Abstract interface to the basic linear algebra routines used by OpenNL internally. More...

#include "nl_private.h"

Classes
struct	NLBlas

Macros
#define	NL_NEW_VECTOR(blas, memtype, dim) (double)blas->Malloc(blas,memtype,(size_t)(dim)sizeof(double))
	Allocates a vector of doubles;. More...

#define	NL_DELETE_VECTOR(blas, memtype, dim, ptr) blas->Free(blas,memtype,(size_t)(dim)*sizeof(double),ptr)
	Deletes a vector of doubles;. More...

Typedefs
typedef struct NLBlas *	NLBlas_t
	A handle to a BLAS abstraction layer.

typedef void (	FUNPTR_malloc) (NLBlas_t blas, NLmemoryType type, size_t size)
	Allocates memory in host or in device. More...

typedef void(*	FUNPTR_free) (NLBlas_t blas, NLmemoryType type, size_t size, void *ptr)
	Frees memory from host or from device. More...

typedef void(*	FUNPTR_memcpy) (NLBlas_t blas, void to, NLmemoryType to_type, void from, NLmemoryType from_type, size_t size)
	Copies a bloc of memory. More...

typedef void(*	FUNPTR_memset) (NLBlas_t blas, void *to, NLmemoryType to_type, int c, size_t size)
	Sets a bloc of memory. More...

typedef void(*	FUNPTR_dcopy) (NLBlas_t blas, int n, const double x, int incx, double y, int incy)
	Copies a vector. More...

typedef void(*	FUNPTR_dscal) (NLBlas_t blas, int n, double a, double *x, int incx)
	Scales a vector. More...

typedef double(*	FUNPTR_ddot) (NLBlas_t blas, int n, const double x, int incx, const double y, int incy)
	Computes the dot product between two vectors. More...

typedef double(*	FUNPTR_dnrm2) (NLBlas_t blas, int n, const double *x, int incx)
	Computes the norm of a vector. More...

typedef void(*	FUNPTR_daxpy) (NLBlas_t blas, int n, double a, const double x, int incx, double y, int incy)
	Computes a linear combination of two vectors. More...

typedef void(*	FUNPTR_dmul) (NLBlas_t blas, int n, const double x, const double y, double *z)
	Component-wise multiplication. More...

typedef void(*	FUNPTR_dgemv) (NLBlas_t blas, MatrixTranspose trans, int m, int n, double alpha, const double A, int ldA, const double x, int incx, double beta, double *y, int incy)
	Computes a matrix-vector product. More...

typedef void(*	FUNPTR_dtpsv) (NLBlas_t blas, MatrixTriangle uplo, MatrixTranspose trans, MatrixUnitTriangular diag, int n, const double AP, double x, int incx)
	Solves a linear system. More...

typedef void(*	FUNPTR_show_stats) (NLBlas_t blas)
	Displays the statistics associated with a BLAS context. More...

typedef void(*	FUNPTR_reset_stats) (NLBlas_t blas)
	Resets the statistics associated with a BLAS context. More...

Enumerations
enum	MatrixTranspose { NoTranspose =0 , Transpose =1 , ConjugateTranspose =2 }
	Specifies whether matrix should be transposed. More...

enum	MatrixTriangle { UpperTriangle =0 , LowerTriangle =1 }
	Specifies which triangular part of a matrix should be used. More...

enum	MatrixUnitTriangular { UnitTriangular =0 , NotUnitTriangular =1 }
	Specifies which triangular part of a matrix should be used. More...

enum	NLmemoryType { NL_HOST_MEMORY , NL_DEVICE_MEMORY }
	Specifies on which type of memory a function should be applied. More...

Functions
NLAPI NLboolean NLAPIENTRY	nlBlasHasUnifiedMemory (NLBlas_t blas)
	Tests whether BLAS device uses the same address space as the CPU. More...

NLAPI void NLAPIENTRY	nlBlasResetStats (NLBlas_t blas)
	Resets the statistics associated with a BLAS context. More...

NLAPI void NLAPIENTRY	nlBlasShowStats (NLBlas_t blas)
	Displays the statistics associated with a BLAS context. More...

