halo_pattern.hpp File Reference

Halo exchange pattern creation from Decomposition. More...

#include <map>
#include <openpfc/core/backend_tags.hpp>
#include <openpfc/core/decomposition.hpp>
#include <openpfc/core/decomposition_neighbors.hpp>
#include <openpfc/core/sparse_vector.hpp>
#include <openpfc/core/types.hpp>
#include <openpfc/core/world.hpp>
#include <vector>

Include dependency graph for halo_pattern.hpp:

Go to the source code of this file.

Typedefs
using	pfc::halo::Int3 = pfc::types::Int3

Enumerations
enum class	pfc::halo::Connectivity { Faces , Edges , All }
	Connectivity pattern for halo exchange.

Functions
template<typename BackendTag = backend::CpuTag>
core::SparseVector< BackendTag, size_t >	pfc::halo::create_send_halo (const decomposition::Decomposition &decomp, int rank, const Int3 &direction, int halo_width)
template<typename BackendTag = backend::CpuTag>
core::SparseVector< BackendTag, size_t >	pfc::halo::create_recv_halo (const decomposition::Decomposition &decomp, int rank, const Int3 &direction, int halo_width)
	Create SparseVector representing receive halo region.
template<typename BackendTag = backend::CpuTag>
std::map< Int3, std::pair< core::SparseVector< BackendTag, size_t >, core::SparseVector< BackendTag, size_t > > >	pfc::halo::create_halo_patterns (const decomposition::Decomposition &decomp, int rank, Connectivity connectivity, int halo_width)
	Create all halo patterns for a rank.

Detailed Description

Halo exchange pattern creation from Decomposition.

Provides functions to create SparseVector objects representing halo regions for exchange between neighbors in a decomposed domain. Handles:

• Automatic index calculation from Decomposition
• Global to local index conversion
• Different connectivity patterns (faces, edges, corners)
• Periodic boundary conditions
• Multiple halo widths

auto decomp = decomposition::create(world, {2, 2, 2});
int rank = 0;
int halo_width = 1;
 
// Create halo pattern for +X face (1 row)
auto send_halo = halo::create_send_halo(decomp, rank, {1, 0, 0}, halo_width);
auto recv_halo = halo::create_recv_halo(decomp, rank, {1, 0, 0}, halo_width);
 
// Exchange indices once (setup)
int neighbor = decomposition::get_neighbor_rank(decomp, rank, {1, 0, 0});
exchange::send(send_halo, rank, neighbor, MPI_COMM_WORLD);
exchange::receive(recv_halo, neighbor, rank, MPI_COMM_WORLD);
 
// Then exchange data repeatedly
exchange::send_data(send_halo, rank, neighbor, MPI_COMM_WORLD);
exchange::receive_data(recv_halo, neighbor, rank, MPI_COMM_WORLD);

See also

core/decomposition.hpp for Decomposition class

core/decomposition_neighbors.hpp for neighbor finding

core/sparse_vector.hpp for SparseVector

core/exchange.hpp for exchange operations

Author

OpenPFC Development Team

Date

2025

Function Documentation

create_halo_patterns()

template<typename BackendTag = backend::CpuTag>

std::map< Int3, std::pair< core::SparseVector< BackendTag, size_t >, core::SparseVector< BackendTag, size_t > > > pfc::halo::create_halo_patterns	(	const decomposition::Decomposition &	decomp,
		int	rank,
		Connectivity	connectivity,
		int	halo_width
	)

Create all halo patterns for a rank.

Creates send and receive SparseVectors for all neighbors based on connectivity pattern. Returns a map from direction to {send_halo, recv_halo}.

Parameters

decomp	Decomposition object
rank	Current rank
connectivity	Connectivity pattern (Faces, Edges, or All)
halo_width	Number of halo rows

Returns

Map from direction Int3 to pair of {send_halo, recv_halo} SparseVectors

Examples

/home/runner/work/OpenPFC/OpenPFC/include/openpfc/core/halo_pattern.hpp.

Here is the call graph for this function:

Here is the caller graph for this function:

create_recv_halo()

template<typename BackendTag = backend::CpuTag>

core::SparseVector< BackendTag, size_t > pfc::halo::create_recv_halo	(	const decomposition::Decomposition &	decomp,
		int	rank,
		const Int3 &	direction,
		int	halo_width
	)

Create SparseVector representing receive halo region.

Extracts indices for the halo region where data should be received from a neighbor. Indices are in local coordinate space (0-based into the local field array, including halo zones).

Parameters

decomp	Decomposition object
rank	Current rank
direction	Direction from neighbor, e.g., {1,0,0} means receiving from +X neighbor
halo_width	Number of halo rows/layers

Returns

SparseVector with local indices of receive halo region

Note

For periodic boundaries, this will wrap around to the opposite side

These indices point into the halo zone of the local field (which should be allocated with extra space for halos)

Examples

/home/runner/work/OpenPFC/OpenPFC/include/openpfc/core/halo_pattern.hpp.

Here is the call graph for this function:

Here is the caller graph for this function: