org.nd4j.linalg.jcublas.util

Class PointerUtil

java.lang.Object

org.nd4j.linalg.jcublas.util.PointerUtil

public class PointerUtil
extends Object

Various methods for pointer based methods (mainly for the jcuda executioner)

Author:

Adam Gibson

Constructor Summary

Constructors

Constructor and Description
PointerUtil()

Method Summary

Modifier and Type	Method and Description
static CUdeviceptr	constructAndAlloc(int length, DataBuffer.Type dType) Construct and allocate a device pointer
static int	getNumBlocks(int n, int maxBlocks, int maxThreads) Compute the number of blocks that should be used for the given input size and limits
static int	getNumThreads(int n, int maxThreads) Compute the number of threads that should be used for the given input size and limits
static Pointer	getPointer(cuComplex x) Get the pointer for a single complex float
static Pointer	getPointer(cuDoubleComplex x) Get the pointer for a single complex float
static Pointer	getPointer(double alpha)
static Pointer	getPointer(float alpha)
static Pointer	getPointer(IComplexDouble x) Get the pointer for a single complex float
static Pointer	getPointer(IComplexFloat x) Get the pointer for a single complex float
static Object	getPointer(ScalarOp scalarOp)
static int	nextPow2(int x) Returns the power of 2 that is equal to or greater than x
static int	sizeFor(DataBuffer.Type dataType)
static double[]	toDoubles(Object[] extraArgs)
static float[]	toFloats(Object[] extraArgs)
static int[]	toShapeInfoBuffer(INDArray arr) Converts a raw int buffer of the layout: rank shape stride offset element wise stride ordering where shape and stride are both straight int pointers Of note here is that offset will be zero automatically because the offset is handled by the pointer object instead
static int[]	toShapeInfoBuffer(INDArray arr, int... dimension) Converts a raw int buffer of the layout: rank shape stride offset element wise stride ordering where shape and stride are both straight int pointers Of note here is that offset will be zero automatically because the offset is handled by the pointer object instead

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

PointerUtil

public PointerUtil()

Method Detail

toDoubles

public static double[] toDoubles(Object[] extraArgs)

toShapeInfoBuffer

public static int[] toShapeInfoBuffer(INDArray arr,
                                      int... dimension)

Converts a raw int buffer of the layout: rank shape stride offset element wise stride ordering where shape and stride are both straight int pointers Of note here is that offset will be zero automatically because the offset is handled by the pointer object instead

toShapeInfoBuffer

public static int[] toShapeInfoBuffer(INDArray arr)

Converts a raw int buffer of the layout: rank shape stride offset element wise stride ordering where shape and stride are both straight int pointers Of note here is that offset will be zero automatically because the offset is handled by the pointer object instead

getPointer

public static Pointer getPointer(IComplexDouble x)

Get the pointer for a single complex float

Parameters:

x - the number ot get the pointer for

Returns:

the pointer for the given complex number

getPointer

public static Pointer getPointer(IComplexFloat x)

Get the pointer for a single complex float

Parameters:

x - the number ot get the pointer for

Returns:

the pointer for the given complex number

getPointer

public static Pointer getPointer(cuDoubleComplex x)

Get the pointer for a single complex float

Parameters:

x - the number ot get the pointer for

Returns:

the pointer for the given complex number

getPointer

public static Pointer getPointer(cuComplex x)

Get the pointer for a single complex float

Parameters:

x - the number ot get the pointer for

Returns:

the pointer for the given complex number

toFloats

public static float[] toFloats(Object[] extraArgs)

getNumBlocks

public static int getNumBlocks(int n,
                               int maxBlocks,
                               int maxThreads)

Compute the number of blocks that should be used for the given input size and limits

Parameters:

n - The input size

maxBlocks - The maximum number of blocks

maxThreads - The maximum number of threads

Returns:

The number of blocks

getNumThreads

public static int getNumThreads(int n,
                                int maxThreads)

Compute the number of threads that should be used for the given input size and limits

Parameters:

n - The input size

maxThreads - The maximum number of threads

Returns:

The number of threads

nextPow2

public static int nextPow2(int x)

Returns the power of 2 that is equal to or greater than x

Parameters:

x - The input

Returns:

The next power of 2

constructAndAlloc

public static CUdeviceptr constructAndAlloc(int length,
                                            DataBuffer.Type dType)

Construct and allocate a device pointer

Parameters:

length - the length of the pointer

dType - the data type to choose

Returns:

the new pointer

sizeFor

public static int sizeFor(DataBuffer.Type dataType)

getPointer

public static Object getPointer(ScalarOp scalarOp)

getPointer

public static Pointer getPointer(double alpha)

getPointer

public static Pointer getPointer(float alpha)