AndaJ / easydk

easy development kit

EN|CN

Cambricon^® Easy Development Kit

Cambricon^® Easy Development Kit is a toolkit, which aims at helping with developing software on Cambricon MLU270/MLU220/MLU370 platform.

Toolkit provides following modules:

• Device: MLU device context operation
• EasyCodec: easy decode and encode on MLU
• EasyInfer: easy inference accelerator on MLU
• EasyTrack: easy track, including feature match track and kcf track
• EasyBang: easy Bang operator

Besides, toolkit contains InferServer component, which aims at simplify developing and deploying High-performance deep learning applications on Cambricon MLU270/MLU220/MLU370 platform. InferServer provides APIs for inference and it provides functions like model loading and management, inference task scheduling and so on.

InferServer contains the following user APIs:

• Model: model loading and management
• Processor: backend processing unit, supports customization
• InferServer: executes inference tasks

Cambricon Dependencies

You can find the cambricon dependencies, including headers and libraries, in the neuware home (installed in /usr/local/neuware by default).

Quick Start

This section introduces how to quickly build instructions on EasyDK and how to develop your own applications based on EasyDK.

Required environments

Before building instructions, you need to install the following software:

• cmake 2.8.7+
• GCC 4.8.5+
• GLog 0.3.4
• Cambricon Neuware Toolkit >= 1.7.0
• CNCV >= 0.4.0 (optional)
• libcurl-dev (optional)

Using the Magicmind backend has the following additional dependencies:

• Cambricon Neuware Toolkit >= 2.4.2
• Magicmind >= 0.5.0

(Magicmind runtime library introduces dependencies on CNNL, CNNL_extra, CNLight. For version requirements, please see Magicmind User Manual)

For decoding or Encoding on MLU370 platform, there are some additional dependencies:

• Cambricon Neuware Toolkit >= 2.6.2
• CNCodec_v3 >= 0.8.2

samples & tests dependencies:

• OpenCV 2.4.9+
• GFlags 2.1.2
• FFmpeg 2.8 3.4 4.2

Ubuntu or Debian

If you are using Ubuntu or Debian, run the following commands:

sudo apt install libgoogle-glog-dev cmake
# optional dependencies
sudo apt install curl libcurl4-openssl-dev
# samples dependencies
sudo apt install libgflags-dev libopencv-dev

Centos

If you are using CentOS, run the following commands:

sudo yum install glog.x86_64 cmake3.x86_64
# optional dependencies
sudo yum install curl libcurl-devel
# samples dependencies
sudo yum install gflags.x86_64 opencv-devel.x86_64 ffmpeg ffmpeg-devel

Build Instructions Using CMake

After finished prerequisite, you can build instructions with the following steps:

1. Run the following command to create a directory for saving the output.

   mkdir build       # Create a directory to save the output.

   A Makefile will be generated in the build folder.

2. Run the following command to generate a script for building instructions.

   cd build
   cmake ${EASYDK_DIR}  # Generate native build scripts.

   Cambricon easydk provides a CMake script (CMakeLists.txt) to build instructions. You can download CMake for free from http://www.cmake.org/.

   ${EASYDK_DIR} specifies the directory where easydk saves for.

   cmake option	range	default	description
   BUILD_SAMPLES	ON / OFF	OFF	build with samples
   BUILD_TESTS	ON / OFF	OFF	build with tests
   WITH_CODEC	ON / OFF	ON	build codec
   WITH_INFER	ON / OFF	ON	build infer
   WITH_TRACKER	ON / OFF	ON	build tracker
   WITH_BANG	ON / OFF	ON	build bang
   WITH_BACKWARD	ON / OFF	ON	build backward
   WITH_TURBOJPEG	ON / OFF	OFF	build with turbo-jpeg
   ENABLE_KCF	ON / OFF	OFF	build with KCF track
   WITH_INFER_SERVER	ON / OFF	ON	build infer-server
   CNIS_WITH_CONTRIB	ON / OFF	ON	build infer-server contrib
   CNIS_RECORD_PERF	ON / OFF	ON	enable recording infer-server perf info
   CNIS_WITH_CURL	ON / OFF	ON	enable curl, to download model from web
   CNIS_USE_MAGICMIND	ON / OFF	OFF	enable infer-server with magicmind backend
   CNIS_WITH_PYTHON_API	ON / OFF	OFF	enable infer-server python api
   SANITIZE_MEMORY	ON / OFF	OFF	check memory
   SANITIZE_ADDRESS	ON / OFF	OFF	check address
   SANITIZE_THREAD	ON / OFF	OFF	check thread
   SANITIZE_UNDEFINED	ON / OFF	OFF	check undefined behavior

   Example:

   cd build
   # build without samples and tests
   cmake ${EASYDK_DIR}      \
        -DBUILD_SAMPLES=ON  \
        -DBUILD_TESTS=ON

3. Run the following command to build instructions:

   make

4. After compilation, the library files are stored in ${EASYDK_DIR}/lib ; the header filers are stored in ${EASYDK_DIR}/include; the header files of InferServer are stores in ${EASYDK_DIR}/infer_server/include .

5. To cross-compile EasyDK, please cross-compile and install third-party dependencies in advance and config CMAKE_TOOLCHAIN_FILE file. For example, on MLU220-SOM platform:

   export NEUWARE_HOME=/your/path/to/neuware
   export PATH=$PATH:/your/path/to/cross-compiler/bin
   cmake ${EASYDK_DIR} -DCMAKE_FIND_ROOT_PATH=/your/path/to/3rdparty-libraries-install-path -DCMAKE_TOOLCHAIN_FILE=${EASYDK_DIR}/cmake/cross-compile.cmake  -DCNIS_WITH_CURL=OFF

   A compilation package is provided at http://video.cambricon.com/models/edge.tar.gz. After decompressing the package, follows the steps in README to compile.

Documentation

Cambricon Forum Docs

For more details, please refer to the documentation on the webpage, including how to use EasyDK, introduction to EasyDK modules and sample codes.

Besides, please refer to the EasyDK-based application porting tutorial , latest user guide and other versions user guide can be found here .