Linking issues with tensorflow-lite.a with android-studio and CMake. #19642

uelordi01 · 2018-05-30T09:48:11Z

System information

Have I written custom code:
OS Platform and Distribution ( ):
Linux Ubuntu 16.04 / Android
TensorFlow installed from:
source
Python version:
2.7
Bazel version (if compiling from source):
0.13
GCC/Compiler version (if compiling from source):
NDK16b

Describe the problem

I am having problems linking libtensorflow-lite.a with CMake and Android Studio.
It seems that the tf-lite library is not compatible with my architecture, but I do not know why this is happening.
I have succesfully compiled the libtensorflow-lite.a library, following these steps:

export NDK_ROOT=<your_ndk>
tensorflow/contrib/lite/download_dependencies.sh
make -f tensorflow/contrib/lite/Makefile TARGET=ANDROID ANDROID_ARCH=armeabi-v7a

The problem comes when I try to link the library with CMake. it seems that even I compiled the armeabi-v7a the linker, I have these errors:

incompatible target

the vtable symbol may be undefined because the class is missing its key function

Source code / logs

This my error log:

Build command failed.
Error while executing process /home/VICOMTECH/uelordi/Android/Sdk/cmake/3.6.4111459/bin/cmake with arguments {--build /home/VICOMTECH/uelordi/projects/SwitchTrackingEngine/kanvas_tracker/kt/tools/Deep-learning-tools/android/tfliteTest/app/.externalNativeBuild/cmake/debug/armeabi-v7a --target native-lib}
[1/1] Linking CXX shared library ../../../../build/intermediates/cmake/debug/obj/armeabi-v7a/libnative-lib.so
FAILED: : && /home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/llvm/prebuilt/linux-x86_64/bin/clang++  --target=armv7-none-linux-androideabi --gcc-toolchain=/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64 --sysroot=/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/sysroot -fPIC -isystem /home/VICOMTECH/uelordi/SDK/android-ndk-r16b/sysroot/usr/include/arm-linux-androideabi -D__ANDROID_API__=21 -g -DANDROID -ffunction-sections -funwind-tables -fstack-protector-strong -no-canonical-prefixes -march=armv7-a -mfloat-abi=softfp -mfpu=vfpv3-d16 -fno-integrated-as -mthumb -Wa,--noexecstack -Wformat -Werror=format-security  -std=c++11 -O0 -fno-limit-debug-info  -Wl,--exclude-libs,libgcc.a -Wl,--exclude-libs,libatomic.a --sysroot /home/VICOMTECH/uelordi/SDK/android-ndk-r16b/platforms/android-21/arch-arm -Wl,--build-id -Wl,--warn-shared-textrel -Wl,--fatal-warnings -Wl,--fix-cortex-a8 -Wl,--no-undefined -Wl,-z,noexecstack -Qunused-arguments -Wl,-z,relro -Wl,-z,now -shared -Wl,-soname,libnative-lib.so -o ../../../../build/intermediates/cmake/debug/obj/armeabi-v7a/libnative-lib.so CMakeFiles/native-lib.dir/src/main/cpp/native-lib.cpp.o  /home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/gen/lib/libtensorflow-lite.a /home/VICOMTECH/uelordi/SDK/android-ndk-r16b/platforms/android-21/arch-arm/usr/lib/liblog.so -latomic -lm "/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/sources/cxx-stl/gnu-libstdc++/4.9/libs/armeabi-v7a/libgnustl_static.a" && :
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x/../../../../arm-linux-androideabi/bin/ld: error: /home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/gen/lib/libtensorflow-lite.a(error_reporter.o): incompatible target
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x/../../../../arm-linux-androideabi/bin/ld: error: /home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/gen/lib/libtensorflow-lite.a(interpreter.o): incompatible target
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x/../../../../arm-linux-androideabi/bin/ld: error: /home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/gen/lib/libtensorflow-lite.a(register.o): incompatible target
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x/../../../../arm-linux-androideabi/bin/ld: error: /home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/gen/lib/libtensorflow-lite.a(model.o): incompatible target
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x/../../../../arm-linux-androideabi/bin/ld: error: /home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/gen/lib/libtensorflow-lite.a(op_resolver.o): incompatible target
/home/VICOMTECH/uelordi/projects/SwitchTrackingEngine/kanvas_tracker/kt/tools/Deep-learning-tools/android/tfliteTest/app/src/main/cpp/native-lib.cpp:23: error: undefined reference to 'tflite::DefaultErrorReporter()'
/home/VICOMTECH/uelordi/projects/SwitchTrackingEngine/kanvas_tracker/kt/tools/Deep-learning-tools/android/tfliteTest/app/src/main/cpp/native-lib.cpp:23: error: undefined reference to 'tflite::FlatBufferModel::BuildFromFile(char const*, tflite::ErrorReporter*)'
/home/VICOMTECH/uelordi/projects/SwitchTrackingEngine/kanvas_tracker/kt/tools/Deep-learning-tools/android/tfliteTest/app/src/main/cpp/native-lib.cpp:27: error: undefined reference to 'tflite::ops::builtin::BuiltinOpResolver::BuiltinOpResolver()'
/home/VICOMTECH/uelordi/projects/SwitchTrackingEngine/kanvas_tracker/kt/tools/Deep-learning-tools/android/tfliteTest/app/src/main/cpp/native-lib.cpp:28: error: undefined reference to 'tflite::InterpreterBuilder::InterpreterBuilder(tflite::FlatBufferModel const&, tflite::OpResolver const&)'
/home/VICOMTECH/uelordi/projects/SwitchTrackingEngine/kanvas_tracker/kt/tools/Deep-learning-tools/android/tfliteTest/app/src/main/cpp/native-lib.cpp:30: error: undefined reference to 'tflite::InterpreterBuilder::operator()(std::unique_ptr<tflite::Interpreter, std::default_delete<tflite::Interpreter> >*)'
/home/VICOMTECH/uelordi/SDK/tensorflow/tensorflow/contrib/lite/op_resolver.h:70: error: undefined reference to 'vtable for tflite::MutableOpResolver'
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/toolchains/arm-linux-androideabi-4.9/prebuilt/linux-x86_64/lib/gcc/arm-linux-androideabi/4.9.x/../../../../arm-linux-androideabi/bin/ld: the vtable symbol may be undefined because the class is missing its key function
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/sources/cxx-stl/gnu-libstdc++/4.9/include/bits/unique_ptr.h:76: error: undefined reference to 'tflite::FlatBufferModel::~FlatBufferModel()'
/home/VICOMTECH/uelordi/SDK/android-ndk-r16b/sources/cxx-stl/gnu-libstdc++/4.9/include/bits/unique_ptr.h:76: error: undefined reference to 'tflite::Interpreter::~Interpreter()'
clang++: error: linker command failed with exit code 1 (use -v to see invocation)
ninja: build stopped: subcommand failed.

