Skip to content

Update all the things to CTK 13.3.#9143

Merged
alliepiper merged 12 commits into
NVIDIA:mainfrom
alliepiper:ctk-13_3-update
May 30, 2026
Merged

Update all the things to CTK 13.3.#9143
alliepiper merged 12 commits into
NVIDIA:mainfrom
alliepiper:ctk-13_3-update

Conversation

@alliepiper
Copy link
Copy Markdown
Contributor

@alliepiper alliepiper commented May 27, 2026

mostly devcontainer updates from the version bump.

Waiting for https://github.com/rapidsai/devcontainers/actions/runs/26529587350 to test.

@copy-pr-bot
Copy link
Copy Markdown
Contributor

copy-pr-bot Bot commented May 27, 2026

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.

@cccl-authenticator-app cccl-authenticator-app Bot moved this from Todo to In Progress in CCCL May 27, 2026
@alliepiper
Copy link
Copy Markdown
Contributor Author

/ok to test

@alliepiper alliepiper linked an issue May 27, 2026 that may be closed by this pull request
1 task
@github-project-automation github-project-automation Bot moved this from In Progress to In Review in CCCL May 27, 2026
@github-actions

This comment has been minimized.

@miscco
Copy link
Copy Markdown
Contributor

miscco commented May 28, 2026

/ok to test 8a20921

@github-actions

This comment has been minimized.

@miscco
Copy link
Copy Markdown
Contributor

miscco commented May 28, 2026

/ok to test f08a271

@github-actions

This comment has been minimized.

…13.3

MSVC 14.44/14.50 + nvcc 13.3 incorrectly fires C4189 on constexpr locals
used only as non-type template parameters; [[maybe_unused]] suppresses it.
Separately, MSVC CTAD synthesis rejects env_t's constructor when the first
parameter type is a private typedef; spelling out any_resource<_Properties...>
directly avoids the private-name access.
@alliepiper
Copy link
Copy Markdown
Contributor Author

/ok to test b5644c3

@github-actions

This comment has been minimized.

@miscco
Copy link
Copy Markdown
Contributor

miscco commented May 29, 2026

pre-commit.ci autofix

@copy-pr-bot
Copy link
Copy Markdown
Contributor

copy-pr-bot Bot commented May 29, 2026

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.

@davebayer
Copy link
Copy Markdown
Contributor

/ok to test 1197d54

@github-actions

This comment has been minimized.

@alliepiper
Copy link
Copy Markdown
Contributor Author

pre-commit.ci autofix

@alliepiper alliepiper marked this pull request as ready for review May 29, 2026 19:38
@alliepiper alliepiper requested review from a team as code owners May 29, 2026 19:38
@alliepiper alliepiper enabled auto-merge (squash) May 29, 2026 19:39
@coderabbitai
Copy link
Copy Markdown
Contributor

coderabbitai Bot commented May 29, 2026

Review Change Stack

📝 Walkthrough

Summary by CodeRabbit

  • Chores

    • Updated development container base images to a newer version across all CUDA and compiler combinations.
    • Updated default CUDA version in CI workflows and documentation from 13.2 to 13.3.
  • Tests

    • Added compiler version compatibility guards to test suites for NVCC 13.3+.
    • Fixed compiler warnings by marking unused test constants with [[maybe_unused]] attributes.

Walkthrough

Devcontainer base image tags (26.0626.08) and CUDA versions updated across configurations. CUDA 13.2 devcontainer variants replaced with CUDA 13.3 equivalents. Workflow defaults and CI matrix updated to reference CUDA 13.3. Test files add compiler-version guards and suppress warnings via [[maybe_unused]]. CUDAX type alias refactoring reduces code repetition.

Changes

Devcontainer Infrastructure Update

Layer / File(s) Summary
Image version bump across CUDA 12.x and 13.0-13.1 variants
.devcontainer/cuda12.0-*/devcontainer.json, .devcontainer/cuda12.9-*/devcontainer.json, .devcontainer/cuda13.0-*/devcontainer.json, .devcontainer/cuda13.1-*/devcontainer.json
Base image tag updated from rapidsai/devcontainers:26.06 to 26.08 across all CUDA 12.0, 12.9, 13.0, 13.1 variants with GCC and LLVM compilers.
CUDA 13.2 → 13.3 migration for devcontainer variants
.devcontainer/cuda13.3-*/devcontainer.json, .devcontainer/devcontainer.json
CUDA 13.3 variants replace 13.2 with coordinated updates: image tag to 26.08-cpp-*-cuda13.3, container name suffix to cuda13.3-*, environment variables (DEVCONTAINER_NAME, CCCL_CUDA_VERSION, CCCL_BUILD_INFIX), and top-level name field.
Workflow and documentation CUDA version defaults
.github/workflows/bench.yml, .github/workflows/build-matx.yml, .github/workflows/build-pytorch.yml, .github/workflows/git-bisect.yml, AGENTS.md, ci/bench.template.yaml, ci/bench.yaml, ci/matrix.yaml
Workflow dispatch input defaults, example commands, and CI configuration updated from --cuda 13.2 to --cuda 13.3; ci/matrix.yaml bumps devcontainer_version from 26.06 to 26.08 and replaces CTK version 13.2 with 13.3.

Test and Compiler Infrastructure Updates

Layer / File(s) Summary
Compiler warning suppressions in tests
cub/test/catch2_test_device_for_each_in_layout.cu, cub/test/catch2_test_device_histogram_custom_policy_hub.cu, cub/test/catch2_test_device_scan_by_key_large_offsets.cu, cub/test/catch2_test_device_segmented_topk_keys.cu, cudax/test/copy/copy_common.cuh
Local test variables and unused constexpr constants marked with [[maybe_unused]] to suppress compiler warnings; test formatting standardized.
NVCC 13.3+ version-conditional test compilation
libcudacxx/test/CMakeLists.txt, libcudacxx/test/libcudacxx/cuda/utility/in_range.pass.cpp, libcudacxx/test/libcudacxx/std/text/charconv/charconv.to.chars/integral/int16.pass.cpp, libcudacxx/test/libcudacxx/std/utilities/memory/allocator.traits/allocator.traits.types/*.pass.cpp
Allocator-traits test cases conditionally excluded for NVCC >= 13.3; in_range assertions guarded for version-specific issues; charconv workaround target condition changed from NV_IS_EXACTLY_SM_120 to NV_IS_DEVICE; CMake tile program version requirement updated to 13.3.
CUDAX type alias refactoring
cudax/include/cuda/experimental/__execution/env.cuh, cudax/include/cuda/experimental/__stf/graph/graph_ctx.cuh, cudax/include/cuda/experimental/__stf/stream/stream_ctx.cuh
Internal classes introduce type aliases to reduce fully-qualified base type references; env_t constructor parameter simplified from __resource alias to explicit public ::cuda::mr::any_resource<_Properties...> type.

Suggested labels

libcu++

Suggested reviewers

  • miscco

Comment @coderabbitai help to get the list of available commands and usage tips.

coderabbitai[bot]

This comment was marked as resolved.

@alliepiper alliepiper disabled auto-merge May 29, 2026 20:15
@github-actions
Copy link
Copy Markdown
Contributor

🥳 CI Workflow Results

🟩 Finished in 4h 29m: Pass: 100%/501 | Total: 4d 10h | Max: 59m 01s | Hits: 99%/631536

See results here.

@alliepiper alliepiper merged commit d7c4233 into NVIDIA:main May 30, 2026
1151 of 1156 checks passed
@github-project-automation github-project-automation Bot moved this from In Review to Done in CCCL May 30, 2026
@alliepiper alliepiper deleted the ctk-13_3-update branch May 30, 2026 01:29
shwina pushed a commit to shwina/cccl that referenced this pull request Jun 1, 2026
* Update all the things to CTK 13.3.
* [MSVC] Avoid unused variable warning
* [libcu++] Disable tests that are technically invalid and are now warned
* [libcu++] Disable runtime issue with `in_range`
  Looks like there is a miscompilation in 13.3 with the result of the comparison when NaN is involved.
* [libcu++] Expand workaround for extreme value
* Also disable buggy test for nvrtc
* [STF] Add missing include
* [CUB] More unused variable warnings
* [STF] Avoid NVCC compiler bug about dependent base classes
* Work around MSVC C4189 false-positives and CTAD regression under CTK 13.3
  MSVC 14.44/14.50 + nvcc 13.3 incorrectly fires C4189 on constexpr locals
  used only as non-type template parameters; [[maybe_unused]] suppresses it.
  Separately, MSVC CTAD synthesis rejects env_t's constructor when the first
  parameter type is a private typedef; spelling out any_resource<_Properties...>
  directly avoids the private-name access.

---------

Co-authored-by: Michael Schellenberger Costa <miscco@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

Archived in project

Development

Successfully merging this pull request may close these issues.

[INFRA]: Update PR testing to use CTK 13.3 as the latest CUDA version

4 participants