Reenable bazel build

602300d6 · Liangliang He · 384099e5 · 602300d6 · 602300d6 · 602300d6
39 changed file
--- a/.gitignore
+++ b/.gitignore
 bazel-*
-tags
-.idea/
-cmake-build-debug/
-*.pyc
-mace/codegen/models/
-mace/codegen/opencl/
-mace/codegen/opencl_bin/
-mace/codegen/tuning/
-mace/codegen/version/
-mace/codegen/engine/
-mace/codegen/lib/
 build/
+cmake-build/
+cmake-build-debug/
 docs/_build/
 *.a

+.idea/
 .vscode/
-builds/
-mace/examples/android/macelibrary/src/main/cpp/mace/
+tags

 \.project/
 *swp
 *~
+*.pyc
 .python-version

-mace/examples/android/macelibrary/src/main/cpp/include/mace/public/
-
-mace/examples/android/macelibrary/src/main/cpp/lib/arm64-v8a/
+mace/codegen/models/
+mace/codegen/opencl/
+mace/codegen/opencl_bin/
+mace/codegen/tuning/
+mace/codegen/version/
+mace/codegen/engine/
+mace/codegen/lib/

-mace/examples/android/macelibrary/src/main/jniLibs/arm64-v8a/
+examples/android/macelibrary/src/main/cpp/mace/
+examples/android/macelibrary/src/main/cpp/include/
+examples/android/macelibrary/src/main/cpp/lib/arm64-v8a/
+examples/android/macelibrary/src/main/jniLibs/arm64-v8a/
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -34,86 +34,70 @@ build_docs:
    paths:
      - docs/_build

-build_android-armeabi-v7a:
+cmake_build_android-armeabi-v7a:
  stage: basic-build
  script:
-    - sh tools/build-android-armeabi-v7a-full.sh
+    - sh tools/cmake-build-android-armeabi-v7a-full.sh

-build_android-arm64-v8:
+cmake_build_android-arm64-v8:
  stage: basic-build
  script:
-    - sh tools/build-android-arm64-v8a-full.sh
+    - sh tools/cmake-build-android-arm64-v8a-full.sh

-code_footprint_check:
+build_android_demo:
  stage: basic-build
+  script:
+    - pushd examples/android/ && bash build.sh static && bash build.sh dynamic && popd
+
+code_footprint_check:
+  stage: smoke-test
  script:
    - echo check shared library size
+    - LIBMACE64_FULL_SIZE=`stat -c%s build/android-arm64-v8a-full/install/lib/libmace.so`
+    - if (( LIBMACE64_FULL_SIZE > 3100000 )) ; then echo "The libmace.so size too large"; exit 1; fi
+    - LIBMACE32_FULL_SIZE=`stat -c%s build/android-armeabi-v7a-full/install/lib/libmace.so`
+    - if (( LIBMACE32_FULL_SIZE > 2200000 )) ; then echo "The libmace.so size too large"; exit 1; fi

-cc_test_smoketest:
+mace_cc_test_smoketest:
  stage: smoke-test
  script:
    - echo tests

-cc_benchmark_smoketest:
+mace_cc_benchmark_smoketest:
  stage: smoke-test
  script:
    - echo benchmark

-build_arm-linux-gnueabihf:
-  stage: build
-  script:
-    - sh tools/build-arm-linux-gnueabihf-full.sh
-  only:
-    - triggers
-
-build_aarch64-linux-gnu:
-  stage: build
-  script:
-    - sh tools/build-aarch64-linux-gnu-full.sh
-  only:
-    - triggers
-
-build_host:
+bazel_build:
  stage: build
  script:
-    - sh tools/build-host.sh
+    - sh tools/bazel-build-standalone-lib.sh
  only:
    - triggers

-build_android_demo:
+cmake_build:
  stage: build
  script:
-    - echo build android demo
+    - sh tools/cmake-build-standalone-lib.sh
  only:
    - triggers

 cc_test:
  stage: test
  script:
-    - echo cc_test
-  only:
-    - triggers
-
-cc_benchmark:
-  stage: test
-  script:
-    - echo cc_benchmark
-  only:
-    - triggers
-
-quantization_test:
-  stage: test
-  script:
-    - echo quantization_test
-  only:
-    - triggers
+    - if [ -z "$TARGET_SOCS" ]; then TARGET_SOCS=random; fi
+    - >
+      if ping -c 1 v9.git.n.xiaomi.com 1>/dev/null 2>&1; then
+        GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@v9.git.n.xiaomi.com:deep-computing/generic-mobile-devices.git
+        DEVICE_CONF_FILE=generic-mobile-devices/devices.yml
+      fi
+    - python tools/bazel_adb_run.py --target="//test/ccunit:mace_cc_test" --device_yml=${DEVICE_CONF_FILE} --run_target=True --stdout_processor=unittest_stdout_processor --target_abis=armeabi-v7a,arm64-v8a,arm64 --target_socs=$TARGET_SOCS

-model_test:
+mace_cc_benchmark:
  stage: test
  script:
-    - echo model_test
-  only:
-    - triggers
+    - if [ -z "$TARGET_SOCS" ]; then TARGET_SOCS=random; fi
+    - python tools/bazel_adb_run.py --target="//test/ccbenchmark:mace_cc_benchmark" --run_target=True --stdout_processor=ops_benchmark_stdout_processor --target_abis=armeabi-v7a,arm64-v8a --target_socs=$TARGET_SOCS --args="--filter=.*SIGMOID.*"

 ndk_compatibility_check:
  stage: extra
@@ -122,20 +106,72 @@ ndk_compatibility_check:
  only:
    - triggers

-python_compatibility_check:
-  stage: extra
+model_tests:
+  stage: test
  script:
-    - echo python3 check
-  only:
-    - triggers
-
-extra_tests:
+    - pwd
+    - rm -rf mace-models
+    - rm -rf generic-mobile-devices
+    - GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@github.com:XiaoMi/mace-models.git
+    - CONF_FILE=mace-models/mobilenet-v1/mobilenet-v1.yml
+    - >
+      if ping -c 1 v9.git.n.xiaomi.com 1>/dev/null 2>&1; then
+        GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@v9.git.n.xiaomi.com:deep-computing/generic-mobile-devices.git
+        DEVICE_CONF_FILE=generic-mobile-devices/devices.yml
+      fi
+    - if [ -z "$TARGET_SOCS" ]; then TARGET_SOCS=random; fi
+    - >
+      python tools/converter.py convert --config=${CONF_FILE}  --target_socs=$TARGET_SOCS --model_graph_format=file --model_data_format=file --cl_mem_type=buffer
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --device_yml=${DEVICE_CONF_FILE} --round=1 --target_abis=armeabi-v7a,arm64 --validate --model_graph_format=file --model_data_format=file
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --device_yml=${DEVICE_CONF_FILE} --example --target_abis=armeabi-v7a,arm64 --round=1 --validate --model_graph_format=file --model_data_format=file
+    - CONF_FILE=mace-models/mobilenet-v2/mobilenet-v2-host.yml
+    - >
+      python tools/converter.py convert --config=${CONF_FILE} --target_socs=$TARGET_SOCS --model_graph_format=file --model_data_format=file
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --round=1 --validate --model_graph_format=file --model_data_format=file --address_sanitizer
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --example --round=1 --validate --model_graph_format=file --model_data_format=file
+      python tools/converter.py benchmark --config=${CONF_FILE} --target_socs=$TARGET_SOCS --round=5 --model_graph_format=file --model_data_format=file
+      python tools/converter.py convert --config=${CONF_FILE} --target_socs=$TARGET_SOCS --model_graph_format=code --model_data_format=file
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --round=1 --validate --model_graph_format=code --model_data_format=file
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --example --round=1 --validate --model_graph_format=code --model_data_format=file
+      python tools/converter.py benchmark --config=${CONF_FILE} --target_socs=$TARGET_SOCS --round=5 --model_graph_format=code --model_data_format=file
+    - rm -rf mace-models
+
+quantization_tests:
+  stage: test
+  script:
+    - pwd
+    - rm -rf mace-models
+    - GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@github.com:XiaoMi/mace-models.git
+    - >
+      if ping -c 1 v9.git.n.xiaomi.com 1>/dev/null 2>&1; then
+        GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@v9.git.n.xiaomi.com:deep-computing/generic-mobile-devices.git
+        DEVICE_CONF_FILE=generic-mobile-devices/devices.yml
+      fi
+    - if [ -z "$TARGET_SOCS" ]; then TARGET_SOCS=random; fi
+    - >
+      for CONF_FILE in mace-models/mobilenet-v1/mobilenet-v1-quantize-friendly.yml mace-models/mobilenet-v1/mobilenet-v1-quantize-retrain-for-check-only.yml mace-models/mobilenet-v1/mobilenet-v1-quantize-retrain-dsp.yml;
+      do
+      python tools/converter.py convert --config=${CONF_FILE} --target_socs=$TARGET_SOCS --model_graph_format=file --model_data_format=file
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --device_yml=${DEVICE_CONF_FILE} --round=1 --validate --model_graph_format=file --model_data_format=file
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --device_yml=${DEVICE_CONF_FILE} --example --round=1 --validate --layers=0 --model_graph_format=file --model_data_format=file
+      done
+    - rm -rf mace-models
+
+dynamic_linking_test:
  stage: extra
  script:
-    - ops_test_disable_neon
-    - ops_test
-    - api_test
-    - extra_tests
-    - dynamic_link_test
-  only:
-    - triggers
+    - pwd
+    - rm -rf mace-models
+    - rm -rf generic-mobile-devices
+    - GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@github.com:XiaoMi/mace-models.git
+    - CONF_FILE=mace-models/mobilenet-v1/mobilenet-v1.yml
+    - >
+      if ping -c 1 v9.git.n.xiaomi.com 1>/dev/null 2>&1; then
+        GIT_SSH_COMMAND="ssh -o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no" git clone git@v9.git.n.xiaomi.com:deep-computing/generic-mobile-devices.git
+        DEVICE_CONF_FILE=generic-mobile-devices/devices.yml
+      fi
+    - if [ -z "$TARGET_SOCS" ]; then TARGET_SOCS=random; fi
+    - >
+      python tools/converter.py convert --config=${CONF_FILE}  --target_socs=$TARGET_SOCS --model_graph_format=file --model_data_format=file --cl_mem_type=buffer
+      python tools/converter.py run --config=${CONF_FILE} --target_socs=$TARGET_SOCS --device_yml=${DEVICE_CONF_FILE} --example --mace_lib_type=dynamic --target_abis=armeabi-v7a,arm64 --round=1 --validate --model_graph_format=file --model_data_format=file
+    - rm -rf mace-models
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -10,7 +10,7 @@ option(MACE_ENABLE_CUDA        "whether to enable CUDA support"             OFF)
 option(MACE_ENABLE_HEXAGON_DSP "whether to enable Hexagon DSP support"      OFF)
 option(MACE_ENABLE_HEXAGON_HTA "whether to enable Hexagon HTA support"      OFF)
 option(MACE_ENABLE_TESTS       "whether to build c++ unit tests"            OFF)
-option(MACE_ENABLE_BENCHMARK   "whether to build c++ micro benchmarks"      OFF)
+option(MACE_ENABLE_BENCHMARKS  "whether to build c++ micro benchmarks"      OFF)
 option(MACE_ENABLE_EXAMPLES    "whether to build examples"                  OFF)
 option(MACE_ENABLE_OPT_SIZE    "whether to build with optimized binary size" ON)
 option(MACE_ENABLE_OBFUSCATE   "whether to build with code obfuscation"      ON)
@@ -110,6 +110,6 @@ if(MACE_ENABLE_EXAMPLES)
  add_subdirectory(examples)
 endif()

-if(MACE_ENABLE_TESTS OR MACE_ENABLE_BENCHMARK)
+if(MACE_ENABLE_TESTS OR MACE_ENABLE_BENCHMARKS)
  add_subdirectory(test)
 endif()
--- a/docker/mace-dev-lite/Dockerfile
+++ b/docker/mace-dev-lite/Dockerfile
@@ -114,8 +114,7 @@ RUN apt-get install -y --no-install-recommends \
 # refer to: https://github.com/jenkinsci/docker/issues/506
 RUN apt-get install -y libltdl7

-RUN pip install --upgrade pip
-RUN pip install -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com setuptools
+RUN pip install -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com --upgrade pip setuptools
 RUN pip install -i http://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com \
    numpy==1.15.4 \
    scipy==1.2.0 \

--- a/docker/mace-dev/Dockerfile
+++ b/docker/mace-dev/Dockerfile
@@ -69,8 +69,15 @@ RUN cd /opt && \
    tar xf gcc-linaro-7.3.1-2018.05-x86_64_aarch64-linux-gnu.tar.xz && \
    rm -rf gcc-linaro-7.3.1-2018.05-x86_64_aarch64-linux-gnu.tar.xz

+# install protoc
+RUN cd opt/ && \
+    wget -q https://github.com/protocolbuffers/protobuf/releases/download/v3.6.1/protoc-3.6.1-linux-x86_64.zip && \
+    unzip protoc-3.6.1-linux-x86_64.zip -d protoc-3.6.1 && \
+    rm -rf protoc-3.6.1-linux-x86_64.zip
+
 ENV CROSS_TOOLCHAIN_PARENT /opt
 ENV ANDROID_NDK_VERSION r19c
 ENV ANDROID_NDK_HOME /opt/android-ndk-${ANDROID_NDK_VERSION}
 ENV LINARO_ARM_LINUX_GNUEABIHF /opt/gcc-linaro-7.3.1-2018.05-x86_64_arm-linux-gnueabihf
 ENV LINARO_AARCH64_LINUX_GNU /opt/gcc-linaro-7.3.1-2018.05-x86_64_aarch64-linux-gnu
+ENV PATH /opt/protoc-3.6.1/bin:${PATH}
--- a/docs/development/how_to_debug.rst
+++ b/docs/development/how_to_debug.rst
@@ -72,7 +72,7 @@ If model's output is suspected to be incorrect, it might be useful to debug your
 or use binary search method until suspicious layer is found.

 You can also specify `--layers` after `--validate` to validate all or some of the layers of the model(excluding some layers changed by MACE, e.g., BatchToSpaceND),
-it only supports TensorFlow now. You can find validation results in `builds/your_model/model/runtime_in_yaml/log.csv`.
+it only supports TensorFlow now. You can find validation results in `build/your_model/model/runtime_in_yaml/log.csv`.

 For quantized model, if you want to check one layer, you can add `check_tensors` and `check_shapes` like in the yaml above. You can only specify
 MACE op's output.
@@ -113,7 +113,7 @@ Debug model conversion
 After model is converted to MACE model, a literal model graph is generated in directory `mace/codegen/models/your_model`.
 You can refer to it when debugging model conversion.

-MACE also provides model visualization HTML generated in `builds` directory, generated after converting model.
+MACE also provides model visualization HTML generated in `build` directory, generated after converting model.


 Debug engine using log

--- a/docs/development/how_to_run_tests.md
+++ b/docs/development/how_to_run_tests.md
@@ -12,10 +12,10 @@ Run unit tests

 MACE use [gtest](https://github.com/google/googletest) for unit tests.

-* Run all unit tests defined in a Bazel target, for example, run `ops_test`:
+* Run all unit tests defined in a Bazel target, for example, run `mace_cc_test`:

  ```sh
-  python tools/bazel_adb_run.py --target="//mace/ops:ops_test" \
+  python tools/bazel_adb_run.py --target="//test/ccunit:mace_cc_test" \
                                --run_target=True
  ```

@@ -23,7 +23,7 @@ MACE use [gtest](https://github.com/google/googletest) for unit tests.
 for example, run `Conv2dOpTest` unit tests:

  ```sh
-  python tools/bazel_adb_run.py --target="//mace/ops:ops_test" \
+  python tools/bazel_adb_run.py --target="//test/ccunit:mace_cc_test" \
                                --run_target=True \
                                --args="--gtest_filter=Conv2dOpTest*"
  ```
@@ -34,10 +34,10 @@ Run micro benchmarks
 MACE provides a micro benchmark framework for performance tuning.

 * Run all micro benchmarks defined in a Bazel target, for example, run all
-`ops_benchmark` micro benchmarks:
+`mace_cc_benchmark` micro benchmarks:

  ```sh
-  python tools/bazel_adb_run.py --target="//mace/ops:ops_benchmark" \
+  python tools/bazel_adb_run.py --target="//test/ccbenchmark:mace_cc_benchmark" \
                                --run_target=True
  ```

@@ -45,7 +45,7 @@ MACE provides a micro benchmark framework for performance tuning.
 micro benchmarks:

  ```sh
-  python tools/bazel_adb_run.py --target="//mace/ops:ops_benchmark" \
+  python tools/bazel_adb_run.py --target="//test/ccbenchmark:mace_cc_benchmark" \
                                --run_target=True \
                                --args="--filter=MACE_BM_CONV_2D_.*_GPU"
  ```
--- a/docs/user_guide/advanced_usage.rst
+++ b/docs/user_guide/advanced_usage.rst
@@ -204,15 +204,15 @@ Convert model(s) to C++ code

            python tools/converter.py convert --config=/path/to/model_deployment_file.yml

-        The command will generate **${library_name}.a** in **builds/${library_name}/model** directory and
-        ** *.h ** in **builds/${library_name}/include** like the following dir-tree.
+        The command will generate **${library_name}.a** in **build/${library_name}/model** directory and
+        ** *.h ** in **build/${library_name}/include** like the following dir-tree.

        .. code::

             # model_graph_format: code
             # model_data_format: file

-             builds
+             build
               ├── include
               │   └── mace
               │       └── public
@@ -225,7 +225,7 @@ Convert model(s) to C++ code
             # model_graph_format: code
             # model_data_format: code

-             builds
+             build
               ├── include
               │   └── mace
               │       └── public
@@ -305,11 +305,11 @@ Tuning for specific SoC's GPU

            python tools/converter.py run --config=/path/to/model_deployment_file.yml --validate

-        The command will generate two files in `builds/${library_name}/opencl`, like the following dir-tree.
+        The command will generate two files in `build/${library_name}/opencl`, like the following dir-tree.

        .. code::

-              builds
+              build
              └── mobilenet-v2
                  ├── model
                  │   ├── mobilenet_v2.data

--- a/docs/user_guide/basic_usage.rst
+++ b/docs/user_guide/basic_usage.rst
@@ -44,17 +44,17 @@ Here we use the mobilenet-v2 model as an example.

        cd path/to/mace
        # Build library
-        # output lib path: builds/lib
-        bash tools/build-standalone-lib.sh
+        # output lib path: build/lib
+        bash tools/bazel-build-standalone-lib.sh


    .. note::

        - This step can be skipped if you just want to run a model using ``tools/converter.py``, such as commands in step 5.

-        - Libraries in ``builds/lib/armeabi-v7a/cpu_gpu/`` means it can run on ``cpu`` or ``gpu`` devices.
+        - Libraries in ``build/lib/armeabi-v7a/cpu_gpu/`` means it can run on ``cpu`` or ``gpu`` devices.

-        - The results in ``builds/lib/armeabi-v7a/cpu_gpu_dsp/`` need HVX supported.
+        - The results in ``build/lib/armeabi-v7a/cpu_gpu_dsp/`` need HVX supported.


    4. Convert the pre-trained mobilenet-v2 model to MACE format model.
@@ -170,7 +170,7 @@ When the deployment file is ready, you can use MACE converter tool to convert yo
    python tools/converter.py convert --config=/path/to/your/model_deployment_file.yml

 This command will download or load your pre-trained model and convert it to a MACE model proto file and weights data file.
-The generated model files will be stored in ``builds/${library_name}/model`` folder.
+The generated model files will be stored in ``build/${library_name}/model`` folder.

 .. warning::

@@ -188,10 +188,10 @@ Or use bazel to build MACE source code into a library.

        cd path/to/mace
        # Build library
-        # output lib path: builds/lib
-        bash tools/build-standalone-lib.sh
+        # output lib path: build/lib
+        bash tools/bazel-build-standalone-lib.sh

-The above command will generate dynamic library ``builds/lib/${ABI}/${DEVICES}/libmace.so`` and static library ``builds/lib/${ABI}/${DEVICES}/libmace.a``.
+The above command will generate dynamic library ``build/lib/${ABI}/${DEVICES}/libmace.so`` and static library ``build/lib/${ABI}/${DEVICES}/libmace.a``.

    .. warning::

@@ -278,7 +278,7 @@ header files.

 .. code::

-    builds
+    build
    ├── include
    │   └── mace
    │       └── public

--- a/docs/user_guide/benchmark.rst
+++ b/docs/user_guide/benchmark.rst
@@ -20,7 +20,7 @@ Usage

    .. code:: bash

-        python tools/bazel_adb_run.py --target="//mace/ops:ops_benchmark" --run_target=True  --args="--filter=.*BM_CONV.*"
+        python tools/bazel_adb_run.py --target="//test/ccbenchmark:mace_cc_benchmark" --run_target=True  --args="--filter=.*BM_CONV.*"

 ======
 Output

--- a/examples/android/build.sh
+++ b/examples/android/build.sh
@@ -20,12 +20,12 @@ fi

 MACE_LINK_TYPE=$1

-pushd ../../../
+pushd ../..

 TARGET_ABI=arm64-v8a
-ANDROID_DEMO_DIR=mace/examples/android/
+ANDROID_DEMO_DIR=examples/android/
 LIBRARY_DIR=$ANDROID_DEMO_DIR/macelibrary/src/main/cpp/
-INCLUDE_DIR=$LIBRARY_DIR/include/mace/public/
+INCLUDE_DIR=$LIBRARY_DIR/include
 LIBMACE_DIR=$LIBRARY_DIR/lib/$TARGET_ABI/
 LIBGNUSTL_SHARED_SO=libgnustl_shared.so
 LIBCPP_SHARED_SO=libc++_shared.so
@@ -44,20 +44,17 @@ else
  exit 1
 fi

-rm -rf $LIBRARY_DIR/include/
-mkdir -p $INCLUDE_DIR
-
-rm -rf $LIBRARY_DIR/lib/
-mkdir -p $LIBMACE_DIR
+python tools/converter.py convert --config=examples/android/mobilenet.yml --target_abis=$TARGET_ABI

+rm -rf $INCLUDE_DIR && mkdir -p $INCLUDE_DIR
+rm -rf $LIBMACE_DIR && mkdir -p $LIBMACE_DIR
 rm -rf $LIBRARY_DIR/model/

-python tools/converter.py convert --config=mace/examples/android/mobilenet.yml --target_abis=$TARGET_ABI
-cp -rf builds/mobilenet/include/mace/public/*.h $INCLUDE_DIR
-cp -rf builds/mobilenet/model $LIBRARY_DIR
+cp -rf include/mace $INCLUDE_DIR
+cp -rf build/mobilenet/include/mace/public/*.h $INCLUDE_DIR/mace/public/
+cp -rf build/mobilenet/model $LIBRARY_DIR

 bazel build --config android --config optimization $BAZEL_LIBMACE_TARGET --define neon=true --define openmp=true --define opencl=true --define quantize=true --cpu=$TARGET_ABI
-cp -rf mace/public/*.h $INCLUDE_DIR
 cp -rf $BAZEL_GEN_LIBMACE_PATH $LIBMACE_DIR

 if [ $MACE_LINK_TYPE == "dynamic" ]; then

--- a/examples/android/macelibrary/CMakeLists.txt
+++ b/examples/android/macelibrary/CMakeLists.txt
@@ -13,6 +13,7 @@ cmake_minimum_required(VERSION 3.4.1)
 #set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${PROJECT_SOURCE_DIR}/../app/libs/${ANDROID_ABI})

 include_directories(${CMAKE_SOURCE_DIR}/)
+include_directories(${CMAKE_SOURCE_DIR}/src/main/cpp)
 include_directories(${CMAKE_SOURCE_DIR}/src/main/cpp/include)
 file(GLOB LIBMACE "${CMAKE_SOURCE_DIR}/src/main/cpp/lib/arm64-v8a/*")
 set(mace_lib ${LIBMACE})

--- a/examples/android/macelibrary/src/main/cpp/image_classify.cc
+++ b/examples/android/macelibrary/src/main/cpp/image_classify.cc
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-#include "src/main/cpp/image_classify.h"
+#include "image_classify.h"

 #include <android/log.h>
 #include <jni.h>
@@ -25,9 +25,8 @@
 #include <vector>
 #include <numeric>

-#include "src/main/cpp/include/mace/public/mace.h"
-#include "src/main/cpp/include/mace/public/mace_engine_factory.h"
 #include "mace/public/mace.h"
+#include "mace/public/mace_engine_factory.h"

 namespace {


--- a/include/BUILD.bazel
+++ b/include/BUILD.bazel
-# Description:
-# MACE public API.
-#
 package(
    default_visibility = ["//visibility:public"],
 )
@@ -8,12 +5,12 @@ package(
 licenses(["notice"])  # Apache 2.0

 cc_library(
-    name = "public",
-    hdrs = [
-        "mace.h",
-    ],
-    srcs = [
-        "status.cc",
-    ],
+    name = "public_headers",
+    hdrs = glob([
+            "mace/public/*.h",
+            "mace/port/*.h",
+            "mace/utils/*.h",
+            ]),
+    strip_include_prefix = "",
    copts = ["-Werror", "-Wextra", "-Wno-missing-field-initializers"],
 )
--- a/include/mace/port/env.h
+++ b/include/mace/port/env.h
@@ -121,13 +121,17 @@ inline MaceStatus GetEnv(const char *name, std::string *value) {
  if (error != 0) {
    return MaceStatus::MACE_RUNTIME_ERROR;
  } else {
-    *value = std::string(val);
-    free(val);
+    if (val != nullptr) {
+      *value = std::string(val);
+      free(val);
+    }
    return MaceStatus::MACE_SUCCESS;
  }
 #else
  char *val = getenv(name);
-  *value = std::string(val);
+  if (val != nullptr) {
+    *value = std::string(val);
+  }
  return MaceStatus::MACE_SUCCESS;
 #endif
 }

--- a/mace/codegen/BUILD.bazel
+++ b/mace/codegen/BUILD.bazel
@@ -8,7 +8,7 @@ package(
 load("//mace:mace.bzl", "mace_version_genrule", "encrypt_opencl_kernel_genrule")

 cc_library(
-    name = "libmodels",
+    name = "generated_models",
    srcs = glob(["models/*/*.cc"]),
    hdrs = glob(["models/*/*.h"]),
    copts = ["-Werror", "-Wextra", "-Wno-missing-field-initializers"],
@@ -17,17 +17,6 @@ cc_library(
    ],
 )

-genrule(
-    name = "generated_models",
-    srcs = [
-        ":libmodels",
-    ],
-    outs = ["libgenerated_models.a"],
-    cmd = "binary=`echo $(locations :libmodels) | xargs -n 1 | grep libmodels.a`;" +
-          "$(STRIP) -x $$binary -o $@",
-    visibility = ["//visibility:public"],
-)
-
 mace_version_genrule()

 encrypt_opencl_kernel_genrule()
@@ -61,7 +50,7 @@ cc_library(
    hdrs = glob(["engine/*.h"]),
    copts = ["-Werror", "-Wextra", "-Wno-missing-field-initializers"],
    deps = [
-        "//mace/public",
+        "//include:public_headers",
    ],
 )


--- a/mace/core/BUILD.bazel
+++ b/mace/core/BUILD.bazel
-# Description:
-# Mace core.
-#
 package(
    default_visibility = ["//visibility:public"],
 )
@@ -30,9 +27,6 @@ cc_library(
            "*.cc",
            "runtime/cpu/*.cc",
        ],
-        exclude = [
-            "*_test.cc",
-        ],
    ) + if_opencl_enabled(glob(
        [
            "runtime/opencl/*.cc",
@@ -118,28 +112,3 @@ cc_library(
        "@opencl_headers//:opencl20_headers",
    ],
 )
-
-cc_library(
-    name = "test_benchmark_main",
-    testonly = 1,
-    srcs = [
-        "testing/test_benchmark.cc",
-        "testing/test_benchmark_main.cc",
-    ],
-    hdrs = [
-        "testing/test_benchmark.h",
-    ],
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ] + if_openmp_enabled(["-fopenmp"]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]),
-    deps = [
-        ":core",
-        "//external:gflags_nothreads",
-        "//mace/ops:test",
-        "//mace/utils",
-    ],
-)
--- a/mace/libmace/BUILD.bazel
+++ b/mace/libmace/BUILD.bazel
@@ -50,7 +50,7 @@ cc_library(
    ]),
    deps = [
        "//mace/ops",
-        "//mace/public",
+        "//include:public_headers",
    ],
    alwayslink = 1,
 )
@@ -91,6 +91,7 @@ cc_library(
 genrule(
    name = "libmace_static",
    srcs = [
+        "//include:public_headers",
        "//mace/codegen:generated_version",
        "//mace/core",
        "//mace/ops:common",
@@ -100,7 +101,6 @@ genrule(
        "//mace/libmace",
        "//mace/port:port_base",
        "//mace/port/posix:port_posix",
-        "//mace/public",
        "//mace/utils",
        "//mace/proto:mace_cc",
        "@com_google_protobuf//:protobuf_lite",
@@ -155,7 +155,6 @@ genrule(
              "$(locations //mace/port/darwin:port_darwin) ",
              default_value = "",
          ) +
-          "$(locations //mace/public:public) " +
          "$(locations //mace/utils:utils) " +
          "$(locations //mace/proto:mace_cc) " +
          "$(locations @com_google_protobuf//:protobuf_lite) " +

--- a/mace/ops/BUILD.bazel
+++ b/mace/ops/BUILD.bazel
@@ -52,35 +52,6 @@ cc_library(
    ],
 )

-cc_library(
-    name = "testing",
-    hdrs = [
-        "testing/test_utils.h",
-    ],
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ] + if_openmp_enabled([
-        "-fopenmp",
-    ]) + if_neon_enabled([
-        "-DMACE_ENABLE_NEON",
-    ]) + if_android_armv7([
-        "-mfpu=neon-fp16",
-        "-mfloat-abi=softfp",
-    ]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]) + if_quantize_enabled([
-        "-DMACE_ENABLE_QUANTIZE",
-    ]) + if_hexagon_enabled([
-        "-DMACE_ENABLE_HEXAGON",
-    ]),
-    deps = [
-        "//mace/core",
-        "@gtest",
-    ],
-)
-
 cc_library(
    name = "ref_kernels",
    srcs = glob(
@@ -178,11 +149,6 @@ cc_library(
        [
            "opencl/*.cc",
            "opencl/**/*.cc",
-            "buffer_transform.cc",
-            "lstm_cell.cc",
-        ],
-        exclude = [
-            "opencl/*_test.cc",
        ],
    ),
    hdrs = glob(
@@ -215,78 +181,6 @@ cc_library(
    ],
 )

-cc_library(
-    name = "arm_neon_kernels_test",
-    srcs = glob(
-        [
-            "arm/fp32/*_test.cc",
-        ],
-    ) + if_quantize_enabled(glob(
-        [
-            "arm/q8/*_test.cc",
-        ],
-    )),
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ] + if_openmp_enabled([
-        "-fopenmp",
-    ]) + if_neon_enabled([
-        "-DMACE_ENABLE_NEON",
-    ]) + if_android_armv7([
-        "-mfpu=neon-fp16",
-        "-mfloat-abi=softfp",
-    ]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]) + if_quantize_enabled([
-        "-DMACE_ENABLE_QUANTIZE",
-    ]) + if_hexagon_enabled([
-        "-DMACE_ENABLE_HEXAGON",
-    ]),
-    deps = [
-        ":arm_neon_kernels",
-        ":ref_kernels",
-        ":testing",
-        "@gtest",
-    ],
-    alwayslink = 1,
-)
-
-cc_library(
-    name = "opencl_kernels_test",
-    srcs = glob(
-        [
-            "opencl/*_test.cc",
-            "opencl/**/*_test.cc",
-        ],
-    ),
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ] + if_openmp_enabled([
-        "-fopenmp",
-    ]) + if_neon_enabled([
-        "-DMACE_ENABLE_NEON",
-    ]) + if_android_armv7([
-        "-mfpu=neon-fp16",
-        "-mfloat-abi=softfp",
-    ]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]) + if_quantize_enabled([
-        "-DMACE_ENABLE_QUANTIZE",
-    ]) + if_hexagon_enabled([
-        "-DMACE_ENABLE_HEXAGON",
-    ]),
-    deps = [
-        ":opencl_kernels",
-        ":ref_kernels",
-        ":testing",
-        "@gtest",
-    ],
-    alwayslink = 1,
-)

 cc_library(
    name = "internal_ops",
@@ -294,32 +188,12 @@ cc_library(
        [
            "*.cc",
        ],
-        exclude = [
-            "*_test.cc",
-            "*_benchmark.cc",
-            "ops_registry.cc",
-            "ops_test_util.cc",
-            "lstm_cell.cc",  # TODO: move it into opencl
-            "buffer_transform.cc",  # TODO: move it into opencl
-            "quantize.cc",
-            "quantization_util.cc",
-            "arm/*_test.cc",  # remove it after refactor
-        ],
-    ) + if_quantize_enabled(
-        glob(
-            [
-                "quantize.cc",
-                "quantization_util.cc",
-            ],
-        ),
    ),
    hdrs = glob(
        [
            "*.h",
        ],
        exclude = [
-            "ops_registry.h",
-            "ops_test_util.h",
            "fixpoint.h",
            "common/gemmlowp_util.h",
            "quantization_util.h",
@@ -363,16 +237,12 @@ cc_library(

 cc_library(
    name = "ops",
-    srcs = glob(
-        [
-            "ops_registry.cc",
+    srcs = [
+        "registry/ops_registry.cc",
        ],
-    ),
-    hdrs = glob(
-        [
-            "ops_registry.h",
+    hdrs = [
+        "registry/ops_registry.h",
        ],
-    ),
    copts = [
        "-Werror",
        "-Wextra",
@@ -396,118 +266,3 @@ cc_library(
        "internal_ops",
    ],
 )
-
-cc_library(
-    name = "test",
-    testonly = 1,
-    srcs = [
-        "ops_test_util.cc",
-    ],
-    hdrs = glob([
-        "*_test_util.h",
-    ]),
-    copts = [
-        "-Werror",
-        "-Wextra",
-    ] + if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
-        "-DMACE_ENABLE_NEON",
-    ]) + if_android_armv7([
-        "-mfpu=neon-fp16",
-    ]) + if_android_armv7([
-        "-mfloat-abi=softfp",
-    ]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]) + if_hexagon_enabled([
-        "-DMACE_ENABLE_HEXAGON",
-    ]),
-    deps = [
-        "ops",
-        "testing",
-        "@gtest",
-    ],
-)
-
-cc_test(
-    name = "ops_test",
-    testonly = 1,
-    srcs = glob(
-        [
-            "*_test.cc",
-            "arm/*_test.cc",  # remove it after refactor
-            "ops_test_util.cc",
-        ],
-        exclude = [
-            "fixpoint_test.cc",
-        ],
-    ) + if_quantize_enabled(glob(
-        [
-            "fixpoint_test.cc",
-        ],
-    )),
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ] + if_openmp_enabled([
-        "-fopenmp",
-    ]) + if_neon_enabled([
-        "-DMACE_ENABLE_NEON",
-    ]) + if_android_armv7([
-        "-mfpu=neon-fp16",
-        "-mfloat-abi=softfp",
-    ]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]) + if_quantize_enabled([
-        "-DMACE_ENABLE_QUANTIZE",
-    ]) + if_hexagon_enabled([
-        "-DMACE_ENABLE_HEXAGON",
-    ]),
-    linkopts = if_openmp_enabled([
-        "-fopenmp",
-    ]),
-    linkstatic = 1,
-    deps = [
-        ":ops",
-        ":test",
-        "@gtest//:gtest_main",
-    ] + if_neon_enabled([
-        ":arm_neon_kernels_test",
-    ]) + if_opencl_enabled([
-        ":opencl_kernels_test",
-    ]),
-)
-
-cc_test(
-    name = "ops_benchmark",
-    testonly = 1,
-    srcs = glob(["*_benchmark.cc"]),
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ] + if_openmp_enabled([
-        "-fopenmp",
-    ]) + if_neon_enabled([
-        "-DMACE_ENABLE_NEON",
-    ]) + if_android_armv7([
-        "-mfpu=neon-fp16",
-        "-mfloat-abi=softfp",
-    ]) + if_opencl_enabled([
-        "-DMACE_ENABLE_OPENCL",
-    ]) + if_quantize_enabled([
-        "-DMACE_ENABLE_QUANTIZE",
-    ]) + if_hexagon_enabled([
-        "-DMACE_ENABLE_HEXAGON",
-    ]),
-    linkopts = if_openmp_enabled([
-        "-fopenmp",
-    ]),
-    linkstatic = 1,
-    deps = [
-        ":ops",
-        "//mace/benchmark:statistics",
-        "//mace/core:test_benchmark_main",
-        "//third_party/eigen3",
-        "@gemmlowp",
-    ],
-)
--- a/mace/port/BUILD.bazel
+++ b/mace/port/BUILD.bazel
@@ -13,42 +13,15 @@ cc_library(
    ],
 )

-cc_library(
-    name = "port_api",
-    hdrs = [
-        "env.h",
-        "file_system.h",
-        "logger.h",
-        "port.h",
-        "port-arch.h",
-    ],
-    deps = [
-        "//mace/public",
-    ],
-)
-
 cc_library(
    name = "port_base",
    srcs = [
        "env.cc",
        "logger.cc",
+        "file_system.cc",
    ],
    deps = [
-        ":port_api",
        "//mace/utils",
-    ],
-)
-
-cc_test(
-    name = "port_test",
-    testonly = 1,
-    srcs = glob([
-        "*_test.cc",
-    ]),
-    linkstatic = 1,
-    deps = [
-        ":port",
-        "@gtest",
-        "@gtest//:gtest_main",
+        "//include:public_headers",
    ],
 )
--- a/mace/python/tools/visualization/visualize_model.py
+++ b/mace/python/tools/visualization/visualize_model.py
@@ -20,7 +20,7 @@ class NPEncoder(json.JSONEncoder):

 class ModelVisualizer(object):
    def __init__(self, model_name, proto):
-        self._output_file = "builds/%s_index.html" % model_name
+        self._output_file = "build/%s_index.html" % model_name
        self._proto = proto

    def render_html(self):

--- a/mace/utils/BUILD.bazel
+++ b/mace/utils/BUILD.bazel
@@ -26,8 +26,7 @@ cc_library(
        "-Wno-missing-field-initializers",
    ],
    deps = [
-        "//mace/port:port_api",
-        "//mace/public",
+        "//include:public_headers",
    ],
 )

@@ -37,9 +36,6 @@ cc_library(
        [
            "*.cc",
        ],
-        exclude = [
-            "*_test.cc",
-        ],
    ),
    copts = [
        "-Werror",
@@ -61,25 +57,3 @@ cc_library(
    ],
    alwayslink = 1,
 )
-
-cc_test(
-    name = "utils_test",
-    testonly = 1,
-    srcs = glob(
-        [
-            "*_test.cc",
-        ],
-    ),
-    copts = [
-        "-Werror",
-        "-Wextra",
-        "-Wno-missing-field-initializers",
-    ],
-    linkstatic = 1,
-    deps = [
-        ":utils",
-        "//mace/port",
-        "@gtest//:gtest",
-        "@gtest//:gtest_main",
-    ],
-)
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -6,6 +6,6 @@ if(MACE_ENABLE_TESTS)
  add_subdirectory(ccunit)
 endif()

-if(MACE_ENABLE_BENCHMARK)
+if(MACE_ENABLE_BENCHMARKS)
  add_subdirectory(ccbenchmark)
 endif()
--- a/test/ccbenchmark/BUILD.bazel
+++ b/test/ccbenchmark/BUILD.bazel
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+licenses(["notice"])  # Apache 2.0
+
+load(
+    "//mace:mace.bzl",
+    "if_android",
+    "if_android_armv7",
+    "if_hexagon_enabled",
+    "if_neon_enabled",
+    "if_opencl_enabled",
+    "if_openmp_enabled",
+    "if_quantize_enabled",
+)
+
+cc_library(
+    name = "benchmark_utils",
+    testonly = 1,
+    hdrs = glob([
+        "mace/benchmark_utils/*.h",
+    ]),
+    srcs = glob([
+        "mace/benchmark_utils/*.cc",
+    ]),
+    copts = [
+        "-Werror",
+        "-Wextra",
+        "-Wno-missing-field-initializers",
+    ],
+    strip_include_prefix = "",
+    deps = [
+        "//mace/core",
+        "//test/ccutils",
+        "//external:gflags_nothreads",
+    ],
+)
+
+cc_test(
+    name = "mace_cc_benchmark",
+    testonly = 1,
+    srcs = glob(
+        [
+            "mace/ops/*.cc",
+        ],
+    ),
+    copts = [
+        "-Werror",
+        "-Wextra",
+        "-Wno-missing-field-initializers",
+        "-fopenmp",
+    ] + if_neon_enabled([
+        "-DMACE_ENABLE_NEON",
+    ]) + if_android_armv7([
+        "-mfpu=neon-fp16",
+        "-mfloat-abi=softfp",
+    ]) + if_opencl_enabled([
+        "-DMACE_ENABLE_OPENCL",
+    ]) + if_quantize_enabled([
+        "-DMACE_ENABLE_QUANTIZE",
+    ]) + if_hexagon_enabled([
+        "-DMACE_ENABLE_HEXAGON",
+    ]),
+    linkopts = [
+        "-fopenmp",
+    ],
+    linkstatic = 1,
+    deps = [
+        "benchmark_utils",
+        "//mace/ops",
+        "//third_party/eigen3",
+        "@gemmlowp",
+    ],
+)
--- a/test/ccunit/BUILD.bazel
+++ b/test/ccunit/BUILD.bazel
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+licenses(["notice"])  # Apache 2.0
+
+load(
+    "//mace:mace.bzl",
+    "if_android",
+    "if_android_armv7",
+    "if_hexagon_enabled",
+    "if_neon_enabled",
+    "if_opencl_enabled",
+    "if_openmp_enabled",
+    "if_quantize_enabled",
+)
+
+cc_test(
+    name = "mace_cc_test",
+    testonly = 1,
+    srcs = glob(
+        [
+            "mace/libmace/*.cc",
+            "mace/ops/*.cc",
+            "mace/port/*.cc",
+            "mace/utils/*.cc",
+        ],
+        exclude = [
+            "mace/ops/fixpoint_test.cc",
+        ],
+    ) + if_neon_enabled(glob(
+        [
+            "mace/ops/arm/fp32/*.cc",
+        ]
+    )) + if_quantize_enabled(glob(
+        [
+            "mace/ops/arm/q8/*.cc",
+            "mace/ops/fixpoint_test.cc",
+        ]
+    )) + if_opencl_enabled(glob(
+        [
+            "mace/ops/opencl/*.cc",
+        ]
+    )),
+    copts = [
+        "-Werror",
+        "-Wextra",
+        "-Wno-missing-field-initializers",
+    ] + if_openmp_enabled([
+        "-fopenmp",
+    ]) + if_neon_enabled([
+        "-DMACE_ENABLE_NEON",
+    ]) + if_android_armv7([
+        "-mfpu=neon-fp16",
+        "-mfloat-abi=softfp",
+    ]) + if_opencl_enabled([
+        "-DMACE_ENABLE_OPENCL",
+    ]) + if_quantize_enabled([
+        "-DMACE_ENABLE_QUANTIZE",
+    ]) + if_hexagon_enabled([
+        "-DMACE_ENABLE_HEXAGON",
+    ]),
+    linkopts = if_openmp_enabled([
+        "-fopenmp",
+    ]),
+    linkstatic = 1,
+    deps = [
+        "//mace/ops",
+        "//test/ccutils",
+        "@gtest//:gtest_main",
+    ],
+)
--- a/test/ccutils/BUILD.bazel
+++ b/test/ccutils/BUILD.bazel
+package(
+    default_visibility = ["//visibility:public"],
+)
+
+licenses(["notice"])  # Apache 2.0
+
+load(
+    "//mace:mace.bzl",
+    "if_android",
+    "if_android_armv7",
+    "if_hexagon_enabled",
+    "if_neon_enabled",
+    "if_opencl_enabled",
+    "if_openmp_enabled",
+    "if_quantize_enabled",
+)
+
+cc_library(
+    name = "ccutils",
+    testonly = 1,
+    srcs = glob([
+        "mace/ops/*.cc",
+    ]),
+    hdrs = glob([
+        "mace/ops/*.h",
+        "mace/ops/testing/*.h",
+    ]),
+    strip_include_prefix = "",
+    copts = [
+        "-Werror",
+        "-Wextra",
+    ] + if_openmp_enabled(["-fopenmp"]) + if_neon_enabled([
+        "-DMACE_ENABLE_NEON",
+    ]) + if_android_armv7([
+        "-mfpu=neon-fp16",
+    ]) + if_android_armv7([
+        "-mfloat-abi=softfp",
+    ]) + if_opencl_enabled([
+        "-DMACE_ENABLE_OPENCL",
+    ]) + if_hexagon_enabled([
+        "-DMACE_ENABLE_HEXAGON",
+    ]),
+    deps = [
+        "//mace/ops",
+        "@gtest",
+    ],
+)
--- a/third_party/opencl-clhpp/opencl-clhpp.BUILD
+++ b/third_party/opencl-clhpp/opencl-clhpp.BUILD
@@ -15,5 +15,6 @@ genrule(
 cc_library(
    name = "opencl_clhpp",
    hdrs = ["include/CL/cl.hpp", "include/CL/cl2.hpp"],
+    strip_include_prefix = "include",
    visibility = ["//visibility:public"],
 )
--- a/third_party/protobuf/protobuf.cmake
+++ b/third_party/protobuf/protobuf.cmake
@@ -56,53 +56,65 @@ add_dependencies(libprotobuf_lite protobuf)

 install(FILES ${PROTOBUF_LITE_LIBRARIES} DESTINATION lib)

-if(NOT APPLE)
-  # Actually this works for iOS build on macOS, but the compatibility is not 
-  # thoroughly tested, so we use the downloaded version instead.
-  set(PROTOC_CMAKE_GENERATOR ${CMAKE_GENERATOR})
-  if(APPLE AND ${CMAKE_GENERATOR} STREQUAL "Xcode")
-    # Force Xcode to build protoc on host
-    set(PROTOC_CMAKE_GENERATOR "Unix Makefiles")
-    execute_process(COMMAND sw_vers -productVersion OUTPUT_VARIABLE MACOS_PRODVER)
-    set(EXTRA_MACOS_CMAKE_ARGS
-      -DCMAKE_OSX_SYSROOT=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk
-      -DCMAKE_OSX_DEPLOYMENT_TARGET=${MACOS_PRODVER}
-    )
-  endif(APPLE AND ${CMAKE_GENERATOR} STREQUAL "Xcode")
+set(BUILD_PROTOC TRUE)
+if(COMMAND protoc)
+  execute_process(COMMAND protoc OUTPUT_VARIABLE PROTOC_VER)
+  if(${PROTOC_VER} STREQUAL "libprotoc 3.6.1")
+    set(PROTOC_BIN protoc CACHE FILEPATH "protoc compiler." FORCE)
+    set(BUILD_PROTOC FALSE)
+    add_custom_target(protoc_bin COMMENT "protoc noop target")
+  endif(${PROTOC_VER} STREQUAL "libprotoc 3.6.1")
+endif(COMMAND protoc)
+
+if(BUILD_PROTOC)
+  if(NOT APPLE)
+    # Actually this works for iOS build on macOS, but the compatibility is not
+    # thoroughly tested, so we use the downloaded version instead.
+    set(PROTOC_CMAKE_GENERATOR ${CMAKE_GENERATOR})
+    if(APPLE AND ${CMAKE_GENERATOR} STREQUAL "Xcode")
+      # Force Xcode to build protoc on host
+      set(PROTOC_CMAKE_GENERATOR "Unix Makefiles")
+      execute_process(COMMAND sw_vers -productVersion OUTPUT_VARIABLE MACOS_PRODVER)
+      set(EXTRA_MACOS_CMAKE_ARGS
+        -DCMAKE_OSX_SYSROOT=/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk
+        -DCMAKE_OSX_DEPLOYMENT_TARGET=${MACOS_PRODVER}
+      )
+    endif(APPLE AND ${CMAKE_GENERATOR} STREQUAL "Xcode")

