t8_time_forest_partition.cxx

t8_time_forest_partition

Overview

The t8_time_forest_partition.cxx benchmark evaluates the performance of t8code's mesh partitioning capabilities. It measures the efficiency of partitioning operations across various mesh configurations and parallel processing scenarios.

---

Execution

Run the benchmark using MPI:

mpirun -n <num_processes> ./t8_time_forest_partition [options]

Replace <num_processes> with the desired number of MPI processes.

---

Command-Line Options

The benchmark accepts various options to configure the mesh and partitioning parameters.

Short	Long	Argument	Description
-h	--help		Show a help message and exit
-o	--no-vtk		Do not produce VTK (visualization) output files
-f	--mshfile	STRING	Use a .msh file (GMSH format, prefix only) as the coarse mesh
-d	--dim	INT	(With -f) Dimension of the mesh: 2 or 3
-c	--cmeshfile	STRING	Use cmesh files created by t8_cmesh_save
-n	--nfiles	INT	Number of cmesh files to use with -c
-s	--stride	INT	MPI rank stride for cmesh file reading (default: 16)
-t	--test-tet		Use built-in tetrahedral connectivity test mesh (disables VTK by default)
-L	--test-linear-cylinder		Use built-in linear cylinder mesh (disables VTK by default)
-O	--test-cad-cylinder		Use built-in CAD (curved) cylinder mesh (disables VTK by default)
-l	--level	INT	Initial uniform refinement level
-r	--rlevel	INT	Number of adaptive refinement levels to add (band refinement)
	--cmesh-level	INT	Starting level for built-in cylinder meshes (default 0, only with -L or -O)
-x	--xmin	REAL	Minimum x-coordinate for the adaptive refinement band
-X	--xmax	REAL	Maximum x-coordinate for the adaptive refinement band
-T	--time	REAL	Simulated time span (end time)
-D	--delta_t	REAL	Timestep size (deprecated; use -C for CFL)
-C	--cfl	REAL	CFL number; if set, delta_t = CFL * 0.64 / 2^level
-g	--ghost		Create ghost elements for parallel overlap
-b	--balance		Enforce 2:1 mesh balancing
-z	--use_cad		Use curved meshes to match original geometry (needs .brep and .msh files)

For a complete list of options and their descriptions, run:

./t8_time_forest_partition -h

---

Sample Usage

mpirun -np 8 ./t8_time_forest_partition -L -l 4 -r 3 -x -0.3 -X -0.4 -T 0.05 -C 0.8 -g -b

What this does:

• Uses the built-in linear cylinder mesh.
• Sets initial uniform refinement level to 4, with up to 3 additional adaptive refinements in the band x ∈ [−0.3, −0.4].
• Runs up to time T = 0.05, with CFL-derived time step.
• Builds ghost layers and enforces mesh balancing.

---

Output

The benchmark outputs timing information for the partitioning operations, including:

• Total time taken for partitioning.
• Average time per trial.
• Minimum and maximum times observed.

This data helps assess the scalability and efficiency of t8code's partitioning algorithms under different configurations.