org.nd4j.linalg.api.blas

Interface Level1

All Known Implementing Classes:

BaseLevel1, SparseBaseLevel1

public interface Level1

Level 1 blas implementations. Incx and other parameters are inferred from the given ndarrays. To avoid boxing, doubles are used in place of normal numbers. The underlying implementation will call the proper data opType. This is a fortran 95 style api that gives us the efficiency and flexibility of the fortran 77 api Credit to: https://www.ualberta.ca/AICT/RESEARCH/LinuxClusters/doc/mkl81/mklqref/blaslev1.htm for the descriptions

Author:

Adam Gibson

Method Summary

Modifier and Type	Method and Description
IComplexNumber	asum(IComplexNDArray arr) computes the sum of magnitudes of all vector elements or, for a complex vector x, the sum
double	asum(INDArray arr) computes the sum of magnitudes of all vector elements or, for a complex vector x, the sum
double	asum(int n, DataBuffer x, int offsetX, int incrX) sum of magnitudes of all elements
void	axpy(int n, double alpha, DataBuffer x, int offsetX, int incrX, DataBuffer y, int offsetY, int incrY) computes a vector-scalar product and adds the result to a vector.
void	axpy(int n, double alpha, INDArray x, INDArray y) computes a vector-scalar product and adds the result to a vector.
void	axpy(int n, IComplexNumber alpha, IComplexNDArray x, IComplexNDArray y) computes a vector-scalar product and adds the result to a vector.
void	copy(IComplexNDArray x, IComplexNDArray y)
void	copy(INDArray x, INDArray y) copy a vector to another vector.
void	copy(int n, DataBuffer x, int offsetX, int incrX, DataBuffer y, int offsetY, int incrY) copy a vector to another vector.
double	dot(int n, DataBuffer dx, int offsetX, int incrX, DataBuffer y, int offsetY, int incrY) Vector-vector dot product
double	dot(int n, double alpha, INDArray X, INDArray Y) computes a vector-vector dot product.
IComplexNumber	dot(int n, IComplexNumber alpha, IComplexNDArray X, IComplexNDArray Y) computes a vector-vector dot product.
int	iamax(IComplexNDArray arr) finds the element of a vector that has the largest absolute value.
int	iamax(INDArray arr) finds the element of a vector that has the largest absolute value.
int	iamax(int n, DataBuffer x, int offsetX, int incrX) Index of largest absolute value
int	iamax(int n, INDArray arr, int stride) finds the element of a vector that has the largest absolute value.
int	iamin(IComplexNDArray arr) finds the element of a vector that has the minimum absolute value.
int	iamin(INDArray arr) finds the element of a vector that has the minimum absolute value.
IComplexNumber	nrm2(IComplexNDArray arr) computes the Euclidean norm of a vector.
double	nrm2(INDArray arr) computes the Euclidean norm of a vector.
void	rot(int N, IComplexNDArray X, IComplexNDArray Y, IComplexNumber c, IComplexNumber s) performs rotation of points in the plane.
void	rot(int N, INDArray X, INDArray Y, double c, double s) performs rotation of points in the plane.
void	rotg(INDArray a, INDArray b, INDArray c, INDArray s) computes parameters for a Givens rotation.
void	rotmg(IComplexNDArray d1, IComplexNDArray d2, IComplexNDArray b1, IComplexNumber b2, IComplexNDArray P) computes the modified parameters for a Givens rotation.
void	rotmg(INDArray d1, INDArray d2, INDArray b1, double b2, INDArray P) computes the modified parameters for a Givens rotation.
void	scal(int N, double alpha, INDArray X) computes a vector by a scalar product.
void	scal(int N, IComplexNumber alpha, IComplexNDArray X) computes a vector by a scalar product.
boolean	supportsDataBufferL1Ops() Can we use the axpy and copy methods that take a DataBuffer instead of an INDArray with this backend?
void	swap(IComplexNDArray x, IComplexNDArray y)
void	swap(INDArray x, INDArray y) swaps a vector with another vector.

Method Detail

dot

double dot(int n,
           double alpha,
           INDArray X,
           INDArray Y)

computes a vector-vector dot product.

Parameters:

n -

alpha -

X -

Y -

Returns:

dot

double dot(int n,
           DataBuffer dx,
           int offsetX,
           int incrX,
           DataBuffer y,
           int offsetY,
           int incrY)

Vector-vector dot product

dot

IComplexNumber dot(int n,
                   IComplexNumber alpha,
                   IComplexNDArray X,
                   IComplexNDArray Y)

computes a vector-vector dot product.

Parameters:

n -

alpha -

X -

Y -

Returns:

nrm2

double nrm2(INDArray arr)

computes the Euclidean norm of a vector.

Parameters:

arr -

Returns:

nrm2

IComplexNumber nrm2(IComplexNDArray arr)

computes the Euclidean norm of a vector.

Parameters:

arr -

Returns:

asum

double asum(INDArray arr)

computes the sum of magnitudes of all vector elements or, for a complex vector x, the sum

Parameters:

arr -

Returns:

asum

double asum(int n,
            DataBuffer x,
            int offsetX,
            int incrX)

sum of magnitudes of all elements

asum

IComplexNumber asum(IComplexNDArray arr)

computes the sum of magnitudes of all vector elements or, for a complex vector x, the sum

Parameters:

arr -

Returns:

iamax

int iamax(INDArray arr)

finds the element of a vector that has the largest absolute value.

Parameters:

arr -

Returns:

iamax

int iamax(int n,
          INDArray arr,
          int stride)

finds the element of a vector that has the largest absolute value.

Parameters:

n - the length to iterate for

arr - the array to get the max index for

stride - the stride for the array

Returns:

iamax

int iamax(int n,
          DataBuffer x,
          int offsetX,
          int incrX)

Index of largest absolute value

iamax

int iamax(IComplexNDArray arr)

finds the element of a vector that has the largest absolute value.

Parameters:

arr -

Returns:

iamin

int iamin(INDArray arr)

finds the element of a vector that has the minimum absolute value.

Parameters:

arr -

Returns:

iamin

int iamin(IComplexNDArray arr)

finds the element of a vector that has the minimum absolute value.

Parameters:

arr -

Returns:

swap

void swap(INDArray x,
          INDArray y)

swaps a vector with another vector.

Parameters:

x -

y -

swap

void swap(IComplexNDArray x,
          IComplexNDArray y)

copy

void copy(INDArray x,
          INDArray y)

copy a vector to another vector.

Parameters:

x -

y -

copy

void copy(int n,
          DataBuffer x,
          int offsetX,
          int incrX,
          DataBuffer y,
          int offsetY,
          int incrY)

copy a vector to another vector.

Parameters:

x -

y -

copy

void copy(IComplexNDArray x,
          IComplexNDArray y)

axpy

void axpy(int n,
          double alpha,
          INDArray x,
          INDArray y)

computes a vector-scalar product and adds the result to a vector.

Parameters:

n -

alpha -

x -

y -

axpy

void axpy(int n,
          double alpha,
          DataBuffer x,
          int offsetX,
          int incrX,
          DataBuffer y,
          int offsetY,
          int incrY)

computes a vector-scalar product and adds the result to a vector. y = a*x + y

Parameters:

n - number of operations

alpha -

x - X

offsetX - offset of first element of X in buffer

incrX - increment/stride between elements of X in buffer

y - Y

offsetY - offset of first element of Y in buffer

incrY - increment/stride between elements of Y in buffer

axpy

void axpy(int n,
          IComplexNumber alpha,
          IComplexNDArray x,
          IComplexNDArray y)

computes a vector-scalar product and adds the result to a vector.

Parameters:

n -

alpha -

x -

y -

rotg

void rotg(INDArray a,
          INDArray b,
          INDArray c,
          INDArray s)

computes parameters for a Givens rotation.

Parameters:

a -

b -

c -

s -

rot

void rot(int N,
         INDArray X,
         INDArray Y,
         double c,
         double s)

performs rotation of points in the plane.

Parameters:

N -

X -

Y -

c -

s -

rot

void rot(int N,
         IComplexNDArray X,
         IComplexNDArray Y,
         IComplexNumber c,
         IComplexNumber s)

performs rotation of points in the plane.

Parameters:

N -

X -

Y -

c -

s -

rotmg

void rotmg(INDArray d1,
           INDArray d2,
           INDArray b1,
           double b2,
           INDArray P)

computes the modified parameters for a Givens rotation.

Parameters:

d1 -

d2 -

b1 -

b2 -

P -

rotmg

void rotmg(IComplexNDArray d1,
           IComplexNDArray d2,
           IComplexNDArray b1,
           IComplexNumber b2,
           IComplexNDArray P)

computes the modified parameters for a Givens rotation.

Parameters:

d1 -

d2 -

b1 -

b2 -

P -

scal

void scal(int N,
          double alpha,
          INDArray X)

computes a vector by a scalar product.

Parameters:

N -

alpha -

X -

scal

void scal(int N,
          IComplexNumber alpha,
          IComplexNDArray X)

computes a vector by a scalar product.

Parameters:

N -

alpha -

X -

supportsDataBufferL1Ops

boolean supportsDataBufferL1Ops()

Can we use the axpy and copy methods that take a DataBuffer instead of an INDArray with this backend?