1 # How to Cross-Compile Arm NN on x86_64 for arm64
3 - [Introduction](#introduction)
4 - [Cross-compiling ToolChain](#cross-compiling-toolchain)
5 - [Install Cmake](#build-cmake)
6 - [Build and install Google's Protobuf library](#build-and-install-google-s-protobuf-library)
7 - [Download Arm NN](#download-arm-nn)
8 - [Build Arm Compute Library](#build-arm-compute-library)
9 - [Build Flatbuffer](#build-flatbuffer)
10 - [Build Onnx](#build-onnx)
11 - [Build TfLite](#build-tflite)
12 - [Build Arm NN](#build-armnn)
13 - [Generate TF Lite Schema](#generate-tflite-schema)
14 - [Build Standalone Sample Dynamic Backend](#build-standalone-sample-dynamic-backend)
15 - [Run Unit Tests](#run-unit-tests)
16 - [Troubleshooting and Errors:](#troubleshooting-and-errors-)
20 These are the step by step instructions on Cross-Compiling Arm NN under an x86_64 system to target an Arm64 Ubuntu Linux system. This build flow has been tested with Ubuntu 18.04 and 20.04 and it depends on the same version of Ubuntu or Debian being installed on both the build host and target machines. The instructions assume you are using a bash shell and show how to build the Arm NN core library, Protobuf, Tflite, Flatbuffer and Compute Libraries.
21 Start by creating a directory to contain all components:
24 mkdir $HOME/armnn-devenv
28 ## Cross-compiling ToolChain
29 * Install the standard cross-compilation libraries for arm64:
31 sudo apt install crossbuild-essential-arm64
35 Cmake 3.19rc3 is required to build TF Lite Delegate.
38 sudo apt-get install libssl-dev
39 wget https://github.com/Kitware/CMake/releases/download/v3.19.0-rc3/cmake-3.19.0-rc3.tar.gz
40 tar -zxvf cmake-3.19.0-rc3.tar.gz
42 ./bootstrap --prefix=$HOME/armnn-devenv/cmake/install
48 ## Build and install Google's Protobuf library
50 We support protobuf version 3.12.0
51 * Get protobuf from here: https://github.com/protocolbuffers/protobuf:
52 (Requires Git if not previously installed: `sudo apt install git`)
54 git clone -b v3.12.0 https://github.com/google/protobuf.git protobuf
56 git submodule update --init --recursive
59 * Build a native (x86_64) version of the protobuf libraries and compiler (protoc):
60 (Requires cUrl, autoconf, llibtool, and other build dependencies if not previously installed: sudo apt install curl autoconf libtool build-essential g++)
64 ../configure --prefix=$HOME/armnn-devenv/google/x86_64_pb_install
68 * Build the arm64 version of the protobuf libraries:
72 CC=aarch64-linux-gnu-gcc \
73 CXX=aarch64-linux-gnu-g++ \
74 ../configure --host=aarch64-linux \
75 --prefix=$HOME/armnn-devenv/google/arm64_pb_install \
76 --with-protoc=$HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc
85 git clone https://github.com/ARM-software/armnn.git
88 * Checkout Arm NN branch:
91 git checkout <branch_name>
94 For example, if you want to check out the 21.11 release branch:
96 git checkout branches/armnn_21_11
100 ## Build Arm Compute Library
101 * Clone Arm Compute Library:
104 cd $HOME/armnn-devenv
105 git clone https://github.com/ARM-software/ComputeLibrary.git
107 * Checkout Arm Compute Library release tag:
110 git checkout <tag_name>
112 Arm NN and Arm Compute Library are developed closely together. If you would like to use the Arm NN 21.11 release you will need the 21.11 release of ACL too. For example, if you want to checkout the 21.11 release tag:
116 Arm NN provides a script that downloads the version of Arm Compute Library that Arm NN was tested with:
118 git checkout $(../armnn/scripts/get_compute_library.sh -p)
120 * Build the Arm Compute Library:
121 (Requires SCons if not previously installed: `sudo apt install scons`)
123 scons arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" -j4
127 * Building Flatbuffer version 1.12.0
129 cd $HOME/armnn-devenv
130 wget -O flatbuffers-1.12.0.tar.gz https://github.com/google/flatbuffers/archive/v1.12.0.tar.gz
131 tar xf flatbuffers-1.12.0.tar.gz
132 cd flatbuffers-1.12.0
136 cmake .. -DFLATBUFFERS_BUILD_FLATC=1 \
137 -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers \
138 -DFLATBUFFERS_BUILD_TESTS=0
142 * Build arm64 version of flatbuffer
147 # Add -fPIC to allow us to use the libraries in shared objects.
148 CXXFLAGS="-fPIC" cmake .. -DCMAKE_C_COMPILER=/usr/bin/aarch64-linux-gnu-gcc \
149 -DCMAKE_CXX_COMPILER=/usr/bin/aarch64-linux-gnu-g++ \
150 -DFLATBUFFERS_BUILD_FLATC=1 \
151 -DCMAKE_INSTALL_PREFIX:PATH=$HOME/armnn-devenv/flatbuffers-arm64 \
152 -DFLATBUFFERS_BUILD_TESTS=0
159 cd $HOME/armnn-devenv
160 git clone https://github.com/onnx/onnx.git
162 git fetch https://github.com/onnx/onnx.git 553df22c67bee5f0fe6599cff60f1afc6748c635 && git checkout FETCH_HEAD
163 LD_LIBRARY_PATH=$HOME/armnn-devenv/google/x86_64_pb_install/lib:$LD_LIBRARY_PATH \
164 $HOME/armnn-devenv/google/x86_64_pb_install/bin/protoc \
165 onnx/onnx.proto --proto_path=. --proto_path=../google/x86_64_pb_install/include --cpp_out $HOME/armnn-devenv/onnx
169 * Arm NN provides a script, armnn/scripts/get_tensorflow.sh, that can be used to check out the version of TensorFlow that Arm NN was tested with:
171 cd $HOME/armnn-devenv
172 git clone https://github.com/tensorflow/tensorflow.git
174 git checkout $(../armnn/scripts/get_tensorflow.sh -p) # Checks out the latest tested version of TF
178 * You will need to download gcc-arm-8.3-2019.03 toolchain and continue building TF Lite as following:
180 curl -LO https://storage.googleapis.com/mirror.tensorflow.org/developer.arm.com/media/Files/downloads/gnu-a/8.3-2019.03/binrel/gcc-arm-8.3-2019.03-x86_64-aarch64-linux-gnu.tar.xz
181 mkdir tflite-toolchains
182 tar xvf gcc-arm-8.3-2019.03-x86_64-aarch64-linux-gnu.tar.xz -C tflite-toolchains
183 mkdir -p tflite/build
185 ARMCC_PREFIX=$HOME/armnn-devenv/tflite-toolchains/gcc-arm-8.3-2019.03-x86_64-aarch64-linux-gnu/bin/aarch64-linux-gnu- \
186 ARMCC_FLAGS="-funsafe-math-optimizations" \
187 cmake -DCMAKE_C_COMPILER=${ARMCC_PREFIX}gcc \
188 -DCMAKE_CXX_COMPILER=${ARMCC_PREFIX}g++ \
189 -DCMAKE_C_FLAGS="${ARMCC_FLAGS}" -DCMAKE_CXX_FLAGS="${ARMCC_FLAGS}" \
190 -DCMAKE_VERBOSE_MAKEFILE:BOOL=ON -DCMAKE_SYSTEM_NAME=Linux \
191 -DTFLITE_ENABLE_XNNPACK=OFF \
192 -DCMAKE_SYSTEM_PROCESSOR=aarch64 \
193 $HOME/armnn-devenv/tensorflow/tensorflow/lite/ \
197 ## Generate TF Lite Schema
199 cd $HOME/armnn-devenv/tflite
200 cp $HOME/armnn-devenv/tensorflow/tensorflow/lite/schema/schema.fbs .
