm_linux.pyx
leofang
leofang requested changes Sep 15, 2025
View reviewed changes
Comment threadcuda_bindings/cuda/bindings/_internal/cufile_linux.pyx

@leofangleofangSep 15, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ralf's comment made me realize... we should also remove get_cuda_version in this file, because it is no longer needed.

@leofangleofangSep 16, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah... I see now why this is challenging: get_cuda_version is a copy-pasta from the extern pxd. It does not seem that we can remove this locally (i.e. without affecting other teams). I'll let Mike make the final call, I run out of good ideas tonight 🙁

@leofangleofangSep 16, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(One silly idea is to split the extern pxd into two, one is always copied/pasted, the other is pasted after the first one only when need_driver is true.)

@mdboommdboomSep 16, 2025

Copy link
Copy Markdown
ContributorAuthor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, get_cuda_version is inserted by the binding generator into every set of bindings. I don't see much need to deal with that, though. It's a static function that the compiler removes if not needed.

Comment threadcuda_bindings/cuda/bindings/_internal/nvjitlink_linux.pyx

@leofangleofangSep 15, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ditto

Comment threadcuda_bindings/cuda/bindings/_internal/nvjitlink_windows.pyx

@leofangleofangSep 15, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ditto

Comment threadcuda_bindings/cuda/bindings/_internal/nvvm_linux.pyx

@leofangleofangSep 15, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ditto

Comment threadcuda_bindings/cuda/bindings/_internal/nvvm_windows.pyx

@leofangleofangSep 15, 2025

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ditto

@mdboom

mdboom commented Sep 16, 2025

Copy link
Copy Markdown
ContributorAuthor

Nice cleanup!

Is there an easy/obvious way to specify libcuda_so = 'libcuda.so.1' in one central place? cuda/bindings/_internal/utils.pyx maybe? Maybe even better, move the entire get_cuda_version() function there? — We could also do that later.

get_cuda_version is needed by other cybind-generated projects, that may not rely on cuda_bindings.

@mdboom
mdboom merged commit 9922034 into NVIDIA:mainSep 16, 2025
49 checks passed
@github-actions

github-actionsBot commented Sep 16, 2025

Copy link
Copy Markdown
Doc Preview CI
Preview removed because the pull request was closed or merged.

mdboom added a commit to mdboom/cuda-python that referenced this pull request Sep 19, 2025
@mdboom
BackportNVIDIA#962andNVIDIA#969to 12.9.x
72607a5
leofang pushed a commit that referenced this pull request Sep 19, 2025
@mdboom
Backport#962and#969to 12.9.x (#992)
553adf7
@cpcloudcpcloud mentioned this pull request Oct 8, 2025
Closed
@mdboom
mdboom deleted the issue926 branch December 9, 2025 16:13
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@rwgkrwgkrwgk approved these changes
@leofangleofangleofang approved these changes
+1 more reviewer
@cpcloudcpcloudcpcloud left review comments
Reviewers whose approvals may not affect merge requirements

Assignees

@mdboommdboom

Labels

bugSomething isn't workingcuda.bindingsEverything related to the cuda.bindings moduleP1Medium priority - Should do

Projects

None yet

Milestone

cuda-python 13.0.2 & 12.9.3

Development

Successfully merging this pull request may close these issues.

Avoid unnecessarily loading and checking driver version

4 participants

@mdboom@cpcloud@rwgk@leofang