public class JCudnn extends Object
Modifier and Type | Field and Description |
---|---|
static double |
CUDNN_BN_MIN_EPSILON
Minimum epsilon allowed to be used in the Batch Normalization formula
|
static int |
CUDNN_DIM_MAX
Maximum supported number of tensor dimensions
|
static int |
CUDNN_MAJOR |
static int |
CUDNN_MINOR |
static int |
CUDNN_PATCHLEVEL |
static int |
CUDNN_SEV_ERROR_EN
Message masks to be used with cudnnSetCallback()
|
static int |
CUDNN_SEV_INFO_EN |
static int |
CUDNN_SEV_WARNING_EN |
static int |
CUDNN_VERSION |
Modifier and Type | Method and Description |
---|---|
static int |
cudnnActivationBackward(cudnnHandle handle,
cudnnActivationDescriptor activationDesc,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
Function to perform backward activation
|
static int |
cudnnActivationForward(cudnnHandle handle,
cudnnActivationDescriptor activationDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform forward activation
|
static int |
cudnnAddTensor(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor aDesc,
Pointer A,
Pointer beta,
cudnnTensorDescriptor cDesc,
Pointer C)
Tensor Bias addition : C = alpha * A + beta * C
|
static int |
cudnnBatchNormalizationBackward(cudnnHandle handle,
int mode,
Pointer alphaDataDiff,
Pointer betaDataDiff,
Pointer alphaParamDiff,
Pointer betaParamDiff,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor dxDesc,
Pointer dx,
cudnnTensorDescriptor dBnScaleBiasDesc,
Pointer bnScale,
Pointer dBnScaleResult,
Pointer dBnBiasResult,
double epsilon,
Pointer savedMean,
Pointer savedInvVariance)
Performs backward pass of Batch Normalization layer.
|
static int |
cudnnBatchNormalizationForwardInference(cudnnHandle handle,
int mode,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor bnScaleBiasMeanVarDesc,
Pointer bnScale,
Pointer bnBias,
Pointer estimatedMean,
Pointer estimatedVariance,
double epsilon)
Performs Batch Normalization during Inference:
y[i] = bnScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + bnBias[k]
with bnScale, bnBias, runningMean, runningInvVariance tensors indexed
according to spatial or per-activation mode.
|
static int |
cudnnBatchNormalizationForwardTraining(cudnnHandle handle,
int mode,
Pointer alpha,
Pointer beta,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor bnScaleBiasMeanVarDesc,
Pointer bnScale,
Pointer bnBias,
double exponentialAverageFactor,
Pointer resultRunningMean,
Pointer resultRunningVariance,
double epsilon,
Pointer resultSaveMean,
Pointer resultSaveInvVariance)
Computes y = BN(x).
|
static int |
cudnnConvolutionBackwardBias(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor dyDesc,
Pointer dy,
Pointer beta,
cudnnTensorDescriptor dbDesc,
Pointer db)
Function to compute the bias gradient for batch convolution
|
static int |
cudnnConvolutionBackwardData(cudnnHandle handle,
Pointer alpha,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx) |
static int |
cudnnConvolutionBackwardFilter(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer beta,
cudnnFilterDescriptor dwDesc,
Pointer dw) |
static int |
cudnnConvolutionBiasActivationForward(cudnnHandle handle,
Pointer alpha1,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer alpha2,
cudnnTensorDescriptor zDesc,
Pointer z,
cudnnTensorDescriptor biasDesc,
Pointer bias,
cudnnActivationDescriptor activationDesc,
cudnnTensorDescriptor yDesc,
Pointer y)
Fused conv/bias/activation operation : y = Act( alpha1 * conv(x) + alpha2 * z + bias )
|
static int |
cudnnConvolutionForward(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnConvolutionDescriptor convDesc,
int algo,
Pointer workSpace,
long workSpaceSizeInBytes,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform the forward pass for batch convolution
|
static int |
cudnnCopyAlgorithmDescriptor(cudnnAlgorithmDescriptor src,
cudnnAlgorithmDescriptor dest) |
static int |
cudnnCreate(cudnnHandle handle) |
static int |
cudnnCreateActivationDescriptor(cudnnActivationDescriptor activationDesc)
Activation functions: All of the form "output = alpha * Op(inputs) + beta * output"
|
static int |
cudnnCreateAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc)
pointer to the returned workspace size
|
static int |
cudnnCreateAlgorithmPerformance(cudnnAlgorithmPerformance[] algoPerf,
int numberToCreate) |
static int |
cudnnCreateConvolutionDescriptor(cudnnConvolutionDescriptor convDesc)
Create an instance of convolution descriptor
|
static int |
cudnnCreateCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc)
Create an instance of a CTC (Connectionist Temporal Classification) loss descriptor
|
static int |
cudnnCreateDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc) |
static int |
cudnnCreateFilterDescriptor(cudnnFilterDescriptor filterDesc)
Create an instance of FilterStruct
|
static int |
cudnnCreateLRNDescriptor(cudnnLRNDescriptor normDesc)
Create an instance of LRN (Local Response Normalization) descriptor
Uses lrnN=5, lrnAlpha=1e-4, lrnBeta=0.75, lrnK=2.0 as defaults from Krizhevsky'12 ImageNet paper
|
static int |
cudnnCreateOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc) |
static int |
cudnnCreatePersistentRNNPlan(cudnnRNNDescriptor rnnDesc,
int minibatch,
int dataType,
cudnnPersistentRNNPlan plan)
Expensive.
|
static int |
cudnnCreatePoolingDescriptor(cudnnPoolingDescriptor poolingDesc)
Create an instance of pooling descriptor
|
static int |
cudnnCreateReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc) |
static int |
cudnnCreateRNNDescriptor(cudnnRNNDescriptor rnnDesc) |
static int |
cudnnCreateSpatialTransformerDescriptor(cudnnSpatialTransformerDescriptor stDesc) |
static int |
cudnnCreateTensorDescriptor(cudnnTensorDescriptor tensorDesc)
Create an instance of a generic Tensor descriptor
|
static int |
cudnnCTCLoss(cudnnHandle handle,
cudnnTensorDescriptor probsDesc,
Pointer probs,
int[] labels,
int[] labelLengths,
int[] inputLengths,
Pointer costs,
cudnnTensorDescriptor gradientsDesc,
Pointer gradients,
int algo,
cudnnCTCLossDescriptor ctcLossDesc,
Pointer workspace,
long workSpaceSizeInBytes)
return the ctc costs and gradients, given the probabilities and labels
|
static int |
cudnnDeriveBNTensorDescriptor(cudnnTensorDescriptor derivedBnDesc,
cudnnTensorDescriptor xDesc,
int mode)
Derives a tensor descriptor from layer data descriptor for BatchNormalization
scale, invVariance, bnBias, bnScale tensors.
|
static int |
cudnnDestroy(cudnnHandle handle) |
static int |
cudnnDestroyActivationDescriptor(cudnnActivationDescriptor activationDesc)
ceiling for clipped RELU, alpha for ELU
|
static int |
cudnnDestroyAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc) |
static int |
cudnnDestroyAlgorithmPerformance(cudnnAlgorithmPerformance[] algoPerf,
int numberToDestroy) |
static int |
cudnnDestroyConvolutionDescriptor(cudnnConvolutionDescriptor convDesc)
Destroy an instance of convolution descriptor
|
static int |
cudnnDestroyCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc) |
static int |
cudnnDestroyDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc) |
static int |
cudnnDestroyFilterDescriptor(cudnnFilterDescriptor filterDesc) |
static int |
cudnnDestroyLRNDescriptor(cudnnLRNDescriptor lrnDesc)
Destroy an instance of LRN descriptor
|
static int |
cudnnDestroyOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc) |
static int |
cudnnDestroyPersistentRNNPlan(cudnnPersistentRNNPlan plan) |
static int |
cudnnDestroyPoolingDescriptor(cudnnPoolingDescriptor poolingDesc)
Destroy an instance of pooling descriptor
|
static int |
cudnnDestroyReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc) |
static int |
cudnnDestroyRNNDescriptor(cudnnRNNDescriptor rnnDesc) |
static int |
cudnnDestroySpatialTransformerDescriptor(cudnnSpatialTransformerDescriptor stDesc) |
static int |
cudnnDestroyTensorDescriptor(cudnnTensorDescriptor tensorDesc)
Destroy an instance of Tensor4d descriptor
|
static int |
cudnnDivisiveNormalizationBackward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int mode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer means,
Pointer dy,
Pointer temp,
Pointer temp2,
Pointer beta,
cudnnTensorDescriptor dXdMeansDesc,
Pointer dx,
Pointer dMeans) |
static int |
cudnnDivisiveNormalizationForward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int mode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer means,
Pointer temp,
Pointer temp2,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
LCN/divisive normalization functions: y = alpha * normalize(x) + beta * y
|
static int |
cudnnDropoutBackward(cudnnHandle handle,
cudnnDropoutDescriptor dropoutDesc,
cudnnTensorDescriptor dydesc,
Pointer dy,
cudnnTensorDescriptor dxdesc,
Pointer dx,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnDropoutForward(cudnnHandle handle,
cudnnDropoutDescriptor dropoutDesc,
cudnnTensorDescriptor xdesc,
Pointer x,
cudnnTensorDescriptor ydesc,
Pointer y,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnDropoutGetReserveSpaceSize(cudnnTensorDescriptor xdesc,
long[] sizeInBytes)
helper function to determine size of the reserve space to be passed to dropout forward/backward calls
|
static int |
cudnnDropoutGetStatesSize(cudnnHandle handle,
long[] sizeInBytes)
helper function to determine size of the states to be passed to cudnnSetDropoutDescriptor
|
static int |
cudnnFindConvolutionBackwardDataAlgorithm(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf[] perfResults) |
static int |
cudnnFindConvolutionBackwardDataAlgorithmEx(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
Pointer dx,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf[] perfResults,
Pointer workSpace,
long workSpaceSizeInBytes) |
static int |
cudnnFindConvolutionBackwardFilterAlgorithm(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor dwDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf[] perfResults) |
static int |
cudnnFindConvolutionBackwardFilterAlgorithmEx(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnTensorDescriptor dyDesc,
Pointer y,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor dwDesc,
Pointer dw,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf[] perfResults,
Pointer workSpace,
long workSpaceSizeInBytes) |
static int |
cudnnFindConvolutionForwardAlgorithm(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf[] perfResults) |
static int |
cudnnFindConvolutionForwardAlgorithmEx(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
Pointer y,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf[] perfResults,
Pointer workSpace,
long workSpaceSizeInBytes) |
static int |
cudnnFindRNNBackwardDataAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor[] dyDesc,
Pointer dy,
cudnnTensorDescriptor dhyDesc,
Pointer dhy,
cudnnTensorDescriptor dcyDesc,
Pointer dcy,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnTensorDescriptor[] dxDesc,
Pointer dx,
cudnnTensorDescriptor dhxDesc,
Pointer dhx,
cudnnTensorDescriptor dcxDesc,
Pointer dcx,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnFindRNNBackwardWeightsAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor[] yDesc,
Pointer y,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes,
cudnnFilterDescriptor dwDesc,
Pointer dw,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnFindRNNForwardInferenceAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes) |
static int |
cudnnFindRNNForwardTrainingAlgorithmEx(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
float findIntensity,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnAlgorithmPerformance[] perfResults,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnGetActivationDescriptor(cudnnActivationDescriptor activationDesc,
int[] mode,
int[] reluNanOpt,
double[] coef)
ceiling for clipped RELU, alpha for ELU
|
static int |
cudnnGetAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc,
int[] algorithm) |
static int |
cudnnGetAlgorithmPerformance(cudnnAlgorithmPerformance algoPerf,
cudnnAlgorithmDescriptor algoDesc,
int[] status,
float[] time,
long[] memory) |
static int |
cudnnGetAlgorithmSpaceSize(cudnnHandle handle,
cudnnAlgorithmDescriptor algoDesc,
long[] algoSpaceSizeInBytes) |
static int |
cudnnGetCallback(int[] mask,
Object udata,
cudnnCallback[] fptr) |
static int |
cudnnGetConvolution2dDescriptor(cudnnConvolutionDescriptor convDesc,
int[] pad_h,
int[] pad_w,
int[] u,
int[] v,
int[] dilation_h,
int[] dilation_w,
int[] mode,
int[] computeType) |
static int |
cudnnGetConvolution2dForwardOutputDim(cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor inputTensorDesc,
cudnnFilterDescriptor filterDesc,
int[] n,
int[] c,
int[] h,
int[] w)
Helper function to return the dimensions of the output tensor given a convolution descriptor
|
static int |
cudnnGetConvolutionBackwardDataAlgorithm_v7(cudnnHandle handle,
cudnnFilterDescriptor filterDesc,
cudnnTensorDescriptor diffDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor gradDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdDataAlgoPerf[] perfResults) |
static int |
cudnnGetConvolutionBackwardDataAlgorithm(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
int preference,
long memoryLimitInBytes,
int[] algo) |
static int |
cudnnGetConvolutionBackwardDataAlgorithmMaxCount(cudnnHandle handle,
int[] count) |
static int |
cudnnGetConvolutionBackwardDataWorkspaceSize(cudnnHandle handle,
cudnnFilterDescriptor wDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor dxDesc,
int algo,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the convolution given an algo
|
static int |
cudnnGetConvolutionBackwardFilterAlgorithm_v7(cudnnHandle handle,
cudnnTensorDescriptor srcDesc,
cudnnTensorDescriptor diffDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor gradDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionBwdFilterAlgoPerf[] perfResults) |
static int |
cudnnGetConvolutionBackwardFilterAlgorithm(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor dwDesc,
int preference,
long memoryLimitInBytes,
int[] algo) |
static int |
cudnnGetConvolutionBackwardFilterAlgorithmMaxCount(cudnnHandle handle,
int[] count) |
static int |
cudnnGetConvolutionBackwardFilterWorkspaceSize(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnTensorDescriptor dyDesc,
cudnnConvolutionDescriptor convDesc,
cudnnFilterDescriptor gradDesc,
int algo,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the convolution given an algo
|
static int |
cudnnGetConvolutionForwardAlgorithm_v7(cudnnHandle handle,
cudnnTensorDescriptor srcDesc,
cudnnFilterDescriptor filterDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor destDesc,
int requestedAlgoCount,
int[] returnedAlgoCount,
cudnnConvolutionFwdAlgoPerf[] perfResults) |
static int |
cudnnGetConvolutionForwardAlgorithm(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
