Skip to content

Cythonize cuda.core more - #1070

Merged
leofang merged 20 commits into
NVIDIA:mainfrom
leofang:cythonize_more
Oct 7, 2025
Merged

Cythonize cuda.core more#1070
leofang merged 20 commits into
NVIDIA:mainfrom
leofang:cythonize_more

Conversation

@leofang

@leofangleofang commented Oct 2, 2025•
edited
Loading

Copy link
Copy Markdown
Member

Description

Close#1065
Close#1063

Checklist

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

@leofangleofang self-assigned this Oct 2, 2025
@copy-pr-bot

copy-pr-botBot commented Oct 2, 2025

Copy link
Copy Markdown
Contributor

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@leofangleofang added enhancement Any code-related improvements triage Needs the team's attention cuda.core Everything related to the cuda.core module blocked This task is currently blocked by other tasks labels Oct 2, 2025
@leofang

This comment was marked as resolved.

Sign in to view
@leofangleofang added this to the cuda.core beta 7 milestone Oct 2, 2025
@leofangleofang removed the triage Needs the team's attention label Oct 2, 2025
This was referenced Oct 3, 2025
Closed
Merged
@leofangleofang added P0 High priority - Must do! and removed blocked This task is currently blocked by other tasks labels Oct 3, 2025
@leofang

leofang commented Oct 3, 2025

Copy link
Copy Markdown
MemberAuthor

/ok to test b713c5c

leofang
leofang commented Oct 3, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_stream.pyx
@github-actions

This comment has been minimized.

Sign in to view
oleksandr-pavlyk
oleksandr-pavlyk reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_device.pyx
leofang
leofang commented Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_device.pyx
shwina
shwina reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx Outdated
shwina
shwina reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx Outdated
rparolin
rparolin reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx
rparolin
rparolin requested changes Oct 6, 2025
View reviewed changes

@rparolinrparolin left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

I left a comment asking about pointer data type changes and pointer value conversions.

rparolin
rparolin reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_device.pyx
kkraus14
kkraus14 reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_utils/cuda_utils.pyx
Comment threadcuda_core/cuda/core/experimental/_utils/cuda_utils.pyx
Comment threadcuda_core/cuda/core/experimental/_device.pyx
Comment threadcuda_core/cuda/core/experimental/_device.pyx
Comment threadcuda_core/cuda/core/experimental/_memory.pyx Outdated
Comment threadcuda_core/cuda/core/experimental/_memory.pyx Outdated
Comment threadcuda_core/cuda/core/experimental/_memory.pyx
Comment threadcuda_core/cuda/core/experimental/_stream.pyx Outdated
Comment threadcuda_core/cuda/core/experimental/_stream.pyx Outdated
cdef int _get_device_and_context(self) except?-1:
cdef cydriver.CUcontext curr_ctx
if self._device_id == cydriver.CU_DEVICE_INVALID:
# TODO: It is likely faster/safer to call cuCtxGetCurrent?

@kkraus14kkraus14Oct 6, 2025

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

I don't think this is universally true, where a context can be non-current without being destroyed. Retrieving the device and context from that stream object would be valid.

@leofangleofangOct 6, 2025

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

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

I did not bother doing a profiling here, just thought that it might be good to reduce a few CUDA calls... 😛

@kkraus14kkraus14Oct 7, 2025

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Can we remove this TODO before merge? It's not logically correct to call cuCtxGetCurrent here as the Stream object could be from a different context.

@leofangleofangOct 7, 2025

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

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

I see why this TODO is confusing. We actually need the current context in order to do a safe cuCtxGetDevice dance (set the stream ctx to current -> get device -> revert to the current ctx). Let me clarify in the code.

@kkraus14kkraus14Oct 7, 2025

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

I think we can just use cuCtxGetDevice_v2 which allows passing a ctx explicitly and doesn't require messing with the current ctx?

@leofangleofangOct 7, 2025

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

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

See commit 22b0e2e

@leofangleofangOct 7, 2025

Copy link
Copy Markdown
MemberAuthor

Choose a reason for hiding this comment

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

I think we can just use cuCtxGetDevice_v2 which allows passing a ctx explicitly and doesn't require messing with the current ctx?

Unfortunately it seems to be CUDA 13 only... 😢

rparolin
rparolin reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx Outdated
oleksandr-pavlyk
oleksandr-pavlyk reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_device.pyx
rparolin
rparolin reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx
rparolin
rparolin reviewed Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx
@rwgkrwgk mentioned this pull request Oct 6, 2025
Merged
@leofang

leofang commented Oct 6, 2025

Copy link
Copy Markdown
MemberAuthor

/ok to test a5e6bcf

leofang
leofang commented Oct 6, 2025
View reviewed changes
Comment threadcuda_core/cuda/core/experimental/_memory.pyx
@leofang

leofang commented Oct 7, 2025

Copy link
Copy Markdown
MemberAuthor

/ok to test 3809a33

kkraus14
kkraus14 previously approved these changes Oct 7, 2025
View reviewed changes

@kkraus14kkraus14 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Approving as the last change I'd like is trivial

@leofang

leofang commented Oct 7, 2025

Copy link
Copy Markdown
MemberAuthor

/ok to test 22b0e2e

@leofang
leofang merged commit 5669118 into NVIDIA:mainOct 7, 2025
71 checks passed
@leofang
leofang deleted the cythonize_more branch October 7, 2025 18:02
@github-actions

github-actionsBot commented Oct 7, 2025

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

Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@oleksandr-pavlykoleksandr-pavlykoleksandr-pavlyk left review comments
@Andy-JostAndy-JostAndy-Jost left review comments
@rparolinrparolinrparolin approved these changes
@mdboommdboomAwaiting requested review from mdboom
@kkraus14kkraus14Awaiting requested review from kkraus14
@shwinashwinaAwaiting requested review from shwina
+1 more reviewer
@cpcloudcpcloudcpcloud left review comments
Reviewers whose approvals may not affect merge requirements

Assignees

@leofangleofang

Labels

cuda.coreEverything related to the cuda.core moduleenhancementAny code-related improvementsP0High priority - Must do!

Projects

None yet

Milestone

cuda.core beta 7

Development

Successfully merging this pull request may close these issues.

cuda-core: Release GIL when calling cimport'd CUDA APIs [FEA]: Update the memory module to use __dealloc__.

8 participants

@leofang@rparolin@mdboom@cpcloud@shwina@kkraus14@oleksandr-pavlyk@Andy-Jost