My CmakeScript:

# For more information about using CMake with Android Studio, read the
# documentation: https://d.android.com/studio/projects/add-native-code.html

# Sets the minimum version of CMake required to build the native library.
`
SET(PROJECT_NAME  native-lib)
cmake_minimum_required(VERSION 3.4.1)
set(PREBUILT_DIR ${TENSORFLOW_ROOT_DIR}/tensorflow/contrib/lite)

add_library(lib_tflite STATIC IMPORTED )
set_target_properties(lib_tflite PROPERTIES IMPORTED_LOCATION
${PREBUILT_DIR}/gen/lib/libtensorflow-lite.a)

INCLUDE_DIRECTORIES(${TENSORFLOW_ROOT_DIR}
    ${PREBUILT_DIR}/downloads/flatbuffers/include)

add_library( # Sets the name of the library.
            ${PROJECT_NAME}
             SHARED
             src/main/cpp/native-lib.cpp )

find_library( log-lib
              log )


target_link_libraries( # Specifies the target library.
                       ${PROJECT_NAME}
                       lib_tflite
                       ${log-lib} )

My build_gradle (I have only copy the relevant part)

android {
    compileSdkVersion 27
    defaultConfig {
        applicationId "com.androidapp.kt.kanvas.tflitetest"
        minSdkVersion 21
        targetSdkVersion 27
        versionCode 1
        versionName "1.0"
        testInstrumentationRunner "android.support.test.runner.AndroidJUnitRunner"
        externalNativeBuild {
            cmake {
                cppFlags "-std=c++11"
                arguments "-DTENSORFLOW_ROOT_DIR=" + getTensorflowDir().toString()
                abiFilters  'armeabi-v7a' 
            }
        }
    }

The text was updated successfully, but these errors were encountered:

reedwm · 2018-06-04T19:45:22Z

/CC @aselle

tensorflowbutler · 2018-06-19T18:43:53Z

Nagging Assignee @aselle: It has been 14 days with no activity and this issue has an assignee. Please update the label and/or status accordingly.

