diff --git a/MKLDNN_README.md b/MKLDNN_README.md new file mode 100644 index 000000000000..43cced49ed0d --- /dev/null +++ b/MKLDNN_README.md @@ -0,0 +1,301 @@ +# Build/Install MXNet with MKL-DNN + +Building MXNet with [Intel MKL-DNN](https://github.com/intel/mkl-dnn) will gain better performance when using Intel Xeon CPUs for training and inference. The improvement of performance can be seen in this [page](https://mxnet.incubator.apache.org/faq/perf.html#intel-cpu). Below are instructions for linux, MacOS and Windows platform. + +

Contents

+ +* [1. Linux](#1) +* [2. MacOS](#2) +* [3. Windows](#3) +* [4. Verify MXNet with python](#4) +* [5. Enable MKL BLAS](#5) +* [6. Support](#6) + +

Linux

+ +### Prerequisites + +``` +sudo apt-get update +sudo apt-get install -y build-essential git +sudo apt-get install -y libopenblas-dev liblapack-dev +sudo apt-get install -y libopencv-dev +sudo apt-get install -y graphviz +``` + +### Clone MXNet sources + +``` +git clone --recursive https://github.com/apache/incubator-mxnet.git +cd incubator-mxnet +``` + +### Build MXNet with MKL-DNN + +``` +make -j $(nproc) USE_OPENCV=1 USE_MKLDNN=1 USE_BLAS=mkl USE_INTEL_PATH=/opt/intel +``` + +If you don't have full [MKL](https://software.intel.com/en-us/intel-mkl) library installed, you can use OpenBLAS by setting `USE_BLAS=openblas`. + +

MacOS

+ +### Prerequisites + +Install the dependencies, required for MXNet, with the following commands: + +- [Homebrew](https://brew.sh/) +- gcc (clang in macOS does not support OpenMP) +- OpenCV (for computer vision operations) + +``` +# Paste this command in Mac terminal to install Homebrew +/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)" + +# install dependency +brew update +brew install pkg-config +brew install graphviz +brew tap homebrew/core +brew install opencv +brew tap homebrew/versions +brew install gcc49 +brew link gcc49 #gcc-5 and gcc-7 also work +``` + +### Clone MXNet sources + +``` +git clone --recursive https://github.com/apache/incubator-mxnet.git +cd incubator-mxnet +``` + +### Enable OpenMP for MacOS + +If you want to enable OpenMP for better performance, you should modify the Makefile in MXNet root dictionary: + +Add CFLAGS '-fopenmp' for Darwin. + +``` +ifeq ($(USE_OPENMP), 1) +# ifneq ($(UNAME_S), Darwin) + CFLAGS += -fopenmp +# endif +endif +``` + +### Build MXNet with MKL-DNN + +``` +make -j $(sysctl -n hw.ncpu) CC=gcc-4.9 CXX=g++-4.9 USE_OPENCV=0 USE_OPENMP=1 USE_MKLDNN=1 USE_BLAS=apple USE_PROFILER=1 +``` + +*Note: Temporarily disable OPENCV.* + +

