1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
|
/*
* Copyright (c) 2019 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#ifndef ARM_COMPUTE_GRAPH_BACKENDS_FUSED_CONVOLUTION_BATCH_NORMAZLIZATION_FUNCTION_H
#define ARM_COMPUTE_GRAPH_BACKENDS_FUSED_CONVOLUTION_BATCH_NORMAZLIZATION_FUNCTION_H
#include "arm_compute/core/Types.h"
#include "arm_compute/runtime/IFunction.h"
namespace arm_compute
{
namespace graph
{
namespace backends
{
/** Wrapper function to first apply {NE, CL}BatchNormalizationLayer on the weights and then run {NE, CL}ConvolutionLayer with the modified weights */
template <typename TargetInfo, typename FusedLayerTypes>
class FusedConvolutionBatchNormalizationFunction : public IFunction
{
public:
using TensorType = typename TargetInfo::TensorType;
using TensorConcreteType = typename TargetInfo::TensorConcreteType;
FusedConvolutionBatchNormalizationFunction(std::shared_ptr<IMemoryManager> memory_manager = nullptr)
: _conv_layer(memory_manager), _fused_batch_norm_layer(), _fused_bias(), _is_prepared(false)
{
}
/** Set the input and output tensors.
*
* @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs.
* Data types supported: QASYMM8/F16/F32.
* @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: Same as @p input.
* @param[in] bias Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
* Data type supported: Should match @p input data type.
* @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
* Data types supported: Same as @p input.
* @param[in] mean Mean values tensor. 1 dimension with size equal to the feature maps [FM]. Data types supported: Same as @p input
* @param[in] var Variance values tensor. 1 dimension with size equal to the feature maps [FM]. Data types supported: Same as @p input
* @param[in] beta Beta values tensor info. 1 dimension with size equal to the feature maps [FM]. If not provided, default value for beta is 0. Data types supported: Same as @p input
* @param[in] gamma Gamma values tensor info. 1 dimension with size equal to the feature maps [FM]. If not provided, default value for gamma is 1. Data types supported: Same as @p input
* @param[in] epsilon Small value to avoid division with zero. Default value is 0.001f.
* @param[in] conv_info Contains padding and stride information described in @ref PadStrideInfo.
* @param[in] num_groups Number of groups when performing a grouped convolution. num_groups != 1 is only supported for NCHW data layout
* @param[in] fast_math Enable fast math computation. In case this flag were set, the function could dispatch the fastest implementation
* available which may introduce a drop of accuracy as well. Default is false
* @param[in] fused_act Activation layer information in case of a fused activation.
*
*/
void configure(TensorType *input,
TensorType *weights,
TensorType *bias,
TensorType *output,
const TensorType *mean,
const TensorType *var,
const TensorType *beta,
const TensorType *gamma,
float epsilon, const PadStrideInfo &conv_info, unsigned int num_groups, bool fast_math, ActivationLayerInfo const &fused_act)
{
// We don't run any validate, as we assume that the layers have been already validated
const bool has_bias = (bias != nullptr);
const TensorType *bias_to_use;
// We check if the layer has a bias. If yes, use it in-place. If not, we need to create one
// as batch normalization might end up with a bias != 0
if(has_bias)
{
_fused_batch_norm_layer.configure(weights, mean, var, nullptr, nullptr, bias, beta, gamma, epsilon);
bias_to_use = bias;
}
else
{
_fused_batch_norm_layer.configure(weights, mean, var, nullptr, &_fused_bias, nullptr, beta, gamma, epsilon);
bias_to_use = &_fused_bias;
}
_conv_layer.configure(input, weights, bias_to_use, output, conv_info, WeightsInfo(), Size2D(1U, 1U), fused_act, fast_math, num_groups);
if(!has_bias)
{
_fused_bias.allocator()->allocate();
}
}
// Inherited methods overridden:
void run()
{
prepare();
_conv_layer.run();
}
void prepare()
{
if(!_is_prepared)
{
_fused_batch_norm_layer.run();
_is_prepared = true;
}
}
private:
typename FusedLayerTypes::ConvolutionLayer _conv_layer;
typename FusedLayerTypes::FuseBatchNormalization _fused_batch_norm_layer;
TensorConcreteType _fused_bias;
bool _is_prepared;
};
} // namespace backends
} // namespace graph
} // namespace arm_compute
#endif /* ARM_COMPUTE_GRAPH_BACKENDS_FUSED_CONVOLUTION_BATCH_NORMAZLIZATION_FUNCTION_H */
|
