nvcc fatal : Unknown option '-mavx'

[shadow150519]

I am configuring the project on wsl2. when I load the CMakeList.txt in the fbgemm_gpu folder, It gives the following error.

Cannot get compiler information:
	Compiler exited with error code 1: \usr\local\cuda\bin\nvcc -Dfbgemm_gpu_py_EXPORTS --generate-code=arch=compute_75,code=[compute_75,sm_75] -Xcompiler=-fPIC -std=c++17 -mavx -mf16c -mfma -mavx2 -fopenmp --dryrun /mnt/c/Users/Lenovo/AppData/Local/Temp/compiler-file16156068507273487186.cu
	nvcc fatal   : Unknown option '-mavx'

[q10]

Could you show the full logs, and let me know the nvcc, gcc, and cmake versions you are using? We don't have official support for building FBGEMM_GPU on WSL2, but if WSL2 works exactly the same as a regular Linux environment, then it may be just an issue with the nvcc and/or gcc versions.

[shadow150519]

Here is the full logs. nvcc version: Cuda compilation tools, release 12.3, V12.3.52 gcc version: gcc (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0 cmake version: cmake version 3.25.3 OS: wsl2 Ubuntu20.4 on Windows10 wsl kernel version: 5.15.133.1-1 wsl version: 2.0.9.0

C:\Windows\system32\wsl.exe --distribution Ubuntu-20.04 --exec /bin/bash -c "export CMAKE_COLOR_DIAGNOSTICS=ON && export CLION_IDE=TRUE && export JETBRAINS_IDE=TRUE && cd /home/wtx/workspace/cpp_project/FBGEMM/fbgemm_gpu/cmake-build-fbgemm-gpu-debug && /usr/bin/cmake -DCMAKE_BUILD_TYPE=Debug -DCMAKE_C_COMPILER=/usr/bin/gcc '-DCMAKE_CXX_COMPILER=/usr/bin/g++' '-DCMAKE_PREFIX_PATH=D:\workspace\cpp_project\libtorch' -G 'CodeBlocks - Unix Makefiles' -S /home/wtx/workspace/cpp_project/FBGEMM/fbgemm_gpu -B /home/wtx/workspace/cpp_project/FBGEMM/fbgemm_gpu/cmake-build-fbgemm-gpu-debug"
================================================================================
Building the CUDA variant of FBGEMM-GPU
================================================================================

================================================================================
Default C++ compiler flags
(values may be overridden by CMAKE_CXX_STANDARD and CXX_STANDARD):


================================================================================

-- The CXX compiler identification is GNU 9.4.0
-- The C compiler identification is GNU 9.4.0
-- The CUDA compiler identification is NVIDIA 12.3.52
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/g++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/gcc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CUDA compiler ABI info
-- Detecting CUDA compiler ABI info - done
-- Check for working CUDA compiler: /usr/local/cuda/bin/nvcc - skipped
-- Detecting CUDA compile features
-- Detecting CUDA compile features - done
-- Performing Test C_HAS_AVX_1
-- Performing Test C_HAS_AVX_1 - Failed
-- Performing Test C_HAS_AVX_2
-- Performing Test C_HAS_AVX_2 - Success
-- Performing Test C_HAS_AVX2_1
-- Performing Test C_HAS_AVX2_1 - Failed
-- Performing Test C_HAS_AVX2_2
-- Performing Test C_HAS_AVX2_2 - Success
-- Performing Test C_HAS_AVX512_1
-- Performing Test C_HAS_AVX512_1 - Failed
-- Performing Test C_HAS_AVX512_2
-- Performing Test C_HAS_AVX512_2 - Failed
-- Performing Test C_HAS_AVX512_3
-- Performing Test C_HAS_AVX512_3 - Failed
-- Performing Test CXX_HAS_AVX_1
-- Performing Test CXX_HAS_AVX_1 - Failed
-- Performing Test CXX_HAS_AVX_2
-- Performing Test CXX_HAS_AVX_2 - Success
-- Performing Test CXX_HAS_AVX2_1
-- Performing Test CXX_HAS_AVX2_1 - Failed
-- Performing Test CXX_HAS_AVX2_2
-- Performing Test CXX_HAS_AVX2_2 - Success
-- Performing Test CXX_HAS_AVX512_1
-- Performing Test CXX_HAS_AVX512_1 - Failed
-- Performing Test CXX_HAS_AVX512_2
-- Performing Test CXX_HAS_AVX512_2 - Failed
-- Performing Test CXX_HAS_AVX512_3
-- Performing Test CXX_HAS_AVX512_3 - Failed
-- Skipping merge_pooled_embeddings sources
-- Configuring done
-- Generating done
-- Build files have been written to: /home/wtx/workspace/cpp_project/FBGEMM/fbgemm_gpu/cmake-build-fbgemm-gpu-debug

Cannot get compiler information:
	Compiler exited with error code 1: \usr\local\cuda\bin\nvcc -Dfbgemm_gpu_py_EXPORTS -g --generate-code=arch=compute_75,code=[compute_75,sm_75] -Xcompiler=-fPIC -std=c++17 -mavx -mf16c -mfma -mavx2 -fopenmp --dryrun /mnt/c/Users/Lenovo/AppData/Local/Temp/compiler-file11626473163701549697.cu
	nvcc fatal   : Unknown option '-mavx'

[q10]

Hmm, some of the tool versions you listed are not the recommended versions - in particular, CUDA should be at 12.1 and GCC should be at 10+.

I did quick Google search cases where users have run into a similar error signature before, and the solution was generally to upgrade the drivers and toolkit. That being said, we highly recommend using building and using FBGEMM in a full Linux environment instead, as WSL may contain intricacies that are beyond our scope for debugging issues like this.

[q10]

@shadow150519 could you try latest main, and run the usual FBGEMM_GPU build command, but with -DWSL_MODE=1 added to the args?

python setup.py  ...<usual args>...   -DWSL_MODE=1

(See here for build instructions in detail)

Let us know how this goes.

[shadow150519]

sorry for the late response, I've moved to a linux machine right now and it works correctly. I'll try this and give you the result as soon as I can. Thank for your help :)

@shadow150519 could you try latest main, and run the usual FBGEMM_GPU build command, but with -DWSL_MODE=1 added to the args?

python setup.py  ...<usual args>...   -DWSL_MODE=1

(See here for build instructions in detail)

Let us know how this goes.