Descriptor for a batch normalization backward propagation primitive. More...
#include <dnnl.hpp>
Collaboration diagram for dnnl::batch_normalization_backward::desc:
Public Member Functions | |
| desc (prop_kind prop_kind, const memory::desc &diff_data_desc, const memory::desc &data_desc, float epsilon, normalization_flags flags) | |
| Constructs a batch normalization descriptor for backward propagation. More... | |
Detailed Description
Descriptor for a batch normalization backward propagation primitive.
Constructor & Destructor Documentation
◆ desc()
|
inline |
Constructs a batch normalization descriptor for backward propagation.
Inputs:
- src (dnnl::primitive_desc_base::src_desc (0))
- mean (dnnl::primitive_desc_base::src_desc (1))
- variance (dnnl::primitive_desc_base::src_desc (2))
- diff_dst (dnnl::primitive_desc_base::diff_dst_desc (0))
- scale_and_shift (dnnl::primitive_desc_base::weights_desc (0)), if dnnl::normalization_flags::use_scale_shift bit-flag is set in
flags - workspace (dnnl::primitive_desc_base::workspace_desc (0)), if dnnl::normalization_flags::fuse_norm_relu bit-flag is set in
flags
Outputs:
- diff_src (dnnl::primitive_desc_base::diff_src_desc (0))
- diff_scale_and_shift (dnnl::primitive_desc_base::diff_weights_desc (0)), if dnnl::normalization_flags::use_scale_shift bit-flag is set in
flagsandprop_kind= dnnl::prop_kind::backward
- Parameters
-
prop_kind Propagation kind. Possible values are dnnl::prop_kind::backward_data and dnnl::prop_kind::backward (diffs for all parameters are computed in this case). diff_data_desc Diff source and diff destination memory descriptor. data_desc Source memory descriptor. epsilon Batch normalization epsilon parameter. flags Batch normalization flags (dnnl::normalization_flags).
The documentation for this struct was generated from the following file:
- include/dnnl.hpp