Generic auxiliary functions. More...

#include <type_traits>
#include "../../../Common/include/option_structure.hpp"
#include "../../../Common/include/parallelization/vectorization.hpp"
#include "../../../Common/include/containers/C2DContainer.hpp"
#include "../../../Common/include/linear_algebra/CSysVector.hpp"
#include "../../../Common/include/linear_algebra/CSysMatrix.hpp"

Classes
struct	CAlignTraits< Type >
	Alignment of the static containers backing a flux value type. More...

struct	CAlignTraits< simd::Array< Scalar_t, N > >

struct	CValueTraits< unsigned long >

struct	CValueTraits< simd::Array< unsigned long, N > >

struct	CLaneTraits< su2double >

struct	CLaneTraits< simd::Array< Scalar_t, N > >

struct	CPair< T >
	Simple pair type for i/j variables. More...

struct	CIdentity< T >
	Blocks a template parameter from participating in argument deduction. More...

struct	EdgeResidual< Double, nVar_ >
	Residual of one edge, accumulated by the convective and the diffusive terms. More...

struct	CGradientBlock< nVarGrad, nDim, Double, Gradient_t >
	Reads the gradient rows of one point as one block, gathered up front. More...

struct	CGradientRows< nDim, Double, Gradient_t >
	Reads the gradient rows of one point one row at a time. More...

Typedefs
using	SparseMatrixType = CSysMatrix< su2mixedfloat >

template<class Type , size_t Size>
using	Vector = C2DContainer< unsigned long, Type, StorageType::ColumnMajor, CAlignTraits< Type >::Align, Size, 1 >
	Static vector and matrix types.

template<class Type , size_t Rows, size_t Cols>
using	Matrix = C2DContainer< unsigned long, Type, StorageType::RowMajor, CAlignTraits< Type >::Align, Rows, Cols >

template<class T >
using	CNonDeduced = typename CIdentity< T >::type

Enumerations
enum class	UpdateType { COLORING , REDUCTION }
	Ways to update vectors and system matrices. COLORING is the typical i/j update, whereas for REDUCTION the fluxes are stored and the matrix diagonal is not modified. More...

enum class	MusclLimiter { NONE , EDGE , POINT }
	How the reconstructed differences are limited. More...

Functions
constexpr size_t	Max (size_t a, size_t b)
	Constexpr version of max.

template<size_t nDim, class ForwardIterator , class T >
auto	dot (ForwardIterator iterator, const T ptr) -> typename std::decay< decltype(ptr)>::type
	Dot product.

template<size_t nDim, class Double , class ForwardIterator >
Double	dot (ForwardIterator iterator, const Vector< Double, nDim > &vector)

template<size_t nDim, class Double >
Double	dot (const Vector< Double, nDim > &a, const Vector< Double, nDim > &b)

template<size_t nDim, class ForwardIterator >
auto	squaredNorm (ForwardIterator iterator) -> typename std::decay< decltype(*iterator)>::type
	Squared norm.

template<size_t nDim, class Double >
Double	squaredNorm (const Vector< Double, nDim > &vector)

template<size_t nDim, class Double >
Vector< Double, nDim >	tangentProjection (const Matrix< Double, nDim, nDim > &tensor, const Vector< Double, nDim > &unitVector)
	Tangential projection.

template<size_t nDim, class Double >
Double	norm (const Vector< Double, nDim > &vector)
	Vector norm.

template<class Int , class Container , class Double = typename CValueTraits<Int>::Double>
Double	gatherVariables (Int iPoint, const Container &vars, size_t iVar=0)
	Gather a single variable, from column iVar (0 by default) of row iPoint of a 2D container, or from index iPoint of a 1D container.

template<size_t nVar, class Int , class Container , class Double = typename CValueTraits<Int>::Double>
Vector< Double, nVar >	gatherVariables (Int iPoint, const Container &vars, size_t iVar=0)
	Gather nVar contiguous variables starting at column iVar (0 by default) of row iPoint of a 2D container.

template<size_t nRows, size_t nCols, class Int , class Container , class Double = typename CValueTraits<Int>::Double>
Matrix< Double, nRows, nCols >	gatherVariables (Int iPoint, const Container &vars, size_t iRow=0)
	Gather an nRows x nCols block of a 3D container, from outer index iPoint and starting at middle index iRow.

template<class Double , size_t Size>
void	setPreaccOut (Vector< Double, Size > &x, size_t nVar)
	Register the leading nVar entries of a static vector as preaccumulation outputs.

template<class Double , size_t nVar>
void	stopPreacc (Vector< Double, nVar > &x)
	Stop the AD preaccumulation.

template<size_t nDim, class Int , class Container , class Double = typename CValueTraits<Int>::Double>
Vector< Double, nDim >	distanceVector (Int iPoint, Int jPoint, const Container &coords)
	Distance vector, from point i to point j of one container.

template<size_t nDim, class Int , class Container , class Double = typename CValueTraits<Int>::Double>
Vector< Double, nDim >	distanceVector (Int iPoint, const Container &coords_i, Int jPoint, const Container &coords_j)
	Distance vector, from point i of one container to point j of another.

template<class Double >
Double	umusclProjection (const Double &gradProj, const Double &delta, const CNonDeduced< Double > &kappa)
	Blended difference for U-MUSCL reconstruction.

template<size_t nVarGrad_, size_t nDim, class Double , class Gradient_t >
auto	gradientReader (typename CLaneTraits< Double >::Int iPoint, const Gradient_t &gradient, size_t iRow)
	Gradient reader of one point, blocked or row by row according to nVarGrad_.

template<MusclLimiter limiterKind, size_t nVarGrad_ = 0, size_t nDim, class Double , class VarType , class Limiter_t , class Gradient_t >
void	muscl (typename CLaneTraits< Double >::Int iPoint, typename CLaneTraits< Double >::Int jPoint, const Vector< Double, nDim > &vector_ij, const Gradient_t &gradient, const Limiter_t &limiter, size_t iRow, CPair< VarType > &V, const CNonDeduced< Double > &kappa, const CNonDeduced< Double > &umusclRamp, size_t nVarGradRuntime)
	U-MUSCL reconstruction of nVarGrad variables, from the gradient rows starting at iRow.

template<size_t nVarGrad_ = 0, size_t nDim, class Double , class VarType , class Limiter_t , class Gradient_t >
void	reconstruct (typename CLaneTraits< Double >::Int iPoint, typename CLaneTraits< Double >::Int jPoint, const Vector< Double, nDim > &vector_ij, const Gradient_t &gradient, const Limiter_t &limiter, LIMITER limiterType, size_t iRow, CPair< VarType > &V, const CNonDeduced< Double > &kappa, const CNonDeduced< Double > &umusclRamp, size_t nVarGradRuntime=0)
	Reconstruct a slice of nVarGrad variables starting at column iRow, dispatching on the limiter type; shared by the flow and the scalar reconstructions so both call one body.

template<size_t nVar, class Double , class Int = typename CLaneTraits<Double>::Int>
void	updateLinearSystem (Int iEdge, Int iPoint, Int jPoint, bool implicit, UpdateType updateType, Double updateMask, const Vector< Double, nVar > &flux, const Matrix< Double, nVar, nVar > &jac_i, const Matrix< Double, nVar, nVar > &jac_j, CSysVector< su2double > &vector, SparseMatrixType &matrix)
	Update the matrix and right-hand-side of a linear system with one conservative flux.

template<size_t nVar, class Double , class Int = typename CLaneTraits<Double>::Int>
void	updateLinearSystem (Int iEdge, Int iPoint, Int jPoint, bool implicit, UpdateType updateType, Double updateMask, const EdgeResidual< Double, nVar > &res, CSysVector< su2double > &vector, CSysVector< su2double > &vectorDiff, SparseMatrixType &matrix)
	Update the matrix and right-hand-side of a linear system with two independent row contributions and four independent Jacobian blocks.

template<class Double , class Int = typename CLaneTraits<Double>::Int>
void	updateEdgeMassFlux (Int iEdge, const Double &massFlux, su2activevector *target)
	Store the (scalar) mass flux of an edge, e.g. for "bounded scalar" transport equations.

Variables
constexpr size_t	Dynamic = size_t(-1)
	Equation count of a model whose value is only known at runtime.

constexpr size_t	MaxScalarVar = 8
	Backing size of the static arrays of a dynamic model.

Detailed Description

Generic auxiliary functions.

Author: P. Gomes

Version: 8.5.0 "Harrier"

SU2 Project Website: https://su2code.github.io

The SU2 Project is maintained by the SU2 Foundation (http://su2foundation.org)

SU2 is free software; you can redistribute it and/or modify it under the terms of the GNU Lesser General Public License as published by the Free Software Foundation; either version 2.1 of the License, or (at your option) any later version.

SU2 is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public License along with SU2. If not, see http://www.gnu.org/licenses/.

Typedef Documentation

◆ CNonDeduced

template<class T >

using CNonDeduced = typename CIdentity<T>::type

◆ Matrix

template<class Type , size_t Rows, size_t Cols>

using Matrix = C2DContainer<unsigned long, Type, StorageType::RowMajor, CAlignTraits<Type>::Align, Rows, Cols>

◆ SparseMatrixType

using SparseMatrixType = CSysMatrix<su2mixedfloat>

◆ Vector

template<class Type , size_t Size>

using Vector = C2DContainer<unsigned long, Type, StorageType::ColumnMajor, CAlignTraits<Type>::Align, Size, 1>

Static vector and matrix types.

Note: These should be used instead of C-style arrays.

Enumeration Type Documentation

◆ MusclLimiter

enum class MusclLimiter

strong

How the reconstructed differences are limited.

Enumerator
NONE
EDGE
POINT

◆ UpdateType

enum class UpdateType

strong

Ways to update vectors and system matrices. COLORING is the typical i/j update, whereas for REDUCTION the fluxes are stored and the matrix diagonal is not modified.

Enumerator
COLORING
REDUCTION

Function Documentation

◆ distanceVector() [1/2]

template<size_t nDim, class Int , class Container , class Double = typename CValueTraits<Int>::Double>

Vector< Double, nDim > distanceVector	(	Int	iPoint,
		const Container &	coords_i,
		Int	jPoint,
		const Container &	coords_j
	)

inline

Distance vector, from point i of one container to point j of another.

Note: The two endpoints of a boundary flux read different containers, the solver's own and the marker's ghost one; the interior edge loop passes the same container twice.

◆ distanceVector() [2/2]

template<size_t nDim, class Int , class Container , class Double = typename CValueTraits<Int>::Double>

Vector< Double, nDim > distanceVector	(	Int	iPoint,
		Int	jPoint,
		const Container &	coords
	)

inline

Distance vector, from point i to point j of one container.

◆ dot() [1/3]

template<size_t nDim, class Double >

Double dot	(	const Vector< Double, nDim > &	a,
		const Vector< Double, nDim > &	b
	)

inline

◆ dot() [2/3]

template<size_t nDim, class ForwardIterator , class T >

auto dot	(	ForwardIterator	iterator,
		const T *	ptr
	)		-> typename std::decay<decltype(*ptr)>::type

inline

Dot product.

◆ dot() [3/3]

template<size_t nDim, class Double , class ForwardIterator >

Double dot	(	ForwardIterator	iterator,
		const Vector< Double, nDim > &	vector
	)

inline

◆ gatherVariables() [1/3]

template<size_t nRows, size_t nCols, class Int , class Container , class Double = typename CValueTraits<Int>::Double>

Matrix< Double, nRows, nCols > gatherVariables	(	Int	iPoint,
		const Container &	vars,
		size_t	iRow = `0`
	)

inline

Gather an nRows x nCols block of a 3D container, from outer index iPoint and starting at middle index iRow.

◆ gatherVariables() [2/3]

template<class Int , class Container , class Double = typename CValueTraits<Int>::Double>

Double gatherVariables	(	Int	iPoint,
		const Container &	vars,
		size_t	iVar = `0`
	)

inline

Gather a single variable, from column iVar (0 by default) of row iPoint of a 2D container, or from index iPoint of a 1D container.

◆ gatherVariables() [3/3]

template<size_t nVar, class Int , class Container , class Double = typename CValueTraits<Int>::Double>

Vector< Double, nVar > gatherVariables	(	Int	iPoint,
		const Container &	vars,
		size_t	iVar = `0`
	)

inline

Gather nVar contiguous variables starting at column iVar (0 by default) of row iPoint of a 2D container.

◆ gradientReader()

template<size_t nVarGrad_, size_t nDim, class Double , class Gradient_t >

auto gradientReader	(	typename CLaneTraits< Double >::Int	iPoint,
		const Gradient_t &	gradient,
		size_t	iRow
	)

inline

Gradient reader of one point, blocked or row by row according to nVarGrad_.

◆ Max()

constexpr size_t Max	(	size_t	a,
		size_t	b
	)

inlineconstexpr

Constexpr version of max.

◆ muscl()

template<MusclLimiter limiterKind, size_t nVarGrad_ = 0, size_t nDim, class Double , class VarType , class Limiter_t , class Gradient_t >

void muscl	(	typename CLaneTraits< Double >::Int	iPoint,
		typename CLaneTraits< Double >::Int	jPoint,
		const Vector< Double, nDim > &	vector_ij,
		const Gradient_t &	gradient,
		const Limiter_t &	limiter,
		size_t	iRow,
		CPair< VarType > &	V,
		const CNonDeduced< Double > &	kappa,
		const CNonDeduced< Double > &	umusclRamp,
		size_t	nVarGradRuntime
	)

inline

U-MUSCL reconstruction of nVarGrad variables, from the gradient rows starting at iRow.

Note: The limiter kind is a template parameter so that the choice is made once, outside the loop, by the dispatching overload below.

Parameters

[in]	iRow	- Starting row of gradient (and column of limiter) to read, for reconstructing a slice of a larger set of gradients (e.g. only the velocity out of the primitives).
[in]	nVarGradRuntime	- Variable count of a Dynamic model, known only at runtime; ignored (falling back to nVarGrad_ or VarType::nVar) when left at its default of 0.

◆ norm()

template<size_t nDim, class Double >

Double norm ( const Vector< Double, nDim > & vector )

inline

Vector norm.

◆ reconstruct()

template<size_t nVarGrad_ = 0, size_t nDim, class Double , class VarType , class Limiter_t , class Gradient_t >

void reconstruct	(	typename CLaneTraits< Double >::Int	iPoint,
		typename CLaneTraits< Double >::Int	jPoint,
		const Vector< Double, nDim > &	vector_ij,
		const Gradient_t &	gradient,
		const Limiter_t &	limiter,
		LIMITER	limiterType,
		size_t	iRow,
		CPair< VarType > &	V,
		const CNonDeduced< Double > &	kappa,
		const CNonDeduced< Double > &	umusclRamp,
		size_t	nVarGradRuntime = `0`
	)

inline

Reconstruct a slice of nVarGrad variables starting at column iRow, dispatching on the limiter type; shared by the flow and the scalar reconstructions so both call one body.

◆ setPreaccOut()

template<class Double , size_t Size>

void setPreaccOut	(	Vector< Double, Size > &	x,
		size_t	nVar
	)

inline

Register the leading nVar entries of a static vector as preaccumulation outputs.

Note: A lane vector is registered one lane at a time, a scalar in one call; a kernel therefore reaches this rather than AD::SetPreaccOut directly, and reads the same whichever value type it is bound to.

◆ squaredNorm() [1/2]

template<size_t nDim, class Double >

Double squaredNorm ( const Vector< Double, nDim > & vector )

inline

◆ squaredNorm() [2/2]

template<size_t nDim, class ForwardIterator >

auto squaredNorm ( ForwardIterator iterator ) -> typename std::decay<decltype(*iterator)>::type

inline

Squared norm.

◆ stopPreacc()

template<class Double , size_t nVar>

void stopPreacc ( Vector< Double, nVar > & x )

inline

Stop the AD preaccumulation.

◆ tangentProjection()

template<size_t nDim, class Double >

Vector< Double, nDim > tangentProjection	(	const Matrix< Double, nDim, nDim > &	tensor,
		const Vector< Double, nDim > &	unitVector
	)

inline

Tangential projection.

◆ umusclProjection()

template<class Double >

Double umusclProjection	(	const Double &	gradProj,
		const Double &	delta,
		const CNonDeduced< Double > &	kappa
	)

inline

Blended difference for U-MUSCL reconstruction.

Parameters

[in]	gradProj	- Gradient projection at point i: dot(grad_i, vector_ij).
[in]	delta	- Centered difference: V_j - V_i.
[in]	kappa	- Blending parameter.

Returns: Blended difference for reconstruction from point i.

◆ updateEdgeMassFlux()

template<class Double , class Int = typename CLaneTraits<Double>::Int>

void updateEdgeMassFlux	(	Int	iEdge,
		const Double &	massFlux,
		su2activevector *	target
	)

inline

Store the (scalar) mass flux of an edge, e.g. for "bounded scalar" transport equations.

Note: No-op if "target" is null. As with CEdge's Nodes/Normal, edges within a SIMD group are contiguous (coloring groups are multiples of the SIMD size), so this is a plain vectorized store starting at iEdge[0], relying on "target" being padded to a multiple of the SIMD size.

◆ updateLinearSystem() [1/2]

template<size_t nVar, class Double , class Int = typename CLaneTraits<Double>::Int>

void updateLinearSystem	(	Int	iEdge,
		Int	iPoint,
		Int	jPoint,
		bool	implicit,
		UpdateType	updateType,
		Double	updateMask,
		const EdgeResidual< Double, nVar > &	res,
		CSysVector< su2double > &	vector,
		CSysVector< su2double > &	vectorDiff,
		SparseMatrixType &	matrix
	)

inline

Update the matrix and right-hand-side of a linear system with two independent row contributions and four independent Jacobian blocks.

Note: It carries a second CSysVector, the target of flux_j under UpdateType::REDUCTION and unused under COLORING, where both rows are written directly.; The residual is the same under both update types, the Jacobian is not: REDUCTION writes the off-diagonal blocks only and CSysMatrix::SetDiagonalAsColumnSum then derives each diagonal block as minus the sum of its column, which equals the jac_ii and jac_jj computed here only where the flux is conservative (jac_ii == -jac_ji). A model whose diffusion coefficients differ between the two orientations of an edge, i.e. one that evaluates a non-conservative term at the point whose row it is writing, therefore converges along a slightly different path under the reducer, to the same solution.

◆ updateLinearSystem() [2/2]

template<size_t nVar, class Double , class Int = typename CLaneTraits<Double>::Int>

void updateLinearSystem	(	Int	iEdge,
		Int	iPoint,
		Int	jPoint,
		bool	implicit,
		UpdateType	updateType,
		Double	updateMask,
		const Vector< Double, nVar > &	flux,
		const Matrix< Double, nVar, nVar > &	jac_i,
		const Matrix< Double, nVar, nVar > &	jac_j,
		CSysVector< su2double > &	vector,
		SparseMatrixType &	matrix
	)

inline

Update the matrix and right-hand-side of a linear system with one conservative flux.

Variable Documentation

◆ Dynamic

constexpr size_t Dynamic = size_t(-1)

constexpr

Equation count of a model whose value is only known at runtime.

◆ MaxScalarVar

constexpr size_t MaxScalarVar = 8

constexpr

Backing size of the static arrays of a dynamic model.

Note: The scalar numerics cap the equation count at this value and error above it, so a configuration that fits them fits these kernels.

Classes

Typedefs

Enumerations

Functions

Variables

Detailed Description

Typedef Documentation

◆ CNonDeduced

◆ Matrix

◆ SparseMatrixType

◆ Vector

Enumeration Type Documentation

◆ MusclLimiter

◆ UpdateType

Function Documentation

◆ distanceVector() [1/2]

◆ distanceVector() [2/2]

◆ dot() [1/3]

◆ dot() [2/3]

◆ dot() [3/3]

◆ gatherVariables() [1/3]

◆ gatherVariables() [2/3]

◆ gatherVariables() [3/3]

◆ gradientReader()

◆ Max()

◆ muscl()

◆ norm()

◆ reconstruct()

◆ setPreaccOut()

◆ squaredNorm() [1/2]

◆ squaredNorm() [2/2]

◆ stopPreacc()

◆ tangentProjection()

◆ umusclProjection()

◆ updateEdgeMassFlux()

◆ updateLinearSystem() [1/2]

◆ updateLinearSystem() [2/2]

Variable Documentation

◆ Dynamic

◆ MaxScalarVar