Aleksei Iupinov [Thu, 12 Apr 2018 14:31:43 +0000 (16:31 +0200)]
Make in-source OpenCL kernel path a per-module parameter
Change-Id: I19cd66b441fec6ceba74eb648bea3409b6521917
Mark Abraham [Fri, 2 Mar 2018 21:08:23 +0000 (22:08 +0100)]
Remove use of where(), if DEBUG etc.
The where() debugging function causes run-time branches in
release-mode builds, and replaces functionality for which one should
use a debugger.
Other such preprocessing clutter should also go away. Since we don't
check whether any of it compiles or produces useful data, we may as
well delete it.
This coincidentally simplifies the call sigatures of some
functionality, too.
Retained some useful essentialdynamics debug code, which is now always
compiled, and written from setup code only at debug level 2.
Refs #2423
Fixes #2122
Change-Id: I2c60c162734f4c2ec1d56153853a36d28e6a66ff
Aleksei Iupinov [Thu, 12 Apr 2018 12:42:48 +0000 (14:42 +0200)]
Remove CUDA_FUNC macros from many functions in PME GPU
As pme-gpu*.cpp files are not compiled during CPU builds anymore,
PME GPU functions that are not exposed to mdrun/tests do not need
to be hidden.
Change-Id: Id39c0bb6e586b508f5bbe8c7501ffd560d3cf14c
Aleksei Iupinov [Tue, 27 Mar 2018 11:18:51 +0000 (13:18 +0200)]
Stop PME GPU code from compiling on non-CUDA builds
On non-CUDA builds: all entrypoints into PME GPU code are
marked as stubs; pme-gpu.cpp and pme-gpu-internal.cpp are
excluded from compiling. Several common functions (such as
check for GPU support at the end of PME initialization)
are moved to pme.cpp, to still work with non-GPU builds.
Change-Id: Iaa9ea7b3e42665c60b788e76510a810457a69f20
Mark Abraham [Mon, 26 Feb 2018 16:47:24 +0000 (17:47 +0100)]
Reorganize energy evaluation for EM
Used aggregate initialized structs to simplify future refactoring.
Refs #1793, #2423
Change-Id: Iedb8bf3b4cb2d7f238c4b674ce459564a3f76a20
Viveca Lindahl [Tue, 10 Apr 2018 13:52:25 +0000 (15:52 +0200)]
Minor text edits and fix of broken link in user guide.
Link to domain decomposition section was broken. Also:
* removed a TODO that was already done
* reordered words in SIMD to make it a definition
Change-Id: I7943e4f59e2de82c2c6e94d15a6089a92f196132
Berk Hess [Wed, 14 Feb 2018 08:46:59 +0000 (09:46 +0100)]
Convert gmx_molblock_t to C++
Changed posres pointers in molblock to std vector.
Also moved the derived helper index data into a separate struct.
That is merged into this change to keep MPI broadcasting simple.
natoms_mol will be moved into the new struct in a child change.
Change-Id: I131fd14ab53cec98800c7e0731d3a92df3f93422
Mark Abraham [Wed, 28 Feb 2018 15:20:03 +0000 (16:20 +0100)]
Simplified handling of simulation runners
Argments to runners are now passed via struct members. That struct is
expected to change over the medium term, as most of its contents
should either be in a container of modules, or data with such a
module. As such, it does not make sense to do a lot of coding to
manage its contents in a sound way. The practical effects of using
this struct are almost the same as the old approach of passing a
forest of arguments - some variables are declared in a scope,
initialized with values, then used once.
The form is chosen so that
* the do_md() etc. functions do not need a large number of useless
textual changes now
* future changes to names and types of the former parameters do not
need to be edited in matching way in multiple places of code and
their doxygen
* we don't need to mark things gmx_unused anywhere
Useless unchecked return values have been removed.
Some excessive includes in integrator.h were removed, which
generates a few minor fixes elsewhere.
Refs #1793
Change-Id: I678598175192c9c68113fdd79fcee17f8e5c504e
Mark Abraham [Wed, 28 Feb 2018 14:54:26 +0000 (15:54 +0100)]
Create gromacs/mdrun module and move code there
Improving how we dispatch the work for do_md and friends requires that
they all compile as part of libgromacs (or all outside of it). The
code currently in src/programs/mdrun is there simply because it hasn't
been moved from the old src/kernel layout, rather than through any
clear design, so we should move it now that we have a reason.
Other code from src/programs/mdrun now needs to move into libgromacs
also, but we currently don't have a module for miscellanous mdrun
features, so we're adding to the abuse of mdlib for now.
Refs #1793
Refs #2423
Change-Id: I4f9ad5d0bf6585675472b49b2d5654f588b7214e
Szilárd Páll [Tue, 29 Mar 2016 20:20:22 +0000 (22:20 +0200)]
Refactor cj preload in the nonbonded OpenCL kernels
This change hides the implementation detail of preload/load operation in
inline functions configured based on the source flavor.
One of the main benefits is that this corrects the "nowarp" force and
pruning kernels removing a warp assumption (that was violated on
hardware with <32 wide wavefronts).
A step towards correctness on alternative GPU hardware, so
Refs #2030
Change-Id: If88d11df6613eadc9b59e5f03c34bac52df3514b
Mark Abraham [Fri, 26 Jan 2018 16:28:01 +0000 (17:28 +0100)]
Quiet checkpoint reading
If reading succeeds, it should write anything of interest to the log
file. If it does not, it should make a fatal error. Code here might
be run as part of mdrun (possibly in a multi-simulation), or grompp,
or compare, or dump, or eventually from an API caller, so we can't
make assumptions about what should be reported to a terminal.
Removed support for issuing a warning about restarting from very old
checkpoint formats, instead issuing a fatal error. Elsewhere, we
anyway observe that checkpoint restarts across major version
differences are not supported.
Change-Id: Id2b084bf0792a6992931609fb78d4a38a5f255fe
Pascal Merz [Mon, 19 Mar 2018 18:59:57 +0000 (12:59 -0600)]
Move checkNumberOfBondedInteractions
In task #1868, we will move the rerun functionality from do_md to
a separate function in a new file. As both do_md and the new
do_rerun need access to checkNumberOfBondedInteractions(), it needs
to be moved to a more central location.
Change-Id: Iae0ace11db2f83e5034fdfae21ccac3e7b9e34d1
Berk Hess [Tue, 13 Feb 2018 13:30:22 +0000 (14:30 +0100)]
Convert gmx_mtop_t to C++
gmx_mtop_t now uses std vectors for moltype and molblock.
gmx_mtop_t, gmx_moltype_t and gmx_molblock_t now have constructors
and destructors.
Changed declarations of gmx_mtop_t* to not use a pointer or to use
std::unique_ptr.
Several sub-structs still use C style pointers and will be converted
later.
Change-Id: Iee802e4382a0a389496bb9395389e5926918d0f8
Mark Abraham [Tue, 3 Apr 2018 09:44:09 +0000 (11:44 +0200)]
Merge "Merge branch release-2018"
Berk Hess [Fri, 23 Mar 2018 10:10:56 +0000 (11:10 +0100)]
Do not calculate COM for pull group 0
Pull group 0 is an absolute reference with coordinates 0,0,0,
so it is clearer to never compute its center of mass.
Refs #2143
Change-Id: Ifa974ab84afe303594c3b65e9567e0568bb151a9
Berk Hess [Fri, 23 Mar 2018 11:14:23 +0000 (12:14 +0100)]
Move file opening out of init_pull()
This is a step towards making a constructor for t_pull.
Change-Id: I24583fa32066503ccfbe0abaef6a412241974b12
Szilárd Páll [Mon, 26 Mar 2018 20:18:07 +0000 (22:18 +0200)]
Tweak/enable icc 18 post-submit tests
SIMD was disabled due to #2335
Change-Id: Ia55742958cfbdfb75e96c8d2b0ba5620a57a6f56
Magnus Lundborg [Thu, 1 Mar 2018 07:39:18 +0000 (08:39 +0100)]
Improved check of if only forces are computed in do_pairs()
Use the check introduced in I67bdb0cba9c6ec20b to see if the
SIMD version can be used.
Change-Id: I3dfd401aa1dc421757bf243809ac85e17a9062b6
Paul Bauer [Wed, 28 Mar 2018 11:33:02 +0000 (13:33 +0200)]
Fix cmake Sphinx detection for versions >1.6.4
The output from sphinx-build --version changed between version 1.6.4 and
1.7.x, leading to the cmake detection script to fail and reporting no
version at all. This fixes detection for both before and after the
change to the output format.
Change-Id: I34e50067f52f8127d137db28e9411420559d44cf
Szilárd Páll [Fri, 2 Mar 2018 21:23:06 +0000 (22:23 +0100)]
Switch CUDA 9.0/gcc 5 test to CUDA 9.1/gcc 6
Change-Id: If3e7277243c986ce96bf31de314edb1a255a44a6
Berk Hess [Tue, 13 Mar 2018 20:00:00 +0000 (21:00 +0100)]
Modularize pull_coord_work_t
Organized the 19 members of pull_coord_work_t into 5 sub-groups
and moved some into a new external PullCoordVectorForces struct.
Added a constructor and changed the pointer to coord in pull_t
to std::vector.
Also fixes #2446
Change-Id: I48d176ddbf3f5649f9072aacfb6bcb2d69785f91
Mark Abraham [Sun, 11 Mar 2018 18:11:31 +0000 (19:11 +0100)]
Free more memory in grompp and mdrun
This will be increasingly useful as we start adding more GoogleTest
cases that call grompp and mdrun repeatedly.
Refs #1793, #1868
Change-Id: Ic0faaa4c3dec2891d0cdc6166561f8ccd4391f44
Berk Hess [Wed, 14 Feb 2018 14:36:49 +0000 (15:36 +0100)]
Remove mols from gmx_mtop_t
The molecule atom begin/end information in gmx_mtop_t was derived
information. Now the mols struct is removed and the information is
derived when necessary.
Introduced BlockRanges, a C++ verion of t_block.
This is preparation for converison of gmx_mtop_t to C++.
Change-Id: I1717a101de82354a7f9cf14fd7f43c7227e554ef
Mark Abraham [Wed, 28 Mar 2018 11:19:26 +0000 (13:19 +0200)]
Merge branch release-2018
Some meaningful conflicts in init_multisim associated with multisim
fixes and clarifications in release-2018 combining with reorganization
in master branch.
Preserved the master-branch pre-submit matrix, since all branches have
separate changes that cope with gpu_id=2 no longer being valid.
Change-Id: Idefe4f1ec3bbbc76dfedf03a81abef8233328de2
Mark Abraham [Tue, 20 Mar 2018 07:39:37 +0000 (08:39 +0100)]
More gcc-8 fixes for POWER
Fixes #2421
Change-Id: I43c13df3a217d0f2154dca0ef215efd9cd27474a
Joe Jordan [Thu, 22 Mar 2018 08:59:25 +0000 (09:59 +0100)]
minor fix to flag strings for gmx potential
Change-Id: I2dadc8d74888dc2ebcd4fcf71e275c21fdd23cd6
Mark Abraham [Wed, 21 Mar 2018 15:04:44 +0000 (16:04 +0100)]
Updated to prepare for 2018.2 release
Change-Id: I1c72e8a3ac1b6b2e29b39c2d52bad5b87a884912
Viveca Lindahl [Thu, 22 Mar 2018 09:36:31 +0000 (10:36 +0100)]
Added cool quote.
Change-Id: I89d86d788f45f0d9124f5d00133709de83a73f38
David van der Spoel [Tue, 20 Mar 2018 21:04:42 +0000 (22:04 +0100)]
Renamed entropy.* to thermochemistry.*
This is a content-less patch to get the renaming history
correct in git.
Change-Id: I34f1d15f2e9136962b1718a60863b3b20a66fda5
Mark Abraham [Fri, 23 Feb 2018 13:20:17 +0000 (14:20 +0100)]
Version 2018.1
Change-Id: I3285c77fc557c1a6f34796c4bfb4843846b0d8a5
Mark Abraham [Wed, 21 Mar 2018 11:07:16 +0000 (12:07 +0100)]
Merge "Merge branch release-2016 into release-2018" into release-2018
Mark Abraham [Tue, 27 Feb 2018 16:10:15 +0000 (17:10 +0100)]
Update release notes for 2018.1
Change-Id: Ia60d523647e1915856ae3b00a0cb9eacd939fc56
Mark Abraham [Tue, 20 Mar 2018 22:17:51 +0000 (23:17 +0100)]
Merge branch release-2016 into release-2018
Conflict in pre-submit matrix, kept 2018 content that already
has the same logical change made.
Change-Id: I40b2db8d149388bed07471fcdb8082b0fc92b9b5
Vedran Miletić [Tue, 20 Mar 2018 11:27:45 +0000 (12:27 +0100)]
Document matrix_convert function in PBC
The brief documentation was in place, but the documentation of the
function parameters was missing. This patch adds it and also fixes
incorrect Doxygen syntax at pbc_dx_d.
Change-Id: Ic897911bf4b8866bb7c4dc3aea982493cc03874c
Berk Hess [Tue, 20 Mar 2018 12:19:01 +0000 (13:19 +0100)]
Fix gmx dump AWH formatting issue
A comma was missing in the list of AWH checkpoint entry string
which caused the naming to be messed up in the gmx dump output.
Change-Id: I85d0722236f3f9559c1117fdcee39766d04b424d
Mark Abraham [Tue, 20 Mar 2018 11:50:17 +0000 (12:50 +0100)]
Update use of gpu_id in Jenkins
We plan to remove one of the GPUs from the slave that had three, so we
need to update the testing configurations. The new configuration might
balance load a little better.
Refs #2408, #2410
Change-Id: I3b5d9c8702c021f0676ea3ca12f9e7f1d3317ed1
Berk Hess [Mon, 19 Mar 2018 13:33:47 +0000 (14:33 +0100)]
Disallow single simulation with multisim
To avoid many untested code paths, we, at least temporarily,
disable a single simulation with mdrun -multisim and -multidir.
Refs #2448
Change-Id: I9befdb677f3c77bcc6dc0c77abc3478f0b83ba8d
Mark Abraham [Tue, 20 Mar 2018 11:51:56 +0000 (12:51 +0100)]
Update use of gpu_id in Jenkins
We plan to remove one of the GPUs from the slave that had three, so we
need to update the testing configurations. The new configuration might
balance load a little better.
Refs #2408, #2410
Change-Id: Ie253ef07998cf45c5fc2e118a5394f0817e66f75
Mark Abraham [Sun, 18 Mar 2018 19:34:53 +0000 (20:34 +0100)]
Update use of gpu_id in Jenkins
We plan to remove one of the GPUs from the slave that had three, so we
need to update the testing configurations. The new configuration might
balance load a little better.
Refs #2408, #2410
Change-Id: Ia63cb701049a02ad86fdd9f9c03388cd11a14bcc
Aleksei Iupinov [Thu, 1 Mar 2018 22:32:03 +0000 (23:32 +0100)]
Clarify the mdrun GPU options docs
Refs #2359
Change-Id: If112077a59f875ffb1c689b2ac67dad7cdf2adb9
Mark Abraham [Tue, 6 Mar 2018 09:53:19 +0000 (10:53 +0100)]
Add helper function to make a vector from an ArrayRef
This syntactic sugar makes it easy to make a vector with a copy of the
contents.
Change-Id: Ib51c8a1d88aee7ca35fe1715d6f094b60e90e0b2
Mark Abraham [Fri, 16 Mar 2018 21:21:48 +0000 (22:21 +0100)]
Fix VSX SIMD with gcc 8
gcc 8 apparently generates different code for the former GROMACS
code, which seems buggy. See
https://bugzilla.redhat.com/show_bug.cgi?id=
1556989#c3
and https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84907
Fixes #2421
Change-Id: I31492cd582b785cdfb42e8b999a165a7339ce4be
Paul Bauer [Wed, 14 Mar 2018 09:15:42 +0000 (10:15 +0100)]
Only free topology when allocated
Fix trjconv so that the topology only gets free'd when it has been
previously allocated.
Fixes #2430
Change-Id: Idc9a4219d988e1794ef86f7cc79f0d00750394ca
Aleksei Iupinov [Thu, 1 Feb 2018 16:16:52 +0000 (17:16 +0100)]
Add a basic TPI sanity test
The floating point exceptions are for now toggled off
in the test body instead of mdrun itself.
Refs #2391
Change-Id: I41b584f4753b9b0e126256a94c1304f0a870aa73
Aleksei Iupinov [Fri, 16 Mar 2018 10:18:47 +0000 (11:18 +0100)]
Stop gmx_feenableexcept() from returning previous exception settings on Apple
This makes the function return value unambiguous.
Change-Id: I3e64e769b27fb0fa59b650d18b2fd099a9181025
David van der Spoel [Fri, 29 Dec 2017 20:56:09 +0000 (21:56 +0100)]
Split off entropy calculation to separate file.
The calculation of the quasiharmonic entropy or the one
through the Schlitter formula has some popularity. In
order to allow computation from a normal mode analysis
as well as from a PCA (essential dynamics analysis) the
code was split off into a separate file. Proper documentation
and five tests are provided as well. Minor cleanups
were done as well.
Change-Id: Ic5e4feab6a304946947d6e69be75c06b46aff75c
Berk Hess [Fri, 16 Mar 2018 10:00:41 +0000 (11:00 +0100)]
Correct PMF sign error in AWH bias sharing
Recent commit
229c8f63 was though to fix a sign error in
the calculation of the deconvolution of the AWH PMF when sharing
the bias over multiple simulations, but this sign was misplaced.
Fixes #2439
Change-Id: I67ec06ff4d832a857467b58b362f9f6f88671aa8
Berk Hess [Thu, 15 Mar 2018 17:09:55 +0000 (18:09 +0100)]
Fix recent mdrun signaling bug
Recent commit
e4deba12 accidentally negated the locality condition
for signaling checkpoint and termination in mdrun.
Fixes #2457
Change-Id: I1b7a4c26637766a44cad2581c9c9ec420aa4ab6e
Aleksei Iupinov [Tue, 23 Jan 2018 11:45:04 +0000 (12:45 +0100)]
Restore the lost mc_temp check in grompp
The expression for checking whether the reference temperature
is defined is moved into a helper function.
Change-Id: Ib331f52e8031a5e5d7c32234cfe936024c49ac28
Berk Hess [Thu, 15 Mar 2018 10:25:50 +0000 (11:25 +0100)]
Fix virial with external pull potential and MPI
When using an external pull potential with domain decomposition,
currently this is only used by AWH, the virial contribution would
be multiplied with the number of MPI ranks.
Fixes #2455
Change-Id: I61452fb9eca6e68fea3256cc25b39033a9c5d8e7
Berk Hess [Wed, 14 Mar 2018 10:59:53 +0000 (11:59 +0100)]
Improve inter-simulation signaling
Use a single boolean for controlling inter-simulation signaling.
This would have prevented issue #2447.
Introduce an inter-simulating signaling frequency. This both
simplifies the code and significantly reduces communiation overhead
for many simulations and/or many ranks per simulation.
Correct the terminimation step print with replica exchange.
Refs #2440 and #2447
Change-Id: I8ed50c5b11564d9a967bcacb15c4fcd3be05caea
Mark Abraham [Thu, 25 Jan 2018 14:47:35 +0000 (15:47 +0100)]
Make a persistent physical node communicator
Rather than make and re-make this communicator multiple times for for
almost all common run configurations, build one, and keep it in a
place distinct from commrec (which has too many responsibilities
already).
Note that the master thread-MPI rank builds a fake one initially,
because MPI is not yet initialized, then coordinates to build the real
one with the newly launched threads. However that communicator is not
actually used until after the execution paths have re-joined.
Simplifies IThreadAffinityAccess while retaining the ability to test
the affinity-setting logic without requiring specific attributes of
the node. Note that some of the tests already required that a
particular MPI environment exists.
Change-Id: Ie0800b07a65f9579ddc8af8f418cd14b85671de7
Berk Hess [Wed, 14 Mar 2018 08:32:23 +0000 (09:32 +0100)]
Fix mdrun signalling with AWH bias sharing
The recent commit
717fd5b8, which should have fixed issue #2440,
disabled checkpointing and termination by disabled signalling
between AWH runs that share a bias.
Fixes #2447
Refs #2440
Change-Id: Ic80ea07cfa1bde1b31fa9d660f00492b794b000f
Berk Hess [Mon, 12 Mar 2018 15:54:04 +0000 (16:54 +0100)]
Fix COM pulling with external potential with #ranks>32
With more than 32 PP-ranks, the pull code could use only a subset
of the PP-ranks. This change forces all ranks to do pulling when
external potentials are present (currently only used by AWH).
Fixes #2441
Change-Id: I8501024b7961600ec79f3707e239ddf25525aa79
Berk Hess [Tue, 13 Mar 2018 13:59:19 +0000 (14:59 +0100)]
Fix pull bug with geometry=direction-periodic
When doing COM pulling with geometry direction-periodic,
the reference location needs to be used to compute the current
distance. This reference was taken from the previous step instead
of the current. In practice this only caused issues at restarts
when the distance was more than half the box length.
Fixes #2446
Change-Id: Ie0c0341ad8c6ce1e78951270cab27ecaf95d65eb
Berk Hess [Tue, 13 Mar 2018 13:18:45 +0000 (14:18 +0100)]
Add AWH fixes to release notes
Change-Id: I17531da77495367b0e62b4e0b0e912bbd16659f9
Mark Abraham [Mon, 26 Feb 2018 16:47:02 +0000 (17:47 +0100)]
Add more const to uses of t_commrec and t_inputrec
Also removed some pointless struct keywords.
Refs #2423
Change-Id: Iacc008fb11a31646e798364e253de322c16ccaee
Szilárd Páll [Tue, 13 Mar 2018 02:24:58 +0000 (03:24 +0100)]
Require OpenCL 1.2
This commit bumps the requirement and removes OpenCL 1.1 API
compatibility support.
Refs #2438
Change-Id: Iaedf5fbdfd24a2d020983847815c2c21bac0f74c
Berk Hess [Mon, 12 Mar 2018 15:43:53 +0000 (16:43 +0100)]
Fix AWH ensemble checkpointing
AWH multi-runs with shared bias could terminate at different steps
leading to checkpointing from which continuation was impossible.
Fixes #2440.
Change-Id: I728c6aea9d030fcfad04efc6082309dafe3408bd
Berk Hess [Mon, 12 Mar 2018 20:36:07 +0000 (21:36 +0100)]
Fix AWH deconvolution when sharing a bias
When sharing an AWH bias over multiple simulations, the PMF output
was garbage due to a missing minus sign in the reduction.
Note that this only affected the PMF output, not the sampling.
Fixes #2439
Change-Id: I688cda045f5e757e58b392d64384705b3ac73a17
Mark Abraham [Wed, 21 Feb 2018 15:02:46 +0000 (16:02 +0100)]
Introduce constrain_shake
Also move various functions and fields of general constraint code and
data to the SHAKE code and data structures. Made some use of inputrec
const, to suit later changes.
Change-Id: I09b7a719b58f8f8192762431edc139674eaf23ae
Szilárd Páll [Mon, 12 Mar 2018 23:23:07 +0000 (00:23 +0100)]
Fix clang 6.0 with CUDA 9.0
This enables clang 6.0 CUDA build for sm_70 and works around an issue
which seems to affect sm_37 support.
Fixes #2443
Change-Id: Ib9c52e4a3a8ac1e05e851c8166eec4a6b5df4825
Mark Abraham [Tue, 13 Mar 2018 08:52:50 +0000 (09:52 +0100)]
Merge "Merge branch release-2018"
Szilárd Páll [Wed, 15 Nov 2017 19:48:50 +0000 (20:48 +0100)]
Avoid icc-specific -march=core-avx2 flag with gcc/clang
This change makes sure that the -march=core-avx2 icc-specific
optimization flag is only used for Intel compilers. It is not only
unnecessary for gcc/clang but also undocumented (and not proven to be
useful, actually potentually harmful with gcc <=6, measured 1-3% lower
perf.).
Change-Id: Ic2f05554ea6e460c53fa307fc1da3de1b6868e2f
Mark Abraham [Fri, 23 Feb 2018 12:20:56 +0000 (13:20 +0100)]
Test gcc-6 and cuda-9.0
Change-Id: Iba9436d9629a404290eb02d897b8d48f2a77b5ee
Berk Hess [Wed, 7 Mar 2018 12:48:42 +0000 (13:48 +0100)]
Correct pdb2gmx -missing option
The help text for the pdb2gmx -missing option only mentioned
missing atoms, but when atoms are missing also bonded interactions
will be missing. Now bonds are mentioned in the help text.
Since -missing allows for missing bonds, it now also allows for
dangling bonds at termini (which can actually be useful in special
cases).
Change-Id: I8d7215691ffc7dbb49f41964af9639c01355d775
Berk Hess [Wed, 7 Mar 2018 07:18:04 +0000 (08:18 +0100)]
Fix memory leak in PME
Now pme_spline_work is freed.
Change-Id: Ia959a85929b794c94a2befcce455151be1b6eca1
Szilárd Páll [Tue, 6 Mar 2018 18:10:55 +0000 (19:10 +0100)]
Extend "Finding out how to run mdrun better" section
Change-Id: I76c75504ef505e2547861015eb20d6d6b1f31383
Aleksei Iupinov [Fri, 2 Mar 2018 17:15:52 +0000 (18:15 +0100)]
Isolate PME GPU spline parameter indexing in inline functions
This makes the spread/gather kernels more readable and allows
to change the spline indexing scheme much easier in
the future. Performance should not be affected.
More TODO regarding the spline indexing scheme is marked.
Change-Id: If735cccf2ce82f46b483c9ada6f309425c51f67e
Berk Hess [Fri, 23 Feb 2018 12:17:10 +0000 (13:17 +0100)]
Convert t_filenm to C++
This change is only refactoring.
Struct members opt and fn are now const char * const.
Replaced nfiles and fns by a vector of strings.
Removed, now unnecessary, copy and delete functions.
Change-Id: I7122f06e7264ad378022321b4c61acf1e3caabd6
Mark Abraham [Tue, 6 Mar 2018 14:07:45 +0000 (15:07 +0100)]
Merge branch release-2018
Minor conflicts from separation of multisim from commrec,
and code movement in constraints code.
Change-Id: Id1fea8a7de0726e34eb2d43b2420b4753631f303
Mark Abraham [Fri, 2 Mar 2018 10:22:42 +0000 (11:22 +0100)]
Fixed dvdlambda for SHAKE + FE
Garbage values have been accumulated to dvdlambda whenever SHAKE was
used.
Refactoring in
513813b514 removed the lam variable. Previously, it was
used as a temporary iteration variable over SHAKE blocks, but that
refactoring to use the main variable changed the logic of the
following loop.
Fixes #2434
Change-Id: Ia9642f50358ab31ec98f8317f7a1dcda8622a9f1
Berk Hess [Mon, 5 Mar 2018 15:34:38 +0000 (16:34 +0100)]
Make multi-atom TPI reproducible
Test particle insertion of molecules of more than one atom called
a rotate function with multiple random calls in the arguments. This
lead to non-reproducible results, at least when switching compiler,
since the language does not define the argument evaluation order.
Note that this issue did not affect physical correctness.
Change-Id: I9ab07566878f381f11284d813d9077fa90e68866
Viveca Lindahl [Mon, 5 Mar 2018 18:31:20 +0000 (19:31 +0100)]
Add checks for non-null log file pointer in AWH
Mdp-option 'awh1-equilibrate-histogram=yes' and certain run setups could
lead to a segmentation fault due to fplog being null.
Fixes #2436
Change-Id: I0110325a8624c9d434af5eeeccace29814d40f2d
Berk Hess [Wed, 28 Feb 2018 13:09:33 +0000 (14:09 +0100)]
Work around gcc bug with group kernel tables
The LJ-only interaction tables for the group kernels were incorrectly
copied with the combination of gcc-6, AVX and -O3.
Since this bug replaced -1/(6 r^6), 1/(12 r^-12) by 1/r,-1/(6 r^6)
for odd table indices, this could not cause silent errors.
Fixes #2424
Change-Id: I021d4f2fe81bd635e6f02686118d661af8c444a6
Aleksei Iupinov [Tue, 27 Feb 2018 16:55:31 +0000 (17:55 +0100)]
Stop using custom FindOpenCL.cmake, per TODO
It was originally taken from cmake 3.1.0 and modified to produce
all-uppercase variable names (with OPENCL_ instead of OpenCL_),
so the surrounding cmake code is now changed to expect exact-case names.
Another (undocumented) modification was that it took into account
CUDA_HOME env.var. when searching for OpenCL, while standard cmake
only looks at seemingly less useful CUDA_PATH. This useful trick is
preserved by temporarily appending $ENV{CUDA_HOME} to CMAKE_PREFIX_PATH.
Change-Id: I5ca60f88e657b064fb6f98b268c2c48e10f00986
Aleksei Iupinov [Thu, 1 Mar 2018 10:31:30 +0000 (11:31 +0100)]
Correct the nbnxn_gpu_compile_kernels() docs
Change-Id: Ieb9d3fd8a55754f6fe04d66b49c1cead1cd4b7c8
Berk Hess [Fri, 23 Feb 2018 13:01:15 +0000 (14:01 +0100)]
Leave nullptr opt in t_filenm
File options that use the default extension of the file type by
setting opt=nullptr in t_filenm are no longer updated by
parse_common_args. Instead the (default) option name is looked up
based on the file type in the utilty functions.
This is preparation for converting t_filenm to C++.
Change-Id: I907c1e88f91ab2e524a72c03c27fd67ad641bc02
Szilárd Páll [Fri, 2 Mar 2018 16:58:41 +0000 (17:58 +0100)]
Add to the contributor section
Change-Id: I1642f1ca247cf0e11781af4848b74e3d0ebd5082
Berk Hess [Fri, 2 Mar 2018 22:45:29 +0000 (23:45 +0100)]
Fix cppcheck warning
Change-Id: Ib188b6da439d071a615158c4889b1a0f6edbda08
Aleksei Iupinov [Thu, 1 Mar 2018 13:14:40 +0000 (14:14 +0100)]
Make TPI use double precision exp
This makes borderline results reproducible with different
compilers (e.g. with MSVC).
Change-Id: I7b7c50c4f857b4fe2966f49a6ef685eab6598ff6
Berk Hess [Thu, 1 Mar 2018 20:58:32 +0000 (21:58 +0100)]
Fix AWH bias-sharing with parallel simulation
Sharing the AWH bias over multiple simulations only worked when
each simulation was running on a single MPI rank. Now parallel
simulations also work.
Fixes #2433.
Change-Id: I71f9069a31b033151c772aac84c9912d91b213a1
Berk Hess [Fri, 23 Feb 2018 14:22:50 +0000 (15:22 +0100)]
Detect AMD SMT topology
On AMD Zen the cpuinfo code detected hyperthreading but put all
threads on different cores in the topology. Now the correct
topology is detected using extended APIC.
Also disabled topology detection for non-AMD, non-x2APIC x86.
Fixes #2388
Change-Id: I194f3e09e669c20d1d62355a36be062e6cce264e
Berk Hess [Thu, 1 Mar 2018 08:30:55 +0000 (09:30 +0100)]
Fix mdrun multisim with separate PME ranks
When running multiple simulations using separate PME ranks,
mdrun would call MPI_Barrier with MPI_COMM_NULL on some ranks.
Note: no new release note needed, since this fixes a fix.
Refs #2403
Fixes #2431
Fixes #2432
Change-Id: I7fe96a15b5030dea3e093d12850f89f00ccc9f48
Mark Abraham [Wed, 21 Feb 2018 11:42:27 +0000 (12:42 +0100)]
Moved and removed constraint code
No functionality changes here.
Removed code that was formerly called by the now-removed
combination of Andersen T coupling and constraints.
Removed crattle declaration from header, and moved its definition to
before the point of use, as normal for functions with file static
linkage.
Moved constr_r_max and support functions to constraintrange.cpp,
because conceptually the functionality is independent of which
algorithm implements the constraints. The current implementation only
works with LINCS, but that's because SHAKE and DD can't be used
together.
Refs #2423
Change-Id: I9e4fa1a4bb57118ca17c55f312103439b38190e3
Mark Abraham [Fri, 16 Feb 2018 17:12:54 +0000 (18:12 +0100)]
Reorganized constraint code
Renamed constraints files after just their algorithm, not the original
programming language, which matches the test file that they
have. Separated header files for each algorithm with matching
names. Removed extern C. Minimized header dependencies. Removed
excessive use of struct keyword.
Removed some header declarations of functions declared static.
Refs #2423
Change-Id: If60ab052ae73ad981fa031c2db6b3e206b380f4c
Aleksei Iupinov [Tue, 27 Feb 2018 13:08:17 +0000 (14:08 +0100)]
Merge branch 'release-2018'
Conflict: multisim barrier handling in src/programs/mdrun/runner.cpp
Amended the test introduced in I16a3705a0d59a to exclude the implicit
solvent output which was disabled in Ib241555ff3d8e60.
Change-Id: Idf382c9c7ba92439097fcc9008cb9cc813daaab7
Mark Abraham [Fri, 23 Feb 2018 14:15:46 +0000 (15:15 +0100)]
Fix GPU utils test
We should only attempt device-side work when we have compatible GPUs.
Fixes #2405 for CUDA
Change-Id: Idc28c7e89a5f08ee1d19943e3663385f2b23ff44
Paul Bauer [Fri, 23 Feb 2018 14:33:33 +0000 (15:33 +0100)]
Change TNG file opening
Change to TNG file opening to return a pointer the same way as open_trx
does. This is needed in a follow-up change so that no memory is lost
there.
Also made sure it is possible to open TNG files if no input file is
specified, or if the input file is not of TNG type.
Change-Id: I47cf1625815e640b399ed5a0ed5bbc9a3ead6e1b
Mark Abraham [Fri, 23 Feb 2018 10:45:56 +0000 (11:45 +0100)]
Permit define .mdp field to assign values
It is intended that the "define" .mdp field can set preprocessor
variables to values, but this got broken in some clean up. Added a
test to cover it.
Fixes #2392
Change-Id: I16a3705a0d59a912a8eb7bfad0b472f12d2dc377
Aleksei Iupinov [Mon, 19 Feb 2018 15:20:54 +0000 (16:20 +0100)]
Advice users on infinite output of TPI
Fixes #2416
Change-Id: I6202154bc2a24377d036e5c115f2c90b1c678dda
Mark Abraham [Tue, 9 Jan 2018 13:41:38 +0000 (14:41 +0100)]
Organize for recording 2016-era release notes and add them
This hierarchical organization will make it easy for us to present to
users that there were two maintained release series at the time of
writing, while making available links to the releases notes of
unmaintained series, and having a straightforward path for migrating
the 2016 folder to the older folder and updating the appropriate
links.
In master branch, we will have stub .rst files in
docs/release-notes/major/ that we can fill out as functionality
changes, and movev to docs/release-notes/2019/major when we
make the release-2019 branch.
Changes to release notes in release branches will merge naturally.
Later, we can make sure that tarballs can have a useful plain-text
version built into them via sphinx also.
Change-Id: Ie35f90a616cb89727f5a9600389fdf228ecd03cb
Berk Hess [Wed, 21 Feb 2018 22:02:32 +0000 (23:02 +0100)]
Fix md integrator with Nose-Hoover coupling
When applying NH T-coupling at an MD step and no PR P-coupling,
the md integrator could apply pressure scaling with an uninitialized
or outdated PR scaling matrix.
Fixes #2418
Change-Id: I835db72776e7782ac044807961bb899e4f8c6c7b
Berk Hess [Thu, 22 Feb 2018 12:40:45 +0000 (13:40 +0100)]
Simplify mdrun tMPI thread spawning
During spawning of the tMPI threads in mdrunner, the mdrunner object
and filename pointer were copied twice. Now they are only copied once
and a const object/pointer is passed around.
Also made thread-MPI take a const argument to enable this and to
ensure correctness, as thread-MPI makes a copy of the argument pointer
for every thread which could cause problems without const.
Added an explicit MPI_Barrier after spawning, with documentation,
in preparation of removing the implicit barrier in MPI_Comm_split
in reinitialize_commrec_for_this_thread().
Change-Id: I31af8a8ff72aec28366c4aa20b357ed96bf95087
Mark Abraham [Wed, 21 Feb 2018 09:24:17 +0000 (10:24 +0100)]
Enhanced docs of pimpl pointer helper
There are constraints on where the destructor can be defined, and
doing it wrongly can provoke confusing compiler errors.
Change-Id: I241d8713d5e9a74f17698397ef871784db9a7292
Szilárd Páll [Wed, 21 Feb 2018 00:12:59 +0000 (01:12 +0100)]
Switch pre-submit CUDA 8.0 test to 9.0
Noted in TODO to add a CUDA 9.1 + gcc 6 config.
Change-Id: I1e9e1a0bcd99237167335f475d65791f6059dbb3
Aleksei Iupinov [Tue, 20 Feb 2018 12:50:06 +0000 (13:50 +0100)]
Remove unused OpenCL debug buffer
Change-Id: I8f5f8a7f1b099f7cf7b058213dedb4d35e35c50c
Szilárd Páll [Fri, 16 Feb 2018 18:48:46 +0000 (19:48 +0100)]
Avoid aborting mdrun when GPU sanity check detects errors
A release assertion was added which assumed that the GPU
compatibility/sanity checks return with a clean CUDA API state.
Consequently, any run that encountered a non-success return value from
the CUDA API would abort the run instead of continuing the run without
using the GPU in question.
This change adds code to handle and issue a note on the error
encountered as well as ensures that the CUDA API error state cleared
at the return of the GPU detection.
Fixes #2415
Change-Id: I5d7ed59ef8e4052a75b51c9a526b8dcb465ff611
Mark Abraham [Wed, 14 Feb 2018 13:26:13 +0000 (14:26 +0100)]
Remove cycle suppression
The dependency was recorded the wrong way around. t_state needs to
have a GPU-aware allocator, so needs to depend on GPU utility
functionality.
Change-Id: I4f67962a564cb6abdd83ba081e08a35fe27af6c3
Mark Abraham [Fri, 26 Jan 2018 16:22:04 +0000 (17:22 +0100)]
Disallowed change of ensemble for checkpoint restart
Change-Id: I63a8330043fddffc9cfc98203f60d71497e6c2f7