NLAPI double NLAPIENTRY	nlBlasGFlops (NLBlas_t blas)
	Gets the number of floating point operations per seconds computed since creation of the BLAS abstaction layer or since the latest call to nlBlasResetStats(). More...

NLAPI NLulong NLAPIENTRY	nlBlasUsedRam (NLBlas_t blas, NLmemoryType type)
	Gets the currently used amount of memory computed since creation of the BLAS abstaction layer or since the latest call to nlBlasResetStats(). More...

NLAPI NLulong NLAPIENTRY	nlBlasMaxUsedRam (NLBlas_t blas, NLmemoryType type)
	Gets the high mark of used amount of memory computed since creation of the BLAS abstaction layer or since the latest call to nlBlasResetStats(). More...

NLAPI NLBlas_t NLAPIENTRY	nlHostBlas (void)
	Gets a pointer to the BLAS abstraction layer for BLAS operation on the host CPU. More...

Detailed Description

Abstract interface to the basic linear algebra routines used by OpenNL internally.

Definition in file nl_blas.h.

Macro Definition Documentation

◆ NL_DELETE_VECTOR

#define NL_DELETE_VECTOR	(	blas,
		memtype,
		dim,
		ptr
	)	blas->Free(blas,memtype,(size_t)(dim)*sizeof(double),ptr)

Deletes a vector of doubles;.

Parameters

[in]	blas	a pointer to the BLAS abstraction layer.
[in]	memtype	one of NL_HOST_MEMORY, NL_DEVICE_MEMORY.
[in]	dim	number of components in the vector.
[in]	ptr	a pointer to the vector to be deleted.

Definition at line 460 of file nl_blas.h.

◆ NL_NEW_VECTOR

#define NL_NEW_VECTOR	(	blas,
		memtype,
		dim
	)	(double)blas->Malloc(blas,memtype,(size_t)(dim)sizeof(double))

Allocates a vector of doubles;.

Parameters

[in]	blas	a pointer to the BLAS abstraction layer.
[in]	memtype	one of NL_HOST_MEMORY, NL_DEVICE_MEMORY.
[in]	dim	number of components in the vector.

Definition at line 450 of file nl_blas.h.

Typedef Documentation

◆ FUNPTR_daxpy

typedef void(* FUNPTR_daxpy) (NLBlas_t blas, int n, double a, const double *x, int incx, double *y, int incy)

Computes a linear combination of two vectors.

In formula: \( y \leftarrow a x + y \)

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	n	dimension of the vectors
[in]	a	scaling coefficient
[in]	x	source vector to be scaled and added
[in]	incx	straddle between two consecutive coefficients of the source vector
[in,out]	y	the vector that `a` `x` should be added to
[in]	incy	straddle between two consecutive coefficients of the destination vector

Definition at line 230 of file nl_blas.h.

◆ FUNPTR_dcopy

typedef void(* FUNPTR_dcopy) (NLBlas_t blas, int n, const double *x, int incx, double *y, int incy)

Copies a vector.

In formula: \( y \leftarrow x \)

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	n	dimension of the vector
[in]	x	source vector
[in]	incx	straddle between two consecutive coefficients of the source vector
[out]	y	destination vector
[in]	incy	straddle between two consecutive coefficients of the destination vector

Definition at line 169 of file nl_blas.h.

◆ FUNPTR_ddot

typedef double(* FUNPTR_ddot) (NLBlas_t blas, int n, const double *x, int incx, const double *y, int incy)

Computes the dot product between two vectors.

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	n	dimension of the vectors
[in]	x	first vector
[in]	incx	straddle between two consecutive coefficients of the first vector
[in]	y	second vector
[in]	incy	straddle between two consecutive coefficients of the second vector

Returns: the dot product between x and y

Definition at line 199 of file nl_blas.h.

◆ FUNPTR_dgemv

typedef void(* FUNPTR_dgemv) (NLBlas_t blas, MatrixTranspose trans, int m, int n, double alpha, const double *A, int ldA, const double *x, int incx, double beta, double *y, int incy)

Computes a matrix-vector product.

performs one of the matrix-vector operations \( y = \alpha A x + \beta y \) or \( y = \alpha A^t x + \beta y \), where \( \alpha \) and \( \beta \) are scalars, x and y are vectors and A is an m by n matrix.

