Fixes where mdrun could behave incorrectly¶
Disabled PME tuning with the group scheme¶
PME tuning with the group cut-off scheme did not work correctly.
Interactions between charge-group pairs at distances between rlist
and rcoulomb can go missing. The group scheme is deprecated, and
this issue would require considerable effort to fix and test, so we
have simply disabled PME tuning with the group scheme.
Fixed value of Ewald shift¶
In all the Ewald short-ranged kernel flavours, the value of the potential at the cutoff is subtracted from the potential at the actual distance, which was done incorrectly (failing to divide the shift value by cutoff distance). Fortunately, the value of that distance is often close to 1, and the inconsistent shifts often cancel in practice, and energy differences computed on neighbour lists of the same size will have the error cancel. The difference doesn’t even show up in the regressiontests, but would if we had a unit test of a single interaction.
Fixed orientation restraint reference¶
The resetting of the COM of the molecule with orientation restraints for fitting to the reference structure was done with the COM of the reference structure instead of the instantaneous structure. This does not affect the restraining (unless ensemble averaging is used), only the printed orientation tensor.
Fixed bugs with setup for orientation restraints¶
The orientation restraint initialization got moved to before the initialization of the domain decomposition, which made the check for domain decomposition fail. Also fixed orientation restraints not working with the whole system as fitting group.
Worked around missing OpenMP implementation in orientation restraints¶
The orientation restraint code is not aware of OpenMP threads and uses some global information. By only running it on the master rank, results are now independent of number of threads used.
Enable group-scheme SIMD kernels on recent AVX extensions¶
The group-scheme code only runs using the feature set of AVX_256, but that is supported on the more recent hardware, so we should have the group scheme run with the maximum suitable SIMD. With previous releases, building AVX_256 binaries was required for best performance with the (deprecated) group scheme.
Fixed COM pull force with SD¶
The reported COM pull force when using the SD integrator was random only. Now the pull force is summed over the systematic and random SD update components. A better solution is to not add the random force at all, but such a change should not be done in a release branch.
Fix PBC bugs in the swap code¶
Fixed flat-bottomed position restraints with multiple ranks¶
Reallocation was never done for flat-bottomed restraints, during domain decomposition, so the indexing could go out of range, leading to segfaults.
Fixed null pointer print in DD¶
Fixed a (rather harmless) print of a null pointer string during
DD initialization. This would only show up with gmx mdrun -dlb yes.
Improved the “files not present” error message¶
It’s possible to use gmx mdrun -deffnm in restarts even if it
wasn’t used in the initial simulation. This can lead to absurd
situations such as:
- Expected output files not present or named differently:
- pullx.xvg pullf.xvg
where pullx.xvg and pullf.xvg are present and named exactly as
listed, but GROMACS expects them to be named as -deffnm requested.
The improved error message suggest to the user to check for that possibility.
Issue 942 (partial workaround)
Fixed LJ-PME + switch grompp error¶
An error call was missing in grompp when LJ-PME was requested in combination with a force or potential switch modifier.
Fixed unused SIMD PME table kernel¶
The Verlet-scheme 2xNN PME kernel with tabulated correction had several issues. This kernel flavor could only be selected manually by setting an environment variable, so no user simulations should be affected.
Fixed bugs in most double-precision Simd4 implementations¶
The double precision version of reduce() and dotProduct() returned a float with AVX2_256, AVX_256, AVX_128_FMA, AVX_512, MIC and IBM_QPX. Only reduce() is used in double, in the PME force gather, and the difference is small.
Avoid inf in SIMD double sqrt()¶
Arguments > 0 and < float_min to double precision SIMD sqrt() would produce inf on many SIMD architectures. Now sqrt() will return 0 for arguments in this range, which is not fully correct, but should be unproblematic.
Fix NVML error messages¶
These should refer to the API calls that failed, e.g. when users lack permissions to change clocks.
Fixed IMD interface malfunctions¶
Fixed initial temperature reporting¶
When continuing a simulation from a checkpoint, mdrun could report
double the intial temperature when nstcalcenergy=1 or nsttcoupl=1.
Note that this only affected reporting, the actual velocities were
correct.
Now the initial temperature is no longer reported for continuation
runs, since at continuation there is no “initial” temperature.
Fix exception in SIMD LJ PME solve¶
Clear SIMD padding elements in solve helper arrays to avoid, otherwise harmles, fp overflow exceptions.
Fixes for gmx tools¶
Fixed memory access issues in gmx solvate¶
- There was out-of-bounds access if
- the solvent configuration was given as a .pdb file, or
- there was more than one type of residue in the solvent (which triggered sorting).
Also fix a memory leak in the sorting routine.
Should fix crashes mentioned in Issue 2148
Fixed a consistency check in gmx make_edi for flooding¶
If one sets up a flooding .edi input file with gmx make_edi,
the code should check that one does not use of the last 6 eigenvectors
of the covariance matrix, which correspond to the rotational and
translational degrees of freedom.
The check that was in the code erroneously checked against the
number of eigenvalues neig that was stored in the .xvg file,
not against the total number of eigenvectors which depends on
the number of atoms nav used in gmx covar. Thus the original
check would always fail if the .xvg eigenvalue file contained
1-6 values only.
Supported quiet trajectory-handling I/O¶
Permits GMX_TRAJECTORY_IO_VERBOSITY=0 to be set to keep frame-reading
code quiet, which is convenient for tools using libgromacs.
Improved documentation¶
Migrated much content from the wiki to the user guide¶
This includes
* expanding the “Performance” section,
* reworking extending simulations, doing restarts and reproducibility,
* adding documentation for mdp option simulation-part.
* adding documentation for issues relating to floating-point arithmetic
* adding documentation for run-time errors
Corrected the PDF manual to reflect that all tools are called gmx <tool>¶
There were still a few occurrences of the old-style g_tool naming,
this patch removes. Deliberately left g_membed as is, because there
was never a gmx membed, but instead it got incorporated into
gmx mdrun.
Clarified gmx editconf help text¶
It is possible that users can confuse -c with -center so this
patch makes it clear that -center doesn’t do anything unless the
user really wants to shift the center of the system away from the
middle of the box.
Added missing .mdp file documentation for the enforced rotation module¶
Fixed parameter description for dihedral_restraints¶
The force-constant parameter for dihedral_restraints was not documented in the table of interaction types.
Replaced instance of “group” by “coord” in pull .mdp documentation¶
Portability enhancements¶
Supported CUDA 9/Volta for nonbonded kernels¶
Implemented production-quality support for Volta GPUs and CUDA 9.
The code was adapted to support changes to the nature of warp synchrony, without disturbing support for older GPUs and/or CUDA. Further improvements may be seen (e.g. in the 2017 release).
Really enabled AVX512 in the GROMACS-managed build of FFTW¶
An earlier attempt to enable AVX512 on GCC 4.9 or newer and Clang 3.9 or newer was wrongly implemented. Now this works on all compilers we officially support (MSVC, GCC, clang, ICC).
Fixed aspects for compiling and running on Solaris¶
Fixed AVX512F compiler flags¶
Avoid using the MIC code generation flags for the Xeon code path.
Fixed compiler flags for using MKL¶
Fixes compilation issues with ARM SIMD¶
ARM_NEON has never supported double precision SIMD, so disabled it with GROMACS double-precision build.
The maskzR* functions used the wrong argument order in the debug-mode pre-masking (and sometimes in a typo-ed syntax).
In the shift operators, the clang-based compilers (including the armclang v6 compiler series) seem to check that the required immediate integer argument is given before inlining the call to the operator function. The inlining seems to permit gcc to recognize that the callers always use an immediate. In theory, the new code might generate code that runs a trifle slower, but we don’t use it at the moment and the cost might be negligible if other effects dominate performance.