Windows

+ +We recommend to build and install MXNet yourself using [Microsoft Visual Studio 2015](https://www.visualstudio.com/vs/older-downloads/), or you can also try experimentally the latest [Microsoft Visual Studio 2017](https://www.visualstudio.com/downloads/). + +**Visual Studio 2015** + +To build and install MXNet yourself, you need the following dependencies. Install the required dependencies: + +1. If [Microsoft Visual Studio 2015](https://www.visualstudio.com/vs/older-downloads/) is not already installed, download and install it. You can download and install the free community edition. +2. Download and Install [CMake 3](https://cmake.org/) if it is not already installed. +3. Download and install [OpenCV 3](http://sourceforge.net/projects/opencvlibrary/files/opencv-win/3.0.0/opencv-3.0.0.exe/download). +4. Unzip the OpenCV package. +5. Set the environment variable ```OpenCV_DIR``` to point to the ```OpenCV build directory``` (```C:\opencv\build\x64\vc14``` for example). Also, you need to add the OpenCV bin directory (```C:\opencv\build\x64\vc14\bin``` for example) to the ``PATH`` variable. +6. If you have Intel Math Kernel Library (MKL) installed, set ```MKL_ROOT``` to point to ```MKL``` directory that contains the ```include``` and ```lib```. If you want to use MKL blas, you should set ```-DUSE_BLAS=mkl``` when cmake. Typically, you can find the directory in +```C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2018\windows\mkl```. +7. If you don't have the Intel Math Kernel Library (MKL) installed, download and install [OpenBLAS](http://sourceforge.net/projects/openblas/files/v0.2.14/). Note that you should also download ```mingw64.dll.zip`` along with openBLAS and add them to PATH. +8. Set the environment variable ```OpenBLAS_HOME``` to point to the ```OpenBLAS``` directory that contains the ```include``` and ```lib``` directories. Typically, you can find the directory in ```C:\Program files (x86)\OpenBLAS\```. + +After you have installed all of the required dependencies, build the MXNet source code: + +1. Download the MXNet source code from [GitHub](https://github.com/apache/incubator-mxnet). Don't forget to pull the submodules: +``` +git clone --recursive https://github.com/apache/incubator-mxnet.git +``` + +2. Copy file `3rdparty/mkldnn/config_template.vcxproj` to incubator-mxnet root. + +3. Start a Visual Studio command prompt. + +4. Use [CMake 3](https://cmake.org/) to create a Visual Studio solution in ```./build``` or some other directory. Make sure to specify the architecture in the +[CMake 3](https://cmake.org/) command: +``` +mkdir build +cd build +cmake -G "Visual Studio 14 Win64" .. -DUSE_CUDA=0 -DUSE_CUDNN=0 -DUSE_NVRTC=0 -DUSE_OPENCV=1 -DUSE_OPENMP=1 -DUSE_PROFILER=1 -DUSE_BLAS=open -DUSE_LAPACK=1 -DUSE_DIST_KVSTORE=0 -DCUDA_ARCH_NAME=All -DUSE_MKLDNN=1 -DCMAKE_BUILD_TYPE=Release +``` + +5. In Visual Studio, open the solution file,```.sln```, and compile it. +These commands produce a library called ```libmxnet.dll``` in the ```./build/Release/``` or ```./build/Debug``` folder. +Also ```libmkldnn.dll``` with be in the ```./build/3rdparty/mkldnn/src/Release/``` + +6. Make sure that all the dll files used above(such as `libmkldnn.dll`, `libmklml.dll`, `libiomp5.dll`, `libopenblas.dll`, etc) are added to the system PATH. For convinence, you can put all of them to ```\windows\system32```. Or you will come across `Not Found Dependencies` when loading mxnet. + +**Visual Studio 2017** + +To build and install MXNet yourself using [Microsoft Visual Studio 2017](https://www.visualstudio.com/downloads/), you need the following dependencies. Install the required dependencies: + +1. If [Microsoft Visual Studio 2017](https://www.visualstudio.com/downloads/) is not already installed, download and install it. You can download and install the free community edition. +2. Download and install [CMake 3](https://cmake.org/files/v3.11/cmake-3.11.0-rc4-win64-x64.msi) if it is not already installed. +3. Download and install [OpenCV](https://sourceforge.net/projects/opencvlibrary/files/opencv-win/3.4.1/opencv-3.4.1-vc14_vc15.exe/download). +4. Unzip the OpenCV package. +5. Set the environment variable ```OpenCV_DIR``` to point to the ```OpenCV build directory``` (e.g., ```OpenCV_DIR = C:\utils\opencv\build```). +6. If you don’t have the Intel Math Kernel Library (MKL) installed, download and install [OpenBlas](https://sourceforge.net/projects/openblas/files/v0.2.20/OpenBLAS%200.2.20%20version.zip/download). +7. Set the environment variable ```OpenBLAS_HOME``` to point to the ```OpenBLAS``` directory that contains the ```include``` and ```lib``` directories (e.g., ```OpenBLAS_HOME = C:\utils\OpenBLAS```). + +After you have installed all of the required dependencies, build the MXNet source code: + +1. Start ```cmd``` in windows. + +2. Download the MXNet source code from GitHub by using following command: + +```r +cd C:\ +git clone --recursive https://github.com/apache/incubator-mxnet.git +``` + +3. Copy file `3rdparty/mkldnn/config_template.vcxproj` to incubator-mxnet root. + +4. Follow [this link](https://docs.microsoft.com/en-us/visualstudio/install/modify-visual-studio) to modify ```Individual components```, and check ```VC++ 2017 version 15.4 v14.11 toolset```, and click ```Modify```. + +5. Change the version of the Visual studio 2017 to v14.11 using the following command (by default the VS2017 is installed in the following path): + +```r +"C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Auxiliary\Build\vcvars64.bat" -vcvars_ver=14.11 +``` + +6. Create a build dir using the following command and go to the directory, for example: + +```r +mkdir C:\build +cd C:\build +``` + +7. CMake the MXNet source code by using following command: + +```r +cmake -G "Visual Studio 15 2017 Win64" .. -T host=x64 -DUSE_CUDA=0 -DUSE_CUDNN=0 -DUSE_NVRTC=0 -DUSE_OPENCV=1 -DUSE_OPENMP=1 -DUSE_PROFILER=1 -DUSE_BLAS=open -DUSE_LAPACK=1 -DUSE_DIST_KVSTORE=0 -DCUDA_ARCH_NAME=All -DUSE_MKLDNN=1 -DCMAKE_BUILD_TYPE=Release +``` + +8. After the CMake successfully completed, compile the the MXNet source code by using following command: + +```r +msbuild mxnet.sln /p:Configuration=Release;Platform=x64 /maxcpucount +``` + +9. Make sure that all the dll files used above(such as `libmkldnn.dll`, `libmklml.dll`, `libiomp5.dll`, `libopenblas.dll`, etc) are added to the system PATH. For convinence, you can put all of them to ```\windows\system32```. Or you will come across `Not Found Dependencies` when loading mxnet. + +

