Documentation: Installation

Example Usages:

• Examples are at config/examples/*.py. We use some of these scripts locally for testing - for example one can update these files and run as:
  • ./config/examples/arch-osx-10.6.py
• Assuming BLAS, LAPACK, MPICH are not currently installed ./configure will download & install BLAS, LAPACK, MPICH if they are not already installed on the system):
  • ./configure --with-cc=gcc --with-cxx=g++ --with-fc=gfortran --download-fblaslapack --download-mpich
• Same as above - but do not have a fortran compiler [and want to use PETSc from C only].
  • ./configure --with-cc=gcc --with-cxx=0 --with-fc=0 --download-f2cblaslapack --download-mpich
• If BLAS, LAPACK, MPI are already installed in default system/compiler locations and mpicc, mpif90, mpiexec are available via PATH - configure does not require any additional options.
  • ./configure
• If BLAS, LAPACK, MPI are already installed in known user location use:
  [Note: Do not specify --with-cc --with-fc etc when using --with-mpi-dir - so that mpicc/mpif90 can be picked up from mpi-dir]
  • ./configure --with-blas-lapack-dir=/usr/local/blaslapack --with-mpi-dir=/usr/local/mpich
  or
  • ./configure --with-blas-lapack-dir=/usr/local/blaslapack --with-cc=/usr/local/mpich/bin/mpicc --with-mpi-f90=/usr/local/mpich/bin/mpif90 --with-mpiexec=/usr/local/mpich/bin/mpiexec
• Build Complex version of PETSc [using c++ compiler] (add the option --with-fortran-kernels=generic to get possibly faster complex number performance on some systems):
  • ./configure --with-cc=gcc --with-fc=gfortran --with-cxx=g++ --with-clanguage=cxx --download-fblaslapack --download-mpich --with-scalar-type=complex
• Install 2 variants of PETSc, one with gnu, the other with intel compilers. Specify different PETSC_ARCH for each build.
  • ./configure PETSC_ARCH=linux-gnu --with-cc=gcc --with-cxx=g++ --with-fc=gfortran --download-mpich
  • make PETSC_ARCH=linux-gnu all test
  • ./configure PETSC_ARCH=linux-gnu-intel --with-cc=icc --with-cxx=icpc --with-fc=ifort --download-mpich --with-blas-lapack-dir=/usr/local/mkl
  • make PETSC_ARCH=linux-gnu-intel all test

PETSC_DIR and PETSC_ARCH are a couple of variables that control the configuration and build process of PETSc. These variables can be set as envirnment variables or specified on the command line [to both configure and make]

• specify enviornment variable for csh/tcsh [can be specified in ~/.cshrc]
  • setenv PETSC_DIR /home/balay/petsc-3.6.0
  • setenv PETSC_ARCH linux-gnu-c-debug
• specify enviornment variable for bash [can be specified in ~/.bashrc]
  • export PETSC_DIR=/home/balay/petsc-3.6.0
  • export PETSC_ARCH=linux-gnu-c-debug
• specify variable on commandline to configure
  • ./configure PETSC_DIR=/home/balay/petsc-3.6.0 PETSC_ARCH=linux-gnu-c-debug [other configure options]
• specify variables on command line to make
  • make PETSC_DIR=/home/balay/petsc-3.6.0 PETSC_ARCH=linux-gnu-c-debug [other make options]

PETSC_DIR: this variable should point to the location of the PETSc installation that is used. Multiple PETSc versions can coexist on the same file-system. By changing PETSC_DIR value, one can switch between these installed versions of PETSc.

PETSC_ARCH: this variable gives a name to a configuration/build. Configure uses this value to stores the generated config makefiles in ${PETSC_DIR}/${PETSC_ARCH}/conf. And make uses this value to determine this location of these makefiles [which intern help in locating the correct include and library files].

Thus one can install multiple variants of PETSc libraries - by providing different PETSC_ARCH values to each configure build. Then one can switch between using these variants of libraries [from make] by switching the PETSC_ARCH value used.

