Mark Abraham [Mon, 17 Mar 2014 17:06:04 +0000 (18:06 +0100)]
Merge "Merge branch release-4-6"
Mark Abraham [Fri, 14 Mar 2014 18:08:48 +0000 (19:08 +0100)]
Merge branch release-4-6
Conflicts:
src/gromacs/pulling/pull_rotation.c
Adopted the bugfixes from release-4-6
Uncrustified incoming code
Change-Id: I7d90bb49ef83750b9607f677e97a94f2b5793eeb
Mark Abraham [Sun, 16 Mar 2014 12:05:28 +0000 (13:05 +0100)]
Fix perturbation to Verlet scheme when not perturbing
A .top that has B-state parameters for non-bonded interactions
should run the same with free-energy = no as a .top that has no
B-state parameters. Renamed variable to better reflect its use,
and used it to correctly trigger the FEP code path.
Change-Id: I47656b29ef4ca2462af0fb587e0424d0eab3d5c4
Teemu Murtola [Sat, 15 Mar 2014 05:00:41 +0000 (07:00 +0200)]
Update manual analysis section
- Reference tools as 'gmx <tool>' instead of the old names.
- Expand the section on selections now that there actually are some
tools that can use the full power.
- Rewrite sections that were referencing obsolete tools (g_bond and
g_sgangle).
- Remove references to tools that aren't there.
Change-Id: Iee99f69e2d4effa585d98930ef1d78c1367e2ec0
Erik Lindahl [Sun, 16 Mar 2014 09:13:11 +0000 (10:13 +0100)]
Use a constant instead of sizeof(int) for SSE2 immediate constant.
Apparently gcc-4.8.2 was not able to evaluate sizeof(int) to the
constant 4 at compile time, but thought it was either a 64-bit
value or a non-const value. Since the register sizes are fixed
we simply hardcode the value 4 bytes instead.
Change-Id: If739b733727b032eb7e5b7fa0a75e7cd51989a0b
Carsten Kutzner [Wed, 12 Mar 2014 15:14:04 +0000 (16:14 +0100)]
Avoid cross product with zero vector in rotational pulling.
Fixes #1431 (rotation/flex-t regression test failing on BG/Q)
In do_flex_lowlevel() we checked (by mistake!) for xj-xcn being
zero, although we need to check for yj0-ycn being zero, since
we use yj0-ycn in a cross product in the following lines of code.
I now also replaced the direct check (0 == norm(...)) by checking
what gmx_numzero(norm(...)) returns. The latter replacement
was also applied in the do_flex2_lowlevel() routine. Note that there
the check for small xj-xcn was and is actually correct.
Change-Id: I972b6d67a81e30f297db286cd2224f66753a20aa
Carsten Kutzner [Fri, 14 Mar 2014 10:27:36 +0000 (11:27 +0100)]
The van der Waals radius must equal the Coulomb radius with Verlet
This patch fixes an issue that can occur when using g_tune_pme
with a Verlet pair-list .tpr input file. If the .tpr itself has
large cutoffs (e.g. 1.2 nm) and one asks g_tune_pme to scale
_down_ the Coulomb radius, the van der Waals radius is not
scaled down with the Coulomb radius (only upscaling worked).
One ends up with an unusable .tpr file because rVdW != rCoul.
This patch ensures that van der Waals and Coulomb radii
are always equal with Verlet pair-lists. Fixes #1460. Thanks
to Joao Rodrigues for reporting the issue!
Change-Id: I5ef30e71a35cd83838040057e16e52e09ea82e9a
Christoph Junghans [Tue, 11 Mar 2014 16:31:22 +0000 (10:31 -0600)]
Install enums.h
Related to #1452
Change-Id: I94db8d0f9661b7398e5924f4adefbfdcc82da483
Mark Abraham [Thu, 6 Mar 2014 17:42:31 +0000 (18:42 +0100)]
Fix MKL linking
Setting the same for CMAKE_STATIC_LINKER_FLAGS broke because that
variable got applied to calls to ar for linking libgmock.a. Probably
not important because Intel probably encourages shared linking for
MKL.
Change-Id: I6816d8ba318c704a0efad7659f5afecaeb48b092
Mark Abraham [Wed, 26 Feb 2014 14:58:50 +0000 (15:58 +0100)]
Avoid array-bounds warnings with gcc 4.8.2
The asserts fix RelWithAssert build and for Relase
build the warning is suppressed.
Change-Id: I0a7e6576014bd32648a1e2f08d88e1f2eef3335e
Rossen Apostolov [Thu, 20 Feb 2014 15:48:29 +0000 (16:48 +0100)]
Removed .itp files from top level directory.
There is no need for these files, the water models
should be included from the corresponding .ff directory.
Since many users are likely to still use those files,
an error message is printed.
Also re-added ions.itp and treated it similarly.
Also update an integration test that was using a one of the top-level
water .itp files that is now removed.
Change-Id: I599e203329340587afa7fc5efced485d09b9ce1c
Roland Schulz [Sat, 1 Mar 2014 06:43:37 +0000 (01:43 -0500)]
Simplify gmxTestLibXml2
Change-Id: Ia652633f529cc1cf6ba09e29159025ff762efad2
Berk Hess [Fri, 28 Feb 2014 22:03:09 +0000 (23:03 +0100)]
PME load balancing now works with LJ-PME
PME load balancing only works with LJ-PME when PME is also used
for electrostatics.
Change-Id: I7145b242fd5dd00690e6b789b9cc6f3e147d430b
Teemu Murtola [Sat, 8 Mar 2014 11:10:34 +0000 (13:10 +0200)]
Fix misc. Doxygen issues
- Fix some missing brief descriptions.
- Fix some function definitions so that Doxygen correctly pairs them
with the corresponding declarations.
- Fix some documentation visibility issues. Not all these are strictly
required, in particular in the SIMD parts, where \cond is used to
exclude full classes. But the presence of \cond is
difficult/expensive to test automatically.
- Also some misc. other fixes.
- Document some of these issues better on the Doxygen page.
- Exclude parser.h from Doxygen, since it doesn't contain any meaningful
comments, and confuses Doxygen at least with the XML output.
Also, explicitly list input file extensions for Doxygen.
This excludes some random Python files from the documentation. At least
for now, we probably want to keep the documentation such that it applies
mainly to the C/C++ code, and not bother how to also document Python
code in a Doxygen-compatible way.
CUDA files are still left out, as they were previously, since they
currently produce several warnings.
Most of these were found by the current work-in-progress checker script.
After these, the checker mainly complains about a lot of missing brief
descriptions in thread-MPI headers and in swapcoords.c.
Change-Id: Ib8a492fde6327ce3d814a23d887b05323bd6d334
Roland Schulz [Fri, 7 Mar 2014 08:38:35 +0000 (03:38 -0500)]
Don't disable MSVC warnings in config.h
All other warning flags are passed as compiler flags, thus this
was inconsistent. Also didn't work correctly if source file
doesn't include config.h.
Also makes ICC warning flags consistent between Linux and
Windows.
Change-Id: Ic52f6eb3c2dc5751b8aac1bba836ad5ff7d7c3cf
Roland Schulz [Sun, 9 Mar 2014 01:00:49 +0000 (20:00 -0500)]
Fix CheckCXXCompilerFlag
ICC on Windows was always failing the CheckCXXCompilerFlag.
This was caused by the warning:
LINK : warning LNK4224: /INCREMENTAL:YES is no longer supported; ignored
being matched by the compiler fail_regex for aix because of a missing
escale for the semicolon. A single escape isn't sufficient. Using double
escape ("\\\;") doesn't produce this problem, but I'm not sure whether it
always matches correctly thus instead a "." is used (works because it is
a regex). This fix doesn't address either the fact that the deprecated
linker flag is used nor does it address the problem that if incorrect
flags are somehow added (either by cmake or the user), this macro
fails without a diagnosible error message or fatal error (to notice it).
Change-Id: Ia45d6cd4fbf4f464bcb03e0d945085987a97aaad
Szilárd Páll [Fri, 28 Feb 2014 12:21:39 +0000 (13:21 +0100)]
Bump required CUDA version to 4.0
- Added warning with CUDA 4.0;
- Fixed some typos in comments;
- Fixed the comment referring to GPU code generation with CUDA v5.0+.
Fixes #1382
Change-Id: Iee182ace765022fc6a179531d42e965594cff104
Roland Schulz [Fri, 7 Mar 2014 06:06:41 +0000 (01:06 -0500)]
Replace inline with gmx_inline
We always try to avoid defining std names in our headers.
Thus we shouldn't define inline for those C compilers (=MSCC)
which don't have it.
Change-Id: Ia2023435c916892cdaf2ce97905b55ec10cd1a08
Mark Abraham [Sun, 9 Mar 2014 21:02:30 +0000 (22:02 +0100)]
Fix dependency on swapcoords enums
state.h needs these enums, but it seems wrong to introduce
a header dependency on the whole enums.h, when so few places need
these enums at all.
Change-Id: Ia8f080d5c0dc73b19a54f5ff77fbf7ef76f16139
Teemu Murtola [Sat, 8 Mar 2014 11:27:59 +0000 (13:27 +0200)]
Remove some unnecessary declarations
- topexcl.h was an outdated copy of gpp_nextnb.h, and wasn't used
anywhere. Rename the source file to gpp_nextnb.c to match the header.
- Remove one declaration from gpp_nextnb.h, which wasn't used outside
the source file. Make it static in the source file.
- Remove duplicate declarations of (init|done)_top().
- Remove unused create_visibility() from src/programs/view/.
Change-Id: I87982e5581aff8caca3dc753492e01ee457ab694
Roland Schulz [Thu, 6 Mar 2014 07:46:51 +0000 (02:46 -0500)]
Remove KERNEL_WIDTH
Wasn't used anywhere
Change-Id: I579297d71225ba3c0360f8419856b0f5e72a4b5a
Rossen Apostolov [Thu, 20 Feb 2014 15:34:45 +0000 (16:34 +0100)]
Removed the deprecated gmx.ff and gmx2.ff.
There are still a few places in the manual that need to be fixed.
Change-Id: I5c265470c1b73cb755918c5a000208da3d1e7b22
Mark Abraham [Thu, 6 Mar 2014 13:51:16 +0000 (14:51 +0100)]
Fix formatting in install guide
Change-Id: I28833dee21220aacbf06ccc7d06f8f5e6a328352
Mark Abraham [Tue, 4 Mar 2014 15:18:48 +0000 (16:18 +0100)]
Fix CMake syntax error
Change-Id: I8a39e50a741a23da3640fe7ccbcd83c844bf0d69
Roland Schulz [Wed, 5 Mar 2014 06:34:53 +0000 (01:34 -0500)]
Use gmx_int64_t for gmx_off_t
We anyhow require 64-bit file support and 64-bit integer
support. Thus the defintion of gmx_off_t can be dont
trivially.
Partial fix for #1452
Change-Id: I6ae70454cfa8ace3ca84a047149b6f4648e84b00
Berk Hess [Tue, 4 Mar 2014 12:01:19 +0000 (13:01 +0100)]
Fixed exit with FEP with PME nodes without LJ-PME
With FEP runs with separate PME nodes without LJ-PME, mdrun would
exit with a gmx_incons.
Reorganized the PME some flags and conditionals for clarity.
Change-Id: I04513a2d84682617b36c0c8ad4afe632ce3f7ee1
ngoga [Tue, 25 Feb 2014 16:27:29 +0000 (17:27 +0100)]
Martini CG - Combined angle-torsion potentials
- Potential important for Martini
- Most Important diffence: bondfree.c
- Rest of files: definition of types
- Different MD groups want to have it
- Avoid the numerical instabilities
of having three beads in a row and
therefore an undefined dihedral
angle. This might not happen easily
at the atomistic level, but for coarse-grain
models this is a real pain.
- Paper at
http://md.chem.rug.nl/~monica/Bulacu_Marrink_2013.pdf
Adopted new tpx_version scheme.
Paper published.
Change-Id: I8ed470c464d843eb0218c87aeea3a44c3a1f64f8
David van der Spoel [Wed, 5 Mar 2014 20:08:18 +0000 (21:08 +0100)]
Six new quotes from classic literature.
Change-Id: Id881baea8d6eeedc70dec3975a4c3e20e788285a
Roland Schulz [Sun, 2 Mar 2014 19:40:47 +0000 (14:40 -0500)]
Add RelWithAssert build type
Works mostly like Release just without -DNDEBUG.
Also generalized some machinery for managing compiler/linker flags so
that we can have an easier time later when we reconsider how to manage
all of our build types.
Also minor corrective changes to the behaviour of other build types.
Change-Id: I70283cb6069cb772c287c0aef329d95b8eacd987
Roland Schulz [Wed, 5 Mar 2014 07:34:17 +0000 (02:34 -0500)]
Remove dependency on TNG in trxio.h
Partial fix for #1452
Change-Id: I59474ea08d40362fa9d10745a5fb1d4e12254b68
Roland Schulz [Thu, 27 Feb 2014 06:41:41 +0000 (01:41 -0500)]
Report MIC SIMD correctly
Avoid cmake warning because SIMD cannot be detected. And
avoid that version information prints "None". This is
needed because MIC doesn't use GMX_SIMD so that the same
setting (__MIC__) can be used both for native and (future)
offload.
Also added installguide section.
Change-Id: I4af3b13678aa5c3c69c60bda499c3691ac83dc53
Mark Abraham [Wed, 5 Mar 2014 06:23:00 +0000 (07:23 +0100)]
Merge "Merge release-4-6 into master"
Berk Hess [Fri, 28 Feb 2014 11:37:29 +0000 (12:37 +0100)]
Buffer estimate now works with rvd-switch=0
With rvdw-switch=0, the Verlet buffer estimation code would divide
by 0. The estimate for vdw-switch functions now use the exact 3rd
derivative of the potential at the cut-off iso estimating the 2nd.
grompp now prints a note with rvdw-switch < 0.5*rvdw.
Updated estimate formula in the manual and added a missing factor
of 2 (the code did use the factor of 2).
Change-Id: Ia259d703515f69b0ed3390e2f7be671e0a15bd5f
Roland Schulz [Thu, 27 Feb 2014 05:53:18 +0000 (00:53 -0500)]
Check zlib can actually be linked
Only use zlib compression in tng if zlib can really be linked.
This is equivalent to
ee26e1264 for libxml2.
It is slightly simpler than the approach in gmxTestLibXml2
and CheckLibraryExists could be used there too.
This does not detect potential linking errors with shared
libraries similar to those solved in #740 for FFTW.
Fixes #1435
Change-Id: I5009e50b6caf810a12e71ecd83bde2edb21a2b32
Roland Schulz [Mon, 3 Mar 2014 23:22:31 +0000 (18:22 -0500)]
Avoid segfault when comparing tpr
Segfault was possible if number of idef didn't match
Change-Id: I62315e3340eec8acf14c01d14917392bc63e46e0
Mark Abraham [Sun, 2 Mar 2014 16:53:40 +0000 (17:53 +0100)]
Add enum and docs for using tpx_version
Documented procedure for having hassle-free feature branches and
ensuring merges into main branch cannot silently cause problems.
Added an enum to make the process work smoothly.
Change-Id: I69b87f9bac0b51784e70f8a618808ef086951343
Roland Schulz [Mon, 3 Mar 2014 20:31:54 +0000 (15:31 -0500)]
Merge release-4-6 into master
Conflicts:
admin/installguide/installguide.tex
3643d2a Add fatal error for Andersen+constraints+DD ignored
because already in master.
Change-Id: I6b3346503036e5b31bcc929daa5bcc9a04dde3bc
Mark Abraham [Mon, 3 Mar 2014 16:34:53 +0000 (17:34 +0100)]
Fix calling SIMD group kernels
The #define of GMX_SIMD*_OR_HIGHER takes place in simd.h, which got
changed late in the evolution of the recent SIMD-module patch, in
order to reduce unwanted module dependencies. However, the explicit
dependency of the group kernel setup code on the SIMD module was not
added. (It seems that the dependency on the old SIMD macros.h was been
satisfied transitively.) In effect, this turned off group-scheme SIMD,
but the silent fallback to C kernels meant there was no red flag until
we got around to measuring performance.
Change-Id: I78842cb191f6188fe4188800399efd2ed99e5543
Teemu Murtola [Fri, 28 Feb 2014 04:33:59 +0000 (06:33 +0200)]
Fix incorrect grid cell size in g_sas -nopbc
Fixes #1445
Change-Id: I798fc8fe96608633f26d9a3500f83f39b44af008
Teemu Murtola [Sun, 12 Jan 2014 18:37:08 +0000 (20:37 +0200)]
Move default library path search to program context
The default library path search is now implemented in
CommandLineProgramContext, and accessible through the generic interface.
This provides the following benefits:
- All the magic for finding stuff from the binary path is in a single
place, under the commandline/ module, where it belongs.
- Code using the library path is now in C++, fixing some potential
buffer overruns and removing the need for some utility code.
- Programs using the Gromacs library can customize how the library data
files are found.
- PATH_SEPARATOR is no longer exposed outside path.cpp, encapsulating
this detail.
- The library path is only searched once during the execution of the
binary, reducing the amount of file I/O done if multiple data files
are accessed.
- The library path is printed in the startup header, making it explicit
where they are read (helpful in particular if there are problems with
it, or the user may have multiple installations). The GMXLIB
override doesn't currently affect this printout; maybe it should, but
I don't want to put the environment variable (and the support for
multiple directories) into the program context.
Change-Id: I4d099215ef7e1f269d4c7a42989835d980bc1566
Mark Abraham [Mon, 24 Feb 2014 16:31:07 +0000 (17:31 +0100)]
Add fatal error for Andersen+constraints+DD
This combination produced a temperature that was 6.5 degrees higher
than the same with one domain, for a 1ns PME lysozyme in water
simulation.
Change-Id: I9f80276c47de955a5053bcabb6fe7c9bfdceaf0e
Christoph Junghans [Mon, 3 Feb 2014 20:52:32 +0000 (13:52 -0700)]
switch ld_seed from int to gmx_int64_t
* allows more than 4,294,967,295 different seeds for ld
* tpx version bumped, used new style for magic version constants
* changed printf format strings to conform to new type
Change-Id: Id661ca331a100dc942ac978cc1b50cfdade09731
Szilard Pall [Tue, 25 Feb 2014 20:39:07 +0000 (21:39 +0100)]
Added CUDA LJ-PME nbnxn kernels
This change implements CUDA non-bonded kernels for LJ-PME introduced
in the Verlet scheme with 99029d.
The CUDA kernels implement geometric as well as Lorentz-Berthelot (LB)
combinations rules (unlike the CPU SIMD) mostly because even though PME
is very slow with LB, it is still beneficial to let the user offload the
non-bondeds to a GPU and potentially bump up the cut-off to further
reduce the CPU PME load.
Note that as now we have 120 kernels compiled for up to four different
target architectures, the nbnxn_cuda module takes a very long time to
build and can become the bottleneck during compilation. We will deal
with this later.
Change-Id: I819b59a8948da0c8492eac6a43d4a7fb6dc98354
Magnus Lundborg [Sat, 1 Mar 2014 11:39:22 +0000 (12:39 +0100)]
Bumped TNG to latest version.
Change-Id: Id3ca6c18c1f51640997829bd8ad251a7ca91d04f
Szilard Pall [Fri, 28 Feb 2014 21:11:08 +0000 (22:11 +0100)]
Improved CUDA non-bonded kernel performance
Some old tweak which was supposed to improve performance had in fact
the opposite effect. Removing this tweak and with it eliminating
shared memory bank conflicts it caused improved performance by up
to 2.5% in the force-only CUDA kernel.
Change-Id: I7fcb24defed2c68627457522c39805afc83b3276
Teemu Murtola [Mon, 27 Jan 2014 18:55:00 +0000 (20:55 +0200)]
Residual SIMD Doxygen improvements
- Add a link to the overview documentation from the module page.
- Remove some trailing spaces.
Some issues noted, but not fixed here:
- \nosubgrouping would make the file pages look more consistent, if it
worked for files...
- Some of the SIMD routines are not included in any named group, making
them appear separately at the end of the documentation.
- The list of members on the SIMD module page is exceedingly long,
making it hard to find things there. Would require some thinking on
how to best solve this.
Change-Id: I628f321172a3c3e8f4874e166c48e12763d1490e
Teemu Murtola [Mon, 27 Jan 2014 18:55:00 +0000 (20:55 +0200)]
Misc. Doxygen configuration changes
- Adjust Doxygen preprocessor behavior such that it actually parses
#ifdefs in the code. Otherwise, some parts of the SIMD code seemed to
get ignored.
- Extract static functions on all levels of documentation for
consistency (and we may end up with some static inline members in
public headers; currently at least vec.h qualifies).
- Remove unnecessary excludes now that VERBATIM_HEADERS=NO already
excludes the unwanted files.
- Some other minor cleanup.
- Move member groups higher up in the layouts, since they generally are
the most interesting part.
Change-Id: Iea04772628716ed12f44cd4465c2027f4670bfae
Teemu Murtola [Sat, 26 Oct 2013 05:25:28 +0000 (08:25 +0300)]
Convert gmx-sas to the C++ framework
An example of converting a non-trivial analysis tool to the new
framework. Most functionality is intact, but the command-line interface
is changed a bit. Changes in behavior:
- Writing a position restraint file is not implemented. Would not be
very difficult to add.
- Solvation free energy estimates are written into a separate file
instead of the main output file, and only if requested with an -odg
option.
- The calculation group is specified as a selection with -surface.
Can be a dynamic selection. The total area of this group is always
written out.
- There can be 0..* extra output groups, specified as selections with
-output. All can be dynamic, and must always be a subset of the
calculation group.
- There is no longer a separate concept of hydrophilic and hydrophobic
surface area. The user can trivially calculate such things by
providing two different output groups. The old hydrophobicity
definition is easily expressed as "... and charge {-0.2 to 0.2}".
- tpr file is no longer required for most of the output.
- Output precision is different from the old tool. The current
precision (fixed to three decimal places) should be sufficient for
most uses, but it could be considered to use %g for formatting like
the old tool if really necessary.
Future TODOs:
- Consider if the legends on the plots could be improved.
- Add unit tests for the actual surface area calculation routine.
Part of #665.
Change-Id: Iee60c13b927b3b63b6e218164cd961971f2f3fce
Teemu Murtola [Thu, 27 Feb 2014 05:02:02 +0000 (07:02 +0200)]
Add support for non-uniform X axis in array data
It is now possible to construct AnalysisArrayData objects that have a
non-uniform X axis. AnalysisDataAverageModule can use this to provide a
more meaningful X axis in the output plot in case the values correspond
to, e.g., a set of atoms.
Change-Id: I578f6944d4ff51c315d068b011b705be2a4be00a
Teemu Murtola [Wed, 26 Feb 2014 09:55:03 +0000 (11:55 +0200)]
Improve support for data sets in C++ tools
- If a data set does not set the number of columns, the framework now
leaves it uninitialized (does not call startData() etc.), and the tool
can check this when processing the frame.
- Make it possible to specify an output file in a tool unit test without
actually testing the contents of the file. This supports testing
tools that only produce data in response to certain output options
being set.
- Add support for specifying a comparison tolerance for a data set in a
unit test.
Change-Id: I49794c93336c22c07f1a745d28d1b4f6f856b25b
Szilard Pall [Fri, 10 Jan 2014 11:46:15 +0000 (12:46 +0100)]
CUDA kernels for switched LJ
This change adds CUDA GPU kernels for force and potential switching,
implemented in the Verlet scheme for CPUs in f3acaf.
Change-Id: I73a83d352924379d9005104fa59f3478d3e684b0
Teemu Murtola [Sun, 23 Feb 2014 19:42:17 +0000 (21:42 +0200)]
Improve floating point comparisons in unit tests
Improve basic floating point comparison support:
- Replace EXPECT_NEAR_TOL with a more flexible alternative that allows
specifying various tolerances.
- Some Doxygen improvements in testasserts.h.
Additionally, make the reference data framework use the new approach:
- Add a tolerance setting to the reference data framework, and make the
default tolerance a lot stricter.
- Store the reference data with enough precision to allow lossless
roundtrips. This makes them more difficult to inspect visually with
the XSLT files, but that can be addressed separately.
- Run FFT tests with GMX_FFT_FLAG_CONSERVATIVE to make the results with
FFTW have less variance. The 3D FFT tests were already using this,
but others weren't. Set the tolerance for the FFT tests based on a
bit of testing of what works. It should now pass even without the
conservative flag, at least most of the time. The tolerance is still
lower than what it was earlier even for single, and a lot stricter for
double.
- Regenerated all reference values from a double-precision build.
For the FFT tests, used fftpack without optimizations.
Change-Id: I187dd9be20fd17cc8b3e779dcd9dbad34043bb00
Roland Schulz [Wed, 26 Feb 2014 22:26:26 +0000 (17:26 -0500)]
Remove Cygwin+MingW from installguide
MingW was never supported (#448). And Cygwin is currently
not working. Until/Unless it is working the guide shouldn't
claim support.
Change-Id: I6558b95bc47b80f7d97a7960587e83e959682990
Berk Hess [Tue, 26 Nov 2013 07:52:16 +0000 (08:52 +0100)]
Free-energy now works with the Verlet scheme
Implemented perturbed non-bonded interactions with the Verlet
by masking them from the nbnxn pair lists and storing them in an
old-fashioned group neighbor list. This way we keep a single
free energy kernel.
The free energy kernel uses OpenMP atomic for reductions.
Also fixed grompp setting nstcalcenergy=0 with nstenergy=0.
Change-Id: I5a6a7e84b46e06250d141e2e08cb3a11077cddab
Alexey Shvetsov [Fri, 29 Nov 2013 18:54:50 +0000 (22:54 +0400)]
Fix -Wmaybe-uninitialized warnings in some files
Also clean-up and clarification of clean_vsite_dihs.
Change-Id: Ica4b1d17039e23bc2274a20fe5abca18295a7593
Signed-off-by: Alexey Shvetsov <alexxy@omrb.pnpi.spb.ru>
Teemu Murtola [Wed, 26 Feb 2014 05:08:53 +0000 (07:08 +0200)]
Misc. Doxygen build system improvements
- Check Doxygen version when running one of the Doxygen targets, and
produce a note that states the expected version if there is a
mismatch.
- Check the result of the Doxygen process and fail the build if the
program returns an error code.
- Add a tag to all warnings that identify the type of documentation that
produces them. This should make it easier to understand the Jenkins
warnings in the case not all documentation types produce a certain
warning.
Change-Id: I030c0b2c69958fe41f0b964399cb1c2e1b250e77
Erik Lindahl [Mon, 24 Feb 2014 23:06:24 +0000 (00:06 +0100)]
Added message to warn that group kernels are deprecated.
Nothing will change for now, but both grompp and mdrun now
warn that we will remove the support for group kernels in
a future version when all interaction forms are supported by
the verlet cutoff scheme (tentatively a year from now).
Change-Id: Id00a67e44e60818cb0e6f1897d6939f53ea01540
Michael Shirts [Tue, 25 Feb 2014 03:28:21 +0000 (22:28 -0500)]
Updating and improving Andersen thermostat documentation
Added mdp options for andersen thermostat, and updated to
reflect changes in 5.0, plus add a citation for one of the assertions.
Change-Id: Ia713a4f2dd39dd5224952be334d8953b6613a66e
Roland Schulz [Wed, 26 Feb 2014 06:20:57 +0000 (01:20 -0500)]
ICC aliasing optimization flags
GCC already uses -fstrict-aliasing (-ansi-alias). Thus it is safe
to also use this for optimization in ICC. const/non-const aliasing
shouldn't be used anywhere either and should be safe too.
Change-Id: I6a1b10c3d00bc2473a25a15e26934f61c046157d
Roland Schulz [Mon, 10 Feb 2014 21:30:50 +0000 (16:30 -0500)]
Use reduce in pme
Use the reduce added to the simd module. Also fixes that
aligned store was used for possible unaligned stack buffer.
Change-Id: I5e08b0a0998fe5371148d1af9668fd6f2ecd5934
Erik Lindahl [Wed, 22 Jan 2014 17:30:10 +0000 (18:30 +0100)]
Redesigned SIMD module and unit tests.
Second part of the commit (after renaming), which contains
the actual functionality changes. This new version
implements a complete interface layer that
is architecture-agnostic, while each architecture has a
separate implementation file with both single, double and
simd4 versions (that can be used simultaneously). simd.h
contains a number of defines that describe the capabilities
of the instruction set, and there is a new documentation
module for Doxygen. This will be used in a later patch
for modularized verlet kernels. With that, we hope to remove
all architecture-specific SIMD code from the rest of Gromacs.
All SIMD math functions have been redesigned so they work even
with instruction sets that do not support integers (and even
on sets that do not support logical operations), and
accuracy has been improved for double precision sincos() by
removing the table implementation. To try to reduce
the size of this relatively large patch I have kept a few
header files (in particular the math files in gromacs/simd)
to avoid touching all the group kernels. With this new kernel
module, 256-bit AVX2 SIMD acceleration will now automatically be
enabled for the verlet kernels. Group kernels will use AVX_256
in this case. Also incorporates changes from Teemu to make
static and gmx_inline library functions appear correctly in
the Doxygen documentation.
Relocation of nbnxn SIMD setup from nb_verlet.h to nbnxn_simd.h
in mdlib by Berk. Now the nbnxn SIMD setup is completely internal.
Replaced the pr4 functions in the nbnxn kernels by simd4.
The nbnxn kernel selection is now nearly architecture agnostic.
Also enabled FMA again for pmecorr SIMD functions in double.
Change-Id: I643da75f346f120500682bcc4bcc1333a635db70
Roland Schulz [Mon, 20 Jan 2014 21:33:02 +0000 (16:33 -0500)]
Replace all mdrun rngs with cycle based rng
The stateful random rumber generator (rng) used previously doesn't
produce reproducible results in parallel for sd/bd and doesn't
produce reproducible results for continutation for replica exchange.
The rng state has been removed from the checkpoint file.
Fixes #995
Change-Id: Id2a5d064cf363c54db3c16a0675cfeba553feeaa
Teemu Murtola [Wed, 26 Feb 2014 07:19:46 +0000 (09:19 +0200)]
Fix GMX_PRIu32 and GMX_PRIu64 for MSVC
Change-Id: Ib7490c66687d98a135338d51de41bfda35b8938d
Mark Abraham [Fri, 18 Oct 2013 15:47:25 +0000 (17:47 +0200)]
Add ThreadSanitizer build type
This might be useful for helping detect misuse of thread
operations.
It does run quite slowly, so use in an automatic per-patch Jenkins
build is probably not a good idea. In any case, it should be run
on a real workload, as well as toy examples.
As discussed in the CMake file that manages the new build type, using
TSan requires that CMake can demand that atomic operations are not
used. Accordingly, the atomic.h functionality and the CMake atomics
testing that uses it have some changes so all the ways atomic.h is
used work correctly. Documentation of how and why things work is
expanded.
A benign race condition with the (unused) return value of
mdrunner_start_threads is fixed.
Change-Id: I0da01c0bedeec2e49d4456d757c54fbe58c56ef4
Berk Hess [Tue, 25 Feb 2014 06:29:54 +0000 (07:29 +0100)]
Fixes seperate PME nodes without LJ-PME
The very recent LJ-PME nbnxn patch dereferenced a NULL pointer
with seperate PME nodes without LJ-PME. Also LJ parameters were
communicated when not needed.
Change-Id: I40bd0c8f96294b54599afee6021139b71cc8c860
Mark Abraham [Wed, 12 Feb 2014 15:57:29 +0000 (16:57 +0100)]
Keep compilers happy on BG/Q
Change-Id: I99439b002f95145504c359fddc61dc701431d8f0
Mark Abraham [Sat, 22 Feb 2014 15:00:59 +0000 (16:00 +0100)]
Clean up options for domain decomposition
Documented mdrun options better. Documented the environment variables.
Renamed the environment variables to better describe their function.
Renamed the helper function because not all of these environment
variables relate to numbers of steps.
Change-Id: I12dba40f8cdf489ea23c3f4dd8810efde54a4aaf
Berk Hess [Thu, 5 Dec 2013 09:27:59 +0000 (10:27 +0100)]
Implemented LJ-PME nbnxn kernels
Implemented nbnxn SIMD and C kernels for LJ-PME. CUDA kernels will
follow later.
The nbnxn LJ-PME kernels subtract the full mesh contribution, using
the mesh C6 coefficients (usually geometrically combined). This gives
much higher accuracy for non-geometric force-fields.
The group kernels will soon be changed to use the same setup.
Change-Id: Ibbc6eef1c62a97ff6b2e3013f19594946581e44e
Berk Hess [Thu, 20 Feb 2014 09:01:22 +0000 (10:01 +0100)]
Fixed bug with LJ pot-switch in 4xN kernels
The recent commit of LJ switch functions in the nbnxn kernels
contained a bug in the 4xN kernels which caused the force on some
pairs not to be switched.
Change-Id: Ib4303223924e92d7e84641071a4a56955de70274
Mark Abraham [Mon, 24 Feb 2014 04:32:06 +0000 (05:32 +0100)]
Permit warning-free use of Andersen thermostat
Simulations using Andersen thermostat don't need to remove center of
mass motion, because that is intrinsic to the algorithm. Using it with
nstcomm > 1 generated a warning that center-of-mass removal is
unnecessary. Using it with nstcomm == 0 generated a warning that
ice-cube artifacts might be generated. Resolved by suppressing the
latter warning for both Andersen thermostats.
Change-Id: I6b2c1594cabd81964b3ba1ebc5dd61e0f1debb5e
Mark Abraham [Sat, 22 Feb 2014 18:13:35 +0000 (19:13 +0100)]
Suppress icc warning for Boost 1.55 use of #pragma once
Change-Id: Ib187cf34ba2c72a2c57fc55cfd26154508e9543b
Rossen Apostolov [Thu, 20 Feb 2014 16:21:50 +0000 (17:21 +0100)]
Removed deprecated Encad FF.
Change-Id: Ib7f5c3d9920307a808dde9ef13b0f042827eb8ca
Roland Schulz [Thu, 20 Feb 2014 08:57:56 +0000 (03:57 -0500)]
Fix load_lj_pair_params2 for MIC
Change-Id: I181bf5c6190d19ee2cb7bcf629c1b81e849e6373
Roland Schulz [Wed, 29 Jan 2014 19:07:28 +0000 (14:07 -0500)]
Updated Boost to latest 1.55.0
Shouldn't have any disadvantage because boost still has support
for e.g. GCC 3.3. Newest version should work better with newer
compilers (e.g. no warnings).
Reduced the number of changes in boost to two:
- Virtual dtor fix is not required anymore
- Removed #error which was aimed at producing compiler errors
for removed dependencies. Because this internal boost is only
meant to be used by Gromacs we shouldn't have to detect
odd usage of BOOST_* defines.
Fixed two comipler warnings:
- Applied nvcc.path from https://svn.boost.org/trac/boost/ticket/9392
- Commented out unsed parameter
These warnings should be fixed in future boost versions but if not and
the changes have to be reapplied, check Gerrit for the diff. Patch Set 1
contains the unmodified version without the fixes.
Change-Id: I09179891f9979a5997171a579a6a552621a693df
Roland Schulz [Fri, 21 Feb 2014 20:16:23 +0000 (21:16 +0100)]
Merge "Merge release-4-6 into master"
Nina Fischer [Thu, 20 Feb 2014 10:48:07 +0000 (11:48 +0100)]
Added new amber99sb-ildn-bsc0 force field.
It extends the original amber99sb-ildn with new
parameters for nucleic acids, while the protein
parameters are the same in both force fields.
Change-Id: I665ea598015428ced103ce18f6665afd0d5ede8c
Szilard Pall [Fri, 7 Feb 2014 18:57:09 +0000 (19:57 +0100)]
portability aspects in install guide + minor tweaks
Added information on portability aspects related to CPU instruction
sets, related to #1428.
Additionally, made several minor updates and tweaks related to
compilers, platforms, cmake, etc.
Change-Id: I621262c939c119e5bdd5e7c91dda0ae3ffc60b7b
Roland Schulz [Fri, 21 Feb 2014 00:53:08 +0000 (19:53 -0500)]
Merge release-4-6 into master
Changed BUILD_CPU_ACCELERATION to GMX_SIMD in
CMakeLists.txt.
Reverts change to src/mdlib/pme_simd4.h. Corresponding
fix for master is under review as I5e08b0a09.
Removed reference to PD in message in shellfc.c.
Conflicts (all trivial):
CMakeLists.txt
src/contrib/fftw/CMakeLists.txt
src/gromacs/gmxlib/gmx_detect_hardware.c
src/gromacs/legacyheaders/shellfc.h
src/gromacs/gmxlib/copyrite.cpp
Change-Id: I2d43520ae8833055eff51d431e47cd8b90d3a687
Mark Abraham [Thu, 20 Feb 2014 21:59:41 +0000 (22:59 +0100)]
Reinstate shell code with DD
Further work on the complex/sw test case in the 5.0 regressiontests
branch reveals that the initial conditions may have been the reason
for the problems observed with DD and more than one node, rather than
the implementation.
Refs #1429
Change-Id: I26ff6d9f8c79605afa794cae4761b5643b712124
Mark Abraham [Sat, 12 Oct 2013 00:27:28 +0000 (02:27 +0200)]
Remove particle decomposition
The paths that are eliminated are those for which MD_PARTDECOMP was
needed, those triggered by PARTDECOMP(cr), and any for which PAR(cr)
&& !cr->dd.
Reviewers, please note
* that for the purposes of this patch, OpenMP is not a form of
parallelism,
* the definition of DOMAINDECOMP(cr) in commrec.h,
* that TPI and NM can run in parallel, but use neither PD nor DD,
* multi-simulations run in parallel but need not use DD, so
* we still need both PAR(cr) and DOMAINDECOMP(cr), and
* I have generally left the indenting alone to make for easy review,
but we will uncrustify before merging
Summary of changes in this patch:
Removed
* code triggered by MD_PARTDECOMP
* code triggered by PARTDECOMP(cr)
* anything decomposition-related with "pd" in the name (but the name
clash with x86 vector intrinsics is unfortunate)
* t_mdatoms field called start (DD does atom number remapping)
Note that bPDvsitecomm was never set anyway!
* manual section
Renamed two functions with "pd" in their name changed to "serial"
because they are still needed there (and moved them?)
Deleted files, functions and function paramters that only supported PD
code paths.
Cleaned up
* use of DOMAINDECOMP(cr)
* a bit of Generalized Born code that should have been static or
was unused
* bcast_state machinery can now be much simpler; still does the right
thing, but does it further behind the scenes and easier to
understand
* made explicit the assumption of replica exchange that the
integrator is dynamical, which makes for some minor
simplifications
Refs #1292
Change-Id: If029f16e6b4b06d58d465afe072a3cde6481479e
Christoph Junghans [Tue, 18 Feb 2014 17:32:47 +0000 (10:32 -0700)]
added gmx_is_{single,double}_precision
* allows easy detection of the precision for cmake, autotools
without parsing the output of gmx_print_version_info (no
cross-compile support) or the output of strings command (unix
only)
* linking against libgmx with/out -DGMX_DOUBLE will lead to
unpredictable segfaults
Change-Id: I472f10ae374a1f42c94c55e156b53f8905bdf098
Roland Schulz [Mon, 17 Feb 2014 22:23:33 +0000 (17:23 -0500)]
Fix aligned store to unaligned memory
Also fixes that unaligned store was used when not necessary.
Change-Id: I44bb222a07ec0af65198667787b8673b3c6cd2e7
Mark Abraham [Fri, 31 Jan 2014 14:06:20 +0000 (15:06 +0100)]
avoid mdrun crash when rdtscp is not supported
When using rdtscp, mdrun now detects at runtime whether the CPU supports
this instruction and if this is not the case, it issues a fatal error
and instructs the user to recompile mdrun for the compute host. Note
that this will happen rarely, only when cross-compiling from a newer
host for a rather old one.
Additionally, when the user manually picks AVX, we also turn on RDTSCP
as all AVX-capable CPUs support it.
Also made CMake advanced cache option for GMX_USE_RDTSCP. This replaces
the previously hidden GMX_DISTRIBUTABLE_BUILD option.
Fixes #1428
Change-Id: I8bc884ef9ea8ea4661626b60490182ae2b302648
Mark Abraham [Tue, 10 Dec 2013 07:02:52 +0000 (18:02 +1100)]
Add integration tests for replica exchange
Only NVT and NPT tested so far.
Added new convenience class for a value-parameterized mdrun
integration-test fixture.
Change-Id: I25211277d6979909f67378dc578e3d388a8abeda
Rossen Apostolov [Wed, 12 Feb 2014 13:14:42 +0000 (14:14 +0100)]
Rename ffopen and ffclose to gmx_ff*.
There are name clashes with e.g. PathScale compiler.
Fixes #1250.
Change-Id: I3e5e85d9335f2695515483d59d440276823624f0
Roland Schulz [Mon, 17 Feb 2014 22:21:19 +0000 (17:21 -0500)]
Fix Reference Build
Rename (
e5e37f2c62c) broke reference build
Change-Id: Id98b61f2af2e4731f57a775ee9c4057baf76587a
Teemu Murtola [Sun, 9 Feb 2014 12:15:32 +0000 (14:15 +0200)]
Add documentation for using Doxygen
- Add a page that provides guidelines and examples for using Doxygen.
Hopefully, this would make simple examples easier to find, and help
people write correct comments from the beginning.
- Remove unnecessary \addtopublicapi and \addtolibraryapi.
Once the contents are there, it is easy to enhance the documentation by
adding content whenever noticing things that are tricky.
Change-Id: I319e30452004998df73c249f46570cb2254daa87
Berk Hess [Thu, 14 Nov 2013 08:59:50 +0000 (09:59 +0100)]
Implemented nbnxn LJ switch functions
The nbnxn C reference and SIMD kernels now support force and
potential switching, which are selected through vdw_modifier.
Both use analytical functions, since tables are expensive with SIMD.
Made the nbnxn kernel naming consistent with the group scheme.
Corresponding CUDA kernels will be added soon.
Dispersion correction always needs table generation, which had always
been a fortuitous side effect of something else, until this patch.
Change-Id: I52a1cda5ec7c3bdce6f3f2735ff7b13303ea5880
Mark Abraham [Tue, 18 Feb 2014 18:03:39 +0000 (19:03 +0100)]
Make some variable names less cryptic
No changes to function
Change-Id: I910abe7d8f42186bfd43706d3389fca2f0995cc6
Mark Abraham [Sun, 16 Feb 2014 22:36:56 +0000 (23:36 +0100)]
Fix outstanding references to "CPU acceleration"
Since we now use GMX_SIMD rather than GMX_CPU_ACCELERATION we
should be consistent.
Change-Id: I6a798870e5b89fc92bdbea18680e017101a2f704
Mark Abraham [Fri, 14 Feb 2014 17:29:34 +0000 (18:29 +0100)]
Fix harmless bug with combination of group and twin-range
The combination of
* group cut-off scheme,
* coulomb-modifier != none,
* rcoulomb > rlist (ie. twin-range), and
* rlistlong == -1
was being treated by the wrong clause in readir.c because the check
for "might be zero at cutoff" did not consider the modifier. The only
behavioural difference was issuing a warning that there will be no
buffering.
The modifier checks have been refactored to make it possible to check
the modifier.
Also removed some useless checks for EVDW_SWITCH
Change-Id: Ic64f3f48be17f6d87230bc4060a31e1fb4343973
Mark Abraham [Mon, 27 Jan 2014 19:43:04 +0000 (20:43 +0100)]
Extend TestFileManager
A test fixture may wish to change the output temporary directory, and
should be able to do so without perturbing the global state from which
the test TestFileManager object is initialized.
The only client of this feature (the setup code) is updated
accordingly.
Minor fixes to existing documentation.
Change-Id: I7ba7a82964c05f12de107710655b6320d42ae99c
Roland Schulz [Tue, 11 Feb 2014 21:30:59 +0000 (16:30 -0500)]
Enable more warnings
- pointer arithmetic on void*
- increase warning level for ICC from 2 (=Wall) to recommended 3.
Includes useful warnings such as:
- statement has no effect
- external/internal linkage conflict
- See gerrit for description of all disabled warnings.
Change-Id: I1c41119ed8bcee8dda8af0cbe2866b3e404a3992
Mark Abraham [Sun, 12 Jan 2014 22:23:11 +0000 (23:23 +0100)]
Fix test reference data accidentally broken
Change-Id: Ib622b126b0331c88acab924aff660211122f9d03
Mark Abraham [Sun, 12 Jan 2014 22:23:11 +0000 (23:23 +0100)]
Remove support for .g87 and .xyz I/O formats
Aspects of support for these have been broken since before GROMACS
4.5. That support is sometimes interactive, and thus a poor fit for
the way GROMACS supports other trajectory and coordinate formats.
Other file-format conversion options exist; GROMACS does not have to
be a Swiss-army knife.
Removed undocumented features of gmx helix that required .g87 support
for no particularly good reason.
Uncrustified files in this commit
Change-Id: Iad4845fd66ad4c55f9a31965d33ded745656db53
Carsten Kutzner [Thu, 13 Feb 2014 17:08:23 +0000 (18:08 +0100)]
Small enhancement of computational electrophysiology manual section.
Added a note about what can be done if large membrane undulations turn out
to be a problem in a Computational Electrophysiology setup.
Change-Id: I5f50b4026ea1a8b88368b10c39d2207070c4740c
Teemu Murtola [Sat, 15 Feb 2014 11:18:01 +0000 (13:18 +0200)]
Improve man page formatting slightly
- Don't add extra space whenever [TT]/[BB]/[II] is specified. This may
have been necessary 15 years ago, or has always been broken, but
currently it adds a lot of forced double spaces and leading spaces
onto man pages. The text is much more readable without it.
- Merge all command line options into a single OPTIONS section. Divide
it into groups using extra paragraphs in between. Could possibly be
formatted nicer, but that would require extra nroff trickery that we
might rather get rid of.
Related to #1437.
Change-Id: I4b27b79604a1830914cd5d14d6581918f33bb107
Teemu Murtola [Sun, 16 Feb 2014 04:48:39 +0000 (06:48 +0200)]
Fix default enum values in help output
nenum() does not return an enum value, but an index into the underlying
array, which means that the first element is 1 and not zero like the
code assumed. The help output was off-by-one, and could error out if
some environment variable was set that was affecting the help output.
Fixes #1438.
Change-Id: Iec2f05c9ced55868902069ce59bc4979a6879f86
Teemu Murtola [Sat, 15 Feb 2014 05:34:44 +0000 (07:34 +0200)]
Fix some broken links in HTML pages
Links to programs need to be prefixed with the relative path, as well as
"gmx-". Also added the "gmx " prefix to the displayed names.
Change-Id: I60bd0a58761a725804aa4c5b02619c3031ae7d26
Magnus Lundborg [Thu, 13 Feb 2014 13:49:39 +0000 (14:49 +0100)]
One more bump of the TNG version.
This is just to include the improved warning flag settings.
Change-Id: Idc1e0b43b963e8431cd943214c424012d09b6a4c