Verify MXNet with python

+ +``` +cd python +sudo python setup.py install +python -c "import mxnet as mx;print((mx.nd.ones((2, 3))*2).asnumpy());" + +Expected Output: + +[[ 2. 2. 2.] + [ 2. 2. 2.]] +``` + +### Verify whether MKL-DNN works + +After MXNet is installed, you can verify if MKL-DNN backend works well with a single Convolution layer. + +``` +import mxnet as mx +import numpy as np + +num_filter = 32 +kernel = (3, 3) +pad = (1, 1) +shape = (32, 32, 256, 256) + +x = mx.sym.Variable('x') +w = mx.sym.Variable('w') +y = mx.sym.Convolution(data=x, weight=w, num_filter=num_filter, kernel=kernel, no_bias=True, pad=pad) +exe = y.simple_bind(mx.cpu(), x=shape) + +exe.arg_arrays[0][:] = np.random.normal(size=exe.arg_arrays[0].shape) +exe.arg_arrays[1][:] = np.random.normal(size=exe.arg_arrays[1].shape) + +exe.forward(is_train=False) +o = exe.outputs[0] +t = o.asnumpy() +``` + +You can open the `MKLDNN_VERBOSE` flag by setting environment variable: +``` +export MKLDNN_VERBOSE=1 +``` +Then by running above code snippet, you probably will get the following output message which means `convolution` and `reorder` primitive from MKL-DNN are called. Layout information and primitive execution performance are also demonstrated in the log message. +``` +mkldnn_verbose,exec,reorder,jit:uni,undef,in:f32_nchw out:f32_nChw16c,num:1,32x32x256x256,6.47681 +mkldnn_verbose,exec,reorder,jit:uni,undef,in:f32_oihw out:f32_OIhw16i16o,num:1,32x32x3x3,0.0429688 +mkldnn_verbose,exec,convolution,jit:avx512_common,forward_inference,fsrc:nChw16c fwei:OIhw16i16o fbia:undef fdst:nChw16c,alg:convolution_direct,mb32_g1ic32oc32_ih256oh256kh3sh1dh0ph1_iw256ow256kw3sw1dw0pw1,9.98193 +mkldnn_verbose,exec,reorder,jit:uni,undef,in:f32_oihw out:f32_OIhw16i16o,num:1,32x32x3x3,0.0510254 +mkldnn_verbose,exec,reorder,jit:uni,undef,in:f32_nChw16c out:f32_nchw,num:1,32x32x256x256,20.4819 +``` + +

