Build pytorch from source error

when building pytorch according to,
I always get an error

xxx/lib/libmagma.a(error.cpp.o): unrecognized relocation (0x2a) in section .text

What’s the reason?
My building steps:

conda install numpy ninja pyyaml mkl mkl-include setuptools cmake cffi
conda install -c pytorch magma-cuda100 #for cuda10.0
git clone --recursive
cd pytorch
# if you are updating an existing checkout
git submodule sync 
git submodule update --init --recursive
# my source code version is f9010d764894f3ba9bc6a129213a0faf42af9b65
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
python build

solved by upgrading binutils from 2.23 to 2.27