sparsedrivers_c.h File Reference

#include <adolc/sparse/sparsedrivers.h>
#include <cstdlib>

Include dependency graph for sparsedrivers_c.h:

Go to the source code of this file.

Functions
int	sparse_jac (short tag, int m, int n, int repeat, const double *x, int *nnz, unsigned int **rind, unsigned int **cind, double **values, int *options)
	Compute and recover a sparse Jacobian using ColPack.
int	sparse_hess (short tag, int n, int repeat, const double *x, int *nnz, unsigned int **rind, unsigned int **cind, double **values, int *options)
	Compute and recover a sparse Hessian using ColPack.
int	jac_pat (short tag, int m, int n, const double *x, unsigned int **JP, int *options)
	Compute a Jacobian sparsity pattern.
int	hess_pat (short tag, int n, const double *x, unsigned int **HP, int *option)
	Compute a scalar function's Hessian sparsity pattern.
void	generate_seed_jac (int m, int n, unsigned int **JP, double ***S, int *p, int *options)
	Generate a ColPack seed matrix for compressed Jacobian recovery.
void	generate_seed_hess (int n, unsigned int **HP, double ***S, int *p, int *options)
	Generate a ColPack seed matrix for compressed Hessian recovery.

Function Documentation

generate_seed_hess()

void generate_seed_hess	(	int	n,
		unsigned int **	HP,
		double ***	S,
		int *	p,
		int *	options
	)

Generate a ColPack seed matrix for compressed Hessian recovery.

Parameters

	n	Number of independent variables.
	HP	Hessian pattern: n rows, each with a count and column indices.
[out]	S	Seed matrix with n rows and p columns.
[out]	p	Number of seed columns.
	options	Two entries; entry 1 selects recovery: 0 indirect or 1 direct.

Note

The caller owns the seed. Delete each row with delete[], then delete the row-pointer array with delete[].

generate_seed_jac()

void generate_seed_jac	(	int	m,
		int	n,
		unsigned int **	JP,
		double ***	S,
		int *	p,
		int *	options
	)

Generate a ColPack seed matrix for compressed Jacobian recovery.

Parameters

	m	Number of dependent variables.
	n	Number of independent variables.
	JP	Jacobian pattern: m rows, each with a count and column indices.
[out]	S	Seed matrix: n x p for column compression or p x m for row compression.
[out]	p	Compressed dimension.
	options	One entry selecting compression: 0 column or 1 row.

Note

The caller owns the seed. Delete each row with delete[], then delete the row-pointer array with delete[].

hess_pat()

int hess_pat	(	short	tag,
		int	n,
		const double *	x,
		unsigned int **	HP,
		int *	option
	)

Compute a scalar function's Hessian sparsity pattern.

Parameters

	tag	Tape identifier.
	n	Number of independent variables (rows and columns).
	x	Evaluation point, with n entries; may be null in safe modes.
[out]	HP	Array of n row pointers. Each allocated row starts with its nonzero count, followed by zero-based column indices.
	option	One entry selecting control flow: 0 safe, 1 tight, 2 old safe, or 3 old tight.

Returns

Nonnegative sweep status on success; negative on failure.

Note

Initialize row pointers to null. Release each allocated row with delete[] before reusing the pointer array; this call resets its entries.

jac_pat()

int jac_pat	(	short	tag,
		int	m,
		int	n,
		const double *	x,
		unsigned int **	JP,
		int *	options
	)

Compute a Jacobian sparsity pattern.

Parameters

	tag	Tape identifier.
	m	Number of dependent variables (rows).
	n	Number of independent variables (columns).
	x	Evaluation point, with n entries; may be null in safe mode.
[out]	JP	Array of m row pointers. Each allocated row starts with its nonzero count, followed by zero-based column indices.
	options	Three entries: method (0 index domains, 1 bit patterns), control flow (0 safe, 1 tight), and bit propagation (0 auto, 1 forward, 2 reverse).

Returns

Nonnegative sweep status on success; negative on failure.

Note

Initialize row pointers to null. Release each allocated row with delete[] before reusing the pointer array; this call resets its entries.

sparse_hess()

int sparse_hess	(	short	tag,
		int	n,
		int	repeat,
		const double *	x,
		int *	nnz,
		unsigned int **	rind,
		unsigned int **	cind,
		double **	values,
		int *	options
	)

Compute and recover a sparse Hessian using ColPack.

Parameters

	tag	Tape identifier for a scalar function.
	n	Number of independent variables.
	repeat	Zero builds and caches the pattern and seed; nonzero reuses them.
	x	Evaluation point, with n entries. Required for numeric recovery.
[in,out]	nnz	Nonzero count set on the first call; must match on reuse.
[in,out]	rind	Zero-based row indices.
[in,out]	cind	Zero-based column indices.
[in,out]	values	Nonzero values.
	options	Two entries: control flow (0 safe, 1 tight, 2 old safe, 3 old tight) and recovery (0 indirect, 1 direct).

Returns

Nonnegative sweep status on success; negative on failure.

Precondition

Pointer arguments other than the three buffer values are non-null.

Note

Supply all three buffers with capacity *nnz to reuse them. Otherwise, the routine deletes any supplied buffers and uses ColPack's unmanaged recovery. Release returned buffers with delete[].

sparse_jac()

int sparse_jac	(	short	tag,
		int	m,
		int	n,
		int	repeat,
		const double *	x,
		int *	nnz,
		unsigned int **	rind,
		unsigned int **	cind,
		double **	values,
		int *	options
	)

Compute and recover a sparse Jacobian using ColPack.

Parameters

	tag	Tape identifier.
	m	Number of dependent variables (rows).
	n	Number of independent variables (columns).
	repeat	Zero builds and caches the pattern and seed; nonzero reuses them.
	x	Evaluation point, with n entries. Required for numeric recovery.
[in,out]	nnz	Recovered nonzero count. On entry, the count of any supplied buffers; they are reused only if all three are non-null and the count matches.
[in,out]	rind	Zero-based row indices.
[in,out]	cind	Zero-based column indices.
[in,out]	values	Nonzero values.
	options	Four entries: method (0 index domains, 1 bit patterns), control flow (0 safe, 1 tight), bit propagation (0 auto, 1 forward, 2 reverse), and compression (0 column, 1 row).

Returns

Nonnegative sweep status on success; negative on failure.

Precondition

nnz, rind, cind, values, and options are non-null.

Note

Set the three buffer pointers to null for allocation. If supplied buffers cannot be reused, the wrapper deletes them and allocates replacements with new[]. Release returned buffers with delete[].