This project using TVM to deploy the mnist module on pc or arm device.
x64
- in
mnist_tvmfolder createbuild_x64and enter into it
$ mkdir build_x64
$ cd build_x64- configure build environment
set the CMAKE_SYSTEM_PROCESSOR=x64to link the x64 runtime dynamic library.
$ cmake .. -DCMAKE_SYSTEM_PROCESSOR=x64- build the source file
this step will generate mnist_test executable file.
$ make -j$(nproc)armv8
- install cross-compile for
aarch64
$ sudo apt-get update
$ sudo apt-get install gcc-aarch64-linux-gnu g++-aarch64-linux-gnu- in
mnist_tvmfolder createbuild_armv8and enter into it.
$ mkdir build_armv8
$ cd build_armv8- configure build environment
$ cmake .. \
-DCMAKE_SYSTEM_NAME=Linux \
-DCMAKE_SYSTEM_VERSION=1 \
-DCMAKE_SYSTEM_PROCESSOR=armv8 \
-DCMAKE_C_COMPILER=/usr/bin/aarch64-linux-gnu-gcc \
-DCMAKE_CXX_COMPILER=/usr/bin/aarch64-linux-gnu-g++ \
-DCMAKE_FIND_ROOT_PATH=/usr/aarch64-linux-gnu \
-DCMAKE_FIND_ROOT_PATH_MODE_PROGRAM=NEVER \
-DCMAKE_FIND_ROOT_PATH_MODE_LIBRARY=ONLY \
-DCMAKE_FIND_ROOT_PATH_MODE_INCLUDE=BOTH \- build the project
$ make -j$(nproc)when running the executable file,three parameter need to be input.
executer [image path] [module dynamic lib path] [module parameter path]when using opencl deploy the module, cpuas host send the data to device. so we need using TVMArrayCopyFromBytesand TVMArrayCopyToBytesto realize memory copy.
- Compilation Tool
- GCC
- CMAKE
- 3rdparty_lib
- ffmpeg
- GTK + 2.x
- libav : libavcodec-dev libavformat-dev libswscale-dev
- optional package
-
libtbb2 libtbb-dev
-
libjpeg-dev, libpng-dev, libtiff-dev, libjasper-dev, libdc1394-22-dev
-
install third-party-lib
$ sudo apt-get install libavcodec-dev libavformat-dev libswscale-dev libv4l-dev libxvidcore-dev libx264-dev libatlas-base-dev gfortran libgtk2.0-dev
$ cmake -D CMAKE_CXX_COMPILER=/usr/bin/arm-linux-gnueabihf-g++ \
-D CMAKE_C_COMPILER=/usr/bin/arm-linux-gnueabihf-gcc \
-D CMAKE_BUILD_TYPE=RELEASE \
-D CMAKE_INSTALL_PREFIX=/usr/local \
-D INSTALL_C_EXAMPLES=ON \
-D INSTALL_PYTHON_EXAMPLES=ON \
-D OPENCV_GENERATE_PKGCONFIG=ON \
-D OPENCV_EXTRA_MODULES_PATH=~/opencv_build/opencv_contrib/modules \
-D BUILD_EXAMPLES=ON ..- GCC install
- ARM32
$ sudo apt-get install g++-arm-linux-gnueabihf gcc-arm-linux-gnueabihf- ARM64
$ sudo apt-get install g++-aarch64-linux-gnu gcc-aarch64-linux-gnuhf: Hard float
base on the actual processor select whether surport hf.
- Using default
cmake
in the folder opencv/platforms/linux, find aarch64-gnu.toolchain.cmake and arm-gnueabi.toolchain.cmake,
- ARMv7
$ cmake -D CMAKE_TOOLCHAIN_FILE="/path/to/opencv/platforms/linux/arm-gnueabi.toolchain.cmake" ..- ARMv8
$ cmake -D CMAKE_TOOLCHAIN_FILE="/path/to/opencv/platforms/linux/aarch64-gnu.toolchain.cmake" ..- System Name
| Linux | Android | QNX | WindowsCE | WindowsPhone | Windows10 |
- System Processor
- arm
- x86
- compiler error
- Disable
BUILD_opencv_freetype=OFFcan solve follow error.
/usr/lib/gcc-cross/aarch64-linux-gnu/7/../../../../aarch64-linux-gnu/bin/ld: cannot find -lfreetype
/usr/lib/gcc-cross/aarch64-linux-gnu/7/../../../../aarch64-linux-gnu/bin/ld: cannot find -lharfbuzz
- add
freetypelib using cross compiler tool