Skip to content

Latest commit

 

History

History
executable file
·
102 lines (87 loc) · 4.87 KB

INSTALL.md

File metadata and controls

executable file
·
102 lines (87 loc) · 4.87 KB

Installtion of gSMFRETda

Although gSMFRETda is coded under Linux firstly, but it designed as a cross-platform application. The core program than runs on GPU nodes, can be compiled and executed under Linux and Windows. The parameter server demo written with python can run on Linux/Windows/MacOS boxes.

If your program runs on the same CUDA_ARCHITECTURES machines as your compiling box, you can add -DCMAKE_CUDA_ARCHITECTURES="" to the cmake configure command. i.e. cmake .. -DCMAKE_CUDA_ARCHITECTURES="" for Linux, or cmake %src_dir% -DCMAKE_CUDA_ARCHITECTURES="" -G "Visual Studio 16" in buildWin.bat for Windows.

Linux

Compile from source

Clone the code

git clone https://github.com/liu-kan/gSMFRETda.git

Submodules are not necessary to pull, them are just listed for FOSSA to analyse dependencies.

Building prerequest

  • CUDA version >= 10
  • libhdf5-dev 1.10 and newer
  • CMake >= 3.11

Install CUDA

Follow the documentation at https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

Build the code

For deb systems, like Debian or Ubuntu

sudo apt install build-essential libhdf5-dev pkg-config protobuf-compiler libprotobuf-dev libnanomsg-dev libboost-dev doxygen libboost-system-dev libboost-serialization-dev cmake gengetopt libboost-filesystem-dev libdlib-dev libgtest-dev libblas-dev liblapack-dev libsqlite3-dev libgsl-dev

For rpm systems, like Fedora, Centos or Redhat

#if centos 8
sudo dnf config-manager --set-enabled PowerTools 
#elseif redhat 8
sudo subscription-manager repos --enable codeready-builder-for-rhel-8-x86_64-rpms
#endif
sudo yum groupinstall "Development Tools" 
#if centos or redhat
sudo yum install https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm # or epel-release-latest-<your_version>.noarch.rpm  
#endif
sudo yum install pkg-config python3-protobuf openssl-devel python3-devel 
sudo dnf install protobuf-devel texinfo hdf5-devel gengetopt 
#If centos or redhat
# Download [nanomsg](https://nanomsg.org/), [gengetopt](http://www.gnu.org/software/gengetopt/), [cmake >=3.14](https://github.com/Kitware/CMake/releases/download/v3.17.4/cmake-3.17.4.tar.gz) and install them.
# If build with ctest, install [remi-release-<version>.rpm](https://rpms.remirepo.net/), then 
sudo dnf install libXext-devel libpng-devel libjpeg-turbo-devel openblas-devel gtest-devel sqlite-devel lapack-devel dlib-devel gsl-devel
# if you don't build with ctest (add -DBUILD_TESTS=OFF when cmake), you can ignore them.
#endif
#if fedora
sudo dnf install nanomsg-devel cmake gtest-devel sqlite-devel lapack-devel dlib-devel gsl-devel
#endif

Then, install Conan. Finally

mkdir gSMFRETda/build
cd gSMFRETda/build
conan install --build=missing ../conanfile.posix
cmake ..
make -j8
sudo ldconfig

If you get error like "No CMAKE_CUDA_COMPILER could be found", try

export CUDA_HOME=/usr/local/cuda # Your cuda install path
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_HOME/lib64:$CUDA_HOME/extras/CUPTI/lib64
export PATH=$PATH:$CUDA_HOME/bin

before cmake. Or add them into ~/.bashrc

Notice

Now, gSMFRETda depend on boost version EXACTly above 1.70.0. If your Linux distribution doesn't ship this version (in some cases), you need install CONAN first, and run

conan install --build=missing ../conanfile.posix

in build directory before you run cmake .. .

Windows

The program can be compiled on Windows natively. And if you wish, it can also run on WSL2 with CUDA with compiling method of Linux mentioned above.

Compiling natively from source

Compiling under Windows depends on conan, in sequence, install Visual Studio 2019 (with C++, CMake for Windows, CMake for Linux & English language pack Component installation. It's highly recommended that installing VS to default path.), CUDA for Windows, vcpkg, conan (Add conan to user path), Git for Windows. Open a "x64 Native Tools Command Prompt for VS 2019" console form Windows Start Menu.

Assume your vcpkg is installed in c:\vcpkg

cd \path\to\build\dir
git clone https://github.com/liu-kan/gSMFRETda.git
cd gSMFRETda
vcpkg install --triplet x64-windows "@.vcpkg_deps.txt"
mkdir build
cd build
..\gSMFRETda\buildWin.bat

The exe files will sit on .\bin\

If the install_dir of Git for Windows is not default sit at "C:/Program Files/Git". Set environment variables patch to full path of patch.exe file,

set patch=F:/Program Files/Git/usr/bin/patch.exe

before run batch file of buildWin.bat/buildWinDebug.bat .