TENO Class Reference

High-order TENO-E reconstruction on unstructured meshes (Liang, Shyy & Fu, J. Sci. Comput. 104:1, 2025; see docs/numerics/teno_e.md). More...

#include <face_reconstruction.h>

Inheritance diagram for TENO:

Classes
struct	Stencils
	Stencil cells and mirror faces of every reconstructed cell, row by row (host copy, for diagnostics). More...

Public Member Functions
	TENO ()
	~TENO ()
void	init (const toml::value &input) override
	Initialize the face reconstruction.
logging::Items	summary () const override
	Display lines for the run log.
uint8_t	n_face_quadrature_points () const override
	Get the number of quadrature points per face.
void	calc_face_values (Kokkos::View< rtype *[N_CONSERVATIVE]> solution, Kokkos::View< rtype **[2][N_CONSERVATIVE]> face_solution) override
	Reconstruct the face values.
std::vector< uint32_t >	cells_independent_of_halo (uint32_t n_owned) const override
	Cells below n_owned whose reconstruction reads no cell at or above n_owned, so that calc_cell_face_values can reconstruct them before the halo is filled. Empty if this reconstruction cannot be split by cells.
void	calc_cell_face_values (const Kokkos::DefaultExecutionSpace &exec, Kokkos::View< rtype *[N_CONSERVATIVE]> solution, Kokkos::View< rtype **[2][N_CONSERVATIVE]> face_solution, Kokkos::View< uint32_t * > cells) override
	Reconstruct the listed cells' sides of their faces, on the given execution space instance, as far as each cell can on its own (see cells_independent_of_halo()). Once every reconstructed cell went through this and those instances are fenced, finish_cell_face_values() completes the face values.
void	finish_cell_face_values (Kokkos::View< rtype *[N_CONSERVATIVE]> solution, Kokkos::View< rtype **[2][N_CONSERVATIVE]> face_solution) override
	Complete the face values after calc_cell_face_values().
bool	cell_gradients (Kokkos::View< rtype *[N_CONSERVATIVE]> solution, Kokkos::View< rtype *[N_CONSERVATIVE][N_DIM]> gradients, uint32_t n_cells) override
	Gradients of W = [rho, u, p] at the centroids of the first n_cells cells from the reconstruction polynomials, if the scheme has higher-order ones than least squares on face neighbors.
void	axisymmetric_source (Kokkos::View< rtype *[N_CONSERVATIVE]> solution, Kokkos::View< rtype *, Kokkos::LayoutStride > mu, Kokkos::View< rtype * > source, uint32_t n_cells) override
	Axisymmetric runs: for each of the first n_cells cells whose reconstruction has a smooth polynomial, replace source(c) by int (p - tau_thetatheta) dA over the planar cell, with p and the velocity of the polynomial at cell quadrature points and the cell viscosity mu(c) (inviscid if mu is empty). Other cells keep their source, which the caller set from cell values.
Stencils	large_stencils () const
std::vector< uint8_t >	stencil_cells () const
	Local cells that some stencil (central or sector) of a reconstructed cell reads.
std::array< double, 9 >	stencil_metric (uint32_t i) const
	Metric (3 x 3, row-major) in which the stencil candidates of cell i are ranked by distance (3D): the identity unless the mesh spacing around the cell is markedly anisotropic.
void	save_cache (uint8_t halo_layers=0)
	Write the precomputed data to the cache file, unless init() loaded it from there. A distributed run calls this once its halo is final, with its halo layers.
void	rebuild_cells (const std::vector< uint32_t > &cells)
	Recompute the tables of the given reconstructed cells and keep every other cell's, after the mesh geometry changed (node coordinates moved, geometry recomputed and on the device). Each cell's tables are a function of the mesh alone, so with the cells of cells_within_reach() the result is bitwise that of a full setup.
std::vector< uint32_t >	cells_within_reach (const std::vector< uint32_t > &changed) const
	Reconstructed cells whose stencil searches reach any of the given cells (within their gather_depth vertex-neighbor layers), i.e. whose tables depend on those cells' geometry.
void	set_mixture (Kokkos::View< rtype *, Kokkos::LayoutStride > gamma, Kokkos::View< rtype * > molar_mass)
	Switch to the mixture variant with these per-cell views (filled every stage).
Public Member Functions inherited from FaceReconstruction
	FaceReconstruction ()
	Construct a new Face Reconstruction object.
virtual	~FaceReconstruction ()
	Destroy the Face Reconstruction object.
void	set_mesh (std::shared_ptr< Mesh > mesh)
	Set the mesh.
void	set_boundaries (const BoundaryData &boundaries)
	Set the boundary data used for ghost states.
FaceReconstructionType	get_type () const
	Get the face reconstruction type.
void	init_face_quadrature_3d (uint8_t degree)
	Set up per-face quadrature (3D): Dunavant rules on triangles and Gauss rules mapped bilinearly onto quadrilaterals, exact for polynomials of the given degree on planar faces (degree <= 1 uses the face centroid). Also maps each quadrature point of a transmissive boundary face to the matching point of its image face (BoundaryData::face_image_quad).
void	init_face_quadrature_2d_curved ()
	Curved meshes in 2D: per-face quadrature points, weights and normals for quadrature_face's Gauss rule on every face, the curved faces' mapped onto their curves (nothing otherwise).

Static Public Member Functions
static uint8_t	cached_halo_layers (const toml::value &input)
	Halo layers recorded in this rank's cache file (0 if there is none, or it was made with other TENO options or rank count), so a distributed run can build its halo at that depth at once.

Public Attributes
uint8_t	degree = 4
uint8_t	n_dof_large = 0
uint16_t	n_stencil_large = 0
uint16_t	n_stencil_small = 10
rtype	stencil_factor = 2.0
rtype	sigma_threshold = 1.0e-3_r
rtype	sigma_upper = 1.0e-2_r
rtype	C_T = -1.0
bool	characteristic = true
bool	bound_preserving = false
bool	curved_mirrors = false
uint8_t	curved_wall_degree = 3
rtype	max_condition = 1.0e8
uint8_t	slice_shift = teno::SLICE_SHIFT
bool	single_tables = false
Kokkos::View< rtype * >	scale
Kokkos::View< rtype ** >	basis_mean
Kokkos::View< uint16_t * >	stencil_large_size
teno::PackedStencils	stencil_large
Kokkos::View< uint16_t ** >	stencil_small_size
teno::PackedStencils	stencil_small
Kokkos::View< rtype ** >	si_matrix
teno::Moments	moments
teno::SmoothnessTerms	si_terms
Kokkos::View< rtype * >	troubled
Kokkos::View< rtype *** >	troubled_coeffs
Kokkos::View< rtype **** >	troubled_small_coeffs
rtype	troubled_capacity = N_DIM == 3 ? 0.5 : 1.0
Kokkos::View< uint32_t * >	troubled_cells
Kokkos::View< uint32_t >	n_troubled
std::vector< uint8_t >	gather_depth
Kokkos::View< rtype *[3]>	cell_rule
std::vector< std::shared_ptr< GrowingBuffer > >	table_memory
bool	setup_on_host = false
uint32_t	setup_batch_cells = 0
bool	primitive = false
Kokkos::View< rtype *, Kokkos::LayoutStride >	cell_gamma
Kokkos::View< rtype * >	cell_molar_mass
Kokkos::View< uint8_t ** >	selection
Public Attributes inherited from FaceReconstruction
Quadrature	quadrature_face
Kokkos::View< rtype *** >	face_quad_points
Kokkos::View< rtype ** >	face_quad_weights
Kokkos::View< rtype *** >	face_quad_normals

Static Public Attributes
static constexpr uint8_t	SELECT_LARGE = 0xFF

Additional Inherited Members
Protected Attributes inherited from FaceReconstruction
FaceReconstructionType	type
std::shared_ptr< Mesh >	mesh
BoundaryData	boundaries

Detailed Description

High-order TENO-E reconstruction on unstructured meshes (Liang, Shyy & Fu, J. Sci. Comput. 104:1, 2025; see docs/numerics/teno_e.md).

Each cell carries one large central stencil (degree r) and one small degree-2 sector stencil per face. Cells flagged smooth by a density-based indicator use the large-stencil polynomial directly on the conservative variables; troubled cells perform TENO stencil selection on characteristic variables per face. Works on triangles and quadrilaterals.

Constructor & Destructor Documentation

TENO()

TENO::TENO

(

)

~TENO()

TENO::~TENO

(

)

Member Function Documentation

axisymmetric_source()

void TENO::axisymmetric_source ( Kokkos::View< rtype *[N_CONSERVATIVE]>  , Kokkos::View< rtype *, Kokkos::LayoutStride >  , Kokkos::View< rtype * >  , uint32_t    )

overridevirtual

Axisymmetric runs: for each of the first n_cells cells whose reconstruction has a smooth polynomial, replace source(c) by int (p - tau_thetatheta) dA over the planar cell, with p and the velocity of the polynomial at cell quadrature points and the cell viscosity mu(c) (inviscid if mu is empty). Other cells keep their source, which the caller set from cell values.

Reimplemented from FaceReconstruction.

cached_halo_layers()

uint8_t TENO::cached_halo_layers ( const toml::value &  input )

static

Halo layers recorded in this rank's cache file (0 if there is none, or it was made with other TENO options or rank count), so a distributed run can build its halo at that depth at once.

calc_cell_face_values()

void TENO::calc_cell_face_values ( const Kokkos::DefaultExecutionSpace &  exec, Kokkos::View< rtype *[N_CONSERVATIVE]>  solution, Kokkos::View< rtype **[2][N_CONSERVATIVE]>  face_solution, Kokkos::View< uint32_t * >  cells  )

overridevirtual

Reconstruct the listed cells' sides of their faces, on the given execution space instance, as far as each cell can on its own (see cells_independent_of_halo()). Once every reconstructed cell went through this and those instances are fenced, finish_cell_face_values() completes the face values.

Parameters

exec	Execution space instance to launch on.
solution	Cell states W = [rho, u_x, u_y, p].
face_solution	Face states W, as in calc_face_values.
cells	Local cells to reconstruct (not empty).

Reimplemented from FaceReconstruction.

calc_face_values()

void TENO::calc_face_values ( Kokkos::View< rtype *[N_CONSERVATIVE]>  solution, Kokkos::View< rtype **[2][N_CONSERVATIVE]>  face_solution  )

overridevirtual

Reconstruct the face values.

Parameters

solution	Cell states W = [rho, u_x, u_y, p].
face_solution	Face states W, indexed (face, quadrature point, side, variable). Side 1 of boundary faces is left untouched.

Implements FaceReconstruction.

cell_gradients()

bool TENO::cell_gradients ( Kokkos::View< rtype *[N_CONSERVATIVE]>  , Kokkos::View< rtype *[N_CONSERVATIVE][N_DIM]>  , uint32_t    )

overridevirtual

Gradients of W = [rho, u, p] at the centroids of the first n_cells cells from the reconstruction polynomials, if the scheme has higher-order ones than least squares on face neighbors.

Returns

Whether gradients were written.

Reimplemented from FaceReconstruction.

cells_independent_of_halo()

std::vector< uint32_t > TENO::cells_independent_of_halo ( uint32_t  n_owned ) const

overridevirtual

Cells below n_owned whose reconstruction reads no cell at or above n_owned, so that calc_cell_face_values can reconstruct them before the halo is filled. Empty if this reconstruction cannot be split by cells.

Reimplemented from FaceReconstruction.

cells_within_reach()

std::vector< uint32_t > TENO::cells_within_reach

(

const std::vector< uint32_t > &

changed

)

const

Reconstructed cells whose stencil searches reach any of the given cells (within their gather_depth vertex-neighbor layers), i.e. whose tables depend on those cells' geometry.

finish_cell_face_values()

void TENO::finish_cell_face_values ( Kokkos::View< rtype *[N_CONSERVATIVE]>  solution, Kokkos::View< rtype **[2][N_CONSERVATIVE]>  face_solution  )

overridevirtual

Complete the face values after calc_cell_face_values().

Reimplemented from FaceReconstruction.

init()

void TENO::init ( const toml::value &  input )

overridevirtual

Initialize the face reconstruction.

Implements FaceReconstruction.

large_stencils()

TENO::Stencils TENO::large_stencils

(

)

const

n_face_quadrature_points()

uint8_t TENO::n_face_quadrature_points ( ) const

overridevirtual

Get the number of quadrature points per face.

Returns

Number of quadrature points per face.

Implements FaceReconstruction.

rebuild_cells()

void TENO::rebuild_cells

(

const std::vector< uint32_t > &

cells

)

Recompute the tables of the given reconstructed cells and keep every other cell's, after the mesh geometry changed (node coordinates moved, geometry recomputed and on the device). Each cell's tables are a function of the mesh alone, so with the cells of cells_within_reach() the result is bitwise that of a full setup.

save_cache()

void TENO::save_cache

(

uint8_t

halo_layers = 0

)

Write the precomputed data to the cache file, unless init() loaded it from there. A distributed run calls this once its halo is final, with its halo layers.

set_mixture()

void TENO::set_mixture	(	Kokkos::View< rtype *, Kokkos::LayoutStride >	gamma,
		Kokkos::View< rtype * >	molar_mass
	)

Switch to the mixture variant with these per-cell views (filled every stage).

stencil_cells()

std::vector< uint8_t > TENO::stencil_cells

(

)

const

Local cells that some stencil (central or sector) of a reconstructed cell reads.

stencil_metric()

std::array< double, 9 > TENO::stencil_metric

(

uint32_t

i

)

const

Metric (3 x 3, row-major) in which the stencil candidates of cell i are ranked by distance (3D): the identity unless the mesh spacing around the cell is markedly anisotropic.

summary()

logging::Items TENO::summary ( ) const

overridevirtual

Display lines for the run log.

Reimplemented from FaceReconstruction.

Member Data Documentation

basis_mean

Kokkos::View<rtype **> TENO::basis_mean

bound_preserving

bool TENO::bound_preserving = false

C_T

rtype TENO::C_T = -1.0

cell_gamma

Kokkos::View<rtype *, Kokkos::LayoutStride> TENO::cell_gamma

cell_molar_mass

Kokkos::View<rtype *> TENO::cell_molar_mass

cell_rule

Kokkos::View<rtype *[3]> TENO::cell_rule

characteristic

bool TENO::characteristic = true

curved_mirrors

bool TENO::curved_mirrors = false

curved_wall_degree

uint8_t TENO::curved_wall_degree = 3

degree

uint8_t TENO::degree = 4

gather_depth

std::vector<uint8_t> TENO::gather_depth

max_condition

rtype TENO::max_condition = 1.0e8

moments

teno::Moments TENO::moments

n_dof_large

uint8_t TENO::n_dof_large = 0

n_stencil_large

uint16_t TENO::n_stencil_large = 0

n_stencil_small

uint16_t TENO::n_stencil_small = 10

n_troubled

Kokkos::View<uint32_t> TENO::n_troubled

primitive

bool TENO::primitive = false

scale

Kokkos::View<rtype *> TENO::scale

SELECT_LARGE

constexpr uint8_t TENO::SELECT_LARGE = 0xFF

staticconstexpr

selection

Kokkos::View<uint8_t **> TENO::selection

setup_batch_cells

uint32_t TENO::setup_batch_cells = 0

setup_on_host

bool TENO::setup_on_host = false

si_matrix

Kokkos::View<rtype **> TENO::si_matrix

si_terms

teno::SmoothnessTerms TENO::si_terms

sigma_threshold

rtype TENO::sigma_threshold = 1.0e-3_r

sigma_upper

rtype TENO::sigma_upper = 1.0e-2_r

single_tables

bool TENO::single_tables = false

slice_shift

uint8_t TENO::slice_shift = teno::SLICE_SHIFT

stencil_factor

rtype TENO::stencil_factor = 2.0

stencil_large

teno::PackedStencils TENO::stencil_large

stencil_large_size

Kokkos::View<uint16_t *> TENO::stencil_large_size

stencil_small

teno::PackedStencils TENO::stencil_small

stencil_small_size

Kokkos::View<uint16_t **> TENO::stencil_small_size

table_memory

std::vector<std::shared_ptr<GrowingBuffer> > TENO::table_memory

troubled

Kokkos::View<rtype *> TENO::troubled

troubled_capacity

rtype TENO::troubled_capacity = N_DIM == 3 ? 0.5 : 1.0

troubled_cells

Kokkos::View<uint32_t *> TENO::troubled_cells

troubled_coeffs

Kokkos::View<rtype ***> TENO::troubled_coeffs

troubled_small_coeffs

Kokkos::View<rtype ****> TENO::troubled_small_coeffs

---

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

• /home/runner/work/mallard/mallard/mallard/src/numerics/face_reconstruction.h
• /home/runner/work/mallard/mallard/mallard/src/numerics/teno.cpp