Use cuda::stream_ref in remaining core libcudf APIs by vyasr · Pull Request #23691 · NVIDIA/cudf

vyasr · 2026-08-17T22:49:48Z

Description

This fifth batch migrates the remaining core libcudf headers and implementations from rmm::cuda_stream_view to cuda::stream_ref.

It intentionally leaves the central stream helpers, such as get_default_stream(), the stream pool, fork_streams, and join_streams, on rmm::cuda_stream_view so this PR does not require Java, pylibcudf, or libcudf_streaming compatibility changes. Those helper updates are split into a downstream follow-up: vyasr#9

Contributes to #23636

Checklist

I am familiar with the Contributing Guidelines.
New or existing tests cover these changes.
The documentation is up to date with these changes.

copy-pr-bot · 2026-08-17T22:49:51Z

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

Contributors can view more details about this message here.

bdice · 2026-08-17T23:25:12Z

  cudf::table_view const& table)
 {
-  auto stream = rmm::cuda_stream_default;
+  auto stream = cuda::stream_ref{};


I think the default constructors like this are deprecated. Please check that. I see this in a few examples.

Thanks, fixed in the split benchmarks/examples PR: #23697. I replaced the nullary cuda::stream_ref{} uses in the examples with explicit cuda::stream_ref{cudaStreamLegacy}.

I think you want cudaStreamDefault and not cudaStreamLegacy. We should respect compiling with PTDS enabled, which changes how cudaStreamDefault is interpreted.

bdice · 2026-08-17T23:28:38Z

  [[nodiscard]] bool may_evaluate_null(table_view const& left,
                                       table_view const& right,
-                                       rmm::cuda_stream_view stream) const override;
+                                       cuda::stream_ref stream) const override;


This file appears to be missing <cuda/stream> includes. Same for cpp/src/ast/jit/expressions.cpp and others in cpp/src/ast. Can you check everything for IWYU?

Done here. I audited the AST/JIT changes and added direct <cuda/stream> includes where cuda::stream_ref is used, including this file and the corresponding AST/JIT implementation files.

Follow-up: I found one remaining IWYU miss while rechecking this thread: cpp/include/cudf/ast/expressions.hpp also declares cuda::stream_ref and was still relying on transitive includes. Fixed in 5ab1aa5855 by adding a direct <cuda/stream> include there as well.

bdice · 2026-08-17T23:29:25Z

    _null_handling{null_handling},
    _mr{std::move(mr)},
-    _d_agg_kinds{0, rmm::cuda_stream_default, cudf::get_current_device_resource_ref()},
+    _d_agg_kinds{0, cuda::stream_ref{}, cudf::get_current_device_resource_ref()},


I think the nullary constructor is deprecated.

Done here. This now uses an explicit legacy stream construction: cuda::stream_ref{cudaStreamLegacy}.

bdice · 2026-08-17T23:34:08Z

 #else
-rmm::cuda_stream_view const default_stream_value{};
+cuda::stream_ref const default_stream_value{};


This should be explicit and not default-constructed. But should it be set to cudaStreamLegacy or cudaStreamDefault? Probably Legacy?

Suggested change

cuda::stream_ref const default_stream_value{};

cuda::stream_ref const default_stream_value{cuda::stream_ref{cudaStreamLegacy}};

Done here. I used the explicit legacy/default-stream spelling for the non-PTDS case: cuda::stream_ref const default_stream_value{cudaStreamLegacy}.

bdice · 2026-08-17T23:36:56Z

 #include <rmm/device_uvector.hpp>
 #include <rmm/exec_policy.hpp>

+#include <cuda/stream_ref>


There are a few files like this with header changes but no changes in the code itself. I don't think the header is used. Can we audit for those files and remove the headers if they're not required?

Done in the split tests PR: #23696. I audited the test-only header additions and removed the unused stream headers.

coderabbitai · 2026-08-18T16:55:10Z

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

@coderabbitai resume to resume automatic reviews.
@coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

▶️ Resume reviews
🔍 Trigger review

📝 Walkthrough

Summary by CodeRabbit

Refactor
- Updated CUDA stream handling across public APIs to use the modern CUDA stream reference interface.
- Updated stream access and synchronization while preserving existing operation behavior.
- Applied consistently across datetime, interop, hashing, quantiles, reshaping, filtering, table, and utility operations.
Documentation
- Updated developer guidance and examples to reflect the current CUDA stream interface.
Chores
- Removed legacy stream-header usage and refreshed copyright notices.

Walkthrough

This change migrates cuDF stream APIs and implementations from rmm::cuda_stream_view to cuda::stream_ref. CUDA calls now use get(), synchronization uses sync(), and stream-related headers use <cuda/stream>.

Changes

CUDA stream API migration

Layer / File(s)	Summary
Public and internal API migration `cpp/include/cudf/...`, `cpp/src/...`	Stream parameters, return types, stored stream members, constructors, dispatchers, and helper APIs now use `cuda::stream_ref`.
Native CUDA stream access `cpp/include/cudf/detail/...`, `cpp/src/...`	Kernel launches, CUB calls, CUDA runtime calls, and event operations now obtain native stream handles with `stream.get()`.
Synchronization and default-stream handling `cpp/include/cudf/utilities/...`, `cpp/src/utilities/...`	Synchronous operations now call `stream.sync()`. Default-stream and stream-pool implementations now use CUDA stream references directly.
Benchmarks, examples, tests, and docs `cpp/benchmarks/...`, `cpp/examples/...`, `cpp/tests/...`, `cpp/doxygen/...`	Benchmark and example code construct `cuda::stream_ref` values, obtain native handles with `.get()`, and synchronize with `.sync()`. Documentation examples show the updated stream type.

Estimated code review effort: 4 (Complex) | ~60 minutes

Merge Risk: 🟠 High · up to c97b1

The stream API migration still contains multiple potential build blockers and a stream-ordering correctness hazard, including outdated synchronization calls, a malformed CUDA error-check statement, and operations that may run on the wrong stream. These issues should be fixed or explicitly accepted before merging.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 18.48% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly identifies the primary change: migrating remaining core libcudf APIs to cuda::stream_ref.
Description check	✅ Passed	The description directly explains the cuda::stream_ref migration and its compatibility scope.

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

_{Comment @coderabbitai help to get the list of available commands.}

vyasr · 2026-08-20T20:47:35Z

/ok to test

vyasr · 2026-08-21T01:33:57Z

/ok to test

bdice · 2026-08-21T08:54:38Z

Follow-up: can we document in the developer guide when to use stream.sync() vs. cudf::detail::sync_stream(stream)?

vyasr · 2026-08-21T16:07:41Z

Follow-up: can we document in the developer guide when to use stream.sync() vs. cudf::detail::sync_stream(stream)?

Yup, happy to do that.

…ef-batch-5 # Conflicts: # cpp/include/cudf/ast/expressions.hpp

This reverts commit c3f0834.

…ef-batch-5 # Conflicts: # cpp/src/interop/from_arrow_host.cu

…ef-batch-5 # Conflicts: # cpp/src/utilities/host_memory.cpp

misiugodfrey

Looks good to me. I see a few new includes, but I assume they were suggested by IWYU since they seem to be required.

vyasr · 2026-08-21T21:04:56Z

/merge

github-actions Bot assigned vyasr Aug 17, 2026

github-actions Bot added libcudf Affects libcudf (C++/CUDA) code. Python Affects Python cuDF API. Java Affects Java cuDF API. pylibcudf Issues specific to the pylibcudf package labels Aug 17, 2026

github-project-automation Bot added this to cuDF Python and cuDF Python Aug 17, 2026

github-project-automation Bot moved this to Todo in cuDF Python Aug 17, 2026

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from fa33c11 to 834a687 Compare August 17, 2026 23:02

vyasr changed the title ~~Use cuda::stream_ref across remaining RAPIDS surfaces~~ Use cuda::stream_ref across remaining libcudf C++ surfaces Aug 17, 2026

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from 834a687 to ee2961a Compare August 17, 2026 23:03

This was referenced Aug 17, 2026

Use cuda::stream_ref in Java native code #23692

Closed

Use cuda::stream_ref in libcudf_streaming #23693

Closed

Use cuda::stream_ref in Python stream shims #23694

Draft

bdice reviewed Aug 17, 2026

View reviewed changes

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from ee2961a to d6cfbf8 Compare August 17, 2026 23:48

vyasr changed the title ~~Use cuda::stream_ref across remaining libcudf C++ surfaces~~ Use cuda::stream_ref across remaining libcudf APIs Aug 17, 2026

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from d6cfbf8 to dd55488 Compare August 17, 2026 23:49

This was referenced Aug 17, 2026

Use cuda::stream_ref in libcudf tests #23696

Closed

Use cuda::stream_ref in benchmarks and examples #23697

Closed

vyasr changed the title ~~Use cuda::stream_ref across remaining libcudf APIs~~ Use cuda::stream_ref in core libcudf APIs Aug 18, 2026

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from c60ef6b to 75513d2 Compare August 18, 2026 14:20

vyasr added improvement Improvement / enhancement to an existing function non-breaking Non-breaking change labels Aug 18, 2026 — with ChatGPT Codex Connector

vyasr marked this pull request as ready for review August 18, 2026 16:43

vyasr requested a review from a team as a code owner August 18, 2026 16:43

vyasr requested review from lamarrr and misiugodfrey August 18, 2026 16:43

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from 6893200 to c54a021 Compare August 20, 2026 20:47

vyasr removed request for a team and mythrocks August 21, 2026 01:15

vyasr removed the request for review from a team August 21, 2026 02:19

vyasr marked this pull request as ready for review August 21, 2026 03:08

vyasr added 5 commits August 20, 2026 22:44

Use cuda::stream_ref across remaining libcudf APIs

51f4fbc

Preserve RMM stream view for pylibcudf slice

55a3262

Preserve RMM stream views in pylibcudf interop

6c5d1d0

Guard stream_ref host synchronization

26c69af

Drop unnecessary pylibcudf stream bridge changes

03a2c15

vyasr force-pushed the codex/cuda-stream-ref-batch-5 branch from 5fe10e0 to 03a2c15 Compare August 21, 2026 05:45

bdice approved these changes Aug 21, 2026

View reviewed changes

vyasr added 4 commits August 21, 2026 09:09

Merge remote-tracking branch 'upstream/main' into codex/cuda-stream-r…

ecc8cc0

…ef-batch-5 # Conflicts: # cpp/include/cudf/ast/expressions.hpp

Preserve PTDS default stream semantics

c3f0834

Revert "Preserve PTDS default stream semantics"

1040417

This reverts commit c3f0834.

Merge remote-tracking branch 'upstream/main' into codex/cuda-stream-r…

dc48949

…ef-batch-5 # Conflicts: # cpp/src/interop/from_arrow_host.cu

lamarrr approved these changes Aug 21, 2026

View reviewed changes

Merge remote-tracking branch 'upstream/main' into codex/cuda-stream-r…

cf17939

…ef-batch-5 # Conflicts: # cpp/src/utilities/host_memory.cpp

misiugodfrey approved these changes Aug 21, 2026

View reviewed changes

Comment thread cpp/include/cudf/detail/utilities/batched_memset.hpp

mythrocks approved these changes Aug 21, 2026

View reviewed changes

vyasr merged commit 8c029c2 into NVIDIA:main Aug 21, 2026
142 of 146 checks passed

github-project-automation Bot moved this to Done in cuDF Python Aug 21, 2026

github-project-automation Bot moved this from Todo to Done in cuDF Python Aug 21, 2026

	cuda::stream_ref const default_stream_value{};
	cuda::stream_ref const default_stream_value{cuda::stream_ref{cudaStreamLegacy}};

Uh oh!

Conversation

vyasr commented Aug 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Checklist

Uh oh!

copy-pr-bot Bot commented Aug 17, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bdice Aug 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

coderabbitai Bot commented Aug 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviews paused

Summary by CodeRabbit

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

vyasr commented Aug 20, 2026

Uh oh!

vyasr commented Aug 21, 2026

Uh oh!

bdice commented Aug 21, 2026

Uh oh!

vyasr commented Aug 21, 2026

Uh oh!

misiugodfrey left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

vyasr commented Aug 21, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

vyasr commented Aug 17, 2026 •

edited

Loading

bdice Aug 19, 2026 •

edited

Loading

coderabbitai Bot commented Aug 18, 2026 •

edited

Loading