Skip to content

Latest commit

 

History

History
182 lines (141 loc) · 6.64 KB

Build.md

File metadata and controls

182 lines (141 loc) · 6.64 KB
Raw

Table of Contents generated with DocToc

#To build

Build Status

Build options

  • If you want to be able to read training/testing data from jpeg files, then please choose BUILD_JPEG_SUPPORT = ON. You will need to provide turbojpeg library and headers, or compatible. Otherwise set to OFF

linux

Pre-requisites

Required:

  • git
  • make
  • cmake
  • cmake-curses-gui
  • gfortran
  • g++ (should support c++11; eg 4.6 or better)
  • An OpenCL-compatible driver installed, and OpenCL-compatible GPU

Optional:

  • libjpeg62 or compatible, eg sudo apt-get install libjpeg-turbo8-dev (libjpeg-turbo is faster than original libjpeg6.2, by around 2-4 times, because it uses SIMD extensions)

Procedure

git clone --recursive https://github.com/hughperkins/DeepCL.git
cd DeepCL
# if you need to use a specific branch, then choose that now, ie:
#    git checkout some-branch-name
#    git submodule update --recursive
mkdir build
cd build
ccmake ..
# in ccmake:
# - press 'c'/configure
# - choose the options you want
# - press 'c' /configure again
# - press 'g' / generate, then `q` / quit
make -j 4 install

The outputs will appear in subdirectories of ../dist

Note:

  • be sure to add --recursive when you clone, else when you build it will complain about OpenCLHelper missing (or clew missing)
    • if you do forget, you can experiment with running git submodule init --recursive, and then git submodule update --recursive
  • you might need to play around with commands such as git submodule update --recursive occasionally, to pull down new OpenCLHelper updates
  • note: recently, moved EasyCL/thirdparty/clew from submodule to simply copying in the files
    • hopefully this makes new clones easier, but for now, if you already have a clone, when you next update, you might need to first remove the EasyCL/thirdparty/clew directory

To activate, setup environment:

Open a bash prompt, and run:

source /path/to/DeepCL/dist/bin/activate.sh

(where you need to modify /path/to/DeepCL appropriately)

Keep the bash open, and go to the next section

To check all is working

Unit-tests:

deepcl_unittests

Most tests should pass, but one or two might fail. Please do feel free to raise an issue for failing tests, even if they fail intermittently.

Commandline training:

deepcl_train numtest=-1 numtrain=10000 datadir=/data/mnist

(change path to wherever the mnist data files are downloaded)

Windows

Pre-requisites

Required:

  • git
  • cmake
  • Visual Studio 2015
  • An OpenCL-compatible driver installed, and OpenCL-compatible GPU

Optional:

Procedure

  • in git, do git clone --recursive https://github.com/hughperkins/DeepCL.git
    • note: recently, moved EasyCL/thirdparty/clew from submodule to simply copying in the files
    • hopefully this makes new clones easier, but for now, if you already have a clone, when you next update, you might need to first remove the EasyCL/thirdparty/clew directory
  • if you need to use a specific branch, then choose that now, ie:
    • git checkout some-branch-name, and then
    • git submodule update --recursive
  • create a subdirectory build in the git cloned DeepCL directory
  • open cmake, point at the DeepCL directory, and set to build in the build subdirectory
    • configure, select 'visual studio 2015' (or as appropriate)
  • choose the options you want, eg turn python on/off, jpeg on/off
  • click generate
  • open visual studio, and load any of the projects in the build directory
    • change release type to Release
    • choose build from the build menu
  • select 'INSTALL' project, right-click and 'Build'

The outputs will appear in the subdirectory 'dist'

To activate, setup environment:

Open a cmd prompt, and run:

call \path\to\DeepCL\dist\bin\activate.bat

(where you need to modify \path\to\DeepCL appropriately)

Keep the cmd open, and go to the next section

To check all is working

First open a cmd prompt, and activate, as above, then:

Unit-tests:

deepcl_unittests

Most tests should pass, but one or two might fail. Please do feel free to raise an issue for failing tests, even if they fail intermittently.

Commandline training:

deepcl_train numtest=-1 numtrain=10000 datadir=c:\data\mnist

(change path to wherever the mnist data files are downloaded)

Linking

If you want to use the DeepCL library from C++, you will need to link with the following libraries:

  • libDeepCL.so (or DeepCL.dll, on Windows)
  • libEasyCL.so (or EasyCL.dll, on Windows)
  • libclew.so / clew.dll
  • libclBLAS.so / clBLAS.dll

Some errors, and possible causes

  • during build, fatal error: CppRuntimeBoundary.h: No such file or directory
    • make sure you ran source ../dist/bin/activate.sh, or similar (see above for exact command similar to this)

Building libjpeg

Using instructoins at https://stackoverflow.com/questions/12652178/compiling-libjpeg/19045485#19045485

  • downloaded http://www.ijg.org/files/jpegsr9b.zip , unzipped
  • downloaded http://www.bvbcode.com/code/f2kivdrh-395674-down , copied to unzipped jpeg-9b directory
  • open cmd
  • run call "C:\Program Files (x86)\Microsoft Visual Studio 14.0\VC\bin\vcvars32.bat"
  • run nmake /f makefile.vc setup-v10
  • open jpeg.sln in msvc2015, and build
  • (dont need apps.sln, ignore it)
  • you've built win32, into release directory
    • copy jpeg.lib somewhere
  • now, redo, but selecting x64 in the dropdown, in the menubar, at hte top of msvc ui
    • you have to do something like right-click, 'add configuration', and select architecture 'x64' to do this
    • this will build into x64\release, instead of simply release

This builds as a static library, so no need to distribute dlls and stuff