int preference,
long memoryLimitInBytes,
int[] algo) |
static int |
cudnnGetConvolutionForwardAlgorithmMaxCount(cudnnHandle handle,
int[] count) |
static int |
cudnnGetConvolutionForwardWorkspaceSize(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor yDesc,
int algo,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the convolution given an algo
|
static int |
cudnnGetConvolutionGroupCount(cudnnConvolutionDescriptor convDesc,
int[] groupCount) |
static int |
cudnnGetConvolutionMathType(cudnnConvolutionDescriptor convDesc,
int[] mathType) |
static int |
cudnnGetConvolutionNdDescriptor(cudnnConvolutionDescriptor convDesc,
int arrayLengthRequested,
int[] arrayLength,
int[] padA,
int[] strideA,
int[] dilationA,
int[] mode,
int[] computeType)
convolution data type
|
static int |
cudnnGetConvolutionNdForwardOutputDim(cudnnConvolutionDescriptor convDesc,
cudnnTensorDescriptor inputTensorDesc,
cudnnFilterDescriptor filterDesc,
int nbDims,
int[] tensorOuputDimA)
Helper function to return the dimensions of the output tensor given a convolution descriptor
|
static int |
cudnnGetCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc,
int[] compType) |
static int |
cudnnGetCTCLossWorkspaceSize(cudnnHandle handle,
cudnnTensorDescriptor probsDesc,
cudnnTensorDescriptor gradientsDesc,
int[] labels,
int[] labelLengths,
int[] inputLengths,
int algo,
cudnnCTCLossDescriptor ctcLossDesc,
long[] sizeInBytes)
return the workspace size needed for ctc
|
static long |
cudnnGetCudartVersion()
Returns CUDA Runtime version statically linked against cudnn
|
static int |
cudnnGetDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc,
cudnnHandle handle,
float[] dropout,
Pointer states,
long[] seed) |
static String |
cudnnGetErrorString(int status)
human-readable error messages
|
static int |
cudnnGetFilter4dDescriptor(cudnnFilterDescriptor filterDesc,
int[] dataType,
int[] format,
int[] k,
int[] c,
int[] h,
int[] w)
width of each input filter
|
static int |
cudnnGetFilterNdDescriptor(cudnnFilterDescriptor filterDesc,
int nbDimsRequested,
int[] dataType,
int[] format,
int[] nbDims,
int[] filterDimA) |
static int |
cudnnGetLRNDescriptor(cudnnLRNDescriptor normDesc,
int[] lrnN,
double[] lrnAlpha,
double[] lrnBeta,
double[] lrnK)
Retrieve the settings currently stored in an LRN layer descriptor
Any of the provided pointers can be NULL (no corresponding value will be returned)
|
static int |
cudnnGetOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc,
int[] opTensorOp,
int[] opTensorCompType,
int[] opTensorNanOpt) |
static int |
cudnnGetPooling2dDescriptor(cudnnPoolingDescriptor poolingDesc,
int[] mode,
int[] maxpoolingNanOpt,
int[] windowHeight,
int[] windowWidth,
int[] verticalPadding,
int[] horizontalPadding,
int[] verticalStride,
int[] horizontalStride) |
static int |
cudnnGetPooling2dForwardOutputDim(cudnnPoolingDescriptor poolingDesc,
cudnnTensorDescriptor inputTensorDesc,
int[] n,
int[] c,
int[] h,
int[] w) |
static int |
cudnnGetPoolingNdDescriptor(cudnnPoolingDescriptor poolingDesc,
int nbDimsRequested,
int[] mode,
int[] maxpoolingNanOpt,
int[] nbDims,
int[] windowDimA,
int[] paddingA,
int[] strideA) |
static int |
cudnnGetPoolingNdForwardOutputDim(cudnnPoolingDescriptor poolingDesc,
cudnnTensorDescriptor inputTensorDesc,
int nbDims,
int[] outputTensorDimA) |
static int |
cudnnGetProperty(int type,
int[] value) |
static int |
cudnnGetReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc,
int[] reduceTensorOp,
int[] reduceTensorCompType,
int[] reduceTensorNanOpt,
int[] reduceTensorIndices,
int[] reduceTensorIndicesType) |
static int |
cudnnGetReductionIndicesSize(cudnnHandle handle,
cudnnReduceTensorDescriptor reduceTensorDesc,
cudnnTensorDescriptor aDesc,
cudnnTensorDescriptor cDesc,
long[] sizeInBytes)
Helper function to return the minimum size of the index space to be passed to the reduction given the input and output tensors
|
static int |
cudnnGetReductionWorkspaceSize(cudnnHandle handle,
cudnnReduceTensorDescriptor reduceTensorDesc,
cudnnTensorDescriptor aDesc,
cudnnTensorDescriptor cDesc,
long[] sizeInBytes)
Helper function to return the minimum size of the workspace to be passed to the reduction given the input and output tensors
|
static int |
cudnnGetRNNBackwardDataAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count) |
static int |
cudnnGetRNNBackwardWeightsAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count) |
static int |
cudnnGetRNNDescriptor(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] hiddenSize,
int[] numLayers,
cudnnDropoutDescriptor dropoutDesc,
int[] inputMode,
int[] direction,
int[] mode,
int[] algo,
int[] dataType) |
static int |
cudnnGetRNNForwardInferenceAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count) |
static int |
cudnnGetRNNForwardTrainingAlgorithmMaxCount(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] count) |
static int |
cudnnGetRNNLinLayerBiasParams(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int pseudoLayer,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
Pointer w,
int linLayerID,
cudnnFilterDescriptor linLayerBiasDesc,
Pointer linLayerBias) |
static int |
cudnnGetRNNLinLayerMatrixParams(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int pseudoLayer,
cudnnTensorDescriptor xDesc,
cudnnFilterDescriptor wDesc,
Pointer w,
int linLayerID,
cudnnFilterDescriptor linLayerMatDesc,
Pointer linLayerMat) |
static int |
cudnnGetRNNMatrixMathType(cudnnRNNDescriptor rnnDesc,
int[] mType) |
static int |
cudnnGetRNNParamsSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnTensorDescriptor xDesc,
long[] sizeInBytes,
int dataType) |
static int |
cudnnGetRNNProjectionLayers(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int[] recProjSize,
int[] outProjSize) |
static int |
cudnnGetRNNTrainingReserveSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
long[] sizeInBytes) |
static int |
cudnnGetRNNWorkspaceSize(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
long[] sizeInBytes)
dataType in weight descriptors and input descriptors is used to describe storage
|
static int |
cudnnGetStream(cudnnHandle handle,
cudaStream_t streamId) |
static int |
cudnnGetTensor4dDescriptor(cudnnTensorDescriptor tensorDesc,
int[] dataType,
int[] n,
int[] c,
int[] h,
int[] w,
int[] nStride,
int[] cStride,
int[] hStride,
int[] wStride) |
static int |
cudnnGetTensorNdDescriptor(cudnnTensorDescriptor tensorDesc,
int nbDimsRequested,
int[] dataType,
int[] nbDims,
int[] dimA,
int[] strideA) |
static int |
cudnnGetTensorSizeInBytes(cudnnTensorDescriptor tensorDesc,
long[] size) |
static long |
cudnnGetVersion() |
static int |
cudnnIm2Col(cudnnHandle handle,
cudnnTensorDescriptor xDesc,
Pointer x,
cudnnFilterDescriptor wDesc,
cudnnConvolutionDescriptor convDesc,
Pointer colBuffer) |
static int |
cudnnLRNCrossChannelBackward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int lrnMode,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
LRN cross-channel backward computation.
|
static int |
cudnnLRNCrossChannelForward(cudnnHandle handle,
cudnnLRNDescriptor normDesc,
int lrnMode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
LRN cross-channel forward computation.
|
static int |
cudnnOpTensor(cudnnHandle handle,
cudnnOpTensorDescriptor opTensorDesc,
Pointer alpha1,
cudnnTensorDescriptor aDesc,
Pointer A,
Pointer alpha2,
cudnnTensorDescriptor bDesc,
Pointer B,
Pointer beta,
cudnnTensorDescriptor cDesc,
Pointer C)
B tensor is ignored for CUDNN_OP_TENSOR_SQRT, CUDNN_OP_TENSOR_NOT.
|
static int |
cudnnPoolingBackward(cudnnHandle handle,
cudnnPoolingDescriptor poolingDesc,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
Function to perform backward pooling
|
static int |
cudnnPoolingForward(cudnnHandle handle,
cudnnPoolingDescriptor poolingDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform forward pooling
|
static int |
cudnnQueryRuntimeError(cudnnHandle handle,
int[] rstatus,
int mode,
cudnnRuntimeTag tag) |
static int |
cudnnReduceTensor(cudnnHandle handle,
cudnnReduceTensorDescriptor reduceTensorDesc,
Pointer indices,
long indicesSizeInBytes,
Pointer workspace,
long workspaceSizeInBytes,
Pointer alpha,
cudnnTensorDescriptor aDesc,
Pointer A,
Pointer beta,
cudnnTensorDescriptor cDesc,
Pointer C)
The indices space is ignored for reduce ops other than min or max.
|
static int |
cudnnRestoreAlgorithm(cudnnHandle handle,
Pointer algoSpace,
long algoSpaceSizeInBytes,
cudnnAlgorithmDescriptor algoDesc) |
static int |
cudnnRestoreDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc,
cudnnHandle handle,
float dropout,
Pointer states,
long stateSizeInBytes,
long seed) |
static int |
cudnnRNNBackwardData(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor[] dyDesc,
Pointer dy,
cudnnTensorDescriptor dhyDesc,
Pointer dhy,
cudnnTensorDescriptor dcyDesc,
Pointer dcy,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnTensorDescriptor[] dxDesc,
Pointer dx,
cudnnTensorDescriptor dhxDesc,
Pointer dhx,
cudnnTensorDescriptor dcxDesc,
Pointer dcx,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnRNNBackwardWeights(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor[] yDesc,
Pointer y,
Pointer workspace,
long workSpaceSizeInBytes,
cudnnFilterDescriptor dwDesc,
Pointer dw,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnRNNForwardInference(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
Pointer workspace,
long workSpaceSizeInBytes) |
static int |
cudnnRNNForwardTraining(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int seqLength,
cudnnTensorDescriptor[] xDesc,
Pointer x,
cudnnTensorDescriptor hxDesc,
Pointer hx,
cudnnTensorDescriptor cxDesc,
Pointer cx,
cudnnFilterDescriptor wDesc,
Pointer w,
cudnnTensorDescriptor[] yDesc,
Pointer y,
cudnnTensorDescriptor hyDesc,
Pointer hy,
cudnnTensorDescriptor cyDesc,
Pointer cy,
Pointer workspace,
long workSpaceSizeInBytes,
Pointer reserveSpace,
long reserveSpaceSizeInBytes) |
static int |
cudnnSaveAlgorithm(cudnnHandle handle,
cudnnAlgorithmDescriptor algoDesc,
Pointer algoSpace,
long algoSpaceSizeInBytes) |
static int |
cudnnScaleTensor(cudnnHandle handle,
cudnnTensorDescriptor yDesc,
Pointer y,
Pointer alpha)
Scale all values of a tensor by a given factor : y[i] = alpha * y[i]
|
static int |
cudnnSetActivationDescriptor(cudnnActivationDescriptor activationDesc,
int mode,
int reluNanOpt,
double coef) |
static int |
cudnnSetAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc,
int algorithm) |
static int |
cudnnSetAlgorithmPerformance(cudnnAlgorithmPerformance algoPerf,
cudnnAlgorithmDescriptor algoDesc,
int status,
float time,
long memory) |
static int |
cudnnSetCallback(int mask,
Object udata,
cudnnCallback fptr) |
static int |
cudnnSetConvolution2dDescriptor(cudnnConvolutionDescriptor convDesc,
int pad_h,
int pad_w,
int u,
int v,
int dilation_h,
int dilation_w,
int mode,
int computeType) |
static int |
cudnnSetConvolutionGroupCount(cudnnConvolutionDescriptor convDesc,
int groupCount) |
static int |
cudnnSetConvolutionMathType(cudnnConvolutionDescriptor convDesc,
int mathType) |
static int |
cudnnSetConvolutionNdDescriptor(cudnnConvolutionDescriptor convDesc,
int arrayLength,
int[] padA,
int[] filterStrideA,
int[] dilationA,
int mode,
int computeType) |
static int |
cudnnSetCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc,
int compType) |
static int |
cudnnSetDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc,
cudnnHandle handle,
float dropout,
Pointer states,
long stateSizeInBytes,
long seed) |
static int |
cudnnSetFilter4dDescriptor(cudnnFilterDescriptor filterDesc,
int dataType,
int format,
int k,
int c,
int h,
int w) |
static int |
cudnnSetFilterNdDescriptor(cudnnFilterDescriptor filterDesc,
int dataType,
int format,
int nbDims,
int[] filterDimA)
width of each input filter
|
static int |
cudnnSetLRNDescriptor(cudnnLRNDescriptor normDesc,
int lrnN,
double lrnAlpha,
double lrnBeta,
double lrnK)
Uses a window [center-lookBehind, center+lookAhead], where
lookBehind = floor( (lrnN-1)/2 ), lookAhead = lrnN-lookBehind-1.
|
static int |
cudnnSetOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc,
int opTensorOp,
int opTensorCompType,
int opTensorNanOpt) |
static int |
cudnnSetPersistentRNNPlan(cudnnRNNDescriptor rnnDesc,
cudnnPersistentRNNPlan plan)
Attaches the plan to the descriptor.
|
static int |
cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor poolingDesc,
int mode,
int maxpoolingNanOpt,
int windowHeight,
int windowWidth,
int verticalPadding,
int horizontalPadding,
int verticalStride,
int horizontalStride) |
static int |
cudnnSetPoolingNdDescriptor(cudnnPoolingDescriptor poolingDesc,
int mode,
int maxpoolingNanOpt,
int nbDims,
int[] windowDimA,
int[] paddingA,
int[] strideA) |
static int |
cudnnSetReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc,
int reduceTensorOp,
int reduceTensorCompType,
int reduceTensorNanOpt,
int reduceTensorIndices,
int reduceTensorIndicesType) |
static int |
cudnnSetRNNAlgorithmDescriptor(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
cudnnAlgorithmDescriptor algoDesc) |
static int |
cudnnSetRNNDescriptor_v5(cudnnRNNDescriptor rnnDesc,
int hiddenSize,
int numLayers,
cudnnDropoutDescriptor dropoutDesc,
int inputMode,
int direction,
int mode,
int dataType) |
static int |
cudnnSetRNNDescriptor_v6(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int hiddenSize,
int numLayers,
cudnnDropoutDescriptor dropoutDesc,
int inputMode,
int direction,
int mode,
int algo,
int dataType)
DEPRECATED routines to be removed next release :
User should use the non-suffixed version (which has the API and functionality of _v6 version)
Routines with _v5 suffix has the functionality of the non-suffixed routines in the CUDNN V6
|
static int |
cudnnSetRNNDescriptor(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int hiddenSize,
int numLayers,
cudnnDropoutDescriptor dropoutDesc,
int inputMode,
int direction,
int mode,
int algo,
int dataType) |
static int |
cudnnSetRNNMatrixMathType(cudnnRNNDescriptor rnnDesc,
int mType) |
static int |
cudnnSetRNNProjectionLayers(cudnnHandle handle,
cudnnRNNDescriptor rnnDesc,
int recProjSize,
int outProjSize) |
static int |
cudnnSetSpatialTransformerNdDescriptor(cudnnSpatialTransformerDescriptor stDesc,
int samplerType,
int dataType,
int nbDims,
int[] dimA) |
static int |
cudnnSetStream(cudnnHandle handle,
cudaStream_t streamId) |
static int |
cudnnSetTensor(cudnnHandle handle,
cudnnTensorDescriptor yDesc,
Pointer y,
Pointer valuePtr)
Set all values of a tensor to a given value : y[i] = value[0]
|
static int |
cudnnSetTensor4dDescriptor(cudnnTensorDescriptor tensorDesc,
int format,
int dataType,
int n,
int c,
int h,
int w) |
static int |
cudnnSetTensor4dDescriptorEx(cudnnTensorDescriptor tensorDesc,
int dataType,
int n,
int c,
int h,
int w,
int nStride,
int cStride,
int hStride,
int wStride)
width of input section
|
static int |
cudnnSetTensorNdDescriptor(cudnnTensorDescriptor tensorDesc,
int dataType,
int nbDims,
int[] dimA,
int[] strideA) |
static int |
cudnnSetTensorNdDescriptorEx(cudnnTensorDescriptor tensorDesc,
int format,
int dataType,
int nbDims,
int[] dimA) |
static int |
cudnnSoftmaxBackward(cudnnHandle handle,
int algo,
int mode,
Pointer alpha,
cudnnTensorDescriptor yDesc,
Pointer y,
cudnnTensorDescriptor dyDesc,
Pointer dy,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx)
Function to perform backward softmax
|
static int |
cudnnSoftmaxForward(cudnnHandle handle,
int algo,
int mode,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Function to perform forward softmax
|
static int |
cudnnSpatialTfGridGeneratorBackward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer dgrid,
Pointer dtheta) |
static int |
cudnnSpatialTfGridGeneratorForward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer theta,
Pointer grid) |
static int |
cudnnSpatialTfSamplerBackward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor dxDesc,
Pointer dx,
Pointer alphaDgrid,
cudnnTensorDescriptor dyDesc,
Pointer dy,
Pointer grid,
Pointer betaDgrid,
Pointer dgrid) |
static int |
cudnnSpatialTfSamplerForward(cudnnHandle handle,
cudnnSpatialTransformerDescriptor stDesc,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer grid,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y) |
static int |
cudnnTransformTensor(cudnnHandle handle,
Pointer alpha,
cudnnTensorDescriptor xDesc,
Pointer x,
Pointer beta,
cudnnTensorDescriptor yDesc,
Pointer y)
Tensor layout conversion helper (y = alpha * x + beta * y)
|
static void |
initialize()
Initializes the native library.
|
static void |
setExceptionsEnabled(boolean enabled)
Enables or disables exceptions.
|
static void |
setLogLevel(LogLevel logLevel)
Set the specified log level for the JCudnn library.
|
public static final int CUDNN_MAJOR
public static final int CUDNN_MINOR
public static final int CUDNN_PATCHLEVEL
public static final int CUDNN_VERSION
public static final int CUDNN_DIM_MAX
public static final double CUDNN_BN_MIN_EPSILON
public static final int CUDNN_SEV_ERROR_EN
public static final int CUDNN_SEV_WARNING_EN
public static final int CUDNN_SEV_INFO_EN
public static void initialize()
public static void setExceptionsEnabled(boolean enabled)
cudnnStatus
from the native methods.
