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
134
135
136
137
138
|
/*
* 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 "ConcatLayer.h"
#include "tensorflow/contrib/lite/kernels/internal/optimized/optimized_ops.h"
#include "kernel/cpu/OperationUtils.h"
namespace neurun
{
namespace kernel
{
namespace cpu
{
ConcatLayer::ConcatLayer()
: _inputDataPtrs(), _outputData(nullptr), _axis(0), _inputShapes(), _outputShape(),
_inputType(OperandType::SCALAR_FLOAT32)
{
// DO NOTHING
}
bool ConcatLayer::concatenationFloat32()
{
uint32_t num_inputs = _inputShapes.size();
tflite::ConcatenationParams op_params;
op_params.axis = _axis;
op_params.inputs_count = num_inputs;
std::vector<::tflite::RuntimeShape *> inputDimsPtr;
std::vector<::tflite::RuntimeShape> inputDims;
inputDimsPtr.reserve(num_inputs);
inputDims.reserve(num_inputs);
for (uint32_t i = 0; i < num_inputs; i++)
{
inputDims.push_back(convertShapeToTFLiteShape(_inputShapes[i]));
inputDimsPtr.push_back(&inputDims[i]);
}
std::vector<const float *> inputFloatPtrs;
for (auto ptr : _inputDataPtrs)
{
inputFloatPtrs.emplace_back(reinterpret_cast<const float *>(ptr));
}
::tflite::optimized_ops::Concatenation<float>(
op_params, inputDimsPtr.data(), inputFloatPtrs.data(),
convertShapeToTFLiteShape(_outputShape), reinterpret_cast<float *>(_outputData));
return true;
}
bool ConcatLayer::concatenationQuant8()
{
int num_inputs = _inputShapes.size();
std::vector<int32_t> input_zeropoints(num_inputs);
std::vector<float> input_scales(num_inputs);
for (uint32_t i = 0; i < num_inputs; i++)
{
input_zeropoints[i] = _inputShapes[i].offset;
input_scales[i] = _inputShapes[i].scale;
}
tflite::ConcatenationParams op_params;
op_params.axis = _axis;
op_params.inputs_count = num_inputs;
op_params.input_zeropoint = input_zeropoints.data();
op_params.input_scale = input_scales.data();
op_params.output_zeropoint = _outputShape.offset;
op_params.output_scale = _outputShape.scale;
std::vector<::tflite::RuntimeShape *> inputDimsPtr;
std::vector<::tflite::RuntimeShape> inputDims;
inputDimsPtr.reserve(num_inputs);
inputDims.reserve(num_inputs);
for (uint32_t i = 0; i < num_inputs; i++)
{
inputDims.push_back(convertShapeToTFLiteShape(_inputShapes[i]));
inputDimsPtr.push_back(&inputDims[i]);
}
::tflite::optimized_ops::Concatenation<uint8_t>(
op_params, inputDimsPtr.data(), _inputDataPtrs.data(),
convertShapeToTFLiteShape(_outputShape), _outputData);
return true;
}
void ConcatLayer::configure(const std::vector<const uint8_t *> &inputDataPtrs,
const std::vector<Shape> &inputShapes, int32_t axis,
uint8_t *outputData, const Shape outputShape)
{
_inputDataPtrs = inputDataPtrs;
for (auto shape : inputShapes)
{
_inputShapes.emplace_back(shape);
_inputType = shape.type;
}
_axis = axis;
_outputData = outputData;
_outputShape = outputShape;
}
void ConcatLayer::run()
{
if (_inputType == OperandType::TENSOR_FLOAT32)
{
concatenationFloat32();
}
else if (_inputType == OperandType::TENSOR_QUANT8_ASYMM)
{
throw std::runtime_error{"ConcatLayer : Not tested for TENSOR_QUANT8_ASYMM"};
// concatenationQuant8();
}
}
} // namespace cpu
} // namespace kernel
} // namespace neurun
|
