oneAPI Deep Neural Network Library (oneDNN)
Performance library for Deep Learning
1.96.0
dnnl_rnn_desc_t Struct Reference

A descriptor for an RNN operation. More...

#include <dnnl_types.h>

Collaboration diagram for dnnl_rnn_desc_t:

Public Attributes

dnnl_primitive_kind_t primitive_kind
 The kind of primitive. More...
 
dnnl_prop_kind_t prop_kind
 The kind of propagation. More...
 
dnnl_alg_kind_t cell_kind
 RNN cell kind. More...
 
dnnl_rnn_direction_t direction
 The direction of RNN primitive execution.
 
dnnl_memory_desc_t src_layer_desc
 Source layer memory descriptor.
 
dnnl_memory_desc_t src_iter_desc
 Source iteration memory descriptor for hidden state.
 
dnnl_memory_desc_t src_iter_c_desc
 Source iteration memory descriptor for cell state.
 
dnnl_memory_desc_t weights_layer_desc
 Weights layer memory descriptor.
 
dnnl_memory_desc_t weights_iter_desc
 Weights iteration memory descriptor.
 
dnnl_memory_desc_t bias_desc
 Bias memory descriptor.
 
dnnl_memory_desc_t dst_layer_desc
 Destination layer memory descriptor.
 
dnnl_memory_desc_t dst_iter_desc
 Destination iter memory descriptor for hidden state.
 
dnnl_memory_desc_t dst_iter_c_desc
 Destination iter memory descriptor for cell state.
 
dnnl_memory_desc_t weights_peephole_desc
 Weights peephole memory descriptor. More...
 
dnnl_memory_desc_t weights_projection_desc
 Weights projection memory descriptor. More...
 
dnnl_memory_desc_t diff_src_layer_desc
 Source gradient layer memory descriptor.
 
dnnl_memory_desc_t diff_src_iter_desc
 Source gradient iter memory descriptor for hidden state.
 
dnnl_memory_desc_t diff_src_iter_c_desc
 Source gradient iter memory descriptor for cell state.
 
dnnl_memory_desc_t diff_weights_layer_desc
 Weights gradient layer memory descriptor.
 
dnnl_memory_desc_t diff_weights_iter_desc
 Weights gradient iter memory descriptor.
 
dnnl_memory_desc_t diff_bias_desc
 Bias gradient memory descriptor.
 
dnnl_memory_desc_t diff_dst_layer_desc
 Destination gradient layer memory descriptor.
 
dnnl_memory_desc_t diff_dst_iter_desc
 Destination gradient iteration memory descriptor for hidden state.
 
dnnl_memory_desc_t diff_dst_iter_c_desc
 Destination gradient iteration memory descriptor for cell state.
 
dnnl_memory_desc_t diff_weights_peephole_desc
 Weights gradient peephole memory descriptor. More...
 
dnnl_memory_desc_t diff_weights_projection_desc
 Weights gradient projection memory descriptor. More...
 
unsigned int flags
 RNN cell flags.
 
dnnl_alg_kind_t activation_kind
 Activation function used for vanilla_rnn cell kind. More...
 

Detailed Description

A descriptor for an RNN operation.

Member Data Documentation

◆ primitive_kind

dnnl_primitive_kind_t dnnl_rnn_desc_t::primitive_kind

The kind of primitive.

Used for self-identifying the primitive descriptor. Must be dnnl_rnn.

◆ prop_kind

dnnl_prop_kind_t dnnl_rnn_desc_t::prop_kind

The kind of propagation.

Possible values: dnnl_forward_training, dnnl_forward_inference, and dnnl_backward.

◆ cell_kind

dnnl_alg_kind_t dnnl_rnn_desc_t::cell_kind

RNN cell kind.

Must be one of dnnl_vanilla_rnn, dnnl_vanilla_lstm, dnnl_vanilla_gru, or dnnl_lbr_gru.

◆ weights_peephole_desc

dnnl_memory_desc_t dnnl_rnn_desc_t::weights_peephole_desc

Weights peephole memory descriptor.

This memory descriptor is equal to zero memory descriptor in case of non-peephole LSTMs and other non-LSTM RNNs.

◆ weights_projection_desc

dnnl_memory_desc_t dnnl_rnn_desc_t::weights_projection_desc

Weights projection memory descriptor.

This memory descriptor is equal to zero memory descriptor in case of non-projection LSTMs and other non-LSTM RNNs.

◆ diff_weights_peephole_desc

dnnl_memory_desc_t dnnl_rnn_desc_t::diff_weights_peephole_desc

Weights gradient peephole memory descriptor.

This memory descriptor is equal to zero memory descriptor in case of non-peephole LSTMs and other non-LSTM RNNs.

◆ diff_weights_projection_desc

dnnl_memory_desc_t dnnl_rnn_desc_t::diff_weights_projection_desc

Weights gradient projection memory descriptor.

This memory descriptor is equal to zero memory descriptor in case of non-projection LSTMs and other non-LSTM RNNs.

◆ activation_kind

dnnl_alg_kind_t dnnl_rnn_desc_t::activation_kind

Activation function used for vanilla_rnn cell kind.

Must be either dnnl_eltwise_relu or dnnl_eltwise_tanh.


The documentation for this struct was generated from the following file: