cmake (c46b2)

Using CMake to install CHARMM

CMake is an alternative to install.com for configuring, compiling,
and installing CHARMM. This file describes how to use cmake and
the configure shell script to build CHARMM.

There are two relevant files in the root CHARMM source distribution.
The CMakeLists.txt file is read by CMake to configure the source
for compilation.
The configure file is a shell script wrapper that invokes CMake.

The configure script provides a more traditional command line switch interface.
Also, if CMake is not available, the script will build and use a cmake binary
from the tool/cmake directory. The script ultimately just builds
a list of options and invokes CMake with the list.

Other methods of using CMake such as the ncurses ccmake GUI and
the Qt GUI are untested and unsupported. Support for ccmake may be added in
the future.


* Workflow | Three Steps to a CMake Install
* Configuration | The First Step
* Compilation | The Second Step
* Installation | The Third and Final Step


Top
Workflow

There are three steps, configuration, compilation, and installation to create
a working install of CHARMM from the source distribution using CMake.
The first step, configuration, uses the configure script,
or the cmake command directly,
to configure the source for compiling.
The configuration step occurs in a new empty directory, the build directory.
The second step is to compile the source in the build directory by invoking
the make command which uses the makefile written by CMake in
the configuration step.
The third step is to install the binaries and support files into a path prefix
designated during configuration by typing 'make install'.


Top
Configuration

To configure the source prior to compilation, make a new directory. For example,
on a GNU/Linux system, I might have the CHARMM source in my home directory under
'charmm'.

$ cd ~
$ ls
charmm

Making a new build directory might look like the following.

$ mkdir charmm-build
$ ls
charmm charmm-build

From the new build directory, invoke the configure script with the desired
options.

$ cd charmm-build
$ ../charmm/configure ...

To see the currently supported list of options, use the --help option.

$ ../charmm/configure --help
Usage: configure [--option1] [--option2] ...

Options:

General options

-p <install path>, --prefix <install path>
install CHARMM into the given directory
-g, --debug
include debug symbols in final exec
-l, --lite
lite build of CHARMM (minimal pref.dat keywords)
-c <path to cmake>, --cmake <path to cmake>
use <path to cmake> to configure the build
-a <keyword1,keyword2,...,keywordN>, --add <keyword1,keyword2,...,keywordN>
add the keywords to the final pref.dat
-r <keyword1,keyword2,...,keywordN>, --remove <keyword1,keyword2,...,keywordN>
remove the keywords from the final pref.dat
-D var1=val1 -D var2=val2 ... -D varN=valN
set CMake variables to the given values

Non-default features

--static
link only static libs for a portable executable
-u, --domdec_gpu
include support for DOMDEC_GPU
--repdstr
include support for REPDSTR
-s, --sccdftb
include support for SCC-DFTB
--stringm
include support for STRINGM
--with-x11
include support for X11 graphics via XDISPLAY
-q, --gamess
include support for GAMESS QM/MM interface

Compiler options

--with-gnu, --with-gcc
finds the first gnu compilers on the path to use in the build
--with-intel
finds the first Intel compilers on the path to use in the build
--with-pgi
finds the first PGI compilers on the path to use in the build
--with-cuda-host-compiler <C/C++ compiler>
the C/C++ host compiler that nvcc will used (passed via -ccbin)

Remove default features

--without-domdec
skips compiling DOMDEC features
--without-fftw
skips linking in the FFTW library
--without-mkl
skips linking in Intel's MKL library
--without-mpi
skips compiling MPI features
--without-openmm
skips compiling OpenMM features

Some notable CMake variables (set with the -D <variable>=<value> option)

CUDA_HOST_COMPILER
the C/C++ host compiler that nvcc will used (passed via -ccbin)

optimization
The optimization level 0, 1, 2, or 3 passed to the compiler via '-O<#>'.
This will be added to the CHARMM_<Lang>_FLAGS_RELEASE CMake variable where
Lang is C, CXX, and Fortran.
It should not normally be necessary to set this flag. A default level is set
in the CMakeLists.txt file.

arch_flag_C
arch_flag_CXX
arch_flag_Fortran
A compiler option to add to the CHARMM_<Lang>_FLAGS_RELEASE CMake variable where
Lang is C, CXX, and Fortran.
It should not normally be necessary to set this flag. A default level is set
in the CMakeLists.txt file.
For example, the default switch for GNU GCC is '-march=native'
in the CMakeLists.txt file.
For example, the default switch for GNU GCC is '-march=native'

nvcc_ptx_target
Set a virtual architechture for CUDA code. The CUDA code will be compiled to PTX
for the given virtual architechture. The CUDA driver will compile the PTX to
a real GPU architecture at runtime. Use the compute number times 10, for example:
32, 35, 53, 75...

nvcc_arch_target
Set a real GPU architechture for CUDA code. The CUDA code will be compiled to
the given real architechture. The CUDA driver will not use the JIT compiler at
runtime. Use the compute number times 10, for example: 32, 35, 53, 75...

nvcc_ptx_target
Set a virtual architechture for CUDA code. The CUDA code will be compiled to PTX
for the given virtual architechture. The CUDA driver will compile the PTX to
a real GPU architecture at runtime. Use the compute number times 10, for example:
32, 35, 53, 75...

nvcc_arch_target
Set a real GPU architechture for CUDA code. The CUDA code will be compiled to
the given real architechture. The CUDA driver will not use the JIT compiler at
runtime. Use the compute number times 10, for example: 32, 35, 53, 75...

nvcc_ptx_target
Set a virtual architechture for CUDA code. The CUDA code will be compiled to PTX
for the given virtual architechture. The CUDA driver will compile the PTX to
a real GPU architecture at runtime. Use the compute number times 10, for example:
32, 35, 53, 75...

nvcc_arch_target
Set a real GPU architechture for CUDA code. The CUDA code will be compiled to
the given real architechture. The CUDA driver will not use the JIT compiler at
runtime. Use the compute number times 10, for example: 32, 35, 53, 75...

nvcc_ptx_target
Set a virtual architechture for CUDA code. The CUDA code will be compiled to PTX
for the given virtual architechture. The CUDA driver will compile the PTX to
a real GPU architecture at runtime. Use the compute number times 10, for example:
32, 35, 53, 75...

nvcc_arch_target
Set a real GPU architechture for CUDA code. The CUDA code will be compiled to
the given real architechture. The CUDA driver will not use the JIT compiler at
runtime. Use the compute number times 10, for example: 32, 35, 53, 75...

For example, to configure compilation using the Intel compiler suite and
including debugging symbols:

$ ../charmm/configure --intel --debug --prefix=/home/jdoe/charmm-install

Please note that you must provide the prefix to successfully complete the final
install phase and the prefix must be the full, literal path.
Do not use shortcuts such as ~ or ..

If the configure script detects the file 'install.com' and
the directory 'build' in the current working directory, then
the script will run the cmake command from the 'build/cmake' directory
creating this directory if necessary.
After configuration, in this case, change to the 'build/cmake' directory
to run the make command or pass the '-C build/cmake' option to the make command.
A relative prefix option will be relative to the
'build/cmake' directory instead of the top-level source directory.
If no prefix option is provided, the executable and plugins will be installed
into the CHARMM source root bin and lib directories.


Top
Compilation

After configuration completes with no errors,
invoke make in the build directory.

$ make -j3


Top
Finally, if compilation completes with no error,
invoke 'make install' to install binaries and support files into
the user specified prefix directory.
The binaries will be installed to <prefix>/bin

To run the test cases from the installed <prefix> test directory,
give test.com the first argument cmake intead of gnu or em64t.

$ cd <prefix>/test
$ ./test.com cmake ...