org.deeplearning4j.nn.layers

Class BaseCudnnHelper

java.lang.Object

org.deeplearning4j.nn.layers.BaseCudnnHelper

Direct Known Subclasses:

CudnnBatchNormalizationHelper, CudnnConvolutionHelper, CudnnDropoutHelper, CudnnLocalResponseNormalizationHelper, CudnnLSTMHelper, CudnnSubsamplingHelper

public abstract class BaseCudnnHelper
extends Object

Functionality shared by all cuDNN-based helpers.

Author:

saudet

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
protected static class	BaseCudnnHelper.CudnnContext
protected static class	BaseCudnnHelper.DataCache
protected static class	BaseCudnnHelper.TensorArray

Field Summary

Fields

Modifier and Type	Field and Description
protected org.bytedeco.javacpp.Pointer	alpha
protected org.bytedeco.javacpp.Pointer	beta
protected int	dataType
protected int	dataTypeSize
protected org.bytedeco.javacpp.SizeTPointer	sizeInBytes
protected static int	TENSOR_FORMAT

Constructor Summary

Constructors

Constructor and Description
BaseCudnnHelper()

Method Summary

Modifier and Type	Method and Description
protected static int[]	adaptForTensorDescr(int[] shapeOrStrides) From CuDNN documentation - "Tensors are restricted to having at least 4 dimensions...
protected static void	checkCuda(int error)
protected static void	checkCudnn(int status)
boolean	checkSupported()
static int	toCudnnDataType(org.nd4j.linalg.api.buffer.DataBuffer.Type type)

Methods inherited from class java.lang.Object

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

Field Detail

TENSOR_FORMAT

protected static final int TENSOR_FORMAT

See Also:

Constant Field Values

dataType

protected int dataType

dataTypeSize

protected int dataTypeSize

alpha

protected org.bytedeco.javacpp.Pointer alpha

beta

protected org.bytedeco.javacpp.Pointer beta

sizeInBytes

protected org.bytedeco.javacpp.SizeTPointer sizeInBytes

Constructor Detail

BaseCudnnHelper

public BaseCudnnHelper()

Method Detail

checkCuda

protected static void checkCuda(int error)

checkCudnn

protected static void checkCudnn(int status)

toCudnnDataType

public static int toCudnnDataType(org.nd4j.linalg.api.buffer.DataBuffer.Type type)

checkSupported

public boolean checkSupported()

adaptForTensorDescr

protected static int[] adaptForTensorDescr(int[] shapeOrStrides)

From CuDNN documentation - "Tensors are restricted to having at least 4 dimensions... When working with lower dimensional data, it is recommended that the user create a 4Dtensor, and set the size along unused dimensions to 1." This method implements that - basically appends 1s to the end (shape or stride) to make it length 4, or leaves it unmodified if the length is already 4 or more. This method can be used for both shape and strides

Parameters:

shapeOrStrides -

Returns: