Skip to content

[RELEASE] rmm v25.10 #2052

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 36 commits into
base: main
Choose a base branch
from
Open

[RELEASE] rmm v25.10 #2052

wants to merge 36 commits into from
+1,700 −338

Conversation

AyodeAwe
Copy link
Contributor

❄️ Code freeze for branch-25.10 and v25.10 release

What does this mean?

Only critical/hotfix level issues should be merged into branch-25.10 until release (merging of this PR).

What is the purpose of this PR?

  • Update documentation
  • Allow testing for the new release
  • Enable a means to merge branch-25.10 into main for the release

AyodeAwe and others added 30 commits July 17, 2025 12:36
@AyodeAwe
DOC v25.10 Updates [skip ci]
078c924
@GPUtester
Merge pull request #1990 from rapidsai/branch-25.08
c40c95f 
Forward-merge branch-25.08 into branch-25.10
@GPUtester
Merge pull request #1991 from rapidsai/branch-25.08
ed44697 
Forward-merge branch-25.08 into branch-25.10
@GPUtester
Merge pull request #1993 from rapidsai/branch-25.08
e447cd5 
Forward-merge branch-25.08 into branch-25.10
@bdice
Allow latest OS in devcontainers (#1997)
c99163d 
This PR removes the OS suffix from devcontainers, allowing the upstream devcontainer images to determine the OS version.

Contributes to rapidsai/build-planning#200.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)

URL: #1997
@vyasr
Use GCC 14 in conda builds. (#1963)
3bab478 
conda-forge is migrating to gcc 14, so this PR is updating for alignment.

See rapidsai/build-planning#188

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - Jake Awe (https://github.com/AyodeAwe)
  - Bradley Dice (https://github.com/bdice)

URL: #1963
@robertmaynard
Update build infra to support new branching strategy (#1994)
1d304e2 
rapids_config will use `RAPIDS_BRANCH` contents to determine what branch to use

Authors:
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - Nghia Truong (https://github.com/ttnghia)
  - Bradley Dice (https://github.com/bdice)

URL: #1994
@vyasr
Use std::for_each instead of std::equal in test (#1999)
6e5b65e 
`std::equal` was basically being used to create a loop, which `std::for_each` is better suited for. See discussion at #1963 (comment)

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Yunsong Wang (https://github.com/PointKernel)

URL: #1999
@bdice
Move more implementations to precompiled shared library (part 2) (#1983)
e1d777a 
Continues from #1980.

Contributes to #1779.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)
  - Rong Ou (https://github.com/rongou)

URL: #1983
@stepsecurity-app
[StepSecurity] Apply security best practices (#2000)
8600dbe 
## Summary

This pull request has been generated by [StepSecurity](https://app.stepsecurity.io/github/rapidsai/actions/dashboard) as part of your enterprise subscription to ensure compliance with recommended security best practices. Please review and merge the pull request to apply these security enhancements.

## Security Fixes

### Harden Runner

Harden-Runner is an open-source security agent for the GitHub-hosted runner to prevent software supply chain attacks. It prevents exfiltration of credentials, detects tampering of source code during build, and enables running jobs without sudo access.

- [GitHub Security Guide](https://docs.github.com/en/actions/security-for-github-actions/security-guides/security-hardening-for-github-actions)
- [The Open Source Security Foundation (OpenSSF) Security Guide](https://github.com/ossf/scorecard/blob/main/docs/checks.md#dangerous-workflow)
### Least Privileged GitHub Actions Token Permissions

The GITHUB_TOKEN is an automatically generated secret to make authenticated calls to the GitHub API. GitHub recommends setting minimum token permissions for the GITHUB_TOKEN.

- [GitHub Security Guide](https://docs.github.com/en/actions/security-guides/automatic-token-authentication#using-the-github_token-in-a-workflow)
- [The Open Source Security Foundation (OpenSSF) Security Guide](https://github.com/ossf/scorecard/blob/main/docs/checks.md#token-permissions)


## Feedback
For bug reports, feature requests, and general feedback; please create an issue in [step-security/secure-repo](https://github.com/step-security/secure-repo) or contact us via [our website](https://www.stepsecurity.io/).

Authors:
  - https://github.com/apps/stepsecurity-app

Approvers:
  - AJ Schmidt (https://github.com/ajschmidt8)

URL: #2000
@bdice
Fix aligned resource adaptor bug when alignment is less than 256. (#2003
fd2b7c5 
)

Fixes #2002.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - David Wendt (https://github.com/davidwendt)
  - Lawrence Mitchell (https://github.com/wence-)

URL: #2003
@robertmaynard
Update to use CUDA runtime APIs that work in CUDA 12 and 13 (#2004)
b6bb091 
CUDA 13 removes the `cudaMemAdvise` and `cudaMemPrefetchAsync` function signatures used by RMM. To continue to support CUDA 12 and 13 we move over to the v2 APIs when in CUDA 12, which are now the unversioned APIs in 13.

Authors:
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #2004
@gforsyth
ci(labeler): update labeler action to @v5 (#2006)
72b1563 
Bumps the version of `actions/labeler` to `@v5` and updates the syntax in
the `labeler.yml` file to account for breaking changes in that version bump.

xref: rapidsai/ops#2968

Authors:
  - Gil Forsyth (https://github.com/gforsyth)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - James Lamb (https://github.com/jameslamb)

URL: #2006
@KyleFromNVIDIA
Update rapids-build-backend to 0.4.1 (#2007)
daf60f2 
Issue: rapidsai/build-planning#207

Authors:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #2007
@GPUtester
Merge pull request #2009 from rapidsai/branch-25.08
9f7a447 
Forward-merge branch-25.08 into branch-25.10
@bdice
Skip test on HMM systems with older CUDA drivers (#1944)
48492f1 
This works around the failures observed in #1935 by disabling system memory resource tests on systems with HMM and CUDA drivers older than 12.8 (R575).

I did some local testing with driver R550, which is old enough to reproduce the bug when used with a Linux kernel new enough to support HMM. I believe the system memory resource works as intended on HMM systems with earlier CUDA drivers, but it appears that [is_device_accessible_memory](https://github.com/rapidsai/rmm/blob/e1d777a5e3dd1306970432a35d0f868ccf67f523/cpp/tests/mr/device/test_utils.hpp#L30-L35) returns false for the system-allocated pointers. This may be a driver bug, as the pointers seem to be device-accessible if, e.g., accessed from a CUDA kernel.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Mark Harris (https://github.com/harrism)
  - Lawrence Mitchell (https://github.com/wence-)
  - Robert Maynard (https://github.com/robertmaynard)

URL: #1944
@bdice
Add numba-cuda to fix errors on driver 580+. (#2013)
1d3a01c 
This adds a dependency on `numba-cuda`. Previously, RMM depended on only `numba`, and the CUDA target in `numba` is [deprecated in favor of `numba-cuda`](https://numba.readthedocs.io/en/stable/cuda/overview.html#cuda-deprecation-status). The fixes needed to support CUDA driver 580+ are in `numba-cuda`, so we adopt it here.

The version pinning chosen here is `>=0.14.0,<0.15.0a0` to align with the current version pinning used elsewhere in RAPIDS (cudf, ucxx, etc.).

See also: numba/numba#10166

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - James Lamb (https://github.com/jameslamb)

URL: #2013
@robertmaynard
Update rapids_config to handle user defined branch name (#2010)
8474569 
rapids_config will use a user defined branch over `RAPIDS_BRANCH` contents

Authors:
  - Robert Maynard (https://github.com/robertmaynard)

Approvers:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)
  - Jake Awe (https://github.com/AyodeAwe)
  - Bradley Dice (https://github.com/bdice)

URL: #2010
@bdice
Update to numba-cuda >=0.18.0,<0.19.0a0 (#2019)
b80ebf7 
This PR updates the numba-cuda version to `>=0.18.0,<0.19.0a0`.

xref: rapidsai/cudf#19604

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Vyas Ramasubramani (https://github.com/vyasr)

URL: #2019
@davidwendt
Change pylibrmm Stream.view() method to noexcept (#2020)
1f69b36 
Changes the pylibrmm `Stream.view()` method from `except *` to `noexcept`.
This prevents  many warnings where the `view()` method. Here is an example:
https://github.com/rapidsai/cudf/actions/runs/17040978623/job/48305629155?pr=19446#step:12:735

```
  performance hint: /__w/cudf/cudf/python/pylibcudf/pylibcudf/reduce.pyx:57:20: Exception check after calling 'view' will always require the GIL to be acquired. Declare 'view' as 'noexcept' if you control the definition and you're sure you don't want the function to raise exceptions.
```

Also changed the `is_default()` to `noexcept` since the underlying librmm function is also declared as `noexcept`.

Authors:
  - David Wendt (https://github.com/davidwendt)

Approvers:
  - Matthew Murray (https://github.com/Matt711)
  - Bradley Dice (https://github.com/bdice)

URL: #2020
@bdice
Add CUDA 13.0.0 devcontainers (#2012)
d5f1641 
This PR adds CUDA 13.0.0 devcontainers.

Contributes to rapidsai/build-planning#208.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - James Lamb (https://github.com/jameslamb)

URL: #2012
@bdice
Minor improvements to README (#2018)
3e00b2f 
Some minor improvements to the README.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Tom Augspurger (https://github.com/TomAugspurger)

URL: #2018
@jameslamb
Build and test with CUDA 13.0.0 (#2021)
2589030 
Contributes to rapidsai/build-planning#208
    
This PR uses CUDA 13.0.0 to build and test.

Authors:
  - James Lamb (https://github.com/jameslamb)
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #2021
@trxcllnt
Use build cluster in devcontainers (#2016)
b6014ee 
RAPIDS has deployed an autoscaling cloud build cluster that can be used to accelerate building large RAPIDS projects. This contributes to rapidsai/build-planning#209.

Authors:
  - Paul Taylor (https://github.com/trxcllnt)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #2016
@bdice
Update dependencies to align with other repos adopting CUDA 13 (#2022)
6fc6831 
Aligns with changes being made in other repos for CUDA 13 bringup.

Contributes to rapidsai/build-planning#208.
Also contributes to rapidsai/build-planning#68.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)

URL: #2022
@bdice
Refactor device capability checks, fix prefetching bug on WSL (#2014)
81d566d 
Refactors device capability checks, add concurrent managed access checks to skip prefetching tests on WSL (not supported).

One motivation for this refactoring is that the previous names really didn't make much sense. We created the `runtime_async_alloc::is_supported()` to check for CUDA 11.2+ support for runtime async allocation. However, we tacked on other features like `is_export_handle_type_supported(cudaMemAllocationHandleType)` and `is_hwdecompress_supported()` into the same struct. Those should have been their own features, not tied to `runtime_async_alloc`. Most of the changes in this PR are structurally similar to that.

Closes #2023.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Tom Augspurger (https://github.com/TomAugspurger)
  - Matthew Murray (https://github.com/Matt711)
  - Mark Harris (https://github.com/harrism)
  - Vyas Ramasubramani (https://github.com/vyasr)

URL: #2014
@bdice
Require numba-cuda>=0.19.1 to get fixes for numba==0.60 support (#2024)
93a84d4 
We need some fixes from NVIDIA/numba-cuda#424 to fully support `numba==0.60` and CUDA 13. This bumps the `numba-cuda` dependency to `>=0.19.1,<0.20.0a0`.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - James Lamb (https://github.com/jameslamb)

URL: #2024
@KyleFromNVIDIA
Update rapids-dependency-file-generator (#2027)
7cef2f5 
This PR updates the rapids-dependency-file-generator hook to get rapidsai/dependency-file-generator#163.

Authors:
  - Kyle Edwards (https://github.com/KyleFromNVIDIA)

Approvers:
  - James Lamb (https://github.com/jameslamb)

URL: #2027
@jameslamb
Use branch-25.10 again (#2030)
e023f81 
Contributes to rapidsai/build-planning#208

Now that rapidsai/shared-workflows#413 is
merged, this converts all GitHub Actions references from `@cuda13.0`
back to `branch-25.10`.

## Notes for Reviewers

This is safe to admin-merge because the change is a no-op... configs on
those 2 branches are identical.
@vyasr
Always release the GIL before any operation that could acquire locks …
50d25c4 
…from the CUDA runtime (#2028)

This PR avoids deadlocks resulting from two Python threads simultaneously contending for both the GIL and the CUDA lock where one is running rmm code and the other is running CUDA code from another Python package calling CUDA functions.

