PETSCPARTITIONERMULTISTAGE#
“multistage” - A PetscPartitioner object using a multistage distribution strategy
Options Database Keys#
-petscpartitioner_multistage_strategy strategy - Either
PETSCPARTITIONER_MS_STRATEGY_NODE, orPETSCPARTITIONER_MS_STRATEGY_MSECTION-petscpartitioner_multistage_node_size nranks - Number of processes per computing node (or
PETSC_DECIDE)-petscpartitioner_multistage_node_interleaved (true|false) - Assign ranks round-robin.
-petscpartitioner_multistage_msection splits - Number of splits per level in recursive m-section splits (use
2for bisection)-petscpartitioner_multistage_tpwgts (true|false) - Use target partition weights in stage partitioner
Notes#
The default multistage strategy use PETSCPARTITIONER_MS_STRATEGY_NODE and automatically discovers node information using MPI_Comm_split_type.
PETSCPARTITIONER_MS_STRATEGY_MSECTION is more for testing purposes.
Options for single stage partitioners are prefixed by -petscpartitioner_multistage_levels_.
For example, to use parmetis in all stages, -petscpartitioner_multistage_levels_petscpartitioner_type parmetis
See Also#
PetscPartitionerType, PetscPartitionerCreate(), PetscPartitionerSetType()
Level#
intermediate
Location#
src/dm/partitioner/impls/multistage/mspart.c
Index of all MatGraphOperations routines
Table of Contents for all manual pages
Index of all manual pages