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
|
/*
* Copyright (c) 2018 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright (C) 2017 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "FullyConnectedLayer.h"
#include <cker/operation/FullyConnected.h>
#include "OperationUtils.h"
namespace neurun
{
namespace backend
{
namespace cpu
{
namespace kernel
{
FullyConnectedLayer::FullyConnectedLayer()
: _inputData(), _weightsData(), _biasData(), _outputData(), _inputDescr(), _weightsDescr(),
_biasDescr(), _outputDescr(), _activation(ir::Activation::NONE),
_inputType(OperandType::FLOAT32)
{
// DO NOTHING
}
void FullyConnectedLayer::fullyConnectedFloat32()
{
float output_activation_min, output_activation_max;
CalculateActivationRangeFloat(_activation, &output_activation_min, &output_activation_max);
nnfw::cker::FullyConnectedParams op_params;
op_params.float_activation_min = output_activation_min;
op_params.float_activation_max = output_activation_max;
nnfw::cker::FullyConnected(op_params, convertToExtendedCkerShape(_inputDescr), _inputData.f,
convertToExtendedCkerShape(_weightsDescr), _weightsData.f,
convertToExtendedCkerShape(_biasDescr), _biasData.f,
convertToExtendedCkerShape(_outputDescr), _outputData.f);
}
// executionMutex is used to protect concurrent access of non-threadsafe resources
// like gemmlowp::GemmContext.
void FullyConnectedLayer::fullyConnectedQuant8()
{
float real_multiplier = 0.0;
int32_t output_multiplier = 0;
int32_t output_shift = 0;
int32_t output_activation_min = 0;
int32_t output_activation_max = 0;
GetQuantizedConvolutionMultiplier(_inputDescr, _weightsDescr, _biasDescr, _outputDescr,
&real_multiplier);
QuantizeMultiplier(real_multiplier, &output_multiplier, &output_shift);
CalculateActivationRangeUint8(_activation, _outputDescr, &output_activation_min,
&output_activation_max);
nnfw::cker::FullyConnectedParams op_params;
op_params.input_offset = -_inputDescr.offset;
op_params.weights_offset = -_weightsDescr.offset;
op_params.output_offset = _outputDescr.offset;
op_params.output_multiplier = output_multiplier;
op_params.output_shift = output_shift;
op_params.quantized_activation_min = output_activation_min;
op_params.quantized_activation_max = output_activation_max;
nnfw::cker::FullyConnected(op_params, convertToExtendedCkerShape(_inputDescr), _inputData.u8,
convertToExtendedCkerShape(_weightsDescr), _weightsData.u8,
convertToExtendedCkerShape(_biasDescr), _biasData.i32,
convertToExtendedCkerShape(_outputDescr), _outputData.u8);
}
void FullyConnectedLayer::configure(uint8_t *inputData, const TensorDescriptor inputDescr,
uint8_t *weightsData, const TensorDescriptor weightsDescr,
uint8_t *biasData, const TensorDescriptor biasDescr,
ir::Activation activation, uint8_t *outputData,
const TensorDescriptor outputDescr)
{
_inputData.u8 = inputData;
_inputDescr = inputDescr;
_inputType = inputDescr.type;
_weightsData.u8 = weightsData;
_weightsDescr = weightsDescr;
_biasData.u8 = biasData;
_biasDescr = biasDescr;
_activation = activation;
_outputData.u8 = outputData;
_outputDescr = outputDescr;
}
void FullyConnectedLayer::run()
{
if (_inputType == OperandType::FLOAT32)
{
fullyConnectedFloat32();
}
else if (_inputType == OperandType::QUANT8_ASYMM)
{
fullyConnectedQuant8();
}
}
} // namespace kernel
} // namespace cpu
} // namespace backend
} // namespace neurun
|
