Can't compile arduino project - undefined reference

TheArheus · December 5, 2019, 1:03pm

Hello, I have some issues in compiling my project with arduino tensorflow lite library.

I can define some of the variables, but if I try to call class methos, or create class instance it just won’t compile.

Here is my code:
    #undef min
    #undef max
    #include "TensorFlowLite.h"
    #include "tensorflow/lite/experimental/micro/micro_mutable_op_resolver.h"
    #include "tensorflow/lite/experimental/micro/kernels/micro_ops.h"
    #include "tensorflow/lite/experimental/micro/kernels/all_ops_resolver.h"
    #include "tensorflow/lite/experimental/micro/micro_error_reporter.h"
    #include "tensorflow/lite/experimental/micro/micro_interpreter.h"
    #include "tensorflow/lite/experimental/micro/testing/micro_test.h"
    #include "tensorflow/lite/schema/schema_generated.h"
    #include "tensorflow/lite/version.h"
    #include "sine_model_data.h"
    //#include "constants.h"

    namespace {
      tflite::ErrorReporter* error_reporter = nullptr;
      const tflite::Model* model = nullptr;
      tflite::MicroInterpreter* interpreter = nullptr;
      TfLiteTensor* input = nullptr;
      TfLiteTensor* output = nullptr;
      int inference_count = 0;

      constexpr int kTensorArenaSize = 2 * 1024;
      uint8_t tensor_arena[kTensorArenaSize];
    } 


    void setup() {
      tflite::MicroErrorReporter mreporter;
      error_reporter = &mreporter;

      model = tflite::GetModel(g_sine_model_data);
      if(model->version() != TFLITE_SCHEMA_VERSION){
        error_reporter->Report("Error"); //this line is couse to error somehow
      }
      static tflite::MicroMutableOpResolver micro_mutable_resolver; //this line is couse to error somehow
      resolver.AddBuiltin(tflite::BuiltinOperator_ADD, tflite::ops::micro::Register_ADD()); //this line is couse to error somehow
}

Build output:

c:/users/1/appdata/local/arduino15/packages/maixduino/tools/riscv64-unknown-elf-gcc/8.2.0_20190409/bin/…/lib/gcc/riscv64-unknown-elf/8.2.0/…/…/…/…/riscv64-unknown-elf/bin/ld.exe: sketch\constants.h.cpp.o: in function `setup’:

c:\Users\1\Documents\PlatformIO\Projects\ff\src/main.ino:31: undefined reference to `vtable for tflite::MicroErrorReporter’

c:/users/1/appdata/local/arduino15/packages/maixduino/tools/riscv64-unknown-elf-gcc/8.2.0_20190409/bin/…/lib/gcc/riscv64-unknown-elf/8.2.0/…/…/…/…/riscv64-unknown-elf/bin/ld.exe: c:\Users\1\Documents\PlatformIO\Projects\ff\src/main.ino:36: undefined reference to `tflite::ErrorReporter::Report(char const*, …)’

c:/users/1/appdata/local/arduino15/packages/maixduino/tools/riscv64-unknown-elf-gcc/8.2.0_20190409/bin/…/lib/gcc/riscv64-unknown-elf/8.2.0/…/…/…/…/riscv64-unknown-elf/bin/ld.exe: sketch\constants.h.cpp.o: in function `_GLOBAL__sub_I_kInferencesPerCycle’:

C:\Users\1\Documents\Arduino\libraries\Arduino_TensorFlowLite\src/tensorflow/lite/experimental/micro/micro_mutable_op_resolver.h:27: undefined reference to `vtable for
tflite::MicroMutableOpResolver’

collect2.exe: error: ld returned 1 exit status

Multiple libraries were found for “TensorFlowLite.h”

Used: C:\Users\1\Documents\Arduino\libraries\Arduino_TensorFlowLite

exit status 1

Thank you for helping

oceansource · December 5, 2019, 4:19pm

Woah! TensorFlow in Arduino!? Cool…

The Maixduino looks awesome too!

Not sure how helpful this is, but I found what looks like a related issue:

github.com/tensorflow/tensorflow

Tensorflow lite undefined reference

opened 10:43AM - 08 Apr 19 UTC

closed 02:11PM - 08 Apr 19 UTC

strattist

**System information** - OS Platform and Distribution: Linux Ubuntu 18.04 - Te…nsorFlow installed from: source - TensorFlow version: 1.12.1 - Bazel version: 0.15.0 - GCC/Compiler version: 7.3.0 (No python related information since I try to compile lite for C/C++ usage) **Describe the problem** **Provide the exact sequence of commands / steps that you executed before running into the problem** First of all, I use CMake through a combo `add_custom_command` and `add_custom_target` to handle the tensorflow build for me. I try to build tensorflow lite for x86 or ARM depending on my needs (I've done it before on the 1.10 tensorflow version). The commands I execute to build tensorflow are in a script: ``` bash configure bazel --output_base=~/.tmp/tensorflow_lite/1.12.1/.Debug_x86 \ build --jobs=4 \ //tensorflow/contrib/lite/experimental/c:libtensorflowlite_c.so ``` The build happens fine, I link `libtensorflowlite_c.so` to the custom CMake target I want to use in my own environment and it shows correctly on my make logs. However I have undefined references for: ``` `tflite::DefaultErrorReporter()' `tflite::Interpreter::AllocateTensors()' `tflite::InterpreterBuilder::InterpreterBuilder(tflite::FlatBufferModel const&, tflite::OpResolver const&)' `vtable for tflite::MutableOpResolver' `tflite::InterpreterBuilder::operator()(std::unique_ptr<tflite::Interpreter, std::default_delete<tflite::Interpreter> >*)' `tflite::InterpreterBuilder::~InterpreterBuilder()' `tflite::Interpreter::Invoke()' `tflite::Interpreter::~Interpreter()' `vtable for tflite::ops::builtin::BuiltinOpResolver' `tflite::Interpreter::SetNumThreads(int)' `tflite::FlatBufferModel::~FlatBufferModel()' `tflite::ops::builtin::BuiltinOpResolver::BuiltinOpResolver()' `tflite::FlatBufferModel::BuildFromFile(char const*, tflite::ErrorReporter*)' ``` I quickly checked the bazel workspace and build directives and it seems my issues are related to the `framework` cc_library bazel is creating and that is linked to `libtensorflowlite_c` through the `c_api`. I don't know bazel a lot, I'm a bit lost with the dependencies. I would say the dependencies should propagate like this `framework -> c_api & c_api_internal -> libtensorflowlite_c.so`. Is it what happen? Should I add `//tensorflow/contrib/lite:framework` and other bazel target to be sure they are built? How bazel handles mutual dependency, is this an acceptable situation? It happens between `context` (`//tensorflow/contrib/lite`) and `c_api_internal` (`//tensorflow/contrib/lite/experimental/c`) and between `frawework` (`//tensorflow/contrib/lite`) and `c_api_internal` (`//tensorflow/contrib/lite/experimental/c`). ``` # context definition, needs c_api_internal cc_library( name = "context", hdrs = ["context.h"], deps = ["//tensorflow/contrib/lite/c:c_api_internal"], ) # framework definition, needs c_api_internal (only deps are shown) cc_library( name = "framework", srcs = [...] + select({...}), hdrs = [...], copts = tflite_copts(), defines = select({...}), linkopts = [] + select({...}), deps = [ ":arena_planner", ":graph_info", ":memory_planner", ":schema_fbs_version", ":simple_memory_arena", ":string", ":util", "//tensorflow/contrib/lite/c:c_api_internal", "//tensorflow/contrib/lite/core/api", "//tensorflow/contrib/lite/kernels:eigen_support", "//tensorflow/contrib/lite/kernels:gemm_support", "//tensorflow/contrib/lite/nnapi:nnapi_lib", "//tensorflow/contrib/lite/profiling:profiler", "//tensorflow/contrib/lite/schema:schema_fbs", ] + select({...}), ) # c_api_internal definition, needs context & framework cc_library( name = "c_api_internal", srcs = ["c_api.h"], hdrs = ["c_api_internal.h"], copts = tflite_copts(), visibility = [ "//tensorflow/contrib/lite/experimental/c:__subpackages__", ], deps = [ "//tensorflow/contrib/lite:context", "//tensorflow/contrib/lite:framework", ], ) # ``` CLOSED: Still have the issue but it's a CMake linkage issue, not related to tensorflow lite build.