[https://nvbugs/5597647][fix] Fix MNNVL Allreduce accuracy issue on Hopper #8612
Conversation
Signed-off-by: Shiyu Li <[email protected]>
Signed-off-by: Shiyu Li <[email protected]>
📝 WalkthroughWalkthroughThe changes expand CUDA inline assembly support in mnnvlTwoShotAllreduceKernels to lower GPU architectures (from 700+) by adopting release-variant intrinsics, and enable a previously-skipped test in the mnnvl_allreduce test suite. Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~12 minutes Pre-merge checks and finishing touches❌ Failed checks (1 warning)
✅ Passed checks (2 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 |
|
/bot run --add-multi-gpu-test |
![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 (2)
cpp/tensorrt_llm/kernels/communicationKernels/mnnvlTwoShotAllreduceKernels.cu(1 hunks)tests/unittest/_torch/multi_gpu/test_mnnvl_allreduce.py(0 hunks)
💤 Files with no reviewable changes (1)
- tests/unittest/_torch/multi_gpu/test_mnnvl_allreduce.py
🧰 Additional context used
📓 Path-based instructions (4)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh}: Namespace closing braces must include a trailing comment with the namespace name (e.g., '} // namespace foo').
Prefer const or constexpr variables over #define for constants.
Declare variables that are not modified after initialization as const.
Avoid magic literals in code; except for 0, nullptr, true, false. Use named constants for comparisons and logic.
Use Allman brace style for formatting.
Place the semicolon of an empty for/while loop on a new line.
Bodies of switch/while/do-while/for must be compound statements (brace-delimited), and if/else must always be followed by brace-delimited statements.
Type names (e.g., classes) must be CamelCase starting with an uppercase letter (e.g., FooBar).
Local variables, methods, and namespaces use lowerCamelCase (e.g., localFooBar).
Non-magic-number global variables that are non-static and not in an anonymous namespace must be lowerCamelCase prefixed with 'g' (e.g., gDontUseGlobalFoos).
Non-magic-number globals that are static or in an anonymous namespace use lowerCamelCase prefixed with 's' (e.g., sMutableStaticGlobal).
Locally visible static variables use lowerCamelCase with 's' prefix (e.g., static std::once_flag sFlag).
Private/protected member variables use 'm' prefix with CamelCase (e.g., mNbFooValues). Public members may omit, but 'm' is encouraged for clarity.
Constants (enums, global constants, static constants, and function-scope magic/literal constants) use uppercase SNAKE_CASE with 'k' prefix (e.g., kDIGIT_NUM).
Function-scope constants that are not magic numbers or literals are named like non-constant variables (e.g., bool const pass = a && b).
If macros are necessary, name them in UPPER_SNAKE_CASE (e.g., FOO_VERSION) and prefer constants over #define.
Use LLVM clang-format; wrap lines at a maximum of 120 columns; use '// clang-format off/on' sparingly with justification.
Use smart pointers for heap allocations; prefer unique_ptr for sole ownership, shared_ptr for shared...
Files:
cpp/tensorrt_llm/kernels/communicationKernels/mnnvlTwoShotAllreduceKernels.cu
**/*.{cpp,cxx,cc,cu,h,hpp,hh,hxx,cuh}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
C++ filenames should be lowerCamelCase (first letter lowercase) and must be case-insensitive unique within a compilation target.
Files:
cpp/tensorrt_llm/kernels/communicationKernels/mnnvlTwoShotAllreduceKernels.cu
**/*.{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:
cpp/tensorrt_llm/kernels/communicationKernels/mnnvlTwoShotAllreduceKernels.cu
**/*.{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:
cpp/tensorrt_llm/kernels/communicationKernels/mnnvlTwoShotAllreduceKernels.cu
⏰ 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
| #elif (defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 700)) | ||
| asm volatile("red.release.global.gpu.add.u32 [%0], %1;" ::"l"(offset_access_ptr), "r"(1) : "memory"); |
There was a problem hiding this comment.
🧩 Analysis chain
Verify PTX instruction support and consider architectural consistency.
The addition of release memory semantics correctly fixes the synchronization issue by ensuring all prior memory writes are visible before the atomic counter increment. This is essential for the Lamport flag protocol's correctness.
However, several points require verification:
-
PTX instruction support: The threshold change from sm_90 to sm_70 extends this path to Volta, Turing, and Ampere architectures. Verify that
red.release.global.gpu.add.u32is supported on sm_70-sm_89. -
Scope broader than PR title: The PR title mentions "Fix MNNVL Allreduce accuracy issue on Hopper", but this change affects all architectures >= sm_70. If the issue exists on these older architectures too, the PR description should reflect this broader impact.
-
Architectural consistency: Other parts of the file still use
__CUDA_ARCH__ >= 900checks (lines 154, 217, 436 forcudaGridDependencySynchronize()andcudaTriggerProgrammaticLaunchCompletion()). Should the synchronization strategy be consistent across these checks?
