mingxiaoh/chainer_for_rebase

iDeep: Intel Deep Learning Extension Package

Intel Deep Learning Extension Package is a module for collection of accelerated deep learning operations like convolution, deconvolution, relu etc. It uses Intel MKL-DNN as acceleration engine.

Requirements

This preview version of iDeep is tested on Ubuntu 16.04 and OS X.

Minimum requirements:

• Cmake3
• GCC 5.3+
• C++ compiler with C++11 standard support
• MKL-DNN 0.1+
• Python 2.7.6+, 3.5.2+, 3.6.0+
• Numpy 1.13
• Swig 3.0.12
• Doxygen 1.8.5

Other requirements:

• Testing utilities
  • Gtest
  • pytest

Installation

iDeep Cpp API

Head file mode to introduce iDeep Cpp APIs:

#include "ideep.hpp"

Pin singleton head file to one Cpp file of your project to instance iDeep singletons.

@@ main.cc
#include "ideep_pin_singletons.hpp"

Install iDeep python package

If you use old setuptools, upgrade it:

pip install -U setuptools

Install iDeep python package(ideep4py) from the source code:

git submodule update --init && mkdir build && cmake .. 
python setup.py install

More information

• MKL-DNN github: https://github.com/01org/mkl-dnn
• iDeep github: https://github.com/intel/ideep.git
• Chainer github: https://github.com/chainer/chainer

License

MIT License (see LICENSE file).