[https://nvbugs/5433581][fix] DeepGEMM installation on SBSA #6588
Conversation
📝 WalkthroughWalkthroughThis change integrates the DeepGEMM module as a new third-party submodule and build target. It updates build scripts, CMake configurations, and packaging logic to support DeepGEMM, including copying source files, adapting imports, and handling Python bindings. Import statements across relevant Python modules and tests are updated to reflect the new package structure. Changes
Sequence Diagram(s)sequenceDiagram
participant Developer
participant BuildScript
participant CMake
participant DeepGEMM Submodule
participant PythonPackage
Developer->>BuildScript: Run build_wheel.py
BuildScript->>CMake: Configure with -DBUILD_DEEP_GEMM=ON
CMake->>DeepGEMM Submodule: Build deep_gemm target
DeepGEMM Submodule->>CMake: Copy/adapt Python & C++ sources, build pybind11 module
CMake->>BuildScript: Build outputs (shared object, Python files)
BuildScript->>PythonPackage: Install deep_gemm_cpp_tllm, copy bindings
BuildScript->>PythonPackage: Generate stubs for deep_gemm_cpp_tllm
Estimated code review effort🎯 4 (Complex) | ⏱️ ~35 minutes Suggested reviewers
Note ⚡️ Unit Test Generation is now available in beta!Learn more here, or try it out under "Finishing Touches" below. 📜 Recent review detailsConfiguration used: .coderabbit.yaml 📒 Files selected for processing (3)
✅ Files skipped from review due to trivial changes (1)
🚧 Files skipped from review as they are similar to previous changes (2)
⏰ 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)
✨ Finishing Touches🧪 Generate unit tests
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. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
Documentation and Community
|
![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: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
requirements.txt(1 hunks)tests/unittest/test_pip_install.py(0 hunks)
💤 Files with no reviewable changes (1)
- tests/unittest/test_pip_install.py
zongfeijing
force-pushed
the
user/zongfeij/fix
branch
from
40760cf to
7a00ae7
Compare
Signed-off-by: Zongfei Jing <20381269+zongfeijing@users.noreply.github.com>
zongfeijing
force-pushed
the
user/zongfeij/fix
branch
from
7a00ae7 to
78830cb
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #14107 [ run ] triggered by Bot |
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (2)
scripts/build_wheel.py (1)
495-495: Fix line length violation.The CMake configure command line exceeds the 120-character limit flagged by static analysis.
Split the long line for better readability:
- cmake_configure_command = ( - f'cmake -DCMAKE_BUILD_TYPE="{build_type}" -DBUILD_PYT="{build_pyt}" -DBINDING_TYPE="{binding_type}" -DBUILD_DEEP_EP="{build_deep_ep}" -DBUILD_DEEP_GEMM="{build_deep_gemm}"' + cmake_configure_command = ( + f'cmake -DCMAKE_BUILD_TYPE="{build_type}" -DBUILD_PYT="{build_pyt}" ' + f'-DBINDING_TYPE="{binding_type}" -DBUILD_DEEP_EP="{build_deep_ep}" ' + f'-DBUILD_DEEP_GEMM="{build_deep_gemm}"'cpp/tensorrt_llm/deep_gemm/CMakeLists.txt (1)
134-135: Consider adding platform-specific optimization flags.The compile options include
-Wno-psabiwhich is GCC-specific. Consider making this conditional for better cross-compiler compatibility.-target_compile_options(deep_gemm_cpp_tllm PRIVATE ${TORCH_CXX_FLAGS} -std=c++20 - -O3 -fPIC -Wno-psabi) +target_compile_options(deep_gemm_cpp_tllm PRIVATE ${TORCH_CXX_FLAGS} -std=c++20 -O3 -fPIC) +if(CMAKE_CXX_COMPILER_ID STREQUAL "GNU") + target_compile_options(deep_gemm_cpp_tllm PRIVATE -Wno-psabi) +endif()
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (13)
.gitignore(1 hunks).gitmodules(1 hunks)3rdparty/deepgemm(1 hunks)cpp/CMakeLists.txt(3 hunks)cpp/tensorrt_llm/CMakeLists.txt(1 hunks)cpp/tensorrt_llm/deep_gemm/CMakeLists.txt(1 hunks)cpp/tensorrt_llm/deep_gemm/deep_gemm_cpp_tllm.version(1 hunks)requirements.txt(0 hunks)scripts/build_wheel.py(5 hunks)setup.py(1 hunks)tensorrt_llm/_torch/modules/fused_moe/fused_moe_deepgemm.py(1 hunks)tensorrt_llm/_torch/modules/linear.py(1 hunks)tests/unittest/_torch/thop/test_fp8_block_scale_gemm.py(1 hunks)
💤 Files with no reviewable changes (1)
- requirements.txt
✅ Files skipped from review due to trivial changes (7)
- .gitmodules
- setup.py
- 3rdparty/deepgemm
- cpp/tensorrt_llm/CMakeLists.txt
- .gitignore
- cpp/tensorrt_llm/deep_gemm/deep_gemm_cpp_tllm.version
- tests/unittest/_torch/thop/test_fp8_block_scale_gemm.py
🧰 Additional context used
🧠 Learnings (1)
📚 Learning: in tensorrt-llm, examples directory can have different dependency versions than the root requirement...
Learnt from: yibinl-nvidia
PR: NVIDIA/TensorRT-LLM#6506
File: examples/models/core/mixtral/requirements.txt:3-3
Timestamp: 2025-08-01T15:14:45.673Z
Learning: In TensorRT-LLM, examples directory can have different dependency versions than the root requirements.txt file. Version conflicts between root and examples dependencies are acceptable because examples are designed to be standalone and self-contained.
Applied to files:
cpp/CMakeLists.txt
🪛 Ruff (0.12.2)
scripts/build_wheel.py
495-495: Line too long (188 > 120)
(E501)
🔇 Additional comments (24)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_deepgemm.py (1)
9-9: LGTM! Import path updated for DeepGEMM submodule integration.The change from direct import to relative import from the
tensorrt_llmpackage is correct and aligns with the broader effort to integrate DeepGEMM as a submodule rather than an external dependency.cpp/CMakeLists.txt (3)
34-34: LGTM! DeepGEMM build option added.The new
BUILD_DEEPGEMMoption follows the same pattern as other build options and is appropriately enabled by default.
203-208: LGTM! Conditional logic properly updated for DeepGEMM.The pybind11 subdirectory inclusion logic is correctly extended to include the new
BUILD_DEEPGEMMoption alongside existing conditions, ensuring pybind11 is available when DeepGEMM is being built.
224-228: LGTM! Include directories logic properly updated.The pybind11 include directories are correctly made available when
BUILD_DEEPGEMMis enabled, maintaining consistency with the subdirectory inclusion logic above.tensorrt_llm/_torch/modules/linear.py (1)
576-576: LGTM! Import path updated for DeepGEMM submodule integration.The change from direct import to relative import from the
tensorrt_llmpackage is consistent with the DeepGEMM submodule integration. The conditional import for SM version 100 remains properly structured.scripts/build_wheel.py (6)
451-451: LGTM: Consistent variable initialization.The
build_deep_gemm = "OFF"initialization follows the same pattern asbuild_deep_epfor cpp_only builds.
453-453: LGTM: Proper target extension.Adding
"deep_gemm"to the targets list alongside"deep_ep"maintains consistency with the existing pattern.
456-456: LGTM: Consistent build flag setting.The
build_deep_gemm = "ON"assignment follows the established pattern for non-cpp_only builds.
642-648: LGTM: Proper deep_gemm directory cleanup.The cleanup logic correctly mirrors the deep_ep pattern, handling both symlinks and directories appropriately.
698-708: LGTM: Consistent deep_gemm installation logic.The installation steps properly mirror the deep_ep pattern, including:
- CUDA architecture file detection
- Conditional pybind11 library installation
- Python bindings directory installation
782-785: LGTM: Consistent stub generation for deep_gemm.The pybind11 stub generation logic correctly follows the same conditional pattern as deep_ep, ensuring stubs are only generated when CUDA architectures are available.
cpp/tensorrt_llm/deep_gemm/CMakeLists.txt (13)
1-1: LGTM: Proper custom target definition.Creating the
deep_gemmcustom target follows CMake best practices.
7-21: LGTM: Robust CUDA architecture filtering.The regex-based CUDA architecture parsing correctly filters for major version >= 9, with proper error handling for invalid formats.
24-26: LGTM: Proper Windows exclusion.Explicitly clearing CUDA architectures on Windows prevents build attempts on unsupported platforms.
32-34: LGTM: Early return for unsupported architectures.The early return when no suitable CUDA architectures are available prevents unnecessary build steps.
43-48: LGTM: Comprehensive submodule validation.The check for DeepGEMM submodule existence with clear error message and remediation instructions is well-implemented.
51-56: LGTM: Dependency validation.Checking for cutlass include directory ensures submodule dependencies are properly initialized.
74-89: LGTM: Smart Python file adaptation.The logic to modify import statements in Python files (
deep_gemm_cpp→tensorrt_llm.deep_gemm_cpp_tllm) and add adaptation headers is well-designed for integration.
97-101: LGTM: Proper dependency tracking.Adding source files to
CMAKE_CONFIGURE_DEPENDSensures CMake reconfigures when source files change.
