 |
 |
 |
| |
|
|
In this page you can learn how to configure, compile, and install PLUMED. For those of you who are impatient, the following might do the job:
> ./configure --prefix=/usr/local > make -j 4 > make doc # this is optional and requires proper doxygen version installed > make install
Notice that make install is not strictly necessary as plumed can be used from the compilation directory. This is very useful so as to quickly test the implementation of new features. However, we strongly recommend to perform a full install.
Once the above is completed the plumed executable should be in your execution path and you will be able to use PLUMED to analyze existing trajectories or play with the Lennard-Jones code that is included. However, because PLUMED is mostly used to bias on the fly simulations performed with serious molecular dynamics packages, you can find instructions about how to patch your favorite MD code so that it can be combined with PLUMED below. Again, if you are impatient, something like this will do the job:
> cd /md/root/dir > plumed patch -p
Then compile your MD code. For some MD codes these instructions are insufficient. It is thus recommended that you read the instructions at the end of this page. Notice that MD codes could in principle be "PLUMED ready" in their official distribution. If your favorite MD code is available "PLUMED ready" you will have to compile PLUMED first, then (optionally) install it, then check the MD codes' manual to discover how to link it.
As of PLUMED 2.10, we require a compiler that supports C++17.
Notice that the ./configure script verifies that your compiler supports C++17. Some compilers do not declare full support, but implement anyway a number of C++17 features sufficient to compile PLUMED. In case you see a warning about C++17 support during ./configure please make sure that PLUMED compiles correctly and, if possible, execute the regtests (using make regtest). Notice that we regularly test a number of compilers on travis-ci, and at least those compilers are guaranteed to be able to compile PLUMED correctly.
The ./configure command just generates a Makefile.conf file and a sourceme.sh file. In PLUMED 2.0 these files were prepared and stored in the directory configurations/. The new ones generated by ./configure are similar to the old ones but are not completely compatible. In particular, some of the -D options have been changed in version 2.2, and several new variables so as to specify the installation directories have been added. For this reason, you now should run ./configure again. Anyway, it should be easy to enforce a similar setup with autoconf by passing the proper arguments on the command line. If you have problems on your architecture, please report them to the mailing list.
Useful command line options for ./configure can be found by typing
> ./configure --help
PLUMED is made up of modules. Some of them are on by default, some others aren't. Since version 2.3, the activation of modules should be made during configuration using the --enable-modules option (see List of modules).
Notice that some of the methods within PLUMED depend on external libraries which are looked for by configure. You can typically avoid looking for a library using the "disable" syntax, e.g.
> ./configure --disable-mpi --disable-gsl
Notice that when MPI search is enabled (by default) compilers such as "mpic++" and "mpicxx" are searched for first. On the other hand, if MPI search is disabled ("./configure --disable-mpi") non-mpi compilers are searched for. Notice that only a few of the possible compiler name are searched. Thus, compilers such as "g++-mp-4.8" should be explicitly requested with the CXX option.
You can better control which compiler is used by setting the variables CXX and CC. E.g., to use Intel compilers use the following command:
> ./configure CXX=icpc CC=icc
Notice that we are using icpc in this example, which is not an MPI compiler as a result MPI will not be enabled. Also consider that this is different with respect to what some other configure script does in that variables such as MPICXX are completely ignored here. In case you work on a machine where CXX is set to a serial compiler and MPICXX to a MPI compiler, to compile with MPI you should use
> ./configure CXX="$MPICXX"
--enable-mpi is perfectly valid but is not needed here. Autoconf will check if a code containing MPI calls can be compiled, and if so it will enable it. --disable-mpi could be used if you are using a compiler that supports MPI but you don't want PLUMED to be compiled with MPI support. Thus the correct way to enable MPI is to pass to ./configure the name of a C++ compiler that implements MPI using the CXX option. In this way, MPI library is treated similarly to all the other libraries that PLUMED tries to link by default.To tune the compilation options you can use the CXXFLAGS variable:
> ./configure CXXFLAGS=-O3
If you are implementing new functionality and want to build with debug flags in place so as to do some checking you can use
> ./configure --enable-debug
This will perform some extra check during execution (possibly slowing down PLUMED) and write full symbol tables in the executable (making the final executable much larger).
The main goal of the automatic configure is to find the libraries. When they are stored in unconventional places it is thus sensible to tell autoconf where to look! To do this there are some environment variable that can be used to instruct the linker which directories it should search for libraries inside. These variables are compiler dependent, but could have been set by the system administrator so that libraries are found without any extra flag. Our suggested procedure is to first try to configure without any additional flags and to then check the log so as to see whether or not the libraries were properly detected.
If a library is not found during configuration, you can try to use options to modify the search path. For example if your gsl libraries is in /opt/local (this is where MacPorts put it) and configure is not able to find it you can try
> ./configure LDFLAGS=-L/opt/local/lib CPPFLAGS=-I/opt/local/include
Notice that PLUMED will first try to link a routine from say gsl without any additional flag, and then in case of failure will retry adding "-lgsl" to the LIBS options. If also this does not work, the gsl library will be disabled and some features will not be available. This procedure allows you to use libraries with custom names. So, if your gsl library is called /opt/local/lib/libmygsl.so you can link it with
> ./configure LDFLAGS=-L/opt/local/lib CPPFLAGS=-I/opt/local/include LIBS=-lmygsl
In this example, the linker will directly try to link /opt/local/lib/libmygsl.so. This rule is true for all the libraries, so that you will always be able to link a specific version of a library by specifying it using the LIBS variable.
Since version 2.3.2, the search for the library functions passing to the linker a flag with the standard library name (in the gsl example, it would be -lgsl) can be skipped by using the option --disable-libsearch. Notice that in this manner only libraries that are explicitly passed using the LIBS option will be linked. For instance
> ./configure --disable-libsearch LIBS=-lgsl
will make sure that only gsl is linked and, for instance, BLAS and LAPACK libraries are not. This might be useful when installing PLUMED within package managers such as MacPorts to make sure that only desired libraries are linked and thus to avoid to introduce spurious dependencies. The only exception to this rule is -ldl, which is anyway a system library on Linux.
libplumed.dylib library, which has an install name hard coded). On the other hand, on Linux it is common practice not to hard code the full path. This means that if you use the LDFLAGS option to specify the path to the libraries you want to link to PLUMED (e.g. ./configure LDFLAGS="-L/path") these libraries might not be found later. The visible symptom is that src/lib/plumed-shared will not be linked correctly. Although the file 'src/lib/plumed-shared' is not necessary, being able to produce it means that it will be possible to link PLUMED dynamically with MD codes later. The easiest solution is to hard code the library search path in this way: > ./configure LDFLAGS="-L/path -Wl,-rpath,/path"Notice that as of PLUMED v2.4 it is possible to use the configure option
--enable-rpath to automatically hard code the path defined in LIBRARY_PATH: > ./configure LIBRARY_PATH=/path --enable-rpathIn this way, the search path used at link time (
LIBRARY_PATH) and the one saved in the libplumed.so library will be consistent by construction. In a typical environment configured using module framework (http://modules.sourceforge.net), LIBRARY_PATH will be a variable containing the path to all the modules loaded at compilation time.PLUMED needs BLAS and LAPACK. These are treated slightly different from other libraries. The search is done in the usual way (i.e., first look for them without any link flag, then add "-lblas" and "-llapack", respectively). As such if you want to use a specific version of BLAS or LAPACK you can make them available to configure by using
> ./configure LDFLAGS=-L/path/to/blas/lib LIBS=-lnameoflib
If the functions of these libraries are not found, the compiler looks for a version with a final underscore added. Finally, since BLAS and LAPACK are compulsory in PLUMED, you can use a internal version of these libraries that comes as part of PLUMED. If all else fails the internal version of BLAS and LAPACK are the ones that will be used by PLUMED. If you wish to disable any search for external libraries (e.g. because the system libraries have problems) this can be done with
> ./configure --disable-external-blas
Notice that you can also disable external LAPACK only, that is use internal LAPACK with external BLAS using
> ./configure --disable-external-lapack
Since typically it is the BLAS library that can be heavily optimized, this configuration should not provide significant slowing down and could be used on systems where native LAPACK libraries have problems.
As a final resort, you can also edit the resulting Makefile.conf file. Notable variables in this file include:
configure LIBS=something will end up in this variable. This is a bit misleading but is required to keep the configuration files compatible with PLUMED 2.0.We tried to keep PLUMED as independent as possible from external libraries and as such those features that require external libraries are optional. However, to have a properly working version of plumed PLUMED you need BLAS and LAPACK libraries. We would strongly recommend you download these libraries and install them separately so as to have the most efficient possible implementations of the functions contained within them. However, if you cannot install BLAS and LAPACK, you can use the internal ones. Since version 2.1, PLUMED uses a configure script to detect libraries. In case system LAPACK or BLAS are not found on your system, PLUMED will use the internal replacement.
We have had a number of emails (and have struggled ourselves) with ensuring that PLUMED can link BLAS and LAPACK. The following describes some of the pitfalls that you can fall into and a set of sensible steps by which you can check whether or not you have set up the configuration correctly.
Notice first of all that the DYNAMIC_LIB variable in the Makefile.conf should contain the flag necessary to load the BLAS and LAPACK libraries. Typically this will be -llapack -lblas, in some case followed by -lgfortran. Full path specification with -L may be necessary and on some machines the BLAS and LAPACK libraries may not be called -llapack and -lblas. Everything will depend on your system configuration.
Some simple to fix further problems include:
PLUMED source code already includes a few selected VMD molfile plugins so as to read a small number of additional trajectory formats (e.g., dcd, gromacs files, pdb, and amber files). If you configure PLUMED with the full set of VMD plugins you will be able to read many more trajectory formats, basically all of those supported by VMD. To this aim, you need to download the SOURCE of VMD, which contains a plugins directory. Adapt build.sh and compile it. At the end, you should get the molfile plugins compiled as a static library libmolfile_plugin.a. Locate said file and libmolfile_plugin.h, they should be in a directory called /pathtovmdplugins/ARCH/molfile (e.g. /pathtovmdplugins/MACOSXX86_64/molfile). Also locate file molfile_plugin.h, which should be in /pathtovmdplugins/include. Then customize the configure command with something along the lines of:
> ./configure LDFLAGS="-L/pathtovmdplugins/ARCH/molfile" CPPFLAGS="-I/pathtovmdplugins/include -I/pathtovmdplugins/ARCH/molfile"
Notice that it might be necessary to add to LDFLAGS the path to your TCL interpreter, e.g.
> ./configure LDFLAGS="-ltcl8.5 -L/mypathtotcl -L/pathtovmdplugins/ARCH/molfile" \
CPPFLAGS="-I/pathtovmdplugins/include -I/pathtovmdplugins/ARCH/molfile"
Then, rebuild plumed.
In order to use machine learning models optimized with PyTorch (as in the PYTORCH module) or specific actions implemented in the PLUMED-ISDB module, one needs to link the LibTorch C++ library. To do so, one can follow these instructions to download the pre-compiled library and configure PLUMED to use it.
Download LibTorch C++ API library
You can download the pre-built LibTorch library from their website. For example, the following script downloads the libtorch-cxx11-abi-shared-with-deps-2.0.0%2Bcpu.zip (2.0.0, CPU, with C++11 ABI compatibility).
wget https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-2.0.0%2Bcpu.zip unzip libtorch-cxx11-abi-shared-with-deps-2.0.0+cpu.zip ;
If you have a GPU, you might want to use the CUDA-accelerated version of LibTorch. For example, the following script downloads the libtorch-shared-with-deps-2.0.0%2Bcu117.zip (2.0.0, GPU, Cuda 11.7, pre-cxx11 ABI binary).
wget https://download.pytorch.org/libtorch/cu117/libtorch-shared-with-deps-2.0.0%2Bcu117.zip unzip libtorch-shared-with-deps-2.0.0+cu117.zip
In both CPU and GPU cases, the location of the include and library files need to be exported in the environment:
LIBTORCH=${PWD}/libtorch
export CPATH=${LIBTORCH}/include/torch/csrc/api/include/:${LIBTORCH}/include/:${LIBTORCH}/include/torch:$CPATH
export INCLUDE=${LIBTORCH}/include/torch/csrc/api/include/:${LIBTORCH}/include/:${LIBTORCH}/include/torch:$INCLUDE
export LIBRARY_PATH=${LIBTORCH}/lib:$LIBRARY_PATH
export LD_LIBRARY_PATH=${LIBTORCH}/lib:$LD_LIBRARY_PATH
Remember to add these lines also in your ~/.bashrc or ~/.bash_profile file.
Configure PLUMED
Once the environment variables are set, we can configure PLUMED with the --enable-libtorch keyword:
> ./configure --enable-libtorch
Notes
--enable-modules=pytorch or --enable-modules=all.--enable-libtorch will first try first to link the CUDA-enabled library and if it does not found it will try to link the CPU-only version.checking libtorch[cpu/cuda] [without extra libs/with -ltorch_cpu ... ]. If any of these commands are succesfull, it will return ... yes. Otherwise, the configure will display a warning (and not an error!) that says: configure: WARNING: cannot enable __PLUMED_HAS_LIBTORCH. In this case, it is recommended to examine the output of the above commands in the config.log file to understand the reason (e.g. it cannot find the required libraries).libtorch-shared-with-deps-2.0.0%2Bcpu.zip) and add the following option to the configure: CXXFLAGS="-D_GLIBCXX_USE_CXX11_ABI=0".PLUMED includes some additional modules that by default are not compiled, but can be enabled during configuration. You can use the option --enable-modules to activate some of them, e.g.
> ./configure --enable-modules=module1name+module2name
For more information on modules see List of modules.
Once configured, PLUMED can be compiled using the following command:
> make -j 4
This will compile the entire code and produce a number of files in the 'src/lib' directory, including the executable 'src/lib/plumed'. When shared libraries are enabled, a shared libraries called 'src/lib/libKernel.so' should also be present. Notice that the extension could be '.dylib' on a Mac.
In case you want to run PLUMED without installing it (i.e. from the compilation directory), you can use the file 'sourceme.sh' that has been created by the configure script in the main PLUMED directory. This file can be "sourced" (presently only working for bash shell) if you want to use PLUMED before installing it (i.e. from the compilation directory). It is a good idea to source it now, so that you can play with the just compiled PLUMED:
> source sourceme.sh
Now a "plumed" executable should be in your path. Try to type
> plumed -h
You can also check if PLUMED is correctly compiled by performing our regression tests. Be warned that some of them fail because of the different numerical accuracy on different machines. As of version 2.4, in order to test the plumed executable that you just compiled (prior to installing it) you can use the following command
> make check
On the other hand, in order to test the plumed executable that you just installed (see Installing PLUMED) you should type
> make installcheck
In addition, similarly to previous versions of PLUMED, you can test the plumed executable that is in your current path with
> cd regtest > make
You can check the exact version they will use by using the command
> which plumed
Thus, you can easily run the test suite using a different version of PLUMED (maybe an earlier version that you already installed), just making sure that it can be found in the path. Clearly, if you test a given version of PLUMED with a test suite from a different version you can expect two possible kinds of innocuous errors:
plumed executable is older than the test suite, the tests might fail since they rely on some feature introduced in PLUMED in a newer version.plumed executable is newer than the test suite, the tests might fail since some non-backward compatible change was made in PLUMED. We try to keep the number of non-backward compatible changes small, but as you can see in the Change Log there are typically a few of them at every new major release.Notice that the compiled executable, which now sits in 'src/lib/plumed', relies on other resource files present in the compilation directory. This directory should thus stay in the correct place. One should thus not rename or delete it. In fact the path to the PLUMED root directory is hard coded in the plumed executable as can be verified using
> plumed info --root
In case you try to use the plumed executable without the compilation directory in place (e.g. you move away the src/lib/plumed static executable and delete or rename the compilation directory) PLUMED will not work correctly and will give you an error message
> plumed help ERROR: I cannot find /xxx/yyy/patches directory
You can force plumed to run anyway by using the option –standalone-executable:
> plumed --standalone-executable help
Many features will not be available if you run in this way. However, this is currently the only way to use the PLUMED static executable on Windows.
It is strongly suggested to install PLUMED in a predefined location. This is done using
> make install
This will allow you to remove the original compilation directory, or to recompile a different PLUMED version in the same place.
To install PLUMED one should first decide the location:
> ./configure --prefix=$HOME/opt > make > make install
As of PLUMED 2.5 you cannot anymore change the location during install. If you didn't specify the --prefix option during configure PLUMED will be installed in /usr/local. The install command should be executed with root permissions (e.g. "sudo make install") if you want to install PLUMED on a system directory.
Notice that upon installation PLUMED might need to relink a library. This was always true until version 2.1, but in version 2.2 libraries should only be relinked if one changes the install prefix during when typing make install. If root user does not have access to compilers, "sudo -E make install" might solve the issue.
Upon install, the executable is copied to $prefix/bin, libraries to $prefix/lib, include files to $prefix/include, and documentation to $prefix/shared/doc/plumed. Additionally, a directory $prefix/lib/plumed is created containing several other files, including patch files, object files (for static patches), etc. Notice also that these path can be further customized using standard autoconf directories (e.g. ./configure --bindir=/usr/bin64).
One should then set the environment properly. We suggest to do it using the module framework (http://modules.sourceforge.net). An ad hoc generated module file for PLUMED can be found in $prefix/lib/plumed/src/lib/modulefile Just edit it as you wish and put it in your modulefile directory. This will also allow you to install multiple PLUMED versions on your machine and to switch among them. If you do not want to use modules, you can still have a look at the modulefile we did so as to know which environment variables should be set for PLUMED to work correctly.
If the environment is properly configured one should be able to do the following things:
As a final note, if you want to install several PLUMED versions without using modules then you should provide a different suffix and/or prefix at configure time:
> ./configure prefix=$HOME/opt --program-suffix=_2.2 --program-prefix=mpi- > make install
This will install a plumed executable named "mpi-plumed_2.2". All the other files will be renamed similarly, e.g. the PLUMED library will be loaded with "-lmpi-plumed_2.2" and the PLUMED header files will be included with "#include <mpi-plumed_2.2/tools/Vector.h>". Notice that you can also use arbitrary scripts to edit the name of the executable with the option –program-transform-name=PROGRAM (see autoconf documentation for more info). These options are useful if you do not want to set up modules, but we believe that using modules as described above is more flexible.
A growing number of MD codes can use PLUMED without any modification. If you are using one of these codes, refer to its manual to know how to activate PLUMED. In case your MD code is not supporting PLUMED already, you should modify it. We provide scripts to adjust some of the most popular MD codes so as to provide PLUMED support. At the present times we support patching the following list of codes:
In the section Code specific notes you can find information specific for each MD code.
To patch your MD code, you should have already installed PLUMED properly. This is necessary as you need to have the command "plumed" in your execution path. As described above this executable will be in your paths if plumed was installed or if you have run sourceme.sh
Once you have a compiled and working version of plumed, follow these steps to add it to an MD code
> plumed patch -pThe script will interactively ask which MD engine you are patching.
There are different options available when patching. You can check all of them using
> plumed patch --help
Particularly interesting options include:
./configure script does its best in this sense, but sometime it cannot solve the problem. Additionally, this patching mode has been reported not to work properly on OSX.Notice that with PLUMED version <2.5 there was no possibility to link PLUMED as a static library (something like 'libplumed.a'). However, starting with PLUMED 2.5, the ./configure script will try to set up the system so that a libplumed.a file is produced. Patching an MD code with --static with try to link against this static library. Creation of the libplumed.a library can be avoided with ./configure --disable-static-archive.
If your MD code is not supported, you may want to implement an interface for it. Refer to the developer manual .
If you are compiling an executable from a different machine, then plumed executable will not be available in the compilation environment. This means that you won't be able to perform regtests on the machine nor to compile the manual. You can try to run the regtests on the computing nodes, but this might require some tweak since often machines where people do cross compiling have architectures with limited capabilities on the compute nodes. Also notice that many of the plumed options (e.g. patch) are implemented as shell scripts launched from within the plumed executable. If the compute nodes have some limitation (e.g. they do not allow to fork new processes) these options will not work. Anyway, the PLUMED library in combination with an MD software should work if both PLUMED and the MD software have been properly compiled.
Also notice that it will not be possible to use the command plumed patch on the machine where you are compiling. You should thus use plumed-patch instead of plumed patch (notice that it should be written as a single word).
Try e.g.:
> plumed-patch --help
This script provides a "shell only" implementation of plumed patch that will skip the launch of the plumed executable.
Notice that other command line tools will be available in the directory prefix/lib/progname/. If configuring with default values this would be /usr/local/lib/plumed/plumed-*. These files are not included in the execution path (prefix/bin) to avoid clashes, but can be executed also when plumed is cross compiled and the main plumed executable cannot be launched.
If you are using a Mac, notice that you can take advantage of a MacPorts package. Installing a working plumed should be as easy as:
sudo port install plumedNotice that plumed comes with many variants that can be inspected with the command
> sudo port info plumed
Plumed uses variants to support different compilers. For instance, you can install plumed with mpich using
> sudo port install plumed +mpich
Using more recent clang instead of native compilers is recommended so as to take advantage of openMP
> sudo port install plumed +mpich +clang50
Notice that support for c++17 with gcc compilers is someway problematic within MacPorts due to impossibility to use the system c++ library. For this reason, only clang compilers are supported (see also this discussion).
Variants can be also used to compile with debug flags (+debug), to pick a linear algebra library (e.g. +openblas) and to enable all optional modules (+allmodules). Notice that the default variant installed with sudo port install plumed is shipped as a compiled binary, which is significantly faster to install.
In addition, we provide a developer version (typically: a later version not yet considered as stable) under the subport plumed-devel that can be installed with
> sudo port install plumed-devel
plumed-devel also supports the same variants as plumed in order to customize the compilation. plumed-devel and plumed cannot be installed at the same time.
It is also possible to install a plumed-patched version of gromacs. For instance, you can use the following command to install gromacs patched with plumed with clang-5.0 compiler and mpich:
> sudo port install plumed +mpich +clang50 > sudo port install gromacs-plumed +mpich +clang50
In case you want to combine gromacs with the unstable version of plumed, use this instead:
> sudo port install plumed-devel +mpich +clang50 > sudo port install gromacs-plumed +mpich +clang50
Notice that gromacs should be compiled using the same compiler variant as plumed (in this example +mpich +clang50). In case this is not true, compilation will fail.
Also notice that gromacs is patched with plumed in runtime mode but that the path of libplumedKernel.dylib in the MacPorts tree is hard coded. As a consequence:
PLUMED_KERNEL environment variable unset (or set to empty), then the MacPorts plumed is used.PLUMED_KERNEL environment variable pointing to another instance of the plumed library, the other instance is used.This is especially useful if you are developing PLUMED since you will be able to install gromacs once for all and combine it with your working version of PLUMED.
If you use the conda package manager you can install a pre-compiled PLUMED binary using the following command:
> conda install -c conda-forge plumed
Similarly, the python wrappers can be installed with
> conda install -c conda-forge py-plumed
These packages are part of conda-forge and as such should be binary compatible with other codes from the same distribution. Notice that it should also be possible to combine the installed plumed kernel with an MD code compiled outside of conda (or within a different conda environment) if plumed is linked in runtime mode. The only variable that you need to set in order to access to the installed plumed kernel is PLUMED_KERNEL (e.g., export PLUMED_KERNEL=/conda/prefix/lib/libplumedKernel.so).
Notice that binaries are only available for Linux and MacOS and that they have a limited number of features. In particular, they do not support MPI and do not include optional modules. However, they can be used to quickly install a working PLUMED version without the need to have a compiler.
Notice that there are additional conda packages on the plumed channel. Those packages are for testing only.
If you are installing PLUMED on a cluster and you want several users to take advantage of it consider the following suggestions.
First of all, we highly recommend using the module file that PLUMED provides to set up the environment. Just edit it as necessary to make it suitable for your environment.
Notice that PLUMED can take advantage of many additional features if specific libraries are available upon compiling it.
Try to patch all MD codes with the --runtime option. This will allow independent update of PLUMED and MD codes. Users will be able to combine any of the installed gromacs/amber/etc versions with any of the installed PLUMED versions. Notice that it is sometime claimed that statically linked codes are faster. In our experience, this is not true. In case you absolutely need a static executable, be ready to face non trivial linking issues. PLUMED is written in C++, thus required the appropriate C++ library to be linked, and might require additional libraries (e.g. libgsl).
Sometime we make small fixes on the patches. For this reason, keep track of which version of PLUMED you used to patch each of the MD code. Perhaps you can call the MD code modules with names such as gromacs/4.6.7p1, gromacs/4.6.7p2 and write somewhere in the module file which version of PLUMED you used. Alternatively, call them something like gromacs/4.6.7p2.2.0. In this way, when we report a bug on the mailing list, users will know if the version they are using is affected by it.
Usually it is not necessary to install both a MPI and a non-MPI PLUMED version. PLUMED library only calls MPI functions when the MD code is compiled with MPI. PLUMED executable calls MPI functions only when it is invoked without --no-mpi. In many machines it is thus sufficient to run the plumed executable on the login node as
> plumed --no-mpi
even though PLUMED was compiled with MPI and the login node does not support MPI. The only case where you might need two different PLUMED installation for compute and login node is when you are cross compiling.
PLUMED needs to be well optimized to run efficiently. If you need a single PLUMED binary to run efficiency on machines with different levels of hardware (e.g.: some of your workstations support AVX and some do not), with intel compiler you can use something like
> ./configure CXX=mpicxx CXXFLAGS="-O3 -axSSE2,AVX"
It will take more time to compile but it will allow you to use a single module. Otherwise, you should install two PLUMED version with different optimization levels.
Using modules, it is not necessary to make the PLUMED module explicitly dependent on the used library. Imagine a scenario where you first installed a module libgsl, then load it while you compile PLUMED. If you provide the following option to configure --enable-rpath, the PLUMED executable and library will remember where libgsl is, without the need to load libgsl module at runtime. Notice that this trick often does not work for fundamental libraries such as C++ and MPI library. As a consequence, usually the PLUMED module should load the compiler and MPI modules.
As of PLUMED 2.5 it is possible to use the PLUMED library through Python wrappers. Notice that this is not something for end users but rather for developers. The interface is very similar to the one used in MD codes linked with PLUMED.
There are two ways to install Python wrappers.
If ./configure finds a python executable that also has the cython module available, Python wrappers will be installed within /prefix/lib/plumed/python. In order to access them, you should add this directory to the environment variable PYTHONPATH. Notice that if your python interpreter has a different name you might have to pass it to ./configure with PYTHON_BIN=python3.6. The whole thing would then be:
Notice that in this manner you will have to commit to a specific python version before installing PLUMED.
If you use multiple python versions, you might find it easier to install the Python wrappers separately from PLUMED. The simplest way is to do it with pip:
Here the --user flag allows you to install the packages on your home. Notice that you don't even need to download PLUMED in order to install the wrappers, but you will need PLUMED in order to use them. You can tell the wrappers where PLUMED is by setting the PLUMED_KERNEL environment variable:
Notice that by installing the wrappers in this manner you will download those that are packaged on Pypi. If you want to install using pip the development version of the wrappers you should download the PLUMED repository and use the following commands:
If you want to install the development version it is recommended to use a virtualenv so that it will not interfere with the released packages.
We here collect a list of suggestions that might be useful on particular machines.
ld -r is not working properly. There is no easy way to detect this at configure time. If during make you receive an error in the form ld: TOC section size exceeds 64kplease configure plumed again with the following flag
> ./configure --disable-ld-r
> export CRAYPE_LINK_TYPE=dynamic
-lmpi_mt -mt_mpi for compiling and linking and the flag -DMPICH_IGNORE_CXX_SEEK for compiling (kindly reported by Abhishek Acharya). You might want to try to configure using > ./configure LDFLAGS=-lmpi_mt CXXFLAGS="-DMPICH_IGNORE_CXX_SEEK -mt_mpi" STATIC_LIBS=-mt_mpiAdding libraries to
STATIC_LIBS uses them for all the linking steps, whereas those in LIBS are only used when linking the PLUMED kernel library. See more at this thread. 
1.8.17