[TRTLLM-12982][perf] remove sync after FlashInfer attention plan()

[ixlmar]

Description

This sync appears not to be necessary.

Test Coverage

pytest -svx "accuracy/test_llm_api_pytorch_multimodal.py::TestNemotron_Nano_12B_V2_VL::test_auto_dtype" still succeeds.

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)

• If PR introduces API changes, an appropriate PR label is added - either api-compatible or api-breaking. For api-breaking, include BREAKING in the PR title.

• Any new dependencies have been scanned for license and vulnerabilities

• CODEOWNERS updated if ownership changes

• Documentation updated as needed

• Update tava architecture diagram if there is a significant design change in PR.

• 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

To see a list of available CI bot commands, please comment /bot help.

Summary by CodeRabbit

• Refactor
  • Optimized internal attention computation performance in the FlashInfer backend.

[ixlmar]

/bot run

[ixlmar]

/bot run

[coderabbitai]

📝 Walkthrough

Walkthrough

This PR removes a single CUDA stream synchronization call from the FlashInfer ragged prefill planning code path that handles the case when no KV cache manager is provided, occurring before the validation and execution of the no-KV ragged planning routine.

Changes

FlashInfer Ragged Prefill Synchronization

Layer / File(s)	Summary
Remove synchronization from ragged prefill planning tensorrt_llm/_torch/attention_backend/flashinfer.py	Deleted a torch.cuda.current_stream().synchronize() call from the no-KV-cache ragged prefill branch before num_contexts validation and _plan_ragged_cudnn_no_kv() invocation.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Possibly related PRs

• NVIDIA/TensorRT-LLM#13428: Introduces FlashInfer MLA planning with explicit synchronization around plan() execution—directly related to CUDA-stream synchronization changes in the same FlashInfer module.

Suggested reviewers

• QiJune
• tburt-nv

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	❓ Inconclusive	The PR description is vague and lacks sufficient detail about the rationale for removing the synchronization call.	Expand the description to explain why the torch.cuda.current_stream().synchronize() call is unnecessary and what performance or correctness implications this change has.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly identifies the main change: removing an unnecessary synchronization call from the FlashInfer attention plan() function for performance improvement.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[tensorrt-cicd]

PR_Github #50553 [ run ] triggered by Bot. Commit: 57838fe Link to invocation

[yechank-nvidia]

Thx for the catch!

[tensorrt-cicd]

PR_Github #50554 [ run ] triggered by Bot. Commit: 57838fe Link to invocation

[tensorrt-cicd]

PR_Github #50553 [ run ] completed with state ABORTED. Commit: 57838fe

Link to invocation

[tensorrt-cicd]

PR_Github #50554 [ run ] completed with state FAILURE. Commit: 57838fe
/LLM/main/L0_MergeRequest_PR pipeline #40055 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[ixlmar]

/bot run

[tensorrt-cicd]

PR_Github #50709 [ run ] triggered by Bot. Commit: 427f7c1 Link to invocation

[tensorrt-cicd]

PR_Github #50709 [ run ] completed with state FAILURE. Commit: 427f7c1
/LLM/main/L0_MergeRequest_PR pipeline #40193 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[ixlmar]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #50755 [ run ] triggered by Bot. Commit: 427f7c1 Link to invocation

[tensorrt-cicd]

PR_Github #50755 [ run ] completed with state FAILURE. Commit: 427f7c1
/LLM/main/L0_MergeRequest_PR pipeline #40234 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[ixlmar]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #50818 [ run ] triggered by Bot. Commit: 427f7c1 Link to invocation

[brb-nv]

LGTM.

[tensorrt-cicd]

PR_Github #50818 [ run ] completed with state FAILURE. Commit: 427f7c1
/LLM/main/L0_MergeRequest_PR pipeline #40288 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[ixlmar]

/bot run

[tensorrt-cicd]

PR_Github #51004 [ run ] triggered by Bot. Commit: 427f7c1 Link to invocation

[tensorrt-cicd]

PR_Github #51004 [ run ] completed with state FAILURE. Commit: 427f7c1
/LLM/main/L0_MergeRequest_PR pipeline #40453 completed with status: 'FAILURE'

CI Report

⚠️ Action Required:

• Please check the failed tests and fix your PR
• If you cannot view the failures, ask the CI triggerer to share details
• Once fixed, request an NVIDIA team member to trigger CI again

CI Agent Failure Analysis

Link to invocation

[ixlmar]

/bot run

[tensorrt-cicd]

PR_Github #51045 [ run ] triggered by Bot. Commit: 427f7c1 Link to invocation

[tensorrt-cicd]

PR_Github #51045 [ run ] completed with state SUCCESS. Commit: 427f7c1
/LLM/main/L0_MergeRequest_PR pipeline #40492 completed with status: 'SUCCESS'

CI Report

Link to invocation