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	trans	one of (NoTranspose, Transpose, ConjugateTranspose), specifies the operation to be performed as follows: NoTranspose: \( y = \alpha A x + \beta y \) Transpose: \( y = \alpha A^t x + \beta y \) ConjugateTranspose: \( y = \alpha A^t x + \beta y \)
[in]	m	number of rows of the matrix A
[in]	n	number of columns of the matrix A
[in]	alpha	the scalar \( \alpha \)
[in]	A	an array of dimension (ldA,n). On entry, the leading m by n part of the array A must contain the array of coefficients.
[in]	ldA	specifies the first dimension of A as declared by the caller. ldA must be at least max(1,m).
[in]	x	array of dimension at least ( 1 + ( n - 1 )abs( incx ) ) when trans = NoTranspose and at least ( 1 + ( m - 1 )abs( incx ) ) otherwise. Before entry, the incremented array X must contain the vector x.
[in]	incx	the increment for the elements of x
[in]	beta	the scalar \( \beta \)
[in,out]	y	an array of dimension ( 1 + ( m - 1 )abs( incy ) ) when trans = NoTranspose and at least ( 1 + ( n - 1 )abs( incy ) ) otherwise. Before entry with beta non-zero, the incremented array y must contain the vector y. On exit, y is overwritten by the updated vector y.
[in]	incy	the increment for the elements of y

Definition at line 285 of file nl_blas.h.

◆ FUNPTR_dmul

typedef void(* FUNPTR_dmul) (NLBlas_t blas, int n, const double *x, const double *y, double *z)

Component-wise multiplication.

In formula: \( z \leftarrow x ** y \)

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	n	dimension of the vectors
[in]	x,y	source vectors
[in]	z	destination vector

Definition at line 244 of file nl_blas.h.

◆ FUNPTR_dnrm2

typedef double(* FUNPTR_dnrm2) (NLBlas_t blas, int n, const double *x, int incx)

Computes the norm of a vector.

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	n	dimension of the vector
[in]	x	the vector
[in]	incx	straddle between two consecutive coefficients of the vector

Returns: the norm of x

Definition at line 213 of file nl_blas.h.

◆ FUNPTR_dscal

typedef void(* FUNPTR_dscal) (NLBlas_t blas, int n, double a, double *x, int incx)

Scales a vector.

In formula: \( x \leftarrow a x \)

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	n	dimension of the vector
[in]	a	scaling coefficient
[in,out]	x	vector to be scaled
[in]	incx	straddle between two consecutive coefficients

Definition at line 182 of file nl_blas.h.

◆ FUNPTR_dtpsv

typedef void(* FUNPTR_dtpsv) (NLBlas_t blas, MatrixTriangle uplo, MatrixTranspose trans, MatrixUnitTriangular diag, int n, const double *AP, double *x, int incx)

Solves a linear system.

Solves one of the systems of equations \( A x = b \) or \( A^t x = b \) where b and x are n element vectors and A is an n by n unit, or non-unit, upper or lower triangular matrix, supplied in packed form. No test for singularity or near-singularity is included in this routine. Such tests must be performed before calling this routine.

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	uplo	one of (UpperTriangle, LowerTriangle), specifies whether A is an upper or lower triangular matrix as follows: UpperTriangle: A is an upper triangular matrix LowerTriangle: A is a lower triangular matrix
[in]	trans	one of (NoTranspose, Transpose, ConjugateTranspose), specifies the equations to be solved as follows: NoTranspose: \( A x = b \) Transpose: \( A^t x = b \) ConjugateTranspose: \( A^t x = b \)
[in]	diag	one of (UnitTriangular, NotUnitTriangular), specifies whether or not A is unit triangular as follows: UnitTriangular: A is assumed to be unit triangular NotUnitTriangular: A is not assumed to be unit triangular
[in]	n	the order of the matrix
[in]	AP	an array of dimension at least ( ( n*( n + 1))/2). if uplo = UpperTriangular, the array AP must contain the upper triangular matrix packed sequentially, column by column, so that AP[0] contains a(1,1), AP[1] and AP[2] contain a(1,2) and a(2,2) respectively, and so on. if uplo = LowerTriangular, the array AP must contain the lower triangular matrix packed sequentially, column by column, so that AP[0] contains a(1,1 ), AP[1] and AP[2] contain a(2,1) and a(3,1) respectively, and so on. Note that when diag = UnitTriangular, the diagonal elements of A are not referenced, but are assumed to be unity.
[in,out]	x	array of dimension at least (1 + (n-1)*abs( incx )). Before entry, the incremented array x must contain the n element right-hand side vector b. On exit, x is overwritten with the solution vector x.
[in]	incx	specifies the increment for the elements of x. Must not be zero.

Definition at line 333 of file nl_blas.h.

◆ FUNPTR_free

typedef void(* FUNPTR_free) (NLBlas_t blas, NLmemoryType type, size_t size, void *ptr)

Frees memory from host or from device.

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	type	whether RAM should be allocated on the CPU (NL_HOST_MEMORY) or the GPU (NL_DEVICE_MEMORY)
[in]	size	the number of bytes of the memory bloc to be allocated.
[in]	ptr	a pointer to the memory to be freed.

Definition at line 120 of file nl_blas.h.

◆ FUNPTR_malloc

typedef void*(* FUNPTR_malloc) (NLBlas_t blas, NLmemoryType type, size_t size)

Allocates memory in host or in device.

Parameters

[in]	blas	a handle to the BLAS abstraction layer
[in]	type	whether RAM should be allocated on the CPU (NL_HOST_MEMORY) or the GPU (NL_DEVICE_MEMORY)
[in]	size	the number of bytes of the memory bloc to be allocated.

Returns: a pointer to the allocated memory.

Definition at line 107 of file nl_blas.h.

◆ FUNPTR_memcpy

typedef void(* FUNPTR_memcpy) (NLBlas_t blas, void *to, NLmemoryType to_type, void *from, NLmemoryType from_type, size_t size)

Copies a bloc of memory.

Parameters

[in]	blas	a handle to the BLAS abstraction layer.
[in]	to,to_type	a pointer to the destination and indication of whether the destination is on CPU RAM (NL_HOST_MEMORY) or GPU RAM (NL_DEVICE_MEMORY).
[in]	from,frm_type	a pointer to the source and indication of whether the source is on CPU RAM (NL_HOST_MEMORY) or GPU RAM (NL_DEVICE_MEMORY).
[in]	size	number of bytes to be copied.

Definition at line 135 of file nl_blas.h.

◆ FUNPTR_memset

typedef void(* FUNPTR_memset) (NLBlas_t blas, void *to, NLmemoryType to_type, int c, size_t size)

Sets a bloc of memory.

Parameters

[in]	blas	a handle to the BLAS abstraction layer.
[in]	to,to_type	a pointer to the destination and indication of whether the destination is on CPU RAM (NL_HOST_MEMORY) or GPU RAM (NL_DEVICE_MEMORY).
[in]	c	value of each byte to be set.
[in]	size	number of bytes to set.

Definition at line 151 of file nl_blas.h.

◆ FUNPTR_reset_stats

typedef void(* FUNPTR_reset_stats) (NLBlas_t blas)

Resets the statistics associated with a BLAS context.

Parameters

[in] blas a pointer to the BLAS abstraction layer.

Definition at line 351 of file nl_blas.h.

◆ FUNPTR_show_stats

typedef void(* FUNPTR_show_stats) (NLBlas_t blas)

Displays the statistics associated with a BLAS context.

Parameters

[in] blas a pointer to the BLAS abstraction layer.

Definition at line 345 of file nl_blas.h.

Enumeration Type Documentation

◆ MatrixTranspose

enum MatrixTranspose

Specifies whether matrix should be transposed.

Used by dtpsv() and dgemv()

Definition at line 69 of file nl_blas.h.

◆ MatrixTriangle

enum MatrixTriangle

Specifies which triangular part of a matrix should be used.

Used by dtpsv()

Definition at line 77 of file nl_blas.h.

