Note

The Deep Neural Network (DNN) component in Intel MKL is deprecated and will be removed in a future release. You can continue to use optimized functions for deep neural networks through Intel Math Kernel Library for Deep Neural Networks.

Intel MKL DNN functions use the following enumerated types:

dnnError_t

Each DNN function returns a value of type dnnError_t that indicates completion status of the function.

Value

Meaning

E_SUCCESS

The operation completed successfully.

E_INCORRECT_INPUT_PARAMETER

The value of one of input parameters is incorrect.

E_MEMORY_ERROR

Allocation of some memory failed.

E_UNSUPPORTED_DIMENSION

The dimension of input or output array is not supported.

E_UNIMPLEMENTED

The operation is not implemented.

dnnResourceType_t

At the execution stage, DNN operations receive a resources array, which contains pointers to the operation resources. The value of type dnnResourceType_t specifies the index of the pointer to a specific resource in this array:

Value

Resource pointed by

resources(Value)

dnnResourceSrc, dnnResourceFrom

Input data.

dnnResourceDst, dnnResourceTo

Output data.

dnnResourceFilter

Filter data.

dnnResourceDiffScaleShift

Scale and shift data.

dnnResourceBias

Bias data.

dnnResourceDiffSrc

Gradient with respect to input data.

dnnResourceDiffFilter

Gradient with respect to filter data.

dnnResourceScaleShift

Gradient with respect to scale and shift.

dnnResourceDiffBias

Gradient with respect to bias data.

dnnResourceDiffDst

Gradient with respect to output data.

dnnResourceWorkspace

Workspace.

dnnResourceMultipleSrc, ... , dnnResourceMultipleSrc+ k

For multiple input data, elements of the input, from first to last, where k [0, 8].

dnnResourceMultipleDst, ... , dnnResourceMultipleDst+ k

For multiple output data, elements of the output, from first to last.

dnnResourceNumber

The number of elements in the resources array.

dnnAlgorithm_t

At the setup stage, the value of type dnnAlgorithm_t species the implementation of convolution and pooling:

Value

Meaning

dnnAlgorithmConvolutionDirect

Direct convolution.

dnnAlgorithmPoolingMax

Maximum pooling.

dnnAlgorithmPoolingMin

Minimum pooling.

dnnAlgorithmPoolingAvgIncludePadding

Average pooling (padded elements contribute to average value).

dnnAlgorithmPoolingAvgExcludePadding

Average pooling (padded elements do not contribute to average value).

dnnBorder_t

At the setup stage, the value of type dnnBorder_t specifies the method to pad the input array if padding is necessary:

Value

Meaning

dnnBorderZeros

Pad with zeros.

dnnBorderZerosAsymm

Asymmetrically pad with zeros.

dnnBorderExtrapolation

Extrapolate based on the input data.

dnnBatchNormalizationFlag_t

At the setup stage, the value of type dnnBatchNormalizationFlag_t specifies the method to perform batch normalization:

Value

Meaning

dnnUseInputMeanVariance

If this value is set, use externally calculated mean and variance as input for the forward propagation step. Otherwise, for backward propagation, use mean and variance output from the forward propagation step.

dnnUseScaleShift

Use learnable gamma and beta parameters.

Для получения подробной информации о возможностях оптимизации компилятора обратитесь к нашему Уведомлению об оптимизации.
Выберите цвет фиксируемой кнопки: 
Orange (only for download buttons)