-  ExternalProject_Add(
-    protoc_bin
-    URL_HASH          "${PROTOBUF_HASH}"
-    URL               "${PROTOBUF_URL}"
-    PREFIX            "${PROTOC_SRCS_DIR}"
-    BUILD_BYPRODUCTS  "${PROTOC_BIN}"
-    SOURCE_DIR        "${PROTOC_SRCS_DIR}/src/protoc"
-    CONFIGURE_COMMAND ${CMAKE_COMMAND} ../protoc/cmake/
-                      -DCMAKE_BUILD_TYPE=Release
-                      -DCMAKE_INSTALL_PREFIX=${PROTOC_INSTALL_DIR}
-                      -DCMAKE_VERBOSE_MAKEFILE=OFF
-                      -Dprotobuf_BUILD_TESTS=OFF
-                      -Dprotobuf_WITH_ZLIB=OFF
-                      -Dprotobuf_BUILD_PROTOC_BINARIES=ON
-                      -DCMAKE_GENERATOR=${PROTOC_CMAKE_GENERATOR}
-                      ${EXTRA_MACOS_CMAKE_ARGS}
+    ExternalProject_Add(
+      protoc_bin
+      URL_HASH          "${PROTOBUF_HASH}"
+      URL               "${PROTOBUF_URL}"
+      PREFIX            "${PROTOC_SRCS_DIR}"
+      BUILD_BYPRODUCTS  "${PROTOC_BIN}"
+      SOURCE_DIR        "${PROTOC_SRCS_DIR}/src/protoc"
+      CONFIGURE_COMMAND ${CMAKE_COMMAND} ../protoc/cmake/
+      -DCMAKE_BUILD_TYPE=Release
+      -DCMAKE_INSTALL_PREFIX=${PROTOC_INSTALL_DIR}
+      -DCMAKE_VERBOSE_MAKEFILE=OFF
+      -Dprotobuf_BUILD_TESTS=OFF
+      -Dprotobuf_WITH_ZLIB=OFF
+      -Dprotobuf_BUILD_PROTOC_BINARIES=ON
+      -DCMAKE_GENERATOR=${PROTOC_CMAKE_GENERATOR}
+      ${EXTRA_MACOS_CMAKE_ARGS}
    )
-else(APPLE)
-  # This is backup protoc when build doesn't work for macOS+iOS
-  # Mirror of "https://github.com/protocolbuffers/protobuf/releases/download/v3.6.1/protoc-3.6.1-osx-x86_64.zip"
-  set(PROTOC_URL  "https://cnbj1.fds.api.xiaomi.com/mace/third-party/protobuf/protoc-3.6.1-osx-x86_64.zip")
-  set(PROTOC_HASH "SHA256=0decc6ce5beed07f8c20361ddeb5ac7666f09cf34572cca530e16814093f9c0c")
-  ExternalProject_Add(
-    protoc_bin
-    URL_HASH          "${PROTOC_HASH}"
-    URL               "${PROTOC_URL}"
-    PREFIX            "${PROTOC_SRCS_DIR}"
-    BINARY_DIR        ""
-    CONFIGURE_COMMAND ""
-    BUILD_COMMAND     ""
-    INSTALL_COMMAND   ""
-    TEST_COMMAND      ""
-  )
-  set(PROTOC_BIN
-    "${PROTOC_SRCS_DIR}/src/protoc_bin/bin/protoc" CACHE FILEPATH "protoc compiler." FORCE)
-endif(NOT APPLE)
+  else(APPLE)
+    # This is backup protoc when build doesn't work for macOS+iOS
+    # Mirror of "https://github.com/protocolbuffers/protobuf/releases/download/v3.6.1/protoc-3.6.1-osx-x86_64.zip"
+    set(PROTOC_URL  "https://cnbj1.fds.api.xiaomi.com/mace/third-party/protobuf/protoc-3.6.1-osx-x86_64.zip")
+    set(PROTOC_HASH "SHA256=0decc6ce5beed07f8c20361ddeb5ac7666f09cf34572cca530e16814093f9c0c")
+    ExternalProject_Add(
+      protoc_bin
+      URL_HASH          "${PROTOC_HASH}"
+      URL               "${PROTOC_URL}"
+      PREFIX            "${PROTOC_SRCS_DIR}"
+      BINARY_DIR        ""
+      CONFIGURE_COMMAND ""
+      BUILD_COMMAND     ""
+      INSTALL_COMMAND   ""
+      TEST_COMMAND      ""
+    )
+    set(PROTOC_BIN
+      "${PROTOC_SRCS_DIR}/src/protoc_bin/bin/protoc" CACHE FILEPATH "protoc compiler." FORCE)
+  endif(NOT APPLE)
+endif(BUILD_PROTOC)
--- a/tools/bazel-build-standalone-lib.sh
+++ b/tools/bazel-build-standalone-lib.sh
+#!/bin/bash
+
+set -e
+
+LIB_DIR=build/lib
+INCLUDE_DIR=build/include
+
+mkdir -p $LIB_DIR
+mkdir -p $INCLUDE_DIR
+
+# copy include headers
+cp -R include/mace $INCLUDE_DIR/
+
+# make directories
+rm -rf $LIB_DIR/armeabi-v7a
+mkdir -p $LIB_DIR/armeabi-v7a/cpu_gpu_dsp
+mkdir -p $LIB_DIR/armeabi-v7a/cpu_gpu
+
+rm -rf $LIB_DIR/arm64-v8a
+mkdir -p $LIB_DIR/arm64-v8a/cpu_gpu_dsp
+mkdir -p $LIB_DIR/arm64-v8a/cpu_gpu
+
+rm -rf $LIB_DIR/linux-x86-64
+mkdir -p $LIB_DIR/linux-x86-64
+
+rm -rf $LIB_DIR/arm_linux_gnueabihf
+mkdir -p $LIB_DIR/arm_linux_gnueabihf/cpu_gpu
+
+rm -rf $LIB_DIR/aarch64_linux_gnu
+mkdir -p $LIB_DIR/aarch64_linux_gnu/cpu_gpu
+
+
+
+# build shared libraries
+echo "build shared lib for armeabi-v7a + cpu_gpu_dsp"
+bazel build --config android --config optimization mace/libmace:libmace_dynamic --define neon=true --define openmp=false --define opencl=true --define hexagon=true --define quantize=true --cpu=armeabi-v7a
+cp bazel-bin/mace/libmace/libmace.so $LIB_DIR/armeabi-v7a/cpu_gpu_dsp/
+cp third_party/nnlib/armeabi-v7a/*so $LIB_DIR/armeabi-v7a/cpu_gpu_dsp/
+
+echo "build shared lib for arm64-v8a + cpu_gpu_dsp"
+bazel build --config android --config optimization mace/libmace:libmace_dynamic --define neon=true --define openmp=false --define opencl=true --define hexagon=true --define quantize=true --cpu=arm64-v8a
+cp bazel-bin/mace/libmace/libmace.so $LIB_DIR/arm64-v8a/cpu_gpu_dsp/
+cp third_party/nnlib/arm64-v8a/*so $LIB_DIR/arm64-v8a/cpu_gpu_dsp/
+
+echo "build shared lib for armeabi-v7a + cpu_gpu"
+bazel build --config android --config optimization mace/libmace:libmace_dynamic --define neon=true --define openmp=false --define opencl=true --define quantize=true --cpu=armeabi-v7a
+cp bazel-bin/mace/libmace/libmace.so $LIB_DIR/armeabi-v7a/cpu_gpu/
+
+echo "build shared lib for arm64-v8a + cpu_gpu"
+bazel build --config android --config optimization mace/libmace:libmace_dynamic --define neon=true --define openmp=false --define opencl=true --define quantize=true --cpu=arm64-v8a
+cp bazel-bin/mace/libmace/libmace.so $LIB_DIR/arm64-v8a/cpu_gpu/
+
+echo "build shared lib for arm_linux_gnueabihf + cpu_gpu"
+bazel build --config arm_linux_gnueabihf --config optimization mace/libmace:libmace_dynamic --define neon=true --define openmp=false --define opencl=true --define quantize=true
+cp bazel-bin/mace/libmace/libmace.so  $LIB_DIR/arm_linux_gnueabihf/cpu_gpu/
+
+echo "build shared lib for aarch64_linux_gnu + cpu_gpu"
+bazel build --config aarch64_linux_gnu  --config optimization mace/libmace:libmace_dynamic  --define neon=true --define openmp=false --define opencl=true --define quantize=true
+cp bazel-bin/mace/libmace/libmace.so  $LIB_DIR/aarch64_linux_gnu/cpu_gpu/
+
+if [[ "$OSTYPE" != "darwin"* ]];then
+	echo "build shared lib for linux-x86-64"
+	bazel build mace/libmace:libmace_dynamic --config optimization --define openmp=false
+	cp bazel-bin/mace/libmace/libmace.so $LIB_DIR/linux-x86-64/
+fi
+
+# build static libraries
+echo "build static lib for armeabi-v7a + cpu_gpu_dsp"
+bazel build --config android --config optimization mace/libmace:libmace_static --config symbol_hidden --define neon=true --define openmp=false --define opencl=true --define hexagon=true --define quantize=true --cpu=armeabi-v7a
+cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/armeabi-v7a/cpu_gpu_dsp/
+cp third_party/nnlib/armeabi-v7a/*so $LIB_DIR/armeabi-v7a/cpu_gpu_dsp/
+
+echo "build static lib for arm64-v8a + cpu_gpu_dsp"
+bazel build --config android --config optimization mace/libmace:libmace_static --config symbol_hidden --define neon=true --define openmp=false --define opencl=true --define hexagon=true --define quantize=true --cpu=arm64-v8a
+cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/arm64-v8a/cpu_gpu_dsp/
+cp third_party/nnlib/arm64-v8a/*so $LIB_DIR/arm64-v8a/cpu_gpu_dsp/
+
+echo "build static lib for armeabi-v7a + cpu_gpu"
+bazel build --config android --config optimization mace/libmace:libmace_static --config symbol_hidden --define neon=true --define openmp=false --define opencl=true --define quantize=true --cpu=armeabi-v7a
+cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/armeabi-v7a/cpu_gpu/
+
+echo "build static lib for arm64-v8a + cpu_gpu"
+bazel build --config android --config optimization mace/libmace:libmace_static --config symbol_hidden --define neon=true --define openmp=false --define opencl=true --define quantize=true --cpu=arm64-v8a
+cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/arm64-v8a/cpu_gpu/
+
+echo "build static lib for arm_linux_gnueabihf + cpu_gpu"
+bazel build --config arm_linux_gnueabihf --config optimization mace/libmace:libmace_static --config symbol_hidden --define neon=true --define openmp=false --define opencl=true --define quantize=true
+cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/arm_linux_gnueabihf/cpu_gpu/
+
+echo "build static lib for aarch64_linux_gnu + cpu_gpu"
+bazel build --config aarch64_linux_gnu --config optimization mace/libmace:libmace_static --config symbol_hidden --define neon=true --define openmp=false --define opencl=true --define quantize=true
+cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/aarch64_linux_gnu/cpu_gpu/
+
+if [[ "$OSTYPE" != "darwin"* ]];then
+	echo "build static lib for linux-x86-64"
+	bazel build mace/libmace:libmace_static --config optimization --define openmp=false
+	cp bazel-genfiles/mace/libmace/libmace.a $LIB_DIR/linux-x86-64/
+fi
+
+echo "LIB PATH: $LIB_DIR"
+echo "INCLUDE FILE PATH: $INCLUDE_DIR"
--- a/tools/bazel_adb_run.py
+++ b/tools/bazel_adb_run.py
@@ -16,7 +16,7 @@
 # python tools/bazel_adb_run.py \
 #     --target_abis=armeabi-v7a \
 #     --target_socs=sdm845
-#     --target=//mace/ops:ops_test
+#     --target=//test/ccunit:mace_cc_test
 #     --stdout_processor=stdout_processor

 import argparse
@@ -43,22 +43,6 @@ def ops_benchmark_stdout_processor(stdout, dev, abi):
    for line in stdout_lines:
        if "Aborted" in line or "Segmentation fault" in line:
            raise Exception("Command failed")
-        line = line.strip()
-        parts = line.split()
-        if len(parts) == 5 and parts[0].startswith("BM_"):
-            metrics["%s.time_ms" % parts[0]] = str(float(parts[1]) / 1e6)
-            metrics["%s.input_mb_per_sec" % parts[0]] = parts[3]
-            metrics["%s.gmac_per_sec" % parts[0]] = parts[4]
-
-    # platform = dev[YAMLKeyword.target_socs]
-    # model = dev[YAMLKeyword.device_name]
-    # tags = {
-    #     "ro.board.platform": platform,
-    #     "ro.product.model": model,
-    #     "abi": abi
-    # }
-    # sh_commands.falcon_push_metrics(server,
-    #    metrics, tags=tags, endpoint="mace_ops_benchmark")


 # TODO: after merge mace/python/tools and tools are merged,

--- a/tools/build-aarch64-linux-gnu-full.sh
+++ b/tools/build-aarch64-linux-gnu-full.sh
@@ -3,7 +3,7 @@
 set -e

 # build for arm linux aarch64
-BUILD_DIR=build/aarch64-linux-gnu-full
+BUILD_DIR=cmake-build/aarch64-linux-gnu-full
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DCROSSTOOL_ROOT=${LINARO_AARCH64_LINUX_GNU} \
      -DCMAKE_TOOLCHAIN_FILE=../../cmake/toolchains/aarch64-linux-gnu.cmake \
@@ -15,5 +15,5 @@ cmake -DCROSSTOOL_ROOT=${LINARO_AARCH64_LINUX_GNU} \
      -DMACE_ENABLE_OBFUSCATE=ON          \
      -DCMAKE_INSTALL_PREFIX=install      \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-android-arm64-v8a-cpu.sh
+++ b/tools/build-android-arm64-v8a-cpu.sh
@@ -3,7 +3,7 @@
 set -e

 # build for android arm64-v8a
-BUILD_DIR=build/android-arm64-v8a-cpu
+BUILD_DIR=cmake-build/android-arm64-v8a-cpu
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DANDROID_ABI="arm64-v8a" \
      -DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_HOME}/build/cmake/android.toolchain.cmake \
@@ -19,5 +19,5 @@ cmake -DANDROID_ABI="arm64-v8a" \
      -DMACE_ENABLE_OBFUSCATE=ON          \
      -DCMAKE_INSTALL_PREFIX=install      \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-android-arm64-v8a-full.sh
+++ b/tools/build-android-arm64-v8a-full.sh
@@ -3,7 +3,7 @@
 set -e

 # build for android arm64-v8a
-BUILD_DIR=build/android-arm64-v8a-full
+BUILD_DIR=cmake-build/android-arm64-v8a-full
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DANDROID_ABI="arm64-v8a" \
      -DCMAKE_TOOLCHAIN_FILE=${ANDROID_NDK_HOME}/build/cmake/android.toolchain.cmake \
@@ -19,5 +19,5 @@ cmake -DANDROID_ABI="arm64-v8a" \
      -DMACE_ENABLE_OBFUSCATE=ON          \
      -DCMAKE_INSTALL_PREFIX=install      \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-android-armeabi-v7a-cpu.sh
+++ b/tools/build-android-armeabi-v7a-cpu.sh
@@ -3,7 +3,7 @@
 set -e

 # build for android armeabi-v7a
-BUILD_DIR=build/android-armeabi-v7a-cpu
+BUILD_DIR=cmake-build/android-armeabi-v7a-cpu
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DANDROID_ABI="armeabi-v7a" \
      -DANDROID_ARM_NEON=ON \
@@ -20,5 +20,5 @@ cmake -DANDROID_ABI="armeabi-v7a" \
      -DMACE_ENABLE_OBFUSCATE=ON          \
      -DCMAKE_INSTALL_PREFIX=install      \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-android-armeabi-v7a-full.sh
+++ b/tools/build-android-armeabi-v7a-full.sh
@@ -3,7 +3,7 @@
 set -e

 # build for android armeabi-v7a
-BUILD_DIR=build/android-armeabi-v7a-full
+BUILD_DIR=cmake-build/android-armeabi-v7a-full
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DANDROID_ABI="armeabi-v7a" \
      -DANDROID_ARM_NEON=ON \
@@ -20,5 +20,5 @@ cmake -DANDROID_ABI="armeabi-v7a" \
      -DMACE_ENABLE_OBFUSCATE=ON          \
      -DCMAKE_INSTALL_PREFIX=install      \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-arm-linux-gnueabihf-full.sh
+++ b/tools/build-arm-linux-gnueabihf-full.sh
@@ -3,7 +3,7 @@
 set -e

 # build for arm linux gnueabihf
-BUILD_DIR=build/arm-linux-gnueabihf-full
+BUILD_DIR=cmake-build/arm-linux-gnueabihf-full
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DCROSSTOOL_ROOT=${LINARO_ARM_LINUX_GNUEABIHF} \
      -DCMAKE_TOOLCHAIN_FILE=cmake/toolchains/arm-linux-gnueabihf.cmake \
@@ -15,5 +15,5 @@ cmake -DCROSSTOOL_ROOT=${LINARO_ARM_LINUX_GNUEABIHF} \
      -DMACE_ENABLE_OBFUSCATE=ON          \
      -DCMAKE_INSTALL_PREFIX=install      \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-host.sh
+++ b/tools/build-host.sh
@@ -3,12 +3,12 @@
 set -e

 # build for host
-BUILD_DIR=build/host
+BUILD_DIR=cmake-build/host
 rm -rf ${BUILD_DIR} && mkdir -p ${BUILD_DIR} && cd ${BUILD_DIR}
 cmake -DMACE_ENABLE_NEON=OFF         \
      -DMACE_ENABLE_QUANTIZE=OFF     \
      -DMACE_ENABLE_OPENCL=ON        \
      -DCMAKE_INSTALL_PREFIX=install \
      ../..
-make -j8 VERBOSE=1 && make install
+make -j6 VERBOSE=1 && make install
 cd ../..
--- a/tools/build-standalone-lib.sh
+++ b/tools/build-standalone-lib.sh
--- a/tools/common.py
+++ b/tools/common.py
@@ -438,7 +438,7 @@ class SystemType:
 PHONE_DATA_DIR = '/data/local/tmp/mace_run'
 DEVICE_DATA_DIR = '/tmp/data/mace_run'
 DEVICE_INTERIOR_DIR = PHONE_DATA_DIR + "/interior"
-BUILD_OUTPUT_DIR = 'builds'
+BUILD_OUTPUT_DIR = 'build'
 BUILD_TMP_DIR_NAME = '_tmp'
 BUILD_DOWNLOADS_DIR = BUILD_OUTPUT_DIR + '/downloads'
 BUILD_TMP_GENERAL_OUTPUT_DIR_NAME = 'general'
@@ -469,7 +469,7 @@ LIBMACE_SO_TARGET = "//mace/libmace:libmace.so"
 LIBMACE_STATIC_TARGET = "//mace/libmace:libmace_static"
 LIBMACE_STATIC_PATH = "bazel-genfiles/mace/libmace/libmace.a"
 MODEL_LIB_TARGET = "//mace/codegen:generated_models"
-MODEL_LIB_PATH = "bazel-genfiles/mace/codegen/libgenerated_models.a"
+MODEL_LIB_PATH = "bazel-bin/mace/codegen/libgenerated_models.a"
 QUANTIZE_STAT_TARGET = "//mace/tools/quantization:quantize_stat"
 BM_MODEL_STATIC_NAME = "benchmark_model_static"
 BM_MODEL_DYNAMIC_NAME = "benchmark_model_dynamic"