114-114: LGTM: Appropriate compilation setting.Disabling interprocedural optimization for CUDA separable compilation is correct for this use case.
123-132: LGTM: Proper pybind11 module configuration.The module configuration with C++20 standard, CUDA architectures, and RPATH settings follows best practices.
149-150: LGTM: Appropriate library linking.Linking against TORCH_LIBRARIES, TORCH_PYTHON_LIB, cuda, and cudart provides the necessary dependencies.
153-155: LGTM: Proper CUDA library directories.Including both lib64 and lib64/stubs directories ensures proper CUDA library resolution.
159-159: LGTM: Correct dependency declaration.Making the deep_gemm target depend on deep_gemm_cpp_tllm ensures proper build ordering.
zongfeijing
force-pushed
the
user/zongfeij/fix
branch
from
78830cb to
109adb6
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #14119 [ run ] triggered by Bot |
|
PR_Github #14107 [ run ] completed with state |
There was a problem hiding this comment.
Actionable comments posted: 0
🔭 Outside diff range comments (1)
scripts/build_wheel.py (1)
774-780: Make deep_gemm stub generation conditional like deep_ep.The deep_gemm stub generation is unconditional while deep_ep checks for CUDA architectures first. This could fail if the deep_gemm module wasn't built.
if deep_ep_cuda_architectures: build_run( f"\"{venv_python}\" -m pybind11_stubgen -o . deep_ep_cpp_tllm --exit-code", env=env_ld) - build_run( - f"\"{venv_python}\" -m pybind11_stubgen -o . deep_gemm_cpp_tllm --exit-code", - env=env_ld) + # Check if deep_gemm was built before generating stubs + if (pkg_dir / "deep_gemm_cpp_tllm.so").exists() or list(pkg_dir.glob("deep_gemm_cpp_tllm.*.so")): + build_run( + f"\"{venv_python}\" -m pybind11_stubgen -o . deep_gemm_cpp_tllm --exit-code", + env=env_ld)
🧹 Nitpick comments (2)
scripts/build_wheel.py (2)
495-495: Split the long CMake command line for better readability.The CMake configuration is correct, but the line exceeds the 120-character limit.
Consider splitting the command across multiple lines:
- f'cmake -DCMAKE_BUILD_TYPE="{build_type}" -DBUILD_PYT="{build_pyt}" -DBINDING_TYPE="{binding_type}" -DBUILD_DEEP_EP="{build_deep_ep}" -DBUILD_DEEP_GEMM="{build_deep_gemm}"' + f'cmake -DCMAKE_BUILD_TYPE="{build_type}" -DBUILD_PYT="{build_pyt}" ' + f'-DBINDING_TYPE="{binding_type}" -DBUILD_DEEP_EP="{build_deep_ep}" ' + f'-DBUILD_DEEP_GEMM="{build_deep_gemm}"'
2-2: Update copyright year to include 2025.The copyright header should be updated to reflect the current year.
-# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-FileCopyrightText: Copyright (c) 2022-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (5)
.gitmodules(1 hunks)3rdparty/DeepGEMM(1 hunks)cpp/CMakeLists.txt(3 hunks)cpp/tensorrt_llm/deep_gemm/CMakeLists.txt(1 hunks)scripts/build_wheel.py(5 hunks)
✅ Files skipped from review due to trivial changes (2)
- 3rdparty/DeepGEMM
- .gitmodules
🚧 Files skipped from review as they are similar to previous changes (2)
- cpp/CMakeLists.txt
- cpp/tensorrt_llm/deep_gemm/CMakeLists.txt
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
**/*.py: The code developed for TensorRT-LLM should conform to Python 3.8+.
Indent Python code with 4 spaces. Do not use tabs.
Always maintain the namespace when importing in Python, even if only one class or function from a module is used.
Python filenames should use snake_case (e.g., some_file.py).
Python classes should use PascalCase (e.g., class SomeClass).
Python functions and methods should use snake_case (e.g., def my_awesome_function():).
Python local variables should use snake_case. Prefix k for variable names that start with a number (e.g., k_99th_percentile).
Python global variables should use upper snake_case and prefix G (e.g., G_MY_GLOBAL).
Python constants should use upper snake_case (e.g., MY_CONSTANT).
Avoid shadowing variables declared in an outer scope in Python.
Initialize all externally visible members of a Python class in the constructor.
For interfaces that may be used outside a file, prefer docstrings over comments in Python.
Comments in Python should be reserved for code within a function, or interfaces that are local to a file.
Use Google style docstrings for Python classes and functions, which can be parsed by Sphinx.
Attributes and variables in Python can be documented inline; attribute docstrings will be rendered under the docstring for the class.
Avoid using reflection in Python when functionality can be easily achieved without reflection.
When using try-except blocks in Python, limit the except to the smallest set of errors possible.
