[None][feat] Fix attention sink load in xqa

[qsang-nv]

Summary by CodeRabbit

• Refactor
  • Enhanced internal data handling in GPU kernel operations for improved stability and reliability.

Description

Test Coverage

PR Checklist

Please review the following before submitting your PR:

• PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.

• PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.

• Test cases are provided for new code paths (see test instructions)

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• The reviewers assigned automatically/manually are appropriate for the PR.

• Please check this after reviewing the above items as appropriate for this PR.

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]

Launch build/test pipelines. All previously running jobs will be killed.

--reuse-test (optional)pipeline-id (OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.

--disable-reuse-test (OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-PyTorch-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--test-backend "pytorch, cpp" (OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests in addition to running L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".

--detailed-log (OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.

--debug (OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in the stage-list parameter to access the appropriate container environment. Note: Does NOT update GitHub check status.

For guidance on mapping tests to stage names, see docs/source/reference/ci-overview.md
and the scripts/test_to_stage_mapping.py helper.

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

[qsang-nv]

/bot run --disable-fail-fast

[coderabbitai]

📝 Walkthrough

Walkthrough

Refactored the loadGmemColWiseVecWithDup function in a CUDA kernel to replace reinterpret_cast-based data extraction with an explicit per-element copy loop that includes index clamping and offset calculations.

Changes

Cohort / File(s)	Summary
Kernel data loading refactor cpp/kernels/xqa/mha_sm90.cu	Changed loadGmemColWiseVecWithDup implementation from cast-based extraction to per-element copy loop with computed clamped indices and base offsets. No external interface changes.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

• Index clamping logic: Verify correctness of computed indices and bounds handling
• Offset calculations: Ensure base offset derivation produces equivalent data access patterns as original cast approach
• Data equivalence: Confirm element-wise copy produces identical results to reinterpret_cast method
• CUDA semantics: Review memory access patterns and potential performance implications in kernel context

Pre-merge checks and finishing touches

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Description Check	⚠️ Warning	The PR description is largely incomplete and fails to meet the repository's template requirements. The Description section (intended to explain the issue and solution) contains only a comment placeholder with no actual content. The Test Coverage section is similarly empty with no tests listed. While the PR Checklist is present and one item is checked off, the absence of substantive information in the critical Description and Test Coverage sections means the description does not provide reviewers with essential context about what was changed or why, or what testing validates the fix.
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	You can run @coderabbitai generate docstrings to improve docstring coverage.

✅ Passed checks (1 passed)

Check name	Status	Explanation
Title Check	✅ Passed	The PR title "[None][feat] Fix attention sink load in xqa" follows the required format with a valid ticket reference (None) and type (feat). The title is concise and specific, clearly identifying the main change location (attention sink load in xqa kernel) which aligns with the changeset that modifies the loadGmemColWiseVecWithDup function in cpp/kernels/xqa/mha_sm90.cu. The title appropriately describes the primary fix (attention sink load) without being vague or overly generic.

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

📜 Recent review details

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 98453d2 and ac3c4b1.

📒 Files selected for processing (1)

• cpp/kernels/xqa/mha_sm90.cu (1 hunks)

🧰 Additional context used

📓 Path-based instructions (4)

**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh}

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh}: Namespace closing braces must include a trailing comment with the namespace name (e.g., '} // namespace foo').
Prefer const or constexpr variables over #define for constants.
Declare variables that are not modified after initialization as const.
Avoid magic literals in code; except for 0, nullptr, true, false. Use named constants for comparisons and logic.
Use Allman brace style for formatting.
Place the semicolon of an empty for/while loop on a new line.
Bodies of switch/while/do-while/for must be compound statements (brace-delimited), and if/else must always be followed by brace-delimited statements.
Type names (e.g., classes) must be CamelCase starting with an uppercase letter (e.g., FooBar).
Local variables, methods, and namespaces use lowerCamelCase (e.g., localFooBar).
Non-magic-number global variables that are non-static and not in an anonymous namespace must be lowerCamelCase prefixed with 'g' (e.g., gDontUseGlobalFoos).
Non-magic-number globals that are static or in an anonymous namespace use lowerCamelCase prefixed with 's' (e.g., sMutableStaticGlobal).
Locally visible static variables use lowerCamelCase with 's' prefix (e.g., static std::once_flag sFlag).
Private/protected member variables use 'm' prefix with CamelCase (e.g., mNbFooValues). Public members may omit, but 'm' is encouraged for clarity.
Constants (enums, global constants, static constants, and function-scope magic/literal constants) use uppercase SNAKE_CASE with 'k' prefix (e.g., kDIGIT_NUM).
Function-scope constants that are not magic numbers or literals are named like non-constant variables (e.g., bool const pass = a && b).
If macros are necessary, name them in UPPER_SNAKE_CASE (e.g., FOO_VERSION) and prefer constants over #define.
Use LLVM clang-format; wrap lines at a maximum of 120 columns; use '// clang-format off/on' sparingly with justification.
Use smart pointers for heap allocations; prefer unique_ptr for sole ownership, shared_ptr for shared...

Files:

• cpp/kernels/xqa/mha_sm90.cu

**/*.{cpp,cxx,cc,cu,h,hpp,hh,hxx,cuh}

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

C++ filenames should be lowerCamelCase (first letter lowercase) and must be case-insensitive unique within a compilation target.

Files:

• cpp/kernels/xqa/mha_sm90.cu

**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

Use only spaces, no tabs; indent with 4 spaces.