Resolves #2026

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #2028
jameslamb and others added 5 commits September 9, 2025 21:11
@jameslamb
update dependencies: cuda-python (>=12.9.2 / >=13.0.1) (#2031)
b0a2a8a 
Contributes to rapidsai/build-planning#214, updating pins to match the rest of RAPIDS:

* `cuda-python` (>=12.9.2 for CUDA 12, >=13.0.1 for CUDA 13)

Authors:
  - James Lamb (https://github.com/jameslamb)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #2031
@bdice
Skip callback memory tests on integrated memory systems, add docs (#2025
163c84a 
)

Closes #1941.

Authors:
  - Bradley Dice (https://github.com/bdice)

Approvers:
  - Yunsong Wang (https://github.com/PointKernel)
  - Lawrence Mitchell (https://github.com/wence-)

URL: #2025
@bdice
Support building with CCCL 3.1.0 (#2017)
de85ed3 
This PR allows building with CCCL 3.1.0. Contributes to #2011.

Authors:
  - Bradley Dice (https://github.com/bdice)
  - Basit Ayantunde (https://github.com/lamarrr)

Approvers:
  - Nghia Truong (https://github.com/ttnghia)
  - Vyas Ramasubramani (https://github.com/vyasr)

URL: #2017
@AyodeAwe
Configure repo for automatic release notes generation (#2037)
4cd5dbf 
This PR configures this repo (adds a `.github/release.yml` file) for
automatic release notes generation. See
https://docs.github.com/en/repositories/releasing-projects-on-github/automatically-generated-release-notes.
This will fix such issues as hotfix releases including wrong content
(e.g [cugraph
v25.04.01](https://github.com/rapidsai/cugraph/releases/tag/v25.04.01)
contains 50+ PRs from the full v25.04.00 release when it should only
contain the single hotfix PR (#5017).)

Example auto-generated release notes:
https://github.com/rapidsai/literate-octo-potato/releases/tag/v25.10.00a
@wence-
Add option to create non-blocking streams (#2036)
278d559 
Add optional flag argument to both `cuda_stream` and `cuda_stream_pool` creation to allow caller to specify whether the resulting stream should be implicitly synchronising with the default stream.

- Closes #2029

Authors:
  - Lawrence Mitchell (https://github.com/wence-)

Approvers:
  - Mark Harris (https://github.com/harrism)
  - Bradley Dice (https://github.com/bdice)

URL: #2036
@AyodeAwe AyodeAwe requested review from a team as code owners September 25, 2025 17:21
@AyodeAwe AyodeAwe requested review from msarahan and removed request for a team September 25, 2025 17:21
@AyodeAwe AyodeAwe requested review from vyasr and shrshi and removed request for a team September 25, 2025 17:21
@copy-pr-bot copy-pr-bot
Copy link

copy-pr-bot bot commented Sep 25, 2025

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.

@github-actions github-actions bot added CMake Python Related to RMM Python API conda ci labels Sep 25, 2025
@bdice
Fix bug in pinned_memory_resource with CCCL 3.1.x. (#2058)
5b46576 
## Description
Closes #2057.

This fix only affects builds with CCCL 3.1.0+.

## Checklist
- [x] I am familiar with the [Contributing
Guidelines](https://github.com/rapidsai/rmm/blob/HEAD/CONTRIBUTING.md).
- [x] New or existing tests cover these changes.
- [x] The documentation is up to date with these changes.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@msarahan msarahan Awaiting requested review from msarahan msarahan is a code owner automatically assigned from rapidsai/packaging-codeowners

@vyasr vyasr Awaiting requested review from vyasr vyasr is a code owner automatically assigned from rapidsai/rmm-cmake-codeowners

@shrshi shrshi Awaiting requested review from shrshi shrshi is a code owner automatically assigned from rapidsai/rmm-cpp-codeowners

Assignees
No one assigned
Labels
ci CMake conda Python Related to RMM Python API
Projects
RMM Project Board
Status: No status
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.