When using try-except blocks to handle multiple possible variable types in Python, keep the body of the try as small as possible, using the else block to implement the logic.
Files:
scripts/build_wheel.py
**/*.{cpp,h,cu,py}
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the current year. This includes .cpp, .h, .cu, .py, and any other source files which are compiled or interpreted.
Files:
scripts/build_wheel.py
🪛 Ruff (0.12.2)
scripts/build_wheel.py
495-495: Line too long (188 > 120)
(E501)
⏰ 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 (3)
scripts/build_wheel.py (3)
451-456: LGTM! Consistent integration of deep_gemm build flag.The
build_deep_gemmflag is properly integrated following the same pattern asbuild_deep_ep, correctly toggling based on thecpp_onlysetting.
642-648: LGTM! Proper cleanup handling for deep_gemm directory.The deep_gemm directory cleanup follows the same robust pattern as deep_ep, correctly handling both symlinks and regular directories.
697-703: Review deep_gemm installation CUDA guardWe’ve noticed a discrepancy between how deep_ep and deep_gemm are handled in scripts/build_wheel.py:
- deep_ep (lines 678–684) is only installed when
deep_ep_cuda_architecturesis non-empty.- deep_gemm (lines 697–703) is installed unconditionally—and there is no
cuda_architectures.txtfor deep_gemm in the CMake setup.Please clarify the intended behavior:
• If deep_gemm should only ship when GPU architectures are specified, introduce a
deep_gemm_cuda_architecturesvariable (by readingbuild_dir/tensorrt_llm/deep_gemm/cuda_architectures.txt) and wrap bothinstall_file(get_binding_lib("deep_gemm", "deep_gemm_cpp_tllm"), pkg_dir) install_tree(... "tensorrt_llm/deep_gemm/python/deep_gemm", deep_gemm_dir, dirs_exist_ok=True)in
if deep_gemm_cuda_architectures:.• Otherwise, please add a comment explaining why deep_gemm is always installed (e.g., CPU fallback or unified binary).
Location to review:
- scripts/build_wheel.py lines 697–703
|
/bot run --disable-fail-fast |
|
PR_Github #14156 [ run ] triggered by Bot |
|
PR_Github #14119 [ run ] completed with state |
|
PR_Github #14156 [ run ] completed with state |
There was a problem hiding this comment.
Actionable comments posted: 4
🧹 Nitpick comments (3)
cpp/tensorrt_llm/deep_gemm/CMakeLists.txt (3)
90-102: Duplicate / conflicting C++ standard flags
CXX_STANDARD 17is already enforced via target properties. Injecting-std=c++17again intarget_compile_optionsis redundant and risks conflicting flags if a toolchain wrapper adds its own-std=option.Simply drop the manual
-std=c++17flag:-target_compile_options(deep_gemm_cpp_tllm PRIVATE ${TORCH_CXX_FLAGS} -std=c++17 - -O3 -fPIC -Wno-psabi) +target_compile_options(deep_gemm_cpp_tllm PRIVATE ${TORCH_CXX_FLAGS} -O3 -fPIC -Wno-psabi)
95-96:CUDA_SEPARABLE_COMPILATIONhas no effect
The target contains only.cppsources; enablingCUDA_SEPARABLE_COMPILATIONis a no-op and can trigger CMake warnings when CUDA language isn’t enabled. Remove it or guard it withif (TARGET CUDA::cuda_driver).
119-122:target_link_directoriesunnecessary once imported CUDA targets are used
If the previous suggestion is applied, manualtarget_link_directoriescan be dropped—CUDA::cudartalready provides the correct paths.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
.gitmodules(1 hunks)3rdparty/DeepGEMM(1 hunks)cpp/tensorrt_llm/deep_gemm/CMakeLists.txt(1 hunks)
✅ Files skipped from review due to trivial changes (1)
- .gitmodules
🚧 Files skipped from review as they are similar to previous changes (1)
- 3rdparty/DeepGEMM
⏰ 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
zongfeijing
force-pushed
the
user/zongfeij/fix
branch
from
f2f30d7 to
54e65c1
Compare
|
/bot run --disable-fail-fast |
Signed-off-by: Zongfei Jing <20381269+zongfeijing@users.noreply.github.com>
zongfeijing
force-pushed
the
user/zongfeij/fix
branch
from
54e65c1 to
f96f760
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #14192 [ run ] triggered by Bot |
|
PR_Github #14193 [ run ] triggered by Bot |
|
PR_Github #14192 [ run ] completed with state |
|
PR_Github #14193 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #14218 [ run ] triggered by Bot |
|
PR_Github #14218 [ run ] completed with state |
5 participants
Summary by CodeRabbit
New Features
Chores
Refactor
Tests
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.
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.