Introduction

Table of Contents

• ArmNN
• License
• Contributor Guide
  • Developer Certificate of Origin (DCO)
  • Releases
  • Developer Repository
  • Code Review
  • Continuous Integration
  • Communications
• Build Instructions
  • How to use the Android NDK to build ArmNN
  • Cross Compilation Build Guide

ArmNN

Arm NN is a key component of the machine learning platform, which is part of the Linaro Machine Intelligence Initiative. For more information on the machine learning platform and Arm NN, see: https://mlplatform.org/, also there is further Arm NN information available from https://developer.arm.com/products/processors/machine-learning/arm-nn

There is a getting started guide here using TensorFlow: https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/configuring-the-arm-nn-sdk-build-environment-for-tensorflow

There is a getting started guide here using TensorFlow Lite: https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/configuring-the-arm-nn-sdk-build-environment-for-tensorflow-lite

There is a getting started guide here using Caffe: https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/configuring-the-arm-nn-sdk-build-environment-for-caffe

There is a getting started guide here using ONNX: https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/configuring-the-arm-nn-sdk-build-environment-for-onnx

There is a guide for backend development here.

License

Arm NN is provided under the MIT license.

MIT License

Copyright (c) 2017-2020 ARM Limited.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Individual files contain the following tag instead of the full license text.

SPDX-License-Identifier: MIT

This enables machine processing of license information based on the SPDX License Identifiers that are available here

Contributor Guide

The Arm NN project is open for external contributors and welcomes contributions. Arm NN is licensed under the MIT license and all accepted contributions must have the same license. For more details on contributing to Arm NN see the Contributing page on the MLPlatform.org website.

Developer Certificate of Origin (DCO)

Before the Arm NN project accepts your contribution, you need to certify its origin and give us your permission. To manage this process we use Developer Certificate of Origin (DCO) V1.1 (https://developercertificate.org/).

To indicate that you agree to the the terms of the DCO, you "sign off" your contribution by adding a line with your name and e-mail address to every git commit message:

Signed-off-by: John Doe john..nosp@m.doe@.nosp@m.examp.nosp@m.le.o.nosp@m.rg

You must use your real name, no pseudonyms or anonymous contributions are accepted.

Releases

Official Arm NN releases are published through the official Arm NN Github repository.

Developer Repository

The Arm NN development repository is hosted on the mlplatform.org git repository hosted by Linaro.

Code Review

Contributions must go through code review. Code reviews are performed through the mlplatform.org Gerrit server. Contributors need to signup to this Gerrit server with their GitHub account credentials.

Only reviewed contributions can go to the master branch of Arm NN.

Continuous Integration

Contributions to Arm NN go through testing at the Arm CI system. All unit, integration and regression tests must pass before a contribution gets merged to the Arm NN master branch.

Communications

We encourage all Arm NN developers to subscribe to the Arm NN developer mailing list.

Build Instructions

Arm tests the build system of Arm NN with the following build environments:

Android NDK Cross compilation from x86_64 Ubuntu to arm64 Linux Native compilation under aarch64 Debian 9

Arm NN is written using portable C++14 and the build system uses CMake, therefore it is possible to build for a wide variety of target platforms, from a wide variety of host environments.

The armnn/tests directory contains tests used during Arm NN development. Many of them depend on third-party IP, model protobufs and image files not distributed with Arm NN. The dependencies of some of the tests are available freely on the Internet, for those who wish to experiment.

The 'armnn/samples' directory contains SimpleSample.cpp, a very basic example of the ArmNN SDK API in use.

The 'ExecuteNetwork' program, in armnn/tests/ExecuteNetwork, has no additional dependencies beyond those required by Arm NN and the model parsers. It takes any model and any input tensor, and simply prints out the output tensor. Run it with no arguments to see command-line help.

