Fix clang vla warning

[cyyever]

Fix the error

/var/lib/jenkins/workspace/third_party/fbgemm/src/FbgemmI8Spmdm.cc:188:41: error: variable length arrays in C++ are a Clang extension [-Werror,-Wvla-cxx-extension]

[netlify]

✅ Deploy Preview for pytorch-fbgemm-docs ready!

Name	Link
🔨 Latest commit	f9846c5
🔍 Latest deploy log	https://app.netlify.com/sites/pytorch-fbgemm-docs/deploys/67c9532d9db59a000886d02d
😎 Deploy Preview	https://deploy-preview-2736--pytorch-fbgemm-docs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[jamesETsmith]

Thanks for working on this @cyyever. I'm still seeing the issue even when manually adding add the flags this PR implements (using clang/clang++ 18.1.8). Do you know if the fixs works for clang 18?

[cyyever]

@jamesETsmith Can you help check compile_commands.json and make sure that the flag is indeed added to the problematic files' command lines?

[jamesETsmith]

@cyyever great idea, it looks like the flag is there (and in stdout during compilation):

From compile_commands.json:

{
  "directory": "/home/james.smith/apps/pytorch/build_qcog",
  "command": "/usr/bin/clang++ -DFBGEMM_STATIC -I/home/james.smith/apps/pytorch/third_party/cpuinfo/include -I/home/james.smith/apps/pytorch/third_party/fbgemm/third_party/asmjit/src -I/home/james.smith/apps/pytorch/third_party/fbgemm/include -I/home/james.smith/apps/pytorch/third_party/fbgemm -isystem /home/james.smith/apps/pytorch/third_party/protobuf/src -isystem /opt/intel/oneapi/mkl/latest/include -isystem /home/james.smith/apps/pytorch/third_party/XNNPACK/include -Wno-vla-cxx-extension -D_GLIBCXX_USE_CXX11_ABI=1 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp=libomp -Wall -Wextra -Werror -Wno-deprecated-declarations -Wimplicit-fallthrough -g -std=c++17 -fPIC -fvisibility=hidden -DMKL_HAS_SBGEMM -DMKL_HAS_SHGEMM -o third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/EmbeddingSpMDM.cc.o -c /home/james.smith/apps/pytorch/third_party/fbgemm/src/EmbeddingSpMDM.cc",
  "file": "/home/james.smith/apps/pytorch/third_party/fbgemm/src/EmbeddingSpMDM.cc",
  "output": "third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/EmbeddingSpMDM.cc.o"
}

From STDOUT:

[26/2892] Building CXX object third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/FbgemmI8Spmdm.cc.o
FAILED: third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/FbgemmI8Spmdm.cc.o 
/usr/bin/ccache /usr/bin/clang++ -DFBGEMM_STATIC -I/home/james.smith/apps/pytorch/third_party/cpuinfo/include -I/home/james.smith/apps/pytorch/third_party/fbgemm/third_party/asmjit/src -I/home/james.smith/apps/pytorch/third_party/fbgemm/include -I/home/james.smith/apps/pytorch/third_party/fbgemm -isystem /home/james.smith/apps/pytorch/third_party/protobuf/src -isystem /opt/intel/oneapi/mkl/latest/include -isystem /home/james.smith/apps/pytorch/third_party/XNNPACK/include -Wno-vla-cxx-extension -D_GLIBCXX_USE_CXX11_ABI=1 -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -fopenmp=libomp -Wall -Wextra -Werror -Wno-deprecated-declarations -Wimplicit-fallthrough -g -std=c++17 -fPIC -fvisibility=hidden -DMKL_HAS_SBGEMM -DMKL_HAS_SHGEMM -MD -MT third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/FbgemmI8Spmdm.cc.o -MF third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/FbgemmI8Spmdm.cc.o.d -o third_party/fbgemm/CMakeFiles/fbgemm_generic.dir/src/FbgemmI8Spmdm.cc.o -c /home/james.smith/apps/pytorch/third_party/fbgemm/src/FbgemmI8Spmdm.cc
/home/james.smith/apps/pytorch/third_party/fbgemm/src/FbgemmI8Spmdm.cc:86:32: error: variable length arrays in C++ are a Clang extension [-Werror,-Wvla-cxx-extension]
   86 |   alignas(64) uint8_t A_buffer[K * 32];

[cyyever]

@jamesETsmith That flag is overwritten by '-Wall -Wextra -Werror' later. That means there is a flag order problem. Check where you added the flag and move it behind Wall

[jamesETsmith]

Thanks for the heads up @cyyever. I'm building FBGEMM through PyTorch and it doesn't seem like I can change the ordering of the CMAKE_CXX_FLAGS from the command line without modifying the cmake source like you have for FBGEMM or PyTorch. I replicated the changes from your pytorch PR and that seems to have worked.

Thanks for the work on this, I hope it gets merged.

[facebook-github-bot]

@q10 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[cyyever]

@q10 Any process towards landing?

[q10]

Hi @cyyever sorry for the delay, there have been some internal issues preventing the PR from being imported into our internal systems, but we are working to import and land this PR.

[facebook-github-bot]

@q10 has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator.

[facebook-github-bot]

@q10 merged this pull request in a1c8ab3.