Scipy build fail on MacOS because LLVM producer != reader

user_1830334 0
Viewed 692

I'm trying to build Scipy from source. However, a linking step...

/usr/bin/clang -bundle -undefined dynamic_lookup -L/Users/gwg/miniconda3/envs/scipy-dev/lib -arch x86_64 -L/Users/gwg/miniconda3/envs/scipy-dev/lib -arch x86_64 -L/usr/local/opt/libomp/lib -lomp -I/usr/local/opt/libomp/include -Xpreprocessor -fopenmp -arch x86_64 build/temp.macosx-10.9-x86_64-3.7/scipy/spatial/src/distance_wrap.o -L/Users/gwg/miniconda3/envs/scipy-dev/lib/python3.7/site-packages/numpy/core/lib -Lbuild/temp.macosx-10.9-x86_64-3.7 -lnpymath -o build/lib.macosx-10.9-x86_64-3.7/scipy/spatial/_distance_wrap.cpython-37m-darwin.so

...fails, and I get the following error:

ld: in /Users/gwg/miniconda3/envs/scipy-dev/lib/python3.7/site-packages/numpy/core/lib/libnpymath.a(npy_math.o), could not parse object file /Users/gwg/miniconda3/envs/scipy-dev/lib/python3.7/site-packages/numpy/core/lib/libnpymath.a(npy_math.o): 'Unknown attribute kind (61) (Producer: 'LLVM10.0.0' Reader: 'LLVM APPLE_1_1000.10.44.4_0')', using libLTO version 'LLVM version 10.0.0, (clang-1000.10.44.4)' for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)

I suspect it is due to the Producer: 'LLVM10.0.0' Reader: 'LLVM APPLE_1_1000.10.44.4_0' mismatch. But I don't know how to investigate this error.

Some details:

  • MacOS: High Sierra, 10.13.6

  • gcc --version:

    Configured with: --prefix=/Library/Developer/CommandLineTools/usr --with-gxx-include-dir=/usr/include/c++/4.2.1
Apple LLVM version 10.0.0 (clang-1000.10.44.4)
Target: x86_64-apple-darwin17.7.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin

  • clang --version:

    Apple LLVM version 10.0.0 (clang-1000.10.44.4)
Target: x86_64-apple-darwin17.7.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin
scipy clang llvm
2 Answers

It seems you're using different clang (judging from the paths like /usr/local/opt/libomp/lib). Likely the one provided by conda. In order to perform build with LTO you need to use the compatible toolchain (e.g. build everything with Apple-provided toolchain or with conda-provided one).

user_1649250

I had the same issue. The suggestions here to use the conda compilers resolved the issue for me:

conda create -n scipy-dev
conda activate scipy-dev
conda config --env --add channels conda-forge
conda install numpy cython pybind11 pytest compilers openblas
python setup.py develop
user_93349620
Related