Enable MKL BLAS

+ +To make it convenient for customers, Intel introduced a new license called [Intel® Simplified license](https://software.intel.com/en-us/license/intel-simplified-software-license) that allows to redistribute not only dynamic libraries but also headers, examples and static libraries. + +Installing and enabling the full MKL installation enables MKL support for all operators under the linalg namespace. + + 1. Download and install the latest full MKL version following instructions on the [intel website.](https://software.intel.com/en-us/mkl) + + 2. Run `make -j ${nproc} USE_BLAS=mkl` + + 3. Navigate into the python directory + + 4. Run `sudo python setup.py install` + +### Verify whether MKL works + +After MXNet is installed, you can verify if MKL BLAS works well with a single dot layer. + +``` +import mxnet as mx +import numpy as np + +shape_x = (1, 10, 8) +shape_w = (1, 12, 8) + +x_npy = np.random.normal(0, 1, shape_x) +w_npy = np.random.normal(0, 1, shape_w) + +x = mx.sym.Variable('x') +w = mx.sym.Variable('w') +y = mx.sym.batch_dot(x, w, transpose_b=True) +exe = y.simple_bind(mx.cpu(), x=x_npy.shape, w=w_npy.shape) + +exe.forward(is_train=False) +o = exe.outputs[0] +t = o.asnumpy() +``` + +You can open the `MKL_VERBOSE` flag by setting environment variable: +``` +export MKL_VERBOSE=1 +``` +Then by running above code snippet, you probably will get the following output message which means `SGEMM` primitive from MKL are called. Layout information and primitive execution performance are also demonstrated in the log message. +``` +Numpy + Intel(R) MKL: THREADING LAYER: (null) +Numpy + Intel(R) MKL: setting Intel(R) MKL to use INTEL OpenMP runtime +Numpy + Intel(R) MKL: preloading libiomp5.so runtime +MKL_VERBOSE Intel(R) MKL 2018.0 Update 1 Product build 20171007 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions 512 (Intel(R) AVX-512) enabled processors, Lnx 2.40GHz lp64 intel_thread NMICDev:0 +MKL_VERBOSE SGEMM(T,N,12,10,8,0x7f7f927b1378,0x1bc2140,8,0x1ba8040,8,0x7f7f927b1380,0x7f7f7400a280,12) 8.93ms CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:40 WDiv:HOST:+0.000 +``` + +

