tensorflow: While compiling external app -> fatal error: unsupported/Eigen/CXX11/Tensor: No such file or directory

Closing due to lack of activity.

Has anyone been able to resolve the issues ?

I have included “tensorflow/third_party/eigen3” but I get below error

tensorflow/third_party/eigen3/unsupported/Eigen/CXX11/Tensor:1:42: error: #include nested too deeply #include “unsupported/Eigen/CXX11/Tensor”

• You don’t have to use Bazel to build tensorflow
• Your code don’t have to place in the Tensorflow repo.

1. Bazel build — success

follow the instruction of https://www.tensorflow.org/api_guides/cc/guide

2. g++ build — success

The following error comes from the wrong eigen header, as mentioned by @leftofcenter

eigen3/Eigen/Core:1:22: error: #include nested too deeply

building script

g++ -std=c++11 example.cc -o example  \
  -I /root/tensorflow-1.10/bazel-genfiles \
  -I /root/tensorflow-1.10  \
  -I /usr/local/lib/python2.7/dist-packages/tensorflow/include/external/eigen_archive  \
  -I /usr/local/lib/python2.7/dist-packages/tensorflow/include \
  -L /root/tensorflow-1.10/bazel-bin/tensorflow \
  -l tensorflow_cc \
  -l tensorflow_framework

Has anyone been able to resolve the issues ?

I have included “tensorflow/third_party/eigen3” but I get below error

tensorflow/third_party/eigen3/unsupported/Eigen/CXX11/Tensor:1:42: error: #include nested too deeply #include “unsupported/Eigen/CXX11/Tensor”

@ashokbugude did you able to solve it? I have a similar problem

When I compile it with GCC I have no problem But when I try to use the same code with ROS package and compiling with catkin_make I get the same error

EDIT: Found the solution in compiler instructions I included the following folders: /usr/local/include/tensorflow/ /usr/local/include/tensorflow/tensorflow/contrib/makefile/downloads/eigen/ /usr/local/include/tensorflow/bazel-tensorflow/tensorflow/contrib/makefile/gen/protobuf/include/ /usr/local/include/tensorflow/bazel-tensorflow/tensorflow/contrib/makefile/downloads/nsync/public/

Regarding OP’s command that failed which was, g++ -I /opt/tensorflow -I /opt/tensorflow/bazel-genfiles loader.cpp I placed the source code in the tensorflow tree that I built from source so paths are different:

g++ -c -std=c++11 \ -I …/…/… -I …/…/…/bazel-genfiles \ -I/usr/local/lib/python2.7/dist-packages/tensorflow/include/external/eigen_archive/ \ loader.cpp

it compiles.

“unsupported/Eigen/CXX11/Tensor” is in the directory of the 3rd -I. The Tensor header is not including itself that has been wrongly assumed by some posters. This assumes you have installed the python .whl file that the build generates.

So I didn’t realise that you can’t actually link with Tensorflow as a normal C++ library from an external tool. You have to use Bazel and your code has to be inside the Tensorflow repo. It does tell you this in the C++ tutorial but it is very unexpected and I hadn’t read that.

Anyway, using Bazel resolved the errors about Eigen. It seems like it downloads it automatically somehow.

Haven’t had a chance to look into the Eigen headers. The documentation being out of date I will fix; it needs to use tf_cc_binary to link in its dependencies (https://github.com/tensorflow/tensorflow/issues/13855).

I have the same issue but it happens when I #include <tensorflow/core/framework/op_kernel.h>:

In file included from ../tensorflow/tensorflow/core/framework/op_kernel.h:23:
In file included from ../tensorflow/tensorflow/core/framework/allocator.h:23:
In file included from ../tensorflow/tensorflow/core/framework/numeric_types.h:21:
../tensorflow/third_party/eigen3/unsupported/Eigen/CXX11/Tensor:1:10: fatal error: 'unsupported/Eigen/CXX11/Tensor' file not found
#include "unsupported/Eigen/CXX11/Tensor"

I added -I../tensorflow/third_party/eigen3 which allows it to find the Tensor header, but … that file seems to #include itself??

$ cd ../tensorflow/third_party/eigen3
$ ls unsupported/Eigen/CXX11/Tensor

#include "unsupported/Eigen/CXX11/Tensor"

#ifdef _WIN32
#ifndef SLEEP_FUNC_HEADER_GUARD
#define SLEEP_FUNC_HEADER_GUARD
inline void sleep(unsigned int seconds) { Sleep(1000*seconds); }
#endif

// On Windows, Eigen will include Windows.h, which defines various
// macros that conflict with TensorFlow symbols. Undefine them here to
// prevent clashes.
#undef DeleteFile
#undef ERROR
#undef LoadLibrary
#endif  // _WIN32

I don’t see how this could ever work. Also that file doesn’t match the current Tensor header. In fact it seems like all of the files in third_party/eigen3 just mirror the structure of Eigen and #include themselves.

In 2016 they actually #included a different directory, e.g.

$ cd ../tensorflow/third_party/eigen3/
$ ls Eigen/Core

#include "eigen-eigen-b4fa9622b809/Eigen/Core"

But in July 2016 Igor Babuschkin changed them all like this:

#include "Eigen/Core"

Can’t see how that is supposed to work given that there is no other copy of Eigen in the source. Am I supposed to have a copy of it on my system? Anyway I guess @ibab would know!