Release Notes

1.7.11

• Fixed mistake in last fix.

1.7.10

• Added runtime check of MPI dimensions to aid debugging.

v1.7.9

• Fixed issue with epsilon calculation code using MPI commands in serial build mode.

v1.7.8

• Reorganisation of some build files and updated copyright.
• Addressed issue with an MPI seg fault on some Windows and Linux systems.

v1.7.7

• Fixed issue resulting in a crash and out of range exception when a single process was used in MPI builds in combination with grid refinement. Issue #4.

v1.7.6

• Updated makefile to resolve issue #3.

v1.7.5

• Fixed build issue due to header conflicts resolving issue #2.

v1.7.4

• Fixed issue with momentum exchange giving negative forces when positive expected.

v1.7.3

• Updated makefile to enable OMP directives.
• Reorganised the FEM implementation to make it easier to add new types of elements.

v1.7.2

• Fix for the issue in 2D where sub-grids had a non-zero size and triggered the initialisation warnings.
• Fixed issue with point cloud reader not communicating across sub-grids on different processes.

v1.7.1

• Added a debug logging function coloured blue in the log file filter.
• Added basic OpenMP parallel for loop for future development.
• Added protection for using corrected sizes without parentheses and better checking.
• Fixed issue with placement of objects on halos incorrectly when periodically wrapped.

v1.7.0

• Updated compilation instructions.
• Bug fixes for IBM.
• Added facility to restart simulation using a different time step along with python utility.
• Updated logging formatting in the main method to match new standard.
• Fixed issue with MPI gather error when using parallel IBM.
• Fixed issue with buried links when using regularised BC which was causing density fluctuations at the corners.
• Fixed issue with incorrect site rejection action for momex.
• Fixed bug when activating point cloud debugging features in serial.
• Fixed issue with incorrect message passing when using a single core in a given dimension.
• Fixed issue with applying regularised BC on corner nodes.
• Fixed issue with incorrect writing of forces for IBM bodies in parallel.
• Updated terminology for output timers.
• Added utility for determining out of domain positions to fix issue with support nodes leaving the domain instead of wrapping.

v1.6.1

• Fixed bug in MPI manager initialisation that caused an overrun exception.

v1.6.0

• Added kernel performance improvements which are especially effective when using BB objects. As a result, a new filled box point cloud is available to add to the domain to effectively switch off unnecessary fluid sites by turning them solid.
• Changed BC setup to make it more flexible in terms of enumerations. Also added slip BC.
• Added ability to ramp up the Reynolds number from the beginning of the simulation to mitigate crashes due to initial transients.
• Parallel IBM implemented.
• Pressure boundary condition added (regularised only).
• Efficiencies introduced for ramping functions.
• Moved setBCPrecedence method from utilities to GridObj and renamed.
• Reworked inlet initialisation.
• Reworked grid initialisation to ensure consistency with declaration of halos.
• Added helper for diagnosing HDF5 hangs.
• Fixed bug with HDF5 hanging when filespaces disagree.
• Added missing condition for applying extrapolation on a recv layer outlet.
• Periodic wrapping of sub-grids is now available by specifying the start and end point to be the same.
• Added a reporting mode to MPI builds to use the smart decomposition and load balancing methods to estimate config performance. Need to compile with the L_MPI_TOPOLOGY_REPORT flag on to switch to the mode.
• Added capability to have objects placed over grid boundaries. Need to specify grid level as -1 in the geometry config file. Only works for BBB for now.
• Added KBC to optimised kernel.
• Fixed missing pointers in restart method that was causing seg faults on load.

v1.5.0

• Updated the HDF5 output and the merge tool to include MPI block labels to make decomposition more apparent.
• Fixed an untracked bug in the HDF5 writer which would cause a hang on certain MPI/sub-grid combos.
• Added some colourmaps for paraview.
• Added script for creating symmetric aerofoil point clouds in 2D and 3D with ends.
• Added MPI buffer size debugging and cross-checking to prevent hidden hangs in future.
• Added ability of LUMA to predict completion time of simulation.
• Merge tool now cuts out solid sites to save memory. Also now uses the CleanToGrid filter from paraview -- extra files supplied with the merge tool. VTU XMLs are written instead of the legacy VTK too. Merge tool incremented to v0.3.0.
• Fixed an issue with the load balancing file output which was given incorrect ops counts.
• Fixed a bug in the smart decomposition method which would cause a seg fault when used on some 3D problems.
• BGK implemented. Set constant in definitions file.
• Upgraded the merge tool to also perform the task of writing out a sorted velocity field from the HDF5 files. Use CLA .
• Reinstated RBC for a fixed velocity.
• Upgraded geometry file and PC reader so objects can easily be centred in the domain if required through keywords in the config.
• Changed dt suggest code so it gives the upper limit for dt instead.
• More intelligent algorithm implemented for constructing MPI block sizes by driving variable boundaries towards most loaded process.

v1.4.2

• Optimisation of PCpts reader so it scales as O(n) rather than O(n^2).
• Construction of GridObjs converted to heap allocations. Nested sub-grids are stored as pointers rather than objects.
• Destructors modified to free memory on parent deletion.
• Added auto sub-grid generation capability. User only needs to specify L1 and then padding and the rest is taken care of by the grid manager. Can be disabled in the definitions file, causing the grid manager to use the values specified in the usual way.

v1.4.1

• Added warning macro and improved some of the documentation in the GridUnits.
• Added debugging capability for Momentum Exchange as well as associated MATLAB tool.
• BFL has been added to the optimised kernel.
• Intersection and Q compute algorithms have been rewritten.
• BFL and IB body classes restructured to reduce duplication.
• Momentum exchange has been upgraded to work with BFL bodies as well as BBB.

v1.4.0

• Wholesale changes to coordinate systems and hence changes in the initialisation machinery.
• Additional bug fixes in labelling and coarse limit interpretation.
• Fixed rare but major bug that occurs with HDF writable region definition when only a corner of a grid is built on a rank exclusively on a halo region. Changed implementation to fix.
• Improvements to initialisation clarity and debug messaging.
• Old boundary condition methods marked as DEPRECATED.
• IBBody constructor for PC added to move away from the "make_body()" wrappers which are messy.
• Created new Enumerations header and added more headers to the stdafx.h header.
• Extended the error function to create an info function as well for more consistent logging.
• User input except L_RHOIN is now expected in dimensionless units.
• Documentation moved to VM @ 130.88.97.222.
• New GEOMETRY CONFIG runtime infrastructure added for adding multiple bodies of different types into a simulation. Replaces the old compile-time mechanisms.
• BBB objects are now able to be read from the new geometry configuration file.
• Probe output method now generalised so it will search for a probe position on the finest grid first and then ascend the hierarchy until point is found.
• The initial velocitiy field can now be read from a file by defining L_INIT_VELOCITY_FROM_FILE in definitions.h
• Fixed bug in 3D serial cases with Grid Manager initialisation that caused the wrong writable region to be set and corrupted HDF5 data.
• User input except L_RHOIN is now expected in dimensionless units. L_RHOIN is expected in LBM units. NOTE: The simulation time and the interval at which the results are written is still in number of time steps (i.e. in LBM units)
• Improved inheritence relationships for the Body class to reduce code duplication.
• Added new Grid Manager class to improve handling of grid information. MPI manager now only should be instantiated when using MPI.
• Fixed issue with compiling with probe output turned on.
• Relaxed restriction requiring uniform decomposition by using a slice generator to decompose.
• Allowed the embedding of sub-grids in walls once again after coordinate system and initialisation change.

v1.3.0

• Reinstated force calculation and included a few other optimisations.
• Added new refinement guide spreadsheet.
• Optimisation for hdf2vtk (h5mgm) conversion tool plus handles missing datasets and has quiet option. Now v0.2.2.
• Added a filled cylinder point cloud to address non-physical transients.
• Added macro to allow marking of functions as deprecated.
• Fixed an issue with the scaling on the cloud reader -- not documented on the tracker.
• Standardised macro definitions.
• Fixed an unlogged issue with the refinement ratio not being initialised correctly.
• H5MGM tool updated to give more information to the user and fix an compilation issue with GNU associated with labels.
• Memory footprint of H5MGM has now been dramatically reduced by removing unnecessary mesh points previously stored.
• Fixed issue with HDF5 writer when TL is completely within a halo region and data is over-trimmed. New implementation for halo trimming.
• Load balance indicator and file write-out added.
• BGK Smagorinsky model added to the optimised kernel.
• Fixed typo in pre-processor directive that prevented the use of the 3D models.
• Replaced use of halo descriptor data with simpler solution for sub-grid striding in slab copies to fix write-out inconsistencies.
• Restart file mechanism changed so that a single file is written out per rank to speed-up the process of restarting large meshes with a large number of cores.
• This removes the capability to restart onto a different topology.
• Compression of old LUMA algorithm into a single loop and marked older kernel and elements as deprecated. 3X speed-up using optimised kernel. Can be selected using the L_USE_OPTIMISED_KERNEL macro. Will reinstate other functionality as required -- raise a ticket.
• Predictor-corrector function deprecated in favour of new kernel function using Li method for IBM.
• Added FGA writer for export of field data to UE4.
• Fix in place for the momentum exchange to resolve issue with scaling and conflict with outlet BC.
• Fixed probe crash when number of probes 1 in a given direction.
• Reinstated D3Q19 velocity model.
• Fixed issue with probe output path.

v1.2.0

• Compatibilty between GNU and MSVC compilers has been improved through header/forward declaration rearrangement.
• IS_NAN() function switched to macro definition.
• LBM_multi routine has been compacted to improve readability.
• Restart functionality has been improved allowing IBM on sub-grids.
• Expensive support node calculation has been removed.
• Gravity force scaling across sub-grids has been fixed.
• Wall thickness can be arbitrarily set for front, back, top and bottom walls.
• Changed the terminal output to be every L_out_every time steps.
• Removed redundant indexToPosition() method.
• Resetting the force vectors is now a separate routine to get rid of the flag for forceGrid().
• LBM_forcegrid method renamed to LBM_forceGrid to match convention.
• IBM_flag renamed to ibmFlag to match convention.
• Removed repeatFlag as unnecessary.
• Refactored the OM methods to use iBody owner rather than passing a GridObj pointer about.
• Now need VTK source files to build the H5MGM tool.
• Fixed erroneous error when using custom MPI sizes.
• Fixed GNU compiler warnings when using HDF5.
• Test Suite capability added.
• Fixed issue with BFL reader not computing Q values correctly.
• Fixed MPI crash on exit on Linux.
• Model-independent version of extrapolation exit on right-hand wall.
• Added grid sizes as GridObj property.
• Some Doxygen documentation added and will involve throughout the releases to contain more information.
• Force calculation for BB objects.
• TL automatic exclusion and reuse of C utility rather than C++ API for HDF5. H5MGM tool v0.2 included in /tools and writes VTK.
• Fixed bug which allowed serial calls of a utility only compatible with MPI builds; also fixes for writing out IBB to restart correctly.
• Patch to fix new point cloud reader where solid bodies were not being inserted correctly. Removal of methods reliant on scaled positions.
• Removed unnecessary virtual wind tunnel definitions and fixed HDF5 issue when embedding sub-grids.
• Fixed crash when using solid block due to erroneous Re definition.
• Added capability for IBM linking to sub-grids.

v1.1.0

• New tool for inlet profile generator -- add a tab for your own type of profile.
• Unification of point cloud readers and added capability for IBB construction from point clouds.
• Removal of VTK and Tecplot writers and implementation of HDF5 writer instead.
• Added enumeration for common options in utility routines, boundary methods and LatTyp for ease of use.
• Explicit casting of unsigned and signed integer types to remove compilation warnings on 64-bit systems.
• Macro convention redefined to avoid current and future conflicts.
• KBC collision operator added in place of MRT operator, Zou-He velocity inlet removed, Uniform Inlet setting removed.

v1.0.3

• Fixed output directory segmentation for large MPI jobs
• Fixed Reynolds number definition for pointcloud/BFL objects

v1.0.2

• Added ability to scale an object with respect to any direction not just x direction
• Fixed issue with BB not being applied when loading an object from a point cloud.

v1.0.1

• Fix added to allow generalised regularised on right hand and upper and back walls to impose velocity with correct direction.
• Fixed do nothing inlet from crashing since wall-embedded sub-grid changes.
• Fixed z-direction position initialisation in MPI builds which addresses hangs witnessed in earlier versions.

v1.0

• Initial release of the code.