Additionally, please search the web for PTX ISA documentation to confirm red.release.global.gpu.add.u32 support:
🏁 Script executed:
#!/bin/bash
# Verify PTX instruction support across architectures
# Search for documentation or comments about PTX instruction requirements
rg -n "red\.release\.global" --type=cu --type=cpp -C3
# Check if there are other uses of architecture-specific synchronization
rg -n "__CUDA_ARCH__.*>=.*(700|900|1000)" --type=cu --type=cpp -C2
# Look for any existing tests or validation of this instruction on different architectures
rg -n "sm_[0-9]+" --type=cu --type=cpp -g "*test*" -g "*mnnvl*" -C2Length of output: 80
🌐 Web query:
PTX ISA red.release.global instruction support compute capability sm_70 sm_80
💡 Result:
Short answer: Yes — PTX reduction instructions with the .release (e.g., red.release.global.*) are supported on sm_70 and on sm_80 (sm_80 ≥ sm_70). The PTX docs show the .sem/.release qualifiers require compute capability sm_70 or higher, and sm_80 is a later target that includes those features. [1][2]
Sources:
- NVIDIA PTX / Parallel Thread Execution target-ISA notes (.sem / .release require sm_70+). [1]
- PTX ISA reference (examples and target notes; sm_80 is a supported target and includes qualifiers introduced for sm_70+). [2]
---
🏁 Script executed:
```shell
#!/bin/bash
# Find the file and examine architecture-specific checks
fd mnnvlTwoShotAllreduceKernels.cu -x cat {} | head -n 300
Length of output: 10462
🏁 Script executed:
#!/bin/bash
# Search for all __CUDA_ARCH__ comparisons in the file
fd mnnvlTwoShotAllreduceKernels.cu -x grep -n "__CUDA_ARCH__" {}Length of output: 297
🏁 Script executed:
#!/bin/bash
# Check for other synchronization primitives and their arch requirements
fd mnnvlTwoShotAllreduceKernels.cu -x grep -n "cudaGridDependencySynchronize\|cudaTriggerProgrammaticLaunchCompletion\|red\." {}Length of output: 429
Architectural mismatch: Grid dependency synchronization still requires sm_90+, making the code incomplete for sm_70-sm_89.
The PTX instruction red.release.global.gpu.add.u32 is confirmed supported on sm_70+, and the instruction change is technically correct.
However, the change reveals a critical architectural inconsistency:
The Lamport flag synchronization now extends to sm_70+ (Volta), but the grid dependency synchronization primitives remain sm_90+ only:
- Line 155:
cudaGridDependencySynchronize()— sm_90+ only - Line 218:
cudaTriggerProgrammaticLaunchCompletion()— sm_90+ only - Line 460:
cudaTriggerProgrammaticLaunchCompletion()— sm_90+ only
On sm_70-sm_89 GPUs, these lines are skipped, leaving the kernel without proper grid-level synchronization. Since cudaGridDependencySynchronize() is called at the start of the kernel (line 155) and cudaTriggerProgrammaticLaunchCompletion() at the end (line 218), skipping them breaks the synchronization protocol.
Required:
Either extend grid dependency synchronization support to sm_70+ to match the Lamport flag change, or explicitly document why this code path should only run on sm_90+. The current state creates undefined behavior on Volta-Ampere architectures.
🤖 Prompt for AI Agents
In cpp/tensorrt_llm/kernels/communicationKernels/mnnvlTwoShotAllreduceKernels.cu
around lines 120-121 (and related grid sync calls at ~155, ~218, ~460): the PTX
lamport flag update was extended to sm_70+, but the grid-dependency APIs
(cudaGridDependencySynchronize / cudaTriggerProgrammaticLaunchCompletion) are
still only enabled for sm_90+, leaving sm_70–sm_89 builds without grid-level
synchronization. Fix by either (A) extending the grid-dependency support to
sm_70+ to match the Lamport change (update the preprocessor/runtime arch checks
so the cudaGridDependencySynchronize and cudaTriggerProgrammaticLaunchCompletion
calls are enabled for >=700, or add a robust runtime arch check and enable them
when supported), or (B) explicitly restrict this whole codepath to sm_90+ and
add a clear compile-time/static-runtime guard plus a comment and error message
explaining that Volta/ampere (sm_70–sm_89) are unsupported; if choosing (A)
provide a safe fallback for devices lacking those APIs (implement equivalent
programmatic-launch completion using the existing Lamport flags/global
atomics/spin-wait protocol) and ensure all three sync points (start at ~155, end
at ~218 and ~460) are consistently handled for the same architectures.
|
PR_Github #22254 [ run ] triggered by Bot. Commit: |
|
PR_Github #22254 [ run ] completed with state |
|
/bot run --add-multi-gpu-test |
|
PR_Github #22312 [ run ] triggered by Bot. Commit: |
|
PR_Github #22312 [ run ] completed with state |
2 participants
Summary by CodeRabbit
Bug Fixes
Tests
Description
This PR fixed the accuracy issue in MNNVL Allreduce kernel on Hopper caused by wrong memory semantic used in the Hopper branch.
Test Coverage
tests/unittest/_torch/multi_gpu/test_mnnvl_allreduce.py
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.