Global data (distribution and block sizes) for tall-and-skinny matrices For very sparse matrices with one very large dimension, storing array data of the same size as the matrix dimensions may require too much memory and we need to compute them on the fly for a given row or column. Hence global array data such as distribution and block sizes are specified as function objects, leaving up to the caller how to efficiently store global data. More...

Data Types
interface	dbt_tas_blk_size_arb
	type for arbitrary block sizes More...

interface	dbt_tas_blk_size_one
	type for blocks of size one More...

interface	dbt_tas_blk_size_repl
	type for replicated block sizes More...

interface	dbt_tas_dist_arb
	type for arbitrary distributions More...

interface	dbt_tas_dist_cyclic
	type for cyclic (round robin) distribution: More...

interface	dbt_tas_dist_repl
	type for replicated distribution More...

type	dbt_tas_distribution

type	dbt_tas_rowcol_data

interface	rowcol_dist
	map matrix rows/cols to distribution rows/cols More...

Functions/Subroutines
type(dbt_tas_dist_arb) function, public	dbt_tas_dist_arb_default (nprowcol, nmrowcol, dbt_sizes)
	Distribution that is more or less cyclic (round robin) and load balanced with different weights for each element. This is used for creating adhoc distributions whenever matrices are mapped to new grids. Only for small dimensions since distribution is created as an array.

subroutine, public	dbt_tas_default_distvec (nblk, nproc, blk_size, dist)
	get a load-balanced and randomized distribution along one tensor dimension

Detailed Description

Global data (distribution and block sizes) for tall-and-skinny matrices For very sparse matrices with one very large dimension, storing array data of the same size as the matrix dimensions may require too much memory and we need to compute them on the fly for a given row or column. Hence global array data such as distribution and block sizes are specified as function objects, leaving up to the caller how to efficiently store global data.

Author: Patrick Seewald

Function/Subroutine Documentation

◆ dbt_tas_dist_arb_default()

type(dbt_tas_dist_arb) function, public dbt_tas_global::dbt_tas_dist_arb_default	(	integer	nprowcol,
		integer(kind=int_8), intent(in)	nmrowcol,
		class(dbt_tas_rowcol_data), intent(in)	dbt_sizes
	)

Distribution that is more or less cyclic (round robin) and load balanced with different weights for each element. This is used for creating adhoc distributions whenever matrices are mapped to new grids. Only for small dimensions since distribution is created as an array.

Parameters

nprowcol	...
nmrowcol	...
dbt_sizes	...

Returns: ...

Author: Patrick Seewald

Definition at line 458 of file dbt_tas_global.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ dbt_tas_default_distvec()

subroutine, public dbt_tas_global::dbt_tas_default_distvec	(	integer, intent(in)	nblk,
		integer, intent(in)	nproc,
		integer, dimension(nblk), intent(in)	blk_size,
		integer, dimension(nblk), intent(out)	dist
	)

get a load-balanced and randomized distribution along one tensor dimension