Section DISTRIBUTION

• can be used used to tune the parallel distribution of the data

• Section path: CP2K_INPUT / FORCE_EVAL / DFT / QS / DISTRIBUTION

• This section cannot be repeated.

• Subsections

  • none

• Keywords

  • 2D_MOLECULAR_DISTRIBUTION
  • BASIC_CLUSTER_OPTIMIZATION
  • BASIC_OPTIMIZATION
  • BASIC_SPATIAL_OPTIMIZATION
  • COST_MODEL
  • SKIP_OPTIMIZATION
  • SYMMETRIC

• Keyword descriptions

  2D_MOLECULAR_DISTRIBUTION
  	2D_MOLECULAR_DISTRIBUTION {Logical}
  	Distribute the atoms so that atoms belonging to a given molecule are on the same CPU for the 2D distribution. This might give rise to a worse distribution but reduces memory needs of finding the optimal distribution.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  BASIC_CLUSTER_OPTIMIZATION
  	BASIC_CLUSTER_OPTIMIZATION {Logical}
  	Creates a distribution with spatial info, using recursively KMEANS clustering.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  BASIC_OPTIMIZATION
  	BASIC_OPTIMIZATION {Logical}
  	Creates a distribution based on a few heuristics using only minimal memory and CPU time.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .TRUE.

  BASIC_SPATIAL_OPTIMIZATION
  	BASIC_SPATIAL_OPTIMIZATION {Logical}
  	Creates a distribution with spatial info, using only minimal memory and CPU time.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  COST_MODEL
  	COST_MODEL {Keyword}
  	The cost model that needs to be minimized
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: BLOCK_COUNT
  	List of valid keywords: BLOCK_COUNT the number of blocks BLOCK_LMAX the number of blocks weighted by the sum of the lmax BLOCK_SURFACE the number of blocks weighted by the number elements per block

  SKIP_OPTIMIZATION
  	SKIP_OPTIMIZATION {Logical}
  	Do not optimize the distribution, go for something very simple. Might be useful if the optimization, which scales quadratically in system size, is too expensive.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .FALSE.

  SYMMETRIC
  	SYMMETRIC {Logical}
  	Take the symmetry of the distribution_2d into account.
  	This keyword cannot be repeated and it expects precisely one logical.
  	The lone keyword behaves as a switch to .TRUE.
  	Default value: .TRUE.