If exceptions are enabled, a CudaException with a detailed error
message will be thrown if a method is about to set a result code
that is not cudnnStatus.CUDNN_STATUS_SUCCESSenabled
- Whether exceptions are enabledpublic static void setLogLevel(LogLevel logLevel)
logLevel
- The log level to use.public static long cudnnGetVersion()
public static long cudnnGetCudartVersion()
public static String cudnnGetErrorString(int status)
public static int cudnnQueryRuntimeError(cudnnHandle handle, int[] rstatus, int mode, cudnnRuntimeTag tag)
public static int cudnnGetProperty(int type, int[] value)
public static int cudnnCreate(cudnnHandle handle)
public static int cudnnDestroy(cudnnHandle handle)
public static int cudnnSetStream(cudnnHandle handle, cudaStream_t streamId)
public static int cudnnGetStream(cudnnHandle handle, cudaStream_t streamId)
public static int cudnnCreateTensorDescriptor(cudnnTensorDescriptor tensorDesc)
public static int cudnnSetTensor4dDescriptor(cudnnTensorDescriptor tensorDesc, int format, int dataType, int n, int c, int h, int w)
public static int cudnnSetTensor4dDescriptorEx(cudnnTensorDescriptor tensorDesc, int dataType, int n, int c, int h, int w, int nStride, int cStride, int hStride, int wStride)
public static int cudnnGetTensor4dDescriptor(cudnnTensorDescriptor tensorDesc, int[] dataType, int[] n, int[] c, int[] h, int[] w, int[] nStride, int[] cStride, int[] hStride, int[] wStride)
public static int cudnnSetTensorNdDescriptor(cudnnTensorDescriptor tensorDesc, int dataType, int nbDims, int[] dimA, int[] strideA)
public static int cudnnSetTensorNdDescriptorEx(cudnnTensorDescriptor tensorDesc, int format, int dataType, int nbDims, int[] dimA)
public static int cudnnGetTensorNdDescriptor(cudnnTensorDescriptor tensorDesc, int nbDimsRequested, int[] dataType, int[] nbDims, int[] dimA, int[] strideA)
public static int cudnnGetTensorSizeInBytes(cudnnTensorDescriptor tensorDesc, long[] size)
public static int cudnnDestroyTensorDescriptor(cudnnTensorDescriptor tensorDesc)
public static int cudnnTransformTensor(cudnnHandle handle, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnAddTensor(cudnnHandle handle, Pointer alpha, cudnnTensorDescriptor aDesc, Pointer A, Pointer beta, cudnnTensorDescriptor cDesc, Pointer C)
public static int cudnnCreateOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc)
public static int cudnnSetOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc, int opTensorOp, int opTensorCompType, int opTensorNanOpt)
public static int cudnnGetOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc, int[] opTensorOp, int[] opTensorCompType, int[] opTensorNanOpt)
public static int cudnnDestroyOpTensorDescriptor(cudnnOpTensorDescriptor opTensorDesc)
public static int cudnnOpTensor(cudnnHandle handle, cudnnOpTensorDescriptor opTensorDesc, Pointer alpha1, cudnnTensorDescriptor aDesc, Pointer A, Pointer alpha2, cudnnTensorDescriptor bDesc, Pointer B, Pointer beta, cudnnTensorDescriptor cDesc, Pointer C)
public static int cudnnCreateReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc)
public static int cudnnSetReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc, int reduceTensorOp, int reduceTensorCompType, int reduceTensorNanOpt, int reduceTensorIndices, int reduceTensorIndicesType)
public static int cudnnGetReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc, int[] reduceTensorOp, int[] reduceTensorCompType, int[] reduceTensorNanOpt, int[] reduceTensorIndices, int[] reduceTensorIndicesType)
public static int cudnnDestroyReduceTensorDescriptor(cudnnReduceTensorDescriptor reduceTensorDesc)
public static int cudnnGetReductionIndicesSize(cudnnHandle handle, cudnnReduceTensorDescriptor reduceTensorDesc, cudnnTensorDescriptor aDesc, cudnnTensorDescriptor cDesc, long[] sizeInBytes)
public static int cudnnGetReductionWorkspaceSize(cudnnHandle handle, cudnnReduceTensorDescriptor reduceTensorDesc, cudnnTensorDescriptor aDesc, cudnnTensorDescriptor cDesc, long[] sizeInBytes)
public static int cudnnReduceTensor(cudnnHandle handle, cudnnReduceTensorDescriptor reduceTensorDesc, Pointer indices, long indicesSizeInBytes, Pointer workspace, long workspaceSizeInBytes, Pointer alpha, cudnnTensorDescriptor aDesc, Pointer A, Pointer beta, cudnnTensorDescriptor cDesc, Pointer C)
public static int cudnnSetTensor(cudnnHandle handle, cudnnTensorDescriptor yDesc, Pointer y, Pointer valuePtr)
public static int cudnnScaleTensor(cudnnHandle handle, cudnnTensorDescriptor yDesc, Pointer y, Pointer alpha)
public static int cudnnCreateFilterDescriptor(cudnnFilterDescriptor filterDesc)
public static int cudnnSetFilter4dDescriptor(cudnnFilterDescriptor filterDesc, int dataType, int format, int k, int c, int h, int w)
public static int cudnnGetFilter4dDescriptor(cudnnFilterDescriptor filterDesc, int[] dataType, int[] format, int[] k, int[] c, int[] h, int[] w)
public static int cudnnSetFilterNdDescriptor(cudnnFilterDescriptor filterDesc, int dataType, int format, int nbDims, int[] filterDimA)
public static int cudnnGetFilterNdDescriptor(cudnnFilterDescriptor filterDesc, int nbDimsRequested, int[] dataType, int[] format, int[] nbDims, int[] filterDimA)
public static int cudnnDestroyFilterDescriptor(cudnnFilterDescriptor filterDesc)
public static int cudnnCreateConvolutionDescriptor(cudnnConvolutionDescriptor convDesc)
public static int cudnnSetConvolutionMathType(cudnnConvolutionDescriptor convDesc, int mathType)
public static int cudnnGetConvolutionMathType(cudnnConvolutionDescriptor convDesc, int[] mathType)
public static int cudnnSetConvolutionGroupCount(cudnnConvolutionDescriptor convDesc, int groupCount)
public static int cudnnGetConvolutionGroupCount(cudnnConvolutionDescriptor convDesc, int[] groupCount)
public static int cudnnSetConvolution2dDescriptor(cudnnConvolutionDescriptor convDesc, int pad_h, int pad_w, int u, int v, int dilation_h, int dilation_w, int mode, int computeType)
public static int cudnnGetConvolution2dDescriptor(cudnnConvolutionDescriptor convDesc, int[] pad_h, int[] pad_w, int[] u, int[] v, int[] dilation_h, int[] dilation_w, int[] mode, int[] computeType)
public static int cudnnGetConvolution2dForwardOutputDim(cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor inputTensorDesc, cudnnFilterDescriptor filterDesc, int[] n, int[] c, int[] h, int[] w)
public static int cudnnSetConvolutionNdDescriptor(cudnnConvolutionDescriptor convDesc, int arrayLength, int[] padA, int[] filterStrideA, int[] dilationA, int mode, int computeType)
public static int cudnnGetConvolutionNdDescriptor(cudnnConvolutionDescriptor convDesc, int arrayLengthRequested, int[] arrayLength, int[] padA, int[] strideA, int[] dilationA, int[] mode, int[] computeType)
public static int cudnnGetConvolutionNdForwardOutputDim(cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor inputTensorDesc, cudnnFilterDescriptor filterDesc, int nbDims, int[] tensorOuputDimA)
public static int cudnnDestroyConvolutionDescriptor(cudnnConvolutionDescriptor convDesc)
public static int cudnnGetConvolutionForwardAlgorithmMaxCount(cudnnHandle handle, int[] count)
public static int cudnnFindConvolutionForwardAlgorithm(cudnnHandle handle, cudnnTensorDescriptor xDesc, cudnnFilterDescriptor wDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor yDesc, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionFwdAlgoPerf[] perfResults)
public static int cudnnFindConvolutionForwardAlgorithmEx(cudnnHandle handle, cudnnTensorDescriptor xDesc, Pointer x, cudnnFilterDescriptor wDesc, Pointer w, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor yDesc, Pointer y, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionFwdAlgoPerf[] perfResults, Pointer workSpace, long workSpaceSizeInBytes)
public static int cudnnGetConvolutionForwardAlgorithm(cudnnHandle handle, cudnnTensorDescriptor xDesc, cudnnFilterDescriptor wDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor yDesc, int preference, long memoryLimitInBytes, int[] algo)
public static int cudnnGetConvolutionForwardAlgorithm_v7(cudnnHandle handle, cudnnTensorDescriptor srcDesc, cudnnFilterDescriptor filterDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor destDesc, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionFwdAlgoPerf[] perfResults)