If configure doesn't find a PETSC_ARCH value [either in env variable or command line option], it automatically generates a default value and uses it. Also - if make doesn't find a PETSC_ARCH env variable - it defaults to the value used by last successful invocation of previous configure.

Return to Installation Instructions

Compilers:

Specify compilers and compiler options used to build PETSc [and perhaps external packages]

• Specify compilers using the options --with-cc --with-cxx --with-fc for c, c++, and fortran compilers
  • --with-cc=mpicc --with-cxx=mpicxx --with-fc=mpif90
  • --with-cc=gcc --with-cxx=g++ --with-fc=gfortran
  • --with-cc=gcc --with-cxx=g++ --with-fc=gfortran --with-clanguage=cxx
  Note that --with-clanguage=cxx means that the PETSc source code is compiled with the C++ compiler. This is not normally needed and we don't recommend it. One can use 'c' build of PETSc from both C and C++. One can also have a complex build with C99.
• If fortran compiler is not available or not needed - then disabling using fortran
  • --with-fc=0
• If a C++ compiler is not available or not needed - disable configuring with it
  • --with-cxx=0
• If no compilers are specified - configure will automatically look for available MPI or regular compilers in the user's PATH
  • mpicc/mpiCC/mpif90 or mpif77
  • gcc/g++/gfortran or g77
  • cc/CC/f77 etc..
• It's best to use MPI compilers as this will avoid the situation where MPI is compiled with one set of compilers [like gcc/g77] and user specified incompatible compilers to PETSc [perhaps icc/ifort]. This can be done by either specifying --with-cc=mpicc or --with-mpi-dir [and not --with-cc=gcc]
  • --with-cc=mpicc --with-cxx=mpicxx --with-fc=mpif90
  • --with-mpi-dir=/opt/mpich2-1.1 [but *no* --with-cc=gcc]
• Configure defaults to building PETSc in debug mode. One can switch to using optimzed mode with the toggle option --with-debugging [defaults to debug enabled]. Additionally one can specify more suitable optimization flags with the options COPTFLAGS, FOPTFLAGS, CXXOPTFLAGS. For ex: when using gnu compilers with corresponding optimization flags:
  • ./configure --with-cc=gcc --with-cxx=g++ --with-fc=gfortran --with-debugging=0 COPTFLAGS='-O3 -march=native -mtune=native' CXXOPTFLAGS='-O3 -march=native -mtune=native' FOPTFLAGS='-O3 -march=native -mtune=native' --download-mpich
• Configure cannot detect compiler libraries for certain set of compilers. In this case one can specify additional system/compiler libraries using the LIBS option
  • ./configure --LIBS='-ldl /usr/lib/libm.a'

Return to Installation Instructions

External Packages:

PETSc provides interfaces to various external packages. Blas/Lapack is a required package, MPI is not required if running sequentially. One can optionally use external solvers like Hypre, MUMPS, etc. from within PETSc aplications.

PETSc configure has the ability to download and install these external packages. Alternatively if these packages are already installed, then configure can detect and use them.

The following modes can be used to install/use external packages with configure.

• --download-PACKAGENAME: Download specified package and install it. Then configure PETSc to use this package.
  • --download-fblaslapack --download-mpich
  • --download-scalapack --download-mumps
• --download-PACKAGENAME=/PATH/TO/package.tar.gz: If ./configure cannot automatically download the package [due to network/firewall issues], one can download the package by alternaive means [perhaps wget or scp via some other machine]. Once the tarfile is downloaded, the path to this file can be specified to configure with this option. Configure will proceed to install this package and then configure PETSc with it.
  • --download-mpich=/home/petsc/mpich2-1.0.4p1.tar.gz
• --with-PACKAGENAME-dir=PATH: If the external package is already installed - specify its location to configure [it will attempt to detect, include, library files from this location.] Normally this corresponds to the top-level installation dir for the package.
  • --with-mpi-dir=/home/petsc/software/mpich2-1.0.4p1
  Note that except for MPI we hightly recommend you have PETSc download and install the external packages rather than you installing them separately first.
• --with-PACKAGENAME-include=INCLUDEPATH --with-PACKAGENAME-lib=LIBRARYLIST: Usually a package is defined completely by its include file location - and library list. [If the package is already installed] - then one can use these two options to specify the package to configure. For eg:
  • --with-superlu-include=/home/petsc/software/superlu/include --with-superlu-lib=/home/petsc/software/superlu/lib/libsuperlu.a
  • --with-parmetis-include=/sandbox/balay/parmetis/include --with-parmetis-lib="-L/sandbox/balay/parmetis/lib -lparmetis -lmetis"
  • --with-parmetis-include=/sandbox/balay/parmetis/include --with-parmetis-lib=[/sandbox/balay/parmetis/lib/libparmetis.a,libmetis.a]

Notes:

• Run ./configure --help to get the list of external packages and corresponding additional options [for example --with-mpiexec for mpich]
• Generally one would use either one of the above 4 modes for any given package - and not mix these. [i.e mixing --with-mpi-dir and --with-mpi-include etc.. should be avoided]
• Some packages might not support certain options like --download-PACKAGENAME or --with-PACKAGENAME-dir. Architectures like Microsoft Windows might have issues with these options. In these cases, --with-PACKAGENAME-include --with-PACKAGENAME-lib options should be prefered.
• It's best to install some external packages like SuperLU_DIST, MUMPS, Hypre, METIS, and ParMETIS with the option --download-PACKAGENAME. [the correct options for these packages are --download-superlu_dist --download-mumps --download-hypre]
  • This will install the COMPATIBLE version of the external package. A generic install of this package might not be compatible with PETSc [perhaps due to version differences - or perhaps due to the requirement of additional patches for it to work with PETSc]
  • Some packages have bug fixes, portability patches, and upgrades for dependent packages that have not yet been included in an upstream release.
  • All packages will be installed with the same set of compilers - this avoids problems [for example weird link time errors] with mixing code compiled with multiple compilers [for example mixing gfortran and ifort compiled code].
• If you want to download a compatible external package manually, then the URL for this package is listed in configure source for this package. For example, check config/config.packages/SuperLU.py for the url for download this package.

Additional options:

• --with-external-packages-dir=PATH: By default, external packages will be unpacked in ${PETSC_DIR}/externalpackages. However one can choose a different location where these packages are unpacked.

Return to Installation Instructions

BLAS/LAPACK

These packages provide some basic numeric kernels used by PETSc.

• Configure will automatically look for blas/lapack in certain standard locations, on most systems you should not need to provide any information about BLAS/LAPACK in the ./configure command.
• One can use the following options to let configure download/install blas/lapack automatically.
  • --download-fblaslapack [when fortran compiler is present]
  • --download-f2cblaslapack [when configuring without a fortran compiler - i.e --with-fc=0]
• Alternatively one can use other options like one of the following.
  • --with-blas-lapack-lib=libsunperf.a
  • --with-blas-lib=libblas.a --with-lapack-lib=liblapack.a
  • --with-blas-lapack-dir=/soft/com/packages/intel/13/079/mkl

Notes:

• Intel MKL

  Intel provides blas/lapack via MKL library. It usually works from GNU/Intel compilers on linux and MS/Intel compilers on Windows. One can specify it to PETSc configure with for eg: --with-blas-lapack-dir=/soft/com/packages/intel/13/079/mkl
  If the above option does not work - one could determine the correct library list for your compilers using Intel MKL Link Line Advisor and specify with the configure option --with-blas-lapack-lib

• IBM ESSL

  Sadly, IBM's ESSL does not have all the routines of BLAS and LAPACK that some packages, such as SuperLU expect; in particular slamch, dlamch and xerbla. In this case instead of using ESSL we suggest --download-fblaslapack. If you really want to use ESSL, see http://www.pdc.kth.se/resources/computers/bluegene/LAPACK-CBLAS/LAPACK-CBLAS-build.

Return to Installation Instructions

MPI

This software provides the parallel functionality for PETSc.

• Configure will automatically look for MPI compilers mpicc/mpif77 etc and use them if found in your PATH.
• One can use the following options to let configure download/install MPI automatically
  • --download-mpich [install and use MPICH]
  • --download-openmpi [Install and use Open MPI]
• See externalpackages for other installation options.

Using MPI Compilers:

• It's best to install PETSc with MPI compiler wrappers (often called mpicc, mpicxx, mpif90) - this way, the SAME compilers used to build MPI are used to build PETSc [this avoids incompatibilities which might crop up - when using libraries compiled with different c or fortran compilers.]. This can be achieved with the following modes.
  • Vendor provided MPI might already be installed. IBM, SGI, Cray etc provide their own:
    ./config/confiure.py --with-cc=mpcc --with-fc=mpf77
  • If using MPICH which is already installed [perhaps using myrinet/gm] then use [without specifying --with-cc=gcc etc.so that configure picks up mpicc from mpi-dir]:
    ./configure --with-mpi-dir=/path-to-mpich-install

Installing without MPI:

• You can build (sequential) PETSc without an MPI. This is useful for quickly installing PETSc [if MPI is not available - for whatever reason]. However - if there is any MPI code in user application, then its best to install a full MPI - even if the usage is currently limited to uniprocessor mode.
  • ./configure --with-mpi=0

Installing with Open MPI with shared MPI libraries:

OpenMPI defaults to building shared libraries for MPI. However, the binaries generated by MPI wrappers mpicc/mpif77 etc require LD_LIBRARY_PATH to be set to the location of these libraries.

Due to this OpenMPI restriction one has to set LD_LIBRARY_PATH correctly [per OpenMPI installation instructions], before running PETSc configure. If you do not set this environmental variables you will get messages when running ./configure such as

            UNABLE to EXECUTE BINARIES for config/configure.py
-------------------------------------------------------------------------------
            Cannot run executables created with C. If this machine uses a batch system
            to submit jobs you will need to configure using/configure.py with the additional option --with-batch.
            Otherwise there is problem with the compilers. Can you compile and run code with your C/C++ (and maybe Fortran) compilers?
        

or when running a code compiled with OpenMPI

-bash-3.3$ ./conftest
./conftest: error while loading shared libraries: libmpi.so.0: cannot open shared object file: No such file or directory

Notes:

• Avoid specifing compilers [with options --with-cc or --with-fc] when using the option --with-mpi-dir. [Option--with-mpi-dir specifies using MPI compilers - so its best to use them - and not overwride them with user specified --with-cc].
• One can specify mpiexec or mpiexec with the options --with-mpiexec

Return to Installation Instructions

Microsoft Windows Installation:

• Are you sure you want to use MS/Windows? We recommend using linux if possible [and minimize troubleshooting windows related issues]. And most externalpackages will not work on windows.
• If you have access to only a windows machine and not a linux machine, consider using linux virtual machine via VirtualBox. One sample tutorial is at http://www.psychocats.net/ubuntu/virtualbox. Google can provide more tutorials.
• You can consider using GNU compilers from cygwin. Our regular [linux] instructions would work with cygwin/gnu compilers. And most externalpackages will also work.

Installation With MS/Intel/Compaq Compilers:

Microsoft Windows OS does not provide the same unix shell enviornment as the other OSes. Also the default MS/Intel/Compaq compilers behave differently than other unix compilers. So to install PETSc on Microsoft Windows - one has to install cygwin [for the unix enviornment] and use win32fe [part of PETSc sources,to interface to MS/Intel/Compaq compilers].

Install Cygwin: Please download and install cygwin package from http://www.cygwin.com. Make sure the following cygwin components are installed.

• python
• make
• [default selection should already have diff and other tools]

Remove Cygwin link.exe: Cygwin link.exe can conflict with Intel ifort and Comapq F90 compilers. If you are using these compilers - please do [from cygwin terminal/bash-shell]:

• mv /usr/bin/link.exe /usr/bin/link-cygwin.exe

Setup cygwin terminal/bash-shell with Working Compilers: We require the compilers to be setup properly in a cygwin bash command shell, so that "cl foo.c" or "ifort foo.f" works from this shell. For example - if using VS2005 C and Intel 10 Fortran one can do:

• Start -> Programs -> Intel Software Development Tools -> Intel Fortran Compiler 10 -> Visual Fortran Build Enviornment [32bit or 64bit depending on your usage]. This should start a 'dos cmd' shell.
• Within this shell - run cygwin terminal/bash-shell mintty.exe as: c:\cygwin\bin\mintty.exe
• verify if the compilers are useable [by running cl, ifort in this cygwin terminal/bash-shell]
• Now run configure with win32fe and then build the libraries with make [as per the usual instructions]

Example Configure usage with Windows Compilers:

Use configure with VC2005 C and Intel Fortran 10 [With MPICH2 installed].

• ./configure --with-cc='win32fe cl' --with-fc='win32fe ifort' --with-cxx='win32fe cl' --download-fblaslapack

If fortran usage is not required, use:

• ./configure --with-cc='win32fe cl' --with-fc=0 --download-f2cblaslapack

Using Compaq F90:

Using Microsoft C/C++ 6.0 & Compaq Fortran 6.0 with MPICH2 configure command to use:

• ./configure --with-cc='win32fe cl' --with-fc='win32fe f90' --download-fblaslapack

Note: MPICH2 mpif.h needs a fix for it to work with Compaq F90 [specifically remove line with MPI_DISPLACEMENT_CURRENT - which uses 'integer*8' - which is unsupported by Compaq F90]

ExternalPackages: The --download-package option does not work with many external packages on Microsoft Windows.

Project Files: We cannot provide Microsoft Visual Studio project files for users as they are specific to the configure options, location of external packages, compiler versions etc. used for any given build of PETSc, so they are potentially different for each build of PETSc. So if you need a project file for use with PETSc - please do the following.

• create an empty project file with one of the examples say src/ksp/ksp/examples/tutorials/ex2.c
• try compiling the example from cygwin bash shell - using makefile - for eg:
  • cd src/ksp/ksp/examples/tutorials
  • make ex2
• if the above works - then make sure all the compiler/linker options used by make are also present in the project file in the correct notation.
• if errors - redo the above step. [if all the options are correctly specified - then the example should compile from MSDev.

Debugger: Running PETSc probrams with -start_in_debugger is not supported on this platform, so debuggers will need to be initiated manually. Make sure your environment is properly configured to use the appropriate debugger for your compiler. The debuggers can be initiated using Microsoft Visual Studio 6: msdev ex1.exe, Microsoft Visual Studio .NET: devenv ex1.exe, Intel Enhanced Debugger: edb ex1.exe, or GNU Debugger gdb ex1.exe.

Using Cygwin gcc/g++/gfortran: One can install and use PETSc with gcc/gfortran compilers from cygwin. In this case follow the regular Unix instructions.

PETSc Win32 front end - win32fe: This tool is used as a wrapper to Microsoft/ Borland/ Intel compilers and associated tools - to enable building PETSc libraries using cygwin make and other UNIX tools. For additional info, run ${PETSC_DIR}/bin/win32/win32fe without any options.

Return to Installation Instructions

Installing PETSc in /usr/local or /opt where sudo or root privileges are required:

If one wants to install PETSc [with sources] in a common system location like /usr/local or /opt, then we suggest creating a dir for PETSc in the required location with user privileges, and then do the PETSc install [as a regular/non-root user]. i.e.

• sudo mkdir /opt/petsc
• sudo chown user:group /opt/petsc
• cd /opt/petsc
• tar -xzf petsc-3.6.0.tar.gz
• cd petsc-3.6.0
• ./configure
• make

Installing with --prefix:

One can also use the GNU --prefix install mode.

• ./configure --prefix=/opt/petsc/petsc-3.6.0 [other configure options]
• make
• [sudo] make install
• [user can now use this PETSc install with:] make PETSC_DIR=/opt/petsc/petsc-3.6.0 ex1

Note: one has to switch to using PETSC_DIR=/opt/petsc/petsc-3.6.0 after install is done - and NOT use PETSC_ARCH anymore.

Using DESTDIR with prefix install:

One can also use the GNU DESTDIR with a prefix install. This mode is generally used by binary packagers. For example

• ./configure --prefix=/opt/petsc/petsc-3.6.0
• make
• make install DESTDIR=/tmp/petsc-pkg
• [package up /tmp/petsc-pkg The package should then be installed at /opt/petsc/petsc-3.6.0]
• [user can now use this PETSc install with:] make PETSC_DIR=/opt/petsc/petsc-3.6.0 ex1

Multiple installs using --prefix and DESTDIR:

Sepcify a different --prefix location/option for each build - at configure time. For eg:

• untar petsc tar ball
• ./configure --prefix=/opt/petsc/petsc-3.6.0-mpich --with-mpi-dir=/opt/mpich
• make
• make install DESTDIR=/tmp/petsc-pkg
• untar petsc tar ball
• ./configure --prefix=/opt/petsc/petsc-3.6.0-openmpi --with-mpi-dir=/opt/openmpi
• make
• make install DESTDIR=/tmp/petsc-pkg

Return to Installation Instructions

PETSc ./configure automatically generates Pkgconfig and module files for each install

These can be found in ${PETSC_DIR}/${PETSC_ARCH}/lib/pkgconfig/PETSc.pc and ${PETSC_DIR}/${PETSC_ARCH}/lib/modules/${PETSC_VERSION}-${PETSC_ARCH}. The module file may need to be edited for your particular system. Note that if --prefix is used then $PETSC_ARCH is not included in the above directories nor is -${PETSC_ARCH} included in the module file name.

Return to Installation Instructions

Installing on machine requiring cross compiler or a job scheduler, such as IBM SP or IBM Blue Gene:

If one has to use a cross compiler - or go through the job scheduler to use MPI on a given machine - use the configure option --with-batch as follows:

• run configure with the additional option --with-batch on the frontend node (compiler server) [perhaps with the additional option --known-mpi-shared-libraries=0]
• the above configure run will create a binary conftest. Run this binary conftest on one compute node using the job scheduler.
• The above run of conftest will create a new python script reconfigure. Run python reconfigure again on the frontend node (compiler server) to complete the configuration process

Return to Installation Instructions

Installing with TAU Instrumentation package:

TAU package and the prerequisite PDT packages need to be installed separately [perhaps with MPI]. Now use tau_cc.sh as compiler to PETSc configure.

• export TAU_MAKEFILE=/home/balay/soft/linux64/tau-2.20.3/x86_64/lib/Makefile.tau-mpi-pdt
• ./configure CC=/home/balay/soft/linux64/tau-2.20.3/x86_64/bin/tau_cc.sh --with-fc=0 PETSC_ARCH=arch-tau

Return to Installation Instructions

Installing PETSc to use NVidia GPUs (aka CUDA)

• Requires CUDA, Thrust and Cusp . Thrust has been included in CUDA since the 4.0 release - so a separate install is not needed.
• Currently using CUDA-4.1 or CUDA-4.2 with Cusp-v0.3.1 or CUDA-5 with Cusp-v0.4.0.
• On Linux - make sure you have compatible NVidia driver installed.
• check config/examples/arch-cuda-double.py for example usage.

Cuda build of PETSc curently works on Apple OSX, Linux [Ubuntu 10.04], Microsoft Windows [Server 2003].

Return to Installation Instructions

Installing PETSc to use GPUs and accelerators via OpenCL (NVIDIA, AMD, Intel MIC)

• Requires ViennaCL and the OpenCL shared library, which is shipped in the vendor graphics driver. Make sure the OpenCL headers are available on your system by e.g. downloading them from the Khronos Group directly. Package managers on Linux provide these headers through a package named 'opencl-headers' or similar.
• Always make sure you have the latest GPU driver installed. There are several known issues with older driver versions.
• check config/examples/arch-viennacl.py for example usage.

OpenCL/ViennaCL builds of PETSc currently work on Apple OSX, Linux, and Microsoft Windows. OpenCL is less a burden on the build system than CUDA, so installation is often easier and works with any host compiler.

Return to Installation Instructions