1 /*
2 * Copyright (c) 2019 Arm Limited.
3 *
4 * SPDX-License-Identifier: MIT
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to
8 * deal in the Software without restriction, including without limitation the
9 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10 * sell copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24 #include "arm_compute/graph/nodes/FusedConvolutionBatchNormalizationNode.h"
25
26 #include "arm_compute/core/Utils.h"
27 #include "arm_compute/graph/Graph.h"
28 #include "arm_compute/graph/INodeVisitor.h"
29 #include "arm_compute/graph/Utils.h"
30
31 namespace arm_compute
32 {
33 namespace graph
34 {
FusedConvolutionBatchNormalizationNode(float epsilon,PadStrideInfo info,unsigned int num_groups,ConvolutionMethod method,FastMathHint fast_math_hint,ActivationLayerInfo fused_activation)35 FusedConvolutionBatchNormalizationNode::FusedConvolutionBatchNormalizationNode(float epsilon, PadStrideInfo info,
36 unsigned int num_groups,
37 ConvolutionMethod method,
38 FastMathHint fast_math_hint,
39 ActivationLayerInfo fused_activation)
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 }
45
set_convolution_method(ConvolutionMethod method)46 void FusedConvolutionBatchNormalizationNode::set_convolution_method(ConvolutionMethod method)
47 {
48 _method = method;
49 }
50
epsilon() const51 float FusedConvolutionBatchNormalizationNode::epsilon() const
52 {
53 return _epsilon;
54 }
55
convolution_method() const56 ConvolutionMethod FusedConvolutionBatchNormalizationNode::convolution_method() const
57 {
58 return _method;
59 }
60
set_fast_math_hint(FastMathHint hint)61 void FusedConvolutionBatchNormalizationNode::set_fast_math_hint(FastMathHint hint)
62 {
63 _fast_math_hint = hint;
64 }
65
fast_math_hint() const66 FastMathHint FusedConvolutionBatchNormalizationNode::fast_math_hint() const
67 {
68 return _fast_math_hint;
69 }
70
convolution_info() const71 PadStrideInfo FusedConvolutionBatchNormalizationNode::convolution_info() const
72 {
73 return _info;
74 }
75
num_groups() const76 unsigned int FusedConvolutionBatchNormalizationNode::num_groups() const
77 {
78 return _num_groups;
79 }
80
fused_activation() const81 ActivationLayerInfo FusedConvolutionBatchNormalizationNode::fused_activation() const
82 {
83 return _fused_activation;
84 }
85
set_fused_activation(ActivationLayerInfo fused_activation)86 void FusedConvolutionBatchNormalizationNode::set_fused_activation(ActivationLayerInfo fused_activation)
87 {
88 _fused_activation = fused_activation;
89 }
90
compute_output_descriptor(const TensorDescriptor & input_descriptor,const TensorDescriptor & weights_descriptor,const PadStrideInfo & info)91 TensorDescriptor FusedConvolutionBatchNormalizationNode::compute_output_descriptor(const TensorDescriptor &input_descriptor,
92 const TensorDescriptor &weights_descriptor,
93 const PadStrideInfo &info)
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 }
113
forward_descriptors()114 bool FusedConvolutionBatchNormalizationNode::forward_descriptors()
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 }
125
configure_output(size_t idx) const126 TensorDescriptor FusedConvolutionBatchNormalizationNode::configure_output(size_t idx) const
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 }
138
type() const139 NodeType FusedConvolutionBatchNormalizationNode::type() const
140 {
141 return FusedConvolutionBatchNormalizationNode::node_type;
142 }
143
accept(INodeVisitor & v)144 void FusedConvolutionBatchNormalizationNode::accept(INodeVisitor &v)
145 {
146 v.visit(*this);
147 }
148 } // namespace graph
149 } // namespace arm_compute
150