M OLPRO

Installation Guide Version 2010.1

H.-J. Werner Institut f r Theoretische Chemie u Universit t Stuttgart a Pfaffenwaldring 55 D-70569 Stuttgart Federal Republic of Germany

P. J. Knowles School of Chemistry Cardiff University Main Building, Park Place, Cardiff CF10 3AT United Kingdom

SHA1 833379180250ebe11052d872bd9ff53639690c67

(Copyright c 2008 University College Cardiff Consultants Limited)

OBTAINING THE DISTRIBUTION MATERIALS

Obtaining the distribution materials

M OLPRO is distributed to licensees on a self-service basis using the world-wide web. Those entitled to the code should obtain it from https://www.molpro.net/download supplying the username and password given to them. The web pages contain both source code and binaries, although not everyone is entitled to source code, and binaries are not available for every platform. Execution of M OLPRO, whether a supplied binary or built from source, requires a valid licence key. Note that the key consists of two components, namely a list of comma-separated key=value pairs, and a password string, and these are separated by &. In most cases the licence key will be automatically downloaded from the website when building or installing the software.

Installation of pre-built binaries

Binaries are given as self-extracting tar archives which are installed by running them on the command line. There are binaries tuned for several architectures. These also support parallel execution. The parallel binaries are built using GA with TCGMSG, and the default connection across nodes is rsh. One can use ssh instead of rsh by changing environment variable TCGRSH (e.g., export TCGRSH=/usr/bin/ssh for bash). There is a generic serial binary which should run on all IA32 architectures. The tar archives are fully relocatable, the location can be changed when running the script interactively, the default is /usr/local. If the script nds a licence key which has been cached in $HOME/.molpro/token from a previous install then that key will be installed with the software. If the script cannot nd a key or automatically download it from the molpro website then the script will prompt that this part of the install has failed. All les of Molpro are installed, but the user must then manually install the key with the library les in a le named .token, e.g.: /usr/local/lib/molpro-mpptype-arch/.token Other conguration options as described in section 3.5 may also be specied in the script le: /usr/local/bin/molpro

3
3.1

Installation from source les

There are usually four distinct stages in installing M OLPRO from source les: Conguration A shell script that allows specication of conguration options is run, and creates a conguration le that drives subsequent installation steps. The program is compiled and linked, and other miscellaneous utilities and les, including the default options le, are built. The essential resulting components are 1. The molpro shell script which launches thge main executable. In serial case one can directly run the main executable. 2. The molpro.exe executable, which is the main program. For parallel computation, multiple copies of molpro.exe are started

Compilation

INSTALLATION FROM SOURCE FILES

ii

by a single instance of molpro shell script using the appropriate system utility, e.g. mpirun, parallel, etc. 3. Machine-ready basis-set, and other utility, libraries. Validation Installation A suite of self-checking test jobs is run to provide assurance that the code as built will run correctly. The program can be run directly from the source tree in which it is built, but it is usually recommended to run the procedure that installs the essential components in standard system directories.

3.2

Prerequisites

The following are required or strongly recommended for installation from source code. 1. A Fortran 90 compiler. Fortran77-only compilers will not sufce. On most systems, the latest vendor-supplied compiler should be used. For IA32 Linux (for example Intel Pentium or AMD athlon) the recommended compilers are the Intel Compiler ifort version 10.1 or higher or the Portland pgf90 compiler version 7.2 or higher. For Opteron and EM64T systems the recommended compilers are Portland version 7.2 or higher, Pathscale compiler pathf90 version 3.1 or higher, or the Intel Compiler version 10.1 or higher. The full list of supported compilers can be found at http://www.molpro.net/supported. 2. GNU make, freely available from http://www.fsf.org and mirrors. GNU make must be used; most system-standard makes do not work. In order to avoid the use of a wrong make, it may be useful to set an alias, e.g., alias make=gmake -s. A recent version of GNU make is required, 3.80 or above. 3. The GNU curl utility for batch-mode http transfers, although not needed for installation, is essential for any subsequent application of patches that implement bug xes. 4. About 10GB disk space (strongly system-dependent; more with large-blocksize le systems, and where binary les are large) during compilation. Typically 100Mb is needed for the nally installed program. Large calculations will require larger amounts of disk space. 5. One or more large scratch le systems, each containing a directory that users may write on. There are parts of the program in which demanding I/O is performed simultaneously on two different les, and it is therefore helpful to provide at least two lesystems on different physical disks if other solutions, such as striping, are not available. The directory names should be stored in the environment variables $TMPDIR, $TMPDIR2, $TMPDIR3,. . .. These variables should be set before the program is installed (preferably in .profile or .cshrc), since at some stages the installation procedures will check for them (cf. section 3.5). 6. If the program is to be built for parallel execution then the Global Arrays toolkit or the MPI-2 library is needed. For building M OLPRO with the Global Arrays toolkit, we recommend the latest stable version (although earlier versions may also work). This is available from http://www.emsl.pnl.gov/docs/global and should be installed prior to compiling M OLPRO. As mentioned in Global Arrays documentation, there are three possible ways for building GA: (1) GA with MPI; (2) GA with TCGMSG-MPI; and (3) GA with TCGMSG. Molpro can work with either of these interfaces. For building M OLPRO with

INSTALLATION FROM SOURCE FILES

iii

the MPI-2 library, we recommend to use the built-in MPI-2 library, which may have advantages of optimization on some platforms. If there is no built-in one on the platform, a fresh MPI-2 library ( e.g.: MPICH2, see http://www.mcs.anl.gov/research/projects/mpich2 ) should be installed prior to compiling M OLPRO. Many MPI-2 libraries, including Intel MPI, Bull MPI, MPICH2, and Open MPI, have been tested, and others untested could also work. 7. The source distribution of M OLPRO, which consists of a base compressed tar archive with a le name of the form molpro.2010.1.tar.gz, together, possibly, with one or more module archives with le names of the form molpro.module.2010.1.tar.gz. The modules contain code which is not generally distributed, or features which are not always required to install the code. An example is the program developers kit (module=develop). The archives can be unpacked using gunzip and tar. All archives must be unpacked in the same directory. It is essential that the base archive is unpacked rst, and advisable that any modules are unpacked before further installation.

3.3

Conguration

Once the distribution has been unpacked, identify the root directory that was created (normally molpro2010.1). In the following description, all directories are given relative to this root. Having changed to the root directory, you should check that the directory containing the Fortran compiler you want to use is in your PATH. Then run the command ./configure -batch which creates the le CONFIG. This le contains machine-dependent parameters, such as compiler options. Normally CONFIG will not need changing, but you should at the least examine it, and change any conguration parameters which you deem necessary. Any changes made to CONFIG will be lost next time ./configure is invoked, so it is best to supply as many of these as possible via the command line. The configure procedure may be given command line options, and, if run without -batch , additionally prompts for a number of parameters: 1. On certain machines it is possible to compile the program to use either 32 or 64 bit integers, and in this case configure may be given a command-line option -i4 or -i8 respectively to override the default behaviour. Generally, the 64-bit choice allows larger calculations (les larger than 2Gb, more than 16 active orbitals), but can be slower if the underlying hardware does not support 64-bit integers. Note that if -i4 is used then large les (greater than 2Gb) are supported on most systems, but even then the sizes of M OL PRO records are restricted to 16 Gb since the internal addressing in M OLPRO uses 32-bit integers. If -i8 is used, the record and le sizes are effectively unlimited. Normally we recommend using the default determined by congure. 2. In the case of building for parallel execution, the option -mpp or -mppx must be given on the command line. For the distinction between these two parallelism modes, please refer to the user manual, section 2. The option -mppbase must also be given followed by the location of the Global Arrays base directory or the MPI-2 library include directory. For the case of using the Global Arrays toolkit, one example can be ./configure -mpp -mppbase /usr/local/ga-4-3-3

INSTALLATION FROM SOURCE FILES

iv

and the -mppbase directory should contain directories include and lib/TARGET where for example on Linux/x86 64 TARGET=LINUX64. If using a Global Arrays build with an MPI library the appropriate MPI executable should appear rst in PATH when more than one is available. Queries regarding Global Arrays installations should be sent directly to the Global Arrays team, any Molpro related queries will assume a fully functional Global Arrays suite with all internal tests run successfully. For the case of using the MPI-2 library, one example can be ./configure -mpp -mppbase /usr/local/mpich2-install/include and the -mppbase directory should contain le mpi.h. Please ensure the built-in or freshly built MPI-2 library fully supports MPI-2 standard and works properly. 3. If any system libraries are in unusual places, it may be necessary to specify them explicitly as the arguments to a -L command-line option. 4. configure asks whether you wish to use system BLAS subroutine libraries. M OLPRO has its own optimised Fortran version of these libraries, and this can safely be used. On most machines, however, it will be advantageous to use a system-tuned version instead. On the command line one can specify the level of BLAS to be used from the system, e.g. -blas2. For example if you specify 2, the system libraries will be used for level 2 and level 1 BLAS, but M OLPROs internal routines will be used for level 3 (i.e., matrixmatrix multiplication). Normally, however, one would choose either 0 or 3, which are the defaults depending upon whether a BLAS library is found. A special situation arises if 64-bit integers are in use (-i8), since on many platforms the system BLAS libraries only supports 32-bit integer arguments. In such cases (e.g., IBM, SGI, SUN) either 0 or 4 can be given for the BLAS level. BLAS=0 should always work and means that the MOLPRO Fortran BLAS routines are used. On some platforms (IBM, SGI, SUN) BLAS=4 will give better performance; in this case some 32-bit BLAS routines are used from the system library (these are then called from wrapper routines, which convert 64 to 32-bit integer arguments. Note that this might cause problems if more than 2 GB of memory is used). For good performance it is important to use appropriate BLAS libraries; in particular, a fast implementation of the matrix multiplication dgemm is very important for M OLPRO. Therefore you should use a system tuned BLAS library whenever available. M OLPRO will automatically detect the most appropriate BLAS library in many cases. In certain cases, in particular when the BLAS library is installed in a non-default location, congure should be directed to the appropriate directory with: ./configure -blaspath /path/to/lib/dir Specication of BLAS libraries can be simplied by placing any relevant downloaded libraries in the directory blaslibs; configure searches this directory (and then, with lower priority, some potential system directories) for libraries relevant to the hardware. For Intel and AMD Linux systems we recommend the following BLAS libraries: MKL ATLAS The Intel Math Kernel Library (MKL) The Automatically Tuned Linear Algebra Software (ATLAS) library. You must use the atlas library specic to your processor:

INSTALLATION FROM SOURCE FILES Pentium III Pentium 4,Xeon AMD Athlon AMD Opteron Linux Linux Linux Linux

v PIIISSE1 P4SSE2 ATHLON HAMMER64SSE2 2 (64 bit)

ACML

When using atlas M OLPRO will automatically compile in the extra lapack subroutines which do not come by default with the package and so the liblapack.a which comes with Atlas is sufcient. For Opteron systems then AMD Core Math Library (ACML) is the preferred blas library.

SGI Altix can use the scsl library is preferred. HP platforms can use the mlib math library. IBM Power platforms can use the essl package. 5. configure prompts for the optional bin directory (INSTBIN) for linking M OLPRO. This directory should be one normally in the PATH of all users who will access M OLPRO, and its specication will depend on whether the installation is private or public. 6. configure prompts for the Molpro installation directory (PREFIX). 7. configure prompts for the destination directory for documentation. This should normally be a directory that is mounted on a worldwide web server. This is only relevant if the documentation is also going to be installed from this directory (see below). The full list of command-line options recognized by configure are: -af90 -aims -auto-ga-mpich2 use Absoft Pro Fortran compiler compile aims code auto-build GA with MPI and MPICH2 (experimental)

-auto-ga-mvapich2 auto-build GA with MPI and MVAPICH2 (experimental) -auto-ga-openmpi auto-build GA with MPI and Open MPI (experimental) -auto-ga-tcgmsg auto-build GA with TCGMSG (experimental) -auto-ga-tcgmsg-mpich2 auto-build GA with TCGMSG-MPI and MPICH2 (experimental) -auto-ga-tcgmsg-mvapich2 auto-build GA with TCGMSG-MPI and MVAPICH2 (experimental) -auto-ga-tcgmsg-openmpi auto-build GA with TCGMSG-MPI and Open MPI (experimental) -auto-mpich2 -auto-mvapich2 -auto-openmpi -batch -blas -blaspath -cc -clearspeed -clearspeedbase auto-build MPICH2 (experimental) auto-build MVAPICH2 (experimental) auto-build Open MPI (experimental) run script non-interactively use external BLAS library specify blas library path use C compiler named cc compile clearspeed code specify clearspeed base path for includes and libraries

INSTALLATION FROM SOURCE FILES try to get settings for compiling CUDA code specify Molpro fortran pre-processor ags use f90 Fortran compiler use Fujitsu C compiler Force linking of main executable use fort Fortran compiler use frt Fortran compiler use G95 Fortran compiler use GNU Compiler Collection C compiler Use settings for mpich2 congured with gforker option use gfortran Fortran compiler use external HDF5 library specify HDF5 library path use settings for i386 machine Makes default integer variables 4 bytes long use settings for i686 machine Makes default integer variables 8 bytes long use Intel C compiler use Intel Fortran compiler append PL to PREFIX when running make install Use settings for Intel MPI with LSF use external LAPACK library specify LAPACK library path specify letter latex paper size use settings for Linux kernel produce parallel Molpro specify mpp base path for includes and libraries produce trivial parallel Molpro use NAG Fortran compiler max number of atoms max number of basis functions Dont use external BLAS library do not compile clearspeed code dont compile CUDA code do not compile C++ code dont use external HDF5 library dont use external LAPACK library Do not use largeles compile without openmp

vi

-cuda -f -f90 -fcc -force-link -fort -frt -g95 -gcc -gforker -gfortran -hdf5 -hdf5path -i386 -i4 -i686 -i8 -icc -ifort -inst-pl -intel-mpi-lsf -lapack -lapackpath -letter -Linux -mpp -mppbase -mppx -nagfor -natom -nbasis -noblas -noclearspeed -nocuda -nocxx -nohdf5 -nolapack -nolargefiles -noopenmp

INSTALLATION FROM SOURCE FILES max number of primitives max number of records max number of states per symmetry max number of state symmetries max number of valence orbitals use NVIDIA CUDA C compiler use Open64 C compiler use Open64 Fortran compiler compile with openmp Use settings for standard openmpi Use settings for openmpi compiled with SGE use Pathscale C compiler use Pathscale Fortran compiler use Portland C compiler use Portland Fortran compiler Specify top-level installation directory Build with RPM settings compile slater code Use settings for sm 13 architecture for CUDA compilation Use settings for sm 20 architecture for CUDA compilation use Sun C compiler use Sun Fortran compiler use trial parse object use settings for 64-bit x86 machine use IBM Fortran compiler

vii

-nprim -nrec -nstate -nsymm -nvalence -nvcc -opencc -openf90 -openmp -openmpi -openmpi-sge -pathcc -pathf90 -pgcc -pgf90 -prefix -rpm -slater -sm 13 -sm 20 -suncc -sunf90 -trial -x86 64 -xlf

3.4

Compilation and linking

After conguration, the remainder of the installation is accomplished using the GNU make command. Remember that the default make on many systems will not work, and that it is essential to use GNU make (cf. section 3.2). Everything needed to make a functioning program together with all ancillary les is carried out by default simply by issuing the command make in the M OLPRO base directory. Most of the standard options for GNU make can be used safely; in particular, -j can be used to speed up compilation on a parallel machine. The program can then be accessed by making sure the bin/ directory is included in the PATH and issuing the command molpro. If MPI library is used for building Global Arrays or building M OLPRO directly, please be aware that some MPI libraries use mpd daemons to launch parallel jobs. In this case, mpd daemons must already be running before make.

INSTALLATION FROM SOURCE FILES

viii

3.5

Adjusting the default environment for M OLPRO

The default running options for M OLPRO are stored in the script bin/molpro. After program installation, either using binary or from source les, this le should be reviewed and adjusted, if necessary, to make system wide changes.

3.6

Tuning

M OLPRO can be tuned for a particular system by running in the root directory the command make tuning This job automatically determines a number of tuning parameters and appends these to the le bin/molpro. Using these parameters, M OLPRO will select the best BLAS routines depending on the problem size. This job should run on an empty system. It may typically take 10 minutes, depending on the processor speed, and you should wait for completion of this run before doing the next steps.

3.7

Testing

At this stage, it is essential to check that the program has compiled correctly. The makele target test (i.e., command make test) will do this using the full suite of test jobs, and although this takes a signicantly long time, it should always be done when porting for the rst time. A much faster test, which checks the main routes through the program, can be done using make quicktest. For parallel installation, it is highly desirable to perform this validation with more than one running process. This can be done conveniently through the make command line as, for example, make MOLPRO OPTIONS=-n2 test If any test jobs fail, the cause must be investigated. It may be helpful in such circumstances to compare the target platform with the lists of platforms on which M OLPRO is thought to function at http://www.molpro.net/supported. If, after due efforts to x problems of a local origin, the problem cannot be resolved, the developers of M OLPRO would appreciate receiving a report. There is a web-based mechanism at https://www.molpro.net/bugzilla at which as many details as possible should be lled in. It may also be helpful to attach a copy of the CONFIG le along with the failing output. Please note that the purpose of such bug reports is to help the developers improve the code, and not for providing advice on installation or running.

3.8

Installing the program for production

Although the program can be used in situ, it is usually convenient to copy only those les needed at run time into appropriate installation directories as specied at conguration time (see section 3.3) and stored in the le CONFIG. To install the program in this way, do make install The complete source tree can then be archived and deleted. The overall effect of this is to create a shell script in the INSTBIN directory. The name should relate to the architecture, type of build, integer etc. Symbolic links relating to the type of build are then made, and nally providing that INSTBIN/molpro is not a le, a symbolic link is created to the new script. In some cases it is preferable to create a localized script in INSTBIN/molpro which will not be over written. The overall effect of this cascade of links is to provide, in the normal case, the

INSTALLATION FROM SOURCE FILES

ix

commands molpro and one or both of molpros (serial) and molprop (parallel) for normal use, with the long names remaining available for explicit selection of particular variants. For normal single-variant installations, none of the above has to be worried about, and the molpro command will be available from directory INSTBIN. During the install process the key from $HOME/.molpro/token is copied to PREFIX/.token so that the key will work for all users of the installed version.

3.9

Getting and applying patches

Normally, the distribution when downloaded is fully up to date, and initial patching is not necessary. However, bug xes and updates may be desired subsequently. The mechanism for updating M OLPRO source code with bug xes and new features is through the provision of selfcontained patch les, which, when applied, replace or add les, and store the replaced code in order to allow later reversion to the original. Those patches that are available can be seen at https://www.molpro.net/download/patch?version=2010.1 whilst a list of those already installed is printed when running the program. Patch les automatically outdate any targets that need rebuilding as a result of the patch; for example, relevant object les are removed. Thus, after all patches have been applied, it is usually necessary to rebuild the program using make. The order in which patches are applied and removed is important. Some patches are prerequisites of others, and some patches are parents of one or more children: the parent and child patches have one or more les in common, but the parent is older than the child. Individual patch scripts will themselves refuse to apply or revert if rules based on these considerations would be violated. In order to deal with this issue smoothly, a program patcher is provided to manage the application and removal of one or more patches. patcher attempts to sort the order in which patches are applied or reverted so as to avoid such conicts; it will also, if necessary, revert and reapply patches. In order to run patcher you should issue the command: make patch This should be sufcient for most purposes. patcher will be built if has not yet been compiled and then it will contact the webserver, apply any available patches and then return the patchlevel that you have reached. If it is necessary to pass arguments to the patcher program then in the top-level directory issue the command ./patcher [--apply | --revert | --list] [--cache-directory] [--user] [--password] [--url] [--local] [--verbose] [--no-action] patch1 patch2 ....

It can operate in one of three possible modes according to the options --apply, -a --revert, -r --list, -l (default) Apply (i.e. install) patches Revert (i.e. remove) patches List available and installed patches

INSTALLATION FROM SOURCE FILES

The list of patches to remove or install can be given on the command line after all options as an explicit list of either patch names or, in the case of application, patch les. Alternatively and usually, for the case of application, one can through options request either all patches that are in a local cache, or all patches that are available. The M OLPRO patches from the central web server (default http://www.molpro.net), are cached by this program in a local directory (default $HOME/.molpro/cache). Access to the web server typically has to be authenticated; the rst time you run this program, you can specify your username and password through command-line options, or else the program will prompt for them. They are then remembered in the le CONFIG in the cache directory. In case of problems, rst consult the le patcher.log, which contains the output from individual patch applications and reversions. The following options can be given. --cache-directory, -c d --verbose, -v --noverbose --url --user, -u u --password, -p p --noaction, -n --local --token, -k --ssl, -s --nossl, -i Examples: patcher Applies all patches that are available, but not yet installed. This is the normal use of the utility in bringing the copy of the source tree up to date with all available updates. patcher -l Lists installed and available patches. patcher -r xx yy Reverts patches xx and yy. patcher -n Loads all uninstalled patches into the cache for later use. patcher --local Applies all patches in the cache; no network connection needed. location of cache directory. Increase amount of information printed. Multiple verbose options can be used. Decrease amount of information printed. URL of web server. Username for web server. Password for web server. No applications or reversions are actually done. Useful for seeing what would happen without doing it. Dont attempt to access the web server, but use only local les. Download your licence key Use SSL when contacting the webserver Turn off SSL use

INSTALLATION FROM SOURCE FILES

xi

3.10

Installation of documentation

The documentation is available on the web at http://www.molpro.net/info/users. It is also included with the source code. The PDF users manual is found in the directory molpro2010.1/doc/manual.pdf, with the HTML version in the directory molpro2010.1/doc/manual (top level le is frames.html). The documentation can be copied to its nal destination as specied in the CONFIG le generated by the configure command. To install the documentation and interactive basis set tool, issue make install in the doc directory. Numerous example input les are included in the manual, and can alternatively be seen in the directory molpro2010.1/examples.

3.11

Fedora 13 (32-bit)

Installing M OLPRO from source on Fedora 13 Linux can be done very easily using only free software. These notes assume a standard installation of Fedora 13. For all builds one should install the following additional packages via yum: gcc-c++ gcc-gfortran zlib-devel provides GNU C and C++ compiler, provides GNU Fortran compiler, provides a library for zipped les.

Optionally one can choose to install: blas-devel lapack-devel provides a BLAS library, provides a LAPACK library,

which will be used instead of compiling the equivalent M OLPRO routines. 3.11.1 Serial M OLPRO

After unpacking M OLPRO one can simply type ./configure -batch -gcc -gfortran make to build M OLPRO. 3.11.2 Parallel M OLPRO using the Global Arrays with MPI2

The following additional packages should be installed via apt-get: mpich2 mpich2-devel Set up mpd by running: mpd provides MPI-2 library les provides MPI-2 include les

INSTALLATION FROM SOURCE FILES

xii

and following the on screen instructions. The latest stable version of the Global Arrays toolkit should be downloaded. Build and test the Global Arrays toolkit with: make TARGET=LINUX FC=gfortran CC=gcc USE_MPI=1 \ MPI_LIB=/usr/lib/mpich2/lib \ MPI_INCLUDE=/usr/include/mpich2-i386 LIBMPI=-lmpich mpirun ./global/testing/test.x Then congure and build M OLPRO with: ./configure -batch -gcc -gfortran \ -mpp -mppbase /path/to/directory/ga-4-3-3 make

3.12

openSUSE 11.3 (32 & 64-bit)

Installing M OLPRO from source on openSUSE 11.3 Linux can be done very easily using only free software. These notes assume a standard installation of openSUSE 11.3. For all builds one should install the following additional packages via YaST: gcc-c++ gcc-fortran make zlib-devel provides GNU C and C++ compiler, provides GNU Fortran compiler, provides GNU make provides a library for zipped les.

Optionally one can choose to install: blas lapack provides a BLAS library, provides a LAPACK library,

which will be used instead of compiling the equivalent M OLPRO routines. 3.12.1 Serial M OLPRO

After unpacking M OLPRO one can simply type ./configure -batch -gcc -gfortran make to build M OLPRO.

INSTALLATION FROM SOURCE FILES Parallel M OLPRO using the Global Arrays with MPI

xiii

3.12.2

The following additional packages should be installed via YaST: mpich mpich-devel provides MPI-1 library provides MPI-1 library

The latest stable version of the Global Arrays toolkit should be downloaded. Build and test the Global Arrays toolkit for 64-bit machines with: make TARGET=LINUX64 FC=gfortran CC=gcc USE_MPI=1 \ MPI_LIB=/opt/mpich/ch-p4/lib64 \ MPI_INCLUDE=/opt/mpich/ch-p4/include LIBMPI=-lmpich /opt/mpich/ch-p4/bin/mpirun ./global/testing/test.x or on 32-bit machines with: make TARGET=LINUX FC=gfortran CC=gcc USE_MPI=1 \ MPI_LIB=/opt/mpich/ch-p4/lib \ MPI_INCLUDE=/opt/mpich/ch-p4/include LIBMPI=-lmpich /opt/mpich/ch-p4/bin/mpirun ./global/testing/test.x Then congure and build M OLPRO with: ./configure -batch -gcc -gfortran \ -mpp -mppbase /path/to/directory/ga-4-3-3 make

3.13

Ubuntu 10.04 (32-bit)

Installing M OLPRO from source on Ubuntu 10.04 Linux can be done very easily using only free software. These notes assume a standard installation of Ubuntu 10.04. For all builds one should install the following additional packages via apt-get: build-essential gfortran curl zlib1g-dev provides GNU C++ compiler, provides GNU Fortran compiler, provides curl for downloading patches provides a library for zipped les.

Optionally one can choose to install: libblas-dev liblapack-dev provides a BLAS library, provides a LAPACK library,

which will be used instead of compiling the equivalent M OLPRO routines.

INSTALLATION FROM SOURCE FILES Serial M OLPRO

xiv

3.13.1

After unpacking M OLPRO one can simply type ./configure -batch -gcc -gfortran make to build M OLPRO. 3.13.2 Parallel M OLPRO using the Global Arrays with MPI

The following additional packages should be installed via apt-get: mpich-bin libmpich1.0-dev openssh-server provides MPI-1 library provides MPI-1 library provides ssh access to localhost

Set up password-less ssh by running the following commands and not entering a password when prompted:: ssh-keygen -t rsa cat /.ssh/id_rsa.pub >> /.ssh/authorized_keys This must be done for each user account which will be running M OLPRO. The latest stable version of the Global Arrays toolkit should be downloaded. Build and test the Global Arrays toolkit with: make TARGET=LINUX FC=gfortran CC=gcc USE_MPI=1 \ MPI_LIB=/usr/lib/mpich/lib \ MPI_INCLUDE=/usr/lib/mpich/include LIBMPI=-lmpich mpirun ./global/testing/test.x Then congure and build M OLPRO with: ./configure -batch -gcc -gfortran \ -mpp -mppbase /path/to/directory/ga-4-3-3 make

3.14

Installation on a Cygwin system

On a Windows machine Cygwin should be installed. In addition to the default package list one should also install the packages listed in table 1. If undertaking development work table 2 contains a list of potentially useful packages. With the above steps Molpro should now congure and will treat Cygwin as a generic Linux system. Once the program has been built it can be run in the normal way under Cygwin.

INSTALLATION FROM SOURCE FILES Package gcc4 gcc4-fortran gcc4-g++ make curl Package Group Devel Devel Devel Devel Web Reason For compiling C les For compiling Fortran les For compiling C++ les need GNU make curl needed for patcher

xv

Table 1: Cygwin requirments for user install Package bison gdb libxslt openssh vim wget Package Group Devel Devel Gnome Net Editors Web Reason bison gdb xsltproc ssh vi wget, alternative to curl for patcher

Table 2: Cygwin packages for developers

3.15

Parallel installation on 64-bit MacOS 10.6.4 standalone machine

This section contains notes and examples for building some of the programs needed for building parallel Molpro. It is not complete or extensive, and in all cases one should refer directly to the documentation of the program in question and forward any queries directly there. 3.15.1 Building MPICH2-1.3

The following instructions are for building MPICH2 with ifort and installing in /usr/local/mpich2-nemesis. ./configure --with-device=ch3:nemesis --with-pm=gforker \ --prefix=/usr/local/mpich2-nemesis --enable-fc FC=ifort \ --enable-cc CC=gcc --enable-cxx CXX=g++ make sudo make install 3.15.2 Building Global Arrays

The following instructions are for building GA using the MPICH2 installation detailed in the previous section. make CC=gcc FC=ifort TARGET=MACX64 USE_MPI=y \ MPI_LIB=/usr/local/mpich2-nemesis/lib \ MPI_INCLUDE=/usr/local/mpich2-nemesis/include \ LIBMPI="-lpmpich -lmpich -lopa -lmpl" export PATH=/usr/local/mpich2-nemesis/bin:$PATH mpiexec -n 2 global/testing/test.x Note that if ___emutls_get_address is reported as an undened reference, then -lgcc_eh must be appended to LIBMPI.