📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

**/*.{cpp,cc,cxx,cu,h,hpp,hh,hxx,cuh}: In C++ and CUDA files, closing braces of namespaces must include a trailing comment naming the namespace (e.g., } // namespace foo)
Prefer const or constexpr variables over #define for constants; variables not modified after initialization must be declared const
Avoid magic literals: except 0, nullptr, true, false, use named constants (e.g., constexpr) instead of inline numeric or string literals
Use Allman brace style; always brace bodies of if/else/switch/while/do/for; put the semicolon of empty loops on a new line
C++ filenames should be camelCase starting lowercase (e.g., thisIsAFilename.cpp) and case-insensitive unique within a build target
Type names are UpperCamelCase; local variables, methods, and namespaces are lowerCamelCase
Global non-magic-number variables: prefix g for non-static globals and s for static or anonymous-namespace globals (e.g., gFoo, sBar)
Locally visible static variables should be lowerCamelCase starting with 's' (e.g., static std::once_flag sFlag)
Member variables use mPrefix (e.g., mCount); public members may omit but using m is encouraged for clarity
Constants (enums, globals, static constants, and function-scope magic-number constants) use uppercase SNAKE_CASE with k prefix (e.g., kMAX_SIZE)
Avoid macros; if necessary, use UPPER_SNAKE_CASE for macro names
Run clang-format (LLVM style) before submitting; maximum line length is 120; use clang-format off/on only for justified exceptions
Use C++ comments (//); C-style /* / only for special inline cases; prefer Doxygen comments: //! and //!<; full-sentence comments are capitalized and punctuated; document public APIs with Doxygen
Disable code with #if/#endif (possibly via a DEBUG macro); do not comment out code; avoid dead code blocks
Do not throw exceptions across library boundaries
Use the least-forceful cast; avoid C-style and functional casts (except void casts); do not remove const/volatile; void to T* via static_cast; reinterpret_cast only a...

Files:

• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp
• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

**/*.{cpp,cc,cxx,cu,h,hpp,hh,hxx,cuh,py}: Use spaces only; no tabs; indent with 4 spaces
Prepend NVIDIA copyright header (current year) to all source files (.cpp, .h, .cu, .py, etc.)

Files:

• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp
• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

**/*.{cpp,cc,cxx,cu}: Prefer smart pointers; use std::unique_ptr for sole ownership, std::shared_ptr for shared ownership; avoid deprecated smart pointers
Do not use assignment in subexpressions (e.g., if (x = y)); avoid chained assignments (x = y = z)
Switch statements: provide cases for all enum values and omit default to catch new values; prohibit fall-through except between empty labels; terminate each case with break or throw; do not end a case with return; place break inside braces when using a compound statement

Files:

• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp

📄 CodeRabbit inference engine (CODING_GUIDELINES.md)

Header include guards are required with macro name TRTLLM_<FILE_NAME_IN_CAPS> (no directories, no leading/trailing underscores)

Files:

• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/BatchedGemmInterface.h

Learnt from: CR
PR: NVIDIA/TensorRT-LLM#0
File: CODING_GUIDELINES.md:0-0
Timestamp: 2025-08-29T08:49:23.221Z
Learning: Applies to **/*.cuh : CUDA headers (.cuh) should also follow the project include-guard convention TRTLLM_<FILE_NAME_IN_CAPS>

Applied to files:

• cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/KernelRunner.cpp

cpp/include/tensorrt_llm/common/cudaUtils.h (1)

• getMultiProcessorCount (393-400)

cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/KernelParams.h (5)

• std (120-565)
• std (189-558)
• std (199-204)
• std (240-259)
• std (251-256)

cpp/tensorrt_llm/kernels/trtllmGenKernels/batchedGemm/trtllmGen_bmm_export/GemmOptions.h (1)

• divUp (499-502)