🐛 Bug

When trying to build pytorch from sources with a c++ toolchain nvcc10.2 + gcc8 + libc++ I'm getting a compiler error

bazel-out/k8-fastbuild-gcc/bin/external/pytorch/aten/src/ATen/native/cpu/MultinomialKernel.cpp.DEFAULT.cpp:52:7:   required from 'void at::native::{anonymous}::multinomial_apply(at::Tensor&, const at::Tensor&, int64_t, bool, c10::optional<at::Generator>) [with scalar_t = c10::Half; int64_t = long int]'
bazel-out/k8-fastbuild-gcc/bin/external/pytorch/aten/src/ATen/native/cpu/MultinomialKernel.cpp.DEFAULT.cpp:134:3:   required from here
/opt/sysroot/opt/libcxx/include/c++/v1/math.h:449:1: error: no type named 'type' in 'struct std::__1::enable_if<false, bool>'

The code that produces the problem is here

To Reproduce

Steps to reproduce the behavior:

Sorry, don't have a good repro with upstream pytorch, since our build uses a custom toolchain and the setup is quite elaborated.

Expected behavior

There should not be a compile error.

Environment

Please copy and paste the output from our
environment collection script
(or fill out the checklist below manually).

You can get the script and run it with:

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

• PyTorch Version (e.g., 1.0):
• OS (e.g., Linux):
• How you installed PyTorch (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information:

Additional context

The problem is that performed checks assumes that libc++ is used with clang only, while gcc+libc++ could be used as well.

I'm going to send a PR with a proposed fix shortly.

cc @malfet @seemethere @walterddr @ngimel