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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
|
/*
* Copyright (c) 2019 Samsung Electronics Co., Ltd. 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 "ONNXHelpers.h"
#include "AttributeHelpers.h"
#include "MirInterpreter.h"
#include "mir/ops/ConstantOp.h"
#include "mir/ShapeRange.h"
#include "mir/Tensor.h"
#include "mir/TensorVariant.h"
#include "mir/Index.h"
namespace mir_onnx
{
const int64_t firstUnknownOpset = 13;
template <typename T> static mir::Shape constantToShapeT(const mir::TensorVariant &t)
{
const mir::Shape &t_shape = t.getShape();
mir::Tensor<T> input(t);
if (t_shape.rank() != 1)
throw std::runtime_error("only 1-d tensors supported as a shape input");
mir::Shape target_shape;
std::int32_t rank = t_shape.dim(0);
target_shape.resize(rank);
for (int i = 0; i < rank; ++i)
target_shape.dim(i) = static_cast<std::int32_t>(input.at(mir::Index{i}));
return target_shape;
}
mir::Shape constantToShape(const mir::ops::ConstantOp *op)
{
const auto &t = op->getValue();
mir::DataType d_type = t.getElementType();
if (t.getType().isQuantized())
throw std::runtime_error("unsupported data type of shape operator");
switch (d_type)
{
case mir::DataType::FLOAT32:
return constantToShapeT<float>(t);
break;
case mir::DataType::FLOAT64:
return constantToShapeT<double>(t);
break;
case mir::DataType::INT32:
return constantToShapeT<int32_t>(t);
break;
case mir::DataType::INT64:
return constantToShapeT<int64_t>(t);
break;
case mir::DataType::UINT8:
return constantToShapeT<uint8_t>(t);
break;
default:
throw std::runtime_error{"Unknown datatype in constant"};
break;
}
}
mir::DataType onnxDataTypeToMirDataType(onnx::TensorProto::DataType type)
{
switch (type)
{
case onnx::TensorProto_DataType_UINT8:
return mir::DataType::UINT8;
break;
case onnx::TensorProto_DataType_INT32:
return mir::DataType::INT32;
break;
case onnx::TensorProto_DataType_INT64:
return mir::DataType::INT64;
break;
case onnx::TensorProto_DataType_DOUBLE:
return mir::DataType::FLOAT64;
break;
case onnx::TensorProto_DataType_FLOAT:
return mir::DataType::FLOAT32;
break;
case onnx::TensorProto_DataType_UNDEFINED:
throw std::runtime_error{"Undefined input data type not supported"};
break;
default:
throw std::runtime_error{"Unsupported tensor element data type"};
}
}
mir::TensorVariant createTensor(const onnx::TensorProto *tensor)
{
mir::DataType type;
const void *src_data;
mir::Shape shape(tensor->dims_size());
for (int i = 0; i < tensor->dims_size(); ++i)
{
shape.dim(i) = tensor->dims(i);
}
if (tensor->float_data_size() != 0)
{
assert(tensor->data_type() == onnx::TensorProto::FLOAT);
type = mir::DataType::FLOAT32;
src_data = tensor->float_data().data();
}
else if (tensor->double_data_size() != 0)
{
assert(tensor->data_type() == onnx::TensorProto::DOUBLE);
type = mir::DataType::FLOAT64;
src_data = tensor->double_data().data();
}
else if (tensor->int32_data_size() != 0)
{
assert(tensor->data_type() == onnx::TensorProto::INT32);
type = mir::DataType::INT32;
src_data = tensor->int32_data().data();
}
else if (tensor->int64_data_size() != 0)
{
assert(tensor->data_type() == onnx::TensorProto::INT64);
type = mir::DataType::INT64;
src_data = tensor->int64_data().data();
}
else if (tensor->has_raw_data())
{
type = onnxDataTypeToMirDataType((onnx::TensorProto_DataType)tensor->data_type());
src_data = tensor->raw_data().data();
}
else
{
throw std::runtime_error("Invalid data in Proto file, investigate");
}
return mir::TensorVariant({type, shape}, src_data);
}
mir::Operation *foldConstants(mir::Graph *graph, mir::Operation *op)
{
if (op->getType() == mir::Operation::Type::constant ||
op->getType() == mir::Operation::Type::input || op->getType() == mir::Operation::Type::output)
{
// don't fold input, output and constant nodes
return op;
}
if (op->getNumOutputs() != 1)
{
// this operation either have more than 1 output or none at all
return op;
}
bool is_foldable =
std::all_of(op->getInputs().begin(), op->getInputs().end(), [](mir::Operation::Output *out) {
return out->getNode()->getType() == mir::Operation::Type::constant;
});
if (!is_foldable)
return op;
mir_interpreter::MIRInterpreter interpreter;
for (mir::Operation::Output *out : op->getInputs())
{
auto *constant = static_cast<mir::ops::ConstantOp *>(out->getNode());
interpreter.setTensor(out, constant->getValue());
}
op->accept(&interpreter);
const mir::TensorVariant &output = interpreter.getTensor(op->getOutput(0));
return graph->create<mir::ops::ConstantOp>(output);
}
} // namespace mir_onnx
|
