Intel® Math Kernel Library 2019 Developer Reference - C
Performs DNN operations. 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.
dnnError_t dnnExecute_F32 (dnnPrimitive_t primitive, void *resources[]);
dnnError_t dnnExecute_F64 (dnnPrimitive_t primitive, void *resources[]);
primitive |
The description of a DNN operation to perform. |
||||||||||||||||||||||||||||||||||||||||||||||||||
resources |
An array of pointers to the resources according to the following table:
|
resources |
An array of pointers to the resources according to the following table:
|
Each dnnExecute function performs a DNN operation on given resources. All the required resources, including the output resources, must point to arrays having a plain layout or to the arrays allocated with a dnnAllocateBuffer function. The dnnExecute functions do not use the remaining elements of the resources array in dereferencing operations.
The following table explains more usage details of dnnExecute for specific operations:
DNN Operation |
Indices of output resources |
---|---|
Local response normalization |
The resource indexed dnnResourceWorkspace is required on forward and backward propagations. On backward propagation, use the buffer allocated for the workspace on forward propagation. |
Pooling |
The resource indexed dnnResourceWorkspace is required on forward and backward propagations. On backward propagation, use the buffer allocated for the workspace on forward propagation. |
Pooling |
The resource indexed dnnResourceWorkspace is required on forward and backward propagations. On backward propagation, use the buffer allocated for the workspace on forward propagation. |
Batch normalization |
Perform backward batch normalization with respect to data before backward batch normalization with respect to scale and shift. The resource indexed dnnResourceWorkspace is required on forward and backward propagations. On backward propagation, use the buffer allocated for the workspace on forward propagation. |
Batch normalization with specified computation method; created with dnnBatchNormalizationCreate_v2 |
The resources indexed dnnResourceMean and dnnResourceVariance are required on forward and backward propagations. On backward propagation, use the buffers allocated for the mean and variance on forward propagation. |