[fbusato]

Description

Remove load/store static assertion that prevents overaligned_flag_value < alignof(T),
i.e. a user can pass a value for the alignment that is less the alignment of the datatype itself, and the functionality is still valid.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Source mode: Auto
• Repository: NVIDIA/cccl
• Branch: main
• File: .coderabbit.yaml
• Commit: 053d3c7
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Enterprise
• Run ID: a995a0e4-12be-4e6d-aec0-5312233abbd4

📥 Commits

Reviewing files that changed from the base of the PR and between 053d3c7 and 12a6da0.

📒 Files selected for processing: 2

• libcudacxx/include/cuda/std/__simd/utility.h
• libcudacxx/test/libcudacxx/std/numerics/simd/simd.loadstore/weaker_overalignment.pass.cpp

💤 Files with no reviewable changes: 1

• libcudacxx/include/cuda/std/__simd/utility.h

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

📝 Summary

Summary by CodeRabbit

• Bug Fixes
  • SIMD load and store operations now accept alignment flags that specify weaker alignment than the value type’s natural alignment. Runtime alignment checks remain in place. The behavior applies to masked and unmasked operations, including partial operations, across several common numeric types and vector widths.

The SIMD load/store alignment check no longer rejects alignment flags weaker than the element type’s natural alignment at compile time. A new test covers loads and stores with weaker flags across several scalar types, widths, masks, and partial counts.

Changes: SIMD Load/Store Alignment

Layer / File(s)	Summary
Allow and test weaker alignment flags libcudacxx/include/cuda/std/__simd/utility.h, libcudacxx/test/libcudacxx/std/numerics/simd/simd.loadstore/weaker_overalignment.pass.cpp	The compile-time assertion that required the flag alignment to meet the element type’s alignment is removed. The new host/device constexpr test checks unmasked, masked, and partial load/store operations with weaker alignment flags. It verifies values and boundary elements across several scalar types and widths.

Possibly related PRs

• NVIDIA/cccl#12206: It concerns alignment computation used by SIMD load/store paths. This PR tests single alignment flags and does not cover the multi-flag behavior described for that PR.

Priority: ⬇️ Low

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 12a6d

Weaker alignment flags appear safe for naturally aligned pointers. No actionable issue remains before normal checks and merge.

- [ ] Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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

[github-actions]

🥳 CI Workflow Results

🟩 Finished in 2h 00m: Pass: 100%/228 | Total: 2d 18h | Max: 1h 19m | Hits: 85%/792460

See results here.