Arm NN 的交叉編譯筆記

---

目錄

---

框架概念 (Concept of Arm NN ML Software)

• 支援 Post-Quantization 與 QAT Model

  • armnn Quantizer

• 格式轉換 (convert to Arm NN format)

  • armnn Converter

  • armnn Serializer

交叉編譯 (Cross Compilation)

在ubuntu常會使用 gcc 或 g++ 指令編譯 c 程式，但是對於嵌入式系統或是開發板而言，並不能直接使用這兩個編譯器生成的執行檔，這是因為底層硬體架構不同，支援的機器語言基本上也會不相同。嵌入式系統或開發板的開發商，沒意外都會提供一套工具鏈 (Toolchain)，用來編譯生成適用於該平台的執行檔。而這樣子的工具鏈其實也是由一般常見的 gcc 或 g++ 指令編譯生成，所以交叉編譯器 (Cross Compiler) 顧名思義是運行在平台A，但它編譯出的程式卻是運行於平台B。

簡單而言，目標平台 (Target Plateform) 如果跟編譯器運行的平台不一樣，這樣就是交叉編譯 (Cross Compilation)。

Arm NN的交叉編譯筆記 (Note for Cross Compilation on Arm NN)

• 如果交叉編譯的平台不是特別平台，不用自行替換toolchain的話，可以參考官方dockerfile

  FROM ubuntu:18.04
  ENV TERM linux
  ENV DEBIAN_FRONTEND noninteractive
  
  # Forward system proxy setting
  # ARG proxy
  # ENV http_proxy $proxy
  # ENV https_proxy $proxy
  
  # Basic apt update
  RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends locales ca-certificates &&  rm -rf /var/lib/apt/lists/*
  
  # Set the locale to en_US.UTF-8, because the Yocto build fails without any locale set.
  RUN locale-gen en_US.UTF-8 && update-locale LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
  ENV LANG en_US.UTF-8
  ENV LC_ALL en_US.UTF-8
  
  # Again, off the certificare
  RUN echo "check_certificate = off" >> ~/.wgetrc
  RUN echo "[global] \n\
  trusted-host = pypi.python.org \n \
  \t               pypi.org \n \
  \t              files.pythonhosted.org" >> /etc/pip.conf
  
  # Get basic packages
  RUN apt-get update && apt-get install -y \
      apparmor \
      aufs-tools \
      automake \
      bash-completion \
      btrfs-tools \
      build-essential \
      cmake \
      createrepo \
      curl \
      dpkg-sig \
      g++ \
      gcc \
      git \
      iptables \
      jq \
      libapparmor-dev \
      libc6-dev \
      libcap-dev \
      libsystemd-dev \
      libyaml-dev \
      mercurial \
      net-tools \
      parallel \
      pkg-config \
      python-dev \
      python-mock \
      python-pip \
      python-setuptools \
      python-websocket \
      golang-go \
      iproute2 \
      iputils-ping \
      vim-common \
      vim \
      wget \
      libtool \
      unzip \
      scons \
      curl \
      autoconf \
      libtool \
      build-essential \
      libssl-dev \
      g++ && rm -rf /var/lib/apt/lists/*
  
  # Install Cross-compiling ToolChain
  RUN apt-get update && apt-get install -y crossbuild-essential-arm64 crossbuild-essential-armhf
  
  # Install Cmake 3.19
  RUN cd $HOME && \
      wget -O cmake-3.19.0.tar.gz https://cmake.org/files/v3.19/cmake-3.19.0.tar.gz && \
      tar -xzf cmake-3.19.0.tar.gz && \
      cd $HOME/cmake-3.19.0 && \
      ./bootstrap && \
      make && \
      make install
  
  # Build and install Google's Protobuf library
  # Download and Extract
  RUN mkdir -p $HOME/google && \
      cd $HOME/google && \
      wget https://github.com/protocolbuffers/protobuf/releases/download/v3.12.0/protobuf-all-3.12.0.tar.gz && \
      tar -zxvf protobuf-all-3.12.0.tar.gz
  
  # Build a native (x86_64) version
  RUN cd $HOME/google/protobuf-3.12.0 && \
      mkdir x86_build && cd x86_build && \
      ../configure --prefix=$HOME/armnn-devenv/google/x86_64_pb_install && \
      make install -j$(nproc)
  
  # Build the arm64 version of the protobuf libraries
  RUN cd $HOME/google/protobuf-3.12.0 && \
      mkdir arm64_build && cd arm64_build && \
      export CC=aarch64-linux-gnu-gcc && \
      export CXX=aarch64-linux-gnu-g++ && \
      ../configure --host=aarch64-linux \
      --prefix=$HOME/armnn-devenv/google/arm64_pb_install \
      --with-protoc=$HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc && \
      make install -j$(nproc)
  
  # Build the arm32 version of the protobuf libraries
  RUN cd $HOME/google/protobuf-3.12.0 && \
      mkdir arm32_build && cd arm32_build && \
      export CC=arm-linux-gnueabihf-gcc && \
      export CXX=arm-linux-gnueabihf-g++ && \
      ../configure --host=arm-linux \
      --prefix=$HOME/armnn-devenv/google/arm32_pb_install \
      --with-protoc=$HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc && \
      make install -j$(nproc)
  
  # Dep Error - Bug ARMNN
  RUN apt-get update && apt-get install -y \
       python-numpy
  
  # Setup Env
  # ENV PATH=$HOME/armnn-devenv/google/x86_64_pb_install/bin/:$PATH
  # ENV LD_LIBRARY_PATH=$HOME/armnn-devenv/google/x86_64_pb_install/lib/:$LD_LIBRARY_PATH
  
  # Download ArmNN 64-bit
  RUN cd $HOME/armnn-devenv && git clone "https://review.mlplatform.org/ml/armnn" && \
      cd armnn/ && git checkout master
  
  # Download ArmNN 32-bit
  RUN cd $HOME/armnn-devenv && git clone "https://review.mlplatform.org/ml/armnn" armnn-32 && \
      cd armnn-32/ && git checkout master
  
  # Build 64-bit Compute Library
  RUN cd $HOME/armnn-devenv/ && git clone https://review.mlplatform.org/ml/ComputeLibrary && \
      cd ComputeLibrary && \
      git checkout $($HOME/armnn-devenv/armnn/scripts/get_compute_library.sh -p) && \
      scons Werror=0 arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" -j$(nproc)
  
  # Build 32-bit Compute Library
  RUN cd $HOME/armnn-devenv/ && mkdir ComputeLibrary_32 && git clone https://review.mlplatform.org/ml/ComputeLibrary ComputeLibrary_32 && \
      cd ComputeLibrary_32 &&\
      git checkout $($HOME/armnn-devenv/armnn/scripts/get_compute_library.sh -p) && \
      scons Werror=0 arch=armv7a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" -j$(nproc)
  
  # Download Tensorflow (Checkout latest tested version of TF using get_tensorflow.sh)
  RUN cd $HOME/armnn-devenv && \
      git clone https://github.com/tensorflow/tensorflow.git && \
      cd tensorflow && \
      git checkout $($HOME/armnn-devenv/armnn/scripts/get_tensorflow.sh -p)
  
  # Build 64-bit TF Lite
  RUN cd $HOME/armnn-devenv && \
      curl -LO https://storage.googleapis.com/mirror.tensorflow.org/developer.arm.com/media/Files/downloads/gnu-a/8.3-2019.03/binrel/gcc-arm-8.3-2019.03-x86_64-aarch64-linux-gnu.tar.xz && \
      mkdir tflite-toolchains &&\
      tar xvf gcc-arm-8.3-2019.03-x86_64-aarch64-linux-gnu.tar.xz -C tflite-toolchains && \
      mkdir -p tflite/build && \
      cd tflite/build && \
      ARMCC_PREFIX=$HOME/armnn-devenv/tflite-toolchains/gcc-arm-8.3-2019.03-x86_64-aarch64-linux-gnu/bin/aarch64-linux-gnu- && \
      ARMCC_FLAGS="-funsafe-math-optimizations" && \
      cmake -DCMAKE_C_COMPILER=${ARMCC_PREFIX}gcc \
            -DCMAKE_CXX_COMPILER=${ARMCC_PREFIX}g++ \
            -DCMAKE_C_FLAGS="${ARMCC_FLAGS}" -DCMAKE_CXX_FLAGS="${ARMCC_FLAGS}" \
            -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON  -DCMAKE_SYSTEM_NAME=Linux \
            -DTFLITE_ENABLE_XNNPACK=OFF \
            -DCMAKE_SYSTEM_PROCESSOR=aarch64 \
            $HOME/armnn-devenv/tensorflow/tensorflow/lite/ && \
      cmake --build .
  
  # Build 32-bit TF Lite
  RUN cd $HOME/armnn-devenv && \
      curl -LO https://storage.googleapis.com/mirror.tensorflow.org/developer.arm.com/media/Files/downloads/gnu-a/8.3-2019.03/binrel/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf.tar.xz && \
      mkdir tflite-toolchains-32 && \
      tar xf gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf.tar.xz -C tflite-toolchains-32 && \
      mkdir -p tflite-32/build && \
      cd tflite-32/build && \
      ARMCC_PREFIX=$HOME/armnn-devenv/tflite-toolchains-32/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf/bin/arm-linux-gnueabihf- && \
      ARMCC_FLAGS="-funsafe-math-optimizations" && \
      cmake -DCMAKE_C_COMPILER=${ARMCC_PREFIX}gcc \
            -DCMAKE_CXX_COMPILER=${ARMCC_PREFIX}g++ \
            -DCMAKE_C_FLAGS="${ARMCC_FLAGS}" -DCMAKE_CXX_FLAGS="${ARMCC_FLAGS}" \
            -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON  -DCMAKE_SYSTEM_NAME=Linux \
            -DTFLITE_ENABLE_XNNPACK=OFF \
            -DCMAKE_SYSTEM_PROCESSOR=armv7a \
            $HOME/armnn-devenv/tensorflow/tensorflow/lite/ && \
      cmake --build .
  
  # Download Flatbuffers
  RUN cd $HOME/armnn-devenv && \
      wget -O flatbuffers-1.12.0.tar.gz https://github.com/google/flatbuffers/archive/v1.12.0.tar.gz && \
      tar xf flatbuffers-1.12.0.tar.gz && \
      cd flatbuffers-1.12.0 && \
      rm -f CMakeCache.txt
  
  # Build native x86_64 version of Flatbuffers
  RUN cd $HOME/armnn-devenv && cd flatbuffers-1.12.0 && \
      mkdir build && \
      cd build && \
      cmake .. -DFLATBUFFERS_BUILD_FLATC=1 \
           -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers \
           -DFLATBUFFERS_BUILD_TESTS=0 && \
      make all install -j$(nproc)
  
  # Build arm64 version of Flatbuffers
  RUN cd $HOME/armnn-devenv && cd flatbuffers-1.12.0 && \
      mkdir build-arm64 && \
      cd build-arm64 && \
      CXXFLAGS="-fPIC" cmake .. -DCMAKE_C_COMPILER=/usr/bin/aarch64-linux-gnu-gcc \
           -DCMAKE_CXX_COMPILER=/usr/bin/aarch64-linux-gnu-g++ \
           -DFLATBUFFERS_BUILD_FLATC=1 \
           -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers-arm64 \
           -DFLATBUFFERS_BUILD_TESTS=0 && \
      make all install -j$(nproc)
  
  # Build arm32 version of Flatbuffers
  RUN cd $HOME/armnn-devenv && cd flatbuffers-1.12.0  && \
      mkdir build-arm32 && \
      cd build-arm32 && \
      CXXFLAGS="-fPIC" cmake .. -DCMAKE_C_COMPILER=/usr/bin/arm-linux-gnueabihf-gcc \
           -DCMAKE_CXX_COMPILER=/usr/bin/arm-linux-gnueabihf-g++ \
           -DFLATBUFFERS_BUILD_FLATC=1 \
           -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers-arm32 \
           -DFLATBUFFERS_BUILD_TESTS=0 && \
      make all install -j$(nproc)
  
  # Build onnx
  RUN cd $HOME/armnn-devenv && git clone https://github.com/onnx/onnx.git && \
      cd onnx && \
      git fetch https://github.com/onnx/onnx.git 553df22c67bee5f0fe6599cff60f1afc6748c635 && git checkout FETCH_HEAD && \
      LD_LIBRARY_PATH=$HOME/armnn-devenv/google/x86_64_pb_install/lib:$LD_LIBRARY_PATH \
      $HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc \
      onnx/onnx.proto --proto_path=. --proto_path=../google/x86_64_pb_install/include --cpp_out $HOME/armnn-devenv/onnx
  
  # Generate TF Lite Schema 64-bit
  RUN cd $HOME/armnn-devenv && \
      cd tflite && \
      cp ../tensorflow/tensorflow/lite/schema/schema.fbs . && \
      ../flatbuffers-1.12.0/build/flatc -c --gen-object-api --reflect-types --reflect-names schema.fbs
  
  # Generate TF Lite Schema 32-bit
  RUN cd $HOME/armnn-devenv && \
      cd tflite-32 && \
      cp ../tensorflow/tensorflow/lite/schema/schema.fbs . && \
      ../flatbuffers-1.12.0/build/flatc -c --gen-object-api --reflect-types --reflect-names schema.fbs
  
  # Build 64-bit ArmNN
  RUN cd $HOME/armnn-devenv && \
      cd armnn && mkdir build && cd build && \
      export CXX=aarch64-linux-gnu-g++ && \
      export CC=aarch64-linux-gnu-gcc && \
      cmake .. \
      -DCMAKE_CXX_FLAGS=-w \
      -DBUILD_TESTS=1 \
      -DARMCOMPUTE_ROOT=$HOME/armnn-devenv/ComputeLibrary \
      -DARMCOMPUTE_BUILD_DIR=$HOME/armnn-devenv/ComputeLibrary/build/ \
      -DARMCOMPUTENEON=1 -DARMCOMPUTECL=1 -DARMNNREF=1 \
      -DONNX_GENERATED_SOURCES=$HOME/armnn-devenv/onnx \
      -DBUILD_ONNX_PARSER=1 \
      -DBUILD_TF_LITE_PARSER=1 \
      -DBUILD_ARMNN_TFLITE_DELEGATE=1 \
      -DTENSORFLOW_ROOT=$HOME/armnn-devenv/tensorflow \
      -DTFLITE_LIB_ROOT=$HOME/armnn-devenv/tflite/build \
      -DTF_LITE_SCHEMA_INCLUDE_PATH=$HOME/armnn-devenv/tflite \
      -DFLATBUFFERS_ROOT=$HOME/armnn-devenv/flatbuffers-arm64 \
      -DFLATC_DIR=$HOME/armnn-devenv/flatbuffers-1.12.0/build \
      -DPROTOBUF_ROOT=$HOME/armnn-devenv/google/x86_64_pb_install \
      -DPROTOBUF_LIBRARY_DEBUG=$HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so.23.0.0 \
      -DPROTOBUF_LIBRARY_RELEASE=$HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so.23.0.0 && \
      make -j$(nproc)
  
  # Build 32-bit ArmNN
  RUN cd $HOME/armnn-devenv && \
      cd armnn-32 && mkdir build && cd build && \
      export CXX=arm-linux-gnueabihf-g++ && \
      export CC=arm-linux-gnueabihf-gcc && \
      cmake .. \
      -DCMAKE_CXX_FLAGS=-w \
      -DBUILD_TESTS=1 \
      -DARMCOMPUTE_ROOT=$HOME/armnn-devenv/ComputeLibrary_32 \
      -DARMCOMPUTE_BUILD_DIR=$HOME/armnn-devenv/ComputeLibrary_32/build/ \
      -DARMCOMPUTENEON=1 -DARMCOMPUTECL=1 -DARMNNREF=1 \
      -DONNX_GENERATED_SOURCES=$HOME/armnn-devenv/onnx \
      -DBUILD_ONNX_PARSER=1 \
      -DBUILD_TF_LITE_PARSER=1 \
      -DBUILD_ARMNN_TFLITE_DELEGATE=1 \
      -DTENSORFLOW_ROOT=$HOME/armnn-devenv/tensorflow \
      -DTFLITE_LIB_ROOT=$HOME/armnn-devenv/tflite-32/build \
      -DTF_LITE_SCHEMA_INCLUDE_PATH=$HOME/armnn-devenv/tflite-32 \
      -DFLATBUFFERS_ROOT=$HOME/armnn-devenv/flatbuffers-arm32 \
      -DFLATC_DIR=$HOME/armnn-devenv/flatbuffers-1.12.0/build \
      -DPROTOBUF_ROOT=$HOME/armnn-devenv/google/x86_64_pb_install \
      -DPROTOBUF_LIBRARY_DEBUG=$HOME/armnn-devenv/google/arm32_pb_install/lib/libprotobuf.so.23.0.0 \
      -DPROTOBUF_LIBRARY_RELEASE=$HOME/armnn-devenv/google/arm32_pb_install/lib/libprotobuf.so.23.0.0 \
      -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON && \
      make -j$(nproc)
  

• 如果需要自行替換toolchain，可以參考下面步驟：

1. Dockerfile

   FROM ubuntu:18.04
   ENV TERM linux
   ENV DEBIAN_FRONTEND noninteractive
   
   # Forward system proxy setting
   # ARG proxy
   # ENV http_proxy $proxy
   # ENV https_proxy $proxy
   
   # Basic apt update
   RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends locales ca-certificates &&  rm -rf /var/lib/apt/lists/*
   
   # Set the locale to en_US.UTF-8, because the Yocto build fails without any locale set.
   RUN locale-gen en_US.UTF-8 && update-locale LC_ALL=en_US.UTF-8 LANG=en_US.UTF-8
   ENV LANG en_US.UTF-8
   ENV LC_ALL en_US.UTF-8
   
   # Again, off the certificare
   RUN echo "check_certificate = off" >> ~/.wgetrc
   RUN echo "[global] \n\
   trusted-host = pypi.python.org \n \
   \t               pypi.org \n \
   \t              files.pythonhosted.org" >> /etc/pip.conf
   
   # Get basic packages
   RUN apt-get update && apt-get install -y \
       apparmor \
       aufs-tools \
       automake \
       bash-completion \
       btrfs-tools \
       build-essential \
       cmake \
       createrepo \
       curl \
       dpkg-sig \
       g++ \
       gcc \
       git \
       iptables \
       jq \
       libapparmor-dev \
       libc6-dev \
       libcap-dev \
       libsystemd-dev \
       libyaml-dev \
       mercurial \
       net-tools \
       parallel \
       pkg-config \
       python-dev \
       python-mock \
       python-pip \
       python-setuptools \
       python-websocket \
       golang-go \
       iproute2 \
       iputils-ping \
       vim-common \
       vim \
       wget \
       libtool \
       unzip \
       scons \
       curl \
       autoconf \
       libtool \
       build-essential \
       libssl-dev \
       g++ && rm -rf /var/lib/apt/lists/*
   
   # Install Cross-compiling ToolChain
   RUN apt-get update && apt-get install -y crossbuild-essential-arm64 crossbuild-essential-armhf
   
   # Install Cmake 3.19
   RUN cd $HOME && \
       wget -O cmake-3.19.0.tar.gz https://cmake.org/files/v3.19/cmake-3.19.0.tar.gz && \
       tar -xzf cmake-3.19.0.tar.gz && \
       cd $HOME/cmake-3.19.0 && \
       ./bootstrap && \
       make && \
       make install
   

   sudo docker build --rm -t imprld01/armnn:v0 . --no-cache
   

   sudo docker run --rm --gpus all --net host -it --shm-size=8g \
                   -v /armnn_scripts/:/armnn_scripts/ \
                   -v /toolchain/:/toolchain/ \
                   imprld01/armnn:v0 bash
   

2. Download Google Protobuf Library amd Build it

   #!/bin/bash
   
   # Build and install Google's Protobuf library
   # Download and Extract
   mkdir -p $HOME/google && \
       cd $HOME/google && \
       wget https://github.com/protocolbuffers/protobuf/releases/download/v3.12.0/protobuf-all-3.12.0.tar.gz && \
       tar -zxvf protobuf-all-3.12.0.tar.gz
   
   # Build a native (x86_64) version
   cd $HOME/google/protobuf-3.12.0 && \
       mkdir x86_build && cd x86_build && \
       ../configure --prefix=$HOME/armnn-devenv/google/x86_64_pb_install && \
       make install -j$(nproc)
   

   #!/bin/bash
   
   # Build the arm32 version of the protobuf libraries
   cd $HOME/google/protobuf-3.12.0 && \
       mkdir arm32_build_myarm && cd arm32_build_myarm && \
       export LD_LIBRARY_PATH=/toolchain/sysroots/unknown-linux/usr/lib/arm-unknown-linux-gnueabi/ && \
       export CC=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-gcc && \
       export CFLAGS="--sysroot=/toolchain/sysroots/unknown-linux-gnueabi" && \
       export CFLAGS_FOR_BUILD="-g -fPIC -fno-short-enums -fno-short-enums -lstdc++" && \
       export CXX=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-g++ && \
       export CXXFLAGS="--sysroot=/toolchain/sysroots/unknown-linux-gnueabi" && \
       export CXXFLAGS_FOR_BUILD="-g -fPIC -fno-short-enums -fno-short-enums -lstdc++" && \
       ../configure --host=arm-unknown-linux \
       --prefix=$HOME/armnn-devenv/google/arm32_pb_install_myarm \
       --with-protoc=$HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc && \
       make install -j$(nproc)
   

3. Download ArmNN

   #!/bin/bash
   
   # Dep Error - Bug ARMNN
   apt-get update && apt-get install -y \
        python-numpy
   
   # Download ArmNN 32-bit
   cd $HOME/armnn-devenv && git clone "https://review.mlplatform.org/ml/armnn" armnn-32-myarm && \
       cd armnn-32-myarm/ && git checkout master
   

4. Download Compute Library and Build it

   #!/bin/bash
   
   # Build 32-bit Compute Library
   cd $HOME/armnn-devenv/ && mkdir ComputeLibrary_myarm && git clone https://review.mlplatform.org/ml/ComputeLibrary ComputeLibrary_myarm && \
       cd ComputeLibrary_myarm && \
       git checkout $($HOME/armnn-devenv/armnn/scripts/get_compute_library.sh -p) && \
       export LD_LIBRARY_PATH=/toolchain/sysroots/unknown-linux/usr/lib/arm-unknown-linux-gnueabi/ && \
       scons Werror=0 arch=armv7a neon=1 opencl=1 embed_kernels=1 cppthreads=1 openmp=0 compiler_prefix="/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-" extra_cxx_flags="-fPIC -mfloat-abi=softfp --sysroot=/toolchain/sysroots/unknown-linux-gnueabi/" extra_link_flags="-fPIC --sysroot=/toolchain/sysroots/unknown-linux-gnueabi/" -j$(nproc)
   

5. Download Tensorflow and Build TFLite

   #!/bin/bash
   
   # Download Tensorflow (Checkout latest tested version of TF using get_tensorflow.sh)
   cd $HOME/armnn-devenv && \
       git clone https://github.com/tensorflow/tensorflow.git && \
       cd tensorflow && \
       git checkout $($HOME/armnn-devenv/armnn/scripts/get_tensorflow.sh -p)
   
   # Build 32-bit TF Lite
   cd $HOME/armnn-devenv && \
       mkdir -p tflite-myarm/build && \
       cd tflite-myarm/build && \
       ARMCC_PREFIX=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi- && \
       ARMCC_FLAGS="-funsafe-math-optimizations -fPIC --sysroot=/toolchain/sysroots/unknown-linux-gnueabi/" && \
       cmake -DCMAKE_C_COMPILER=${ARMCC_PREFIX}gcc \
             -DCMAKE_CXX_COMPILER=${ARMCC_PREFIX}g++ \
             -DCMAKE_C_FLAGS="${ARMCC_FLAGS}" -DCMAKE_CXX_FLAGS="${ARMCC_FLAGS}" \
             -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON  -DCMAKE_SYSTEM_NAME=Linux \
             -DTFLITE_ENABLE_XNNPACK=OFF \
             -DCMAKE_SYSTEM_PROCESSOR=armv7a \
             $HOME/armnn-devenv/tensorflow/tensorflow/lite/ && \
       cmake --build .
   

6. Download Flatbuffers and Build it

   #!/bin/bash
   
   # Download Flatbuffers
   cd $HOME/armnn-devenv && \
       wget -O flatbuffers-1.12.0.tar.gz https://github.com/google/flatbuffers/archive/v1.12.0.tar.gz && \
       tar xf flatbuffers-1.12.0.tar.gz && \
       cd flatbuffers-1.12.0 && \
       rm -f CMakeCache.txt
   
   # Build native x86_64 version of Flatbuffers
   cd $HOME/armnn-devenv && cd flatbuffers-1.12.0 && \
       mkdir build && \
       cd build && \
       cmake .. -DFLATBUFFERS_BUILD_FLATC=1 \
            -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers \
            -DFLATBUFFERS_BUILD_TESTS=0 && \
       make all install -j$(nproc)
   

   修改flatbuffers-1.12.0下面的CMakeLists.txt，加上：

   set(CMAKE_C_COMPILER_FORCED TRUE)
   set(CMAKE_CXX_COMPILER_FORCED TRUE)
   

   # Build arm32 version of Flatbuffers
   cd $HOME/armnn-devenv && cd flatbuffers-1.12.0  && \
       cp /root/armnn_envs_x86_64/flatbuffers-1.12.0/CMakeLists.txt . && \
       mkdir build-arm32-myarm && \
       cd build-arm32-myarm && \
       CXXFLAGS="-fPIC --sysroot=/toolchain/sysroots/unknown-linux-gnueabi" cmake .. -DCMAKE_C_COMPILER=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-gcc \
            -DCMAKE_CXX_COMPILER=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-g++ \
            -DFLATBUFFERS_BUILD_FLATC=1 \
            -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers-arm32-myarm \
            -DFLATBUFFERS_BUILD_TESTS=0 && \
       make all install -j$(nproc)
   

7. Download ONNX and Build it

   #!/bin/bash
   
   # Build onnx
   cd $HOME/armnn-devenv && git clone https://github.com/onnx/onnx.git && \
       cd onnx && \
       git fetch https://github.com/onnx/onnx.git 553df22c67bee5f0fe6599cff60f1afc6748c635 && git checkout FETCH_HEAD && \
       LD_LIBRARY_PATH=$HOME/armnn-devenv/google/x86_64_pb_install/lib:$LD_LIBRARY_PATH \
       $HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc \
       onnx/onnx.proto --proto_path=. --proto_path=../google/x86_64_pb_install/include --cpp_out $HOME/armnn-devenv/onnx
   

8. Generate TFLite Schema

   #!/bin/bash
   
   # Generate TF Lite Schema 32-bit
   cd $HOME/armnn-devenv && \
       cd tflite-myarm && \
       cp ../tensorflow/tensorflow/lite/schema/schema.fbs . && \
       ../flatbuffers-1.12.0/build/flatc -c --gen-object-api --reflect-types --reflect-names schema.fbs
   

9. Build Arm NN

   修改armnn-32-myarm下面的CMakeLists.txt，加上：

   set(CMAKE_C_COMPILER_FORCED TRUE)
   set(CMAKE_CXX_COMPILER_FORCED TRUE)
   

   #!/bin/bash
   
   # Build 32-bit ArmNN
   cd $HOME/armnn-devenv && \
       cd armnn-32-myarm && \
       cp /root/armnn_envs_x86_64/armnn-32-myarm/CMakeLists.txt . && \
       mkdir build && cd build && \
       export LD_LIBRARY_PATH=/toolchain/sysroots/unknown-linux/usr/lib/arm-unknown-linux-gnueabi/ && \
       export CXX=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-g++ && \
       export CC=/toolchain/sysroots/unknown-linux/usr/bin/arm-unknown-linux-gnueabi/arm-unknown-linux-gnueabi-gcc && \
       cmake .. \
       -DCMAKE_CXX_FLAGS="-w -fPIC -pthread -latomic --sysroot=/toolchain/sysroots/unknown-linux-gnueabi" \
       -DBUILD_TESTS=1 \
       -DARMCOMPUTE_ROOT=$HOME/armnn-devenv/ComputeLibrary_myarm \
       -DARMCOMPUTE_BUILD_DIR=$HOME/armnn-devenv/ComputeLibrary_myarm/build/ \
       -DARMCOMPUTENEON=1 -DARMCOMPUTECL=1 -DARMNNREF=1 \
       -DONNX_GENERATED_SOURCES=$HOME/armnn-devenv/onnx \
       -DBUILD_ONNX_PARSER=1 \
       -DBUILD_TF_LITE_PARSER=1 \
       -DBUILD_ARMNN_TFLITE_DELEGATE=1 \
       -DTENSORFLOW_ROOT=$HOME/armnn-devenv/tensorflow \
       -DTFLITE_LIB_ROOT=$HOME/armnn-devenv/tflite-myarm/build \
       -DTF_LITE_SCHEMA_INCLUDE_PATH=$HOME/armnn-devenv/tflite-myarm \
       -DFLATBUFFERS_ROOT=$HOME/armnn-devenv/flatbuffers-arm32-myarm \
       -DFLATC_DIR=$HOME/armnn-devenv/flatbuffers-1.12.0/build \
       -DPROTOBUF_ROOT=$HOME/armnn-devenv/google/x86_64_pb_install \
       -DPROTOBUF_LIBRARY_DEBUG=$HOME/armnn-devenv/google/arm32_pb_install_myarm/lib/libprotobuf.so.23.0.0 \
       -DPROTOBUF_LIBRARY_RELEASE=$HOME/armnn-devenv/google/arm32_pb_install_myarm/lib/libprotobuf.so.23.0.0 \
       -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON && \
       make -j$(nproc)
   

Further Reading

1. Note of Compilation of Arm NN by Android NDK - 使用 Android NDK 編譯 Arm NN

Reference

1. https://github.com/ARM-software/armnn
2. 交叉编译ARM平台、移植ArmNN_yiyayiya557的博客-CSDN博客_arm nn
3. 为Tensorflow、Tensorflow lite配置Arm NN SDK编译环境_我是木林森呀的博客-CSDN博客_armnn tensorflow
4. Configuring the Arm NN SDK build environment for Caffe
5. 在imx8qm上开心把玩Machine Learning_AIA-XP的博客-CSDN博客_imx8qm
6. ATU Book i.MX8系列 - ArmNN 使用介紹
7. armnn/BuildGuideCrossCompilation.md at branches/armnn_20_02 · ARM-software/armnn

---

• ← Previous Post
• Next Post →

comments powered by Disqus