Skip to content

[None][fix] gate NCCL window autotune pre-hook - #14566

Merged
liji-nv merged 1 commit into
NVIDIA:feat/deepseek_v4from
liji-nv:fix/nccl-window-autotune-gating
May 27, 2026
Merged

[None][fix] gate NCCL window autotune pre-hook#14566
liji-nv merged 1 commit into
NVIDIA:feat/deepseek_v4from
liji-nv:fix/nccl-window-autotune-gating

Conversation

@liji-nv

@liji-nv liji-nv commented May 26, 2026

Copy link
Copy Markdown
Collaborator

The allreduce autotuner can profile with dummy tensors. When the real input is not an NCCL symmetric window buffer, registering a dummy window input during autotune lets some fallback tactic get incorrect perf result.

Add an NCCL-window buffer detector and include the real input window state in the allreduce runner identity. Only install the autotune pre-hook that allocates a symmetric window input when the real allreduce input already uses the NCCL symmetric memory window.

@coderabbitai summary

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)

  • 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.

The allreduce autotuner can profile with dummy tensors. When the real input is not an NCCL symmetric window buffer, registering a dummy window input during autotune lets NCCL_SYMMETRIC look valid even though runtime later sees a regular tensor and pays the extra copy or falls back.

Add an NCCL-window buffer detector and include the real input window state in the allreduce runner identity. Only install the autotune pre-hook that allocates a symmetric window input when the real allreduce input already uses the NCCL symmetric memory window.

Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
@liji-nv
liji-nv requested a review from a team as a code owner May 26, 2026 09:37
@liji-nv
liji-nv requested review from yizhang-nv and removed request for a team May 26, 2026 09:37
@liji-nv
liji-nv requested review from hyukn and nv-lschneider May 26, 2026 09:38
@liji-nv

liji-nv commented May 26, 2026

Copy link
Copy Markdown
Collaborator Author

/bot run --disable-fail-fast --add-multi-gpu-test

@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50339 [ run ] triggered by Bot. Commit: 4440857 Link to invocation

@nv-lschneider nv-lschneider left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for upgrading the logic for the auto-tuning.

I think we should strive to get more inputs to be window backed if possible, but this is the correct logic for all inputs we haven't upgraded yet.

Comment thread cpp/tensorrt_llm/thop/allreduceOp.cpp
@tensorrt-cicd

Copy link
Copy Markdown
Collaborator

PR_Github #50339 [ run ] completed with state SUCCESS. Commit: 4440857
/LLM/main/L0_MergeRequest_PR pipeline #39866 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

CI Report

Link to invocation

@liji-nv
liji-nv merged commit 435a0a2 into NVIDIA:feat/deepseek_v4 May 27, 2026
10 of 11 checks passed
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 10, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 11, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 12, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jun 25, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
(cherry picked from commit 435a0a2)
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jul 2, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
(cherry picked from commit 435a0a2)
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jul 2, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
(cherry picked from commit 435a0a2)
lfr-0531 pushed a commit to lfr-0531/TensorRT-LLM that referenced this pull request Jul 2, 2026
Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
(cherry picked from commit 435a0a2)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants