Hi:
I want to build pytorch libs with _GLIBCXX_USE_CXX11_ABI=1 in branch v1.1.0. And I reivise TorchConfig.cmake.in file. But it seems not to work. Is there some suggents?
thanks
Hi:
I want to build pytorch libs with _GLIBCXX_USE_CXX11_ABI=1 in branch v1.1.0. And I reivise TorchConfig.cmake.in file. But it seems not to work. Is there some suggents?
thanks
You can add set(GLIBCXX_USE_CXX11_ABI 1)
in the beginning of pytorch/CMakeLists.txt
.
We now have new CXX11 ABI binaries for libtorch. They can be found on
http://pytorch.org, or at https://github.com/pytorch/pytorch/issues/17492#issuecomment-524692441.
Seems like, the pip packages on linux use the GLIBCXX_USE_CXX11_ABI=0
some say the conda ones use GLIBCXX_USE_CXX11_ABI=1 . Under windows its always
GLIBCXX_USE_CXX11_ABI=0` for cpu and cuda builds it seems. ( I tested the 1.6 version using pip )
This is important for people like me that have a python extension which uses libtorch .
This is still an issue in 1.6cpu! that I tested
Update, it seems the conda package on Linux is also built using GLIBCXX_USE_CXX11_ABI=0
so does the package on windows:
In [2]: torch._C._GLIBCXX_USE_CXX11_ABI
Out[2]: False
so basically all Pytorch packages are shipped with GLIBCXX_USE_CXX11_ABI=0
but why?