WeiboXu · 2018-06-26T06:27:23Z

@uelordi01 Have you resolved this issue? I also meet the similar problem.

uelordi01 · 2018-06-26T11:57:30Z

Unfortunately not, but I think my error is related to the architecture mismatch, actually when I tried to make objdump in the tf-lite static library it says that is x86-64 architecture.

I think this make -f tensorflow/contrib/lite/Makefile TARGET=ANDROID ANDROID_ARCH=armeabi-v7a is it does not really take into account the android architecture.

Furthermore I tried to use this issue explanations:
#14688
to build armv7 target abi with no success.

So it seems that we need an android toolchain to say to gcc or clang compiler the correct compilation flgs.

tensorflowbutler · 2018-07-11T12:51:11Z

Nagging Assignee @aselle: It has been 14 days with no activity and this issue has an assignee. Please update the label and/or status accordingly.

shashishekhar · 2018-07-17T15:17:24Z

@uelordi01 any reason you can use the bazel build?

shashishekhar · 2018-07-19T01:39:04Z

@uelordi01 : Let me know if you are still having issues, reopen the issue.

zhyj3038 · 2018-09-19T04:53:29Z

hi, have you solved the "incompatible target" problem ?

uelordi01 · 2018-09-19T07:54:02Z

Currently I am able to compile the tensorflow-lite.a through bazel, and also I can build shared lib adding in tese lines to build_file:

cc_binary(
name = "libtensorflowLite.so",

linkopts=["-shared", "-Wl"],
linkshared=1,

copts = tflite_copts(),
deps = [
    ":framework",
    "//tensorflow/contrib/lite/kernels:builtin_ops",
],

zhyj3038 · 2018-10-12T12:51:13Z

@uelordi01 hi, follow your instruction, I get a libtensorflowLite.so and run it on mobile phone. But the size is nearly 37M, is it normal ?

uelordi01 · 2018-10-16T12:57:34Z

My tensorflow-lite library size is 34.7 MB, so I guess depending on the version of the tensorflow-lite is the reason that you have 37 MB size library.
Probably if you link a static library, you will reduce your final library size. But I couln't compile for android ndk, and now I am not on it.

zhyj3038 · 2018-10-17T03:27:30Z

@uelordi01 yes, now I have integrate th dynamic library in my android project.

holyhao · 2018-11-09T03:00:54Z

@zhyj3038 Hi, how can you use the libtensorflowLite.so in an android project? which head files do you include ？

zhyj3038 · 2018-11-09T08:07:31Z

@holyhao the whole tflite folder

holyhao · 2018-11-09T09:10:00Z

@zhyj3038 Thans for your reply, i have some trouble about compiling to the dynamic library. If any possible, would you please share it ?

zhyj3038 · 2018-11-09T09:21:40Z

just add

cc_binary(
name = "libtensorflowLite.so",

linkopts=["-shared", "-Wl"],
linkshared=1,

copts = tflite_copts(),
deps = [
":framework",
"//tensorflow/contrib/lite/kernels:builtin_ops",
],

to any positon of tensorflow's BUILD file. the BUILD file is in tensorflow's root folder

holyhao · 2018-11-09T09:59:25Z

@zhyj3038 Because of some proxy problem , i failed compiling with bazel .
The following are the error message, do you have some idea about this?:
Encountered error while reading extension file 'closure/defs.bzl': no such package '@io_bazel_rules_closure//closure': Error downloading [https://mirror.bazel.build/github.com/bazelbuild/rules_closure/archive/dbb96841cc0a5fb2664c37822803b06dab20c7d1.tar.gz, https://github.com/bazelbuild/rules_closure/archive/dbb96841cc0a5fb2664c37822803b06dab20c7d1.tar.gz] to /home/haolei/.cache/bazel/_bazel_haolei/3a2d9699e4c676169595171b9b4d83da/external/io_bazel_rules_closure/dbb96841cc0a5fb2664c37822803b06dab20c7d1.tar.gz: All mirrors are down: []

holyhao · 2018-11-20T02:14:06Z

@uelordi01 hi, I get a libtensorflowLite.so sucessfully, But the size is almost 40MB. I know that the offical libtensorflowlite_jni.so only 2.04MB, how could it be so different, do you have some idea?

uelordi01 · 2018-11-20T15:00:04Z

I cannot assure, but this can be due to the lite jni library could work as a wrapper to make calls to tensorflow-lite core functions, so in concluision, probably in this library are placed only wrapper functions.

