[#8921][feat] Added symetric memory AllReduce strategy #8919
Conversation
📝 WalkthroughWalkthroughIntroduces symmetric memory-based AllReduce support for H100+ GPUs via a new Changes
Sequence DiagramssequenceDiagram
participant User
participant AllReduce
participant SymmMemAR as SymmetricMemoryAllReduce
participant MNNVL
participant NCCL
User->>AllReduce: forward(input)
alt fusion_op is NONE and symm_mem_allreduce available
AllReduce->>SymmMemAR: forward(input)
SymmMemAR->>SymmMemAR: should_use_symm_mem(input)?
alt tensor compatible
SymmMemAR->>SymmMemAR: copy to symm_mem buffer
SymmMemAR->>SymmMemAR: multimem_all_reduce_ or fallback
SymmMemAR->>SymmMemAR: copy result to output
SymmMemAR-->>AllReduce: return output
else tensor incompatible
SymmMemAR-->>AllReduce: return None
end
end
alt symm_mem_allreduce returned output
AllReduce-->>User: return output (early exit)
else symm_mem_allreduce unavailable or returned None
alt strategy is MNNVL
AllReduce->>MNNVL: forward(input)
MNNVL-->>AllReduce: return output
else strategy is AUTO or fallback
AllReduce->>NCCL: forward(input)
NCCL-->>AllReduce: return output
end
AllReduce-->>User: return output
end
sequenceDiagram
participant Init as Initialization
participant PyTorch
participant Device
participant PG as ProcessGroup
participant SymmMem as Symmetric Memory
Init->>Init: Check PyTorch symm_mem availability
alt available
Init->>Device: Detect SM version & capability
alt H100+ (SM >= threshold)
Init->>Init: Validate world_size support
alt valid
Init->>PG: Create/use TP process group
Init->>SymmMem: Allocate buffer (dtype-sized)
Init->>SymmMem: Create rendezvous handle
Init->>Init: Determine use_multimem capability
Init-->>Init: SymmetricMemoryAllReduce ready ✓
else invalid
Init-->>Init: Disabled (unsupported world_size)
end
else older GPU
Init-->>Init: Disabled (insufficient capability)
end
else unavailable
Init-->>Init: Disabled (PyTorch support missing)
end
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes
Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ 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: 1
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/_torch/distributed/ops.py(5 hunks)tensorrt_llm/_torch/distributed/symm_mem_allreduce.py(1 hunks)tensorrt_llm/functional.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/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.pytensorrt_llm/functional.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/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.pytensorrt_llm/functional.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/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.pytensorrt_llm/functional.py
🧠 Learnings (8)
📚 Learning: 2025-08-14T06:36:40.701Z
Learnt from: timlee0212
Repo: NVIDIA/TensorRT-LLM PR: 6886
File: tensorrt_llm/_torch/models/modeling_deepseekv3.py:0-0
Timestamp: 2025-08-14T06:36:40.701Z
Learning: In DeepSeek V3 model (tensorrt_llm/_torch/models/modeling_deepseekv3.py), the disagreement between AllReduce.__init__ guard and _compute_mlp_tp_size logic for MNNVL usage is expected by design. The AllReduce component and MLP TP-size computation intentionally use different criteria for MNNVL availability decisions.
Applied to files:
tensorrt_llm/_torch/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.py
📚 Learning: 2025-09-24T03:31:28.908Z
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 7520
File: tensorrt_llm/_torch/pyexecutor/resource_manager.py:605-613
Timestamp: 2025-09-24T03:31:28.908Z
Learning: In TensorRT-LLM Ray orchestrator mode, ProcessGroups are initialized with both Gloo and NCCL backends (e.g., "cuda:nccl,cpu:gloo"), allowing PyTorch distributed to automatically route CPU tensors through Gloo and GPU tensors through NCCL. This eliminates the need for manual device placement when performing allreduce operations on base types.
Applied to files:
tensorrt_llm/_torch/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.py
📚 Learning: 2025-10-13T19:45:03.518Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: tests/unittest/_torch/multi_gpu/test_nccl_device.py:138-149
Timestamp: 2025-10-13T19:45:03.518Z
Learning: In test_nccl_device.py, the NCCL device AllReduce implementation compares the entire residual tensor on each rank, unlike the UB implementation which compares per-rank chunks. The residual chunking calculations in the test are intentionally overridden to reflect this design difference.
Applied to files:
tensorrt_llm/_torch/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.py
📚 Learning: 2025-09-23T15:12:38.312Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/thop/allreduceOp.cpp:352-446
Timestamp: 2025-09-23T15:12:38.312Z
Learning: In TensorRT-LLM NCCL device allreduce implementation (cpp/tensorrt_llm/thop/allreduceOp.cpp), the goto pattern in runNCCLAllReduceDeviceFusion is intentionally used for future extensibility, allowing multiple switch cases to fallback to the default handler. While not aesthetically ideal, this pattern supports adding more fusion cases later that can reuse the same fallback logic.
Applied to files:
tensorrt_llm/_torch/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.pytensorrt_llm/functional.py
📚 Learning: 2025-09-23T15:12:38.312Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/thop/allreduceOp.cpp:352-446
Timestamp: 2025-09-23T15:12:38.312Z
Learning: In TensorRT-LLM NCCL device implementation, NCCL version 2.28+ requirements are handled at runtime in the nccl_device/config layer rather than with compile-time guards. This allows the allreduceOp to remain version-agnostic and delegates version compatibility validation to the appropriate lower-level components that can gracefully handle unsupported configurations.
Applied to files:
tensorrt_llm/_torch/distributed/symm_mem_allreduce.pytensorrt_llm/_torch/distributed/ops.py
📚 Learning: 2025-09-02T13:42:44.885Z
Learnt from: pcastonguay
Repo: NVIDIA/TensorRT-LLM PR: 7455
File: tensorrt_llm/_torch/pyexecutor/py_executor.py:1852-1860
Timestamp: 2025-09-02T13:42:44.885Z
Learning: In MPI communication within TensorRT-LLM pipeline parallelism, different communication types (tokens, logits, termination sync) must use disjoint tag namespaces to avoid message routing collisions when using the same source/destination patterns.
Applied to files:
tensorrt_llm/_torch/distributed/ops.py
📚 Learning: 2025-09-16T09:30:09.716Z
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 7763
File: cpp/tensorrt_llm/CMakeLists.txt:297-301
Timestamp: 2025-09-16T09:30:09.716Z
Learning: In the TensorRT-LLM project, NCCL libraries are loaded earlier by PyTorch libraries or the bindings library, so the main shared library doesn't need NCCL paths in its RPATH - the libraries will already be available in the process address space when needed.
Applied to files:
tensorrt_llm/_torch/distributed/ops.py
📚 Learning: 2025-09-23T14:58:05.372Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/kernels/nccl_device/config.cu:42-49
Timestamp: 2025-09-23T14:58:05.372Z
Learning: In TensorRT-LLM NCCL device kernels (cpp/tensorrt_llm/kernels/nccl_device/), the token partitioning intentionally uses ceil-like distribution (same token_per_rank for all ranks) to ensure all ranks launch the same number of blocks. This is required for optimal NCCL device API barrier performance, even though it may launch extra blocks for non-existent tokens on later ranks. Runtime bounds checking in the kernel (blockID validation) handles the overshoot cases.
Applied to files:
tensorrt_llm/_torch/distributed/ops.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/distributed/symm_mem_allreduce.py (2)
tensorrt_llm/mapping.py (1)
Mapping(336-493)tensorrt_llm/_torch/distributed/ops.py (3)
forward(432-503)forward(621-709)forward(747-800)
tensorrt_llm/_torch/distributed/ops.py (2)
tensorrt_llm/_torch/distributed/symm_mem_allreduce.py (1)
SymmetricMemoryAllReduce(30-221)tensorrt_llm/functional.py (6)
AllReduceStrategy(3876-3886)dtype(255-259)dtype(262-267)shape(270-274)shape(277-282)shape(2056-2096)
🪛 Ruff (0.14.3)
tensorrt_llm/_torch/distributed/symm_mem_allreduce.py
45-48: Mutable class attributes should be annotated with typing.ClassVar
(RUF012)
51-61: Mutable class attributes should be annotated with typing.ClassVar
(RUF012)
160-160: Do not catch blind exception: Exception
(BLE001)
tensorrt_llm/_torch/distributed/ops.py
589-589: f-string without any placeholders
Remove extraneous f prefix
(F541)
591-591: Do not catch blind exception: Exception
(BLE001)
⏰ 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 (1)
tensorrt_llm/functional.py (1)
3886-3887: Enum addition verified—existing TRT safeguards already in place.The SYMM_MEM enum addition (lines 3886–3887) is safe and properly commented. The TRT path is already protected: the
distributed/ops.pylayer (lines 679–682) remaps SYMM_MEM to AUTO before the TRT allreduceOp plugin, and the torch runtime correctly handles it via SymmetricMemoryAllReduce (line 578).The suggested guard in
allreduce()is architecturally unnecessary—the safeguard is appropriately delegated to the lower-level distributed layer, consistent with TensorRT-LLM's runtime-handling pattern. No critical issues detected.
MrGeva
changed the title
MrGeva
changed the title
MrGeva
changed the title
|
/bot run |
|
PR_Github #23543 [ run ] triggered by Bot. Commit: |
|
PR_Github #23543 [ run ] completed with state |
Signed-off-by: Eran Geva <[email protected]>
Signed-off-by: Eran Geva <[email protected]>
|
/bot run |
|
PR_Github #23584 [ run ] triggered by Bot. Commit: |
|
PR_Github #23584 [ run ] completed with state |
Signed-off-by: Eran Geva <[email protected]>
|
/bot run |
|
PR_Github #23625 [ run ] triggered by Bot. Commit: |
|
PR_Github #23625 [ run ] completed with state |
Signed-off-by: Eran Geva <[email protected]>
|
/bot run |
|
PR_Github #23663 [ run ] triggered by Bot. Commit: |
| # Use SYMM_MEM strategy (tries symmetric memory first, falls back to AUTO if needed) | ||
| _allreduce_cache[cache_key] = AllReduce( | ||
| mapping=p_config, strategy=AllReduceStrategy.AUTO, dtype=tensor.dtype | ||
| mapping=p_config, strategy=AllReduceStrategy.SYMM_MEM, dtype=tensor.dtype |
There was a problem hiding this comment.
@MrGeva : we should make this configurable and users should be able to specify the choice through inference optimizer config. Would you be able to take on the task of plumbing the AllReduceStrategy from the top level default.yaml -> sharding transformation -> inserting the all reduce op with the specified strategy?
3 participants
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
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.