Next Steps and Support

+ +- For questions or support specific to MKL, visit the [Intel MKL](https://software.intel.com/en-us/mkl) + +- For questions or support specific to MKL, visit the [Intel MKLDNN](https://github.com/intel/mkl-dnn) + +- If you find bugs, please open an issue on GitHub for [MXNet with MKL](https://github.com/apache/incubator-mxnet/labels/MKL) or [MXNet with MKLDNN](https://github.com/apache/incubator-mxnet/labels/MKLDNN) diff --git a/MKL_README.md b/MKL_README.md deleted file mode 100644 index a5c63b097c5e..000000000000 --- a/MKL_README.md +++ /dev/null @@ -1,77 +0,0 @@ -## Build/Install MXNet with a full MKL installation: - -To make it convenient for customers, Intel introduced a new license called [Intel® Simplified license](https://software.intel.com/en-us/license/intel-simplified-software-license) that allows to redistribute not only dynamic libraries but also headers, examples and static libraries. - -Installing and enabling the full MKL installation enables MKL support for all operators under the linalg namespace. - - 1. Download and install the latest full MKL version following instructions on the [intel website.](https://software.intel.com/en-us/mkl) - - 2. Run 'make -j ${nproc} USE_BLAS=mkl' - - 3. Navigate into the python directory - - 4. Run 'sudo python setup.py install' - - -## Build/Install MXNet with MKLDNN on Windows: - -To build and install MXNet yourself, you need the following dependencies. Install the required dependencies: - -1. If [Microsoft Visual Studio 2015](https://www.visualstudio.com/vs/older-downloads/) is not already installed, download and install it. You can download and install the free community edition. -2. Download and Install [CMake](https://cmake.org/) if it is not already installed. -3. Download and install [OpenCV](http://sourceforge.net/projects/opencvlibrary/files/opencv-win/3.0.0/opencv-3.0.0.exe/download). -4. Unzip the OpenCV package. -5. Set the environment variable ```OpenCV_DIR``` to point to the ```OpenCV build directory``` (```C:\opencv\build\x64\vc14``` for example). Also, you need to add the OpenCV bin directory (```C:\opencv\build\x64\vc14\bin``` for example) to the ``PATH`` variable. -6. If you have Intel Math Kernel Library (MKL) installed, set ```MKL_ROOT``` to point to ```MKL``` directory that contains the ```include``` and ```lib```. If you want to use MKL blas, you should set ```-DUSE_BLAS=mkl``` when cmake. Typically, you can find the directory in -```C:\Program Files (x86)\IntelSWTools\compilers_and_libraries_2018\windows\mkl```. -7. If you don't have the Intel Math Kernel Library (MKL) installed, download and install [OpenBLAS](http://sourceforge.net/projects/openblas/files/v0.2.14/). Note that you should also download ```mingw64.dll.zip`` along with openBLAS and add them to PATH. -8. Set the environment variable ```OpenBLAS_HOME``` to point to the ```OpenBLAS``` directory that contains the ```include``` and ```lib``` directories. Typically, you can find the directory in ```C:\Program files (x86)\OpenBLAS\```. - -After you have installed all of the required dependencies, build the MXNet source code: - -1. Download the MXNet source code from [GitHub](https://github.com/apache/incubator-mxnet). Don't forget to pull the submodules: -``` - git clone https://github.com/apache/incubator-mxnet.git --recursive -``` - -2. Copy file `3rdparty/mkldnn/config_template.vcxproj` to incubator-mxnet root. - -3. Start a Visual Studio command prompt. - -4. Use [CMake](https://cmake.org/) to create a Visual Studio solution in ```./build``` or some other directory. Make sure to specify the architecture in the -[CMake](https://cmake.org/) command: -``` - mkdir build - cd build - cmake -G "Visual Studio 14 Win64" .. -DUSE_CUDA=0 -DUSE_CUDNN=0 -DUSE_NVRTC=0 -DUSE_OPENCV=1 -DUSE_OPENMP=1 -DUSE_PROFILER=1 -DUSE_BLAS=open -DUSE_LAPACK=1 -DUSE_DIST_KVSTORE=0 -DCUDA_ARCH_NAME=All -DUSE_MKLDNN=1 -DCMAKE_BUILD_TYPE=Release -``` - -5. In Visual Studio, open the solution file,```.sln```, and compile it. -These commands produce a library called ```libmxnet.dll``` in the ```./build/Release/``` or ```./build/Debug``` folder. -Also ```libmkldnn.dll``` with be in the ```./build/3rdparty/mkldnn/src/Release/``` - -6. Make sure that all the dll files used above(such as `libmkldnn.dll`, `libmklml.dll`, `libiomp5.dll`, `libopenblas.dll`, etc) are added to the system PATH. For convinence, you can put all of them to ```\windows\system32```. Or you will come across `Not Found Dependencies` when loading mxnet. - -## Install MXNet for Python - -1. Install ```Python``` using windows installer available [here](https://www.python.org/downloads/release/python-2712/). -2. Install ```Numpy``` using windows installer available [here](http://scipy.org/install.html). -3. Next, we install Python package interface for MXNet. You can find the Python interface package for [MXNet on GitHub](https://github.com/dmlc/mxnet/tree/master/python/mxnet). - -```CMD - cd python - python setup.py install -``` -Done! We have installed MXNet with Python interface. Run below commands to verify our installation is successful. -```CMD - # Open Python terminal - python - - # You should be able to import mxnet library without any issues. - >>> import mxnet as mx; - >>> a = mx.nd.ones((2, 3)); - >>> print ((a*2).asnumpy()); - [[ 2. 2. 2.] - [ 2. 2. 2.]] -``` -We actually did a small tensor computation using MXNet! You are all set with MKLDNN MXNet on your Windows machine.