Field distributed over a level: decomposition (BoxArray) + distribution (DistributionMapping) + ncomp components + ngrow ghosts. More...

#include <multifab.hpp>

Collaboration diagram for pops::MultiFab:

Public Member Functions
	MultiFab ()=default

	MultiFab (BoxArray ba, DistributionMapping dm, int ncomp, int ngrow)
	Builds the field: allocates one Fab2D (ncomp components, ngrow ghosts) for EACH box that this rank owns according to dm.

const BoxArray &	box_array () const
	GLOBAL decomposition of the level (all boxes, all ranks).

const DistributionMapping &	dmap () const
	GLOBAL distribution (owner rank per box).

int	ncomp () const
	Number of components.

int	n_grow () const
	Number of ghost layers.

int	local_size () const
	Number of fabs OWNED by this rank (bound on local indices).

Fab2D &	fab (int li)
	Local fab at index li (0 <= li < local_size()), for writing.

const Fab2D &	fab (int li) const
	Local fab at index li, for reading.

const Box2D &	box (int li) const
	VALID box of local fab li.

int	global_index (int li) const
	GLOBAL index (in box_array) of local fab li.

int	local_index_of (int global) const
	LOCAL index of the global box `global`, or -1 if it is not owned by this rank.

void	sync_host ()
	Makes the HOST residence valid (before a host access: operator(), loop, set_val).

void	sync_device ()
	Marks a DEVICE residence (before a kernel). No-op under unified memory.

void	set_val (Real v)
	Fills all cells (valid + ghosts) of every local fab with v.

HaloScheduleCache &	halo_cache () const
	Internal (ADC-260): memoized halo-exchange schedule used by fill_boundary.

Detailed Description

Field distributed over a level: decomposition (BoxArray) + distribution (DistributionMapping) + ncomp components + ngrow ghosts.

Allocates only the fabs owned by THIS rank; iteration runs over local_size() (LOCAL indices). global_index/local_index_of bridge local <-> global.

Constructor & Destructor Documentation

◆ MultiFab() [1/2]

pops::MultiFab::MultiFab ( )

default

◆ MultiFab() [2/2]

pops::MultiFab::MultiFab	(	BoxArray	ba,
		DistributionMapping	dm,
		int	ncomp,
		int	ngrow
	)

inline

Builds the field: allocates one Fab2D (ncomp components, ngrow ghosts) for EACH box that this rank owns according to dm.

Boxes belonging to other ranks are not allocated here.

Here is the call graph for this function:

Member Function Documentation

◆ box()

const Box2D & pops::MultiFab::box ( int li ) const

inline

VALID box of local fab li.

Here is the caller graph for this function:

◆ box_array()

const BoxArray & pops::MultiFab::box_array ( ) const

inline

GLOBAL decomposition of the level (all boxes, all ranks).

Here is the caller graph for this function:

◆ dmap()

const DistributionMapping & pops::MultiFab::dmap ( ) const

inline

GLOBAL distribution (owner rank per box).

Here is the caller graph for this function:

◆ fab() [1/2]

Fab2D & pops::MultiFab::fab ( int li )

inline

Local fab at index li (0 <= li < local_size()), for writing.

◆ fab() [2/2]

const Fab2D & pops::MultiFab::fab ( int li ) const

inline

Local fab at index li, for reading.

◆ global_index()

int pops::MultiFab::global_index ( int li ) const

inline

GLOBAL index (in box_array) of local fab li.

◆ halo_cache()

HaloScheduleCache & pops::MultiFab::halo_cache ( ) const

inline

Internal (ADC-260): memoized halo-exchange schedule used by fill_boundary.

Lazily created on first use. The schedule is a pure function of (box_array, dmap, n_grow) for a given (Periodicity, domain); since none of ba_/dm_/ngrow_ has an in-place setter, the cache can only go stale through whole-object (re)assignment (e.g. AMR regrid builds a fresh MultiFab and move-assigns it over the level slot), which drops the cache with the object. It is shared on copy (a copy has the same layout), which keeps copies consistent. Not part of the public numerical API. Returned by reference so fill_boundary can populate it.

Here is the caller graph for this function:

◆ local_index_of()

int pops::MultiFab::local_index_of ( int global ) const

inline

LOCAL index of the global box global, or -1 if it is not owned by this rank.

Here is the caller graph for this function:

◆ local_size()

int pops::MultiFab::local_size ( ) const

inline

Number of fabs OWNED by this rank (bound on local indices).

Here is the caller graph for this function:

◆ n_grow()

int pops::MultiFab::n_grow ( ) const

inline

Number of ghost layers.

Here is the caller graph for this function:

◆ ncomp()

int pops::MultiFab::ncomp ( ) const

inline

Number of components.

Here is the caller graph for this function:

◆ set_val()

void pops::MultiFab::set_val ( Real v )

inline

Fills all cells (valid + ghosts) of every local fab with v.

Synchronizes host residence first (a kernel may have written these fabs).

Here is the call graph for this function:

Here is the caller graph for this function:

◆ sync_device()

void pops::MultiFab::sync_device ( )

inline

Marks a DEVICE residence (before a kernel). No-op under unified memory.

Here is the call graph for this function:

◆ sync_host()

void pops::MultiFab::sync_host ( )

inline

Makes the HOST residence valid (before a host access: operator(), loop, set_val).

Under unified memory = a targeted device_fence().

Here is the call graph for this function:

Here is the caller graph for this function:

The documentation for this class was generated from the following file:

include/pops/mesh/storage/multifab.hpp

MultiFab Class Reference¶

Public Member Functions

Detailed Description

Constructor & Destructor Documentation

◆ MultiFab() [1/2]

◆ MultiFab() [2/2]

Member Function Documentation

◆ box()

◆ box_array()

◆ dmap()

◆ fab() [1/2]

◆ fab() [2/2]

◆ global_index()

◆ halo_cache()

◆ local_index_of()

◆ local_size()

◆ n_grow()

◆ ncomp()

◆ set_val()

◆ sync_device()

◆ sync_host()