The 'ArmnnConverter' program, in armnn/src/armnnConverter, has no additional dependencies beyond those required by Arm NN and the model parsers. It takes a model in TensorFlow format and produces a serialized model in Arm NN format. Run it with no arguments to see command-line help. Note that this program can only convert models for which all operations are supported by the serialization tool src/armnnSerializer.

The 'ArmnnQuantizer' program, in armnn/src/armnnQuantizer, has no additional dependencies beyond those required by Arm NN and the model parsers. It takes a 32-bit float network and converts it into a quantized asymmetric 8-bit or quantized symmetric 16-bit network. Static quantization is supported by default but dynamic quantization can be enabled if CSV file of raw input tensors is specified. Run it with no arguments to see command-line help.

note that Arm NN needs to be built against a particular version of ARM's Compute Library. The get_compute_library.sh in the scripts subdirectory will clone the compute library from the review.mlplatform.org github repository into a directory alongside armnn named 'clframework' and checks out the correct revision.

How to use the Android NDK to build ArmNN

• Introduction
• Download the Android NDK and make a standalone toolchain
• Build the Boost C++ libraries
• Build the Compute Library
• Build Google's Protobuf library
• Download TensorFlow
• Build ArmNN
• Run ArmNN UnitTests on an Android device

Introduction

These are step by step instructions for using the Android NDK to build ArmNN. They have been tested on a clean install of Ubuntu 18.04, and should also work with other OS versions. The instructions show how to build the ArmNN core library and the optional TensorFlow parser. All downloaded or generated files will be saved inside the ~/armnn-devenv directory.

Download the Android NDK and make a standalone toolchain

Download the Android NDK from the official website:

mkdir -p ~/armnn-devenv/toolchains
cd ~/armnn-devenv/toolchains

#For Mac OS, change the NDK download link accordingly.
wget https://dl.google.com/android/repository/android-ndk-r17b-linux-x86_64.zip
unzip android-ndk-r17b-linux-x86_64.zip
export NDK=~/armnn-devenv/toolchains/android-ndk-r17b

 You may want to append `export NDK=~/armnn-devenv/toolchains/android-ndk-r17b` to your `~/.bashrc` (or `~/.bash_profile` in Mac OS).

Make a standalone toolchain:

 (Requires python if not previously installed: `sudo apt install python`)

# Create an arm64 API 26 libc++ toolchain.
$NDK/build/tools/make_standalone_toolchain.py \
    --arch arm64 \
    --api 26 \
    --stl=libc++ \
    --install-dir=$HOME/armnn-devenv/toolchains/aarch64-android-r17b
export PATH=$HOME/armnn-devenv/toolchains/aarch64-android-r17b/bin:$PATH

 You may want to append `export PATH=$HOME/armnn-devenv/toolchains/aarch64-android-r17b/bin:$PATH` to your `~/.bashrc` (or `~/.bash_profile` in Mac OS).

Build the Boost C++ libraries

Download Boost version 1.64:

mkdir ~/armnn-devenv/boost
cd ~/armnn-devenv/boost
wget https://dl.bintray.com/boostorg/release/1.64.0/source/boost_1_64_0.tar.bz2
tar xvf boost_1_64_0.tar.bz2

Build:

(Requires gcc if not previously installed: `sudo apt install gcc`)

echo "using gcc : arm : aarch64-linux-android-clang++ ;" > $HOME/armnn-devenv/boost/user-config.jam
cd ~/armnn-devenv/boost/boost_1_64_0
./bootstrap.sh --prefix=$HOME/armnn-devenv/boost/install
./b2 install --user-config=$HOME/armnn-devenv/boost/user-config.jam \
 toolset=gcc-arm link=static cxxflags=-fPIC --with-filesystem \
 --with-test --with-log --with-program_options -j16

Build the Compute Library

Clone the Compute Library:

(Requires Git if not previously installed: `sudo apt install git`)

cd ~/armnn-devenv
git clone https://github.com/ARM-software/ComputeLibrary.git

Build:

(Requires SCons if not previously installed: `sudo apt install scons`)

cd ComputeLibrary
scons arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" \
 benchmark_tests=0 validation_tests=0 os=android -j16

Build Google's Protobuf library

Clone protobuf:

mkdir ~/armnn-devenv/google
cd ~/armnn-devenv/google
git clone https://github.com/google/protobuf.git
cd protobuf
git checkout -b v3.5.2 v3.5.2

Build a native (x86) version of the protobuf libraries and compiler (protoc):

(Requires cUrl, autoconf, llibtool, and other build dependencies if not previously installed: `sudo apt install curl autoconf libtool build-essential g++`)

./autogen.sh
mkdir x86_build
cd x86_build
../configure --prefix=$HOME/armnn-devenv/google/x86_pb_install
make install -j16
cd ..

Build the arm64 version of the protobuf libraries:

mkdir arm64_build
cd arm64_build
CC=aarch64-linux-android-clang \
  CXX=aarch64-linux-android-clang++ \
  CFLAGS="-fPIE -fPIC" LDFLAGS="-pie -llog" \
   ../configure --host=aarch64-linux-android \
   --prefix=$HOME/armnn-devenv/google/arm64_pb_install \
   --with-protoc=$HOME/armnn-devenv/google/x86_pb_install/bin/protoc
make install -j16
cd ..

Download TensorFlow

Clone TensorFlow source code:

cd ~/armnn-devenv/google/
git clone https://github.com/tensorflow/tensorflow.git
cd tensorflow/
git checkout a0043f9262dc1b0e7dc4bdf3a7f0ef0bebc4891e

You need tensorflow/contrib/makefile/tf_proto_files.txt from TensorFlow to generate TensorFlow protobuf definitions. This file is not available in TensorFlow master branch.

Build ArmNN

Clone ArmNN source code:

cd ~/armnn-devenv/
git clone https://github.com/ARM-software/armnn.git

Generate TensorFlow protobuf definitions:

cd ~/armnn-devenv/google/tensorflow
~/armnn-devenv/armnn/scripts/generate_tensorflow_protobuf.sh \
 $HOME/armnn-devenv/google/tf_pb $HOME/armnn-devenv/google/x86_pb_install

Build ArmNN:

(Requires CMake if not previously installed: `sudo apt install cmake`)

mkdir ~/armnn-devenv/armnn/build
cd ~/armnn-devenv/armnn/build
CXX=aarch64-linux-android-clang++ \
CC=aarch64-linux-android-clang \
CXX_FLAGS="-fPIE -fPIC" \
cmake .. \
-DCMAKE_SYSTEM_NAME=Android \
-DCMAKE_ANDROID_ARCH_ABI=arm64-v8a \
-DCMAKE_ANDROID_STANDALONE_TOOLCHAIN=$HOME/armnn-devenv/toolchains/aarch64-android-r17b/ \
-DCMAKE_EXE_LINKER_FLAGS="-pie -llog" \
-DARMCOMPUTE_ROOT=$HOME/armnn-devenv/ComputeLibrary/ \
-DARMCOMPUTE_BUILD_DIR=$HOME/armnn-devenv/ComputeLibrary/build \
-DBOOST_ROOT=$HOME/armnn-devenv/boost/install/ \
-DARMCOMPUTENEON=1 -DARMCOMPUTECL=1 -DARMNNREF=1 \
-DTF_GENERATED_SOURCES=$HOME/armnn-devenv/google/tf_pb/ -DBUILD_TF_PARSER=1 \
-DPROTOBUF_ROOT=$HOME/armnn-devenv/google/arm64_pb_install/
make -j16

Run the ArmNN unit tests on an Android device

Push the build results to an Android device and make symbolic links for shared libraries:

Currently adb version we have used for testing is 1.0.41.

adb push libarmnnTfParser.so /data/local/tmp/
adb push libarmnn.so /data/local/tmp/
adb push UnitTests /data/local/tmp/
adb push $NDK/sources/cxx-stl/llvm-libc++/libs/arm64-v8a/libc++_shared.so /data/local/tmp/
adb push $HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so /data/local/tmp/libprotobuf.so.15.0.1
adb shell 'ln -s libprotobuf.so.15.0.1 /data/local/tmp/libprotobuf.so.15'
adb shell 'ln -s libprotobuf.so.15.0.1 /data/local/tmp/libprotobuf.so'

Push the files needed for the unit tests (they are a mix of files, directories and symbolic links):

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/testSharedObject
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/testSharedObject/* /data/local/tmp/src/backends/backendsCommon/test/testSharedObject/

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/testDynamicBackend
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/testDynamicBackend/* /data/local/tmp/src/backends/backendsCommon/test/testDynamicBackend/

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath1
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/backendsTestPath1/* /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath1/

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/backendsTestPath2/Arm_CpuAcc_backend.so /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2/
adb shell ln -s Arm_CpuAcc_backend.so /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2/Arm_CpuAcc_backend.so.1
adb shell ln -s Arm_CpuAcc_backend.so.1 /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2/Arm_CpuAcc_backend.so.1.2
adb shell ln -s Arm_CpuAcc_backend.so.1.2 /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2/Arm_CpuAcc_backend.so.1.2.3
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/backendsTestPath2/Arm_GpuAcc_backend.so /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2/
adb shell ln -s nothing /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath2/Arm_no_backend.so

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath3

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath5
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/backendsTestPath5/* /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath5/

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath6
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/backendsTestPath6/* /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath6/

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath7

adb shell mkdir -p /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath9
adb push -p ~/armnn-devenv/armnn/build/src/backends/backendsCommon/test/backendsTestPath9/* /data/local/tmp/src/backends/backendsCommon/test/backendsTestPath9/

adb shell mkdir -p /data/local/tmp/src/backends/dynamic/reference
adb push -p ~/armnn-devenv/armnn/build/src/backends/dynamic/reference/Arm_CpuRef_backend.so /data/local/tmp/src/backends/dynamic/reference/

Run ArmNN unit tests:

adb shell 'LD_LIBRARY_PATH=/data/local/tmp /data/local/tmp/UnitTests'

If libarmnnUtils.a is present in `~/armnn-devenv/armnn/build/` and the unit tests run without failure then the build was successful.

Cross Compilation Build Guide

• Introduction
• Cross-compiling ToolChain
• Build and install Google's Protobuf library
• Build Caffe for x86_64
• Build Boost library for arm64
• Build Compute Library
• Build ArmNN
• Run Unit Tests
• Troubleshooting and Errors

Introduction

These are the step by step instructions on Cross-Compiling ArmNN under an x86_64 system to target an Arm64 system. This build flow has been tested with Ubuntu 16.04. The instructions show how to build the ArmNN core library and the Boost, Protobuf, Caffe and Compute Libraries necessary for compilation.

Cross-compiling ToolChain

Install the standard cross-compilation libraries for arm64:

sudo apt install crossbuild-essential-arm64

Build and install Google's Protobuf library

Get protobuf-all-3.5.1.tar.gz from here.

Extract:

tar -zxvf protobuf-all-3.5.1.tar.gz
cd protobuf-3.5.1

Build a native (x86_64) version of the protobuf libraries and compiler (protoc):

(Requires cUrl, autoconf, llibtool, and other build dependencies if not previously installed: sudo apt install curl autoconf libtool build-essential g++)

mkdir x86_64_build
cd x86_64_build
../configure --prefix=$HOME/armnn-devenv/google/x86_64_pb_install
make install -j16
cd ..

Build the arm64 version of the protobuf libraries:

mkdir arm64_build
cd arm64_build
CC=aarch64-linux-gnu-gcc \
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 -j16
cd ..

Build Caffe for x86_64

Ubuntu 16.04 installation. These steps are taken from the full Caffe installation documentation here

Install dependencies:

sudo apt-get install libleveldb-dev libsnappy-dev libopencv-dev libhdf5-serial-dev
sudo apt-get install --no-install-recommends libboost-all-dev
sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev
sudo apt-get install libopenblas-dev
sudo apt-get install libatlas-base-dev

Download Caffe-Master from here.

git clone https://github.com/BVLC/caffe.git
cd caffe
cp Makefile.config.example Makefile.config

Adjust Makefile.config as necessary for your environment, for example:

#CPU only version:
CPU_ONLY := 1

#Add hdf5 and protobuf include and library directories (Replace $HOME with explicit /home/username dir):
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/ $HOME/armnn-devenv/google/x86_64_pb_install/include/
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial/ $HOME/armnn-devenv/google/x86_64_pb_install/lib/

Setup environment:

export PATH=$HOME/armnn-devenv/google/x86_64_pb_install/bin/:$PATH
export LD_LIBRARY_PATH=$HOME/armnn-devenv/google/x86_64_pb_install/lib/:$LD_LIBRARY_PATH

Compilation with Make:

make all
make test
make runtest

These should all run without errors

caffe.pb.h and caffe.pb.cc will be needed when building ArmNN's Caffe Parser

Build Boost library for arm64

Build Boost library for arm64

Download Boost version 1.64 from here. Using any version of Boost greater than 1.64 will fail to build ArmNN, due to different dependency issues.

tar -zxvf boost_1_64_0.tar.gz
cd boost_1_64_0
echo "using gcc : arm : aarch64-linux-gnu-g++ ;" > user_config.jam
./bootstrap.sh --prefix=$HOME/armnn-devenv/boost_arm64_install
./b2 install toolset=gcc-arm link=static cxxflags=-fPIC --with-filesystem --with-test --with-log --with-program_options -j32 --user-config=user_config.jam

Build Compute Library

Building the Arm Compute Library:

git clone https://github.com/ARM-software/ComputeLibrary.git
cd ComputeLibrary/
scons arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" -j8 internal_only=0

Build ArmNN

Compile ArmNN for arm64:

git clone https://github.com/ARM-software/armnn.git
cd armnn
mkdir build
cd build

Use CMake to configure your build environment, update the following script and run it from the armnn/build directory to set up the armNN build:

#!/bin/bash
CXX=aarch64-linux-gnu-g++ \
CC=aarch64-linux-gnu-gcc \
cmake .. \
-DARMCOMPUTE_ROOT=$HOME/armnn-devenv/ComputeLibrary \
-DARMCOMPUTE_BUILD_DIR=$HOME/armnn-devenv/ComputeLibrary/build/ \
-DBOOST_ROOT=$HOME/armnn-devenv/boost_arm64_install/ \
-DARMCOMPUTENEON=1 -DARMCOMPUTECL=1 -DARMNNREF=1 \
-DCAFFE_GENERATED_SOURCES=$HOME/armnn-devenv/caffe/build/src \
-DBUILD_CAFFE_PARSER=1 \
-DPROTOBUF_ROOT=$HOME/armnn-devenv/google/x86_64_pb_install/ \
-DPROTOBUF_LIBRARY_DEBUG=$HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so.15.0.1 \
-DPROTOBUF_LIBRARY_RELEASE=$HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so.15.0.1

Run the build

make -j32

Run Unit Tests

Copy the build folder to an arm64 linux machine

Copy the libprotobuf.so.15.0.1 library file to the build folder

cd to the build folder on your arm64 machine and set your LD_LIBRARY_PATH to its current location:

cd build/
export LD_LIBRARY_PATH=`pwd`

Create a symbolic link to libprotobuf.so.15.0.1:

ln -s libprotobuf.so.15.0.1 ./libprotobuf.so.15

Run the UnitTests:

./UnitTests
Running 567 test cases...

    No errors detected

Troubleshooting and Errors:

Error adding symbols: File in wrong format

When building armNN:

/usr/local/lib/libboost_log.a: error adding symbols: File in wrong format
collect2: error: ld returned 1 exit status
CMakeFiles/armnn.dir/build.make:4028: recipe for target 'libarmnn.so' failed
make[2]: *** [libarmnn.so] Error 1
CMakeFiles/Makefile2:105: recipe for target 'CMakeFiles/armnn.dir/all' failed
make[1]: *** [CMakeFiles/armnn.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2

Boost libraries are not compiled for the correct architecture, try recompiling for arm64

Virtual memory exhausted

When compiling the boost libraries:

virtual memory exhausted: Cannot allocate memory

Not enough memory available to compile. Increase the amount of RAM or swap space available.

Unrecognized command line option '-m64'

When compiling the boost libraries:

aarch64-linux-gnu-g++: error: unrecognized command line option ‘-m64’

Clean the boost library directory before trying to build with a different architecture:

sudo ./b2 clean

It should show the following for arm64:

- 32-bit                   : no
- 64-bit                   : yes
- arm                      : yes

Missing libz.so.1

When compiling armNN:

/usr/lib/gcc-cross/aarch64-linux-gnu/5/../../../../aarch64-linux-gnu/bin/ld: warning: libz.so.1, needed by /home/<username>/armNN/usr/lib64/libprotobuf.so.15.0.0, not found (try using -rpath or -rpath-link)

Missing arm64 libraries for libz.so.1, these can be added by adding a second architecture to dpkg and explicitly installing them:

sudo dpkg --add-architecture arm64
sudo apt-get install zlib1g:arm64
sudo apt-get update
sudo ldconfig

If apt-get update returns 404 errors for arm64 repos refer to section 5 below.

Alternatively the missing arm64 version of libz.so.1 can be downloaded and installed from a .deb package here

sudo dpkg -i zlib1g_1.2.8.dfsg-2ubuntu4_arm64.deb

Unable to install arm64 packages after adding arm64 architecture

Using sudo apt-get update should add all of the required repos for arm64 but if it does not or you are getting 404 errors the following instructions can be used to add the repos manually:

From stackoverflow here.

Open /etc/apt/sources.list with your preferred text editor.

Mark all the current (default) repos as [arch=], e.g.

deb [arch=amd64] http://archive.ubuntu.com/ubuntu/ xenial main restricted

Then add the following:

deb [arch=arm64] http://ports.ubuntu.com/ xenial main restricted
deb [arch=arm64] http://ports.ubuntu.com/ xenial-updates main restricted
deb [arch=arm64] http://ports.ubuntu.com/ xenial universe
deb [arch=arm64] http://ports.ubuntu.com/ xenial-updates universe
deb [arch=arm64] http://ports.ubuntu.com/ xenial multiverse
deb [arch=arm64] http://ports.ubuntu.com/ xenial-updates multiverse
deb [arch=arm64] http://ports.ubuntu.com/ xenial-backports main restricted universe multiverse

Update and install again:

sudo apt-get install zlib1g:arm64
sudo apt-get update
sudo ldconfig

Undefined references to google::protobuf:: functions

When compiling armNN there are multiple errors of the following type:

`` libarmnnCaffeParser.so: undefined reference togoogle::protobuf:* ```

Missing or out of date protobuf compilation libraries.

Use the command 'protoc –version' to check which version of protobuf is available (version 3.5.1 is required). Follow the instructions above to install protobuf 3.5.1 Note this will require you to recompile Caffe for x86_64

Errors on strict-aliasing rules when compiling the Compute Library

When compiling the Compute Library there are multiple errors on strict-aliasing rules:

cc1plus: error: unrecognized command line option ‘-Wno-implicit-fallthrough’ [-Werror]

Add Werror=0 to the scons command:

scons arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" -j8 internal_only=0 Werror=0