◆ MatrixUnitTriangular

enum MatrixUnitTriangular

Specifies which triangular part of a matrix should be used.

Used by dtpsv()

Definition at line 85 of file nl_blas.h.

◆ NLmemoryType

enum NLmemoryType

Specifies on which type of memory a function should be applied.

NL_HOST_MEMORY refers to CPU RAM and NL_DEVICE_MEMORY to GPU RAM.

Definition at line 94 of file nl_blas.h.

Function Documentation

◆ nlBlasGFlops()

NLAPI double NLAPIENTRY nlBlasGFlops ( NLBlas_t blas )

Gets the number of floating point operations per seconds computed since creation of the BLAS abstaction layer or since the latest call to nlBlasResetStats().

Returns: the number of GFlops.

◆ nlBlasHasUnifiedMemory()

NLAPI NLboolean NLAPIENTRY nlBlasHasUnifiedMemory ( NLBlas_t blas )

Tests whether BLAS device uses the same address space as the CPU.

Parameters

[in] blas a pointer to the BLAS abstraction layer.

Returns: NL_TRUE if device and CPU use the same address space.

Return values

NL_FALSE otherwise.

◆ nlBlasMaxUsedRam()

NLAPI NLulong NLAPIENTRY nlBlasMaxUsedRam	(	NLBlas_t	blas,
		NLmemoryType	type
	)

Gets the high mark of used amount of memory computed since creation of the BLAS abstaction layer or since the latest call to nlBlasResetStats().

Parameters

[in]	blas	a pointer to the BLAS abstraction layer.
[in]	type	one of NL_HOST_MEMORY, NL_DEVICE_MEMORY.

Returns: the high mark of the used memory, in bytes.

◆ nlBlasResetStats()

NLAPI void NLAPIENTRY nlBlasResetStats ( NLBlas_t blas )

Resets the statistics associated with a BLAS context.

Parameters

[in] blas a pointer to the BLAS abstraction layer.

◆ nlBlasShowStats()

NLAPI void NLAPIENTRY nlBlasShowStats ( NLBlas_t blas )

Displays the statistics associated with a BLAS context.

Parameters

[in] blas a pointer to the BLAS abstraction layer.

◆ nlBlasUsedRam()

NLAPI NLulong NLAPIENTRY nlBlasUsedRam	(	NLBlas_t	blas,
		NLmemoryType	type
	)

Gets the currently used amount of memory computed since creation of the BLAS abstaction layer or since the latest call to nlBlasResetStats().

Parameters

[in]	blas	a pointer to the BLAS abstraction layer.
[in]	type	one of NL_HOST_MEMORY, NL_DEVICE_MEMORY.

Returns: the amount of used memory, in bytes.

◆ nlHostBlas()

NLAPI NLBlas_t NLAPIENTRY nlHostBlas ( void )

Gets a pointer to the BLAS abstraction layer for BLAS operation on the host CPU.

Returns: a pointer to the BLAS abstraction layer.

Classes

Macros

Typedefs

Enumerations

Functions

Detailed Description

Macro Definition Documentation

◆ NL_DELETE_VECTOR

◆ NL_NEW_VECTOR

Typedef Documentation

◆ FUNPTR_daxpy

◆ FUNPTR_dcopy

◆ FUNPTR_ddot

◆ FUNPTR_dgemv

◆ FUNPTR_dmul

◆ FUNPTR_dnrm2

◆ FUNPTR_dscal

◆ FUNPTR_dtpsv

◆ FUNPTR_free

◆ FUNPTR_malloc

◆ FUNPTR_memcpy

◆ FUNPTR_memset

◆ FUNPTR_reset_stats

◆ FUNPTR_show_stats

Enumeration Type Documentation

◆ MatrixTranspose

◆ MatrixTriangle

◆ MatrixUnitTriangular

◆ NLmemoryType

Function Documentation

◆ nlBlasGFlops()

◆ nlBlasHasUnifiedMemory()

◆ nlBlasMaxUsedRam()

◆ nlBlasResetStats()

◆ nlBlasShowStats()

◆ nlBlasUsedRam()

◆ nlHostBlas()