Another idea could be because the jni lit library linking could be done through static library where this library would have had only the operations that the jni library.
However, I think is the first option.

Timshel · 2018-11-27T17:38:44Z

@uelordi01, hi, I have the same problem when using make the resulting library appear to be for x86-64.
On 19 Sep You said :

Currently I am able to compile the tensorflow-lite.a through bazel

Would you mind sharing this ? :) (I'm able to build a shared lib but due to external limitations I need a static lib).

This target exports the interpreter C++ API and the kernel C API, and includes the builtin set of ops. Fixes #25278 and fixes #19642. PiperOrigin-RevId: 232396927

gargn added the comp:lite TF Lite related issues label May 31, 2018

tensorflowbutler assigned reedwm Jun 2, 2018

reedwm assigned aselle and unassigned reedwm Jun 4, 2018

shashishekhar assigned shashishekhar and unassigned aselle Jul 17, 2018

shashishekhar added the stat:awaiting response Status - Awaiting response from author label Jul 17, 2018

shashishekhar closed this as completed Jul 19, 2018

holyhao mentioned this issue Jan 29, 2019

Reducing the binary size of tflite library for android #25278

Closed

tensorflow-copybara pushed a commit that referenced this issue Feb 5, 2019

Add libtensorflowlite.so target

0300883

This target exports the interpreter C++ API and the kernel C API, and includes the builtin set of ops. Fixes #25278 and fixes #19642. PiperOrigin-RevId: 232396927

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Linking issues with tensorflow-lite.a with android-studio and CMake. #19642

Linking issues with tensorflow-lite.a with android-studio and CMake. #19642

uelordi01 commented May 30, 2018

reedwm commented Jun 4, 2018

tensorflowbutler commented Jun 19, 2018

WeiboXu commented Jun 26, 2018

uelordi01 commented Jun 26, 2018

tensorflowbutler commented Jul 11, 2018

shashishekhar commented Jul 17, 2018

shashishekhar commented Jul 19, 2018

zhyj3038 commented Sep 19, 2018

uelordi01 commented Sep 19, 2018

zhyj3038 commented Oct 12, 2018

uelordi01 commented Oct 16, 2018 •

edited

zhyj3038 commented Oct 17, 2018

holyhao commented Nov 9, 2018

zhyj3038 commented Nov 9, 2018

holyhao commented Nov 9, 2018

zhyj3038 commented Nov 9, 2018

holyhao commented Nov 9, 2018

holyhao commented Nov 20, 2018

uelordi01 commented Nov 20, 2018

Timshel commented Nov 27, 2018

Linking issues with tensorflow-lite.a with android-studio and CMake. #19642

Linking issues with tensorflow-lite.a with android-studio and CMake. #19642

Comments

uelordi01 commented May 30, 2018

System information

Describe the problem

Source code / logs

reedwm commented Jun 4, 2018

tensorflowbutler commented Jun 19, 2018

WeiboXu commented Jun 26, 2018

uelordi01 commented Jun 26, 2018

tensorflowbutler commented Jul 11, 2018

shashishekhar commented Jul 17, 2018

shashishekhar commented Jul 19, 2018

zhyj3038 commented Sep 19, 2018

uelordi01 commented Sep 19, 2018

zhyj3038 commented Oct 12, 2018

uelordi01 commented Oct 16, 2018 • edited

zhyj3038 commented Oct 17, 2018

holyhao commented Nov 9, 2018

zhyj3038 commented Nov 9, 2018

holyhao commented Nov 9, 2018

zhyj3038 commented Nov 9, 2018

holyhao commented Nov 9, 2018

holyhao commented Nov 20, 2018

uelordi01 commented Nov 20, 2018

Timshel commented Nov 27, 2018

uelordi01 commented Oct 16, 2018 •

edited