Building TF Lite C++ shared library for macOS, linux, iOS and Android

[ebraraktas]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): macOS Mojave (10.14)
• TensorFlow installed from (source or binary): source
• TensorFlow version: r2.2
• Python version: 3.7
• Installed using virtualenv? pip? conda?: pip
• Bazel version (if compiling from source): 3.1.0
• GCC/Compiler version (if compiling from source): Apple clang version 11.0.0 (clang-1100.0.33.17)
• CUDA/cuDNN version: -
• GPU model and memory: -

I am trying to develop a C++ library on macOS and ubuntu to use in Android and iOS. Therefore, I am trying to build TFLite C++ shared library to link it my own C++ shared library using CMake. I have tried to build that using command bazel build -c opt //tensorflow/lite:tensorflowlite and got libtensorflowlite.dylib file as bazel-output. However CMake fails to link it and gives the following error:

Undefined symbols for architecture x86_64:
  "tflite::FlatBufferModel::BuildFromFile(char const*, tflite::ErrorReporter*)", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
  "tflite::FlatBufferModel::~FlatBufferModel()", referenced from:
      std::__1::default_delete<tflite::FlatBufferModel>::operator()(tflite::FlatBufferModel*) const in my_class.cpp.o
  "tflite::DefaultErrorReporter()", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
  "tflite::ops::builtin::BuiltinOpResolver::BuiltinOpResolver()", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
  "tflite::impl::Interpreter::AllocateTensors()", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
      MyClass::run_model(std::__1::vector<std::__1::vector<float, std::__1::allocator<float> >, std::__1::allocator<std::__1::vector<float, std::__1::allocator<float> > > > const&) in my_class.cpp.o
  "tflite::impl::Interpreter::ResizeInputTensor(int, std::__1::vector<int, std::__1::allocator<int> > const&)", referenced from:
      MyClass::run_model(std::__1::vector<std::__1::vector<float, std::__1::allocator<float> >, std::__1::allocator<std::__1::vector<float, std::__1::allocator<float> > > > const&) in my_class.cpp.o
  "tflite::impl::Interpreter::Invoke()", referenced from:
      MyClass::run_model(std::__1::vector<std::__1::vector<float, std::__1::allocator<float> >, std::__1::allocator<std::__1::vector<float, std::__1::allocator<float> > > > const&) in my_class.cpp.o
  "tflite::impl::Interpreter::~Interpreter()", referenced from:
      std::__1::default_delete<tflite::impl::Interpreter>::operator()(tflite::impl::Interpreter*) const in my_class.cpp.o
  "tflite::impl::InterpreterBuilder::InterpreterBuilder(tflite::FlatBufferModel const&, tflite::OpResolver const&)", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
  "tflite::impl::InterpreterBuilder::~InterpreterBuilder()", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
  "tflite::impl::InterpreterBuilder::operator()(std::__1::unique_ptr<tflite::impl::Interpreter, std::__1::default_delete<tflite::impl::Interpreter> >*)", referenced from:
      MyClass::MyClass(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, int, int, int) in my_class.cpp.o
  "vtable for tflite::MutableOpResolver", referenced from:
      tflite::MutableOpResolver::~MutableOpResolver() in my_class.cpp.o

But, If I try to link shared tflite library(libtensorflowlite.dylib) with my static library, it builds without error.

I am fairly new to bazel and CMake, however I thought it is caused by building the libtensorflowlite.dylib with missing x86_64 architecture and tried below commands and got following errors.:

bazel build -c opt --config=ios_fat  //tensorflow/lite:tensorflowlite
ERROR: /Users/username/tensorflow/tensorflow/lite/kernels/internal/BUILD:687:1: C++ compilation of rule '//tensorflow/lite/kernels/internal:kernel_utils' failed (Exit 1)
clang: error: invalid version number in '-mmacosx-version-min=13.2'

and

bazel build -c opt --cpu=x86_64  //tensorflow/lite:tensorflowlite
ERROR: /private/var/tmp/_bazel_username/ec3ee607500668f8e05b1977bf58c2f5/external/local_config_cc/BUILD:41:1: in cc_toolchain_suite rule @local_config_cc//:toolchain: cc_toolchain_suite '@local_config_cc//:toolchain' does not contain a toolchain for cpu 'x86_64'
ERROR: Analysis of target '//tensorflow/lite:tensorflowlite' failed; build aborted: Analysis of target '@local_config_cc//:toolchain' failed

Finally, I thought it is related to #36178 , #39876 and #35386, but couldn't resolve it.

[ebraraktas]

As I said, I was (and still am) new to CMake. And CMakeList.txt of my project was missing some link parameters. I added them and it is working. Therefore, there was not an issue at all. Sorry to bother.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No

[wisonye]

And CMakeList.txt of my project was missing some link parameters. I added them and it is working. Therefore, there was not an issue at all. Sorry to bother.

@ebraraktas Hi there, could u list all the details how you solve it plz? Definitely need your help on this:)