[cudax] Delete group's copy/move cons + assign op

[davebayer]

It doesn't make sense to have a copy constructor for a group, because you wouldn't be able to supply a new synchronizer for the group.

Moving a group could work, but it brings problems regarding conditional synchronizer destruction and synchronization, because a synchronizer can be in a moved-from state, which would cost some performance. And I don't have any use case for it yet.

Assigning a group is problematic, because basically every group has it's own signature, so it would work only under some very specific conditions.

So my idea is that if you need to pass a group, pass it by reference. If you need to pass it by copy, just pass a group_view.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4bd9f37c-20eb-4eaa-8dd8-defa77928dd4

📥 Commits

Reviewing files that changed from the base of the PR and between 0941c14 and bd5a525.

📒 Files selected for processing (1)

• cudax/include/cuda/experimental/__group/synchronizer/barrier_synchronizer.cuh

Included review availability: Your plan provides up to 12 included reviews per hour; 10 remain after this review.

---

📝 Summary

Summary by CodeRabbit

• API Changes
  • Group-based reduction and summation operations now accept group objects by reference.
  • Group objects and barrier synchronizers can no longer be copied, moved, or assigned.
  • Barrier synchronizers now provide a dedicated non-owning view.
• Behavior
  • Reduction supports groups without statically known sizes.
  • Reduction results and synchronization behavior remain unchanged.
• Tests
  • Updated coverage verifies group references and barrier synchronizer view behavior.

Walkthrough

Changes

Cooperative group types are now non-copyable. Reduction and sum overloads accept groups by const reference. Warp reduction uses group_view, and barrier synchronizers provide non-owning views with updated deinitialization behavior.

Cooperative group API

Layer / File(s)	Summary
Group object lifetime and barrier views cudax/include/cuda/experimental/__group/group.cuh, cudax/include/cuda/experimental/__group/this_group.cuh, cudax/include/cuda/experimental/__group/virtual_group.cuh, cudax/include/cuda/experimental/__group/synchronizer/barrier_synchronizer.cuh, cudax/test/group/synchronizer/barrier_synchronizer.cu	Group types and owning barrier synchronizers delete copy and move operations. Barrier synchronizers return a dedicated non-owning view. Deinitialization validates the mapping and destroys the barrier for the designated thread. Tests expect the new view type.
Reduction reference API cudax/include/cuda/experimental/__coop/reduce.cuh	Internal and public reduction overloads accept groups by const reference. Warp reduction uses group_view for static warp-count lookup. Public static-size assertions are removed.
Sum overload alignment cudax/test/group/cooperative_algorithm.cu	Thread, warp, block, cluster, and generic sum overloads use const-reference group parameters.

Merge Risk: ⚪ Minimal · up to bd5a5

Groups and owning synchronizers are now non-copyable, with reductions borrowing groups by reference and views providing non-owning synchronization access. The remaining deinitialization concern is addressed by requiring a valid mapping, so no merge-blocking current-head risk remains.

---

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

[coderabbitai]

Actionable comments posted: 2

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 8ab2b6dc-2327-435f-8422-c67305be524f

📥 Commits

Reviewing files that changed from the base of the PR and between c6cdc15 and 3067bb6.

📒 Files selected for processing (6)

• cudax/include/cuda/experimental/__coop/reduce.cuh
• cudax/include/cuda/experimental/__group/group.cuh
• cudax/include/cuda/experimental/__group/synchronizer/barrier_synchronizer.cuh
• cudax/include/cuda/experimental/__group/this_group.cuh
• cudax/include/cuda/experimental/__group/virtual_group.cuh
• cudax/test/group/cooperative_algorithm.cu

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

[coderabbitai]

Actionable comments posted: 1

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4bc342df-1e5b-4ec7-a718-e794702e004b

📥 Commits

Reviewing files that changed from the base of the PR and between 3067bb6 and 0941c14.

📒 Files selected for processing (3)

• cudax/include/cuda/experimental/__coop/reduce.cuh
• cudax/include/cuda/experimental/__group/synchronizer/barrier_synchronizer.cuh
• cudax/test/group/synchronizer/barrier_synchronizer.cu

Included review availability: Your plan provides up to 12 included reviews per hour; 11 remain after this review.

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 1h 08m: Pass: 100%/63 | Total: 12h 46m | Max: 48m 58s | Hits: 63%/56982

See results here.