Preparing a Linux machine for Bilder » History » Revision 56
Revision 55 (David Alexander, 04/09/2019 01:08 PM) → Revision 56/86 (John Cary, 08/14/2019 06:37 AM)
# Preparing a Linux machine for Bilder __If all you need is python and sphinx__ (e.g., for builds of documentation), then you need only * CMake (probably present) * Python (probably present) * Python-dev or Python-devel (on Fedora) * Latex as installed below * Perl-dev or Perl-devel with perl-ExtUtils-Embed and then you can add the arguments, -W python,cmake, and your installed Python and CMake will be used if your path is correct. ## Java At least one package (Babel) needs Oracle's java, so we recommend installing it. ## GCC It is often useful to get a newer gcc that has mmintrinsics. This may break compatibility with some versions of [[ CUDAgccCompatibility | CUDA ]]. Below are the instructions for building gcc-6.8.34.0. gcc-6.4.0. [[ BilderKnownIssuesGcc730 | (As of this writing 8.3 6.4 is the latest compiler to work with CUDA-10.2, CUDA-9.1, and gcc-5.3.1 will not build QT-5.10.0) ]] ### Make sure 32bit glibc-devel is installed Do the applicable of ~~~~~~ yum -y install glibc-devel.i686 dnf -y install glibc-devel.i686 zypper install glibc-devel-32bit (SLES 11 SP3) ~~~~~~ ### Get and unpack the source package ~~~~~~ export GCCVER=8.3.0 GCCVER=6.4.0 wget ftp://ftp.gnu.org/gnu/gcc/gcc-${GCCVER}/gcc-${GCCVER}.tar.xz tar xf gcc-${GCCVER}.tar.xz && cd gcc-${GCCVER} ~~~~~~ ### Patch as needed If your system has glibc 2.26 or later, you will need to fix the source by making the changes specified in https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=72edc2c02f8b4768ad660f46a1c7e2400c0a8e06 and also ~~~~~~ for i in libgcc/config/*/linux-unwind.h; do if grep -q "struct ucontext" $i; then sed -i.bak 's/struct ucontext/ucontext_t/' $i fi done ~~~~~~ ### Build Define CONTRIB_DIR to be your tarball installation directory, e.g., ~~~~~~ export CONTRIB_DIR=/contrib ~~~~~~ Usually CONTRIB_DIR=/contrib, but if one does not have root access, one has to put this in a user area. If one does have root access, one should execute the following commands: ~~~~~~ mkdir /internal /contrib /volatile chmod 775 /internal /contrib /volatile ~~~~~~ Unpack gcc, go into the gcc source directory, and build according to the instructions below. **gcc:** ~~~~~~ ./contrib/download_prerequisites mkdir build && cd build ../configure --prefix=$CONTRIB_DIR/gcc-${GCCVER} --enable-languages=c,c++,fortran --with-pic env LD_RUN_PATH=$CONTRIB_DIR/gcc-${GCCVER}/lib64 make -j 4 make install # Alias as desired. mkdir -p $CONTRIB_DIR/bin && cd $CONTRIB_DIR/bin ln -s $CONTRIB_DIR/gcc-${GCCVER}/bin/{c,g}* . ~~~~~~ ### Fix permissions ~~~~~~ chmod -R g+rwX $CONTRIB_DIR/gcc-${GCCVER} chmod -R o+rX $CONTRIB_DIR/gcc-${GCCVER} find $CONTRIB_DIR/gcc-${GCCVER} -type d -exec chmod g+rs '{}' \; ~~~~~~ ## Java Install Oracle's Java. Once it is installed, you will need to modify your path to include /usr/java/default/bin before /usr/bin or use the alternatives package: ~~~~~~ for i in java jar javac; do alternatives --install /usr/bin/$i j$i /usr/java/jdk1.6.0_16/bin/$i 3 done alternatives --config java # Pick 3 ~~~~~~ On SLES , Install using the tar package from Oracle, then change into the installation directory ~~~~~~ for i in java jar javac; do /usr/sbin/update-alternatives --install /usr/bin/$i $i <installation directory>/bin/$i 3 done ~~~~~~ where <installation directory\> is the absolute path to where Java was just installed ## Version control systems Make sure git, mercurial, and subversion are all installed. For building VisIt, install git large file storage: https://git-lfs.github.com/ Download the tarball from that site, untar, and run ~~~~~~ git lfs install --force --skip-smudge ~~~~~~ ## LaTeX LaTeX is needed for building documentation using Sphinx. Bilder prefers the texlive-latex package. For gui usage, see texworks: http://www.tug.org/texworks If you cannot install in the system area, you can download ~~~~~~ wget http://mirror.ctan.org/systems/texlive/tlnet/install-tl-unx.tar.gz tar xzf install-tl-unx.tar.gz cd install-tl-20150411 (date might be different) ./install-tl -no-gui ~~~~~~ This is now interactive, so set * Set scheme to basic ** To do this type 'S' and hit enter. ** Then enter 'd' to set the scheme to basic. ** Type 'R' and hit enter to return to the main configuration page. * Set TEXDIR = $CONTRIB_DIR/texlive ** Start by typing 'D' and hit enter. ** Then enter '1' to change TEXDIR. ** Type $CONTRIB_DIR/texlive and hit return. ** To return to configuration type 'R' and return. * Set page size to Letter size ** Type 'O' and hit enter. ** Enter 'P' ** Return to main menu by typing 'R' and enter. * Install ** Type 'I' and hit enter to install. * Put your texlive in your PATH. ** Edit you bashrc (vim ~/.bashrc) ** Add "export PATH=$PATH:<$CONTRIB_DIR>/texlive/bin/x86_64-linux/" (enter the full CONTRIB_DIR value) ** Exit and save. Then source ~/.bashrc. To ensure you are using the correct texlive installation type "which pdflatex" and make sure it the one in $CONTRIB_DIR/texlive/bin/x86_64-linux/. If it is not the correct pdflatex, you may have to remove the system version (sudo yum remove texlive). and then install the packages noted at [[LaTeX packages needed by Bilder]] ## Additional packages Other parts of the toolchain introduce other package requirements. Below are the initial lists. ### Fedora The following packages (cut and paste after dnf install or yum install) are needed for Fedora, Redhat, CentOS ~~~~~~ bc bison glib2-devel imake lapack-devel libblas-devel libcurl-devel libxml2-devel openssl-devel patch redhat-lsb zlib-devel ~~~~~~ For more recent versions of Fedora, you will need. ~~~~~~ egl-wayland mesa-libEGL-devel redhat-rpm-config ~~~~~~ For centos7 you will also need. ~~~~~~ imake xorg-x11-utils.x86_64 ~~~~~~ ### Ubuntu The following packages are needed for Ubuntu ~~~~~~ ace-netsvcs bc bison flex gfortran graphviz libblas-dev libfreetype6-dev libgl1-mesa-dev libglu1-mesa-dev libgraphviz-dev liblapack-dev libntlm0 libpng-dev libxext-dev libxrender-dev libxtst-dev libzzip-dev mailutils python2.7 python2.7-dev zlib1g-dev ~~~~~~ * You will probably want a lightweight command line editor other than gedit, so "sudo apt-get install vim" or "sudo apt-get install emacs" * zlib-dev doesn't exist on ubuntu 16.04 - use zlib1g-dev instead. mailutils is necessary for configuring the system for jenkins. You'll be offered to choose: * Internet site * Internet with smarthost * Satellite system * Local only Internet with smarthost is recommended. The use of the following commands are helpful. To search for an available package dealing with a keyword, use: ~~~~~~ $ sudo apt-cache search keyword ~~~~~~ To install a particular package, use: ~~~~~~ $ sudo apt-get install libblah-dev ~~~~~~ Note: that development packages names typically end with "dev" not "devel" as in other flavors of linux. ### Suse Linux Enterprise Server (SLES) The following packages (in cut/paste for zypper form) are needed on SLES ~~~~~~ bc bison blas-devel blas-devel-static libblas3 flex freetype2-devel gcc gcc-32bit gcc-fortran git glibc-devel glibc-devel-32bit graphviz graphviz-devel libbz2-devel lapack-devel lapack-devel-static liblapack3 libpng-devel libssh-devel libxml2-devel mailx mercurial Mesa-devel python-devel xorg-x11-devel xorg-x11-libXext-devel xorg-x11-libXrender-devel xorg-x11-libs zlib-devel ~~~~~~ The following are needed to build subversion ~~~~~~ libapr-util1 libapr1 libneon27 ~~~~~~ To get blas-devel, blas-devel-static and lapack-devel, lapack-devel-static on SLES 11 SP3 , add the Science repository ~~~~~~ zypper addrepo http://download.opensuse.org/repositories/science/SLE_11_SP3/science.repo zypper refresh zypper install blas-devel blas-devel-static lapack-devel lapack-devel-static ~~~~~~ ## Qt Prerequisites Qt-5 must be built with QtWebEngine, which particularly introduces requirements for other packages as noted at * https://doc.qt.io/qt-5.10/qtwebengine-platform-notes.html * https://wiki.qt.io/QtWebEngine/How_to_Try * https://www.ics.com/blog/building-qtwebengine * http://www.linuxfromscratch.org/blfs/view/svn/x/qtwebengine.html Each of the above websites appears to have made assumptions about what is already installed, so as one encounters more workstations, one is likely to find yet another package missing. This is our list so far. ### Fedora ~~~~~~ sudo yum install alsa-lib-devel bison cups-devel dbus-devel flex fontconfig-devel gperf libicu-devel sudo yum install libcap-devel libgcrypt-devel libgcrypt libgudev1-devel libsndfile-devel libwebp-devel sudo yum install libXcomposite-devel libXcursor-devel libXi-devel libXrandr-devel libxslt-devel libXtst-devel sudo yum install mesa-libEGL-devel nss-devel opus-devel pciutils-devel pulseaudio-libs-devel re2c ruby systemd-devel ~~~~~~ ## Vendor-specific version of graphics driver If you are preparing a Linux machine for the VisIt Visualization package, the vendor-specific graphics driver for your graphics card should be installed. By default, a generic graphics driver is installed. On Fedora, the generic driver is "nouveau". To determine if the nouveau driver is installed on your system, enter the command ~~~~~~ lsmod | grep nouv ~~~~~~ You will need to download and install the appropriate driver from your graphics card vendor. This generally uninstalls the nouveau drivers. ## NVIDIA CUDA Toolkit If you have a compatible GPU and wish to compile with nvcc code, you must first install the CUDA Toolkit. The linux installation page is ~~~~~~ https://developer.nvidia.com/cuda-toolkit-70 ~~~~~~ Download the applicable file. Your Linux distribution may or may not be natively supported for version 7.0. You may alternatively use version 7.5 if supported or an older version if necessary. As an example, we will include the steps known to work for Fedora 22. On Fedora 22, download the Fedora 21 Local RPM package. ~~~~~~ cd <location of download> sudo rpm -i cuda-repo-fedora21-7-0-local-7.0-28.x86_64.rpm #<or whatever file you downloaded> sudo dnf install cuda ~~~~~~ Different Linux distributions will have different package handlers (here we use dnf, ubuntu has apt-get, fedora < 22 uses yum). You should now have the directory /usr/local/cuda-7.0. Add /usr/local/cuda-7.0/bin to your path. To ensure you have gotten the necessary components, type: ~~~~~~ which nvcc ~~~~~~ The output should be /usr/local/cuda-7.0/bin/nvcc. This will allow bilder to build GPU c++ code. IMPORTANT NOTE: CUDA versions only support particular GCC versions, and the latest GCC may not support your CUDA.[[ CUDAgccCompatibility | See this wiki page for details ]]