Compute Library
 21.02
FusedConvolutionBatchNormalizationNode Class Referencefinal

Batch Normalization node. More...

#include <FusedConvolutionBatchNormalizationNode.h>

Collaboration diagram for FusedConvolutionBatchNormalizationNode:
[legend]

Public Member Functions

 FusedConvolutionBatchNormalizationNode (float epsilon, PadStrideInfo info, unsigned int num_groups=1, ConvolutionMethod method=ConvolutionMethod::Default, FastMathHint fast_math_hint=FastMathHint::Disabled, ActivationLayerInfo fused_activation=ActivationLayerInfo())
 Constructor. More...
 
float epsilon () const
 Epsilon parameter accessor. More...
 
ActivationLayerInfo fused_activation () const
 Returns fused activation. More...
 
void set_fused_activation (ActivationLayerInfo fused_activation)
 Sets fused activation. More...
 
void set_convolution_method (ConvolutionMethod method)
 Sets the convolution layer method to use. More...
 
unsigned int num_groups () const
 Number of groups in convolution accessor. More...
 
ConvolutionMethod convolution_method () const
 Convolution layer method accessor. More...
 
void set_fast_math_hint (FastMathHint hint)
 Sets the fast math fast hint. More...
 
FastMathHint fast_math_hint () const
 Fast math hint accessor. More...
 
PadStrideInfo convolution_info () const
 Convolution metadata accessor. More...
 
NodeType type () const override
 Returns node's type. More...
 
bool forward_descriptors () override
 Forwards descriptor information to outputs if possible. More...
 
TensorDescriptor configure_output (size_t idx) const override
 Calculates output configuration. More...
 
void accept (INodeVisitor &v) override
 Accepts a node visitor. More...
 
- Public Member Functions inherited from INode
 INode ()
 Constructor. More...
 
virtual ~INode ()=default
 Destructor. More...
 
 INode (const INode &)=delete
 Prevent instances of this class from being copied (As this class contains pointers) More...
 
INodeoperator= (const INode &)=delete
 Prevent instances of this class from being copy assigned (As this class contains pointers) More...
 
 INode (INode &&)=default
 Allow instances of this class to be moved. More...
 
INodeoperator= (INode &&)=default
 Allow instances of this class to be move assigned. More...
 
virtual Status validate () const
 Validate node. More...
 
std::string name () const
 Returns node's name. More...
 
NodeID id () const
 Returns node's ID. More...
 
const Graphgraph () const
 Returns node's Graph. More...
 
Graphgraph ()
 Returns node's Graph. More...
 
void set_graph (Graph *g)
 Sets the graph that this node is registered to. More...
 
void set_id (NodeID id)
 Sets the node id. More...
 
void set_common_node_parameters (NodeParams common_params)
 Sets common node parameters. More...
 
void set_requested_target (Target target)
 Sets target preference. More...
 
void set_assigned_target (Target target)
 Sets the final execution target. More...
 
void set_output_tensor (TensorID tid, size_t idx)
 Sets the output tensor of at a given index. More...
 
const std::vector< TensorID > & inputs () const
 Returns inputs of the node. More...
 
const std::vector< TensorID > & outputs () const
 Returns outputs of the node. More...
 
const std::vector< EdgeID > & input_edges () const
 Returns input edge set. More...
 
const std::set< EdgeID > & output_edges () const
 Returns output edge set. More...
 
TensorID input_id (size_t idx) const
 Returns the tensor ID of a given input of the node. More...
 
TensorID output_id (size_t idx) const
 Returns the tensor ID of a given output of the node. More...
 
Tensorinput (size_t idx) const
 Returns the tensor of a given input of the node. More...
 
Tensoroutput (size_t idx) const
 Returns the tensor of a given output of the node. More...
 
EdgeID input_edge_id (size_t idx) const
 Returns the edge ID of a given input of the node. More...
 
Edgeinput_edge (size_t idx) const
 Returns the edge of a given input of the node. More...
 
size_t num_inputs () const
 Returns number of inputs of the node. More...
 
size_t num_outputs () const
 Returns number of outputs of the node. More...
 
