feat: Add non UB AR + Residual + Norm + Quant fusion #6320
Conversation
Signed-off-by: Jin Li <[email protected]>
📝 WalkthroughWalkthroughThe changes consolidate the registration of allreduce residual normalization and userbuffer fusion patterns into a single function, add new pattern matchers for fused quantization operations, and remove a dedicated userbuffer allreduce pattern file. Test assertions are updated to reflect the increased number of AR_NORM replacements due to expanded pattern coverage. Changes
Sequence Diagram(s)sequenceDiagram
participant Backend
participant PatternRegistry
participant CustomPasses
Backend->>PatternRegistry: register_ar_fusions(custom_passes, enable_ub)
alt enable_ub is True
PatternRegistry->>CustomPasses: Register UB and non-UB AR norm + quant patterns
else enable_ub is False
PatternRegistry->>CustomPasses: Register only non-UB AR norm + quant patterns
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~15–25 minutes Note ⚡️ Unit Test Generation is now available in beta!Learn more here, or try it out under "Finishing Touches" below. ✨ Finishing Touches
🧪 Generate unit tests
🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. Note: Be mindful of the bot's finite context window. It's strongly recommended to break down tasks such as reading entire modules into smaller chunks. For a focused discussion, use review comments to chat about specific files and their changes, instead of using the PR comments. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
Documentation and Community
|
|
/bot run |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (2)
tensorrt_llm/_torch/compilation/patterns/ar_residual_norm.py (2)
120-283: FP8 quantization fusion patterns correctly implemented.The patterns properly identify allreduce + quantization sequences and replace them with fused operations. The distinction between "out" and regular variants provides flexibility for different use cases.
Consider extracting common pattern elements (e.g., the allreduce CallFunction definition) into shared variables to reduce code duplication between the two FP8 pattern functions.
285-437: FP4 quantization patterns maintain consistency with FP8 structure.The FP4 patterns correctly handle the different quantization function and output structure. The fixed block size of 16 aligns with NVFP4 format requirements.
Similar to FP8 patterns, consider extracting common elements to reduce duplication between the "out" and regular FP4 variants.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
tensorrt_llm/_torch/compilation/backend.py(2 hunks)tensorrt_llm/_torch/compilation/patterns/ar_residual_norm.py(3 hunks)tensorrt_llm/_torch/compilation/patterns/ub_allreduce.py(0 hunks)tests/unittest/_torch/multi_gpu/test_user_buffers.py(2 hunks)
💤 Files with no reviewable changes (1)
- tensorrt_llm/_torch/compilation/patterns/ub_allreduce.py
🧰 Additional context used
🧠 Learnings (1)
tensorrt_llm/_torch/compilation/patterns/ar_residual_norm.py (1)
Learnt from: yechank-nvidia
PR: #6254
File: tensorrt_llm/_torch/pyexecutor/model_engine.py:1201-1204
Timestamp: 2025-07-22T09:22:14.726Z
Learning: In TensorRT-LLM's multimodal processing pipeline, shared tensor recovery using from_shared_tensor() is only needed during the context phase. Generation requests reuse the already-recovered tensor data and only need to call strip_for_generation() to remove unnecessary multimodal data while preserving the recovered tensors. This avoids redundant tensor recovery operations during generation.
🔇 Additional comments (7)
tests/unittest/_torch/multi_gpu/test_user_buffers.py (2)
458-463: Test assertions correctly updated to reflect new pattern match behavior.The update from 1 to 3 AR_NORM replacements aligns with the consolidated pattern registration in the backend, where
register_ar_fusionsnow handles both AR residual norm and UB-related patterns.
1014-1019: Consistent test update for FP4 quantization test.The FP4 test assertions are correctly updated to match the same pattern count changes as the FP8 test, maintaining consistency across quantization formats.
tensorrt_llm/_torch/compilation/backend.py (2)
16-16: Import update reflects consolidated pattern registration approach.The change from
register_ar_residual_normtoregister_ar_fusionsaligns with the new unified registration mechanism that handles both AR residual norm and UB patterns.
78-80: Clean consolidation of pattern registration logic.The refactoring to compute
ub_enabledonce and pass it toregister_ar_fusionsimproves code maintainability by consolidating related pattern registrations into a single function call with explicit configuration.tensorrt_llm/_torch/compilation/patterns/ar_residual_norm.py (3)
101-118: Well-designed helper functions for input validation.The helper functions properly validate input constraints with appropriate type checking before accessing attributes. Good defensive programming practice.
440-717: Comprehensive UB pattern registration with proper operation coverage.The UB patterns correctly handle:
- Conversion of supported AR operations to UB strategy with proper copy/finalize
- Prologue optimizations for various compute operations (scaled_mm, nvfp4_gemm, mm, add)
- Finalize pattern to eliminate redundant operations
The extra checks appropriately limit UB conversion to supported fusion operations.
719-733: Well-structured pattern registration with appropriate conditional logic.The main registration function properly orchestrates pattern registration:
- Base AR residual norm is always registered
- Quantization patterns are added in a separate pass for better organization
- "Out" variants are correctly excluded when UB is enabled (as noted in comment about kernel support)
- UB patterns are conditionally registered based on the enable flag
|
PR_Github #12791 [ run ] triggered by Bot |
|
PR_Github #12791 [ run ] completed with state |
Signed-off-by: Jin Li <[email protected]> Signed-off-by: Shreyas Misra <[email protected]>
Signed-off-by: Jin Li <[email protected]> Signed-off-by: Ransiki Zhang <[email protected]>
Signed-off-by: Jin Li <[email protected]> Signed-off-by: Lanyu Liao <[email protected]>
3 participants
Summary by CodeRabbit
New Features
Bug Fixes
Refactor
Description
Test Coverage
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.
Details
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 thestage-listparameter 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.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip 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-pipelineReuse 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.