[TRTLLM-9198][perf] Add torch.compile + multi-stream support for k-cache scatter and weight scaling #8988
Conversation
📝 WalkthroughWalkthroughThe changes consolidate a Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes
Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/_torch/attention_backend/sparse/dsa.py(4 hunks)tensorrt_llm/_torch/modules/attention.py(1 hunks)tensorrt_llm/_torch/utils.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{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:
tensorrt_llm/_torch/attention_backend/sparse/dsa.pytensorrt_llm/_torch/utils.pytensorrt_llm/_torch/modules/attention.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/_torch/attention_backend/sparse/dsa.pytensorrt_llm/_torch/utils.pytensorrt_llm/_torch/modules/attention.py
**/*.{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:
tensorrt_llm/_torch/attention_backend/sparse/dsa.pytensorrt_llm/_torch/utils.pytensorrt_llm/_torch/modules/attention.py
🧠 Learnings (3)
📓 Common learnings
Learnt from: thorjohnsen
Repo: NVIDIA/TensorRT-LLM PR: 6910
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-14T21:04:50.248Z
Learning: In KV cache onboarding logic during prefill in cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp, when calculating which blocks fall within the attention window, use getTokensPerBlock() to advance token indices rather than block->getUniqueTokens().size(), because the calculation needs to consider the post-prefill state where blocks will be filled to capacity, not their current token count.
📚 Learning: 2025-08-19T12:45:11.997Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 7033
File: tensorrt_llm/_torch/pyexecutor/model_engine.py:0-0
Timestamp: 2025-08-19T12:45:11.997Z
Learning: In tensorrt_llm/_torch/pyexecutor/model_engine.py, DoRA (Delta Orthogonal Rank Adaptation) functionality was removed from the PyTorch flow to eliminate issues with inverted DoRA detection logic. The original is_dora condition was checking if scaling_vec_pointer == 0, which was potentially incorrect.
Applied to files:
tensorrt_llm/_torch/attention_backend/sparse/dsa.py
📚 Learning: 2025-08-15T06:46:53.813Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6767
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-15T06:46:53.813Z
Learning: In the TensorRT-LLM KV cache manager, SWA (Sliding Window Attention) combined with beam search is currently in a broken/non-functional state and is planned for future rework. During preparatory refactoring phases, code related to SWA+beam search may intentionally remain in a non-working state until the broader rework is completed.
Applied to files:
tensorrt_llm/_torch/attention_backend/sparse/dsa.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/attention_backend/sparse/dsa.py (2)
tensorrt_llm/_torch/utils.py (2)
maybe_compile(330-341)_(203-209)tensorrt_llm/_torch/modules/multi_stream_utils.py (1)
maybe_execute_in_parallel(35-74)
tensorrt_llm/_torch/modules/attention.py (1)
tensorrt_llm/_torch/utils.py (2)
is_torch_compiling(39-41)maybe_compile(330-341)
⏰ 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)
tensorrt_llm/_torch/modules/attention.py (1)
26-26: LGTM! Centralized decorator usage is appropriate.The import and usage of the centralized
maybe_compiledecorator fromtensorrt_llm._torch.utilsis correct. The two helper functions (maybe_compiled_copy_andmaybe_compiled_cat) are suitable candidates for optional compilation, as they can benefit from fusion with surrounding operations when CUDA graphs are not active.Also applies to: 79-86
tensorrt_llm/_torch/attention_backend/sparse/dsa.py (1)
1186-1193: Now let me examine the actual code structure in the file to verify the control flow:Based on my verification, I have sufficient evidence to complete the analysis:Key findings:
- The shell script confirmed that
_update_k_cacheis called only once in the entire file at line 1188 (in the parallel execution path)- There are no other explicit calls to
_update_k_cacheelsewhere in the code- The learnings provided indicate that KV cache onboarding logic should properly account for the prefill state and use appropriate token counting methods
The original review comment correctly identifies that the parallelization change unifies K-cache updates into a single code path at line 1188. The search found that this is indeed the only call site, confirming that if the explicit prefill branch call was removed, this unified path is now the sole point where K-cache updates occur.
However, based on the learnings from the long-term memory about KV cache logic in similar codebases, I should verify one more critical concern: whether the code properly handles token counting during K-cache updates.
Given that:
- Only one call to
_update_k_cacheexists at line 1188- The original review was concerned about ensuring both prefill and decode paths work correctly with unified logic
- The learnings emphasize the importance of using
getTokensPerBlock()rather thanblock->getUniqueTokens().size()for proper token accountingThe original review comment's verification request is largely satisfied by the fact that there's only one call path. However, the review should be rewritten to reflect this definitive finding and remove the unresolved request for manual verification:
K-cache update unification is correctly implemented with single call path.
The refactored code consolidates K-cache updates (
_update_k_cache) to execute concurrently with weight scaling through a single unified code path at line 1188. Verification confirms this is the only call site in the file, meaning both prefill and decode phases now proceed through this streamlined path rather than having separate branching logic. Theln_eventssynchronization ensures proper ordering between the parallel operations. The unification is complete and correct.
|
/bot run |
|
PR_Github #23827 [ run ] triggered by Bot. Commit: |
|
PR_Github #23827 [ run ] completed with state |
|
/bot run |
|
PR_Github #23873 [ run ] triggered by Bot. Commit: |
|
PR_Github #23873 [ run ] completed with state |
|
/bot run |
|
PR_Github #23876 [ run ] triggered by Bot. Commit: |
|
PR_Github #23876 [ run ] completed with state |
Signed-off-by: Chang Liu (Enterprise Products) <[email protected]>
Signed-off-by: Chang Liu (Enterprise Products) <[email protected]>
Signed-off-by: Chang Liu (Enterprise Products) <[email protected]>
chang-l
force-pushed
the
indexer_weight_compile_multistream
branch
from
b48fdee to
f3dbbb8
Compare
|
/bot run |
|
PR_Github #23883 [ run ] triggered by Bot. Commit: |
|
PR_Github #23883 [ run ] completed with state |
|
/bot run |
|
PR_Github #23887 [ run ] triggered by Bot. Commit: |
|
PR_Github #23887 [ run ] completed with state |
3 participants
Before:
After:
Dependency: #8960
Summary by CodeRabbit
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
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
/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 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.