External field omega problem using DynOmegaD2Q9Descriptor and MPI compiler

[guojuw]

Dear OLB developers:

Here is how I set up the program using fresh download codes:

1. Go to ‘$openlbRoot/config.mk’, and switch “CXX:=mpic++”, “CXXFLAGS := $(OPTIM)”,”PARALLEL_MODE := MPI”, “BUILDTYPE := generic“。

2. Go to “$openlbRoot/examples/laminar/powerLaw2d/powerLaw2d.cpp”,
add macro “#define PARALLEL_MODE_MPI”. Within ‘void getResults()’ function, add the following lines to obtain the relaxation frequency: ” SuperLatticeField2D<T,DESCRIPTOR,olb::descriptors::OMEGA> omegaLB (sLattice);

vtmWriter.addFunctor( omegaLB );”

and ”
BlockReduction2D2D<T> planeReduction2( omegaLB, 600, BlockDataSyncMode::ReduceOnly );
heatmap::write(planeReduction, iT);”

3. Compile the source code using “make” command and execute the program using “mpirun -np 1 powerLaw2d”, “mpirun -np 2 powerLaw2d” and “mpirun -np 4 powerLaw2d”.

4. In these three cases with different numbers of MPI processes, the results of Omega field were

5. The problem is that, the Omega at cuboid block edges were discontinued, i.e., in the second map, there was a vertical line in the center, and in the third map, there were a vertical line and a horizontal line at the center.

I don’t know whether it is a problem within MPI post-processing or DynOmegaD2Q9Descriptor.

Thanks,
Junwei Guo

[mathias]

Dear Junwei,

that is not a bug. A communication step is missing. Add

SuperExternal3D sExternal( superGeometry, sLattice, sLattice.getOverlap() );

sExternal.communicate();

and it will work!

Best
Mathias

[Author]

Posts