public static int cudnnGetConvolutionForwardWorkspaceSize(cudnnHandle handle, cudnnTensorDescriptor xDesc, cudnnFilterDescriptor wDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor yDesc, int algo, long[] sizeInBytes)
public static int cudnnConvolutionForward(cudnnHandle handle, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, cudnnFilterDescriptor wDesc, Pointer w, cudnnConvolutionDescriptor convDesc, int algo, Pointer workSpace, long workSpaceSizeInBytes, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnConvolutionBiasActivationForward(cudnnHandle handle, Pointer alpha1, cudnnTensorDescriptor xDesc, Pointer x, cudnnFilterDescriptor wDesc, Pointer w, cudnnConvolutionDescriptor convDesc, int algo, Pointer workSpace, long workSpaceSizeInBytes, Pointer alpha2, cudnnTensorDescriptor zDesc, Pointer z, cudnnTensorDescriptor biasDesc, Pointer bias, cudnnActivationDescriptor activationDesc, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnConvolutionBackwardBias(cudnnHandle handle, Pointer alpha, cudnnTensorDescriptor dyDesc, Pointer dy, Pointer beta, cudnnTensorDescriptor dbDesc, Pointer db)
public static int cudnnGetConvolutionBackwardFilterAlgorithmMaxCount(cudnnHandle handle, int[] count)
public static int cudnnFindConvolutionBackwardFilterAlgorithm(cudnnHandle handle, cudnnTensorDescriptor xDesc, cudnnTensorDescriptor dyDesc, cudnnConvolutionDescriptor convDesc, cudnnFilterDescriptor dwDesc, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionBwdFilterAlgoPerf[] perfResults)
public static int cudnnFindConvolutionBackwardFilterAlgorithmEx(cudnnHandle handle, cudnnTensorDescriptor xDesc, Pointer x, cudnnTensorDescriptor dyDesc, Pointer y, cudnnConvolutionDescriptor convDesc, cudnnFilterDescriptor dwDesc, Pointer dw, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionBwdFilterAlgoPerf[] perfResults, Pointer workSpace, long workSpaceSizeInBytes)
public static int cudnnGetConvolutionBackwardFilterAlgorithm(cudnnHandle handle, cudnnTensorDescriptor xDesc, cudnnTensorDescriptor dyDesc, cudnnConvolutionDescriptor convDesc, cudnnFilterDescriptor dwDesc, int preference, long memoryLimitInBytes, int[] algo)
public static int cudnnGetConvolutionBackwardFilterAlgorithm_v7(cudnnHandle handle, cudnnTensorDescriptor srcDesc, cudnnTensorDescriptor diffDesc, cudnnConvolutionDescriptor convDesc, cudnnFilterDescriptor gradDesc, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionBwdFilterAlgoPerf[] perfResults)
public static int cudnnGetConvolutionBackwardFilterWorkspaceSize(cudnnHandle handle, cudnnTensorDescriptor xDesc, cudnnTensorDescriptor dyDesc, cudnnConvolutionDescriptor convDesc, cudnnFilterDescriptor gradDesc, int algo, long[] sizeInBytes)
public static int cudnnConvolutionBackwardFilter(cudnnHandle handle, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnConvolutionDescriptor convDesc, int algo, Pointer workSpace, long workSpaceSizeInBytes, Pointer beta, cudnnFilterDescriptor dwDesc, Pointer dw)
public static int cudnnGetConvolutionBackwardDataAlgorithmMaxCount(cudnnHandle handle, int[] count)
public static int cudnnFindConvolutionBackwardDataAlgorithm(cudnnHandle handle, cudnnFilterDescriptor wDesc, cudnnTensorDescriptor dyDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor dxDesc, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionBwdDataAlgoPerf[] perfResults)
public static int cudnnFindConvolutionBackwardDataAlgorithmEx(cudnnHandle handle, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor dxDesc, Pointer dx, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionBwdDataAlgoPerf[] perfResults, Pointer workSpace, long workSpaceSizeInBytes)
public static int cudnnGetConvolutionBackwardDataAlgorithm(cudnnHandle handle, cudnnFilterDescriptor wDesc, cudnnTensorDescriptor dyDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor dxDesc, int preference, long memoryLimitInBytes, int[] algo)
public static int cudnnGetConvolutionBackwardDataAlgorithm_v7(cudnnHandle handle, cudnnFilterDescriptor filterDesc, cudnnTensorDescriptor diffDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor gradDesc, int requestedAlgoCount, int[] returnedAlgoCount, cudnnConvolutionBwdDataAlgoPerf[] perfResults)
public static int cudnnGetConvolutionBackwardDataWorkspaceSize(cudnnHandle handle, cudnnFilterDescriptor wDesc, cudnnTensorDescriptor dyDesc, cudnnConvolutionDescriptor convDesc, cudnnTensorDescriptor dxDesc, int algo, long[] sizeInBytes)
public static int cudnnConvolutionBackwardData(cudnnHandle handle, Pointer alpha, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnConvolutionDescriptor convDesc, int algo, Pointer workSpace, long workSpaceSizeInBytes, Pointer beta, cudnnTensorDescriptor dxDesc, Pointer dx)
public static int cudnnIm2Col(cudnnHandle handle, cudnnTensorDescriptor xDesc, Pointer x, cudnnFilterDescriptor wDesc, cudnnConvolutionDescriptor convDesc, Pointer colBuffer)
public static int cudnnSoftmaxForward(cudnnHandle handle, int algo, int mode, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnSoftmaxBackward(cudnnHandle handle, int algo, int mode, Pointer alpha, cudnnTensorDescriptor yDesc, Pointer y, cudnnTensorDescriptor dyDesc, Pointer dy, Pointer beta, cudnnTensorDescriptor dxDesc, Pointer dx)
public static int cudnnCreatePoolingDescriptor(cudnnPoolingDescriptor poolingDesc)
public static int cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor poolingDesc, int mode, int maxpoolingNanOpt, int windowHeight, int windowWidth, int verticalPadding, int horizontalPadding, int verticalStride, int horizontalStride)
public static int cudnnGetPooling2dDescriptor(cudnnPoolingDescriptor poolingDesc, int[] mode, int[] maxpoolingNanOpt, int[] windowHeight, int[] windowWidth, int[] verticalPadding, int[] horizontalPadding, int[] verticalStride, int[] horizontalStride)
public static int cudnnSetPoolingNdDescriptor(cudnnPoolingDescriptor poolingDesc, int mode, int maxpoolingNanOpt, int nbDims, int[] windowDimA, int[] paddingA, int[] strideA)
public static int cudnnGetPoolingNdDescriptor(cudnnPoolingDescriptor poolingDesc, int nbDimsRequested, int[] mode, int[] maxpoolingNanOpt, int[] nbDims, int[] windowDimA, int[] paddingA, int[] strideA)
public static int cudnnGetPoolingNdForwardOutputDim(cudnnPoolingDescriptor poolingDesc, cudnnTensorDescriptor inputTensorDesc, int nbDims, int[] outputTensorDimA)
public static int cudnnGetPooling2dForwardOutputDim(cudnnPoolingDescriptor poolingDesc, cudnnTensorDescriptor inputTensorDesc, int[] n, int[] c, int[] h, int[] w)
public static int cudnnDestroyPoolingDescriptor(cudnnPoolingDescriptor poolingDesc)
public static int cudnnPoolingForward(cudnnHandle handle, cudnnPoolingDescriptor poolingDesc, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnPoolingBackward(cudnnHandle handle, cudnnPoolingDescriptor poolingDesc, Pointer alpha, cudnnTensorDescriptor yDesc, Pointer y, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor dxDesc, Pointer dx)
public static int cudnnCreateActivationDescriptor(cudnnActivationDescriptor activationDesc)
public static int cudnnSetActivationDescriptor(cudnnActivationDescriptor activationDesc, int mode, int reluNanOpt, double coef)
public static int cudnnGetActivationDescriptor(cudnnActivationDescriptor activationDesc, int[] mode, int[] reluNanOpt, double[] coef)
public static int cudnnDestroyActivationDescriptor(cudnnActivationDescriptor activationDesc)
public static int cudnnActivationForward(cudnnHandle handle, cudnnActivationDescriptor activationDesc, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnActivationBackward(cudnnHandle handle, cudnnActivationDescriptor activationDesc, Pointer alpha, cudnnTensorDescriptor yDesc, Pointer y, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor dxDesc, Pointer dx)
public static int cudnnCreateLRNDescriptor(cudnnLRNDescriptor normDesc)
Create an instance of LRN (Local Response Normalization) descriptor Uses lrnN=5, lrnAlpha=1e-4, lrnBeta=0.75, lrnK=2.0 as defaults from Krizhevsky'12 ImageNet paper
public static int cudnnSetLRNDescriptor(cudnnLRNDescriptor normDesc, int lrnN, double lrnAlpha, double lrnBeta, double lrnK)
Uses a window [center-lookBehind, center+lookAhead], where lookBehind = floor( (lrnN-1)/2 ), lookAhead = lrnN-lookBehind-1. Values of double parameters cast to tensor data type.