NodeParams common_node_params () const
 Returns common node parameters. More...
 
Target requested_target () const
 Returns requested target for this node. More...
 
Target assigned_target () const
 Returns assigned target for this node. More...
 

Static Public Member Functions

static TensorDescriptor compute_output_descriptor (const TensorDescriptor &input_descriptor, const TensorDescriptor &weights_descriptor, const PadStrideInfo &info)
 Computes convolution output descriptor. More...
 

Static Public Attributes

static constexpr NodeType node_type = NodeType::FusedConvolutionBatchNormalizationLayer
 

Detailed Description

Batch Normalization node.

Definition at line 34 of file FusedConvolutionBatchNormalizationNode.h.

Constructor & Destructor Documentation

◆ FusedConvolutionBatchNormalizationNode()

FusedConvolutionBatchNormalizationNode ( float  epsilon,
PadStrideInfo  info,
unsigned int  num_groups = 1,
ConvolutionMethod  method = ConvolutionMethod::Default,
FastMathHint  fast_math_hint = FastMathHint::Disabled,
ActivationLayerInfo  fused_activation = ActivationLayerInfo() 
)

Constructor.

Parameters
[in]epsilonEpsilon parameter.
[in]infoConvolution layer attributes.
[in]num_groups(Optional) Number of groups (Defaults to 1)
[in]method(Optional) Convolution method to use
[in]fast_math_hint(Optional) Fast math hint
[in]fused_activation(Optional) Fused activation layer. Disabled if not specified

Definition at line 35 of file FusedConvolutionBatchNormalizationNode.cpp.

40  : _epsilon(epsilon), _info(std::move(info)), _num_groups(num_groups), _method(method), _fast_math_hint(fast_math_hint), _fused_activation(fused_activation)
41 {
42  _input_edges.resize(7, EmptyEdgeID);
43  _outputs.resize(1, NullTensorID);
44 }
unsigned int num_groups() const
Number of groups in convolution accessor.
ActivationLayerInfo fused_activation() const
Returns fused activation.
constexpr EdgeID EmptyEdgeID
Definition: Types.h:75
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
constexpr TensorID NullTensorID
Constant NodeID specifying an equivalent of null node.
Definition: Types.h:71

Member Function Documentation

◆ accept()

void accept ( INodeVisitor v)
overridevirtual

Accepts a node visitor.

Parameters
[in]vVisitor to accept

Implements INode.

Definition at line 144 of file FusedConvolutionBatchNormalizationNode.cpp.

References INodeVisitor::visit().

145 {
146  v.visit(*this);
147 }

◆ compute_output_descriptor()

TensorDescriptor compute_output_descriptor ( const TensorDescriptor input_descriptor,
const TensorDescriptor weights_descriptor,
const PadStrideInfo info 
)
static

Computes convolution output descriptor.

Parameters
[in]input_descriptorInput descriptor
[in]weights_descriptorWeights descriptor
[in]infoConvolution operation attributes
Returns
Output descriptor

Definition at line 91 of file FusedConvolutionBatchNormalizationNode.cpp.

References arm_compute::CHANNEL, arm_compute::test::validation::data_layout, arm_compute::graph::get_dimension_idx(), arm_compute::graph::get_dimension_size(), arm_compute::HEIGHT, input_height, input_width, TensorDescriptor::layout, arm_compute::scaled_dimensions(), TensorShape::set(), TensorDescriptor::shape, and arm_compute::WIDTH.

Referenced by FusedConvolutionBatchNormalizationNode::configure_output().

94 {
95  unsigned int output_width = 0;
96  unsigned int output_height = 0;
97 
98  const unsigned int input_width = get_dimension_size(input_descriptor, DataLayoutDimension::WIDTH);
99  const unsigned int input_height = get_dimension_size(input_descriptor, DataLayoutDimension::HEIGHT);
100  const unsigned int kernel_width = get_dimension_size(weights_descriptor, DataLayoutDimension::WIDTH);
101  const unsigned int kernel_height = get_dimension_size(weights_descriptor, DataLayoutDimension::HEIGHT);
102 
103  std::tie(output_width, output_height) = scaled_dimensions(input_width, input_height, kernel_width, kernel_height, info);
104 
105  const DataLayout data_layout = input_descriptor.layout;
106  TensorDescriptor output_descriptor = input_descriptor;
107  output_descriptor.shape.set(get_dimension_idx(data_layout, DataLayoutDimension::WIDTH), output_width);
108  output_descriptor.shape.set(get_dimension_idx(data_layout, DataLayoutDimension::HEIGHT), output_height);
109  output_descriptor.shape.set(get_dimension_idx(data_layout, DataLayoutDimension::CHANNEL), weights_descriptor.shape[3]);
110 
111  return output_descriptor;
112 }
size_t get_dimension_size(const TensorDescriptor &descriptor, const DataLayoutDimension data_layout_dimension)
Get size of a tensor&#39;s given dimension depending on its layout.
Definition: Utils.cpp:123
const DataLayout data_layout
Definition: Im2Col.cpp:151
std::pair< unsigned int, unsigned int > scaled_dimensions(int width, int height, int kernel_width, int kernel_height, const PadStrideInfo &pad_stride_info, const Size2D &dilation=Size2D(1U, 1U))
Returns expected width and height of output scaled tensor depending on dimensions rounding mode...
Definition: Utils.cpp:419
ScaleKernelInfo info(interpolation_policy, default_border_mode, PixelValue(), sampling_policy, false)
DataLayout
[DataLayout enum definition]
Definition: Types.h:120
size_t get_dimension_idx(DataLayout data_layout, const DataLayoutDimension data_layout_dimension)
Get index of a tensor&#39;s given dimension depending on its layout.
Definition: Utils.cpp:129

◆ configure_output()

TensorDescriptor configure_output ( size_t  idx) const
overridevirtual

Calculates output configuration.

Parameters
[in]idxOutput index to configure
Returns
Output descriptor configuration

Implements INode.

Definition at line 126 of file FusedConvolutionBatchNormalizationNode.cpp.

References ARM_COMPUTE_ERROR_ON, ARM_COMPUTE_UNUSED, FusedConvolutionBatchNormalizationNode::compute_output_descriptor(), Tensor::desc(), INode::input(), arm_compute::test::validation::output_info, and arm_compute::test::validation::src.

Referenced by FusedConvolutionBatchNormalizationNode::forward_descriptors().

127 {
128  ARM_COMPUTE_UNUSED(idx);
129  const Tensor *src = input(0);
130  const Tensor *weights = input(1);
131 
132  ARM_COMPUTE_ERROR_ON(src == nullptr || weights == nullptr);
133 
134  TensorDescriptor output_info = compute_output_descriptor(src->desc(), weights->desc(), _info);
135 
136  return output_info;
137 }
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:466
SimpleTensor< float > src
Definition: DFT.cpp:155
#define ARM_COMPUTE_UNUSED(...)
To avoid unused variables warnings.
Definition: Error.h:152
static TensorDescriptor compute_output_descriptor(const TensorDescriptor &input_descriptor, const TensorDescriptor &weights_descriptor, const PadStrideInfo &info)
Computes convolution output descriptor.
Tensor * input(size_t idx) const
Returns the tensor of a given input of the node.
Definition: INode.cpp:150

◆ convolution_info()

PadStrideInfo convolution_info ( ) const

Convolution metadata accessor.

Returns
Convolution information

Definition at line 71 of file FusedConvolutionBatchNormalizationNode.cpp.

72 {
73  return _info;
74 }

◆ convolution_method()

ConvolutionMethod convolution_method ( ) const

Convolution layer method accessor.

Note
This is an indication on which convolution layer implementation to use, if it fails to be created the library's heuristic approach will be used
Returns
Convolution layer method to be used by the node

Definition at line 56 of file FusedConvolutionBatchNormalizationNode.cpp.

57 {
58  return _method;
59 }

◆ epsilon()

float epsilon ( ) const

Epsilon parameter accessor.

Returns
Epsilon parameter

Definition at line 51 of file FusedConvolutionBatchNormalizationNode.cpp.

52 {
53  return _epsilon;
54 }

◆ fast_math_hint()

FastMathHint fast_math_hint ( ) const

Fast math hint accessor.

Returns
Fast math hint to be used by the node

Definition at line 66 of file FusedConvolutionBatchNormalizationNode.cpp.

67 {
68  return _fast_math_hint;
69 }

◆ forward_descriptors()

bool forward_descriptors ( )
overridevirtual

Forwards descriptor information to outputs if possible.

Returns
True if descriptor information could be forwarded otherwise false

Implements INode.

Definition at line 114 of file FusedConvolutionBatchNormalizationNode.cpp.

References ARM_COMPUTE_ERROR_ON, FusedConvolutionBatchNormalizationNode::configure_output(), Tensor::desc(), arm_compute::test::validation::dst, INode::input_id(), arm_compute::graph::NullTensorID, INode::output(), and INode::output_id().

115 {
116  if((input_id(0) != NullTensorID) && (input_id(1) != NullTensorID) && (output_id(0) != NullTensorID))
117  {
118  Tensor *dst = output(0);
119  ARM_COMPUTE_ERROR_ON(dst == nullptr);
120  dst->desc() = configure_output(0);
121  return true;
122  }
123  return false;
124 }
TensorDescriptor configure_output(size_t idx) const override
Calculates output configuration.
#define ARM_COMPUTE_ERROR_ON(cond)
If the condition is true then an error message is printed and an exception thrown.
Definition: Error.h:466
Tensor * output(size_t idx) const
Returns the tensor of a given output of the node.
Definition: INode.cpp:158
TensorID input_id(size_t idx) const
Returns the tensor ID of a given input of the node.
Definition: INode.cpp:137
TensorID output_id(size_t idx) const
Returns the tensor ID of a given output of the node.
Definition: INode.cpp:144
constexpr TensorID NullTensorID
Constant NodeID specifying an equivalent of null node.
Definition: Types.h:71

◆ fused_activation()

ActivationLayerInfo fused_activation ( ) const

Returns fused activation.

Returns
Fused activation

Definition at line 81 of file FusedConvolutionBatchNormalizationNode.cpp.

Referenced by FusedConvolutionBatchNormalizationNode::set_fused_activation().

82 {
83  return _fused_activation;
84 }

◆ num_groups()

unsigned int num_groups ( ) const

Number of groups in convolution accessor.

Returns
Number of groups in convolution

Definition at line 76 of file FusedConvolutionBatchNormalizationNode.cpp.

77 {
78  return _num_groups;
79 }

◆ set_convolution_method()

void set_convolution_method ( ConvolutionMethod  method)

Sets the convolution layer method to use.

Parameters
[in]methodMethod to use for convolution

Definition at line 46 of file FusedConvolutionBatchNormalizationNode.cpp.

47 {
48  _method = method;
49 }

◆ set_fast_math_hint()

void set_fast_math_hint ( FastMathHint  hint)

Sets the fast math fast hint.

Parameters
[in]hintHint to use for convolution

Definition at line 61 of file FusedConvolutionBatchNormalizationNode.cpp.

62 {
63  _fast_math_hint = hint;
64 }

◆ set_fused_activation()

void set_fused_activation ( ActivationLayerInfo  fused_activation)

Sets fused activation.

Parameters
[in]fused_activationFused activation to set

Definition at line 86 of file FusedConvolutionBatchNormalizationNode.cpp.

References FusedConvolutionBatchNormalizationNode::fused_activation().

87 {
88  _fused_activation = fused_activation;
89 }
ActivationLayerInfo fused_activation() const
Returns fused activation.

◆ type()

NodeType type ( ) const
overridevirtual

Field Documentation

◆ node_type


The documentation for this class was generated from the following files: