Description

Internal GROMACS namespace.

This namespace is used to contain some implementation-specific functions and classes. These are not meant for direct user access, but typically reside in public headers because of implementation reasons.

Namespaces
	anonymous_namespace{selection.cpp}

Classes
class	AnalysisDataHandleImpl
	Private implementation class for AnalysisDataHandle. More...

class	AnalysisDataStorageImpl
	Private implementation class for AnalysisDataStorage. More...

class	AnalysisDataStorageFrameData
	Internal representation for a single stored frame. More...

class	BasicAverageHistogramModule
	Implements average histogram module that averages per-frame histograms. More...

class	BasicHistogramImpl
	Base class for private implementation classes for histogram modules. More...

class	EnumIndexStore
	Type-specific implementation for IOptionValueStore for an enum option. More...

class	OptionSectionImpl
	Internal implementation class for storing an option section. More...

class	OptionsImpl
	Private implementation class for Options. More...

class	SelectionData
	Internal data for a single selection. More...

struct	SimdTraits
	Simd traits. More...

class	SimdArrayRef
	STL-like container for aligned SIMD type. Used as ArrayRef<SimdReal>. More...

class	IExceptionInfo
	Base class for ExceptionInfo. More...

Typedefs
typedef std::unique_ptr < AnalysisDataStorageFrame >	AnalysisDataFrameBuilderPointer
	Smart pointer type for managing a storage frame builder.

template<typename T >
using	SimdTraitsT = typename SimdTraits< T >::type

template<typename T >
using	Simd4TraitsT = typename Simd4Traits< T >::type

typedef std::vector < std::exception_ptr >	NestedExceptionList
	Internal container type for storing a list of nested exceptions.

typedef std::unique_ptr < IExceptionInfo >	ExceptionInfoPointer
	Smart pointer to manage IExceptionInfo ownership.

Functions
template<typename ValueType >
sycl::event	fillSyclBufferWithNull (sycl::buffer< ValueType, 1 > &buffer, size_t startingOffset, size_t numValues, sycl::queue queue)
	Helper function to clear device buffer. More...

template<>
sycl::event	fillSyclBufferWithNull (sycl::buffer< Float3, 1 > &buffer, size_t startingOffset, size_t numValues, sycl::queue queue)
	Helper function to clear device buffer of type Float3.

template<typename ValueType >
sycl::event	fillSyclUsmWithNull (ValueType *buffer, size_t startingOffset, size_t numValues, sycl::queue queue)

template<>
sycl::event	fillSyclUsmWithNull (Float3 *buffer, size_t startingOffset, size_t numValues, sycl::queue queue)
	Helper function to clear device memory of type Float3.

AbstractOptionStorage *	createEnumOptionStorage (const AbstractOption &option, const char const enumValues, int count, int defaultValue, int defaultValueIfSet, std::unique_ptr< IOptionValueStore< int >> store)
	Helper to create EnumOptionStorage instances. More...

void	throwUnlessDerivativeIsConsistentWithFunction (const std::function< double(double)> &function, const std::function< double(double)> &derivative, const std::pair< real, real > &range)
	Ensure analytical derivative is the derivative of analytical function. More...

void	throwUnlessDerivativeIsConsistentWithFunction (ArrayRef< const double > function, ArrayRef< const double > derivative, double inputSpacing, const std::pair< real, real > &range)
	Ensure vector of derivative values is the derivative of function vector. More...

static double	quotientOfFunctionAndSecondDerivative (double previousPoint, double thisPoint, double nextPoint, double spacing)
	Calculate absolute quotient of function and its second derivative. More...

real	findSmallestQuotientOfFunctionAndSecondDerivative (const std::function< double(double)> &f, const std::pair< real, real > &range)
	Find smallest quotient between analytical function and its 2nd derivative. More...

real	findSmallestQuotientOfFunctionAndSecondDerivative (ArrayRef< const double > function, double inputSpacing, const std::pair< real, real > &range)
	Find smallest quotient between vector of values and its 2nd derivative. More...

static double	quotientOfFunctionAndThirdDerivative (double previousPreviousPoint, double previousPoint, double thisPoint, double nextPoint, double nextNextPoint, double spacing)
	Calculate absolute quotient of function and its third derivative. More...

real	findSmallestQuotientOfFunctionAndThirdDerivative (const std::function< double(double)> &f, const std::pair< real, real > &range)
	Find smallest quotient between analytical function and its 3rd derivative. More...

real	findSmallestQuotientOfFunctionAndThirdDerivative (ArrayRef< const double > function, double inputSpacing, const std::pair< real, real > &range)
	Find smallest quotient between function and 2nd derivative (vectors) More...

std::vector< double >	vectorSecondDerivative (ArrayRef< const double > f, double spacing)
	Calculate second derivative of vector and return vector of same length. More...

template<class T , class U >
void	fillMultiplexedTableData (const T inputData, U *multiplexedOutputData, std::size_t valuesPerTablePoint, std::size_t numTables, std::size_t thisTableIndex)
	Copy (temporary) table data into aligned multiplexed vector. More...

void	printFatalErrorHeader (FILE fp, const char title, const char func, const char file, int line)
	Formats a common header for fatal error messages. More...

void	printFatalErrorMessageLine (FILE fp, const char text, int indent)
	Formats a line of fatal error message text. More...

void	printFatalErrorFooter (FILE *fp)
	Formats a common footer for fatal error messages. More...

void	assertHandler (const char condition, const char msg, const char func, const char file, int line)
	Called when an assert fails. More...

template<typename T >
static void	ignoreValueHelper (const T &)
	Helper for ignoring values in macros. More...

void	current_function_helper ()
	Helper for defining GMX_CURRENT_FUNCTION.

Function Documentation

template<class T , class U >

void gmx::internal::fillMultiplexedTableData	(	const T	inputData,
		U *	multiplexedOutputData,
		std::size_t	valuesPerTablePoint,
		std::size_t	numTables,
		std::size_t	thisTableIndex
	)

Copy (temporary) table data into aligned multiplexed vector.

This routine takes the temporary data generated for a single table and writes multiplexed output into a multiple-table-data vector. If the output vector is empty we will resize it to fit the data, and otherwise we assert the size is correct to add out input data.

Template Parameters

T	Type of container for input data
U	Type of container for output data

Parameters

[in]	inputData	Input data for single table
[in,out]	multiplexedOutputData	Multiplexed output vector, many tables.
[in]	valuesPerTablePoint	Number of real values for each table point, for instance 4 in DDFZ tables.
[in]	numTables	Number of tables mixed into multiplexed output
[in]	thisTableIndex	Index of this table in multiplexed output

Note: The output container type can be different from the input since the latter sometimes uses an aligned allocator so the data can be loaded efficiently in the GROMACS nonbonded kernels.

template<typename ValueType >

sycl::event gmx::internal::fillSyclBufferWithNull	(	sycl::buffer< ValueType, 1 > &	buffer,
		size_t	startingOffset,
		size_t	numValues,
		sycl::queue	queue
	)

Helper function to clear device buffer.

Not applicable to GROMACS's Float3 (a.k.a. gmx::RVec) and other custom types. From SYCL specs: "T must be a scalar value or a SYCL vector type."

real gmx::internal::findSmallestQuotientOfFunctionAndSecondDerivative	(	const std::function< double(double)> &	f,
		const std::pair< real, real > &	range
	)

Find smallest quotient between analytical function and its 2nd derivative.

Used to calculate spacing for quadratic spline tables. This function divides the function value by the second derivative (or a very small number when that is zero), and returns the smallest such quotient found in the range.

Our quadratic tables corresponds to linear interpolation of the derivative, which means the derivative will typically have larger error than the value when interpolating. The spacing required to reach a particular relative tolerance in the derivative depends on the quotient between the first derivative and the third derivative of the function itself.

You should call this routine with the analytical derivative as the "function" parameter, and the quotient between "function and second derivative" will then correspond to the quotient bewteen the derivative and the third derivative of the actual function we want to tabulate.

Since all functions that can be tabulated efficiently are reasonably smooth, we simply check 1,000 points in the interval rather than bother about implementing any complicated global optimization scheme.

Parameters

f	Analytical function
range	Interval

Returns: Smallest quotient found in range.

Note: The function is always double-valued to avoid accuracy loss.

real gmx::internal::findSmallestQuotientOfFunctionAndSecondDerivative	(	ArrayRef< const double >	function,
		double	inputSpacing,
		const std::pair< real, real > &	range
	)

Find smallest quotient between vector of values and its 2nd derivative.

Used to calculate spacing for quadratic spline tables. This function divides the function value by the second derivative (or a very small number when that is zero), and returns the smallest such quotient found in the range.

Our quadratic tables corresponds to linear interpolation of the derivative, which means the derivative will typically have larger error than the value when interpolating. The spacing required to reach a particular relative tolerance in the derivative depends on the quotient between the first derivative and the third derivative of the function itself.

You should call this routine with the analytical derivative as the "function" parameter, and the quotient between "function and second derivative" will then correspond to the quotient bewteen the derivative and the third derivative of the actual function we want to tabulate.

Parameters

function	Vector with function values
inputSpacing	Spacing between function values
range	Interval to check

Returns: Smallest quotient found in range.

Note: The function vector and input spacing are always double-valued to avoid accuracy loss.

real gmx::internal::findSmallestQuotientOfFunctionAndThirdDerivative	(	const std::function< double(double)> &	f,
		const std::pair< real, real > &	range
	)

Find smallest quotient between analytical function and its 3rd derivative.

Used to calculate table spacing. This function divides the function value by the second derivative (or a very small number when that is zero), and returns the smallest such quotient found in the range.

Our quadratic tables corresponds to linear interpolation of the derivative, which means the derivative will typically have larger error than the value when interpolating. The spacing required to reach a particular relative tolerance in the derivative depends on the quotient between the first derivative and the third derivative of the function itself.

You should call this routine with the analytical derivative as the "function" parameter, and the quotient between "function and second derivative" will then correspond to the quotient bewteen the derivative and the third derivative of the actual function we want to tabulate.

Since all functions that can be tabulated efficiently are reasonably smooth, we simply check 1,000 points in the interval rather than bother about implementing any complicated global optimization scheme.

Parameters

f	Analytical function
range	Interval

Returns: Smallest quotient found in range.

Note: The function is always double-valued to avoid accuracy loss.

real gmx::internal::findSmallestQuotientOfFunctionAndThirdDerivative	(	ArrayRef< const double >	function,
		double	inputSpacing,
		const std::pair< real, real > &	range
	)

Find smallest quotient between function and 2nd derivative (vectors)

Used to calculate table spacing. This function divides the function value by the second derivative (or a very small number when that is zero), and returns the smallest such quotient found in the range.

Our quadratic tables corresponds to linear interpolation of the derivative, which means the derivative will typically have larger error than the value when interpolating. The spacing required to reach a particular relative tolerance in the derivative depends on the quotient between the first derivative and the third derivative of the function itself.

You should call this routine with the analytical derivative as the "function" parameter, and the quotient between "function and second derivative" will then correspond to the quotient bewteen the derivative and the third derivative of the actual function we want to tabulate.

Parameters

function	Vector with function values
inputSpacing	Spacing between function values
range	Interval to check

Returns: Smallest quotient found in range.

Note: The function vector and input spacing are always double-valued to avoid accuracy loss.

static double gmx::internal::quotientOfFunctionAndSecondDerivative	(	double	previousPoint,
		double	thisPoint,
		double	nextPoint,
		double	spacing
	)

static

Calculate absolute quotient of function and its second derivative.

This is a utility function used in the functions to find the smallest quotient in a range.

Parameters

[in]	previousPoint	Value of function at x-h.
[in]	thisPoint	Value of function at x.
[in]	nextPoint	Value of function at x+h.
[in]	spacing	Value of h.

Returns: The absolute value of the quotient. If either the function or second derivative is smaller than sqrt(GMX_REAL_MIN), they will be set to that value.

static double gmx::internal::quotientOfFunctionAndThirdDerivative	(	double	previousPreviousPoint,
		double	previousPoint,
		double	thisPoint,
		double	nextPoint,
		double	nextNextPoint,
		double	spacing
	)

static

Calculate absolute quotient of function and its third derivative.

This is a utility function used in the functions to find the smallest quotient in a range.

Parameters

[in]	previousPreviousPoint	Value of function at x-2h.
[in]	previousPoint	Value of function at x-h.
[in]	thisPoint	Value of function at x.
[in]	nextPoint	Value of function at x+h.
[in]	nextNextPoint	Value of function at x+2h.
[in]	spacing	Value of h.

Returns: The absolute value of the quotient. If either the function or third derivative is smaller than sqrt(GMX_REAL_MIN), they will be set to that value.

void gmx::internal::throwUnlessDerivativeIsConsistentWithFunction	(	const std::function< double(double)> &	function,
		const std::function< double(double)> &	derivative,
		const std::pair< real, real > &	range
	)

Ensure analytical derivative is the derivative of analytical function.

This routine evaluates the numerical derivative of the function for a few (1000) points in the interval and checks that the relative difference between numerical and analytical derivative is within the expected error for the numerical derivative approximation we use.

The main point of this routine is to make sure the user has not made a mistake or sign error when defining the functions.

Parameters

function	Analytical function to differentiate
derivative	Analytical derivative to compare with
range	Range to test

Exceptions

If	the provided derivative does not seem to match the function.

Note: The function/derivative are always double-valued to avoid accuracy loss.

void gmx::internal::throwUnlessDerivativeIsConsistentWithFunction	(	ArrayRef< const double >	function,
		ArrayRef< const double >	derivative,
		double	inputSpacing,
		const std::pair< real, real > &	range
	)

Ensure vector of derivative values is the derivative of function vector.

This routine differentiates a vector of numerical values and checks that the relative difference to a provided vector of numerical derivatives is smaller than the expected error from the numerical differentiation.

The main point of this routine is to make sure the user has not made a mistake or sign error when defining the functions.

To avoid problems if the vectors change from zero to finite values at the start/end of the interval, we only check inside the range requested.

Parameters

function	Numerical function value vector to differentiate
derivative	Numerical derivative vector to compare with
inputSpacing	Distance between input points
range	Range to test

Exceptions

If	the provided derivative does not seem to match the function.

Note: The function/derivative vectors and spacing are always double-valued to avoid accuracy loss.

std::vector< double > gmx::internal::vectorSecondDerivative	(	ArrayRef< const double >	f,
		double	spacing
	)

Calculate second derivative of vector and return vector of same length.

5-point approximations are used, with endpoints using non-center interpolation.

Parameters

f	Vector (function) for which to calculate second derivative
spacing	Spacing of input data.

Exceptions

If	the input vector has fewer than five data points.

Note: This function always uses double precision arguments since it is meant to be used on raw user input data for tables, where we want to avoid accuracy loss (since differentiation can be numerically fragile).

Description

Namespaces

Classes

Typedefs

Functions

Function Documentation