public static int cudnnGetLRNDescriptor(cudnnLRNDescriptor normDesc, int[] lrnN, double[] lrnAlpha, double[] lrnBeta, double[] lrnK)
Retrieve the settings currently stored in an LRN layer descriptor Any of the provided pointers can be NULL (no corresponding value will be returned)
public static int cudnnDestroyLRNDescriptor(cudnnLRNDescriptor lrnDesc)
public static int cudnnLRNCrossChannelForward(cudnnHandle handle, cudnnLRNDescriptor normDesc, int lrnMode, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnLRNCrossChannelBackward(cudnnHandle handle, cudnnLRNDescriptor normDesc, int lrnMode, Pointer alpha, cudnnTensorDescriptor yDesc, Pointer y, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor dxDesc, Pointer dx)
public static int cudnnDivisiveNormalizationForward(cudnnHandle handle, cudnnLRNDescriptor normDesc, int mode, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer means, Pointer temp, Pointer temp2, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnDivisiveNormalizationBackward(cudnnHandle handle, cudnnLRNDescriptor normDesc, int mode, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer means, Pointer dy, Pointer temp, Pointer temp2, Pointer beta, cudnnTensorDescriptor dXdMeansDesc, Pointer dx, Pointer dMeans)
public static int cudnnDeriveBNTensorDescriptor(cudnnTensorDescriptor derivedBnDesc, cudnnTensorDescriptor xDesc, int mode)
Derives a tensor descriptor from layer data descriptor for BatchNormalization scale, invVariance, bnBias, bnScale tensors. Use this tensor desc for bnScaleBiasMeanVarDesc and bnScaleBiasDiffDesc in Batch Normalization forward and backward functions.
public static int cudnnBatchNormalizationForwardTraining(cudnnHandle handle, int mode, Pointer alpha, Pointer beta, cudnnTensorDescriptor xDesc, Pointer x, cudnnTensorDescriptor yDesc, Pointer y, cudnnTensorDescriptor bnScaleBiasMeanVarDesc, Pointer bnScale, Pointer bnBias, double exponentialAverageFactor, Pointer resultRunningMean, Pointer resultRunningVariance, double epsilon, Pointer resultSaveMean, Pointer resultSaveInvVariance)
public static int cudnnBatchNormalizationForwardInference(cudnnHandle handle, int mode, Pointer alpha, Pointer beta, cudnnTensorDescriptor xDesc, Pointer x, cudnnTensorDescriptor yDesc, Pointer y, cudnnTensorDescriptor bnScaleBiasMeanVarDesc, Pointer bnScale, Pointer bnBias, Pointer estimatedMean, Pointer estimatedVariance, double epsilon)
Performs Batch Normalization during Inference: y[i] = bnScale[k]*(x[i]-estimatedMean[k])/sqrt(epsilon+estimatedVariance[k]) + bnBias[k] with bnScale, bnBias, runningMean, runningInvVariance tensors indexed according to spatial or per-activation mode. Refer to cudnnBatchNormalizationForwardTraining above for notes on function arguments.
public static int cudnnBatchNormalizationBackward(cudnnHandle handle, int mode, Pointer alphaDataDiff, Pointer betaDataDiff, Pointer alphaParamDiff, Pointer betaParamDiff, cudnnTensorDescriptor xDesc, Pointer x, cudnnTensorDescriptor dyDesc, Pointer dy, cudnnTensorDescriptor dxDesc, Pointer dx, cudnnTensorDescriptor dBnScaleBiasDesc, Pointer bnScale, Pointer dBnScaleResult, Pointer dBnBiasResult, double epsilon, Pointer savedMean, Pointer savedInvVariance)
public static int cudnnCreateSpatialTransformerDescriptor(cudnnSpatialTransformerDescriptor stDesc)
public static int cudnnSetSpatialTransformerNdDescriptor(cudnnSpatialTransformerDescriptor stDesc, int samplerType, int dataType, int nbDims, int[] dimA)
public static int cudnnDestroySpatialTransformerDescriptor(cudnnSpatialTransformerDescriptor stDesc)
public static int cudnnSpatialTfGridGeneratorForward(cudnnHandle handle, cudnnSpatialTransformerDescriptor stDesc, Pointer theta, Pointer grid)
public static int cudnnSpatialTfGridGeneratorBackward(cudnnHandle handle, cudnnSpatialTransformerDescriptor stDesc, Pointer dgrid, Pointer dtheta)
public static int cudnnSpatialTfSamplerForward(cudnnHandle handle, cudnnSpatialTransformerDescriptor stDesc, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer grid, Pointer beta, cudnnTensorDescriptor yDesc, Pointer y)
public static int cudnnSpatialTfSamplerBackward(cudnnHandle handle, cudnnSpatialTransformerDescriptor stDesc, Pointer alpha, cudnnTensorDescriptor xDesc, Pointer x, Pointer beta, cudnnTensorDescriptor dxDesc, Pointer dx, Pointer alphaDgrid, cudnnTensorDescriptor dyDesc, Pointer dy, Pointer grid, Pointer betaDgrid, Pointer dgrid)
public static int cudnnCreateDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc)
public static int cudnnDestroyDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc)
public static int cudnnDropoutGetStatesSize(cudnnHandle handle, long[] sizeInBytes)
public static int cudnnDropoutGetReserveSpaceSize(cudnnTensorDescriptor xdesc, long[] sizeInBytes)
public static int cudnnSetDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc, cudnnHandle handle, float dropout, Pointer states, long stateSizeInBytes, long seed)
public static int cudnnRestoreDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc, cudnnHandle handle, float dropout, Pointer states, long stateSizeInBytes, long seed)
public static int cudnnGetDropoutDescriptor(cudnnDropoutDescriptor dropoutDesc, cudnnHandle handle, float[] dropout, Pointer states, long[] seed)
public static int cudnnDropoutForward(cudnnHandle handle, cudnnDropoutDescriptor dropoutDesc, cudnnTensorDescriptor xdesc, Pointer x, cudnnTensorDescriptor ydesc, Pointer y, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnDropoutBackward(cudnnHandle handle, cudnnDropoutDescriptor dropoutDesc, cudnnTensorDescriptor dydesc, Pointer dy, cudnnTensorDescriptor dxdesc, Pointer dx, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnCreateRNNDescriptor(cudnnRNNDescriptor rnnDesc)
public static int cudnnDestroyRNNDescriptor(cudnnRNNDescriptor rnnDesc)
public static int cudnnGetRNNForwardInferenceAlgorithmMaxCount(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int[] count)
public static int cudnnFindRNNForwardInferenceAlgorithmEx(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, Pointer x, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor cxDesc, Pointer cx, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor[] yDesc, Pointer y, cudnnTensorDescriptor hyDesc, Pointer hy, cudnnTensorDescriptor cyDesc, Pointer cy, float findIntensity, int requestedAlgoCount, int[] returnedAlgoCount, cudnnAlgorithmPerformance[] perfResults, Pointer workspace, long workSpaceSizeInBytes)
public static int cudnnGetRNNForwardTrainingAlgorithmMaxCount(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int[] count)
public static int cudnnFindRNNForwardTrainingAlgorithmEx(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, Pointer x, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor cxDesc, Pointer cx, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor[] yDesc, Pointer y, cudnnTensorDescriptor hyDesc, Pointer hy, cudnnTensorDescriptor cyDesc, Pointer cy, float findIntensity, int requestedAlgoCount, int[] returnedAlgoCount, cudnnAlgorithmPerformance[] perfResults, Pointer workspace, long workSpaceSizeInBytes, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnGetRNNBackwardDataAlgorithmMaxCount(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int[] count)
public static int cudnnFindRNNBackwardDataAlgorithmEx(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] yDesc, Pointer y, cudnnTensorDescriptor[] dyDesc, Pointer dy, cudnnTensorDescriptor dhyDesc, Pointer dhy, cudnnTensorDescriptor dcyDesc, Pointer dcy, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor cxDesc, Pointer cx, cudnnTensorDescriptor[] dxDesc, Pointer dx, cudnnTensorDescriptor dhxDesc, Pointer dhx, cudnnTensorDescriptor dcxDesc, Pointer dcx, float findIntensity, int requestedAlgoCount, int[] returnedAlgoCount, cudnnAlgorithmPerformance[] perfResults, Pointer workspace, long workSpaceSizeInBytes, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnGetRNNBackwardWeightsAlgorithmMaxCount(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int[] count)
