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
|
/*
* Copyright (c) 2020 Samsung Electronics Co., Ltd. All Rights Reserved
* Copyright 2019 The TensorFlow Authors. All Rights Reserved.
*
* 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 "kernels/Concatenation.h"
#include "kernels/Utils.h"
#include <tensorflow/lite/kernels/internal/reference/reference_ops.h>
#include <stdexcept>
namespace luci_interpreter
{
namespace kernels
{
Concatenation::Concatenation(std::vector<const Tensor *> inputs, Tensor *output,
const ConcatenationParams ¶ms)
: KernelWithParams<ConcatenationParams>(std::move(inputs), {output}, params)
{
}
void Concatenation::configure()
{
const int num_inputs = _inputs.size();
assert(num_inputs > 0);
const Tensor *t0 = _inputs[0];
int axis = _params.axis;
if (axis < 0)
axis += t0->shape().num_dims();
assert(axis >= 0 && axis < t0->shape().num_dims());
int32_t sum_axis = t0->shape().dim(axis);
for (int i = 1; i < num_inputs; ++i)
{
const Tensor *tensor = _inputs[i];
assert(tensor->element_type() == t0->element_type());
assert(tensor->shape().num_dims() == t0->shape().num_dims());
for (int d = 0; d < t0->shape().num_dims(); ++d)
{
if (d == axis)
{
sum_axis += tensor->shape().dim(axis);
}
else
{
assert(tensor->shape().dim(d) == t0->shape().dim(d));
}
}
}
Shape output_shape = t0->shape();
output_shape.dim(axis) = sum_axis;
// TODO S8 type needs more checking: quantization parameters of all input tensors and the output
// tensor should be the same. Note that there is no such requirement for U8 type.
if (t0->element_type() == DataType::S8)
throw std::runtime_error("Unsupported type.");
output()->resize(output_shape);
}
void Concatenation::execute() const
{
switch (_inputs[0]->element_type())
{
case DataType::FLOAT32:
evalGeneric<float>();
break;
case DataType::U8:
evalQuantized();
break;
case DataType::S8:
evalGeneric<int8_t>();
break;
case DataType::S32:
evalGeneric<int32_t>();
break;
case DataType::S64:
evalGeneric<int64_t>();
break;
default:
throw std::runtime_error("Unsupported type.");
}
}
template <typename T> void Concatenation::evalGeneric() const
{
int axis = _params.axis;
if (axis < 0)
axis += output()->shape().num_dims();
VectorOfTensors<T, true> inputs(_inputs);
tflite::ConcatenationParams params{};
params.axis = axis;
params.inputs_count = _inputs.size();
tflite::reference_ops::Concatenation(params, inputs.shapes(), inputs.data(),
getTensorShape(output()), getTensorData<T>(output()));
}
void Concatenation::evalQuantized() const
{
int axis = _params.axis;
if (axis < 0)
axis += output()->shape().num_dims();
VectorOfQuantizedTensors<true> inputs(_inputs);
tflite::ConcatenationParams params{};
params.axis = axis;
params.input_zeropoint = inputs.zero_point();
params.input_scale = inputs.scale();
params.inputs_count = _inputs.size();
params.output_zeropoint = output()->zero_point();
params.output_scale = output()->scale();
tflite::reference_ops::ConcatenationWithScaling(params, inputs.shapes(), inputs.data(),
getTensorShape(output()),
getTensorData<uint8_t>(output()));
}
} // namespace kernels
} // namespace luci_interpreter
|
