9 Introduction to building |Gromacs|
10 ==================================
12 These instructions pertain to building |Gromacs|
13 |version|. You might also want to check the `up-to-date installation instructions`_.
15 Quick and dirty installation
16 ----------------------------
17 1. Get the latest version of your C and C++ compilers.
18 2. Check that you have CMake version |GMX_CMAKE_MINIMUM_REQUIRED_VERSION| or later.
19 3. Get and unpack the latest version of the |Gromacs| tarball.
20 4. Make a separate build directory and change to it.
21 5. Run ``cmake`` with the path to the source as an argument
22 6. Run ``make``, ``make check``, and ``make install``
24 Or, as a sequence of commands to execute:
28 tar xfz gromacs-|version|.tar.gz
32 cmake .. -DGMX_BUILD_OWN_FFTW=ON -DREGRESSIONTEST_DOWNLOAD=ON
36 source /usr/local/gromacs/bin/GMXRC
38 This will download and build first the prerequisite FFT library
39 followed by |Gromacs|. If you already have FFTW installed, you can
40 remove that argument to ``cmake``. Overall, this build of |Gromacs| will
41 be correct and reasonably fast on the machine upon which ``cmake``
42 ran. If you want to get the maximum value for your hardware with
43 |Gromacs|, you will have to read further. Sadly, the interactions of
44 hardware, libraries, and compilers are only going to continue to get
49 As above, and with further details below, but you should consider
50 using the following `CMake options`_ with the
51 appropriate value instead of ``xxx`` :
53 * ``-DCMAKE_C_COMPILER=xxx`` equal to the name of the C99 `Compiler`_ you wish to use (or the environment variable ``CC``)
54 * ``-DCMAKE_CXX_COMPILER=xxx`` equal to the name of the C++98 `compiler`_ you wish to use (or the environment variable ``CXX``)
55 * ``-DGMX_MPI=on`` to build using `MPI support`_
56 * ``-DGMX_GPU=on`` to build using nvcc to run with an NVIDIA `native GPU acceleration`_
57 * ``-DGMX_SIMD=xxx`` to specify the level of `SIMD support`_ of the node on which mdrun will run
58 * ``-DGMX_BUILD_MDRUN_ONLY=on`` for `building only mdrun`_, e.g. for compute cluster back-end nodes
59 * ``-DGMX_DOUBLE=on`` to run |Gromacs| in double precision (slower, and not normally useful)
60 * ``-DCMAKE_PREFIX_PATH=xxx`` to add a non-standard location for CMake to `search for libraries, headers or programs`_
61 * ``-DCMAKE_INSTALL_PREFIX=xxx`` to install |Gromacs| to a non-standard location (default ``/usr/local/gromacs``)
62 * ``-DBUILD_SHARED_LIBS=off`` to turn off the building of shared libraries to help with `static linking`_
63 * ``-DGMX_FFT_LIBRARY=xxx`` to select whether to use ``fftw``, ``mkl`` or ``fftpack`` libraries for `FFT support`_
64 * ``-DCMAKE_BUILD_TYPE=Debug`` to build |Gromacs| in debug mode
66 Building older versions
67 -----------------------
68 For installation instructions for old |Gromacs| versions, see the
69 documentation for installing
70 `GROMACS 4.5 <http://www.gromacs.org/Documentation/Installation_Instructions_4.5>`_,
71 `GROMACS 4.6 <http://www.gromacs.org/Documentation/Installation_Instructions_4.6>`_,
73 `GROMACS 5.0 <http://www.gromacs.org/Documentation/Installation_Instructions_5.0>`_.
79 |Gromacs| can be compiled for many operating systems and architectures.
80 These include any distribution of Linux, Mac OS X or Windows, and
81 architectures including x86, AMD64/x86-64, PPC, ARM v7 and SPARC VIII.
85 Technically, |Gromacs| can be compiled on any platform with an ANSI C99
86 and C++98 compiler, and their respective standard C/C++ libraries.
87 We use only a few C99 features, but note that the C++ compiler also needs to
88 support these C99 features (notably, int64_t and related things), which are not
89 part of the C++98 standard.
90 Getting good performance on an OS and architecture requires choosing a
91 good compiler. In practice, many compilers struggle to do a good job
92 optimizing the |Gromacs| architecture-optimized SIMD kernels.
94 For best performance, the |Gromacs| team strongly recommends you get the
95 most recent version of your preferred compiler for your platform.
96 There is a large amount of |Gromacs| code that depends on effective
97 compiler optimization to get high performance. This makes |Gromacs|
98 performance sensitive to the compiler used, and the binary will often
99 only work on the hardware for which it is compiled.
101 * In particular, |Gromacs| includes a lot of explicit SIMD (single
102 instruction, multiple data) optimization that can use assembly
103 instructions available on most modern processors. This can have a
104 substantial effect on performance, but for recent processors you
105 also need a similarly recent compiler that includes support for the
106 corresponding SIMD instruction set to get this benefit. The
107 configuration does a good job at detecting this, and you will
108 usually get warnings if |Gromacs| and your hardware support a more
109 recent instruction set than your compiler.
111 * On Intel-based x86 hardware, we recommend you to use the GNU
112 compilers version 4.7 or later or Intel compilers version 12 or
113 later for best performance. The Intel compiler has historically been
114 better at instruction scheduling, but recent gcc versions have
115 proved to be as fast or sometimes faster than Intel.
117 * The Intel and GNU compilers produce much faster |Gromacs| executables
118 than the PGI and Cray compilers.
120 * On AMD-based x86 hardware up through the "K10" microarchitecture
121 ("Family 10h") Thuban/Magny-Cours architecture (e.g. Opteron
122 6100-series processors), it is worth using the Intel compiler for
123 better performance, but gcc version 4.7 and later are also
126 * On the AMD Bulldozer architecture (Opteron 6200), AMD introduced
127 fused multiply-add instructions and an "FMA4" instruction format not
128 available on Intel x86 processors. Thus, on the most recent AMD
129 processors you want to use gcc version 4.7 or later for best
130 performance! The Intel compiler will only generate code for the
131 subset also supported by Intel processors, and that is significantly
134 * If you are running on Mac OS X, the best option is the Intel
135 compiler. Both clang and gcc will work, but they produce lower
136 performance and each have some shortcomings. Current Clang does not
137 support OpenMP. This may change when clang 3.5 becomes available.
139 * For all non-x86 platforms, your best option is typically to use the
140 vendor's default or recommended compiler, and check for specialized
143 Compiling with parallelization options
144 --------------------------------------
145 |Gromacs| can run in parallel on multiple cores of a single
146 workstation using its built-in thread-MPI. No user action is required
147 in order to enable this.
151 If you wish to use the excellent native GPU support in |Gromacs|,
152 NVIDIA's CUDA_ version |REQUIRED_CUDA_VERSION| software development kit is required,
153 and the latest version is strongly encouraged. NVIDIA GPUs with at
154 least NVIDIA compute capability |REQUIRED_CUDA_COMPUTE_CAPABILITY| are
155 required, e.g. Fermi or Kepler cards. You are strongly recommended to
156 get the latest CUDA version and driver supported by your hardware, but
157 beware of possible performance regressions in newer CUDA versions on
158 older hardware. Note that while some CUDA compilers (nvcc) might not
159 officially support recent versions of gcc as the back-end compiler, we
160 still recommend that you at least use a gcc version recent enough to
161 get the best SIMD support for your CPU, since |Gromacs| always runs some
162 code on the CPU. It is most reliable to use the same C++ compiler
163 version for |Gromacs| code as used as the back-end compiler for nvcc,
164 but it could be faster to mix compiler versions to suit particular
171 If you wish to run in parallel on multiple machines across a network,
172 you will need to have
174 * an MPI library installed that supports the MPI 1.3
176 * wrapper compilers that will compile code using that library.
178 The |Gromacs| team recommends OpenMPI_ version
179 1.6 (or higher), MPICH_ version 1.4.1 (or
180 higher), or your hardware vendor's MPI installation. The most recent
181 version of either of these is likely to be the best. More specialized
182 networks might depend on accelerations only available in the vendor's
183 library. LAMMPI_ might work, but since it has
184 been deprecated for years, it is not supported.
186 Often OpenMP_ parallelism is an
187 advantage for |Gromacs|, but support for this is generally built into
188 your compiler and detected automatically.
190 In summary, for maximum performance you will need to examine how you
191 will use |Gromacs|, what hardware you plan to run on, and whether you
192 can afford a non-free compiler for slightly better
193 performance. Unfortunately, the only way to find out is to test
194 different options and parallelization schemes for the actual
195 simulations you want to run. You will still get *good*,
196 performance with the default build and runtime options, but if you
197 truly want to push your hardware to the performance limit, the days of
198 just blindly starting programs with ``mdrun`` are gone.
202 |Gromacs| uses the CMake build system, and requires
203 version |GMX_CMAKE_MINIMUM_REQUIRED_VERSION| or higher. Lower versions
204 will not work. You can check whether CMake is installed, and what
205 version it is, with ``cmake --version``. If you need to install CMake,
206 then first check whether your platform's package management system
207 provides a suitable version, or visit the `CMake installation page`_
209 binaries, source code and installation instructions. The |Gromacs| team
210 recommends you install the most recent version of CMake you can.
214 Fast Fourier Transform library
215 ------------------------------
216 Many simulations in |Gromacs| make extensive use of fast Fourier
217 transforms, and a software library to perform these is always
218 required. We recommend FFTW_ (version 3 or higher only) or
219 Intel MKL_. The choice of
220 library can be set with ``cmake -DGMX_FFT_LIBRARY=<name>``, where
221 ``<name>`` is one of ``fftw``, ``mkl``, or ``fftpack``. FFTPACK is bundled
222 with |Gromacs| as a fallback, and is acceptable if mdrun performance is
227 FFTW_ is likely to be available for your platform via its package
228 management system, but there can be compatibility and significant
229 performance issues associated with these packages. In particular,
230 |Gromacs| simulations are normally run in "mixed" floating-point
231 precision, which is suited for the use of single precision in
232 FFTW. The default FFTW package is normally in double
233 precision, and good compiler options to use for FFTW when linked to
234 |Gromacs| may not have been used. Accordingly, the |Gromacs| team
237 * that you permit the |Gromacs| installation to download and
238 build FFTW from source automatically for you (use
239 ``cmake -DGMX_BUILD_OWN_FFTW=ON``), or
240 * that you build FFTW from the source code.
242 If you build FFTW from source yourself, get the most recent version
243 and follow the `FFTW installation guide`_. Note that we have recently
244 contributed new SIMD optimization for several extra platforms to
245 FFTW, which will appear in FFTW-3.3.5 (for now it is available in the
246 FFTW repository on github, or you can find a very unofficial prerelease
247 version at ftp://ftp.gromacs.org/pub/prerequisite_software ).
248 Choose the precision for FFTW (i.e. single/float vs. double) to
249 match whether you will later use mixed or double precision for
250 |Gromacs|. There is no need to compile FFTW with
251 threading or MPI support, but it does no harm. On x86 hardware,
252 compile with *both* ``--enable-sse2`` and ``--enable-avx`` for
253 FFTW-3.3.4 and earlier. As of FFTW-3.3.5 you should also add
254 ``--enable-avx2``. FFTW will create a fat library with codelets
255 for all different instruction sets, and pick the fastest supported
256 one at runtime. On IBM Power8, you definitely want the upcoming
257 FFTW-3.3.5 and use ``--enable-vsx`` for SIMD support. If you are
258 using a Cray, there is a special modified (commercial) version of
259 FFTs using the FFTW interface which might be faster, but we have
260 not yet tested this extensively.
264 Using MKL_ with the Intel Compilers version 11 or higher is very
265 simple. Set up your compiler environment correctly, perhaps with a
266 command like ``source /path/to/compilervars.sh intel64`` (or consult
267 your local documentation). Then set ``-DGMX_FFT_LIBRARY=mkl`` when you
268 run cmake. In this case, |Gromacs| will also use MKL for BLAS and LAPACK
269 (see `linear algebra libraries`_). Generally,
270 there is no advantage in using MKL with |Gromacs|, and FFTW is often
273 Otherwise, you can get your hands dirty and configure MKL by setting
277 -DGMX_FFT_LIBRARY=mkl
278 -DMKL_LIBRARIES="/full/path/to/libone.so;/full/path/to/libtwo.so"
279 -DMKL_INCLUDE_DIR="/full/path/to/mkl/include"
281 where the full list (and order!) of libraries you require are found in
282 Intel's MKL documentation for your system.
284 Optional build components
285 -------------------------
286 * Compiling to run on NVIDIA GPUs requires CUDA_
287 * An external Boost library can be used to provide better
288 implementation support for smart pointers and exception handling,
289 but the |Gromacs| source bundles a subset of Boost 1.55.0 as a fallback
290 * Hardware-optimized BLAS and LAPACK libraries are useful
291 for a few of the |Gromacs| utilities focused on normal modes and
292 matrix manipulation, but they do not provide any benefits for normal
293 simulations. Configuring these are discussed at
294 `linear algebra libraries`_.
295 * The built-in |Gromacs| trajectory viewer ``gmx view`` requires X11 and
296 Motif/Lesstif libraries and header files. You may prefer to use
297 third-party software for visualization, such as VMD_ or PyMol_.
298 * An external TNG library for trajectory-file handling can be used,
299 but TNG 1.7.3 is bundled in the |Gromacs| source already
300 * zlib is used by TNG for compressing some kinds of trajectory data
301 * Running the |Gromacs| test suite requires libxml2
302 * Building the |Gromacs| documentation requires ImageMagick, pdflatex,
303 bibtex, doxygen, python, sphinx and pygments.
304 * The |Gromacs| utility programs often write data files in formats
305 suitable for the Grace plotting tool, but it is straightforward to
306 use these files in other plotting programs, too.
308 Doing a build of |Gromacs|
309 ==========================
310 This section will cover a general build of |Gromacs| with CMake_, but it
311 is not an exhaustive discussion of how to use CMake. There are many
312 resources available on the web, which we suggest you search for when
313 you encounter problems not covered here. The material below applies
314 specifically to builds on Unix-like systems, including Linux, and Mac
315 OS X. For other platforms, see the specialist instructions below.
317 Configuring with CMake
318 ----------------------
319 CMake will run many tests on your system and do its best to work out
320 how to build |Gromacs| for you. If your build machine is the same as
321 your target machine, then you can be sure that the defaults will be
322 pretty good. The build configuration will for instance attempt to
323 detect the specific hardware instructions available in your
324 processor. However, if you want to control aspects of the build, or
325 you are compiling on a cluster head node for back-end nodes with a
326 different architecture, there are plenty of things you can set
329 The best way to use CMake to configure |Gromacs| is to do an
330 "out-of-source" build, by making another directory from which you will
331 run CMake. This can be outside the source directory, or a subdirectory
332 of it. It also means you can never corrupt your source code by trying
333 to build it! So, the only required argument on the CMake command line
334 is the name of the directory containing the ``CMakeLists.txt`` file of
335 the code you want to build. For example, download the source tarball
340 tar xfz gromacs-|version|.tgz
346 You will see ``cmake`` report a sequence of results of tests and
347 detections done by the |Gromacs| build system. These are written to the
348 ``cmake`` cache, kept in ``CMakeCache.txt``. You can edit this file by
349 hand, but this is not recommended because you could make a mistake.
350 You should not attempt to move or copy this file to do another build,
351 because file paths are hard-coded within it. If you mess things up,
352 just delete this file and start again with ``cmake``.
354 If there is a serious problem detected at this stage, then you will see
355 a fatal error and some suggestions for how to overcome it. If you are
356 not sure how to deal with that, please start by searching on the web
357 (most computer problems already have known solutions!) and then
358 consult the gmx-users mailing list. There are also informational
359 warnings that you might like to take on board or not. Piping the
360 output of ``cmake`` through ``less`` or ``tee`` can be
363 Once ``cmake`` returns, you can see all the settings that were chosen
364 and information about them by using e.g. the curses interface
370 You can actually use ``ccmake`` (available on most Unix platforms,
371 if the curses library is supported) directly in the first step, but then
372 most of the status messages will merely blink in the lower part
373 of the terminal rather than be written to standard out. Most platforms
374 including Linux, Windows, and Mac OS X even have native graphical user interfaces for
375 ``cmake``, and it can create project files for almost any build environment
376 you want (including Visual Studio or Xcode).
377 Check out `running CMake`_ for
378 general advice on what you are seeing and how to navigate and change
379 things. The settings you might normally want to change are already
380 presented. You may make changes, then re-configure (using ``c``), so that it
381 gets a chance to make changes that depend on yours and perform more
382 checking. It may take several configuration passes to reach the desired
383 configuration, in particular if you need to resolve errors.
385 A key thing to consider here is the setting of
386 ``CMAKE_INSTALL_PREFIX``. You will need to be able to write to this
387 directory in order to install |Gromacs| later.
388 So if you do not have super-user privileges on your
389 machine, then you will need to choose a sensible location within your
390 home directory for your |Gromacs| installation. Even if you do have
391 super-user privileges, you should use them only for the installation
392 phase, and never for configuring, building, or running |Gromacs|!
394 When you have reached the desired configuration with ``ccmake``, the
395 build system can be generated by pressing ``g``. This requires that the previous
396 configuration pass did not reveal any additional settings (if it did, you need
397 to configure once more with ``c``). With ``cmake``, the build system is generated
398 after each pass that does not produce errors.
400 You cannot attempt to change compilers after the initial run of
401 ``cmake``. If you need to change, clean up, and start again.
405 Using CMake command-line options
406 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
407 Once you become comfortable with setting and changing options, you may
408 know in advance how you will configure |Gromacs|. If so, you can speed
409 things up by invoking ``cmake`` and passing the various options at once
410 on the command line. This can be done by setting cache variable at the
411 cmake invocation using the ``-DOPTION=VALUE``; note that some
412 environment variables are also taken into account, in particular
413 variables like ``CC``, ``CXX``, ``FCC`` (which may be familiar to autoconf users).
415 For example, the following command line
419 cmake .. -DGMX_GPU=ON -DGMX_MPI=ON -DCMAKE_INSTALL_PREFIX=/home/marydoe/programs
421 can be used to build with GPUs, MPI and install in a custom
422 location. You can even save that in a shell script to make it even
423 easier next time. You can also do this kind of thing with ``ccmake``,
424 but you should avoid this, because the options set with ``-D`` will not
425 be able to be changed interactively in that run of ``ccmake``.
429 |Gromacs| has extensive support for detecting and using the SIMD
430 capabilities of many modern HPC CPU architectures. If you are building
431 |Gromacs| on the same hardware you will run it on, then you don't need
432 to read more about this, unless you are getting configuration warnings
433 you do not understand. By default, the |Gromacs| build system will
434 detect the SIMD instruction set supported by the CPU architecture (on
435 which the configuring is done), and thus pick the best
436 available SIMD parallelization supported by |Gromacs|. The build system
437 will also check that the compiler and linker used also support the
438 selected SIMD instruction set and issue a fatal error if they
441 Valid values are listed below, and the applicable value with the
442 highest number in the list is generally the one you should choose:
444 1. ``None`` For use only on an architecture either lacking SIMD,
445 or to which |Gromacs| has not yet been ported and none of the
446 options below are applicable.
447 2. ``SSE2`` This SIMD instruction set was introduced in Intel
448 processors in 2001, and AMD in 2003. Essentially all x86
449 machines in existence have this, so it might be a good choice if
450 you need to support dinosaur x86 computers too.
451 3. ``SSE4.1`` Present in all Intel core processors since 2007,
452 but notably not in AMD magny-cours. Still, almost all recent
453 processors support this, so this can also be considered a good
454 baseline if you are content with portability between reasonably
456 4. ``AVX_128_FMA`` AMD bulldozer processors (2011) have this.
457 Unfortunately Intel and AMD have diverged the last few years;
458 If you want good performance on modern AMD processors
459 you have to use this since it also allows the reset of the
460 code to use AMD 4-way fused multiply-add instructions. The drawback
461 is that your code will not run on Intel processors at all.
462 5. ``AVX_256`` This instruction set is present on Intel processors
463 since Sandy Bridge (2011), where it is the best choice unless
464 you have an even more recent CPU that supports AVX2. While this
465 code will work on recent AMD processors, it is significantly
466 less efficient than the ``AVX_128_FMA`` choice above - do not be
467 fooled to assume that 256 is better than 128 in this case.
468 6. ``AVX2_256`` Present on Intel Haswell processors released in 2013,
469 and it will also enable Intel 3-way fused multiply-add instructions.
470 This code will not work on AMD CPUs.
471 7. ``IBM_QPX`` BlueGene/Q A2 cores have this.
472 8. ``Sparc64_HPC_ACE`` Fujitsu machines like the K computer have this.
474 The CMake configure system will check that the compiler you have
475 chosen can target the architecture you have chosen. mdrun will check
476 further at runtime, so if in doubt, choose the lowest setting you
477 think might work, and see what mdrun says. The configure system also
478 works around many known issues in many versions of common HPC
479 compilers. However, since the options also enable general compiler
480 flags for the platform in question, you can end up in situations
481 where e.g. an ``AVX_128_FMA`` binary will just crash on any
482 Intel machine, since the code will try to execute general illegal
483 instructions (inserted by the compiler) before mdrun gets to the
484 architecture detection routines.
486 A further ``GMX_SIMD=Reference`` option exists, which is a special
487 SIMD-like implementation written in plain C that developers can use
488 when developing support in |Gromacs| for new SIMD architectures. It is
489 not designed for use in production simulations, but if you are using
490 an architecture with SIMD support to which |Gromacs| has not yet been
491 ported, you may wish to try this option instead of the default
492 ``GMX_SIMD=None``, as it can often out-perform this when the
493 auto-vectorization in your compiler does a good job. And post on the
494 |Gromacs| mailing lists, because |Gromacs| can probably be ported for new
495 SIMD architectures in a few days.
497 CMake advanced options
498 ^^^^^^^^^^^^^^^^^^^^^^
499 The options that are displayed in the default view of ``ccmake`` are
500 ones that we think a reasonable number of users might want to consider
501 changing. There are a lot more options available, which you can see by
502 toggling the advanced mode in ``ccmake`` on and off with ``t``. Even
503 there, most of the variables that you might want to change have a
504 ``CMAKE_`` or ``GMX_`` prefix. There are also some options that will be
505 visible or not according to whether their preconditions are satisfied.
507 .. _search for libraries, headers or programs:
509 Helping CMake find the right libraries, headers, or programs
510 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
511 If libraries are installed in non-default locations their location can
512 be specified using the following environment variables:
514 * ``CMAKE_INCLUDE_PATH`` for header files
515 * ``CMAKE_LIBRARY_PATH`` for libraries
516 * ``CMAKE_PREFIX_PATH`` for header, libraries and binaries
517 (e.g. ``/usr/local``).
519 The respective ``include``, ``lib``, or ``bin`` is
520 appended to the path. For each of these variables, a list of paths can
521 be specified (on Unix, separated with ":"). Note that these are
522 enviroment variables (and not ``cmake`` command-line arguments) and in
523 a ``bash`` shell are used like:
527 CMAKE_PREFIX_PATH=/opt/fftw:/opt/cuda cmake ..
529 Alternatively, these variables are also ``cmake`` options, so they can
530 be set like ``-DCMAKE_PREFIX_PATH=/opt/fftw:/opt/cuda``.
532 The ``CC`` and ``CXX`` environment variables are also useful
533 for indicating to ``cmake`` which compilers to use, which can be very
534 important for maximising |Gromacs| performance. Similarly,
535 ``CFLAGS``/``CXXFLAGS`` can be used to pass compiler
536 options, but note that these will be appended to those set by
537 |Gromacs| for your build platform and build type. You can customize
538 some of this with advanced options such as ``CMAKE_C_FLAGS``
541 See also the page on `CMake environment variables`_.
543 Native GPU acceleration
544 ^^^^^^^^^^^^^^^^^^^^^^^
545 If you have the CUDA_ Toolkit installed, you can use ``cmake`` with:
549 cmake .. -DGMX_GPU=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda
551 (or whichever path has your installation). In some cases, you might
552 need to specify manually which of your C++ compilers should be used,
553 e.g. with the advanced option ``CUDA_HOST_COMPILER``.
555 By default, optimized code will be generated for CUDA architectures
556 supported by the nvcc compiler (and the |Gromacs| build system).
557 However, it can be beneficial to manually pick the specific CUDA architecture(s)
558 to generate code for either to reduce compilation time (and binary size) or to
559 target a new architecture not yet supported by the |GROMACS| build system.
560 Setting the desired CUDA architecture(s) and virtual architecture(s)
561 can be done using the ``GMX_CUDA_TARGET_SM`` and ``GMX_CUDA_TARGET_COMPUTE``
562 variables, respectively. These take a semicolon delimited string with
563 the two digit suffixes of CUDA (virtual) architectures names
564 (for details see the "Options for steering GPU code generation" section of the
565 nvcc man / help or Chapter 6. of the nvcc manual).
567 The GPU acceleration has been tested on AMD64/x86-64 platforms with
568 Linux, Mac OS X and Windows operating systems, but Linux is the
569 best-tested and supported of these. Linux running on ARM v7 (32 bit)
574 Dynamic linking of the |Gromacs| executables will lead to a
575 smaller disk footprint when installed, and so is the default on
576 platforms where we believe it has been tested repeatedly and found to work.
577 In general, this includes Linux, Windows, Mac OS X and BSD systems.
578 Static binaries take much more space, but on some hardware and/or under
579 some conditions they are necessary, most commonly when you are running a parallel
580 simulation using MPI libraries (e.g. BlueGene, Cray).
582 * To link |Gromacs| binaries statically against the internal |Gromacs|
583 libraries, set ``-DBUILD_SHARED_LIBS=OFF``.
584 * To link statically against external (non-system) libraries as well,
585 the ``-DGMX_PREFER_STATIC_LIBS=ON`` option can be used. Note, that in
586 general ``cmake`` picks up whatever is available, so this option only
587 instructs ``cmake`` to prefer static libraries when both static and
588 shared are available. If no static version of an external library is
589 available, even when the aforementioned option is ``ON``, the shared
590 library will be used. Also note, that the resulting binaries will
591 still be dynamically linked against system libraries on platforms
592 where that is the default. To use static system libraries,
593 additional compiler/linker flags are necessary, e.g. ``-static-libgcc
595 * To attempt to link a fully static binary set
596 ``-DGMX_BUILD_SHARED_EXE=OFF``. This will prevent CMake from explicitly
597 setting any dynamic linking flags. This option also sets
598 ``-DBUILD_SHARED_LIBS=OFF`` and ``-DGMX_PREFER_STATIC_LIBS=ON`` by
599 default, but the above caveats apply. For compilers which don't
600 default to static linking, the required flags have to be specified. On
601 Linux, this is usually ``CFLAGS=-static CXXFLAGS=-static``.
605 Here, we consider portability aspects related to CPU instruction sets,
606 for details on other topics like binaries with statical vs dynamic
607 linking please consult the relevant parts of this documentation or
608 other non-|Gromacs| specific resources.
610 A |Gromacs| build will normally not be portable, not even across
611 hardware with the same base instruction set like x86. Non-portable
612 hardware-specific optimizations are selected at configure-time, such
613 as the SIMD instruction set used in the compute-kernels. This
614 selection will be done by the build system based on the capabilities
615 of the build host machine or based on cross-compilation information
616 provided to ``cmake`` at configuration.
618 Often it is possible to ensure portability by choosing the least
619 common denominator of SIMD support, e.g. SSE2 for x86, and ensuring
620 the you use ``cmake -DGMX_USE_RDTSCP=off`` if any of the target CPU
621 architectures does not support the ``RDTSCP`` instruction. However, we
622 discourage attempts to use a single |Gromacs| installation when the
623 execution environment is heterogeneous, such as a mix of AVX and
624 earlier hardware, because this will lead to programs (especially
625 mdrun) that run slowly on the new hardware. Building two full
626 installations and locally managing how to call the correct one
627 (e.g. using the module system) is the recommended
628 approach. Alternatively, as at the moment the |Gromacs| tools do not
629 make strong use of SIMD acceleration, it can be convenient to create
630 an installation with tools portable across different x86 machines, but
631 with separate mdrun binaries for each architecture. To achieve this,
632 one can first build a full installation with the
633 least-common-denominator SIMD instruction set, e.g. ``-DGMX_SIMD=SSE2``,
634 then build separate mdrun binaries for each architecture present in
635 the heterogeneous environment. By using custom binary and library
636 suffixes for the mdrun-only builds, these can be installed to the
637 same location as the "generic" tools installation.
638 `Building just the mdrun binary`_ is possible by setting the
639 ``-DGMX_BUILD_MDRUN_ONLY=ON`` option.
641 Linear algebra libraries
642 ^^^^^^^^^^^^^^^^^^^^^^^^
643 As mentioned above, sometimes vendor BLAS and LAPACK libraries
644 can provide performance enhancements for |Gromacs| when doing
645 normal-mode analysis or covariance analysis. For simplicity, the text
646 below will refer only to BLAS, but the same options are available
647 for LAPACK. By default, CMake will search for BLAS, use it if it
648 is found, and otherwise fall back on a version of BLAS internal to
649 |Gromacs|. The ``cmake`` option ``-DGMX_EXTERNAL_BLAS=on`` will be set
650 accordingly. The internal versions are fine for normal use. If you
651 need to specify a non-standard path to search, use
652 ``-DCMAKE_PREFIX_PATH=/path/to/search``. If you need to specify a
653 library with a non-standard name (e.g. ESSL on AIX or BlueGene), then
654 set ``-DGMX_BLAS_USER=/path/to/reach/lib/libwhatever.a``.
656 If you are using Intel MKL_ for FFT, then the BLAS and
657 LAPACK it provides are used automatically. This could be
658 over-ridden with ``GMX_BLAS_USER``, etc.
660 On Apple platforms where the Accelerate Framework is available, these
661 will be automatically used for BLAS and LAPACK. This could be
662 over-ridden with ``GMX_BLAS_USER``, etc.
664 Changing the names of |Gromacs| binaries and libraries
665 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
666 It is sometimes convenient to have different versions of the same
667 |Gromacs| programs installed. The most common use cases have been single
668 and double precision, and with and without MPI. This mechanism can
669 also be used to install side-by-side multiple versions of mdrun
670 optimized for different CPU architectures, as mentioned previously.
672 By default, |Gromacs| will suffix programs and libraries for such builds
673 with ``_d`` for double precision and/or ``_mpi`` for MPI (and nothing
674 otherwise). This can be controlled manually with ``GMX_DEFAULT_SUFFIX
675 (ON/OFF)``, ``GMX_BINARY_SUFFIX`` (takes a string) and ``GMX_LIBS_SUFFIX``
676 (also takes a string). For instance, to set a custom suffix for
677 programs and libraries, one might specify:
681 cmake .. -DGMX_DEFAULT_SUFFIX=OFF -DGMX_BINARY_SUFFIX=_mod -DGMX_LIBS_SUFFIX=_mod
683 Thus the names of all programs and libraries will be appended with
686 Changing installation tree structure
687 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
688 By default, a few different directories under ``CMAKE_INSTALL_PREFIX`` are used
689 when when |Gromacs| is installed. Some of these can be changed, which is mainly
690 useful for packaging |Gromacs| for various distributions. The directories are
691 listed below, with additional notes about some of them. Unless otherwise noted,
692 the directories can be renamed by editing the installation paths in the main
696 The standard location for executables and some scripts.
697 Some of the scripts hardcode the absolute installation prefix, which needs
698 to be changed if the scripts are relocated.
700 The standard location for installed headers.
702 The standard location for libraries. The default depends on the system, and
703 is determined by CMake.
704 The name of the directory can be changed using ``GMX_LIB_INSTALL_DIR`` CMake
707 Information about the installed ``libgromacs`` library for ``pkg-config`` is
708 installed here. The ``lib/`` part adapts to the installation location of the
709 libraries. The installed files contain the installation prefix as absolute
712 CMake package configuration files are installed here.
714 Various data files and some documentation go here.
715 The ``gromacs`` part can be changed using ``GMX_DATA_INSTALL_DIR``. Using this
716 CMake variable is the preferred way of changing the installation path for
717 ``share/gromacs/top/``, since the path to this directory is built into
718 ``libgromacs`` as well as some scripts, both as a relative and as an absolute
719 path (the latter as a fallback if everything else fails).
721 Installed man pages go here.
723 Compiling and linking
724 ---------------------
725 Once you have configured with ``cmake``, you can build |Gromacs| with ``make``.
726 It is expected that this will always complete successfully, and
727 give few or no warnings. The CMake-time tests |Gromacs| makes on the settings
728 you choose are pretty extensive, but there are probably a few cases we
729 have not thought of yet. Search the web first for solutions to
730 problems, but if you need help, ask on gmx-users, being sure to
731 provide as much information as possible about what you did, the system
732 you are building on, and what went wrong. This may mean scrolling back
733 a long way through the output of ``make`` to find the first error
736 If you have a multi-core or multi-CPU machine with ``N``
737 processors, then using
743 will generally speed things up by quite a bit. Other build generator systems
744 supported by ``cmake`` (e.g. ``ninja``) also work well.
746 .. _building just the mdrun binary:
750 Past versions of the build system offered "mdrun" and "install-mdrun"
751 targets (similarly for other programs too) to build and install only
752 the mdrun program, respectively. Such a build is useful when the
753 configuration is only relevant for mdrun (such as with
754 parallelization options for MPI, SIMD, GPUs, or on BlueGene or Cray),
755 or the length of time for the compile-link-install cycle is relevant
758 This is now supported with the ``cmake`` option
759 ``-DGMX_BUILD_MDRUN_ONLY=ON``, which will build a cut-down version of
760 ``libgromacs`` and/or the mdrun program (according to whether shared
761 or static). Naturally, now ``make install`` installs only those
762 products. By default, mdrun-only builds will default to static linking
763 against |Gromacs| libraries, because this is generally a good idea for
764 the targets for which an mdrun-only build is desirable. If you re-use
765 a build tree and change to the mdrun-only build, then you will inherit
766 the setting for ``BUILD_SHARED_LIBS`` from the old build, and will be
767 warned that you may wish to manage ``BUILD_SHARED_LIBS`` yourself.
771 Finally, ``make install`` will install |Gromacs| in the
772 directory given in ``CMAKE_INSTALL_PREFIX``. If this is a system
773 directory, then you will need permission to write there, and you
774 should use super-user privileges only for ``make install`` and
775 not the whole procedure.
777 .. _getting access to GROMACS:
779 Getting access to |Gromacs| after installation
780 ----------------------------------------------
781 |Gromacs| installs the script ``GMXRC`` in the ``bin``
782 subdirectory of the installation directory
783 (e.g. ``/usr/local/gromacs/bin/GMXRC``), which you should source
788 source /your/installation/prefix/here/bin/GMXRC
790 It will detect what kind of shell you are running and set up your
791 environment for using |Gromacs|. You may wish to arrange for your
792 login scripts to do this automatically; please search the web for
793 instructions on how to do this for your shell.
795 Many of the |Gromacs| programs rely on data installed in the
796 ``share/gromacs`` subdirectory of the installation directory. By
797 default, the programs will use the environment variables set in the
798 ``GMXRC`` script, and if this is not available they will try to guess the
799 path based on their own location. This usually works well unless you
800 change the names of directories inside the install tree. If you still
801 need to do that, you might want to recompile with the new install
802 location properly set, or edit the ``GMXRC`` script.
804 Testing |Gromacs| for correctness
805 ---------------------------------
806 Since 2011, the |Gromacs| development uses an automated system where
807 every new code change is subject to regression testing on a number of
808 platforms and software combinations. While this improves
809 reliability quite a lot, not everything is tested, and since we
810 increasingly rely on cutting edge compiler features there is
811 non-negligible risk that the default compiler on your system could
812 have bugs. We have tried our best to test and refuse to use known bad
813 versions in ``cmake``, but we strongly recommend that you run through
814 the tests yourself. It only takes a few minutes, after which you can
817 The simplest way to run the checks is to build |Gromacs| with
818 ``-DREGRESSIONTEST_DOWNLOAD``, and run ``make check``.
819 |Gromacs| will automatically download and run the tests for you.
820 Alternatively, you can download and unpack the GROMACS
821 regression test suite |gmx-regressiontests-package| tarball yourself
822 and use the advanced ``cmake`` option ``REGRESSIONTEST_PATH`` to
823 specify the path to the unpacked tarball, which will then be used for
824 testing. If the above does not work, then please read on.
826 The regression tests are also available from the download_ section.
827 Once you have downloaded them, unpack the tarball, source
828 ``GMXRC`` as described above, and run ``./gmxtest.pl all``
829 inside the regression tests folder. You can find more options
830 (e.g. adding ``double`` when using double precision, or
831 ``-only expanded`` to run just the tests whose names match
832 "expanded") if you just execute the script without options.
834 Hopefully, you will get a report that all tests have passed. If there
835 are individual failed tests it could be a sign of a compiler bug, or
836 that a tolerance is just a tiny bit too tight. Check the output files
837 the script directs you too, and try a different or newer compiler if
838 the errors appear to be real. If you cannot get it to pass the
839 regression tests, you might try dropping a line to the gmx-users
840 mailing list, but then you should include a detailed description of
841 your hardware, and the output of ``mdrun -version`` (which contains
842 valuable diagnostic information in the header).
844 A build with ``-DGMX_BUILD_MDRUN_ONLY`` cannot be tested with
845 ``make check`` from the build tree, because most of the tests
846 require a full build to run things like ``grompp``. To test such an
847 mdrun fully requires installing it to the same location as a normal
848 build of |Gromacs|, downloading the regression tests tarball manually
849 as described above, sourcing the correct ``GMXRC`` and running the
850 perl script manually. For example, from your |Gromacs| source
857 cmake .. -DCMAKE_INSTALL_PREFIX=/your/installation/prefix/here
861 mkdir build-mdrun-only
863 cmake .. -DGMX_MPI=ON -DGMX_GPU=ON -DGMX_BUILD_MDRUN_ONLY=ON -DCMAKE_INSTALL_PREFIX=/your/installation/prefix/here
866 cd /to/your/unpacked/regressiontests
867 source /your/installation/prefix/here/bin/GMXRC
868 ./gmxtest.pl all -np 2
870 If your mdrun program has been suffixed in a non-standard way, then
871 the ``./gmxtest.pl -mdrun`` option will let you specify that name to the
872 test machinery. You can use ``./gmxtest.pl -double`` to test the
873 double-precision version. You can use ``./gmxtest.pl -crosscompiling``
874 to stop the test harness attempting to check that the programs can
875 be run. You can use ``./gmxtest.pl -mpirun srun`` if your command to
876 run an MPI program is called ``srun``.
878 The ``make check`` target also runs integration-style tests that may run
879 with MPI if ``GMX_MPI=ON`` was set. To make these work, you may need to
880 set the CMake variables ``MPIEXEC``, ``MPIEXEC_NUMPROC_FLAG``, ``NUMPROC``,
881 ``MPIEXEC_PREFLAGS`` and ``MPIEXEC_POSTFLAGS`` so that
882 ``mdrun-mpi-test_mpi`` would run on multiple ranks via the shell command
884 $ ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${NUMPROC} ${MPIEXEC_PREFLAGS} \
885 mdrun-mpi-test_mpi ${MPIEXEC_POSTFLAGS} -otherflags
887 Typically, one might use variable values ``mpirun``, ``-np``, ``2``, ``''``,
888 ``''`` respectively, in order to run on two ranks.
891 Testing |Gromacs| for performance
892 ---------------------------------
893 We are still working on a set of benchmark systems for testing
894 the performance of |Gromacs|. Until that is ready, we recommend that
895 you try a few different parallelization options, and experiment with
896 tools such as ``gmx tune_pme``.
900 You are not alone - this can be a complex task! If you encounter a
901 problem with installing |Gromacs|, then there are a number of
902 locations where you can find assistance. It is recommended that you
903 follow these steps to find the solution:
905 1. Read the installation instructions again, taking note that you
906 have followed each and every step correctly.
908 2. Search the |Gromacs| webpage_ and users emailing list for information
910 ``site:https://mailman-1.sys.kth.se/pipermail/gromacs.org_gmx-users``
911 to a Google search may help filter better results.
913 3. Search the internet using a search engine such as Google.
915 4. Post to the |Gromacs| users emailing list gmx-users for
916 assistance. Be sure to give a full description of what you have
917 done and why you think it did not work. Give details about the
918 system on which you are installing. Copy and paste your command
919 line and as much of the output as you think might be relevant -
920 certainly from the first indication of a problem. In particular,
921 please try to include at least the header from the mdrun logfile,
922 and preferably the entire file. People who might volunteer to help
923 you do not have time to ask you interactive detailed follow-up
924 questions, so you will get an answer faster if you provide as much
925 information as you think could possibly help. High quality bug
926 reports tend to receive rapid high quality answers.
928 Special instructions for some platforms
929 =======================================
933 Building on Windows using native compilers is rather similar to
934 building on Unix, so please start by reading the above. Then, download
935 and unpack the |Gromacs| source archive. Make a folder in which to do
936 the out-of-source build of |Gromacs|. For example, make it within the
937 folder unpacked from the source archive, and call it ``build-gromacs``.
939 For CMake, you can either use the graphical user interface provided on
940 Windows, or you can use a command line shell with instructions similar
941 to the UNIX ones above. If you open a shell from within your IDE
942 (e.g. Microsoft Visual Studio), it will configure the environment for
943 you, but you might need to tweak this in order to get either a 32-bit
944 or 64-bit build environment. The latter provides the fastest
945 executable. If you use a normal Windows command shell, then you will
946 need to either set up the environment to find your compilers and
947 libraries yourself, or run the ``vcvarsall.bat`` batch script provided
948 by MSVC (just like sourcing a bash script under Unix).
950 With the graphical user interface, you will be asked about what
951 compilers to use at the initial configuration stage, and if you use
952 the command line they can be set in a similar way as under UNIX. You
953 will probably make your life easier and faster by using the new
954 facility to download and install FFTW automatically.
956 For the build, you can either load the generated solutions file into
957 e.g. Visual Studio, or use the command line with ``cmake --build`` so
958 the right tools get used.
962 |Gromacs| builds mostly out of the box on modern Cray machines, but
964 * you may need to specify the use of static binaries
965 with ``-DGMX_BUILD_SHARED_EXE=off``,
966 * you may need to set the F77 environmental variable to ``ftn`` when
974 There is currently native acceleration on this platform for the Verlet
975 cut-off scheme. There are no plans to provide accelerated kernels for
976 the group cut-off scheme, but the default plain C kernels will work
979 Only static linking with XL compilers is supported by |Gromacs|. Dynamic
980 linking would be supported by the architecture and |Gromacs|, but has no
981 advantages other than disk space, and is generally discouraged on
982 BlueGene for performance reasons.
984 Computation on BlueGene floating-point units is always done in
985 double-precision. However, mixed-precision builds of |Gromacs| are still
986 normal and encouraged since they use cache more efficiently. The
987 BlueGene hardware automatically converts values stored in single
988 precision in memory to double precision in registers for computation,
989 converts the results back to single precision correctly, and does so
990 for no additional cost. As with other platforms, doing the whole
991 computation in double precision normally shows no improvement in
992 accuracy and costs twice as much time moving memory around.
994 You need to arrange for FFTW to be installed correctly, following the
997 MPI wrapper compilers should be used for compiling and linking. Both
998 xlc and bgclang are supported back ends - either might prove to be
999 faster in practice. The MPI wrapper compilers can make it awkward to
1000 attempt to use IBM's optimized BLAS/LAPACK called ESSL (see the
1001 section on `linear algebra libraries`_. Since mdrun is the only part
1002 of |Gromacs| that should normally run on the compute nodes, and there is
1003 nearly no need for linear algebra support for mdrun, it is recommended
1004 to use the |Gromacs| built-in linear algebra routines - this is never
1005 a problem for normal simulations.
1007 The recommended configuration is to use
1011 cmake .. -DCMAKE_C_COMPILER=mpicc \
1012 -DCMAKE_CXX_COMPILER=mpicxx \
1013 -DCMAKE_TOOLCHAIN_FILE=Platform/BlueGeneQ-static-XL-CXX.cmake \
1014 -DCMAKE_PREFIX_PATH=/your/fftw/installation/prefix \
1016 -DGMX_BUILD_MDRUN_ONLY=ON
1020 which will build a statically-linked MPI-enabled mdrun for the compute
1021 nodes. Or use the Platform/BlueGeneQ-static-bgclang-cxx
1022 toolchain file if compiling with bgclang. Otherwise, |Gromacs| default configuration
1025 It is possible to configure and make the remaining |Gromacs| tools with
1026 the compute-node toolchain, but as none of those tools are MPI-aware
1027 and could then only run on the compute nodes, this would not normally
1028 be useful. Instead, these should be planned to run on the login node,
1029 and a separate |Gromacs| installation performed for that using the login
1030 node's toolchain - not the above platform file, or any other
1031 compute-node toolchain.
1033 Note that only the MPI build is available for the compute-node
1034 toolchains. The |Gromacs| thread-MPI or no-MPI builds are not useful at
1039 There is currently no SIMD support on this platform and no plans to
1040 add it. The default plain C kernels will work.
1044 This is the architecture of the K computer, which uses Fujitsu
1045 Sparc64VIIIfx chips. On this platform, |Gromacs| has
1046 accelerated group kernels using the HPC-ACE instructions, no
1047 accelerated Verlet kernels, and a custom build toolchain. Since this
1048 particular chip only does double precision SIMD, the default setup
1049 is to build |Gromacs| in double. Since most users only need single, we have added
1050 an option GMX_RELAXED_DOUBLE_PRECISION to accept single precision square root
1051 accuracy in the group kernels; unless you know that you really need 15 digits
1052 of accuracy in each individual force, we strongly recommend you use this. Note
1053 that all summation and other operations are still done in double.
1055 The recommended configuration is to use
1059 cmake .. -DCMAKE_TOOLCHAIN_FILE=Toolchain-Fujitsu-Sparc64-mpi.cmake \
1060 -DCMAKE_PREFIX_PATH=/your/fftw/installation/prefix \
1061 -DCMAKE_INSTALL_PREFIX=/where/gromacs/should/be/installed \
1063 -DGMX_BUILD_MDRUN_ONLY=ON \
1064 -DGMX_RELAXED_DOUBLE_PRECISION=ON
1070 |Gromacs| has preliminary support for Intel Xeon Phi. Only symmetric
1071 (aka native) mode is supported. |Gromacs| is functional on Xeon Phi, but
1072 it has so far not been optimized to the same level as other
1073 architectures have. The performance depends among other factors on the
1074 system size, and for
1075 now the performance might not be faster than CPUs. Building for Xeon
1076 Phi works almost as any other Unix. See the instructions above for
1077 details. The recommended configuration is
1081 cmake .. -DCMAKE_TOOLCHAIN_FILE=Platform/XeonPhi
1087 While it is our best belief that |Gromacs| will build and run pretty
1088 much everywhere, it is important that we tell you where we really know
1089 it works because we have tested it. We do test on Linux, Windows, and
1090 Mac with a range of compilers and libraries for a range of our
1091 configuration options. Every commit in our git source code repository
1092 is currently tested on x86 with gcc versions ranging from 4.4 through
1093 4.7, and versions 12 and 13 of the Intel compiler as well as Clang
1094 version 3.1 through 3.4. For this, we use a variety of GNU/Linux
1095 flavors and versions as well as recent version of Mac OS X. Under
1096 Windows we test both MSVC and the Intel compiler. For details, you can
1097 have a look at the `continuous integration server used by GROMACS`_,
1098 which runs Jenkins_.
1100 We test irregularly on ARM v7, BlueGene/Q, Cray, Fujitsu PRIMEHPC, Google
1101 Native Client and other environments, and with other compilers and
1102 compiler versions, too.