Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Bump CCCL version to include cuda::std::span fix (#631)
This PR updates the CCCL version to include a fix for `cuda::std::span` which is required for cuCollections to work properly with CCCL 2.5.0. Most of the changes between the last CCCL version bump (#607) and this one were related to doc updates and unit test fixes, so I don't expect much functional impact for RAPIDS. After this PR we likely have to bump the cuco version again to include the new changes. ### CCCL PR: - NVIDIA/cccl#1836 ### CUCO PR: - NVIDIA/cuCollections#502 ### RAPIDS PRs: - [x] rapidsai/cudf#15986 - [error during docs-build](https://github.com/rapidsai/cudf/actions/runs/9406273871/job/25911619452?pr=15946#step:9:2243) but seems unrelated - [x] rapidsai/cugraph#4483 - some CI jobs ran into a network timeout -> rerunning - [x] rapidsai/cuml#5924 - successful apart from some optional conda python tests - [x] rapidsai/raft#2358 - This one is weird as it [shows an error](https://github.com/rapidsai/raft/actions/runs/9475150045/job/26109283214?pr=2358#step:7:824) that is similar to the one we found in cuco when bumping the CCCL version to 2.5.0. However, we thought the problem was on cuco's end and pushed a fix that resolved the issue (see [Slack thread](https://nvidia.slack.com/archives/CCP05T27R/p1718060955876199)) - [x] rapidsai/rmm#1584 Authors: - Daniel Jünger (https://github.com/sleeepyjack) Approvers: - Bradley Dice (https://github.com/bdice) URL: #631
- Loading branch information