Files:

• cpp/kernels/xqa/mha_sm90.cu

**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).

Files:

• cpp/kernels/xqa/mha_sm90.cu

🧠 Learnings (2)

📓 Common learnings

Learnt from: MatthiasKohl
Repo: NVIDIA/TensorRT-LLM PR: 6904
File: cpp/tensorrt_llm/kernels/mlaKernels.cu:436-439
Timestamp: 2025-08-14T15:36:37.610Z
Learning: CUDA kernels prioritize performance and should avoid runtime bounds checking or conditional operations that cause branching/warp divergence. Input validation should be done at the host level before kernel launch, not per-thread in the kernel.

Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4616-4626
Timestamp: 2025-08-19T03:35:20.866Z
Learning: In the MOE profiler TMA workspace preparation (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu), the overlapping of TMA WS regions for NONE and FINALIZE variants is deliberate design to save memory space, as confirmed by djns99. The comment "reuse the same pointers to save space" reflects this intentional behavior.

📚 Learning: 2025-08-19T03:35:20.866Z

Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4616-4626
Timestamp: 2025-08-19T03:35:20.866Z
Learning: In the MOE profiler TMA workspace preparation (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu), the overlapping of TMA WS regions for NONE and FINALIZE variants is deliberate design to save memory space, as confirmed by djns99. The comment "reuse the same pointers to save space" reflects this intentional behavior.

Applied to files:

• cpp/kernels/xqa/mha_sm90.cu

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)

• GitHub Check: Pre-commit Check

🔇 Additional comments (2)

cpp/kernels/xqa/mha_sm90.cu (2)

2081-2087: The review comment is based on incorrect assumptions about the code structure and should be disregarded.

The evidence shows:

1. GmmaAccCoreMat::cols = 2 (not variable or large as implied). This is a fixed compile-time constant defining a 2×2 float matrix.

2. The inner loop is fully unrolled: #pragma unroll for (uint32_t j = 0; j < 2; j++) compiles to exactly 2 direct array accesses, not a general "element-wise access pattern."

3. Dynamic bounds checking is necessary for correctness: The call site (line 2950) passes headGrpSize - 1 as the bound parameter. The original reinterpret_cast approach could not respect this dynamic bound, which is why the correctness issue exists (confirmed by git commit "ac3c4b12 fix attention sink load").

4. Performance impact is negligible: A 2-iteration unrolled loop with bounds clamping has minimal overhead compared to unsafe vectorized access. Modern compilers easily optimize this pattern.

The original approach traded off correctness (by ignoring bounds) for an unsafe "optimization." The fix properly restores correctness with negligible performance cost.

Likely an incorrect or invalid review comment.

---

2081-2087: Move bounds validation to host level and remove per-thread clamping in the kernel.

The current implementation performs runtime bounds checking per-thread at line 2081 (clampedIdx = mha::min(i * nbThrdsPerInstNBase + idx, bound)), which introduces unnecessary branching and potential warp divergence. CUDA kernels should validate input bounds at the host level before kernel launch, not within per-thread operations.

Instead, ensure that the host validates bound * GmmaAccCoreMat::cols + (GmmaAccCoreMat::cols - 1) < ctaNbQHeads before calling loadGmemColWiseVecWithDup(attentionSinksVec[0], headGrpSize - 1) at line 2950. If this invariant is guaranteed, remove the per-thread clamping logic at line 2081 and access gmemVec[i * nbThrdsPerInstNBase + idx * GmmaAccCoreMat::cols + j] directly.

⛔ Skipped due to learnings

Learnt from: MatthiasKohl
Repo: NVIDIA/TensorRT-LLM PR: 6904
File: cpp/tensorrt_llm/kernels/mlaKernels.cu:436-439
Timestamp: 2025-08-14T15:36:37.610Z
Learning: CUDA kernels prioritize performance and should avoid runtime bounds checking or conditional operations that cause branching/warp divergence. Input validation should be done at the host level before kernel launch, not per-thread in the kernel.

Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4616-4626
Timestamp: 2025-08-19T03:35:20.866Z
Learning: In the MOE profiler TMA workspace preparation (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu), the overlapping of TMA WS regions for NONE and FINALIZE variants is deliberate design to save memory space, as confirmed by djns99. The comment "reuse the same pointers to save space" reflects this intentional behavior.

Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6915
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:4010-4012
Timestamp: 2025-08-14T23:23:27.449Z
Learning: For MOE (Mixture of Experts) code reviews in TensorRT-LLM, avoid repeatedly suggesting finalize fusion validation checks and safety assertions. The user djns99 has indicated these suggestions are repetitive and unwanted across multiple MOE-related changes.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[tensorrt-cicd]

PR_Github #23145 [ run ] triggered by Bot. Commit: ac3c4b1

[tensorrt-cicd]

PR_Github #23145 [ run ] completed with state SUCCESS. Commit: ac3c4b1
/LLM/main/L0_MergeRequest_PR pipeline #17451 completed with status: 'FAILURE'

[qsang-nv]

/bot run --stage-list "A10-PyTorch-2, A10-PackageSanityCheck-PY310-UB2204-CU12"

[tensorrt-cicd]

PR_Github #23187 [ run ] triggered by Bot. Commit: ac3c4b1

[tensorrt-cicd]

PR_Github #23187 [ run ] completed with state SUCCESS. Commit: ac3c4b1
/LLM/main/L0_MergeRequest_PR pipeline #17478 (Partly Tested) completed with status: 'FAILURE'

[qsang-nv]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #23197 [ run ] triggered by Bot. Commit: ac3c4b1

[tensorrt-cicd]

PR_Github #23197 [ run ] completed with state SUCCESS. Commit: ac3c4b1
/LLM/main/L0_MergeRequest_PR pipeline #17484 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.