public static int cudnnFindRNNBackwardWeightsAlgorithmEx(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, Pointer x, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor[] yDesc, Pointer y, float findIntensity, int requestedAlgoCount, int[] returnedAlgoCount, cudnnAlgorithmPerformance[] perfResults, Pointer workspace, long workSpaceSizeInBytes, cudnnFilterDescriptor dwDesc, Pointer dw, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnCreatePersistentRNNPlan(cudnnRNNDescriptor rnnDesc, int minibatch, int dataType, cudnnPersistentRNNPlan plan)
public static int cudnnSetPersistentRNNPlan(cudnnRNNDescriptor rnnDesc, cudnnPersistentRNNPlan plan)
public static int cudnnDestroyPersistentRNNPlan(cudnnPersistentRNNPlan plan)
public static int cudnnSetRNNDescriptor(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int hiddenSize, int numLayers, cudnnDropoutDescriptor dropoutDesc, int inputMode, int direction, int mode, int algo, int dataType)
public static int cudnnSetRNNProjectionLayers(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int recProjSize, int outProjSize)
public static int cudnnGetRNNProjectionLayers(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int[] recProjSize, int[] outProjSize)
public static int cudnnSetRNNAlgorithmDescriptor(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, cudnnAlgorithmDescriptor algoDesc)
public static int cudnnGetRNNDescriptor(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int[] hiddenSize, int[] numLayers, cudnnDropoutDescriptor dropoutDesc, int[] inputMode, int[] direction, int[] mode, int[] algo, int[] dataType)
public static int cudnnSetRNNMatrixMathType(cudnnRNNDescriptor rnnDesc, int mType)
public static int cudnnGetRNNMatrixMathType(cudnnRNNDescriptor rnnDesc, int[] mType)
public static int cudnnGetRNNWorkspaceSize(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, long[] sizeInBytes)
public static int cudnnGetRNNTrainingReserveSize(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, long[] sizeInBytes)
public static int cudnnGetRNNParamsSize(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, cudnnTensorDescriptor xDesc, long[] sizeInBytes, int dataType)
public static int cudnnGetRNNLinLayerMatrixParams(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int pseudoLayer, cudnnTensorDescriptor xDesc, cudnnFilterDescriptor wDesc, Pointer w, int linLayerID, cudnnFilterDescriptor linLayerMatDesc, Pointer linLayerMat)
public static int cudnnGetRNNLinLayerBiasParams(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int pseudoLayer, cudnnTensorDescriptor xDesc, cudnnFilterDescriptor wDesc, Pointer w, int linLayerID, cudnnFilterDescriptor linLayerBiasDesc, Pointer linLayerBias)
public static int cudnnRNNForwardInference(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, Pointer x, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor cxDesc, Pointer cx, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor[] yDesc, Pointer y, cudnnTensorDescriptor hyDesc, Pointer hy, cudnnTensorDescriptor cyDesc, Pointer cy, Pointer workspace, long workSpaceSizeInBytes)
public static int cudnnRNNForwardTraining(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, Pointer x, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor cxDesc, Pointer cx, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor[] yDesc, Pointer y, cudnnTensorDescriptor hyDesc, Pointer hy, cudnnTensorDescriptor cyDesc, Pointer cy, Pointer workspace, long workSpaceSizeInBytes, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnRNNBackwardData(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] yDesc, Pointer y, cudnnTensorDescriptor[] dyDesc, Pointer dy, cudnnTensorDescriptor dhyDesc, Pointer dhy, cudnnTensorDescriptor dcyDesc, Pointer dcy, cudnnFilterDescriptor wDesc, Pointer w, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor cxDesc, Pointer cx, cudnnTensorDescriptor[] dxDesc, Pointer dx, cudnnTensorDescriptor dhxDesc, Pointer dhx, cudnnTensorDescriptor dcxDesc, Pointer dcx, Pointer workspace, long workSpaceSizeInBytes, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnRNNBackwardWeights(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int seqLength, cudnnTensorDescriptor[] xDesc, Pointer x, cudnnTensorDescriptor hxDesc, Pointer hx, cudnnTensorDescriptor[] yDesc, Pointer y, Pointer workspace, long workSpaceSizeInBytes, cudnnFilterDescriptor dwDesc, Pointer dw, Pointer reserveSpace, long reserveSpaceSizeInBytes)
public static int cudnnCreateCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc)
public static int cudnnSetCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc, int compType)
public static int cudnnGetCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc, int[] compType)
public static int cudnnDestroyCTCLossDescriptor(cudnnCTCLossDescriptor ctcLossDesc)
public static int cudnnCTCLoss(cudnnHandle handle, cudnnTensorDescriptor probsDesc, Pointer probs, int[] labels, int[] labelLengths, int[] inputLengths, Pointer costs, cudnnTensorDescriptor gradientsDesc, Pointer gradients, int algo, cudnnCTCLossDescriptor ctcLossDesc, Pointer workspace, long workSpaceSizeInBytes)
public static int cudnnGetCTCLossWorkspaceSize(cudnnHandle handle, cudnnTensorDescriptor probsDesc, cudnnTensorDescriptor gradientsDesc, int[] labels, int[] labelLengths, int[] inputLengths, int algo, cudnnCTCLossDescriptor ctcLossDesc, long[] sizeInBytes)
public static int cudnnCreateAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc)
public static int cudnnSetAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc, int algorithm)
public static int cudnnGetAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc, int[] algorithm)
public static int cudnnCopyAlgorithmDescriptor(cudnnAlgorithmDescriptor src, cudnnAlgorithmDescriptor dest)
public static int cudnnDestroyAlgorithmDescriptor(cudnnAlgorithmDescriptor algoDesc)
public static int cudnnCreateAlgorithmPerformance(cudnnAlgorithmPerformance[] algoPerf, int numberToCreate)
public static int cudnnSetAlgorithmPerformance(cudnnAlgorithmPerformance algoPerf, cudnnAlgorithmDescriptor algoDesc, int status, float time, long memory)
public static int cudnnGetAlgorithmPerformance(cudnnAlgorithmPerformance algoPerf, cudnnAlgorithmDescriptor algoDesc, int[] status, float[] time, long[] memory)
public static int cudnnDestroyAlgorithmPerformance(cudnnAlgorithmPerformance[] algoPerf, int numberToDestroy)
public static int cudnnGetAlgorithmSpaceSize(cudnnHandle handle, cudnnAlgorithmDescriptor algoDesc, long[] algoSpaceSizeInBytes)
public static int cudnnSaveAlgorithm(cudnnHandle handle, cudnnAlgorithmDescriptor algoDesc, Pointer algoSpace, long algoSpaceSizeInBytes)
public static int cudnnRestoreAlgorithm(cudnnHandle handle, Pointer algoSpace, long algoSpaceSizeInBytes, cudnnAlgorithmDescriptor algoDesc)
public static int cudnnSetCallback(int mask, Object udata, cudnnCallback fptr)
public static int cudnnGetCallback(int[] mask, Object udata, cudnnCallback[] fptr)
public static int cudnnSetRNNDescriptor_v6(cudnnHandle handle, cudnnRNNDescriptor rnnDesc, int hiddenSize, int numLayers, cudnnDropoutDescriptor dropoutDesc, int inputMode, int direction, int mode, int algo, int dataType)
DEPRECATED routines to be removed next release : User should use the non-suffixed version (which has the API and functionality of _v6 version) Routines with _v5 suffix has the functionality of the non-suffixed routines in the CUDNN V6
public static int cudnnSetRNNDescriptor_v5(cudnnRNNDescriptor rnnDesc, int hiddenSize, int numLayers, cudnnDropoutDescriptor dropoutDesc, int inputMode, int direction, int mode, int dataType)
Copyright © 2018. All rights reserved.