Skip to content

[auto-merge] bot-auto-merge-release/26.08 to main [skip ci] [bot] - #4941

Merged
nvauto merged 2 commits into
mainfrom
bot-auto-merge-release/26.08
Aug 4, 2026
Merged

[auto-merge] bot-auto-merge-release/26.08 to main [skip ci] [bot]#4941
nvauto merged 2 commits into
mainfrom
bot-auto-merge-release/26.08

Conversation

@nvauto

@nvauto nvauto commented Aug 4, 2026

Copy link
Copy Markdown
Collaborator

auto-merge triggered by github actions on bot-auto-merge-release/26.08 to create a PR keeping main up-to-date. If this PR is unable to be merged due to conflicts, it will remain open until manually fix.

res-life and others added 2 commits August 4, 2026 07:44
Fixes #4937.

## Summary

- restore the JNI build toolchain to GCC 14
- replace the `shuffle_split` `std::pair` return type with an aggregate
result
- preserve structured-binding behavior and update the explicit test
container type

## Root cause

CUDA 12.9 `cudafe++` crashes while instantiating GCC 14's concepts-based
`std::pair`
constructor constraints for the shuffle result types. This previously
forced the JNI build back to
GCC 13. ARM64 nvCOMP 5.3 is built with GCC 14 and requires
`__cxa_call_terminate`, which the GCC 13
link does not provide.

Using an aggregate avoids the compiler crash without changing the
shuffle algorithm. GCC 14 can
then link nvCOMP 5.3 and supplies `__cxa_call_terminate` through its
nonshared compatibility
library, without adding a newer dynamic libstdc++ requirement.

This replaces the GCC 13 toolchain workaround introduced by #4468 while
retaining compatibility
with CUDA 12.9.

## Validation

- reproduced the ARM64 GCC 13 final-link failure: nvCOMP 5.3.0.16
contains 20 undefined
`__cxa_call_terminate` references, and GCC 13 does not provide the
symbol
- passed the matching ARM64 Rocky Linux 8 GCC 14 final-link probe; GCC
14 provides
  `__cxa_call_terminate` through `libstdc++_nonshared.a`
- reproduced the `shuffle_split.cu` crash with CUDA 12.9.86 and GCC
14.2.1 before the change
- compiled and linked the JNI native libraries with CUDA 12.9.86, GCC
14.2.1, and nvCOMP 5.3.0.16
  after the change
- verified the resulting `libcudf.so` has no unresolved relocations with
Rocky Linux 8
  `libstdc++.so.6.0.25`
- completed a clean x86_64 package build with CUDA 12.9.86 and GCC
14.2.1
- passed all 18 `SHUFFLE_SPLIT` native tests, including `MixedValidity`
- repository-pinned clang-format v20.1.8 passes

---------

Signed-off-by: Chong Gao <chongg@nvidia.com>
Co-authored-by: Chong Gao <res_life@163.com>
Signed-off-by: spark-rapids automation <70000568+nvauto@users.noreply.github.com>
@nvauto
nvauto requested a review from a team as a code owner August 4, 2026 14:44
@nvauto
nvauto merged commit 3dfb27f into main Aug 4, 2026
1 check was pending
@nvauto

nvauto commented Aug 4, 2026

Copy link
Copy Markdown
Collaborator Author

SUCCESS - auto-merge

@nvauto
nvauto deleted the bot-auto-merge-release/26.08 branch August 4, 2026 14:44
@greptile-apps

greptile-apps Bot commented Aug 4, 2026

Copy link
Copy Markdown
Contributor

Greptile Summary

This automated upmerge advances the cuDF submodule and related dependency pins toward main/26.10, adopts GCC 14 in the build environment, and replaces the shuffle split std::pair return value with a CUDA-compiler-friendly aggregate.

  • Updates the cuDF submodule branch, commit, rapids-cmake SHA, RMM pin, and KvikIO pin.
  • Installs and activates GCC Toolset 14 in CI and local build scripts.
  • Introduces shuffle_split_output while preserving structured-binding compatibility and updating pair-specific test access.

Confidence Score: 5/5

The PR appears safe to merge, with no concrete changed-code failure identified.

The shuffle return-type migration covers pair-specific callers while preserving structured bindings, the compiler package and activation changes are coordinated, and the dependency-pin updates reveal no established incompatibility.

Important Files Changed

Filename Overview
.gitmodules Tracks the cuDF submodule’s main branch while retaining an exact commit pin.
build/env.sh Changes the default compiler activation command from GCC Toolset 13 to 14.
ci/Dockerfile Installs GCC Toolset 14 consistently with the updated build environment default.
src/main/cpp/src/shuffle_split.hpp Adds an aggregate shuffle result type to avoid CUDA 12.9/GCC 14 std::pair constraint instantiation issues.
src/main/cpp/src/shuffle_split.cu Updates the shuffle split definition to return the new aggregate without changing its result construction semantics.
src/main/cpp/tests/shuffle_split.cu Updates pair-specific storage and member access to use shuffle_split_output; existing structured-binding callers remain compatible.
thirdparty/cudf Advances the exact cuDF submodule commit as part of the automated upmerge.
thirdparty/cudf-pins/rapids-cmake.sha Advances the rapids-cmake pin alongside the cuDF update.
thirdparty/cudf-pins/versions.json Advances the KvikIO and RMM commits and versions from 26.08 to 26.10.

Reviews (1): Last reviewed commit: "Auto-merge use main versions" | Re-trigger Greptile

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

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants