|
strong |
|
strong |
Propagation kind.
Enumerator | |
---|---|
undef | Undefined propagation kind. |
forward_training | Forward data propagation (training mode). In this mode primitives perform computations necessary for subsequent backward propagation. |
forward_inference | Forward data propagation (inference mode). In this mode primitives perform only computations that are necessary for inference and omit computations that are necessary only for backward propagation. |
forward_scoring | Forward data propagation, alias for dnnl::prop_kind::forward_inference. |
forward | Forward data propagation, alias for dnnl::prop_kind::forward_training. |
backward | Backward propagation (with respect to all parameters). |
backward_data | Backward data propagation. |
backward_weights | Backward weights propagation. |
backward_bias | Backward bias propagation. |
|
strong |
Kinds of algorithms.
Enumerator | |
---|---|
convolution_auto | Convolution algorithm(either direct or Winograd) is chosen just in time. |
convolution_direct | Direct convolution. |
convolution_winograd | Winograd convolution. |
deconvolution_direct | Direct deconvolution. |
deconvolution_winograd | Winograd deconvolution. |
eltwise_relu | Eltwise: ReLU. |
eltwise_tanh | Eltwise: hyperbolic tangent non-linearity (tanh) |
eltwise_elu | Eltwise: parametric exponential linear unit (elu) |
eltwise_square | Eltwise: square. |
eltwise_abs | Eltwise: abs. |
eltwise_sqrt | Eltwise: square root. |
eltwise_swish | Eltwise: x*sigmoid(a*x) |
eltwise_linear | Eltwise: linear. |
eltwise_bounded_relu | Eltwise: bounded_relu. |
eltwise_soft_relu | Eltwise: soft_relu. |
eltwise_logistic | Eltwise: logistic. |
eltwise_exp | Eltwise: exponent. |
eltwise_gelu | Eltwise: gelu. |
lrn_across_channels | Local response normalization (LRN) across multiple channels. |
lrn_within_channel | LRN within a single channel. |
pooling_max | Max pooling. |
pooling_avg | Average pooling exclude padding, alias for dnnl::algorithm::pooling_avg_include_padding. |
pooling_avg_include_padding | Average pooling include padding. |
pooling_avg_exclude_padding | Average pooling exclude padding. |
vanilla_rnn | RNN cell. |
vanilla_lstm | LSTM cell. |
vanilla_gru | GRU cell. |
lbr_gru | GRU cell with linear before reset. Modification of original GRU cell. Differs from #dnnl_vanilla_gru in how the new memory gate is calculated: \f[ c_t = tanh(W_c*x_t + b_{c_x} + r_t*(U_c*h_{t-1}+b_{c_h})) \f] Primitive expects 4 biases on input: \f$[b_{u}, b_{r}, b_{c_x}, b_{c_h}]\f$ |
binary_add | Binary add. |
binary_mul | Binary mul. |
|
strong |
Flags for batch normalization primitive.
|
strong |
Primitive descriptor query specification.
In general should be used from C++ API since required queries are directly implemented as class members (for instance, a query for source memory descriptor).
For more information see dnnl_query_t.