201 ../flatbuffers-1.12.0/build/flatc -c --gen-object-api --reflect-types --reflect-names schema.fbs
205 * Compile Arm NN for arm64:
207 cd $HOME/armnn-devenv/armnn
212 * Use CMake to configure your build environment, update the following script and run it from the armnn/build directory to set up the Arm NN build:
215 CXX=aarch64-linux-gnu-g++ CC=aarch64-linux-gnu-gcc cmake .. \
216 -DARMCOMPUTE_ROOT=$HOME/armnn-devenv/ComputeLibrary \
217 -DARMCOMPUTE_BUILD_DIR=$HOME/armnn-devenv/ComputeLibrary/build/ \
218 -DARMCOMPUTENEON=1 -DARMCOMPUTECL=1 -DARMNNREF=1 \
219 -DONNX_GENERATED_SOURCES=$HOME/armnn-devenv/onnx \
220 -DBUILD_ONNX_PARSER=1 \
221 -DBUILD_TF_LITE_PARSER=1 \
222 -DTENSORFLOW_ROOT=$HOME/armnn-devenv/tensorflow \
223 -DTF_LITE_SCHEMA_INCLUDE_PATH=$WORKING_DIR/tflite \
224 -DFLATBUFFERS_ROOT=$HOME/armnn-devenv/flatbuffers-arm64 \
225 -DFLATC_DIR=$HOME/armnn-devenv/flatbuffers-1.12.0/build \
226 -DPROTOBUF_ROOT=$HOME/armnn-devenv/google/x86_64_pb_install \
227 -DPROTOBUF_LIBRARY_DEBUG=$HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so.23.0.0 \
228 -DPROTOBUF_LIBRARY_RELEASE=$HOME/armnn-devenv/google/arm64_pb_install/lib/libprotobuf.so.23.0.0
231 * If you want to include standalone sample dynamic backend tests, add the argument to enable the tests and the dynamic backend path to the CMake command:
233 -DSAMPLE_DYNAMIC_BACKEND=1 \
234 -DDYNAMIC_BACKEND_PATHS=$SAMPLE_DYNAMIC_BACKEND_PATH
236 * If you want to build Arm NN TF Lite Delegate, add the arguments:
238 -DTFLITE_LIB_ROOT=$HOME/armnn-devenv/tflite/build \
239 -DBUILD_ARMNN_TFLITE_DELEGATE=1
246 ## Build Standalone Sample Dynamic Backend
247 * The sample dynamic backend is located in armnn/src/dynamic/sample
249 cd $HOME/armnn-devenv/armnn/src/dynamic/sample
254 * Use CMake to configure your build environment, update the following script and run it from the armnn/src/dynamic/sample/build directory to set up the Arm NN build:
257 CXX=aarch64-linux-gnu-g++ CC=aarch64-linux-gnu-gcc cmake .. \
258 -DCMAKE_CXX_FLAGS=--std=c++14 \
259 -DARMNN_PATH=$HOME/armnn-devenv/armnn/build/libarmnn.so
268 * Copy the build folder to an arm64 linux machine
269 * Copy the libprotobuf.so.23.0.0 library file to the build folder
270 * If you enable the standalone sample dynamic tests, also copy libArm_SampleDynamic_backend.so library file to the folder specified as $SAMPLE_DYNAMIC_BACKEND_PATH when you build Arm NN
271 * cd to the build folder on your arm64 machine and set your LD_LIBRARY_PATH to its current location:
277 * Create a symbolic link to libprotobuf.so.23.0.0:
280 ln -s libprotobuf.so.23.0.0 ./libprotobuf.so.23
286 LD_LIBRARY_PATH=./:$LD_LIBRARY_PATH ./UnitTests
287 [doctest] doctest version is "2.4.6"
288 [doctest] run with "--help" for options
289 ===============================================================================
290 [doctest] test cases: 4817 | 4817 passed | 0 failed | 0 skipped
291 [doctest] assertions: 807634 | 807634 passed | 0 failed |
292 [doctest] Status: SUCCESS!
295 * Run the Delegate UnitTests:
298 LD_LIBRARY_PATH=./:$LD_LIBRARY_PATH ./delegate/DelegateUnitTests
301 ## Troubleshooting and Errors:
303 * When Cross-compiling Arm NN if you are getting error
305 [ 62%] Generating SchemaText.cpp
306 /bin/sh: 1: xxd: not found
307 make[2]: *** [armnn/CMakeFiles/UnitTests.dir/build.make:63: armnn/SchemaText.cpp] Error 127
308 make[1]: *** [CMakeFiles/Makefile2:674: armnn/CMakeFiles/UnitTests.dir/all] Error 2
309 make[1]: *** Waiting for unfinished jobs....
311 * Missing xxd, this can be fixed by installing it:
313 sudo apt-get install xxd
318 ### Missing libz.so.1
319 * When compiling armNN:
321 /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.23.0.0, not found (try using -rpath or -rpath-link)
324 * Missing arm64 libraries for libz.so.1, these can be added by adding a second architecture to dpkg and explicitly installing them:
326 sudo dpkg --add-architecture arm64
327 sudo apt-get install zlib1g:arm64
331 * If apt-get update returns 404 errors for arm64 repos refer to section 5 below.
332 * Alternatively the missing arm64 version of libz.so.1 can be downloaded and installed from a .deb package here:
333 https://launchpad.net/ubuntu/wily/arm64/zlib1g/1:1.2.8.dfsg-2ubuntu4
335 sudo dpkg -i zlib1g_1.2.8.dfsg-2ubuntu4_arm64.deb
339 ### Unable to install arm64 packages after adding arm64 architecture
340 * 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:
341 * From stackoverflow:
342 https://askubuntu.com/questions/430705/how-to-use-apt-get-to-download-multi-arch-library/430718
343 * Open /etc/apt/sources.list with your preferred text editor.
345 * Mark all the current (default) repos as \[arch=<current_os_arch>], e.g.
347 deb [arch=amd64] http://archive.ubuntu.com/ubuntu/ xenial main restricted
349 * Then add the following:
351 deb [arch=arm64] http://ports.ubuntu.com/ xenial main restricted
352 deb [arch=arm64] http://ports.ubuntu.com/ xenial-updates main restricted
353 deb [arch=arm64] http://ports.ubuntu.com/ xenial universe
354 deb [arch=arm64] http://ports.ubuntu.com/ xenial-updates universe
355 deb [arch=arm64] http://ports.ubuntu.com/ xenial multiverse
356 deb [arch=arm64] http://ports.ubuntu.com/ xenial-updates multiverse
357 deb [arch=arm64] http://ports.ubuntu.com/ xenial-backports main restricted universe multiverse
359 * Update and install again:
361 sudo apt-get install zlib1g:arm64
367 ### Undefined references to google::protobuf:: functions
368 * Missing or out of date protobuf compilation libraries.
369 Use the command 'protoc --version' to check which version of protobuf is available (version 3.12.0 is required).
370 Follow the instructions above to install protobuf 3.12.0
373 ### Errors on strict-aliasing rules when compiling the Compute Library
374 * When compiling the Compute Library there are multiple errors on strict-aliasing rules:
376 cc1plus: error: unrecognized command line option ‘-Wno-implicit-fallthrough’ [-Werror]
378 * Add Werror=0 to the scons command:
380 scons arch=arm64-v8a neon=1 opencl=1 embed_kernels=1 extra_cxx_flags="-fPIC" -j8 Werror=0