Building

https://travis-ci.org/DiffusionMRITool/dmritool.svg?branch=master

Matlab codes

The codes in folder ${DMRITOOL_SOURCE_DIR}/Matlab are purely in matlab. Thus you do not need to build c++ codes to run codes in that folder. The codes in folder ${DMRITOOL_SOURCE_DIR}/Wrapping/Matlab are matlab mex files, which depend on the c++ codes.

See the matlab function list, where these functions starting with mex prefix need to be built using cmake.

Dependent Packages

Required Packages

Optional Packages

  • Qt: needed for building QTApplications.
  • Doxygen: needed when building doxygen documents.

Third Party Packages (included in dmritool)

  • GTest: GTest can be automatically downloaded and built when building tests.
  • SPAMS: included in ThirdParty/Spams

Build Souce Code

Linux and Mac

  1. build/install CMake, GCC, Git, GSL, VTK, ITK, Qt (optional), Doxygen (optional).

  2. build/install OpenBlas + Lapack or MKL .

    MKL is free for students, educators, academic researchers and open source contributors. See the free download link. OpenBlas and Lapack are always free and easy to use.

    If you choose OpenBlas + Lapack, you can use them from system repositories. In this case, you need to set environment variables correctly. See FAQ of OpenBlas.

    export OPENBLAS_NUM_THREADS=1
    

    We suggest that you build OpenBlas from its source codes. Suggested building command for OpenBlas used in dmritool is:

    git clone https://github.com/xianyi/OpenBLAS.git
    cd OpenBlas
    git checkout v0.2.13
    make BINARY=64 USE_THREAD=0 USE_OPENMP=1 cc=gcc  FC=gfortran
    make install
    
  3. build SlicerExecutionModel

    git clone https://github.com/Slicer/SlicerExecutionModel.git
    mkdir SlicerExecutionModel-build
    cd SlicerExecutionModel-build
    ccmake ../SlicerExecutionModel
    make
    
  4. build DMRITool source

    git clone https://github.com/DiffusionMRITool/dmritool
    mkdir dmritool-build
    cd dmritool-build
    ccmake ../dmritool
    make
    
  • When building DMRITool with ccmake, GenerateCLP_DIR needs to be set as SlicerExecutionModel-build/GenerateCLP.
  • The code uses OpenBlas + Lapack if DMRITOOL_USE_MKL is set OFF, it uses MKL if DMRITOOL_USE_MKL is ON.
  • If DMRITOOL_USE_FASTLAPACK=ON, then the codes use fast versions for SVD and eigen-decomposition, but it may have errors for openblas. You can check whether utlVNLLapackGTest and utlVNLBlasGTest can successfully pass. If these two tests do not passed, you may need to build OpenBlas manually or build DMRITool with DMRITOOL_USE_FASTLAPACK=OFF
  • If you want to build matlab mex files, please set DMRITOOL_WRAP_MATLAB=ON.
  • If BUILD_TESTING is set ON, some tests based on GTest will be built. We suggest you set it ON. GTest will be automatically downloaded and built if you do not have it in system.
  • Qt is needed if BUILD_QT_APPLICATIONS is set ON.
  • VERBOSITY_LEVEL is used for debug. The default value is 0. If you set it as 1, the routines can provide more logging information and perform more condition checking, but it may make the built executables a little bit slower.
  • If you use an old version of ITK (<4.9), you may have to set CMAKE_CXX_FLAGS=-fpermissive to build the codes. That is a known issue of ITK with c++11 support, which is solved after ITK 4.9.
  1. build doxygen document (optional)

    make doxygen
    
  2. Set environments.

  • You can add dmritool-build/bin into your PATH to use the binary executables easily.

  • When DMRITOOL_WRAP_MATLAB is set ON, the built mex executables are in dmritool-build/Wrapping/Matlab/bin. Then to use these mex executables in matlab, you need to put the folder into matlab path.

  • For matlab mex files, you also need to set BLAS_VERSION as the openblas lib or mkl lib. If you use mkl, then set blas_version as

    export BLAS_VERSION="MKL_DIR/lib/intel64/libmkl_rt.so"
    

    If you use openblas, then set blas_version as

    export BLAS_VERSION="/usr/local/lib/libopenblas.so"
    
  1. build tests (optional, but suggested)

    make test
    
  • Set BUILD_TESTING ON to build tests.
  • Make sure all tests are successfully passed. Otherwise the routines may not run correctly.

Windows

We did not test the building of DMRITool in windows. You can try it by yourself.