fatal error: ATen/autocast_mode.h: No such file or directory compilation terminated.

🐛 Describe the bug

error information:

Building wheel torchvision-0.13.0a0+7bb5e41
PNG found: True
libpng version: 1.6.37
Building torchvision with PNG image support
libpng include path: /home/chy/anaconda3/envs/maskrcnn_benchmark/include/libpng16
Running build on conda-build: False
Running build on conda: True
JPEG found: True
Building torchvision with JPEG image support
NVJPEG found: False
FFmpeg found: True
ffmpeg include path: ['/home/chy/anaconda3/envs/maskrcnn_benchmark/include', '/home/chy/anaconda3/envs/maskrcnn_benchmark/include/x86_ 64-linux-gnu']
ffmpeg library_dir: ['/home/chy/anaconda3/envs/maskrcnn_benchmark/lib', '/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/x86_64-linux- gnu']
video codec found: False
The installed version of ffmpeg is missing the header file 'bsf.h' which is required for GPU video decoding. Please install the latest ffmpeg from conda-forge channel: conda install -c conda-forge ffmpeg.
running install
running bdist_egg
running egg_info
writing torchvision.egg-info/PKG-INFO
writing dependency_links to torchvision.egg-info/dependency_links.txt
writing requirements to torchvision.egg-info/requires.txt
writing top-level names to torchvision.egg-info/top_level.txt
reading manifest file 'torchvision.egg-info/SOURCES.txt'
reading manifest template 'MANIFEST.in'
warning: no previously-included files matching 'pycache' found under directory ''
warning: no previously-included files matching '.py[co]' found under directory '*'
writing manifest file 'torchvision.egg-info/SOURCES.txt'
installing library code to build/bdist.linux-x86_64/egg
running install_lib
running build_py
copying torchvision/version.py -> build/lib.linux-x86_64-3.6/torchvision
running build_ext
building 'torchvision.C' extension
gcc -pthread -B /home/chy/anaconda3/envs/maskrcnn_benchmark/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wa ll -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/chy/github/vision/torchvision/csrc -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/ python3.6/site-packages/torch/include -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/python3.6/site-packages/torch/include/torch/cs rc/api/include -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/python3.6/site-packages/torch/include/TH -I/home/chy/anaconda3/envs/m askrcnn_benchmark/lib/python3.6/site-packages/torch/include/THC -I/usr/local/cuda-9.0.176/include -I/home/chy/anaconda3/envs/maskrcnn benchmark/include/python3.6m -c /home/chy/github/vision/torchvision/csrc/ops/autocast/deform_conv2d_kernel.cpp -o build/temp.linux-x86 64-3.6/home/chy/github/vision/torchvision/csrc/ops/autocast/deform_conv2d_kernel.o -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION NAME=_C -D_GLIBCXX_USE_CXX11_ABI=0 -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=image -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++11
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
:0:0: warning: "TORCH_EXTENSION_NAME" redefined
:0:0: note: this is the location of the previous definition
/home/chy/github/vision/torchvision/csrc/ops/autocast/deform_conv2d_kernel.cpp:3:32: fatal error: ATen/autocast_mode.h: No such file o r directory
compilation terminated.
error: command 'gcc' failed with exit status 1
(maskrcnn_benchmark) chy@guest-PR4908P:~/github/vision$ python setup.py build
Building wheel torchvision-0.13.0a0+7bb5e41
PNG found: True
libpng version: 1.6.37
Building torchvision with PNG image support
libpng include path: /home/chy/anaconda3/envs/maskrcnn_benchmark/include/libpng16
Running build on conda-build: False
Running build on conda: True
JPEG found: True
Building torchvision with JPEG image support
NVJPEG found: False
FFmpeg found: True
ffmpeg include path: ['/home/chy/anaconda3/envs/maskrcnn_benchmark/include', '/home/chy/anaconda3/envs/maskrcnn_benchmark/include/x86_64-linux-gnu']
ffmpeg library_dir: ['/home/chy/anaconda3/envs/maskrcnn_benchmark/lib', '/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/x86_64-linux-gnu']
video codec found: False
The installed version of ffmpeg is missing the header file 'bsf.h' which is required for GPU video decoding. Please install the latest ffmpeg from conda-forge channel: conda install -c conda-forge ffmpeg.
running build
running build_py
copying torchvision/version.py -> build/lib.linux-x86_64-3.6/torchvision
running build_ext
building 'torchvision._C' extension
gcc -pthread -B /home/chy/anaconda3/envs/maskrcnn_benchmark/compiler_compat -Wl,--sysroot=/ -Wsign-compare -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes -fPIC -DWITH_CUDA -I/home/chy/github/vision/torchvision/csrc -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/python3.6/site-packages/torch/include -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/python3.6/site-packages/torch/include/torch/csrc/api/include -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/python3.6/site-packages/torch/include/TH -I/home/chy/anaconda3/envs/maskrcnn_benchmark/lib/python3.6/site-packages/torch/include/THC -I/usr/local/cuda-9.0.176/include -I/home/chy/anaconda3/envs/maskrcnn_benchmark/include/python3.6m -c /home/chy/github/vision/torchvision/csrc/ops/autocast/deform_conv2d_kernel.cpp -o build/temp.linux-x86_64-3.6/home/chy/github/vision/torchvision/csrc/ops/autocast/deform_conv2d_kernel.o -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=_C -D_GLIBCXX_USE_CXX11_ABI=0 -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=image -D_GLIBCXX_USE_CXX11_ABI=0 -std=c++11
cc1plus: warning: command line option ‘-Wstrict-prototypes’ is valid for C/ObjC but not for C++
:0:0: warning: "TORCH_EXTENSION_NAME" redefined
:0:0: note: this is the location of the previous definition
/home/chy/github/vision/torchvision/csrc/ops/autocast/deform_conv2d_kernel.cpp:3:32: fatal error: ATen/autocast_mode.h: No such file or directory
compilation terminated.

Versions

cuda 9.0
pytorch 1.1.0
Linux Ubuntu 16.04
install torchvision 0.3.0

Start with torchvision/csrc/ops/autocast/deform_conv2d_kernel.cpp and inspect how the build command includes headers from the installed PyTorch package. Reproduce the failure with python setup.py build using the reported CUDA, PyTorch, and torchvision versions. Done means the extension builds without the missing ATen/autocast_mode.h error.